Demoting the Agent: Passive, middle and other voice phenomena

Demoting the Agent Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a platfo...

Author: Benjamin Lyngfelt | Torgrim Solstad

31 downloads 1120 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Demoting the Agent

Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.

Series Editors Werner Abraham

Elly van Gelderen

University of Vienna

Arizona State University

Advisory Editorial Board Cedric Boeckx

Ian Roberts

Harvard University

Cambridge University

Guglielmo Cinque

Ken Safir

University of Venice

Rutgers University, New Brunswick NJ

Günther Grewendorf

Lisa deMena Travis

J.W. Goethe-University, Frankfurt

McGill University

Liliane Haegeman

Sten Vikner

University of Lille, France

University of Aarhus

Hubert Haider

C. Jan-Wouter Zwart

University of Salzburg

University of Groningen

Christer Platzack University of Lund

Volume 96 Demoting the Agent: Passive, middle and other voice phenomena Edited by Benjamin Lyngfelt and Torgrim Solstad

Demoting the Agent Passive, middle and other voice phenomena

Edited by

Benjamin Lyngfelt Göteborg University

Torgrim Solstad University of Oslo

John Benjamins Publishing Company Amsterdam/Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Demoting the agent : passive, middle and other voice phenomena / edited by Benjamin Lyngfelt and Torgrim Solstad. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 96) Includes bibliographical references and indexes. 1. Grammar, Comparative and general--Voice. 2. Grammar, Comparative and general--Syntax. I. Lyngfelt, Benjamin. II. Solstad, Torgrim. P281.D45 2006 415--dc22 isbn 90 272 3360 8 (Hb; alk. paper)

2006047803

© 2006 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents

List of contributors Preface Perspectives on demotion: Introduction to the volume Torgrim Solstad and Benjamin Lyngfelt Semantic and syntactic patterns in Swedish passives Elisabet Engdahl

vii ix 1 21

The Eastern Khanty locative-agent constructions: A functional discourse-pragmatic perspective Andrey Filtchenko

47

Agent back-grounding as a functional domain: Reflexivization and passivization in Czech and Russian Mirjam Fried

83

Invisible arguments: Effects of demotion in Estonian and Finnish Elsi Kaiser and Virve-Anneli Vihman

111

Argument demotion as feature suppression Dalina Kallulli

143

A comparative view of the requirement for adverbial modification in middles 167 Marika Lekakou From passive to active: Syntactic change in progress in Icelandic Joan Maling

197

The relation between information structure, syntactic structure and passive Anneliese Pitz

225

Syntax and semantics of the deontic WANT-passive in Italo-Romance Eva-Maria Remberger

249

Agentivity and the virtual reflexive construction Nola M. Stephens

275

Arguments in middles Thomas Stroik

301



Table of contents

Language index

327

Name index

329

Subject index

331

JB[v.20020404] Prn:18/10/2006; 12:44

F: LA96LC.tex / p.1 (45-222)

List of contributors

Torgrim Solstad University of Oslo ILOS P.O. Box 1003, Blindern N 0315 Oslo Norway [email protected] Benjamin Lyngfelt Göteborg University Dept. of Swedish P.O. Box 200 S 40530 Göteborg Sweden [email protected] Elisabet Engdahl Göteborg University Dept. of Swedish P.O. Box 200 S 40530 Göteborg Sweden [email protected] Andrey Filtchenko Rice University 6100 Main Dept. of Linguistics MS 23 Houston TX 77005 USA [email protected] Dr. Mirjam Fried Princeton University Dept. of Slavic Languages and Literatures East Pyne Hall Princeton NJ 08544 USA [email protected]

Elsi Kaiser USC Dept. of Linguistics 3601 Watt Way, GFS 301 Los Angeles CA 90089-1693 USA [email protected] Virve-Anneli Vihman University of Tartu Ülikooli 18 50090 Tartu Estonia [email protected] Dalina Kallulli Dept. of Linguistics Universität Wien Berggasse 11/3 A-1090 Wien Austria [email protected] Marika Lekakou Joan Muyskenweg 25 NL-1096 CJ Amsterdam The Netherlands [email protected] Joan Maling Division of Behavioral & Cognitive Sciences National Science Foundation, Room 995 4201 Wilson Blvd, Arlington VA 22230 USA [email protected]

JB[v.20020404] Prn:18/10/2006; 12:44

F: LA96LC.tex / p.2 (222-279)

 List of contributors

Anneliese Pitz University of Oslo ILOS P.O. Box 1003, Blindern N 0315 Oslo Norway [email protected] Eva-Maria Remberger Romance Languages Free University Berlin Habelschwerdter Alle 45 D 14195 Berlin Germany [email protected]

Nola Stephens Margaret Jacks Hall, Bldg. 460 Stanford CA 94305-2150 USA [email protected] Thomas Stroik Dept. of English 106 Cockefair-Hall University of Missouri-Kansas City Kansas City MO 64110 USA [email protected]

JB[v.20020404] Prn:18/10/2006; 12:47

F: LA96PR.tex / p.1 (46-121)

Preface

In November 2004 we organised a workshop called “Demoting the agent: Passive and other voice-related phenomena”, hoping to attract researchers from different areas of linguistics with a common interest in passives, middles, implicit agents, etc. Our aim was to provide a forum for bringing together different perspectives on voice and related topics – different theoretical viewpoints, different languages, and different kinds of voice phenomena. As it turned out, the workshop became a great success, thanks to a variety of good talks and to the spirited discussions these prompted. We got exactly the kind of melting-pot we were hoping for. We were also offered to turn the workshop into a book, and here is the result. This volume contains eleven articles based on talks at the workshop. They each make a valuable contribution to the field, offering new data, new analyses, and bringing new light to long-standing issues. In combination, they present a multifaceted and yet coherent picture of the topics at hand. As with any other book, several people apart from the authors and editors were involved in different stages of the publication process. First of all, we would like to thank the editors of the series Linguistik Aktuell/Linguistics Today, Werner Abraham and Elly van Gelderen, for giving us the opportunity to publish this volume and for valuable feedback throughout the whole process. We would further like to thank Kees Vaes at John Benjamins for helping us with all matters practical. Our warm thanks go to Maria Filiouchkina Krave, who has assisted us in this project since we started organising the workshop. Among other things, she provided invaluable help in the final stages of proof-reading. Thanks also to Barbara Clare Dalton for proof-reading some of the manuscripts. Last but not least, the following colleagues have provided invaluable help in the review process: Werner Abraham, Elisabet Engdahl, Mirjam Fried, Elly van Gelderen, Arthur Holmer, Gunnar Hrafn Hrafnbjargarson, Elsi Kaiser, Dalina Kallulli, Marika Lekakou, Joan Maling, Kjartan Ottósson, Anneliese Pitz, Eva-Maria Remberger, Peter Sells, Thomas Stroik, Kjell Johan Sæbø, Ludmila Veselovská and Virve-Anneli Vihman. Thanks are also due to the research project SPRIK – “Språk i kontrast” (Eng. Languages in contrast) – for funding the workshop on which the contributions of this volume are based. Benjamin Lyngfelt’s participation was funded by The Bank of Sweden Tercentenary Foundation through the project “Svenska konstruktioner”

JB[v.20020404] Prn:18/10/2006; 12:47



F: LA96PR.tex / p.2 (121-129)

Preface

(Eng. Swedish constructions). We are also grateful for the financing of our subsequent collaboration by the Faculty of Arts at Göteborg University, Sweden and the Faculty of Humanities at the University of Oslo, Norway. Göteborg and Oslo, September 6th 2006 Benjamin Lyngfelt and Torgrim Solstad

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.1 (49-204)

Perspectives on demotion* Introduction to the volume Torgrim Solstad and Benjamin Lyngfelt

.

Background

The passive has constituted a core area of research in modern linguistics at least since the introduction of transformational grammar. Though there is little controversy of what constitutes the relevant data, several quite different perspectives on passive constructions have been developed. Whereas formal syntacticians and semanticists have focused on the contrast in argument structure and its realisation with regard to the active-passive diathesis, functionalist approaches have mostly dealt with discourse participant prominence and other distinctions related to what one might term information structure. There has also been an increasing interest in more typologically diverse data which has led to research in other voice-related phenomena such as medial constructions and antipassives. These phenomena, too, have prompted a wealth of various analyses. In this volume, we aim to bring together different perspectives on voice, centering around the notion of agent demotion. The articles are all based on talks from the workshop “Demoting the Agent: Passive and Other Voice-Related Phenomena” (Solstad et al. 2004), which was held in November 2004 at the University of Oslo. Some of the papers address these issues from formal perspectives and some adopt a functional view. They deal with a wide range of data, such as passives, middles, reflexives, impersonals and anti-causatives, from a variety of languages. In this introductory article, we do not attempt to provide a state-of-the-art overview neither of voice research in general, nor of any of the subdomains which are often subsumed under that term, such as passives, middles etc. Neither do we provide abstracts of the articles included in this volume, although we refer to the * We would like to thank Werner Abraham, Maia Andréasson, Elisabet Engdahl, Cathrine Fabricius-Hansen, Elly van Gelderen, Elena Karagjosova, Joan Maling, Kerstin Norén, Anneliese Pitz and Eva-Maria Remberger for valuable comments.

JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.2 (204-264)

Torgrim Solstad and Benjamin Lyngfelt

articles where we touch on similar topics. Abstracts are instead included with the papers. What we aim to do is to briefly introduce the topic of agent demotion as such, by highlighting some of its important aspects and present some of the relevant concepts and terminology. Since these have been used in quite diverse ways in the literature, including this volume, it will be of some importance to outline the dimensions along which they are understood. Of course, the views presented here are our own, and we make no claims that they are reflected in all the papers. In Sections 2–4, which are mainly introductory, we relate voice phenomena to more general mechanisms of reduced transitivity. Section 2 introduces some central aspects of transitivity and argument structure; Section 3 gives a brief descriptive overview of how these may be realised within a language, in this case, Swedish; and Section 4 addresses the issue of terminology, focusing on the core notions of agent and demotion. In Sections 5–6, we focus on the passive, first from the viewpoint of information structure (Section 5), and then from a semantic perspective (Section 6). In Section 7, finally, we bring them together and argue in favor of taking both perspectives into account.

. Between active and passive: Reduced transitivity As has become increasingly evident the last few decades, agent demotion and voice concern more than just passivisation. Phenomena that do not quite qualify as either (prototypical) active or passive, including e.g. reflexive- and middle constructions, may be viewed as situated somewhere in between. This domain is sometimes referred to as the middle voice (cf. e.g. Kemmer 1993). However, the notions related to this domain are usually discussed more in relation to event semantics than in terms of voice. A key concept in this regard is transitivity, in the broad sense of e.g. Hopper & Thompson (1980). In addition to the traditional criterion of involving more than one participant, transitivity in this sense also involves factors like action, telicity, volitionality, and affectedness. Such factors, in turn, serve to distinguish other grammatical categories within the general area between active and passive. In this context, also note the close relation between transitivity and voice: Prototypical active sentences are transitive and vice versa, and transitivity is often considered a prerequisite for the active-passive diathesis. Thus, compared to a prototypical active transitive sentence, where an animate agent in control of the situation volitionally performs a specific action on an affected patient, there are several ways in which a sentence could be less transitive without actually being passive. In some cases, this is simply a matter of different argument structures for different verbs. In others, various grammatical devices are

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.3 (264-343)

Perspectives on demotion

employed to mark structures that exhibit low transitivity. In the following, we will give a brief overview of some central notions and typical constructions. Starting with constructions that are intransitive in the traditional sense, i.e. objectless, it is common to distinguish between so-called unergative (1a) and unaccusative (1b) verbs/sentences:1 (1) a. Sandy is swimming. b. Chris is drowning.

The crucial difference between the verbs in (1a) and (1b) is that unergatives have agentive (or actor/causer) subjects, whereas the subject of unaccusatives is typically a patient (or undergoer). Thus, with respect to the active-passive dimension, unergatives are more active-like and unaccusatives are more passive-like. Note, for instance, the so-called unaccusative alternation in (2), which resembles passivisation (3) in that the patient argument is promoted to subject. A crucial difference, however, is that the passive implies an agent, or at least a causer, whereas the unaccusative is not inherently causative in this sense (Levin & Rappaport Hovav 1995; Chierchia 2004; but see also Kallulli, this volume). (2) a. Kim broke the window. b. The window broke. (3) The window was broken (by Kim).

The unaccusative alternation in (2) visibly concerns only argument structure, whereas many other detransitivising operations typically involve some kind of morphological marker.2 For instance, Albanian anti-causatives involve non-active morphology, which seems to be a cross-linguistically common way to indicate reduced transitivity, at least among Indo-European languages (Kallulli, this vol. The rationale behind this terminology, stemming from Perlmutter (1978), is indirectly based on the well-known differencies between ergative-absolutive and nominative-accusative case systems. Very roughly, one might say that unergatives have agent arguments without ergative marking, and unaccusatives have patient arguments without accusative marking. Note, however, that there is also a tradition where the term ergative is used instead of unaccusative (e.g. Burzio 1986; Pitz, this volume). For an overview of the development of these notions and terms, see Kuno & Takami (2004: Chapter 1). . As an example of the opposite, i.e. morphological marking of a more transitive variant, consider lexical pairs such as sjunka (‘sink-non-causative’) and sänka (‘sink-causative’) in Swedish, where a normally non-causative verb is rendered causative through a vowel shift. (i)

Skeppet sjönk. ‘The ship sunk’

(ii) Ubåten/läckan sänkte skeppet. ‘The submarine/leak sunk the ship’



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.4 (343-405)


ume). Another common element in detransitivising constructions are reflexive morphemes. Both types are often multi-functional, being used in several different constructions. This is illustrated in the following section by a brief overview of some uses of passive- and reflexive morphemes in Swedish. For a comparison between some reflexive- and passive constructions in Czech and Russian, see Fried (this volume).

. S-verbs and reflexives in Swedish Passive in Swedish is expressed either periphrastically, with the copula verb bli (‘become’), or morphologically, with an s-suffix (see Engdahl, this volume). It is this s-suffix which concerns us here, since it also has a number of other uses, some of which are fairly close to the passive and others further away from it. These functions exhibit a considerable overlap with those of the reflexive pronoun sig. Since the s-passive developed historically from the reflexive, such an overlap is only to be expected. To begin with the more passive-like uses, both the s-form and the reflexive are used in a number of constructions that may be called medial.3 These have been characterised in the literature as having a patient or patient-like subject, but – unlike passives – no perceived agent (Teleman et al. 1999: Vol. 1: 197 & Vol. 2: 556).4 They differ from unaccusatives in that they require morphological marking – in Swedish either the s-suffix or a reflexive, as illustrated in (4):5 (4) (Grundprincipen är [...] ett enkelt och inledande penseldrag) ‘The basic principle is [...] one simple commencing stroke of the brush’

This type of alternation is no longer productive in Swedish, but has been lexicalised for a number of verbs. The morphologically marked form is the causative, which also derives historically from the non-causative variant. . An alternative term would be middle. That term, however, is usually associated with a specific kind of medial construction, which is treated extensively by Lekakou (this volume) and Stroik (this volume). An English example is given in (i): (i)

This bread cuts easily.

. This does not quite hold for medial constructions such as middles (Lekakou, this volume; Stroik, this volume) and virtual reflexives (Stephens, this volume), which arguably include agents. What the criterion really amounts to is the possibility of an agentive adjunct, such as a by-phrase (av-phrase in Swedish), which is a typical test for passives (see e.g. Maling, this volume). . In the following examples, reflexive morphemes will be glossed sig, which is the unmarked form for the reflexive in Swedish.

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.5 (405-472)


a.

som sedan ska förgrenas i det oändliga. which then shall ramify-s in the infinite b. som sedan ska förgrena sig i det oändliga. which then shall ramify sig in the infinite ‘which then ramifies / is ramified infinitely.’

Swedish also allows so-called impersonal passives, as in (5). Another kind of impersonal use of s-forms and reflexives is exemplified in (6), and regular transitive and passive uses of the same verb vattna (‘water’) are shown in (7): (5) Det snackas mycket. expletive talk-s much ‘There’s a lot of talk.’ (6) a.

Det vattnas i munnen. expl water-s in the.mouth b. Det vattnar sig i munnen. expl water sig in the.mouth ‘My mouth waters.’ / ‘It makes my mouth water.’

(7) a.

Jag vattnade blommorna igår. I watered the.flowers yesterday ‘I watered the flowers yesterday.’ b. Blommorna vattnades igår. the.flowers watered-s yesterday ‘The flowers got watered yesterday.’

The medial constructions in (4) and (6) can be viewed as a blend between passives and unaccusatives, and may correspond to either, depending on the context. It should be emphasized that the distributional overlap between reflexives and s-verbs in Swedish medials is only partial. Their interpretations are sometimes basically synonymous, sometimes quite distinct, to some extent lexically determined by the verb. A medial construction exclusive to reflexives is the so-called virtual reflexive (for a treatment of the corresponding construction in English, see Stephens, this volume). Unlike most other reflexives in Swedish, this construction requires the pronoun själv (‘self ’), which also has to be stressed. (8) Det var en sådan pjäs som skrev sig själv, som en blixt bara, över It was a such play that wrote sig self, like a flash only, over en vår. a spring ‘It was the kind of play that wrote itself, just like a flash, over spring.’

Turning to more active uses, both reflexives and s-forms may indicate reciprocity, as in (9) below. There are also quite a few lexicalised reflexive verbs and s-verbs,



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.6 (472-528)


such as förhasta sig (‘be hasty’), bestämma sig (‘decide’), finnas (‘exist’, literally ‘find-s’) and hoppas (‘hope’). (9) a.

Parterna har svårt att enas i frågan. the.parties have difficult to unite-s in the.issue b. Parterna har svårt att ena sig i frågan. the.parties have difficult to unite sig in the.issue ‘It is hard for the parties to agree on this issue.’

In addition, reflexives and s-forms both appear in several more or less objectsuppressing constructions, although they exhibit less co-variance among these than among the medial constructions. For instance, s-forms are often used with a so-called absolute function (cf. Teleman et al. 1999: Vol. II, pp. 555f.): (10) Kalle retas. Kalle tease-s a. ‘Kalle is teasing someone/me.’ b. ‘Kalle is a tease.’

Characteristic of this absolute construction is object suppression. Whereas the corresponding verb without -s requires an object (reta någon – ‘tease someone’), the s-form is intransitive in this respect. Also note that this construction can either refer to specific events (10a) or receive a dispositional reading (10b). As for reflexive elements, those that correspond to arguments (typically a patient or a benefactor argument coreferent with the subject) are of course more transitive than non-argument reflexives.6 Both are very common in Swedish, both in the sense that they are employed in many different constructions and in the sense that several constructions are considerably more productive with reflexives than with full NP arguments, or even other pronouns, for instance the resultative constructions in (11): (11) a.

Han försökte köpa sig / henne / marken fri. he tried buy sig / her / the.land free ‘He tried to buy himself/her/the land free.’ (cf. Jansson 2006) b. Han ölade sig / ?? henne / ?? kvällen odödlig. he aled SIG / her / the.night immortal ‘He drunk himself immortal on ale.’ (not ?? her or ?? the night)

There is also a number of exclusively reflexive constructions, such as the Swedish correspondent to the English way-construction (cf. Goldberg 1995), illustrated in (12): . Note, however, that even argument reflexives may be viewed as less transitive than ordinary object constructions, since subject and object are referentially non-distinct.

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.7 (528-605)


(12) Han grävde sig ut. He dug sig out ‘He dug his way out.’

A semi-productive construction bordering between argument and non-argument reflexives is illustrated in (13), where the reflexive obscures the reference of the object without eliminating it (cf. the notion of deobjectification, e.g. Fried, this volume). However, we will not explore the wealth of agentive reflexive constructions in Swedish further, since they do not involve agent demotion. (13) Han hade svårt att uttrycka sig / vad han kände. He had difficult to express sig / what he felt ‘He had a hard time expressing himself / how he felt.’

Swedish also employs a version of the much discussed middle construction, although perhaps restricted to just one verb: sälja (‘sell’), illustrated in (14). However, most instances lack the generic or dispositional property typical of middles (cf. Lekakou, this volume; Stroik, this volume), and might therefore be more accurately characterised as unaccusatives.7 (14) Sådana böcker säljer bra. ‘Such books sell well’

On the other hand, there seems to be different kinds of middles. Although middles in English lack morphological marking, German middles require a reflexive and some Romance languages employ imperfective morphology. Also, Greek middles differ from the Germanic middles in allowing agentive adjuncts (Lekakou, this volume). Finally, it should be mentioned that the s-passive also has acquired other uses, such as the modal passive in (15) (see also Engdahl, this volume). For an analysis of modal passives in Italoromance, see Remberger (this volume). In Icelandic, there even seems to be a passive construction on its way to develop into an active one (Maling, this volume). (15) Mjölken förvaras i högst +8◦ C. the.milk keep-s in at.most +8◦ C ‘The milk should not be kept warmer than +8◦ C.’

. Whether such cases fit within the (somewhat wider) notion of middle voice, whose main characteristic has been claimed to be subject affectedness (cf. Kemmer 1993; Lyons 1968: 373) is of course a different issue (are the books in (14) ‘affected’ in the relevant sense?) which will not be explored here.



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.8 (605-656)


. A note on terminology: What does agent demotion mean? The constructions in the former section are all less transitive than typical active sentences. Whether they are also cases of agent demotion, however, is quite another matter. For instance, the absolute construction in (10) and the reflexives in (11)–(13) are not, since they all have agentive subjects. The reduction in transitivity applies to the patient (or, in the dispositional reading, to the action as such) – not to the agent. As for the passive and medial constructions, the issue is a matter of definition. It all depends on one’s notion of both demotion and agentivity. Starting with demotion, this term may apply to several different levels. In Relational Grammar, demotion applies to a hierarchy of grammatical functions (as the opposite of advancement). Thus, passives involve agent demotion, since the agent is demoted from subject to (optional) oblique (cf. e.g. Perlmutter & Postal 1983).8 Non-agentive constructions, however, do not. Unaccusatives and the medial constructions in (4) and (6) simply have no agent arguments to demote. On the other hand, reduction from presence to absence may also be called demotion, with respect to semantics as well as surface realisation. On this view, the unaccusative alternation illustrated in (2) involves agent demotion on a semantic level, through a shift from agentive to non-agentive. This form of demotion does not apply to passives, where the agent in general is not eliminated semantically (this will be modified somewhat in Section 6). Concerning surface representation, rendering the agent implicit may also be seen as a form of demotion. Kaiser & Vihman (this volume) effectively treat demotion as absence, in the sense of being available for various linguistic operations. They apply several tests to determine whether implicit arguments in some Finnic constructions are demoted on the semantic, syntactic, or discourse level. The next question concerns what is actually demoted (in whatever sense). An Agent may be roughly defined as an animate entity that intentionally performs an action. However, not all demoted arguments are Agents in this sense. In passives, for instance, the argument demoted from subject to (optional) oblique may be a Cause (or an Instrument) as well as an Agent (cf. e.g. Marantz 1984; Williams 1981). The crucial difference is that a Cause, which may be animate or inanimate,

. In impersonal passives, the demotion goes further in some languages, where the agent cannot even appear in an agent phrase, as in English by-phrases or Swedish av-phrases, although it is semantically present and may appear in a circumstantial adverbial. In other languages, like Dutch, German and Norwegian, however, agent phrases may occur in impersonal passives. A more moderate demotion occurs in presentational constructions, where the subject is demoted to object position (with an expletive in the subject position) but retains its subject function (see Engdahl, this volume).

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.9 (656-712)


does not involve intentionality – at least not with respect to the event at hand (Kallulli, this volume). by my neighbour (16) The car was destroyed by a stone In example (16) my neighbour may be either an Agent or a Cause and a stone may be either an Instrument or a Cause, depending on whether the destruction is intentional or not. What complicates matters is that the term agent is sometimes used in a broader sense, which subsumes both Agent and Cause. This is especially common when the term does not refer to the semantic role as such but to the argument fulfilling that role: Causes are more commonly included (implicitly) in utterances like “In passive constructions, the agent is demoted to an optional by-phrase” than in “This verb requires its subject to be an agent”. To complicate things further, these notions (or very similar notions) are also referred to by other terms, like Actor, Agency, or Causer, and all of these terms are used somewhat differently by different authors. As a superordinate term which subsumes Agent, Cause and, where relevant, Instrument, Dowty (1991) introduced the notion of proto-agent. From a semantic perspective, the distinction between Agent and Cause may be viewed as roughly corresponding to the distinction between individuals (or entities) and events. An Agent is an individual that performs an action, whereas a Cause is an event that brings about another event (or, rather, a relation between events).9 This view will be explored further in Section 6. In the following sections, we will turn to the passive, as viewed from an information structural perspective in Section 5 and from a semantic perspective in Section 6. Finally, we will try to bring these two together in Section 7.

. Passive and information structure A common view on passive and in many cases on voice, most prominently in functional accounts, is to relate it to information structure. Especially in languages with a relatively fixed word order, the passive allows – compared to the active – a different linearisation of the arguments of a predicate. Therefore, in terms of information structure, demoting the agent and promoting the patient has been seen as two faces of the same phenomenon. Put differently: From this perspective, the passive is a means of relating the information of a sentence to its context in a dif. Note that this is the case even when the causer is animate (as in Chris (accidentally) squished the toad), since the Cause from the semantic perspective is not the animate entity as such, but some causing action performed by that entity.



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.10 (712-773)


ferent way than a corresponding active sentence. A quotation from Judith Aissen (Aissen 1999: 687) sums up this view: There is general agreement in the functional syntax literature that the choice between active and passive in English is primarily determined by the relative prominence of agent and patient. Crudely put: The passive is used if the patient is more prominent than the agent; the active is used if either the agent is more prominent than the patient, or if neither has greater prominence.

We will not explore the notion of prominence here, but merely conclude that agent demotion and patient promotion, from the information structural point of view, are processes which take place because of the demands of context.10 Demotion and promotion are first and foremost determined by linearisation and discourse processing. If a discourse referent was introduced in the preceding context, is to be kept in the center of attention, it might be preferable to refer to this referent in a syntactic topicalisation position, often influenced also by factors such as definiteness, animacy, person etc. (cf. e.g. Aissen 1999). This places the passive in a paradigm with other mechanisms of information partitioning such as clefting, inversion, syntactic topicalisation and intonational focusing. See Filtchenko (this volume) and Pitz (this volume) for a discussion of some of the parallels and differences here. The following examples are intended to illustrate how the passive interacts with various mechanisms of information partitioning as mentioned above. In (17), from a Norwegian original text, the passive has to be chosen in order to make the sequence of sentences coherent (the relevant parts are italicised in the original and its corresponding translation): (17) Foran den svartsvidde branntomten var to kvinner avbildet [. . . ]. Billedteksten lød: Kontordame Elise Blom og sekretær Alvhilde Pedersen, som begge kom uskadete fra ulykken, foran den nedbrente fabrikkbygningen på Fjøsangerveien. – Fotografiet ble ledsaget av intervjuer, blant annet med de to på bildet, der alle var enige om at eksplosjonen kom fullstendig uventet og “som et sjokk”, ifølge frøken Pedersen. ‘Two women were photographed standing in front of the blackened site of the fire [. . . ]. The caption read: “Office clerk Elise Blom and secretary Alvhilde Pedersen, who both escaped from the fire without injury, in front of the burned-down factory building on Fjøsangerveien.” The photograph was accompanied by interviews with, among others, the two women in the photo-

. The direction of causality has been the matter of some dispute, though, i.e. whether agent demotion is the consequence of patient promotion or vice versa, cf. Comrie (1977) and Blevins (2003).

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.11 (773-834)


graph; all those interviewed confirmed unanimously that the explosion had been completely unexpected and a shock, according to Miss Pedersen.’ a. Alternatively: [. . . ] ?? Interviews accompanied the photograph . . .

Here, an active sentence as indicated in (17a) simply would not constitute a coherent continuation for the definite topic fotografiet (‘the photograph’).11 This is the kind of passivisation addressed by the generalisation in the above quote from Aissen. However, other means of information structuring may lead the demoted agent to be more prominent than the topicalised promoted patient. The German example in (18a) shows a case where the demoted agent is made prominent through intonational focusing (marked by small capitals): (18) a.

Von seinem bruder wurde Abel ermordet, und nicht von seinem By his brother was Abel murdered and not by his Vater. father. ‘Abel was murdered by his brother, not by his father.’ b. *Sein Bruder ermordete Abel, nicht sein Vater. His brother.nom murdered Abel, not his father. Intended: ‘Abel was murdered by his brother.’

In a context where the agent phrase including a possessive pronoun in determiner position is to be topicalised, the alternative in (18b) is excluded where (18a) may naturally occur, because there can be no correct binding for the possessive pronoun sein (‘his’) in (18b). There are also examples where the prominence differentiation as defined above does not apply. One such case is the passivisation of one-place predicates like German tanzen (‘dance’), where there is no difference in relative prominence between an agent and a patient: (19) a.

Es wurde die ganze Nacht getanzt. It was the whole night danced. ‘There was dancing going on all night.’

As discussed by e.g. Engdahl (this volume), it is reasonable to assume that for verbs like dance, the agent is less prominent in a passive sentence when not overtly expressed than in a corresponding active sentence, where it has to occur. Thus, the notion of prominence would also be relevant when comparing different realisations of one and the same participant and not only in competition with other participants. In such cases, the passive can be said to have the effect of focusing on the event in question. . Though the noun photograph does not occur in the preceding context, it can be seen as a topic due to a bridging effect (Clark 1977; Asher & Lascarides 1998).



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.12 (834-898)


In terms of demotion, the Agent will still be present in such cases, although not in focus of the event description provided by the passive. In the next section, we will broaden the perspective on how demotion relates to the semantics of the predicate in question.

. On the semantics of voice Traditionally, much of both functional and formal lingustic approaches have viewed passives and actives as semantically equal. In formal analyses, it is usually assumed that actives and passives do not differ in their truth conditions (Kratzer 1996), whereas in the functional literature, the absence or presence of an agent has been interpreted in pragmatic terms. Thus, Givón (1990: 567–569) gives a list of circumstances under which an agent is often omitted. The agent might for example be unknown (20a), anaphorically predictable (20b), stereotypical (20c), or predictable as the author of the text (20d): (20) a. He was shot dead in the big rally. b. The children ran into the room, and soon all the toys were being played with. c. The airplane was landed safely. d. It was shown in Chapter 4 that the semantics of cause . . .

See Kaiser & Vihman (this volume) for a discussion of how the implicitness of an agent may have different consequences for its discourse referent status. Since the importance or activation status of an agent does not affect the truth conditions of a sentence, the lack of an agent as in the above sentences would be seen as a pragmatic phenomenon within a formal approach as well. Though no agent is mentioned, there is a sense in which it is interpreted as necessarily present at some level in all the sentences (20a–d). Some of the standard arguments for such a view, mainly stemming from the formal literature, rely on facts found with binding phenomena and purpose clauses. Concerning binding, it can be noted that even though passives allow us not to mention agents explicitly, an example like the following can be seen as evidence that they are present anyway (Frey 1993: 158): (21) a.

Und da wurde einander nicht nur zugeprostet. And there was reciprocal not only toasted ‘And they didn’t only toast at each other.‘

Although the agent is not present on the surface in (21a), the reciprocal element einander (‘each other’) demands an antecedent for its meaningful interpretation, and thus an agent has to be assumed to be implicitly present.

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.13 (898-960)


Another such test commonly cited in the literature involves combination with purpose clauses (see Kaiser & Vihman, this volume; Kallulli, this volume), which is meant to show that actives and passives (22a-b) differ from unaccusatives (22c) concerning the availability of an agent: (22) a. Theyj sank the ship, to proj collect the insurance money. b. The shipi was sunk (by themj ), to proj collect the insurance money. c. *The ship sank, to pro? collect the insurance money.

In (22b), pro in the purpose clause is obligatorily coreferent with the agent of the passive matrix verb, even when the agent is implicit.12 It is clear that whenever a purpose clause occurs in an acceptable sentence, there has to be an agent controlling the eventuality described in the purpose clause. We cannot discuss these data in any detail here, but merely want to point to evidence which indicates that the purpose clause examples cannot always be applied straightforwardly to show that a matrix clause includes an agent of its own, since the origin of the agent can not always be assumed to be the matrix clause, cf. (23): (23) Dieses Auto hat den niedrigsten Luftwiderstandsbeiwert in seiner Klasse. This car has the lowest wind resistance of its class. Windschutzscheibe und Rückfenster sind stark geneigt. Das Heck Wind shield and rear window are strongly sloped. The rear end hingegen ist hoch, um die Aerodynamik weiter zu verstärken. however is high, for the aerodynamics further to strengthen. ‘This car has the lowest wind resistance of its class. The wind shield and the rear window have a considerable angle of slope. The rear end, however, is high, in order to add further to the aerodynamics.’

For a stative predicate such as (be) high, it is not very plausible to assume the existence of an (implicit) agent. Still, as mentioned, the acceptability of (23) demands that there is an agent in a control relation with (the pro of) the purpose clause. The state described in the matrix clause is interpreted as the resultant state of some causing event including an agent. In the case of (23), it can be argued that the purpose clause itself is the most plausible candidate for introducing an agent in the matrix clause. From a semantic viewpoint, a possible analysis for the introduction of an agent in examples such as (23) is to assume some mechanism of accommodation (Sæbø 1991: 931), whereby the presence of a purpose clause with a stative predicate such as in (23) leads to a context modification involving . As for the syntactic representation of implicit agents in passives, we do not propose an analysis in this introduction. See Sternefeld (1995) for an analysis where it is represented as a (little) pro, whereas Remberger (this volume) assumes it to be a (big) pro. A more neutral approach is pursued in Lyngfelt (2002).



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.14 (960-1025)


the interpretation of an implict agent having intentionally brought about the state in question.13 Still, the view that passives always include an agent is in general relatively unproblematic, as long as one disregards the distinction between Agent and Cause. However, this distinction is highly relevant for several aspects of the passive, both theoretical and distributional. Passives where the demoted element is a Cause rather than an Agent may be termed event-type passives (Solstad forthcoming; Sæbø & Solstad 2005). Such passives may lack an Agent altogether. The following examples involving causative predicates illustrate this: (24) a.

Before he could grab his parachute, the plane exploded and he was thrown into the air. b. As the silk began to stream out, he was flung into the top branches of an oak tree. c. The car was destroyed (by the blast).

In (24), no agent can be made out as immediately responsible for the caused events. In (24a–b) it is even difficult to name the exact event responsible for the caused events, which might be a motivation for choosing the passive in such a context. Of course, if it is possible to single out such an event and refer to it by a noun phrase, it can be included, as in (24c). The following point can be made concerning the distinction between Agent and Cause as described in Section 4 and its relation to event-type passives. If a by-phrase in combination with causative predicates would be possible in English, but is left out, it is not always possible to predict whether the implicit element should be seen as an Agent or a Cause. Thus there is no straightforward semantic correspondence to the syntactic level, where the various kinds of individuals and events being able to occur both as subjects in actives and in the by-phrase in passives behave similarly. This has to do with the fact that the notion of Agent involves an individual, whereas a Cause may include both individuals and events.14 As an illustration of the difficulties which the event-type passive constitutes for syntactic and semantic theories of voice, consider the following (neo-Davidsonian) representation for the semantics of agent assumed by Kratzer (1996) to be the same for passives and actives: (25) λxλe.agent(x)(e) . More accurately, the stative predicate is coerced into a resultant state, whereas the agent is accommodated. We cannot go into a discussion of the formal basis and ordering of these processes here. . Note the difference to the formal-semantic cause, which is assumed to be a relation between events or propositions (cf. Dowty 1979).

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.15 (1025-1104)


If one allows the agent element to be as broadly defined as including the individuals and objects in the scope of Dowty’s (1991) proto-agent, this can be applied in an analysis of examples where the individual is either an Agent, a Cause or an Instrument as described in Section 4. However, the representation in (25) cannot be applied in the case of event-type passives. The reason for this is that the representation in (25) takes entities (including individuals and objects) as arguments (represented by the variable x), whereas the argument in the event-type passives such as in (24) is an event. Importantly, the class of event-type passives is only found with causative predicates. It is, of course, possible to find passives of destroy with an individual as agent. What is important regarding the above discussion is that it is also possible for destroy to be interpreted as not having an agent at all. The reason for these two possibilities being available to causatives is that a causative expresses a relation between events, and that certain events may be construed as not having an agent. In relation to this, it can be observed that causative predicates can be divided into (at least) three classes (Solstad forthcoming), inherently agentive (26), nonagentive (27) and agentivity-neutral (28):    by the soldiers  (26) The enemy was executed ∗ by a torpedo ∗  by the lightning ∗   by the soldiers  (27) The case was washed ashore ∗ by a torpedo   by the undercurrents   by the soldiers  (28) The ship was destroyed by a torpedo   by the undercurrents Inherently agentive predicates like execute (26) can only be interpreted as including an intentionally acting agent. This lies in the nature of the event of executing. Such predicates thus never allow a reading where some non-controlled event causes the death of the patient. On the other hand, a non-agentive predicate like wash ashore (a directional) as in (27), can never be intepreted as including an agent. Only natural forces, which are seen as events here, can occur as active subjects and in the passive by-phrase, specifying the cause relation of the predicate. Only event-type passives are allowed with these predicates. In the third, agentivity-neutral class, illustrated in (28), the predicates may vary: they may include intentionally or unintentionally acting agents, and they may occur in event-type passives. The differentiation outlined above between event-type and non-event-type passives can be illustrated by means of German data (Solstad forthcoming). One of the main functions of the German preposition durch (‘by’/‘by means of ’) is to mark its complement as the causing event in a causal relation. Durch may also be



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.16 (1104-1151)


used as an equivalent to the passive by-phrase in English, with a limited distribution, von being the unmarked agent phrase preposition in German. Importantly, the complement noun phrase of durch always receives an event interpretation: (29) a.

Er zerstörte das Auto durch einen Sprengkörper. He destroyed the car through indef explosive body. ‘He destroyed the car by means of an explosive body.’ b. Das Auto wurde von ihm durch einen Sprengkörper zerstört. The car was by him through indef explosive body destroyed. ‘The car was destroyed by him by means of an explosive body.’ c. Das Auto wurde durch einen Sprengkörper zerstört. The car was through indef explosive body destroyed. by ‘The car was destroyed an explosive body.’ by means of

In (29c), it can be observed that durch allows an interpretational variance, which is not available in the active sentence (29a) or the passive one including an agent (29b). The differentiation in the translation as being either by or by means of is intended to reflect the double nature of the durch-phrase, by means of presuming intentionality, and by allowing lack of intentionality.15 Example (29c) has a possible interpretation where no agent is involved, where the event of an explosion alone causes the destruction. In (29c), an Agent can be interpreted as being present, but what is important here is that it does not have to. If we demand that an agent (of the entity type) always has to be present implicitly, this behaviour cannot be explained.

. Semantics and pragmatics It was noted in Section 5 that the information structural or pragmatic aspect of the passive sets it in a context where it interacts with other mechanisms of discourse prominence, such as phonological stress, clefting, topicalisation etc. In example (18), the interplay of voice and intonational focus was illustrated. From a semantic perspective, however, the passive competes with reflexives and other constructions where no agent is expressed obligatorily, such as anticausatives. Anticausatives are predicates which are often thought to be derived from causative ones (cf. e.g. Wunderlich 1997: 54ff.), where the anticausative expresses a change of state without a causing event. In many Indo-European languages, they have reflexive morphology, cf. (31a) vs. (31b). These have not been . In terms of semantic roles, the different readings of (29c) concerns whether the explosive body is interpreted as a Cause or an Instrument.

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.17 (1151-1217)


studied extensively from an information structure perspective, but have rather been seen as agent demotion constructions from a more semantic viewpoint, where genericity and the aforementioned lack of intentionality (as with the impossibility of combining the anticausative variant of sink with a purpose sentence as in (22c)) are of importance. The relation between passives and reflexives can be illustrated by means of the following examples: (30) a.

Das Haus wurde durch eine Explosion zerstört. The house was through an explosion destroyed. ‘The house was destroyed by an explosion.’ b. *Das Haus zerstörte sich. The house destroyed refl. Intended: ‘The house destroyed itself.’

(31) a.

Die Tür wurde geöffnet (um zu lüften). The door was opened for to let in some air. ‘The door was opened to air the room.’ b. Die Tür öffnete sich (*um zu lüften). The door opened refl for to let in some air. Intended: ‘The door opened to air the room.’

For predicates where no reflexive or anticausative variant is available, like zerstören (‘destroy’) in (30), we expect the passive to be more open to non-agentive readings, whereas a passive for a verb like öffnen (‘open’) in (31) should be more agentive, since an anticausative variant is available as a possible way of expression. Though example (31a), without the purpose clause extension, allows non-intentional readings, (30a) has a stronger tendency for such an interpretation. We would claim that this is related to the fact that there is a clearly non-intentional expression available for the predicate öffnen, namely the reflexive, which is not possible for zerstören. What is important to note here is that while it is not the case that the passive variant of öffnen cannot be used to express non-volitional or non-controlled causation, the anticausative variant can only have such a non-volitional intepretation. Thus, in a context where there could be doubt as to whether something was caused intentionally by someone or only by some force of nature, an agentless passive would be a more natural choice, whereas if one wants to express that the opening event was indeed not caused by someone, but by some unknown or unspecified force, one should use the reflexive variant, since the passive allows an intentional, controlled reading. For öffnen, there is a reflexive variant available for the non-intentional reading, and thus the passive is dispreferred in such cases. Zerstören, however, lacks that possibility, and therefore the passive will have to do. Consequently in passive



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.18 (1217-1311)


clauses, the non-intentional reading is more readily available for zerstören than for öffnen. Note that this is seen as a somewhat different discussion than the one pertaining to the question why there can’t be a (reflexive) anticausative variant of zerstören (Härtl 2003). What is relevant here, is that the perfectly acceptable passive variant of the two predicates has different interpretational preferences and ranges due to the lack or availability of a reflexive variant. A bidirectional optimalitytheoretic approach (Blutner 2000; Wilson 2001) seems to suggest itself here, in which there is competition between forms of expression and their corresponding interpretations (Solstad forthcoming). We have tried to illustrate here how passive – and voice in general – belongs to two different paradigms, one of a pragmatic nature and one semantic. These two sides may seem to relate to different phenomena: either as part of an information distributional paradigm including syntactic topicalisation, intonational patterns, inversion etc., or as part of an event-semantic paradigm including middles and other non-transitive constructions. However, these may involve both semantic and pragmatic factors as well. Although our illustration of this claim focused on the passive, it seems likely that topicalisation etc. involves semantics as well as information structure – and that middles, reflexives etc. are often contextually motivated, not just issues of event semantics. There has been a tendency in the literature to focus either on pragmatic or semantic aspects. We believe that approaches taking both perspectives into consideration would benefit the understanding of voice and related phenomena.

References Aissen, Judith (1999). Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory, 17(4), 673–711. Asher, Nicholas & Alex Lascarides (1998). Bridging. Journal of Semantics, 15(1), 83–113. Blevins, James P. (2003). Passives and impersonals. Journal of Linguistics, 39(3), 473–520. Blutner, Reinhard (2000). Some aspects of optimality in natural language interpretation. Journal of Semantics, 17(3), 189–216. Burzio, Luigi (1986). Italian Syntax. A Government and Binding Approach. Dordrecht: Foris. Chierchia, Gennaro (2004). A semantics for unaccusatives and its syntactic consequences. In Artemis Alexiadou, Elena Anagnostopoulou & Martin Everaert (Eds.), The Unaccusativity Puzzle: Explorations of the Syntax–Lexicon Interface (pp. 22–59). Oxford: Oxford University Press. Clark, Herbert (1977). Bridging. In Philip N. Johnson-Laird & P. C. Wason (Eds.), Thinking: Readings in Cognitive Science (pp. 98–113). Cambridge: Cambridge University Press. Comrie, Bernard (1977). In defense of spontaneous demotion: The impersonal passive. In Peter Cole & Jerrold M. Sadock (Eds.), Grammatical Relations [= Syntax & Semantics 8] (pp. 47–58). New York: Academic Press. Dowty, David R. (1979). Word Meaning and Montague Grammar. Dordrecht: D. Reidel.

JB[v.20020404] Prn:19/10/2006; 13:49

F: LA9601.tex / p.19 (1311-1470)


Dowty, David R. (1991). Thematic proto-roles and argument selection. Language, 67(3), 547– 619. Frey, Werner (1993). Syntaktische Bedingungen für die semantische Interpretation (Syntactic conditions on semantic interpretation). Berlin: Akademie-Verlag. Givón, Talmy (1990). Syntax: A Functional-Typological Introduction, Vol. II. Amsterdam: John Benjamins. Goldberg, Adele (1995). Constructions. A Construction Grammar Approach to Argument Structure. Chicago, IL: University of Chicago Press. Härtl, Holden (2003). Conceptual and grammatical characteristics of argument alternations: The case of decausative verbs. Linguistics, 41(5), 883–916. Hopper, Paul J. & Sandra A. Thompson (1980). Transitivity in grammar and discourse. Language, 56(2), 251–299. Jansson, Håkan (2006). Har du ölat dig odödlig? En undersökning av resultativkonstruktioner i svenskan (Have you aled yourself immortal? An investigation of resultative constructions in Swedish). M.A. thesis, Dept. of Swedish, Göteborg University. Kemmer, Suzanne (1993). The Middle Voice [= Typological Studies in Language 23]. Amsterdam: John Benjamins. Kratzer, Angelika (1996). Severing the external argument from its verb. In Johan Rooryck & Laurie Zaring (Eds.), Phrase Structure and the Lexicon (pp. 109–137). Dordrecht: Kluwer. Kuno, Susumo & Ken-ichi Takami (2004). Functional Constraints in Grammar. On the Unergative–Unaccusative Distinction. Amsterdam: John Benjamins. Levin, Beth & Malka Rappaport Hovav (1995). Unaccusativity: At the Syntax–Lexical Semantics Interface. Cambridge, MA: The MIT Press. Lyngfelt, Benjamin (2002). Kontroll i svenskan. Den optimala tolkningen av infinitivers tankesubjekt (Control in Swedish. The optimal interpretation of implicit subjects in infinitival clauses). Göteborg: Acta Universitatis Gothoburgensis. Lyons, John (1968). Introduction to Theoretical Linguistics. London: Cambridge University Press. Marantz, Alec (1984). On the Nature of Grammatical Relations. Cambridge, MA: The MIT Press. Perlmutter, David M. (1978). Impersonal passives and the unaccusative hypothesis. In J. Jaeger et al. (Eds.), Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society (pp. 157– 189). University of California at Berkeley. Perlmutter, David M. & Paul M. Postal (1983). Some proposed laws of basic clause structure. In David M. Perlmutter (Ed.), Studies in Relational Grammar, Vol. 1 (pp. 81–128). Chicago, IL: The University of Chicago Press. Solstad, Torgrim (Forthcoming). Mehrdeutigkeit und Kontexteinfluss: Die Spezifikation kausaler Relationen am Beispiel von ‘durch’ (Polysemy and the influence of context: The specification of causal relations). Ph.D. dissertation, University of Oslo. Solstad, Torgrim, Benjamin Lyngfelt, & Maria F. Krave (Eds.). (2004). Demoting the Agent: Passive and Other Voice-Related Phenomena. Pre-proceedings from the 2004 workshop with the same title. University of Oslo. Sternefeld, Wolfgang (1995). Voice phrases and their specifiers. FAS Papers in Linguistics, 3, 48– 85. Sæbø, Kjell Johan (1991). Causality and finality. In Arnim von Stechow & Dieter Wunderlich (Eds.), Semantics: An International Handbook of Contemporary Research (pp. 930–939). Berlin: Mouton de Gruyter. Sæbø, Kjell Johan & Torgrim Solstad (2005). The mystery of the missing agent: Reconstructing passives. Handout from a talk given at the 14th Wuppertaler Linguistic Colloquium.



JB[v.20020404] Prn:19/10/2006; 13:49



F: LA9601.tex / p.20 (1470-1497)


Teleman, Ulf, Staffan Hellberg, & Erik Andersson (1999). Svenska Akademiens grammatik (The Swedish Academy grammar). Stockholm: Norstedts. Williams, Edwin (1981). Argument structure and morphology. The Linguistic Review, 1, 81–114. Wilson, Colin (2001). Bidirectional optimization and the theory of anaphora. In Géraldine Legendre, Jane Grimshaw, & Sten Vikner (Eds.), Optimality-Theoretic Syntax (pp. 466–507). Cambridge, MA: The MIT Press. Wunderlich, Dieter (1997). Cause and the structure of verbs. Linguistic Inquiry, 28(1), 27–68.

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.1 (46-136)

Semantic and syntactic patterns in Swedish passives* Elisabet Engdahl Swedish has two productive ways of forming the passive, a morphological way using the suffix -s and a periphrastic way using the copula bli (become) together with a passive participle. This article discusses the way the two passive forms are used in written Swedish. A quantitative study reveals substantial differences in the distribution of the two forms with the s-form emerging as the most common. The s-passive is shown to be the unmarked passive form in Swedish in the sense that it is used more often, in more contexts and with more verbs. The bli-passive is a marked option and often requires that the subject has some control over the described situation. The bli-passive is also used when the inception or the completion of the event are in focus. The article also discusses the use of passives in impersonal constructions which are common in Swedish. In these constructions, the agent may not be expressed overtly but is understood to be human.

Most languages have a way of forming passives. The mainland Scandinavian languages, Danish, Norwegian and Swedish, actually have two ways, both used in both spoken and written language. It turns out, not very surprisingly, that the two forms are used under different conditions. In this article I will concentrate on how the two forms are used in present-day written Swedish, occasionally comparing the Swedish patterns to the ones found in Danish and Norwegian. After having introduced the two passives in Section 1, I turn to a discussion of when each form is used in Section 2. Then follows a quantitative overview of the distribution of the two passives, Section 3, and a discussion of certain restrictions pertaining to the * Most of the original research for this article was carried out in the NORDSEM project funded by the Joint Committee for the Nordic Research Councils for the Humanities. I have benefitted from discussions with other project members, in particular Robin Cooper, Kjell Johan Sæbø, Finn Sørensen (†) and Carl Vikner, and other Scandinavian colleagues, in particular Hans-Olav Enger and Lars Heltoft. I thank Jan Terje Faarlund and Helge Lødrup for valuable comments in Oslo and Peter Andersson, Ida Larsson, Benjamin Lyngfelt, Torgrim Solstad, Lilja Øvrelid and two anonymous reviewers for helpful comments on earlier versions.

JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.2 (136-197)

Elisabet Engdahl

subject, Section 4. In Section 5, I look closer at some cases where the two forms alternate and in Section 6, I present different ways that the demoted agent may or may not be realised in passive and impersonal constructions. In the concluding section, I briefly address the issue of how the choice of passive form can be accounted for in two theoretical frameworks, HPSG and Construction Grammar. For reasons of space I cannot discuss all relevant properties of the two passives. Readers who understand Swedish will find additional relevant information in Svenska Akademiens grammatik (SAG), the main reference grammar of Swedish, edited by Teleman, Hellberg and Andersson, which appeared in 1999. Some of the examples in this article are taken from SAG. Most of the data discussed comes from the corpora of Swedish available at Språkbanken (the Bank of Swedish).1

.

S-passive and bli-passive

The so-called s-passive is formed morphologically by adding -s to the verb form, which may be inflected for tense. (1) På den tiden tala-de-s franska vid hovet. at that time spoke-S French at the.court “At that time, French was spoken at the court.”

The passive -s can be added to infinitives, present and preterite forms as well as to the so-called supine form which is used in the compound tenses perfect and pluperfect. Table 1 gives an overview of the s-forms in the regular first conjugation. Irregular verbs also have s-forms. I refer to the forms with the suffix -s as s-forms and gloss them as S in the examples.2 The other common passive form is formed by the copula verb bli (‘become’) followed by a participle which agrees with the subject in number and gender. 3

. . The s-form has developed from the Old Swedish reflexive pronoun sik which presumably cliticized onto the verb. This explains the apparent unusual word formation pattern in which a derivational morpheme appears further from the stem than inflectional morphemes. S-passives are attested in runic inscriptions from the 11th century. S-forms also have other uses such as reciprocal and middle, but these will not be discussed here. See Kemmer (1993), Enger (2001) and Solstad & Lyngfelt (this volume) for relevant discussion. . The verb bli(va) was borrowed from Low German bleiben in the 14th century. Its use increased rapidly and replaced varda (‘become’) which was the common passive auxiliary until then. In some dialects, the preterite form vart is still used as a passive auxiliary.

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.3 (197-264)

Semantic and syntactic patterns in Swedish passives

Table 1. Swedish verb inflections

Infinitive Present Preterite Perfect Pluperfect

Active

Passive

tala tala-r tala-de har tala-t hade tala-t

tala-s tala-s tala-de-s har tala-t-s hade tala-t-s

(2) Talar-na blev avbrut-na flera gånger. the.speakers became interrupted-PL several times “The speakers were interrupted several times.”

I will refer to this passive form as the bli-passive.4 The bli-passive is typically used when the clause describes a change of state. When the result of the change rather than the change itself is relevant, the stative copula verb vara (‘be’) tends to be used as illustrated by the following example. (3) Fången var redan avrättad. the.prisoner was already executed “The prisoner had already been executed.”

The participle used is traditionally called a perfect participle since it is derived from the perfect stem. However, for transitive verbs it is more appropriate to refer to it as a passive participle, since it has ‘passive’ meaning in the sense that the subject argument is necessarily interpreted as being affected by the verb.5 The subject (or external argument) of the passive participle has the semantic role that the corresponding active verb assigns to its object as shown in (4a), whereas the supine form used in the perfect tense has active meaning (4b). The supine is invariant and is never used with copula verbs, as shown in (4c).6 (4) a.

Älg-en blev/var skjut-en/*skjut-et. the.moose BLEV/was shot-ppcp,com/shot-ppcp,neut “The moose was/had been shot” b. Jägaren har skjut-it älgen. the.hunter has shot-sup the.moose “The hunter has shot the moose”

. I will gloss the bli-forms as BLI in the present tense, BLEV in the preterite tense and BLIVIT in the perfect tense. . The present participle, on the contrary, is inherently active and can not be used in periphrastic passive constructions. . The following abbreviations are used in the glosses: ppcp passive participle; com common gender; neut neuter gender; sup supine.



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.4 (264-343)

Elisabet Engdahl

c. *Älgen blev/var skjut-it. the.moose BLEV/was shot-sup

The passive meaning thus arises from the use of the passive participle.7 The auxiliary verb contributes tense and aspectual information and essentially functions as a raising verb.

. Conditions of use In this section we will look at some factors that affect the choice of passive form. In some contexts only the s-passive can be used. General rules, recipes and prescriptions are expressed using the s-form, see (5)–(7). This is also the normal option for generic statements as in (8). (5) a.

sign

(6) a.

announcement

Dörren öppnas utåt. the.door open-S outward “The door opens outward.” b. *Dörren blir öppnad utåt.

Dörrarna stängs! the.doors close-S “The doors are about to be closed.” b. *Dörrarna blir stängda.

rule I bisatser placeras inte framför det finita verbet. in subordinate clauses place-S not before the finite verb “In subordinate clauses, ‘not’ should be placed in front of the finite verb.” recipe b. Mjölet tillsätts under omrörning. the.flour add-S under stirring “Add the flour while stirring.” generic (8) a. Fågelfrö köps i djuraffären. (SAG 4: 398) bird seed buy-S in the.pet shop “Bird seed can be bought in the pet shop.” b. Det talas inte längre danska i Skåne. it speak-S no longer Danish in Skåne “Danish is no longer spoken in Skåne.”

(7) a.

Outside these contexts of use, both the s-form and the bli-form can be used and we will now look at other factors that affect the choice. . Perfect participles of unaccusative verbs like anlända (arrive) and vissna (wilt) have the same meaning as the active verb. These verbs are not used in bli-passives or s-passives (cf. Platzack 1998: 118ff.).

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.5 (343-394)


. Types of events vs. specific events In descriptive Swedish grammars from the 20th century it is common to find two claims concerning the choice of passive. The first claim is that the s-form is used when referring to types of events, without localising them in time and that the bli-form is used when talking about particular events, in the past, present or future (see e.g. Thorell 1973: 135f.). To some extent, this is correct. As we have seen, the s-form is the only possible form in rules, generic statements, announcements, recipes etc. which describe types of events. However, the claim that the bli-passive is used systematically when referring to specific events is not correct for Swedish. The claim is actually more appropriate for Danish and Norwegian, where the choice of passive to a large extent reflects precisely this distinction between generic and specific events. Consider the Norwegian minimal pair in (9), taken from the Norwegian reference grammar edited by Faarlund et al. (1997: 514).8 (9) a.

Oppgavene leveres hver uke. the.exercises hand-S in every week “The exercises are normally handed in every week.” b. Oppgaven ble levert for seint. the.exercise BLEV handed in too late “The exercise was handed in too late.”

(N) generic

specific event

Heltoft & Falster Jakobsen (1996: 7) argue that the use of the s-passive in the Danish example in (10a) indicates that this is to be understood as an objective report of a fact or as a rule, whereas the bli-passive in (10b) reports a subjective prediction of what will happen in the case of a specific writing project. (10) a.

Indledningen skrives til sidst. the.introduction write-S last “The introduction is to be written last.” b. Indledningen bliver skrevet til sidst. the.introduction BLI written last “The introduction will be written last.”

(D) objective or rule

subjective prediction of specific event

In Swedish, however, the passive forms don’t contrast in this way. The Swedish versions of the Norwegian examples in (9) show that the s-passive is used both in the generic case and in the specific case. The bli-form in (11c) is very marked. (11) a.

Uppgifterna lämnas in varje vecka. the.exercises hand-S in every week

generic

. See also Lødrup (2000) for more details on restrictions on Norwegian passives which involve thematic roles and aspect.



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.6 (394-456)

Elisabet Engdahl

b. Uppgiften the.exercise c. ??Uppgiften the.exercise

specific event

lämnades in för sent. handed-S in too late blev inlämnad för sent. BLEV handed-S in too late

Similarly in (12) where we find the Swedish counterparts to the Danish example. The s-passive can be understood both as an objectively stated fact or rule and as a subjective prediction of a specific event. The bli-form is not acceptable here. (12) a.

Inledningen skrivs sist. the.introduction write-S last “The introduction is to be written last.” “The introduction will be written last.” blir skriven sist. b. ??Inledningen the.introduction BLI written last

objective, rule or subjective prediction

. Aktionsart The second claim often found in descriptive grammars like Thorell and SAG is that the s-form is supposed to be used when the focus is the event itself, whereas the bli-form is used when the inception or completion of the event is foregrounded. Again, this claim is correct to some extent. The copula bli retains its lexical meaning ‘become’ also when it is used in the passive. Consequently, bli-passives tend to focus the inception of a new state or the result of the changed state. However, this does not mean that the aktionsart of the verb actually determines which passive form is used, as pointed out in SAG (4: 381f.). Both atelic and telic verb phrases can be expressed using the s-passive as shown in the examples in (13)–(14), taken from SAG. (13) a.

Polisen höll tjuven under uppsikt i/*på tre timmar. atelic “The police kept watching the thief for/*in three hours” b. Tjuven hölls under uppsikt i/*på tre timmar (av polisen). the.thief kept-S under surveillance for/*in three hours “The thief was kept under surveillance for three hours.”

(14) a.

En journalist skrev boken på/*i några dagar. “A journalist wrote the book in/*for a few days.” b. Boken skrevs på/*i några dagar (av en journalist). the.book wrote-S in/*for a few days “The book was written in a few days (by a journalist).”

telic

Similarly, the periphrastic passive is used with both telic and atelic verb phrases, but is more common with telic verbs. As mentioned earlier, the choice of copula reflects whether the focus is on the change of state (bli is preferred) or on the

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.7 (456-533)


resultant state (vara is preferred). In (15a) a singular telic event is described. (15b) can be used to refer to iterated telic events, whereas (15c) describes the resultant state of a telic event and in (15d) a telic verb is used in a generic description. telic Han blev rakad och klippt på några minuter. he BLEV shaved and cut in a few minutes “He was shaved and had his hair cut in a few minutes.” telic, iterated b. Hon blev arresterad flera gånger. she BLEV arrested several times “She was arrested several times.” telic, resultant state c. Staden var förstörd (*flera gånger). the.town was destroyed (*several times) “The town had been permanently destroyed.” telic, generic d. Här blir man fort glömd. here BLI one quickly forgotten “Here you are quickly forgotten.”

(15) a.

SAG notes that when the bli-passive is used with an atelic verb, the focus is often on the beginning of the new state (4: 392). This can be shown by the fact that a time adverbial in such clauses is interpreted as measuring the time it takes for the new state to come into existence, see (16a). If the verb denotes a process, a durative adverbial is understood as measuring the duration of the activity (16b). (16) a.

Lisa blev på kort tid uppskattad av hela personalen. Lisa BLEV in short time appreciated by all the.staff “Lisa soon became appreciated by all the staff.” b. Patienten blev hela tiden övervakad av vår personal. the.patient BLEV all the.time monitored by our staff “The patient was being monitored by our staff all the time.”

atelic

atelic

We have seen that the two claims concerning the use of the two passives are partially correct. They account for the preference for s-passives in generic, rule-like sentences and for the focus on the inception of a new state in (some) bli-passives. However, the claims are too strong since they wrongly predict that examples like (11b), (13b) and (14b) should not occur in Swedish since the s-passive is here used to describe singular events, nor should (15d) occur since the bli-passive is here used in a sentence describing what normally happens in some place. The claims are probably more appropriate for Danish and Norwegian than for presentday Swedish.



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.8 (533-597)

Elisabet Engdahl

Table 2. Proportion of passive clauses in a newspaper corpus from 2003, after Laanemets (2004: 87).

Clauses Passive clauses Percentage

Danish

Norwegian

Swedish

Total

4028 504 12,5%

4015 513 12,8%

4029 562 13,9%

12072 1579 13,1%

Table 3. Percentage of different passive forms. Data from Laanemets’ study of newspapers (2004: 91).

s-passive bli-passive vara-passive

Danish

Norwegian

Swedish

55% 31% 14%

53% 32% 15%

91% 1% 7%

. Frequencies of the two passives Before pursuing the question when the two forms are used, let us look at how often the two passives are used in modern Swedish. Anu Laanemets has carried out a corpus study of the frequencies of the different passive forms in Danish, Norwegian and Swedish newspapers (Laanemets 2004) and the numbers in the tables in this section are taken from her work.9 Her materials consisted of columns written in March 2003 by regular columnists in two national newspapers in each country. She chose this specific kind of article in order to avoid stylistic and genre-based variation. The style can be characterised as carefully worded prose. She then extracted the first 4000 finite clauses in each subcorpus and identified the passive clauses. In addition to s-passives and bli-passives, Laanemets included vara-passives and få-passives (get-passives) although these were very few. The proportion of passive clauses of all types is shown in Table 2. On average, 13% of the finite clauses in the corpus were passive. This is a rather high figure and probably reflects the somewhat elevated style of the columns. The Swedish newspapers had the highest frequency of passive clauses, but the difference was not great. However, when we look at the frequencies of the different passive forms separately, a big difference emerges between Danish and Norwegian on one hand and Swedish on the other. The percentages for Danish and Norwegian are very similar. A little over half the number of passive clauses are s-passives, a third are bli-passives and the rest are . Laanemets’ masters thesis, written in Danish, also contains a useful survey of the way the two passives are described in reference grammars for the three languages. The Swedish newspapers she used were Dagens Nyheter (DN) and Svenska Dagbladet (SvD).

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.9 (597-695)


Table 4. The distribution of s-passives in different tenses. Data from Laanemets’ study of ∼ 500 passive sentences in each language (2004: 97f.).

Infinitive Present Past Perfect + pluperf. Total

Danish

Norwegian

Swedish

109 139 14 0 262

129 120 0 0 249

144 188 82 77 491

Table 5. Distribution of bli-passives in different tenses. Data from Laanemets’ study of ∼500 passive sentences in each language (2004: 97f.).

Infinitive Present Past Perfect + pluperf. Total

Danish

Norwegian

Swedish

28 25 73 27 153

12 40 73 24 149

1 4 1 1 7

vara-passives. In Swedish an overwhelming 91% are s-passives and only 1% are bli-passives. The Danish and Norwegian distribution seems to fit quite well with the claims discussed above, viz. that there are certain contexts where only s-passives can be used. Elsewhere there is a choice between s-passive and bli-passive that reflects the kind of event being talked about; generic or specific, according to Faarlund et al., whether the reporting is objective or subjective, according to Heltoft and Falster Jakobsen (1996) (see also Heltoft 1994). However, the Swedish distribution does not fit this picture. The very small proportion of bli-passives suggests that they are used rather restrictively. If we look in more detail at the distribution of the passive forms in different tenses a further difference between Danish and Norwegian on one hand and Swedish on the other becomes evident. Table 4 shows the distribution of s-passives and Table 5 the distribution of bli-passives in the different tenses in the three languages. Looking first at Table 4, we see that there are almost twice as many s-passive forms in Swedish as in Danish and Norwegian. There are more Swedish spassive infinitives and present tense forms than in Danish and Norwegian, but the strongest contrast is in the past and perfect. Here the s-forms are used very sparingly in Danish and not at all in Norwegian, whereas they appear to be frequent in Swedish. In Table 5 we see that a majority of the Danish and Norwegian bli-forms are used in the past and perfect. There are approximately the same number of past



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.10 (695-776)

Elisabet Engdahl

and perfect tense clauses in the three languages but they are expressed differently. In Danish and Norwegian, past and perfect tenses are almost exclusively expressed using the bli-passive, whereas in Swedish, the s-passive is overwhelmingly more common. Again the Danish and Norwegian pattern fits with the claims discussed above. If bli-passive is used primarily for specific events and s-passive for generics, then it is not surprising that bli-passives are more common in the past and perfect tenses and s-passives more common in the present tense and in the infinitive. Swedish, again, behaves differently. One may wonder whether the absence of s-forms in the past and perfect in Norwegian means that these forms are not used in the language, i.e. that there is a morphological gap which would explain the distribution. If there is no s-form in the past tense, then clearly there is no choice but to use the bli-passive. But there are past and perfect s-forms. For instance the verb møte (‘meet’) is used in the sform both with a passive and a reciprocal meaning. Interestingly, in the past tense only the reciprocal reading is available. (18) Vi møttes på stasjonen. we meet-PAST-S at the.station “We met (each other) at the station.” *“We were met at the station.”

(N)

The Danish and Norwegian distributional patterns shown in Tables 4 and 5 fit very well with the descriptions we find in grammars from the beginning of the 20th century such as Mikkelsen (1911) and Western (1921) according to which the bli-passive is the dominant form in the past and perfect tenses.

. The nature of the subject In the previous section we saw that of the 562 Swedish passive clauses examined by Laanemets, only seven, about 1%, were bli-passives. In this section we will try to establish when bli-passives are used in Swedish. If we look back at the examples used so far in this article and look at the nature of the subject, we find that in all of them the grammatical subject of the bli-passive is animate. See e.g. the two examples repeated here in (19) together with one example from Laanemets. (19) a.

Talarna blev avbrutna flera gånger. “The speakers were interrupted several times” b. Lisa blev på kort tid uppskattad av hela personalen. “Lisa soon became appreciated by all the staff.” c. Om han till slut blir övertygad, ger han sitt medgivande. if he to end BLI convinced gives he his-REFL permission “If he finally is convinced, he gives his permission.” (DN)

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.11 (776-828)


Table 6. Percentage of animate subjects in Swedish and Norwegian passive clauses.

s-passive bli-passive

Swedish

Norwegian

31% 88%

40% 63%

This is no coincidence. In a previous study (Engdahl 2001), I investigated a random sample of 100 s-passive and 100 bli-passive examples in Swedish and Norwegian.10 The results are shown in Table 6. In Swedish almost 90% of the subjects of bli-passives are animate, compared with a third of the subjects in s-passives. In Norwegian, too, there are more animate subjects in bli-passives than in s-passives, but the difference is not as striking as in Swedish. When I looked at which verbs were used in the two passives, I discovered that high frequency verbs that only take inanimate objects such as öppna (open), stänga (close), tala (talk) and dricka (drink) had no occurrences in the bli-passive in the corpus.11 What about the approximately 10% bli-passives with inanimate subjects? Two of the examples in Laanemets’ study had inanimate subjects. They are shown in (20). (20) a.

Blir säkerhetsrådet återigen permanent förlamat som under det kalla kriget när världens enda supermakt vänder det ryggen?12 “Will the Security Council be permanently paralysed again as (it was) during the cold war when the only super power in the world turns its back against it?” b. Det är tänkt att de ska betala in hyran själva, men för säkerhets skull avdelas personal att gå med och se till att det blir gjort. “They are supposed to pay in the rent themselves, but as a precaution, staff are assigned to come along and make sure that it gets done.”

The subject in (20a), säkerhetsrådet, is inanimate, but more importantly it denotes a group of representatives which are animate. Nouns that refer to organisations and institutions are often used metonymically for the people responsible. I think . The Norwegian examples came from the Oslo Bokmål corpus and the Swedish examples from the PAROLE corpus. . See Engdahl (2001) for a more detailed discussion of the types of verbs involved. The fact that certain common verbs are never used in the bli-passive is overlooked in SAG. The authors mention in a footnote (SAG 4: 390) that bli-passives show a statistical tendency to take animate subjects. . In Swedish, the future tense is often expressed by the present tense, in this case blir.



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.12 (828-903)

Elisabet Engdahl

it is appropriate to regard such subjects as animate.13 The passive in (20b), det blir gjort (it gets done), is a very common expression, used when you want to emphasise the importance of the event being completed. Consider also the example in (21a), taken from the PAROLE corpus. (21) a.

Det var brev som skulle skrivas men de blir inte skrivna. there were letters that should write-S but they BLI not written “There were letters that were supposed to be written but they never get written.” b. ??Det var brev som skulle bli skrivna men de skrivs inte. there were letters that should BLI written but they write-S not

Here the author is deliberately contrasting the s-passive and the bli-passive, thereby achieving a rhetorical effect. Note that if we switch the order of the passives as in (21b), the result sounds strange. There may be several reasons for this; deontic modals (first conjunct) are often used together with s-passives. There is a preference for bli-passives when you focus on the effects of the event (second conjunct), as we have seen. The s-form de skrivs inte implies that there is no writing activity going on, whereas the author presumably intends to highlight the result, the absence of written letters. We have seen that there is a high proportion of animate subjects in bli-passives. However, it is probably not animacy per se that plays a role here. There are some cases where the bli-form is actually preferred and they typically involve situations where the subject of the passive clause may influence the situation described or has some control over it. (22) a.

Representanten försökte bli omvald. the.representative tried BLI reelected “The representative tried to be reelected.” b. ??Representanten försökte omväljas. the.representative tried reelect-S c. *Representanten försökte ha bruna ögon. the.representative tried have brown eyes

As noticed by Zaenen (1993), a subject who tries to bring about a situation has to have some control over this situation, cf. the ungrammaticality of (22c). When the subject is a possible controller and the situation can be controlled or influenced, as in the case of elections, it is often preferable to use a bli-passive in Swedish. Since controllers are normally animate, a high proportion of animate subjects is what we would expect. . In her study of factors that affect whether an initial NP is interpreted as subject or object, Øvrelid (2003) similarly treats metonymic inanimate nouns as animate.

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.13 (903-945)


The connection between this type of volitional control and passives was noted by Sundman (1983) who also was able to explain why only bli-passives can be used in imperatives. Normally imperatives are instructions to the listener to bring about a situation. The understood subject of an imperative thus has to be an agent or a cause. But since the agent has been demoted in passives, passive imperatives should not occur. See (23) where the active imperative of the verb råna (rob) is fine, but the s-passive imperative cannot be interpreted as a passive.14 (23) a.

Råna mig inte! rob me not “Don’t rob me!” b. *Rånas inte! rob-S not

Imperative bli-passives are much better, especially if they are negated. Although the understood subject is not agentive – it is the undergoer of the robbing event – the imperative is fine, as shown in (24a). (24) a.

Bli inte rånad i Chicago! BLI not robbed in Chicago “Don’t get robbed in Chicago!” ? b. Bli rånad i Chicago! BLI robbed in Chicago c. Försök att inte bli rånad i Chicago! try to not BLI robbed in Chicago “Try to avoid being robbed in Chicago!”

The negative imperative in (24a) is interpreted as ’Avoid putting yourself in a situation where you are likely to get robbed’. Although the subject is not expressed, the use of the bli-passive suggests that it has to have some control over the situation, cf. (24c). The positive imperative in (24b) is stranger, as is to be expected, since the addressee presumably cannot control when the robbers in Chicago are likely to strike. Positive bli-imperatives with expressed agents can be found in advertis. The only way (23b) can be interpreted is as an absolute construction with the meaning ‘Don’t perform robberies!’. In an absolute construction the verb is in the s-form but still active, with a dispositional meaning. Subjects of absolute constructions are agentive and objects are omitted. (i) is a typical example. (i)

Hunden bits the.dog bite-S “The dog tends to bite.”

See also Solstad & Lyngfelt (this volume).



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.14 (945-1004)

Elisabet Engdahl

ing as in (25) from SAG (4:390). The message here is presumably an invitation to avail yourself of the services of the firm. (25) Bli omskött av världens mest serviceinriktade personal! BLI pampered by the.world’s most serviceminded staff “Allow yourself to be pampered by the most serviceminded staff in the world.”

In the sample of Swedish bli-passives investigated, there was a large proportion of verbs which describe events with detrimental effects on the passive subject, verbs like döda (kill), förstöra (destroy) and mörda (murder). In more than 50% of the bli-passives, the subject was adversely affected by the event. In this respect the Swedish bli-passives resemble the so-called get-passive in English. Compare the following examples from the Brown corpus discussed in Gronemeyer (1999): (26)–(27). (26) “You’re gonna get caught.”

(Hebson, 1961 The Lattimer Legend)

(27) “Don’t get yourself killed for something that doesn’t concern you.” (Chadwick, 1961 No Land Is Free)

Gronemeyer notes that out of the 46 get-passives in the Brown corpus close to half were adversative (see also Quirk et al. 1985: §3.66). In this section, I have brought up some examples where bli-passives are either the preferred or the only option and we have seen that the subject’s ability to control the situation seems to be important.15 However, in many cases, animate subjects are expressed with s-passives. Recall that about a third of the s-passives in Table 3 had animate subjects. This shows that animacy is not a sufficient condition for triggering bli-passive. Apart from the clear control cases shown in (22) and the passive imperatives in (24), many bli-passives can equally well be expressed using an s-passive. On the contrary, replacing an s-passive with a bli-passive often leads to awkward or ungrammatical sentences. This confirms Heltoft’s suggestion that the s-passive is the unmarked passive in modern Swedish (1994).

. Alternation between s-passive and bli-passive After having looked at the overall distribution of the two passives and identified some important factors affecting the choice between them, let us next consider . Concerning the vara-passives, my impression is that there are no particular restrictions on the subjects. In Laanemets’ study there were 35 Swedish vara-passives. Five of them (14%) had animate subjects, but the sample is too small to say whether this is typical. It may be the case that the number of animate subjects is smaller in her sample of newspaper columns than in the sample of passive clauses that I used for Table 3 which were drawn mainly from fiction.

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.15 (1004-1073)


some cases where the two forms alternate. I concluded at the end of the last section that the s-passive is the unmarked form which in most contexts can be used with most verbs. Consequently, the use of the bli-passive is a marked option which the speaker may choose and in doing so s/he often conveys something in addition to reporting what happened to the passive subject. Consider the following excerpt from a Swedish newspaper where the verb döda (kill) is used first in the s-passive form and then in the bli-passive. (28) Våldsmännen som inte stoppas Varje år dödas 25 svenska kvinnor av män som de delat sina liv med – trots kvinnofridsreformen. GP har granskat två fall i Västsverige. I båda fallen kände myndigheterna till att kvinnorna misshandlades av sina män. Ändå blev båda kvinnorna dödade. (Göteborgs-Posten December 5, 2000) “The violent men who are not stopped (stop-S) Every year 25 Swedish women are killed (killed-S) by men that they have shared their lives with – despite the ‘Peace for woman’ reform. GP has looked into two cases in the west of Sweden. In both cases the authorities knew that the women were being battered (battered-S) by their husbands. Still both women were killed (BLI killed).”

In this paragraph there are four occurrences of passive verbs. The three first are in the s-passive and the final in the bli-passive. The heading, and the verb in the first sentence are in the present tense, describing the current situation with respect to domestic violence. The use of the s-forms, stoppas (stop-S) and dödas (kill-S), is natural in a description of the circumstances. The s-form is also used when the reporter talks about the repeated instances of battering, misshandlades (battered-S) that were known to the authorities. The last sentence is introduced by an adversative adverb, ändå (still), which has the rhetorical effect of highlighting the final result, viz. that the women were killed. The switch to the bli-passive in the final sentence can be seen as a way of emphasizing the outcome of the ongoing violence. We saw in Section 2 that when the verb is atelic, the bli-passive is preferred if the focus is on the beginning of the new state, cf. (16a) repeated here. (16) a.

Lisa blev på kort tid uppskattad av hela personalen. Lisa BLEV in short time appreciated by all the.staff “Lisa soon became appreciated by all the staff.”

atelic

In this example the subject is animate. However, similar considerations apply to examples with inanimate subjects. Whereas the verb läsa (read) is almost only used in the s-passive and this form is normally preferred over the bli-passive, as shown in (29a), the bli-passive is possible, and for some speakers even preferred, in the presence of a temporal adverb measuring the duration of the inception of the new state, as in (29b).



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.16 (1073-1164)

Elisabet Engdahl

Boken lästes/ ? blev läst över hela jorden. the.book read-S/BLEV read over whole the.earth “The book was read all over the earth.” ? b. Boken lästes/ blev på kort tid läst över hela jorden. the.book read-S/BLEV in short time read over whole the.earth “The book was soon read all over the earth.”

(29) a.

Although no instances of bli-passives of verbs like öppna (open), stänga (close) and tala (talk) were found in the corpus, examples can of course be constructed. The examples in (30) were suggested by B. Lyngfelt. (30) a.

När kassaskåpet äntligen blivit öppnat hörde vi nattvakten. when the.safe finally BLIVIT opened heard we the.nightguard “When the safe finally had been opened, we heard the night guard.” b. Det här låset lär inte bli öppnat i första taget. this here lock suppose not BLI opened in first the.take “This lock definitely won’t be opened at the first try.”

Both examples sound somewhat odd but they are not ungrammatical. They could conceivably be uttered, provided the context motivates the use of the marked blipassive. A plausible context for (30a) might be a robbery. After having worked on the safe for a while and finally getting it open, the robbers hear the night guard arrive and have to leave. In retelling the event, one of the robbers might use (30a) as it conveys that the resulting state, the open safe, finally came about after some effort. (30b) could be uttered by the bank’s head of security who has just installed a new secure lock and who is confident that the resulting open state won’t come about. However, there are also plenty of cases where the use of the bli-passive does not seem to add anything extra, as in the following examples involving coordinated passives where the first conjunct is an s-passive and the second a bli-passive. (31a) is from the Swedish radio and (31b) from Laanemets’ corpus. (31) a.

Vid olyckan dödades åtta passagerare och fjorton blev in the.accident killed-S eight passengers and fourteen BLEV skadade. hurt “Eight passengers were killed in the accident and fourteen were hurt.” b. ...att varje barn möts och blir respekterat för hurdan hon/han that every child meet-S and BLI respected for how she/he är. is “...that every child is met and is respected for what she or he is.”

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.17 (1164-1201)


To me these examples could equally well be expressed with two s-passives or two bli-passives. The reason for this alternation might be stylistic, viz. a wish to vary the form of expression.16

. What happens to the demoted agent? So far we have mainly looked at how passive verbs and participles behave and at possible restrictions on the kinds of subjects they take. In this section, we will look at to what extent a demoted agent is nevertheless present, syntactically and/or semantically. . Av-phrases All through the discussion we have been assuming that there is an agent in the passive clauses. The situation described by the passive clause has not arisen by itself but has been brought about by some agent or cause. Sometimes this agent or cause is expressed as an av-phrase (by-phrase). In the sample of passive clauses from the PAROLE corpus used to establish the proportion of animate subjects in Table 3, only 13% of the s-passives and 11% of the bli-passives had an explicit av-phrase. In the vast majority of passive clauses, the agent or cause thus remains unexpressed. About half the av-phrases introduced an animate agent, the rest were inanimate causes and instruments. There were no clear differences between s-passives and blipassives in this respect. A few examples from the PAROLE corpus are given in (32). . In addition to the types of alternations illustrated in the text, I will mention one further type. For certain homonymous verbs, the different senses tend to be expressed with different passive forms. For instance, the verb anta is used predominantly in the s-form when it has the meaning accept as in (i) and in the bli-form when it corresponds to admit, as in (ii). When the meaning corresponds to assume, only the s-form can be used, (iii). The examples are from PAROLE. See Engdahl (2000) for discussion. (i)

Programmet antogs av kongressen. the.programme accepted-S by the.congress “The programme was accepted by the congress.”

(ii) För tre år sedan blev hon antagen till kursen. for three years since BLEV she admitted to the.course “Three years ago she was admitted to the course. ” (iii) Inlärningen antas bli mera effektiv om . . . the.learning assume-S become more efficient if . . . “The learning is assumed to be more efficient if . . .”



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.18 (1201-1276)

Elisabet Engdahl

(32) a.

Den ackompanjeras av ostämda gitarrtoner. it accompany-S by untuned guitar.tones “It is accompanied by the sound of an out-of-tune guitar.” b. Han blir aldrig nockad av något stenskott. he BLI never knocked-out by some stone-shot “He is never brought out of action because of a stone hitting the windscreen.”

Av-adverbials are obligatory with stative passive verbs denoting properties, identity, temporal, spatial and causal relations as well as possession, cf. SAG (4: 373ff.). (33) a.

Månaden dominerades *(av ett köldhögtryck). the.month dominated-S by a cold.high.pressure “The month was dominated by a cold high pressure system.” b. Marken ägdes fortfarande *(av Danviks hospital). the.ground owned-S still by Danvik’s hospital “The ground was still owned by the Danvik hospital.”

. Impersonal passives The most radical way of demoting an agent is found in the so-called impersonal passive constructions where agentive intransitive and transitive verbs are used in the s-passive, as shown in (34) with intransitive verbs and in (35) with a transitive verb. (34) a.

Det sjöngs i ladorna (*av ungdomarna). it sang-S in the.barns (*by the young people) “There was singing (by people) in the barns.” b. Det arbetas från morgon till kväll. it work-S from morning to evening “People work all day.”

(35) a.

Det dricks mycket öl (*av studenterna). it drink-S much beer (*by the.students) “There is a lot of beer drinking.” b. Det dricks mycket öl bland studenterna. it drink-S much beer among the.students “There is a lot of beer drinking among the students.”

In impersonal passives the agent may not be expressed in an av-phrase in the core clause. However, it is possible to refer to the agent by a circumstantial adverbial, as shown in (35b), which suggests that the restriction on av-phrases is syntactic

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.19 (1276-1313)


rather than semantic.17 Interestingly enough, the understood agent can only be interpreted as human, as noted by Siewierska (1984: 100f.).18 (34a) cannot be used to describe a situation where for instance a lot of birds were singing in the barns. When the understood agent is not human, an impersonal active construction is preferred, as shown in (36).19 As far as I am aware, this construction is only used with verbs that denote events that can be perceived through the senses. The source of the perception may be indirectly expressed as a circumstantial adverbial. (36) a.

Det sjunger i telefonledningarna. it sings in the.telephone.lines “The telephone lines are singing.” b. Det viskar i buskarna. it whispers in the.shrubs “There are whispering noises coming from the shrubs.” *“There are people whispering in the shrubs.”

In Swedish only the s-passive is used productively in these impersonal constructions. Impersonal bli-passives are restricted to examples like in (37) which typically contain a negative and/or quantitative phrase which precedes the participle and occurs in the normal position for sentential adverbials. Note that (37a) contains the same expression as in (20b). (37) a.

Det blev ingenting gjort. it BLEV nothing done “We got nothing done.” b. Det blev inte mycket sagt. it BLEV not much said “Not much was said.”

. Impersonal passives should be distinguished from examples like (i) which involve an extraposed clausal subject and an expletive det. In these cases the agent may be overtly expressed by an av-phrase in the core clause. (i)

Det påstods av talesmannen att detta inte skulle få några it claimed-S by the.spokesperson that this should not get any oönskade konsekvenser. undesirable consequences “It was claimed by the spokesperson that this would not get any undesirable consequences.”

. Impersonal passives in Icelandic also display this property as shown by Maling (this volume). Maling further notes that an understood agent is interpreted as human also in personal passives in Icelandic, but not in the other Scandinavian languages. . Åfarli (2006) discusses Norwegian data concerning impersonal active and passive constructions.



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.20 (1313-1369)

Elisabet Engdahl

In Danish and Norwegian, both passives are possible in impersonal constructions and the choice depends on the type of event described, as discussed in Section 2. A generic sentence like (34b) would probably be expressed using an s-passive. If (34a) is used to describe a single past event, the bli-passive would be used. However, if (34a) is used to describe a previous custom, the s-passive could be used, at least in Danish. We have already seen that the agent is suppressed in impersonal passives (34)–(35). They are used when the identity of the agent is not at stake and are hence a convenient device for avoiding reference.20 Although the agent can not be expressed as an av-phrase, the understood agent is semantically accessible in impersonal passives as shown by the fact that agent oriented adverbials are possible (Anward 1981: 10f.).21 (38) a.

Det arbetas frivilligt/motvilligt (*av lärarna) på universitetet. it work-S voluntarily/reluctantly by the.teachers at the.university “People work voluntarily / reluctantly at the university.” b. Det koms för sent (*av studenterna) med flit. it come-S too late by the.students on purpose “People are late on purpose.”

. Presentational constructions It is interesting to compare the impersonal passives with presentational constructions which provide a somewhat more moderate way of demoting the agent. In Swedish, presentational constructions can be used to introduce an agentive subject which is demoted and appears in the ordinary object position while the subject position is filled by an expletive pronoun, det. (39) a.

Det arbetar 5000 lärare på universitetet. it work 5000 teachers at the.university “There are 5000 teachers working at the university.”

. The impersonal passive offers a convenient way of avoiding directly addressing the hearer. In certain situations, some Swedish speakers find it difficult to know whether to use the informal du (you, singular) or the more formal ni (you, plural) when addressing a single person and then resort to utterances like in (i). (i)

Dricks det kaffe? drink-S it coffee “Would you like some coffee?”

. The verb used in (38b) is the passive of the unaccusative verb komma (come). Unaccusative verbs are not normally used in the passive but do occur in impersonal passives. The same happens in Icelandic (Maling this volume) and in Dutch (Zaenen 1993).

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.21 (1369-1441)


b. Ringde det en student och frågade efter mig? phoned it a student and asked after me “Did a student phone and ask for me?”

The clause containing a demoted agentive subject can be coordinated with a verb phrase, as shown in (39b), which suggests that the demoted phrase still behaves syntactically as a subject.22 Somewhat surprisingly, no agent oriented adverbial is possible in a presentational construction. (40) a. *Det arbetar motvilligt 5000 lärare på universitetet. it work reluctantly 50000 teachers at the.university b. *Det arbetar 5000 lärare motvilligt på universitet. it work 5000 teachers reluctantly at the.university c. 5000 lärare arbetar motvilligt på universitetet. 5000 teachers work reluctantly at the.university.

Despite the fact that the agent is overt in (40a, b), agent oriented adverbials are not possible. Only if the subject appears outside the verb phrase is such an adverbial possible (40c). This is rather surprising, especially given the fact that agent oriented adverbials are possible in impersonal passives, as shown above. Just as in the case of impersonal passives, the restriction barring agent oriented adverbials in presentational constructions must be a syntactic constraint on what may appear in the core clause. The restriction can not be semantic, since it is possible to add such an adverbial in a following reduced clause, as in (41).23 (41) Det arbetar 5000 lärare på universitetet, flera av dem ganska it work 5000 teachers at the.university several of them rather motvilligt. reluctantly “5000 teachers work at the university, some of them rather reluctantly.”

While discussing locative inversion, Levin and Rappaport Hovav (1995: 259) mention that adverbs like voluntarily are infelicitous in English presentational constructions. They suggest that this infelicity follows from the discourse function of the presentational construction since the adverbs ‘typically assert information about the postverbal NP that is incompatible with the relative newness of this NP’. This is not entirely convincing. The referent of the postverbal NP is presented as new to the hearer, but it may very well be the case that the speaker has more information about the referent, information that s/he may convey e.g. using a PP or a . Lødrup (1999) assumes that the postverbal NP in similar Norwegian examples is not a demoted subject but an agentive object. . Robin Cooper, p.c.



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.22 (1441-1478)

Elisabet Engdahl

relative clause. Nevertheless, the speaker may not convey information concerning the attitudes of the postverbal subjects. (42) a.

Det arbetar två lärare från Finland (*motvilligt) på min it work two teachers from Finland (reluctantly) on my institution. department “There are two teachers from Finland working in my department.” b. Det arbetar en lärare som är från Finland (*motvilligt) på min it work a teacher that is from Finland (reluctantly) on my institution. department “There is a teacher who is from Finland working in my department.”

It is the presence of a subject oriented adverb that makes these sentences ungrammatical and judgments are quite clear. This supports the view that we are dealing with a syntactic restriction rather than with a restriction following from the discourse function. As regards the verbs that may be used, Levin & Rappaport Hovav correctly point out that they must denote a ‘characteristic activity’ of the postverbal subject.

. Concluding remarks In this article I have discussed some syntactic and semantic factors that are implicated in the choice of passive form in modern Swedish. We have seen that the pattern in Swedish differs from the Danish and Norwegian patterns to a considerable extent. Whereas factors like event type and mood seem highly relevant for predicting which passive form is used in Norwegian and Danish, this is not so in Swedish. In this language, the s-passive is clearly the unmarked passive and the bli-passive is marked, meaning that it is subject to additional restrictions. For instance, the passive subject’s ability to influence the situation seems to play a role. It is plausible that the use of passives in Swedish has changed more than in Danish and Norwegian. In 1904, Ebba Björnström assumed that the pattern of usage in Danish and Norwegian held for Swedish as well, but noted some unexpected uses of s-passives. Heltoft (1994) suggests that the Swedish s-passive has evolved to become the unmarked passive, whereas the bli-passive is only used in marked conditions. Enger (2001) considers the development of the Swedish s-passive to be a case of grammaticalization and this is clearly an interesting topic for further research. Further questions concern the emergence of s-passives in the past and perfect and whether there is a connection with the development of the special supine form during the 17th century. Another interesting topic is whether

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.23 (1478-1534)


the patterns of usage we have found in written Swedish also hold in the spoken language. All through this article I have talked about two passive forms and tried to identify the conditions under which they are used. In an earlier article (Engdahl 2001), I outlined an analysis in Head-driven Phrase Structure Grammar (HPSG). The basic analytic notion in HPSG is the sign, in which the phonological, morphological, syntactic, semantic and pragmatic properties of a word, phrase or clause are represented. This means that dependencies between e.g. morphological form, semantic interpretation and context of use can be stated locally, in a single structure. The choice of form may entail certain constraints on syntactic structure, on the interpretation and on the use and vice versa. Productive syntactic alternations such as passive formation are captured through lexical rules which change the valence requirements of the verb. In addition, the lexical rule can specify the concomitant restrictions on the use of the particular passive form. For instance, the lexical rule for s-passives in Danish might impose a restriction on the type of situation that can be described by the clause in which the verb is used, viz. that it be a generic situation. In Swedish, the lexical rule for bli, which enables this verb to function as a passive copula, might simultaneously impose a restriction on the type of subject that it can combine with as well as what type of situation the passive participle may describe. The HPSG analysis is lexically driven in that the form-interpretation-function correlations are captured in the signs of individual lexical items, albeit by way of general lexical rules. An interesting alternative is to analyse the two passive forms as two distinct constructions in the sense of Construction Grammar (see Fillmore et al. 1988; Goldberg 1995). This approach assumes that over and above the lexical items there are ‘syntactic frames which have constructional interpretations of their own’ to use a quote from Åfarli (2006).24 This seems to be a plausible approach to some of the examples discussed in this article. One candidate construction is the impersonal passive in Swedish, where, as we have seen, the use of the construction entails a restriction on the understood subject, viz. that it be human, which is not present when the verb is used outside this construction. Presentational sentences is another case where a construction approach might provide the right locus for expressing the various syntactic, semantic and pragmatic constraints. In a similar fashion we could try to capture the interpretational regularities associated with e.g. the frame bli + passive participle in Swedish in a periphrastic passive construction. Whether the morphological s-passive in Danish and Norwegian is best analysed as a construction is less clear, but worth exploring. . Åfarli’s paper expresses similar ideas to the ones found in Construction Grammar but his notion of syntactic frame is closer to the kind of structure assumed in the Minimalist program.



JB[v.20020404] Prn:19/10/2006; 13:51



F: LA9602.tex / p.24 (1534-1636)

Elisabet Engdahl

References Åfarli, Tor (2006). Passive and argument structure. In Werner Abraham & Larisa Leisiö (Eds.), Passivization and Typology, Form and Function (pp. 373–382). Amsterdam: John Benjamins. Anward, Jan (1981). Functions of Passive and Impersonal Constructions. A Case Study from Swedish. Ph.D. dissertation. Department of Linguistics, Uppsala. Björnström, Ebba (1904). Om några olika sätt att uttrycka passiv betydelse i nusvenskan (On some ways of expressing passive meaning). Språk och stil, 4, 198–237. Engdahl, Elisabet (2000). Valet av passivform i modern svenska (The choice of passive form in modern Swedish). Svenskans beskrivning 24. Linköping Electronic Conference Proceedings. URL: http://www.ep.liu.se/ecp/006/007/index.html Engdahl, Elisabet (2001). Scandinavian passives in HPSG. In A. Holmer, J.-O. Svantesson, & Å. Viberg (Eds.), Proceedings of the 18th Scandinavian Conference of Linguistics, Vol. 2 [Travaux de l’institut de linguistique de Lund 39:2] (pp. 23–36). Lund: Universitetstryckeriet. Enger, Hans-Olav (2001). Om s-passivens relasjoner til modalitet, aspekt og kasus (On the relationships between s-passive, modality, aspect and case). Tijdschrift voor Skandinavistiek, 22, 411–433. Faarlund, Jan Terje, Svein Lie, & Kjell Ivar Vannebo (1997). Norsk referansegrammatikk. Oslo: Universitetsforlaget. Fillmore, Charles, Paul Kay, & Mary Catherine O’Connor (1988). Regularity and idiomaticity in grammatical constructions: The case of let alone. Language, 64, 501–538. Goldberg, Adele (1995). Constructions. A Construction Grammar Approach to Argument Structure. Chicago, IL: Chicago University Press. Gronemeyer, Claire (1999). On deriving complex polysemy: The grammaticalization of get. English Language and Linguistics, 3, 1–39. Heltoft, Lars (1994). S-modus og perifrastisk modus (S-mood and periphrastic mood). Språkbruk, grammatik och språkförändring. En festskrift till Ulf Teleman. Department of Scandinavian Languages, Lund University. Heltoft, Lars, & Lisbeth Falster Jakobsen (1996). Danish passives and subject positions as a mood system. In Elisabeth Engberg-Pedersen et al. (Eds.), Content, Expression, and Structure: Studies in Danish Functional Grammar (pp. 199–234). Amsterdam: John Benjamins. Kemmer, Suzanne (1993). The Middle Voice. Amsterdam: John Benjamins. Laanemets, Anu (2004). Dannelse og anvendelse af passiv i dansk, norsk og svensk [Nordistica Tartuensia 11]. Tartu University. Levin, Beth, & Malka Rappaport Hovav (1995). Unaccusativity. At the Syntax–Lexical Semantics Interface. Cambridge, MA: The MIT Press. Lødrup, Helge (1999). Linking and optimality in the Norwegian presentational focus construction. Nordic Journal of Linguistics, 22, 205–229. Lødrup, Helge (2000). Exceptions to the Norwegian passive: Unaccusativity, aspect and thematic roles. Norsk Lingvistisk Tidskrift, 18, 37–54. Mikkelsen, Kristian (1975 [1911]). Dansk ordføjningslære (Danish syntax). København: Hans Reitzells forlag. Øvrelid, Lilja (2003). Subject or Object? Syntactic Disambiguation in Norwegian, Stochastic Optimality Theory and Application in an Automatic System. Cand. philol. thesis. Department of Linguistics, University of Oslo. Platzack, Christer (1998). Svenskans inre grammatik – det minimalistiska programmet (The internal grammar of Swedish – the minimalist programme). Lund: Studentlitteratur.

JB[v.20020404] Prn:19/10/2006; 13:51

F: LA9602.tex / p.25 (1636-1681)


Quirk, R., S. Greenbaum, G. Leech, & J. Svartvik (1985). A Comprehensive Grammar of the English Language. London: Longman. Siewierska, Anna (1984). The Passive. A Comparative Linguistic Analysis. London: Croom Helm. Sundman, Marketta (1983). Control, subject, and voice in Swedish. In F. Karlsson (Ed.), Papers from the Seventh Scandinavian Conference of Linguistics (pp. 101–112). Dept. of General Linguistics, University of Helsinki. Teleman, Ulf, Staffan Hellberg, & Erik Andersson (1999). Svenska Akademiens Grammatik (SAG). Stockholm: Norstedts. Thorell, Olof (1973). Svensk grammatik (Swedish grammar). Stockholm: Scandinavian University Books. Western, August (1921). Norsk riksmåls-grammatikk for studerende og lærere. Kristiania: Aschehoug. Zaenen, Annie (1993). Unaccusativity in Dutch: Integrating syntax and lexical semantics. In J. Pustejovsky (Ed.), Semantics and the Lexicon (pp. 129–161). Dordrecht: Kluwer.

Corpora Oslo Corpus of Bokmål: Swedish PAROLE Corpus:



JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.1 (46-138)

The Eastern Khanty locative-agent constructions* A functional discourse-pragmatic perspective Andrey Filtchenko Eastern Khanty narratives provide evidence to the claim that a wide cognitive faculty, facilitating the structuring of information and specifying the roles and interrelation of participants, governs the utilization of the available system’s grammatical resources. It follows from the analysis of interrelated discourse-pragmatic, semantic and grammatical features of the participants in their discourse interaction that certain grammatical resources identify with certain pragmatic-semantic properties. Eastern Khanty passives illustrate the promotion of the Target referent and demotion of the Agent, corresponding to a temporary altering of the discourse status of the referents. “Ergative” displays the Locative-marked Agent, marking a temporary referents’ discourse status shift – the parenthetically backgrounded primary topic. The Loc-Agent constructions mark multiple topicality or shift events, specifying the degree of the participants’ agentivity, control and volition.

.

Introduction

The paper outlines possible types of discourse-pragmatic functions and kinds of propositional-semantic content associated with voice constructions in Eastern Khanty. Selected methodology includes a contrastive morpho-syntactic and contextual analysis of a narrative corpus in a general cognitive-functional framework. Khanty is one of the endangered native languages of the indigenous huntergatherer minorities of north-western Siberia. Together with Mansi, it comprises the Ob-Ugric branch of Finno-Ugric languages, which united with the Samoyedic languages form the Uralic language family. The language falls into two dialectal groups – western and eastern, which are arguably mutually incomprehensible. Most of the descriptive work that has been done is on the western dialects. The * The material for this publication was collected with the support of the National Science Foundation grant No. 0416607.

JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.2 (138-196)

Andrey Filtchenko

dialects of interest in this study are two immediately related river dialects of Yugan and Vasyugan Khanty (approx. 500 speakers). These dialects are particularly interesting as they represent a reportedly underdescribed, archaic and richer system, in morphosyntactic terms (Gulya 1970; Honti 1984; Kulonen 1989; Décsy 1990). The synchronic analysis of the dialects of the Yugan and Vasyugan Khanty river communities has historical and linguistic relevance in light of the prominent hypothesis implying a route of historical migration of the Khanty from the Yugan river basin to the south-east/east of the Vasyugan river basin. This hypothesis is consistent with the supposition of a gradual cross-dialectal distribution of linguistic features in a general west-to-east direction and the increase in the grammaticalization of a number of categories and relations: case, Tense-AspectModality, split-ergativity; preservation of features attributed to the proto-Ugric state (Décsy 1990), etc.

. The basics of the Eastern Khanty clause structure The basis for the discussion in this paper is a narrative corpus of Eastern Khanty texts from Vasyugan and Yugan, collected and transcribed between 2000 and 2004. In our analysis of Eastern Khanty narratives, we differentiate between the grammatical relations and semantic roles of the arguments of propositions, and pragmatic functions and properties of the referents in the discourse expressed by the above arguments. Similar to Dixon (1994) we indicate the grammatical relations as follows: S – intransitive subject; A – transitive subject; O – transitive non subject.

We also recognize a variety of semantic properties and roles that can be mapped onto the above syntactic relations, identifying these in accordance with the extent of their participation, control, involvement, affectedness, etc. in every particular situation. After Dixon, we understand that the role typically mapped onto the A relation will be the one of ‘Agent’, an entity that acts, either volitionally or not, in the situation, originating or causing the event, and is most likely relevant to the success of an event. The roles mapped onto the O relation are typically the ones saliently affected by an event. We shall generically refer to these as the ‘Target’, which will generalize both animate referents and those with a low (null) animacy status. The role mapped onto S is understood as that of a single core NP of an intransitive verb. The facts of Eastern Khanty clause structure most relevant here include: 1. Khanty clauses are generally verb-final:

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.3 (196-286)

The Eastern Khanty locative-agent constructions

(1)

‘I caught a pike-fish’

2. Grammatical functions can be distinguished by word order, case marking and verbal co-referential inflection (cf. (1) vs. (2)). (2)

‘I pulled my dog closer’

Cases essential for distinguishing grammatical relations are: Nom(zero-marked), Acc, Instr, Lat and Loc. In (1) and (2) the referent with the semantic role of Agent appears clauseinitially (SOV), expressed by an S/A argument in the Nom. case, that controls co-referential S/A-V agreement (arrows in (1) and (2)) on the predicate. The referent with the semantic role of Target, usually expressed by a noun Ø-marked for Acc. case, in the O grammatical relation, appears in pre-V position after the S/A argument (SOV).3 It may trigger O-V agreement inflection on the predicate (arrows in (2)). Agreement is obligatory between the grammatical relation of S/A and the predicate V (1)–(2), and agreement between the grammatical relation of O and transitive predicate is contingent upon the pragmatic properties of the O (2). More precisely, transitive verbs, while always agreeing with the A argument (1), may also have agreement with the O argument, as in (2), expressing pragmatic identifiability, accessibility and a high degree of activation of this referent in the interlocutors’ discourse universe. Contextually, it would imply that this referent was recently mentioned, discussed or is unambiguous in the situation (here understood as being cognitively available to interlocutors). Conversely, in (1), absence of agreement . 1st person-singular. See list of abbreviations towards the end of the paper. . This is the so called object conjugation marking the person-number of S/A (1sg) and the number of O (sg). . The Accusative case in Eastern Khanty nouns is zero-marked, i.e. morphologically indistinguishable from the Nominative, in Eastern Khanty pronouns, however, the Accusative has a marker /-t/: ma kuS tu>-6m 1sg fish bring-1sg ‘I brought a fish’

amp man-t por dog 1sg-Acc bite (3sg) ‘A dog bit me’



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.4 (286-338)

Andrey Filtchenko

between the predicate and the O argument manifests pragmatic unidentifiability, inactiveness of the O referent; that is, this referent is unavailable to interlocutors, either from context, or from the situation. This is evident from the pragmatic context of such a clause, where a new or unidentifiable referent is introduced into the discourse expressed by an O argument in the part of the proposition that asserts new information. Thus (3), which differs from (1) only in the choice of agreement, will be unacceptable in its original pragmatic sense of a Comment asserting new information containing a brand-new referent. The O argument would have to code an identifiable, accessible discourse referent with relatively high degree of pragmatic activation: (3)

NOT ‘I caught a pike-fish’, but rather ‘I caught the pike-fish’.

In case of marked O-V agreement, the S/A agreement markers and the O agreement markers form an individual set of portmanteau morphemes – distinct from the S/A agreement set (1) vs. (2). Hence, there is a tradition of describing the Khanty verbal agreement in terms of two paradigmatic sets or conjugations. Absence of agreement marking between the V and the O with an overtly present O is traditionally considered a way of formally expressing the indefiniteness of the O participant, hence the reference to the Subjective Indefinite conjugation (Tereshkin 1961; Gulya 1970). In contrast, presence of the agreement with the O participant in number is a means of identifying the definiteness of the O, hence the reference to the Objective Definite conjugation. In the Finno-Ugric literature, the most frequently cited property of the O that is essential in triggering the O-V agreement on the predicate is definiteness and specificity, which is commonly understood as having to do with the formal grammatical properties of this argument: possessive constructions, pronouns, constructions with demonstratives, embedded clauses, elided/zero objects (Tereshkin 1961; Gulya 1970; Honti 1984; among others). However, while these properties are indeed typical of the O arguments that co-occur with the O-V agreement in Eastern Khanty, they do not always trigger such agreement, i.e. numerous examples are attested where O-arguments with these above “definiteness” properties do not necessarily co-occur with the O-V agreement. For example, the definite and specific nominal O argument preceded by the Dem in (4) and the definite, specific possession-marked O in (5) do not trigger O-V agreement: (4) mä twu qul wel->äl-6m 1sg Dem fish kill-Pst1-1sg ‘I caught that fish’

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.5 (338-404)


(5) voj6> u>-6S nok twutw pan6 nur6>S6> animal head-POSS.3sg up turn.3sg and run.3sg ‘The animal turned up his head and ran away’

It appears that a host of pragmatic and semantic properties of the referent expressed by the O argument are relevant here. Such properties may be revealed while observing the syntactic behavior of the arguments in the narrative. The O arguments that co-occur with the O-V predicate agreement are more flexible in their constituent position (6b), or altogether elided (6c), whereas the O argument without O-V agreement is fixed in its overt SOV position (6a).4 mä sart wel-s-6m ¨fll¨6 1sg pike kill-Pst2-1sg big ‘I caught a pike fish, big one’ b. ¨fll¨6 sart ratw män-n6 lö>öli-s-im Big pike old man 1sg-Loc get ready-Pst2-1sg/sg ‘I got the big pike ready’ iwes-n6 c. terkä-s-im fry-Pst2-1sg/sg stick-Loc ‘I fried it on sticks’

(6) a.

Reflexives can be bound by the S/A argument or the O argument that co-occurs with the O-V agreement, while the O argument without the O-V agreement cannot control it. The reflexive/possessive affix on the locative ‘house’ in (7a) may refer both to the S/A ‘bear’ and the O ‘dog’, accompanied with O-V agreement, whereas in (7b) this affix may only refer to the S/A ‘bear’, and not the O ‘dog’, accompanied with S-V agreement. (7) a.

waja>-n6 ämp jo> nirimtä-s-tä twimin animal-Loc dog home pull-Pst2-3sg/sg there ont-kat-al inside-house-POSS.3sg ‘The bear hid the dog inside his (bear’s) / (dog’s) house.’ b. waja>-n6 ämp jo> nirimtä-s-¨6>¨6n twimin animal-Loc dog home pull-Pst2-3sg there ont-kat-al inside-house-POSS.3sg ‘The bear hid a dog inside his (bear’s) house / *(dog’s) house.’

. Though Khanty is an SOV language, the position of the O constituent may vary contingent upon its pragmatic properties; that is, while brand new, inactive, unidentifiable O referents are always rigidly fixed in SOV order clauses, in other cases pragmatically active and identifiable O referents may cause OSV and occasional SVO order, frequently originating from the increasing Russian interference.



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.6 (404-463)

Andrey Filtchenko

In O argument-focus clauses, WH-question/answer sequences, only S-V agreement on the predicate and no O-V agreement is attested. (8) a.

ämp kojo>i por / *por-6tt6 dog who bite.3sg / *bite-3sg/sg ‘Who did the dog bite?’ b. ämp (tam) iki por / *por-6tt6 dog (Dem) man bite.3sg / *bite-3sg/sg ‘Dog bit a (this) man’.

The combination of the above functional features, and especially the latter collocation of O-focus clauses and S-V agreement, indicate the importance of the discourse-pragmatic function of the O argument in triggering the O-V agreement. That is, the correlation of the syntactic flexibility, omissibility, control over reflexivization, inferability (pragmatic identifiability, activation through a relation, such as possession, to an already active referent) are consistent with overall pragmatic prominence (degree of topicality) of the O argument in triggering the O-V agreement. On the other hand, the correlation of syntactic rigidity, obligatory overtness, inability to control reflexivization and uninferability is consistent with the pragmatic function of focus of the O argument not triggering O-V agreement (Lambrecht 1994; Nikolaeva 1999; Givón 2001). Verbal agreement is instrumental in the elision of clause constituents. The S/A argument is frequently omitted as a free clause constituent, being expressed by the co-referential person-number inflection (2) on the predicate or by zero agreement in the case of a 3sg argument: (9) män-a tG>pGl nöri>-w6l 1sg-Lat Dem swim-PST0.3sg ‘(S)he swam towards me here’

Whenever the S/A argument is overt, it is seen to be emphasized (pragmatically marked) as an established brand-new or reactivated discourse referent, for example in a topic-shift (1) and (4) vs. (2); or thetic or presentational clause type, as in (10): 5 (10) 6j m6ta Sat-n6 mi] ik6m-nat ju> ont-nam nGn-ta One IndPn time-Loc 1Du husband-Com forest Inside-Lat rest-INF m6n-ta j6>-m6n go-INF become-1Du ‘Once, me and my husband decided to go to the woods to relax’ . The thetic/presentational clause type (Lambrecht 1994) has as its purpose not a predication of a property of an argument, but the introduction of a new referent into the discourse for subsequent predication.

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.7 (463-517)


Here, in what is the first clause of the narrative, the discourse participants are introduced for the first time as the central referents of the narrative, making them available for predication in subsequent discourse, i.e. making them discourseactive. This type of clause is commonly associated with indefinite temporal/spatial adverbial phrases in a conventionalized discourse-setting, such as once (upon a time) etc. Hence, clauses in Khanty narratives are commonly devoid of an overt S/A argument, and the S/A information is formally accessible from the co-referential S/A-V inflection on the predicate. Elision of the O argument is attested whenever the context provides enough information about this pragmatically active referent, and then the clause can consist of only a VP as (6c) above. The elision of the identifiable and active referent in the O relation is expected only when licensed by the definite (objective) conjugation, i.e. O-V agreement inflection. Such elision of the O is naturally far less frequent than S/A omission, as often the O argument is part of the pragmatic assertion – new information, rather than part of the presupposition; and even once identifiable, is still more often than not overtly present. Formal presence of the O argument is motivated by the very nature of this grammatical relation, which differs from that of the S/A argument by a host of properties, such as lack of autonomy – i.e. greater dependence of the action or property expressed by the predicate on the argument, etc.

. Information structure As a general observation on the information structuring in the Khanty narratives, it can be noted that the most frequent way of introducing a new referent in the beginning of the discourse is by a full NP or a free pronoun in a thetic-type clause (10). The referent, which is thereby discourse-identifiable and -active, is from then onwards marked by an unaccented pronominal co-reference, or a co-referential zero agreement (in case of a 3sg. argument) on the predicate (9). In the following section, Vasyugan Khanty text examples are used to corroborate the claims about the formal and semantic features of the established S/A participant contrasted with the new S/A participant, in relation to their discoursepragmatic features. In the selected narrative, in clause (12), the central referent – 1Pl in the S/A relation is activated and maintained (8 clauses in the narrative) as an unaccented co-referential inflection on the predicate:



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.8 (517-584)

Andrey Filtchenko

(12) nu ej to>Áoj-n6 ... to>Áoj-n6 men-s-6w qul-kant-wa-tati OK one spring-Loc spring-Loc go-PST2-1Pl fish-find-INF-Sup Gll-atG down-Lat ‘Once in spring..., in spring we went down to fish’

Formal absence of this referent, as a free pronoun, may be attributed to a certain cataphoricity effect, as most of the participants, implied under 1Pl here, are identified by full NPs and free accented pronouns further in the discourse. This 1Pl central discourse referent is also inclusive of the author of the text (1sg SAP), which is a feature of the genre of auto-biographical narrative, and thus, already has high situational accessibility. Once the referent is identifiable in the discourse and has high activation status in the interlocutors’ discourse universe, it can undergo temporary or terminal deactivation as a result of introduction or activation of a new or competing active referent in the proposition: (13) nu jemaki kall-6w – Ok spend night-1Pl ‘Ok, so we stay for one more night’ (14) mä sart wel-s-6m, ¨fll¨6 1sg pike kill-PST2-1sg big ‘I caught a pike fish, big one’ (15) ¨fll¨6 sart ratw män-n6 lö>ö-li-s-im Big pike oldman 1sg-Loc get ready-Intr-PST2-1sg/sg ‘I got the big pike ready’

Clause (14) demonstrates a change of the topical 1Pl we referent in the S/A relation, which is active and pragmatically central in (12), and is expressed by the verbal agreement with the 1sg S/A in (14). As anticipated by the pattern, the change is marked by a free accented 1sg pronoun. The 1sg referent is maintained as topical further on, expressed only by the 1sg predicate inflection. The stretch of discourse (16)–(20a) represents a sequence of changes of the central discourse referent from ‘bear’ to the 1sg S/A ‘I’, and it deviates from the established pattern in that the change is not marked by a full NP or free pronoun. Rather, each referent takes turns being maintained as a topic by elision and verbal inflection. In (16) the 3sg S/A referent ‘he/she/it’ assumes the topic relation, marked by a free pronoun, as expected by the pattern, and is maintained as the center in (17) marked by elision and an absent S-V agreement (in the case of 3sg argument). In (18), the 1sg ‘I’ reappears as the center, not marked by a free pronoun, as expected, but only by elision and verbal inflection.

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.9 (584-652)


(16) ju>a jG>G jor-n6 nu> lo>G-m-a>G 3sg river middle-Loc up lie-PP-Pred ‘In the middle of the river he floats, just resting there’ (17) lo>G-w6l Lie-PRST.3sg ‘He stayed there (on the water)’ (18) twila>ta-s-Gm˜ rut’ sa>G “medwed” cry-PST2-1sg/sg Russian along/way “bear” ‘I cried in Russian “bear!” ’ (19) “moŠet jG>G-n6 kol-wa>ta-l-il” “maybe” 3Pl-Loc hear-Attenuat-PRST-3Pl/sg ‘ “Maybe they would hear it” ’ (20) a.

nu jemaki, ji>ata-l-Gm, – good look-PRST-1sg/sg b. “aha, waja>.” OK, animal ‘Ok, I look “OK, there it is” ’

The absence of overt expression of both of the participants (‘man’ and ‘bear’) suggests that, at this point in the narrative, they together constitute the foreground in the narrative. As both the literal (spatial) and pragmatic distance between them decreases, they are simultaneously on the stage. This also allows them to maintain a certain economy and dynamic in the narrative that is pertinent to the particular described situation. This is supported by the fact that this kind of “dynamic pragmatic alternation” is used again in the subsequent discourse for these same referents. (19) presents an inner speech quotation – the thought of the narrator, where the S/A argument 3Pl ‘they’ is marked with Loc case. The predicate in (19) is inflected for a 3Pl S/A argument and for a definite sg O argument, i.e. ‘it’, the bear. Since (19) is outside the narrative stream, it does not displace the 1sg S/A, the narrator from the central position, and he continues as the topic in (20a) appropriately expressed by elision and 1sg verb inflection. Similarly, (20b) quotes the inner speech of the narrator. Thus, we can sum up with regard to the pragmatic organization of the Khanty narrative: 1. the way to introduce a brand-new referent into the discourse, or to reactivate it as a participant in the discourse with the topical pragmatic status, is by marking it with a full NP or a free pronoun in the S/A relation and by the respective S/A-V agreement inflection. Inner speech quotations, asides and such, do not alter the pragmatic status of the arguments and do not require their formal reestablishment;



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.10 (652-735)

Andrey Filtchenko

2. once the referent is identifiable as the topic at the current stretch in the discourse, it is expressed by an elision and objective verbal agreement inflection; 3. several discourse referents can have compatible prominence, representing closely associated participants simultaneously occupying the stage in the narrative. They can alternate in the S/A relation in conjoined or adjoined clauses being marked only by elision and respective verbal agreement inflection. In these cases, the alteration of such pragmatic “equilibrium” is signaled by the reversal to marking of the current topical referent with an NP or free pronoun in the S/A relation and the respective verbal agreement inflection. The exceptions to the above mapping from the pragmatic functions to the semantic roles and to grammatical relation are: 1. clauses with no clear topical referent (background setting, thetic, event reporting); 2. clauses with no clear agentive argument, or with an inanimate referent in S/A relation; 3. clauses with the topical referent expressed by an argument with the semantic role of Target in the S grammatical relation. The above correlation between the pragmatic function and grammatical relation prevails over that between the semantic role and grammatical relation. The latter is evident from the fact that, although the correlation between the pragmatic topicality, the semantic role of Agent and the grammatical relation of S/A has, by far, overall prevalence, as in some instances arguments with the semantic role of Target correlate with a topic and with the grammatical relation of S. At the same time, in such constructions, arguments with the semantic role of Agent are oblique-case marked and assume a non-S relation. We can re-affirm universal correlation (Lambrecht 1994) of the pragmatic status of a referent’s formal complexity, in that the continuum between zero and maximal morphological explicitness is counter proportionate to the continuum between pragmatic centrality, activation and unidentifiability, inactiveness (Table 1). That is, in Eastern Khanty, the S/A argument normally has high activation status – is a topic of the proposition, typically expressed by a elision and verbal agreement, whereas a brand-new discourse referent has low discourse status typically expressed by a full NP and S/A verbal agreement inflection. Table 1. Morphological marking NP(+agreement) (–)

→ ←

pronoun(+agreement) central/active

→ →

zero argument (+agreement) (+)

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.11 (735-782)


. Non-canonical constructions There are a number of clauses in the narrative that consistently deviate from the above prototype in their formal marking of the grammatical relations of the arguments in the proposition. These include the Eastern Khanty agented passive, representing around 8% of the clauses in the narratives (all passives in the narratives average 16%), and the so-called “ergative” constructions, averaging 12% of the clauses. The structural properties of these constructions are neither typologically unique, nor previously undescribed (cf. Kulonen 1989). However, what somehow eluded previous accounts is any depth of identification of the motivation of this grammaticalized correspondence of form to function, i.e. motivation of the choice of every of these particular non-canonical construction type. In the following part of the paper, we examine in more detail the instances of these construction types in their narrative discourse environment in an attempt to obtain a satisfactory answer to this question. . Structural properties of the non-canonical constructions .. Structural characteristics of the Eastern Khanty agented passive The functional domain of passive is shared in Khanty by two types of constructions with regard to their formal and semantic properties. Constructions of the first kind, referred to as stative passive, express situations that represent a state after the action/event, structurally represent a nonfinite verb form in an attributive function (participle) in the perfective form, also often generally associated with passive semantics. These are also occasionally referred to as resultative passive constructions. This correlation between the resultative and the perfective on one hand and the passive on the other, is well attested and described as resulting from the “inactive nature of the passive subject (patient)...”, which has “the effect of shifting the perspective from the agent’s side to the patient’s, and accordingly from the beginning to the end of the event” (Shibatani 1985). (21) min atwIsa-k6n men-kal-men wal-m-aw qat-a 1Du brother-Du go-PST1-1Du live-PP-POSS.1Pl house-Lat ‘We, two brothers went to the house where we lived’

Khanty participial/non-finite constructions have been described in adequate depth and representativeness by various Siberian scholars (Skribnik & Kovgan 1991; Cheremisina & Koshkareva 1991; Kovgan 1991). The kind of passive construction that is one of the foci of attention in this paper is the dynamic passive, structurally representing a morphologically passive finite verb form and expressing events that are inactive from the point of view of the S argument, in this case, the semantic role of Target:



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.12 (782-849)

Andrey Filtchenko

(22)

‘My younger brother and I were taken by our parents from the boarding school for holidays’

This group of constructions can be differentiated into subgroups based on the status of the Agent referent mapped onto O grammatical relation. The first type of agented constructions may describe situations with either one or more event participants. The affecting referent is lacking control, authority, volition and can be even inanimate, typical of fairy tales with supernatural agents/events and idiomatic expressions: (23)

‘The eyes are smarting with the smoke’ (Lit. the eyes are eaten by something)

The second type has no overtly expressed Agent argument: (24)

‘(We) put some petrol in the snowmobile’ (Lit. the snowmobile was filled with gas (by us))

In cases when the agent, in the S/A relation of an active-direct clause, is an inanimate entity or is devoid of control, authority, volition, it obtains the semantic quality of Force, i.e. situations without a typical Agent, corresponding to “process”, describing an event that occurs without a visible cause, i.e. ‘automatically’ (25), which makes them in a sense similar to anti-causative constructions.6 (25) pami sö> pä wetw->el->äl-i hay bunch also light up-Rfl-PST-PS.3sg ‘The hay bunch also caught fire’ (Lit. hay bunch also got lit up)

(22) and (23) testify to a S-V agreement pattern in the passive construction, though the S referent is not always overtly expressed by a free constituent, as in (24). . These constructions are referred to as “Automative use of causatives” by Kulonen (1989).

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.13 (849-893)


There are apparently two groups of verbs that are commonly found in these constructions: (a) verbs occurring in both active and passive forms, i.e. having, in the active, a causative/transitive sense, and in the passive describing an event with the S relation taken by the Target role and an Agent in the non-A relation (25) (Kulonen 1989); (b) automative verbs, often only occurring in the passive or the inactive semantics. The function of these is similar to that of medial verbs (Kulonen 1989), in that they cannot be used to express an action with an Agent, but in “process” semantics instead. The agentless passive clause with an inanimate agent and automative verbs are also frequently associated with a certain spontaneity of the event, which “is highly germane to passive defocusing (demotion) of an agent, as an event that is brought about by an unknown/unmentioned agent is perceived as spontaneous” (Shibatani 1985: 838). There is also a group of passive constructions with motion/posture verbs that can be revealing for the notion of transitivity in Khanty. These verbs are apparently neither prototypically intransitive, as they are associated with an O argument (location), nor are they prototypically transitive, as the second referent with a central role in the event, O, is usually marked by one of the oblique cases and has locative semantics, as in (26) below: (26) tüt pirn6 ju> ont-nam ti m6n-i Dem after forest inside-Lat1 Dem go-Pst0.PS/3sg ‘After this we went to the woods’ (Lit. after that into the forest it was gone)

(26) may also be seen as exemplifying the impersonal passive construction, traditionally describing events, where the demotion of an Agent referent (Shibatani 1985) in the A relation of the active-direct clause happens without the promotion of another referent, (while in the personal passive there is a promotion of a referent from the O relation in the active-direct to the S relation in the passive) (Comrie 1977). So impersonal passives just demonstrate demotion of the Agent (Shibatani 1985) expressing the action of an indefinite Agent unaccompanied by promotion of an inactive referent. The event structure here is underlyingly the same in the passive and the active. The motion verbs are seen to behave here in a way transitively, having the locative referent as one of the core arguments – in the O relation and occasionally displaying O-V co-referential inflection. Such “transitive” behavior is attested not exclusively in the domain of the passive, but also in the active-direct (Kulonen 1989): (27)



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.14 (893-961)

Andrey Filtchenko

‘the dogs chased the elk and caught it’ (Lit. the dogs chased the elk and made it come)

Both verbs in the clause show objective/definite agreement inflection, providing information both on the S/A and O referents of the proposition (3Pl=A, sg=O), thus demonstrating the features typical of transitive verbs. These occurrences are rare, demonstrating that in Khanty a participant (even that of a prototypically intransitive situation) can be made part of the semantic core for pragmatic purposes, “if the Loc/Ben constituent is important for the situation and bears firm relation to the verb, its status can be emphasized grammatically by transitivization” (Kulonen 1989). (28) Ø äj-ku-n6 os jo>t-i (3sg) young-man-Loc again come-PS/3sg ‘The young man came to him again’ (Lit. (he) was come to by the young man)

These motion verb passives, i.e. passivization of the locatives in Khanty, are conceptually feasible, as in these cases the locatives denoting a certain landmark of motion (source, goal, relative orientation) are important, if not inseparable, parts of the semantic frame of the event expressed by the proposition. And “all entities which correspond to the elements of a semantic frame or valence can be considered as focused to some extent ..., that is, they are singled out as essential elements, requiring the listener’s attention in decoding the message; they are highlighted against the background of all other entities which may be in the consciousness of the speech-act participants, but are not semantically coded” (Shibatani 1985: 832). .. The structure of Eastern Khanty Loc-S/A constructions Another type of a non-canonical clause is the so-called Eastern Khanty “ergative”7 construction type: (29) ¨fll¨6 sart ratw män-n6 ö>öli-s-im big pike oldman 1sg-Loc prepare-PST2-1sg/sg ‘I got the big pike ready’

As will be shown below and throughout the paper (cf. sections 4.3.2 and 5.), “ergative” may not be the most precise terminological choice for this construction type. For our purposes we will hereafter refer to this construction type as Loc-S/A clauses. . The term “ergative” in reference to the described construction type is inherited from the previous descriptions (Balandin, Comrie, Kulonen). The type of constructions should be structurally and functionally similar to what Dixon (1994) referred to as marked nominative constructions, where the S/A has a marked case instead of the O.

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.15 (961-1042)


At first glance these constructions demonstrate great structural similarity to the canonical active-direct clause type, with, perhaps, two important exceptions: the S/A argument is always overt, expressed by a full NP or a free pronoun, and it is always inflected with Loc case. This Locative marking of the agent referent is, although not unique, quite rare typologically. The more complete list of formal properties of this construction type is as follows: 1. Frequency: average 10%. 2. The argument with the semantic role of Agent is overtly present in the proposition, expressed by a free pronominal NP or a full NP (fairly high on agentivity properties: identifiable, accessible, predominantly human/animate), marked with the Loc case. 3. This Agent argument marked for Loc case is mapped to the grammatical relation of S/A showing the co-referential verbal inflection on the predicate. 4. The predicate of the proposition is a verb normally expressing a perfective action frequently with an unclear affect on the Target referent, i.e. underlyingly de-transitive (‘take aim’, ‘look’, motion/posture, ‘shoot’ not implying ‘hit’, body part manipulation, which are mainly intransitive events). 5. When present, the second core argument of the proposition with the semantic role of Target is expressed by a full Ø-marked NP, mainly definite (possessive-marked, identifiable, accessible contextually); locus of motion or state, direction (object of looking or target of shooting). A brief outline of the statistical, pragmatic, structural and semantic features of the Loc-S/A constructions are laid out in Table 2 below. Table 2. Loc-S/A construction features S/A-Loc – Prs.Pronoun: 71% – Proper Nouns: 16% – Nouns (identifiable/ accessible): 18% – + NP animate/ non human 16% – + NP inanimate 6%

V

O

– motion verb: 6% (omitted) + Loc of motion direction – posture (sit/lie): 16% + Loc of whereabouts – perception (look/see, hear): 18% + what is perceived (seen/heard) is a separate clause – ‘take aim’ – intransitive: 16% – ‘shoot’ – 16% + 18% (omitted) + Lat/direction of shooting – body parts (head/nose) manipulation (shake/hide/move/stick/ put): 30% – de-transitive/anticausative (cut): 6% + definite object NP (fish)

– no O argument: 18% – Loc/Lat O: 41% – locus of motion/posture – direction of looking/ shooting – definite NP=O: 41% – body part: 30% – “message” of saying: 6% – definite object NP (fish): 6%



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.16 (1042-1109)

Andrey Filtchenko

It can be seen that nothing in the grammar of these propositions precludes the use of the canonical active-direct clause type to formally express the same semantic content. The question that arises then is: What motivates the choice of this particular non-canonical construction type? . Eastern Khanty non-canonical constructions in the narrative We posit that such notions as discourse-pragmatic status, pragmatic activation, and topicality are central to the discussion of the functional characteristics of these non-canonical constructions. We define a referent as topical based on a host of properties such as: its belonging to the presuppositional part of the proposition, its being contextually accessible and active, in dislocation tests (“as for” and “about”) it produces the target clause (Kuno 1972; Gundel 1976; Lambrecht 1994), it normally does not carry the clause accent, and the rest of the proposition appears to carry a relation of “aboutness” towards it (Lambrecht 1994). .. Passive constructions in the narrative The first non-canonical construction type to be reviewed in its natural environment, is the Eastern Khanty agented passive, i.e. those passive clauses having explicit agents. In the narrative, clause (30) below maintains the referent ‘we’ as topical, according to the established pattern, by elision of the Agent in the S/A relation and by verbal agreement inflection, which is then altered in (31)–(32), by allowing the referent with non-agentive properties, typical of the semantic role of Target/Undergoer, to be mapped onto the S relation, and controlling the verbal agreement inflection (3sg zero agreement). (30) juta put w6-jm6n pan6 lit6 ot together kettle take-1Du/sg and food thing ‘We took with us a kettle and some food’ (Lit.: with us a kettle was taken and some food) (31) a.

puran pensin-nat pon-i, snowmobile gasoline-Com put-PS/3sg b. aw6t-at jur-i, sled-Com tie-PS/3sg

(32) juta amp->6n w6-jin together dog-Du take-1Du ‘We put some petrol in the snowmobile, tied sled to it, and took two dogs with us’ (Lit.: The snowmobile was fueled, sledge attached and dogs tied up)

In clauses (31)–(32), the topical S/A referent of (30) is not overtly present, whereas a referent with the semantic role of Target, ‘snowmobile’, is expressed by the full NP in (31a) appearing as the topic and assuming the S relation, showing 3sg zero

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.17 (1109-1168)


agreement on the predicate. Such re-arrangement, in this stretch of discourse, is necessitated by the pragmatic demand to establish and maintain a non-agentive referent as pragmatically prominent, while the agentive referent, previously established as topical, is temporarily demoted as contextually obvious or of lesser interest. The S argument is elided in (31b) and is still controlling zero agreement on the predicate. The 1Du agentive referent, however, effortlessly resumes as a topic as soon as the “special” passive re-arrangement is no longer maintained in the active-direct (32), expressed by elision and verbal inflection. Example (33) represents a sequence of passive and active-direct clauses, where in the passive (33a) the agentive referent is expressed by the full NP ‘my man’ marked for Loc case and obtaining the non-S relation, whereas a Target referent ‘snowmobile’ appears in the S relation being expressed by the full NP and 3sg zero agreement. (33) a.

ej pG>r-6m-n6 puran nö>t-i pan6 one man-POSS.1sg-Loc snowmobile pull-PS/3sg and b. sar-nam m6n ahead-Lat1 go.3sg ‘My man turned the snowmobile on and went on forward’ (Lit: By my man the snowmobile was started and he went forward)

In the coordinated (33b) conjoined by pan6, however, the agent of (33a) ‘my man’ appears in the S/A relation expressed by elision and 3sg zero agreement, which, according to the previously established pattern signifies topicality of the discourse referent. The possible question – which argument in (33a) the predicate in (33b) actually agrees with (as both the oblique-marked Agent and the Target in the S position are 3sg arguments) – can be clarified by (33d) and (33e) below, which are in many ways similar. In the active-direct (33c) the topical referent is expressed by the pronominal argument 1sg with the role of Agent in the S/A relation and verbal inflection. It is then de-activated and pragmatically backgrounded in (33d) taking the O relation, expressed by the Loc-marked 1sg agent pronoun, while a referent with properties testifying to its relatively lower agentivity status is pragmatically foregrounded, necessitating the special passive re-arrangement where it is expressed by the full NP in the S relation controlling the verbal agreement. os mä avet-a 6nt6 Gmt-6m but 1sg sled-Lat NEG sit-1sg d. aj amp-6li män-n6 kur->6t-i small dog-Dim 1sg-Loc leg-Pl-Lat e. pan6 puran pGr-i ti and snowmobile back-Lat2 Dem

(33) c.

katS-i hold-PS/3sg qu>t-6m, trod-1sg



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.18 (1168-1237)

Andrey Filtchenko

‘But I did not sit in the sledge, with the small dog held by me by the legs, I ran behind the sledge’

In the conjoined (33e), however, similar to (33b) above, the 1sg agent referent appears in the S/A relation of the non-finite clause. The agent referent consequently reemerges as the topical S/A referent in the active-direct (33e) once passive non-agent foregrounding stops. Both sequences (33a–b) and (33c–e) contain active-passive-active clause combinations. In each of the passive clauses, the referent with the semantic role of Agent is expressed by the oblique-marked NP (nominal or pronominal). This referent, however, still enjoys a degree of pragmatic activation strong enough to allow this referent to appear as topical, expressed by elision and the verbal agreement inflection in the immediately following discourse. That is, the Target referent in the S relation in the passive clauses increased their pragmatic activation during the passive clause(s), but it did not interrupt the topicality of the Agent referent at the overall discourse level. This all leads to the conclusion that passive is not used to establish a new discourse topic. It is capable of setting the stage but does not promote the discourse pragmatic status of a referent. Notably, one of the structural distinctions between (31a–b) and (33a, d) is the overt expression of the agentive referent. We posit that the question of what requires the overt expression of the agent referent in some passive clauses could be approached via the analysis of the information structure in the narrative. It is evident from observing this stretch of discourse that the “agentless” passives (lacking an overt agent referent) are preceded and followed by active-direct clauses with the same agent referent (1sg), which testifies to the identifiability and topical status of this referent in the interlocutors’ universe. It is also evident, however, that in cases of “agented” passives this agent referent is an essential participant, and if elided, would render the proposition ambiguous. The overt presence of the agent is thus needed for the adequate/intended processing of both the information/locution content and communicative/perlocution effect (humor) of the proposition. The analysis of the following sequence (34a–b) reveals a similar pattern. In the (34a) the topical discourse referent, agentive 1sg S/A, is expressed predictably by elision. (34) a.

op6l-6m qot mutw6 puran pGr-i qu>t-6m, sister-POSS.1sg/sg home until snowmobile back-Lat2 trod-1sg b. tü lat-n6 aj amp-6li ma-n6 iS6-ti asS-i Dem time-Loc small dog-Dimin 1sg-Loc front-Lat let go-PS/3sg ‘I ran behind the snowmobile all the way to my sister’s house and the doggy was let go by me’

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.19 (1237-1293)


In the adjoined passive (34b), the Target referent ‘small doggy’ is expressed by the full NP in the S relation controlling 3sg verbal null agreement, whereas the Agent referent is expressed by the free Loc-marked pronoun in the non-S relation. The topic in (34b) is the S/A argument ‘small dog’, whereas the 1sg agent referent is apparently temporarily backgrounded. However, the relevance of the agent in the proposition (34b) is still manifested through its overt presence, presumably to minimize ambiguity (since the dog is “held/bound” on both sides, by the sled and 1sg human referent). In the active-direct (35), the 1sg agentive in the S/A relation remains topical, expressed as expected by elision and verbal agreement, as well as controlling the S/A relation over the non-finite modifiers (typically also inflected for 1sg). (35) puran pGr-i qot-m-am-n6 näxt-6m, otj Si>pil snowmobile back-Lat2 trod-PP-1sg-Loc laugh-1sg yard inside Sa]-m-am-a imat sar-nam nä>-ta j6>-6m enter-PP-1sg-Lat more ahead-Lat1 laugh-INF become-1sg ‘While running behind the snowmobile I laughed and having entered the front yard, I laughed even more’

Finally, (36) demonstrates the so-called Khanty passivization of motion verbs. It should be noted that both in active and passive verbal clauses motion verbs reveal transitive features, namely possible O-V agreement with the NP[location] argument in active clauses and allowing passivization, as in (36), where the referent expressed by the NP[locative] appears as the only overt referent present. The verbal agreement inflection in this case is the 3sg null agreement. (36) tüt pGrn6 ju> ont-nam ti m6n-i Dem after forest inside-Lat Dem go-PS/3sg ‘After this we went to the woods’ (Lit.: the woods were went to)

The referent with the semantic role of Agent is again elided, but is identifiable and has a high degree of pragmatic activation, accessible both situationally and textually. We can now outline the discourse-pragmatic based generalizations regarding the use of passive in the Eastern Khanty narrative: 1. clause type frequency in the narratives ∼ 14%; 2. the passive is the clause type showing a deviation from the canonical activedirect arrangement, in that the referent with the semantic role of Target is mapped onto the S grammatical relation, controlling the S-V agreement on the predicate; 3. the referent with the semantic role of Agent is mapped onto a non-S/A grammatical relation, being marked by oblique case morphology, or elided;



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.20 (1293-1357)

Andrey Filtchenko

4. semantically, the agentivity status of the S argument (Target) of the passive is typically relatively lower than that of the non-S argument (Agent); 5. pragmatically, the passive marks a change in the degree of topicality of the referents, temporarily foregrounding the status of the non-agent referent, rendering it in the S relation; and backgrounding the status of the agent referent rendering it in the non-S relation; 6. however, while at the clausal level the pragmatic status of the referents is altered by the passive, at the level of overall discourse the agentive referent, temporarily demoted when relevant, or omitted when obvious/irrelevant, maintains high activation status, which follows from its canonical expression by elision and verbal agreement in the passive-active clause sequences; 7. (active-)passive-active clause sequences such as (33), when contrasted with the active-active ones, testify that passive is a marked construction type, requiring a special arrangement of the referents, which is outside the general pattern of mapping pragmatic function – to semantic role – to grammatical relation; 8. all reviewed passive constructions demonstrate, that the alignment of <pragmatic function=grammatical relation> in eastern Khanty is the strongest, overriding that of <pragmatic function=semantic role> or that of <semantic role=grammatical relation>. That is, the canonical active alignment of the active-direct construction is altered in the passive construction to be . With regard to information structure, it was noted above that the preferred topic expression in Eastern Khanty is by unaccented co-referential pronominal inflection on the predicate, including zero agreement in the case of a 3sg S/A argument. It was also noted that applying a variety of topicality tests it is evident that the referent with the role of Target, in the S relation in the passive clause, has the relation of aboutness in the proposition. However, it is seen that the referent with the role of Agent, demoted in the passive clause from the A relation, retains, to a large extent, the set of pragmatic properties that allows it to emerge as having the function of topic in the immediately subsequent discourse (after passive) without any special topic promotion means, i.e. expressed by elision and verbal inflection – the preferred topic expression. Thus, the topicality of the agent referent appears to have been maintained in the course of the passive. This above “residue topicality” of the agent referent in the passive constructions correlates with the fact of retention by the agent argument of some of the grammatical features normally assigned to the S/A relation. That is, in the Eastern Khanty examples, there is a certain distribution of properties that is characteristic of the S/A relation (subject properties) between the promoted non-Agent/Target referent on one hand and the demoted elided or overt oblique-marked agent ar-

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.21 (1357-1434)


gument on the other. Defining the subject as an element of the clause possessing a convergence of properties characteristic of subjecthood, such as: control of coreferential agreement in conjoined or adjoined clauses; control over non-finite clauses embedded within the matrix clause; control over reflexivization, we see that in Eastern Khanty narratives, in passive clauses with an overt agent argument these properties characterise the demoted Loc-marked agent argument equally well as they do the promoted Target referent in the S relation. With respect to control of co-reference, we see that though the control over verbal inflection in a passive clause is by the promoted Target referent, the control over an embedded non-finite clause is by the demoted Loc-marked Agent, rather than the promoted Target. In the sequence (34b)–(35), in (34b) the passive demoted Loc-marked agent associates both with the embedded non-finite clauses and with the 1sg S/A co-reference in (35). Control over reflexivization can be maintained both by the S/A argument and the O argument that co-occurs with the O-V agreement on the predicate. The reflexive/possessive affix on the locative ‘house’ in (37a) may refer both to the S/A ‘bear’ and the O ‘dog’, when accompanied by O-V agreement. (37) a.

waja>-n6 amp jo> nirimtä-s-tä twimin ont-kat-al animal-Loc dog home pull-Pst2-3sg/sg there inside-house-3sg ‘The bear hid the dog inside his (bear’s) / (dog’s) house.’

But retaining of control over reflexivization by the demoted agent referents in passive clauses is, perhaps, not a valid test in Eastern Khanty, as both Agents and Targets can bind the reflexives. Thus, in Eastern Khanty passive constructions there is a certain curious distribution of grammatical and pragmatic properties, commonly assigned to subjecthood, between the demoted oblique-marked agent and the promoted unmarked target argument. Pragmatically, we can also speak of the continuum of topicality (or nontopicality for that matter) or a certain foreground/background dynamic, where, in the passive clause, the non-Agent (Target) referent in the S relation is temporarily foregrounded against the background of the active-direct discourse with a primary discourse topical referent, which is expressed in the passive clause by the demoted Agent referent. The above-mentioned possible distribution of both pragmatic and grammatical features between the two arguments is most relevant for just one type of the passive clause, namely the one with the overtly expressed oblique-marked Agent referent in the non-S relation (agented passive) and less relevant for passive clauses with an unexpressed/obvious/irrelevant (agentless) or unknown (impersonal) agent referent.



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.22 (1434-1472)

Andrey Filtchenko

As shown above, all the predicates in the non-canonical examples are prototypically transitive verbs implying two core arguments, one of which has high agentivity status, volition and control properties while the other is lacking in such properties and is affected in the event. This pragmatic and semantic context for passivization appears to resonate with prototypical passive features by Shibatani’s passive prototype (1985: 839) and Givón’s “promotional detransitivization” (Givón 2001: 126–128) such as: – agent marked by an oblique case, conceptualised, but defocused at the level of syntactic encoding; – the syntax of the passive diverges from that of the active-direct; – the range of non-agent roles that can be promoted to subjecthood in a passive clause tends to be restricted to semantic Target or grammatical O; and – passives tend to be restricted to semantically transitive verbs, those having both agent and patient in the semantic frame, but syntactically, are typically intransitive. This list of prototypical features places the Eastern Khanty passive clauses with overtly present agent arguments somewhat at the periphery, as such agented passive clauses appear to deviate from the prototype. Particularly, overt syntactic presence of an agent referent in a passive clause upsets some of the prototypical features, however, these clauses appear to resonate with the general tendency of the passive, that is, its fundamental function of having “to do with defocusing of agents” (Shibatani 1985: 831). The functional/pragmatic features that are commonly associated with and are motivating the agentless passives, such as: agent-unknown, agent superfluous/obvious, agent absent for specific reasons (tact/delicacy), agent is of lesser interest, spontaneity of the event, etc. (Shibatani 1985; Givón 2001), appear to apply, to various extent to all the Khanty constructions that we have assigned to the domain of passive earlier. Thus, based on the narrative discourse analysis, it can be observed, that: (i) agentless passives (32a–b), (36) describe a prototypically transitive event that implies the agent, whose omission is contextually motivated by its superfluousness/lesser interest, which follows from its (the agent’s) topicality in the immediately preceding and following active-direct clauses (30) and (32), expressed by elision and verbal inflection. However, the other features (agent unknown or absent for tact or delicacy, spontaneity, etc.) hardly apply to the agentless passive clauses in the narratives above, as in the majority of the cases the events described by agentless passives are volitional, authoritative, deliberate or premeditated (‘fuelling’, ‘tying together’, ‘tying up dogs’, ‘going to the forest’) and generally devoid of adverse situation connotation; (ii) agented passives (33a, d), (34b), describe events where the agent referent is of high relevance: a new agentive referent replacing the previously established one, or important for the unambiguous interpretation of the proposition/event (33d), (34a). Incidentally, when contrasted, the events described by the agented passives appear to imply more spontaneity and affectedness of the Target, than those described by the agentless passives (cf. (31a–b), (30), (36) and (33a, d), (34b). That

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.23 (1472-1529)


is, a more prominent/new agentive referent of the agented passive clause implies a change, dynamicity, affectedness for the Target referent, whose perspective, according to common interpretations, is dominant in the passive arrangement, whereas a known/superfluous agent of the agentless passive is unlikely to imply spontaneity. This is consistent with the information structure pattern established at the onset of this paper, i.e. elision/omission corresponds to ‘known/topical’, whereas overt expression (full NP, free pronoun) – to shift/change – ‘new’ (cf. 3. Information structure). It appears reasonable then, to correlate the above-mentioned pragmatic continuum of topicality (foregrounding/backgrounding) with a certain formal/grammatical continuum, in the sense of considering various related forms as “... passives to the extent that they share properties of the passive prototype”, placing all these forms “... along the continuum between the passive and active-direct” (Shibatani 1985: 844). Such “fuzzy” treatment of the attested formal possibilities allows for a degree of flexibility and dynamicity typical of the language as a system, resonant with Shibatani’s “change in progress” exhibited by the language forms, which “a discrete analysis of grammatical structure does not allow us to capture” (1985: 846). .. Loc-S/A (“ergative”) constructions in the narrative Clause (38) demonstrates establishment of a new topical discourse referent – 1sg S/A controlling the verbal inflection. (38) mä sart wel-s-6m, ¨fll¨6 1sg pike kill-PST2-1sg big ‘I caught a pike-fish, a big one’

Clause (39) is the first instance of the Loc-S/A (ergative) clause type. (39) ¨fll¨6 sart män-n6 lö>ö-li-s-im big pike 1sg-Loc ready-Intr-PST2-1sg/sg ‘I prepared the big pike’

As will be seen in the analysis of all the Loc-S/A clauses in the narrative, (39) is anomalous in having the O argument in clause initial position. Such a position, preceding the S/A agentive argument, is an evidence of the increased degree of pragmatic prominence of this referent. The referent ‘pike’ is identifiable and highly accessible textually, which is also evident from the marked O-V agreement on the predicate. This results in a situation where there are two referents with a compatibly high degree of pragmatic activation. There are still, however, marked differences in the semantic properties of these referents, such as animacy/agentivity status, as well as their comparative discourse function. The 1sg, apart from being the SAP, is the author of the narrative and a central participant in the event. In (38)



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.24 (1529-1606)

Andrey Filtchenko

this human agentive referent appears in the S/A relation controlling S/A-V agreement inflection. The above-established pattern would predict further maintenance of this referent as the topic by elision and verbal inflection. However, counter to this expectation, in (39) this referent appears expressed by the free Loc-marked pronoun in the S/A grammatical relation and verbal inflection. There are other properties that make this clause stand out in the otherwise canonical active-direct narrative discourse flow, namely the intransitive/anticausative verbal affixation that contributes to the distinctness of the proposition affecting the sense of transitivity/subjecthood in the event. After this temporary alteration by the non-canonical (39) clause, the narrative discourse resumes in the expected canonical way in the immediately following active-direct (40), where the topicality of the 1sg referent is “canonically” expressed as an active-direct S/A, by the elision and co-reference on the predicate. (40) terkä-s-im iwes-n6 fry-PST2-1sg/sg stick-Loc ‘I fried it on sticks’

The next Loc-S/A clause (42a), while demonstrating familiar formal characteristics, i.e. a Loc-marked overt agentive S/A, manifests the beginning of a repair describing a temporally displaced coherent episode, licensed by the adverbial ‘before that’ in (42a) and completed by the logical return to the main narrative sequence, 7 clauses down in the narrative. (41) wäsk-äti k6m nö>ü-lök jG>a-ta-tG duck-Sup Conj(or) moose-track look/see-INF-Supine ‘Either to look for ducks or moose tracks’ twu Gl pele>-n6 mGn-n6 amas-ta ... koll6m kasi Dem ahead/down half/side-Loc 1Du-Loc sit-INF three man amas-l-6w sit-PRST-1Pl ‘Before that, we sit ..., the three of us are sitting (and see) b. jaqk6n ä saqq6n nuru>te-l-k6n bear-Du Conj. cub-Du run-PRST-3Du . . . a bear with a cub running’

(42) a.

(42a) is also peculiar as it contains a self-initiated repair from ‘we-two sit’ to ‘(we) three people are sitting’, where, interestingly, in the repair part an active-direct arrangement substitutes for the Loc-S/A clause; that is, a Loc-marked S/A in the initial part is changed into Nom S/A in the repaired proposition. Essentially, the number of the S/A participant is repaired from Du ‘we’ to Pl ‘us, three men’. One of the motivations of the change from Loc-S/A clause grammar (S/A=Loc) to activedirect (S/A=Ø) may stem from the alteration of pragmatic content in this repair,

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.25 (1606-1674)


that is, the change of referentiality of the participant (from pronomial-1Du to full NP-Pl), and a decrease in its pragmatic activation. (43) twila>t-6t-6m “rut sa>G ˜ medwed !” cry-IP-1sg Russian along/way “bear” ‘I cried in Russian “bear!” ’ (44) moŠet jG>G-n6 kol-wa>ta-l-il “maybe” 3Pl-Loc hear-Attenuat-PRST-1Pl/sg ‘Maybe they would hear it’ (45) a.

nu jemaki, ji>ata-l-Gm, – good look-PRST-1sg/sg b. a>a, waja>. OK, animal ‘Ok, I look, there is the animal’

In the next example of the Loc-S/A clause, the canonical clause type (43) with the topical discourse referent – the Agent in the S/A relation expressed by elision and S/A-V verbal co-reference, is interrupted by the Loc-S/A clause type in (44) with a new referent temporarily foregrounded as a topic, expressed by the Locinflected pronominal agentive S/A. It also represents a type of an aside containing the inner speech of the narrator. The central discourse referent (1sg S/A) is consequently re-activated as topical in the following canonical active-direct clause (45a) ‘appropriately’ expressed by elision and verbal inflection. The Loc-S/A clause (47a) is yet another instance of the described noncanonical pattern. It (47a) temporarily establishes a new prominent referent, by the Loc-marked pronominal Agent in S/A relation, against the background of the preceding active-direct (46) and the following active-direct (47b) with the topical discourse referent expressed by elision and S/A-V zero agreement. (46) pestätGl tom to>-6m-al pelk-a nörö>-w6l ... i fast Dem go-PP-POSS.3sg side-Lat swim-PRST.3sg “and” twel-w6l cry-PRST.3sg ‘He swims to the other side ... and yells’ män-n6 6wo ... jo>o-ta 6nt6 uspet wer-s-äm, 1sg-Loc again... shoot-INF NEG “be on time” do-PST2-1sg b. twerä nörö>-w¨6l Dem swim-PRST.3sg ‘I didn’t shoot in time, it swam so fast’

(47) a.

Thus, as a generalization of the use of Loc-S/A clauses in the narrative, it can be concluded that in addition to the listed structural features above, the following holds true:



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.26 (1674-1735)

Andrey Filtchenko

1. Pragmatically, the Loc-S/A clause marks a temporary alteration of the discourse topic, parenthetically establishing a new topic expressed by the Locativemarked S/A argument. This can also be viewed as an act of temporary foregrounding of a referent other than the current discourse topic of the narrative. 2. In every case, the topical discourse referent in the preceding active-direct discourse, which was altered by the Loc-S/A clause foregrounding act, reappears after being temporary backgrounded in the Loc-S/A clause, expressed by elision and verbal inflection, thus maintaining the topic, and continues as a topical agentive S/A argument in the consequent canonical active-direct discourse. .. Comparative analysis of active-direct vs. Loc-S/A (ergative) constructions Seeking a functional-pragmatic motivation of the choice between the Loc-S/A clause and active-direct clause we contrast below the instances of each construction in their narrative discourse environment. The first pair readily yields itself to a contrastive analysis, in the above mentioned sequence (38)–(39), the active-direct clause (38) is immediately followed by the first Loc-S/A clause (39). Having the same pronominal 1sg agentive S/A referent ‘I’ both clauses display such features as: S/A argument controlling verbal agreement inflection, high agentivity of the S/A referent, verb-final word order and perfective aspect of the transitive verb. However, there are differentiating features as well: 1. Ø-marked Agent-S/A in active-direct (38) vs. Loc-marked Agent-S/A in the Loc-S/A (39), 2. subjective verb conjugation in (38) vs. objective verb conjugation in (39), 3. SOV order of (38) vs. OSV8 order of (39), 4. brand-new unidentifiable O in (38) vs. identifiable/active O in (39), 5. accented 1sg S/A pronoun in (38) vs. unaccented Loc-marked 1sg S/A in (39), 6. transitive event with affected 3sg Target referent in (38) vs. de-transitivized verb in (39). These all correlate to the general pragmatics of topic-comment (38) with a 1sg agentive S/A Nom pronoun vs. marked/focused (39) with a de-emphasized Locmarked S/A pronoun. Here, the reference to the cultural frame may be of import. When caught, a fish has to be immediately processed (scaled, gutted and pickled in salt-brine) to avoid spoiling and attracting predators. Thus, in a sense, catching the

. OSV is discussed above, as associated with the status of the O – definite, triggering O-V agreement.

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.27 (1735-1815)


fish is part of a conventionalized routine. In this, the participant of the routinized event is to an extent topical. The second contrasted pair includes clauses with a full NP S/A argument (49) and (52). In the active-direct (49) the previous topical 1sg S/A referent, marked by elision and verbal inflection, is replaced by the referents expressed by the full NP; the agent in the S/A relation takes the status of topic, controlling S/A-V verbal agreement inflection, and proceeding further as the new topical participant in (50). (48) qulum jin-s-im ni> fish kebab stand-PST2-1sg/sg up ‘I put fish kebabs in(to the fire)’ (49) tom kät-k6n ˜ sidor igorenka sashka Gl kGl-kam tom Dem two-Du Sidor Igorenko Sashka ahead/down IndPn-Lat Dem m6n-s-6k6n ill6-pa down-Lat go-PST2-3Du ‘These two: Sidor and Igorenko Sashka, went somewhere down there’ (50) wäsk-äti k6m nü>ü-lök jG>a-ta-tG duck-Sup or moose-track look/see-INF-Sup ‘Either to look for ducks or moose tracks’

In the Loc-S/A clause (52), as mentioned earlier, a new referent expressed by the Loc-marked S/A NP ‘gun’ signifies a temporary interruption from the core discourse line, where this new referent appears as a temporary focus of attention for the length of (52)–(53). (51) nu>-la paja>-inta-s-im ämp-äm ajrit-na-ki up-Lat go-Trans-PST2-1sg/sg dog-POSS1sg/sg canoe-Loc-Pred ‘I went upstream, the dog is in the boat’ (52) sidar twö]wä pötwkän-n¨6 tow6 ajrit-n6 ola>-w6l Sidor late gun-Loc also canoe-Loc lie-PRST.3sg ‘Late Sidor’s gun is also in the boat, it lay there’ (53) a.

twenati kaliberam ... ola>-w6l pon-am trop-na ... twelve calibre lay-PRST.3sg load-PP buckshot-Com b. m6n-n6pa trop-na 1sg-Instr buckshot-Com ‘Twelve calibre lay here, loaded with buckshot... by me (loaded) with buckshot’

(54) a.

jemaki nu>la-pa ja>G-nta-l-Gm, good up-Lat paddle-Trns-PRST-1sg/sg ‘Ok, I paddled upstream ...’



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.28 (1815-1875)

Andrey Filtchenko

However, in (54a) the temporary backgrounded topical discourse referent, the 1sg narrator, regains/further maintains its status as the primary topic in the discourse without any special referent promotion means, i.e. only by elision and verbal inflection. Contextually, Loc-S/A clauses appear to have parenthetical, consequential semantics, that is, representing an action-reaction (effect) causal dependence on the preceding event, expressed by an active-direct clause immediately prior to the adjoined Loc-S/A clause (that is representing an implicature, not merely an entailment). In (55) and (56) complex clauses, the events of the coordinated Loc-S/A clauses are in consequential relation to those of the preceding active clauses. (55) os lG>6lta-l-Gm, ju>-n6 os o>-6l Gl again take_aim-PRST-1sg 3sg-Loc again head-POSS.3sg down wer-w6l-6t6 do-PRST-3sg/sg ‘I tried to take aim again, he had his head down again’ (56) twa o>-ol nu> al6m-s-6tt6, män-n6 twa then head-POSS.3sg up move-PST2-3sg/sg 1sg-Loc then lG>6lta-s-6m take_aim-PST2-1sg ‘Then he got his head out, I raised the gun’

Equally in clause sequence (57)–(58), the event in (58) is a consequence of the adjoined (57). (57) twinam jo>o-s-Gm there shoot-PST2-1sg/sg ‘I threw it (a stick) there’ (58) ämp-äli-n¨6 twinam pü]k-6l warta-kata-s-ta doggy-Dim-Loc there nose-POSS.3sg stick-Inch-PST2-3sg/sg ‘The dog started to stick its nose there (in the hole from the stick)’

It is thus justified to posit that the events described by the Loc-S/A clauses are in a consequential relation, reactive in their nature, dependent upon the events in the preceding discourse. By extension, it is reasonable to hypothesize that the Loc-marked S/A referents of these clauses, although mainly semantically agentive (definite human/animate), are deprived, at least in part, of the control, volition properties in the above reactive sense. This correlates appropriately with the oblique case marking – non-canonical for the S/A relation, and ultimately and importantly, with the above-mentioned discourse pragmatic failure of these Loc-S/A arguments to assume discourse topicality status.

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.29 (1875-1942)


.. Cross-linguistic perspective on the Eastern Khanty Loc-S/A constructions As it was already explored above, the Eastern Khanty Loc-marked agent demonstrates a mixture of features typical of both subject and non-subject arguments. Though demonstrating regular S/A-V agreement patterns, and agencysubjecthood features, its Locative case marking aligns this argument with the non-S/A locative arguments of motion/posture/state propositions – essentially intransitive in their nature. This widely correlates with the cross-linguistic observations for non-canonically marked S/A arguments, namely the fact that among the predicates requiring the non-canonical S/A marking, those expressing uncontrollable activities are numerous to the extent that the non-control vs. control may be “a generally applicable semantic feature with regard to predicates requiring non-canonically marked S/A” (Onishi 2001). It is also observed cross-linguistically that in general, noncanonical oblique case marking of core arguments “reflects decreased transitivity status of the whole clause” (Onishi 2001) owing to a set of multilevel factors, such as: the valency of predicates; the referential status of the NPs; and clause TAM, polarity, genericity, etc. in their integration (Shibatani 1985). These tendencies could be represented in the following adaptation of the Onishi continuum (Aikhenvald et al. 2001) (Table 3). Such observed tendency for non-canonical clauses to manifest the reduced control/volition of the agent referent in the described event finds interesting resonance with the early descriptive approaches to this phenomenon in some FinnoUgric languages. Observations were made about such constructions representing a “logically impersonal sentence” with a “covert subject”, when events were conceptualized by speakers as caused by “. . . other (mystical forces) – true agents”. A human – an apparent agent of change, is not granted agentive status, “merely representing a locus of an event”, whereas “the causative effect of outside forces is revealed, and the agent appears in essence a mere semi-responsible performer” (Balandin 1967). Extra-linguistically this approach finds strong standing in case of Eastern Khanty. It is notable that ‘bear’ is an extremely frequent participant in the events described by these non-canonical constructions. Given that ‘bear’ is a highly ritualized and tabooed cultural agent for the Khanty, and there are strongly observed Table 3. The Onishi continuum (Aikhenvald et al. 2001) (+) agent’s subjecthood (–)agent’s subjecthood (control/volition) (control/volition) (+) Clause/event transitivity Clause/event transitivity (–) Nominative Locative ←––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––→ ← canonical → ← non-canonical →



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.30 (1942-1992)

Andrey Filtchenko

rituals/habits associated with bear hunting and feasting such as use of masks and nicknames by hunters to conceal the identity of those responsible for the bear’s death, the tendency to demote or ‘de-agentivize’ (‘de-subjectivize’) particular referents in these constructions appears to be pragmatically reasonable. In accordance with conventional definitions, ergativity is a reference term for a system of organisation of the noun case marking where there is a morphological continuity between the subject of an intransitive verb and an object of a transitive verb, commonly accompanied by syntactic correspondences between S and O. The well documented (Comrie 1978; Dixon 1979, 1994) variety of manifestations of ergativity introduces less discreteness to the category allowing for observations of attested ‘ergative-like’ behaviour in otherwise prototypically nominative languages, making ergativity less a category than a language-organisation pattern. Generally, the examples of multifaceted ergativity manifestations and simultaneous features of the nominative and ergative organisation contribute to the perception of a somewhat continuous character of ergativity, prompting an absence of discreteness in the subdivision of languages into Ergative/Absolutive vs. Nominative/Accusative in an Aristotelian criteria-attribute manner. Ergativity or Nominativity might be construed then as a scalar (Comrie 1978; Dixon 1979), a dynamic prototype feature of structuring grammatical relations and of language functioning, which can be co-present, to a diachronically varying extent at different levels of organisation of the language system. We find that this pertains to the Eastern Khanty Loc-S/A constructions discussed above, as they represent, obviously somewhat marginal, pragmatically motivated ‘ergative-like’ behaviour in the otherwise nominative-accusative Eastern Khanty system. Instances of the co-occurrence of both systems of organisation of grammatical relations in the same language and even at the same levels of a language grammar are exemplified by many instances of split ergativity. Among the factors that underlie the dominant ergative, nominative or split organisation, it is often suspected that the information structure or larger discourse pragmatic and/or semantic considerations might be the conditioning factors, namely the correlation of the degree of volition/control/agentivity, exercised by the agent, with the ergativity/nominativity of the grammatical relations,9 and it is, in our view, exactly the . Nominalization, where S and A – arguments with subject properties – are replaced by oblique and O remain unaffected, is occasionally offered as another possible source of the origin of ergativity. Cross-linguistic observations concerning nominalization (incipient passivity of past participles, as well as non-Nom of the subject implying possible inversion of morphological complexity principle, i.e. – Nom more complex than Acc), are generally consistent with the Khanty data observations, where Loc is a frequent marker of the subject of the perfective participle construed as stative/resultative passive, which is used in the language system in abundance and in a multitude of functions.

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.31 (1992-2052)


motivation informing the Function element of the Form-Function correlation of the Eastern Khanty Loc-S/A (Ergative) constructions. The internal motivation of case-marking in general and ergative/absolute in particular, is seen in the identification of key participants of the proposition and their exact interrelations. A general claim holds that, for nominative languages the differentiation is made by prototypically making O more morphologically complex than S/A, and for ergative – making S/A more complex than O (Comrie 1978; Dixon 1979). This may seem superficially counter to our Eastern Khanty “ergative” or Loc-S/A clauses data observations, where the event expressed by the “ergative” construction is construed as largely intransitive. It should be born in mind that these Eastern Khanty Loc-S/A constructions are not the most prototypical ergative per se. However, morphological complexity of the S argument of the eastern Khanty Loc-S/A clause quite consistently correlates to prototypical ergative continuity between S and O at the deeper level, i.e. the S of the Loc-S/A clauses approximates the O of the active in its pragmatic and semantic features: decreased topicality and activity, thus approaching the semantics of the Experiencer in the event, where affectedness of the O, if present in the proposition, is unspecified.

. Conclusions on Eastern Khanty non-canonical constructions Lexical semantics, i.e., the degree of animacy, corresponding to potential agentivity and subjectivity of the referent, may provide enough information to prevent any potential ambiguity of the proposition and thus, cancel the need for additional explicit labelling of the roles. And indeed, this is the case in the Eastern Khanty canonical active-direct constructions, where NPs in the S/A and O relation are unmarked for case. However, specific pragmatic pressures may cause a language system to resort to specific marking of the arguments, particularly of Agent referents in cases of an unpredictable distribution of roles, where there is compatible pragmatic prominence and agentivity of the referents, or where there is an unpredictable discrepancy between the pragmatic status of referents, their high animacy and their decreased or de-emphasised agentivity. Such discrepancies may necessitate a special assignment of grammatical resources to semantic roles or grammatical relations, agented passive and the non-canonical Loc-S/A clause arrangements. The prevalent low morphological complexity of the nominative compared to other cases and the absolutive compared to the ergative, concurrent with a certain similarity in correlation between (full) NPs and pronouns in passive nominative and Loc-S/A clauses (Eastern Khanty non-canonical constructions), is potentially interpretable as a ‘diachronic relationship’ between passive and ergative organisations. The dynamics of assignment of subjecthood properties to arguments may



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.32 (2052-2168)

Andrey Filtchenko

be perceived as supporting the hypothesis of the process of the ‘gradual’ evolution of nominative languages to ergative ones via the switch of subject properties from the morphologically simple P of the passive to the morphologically complex A (Comrie 1978). It is thus possible to posit that Eastern Khanty may obtain at a stage of the evolution of a language system, where the predominant nominative features of the organisation of the grammatical relations are concurrent with the use, restricted to a particular functional domain, of the available grammatical resources in the way that is characteristic of the ergative organisation. The directionality of this dynamic combination of features may be determined in further historical-reconstruction investigations. In course of the discussion, we pursued the question of what are the types of discourse-pragmatic functions and kinds of propositional-semantic content associated with the non-canonical (Loc-marked Agent) construction types in Eastern Khanty. Based on the analysis of Eastern Khanty narratives, we sought to support the hypothesis that the choice of the non-canonical construction is motivated by pragmatic pressures/intentions in the discourse. The need to identify, maintain and change the pragmatic properties and functions, and interrelation of the referents in the discourse utilizes the structural means (the grammatical resources) available in the system of Eastern Khanty, finding formal expression in the variety of available construction types. The monostratal analysis of the host of pragmatic-functional, semantic and structural properties of all the participants in their interaction and the proposition as a whole is found as particularly revealing in approaching the issue of non-canonical constructions. The comparison of the main relevant pragmatic, semantic and structural features of the core constituents, the verbal predicate, as well as the whole event expressed by the non-canonical constructions is represented in Table 4. It follows from the analysis of a host of interrelated features of the noncanonical constructions that specific sets of the system’s grammatical resources identify with certain pragmatic-functional properties. More specifically: 1. In the Loc-S/A clauses, the overt Loc-marked S relation of the human/animate Agent, low transitivity of the morphologically active verbal predicate, SV agreement, parenthetical character (1 clause-length followed by canonical active-direct clauses with the continuing topic expressed by elision) – correlates to temporary pragmatic prominence (topicality) of the low control/volition Agent in the consequential dependence of the event, where the agentive (causer) nature of the agent participant is de-emphasised, consistent with specific cultural conventions and practices (taboo, cosmology); 2. In the agented passive clauses, the overt Loc-marked human/animate Agent demoted to the non-S relation, high semantic transitivity of the morphologically passive verb, less animate (inanimate) Ø-marked Target promoted to

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.33 (2168-2201)


the S relation, S-V agreement, parenthetical character (1–2 clause-length followed by a canonical active-direct clause with the continuing pragmatic center expressed by elision) – correlate to temporary topicality of the Target in the spontaneous/consequential/dependent event, where the topicality and causer nature of the Agent is de-emphasised, but whose identity is relevant; 3. In the agentless passive clauses, no overt agent/causer, high semantic transitivity of the morphologically passive verb, the S relation of the less animate/inanimate Ø-marked Target, S-V agreement, parenthetical character (1 clause-length followed by canonical active-direct clause(s) with the continuing topic expressed by elision) – correlate to temporary topicality of the Target in the event, where the agent/causer is either irrelevant (agentless), rendering the event less spontaneous or unknown (impersonal), rendering the event in a somewhat automatic sense. Eastern Khanty non-canonical constructions – agented passive and Loc-S/A clauses, are similar insofar as they manifest a parenthetical establishment of an alternative, a secondary topic, i.e. a discourse-prominent referent, whose activation (topicality) status is briefly competing with that of the primary discourse-topical referent, which is expressed by the temporary promotion (foregrounding) of such Table 4. Non-canonical clause features Loc-S/A (“Ergative”)

Agented Passive Agentless

Agent Known Human/animate Definite Loc-marked S/A Foregrounded Target None/Definite None/inanimate Nom-marked non-S – Affectedness (–) / ? Predic. Intrans./less trans. Imperf./Perfect. Event Parenthetical/disrupting but not interrupting (+) Consequential (+) Spontaneous (–) Control/volition Less agentive Highlights/foregr.Agent

Known Human/animate Definite/Indefinite Loc-marked non-S Backgrounded Definite/Indefinite Less animate Nom-marked S Foregrounded Affectedness (+) / ? Transitive Perfective/Imperfective Parenthetical/disrupting but not interrupting (+) Consequential (+) Spontaneous (–) Control/volition Less agentive Highlights/foregr.Target

Agentless Passive Impersonal

Known Human/animate Definite/Indefinite – non-S Backgrounded Definite/Indefinite Less animate Nom-marked S Foregrounded Affectedness (+) / ?

Parenthetical/disrupting but not interrupting (–) Consequential (–) Spontaneous (+) Control/volition – Highlights/foregr.Target

Unknown – – – non-S –



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.34 (2201-2258)

Andrey Filtchenko

a referent to the S grammatical relation. This secondary discourse-prominent referent is expressed by a full NP or a free pronoun, which, as shown at the onset of this paper, is not a preferred primary topic expression in Eastern Khanty.10 The second feature of the Eastern Khanty non-canonical constructions is that the agent referent is marked with an oblique, Loc case. Such Loc-Agent non-canonical constructions (Loc-S/A clauses and agented passive clauses) differ in the semantic properties of the referents promoted to the S relation. In the agented passive clauses this referent is always lower in the animacy, agentivity hierarchy and has semantic properties of the Target, whereas in the case of the Loc-S/A clauses, the agentivity status of the referent is normally very high. Identifying more broadly, within the Eastern Khanty system, non-canonical agent marking constructions vs. canonical Nom/Abs-marking, a general consistency of the Loc marking of the Agent with the particular pragmatic and semantic environments can be observed. This identification of the non-canonical constructions based on pragmatic functional parameters is supported by the indications of what appears to be a complementary distribution of the Loc-S/A clauses and passive constructions in the narrative discourse. That is, these constructions have compatible type frequency in the narratives (10% and 12% respectively), however, they appear to show counter-proportional or exclusive frequency in the same narratives, as evident in the discussed corpus and from prior studies (Kulonen 1989). It is thus this study’s prediction that the instances of the described noncanonical constructions are going to be associated with the above common pragmatic-functional environments: decreased activeness/control/volition and unspecified completion/affectedness; but individually restricted by the semantic properties of the proposition: role of the participants and underlying transitivity of the event.

Abbreviations Acc Attenuat Car Com Dem DET Du

Accusative case Attenuative Carritive Comitative Demonstrative Determiner Dual

. The preferred topic expression is by elision and S/A-V agreement inflection (3. Information Structure).

JB[v.20020404] Prn:6/11/2006; 9:17

F: LA9603.tex / p.35 (2258-2319)


IMPP IMPRF Instr GER Lat Loc Nom Pl/3sg Mod. PP Pred PRST PST POSS.1sg PS SAP TR

imperfective participle Imperfect Instrumental case gerundial non finite verb form Lative case Locative case Nominative case agreement in num. of the O=(Pl) and pers/num. of the S/A=(3sg) modifier (nominal, verbal) perfective participle Predicator Present Past 1sg Possessor Passive speech act participant Transitivizer (verbal affix)

References Aikhenvald, Alexandra Y., Robert M. W. Dixon, & Onishi Masayoshi (Eds.). (2001). NonCanonical Marking of Subjects and Objects. Amsterdam: John Benjamins. Balandin, A. N. (1967). Obsko-ugorskie konstruktsii glagolnogo predlozhenia so “skrytym subjektom” (Ob-Ugric constructions in verbal clauses with “covert subject”). In V. M. Zirmunskij & S. D. Kacnel’son (Eds.), Ergativnaja konstruktsija predlozhenija v jazykakh razlichnikh tipov (Ergative clause type in languages of various typology) (pp. 298–304). Moscow. Cheremisina, Maya I., & Natalia B. Koshkareva (1991). Slozhnoe i oslozhnennoe predlozhenie v khantyjskom jazyke (Complex and extended clause in the Khanty language). Metodicheskie ukazanija i laboratornyje raboty k kursu “Obshee Jazykoznanije”. Novosibirsk. Comrie, Bernard (1977). In defence of spontaneous demotion: The impersonal passive. In Peter Cole & Jerrold M. Sadock (Eds.), Grammatical Relations [Syntax and Semantics 8] (pp. 47– 58). New York, NY: Academic Press. Comrie, Bernard (1978). Ergativity. In Winfred P. Lehmann (Ed.), Syntactic Typology (pp. 329– 394). Austin, TX: University of Texas Press. Décsy, Gulya (1990). The Uralic Protolanguage: A Comprehensive Reconstruction. Bloomington, IN: Eurolingua. Dixon, Robert M. W. (1979). Ergativity. Language, 55(1), 59–138. Dixon, Robert M. W. (1994). Ergativity. Cambridge: Cambridge University Press. Givón, Talmy (2001). Syntax: An Introduction. Amsterdam: John Benjamins. Gulya, Janos (1970). Aktiv, Ergativ und Passiv im Vakh-Ostjakischen. Symposium über die Syntax der uralischen Sprachen, 80–83.



JB[v.20020404] Prn:6/11/2006; 9:17



F: LA9603.tex / p.36 (2319-2378)

Andrey Filtchenko

Gundel, Jeanette K. (1976). Topic-comment structure and the use of tože and takže. Slavic and East European Journal, 19, 174–176. Honti, László (1984). Chrestomathia Ostiacica. Budapest: Tankönyvkiadó. Kovgan, Elena V. (1991). Opredelitel’nye konstruktsii v zapadnyx dialektakh khantyjskogo jazyka [Attributive clauses in western dialects of Khanty]. “Kandidata Filologicheskikh Nauk” Thesis. Novosibirsk. Kulonen, Ulla-Maija (1989). The Passive in Ob-Ugrian. Helsinki: Finno-Ugrian Society. Kuno, Susumo (1972). Functional sentence perspective. Linguistic Inquiry, 3, 269–320. Lambrecht, Knud (1994). Information Structure and Sentence Form. Cambridge: Cambridge University Press. Li, Charles N., & Sandra A. Thompson (1976). Subject and topic: A new typology of languages. In Charles N. Li (Ed.), Subject and Topic (pp. 457–90). New York, NY: Academic Press. Nikolaeva, Irina (1999). Ostyak. München: Lincom. Onishi, Masayoshi (2001). Introduction: Non-canonical subjects and objects: Parameters and properties. Aikhenvald et al. (2001), 1–51. Shibatani, Masayoshi (1985). Passives and related constructions. Language, 61(4), 821–848. Skribnik Elena K., & Elena V. Kovgan (1991). Sistema prichastnyx opredelitel’nyx konstruktsij v obsko-ugorskix jazykax (The system of attributive participial constructions in ObUgric languages). In E. I. Ubrjatova & M. I. Cheremisina (Eds.), Jazyki narodov Sibiri. Grammaticheskie issledovanija (The languages of Siberian peoples. Studies in grammar) (pp. 84–108). Novosibirsk: Nauka. Tereshkin, Nikolaj I. (1961). Ocerki dialektov khantijskogo jazika. Vakhovskij dialect (Overview of the Khanty dialects. Vakh dialect). Moscow-Leningrad: Izdatelstvo Akademii Nauk.

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.1 (46-109)

Agent back-grounding as a functional domain Reflexivization and passivization in Czech and Russian Mirjam Fried This study examines the relationship between the be-passive and the passive reflexive in Czech and Russian, challenging the common assumption that both forms are just two different expressions of a passive diathesis. The analysis is centered on the functional notion of agent back-grounding, understood broadly as making the agent referentially obscure, and leads to the conclusion that what may resemble passives on the basis of superficial formal features amounts to distinct constructions in the sense of Construction Grammar. Overall, the crucial distinction is one between highlighting the end-result of an action (be-passive in both languages and Russian reflexive) vs. an existential, event-reporting function (the Czech reflexive).

.

Introduction

. Agent demotion and passive The topic of passiveness has commanded linguists’ sustained interest for a long time. Yet, a consistent definition – and, by implication, a full understanding – of the phenomenon remains elusive, especially when considered in relation to the more general question of agent demotion. The vast literature covers a wide range of approaches (syntactic, semantic, morphological, functional) and offers a wealth of often conflicting and confusing terminology, which only compounds the challenge of understanding the nature of the various patterns that have been labeled passive. This is perhaps to be expected in broad cross-linguistic comparisons, but capturing the essence of passive-like patterns or agent demotion in their various formal and functional manifestations is not a trivial task even within a single language. Taking a usage-based approach, I will examine these two notions from a functional perspective. In order to avoid terminological confusion with the syntactically defined agent demotion, and anticipating the distinctions that will be

JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.2 (109-176)

Mirjam Fried

substantiated in the body of this paper, I prefer the label “agent back-grounding”. One of my aims is to elucidate (some of) the factors that determine what may count as passive, in contrast to other kinds of agent back-grounding, in a given language. Slavic languages are known for their rich inventories of agent back-grounding constructions, including those that have been traditionally treated simply as types of passive: the periphrastic be-passive and the so-called passive reflexive, exemplified by the Russian [R] sentences in (1a) and (1b), respectively. The patient is in the nominative, agreeing in number with the verb, and the agent is in the instrumental; note the reflexive postfix -sja on the verb in (1b).1 (1) a.

Kem takoj dokument budet who.ins such.nom.sg.m document.nom.sg.m be.fut.3sg [R; Exo Moskvy 13/10/97] podpisan? sign.pf.pass.sg.m ‘By whom will such a document be signed?’ b. insulin. . . , kotoryj mnogimi ljud’mi ne insulin which.nom.sg.m many.ins.pl people.ins.pl neg [R; Sputnik 1986/5, 1.506] vosprinimajetsja take.ipf.pres.3sg.rf ‘insulin, which is not taken by a lot of people. . .’

Although selected aspects of these two patterns, including their so-called impersonal variants, have been under much discussion both in Slavic linguistics and in various typological studies (e.g., Jakobson 1957; Babby 1983; Siewierska 1984, 1988; Croft et al. 1987; Schenker 1988; Haspelmath 1990; Janda 1993; Kemmer 1993; Croft 2001), there is no general consensus about the exact formal and functional nature of the clause types in which each form occurs, whether in individual Slavic languages or in the family as a whole. The problem is rooted in the way the issue is traditionally framed. We are confronted with two morphologically distinct forms (passive participle vs. a reflexive form of an otherwise non-reflexive verb) that appear to have a comparable effect on the argument structure of transitive verbs and its expression in a sentence (patient promotion, agent demotion). At the same time, the very fact that they are morphologically wholly unrelated to each other suggests, at least hypothetically, that speakers could be using each form to encode a different communicative . Abbreviations: nom ‘nominative’, gen ‘genitive’, acc ‘accusative’, dat ‘dative’, ins ‘instrumental’, m/f/n ‘masculine/feminine/neuter’, sg/pl ‘singular/ plural’, rf ‘reflexive’, (i)pf ‘(im)perfective’, pres ‘present tense’, fut ‘future’, imp ‘imperative’, ppl ‘past participle’, pass ‘passive participle’, inf ‘infinitive’, neg ‘negative’, aux ‘auxiliary’, cond ‘conditional’, purp ‘purposive’, dp ‘discourse particle’.

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.3 (176-231)

Reflexivization and passivization in Czech and Russian

content. Once we allow ourselves to pose the question of their similarities and differences in this way, we may very well arrive at a different picture: one that moves beyond the issue of transitivity and syntactic patterning as the central defining characteristics and, instead, focuses on the functional domains associated with each form as the primary source of explanation for their overall behavior. It is the latter approach that frames the present study. I will assume a general functional domain, labeled agent back-grounding (in the pragmatically motivated sense of casting an agent referent as less prominent or distinct than the lexical meaning of the verb would suggest) and examine how different morphosyntactic patterns fit within that space. Specifically, I will study the relationship between the be-passive and the so-called passive reflexive in Czech and Russian. I will show that the Czech patterns differ from their Russian counterparts, which are generally better known and, with the exception of Siewierska’s (1984) study, uniformly (but mistakenly) taken to represent Slavic languages in general. The goal of the paper is thus two-fold. First, I present evidence that differentiating between individual agent back-grounding patterns, even within this restricted inventory, involves much more than syntactic promotion/demotion (Comrie 1977; Siewierska 1988; Haspelmath 1990), change in transitivity or argument structure (Babby 1983; Haiman 1983; Keenan 1985; Croft et al. 1987; Kemmer 1993; Givón 1994a), relative topicality of agents and patients and generally information-structural motivations (Shibatani 1988a; Givón 1994a; Hidalgo 1994), or subject affectedness (Siewierska 1988; Janda 1993). I will draw attention to a number of semantic and pragmatic constraints that help differentiate between the be-passive and the reflexive and situate each of them within the grammar of a given language. Second, I will argue that we cannot formulate any meaningful generalizations – language-specific or typological – without first establishing a clear understanding of the properties (morphosyntactic, semantic, pragmatic, discourse-based) that collectively shape the speakers’ native-like knowledge of each pattern. The analysis leads to the conclusion that what may resemble passive in superficial formal features (patient-subject, optional agent-oblique) amounts to distinct, albeit partially overlapping, communicative patterns, not all of which are truly passive. Rather, they occupy different parts of the agent back-grounding functional space and must be represented as such: as conventionalized associations between particular formal, semantic, and pragmatic features, i.e., as ‘constructions’ in the sense of Construction Grammar (esp. Fillmore 1989; Croft 2001; Fried & Östman 2004). The criteria that will emerge as relevant include agent animacy, indefiniteness vs. genericity of the back-grounded agent, inherent verb semantics, distribution of aspect, and the potential for semantic extensions. Overall, the crucial contrast is one between highlighting the end-result of an action vs. an existential, event-reporting function, both of which, however, presuppose a referentially



JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.4 (231-306)

Mirjam Fried

relatively obscure (syntactically demoted) agent. As a step toward developing a full constructional representation of the patterns in question, I will draw a constructional map in which their mutual relations can be arranged systematically, much the same way in which generalizations about grammatical categories are represented by the semantic maps used in cognitively oriented typological research (Croft et al. 1987; Kemmer 1993; Haspelmath 2003; Croft 2001). . Data My primary focus will be on the relevant clause types in Czech, using mostly data from the Czech National Corpus (a 100,000,000-word electronic corpus of both spoken and written contemporary Czech)2 and supplementing the discussion by Russian corpus examples in order to bring out the differences between the two languages. The introductory example in (2) below briefly illustrates the fact that the Slavic reflexive is a richly polysemous category; the Czech [Cz] example in (2b) shows some, thought not all, of the well-known functions of the reflexive clitic se, in contrast to the active transitive form of the same verb in (2a). (2) a.

Zavˇrel jsem branku. close.pf.ppl.sg.m aux.1sg gate.acc.sg.f ‘I closed a/the gate.’ b. Zavˇrela se. close.pf.ppl.sg.f rf (i) ‘She closed herself [in].’ (ii) ‘She/it got closed [in] / One closed her/it.’ (iii) ‘It closed.’

[Cz]

[Cz]

In the absence of any overtly expressed arguments, the reflexive can be ambiguous between several readings; here I will only be concerned with the interpretation in (2b-ii), which is the one directly related to the issue of passiveness (the full family of Czech syntactic reflexives is treated in Fried 2004, to appear). Corpus examples corresponding to (2b-ii) are in (3), showing four transitive verbs (uzavírat ‘close/finalize’, slavit ‘celebrate’, studovat ‘study/attend [a school]’, poznat ‘iden-

. Relying on corpus data as much as possible is consistent with the usage-based nature of constructional approaches to linguistic analysis, which are motivated by the generally recognized need to work with authentic utterances if we are serious about capturing linguistic phenomena in all their natural richness. The acceptability judgments and made-up examples are based on the author’s native knowledge of Czech and on the intuitions of several other native speakers.

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.5 (306-340)


tify’); the patient is in the nominative, agreeing in number (and, in the past tense, gender) with the verb, just like in the Russian example in (1b). (3) a.

celá ˇrada tˇech manželství se whole.nom.sg.f series.nom.sg.f those.gen.pl marriage.gen.pl.n rf uzavírala [. . . ] z nutnosti close.ipf.ppl.sg.f from necessity.gen.sg.f ‘a large number of these marriages were formed [. . .] out of necessity’ [Cz; PMK 185-#17520] b. na to se studujou vysoký školy3 for it rf study.ipf.pres.3pl university.nom.pl.f ‘people go to college to study it [Management & Planning]’ [Cz; PMK 199-#6602] c. (prosim vás pˇeknˇe) a jak se to pozná ? and how rf it.nom.sg identify.pf.pres.3sg ‘(please be so kind), how can one tell?’ [Cz; PMK 401-#77624]

The same syntactic pattern is associated with be-passives, shown in (4) with the transitive verbs vybavit ‘equip’ and obsadit ‘occupy’. (4) a.

kdyby zvˇeˇr nebyla pˇrírodou if game[animals].nom.sg.f neg.be.ppl.sg.f nature.ins.sg.f lépe vybavena než ˇclovˇek. . . better equip.pf.pass.sg.f than man.nom.sg.m ‘if animals weren’t equipped by [mother] nature better than man. . .’ [Cz; SYN 2-s4953] b. Šli kolem laviˇcky, která byla go.ppl.pl.m around bench.gen.sg.f which.nom.sg.f be.ppl.sg.f obsazena lázeˇnskými hosty occupy.pf.pass.sg.f spa.adj.ins.pl.m guest.ins.pl.m ‘They walked past a bench that was occupied by spa guests.’ [Cz; SYN 1-s3832]

Because of the morphosyntactic similarity, together with the fact that the reflexive form sometimes appears to report a transitive situation in a way that is roughly comparable to what the be-passive expresses, as in (3a), analyzing the reflexive morpheme as a passive voice marker has been the preferred option in most accounts (Kopeˇcný 1962; Parolková 1967; Králíková 1981; Babby 1983; Haiman . In the corpus examples, I preserve the PMK transcriptions, which often deviate from standard spelling, as a reflection of actual pronunciation. Thus we find shortening of certain vowels, as in (3c) and elsewhere, simplified segmental representation, as in spoken tˇrea instead of written tˇreba (8b), idiosyncratic adjectival morphology (3b), etc. None of these features affect the discussion of our constructions.



JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.6 (340-401)

Mirjam Fried

1983; Croft et al. 1987; Siewierska 1988; Kemmer 1993; Grepl & Karlík 1998). However, the Czech reflexive differs from the be-passive, and also from its Russian cognate, along a number of criteria that prove the traditional analysis overly reductionist and, hence, inadequate. The paper is organized as follows. Section 2 is a brief review of the features that characterize the Czech and Russian be-passive. Section 3 contains a detailed description of the reflexive in Czech, contrasting it with its Russian cognate. Section 4 brings both forms in both languages together in a feature-by-feature comparison and summarizes the crucial functional distinction between them in terms of different event structures: marking resultant state vs. reporting an event brought about by an anonymous agent. Section 5 relates the analysis to a broader typology of passives and suggests a representation of the contrast between the Czech and Russian systems, in the form of a network of partially overlapping grammatical patterns. Section 6 briefly concludes the paper.

. Be-passive Let us start by summarizing the main (all well-known) properties of the periphrastic passive. Based on the patient promotion and the verb agreement pattern described above, with the attendant reconfiguration of the argument structure, the be-passive is analyzed, uncontroversially, as a diathetic shift, in which the agent has the syntactic status of an optional adjunct, in Czech and Russian marked by the instrumental case, as shown in (4) for Czech and in (1a) for Russian. The quintessential pragmatic function of the passive is to draw attention to the endpoint/result of a transitive event (cf. also Givón 1979; Haspelmath 1990); transitivity is understood here in the sense of Hopper and Thompson’s (1980) semantically based prototype structure. The sentence in (4a), for example, is part of a tongue-in-cheek narrative about the differences between wild animals and their hunters; the center of attention is the effect mother nature leaves on its creation, so to speak. Similarly in (4b), where the use of a relative clause makes it explicit that the focus of interest is the bench and the state it was in as a result of some event. (This aboutness, however, is not to be understood in the topic-focus sense necessarily; I will return to this issue in Section 4.) The rest of the commonly acknowledged properties of this morphosyntactic pattern follow from its function, then: the agent is cast in a less prominent role (indicated by the instrumental case) and the be-passive shows strong preference for perfective verbs (highlighting the resultative flavor of the be-passive) and an even more pronounced preference for semantically strongly transitive verbs, since such verbs readily provide an (affected) endpoint (cf. Hopper & Thompson 1980: 252).

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.7 (401-462)


It is important to stress that this cannot be just a matter of syntactic transitivity since not all verbs that fit the nominative-accusative case marking pattern in their active form can be used in the be-passive. Verbs of perception and cognition, such as vidˇet ‘see’, slyšet ‘hear’, cítit ‘feel’, vˇedˇet ‘know’, etc., do not passivize, as shown in (5b), even though in an active sentence they mark their second argument by the accusative (5a). Notice further that the impossibility of (5b) cannot be explained by appealing to any aspectual preferences either: the presence of the perfectivizing prefix u-, which presumably should improve the verb’s compatibility with the passive on aspectual grounds, does not change anything about the severe ungrammaticality of (5b). (5) a.

Pak jsem (u)slyšel ránu. then aux.1sg (pf)hear.ppl.sg.m bang.acc.sg.f ‘Then I heard a bang.’ b. *Pak byla (u)slyšena rána. then be.ppl.sg.f (pf)hear.pass.sg.f bang.nom.sg.f ‘Then a bang was heard.’

[Cz]

[Cz]

The properties listed above hold equally for Czech and Russian. However, it is potentially significant that the be-passive is not exactly the same in both languages in other respects. One difference is the aspectual restriction, which is almost absolute in Russian (Janda 1993; Israeli 1997), while in Czech it is only a relatively strong tendency, allowing imperfective verbs to appear much more commonly than we find in Russian; an imperfective Czech example is in (6) and also in (17a), in contrast to the perfectives in (4). (6) je tím rušena zvˇeˇr be.pres.3sg it.ins.sg disturb.ipf.pass.sg.f game[animals].nom.sg.f ‘it disturbs the animals [lit. ‘by that the ANIMALSFocus are disturbed]’ [Cz; SYN 1-s2869]

Another difference has to do with the range of referents that the demoted agent can designate. In Russian, the agent is semantically unrestricted and freely includes even discourse participants (speaker or hearer). The Czech be-passive excludes the speaker or hearer as possible agents, although otherwise does not restrict the agent semantics either: both animate (4b) and inanimate referents (4a, 6) are permitted. The Russian usage is illustrated in (7a), where the agent is the speaker, as compared to its Czech translation in (7b), where the 1st pers. pronoun is not possible. (7) a.

Samyje rezkije slova byli mnoj skazany most harsh word.nom.pl.n be.ppl.pl 1sg.ins say.pf.pass.pl ‘The harshest words were said by me [. . .].’ [R; Ogonek 16/6/1997]



JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.8 (462-518)

Mirjam Fried

b. Jen ta nejprudší slova byla (*mnou) only those harshest word.nom.pl.n be.ppl.pl 1sg.ins použita. use.pf.pass.pl.n ‘Only the harshest words were used (*by me).’

[Cz]

Crucially, in the absence of any other NP, the only interpretation available in (7b) is ‘by someone’ (some unidentified third party), not ‘by me/us/you’, regardless of context. I will return to this point in Section 3.1, in a comparison to the reflexive form.

. Anonymous-agent reflexive Traditional accounts make a distinction between passive and impersonal reflexives. The latter contains no nominative NP with which the verb could agree; instead, the verb is in the 3rd pers. sg. neuter form. Czech examples of the impersonal form are in (8), showing the verbs pracovat ‘work’ (8a) and kˇriknout ‘yell’ and jít/šlo ‘go/went’ (8b). (8) a.

(V tý práci trávim asi jednu tˇretinu dne,) jako dyž se pracuje vosum hodin like when rf work.ipf.pres.3sg eight hour.gen.pl.f ‘(I spend about one third of the day at work,) if one, like, works eight hours’ [Cz; PMK 271-#94890] b. (ty dˇeti zlobˇej), tak se na nˇe kˇrikne, místo aby se so rf on 3pl.acc yell.pf.pres.3sg instead.of rf tˇrea k nim šlo maybe to 3pl.dat go.ppl.sg.n ‘(the kids get naughty), so they get yelled at, instead of, I dunno, [somebody] coming over to them’ [Cz; PMK 329-#52346]

Interestingly, the formation of corresponding impersonal reflexives based on oneplace predicates is not possible in Russian, as shown in (9).4 . Intransitive verbs can appear in Russian productively only in the so-called dispositional reading, such as in (i); for discussion of this special usage, cf. Janda (1993: 313) and Israeli (1997: Ch. 5); Fried (2004) for Czech. (i)

mne ne spitsja 1sg.dat neg sleep.pres.3sg.rf ‘I don’t feel like sleeping/I can’t sleep.’

[R]

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.9 (518-596)


(9) *zdes’ ne rabotajetsja here neg work.pres.3sg.rf ‘One doesn’t work here/there’s no working going on here’

[R]

I will argue, however, that the emphasis on the personal/impersonal sentence structure as a defining distinction draws the dividing line in the wrong place: it is based purely on the superficial difference in verb form (ultimately reflecting the difference in syntactic transitivity of the verb) and obscures the fact that in Czech both types of reflexive have something important in common that sets them apart as a functionally unified grammatical category, different from the be-passive.5 The rest of this section discusses some of the ways in which both Czech reflexives, personal and impersonal, are alike, and together differ from the be-passive (cf. also Fried 1990). I will refer to both reflexive forms collectively as “anonymous-agent reflexive” (AR), for reasons that will become clear in the subsequent section. The fact that the reflexive can be formed from both transitive and intransitive verbs with equal ease contrasts directly with the be-passive, as inherently intransitive verbs (and especially verbs of motion) do not passivize at all. This restriction concerns not only monovalent verbs, such as kˇriknout ‘yell’ or jít ‘go’ in (8b), but also all polyvalent verbs that are not of the nominative-accusative type. One example is in (10) below, showing the verb ˇríkat nˇeˇcemu X ‘call something.dat X’: (10a) illustrates its non-reflexive use, (10b) is a corpus example of a reflexive use, and (10c) is an unsuccessful attempt to use the be-passive. [Cz] (10) a. ˇcesky mu ˇríkáme sada. in.Czech 3sg.dat.m say.pres.1pl set.nom.sg.f ‘[a set in English], in Czech we call it sada’ [Cz] b. ˇcesky se mu ˇríká sada. in.Czech rf 3sg.dat.m say.pres.3sg set.nom.sg.f ‘. . . in Czech yougeneric call it sada’ [Cz; PMK 273-#99356] [Cz] c. *ˇcesky je mu ˇríkáno sada in.Czech be.pres.3sg 3sg.dat.m say.pass.sg.n set.nom.sg.f ‘. . . in Czech it’s called sada.’

Moreover, most of the personal reflexives cannot be naturally paraphrased by the be-passive. In the data presented here, only (3a) would sound reasonably acceptable in such a paraphrase. It must also be noted that this cannot be attributed simply to the potential clash in aspect: consider (3a), here repeated as (11a), which has an imperfective verb but could be paraphrased by the be-passive in (11b), compared to (3c), here repeated as (12a), which contains a perfective verb and yet, . As Geniušien˙e (1987) argues and amply documents, the similarity between the personal and impersonal reflexive is not a quirk of Czech but represents a cross-linguistically common pattern.



JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.10 (596-651)

Mirjam Fried

cannot be interpreted as reporting a resultant state of the kind be-passives encode; the failed attempt at a be-passive paraphrase of (12a) is in (12b). (11) a.

celá ˇrada tˇech manželství se whole.nom.sg.f series.nom.sg.f those.gen.pl marriage.gen.pl.n rf [Cz] uzavírala [. . . ] z nutnosti close.ipf.ppl.sg.f from necessity.gen.sg.f ‘a large number of these marriages were formed [. . .] out of necessity’ b. celá ˇrada tˇech manželství byla uzavírána z nutnosti be.ppl.sg.f close.ipf.pass.sg.f

(12) a.

(prosim vás pˇeknˇe) a jak se to pozná? and how rf it.nom.sg identify.pf.pres.3sg ‘(please be so kind) – how can one tell?’ b. *a jak je to poznáno? and how be.pres.3sg it.nom.sg recognize.pf.pass.sg.n ‘and how is it recognized/known?’

[Cz]

Nor is this an issue of forming the passive participle per se; the verb poznat can be used in the be-passive, as shown in (13): (13) byl mezi d˚ustojníky pˇrítomnými 24. dubna be.ppl.sg.m among officer.ins.pl.m present.ins.pl.m 24th.April poznán i nadporuˇcík identify.pf.pass.sg.m also first.lieutenant.nom.sg.m (tajné policie, který se pˇredstavil jako Dan. . .) ‘even a secret service lieutenantFocus was identified [as being] among the officers present on April 24th; (he introduced himself as Dan. . .)’ [Cz; SYN – Respekt9011]

The problem is the overall meaning expressed in (12a): it is active, just like in all the other reflexive examples (cf. also Grepl & Karlík 1998: 135 on the active reading of AR). Notice that to render the meaning of these sentences in English, it is often necessary to use the generic pronouns one or you, rather than the periphrastic passive. In fact, the communicative function of AR is best understood as drawing attention to the event itself – the process of forming a marriage (11a), attending college (3b), identifying something (12a), working for eight hours (8a), yelling out to the kids (8b), etc. – rather than highlighting a participant to which something happens (marriages being formed, college being attended, something being identified, kids being yelled at), which draws attention to the endpoints of events/processes and is the functional domain of the be-passive. It is not surprising, then, that transitivity may not be relevant in AR, since attention to the result or endpoint is not what the reflexive is primarily concerned with.

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.11 (651-722)


This event-reporting characteristic of AR is further confirmed by the conditions that constrain the encoding of the arguments, both agents and patients (if present). . Constraints on the agent The most readily observable difference consists in the fact that the Czech reflexive does not allow the agent to be expressed at all, in contrast both to the Czech and Russian be-passive (1a, 4, 6, 7a), and to the Russian reflexive as well. One example of the latter is in (1b), which contains the instrumental-marked agent mnogimi ljud’mi ‘by many people’, and another in (14), with the phrase vsemi graždanami ‘by all citizens’. (14) (sejˇcas, kogda Konstitucija prinjata,) ona dolžna ispolnjat’sja 3sg.f.nom oblig.sg.f uphold.inf.rf vsemi graždanami all.ins.pl citizen.ins.pl.m ‘(now that the Constitution has been accepted,) it must be upheld by all citizens’ [R; Izvestija 15/10/1988]

In addition, the identity of the agent in AR is restricted to human beings, which is not the case in the be-passive, as evidenced by (4a) or (6). In contrast to both the Czech be-passive and the Russian reflexive (cf. Israeli 1997: 187), the Czech AR allows the agent to be interpreted as a discourse participant, i.e. the speaker or the hearer. Recall that explicit reference to these discourse roles in the Czech be-passive is impossible, as was shown in (7b) by the prohibition on the use of the 1st or 2nd person pronouns in the instrumental-marked agent phrase; we also noted that if an agent is left out altogether, the passive can never be used to imply the speaker or the hearer as being the understood agent. AR, on the other hand, is very commonly used in contexts in which the speaker has reasons for keeping a communicative distance from the interlocutors’ involvement in the reported event, and for highlighting, instead, the event itself. Consider (15) below, which is taken from a conversation about a meeting to which the speaker and his fellow students do not want to go and are discussing who else they could persuade to skip it. (15) (ta Hrachová, tý by staˇcilo ˇríct nechod’ tam, jako–) ˇreklo by se jí vo co de, say.ppl.sg.n cond rf 3sg.dat.f about what go.pres.3sg (v tu chvíli by byla s náma) ‘(that Hrachová [woman], all one would need to do would be tell her, like, “don’t go there” –) one would explain to her what the deal is (and she’d be on our side immediately)’ [Cz; PMK 189-#526526]



JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.12 (722-774)

Mirjam Fried

By using the AR, the speaker in (15) distances himself from taking a direct stance or from assigning any explicit responsibility for the event of telling, although it is clear that the agent would be the speaker and his fellow students (‘we’d tell her and she’d understand and join us’). Similarly in (12a), which is part of a phone conversation in which the speaker is asking for hints about how to identify the tour guide assigned to the particular group in which the speaker belongs; the referent of the unexpressed agent in (12a) is clearly the speaker herself, along the lines of ‘how will I know/tell. . .?’. In this respect, AR also constitutes one of the syntactic strategies Czech speakers use for encoding indirectness, motivated by general social conventions for expressing politeness; a comparable discourse-functional extension is not associated with the be-passive. A typical example of a polite directive generalized to any potential audience is in (16a), which comes from an instructional text about the DOs and DON’Ts of seating arrangements for various occasions. (16b), using the same verb, illustrates an indirect order from an aunt to her nephew with whom she is displeased at the moment – here it is intended specifically for her interlocutor, as a less harsh alternative to the direct imperative that precedes it: (16) a.

(U stolu, natož u slavnostní tabule,) se nesedí libovolnˇe. rf neg.sit.ipf.pres.3sg randomly ‘Yougeneric can’t [let people] sit around (the [dinner] table, let alone a festive one,) in a random order.’ [Cz; SYN 11866-s2329267] b. Sed’ poˇrádnˇe ! Takhle se nesedí. sit.imp.2sg properly this.way rf neg.sit.ipf.pres.3sg ‘Sit properly! You can’t sit like this.’ [Cz; SYN 54-s91323]

Finally, the restriction to human agents also brings out the AR’s relationship to another agent-demoting non-passive construction known in both Czech and Russian: the so-called generic agent sentences (Panevová 1973) in which the verb is in the 3rd pers. plural and the agent, necessarily human, is obligatorily unexpressed and interpreted as ‘folks that can be expected to do such things’. An illustrative example of the similarity between this sentence type and AR is in (17). Both in the false start, with the AR form of the verb vyrábˇet ‘manufacture’, and in the follow-up with the verb montovat ‘put together’, which takes the form of the generic-agent clause type, the agent is obligatorily left unexpressed and refers to some generic ‘they’ – the same unidentified group of humans in both cases. (17) a tam se dˇrív vyrábˇelo – tam montovali and there rf earlier make.ipf.ppl.sg.n there put.together.ipf.ppl.pl.m nˇ áký auta some car.acc.pl.n

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.13 (774-851)


‘and in the old days, [that place was used for] manufacturing – making cars [is what went on] there’ [Cz; PMK 276-#107902]

It is worth noting that in Russian this sentence type is the most common way of back-grounding the speaker as the agent, thus functionally similar to the Czech AR (12a, 13); a Russian example is in (18): (18) Vam ved’ govorjat russkim jazykom. 2pl.dat dp speak.pres.3pl Russian.adj.ins.sg.m language.ins.sg.m ‘Why, I’m telling you in plain Russian.’ [lit. ‘theygeneric are telling. . .’] [R; Israeli (1997: 189, ex. 92)]

To summarize, the human agent of AR can be, depending on context, interpreted as generic, indefinite, or even as a specific discourse participant, but its referent can never be named directly. This is not a surprising feature, though: it is fully consistent with the generalizing function of AR, and it also motivates additional pragmatic functions the AR exhibits, such as its well-known shifts into various modal interpretations, especially possibility (12a) and obligation/prohibition (16). . Constraints on the patient As the following examples illustrate, there is also a telling restriction on the patient referent in AR with transitive verbs: the reflexive does not tolerate patients that are referred to by proper nouns (19b), while the be-passive shows no such constraint (19a). This suggests that AR not only suppresses the identity of the agent, but there is a threshold for the degree of individuation on the patient as well. Highly specific, highly individuated patients give too much prominence to the endpoint of the reported situation and thus clash with the event-focused semantics of the reflexive. (A similar observation has been made about comparable Spanish reflexives by Hidalgo 1994: 176.) ˇ (Ivanka je velmi ˇcilá . . .) Casto bývá napomínána Ivanka.nom.sg.f . . . often be.pres.3sg reprimand.ipf.pass.sg.f ‘(Little Ivanka is very active. . .) She is often reprimanded (for hyperactivity).’ [Cz; SYN 4285-s1561257] [Cz] b. *Ivanka se ˇcasto napomíná. Ivanka.nom.sg.f rf often reprimand.ipf.pres.3sg ‘Little Ivanka gets reprimanded often.’

(19) a.

Similarly, patients that refer to discourse participants are highly unusual in AR, again in contrast to the be-passive. The example in (20a) below is syntactically unproblematic and the verb is semantically appropriate, but the sentence is very odd pragmatically, as indicated by the symbol #. Note that a passive form, shown



JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.14 (851-909)

Mirjam Fried

in (20b), would be perfectly natural with a 2nd person patient (the same holds for 1st person as well). (20) a.

#Vyzýváte

se, abyste (dlužnou ˇcástku splatil. . . ). request.ipf.pres.2pl rf purp (. . . ) ‘It is [hereby] requested of you that (you pay your bill . . .).’ b. Byl jste vyzván, abyste (. . . ). be.ppl.sg.m aux.2pl request.pf.pass.sg.m purp (. . . ) ‘You’ve been asked to (pay your bill . . . ).’

[Cz]

[Cz]

This usage is limited to a particular type of institutional discourse, and serves essentially as a performative. As such, it constitutes a distinct formal and functional subtype of AR. This analysis of low saliency of the patient is also consistent with cases of a non-agreeing AR occasionally found in spoken Czech, where the patient is formally promoted but fails to trigger agreement on the verb, as in (21a). The standard, promotional form would require the finite verb to agree in gender with the nominative, as shown in (21b). (21) a.

možná by se tam dalo udˇelat nˇ áká maybe cond rf there give.ppl.sg.n make.inf some.nom.sg.f [Cz; PMK 194-#651263] díra hole.nom.sg.f ‘maybe one could make some sort of a hole there’ [Cz] b. možná by se tam dala udˇelat . . . díra maybe cond rf there give.ppl.sg.f make.inf . . . hole.nom.sg.f

In fact, we could take the form exemplified in (21a) as an iconic expression of the basic communicative function of the Czech AR, which is to report processes brought about by human agents, whose identity must remain anonymous. The non-agreeing pattern highlights the event by de-emphasizing both its agent and its target (when such a participant is required by the meaning of the verb): it leaves the patient argument only loosely associated with the rest of the sentence, indicating possibly an afterthought status of the patient, or a case of elaboration after the event as a whole has been named (further discussion of this pattern can be found in Fried, to appear). Granted, the non-agreeing pattern is stylistically marked and not accepted by all speakers (let alone by the prescriptive grammars), as one reviewer also noted. Its marked status, however, does not constitute evidence that it should, therefore, be excluded from analysis, nor simply dismissed as a random and inexplicable performance error. It is in the nature of any usage-based approach to language that it does not reject peripheral examples, and the point of the present analysis is that the marginal uses (such as (20a) or (21a)) not only can be easily motivated by

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.15 (909-964)


the general constraints on the standard pattern and interpreted in the spirit of AR, but they actually help further illuminate the general phenomenon as well. It is a built-in advantage of a constructional analysis that both standard and exceptional uses can be accounted for by appealing to a single analytic mechanism, without drawing any arbitrary line between what is presumed to be an inherently more significant domain of grammatical structure, and what should remain outside of grammarians’ interest.

. Resultant state vs. agent anonymity To summarize, the Czech AR, whether personal or impersonal, can be contrasted with the be-passive in terms of distinct communicative functions: the be-passive highlights the endpoint of a (necessarily) transitive event, while the reflexive pattern serves primarily to cast the agent as anonymous, independently of the status of the patient (consequently, the patient can be downplayed as well). Each function is associated with specific semantic and formal restrictions that need not be stipulated but follow directly from the pragmatic function. In the preceding sections, I described this difference informally in terms of aboutness (be-passive being about the patient, AR about events) but we can now make this notion more precise. There is, of course, a correlation between patient topicality/agent nontopicality and the use of the be-passive, as has been repeatedly noted in various analyses of passives cross-linguistically (Shibatani 1988a; Givón 1994a; Hidalgo 1994), but these two phenomena are inherently independent of each other, as Slavic data easily demonstrate. For example, the clause-final patient in (6) and (13) is actually the discourse focus, which means that these passive sentences still follow the most neutral word order pattern in Czech (Theme-first, Rheme-last). Examples of this kind are not difficult to find, either in Czech (22) or Russian, shown in (23) and also in the information question in (1a). (22) proˇc bylo ministerstvem schváleno [. . . ] osm why be.ppl.sg.n ministry.ins.sg.m approve.pf.pass.sg.n eight.nom.sg.n nových pojišt’oven. . . ? new.gen.pl insurance.company.gen.pl.f ‘why did the ministry approve eight new insurance companies?’ [lit. ‘why were 8 new insurance companiesFocus approved by the ministry?’] [Cz; SYN hnh5-s4567135] (23) (Režiser . . . iskal aktera na rol’ Sergeja Esenina . . . . Etim akterom, blagodarja steˇceniju sˇcaslivych obstojatel’stv), byl izbran ja. be.ppl.sg.m select.pf.pass.sg.m 1sg.nom



JB[v.20020404] Prn:6/11/2006; 9:24



F: LA9604.tex / p.16 (964-1051)

Mirjam Fried

‘(The director was looking for an actor to play Sergej Jesenin. . . To be that actor), he selected (thanks to a lucky coincidence) ME.’ [R; AIFnol 1-2jan98 iskusstva]

These uses suggest that the passive is at least as much a marker of shifted event structure (highlighting the endpoint of a reported event) as a marker of shifted discourse structure (giving the endpoint a particular status in the flow of information). The relative order in AR is equally flexible: the nominative patient (if present) can be found sentence-initially as a topical element, as in (11a), (12a), and sentence-finally as part of the focus (3b), (21). The be-passive and AR thus have to be contrasted in terms of the event structure each encodes, not (just) in terms of information structure. While these two dimensions interact with each other and may often converge, they constitute two independent domains and must be considered separately in teasing out subtle communicative distinctions encoded by specific grammatical patterns. The reflexive holds a distinctly different status in each language, both relative to each other and relative to the be-passive, along a number of criteria: restrictions on agents and patients, preferences with respect to aspect, verb class and degree of transitivity, and the potential for semantic and pragmatic extensions of the patterns each form occurs in. The differences and overlaps are summarized in Table 1. This feature-by-feature comparison confirms the standard analyses of Russian: the Russian reflexive is more truly passive, the difference between the two constructions being simply aspectual. The Czech AR, on the other hand, is not a type of passive, whether formally, semantically, or pragmatically; while the bepassive serves to profile the patient (profiling in the sense of Langacker 1985, 1993) and focus attention on the result of the event denoted by the verb (hence also its Table 1. be-passive vs. reflexive in Russian and Czech be-passive Russian Czech Agent

Patient Aspect Verb class

Extensions

AdjunctINS Speaker/Hearer + human Proper N High saliency VTRANS. VINTRANS. V of motion Modal Politeness

– perf. – – – –

– – perf. favored – – – –

Russian – – impf. – – – –

reflexive Czech – – – impf. favored

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.17 (1051-1112)


unavailability for intransitive predicates), the AR, personal or impersonal, functions as a signal of unexpected (non-explicit, not highly individuated) referential status of the agent and is, therefore, better understood as being about the event itself, rather than about its participants. By not forcing the reflexive pattern into a predefined expectation about its passiveness due to some of its formal features, we can not only explain why we do not find the non-agreeing patterns of the kind in (19) in the be-passive, but also account for the fact that the AR is by far the most common strategy for expressing generalizations, including various modal and performative interpretations, while the be-passive never serves such functions. This contrast could also be articulated in terms of the morphosyntactically defined typological contrast between promotional vs. demotional passive (Comrie’s 1977): the Czech and Russian be-passive can be classified as promotional both formally and functionally (potentially greater communicative prominence of patient, optional presence of oblique-marked agent, no semantic restrictions on the agent or patient, restriction to syntactically transitive verbs, resultative interpretation, preference for perfective aspect), whereas the Czech AR (but not its Russian cognate) presents a more complicated picture: formally it can be promotional (to the extent that a suitable argument is present in the first place), but functionally falls squarely in the demotional category, without placing any particular expectations on the (potential) patient as coming into greater focus. The difference in function might also help explain the relative frequency of the reflexive (very high) vs. be-passive (practically non-existent) in the spoken language: in Czech, discourse-motivated promotion of the patient (or, for that matter, of any other constituent) is most naturally accommodated by changes in word order, but diminishing the role of the agent in an event seems to require explicit morphosyntactic marking.6 Put differently, the verb meaning (for syntax mediated by a valence) suggests the presence of an agent (exclusively human at that, hence highly expected to be coded prominently), but no agent is covertly acknowledged by the speaker. There are good reasons, cognitive and communicative, for having a special morphosyntactic pattern to indicate such an unpredictable, unexpected state of affairs. In Czech, the AR serves this special function. Russian, as we shall see in Section 5, employs a different construction for this purpose.

. It is worth pointing out that while Czech is known for its diglossia, the spoken and written varieties of Czech do not actually display any significant differences with respect to syntactic reflexives (including AR), which are all highly productive and naturally occurring in both. Contrary to Siewierska’s (1984: 184) claim that the “reflexive passive in Czech [. . .] is primarily associated with the colloquial language”, it is, in fact, the be-passive that is stylistically highly marked and limited to certain written genres (also Grepl & Karlík 1998: 135).



JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.18 (1112-1150)

 Mirjam Fried

. Networks of typologically related patterns A close study of the Czech facts helps further clarify the relationship between passiveness and the reflexive from a more general typological perspective, contradicting Haspelmath’s (1990: 59f.) proposal of “inactivization” (in the sense of “suffering”) as the original function of ARs. Grammaticization of the reflexive pronoun into a marker of passive (through other stages, such as anticausative) is, of course, a common evolutionary path. However, the inactivization analysis that is supposed to explain that path seems to gloss over the fact that the function of the Slavic reflexive pronoun is not to cast an active verb in an inactive (passive) situation, but to mark the lack of referential distinctness between the agent and the patient (Timberlake 1980; similarly Langacker & Munro 1975), signaling that the non-explicit identity of the agent goes against the expectation raised by the lexical meaning and the valence of the verb, which presupposes two distinct participants, as exemplified in (24): the verb vidˇet ‘see’ requires two entities, the perceiver and the perceptum, but the reflexive pronoun marks them as having a single real-world referent. (24) Vidím se v zrcadle. see.pres.1sg self.acc in mirror.loc.sg.n ‘I see myself in the mirror.’

[Cz]

AR can thus be understood as a particular extension of this semantic/pronominal reflexive in that it also marks an unexpected referential status of the agent (anonymity as an extreme manifestation of low individuation). Such an interpretation of AR has several explanatory advantages over the inactivization analysis for understanding the variety of agent back-grounding functions associated with reflexives. First, the existential, event-centered analysis argued for in this study motivates the preference for imperfective aspect, which goes hand in hand with the function of defocusing the endpoints of events; it is also consistent with the fact that AR is by far the most common form of expressing generalizations in Czech. Second, this kind of reflexive is not actually inactive (whether we understand this as passive or simply stative), but merely keeps the agent obligatorily anonymous. It is thus no coincidence that a natural paraphrase for most instances of the Czech AR are the generic-agent sentences, which keep active morphology, as discussed in Section 3.1. And third, we have a plausible explanation for the otherwise puzzling “deobjective” uses found in some languages (Haspelmath 1990: 55), in which the reflexive form is associated with de-emphasizing the patient to the point of leaving it unspecified. Haspelmath has no explanation for this relationship. However, if we acknowledge that the reflexive serves to focus on the event rather than on its participants (including the patient), then the reflexive clearly is not the same as

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.19 (1150-1220)

Reflexivization and passivization in Czech and Russian 

passive and a deobjective function need not be incongruous at all; Czech actually seems to suggest a stage that would lead toward such a development. Both the restrictions on the extreme specificity of the patient referent (19)– (20) and the non-agreeing form (21) can be seen as an intermediate stage in deobjectification. But the full expression of that function is also commonly attested in far more mundane cases: in transitive verbs whose patient argument is left unexpressed, resulting in an impersonal variant of AR (e.g. jedlo se ‘eating was going on’ vs. jedla se veˇceˇreNOM ‘(every)one was eating dinner’; or bude se uklízet ‘cleaning will be going on/one will be cleaning’). Particularly illustrative examples are in (25)–(26). (25) shows the transitive verb týrat ‘abuse’ both with its object in the non-reflexive form (týraj své vlastní dˇeti ‘[they] abuse their own children’, underlined) and without it in the AR form. (25) (lidé, že jo, kteˇrí teda týraj své vlastní dˇeti sou samozˇrejmˇe velice hluboce neštastnými rodiˇci –) [Cz; PMK 400-#75817] a bohužel se týrá, and unfortunately rf abuse.pres.3sg ‘(people, y’know, who abuse their own children are of course deeply unhappy as parents –) and unfortunately abuse does exist/goes on’

In (26), we have the verb závidˇet ‘envy’ in an excerpt from comparing the relative degree of personally directed gossip in a small village and in a big city (Prague), in which the problem is assessed by the speaker as less personal (people still engage in envy, but do not dwell on the specifics as much because they do not know enough about each other): (26) (. . . urˇcitˇe . . . tˇreba se tam víc pomlouvají, že jo . . . ) tak v Praze se taky hodnˇe závidí, že jo so in Prague.loc rf also a.lot envy.pres.3sg dp (ale tim že ty lidi na tý vesnici se hodnˇe znají. . .) ‘(I’m sure that for example they gossip about each other there [=in a village] more, right? –) I mean, there’s lots of envy in Prague too, right? (but because people in a small village know each other much better, [they say bad things about each other. . . ])’ [Cz; PMK 148-#551206]

Uses of this sort are very easy to find in Czech and, to my mind, they only strenghthen the argument in favor of not separating AR into passive and impersonal reflexives as if they were two contrasting patterns based on the inherent transitivity of the verb. Whether we end up with a personal or impersonal reflexive form in Czech is a fairly arbitrary issue, depending not only on the transitivity of the verb but also on the speaker’s decision whether to express a transitive patient overtly, or leave it only implicit in a given utterance. The examples in (25) and (26) also demonstrate that the deobjective usage is not limited to verbs that may leave

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.20 (1220-1290)

 Mirjam Fried

out the patient when its referent is implied under various conventionalized interpretations (e.g., general with jíst ‘eat’, ˇcíst ‘read’, or fixed in particular meanings, as in the case of pít ‘drink [alcohol]’). The verbs týrat or závidˇet do not generally allow omission of the patient, but the patient’s absence in an AR use is not semantically odd; it just results in a distinctly existential flavor (notice also the English translations). Finally, comparing Czech with Russian, we must conclude that the passive character of the Russian reflexive must be seen as a secondary development, whereby the generalization function of AR, inherited from Proto-Slavic, has been marginalized in favor of incorporating the reflexive form into the grammatical voice opposition (active-passive) along the aspectual distinction: the be-passive form for perfective stems, the reflexive for imperfective stems. It further follows from this difference that the Russian reflexive overwhelmingly favors transitive verbs and is fairly rigid about excluding perfective verbs, while its Czech counterpart has completely obliterated the transitive/intransitive distinction and is more permissive with respect to the distribution of aspect. We can summarize the differences by representing them within a network of grammatical patterns that are organized around shared features within a particular functional space. As we have seen, an adequate description of the patterns must incorporate their formal, semantic, and pragmatic features as they co-occur in specific conventional combinations. In order to capture this unity of form and function, I elaborate the notion of a semantic map (used by cognitively oriented typologists) into a constructional map, treating the patterns as constructions in the sense of Construction Grammar, i.e. representations in which all the relevant dimensions are specified as equal contributors to a given pattern. Possible maps of the Czech and Russian constructions discussed in this study are in Diagrams 1 and 2, respectively. In order to keep the diagrams as uncluttered as possible, I abstract away from a number of details that a full constructional representation (and especially its fully formalized statement) would have to include. For our purposes, only certain features need to be listed, giving us a relatively high-level overview. I turn to Czech first. The functional space, labeled “agent back-grounding” at the top, provides two clusters of features that motivate the different constructions. The cluster on the left says that transitive events are favored, with affected patients which are marked by the nominative, and that the overall interpretation tends to be one reporting a (resultant) state rather than an action. The cluster on the right delimits a particular kind of back-grounded agents (only human referents, obligatorily null in syntax) and associates them with an active interpretation of the patterns in which they figure; note that this cluster need not specify any requirements regarding the verb’s transitivity. Both of these clusters overlap (indicated by the top two ovals) as parts of what it could mean that an agent is back-grounded. And each cluster then

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.21 (1290-1319)

Reflexivization and passivization in Czech and Russian 

Diagram 1. Constructional map of (some) Czech agent back-grounding patterns

overlaps with additional configurations of formal, semantic, and pragmatic features that give us the be-passive construction (on the left) and the AR construction (on the right). The rectangle to the right of AR indicates the overlap between AR and the non-reflexive generic-agent construction mentioned in Section 3.1; they both clearly share some functional features as well as the fact that they both focus on reporting events without putting much emphasis on their participants. Their “event-centered” character puts them both in contrast to the “result-centered” be-passive and various other constructions not discussed in this study (e.g., the decausative pattern Dveˇre se zavˇrely ‘The door closed’, cf. Fried 1990, 2004). Overall, the network in Diagram 1 captures clearly that the Czech AR is primarily not about emphasizing an affected argument but about obscuring a human agent; hence, AR is not really passive (semantically, pragmatically, and thus not necessarily in form either). In contrast, the main criterion for establishing the reflexive in relation to the be-passive in Russian is aspect, as summarized in Diagram 2. The Russian reflexive serves as the preferentially imperfective counterpart to be-passive (preferentially perfective), otherwise both types share passive properties, both semantic/pragmatic and structural. This shift away from the agentrelated cluster toward the patient-related one is indicated in Diagram 2 by the dashed line enclosing the reflexive, which also cannot be labeled AR the way it

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.22 (1319-1346)

 Mirjam Fried

Diagram 2. Constructional map of (some) Russian agent-back-grounding patterns

is in Diagram 1; the differences from the Czech patterns are indicated by a line through the relevant features. The Russian reflexive thus joins the inventory of result-centered constructions, leaving the event-centered domain to the genericagent construction at the far right. It is also worth noting that the latter is fully productive in Russian, while its productivity in Czech is highly restricted. These maps make connections to other maps, which is to say, to other functional spaces, and by elaborating further on these overlapping networks, we can incorporate additional relationships, both across various functional domains and across additional constructions. The space of this study does not permit me to provide any more detail, but I can at least enumerate several such connections that would lead to further elaboration on the present proposal and to additional generalizations. The agent back-grounding space must be invoked in accounting for a number of other constructions in both Czech and Russian, both reflexive and nonreflexive, personal and impersonal, such as the following (the list is not intended to be exhaustive): –

the decausative reflexive, already mentioned: the type Dveˇre se zavˇrely [Cz] ‘The door closed’;

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.23 (1346-1422)

Reflexivization and passivization in Czech and Russian 

–

–

–

the usage variously known as middle, dispositional, modal deagentive, or virtual (Stephens, this volume); in Slavic it is reflexive in form: the type Žilo se tam dobˇre [Cz] ‘Life was good there’, lit. ‘[itnon-ref. ] lived-rf well/easily there’, or the Russian example in Footnote 4; various generic-agent active sentence types (discussed in detail in Panevová 1973), which obligatorily leave out any subject NP and the verb takes on various personal forms: 3rd pers. plural (17)–(18); 2nd pers. sg.; 1st per. plural (10a). the non-reflexive construction known as “adversative impersonal” (Babby 1994) or more generally as “uncontrolled impersonals” (Davies 2005), exemplified by the following Russian sentence: (27) Ego zasypalo podarkami. 3sg.acc.m bombard.pf.ppl.sg.n present.ins.pl.m ‘He got bombarded with presents.’

[R]

At the same time, each of the constructions within this functional domain displays properties that belong to some other domain(s) as well and shares those features with other constructions that may not have anything to do with agent back-grounding.7 We noted, for example, that the AR has modal and performative extensions, which means that it shares features associated with a functional space that delimits modality and attitude, as well as features that define different speech acts (another set of functional domains). A particularly salient function that may overlap with agent back-grounding is that of indirect communication or distancing (with or without the attendant politeness status), which subsumes certain reflexives (AR) and to some degree the be-passive as well, but also includes forms that encode indirectness without manipulating the agent or the valence in general (various discourse particles, conditional syntax, etc.). In short, in the approach advocated in this study, grammar is seen as overlapping networks of grammatical patterns, where the sharing across constructions and across functional domains occurs along multiple dimensions, both formal and functional, and none of the dimensions is presupposed to be fully autonomous. Finally, the same functionally based analysis can be applied to other languages, as the contrasts identified in this study are not just a quirk of Slavic. Existing analyses of comparable phenomena in other languages generally do not offer a . By the same token, not all reflexive patterns involve agent back-grounding. For example, semantic reflexives (24) or lexical reflexives do not fall into the AR category. Construction Grammar does not assume any kind of universal alignment between a particular function and a particular form, nor does its architecture require any such assumption; just like it is not the case that agent back-grounding is exclusively the domain of, say, reflexives, the reverse is not the case either: reflexivization is not a device dedicated exclusively to back-grounding agents.

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.24 (1422-1462)

 Mirjam Fried

detailed comparison of functionally different types of passives (largely due to the assumption that everything is simply passive as long as it shows the same morphosyntax regarding the diathetic shift in expressing agent-patient relations), but it is clear from the description of individual constructions in various languages that they parallel the patterns described here in many ways, whether they involve the formal distinction between passive and reflexive, as in Spanish (Hidalgo 1994), German, Romance, and many other Indo-European languages (e.g., Siewierska 1984; Geniušien˙e 1987), or the distinction between the periphrastic, English-like passive and some other form used for agent back-grounding (such as the infinitival -al- construction in Kannada, cf. Sridhar 1979), or altogether different formal expressions in other non-Indo-European languages (cf. Keenan 1985; Haspelmath 1990; Blevins 2003; Kaiser & Vihman, this volume), in which the form does not necessarily involve be-passive or reflexive, but the functional issues remain the same as long as there are multiple, competing forms to encode the general concept of agent back-grounding. The prediction of the model employed here is that different languages may show slightly different partitioning of the same functional space(s), each partition associated with a particular grammatical construction.

. Conclusions If by changes in voice we mean changes in the hierarchical arrangement of the same participants in a given event, then both the be-passive and the reflexive forms are, indeed, instances of diathetic changes, i.e. voice. The point of difference between them is the semantic and pragmatic function of each diathesis and the range of argument structures that can undergo such reconfigurations (i.e., it is not just a question of de-transitivization). They overlap in that both involve agent backgrounding and for transitive verbs also patient promotion (with telling restrictions in the Czech AR, though). In the rest of their properties, however, we clearly have to allow for a more discriminating analysis of each grammatical pattern. Specifically, AR highlights the event, rather than its participants, and this feature is motivated by the pragmatically grounded function of the semantic reflexive, which is a marker of unexpected (diminished) referential status of the agent and is inherently independent of the quintessential passive function of reassigning relative prominence between agents and patients vis-à-vis active voice. The analysis also shows that greater precision is required in what we mean by demoting or de-emphasizing or back-grounding the agent. On the one hand, it is necessary to distinguish between the concept of back-grounding (here understood as giving the agent referentially less prominent status than would be normally expected in reporting an agent-based event), and the encoding of that concept. The latter can range from complete absence (no overt expression) to

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.25 (1462-1530)

Reflexivization and passivization in Czech and Russian 

using less direct coding (e.g., a non-nominative case) and it is thus important to keep in mind that simple presence in a given sentence, without further consideration of its morphological shape, does not imply lack of back-grounding in the conceptual sense (manipulating the event structure). Furthermore, these two dimensions interact with information structure, in that the diminished referential prominence signaled formally may, but does not have to, be isomorphic with lower discourse prominence. Both Czech and Russian provide clear evidence that the issue of agent demotion is far more complex than a purely formal/syntactic definition can accommodate. On a more general level, the study makes a case for a constructional approach to linguistic analysis, which sees grammatical patterns in terms of conventionalized complex pairings between formal features, verb semantics, inherent semantics of its arguments, and a particular meaning or communicative function of the pattern as a whole. Such an approach provides useful tools both for representing individual linguistic facts and for drawing typologically interesting generalizations. I suggest capturing the relationships between grammatical patterns in the form of constructional maps that allow a systematic representation of both the overlaps and the differences, whether within a language (here, be-passive and AR in Czech) or across languages (here, Czech and Russian).

Sources of data ˇ ˇ Ceský národní korpus (PMK, SYN2000). Ústav Ceského národního korpusu FF UK, Praha 2000. . The Uppsala Corpus of Russian, .

References Babby, Leonard (1983). The relation between causative and voice: Russian vs. Turkish. Wiener Slawistischer Almanach, II, 61–88. Babby, Leonard (1994). A theta-theoretic analysis of adversity impersonal sentences in Russian. Formal Approaches to Slavic Languages, 3, 25–67. Blevins, James P. (2003). Passives and impersonals. Journal of Linguistics, 39, 473–520. Comrie, Bernard (1977). In defense of spontaneous demotion: The impersonal passive. In Peter Cole & Jerrold M. Sadock (Eds.), Grammatical Relations [Syntax & Semantics 8] (pp. 47– 58). New York, NY: Academic Press. Croft, William (2001). Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press.

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.26 (1530-1638)

 Mirjam Fried

Croft, William, Hava Bat-Zeev Shyldkrot, & Susanne Kemmer (1987). Diachronic semantic processes in the middle voice. In Anna Giacalone Ramat, Onofrio Carruba, & Guiliano Bernini (Eds.), Papers from the Seventh International Conference on Historical Linguistics (pp. 179–192). Amsterdam: John Benjamins. Davies, Laura (2005). Redefining the Russian ‘adversity’ impersonal: A construction-based account. In Steven Franks, Frank Y. Gladney, & Mila Tasseva-Kurktchieva (Eds.), Workshop on Formal Approaches to Slavic Linguistics: The Columbia Meeting 2004 (pp. 92–103). Ann Arbor, MI: Michigan Slavic Publications. Fillmore, Charles J. (1989). Grammatical construction theory and the familiar dichotomies. In Rainer Dietrich & Carl F. Graumann (Eds.), Language Processing in Social Context (pp. 17– 38). Amsterdam: North-Holland/Elsevier. Fried, Mirjam (1990). Reflexives as grammatical constructions: A case study in Czech. Berkeley Linguistic Society, 16, 127–139. Fried, Mirjam (2004). Czech reflexivization and the invariance principle revisited. Slavic and East European Journal, 84(4), 625–650. Fried, Mirjam (To appear). Constructing grammatical meaning: Isomorphism and polysemy in Czech reflexivization. Studies in Language. Fried, Mirjam, & Jan-Ola Östman (2004). Construction Grammar: A thumbnail sketch. In Mirjam Fried & Jan-Ola Östman (Eds.), Construction Grammar in a Cross-Language Perspective [Constructional Approaches to Language 2] (pp. 11–86). Amsterdam: John Benjamins. Geniušien˙e, Emma (1987). The Typology of Reflexives. Berlin: Mouton de Gruyter. Givón, Talmy (1979). On Understanding Grammar. New York, NY: Academic Press. Givón, Talmy (1994a). The pragmatics of de-transitive voice: Functional and typological aspects of inversion. In Givón (1994b), 3–44. Givón, Talmy (Ed.). (1994b). Voice and Inversion [Typological Studies in Language 28] Amsterdam: John Benjamins. Grepl, Miroslav, & Petr Karlík (1998). Skladba spisovné ˇceštiny (Czech syntax). Olomouc: Votobia. Haiman, John (1983). Iconic and economic motivation. Language, 59, 781–819. Haspelmath, Martin (1990). Grammaticization of passive morphology. Studies in Language, 14, 25–70. Haspelmath, Martin (2003). The geometry of grammatical meaning: Semantic maps and cross-linguistic comparison. In Michael Tomasello (Ed.), The New Psychology of Language (pp. 155–175). Mahwah, NJ: Lawrence Erlbaum Associates. Hidalgo, Raquel (1994). The pragmatics of de-transitive voice in Spanish: From passive to inverse? In Givón (1994b), 169–186. Hopper, Paul J., & Sandra A. Thompson (1980). Transitivity in grammar and discourse. Language, 56, 251–299. Hopper, Paul J., & Sandra A. Thompson (1984). The discourse basis for lexical categories in universal grammar. Language, 60, 703–752. Israeli, Alina (1997). Semantics and Pragmatics of the “Reflexive” Verbs in Russian. Munich: Verlag Otto Sagner. Jakobson, Roman (1971 [1957]). Shifters, Verbal Categories, and the Russian Verb. Cambridge, MA: Dept. of Slavic Languages and Literatures, Harvard University. Janda, Laura A. (1993). The semantics of Russian and Czech reflexives. In Robert A. Maguire & Alan Timberlake (Eds.), American Contributions to the Eleventh International Congress of Slavists, Bratislava (pp. 310–319). Collumbus, OH: Slavica.

JB[v.20020404] Prn:6/11/2006; 9:24

F: LA9604.tex / p.27 (1638-1718)

Reflexivization and passivization in Czech and Russian 

Keenan, Edward (1985). Passive in the world’s languages. In Timothy Shopen (Ed.), Language Typology and Syntactic Description, Vol. I (pp. 243–281). Cambridge: Cambridge University Press. Kemmer, Susanne (1993). The Middle Voice. Amsterdam: John Benjamins. Kopeˇcný, František (1962). Základy ˇceské skladby (The foundations of Czech syntax). Praha: SPN. Králíková, Kvˇeta (1981). Reflexívnost sloves z hlediska automatické analýzy cˇ eštiny (Reflexivity and machine-based analysis of Czech). Slovo a slovesnost, XLII, 291–298. Langacker, Ronald (1985). Observations and speculations on subjectivity. In John Haiman (Ed.), Iconicity in Syntax [Typological Studies in Language 6] (pp. 109–150). Amsterdam: John Benjamins. Langacker, Ronald (1993). Universals of construal. Berkeley Linguistic Society, 19, 447–463. Langacker, Ronald W., & Pamela Munro (1975). Passives and their meaning. Language, 51, 789– 830. Panevová, Jarmila (1973). Vˇety se všeobecným konatelem (Generic-agent sentences). Studia Slavica Pragensia, 133–144. Parolková, Olga (1967). K problematice zvratných sloves a tzv. zvratného pasíva v souˇcasné spisovné ruštinˇe a cˇ eštinˇe (On the problem of reflexive verbs and the so-called reflexive passive in literary Russian and Czech). Slavia, 36, 33–46. Schenker, Alexander M. (1988). Slavic reflexive and Indo-European Middle: A typological study. In Alexander M. Schenker (Ed.), American Contributions to the Tenth International Congress of Slavists, Sofia: Linguistics (pp. 363–383). Columbus, OH: Slavica. Shibatani, Masayoshi (1988a). Voice in Philippine languages. In Shibatani (1988b), 85–142. Shibatani, Masayoshi (Ed). (1988b). Passive and Voice. Amsterdam: John Benjamins. Siewierska, Anna (1984). The Passive. A Comparative Linguistic Analysis. London: Croom Helm. Siewierska, Anna (1988). The passive in Slavic. In Shibatani (1988b), 243–289. Sridhar, S. N. (1979). New evidence for spontaneous demotion. International Journal of Dravidian Linguistics, 8(2), 312–322. Timberlake, Alan (1980). Reference conditions on Russian reflexivization. Language, 56, 777– 796.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.1 (46-147)

Invisible arguments* Effects of demotion in Estonian and Finnish Elsi Kaiser and Virve-Anneli Vihman This paper investigates the syntactic, semantic and discourse-level properties of the implicit argument in two constructions in Finnish and Estonian in which the agent is left unspecified: the impersonal construction and the zero person construction. In light of data from a number of tests, we conclude that the implicit argument is present on the semantic level in both constructions in both Finnish and Estonian, but fails to project a salient discourse entity. On the syntactic level, data show that the implicit argument is syntactically present in the zero person construction in both languages. However, the picture is more complex for the impersonal constructions. We hypothesize that the Estonian impersonal construction removes the highest argument on the discourse level but leaves it at least partially syntactically intact, whereas the Finnish impersonal construction seems to downgrade the agent on the level of syntax as well as on the discourse level.

.

Introduction

In this paper we investigate two constructions in Finnish and Estonian in which the agent is left unspecified, namely the impersonal construction and what has been called the zero person construction.1 Although these constructions resemble each * We would like to thank two anonymous reviewers for their comments, as well as the organisers and participants at the Demotion of the Agent workshop at the University of Oslo for their questions and comments. Many thanks also to the other members of the Finnic passives group, Diane Nelson, Satu Manninen, and Katrin Hiietam, and to the British Academy for funding the ‘Syntax and Semantics of the Finnic Passive’ project led by Diane and Satu. In addition, thanks to Bridget Copley and Christine Gunlogson for helpful comments and references. We are also grateful to the ESRC for its support (postdoctoral fellowship award PTA-026-27-0076 to V.-A. Vihman). . We are following the literature on Finnish here, which uses the term nollapersoona for this construction (e.g. Laitinen 1995; Löflund 1998; Hakulinen et al. 2004). The term ‘missing person’ has also been used in the Finnish literature (Hakulinen & Karttunen 1973). The construction has been less studied in Estonian, but the main academic grammar categorizes it as a

JB[v.20020404] Prn:19/10/2006; 13:57



F: LA9605.tex / p.2 (147-184)

Elsi Kaiser and Virve-Anneli Vihman

other in containing an unspecified agent, the existing literature has noted that they differ from each other semantically, syntactically and pragmatically. With the aims of (i) gaining a better understanding of the similarities and differences between the impersonal and zero person constructions and (ii) learning more about the (dis)similarities between two closely related languages, we investigate the syntactic, semantic and discourse-level status of the implicit argument in each of these constructions. We apply a number of tests that are commonly used in the generative linguistics tradition to probe for implicit arguments, with the secondary purpose of gauging whether these tests, primarily designed with Indo-European languages in mind, can be used for typologically different languages. Based on the results, we conclude that the implicit argument is present on the level of semantics in both languages in both constructions, but does not project a full-fledged discourse entity. The data suggest that the implicit argument is syntactically present in the zero person construction in both languages, but that its status in impersonal constructions is less clear. More specifically, given the data discussed in this paper, we hypothesize that the Estonian impersonal construction demotes the highest argument on the discourse level but leaves it at least partially syntactically intact, whereas the Finnish impersonal construction seems to demote the agent on the level of syntax as well as in the discourse. In this paper, we use the term ‘demotion’ to mean, roughly speaking, the removal or ‘downgrading’ of an argument from the representation of the sentence in a particular linguistic domain (e.g. syntax, semantics, discourse). We leave open for future research the intriguing question of whether demotion is always absolute or whether demotion can be partial, i.e., whether there can exist different degrees of demotion in certain domains (e.g. on the discourse level). Although further research on Finnic impersonal and zero person constructions is still needed, our findings raise interesting questions regarding the strength of the connections and correlations that hold between different linguistic domains. In particular, they suggest that dissociation between domains is possible, such that an argument can be demoted – that is, at least partially removed from the representation – in one domain while remaining intact in another. In future work we aim to investigate the feasibility and implications of this tentative ‘dissociation hypothesis’ in more detail by means of a formal theoretical analysis. The aims of the present paper are more modest: we focus on presenting data regarding the syntactic, semantic and discourse-level properties of the implicit arguments in zero person and impersonal constructions, with the goal of gaining a better empirical understanding of the nature of the implicit arguments in these two constructions in Finnish and Estonian. ‘defective clause’ and calls it üldisikuline ‘general person’ (Erelt et al. 1993: 227); it has also been referred to as a ‘generic apersonal’ construction (Vihman 2004: 90).

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.3 (184-257)

Effects of demotion in Estonian and Finnish

. Impersonals in Estonian and Finnish Finnish and Estonian impersonals use impersonal verbal morphology, distinct from their personal counterparts: the present tense and the simple past of the impersonal are expressed by synthetic verb forms (Finnish -taan/-tiin, Estonian -takse/-ti), while the perfect tenses are periphrastic. The impersonals of both languages have common historical origins (Laakso 2001; Laanest 1975), and in fact, in many respects Finnish and Estonian impersonals look virtually identical.2 In both languages, transitive and intransitive verbs can be impersonalized (as shown in example (1)), as can modals, auxiliaries and even unaccusatives (Löflund 1998; Torn 2002; Blevins 2003). The only general restrictions on which verbs can occur in an impersonal construction are that (i) the verb must have at least one (non-experiencer) argument in its underlying argument structure, and (ii) the impersonal referent must be compatible with a human interpretation. The Finnic impersonals have no direct English counterpart, hence translations in our examples vary between passives (limited in English to transitive verbs), the impersonal one, and general terms like ‘people’. (1) a.

Siellä nukutaan. (Finnish)3 there sleep.imp.prs4 ‘People are sleeping there.’

. We refer to the constructions in question as ‘impersonals’. In the present paper, we focus primarily on the status of the implicit argument, and do not directly investigate the syntactic formation of impersonal or zero person constructions, nor the promotion or non-promotion of the arguments. Thus, in the present paper, we remain agnostic on the issue of whether the structure we call ‘impersonal’ represents a form of impersonal passivization (Manninen & Nelson 2004; Hiietam & Manninen 2005) or true impersonalization (Blevins 2003). . In many dialects of Colloquial Finnish, the first person plural verb, formed with the ending -mme, has been replaced with the impersonal form, and is usually used with an overt preverbal pronoun (e.g. me nukutaan ‘we are sleeping’). These verbs are clearly not semantically impersonal or passive in any way. This extended usage of impersonal morphology is beyond the scope of this paper, since it is not a form of agent demotion. In the impersonal construction proper, no overt pronoun can be used to refer to the impersonal argument referent and no element other than the impersonal verbal inflection is used to indicate impersonalization. Note, also, that the 1pl development of the Finnish impersonal verb form does not appear to be quite the same as the colloquial French Nous on s’en va, where the 1pl emphatic pronoun seems to identify the scope of the impersonal argument referent, rather than replacing it with a personal referent. . The following abbreviations are used in glosses: imp-impersonal, prs-present, pst-past, infinfinitive, qtv-quotative, neg-negative, sup-supine, inst-instructive, nom-nominative, parpartitive, gen-genitive, acc-accusative, ill-illative, ine-inessive, ela-elative, all-allative, adeadessive, abl-ablative, ess-essive, com-comitative, ref-reflexive, rec-reciprocal, cl-clitic.



JB[v.20020404] Prn:19/10/2006; 13:57



F: LA9605.tex / p.4 (257-320)


b. Seal magatakse. (Estonian) there sleep.imp.prs ‘People are sleeping there.’ c. Pekka pidätettiin. (F.) Pekka.nom arrest.imp.pst d. Pekka võeti kinni. (E.) Pekka.nom take.imp.pst closed ‘Pekka was arrested.’

There is no agreement between the internal argument and the impersonal verb in either Finnish or Estonian. Compare examples (1c–d), with singular patient arguments, to (2a–b), where the patient is plural but the verbs (in Finnish, (2a), and Estonian, (2b)) are unchanged from (1). Kaikki varkaat pidätettiin.5 (F.) all.nom burglar.nom.pl arrest.imp.pst ‘All burglars were arrested.’ b. Kütusevargad võeti kinni.6 (E.) fuel-burglar.nom.pl take.imp.pst closed ‘The gas thieves were arrested.’

(2) a.

The internal argument usually surfaces with nominative or partitive case marking, depending on verbal semantics and aspect. With a telic impersonal verb and perfective clause, a totally affected internal argument takes nominative case (the counterpart of accusative case-marking on the object in the active voice), whereas an atelic verb, imperfective clause or partially affected argument results in partitive case-marking in both the active voice and in the impersonal construction (see e.g. Erelt et al. 1993: 51–53, Manninen & Nelson 2004).7 Despite (or because of) these case-marking patterns, there is no consensus in the literature regarding the status of the internal argument and the promotional or non-promotional nature of the construction (Blevins 2003; Manninen & Nelson 2004). The interpretation of the implicit argument involves a human (or at least an animate referent that is personified, see e.g. Vilkuna (1996), Löflund (1998: 45) on . From the discussion group sfnet.keskustelu.laki, 8.2.2001. . From <www.sloleht.ee/index.aspx?d=20.09.03&r=4&id=146223>. . Note that nominative case in itself is not necessarily a signal of subject status in Finnish and Estonian. The fully affected internal argument also receives nominative case-marking in imperatives and certain infinitival phrases (see e.g. Erelt et al. 1993: 53). On a related note, the reader should also be aware of the existence of a well-known exception in Finnish impersonals: direct object human personal pronouns marked with accusative case in the active voice are also accusative in the impersonal construction. Unlike full nouns, these pronouns do not surface with nominative case in the impersonal.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.5 (320-357)


Finnish, Erelt et al. (1993: 227) and Torn (2002: 95) on Estonian), often agentive,8 usually plural referent in both languages (Hakulinen & Karlsson 1988; Shore 1988; Vilkuna 1996; Nelson 1998; Rajandi 1999; Blevins 2003; Vihman 2004). The restriction regarding the human/highly personified interpretation is shown in (3). As Vilkuna (1996: 140) points out, example (3a) is felt by speakers to be amusing. She notes that it can suggest an image of someone peering into a pot of potatoes and speaking to the potatoes as if they were human. Example (3b) is given by Blevins (2003), who notes that “since non-animate verbs like aeguma ‘to expire, elapse’, cannot be assigned a metaphorical interpretation, the impersonal forms. . . are unacceptable” (2003: 484). (3) a.

#No,

jokos täällä kiehutaan? (F.) Well, already here boil.prs.imp ‘Well, is there any boiling going on here yet?’ b. *Aegutakse/Aeguti. (E.) expire.imp.prs/pst ‘One expires/ expired.’

Within the bounds of this basic restriction that the implicit argument must be human, its interpretation can vary depending on factors such as tense, lexical semantics of the verb, and discourse context (Shore 1988; Nelson & Vihman 2004; Löflund 1998). In both Finnish and Estonian, the implicit impersonal argument can be used with a narrow existential interpretation or a general universal reading.

. The ‘zero person’ construction in Finnish and Estonian The Finnic impersonal construction, discussed in the preceding section, provides a means of describing an event (or state) without specifying the identity of the referent of the actor (or undergoer). Although the particular identity remains unspecified, the interpretation of the impersonal includes a referent of some sort, and beginning in Section 4 we explore the semantic, syntactic and discourse-level status of this referent. In this section, however, we turn to another means – available in both Finnish and Estonian – of leaving the highest argument unspecified, namely the ‘zero person’ construction. Although anticausative, personal/resultative passive, and ‘zero person’ constructions all exist in some form as agent-demoting devices in both languages (e.g. Vihman, forthcoming, for details on Estonian), the zero person construction bears interesting similarities as well as important dif. It is worth noting that unaccusatives (e.g. to die) can also be impersonalized in both languages, as long as the implicit argument is human/highly personified (see e.g. Manninen & Nelson 2004 on Finnish).



JB[v.20020404] Prn:19/10/2006; 13:57



F: LA9605.tex / p.6 (357-426)


ferences to the impersonal. This paper compares only the impersonal and the zero person, leaving the others unexamined. The zero person employs a (default, non-agreeing) third person singular verb form with no special morphological marking and no overt subject, as shown in example (4). As the zero person construction has been less widely studied than the impersonal, the following discussion is meant to introduce the construction, but not as an exhaustive description. Furthermore, in the present discussion, we focus on the commonalities between the Finnish and Estonian zero person constructions and do not explore in any detail the question of whether there are some fine-grained differences between them. Sunnuntaina voi nukkua pitkään.9 (F.) Sunday-on can.prs.3sg sleep.inf long b. Pühapäeviti saab sisse magada. (E.) on-Sundays can.prs.3sg in.ill sleep.inf ‘On Sundays you/one can sleep in.’

(4) a.

This construction lacks an overt nominative subject; if there is an object present, it retains its object case and position, leaving no doubt that the object remains unpromoted. This is in contrast to the impersonal construction, in which full NP internal arguments occurring with accusative-assigning verbs show up with nominative case, and the (non)promotional nature of the construction remains a source of disagreement. It has been argued that the zero person construction contains a covert argument in subject position, and it is important to note that it is not accurate to describe this construction as being derived by syntactic demotion (see Vainikka & Levy 1999; Holmberg 2004 for more examples and further details). We refer to the null subject of the zero person as an implicit argument, and we use the term ‘implicit argument’ to refer to the ‘non-overt’ argument in both the impersonal and zero person constructions. The use of this term is for expository convenience, and should not be regarded as theoretically significant. The distinguishing semantic features of the zero person include a generic interpretation (Laitinen 1995; Erelt et al. 1993: 227) where the “action referred to is . . . generally applicable” (Penttilä 1963: 464, cited by Löflund 1998: 150, our translation), and the referent is interpreted as ‘whoever, anyone’ (Erelt et al. 1993). These constructions often have a modal interpretation, with the action interpreted as one which ‘can be’ or ‘must be’ performed (Hakulinen et al. 2004; Erelt et al. 1993). The zero person tends to occur with stative verbs, with the implicit argument in an experiencer rather than an agentive role (Löflund 1998: 154); this construc. Example from Shore (1988: 156).

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.7 (426-466)


tion is used mostly in the present tense, either indicative or conditional, with occasional occurrences of imperfect and perfect tenses (Löflund 1998:155; Vihman 2004: 92). In Estonian, the zero person allows active verbs mostly by pairing them with modals to give a stative reading. The zero person tends to refer to the state of potential for action rather than a dynamic event. In Finnish, in addition to a preference for appearing with modal verbs, zero person constructions often occur in conditional/hypothetical contexts (Vilkuna 1996; Hakulinen et al. 2004). Use of a conditional if/then construction makes agentive verbs possible in Finnish (e.g. Hakulinen & Karttunen 1973; Laitinen 1995; Löflund 1998: 155), as shown in example (5a), and (5b–c) show the zero person with a modal verb in both languages. Given that both conditionals and modals have been argued to involve quantification over possible worlds or situations (e.g. Kratzer 1986, see also Stalnaker 1968), the occurrence of the zero person in conditional and modal sentences in Finnish may be related to this kind of quantification. Jos ei kuuntele eikä tee tehtäviä, ei opi.10 (F.) 11 If neg listen neg-and do exercises neg learn ‘If one doesn’t listen and do homework, one doesn’t learn.’ b. Kotona voi myös testata erilaisia täytettyjä pastoja.12 (F.) at-home can also test.inf different.par filled.par pastas.par ‘At home one can also test different kinds of filled pasta.’ c. Oma tööd peab armastama.13 (E.) self.gen work.par.sg must.prs.3sg love.inf ‘One has to love one’s work/ you have to love your work.’

(5) a.

The referent of the implicit zero person argument patterns as a singular entity, unlike the implicit argument in an impersonal (Vilkuna 1996: 141). According to Löflund (1998: 156), the Finnish zero person refers distributively to one person at a time, whereas the impersonal can refer collectively to multiple people. The zero person governs only singular agreement (6a), whereas the impersonal can govern either singular or plural agreement (7a). The same holds for Estonian, as illustrated in (6b) and (7b).

. Example from Vilkuna (1996: 140). . In negative sentences in the present tense in Finnish and Estonian, the main verb is a bare inflectional stem without an ending. In Finnish, negation is an auxiliary and agrees with the subject in person and number. . From <www.soneraplaza.fi/ellit/artikkeli>. . Example from Erelt et al. (1993: 40).



JB[v.20020404] Prn:19/10/2006; 13:57



F: LA9605.tex / p.8 (466-532)


(6) a.

Jos ei ole tarpeeksi hieno (/*hienoja), ei pääse sisään. (F.) if neg be sufficiently fancy.sg (*fancy.pl), neg get in b. Kui ei ole piisavalt esinduslik(/*-ud), siis ei pääse sisse. (E.) if neg be sufficiently fancy.sg (*pl), then ø neg get in ‘If one isn’t sufficiently well-dressed, one won’t get in.’ Ollaanpas sitä taas hienoja(/hieno). (F.)14 15 be.imp.cl it.par again fancy.par.pl(/sg) ‘Well, aren’t you all (isn’t everyone) well-dressed again.’ b. Ollakse rõõmsad(/rõõmus), kui lapselapsed külla be.imp.prs joyful.nom.pl(/sg) when grandchildren visit.ill tulevad.16 (E.) come.3pl ‘People are happy when their grandchildren come to visit.’

(7) a.

. Probing for the implicit argument In the following sections, we investigate the semantic, syntactic and discourselevel status of the implicit argument in Finnish and Estonian impersonals and zero person constructions. In Sections 5 and 6, we investigate whether the implicit argument is available for different kinds of anaphoric reference, namely (i) cross-sentential definite pronouns, and (ii) intra-sentential referential reflexives, possessives and reciprocals. We then turn, in Section 7, to the classic diagnostic tests for probing the presence or absence of implicit arguments, including “licensing of by-phrases, the ability to control, and compatibility with adverbs like ‘deliberately”’ (Bhatt & Pancheva 2004: 4). In this section, before looking at the results of any of the tests, we consider what linguistic domain the tests are probing. Let us start with the tests that investigate the discourse properties of the implicit argument by testing whether it can act as the antecedent for a personal pronoun (Section 5). Here, we follow Koenig (1999) and Koenig & Mauner (2000), who investigate the discourse-level properties of the French impersonal pronoun on ‘one’ by testing whether it can be referred back to with a personal pronoun. In addition to investigating intersentential pronouns, in Section 6 we probe for the presence of the implicit argument on what we assume to be the syntactic level by testing whether the implicit argument of Finnish and Estonian impersonals and zero person constructions can . Examples from Vilkuna (1996: 140). . The partitive form of ‘it’ can be regarded as a kind of expletive, see Holmberg & Nikanne (2002). . Original (plural) example from <www.kirikuleht.ee>.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.9 (532-595)


serve as the antecedent for reflexives and other anaphoric forms within a sentence (see also Blevins 2003: 485, Maling this volume). In Section 7, we turn to the three classic tests discussed by Bhatt and Pancheva (2004, see also Maling this volume). As Bhatt & Pancheva point out, it is not clear whether the acceptability of adverbials like ‘deliberately’ requires a syntactically represented argument; it may well be the case that only a semantically realized agent is necessary. (In light of the observation that such adverbs are incompatible with unaccusatives, at least in English, Finnish and Estonian, we assume that they cannot introduce an implicit agent by accommodation.) A similar argument, Bhatt and Pancheva suggest, could be applied to by-phrases: perhaps all that is needed is that there be a syntactically unexpressed, semantically present agent.17 The third well-known test, control – in particular, control of rationale clauses – has also been claimed not to be a conclusive test of a syntactically realized implicit argument (see Williams 1985; Bhatt & Pancheva 2004). Since one could argue that these three tests are not conclusive in favor of a syntactically realized implicit argument, but given that passives clearly do have an implicit agent in their semantics and pass these tests, we will make the minimal assumption that these tests are sensitive to the presence of an implicit agent at least on the semantic level. In other words, in this paper we will treat adverbials like deliberately, by-phrases and control as semantic, not syntactic, tests.18

. Anaphoric reference: Definite pronouns Our investigation of the discourse status of the implicit argument is based on work by Koenig (1999) and Koenig & Mauner (2000) regarding the French impersonal pronoun on ‘one’. Koenig and Mauner, whose analysis is within the framework of . On a related note, Bhatt and Pancheva point out (and many others have also noticed) that middles cannot occur with by-phrases (examples from Keyser & Roeper 1984: 406). a. Bureaucrats were bribed by managers. b. *Bureaucrats bribe easily by managers. Bhatt and Pancheva also cite other tests for agentivity that middles fail (e.g. they cannot occur with rationale clauses or subject-oriented adverbs, Baker, Johnson & Roberts 1989), and note that researchers who argue in favor of middles having an implicit agent need to explain why middles fail these tests that the English passive passes. Moreover, as Rapoport (1999) has noted, other tests that have been used to argue in favor of middles having an implicit agent do not hold for all middles (see also Ackema & Schoorlemmer 1995). . These may, in fact, turn out to be syntactic tests in the end. Our choice to treat them as semantic diagnostics simply reflects the current lack of clear evidence in favor of a syntactic approach.



JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.10 (595-649)

 Elsi Kaiser and Virve-Anneli Vihman

Discourse Representation Theory (DRT), argue that although the French impersonal pronoun on ‘one’ is truth-conditionally equivalent to the indefinite pronoun quelqu’un ‘someone’, the two forms differ in that on does not introduce a ‘discourse marker’ into the discourse structure, while quelqu’un does. As a consequence, on is not available for certain types of anaphoric reference which are available with the indefinite pronoun. For example, as shown below, the definite pronoun il ‘he’ can be used to refer back to quelqu’un ‘someone’, but not to the impersonal on ‘one’: (8) a.

#On

a assassiné la présidente. Ili était du Berry, paraît-il. ‘One has assassinated the (female) president. He was from.the Berry, it seems.’ b. Quelqu’uni a assassiné la présidente. Ili était du Berry, paraît-il. ‘Someone has assassinated the (female) president. He was from.the Berry, it seems.’ i

Thus, underlying this test is the idea that if a referent cannot be referred to with a pronoun, it is not saliently realized on the discourse level. Like French on, neither Estonian nor Finnish impersonal constructions permit reference to the implicit argument in the subsequent sentence with a singular or plural definite pronoun (9).19 While the French examples use gender for disambiguating the pronoun referent, we use number, as neither Estonian nor Finnish encode gender in the pronominal system. (9) a.

Presidenttii murhattiinj . (F.) president.nom murder.imp.pst Häni/#j /He#i/#j on/ovat kotoisin Helsingistä. 3sg/3pl.nom be.3sg/pl.prs originally H.-ela ‘The presidenti was murdered Øj . Hei/#j is / They#i/#j are from Helsinki.’ b. Presidenti tapetij . (E.) president.nom kill.imp.pst Tai/#j /Nad??i/#j olevat Tallinnast pärit. 3sg/3pl.nom be.qtv T.ela from ‘The presidenti was killed Øj . Hei/#j is/ They??i/#j are apparently from Tallinn.’

Native speaker judgments indicate that, in both Estonian and Finnish, the plural pronoun in this context is infelicitous, and the singular pronoun can felicitously pick up only the patient (‘the president’) as its antecedent. If it were possible to refer to the implicit argument by a pronominal anaphor, we might expect the plural pronoun to be used rather than the singular, since the impersonal is typically . Subscripts on the verb in all examples index the implicit argument.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.11 (649-706)


interpreted as having a plural referent (Shore 1988; Rajandi 1999), and also because the singular pronoun shows such a strong preference for the singular patient. However, native speaker judgments indicate that neither a plural nor a singular pronoun can felicitously pick out the implicit argument (see Vihman 2004 for Estonian judgments). Likewise, Finnish and Estonian zero person constructions do not allow the implicit argument to be picked up by a singular or plural pronoun, as the examples below illustrate. Given that personal pronouns pick out specific discourse referents, the data in (10) fit in well with the description of the zero person construction as one that does not refer to any specific person but rather to a non-specific, generic human referent. (10) a.

Jos øi murhaa presidentinj , hänj/#i joutuu/hek/#j/#i joutuvat if murders president-acc, s/he lands/they land vankilaan. (F) jail.ill ‘If onei murders the presidentj , s/hej/#i / theyk/#i/#j end up in jail.’ b. Kui øi tapab presidendij , taj/#i läheb/nadk/#j/#i if murder.prs.3sg president.gen 3sg go.prs.3sg/3pl lähevad vangi. (E.) go.prs.3pl jail.ill ‘If onei murders the presidentj , s/hej/#i / theyk/#i/#j go to jail.’

Koenig and Mauner (2000) use Discourse Representation Theory (DRT) in their analysis to underline the distinction between referential and role-related properties of NPs. They claim that so-called ‘a-definites’, such as the impersonal pronoun in French, satisfy an argument position (fill a role in the predicate) without introducing a discourse marker. In a similar vein, the data in (9) and (10) above suggest that the Finnish and Estonian impersonal and zero person constructions do not introduce a salient discourse referent, but nevertheless do fill an argument position in the predicate. The implicit argument makes a semantic contribution in that it is interpreted as human and it satisfies the highest argument position, but – as the pronoun data show – it is not as salient, on the discourse level, as an entity that is explicitly mentioned (i.e., in the examples above, ‘the president’).

. Anaphoric reference: Possessives, reflexives, reciprocals In this section we investigate the syntactic properties of the implicit arguments of impersonal and zero person constructions by probing their abilities to act as antecedents for possessives, reflexives and reciprocals in Finnish and Estonian. With the impersonal construction, we will see that the binding abilities of the



JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.12 (706-768)

 Elsi Kaiser and Virve-Anneli Vihman

implicit argument are more limited in Finnish than in Estonian, but that the evidence regarding syntactic realization or demotion of the implicit argument is not definitive in either language. In contrast, in the zero person construction, the binding data indicate syntactic realization of the implicit argument in both Finnish and Estonian. . Binding in the impersonal construction Let us start by considering the impersonal construction, where a detailed look at the data reveals differences in the behavior of the implicit arguments. Even though the implicit argument in the impersonal construction is not salient enough to be picked up by a pronoun in a subsequent sentence, in Estonian it can nevertheless participate in other referential relations. The Estonian implicit argument can bind the (non-person-specific) subject-oriented possessive oma (11a), as well as the reflexive (ise)end ‘self ’ (11b) (see also Blevins 2003), and the reciprocal üksteist ‘one-another’ (11c), which also tend to be subject-oriented. Kui sageli pestii omai (/temaj ) keha üleni?20 (E.) how often wash.imp.pst self/3sg.gen body.gen.sg overall ‘How often did onei wash one’si (his/herj ) whole body?’ b. Kord päevas pestii endi üleni külma veega.21 (E.) once day.ine wash.imp.pst ref.par overall cold.gen water.com ‘Once a day one washed oneself in cold water.’ c. Üksteisti hoitii asjade käiguga one-another.par hold.imp.pst thing.gen.pl process.com kursis.22 (E.) course.ine ‘One/people held one another informed of the course of events.’

(11) a.

In Estonian, the possessive pronoun oma (not marked for person) refers to a subject antecedent, in clear contrast to tema, which is a genitive third-person pronoun referring to a non-subject referent (see 11a). Only the subject-oriented possessive oma, not tema, can be used to refer to the impersonal argument. In (11a), with no alternative referent provided, tema can only refer to an unmentioned sentenceexternal referent. Like oma, reflexive and reciprocal pronouns are also primarily subject-oriented, but are also sensitive to the agency of the antecedent (e.g. data given in Erelt et al. 1993: 201). . From <www.erm.ee/?node=276>. . From <www.miksike.ee/lisa/6klass/5kreeka/ateena_ref.htm>. . From .

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.13 (768-830)

Effects of demotion in Estonian and Finnish 

The data in (11) seem to suggest that the implicit impersonal agent patterns with subjects, acting as the antecedent for subject-oriented possessive, reflexive and reciprocal pronouns. However, an implicit impersonal agent is not necessarily a stronger antecedent than an overt patient. In some contexts, the implicit argument can be outranked as anaphor antecedent if the sentence contains a featurally compatible overt argument. As pointed out by Erelt et al. (1993: 12) and Hiietam (2003), in sentences like (12a–b) that contain a preverbal patient, the patient can act as the antecedent of the anaphor (see also Hiietam & Manninen 2005). (12) a.

Lapsi pandij laua äärde omai koha peale child.nom put.pst.imp table.gen at.ill own.gen seat.gen on.all istuma. (E.) sit.sup ‘One sat the child at the table at his/her own seat.’ pandij rääkima iseendasti . (E.) b. Lapsi child.nom put.pst.imp talk.inf self.gen ‘One made the child talk about his/herself.’

In the examples in (12), the implicit argument ‘loses out’ to an overtly realized preverbal argument. The implicit argument here appears be a weaker binder than an overt, full-fledged constituent.23 However, the picture is further complicated by the observation that the overt patient is not always a stronger binder than the implicit argument. Although the possessive pronoun oma is unambiguously used in active sentences to refer to the subject, in certain impersonal constructions it is actually ambiguous between the implicit actor argument (13a) and the overt patient argument (13b). (13) a.

Relvadi võetij omaj elude hinnaga weapons.nom take.imp.pst own life.gen.pl price.com vaenlaselt.24 (E) enemy.abl ‘Weapons were taken from the enemy at the cost of one’s own life.’ b. Pallidi pannaksej omai algasetusse tagasi.25 (E) ball.nom.pl put.imp.prs own.gen start-position.ill back ‘The balls are put back into their (‘own’) starting position.’

. More research is needed in order to fully understand the interaction of word order and binding. In this paper, we aim to – at least partially – control the effects of word order by focusing mostly on examples where the anaphor occurs towards the end of the sentence. . From . . From <www.piljard.ee/reeglid.php>.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.14 (830-897)

 Elsi Kaiser and Virve-Anneli Vihman

In (13a), the possessive pronoun oma refers to the implicit actor (the people taking weapons) despite the compatible overt preverbal patient relvad ‘weapons’, whereas in (13b), oma refers to the overt patient pallid ‘balls’. The observation that subjectoriented oma can be used to refer to the nominative preverbal argument in some cases and to the implicit agent in other cases, combined with the data in (11), highlights the syntactically unclear status of the implicit agent. Even though the implicit argument can act as the antecedent for canonically subject-oriented possessives, reflexives and reciprocals, and can override an overt nominative patient as referent for the subject-oriented pronoun, it does not always do so. In fact, the subsequent discussion suggests that one could regard the implicit argument as being more syntactically active in Estonian than in Finnish, but even the Estonian implicit agent is not comparable to an overt argument. Before turning to the Finnish data, it is worth noting that, whereas Estonian has independent lexical items for possessives, reflexives and reciprocals, Finnish relies on a system of possessive suffixes that attach to (i) the possessed noun, in the case of possessive constructions, (ii) the reflexive stem itse ‘self ’ in the case of reflexives, and (iii) the reciprocal stem toinen ‘other’ in the case of reciprocals. In third person possessive constructions (e.g. ‘her book’), the possessive suffix ([-nsA] or [-An], capitalization indicates that the vowel surfaces as a or ä, depending on vowel harmony) is present on the possessed noun. If the possessor is the subject of the sentence, an overt possessive pronoun is normally not present (Vilkuna 1996: 228–230; Nelson 1998: 13). When an overt possessive pronoun (hänen s/he.gen) is present, the possessor is normally interpreted to be someone other than the subject (Nelson 1998: 13). In light of these kinds of data, Vainikka (1989), Nelson (1998) and Trosterud (1993) claim that third person possessive suffixes are anaphors which must be locally bound by the subject of the sentence or by a third person possessive pronoun (e.g. Nelson 1998: 187–188). It is important to note, however, that some exceptions exist. In certain cases, it is possible for a nonsubject to be the antecedent for a possessive suffix, especially if pragmatic factors provide a bias towards such an interpretation (e.g. Vilkuna 1996: 232). In reflexive constructions, the possessive suffix attaches to the reflexive stem itse ‘self ’. Given what we already know about the possessive suffix, it is not surprising that the third-person reflexive itsensä is basically subject-oriented (Hakulinen & Karlsson 1988; Trosterud 1993).26 The morphology of reciprocals is somewhat more complex, and we discuss it in more detail below. In Finnish, as examples (14a, b) show, possessives and reflexives cannot be bound by the implicit argument (Manninen & Nelson 2004, see also Hakulinen & . In both Finnish and Estonian, embedded infinitivals often lead to ambiguity, providing two potential antecedents (see e.g. Vilkuna 1996: 233 on Finnish; Erelt et al. 1993: 200–201 on Estonian).

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.15 (897-962)

Effects of demotion in Estonian and Finnish 

Karlsson 1988; Vainikka 1989).27 Importantly, adding the possessive pronoun hänen (s/he.gen) to (14a) has the effect of inducing a reading where hänen refers to a sentence-external third person referent, but not to the implicit argument. Given that hänen noun+Px constructions are known to allow reference to non-subject constituents in Finnish, this indicates that the implicit argument does not have the syntactic status of a (sentence-internal) non-subject constituent either. This suggests that, for purposes of providing an antecedent for a possessive construction, the implicit argument does not pattern like overt constituents. (14) a. *Suihkussa pestiini hiuksiaani . (F.) shower.ine wash.imp.pst hair.par.3-poss ‘In the shower one washed one’s hair.’ (intended meaning) itseääni . (F.) b. *Suihkussa pestiini shower.ine wash.imp.pst self.par.3-poss ‘One washed oneself in the shower.’ (intended meaning)

The situation is somewhat more complex for reciprocals. The basic Finnish reciprocal form is toinen ‘other’, which is either (i) used alone, in the plural form with a possessive suffix (see (15a)), or (ii) ‘doubled’ (15b), with the first occurrence being indeclinable and the second in the singular (sometimes plural) with a possessive suffix (examples from Sulkala & Karjalainen 1992: 146). (15) a.

Tytöt katsoivat toisiaan. (F.) Girls.nom looked other.pl.par.3-poss ‘The girls looked at each other.’ b. Lapset tuijottivat toinen toistaan. (F.) Children.nom stared other other.par.3-poss ‘The children stared at each other.’

‘single form’

‘double’

The binding possibilities for reciprocals appear to differ from those of reflexives and possessives. Hakulinen et al. (2004: 1264) provide examples of the ‘doubled’ form being bound by the implicit agent in the impersonal construction. This is also illustrated in (16). The ‘non-doubled’ form, however, seems to sound more

. In their extensive corpus study, Hakulinen et al. (2004) did find some occurrences of reflexives and possessives bound by the implicit argument of an impersonal construction. However, they comment on the markedness of such examples and emphasize that they are unusual. In fact, they describe the impersonal as a structure that does not fit together with (subjectbound) reflexives and possessives, which matches the negative reactions of native speakers that we have observed in response to impersonal sentences with reflexives/possessives. Furthermore, Hakulinen et al. point out that the impersonal differs from the zero person in that the implicit argument of the zero person can indeed bind reflexives and possessives without any problems (see also example (17) below).

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.16 (962-1012)

 Elsi Kaiser and Virve-Anneli Vihman

marked to native speakers – for some speakers almost verging on ungrammatical – in such contexts. (16) Minun mielestäni ystävyys on sitä, my.gen mind.ela.1-poss friendship.nom is it.par että tuetaan toinen toistaan that support.imp other other.par.3-poss ja molempien asiat ovat yhtä tärkeitä.28 (F.) and both.pl.gen things are equally important.par.pl ‘In my opinion friendship means that you support each other and that both people’s concerns are equally important.’

The contrast between the ‘non-doubled’ and the ‘doubled’ forms is not surprising if the structure of constructions like toinen toistaan is such that toinen can act as the binder for the suffix on toistaan. (As mentioned above, it is often assumed for Finnish that the possessive suffix itself – not the stem to which it attaches – is the anaphor.) In other words, in examples like (16), we may be dealing with ‘internal’ binding within the doubled reciprocal form (see also Trosterud 1993 on Finnish possessive suffixes being bound DP-internally by an overt pronoun within the same DP).29 If this is the case, i.e. if the relevant binding is happening DPinternally, then the doubled reciprocal cannot tell us about the syntactic status of the implicit agent. In order to see if this ‘internal binding’ explanation is correct, further work needs to be done on the internal structure of reciprocal forms like toinen toistaan. Moreover, it has been suggested that the grammaticality of some of the reciprocal forms may be connected to the fact that reciprocals require plural antecedents, and the implicit argument in Finnish impersonals tends to be interpreted as referring to a plural entity (Hakulinen et al. 2004: 1263). However, it is not clear how such a plurality account fits in with the fact that a possessive construction containing an overt plural genitive pronoun does not seem to be able to refer to the implicit argument. Clearly, more research is needed in this area in order to better understand the differences between the Finnish possessive suffixes and the Estonian anaphors and how they relate to the observation that in Finnish, the binding abilities of the implicit argument are even more reduced than in Estonian. Given the morphological differences between Finnish and Estonian, it could be the case that the differences in the binding patterns (in particular, the inability of the implicit argument in the Finnish impersonal to act as an antecedent) have something to do with the suffixal . From <www.iltasanomat.fi/arkisto> (keskustelu), 5.4.2004. . As a reviewer notes, this doubling might be somewhat similar to the English construction each. . . the other.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.17 (1012-1063)

Effects of demotion in Estonian and Finnish 

nature of the Finnish system. However, it is important to note that the suffixes can be bound by overt arguments that are present in the syntax, and thus it is not clear why a syntactically present but covert argument should not be able to do so (see also ex. (17) below). In other words, if the implicit argument in the impersonal were syntactically present, we would expect it to be able to act as an antecedent, at least for the possessive constructions with genitive possessive pronouns (which can have non-subject antecedents). The fact that it cannot do so suggests that, when it comes to these kinds of anaphoric relations, the implicit argument in the Finnish impersonal does not have the same syntactic status as overt (subject or non-subject) arguments. Overall, we conclude that when it comes to anaphor binding, the implicit argument of impersonals is not directly comparable to a ‘regular’ overt argument in either Finnish or Estonian. In other words, we do not have clear evidence in favor of the claim that the implicit argument is fully realized on the syntactic level in either language. . Binding in the zero person construction Turning now to the zero person construction in Finnish and Estonian, we see a new set of binding possibilities. In Finnish, in zero person constructions, the implicit argument can bind both possessives and reflexives (e.g. Vainikka 1989; Holmberg 2004; Hakulinen et al. 2004). . . . jos pesee vaatteitaan matkalla.30 (F.) if wash.prs.3sg clothes.par.3 trip.ade ‘. . .if one washes one’s clothes while traveling.’ b. . . . jos pesee itsensä suihkussa jollain if wash.prs.3sg self.acc.3 shower.ine some.ade hierontakintaalla.31 (F.) exfoliating-mitten.ade ‘. . .if one washes oneself in the shower using a kind of exfoliating glove.’

(17) a.

Not surprisingly, if the overt genitive possessive pronoun hänen (s/he.gen) is added to the examples above, the implicit argument cannot bind the possessive anymore. This shows that the implicit argument in the zero person construction patterns like a syntactic subject in Finnish. Similarly, in Estonian zero person constructions, both possessives and reflexives can be bound by the implicit argument, as shown below. As with the . From <www.vr.fi/heo/lansi/vinkit.htm>. . From <www.keskustelu.suomi24.fi>, posted 11.6.2004.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.18 (1063-1126)

 Elsi Kaiser and Virve-Anneli Vihman

impersonal construction, replacing the subject-oriented possessive oma with the non-subject-oriented tema results in a different reading with sentence-external reference. In other words, the implicit argument in Estonian zero person constructions patterns with syntactic subjects, as in Finnish. (18) a.

Võib isennast süüdistada. (E.) can.prs.3sg ref.par blame.inf ‘One can blame oneself.’ b. Kanuumatkale võib oma lemmiklooma kaasa canoe-trip.all can.prs.3sg self.gen pet.par.sg along 32 võtta. (E.) take.inf ‘One can take one’s pet along on the canoe trip.’

In both languages, the implicit argument can act as the binder even if another featurally compatible referent is present in the sentence. This is illustrated for Finnish possessives and reflexives in (19a) and (19b) respectively, and for Estonian reflexives in (19c). This is in contrast to the Estonian data in (12) above. (19) a.

Jos panee lapsen sänkyynsä . . . (F.) if puts child.acc bed.ill.3-poss ‘If one puts the child into one’s bed. . .’ b. Jos puhuu naapurille itsestään. . . (F.) If speaks neighbor.all self.ela.3-poss ‘If one speaks to a neighbor about oneself. . .’ naabrile iseendast. (E.) c. Räägib talk.prs.3sg neighbor.all self.ela ‘One talks (could talk) to the neighbor about oneself.’

Having considered possessives and reflexives, let us now turn to reciprocals in zero person constructions. In Estonian, the implicit argument of the zero person construction can act as the antecedent of a reciprocal, as shown in (20a). In contrast, the implicit argument is not available for reference with a (plural) reciprocal pronoun in Finnish (20b). (20) a.

Võib teineteist süüdistada. (E.) can.prs.3sg rec.par blame.inf ‘People (‘one’) can blame each other.’ b. *Jos auttaa toisiaan. . . (F.) if help other.par.pl.3-poss ‘If you help each other. . . ’

. From <www.soomaa.com>.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.19 (1126-1185)

Effects of demotion in Estonian and Finnish 

The explanation for the compatibility of the zero person with the reciprocal pronoun in Estonian (20a) is unclear. It appears that, while the Estonian zero person can only take singular adjective agreement, it can nevertheless have a distributive reading, as shown with the reciprocal pronoun (cf. Löflund’s comment (1998: 156) that the zero person functions distributively). In light of the plural marking on the Finnish reciprocal, and the fact that the null subject of the zero person construction is syntactically singular (e.g. ex. (6)), the ungrammaticality in Finnish is not surprising (Hakulinen et al. 2004: 1286, see also Heim, Lasnik, & May 1991 on plurality and reciprocals). However, the basic word toista (other-part) – with no plural marking and no possessive suffix – can be used in zero person constructions (see Hakulinen et al. 2004). In such uses, though, it is not clear whether, in the absence of a possessive suffix, this form can still be treated as a reciprocal or whether it is simply the word meaning ‘other (person)’. In sum, in this section we investigated the binding abilities of the implicit argument in impersonal and zero person constructions in Finnish and Estonian. We saw that, in the impersonal construction, the binding abilities of the implicit argument are more reduced in Finnish than in Estonian, but the implicit argument of impersonals is not directly comparable to a ‘regular’ overt argument in either language. In contrast, the implicit argument in the zero person construction acts as a binder in both Finnish and Estonian, much like an overt subject would. So, while we do not have clear evidence that the implicit argument is fully realized on the syntactic level in the impersonal in either language, binding data show that it is syntactically realized in the zero person construction (see also Holmberg 2004).33 . Semantic tests In this section, we investigate the semantic status of the implicit argument in impersonals and zero person constructions in Finnish and Estonian. A range of different tests indicate that the implicit argument is present on the semantic level in both constructions in both Finnish and Estonian. . A reviewer mentions a possible semantic account of the binding data, namely that the agent of the impersonal construction is not sufficiently individuated (because it tends to be interpreted as plural) to be picked up as an antecedent, whereas the agent of the zero person construction is referentially stronger (singular, more individuated) and can thus function as an antecedent. Although this is an interesting idea, we do not discuss it in the present paper for reasons of brevity. Furthermore, it is worth noting that in both Finnish and Estonian, the impersonal construction can also be used in contexts where the agent is a particular, known individual (see e.g. Löflund 1998: 84 on Finnish). This observation does not seem compatible with the idea that it is low individuation that prevents the impersonal implicit agent from acting as an antecedent.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.20 (1185-1254)

 Elsi Kaiser and Virve-Anneli Vihman

. Adverbials Agent-oriented adverbials such as ‘intentionally’ or ‘on purpose’ can be used to test for the semantic presence of a volitional agent. These are felicitous with impersonalized agentive verbs (21)–(22) as well as zero person constructions (23)–(24) in both Finnish and Estonian. In all the examples below, the agentive adverbial is interpreted as referring to the implicit argument. (21) Kolmihenkisen perheen päälle ajettiin tahallaan three-person.gen family.gen over drive.imp.pst on-purpose keskiviikkona klo 14.40 Helsingin Malmilla.34 (F.) Wednesday.ess o’clock 14:40 Helsinki.gen Malmi.all ‘Someone intentionally/on purpose drove over a three-person family at 14:40 in the Malmi region of Helsinki.’ (22) Suur osa kirjavigadest tehakse vist large part.nom spelling-errors.ela make.imp.prs apparently meelega.35 (E.) on-purpose ‘A large portion of the spelling errors seem to be made on purpose.’ (23) Eri asia on jos tahallaan kävelee suoraan Different.nom thing.nom is if on-purpose walks directly latu-urien päällä.36 (F.) skiing-tracks.gen on-top ‘It’s a different matter if one walks directly on top of the skiing tracks on purpose.’ (24) Võib osta meelega liiga palju õlut. (E.) can.prs.3sg buy.inf on-purpose too much beer.par.sg ‘One could buy too much beer on purpose.’

We interpret the fact that agent-oriented adverbials in both Finnish and Estonian can refer to the implicit argument as evidence of its semantic presence in both of these constructions (see also Bhatt & Pancheva 2004).37 . From <www.samikarjalainen.fi/hullu_maailma.html> (an excerpt appeared in Helsingin Sanomat, 8.12.01). . From the discussion forum . . From <www.jyvaskyla.fi/kysy/kysymys.php/2267>. . Finnish and Estonian also possess some infinitival (F.) and gerundive (E.) constructions that might shed light on the status of the implicit argument. However, these constructions have not received much attention in the existing literature and as a result their basic syntactic and semantic properties are not fully understood. Thus, they are not (yet) well-suited for probing

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.21 (1254-1313)


. Control structures Finnish and Estonian both have subject control constructions (25a), (26a), and in both languages, the implicit argument of the impersonal can control PRO in this construction, as shown in (25b) and (26b). The implicit argument of the zero person construction is also capable of control (27) and (28). (25) a.

Matti yritti oppia järjestystä. (F.) Matti.nom tried PRO learn orderliness ‘Matti tried to learn orderliness.’ (to be orderly) b. Siellä yritettiin oppia järjestystä.38 (F.) there try.imp learn orderliness.prt ‘One tried to learn orderliness there.’

(26) a.

Jaan lubas ära minna. (E.) Jaan.nom promised PRO away go ‘John promised to leave.’ b. Lubati ära minna. (E.) promised.imp away go ‘They/ people promised to leave.’

(27) Tämä saattaa tuntua ärsyttävältä, jos yrittää oppia hollantia. . . (F.) This.nom might feel annoying, if tries learn Dutch.par ‘This [the fact that everyone speaks English] might feel annoying if one is trying to learn Dutch. . .’39 (28) Proovib rääkida kogu õhtu ainult inglise keeles. (E.) try.prs.3sg talk.inf all evening only English.ine ‘Let’s try to speak only in English all evening.’

We interpret these data as corroborating the adverbial test, showing that the implicit arguments of impersonals and zero person constructions are represented at least on the semantic level. We follow Williams (1985) (see also Bhatt & Pancheva 2004) in not taking control to be a fully conclusive test for a syntactically present implicit argument. . By-phrases It is a well-known observation that in English, passives allow for by-phrases, whereas middles and unaccusatives do not (e.g. Roeper 1987). This has been taken the status of the implicit argument. Hopefully future research will further our understanding of these structures. . From Vilkuna (1996: 275). . From <www.uta.fi/opiskelu/kv-asiat/tarinat/slk2000_haag.html>.



JB[v.20020404] Prn:19/10/2006; 13:57



F: LA9605.tex / p.22 (1313-1383)


to indicate that passives have implicit agents that have been demoted but that can be re-established by means of a by-phrase, whereas middles and unaccusatives have no implicit agents and are thus incompatible with by-phrases. (29) a. The ship was sunk by Bill.40 b. *The ship sank by Bill.

English ‘one’-constructions, similar in some respects to the Finnish and Estonian zero person construction, do not allow by-phrases, presumably because the agent is already overtly expressed in the sentence: (30) One would think (*by computer users) that Windows was enough of a bad example that nobody would ever try that layout again.41

If we apply the by-phrase test to Finnish and Estonian impersonals, the results are less clear than with English. Although Estonian allows the identity of the implicit argument to be established in an oblique phrase, there are restrictions on its acceptability. An adverbial phrase identifying the agent referent is acceptable, especially if the semantics of the agent are compatible with the semantics associated with the impersonal: generalized, unspecific groups tend to be considered grammatical (e.g. ‘by the county council’ in (31a)), but specific, identifiable individuals (‘by us’ in (31a)) are not. Specific names or pronouns used in agentive oblique phrases are generally deemed awkward. In present-day Estonian however, this distinction can be blurred to a certain degree: it is becoming more common to find examples such as that in (31b), where the agentive adverbial refers to a specific, singular individual. However, the singular referent of the agentive phrase in example (31b) has institutional relevance, and could be argued to be acceptable here only through the symbolic level of the mayor representing the town or the local government. (31) a.

Komisjon kinnitatakse vallavolikogu (/*meie) commission.nom confirm.imp.prs county-council.gen (/ we.gen) poolt.42 (E) by ‘The commission is confirmed by the county council (*by us).’ b. Diplom, mis kirjutatakse alla linnapea poolt...43 (E) diploma.nom.sg what.nom write.imp.prs under mayor.gen by ‘The diploma, which is signed by the mayor...’

. Examples in (29) are from Roeper (1987). . Modified from a comment posted 7.1.2005 at comp.os.linux.advocacy. . From . . From .

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.23 (1383-1444)


Although judgments on by-phrases such as that in (31b) in Estonian impersonals vary, the fact that names and pronouns are still generally found to be awkward shows that the specific/nonspecific distinction has not been entirely blurred. The varying judgments are not surprising if we keep in mind that the use of by-phrases with impersonals is something that was ‘imported’ into Estonian from IndoEuropean. Nemvalts (1998) notes that ‘poolt-phrases’ represent an Indo-European influence in Estonian (1998: 63), and Blevins (2003: 485–486) lists authors who point to the by-phrases as gaining acceptability but feel that the agentive by-phrase is somehow ‘intrusive’ when used in an impersonal (including Torn 2002 and Tuldava 1994). In Finnish, the situation is even less clear. Adverbial clauses formed with NPGEN toimesta ‘by NP’ do occur in Finnish, but they are often regarded by native speakers, linguists and prescriptive grammarians alike as sounding awkward, part of a ‘legalese’-type official register, and not directly comparable to English-style byphrases (Länsimäki 1999; Vilkuna 1996: 147 inter alia). Semantically, as Hakulinen et al. (2004) point out, they often refer to institutions or collectives that are interpreted not as the actual direct agent of an action but as the entity that initiated or ordered an action to be carried out, as shown in (32), (see also Löflund 1998: 28). We will refer to these as ‘indirect agents’. This generalization may also be relevant for Estonian, given that collective, institutional entities like the ‘county council’ can be used in poolt-phrases, but actual individual entities sound much more marked for many speakers. (32) Vääräuskoiset hugenotit murhataan kuninkaan Heretic huguenot.nom.pl murder.imp.prs king.gen toimesta.44 (F.) by #‘The heretic huguenots are murdered by the king.’ ok ‘The king has the heretic huguenots murdered.’

Example (32) is most naturally interpreted as meaning that the murders were carried out under the king’s orders, but the king was not the actual agent of murder (see also Länsimäki 1999). These structures are thus reminiscent of English had-constructions such as ‘The king had the heretics murdered.’ The claim that toimesta-phrases are not directly comparable to English-style by-phrases is also supported by the fact that toimesta-phrases can surface in active sentences with overt agents, as shown in (33). It is clear that in this example, the

. From <www.leffa-arviot.com/arvostelut> (in a review of D. W. Griffith’s movie “Intolerance”).



JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.24 (1444-1498)

 Elsi Kaiser and Virve-Anneli Vihman

two men mentioned in the toimesta-phrase are not the agents of the change: the actual agent is the government of the theater school. (33) Jouko Turkan ja Jussi Parviaisen toimesta teatterikoulun Jouko Turkka.gen and Jussi Parvianen.gen by theater-school.gen hallitus muutti keväistä board-of-directors.nom changed spring-time rangaistuspäätöstään.45 (F.) punishment-decision.par.3-poss ‘As a result of the actions of Jouko Turkka and Jussi Parvianen, the board of directors of the theater school changed the decision it had made in the spring regarding punishment.’

However, there do exist in Finnish naturally-occurring examples of toimestastructures being used to refer to direct agents, as shown in (34). It seems that variation exists in native speaker judgments, with some speakers only accepting toimesta-adverbials for indirect agents/initiators and others extending their use to direct agents as well. (34) Seurannut kulmapotku annettiin Mikan toimesta subsequent corner-kick.nom give.imp.pst Mika.gen by vasemmalta puolelta.46 (F.) left.abl side.abl ‘The next corner kick was given by Mika, from the left side.’

It may well be that the Finnish system is currently in a state of transition. What is clear, however, is that in general, toimesta-adverbials are perceived to be more marked than and functionally different from by-phrases in languages like English and German. As with Estonian, given that the use of toimesta-adverbials with impersonals has been ‘imported’ into Finnish from Indo-European (e.g. Häkkinen 1994), it is not surprising that it patterns differently from English-style by-phrases. In fact, it seems that the importation process has progressed less in Finnish than in Estonian, since Estonian seems more flexible than Finnish in its use of pooltadverbials, but even Estonian does not accept oblique phrases expressing the agent to the same extent that English passives do. In our opinion, the non-Finnic origin of the Finnish and Estonian by-phrases, combined with the resulting semantic restrictions, means that these structures cannot be directly compared to English-style by-phrases. In other words, since neither Finnish nor Estonian has a ‘native’ way of expressing the agent in impersonal . From Vilkuna (1996: 147). . From .

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.25 (1498-1553)


structures, and since the imported versions are subject to semantic restrictions regarding the degree of direct agentivity (vs. mere ‘initiatorship’) they are able to express, we conclude that this is a test that cannot be extended from IndoEuropean to Finnic: in our view, toimesta- and poolt-phrases cannot be reliably used to probe for the presence of a demoted implicit agent in Finnish and Estonian impersonals. Interestingly, if we turn to the zero person construction in Finnish and Estonian, the data are much clearer. Neither language allows by-phrases with this construction, as shown below. This result does not seem to depend on the semantics of the by-phrase. For example, a universal reading does not make the by-phrase more grammatical in the zero person construction. (35) a.

Täältä näkee koko kaupungin (*turistien /jokaisen toimesta). (F.) here.abl see whole city.acc (tourists.gen/everyone.gen by) ‘One can see the whole city from here (*by the tourists/everyone).’ b. Siit näeb tervet linna (*turistide/kõigi poolt). (E.) here.ela see.prs.3sg whole.par city.par (tourists/everyone.gen by) ‘You can see the whole city from here (*by the tourists/everyone).’

Given the generic interpretation of the zero person, these data are not surprising: it is not possible to spell out the agent of a generic person, since reference to a generic person is by definition a generalization, not an episodic event with a specific agent. To put it differently, if one of the functions of the zero person construction is avoidance of mentioning an agent (any agent), then it is not unexpected that a by-phrase spelling out the agent (even a universal one) is incompatible with this construction.47 Even if we assume that toimesta- and poolt-phrases spell out only an initiator of the action, this too conflicts with the generic nature of this construction. Because of this semantic conflict, the by-phrase test cannot be used to probe for the semantic (or syntactic) presence of the implicit argument in the zero person.

. Bringing the data back home In the discussion so far, we have attempted to locate the implicit argument in both impersonals and zero person constructions with various tests and uncovered quite a diverse set of results. As we have looked at two different constructions across two different languages, let us pause to pull the data together into a concise summary . The question of how the generic/episodic distinction relates to the universal/existential distinction is interesting, but beyond the scope of this paper. See Condoravdi (1989) and AlonsoOvalle (2000) for evidence that these two distinctions do not coincide.



JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.26 (1553-1626)

 Elsi Kaiser and Virve-Anneli Vihman

Table 1. Summary of tests probing the status of the implicit argument Impersonal

Definite pronouns Binding – reflexives Binding – possessives Binding – reciprocals Agentive adverbials Subject control By-phrases

Zero person

Finnish

Estonian

Finnish

Estonian

* * * *? *?

* ()

* *

* *

(see Table 1), before examining what these results tell us about the status of the implicit argument in each of the contexts examined. As Table 1 shows, intersentential definite pronouns cannot refer to the implict argument in any of the four constructions. The intrasentential binding data distinguishes the Finnish impersonal from both the Estonian impersonal as well as the zero person in both languages. In future research, we hope to address the possibility that this result could be derived from the morphosyntactic differences that distinguish the Finnish and Estonian anaphor systems, rather than the status of the impersonal argument. Although the data from adverbials and control indicate that the implicit argument is present in the semantics of both constructions in both languages, the data from agentive by-phrases reveal a difference between the Finnish and Estonian impersonal. Although Estonian has not traditionally supported by-phrases, the impersonal seems to have drifted to a certain degree (presumably under a strong Indo-European influence), now accepting by-phrases more readily than the Finnish impersonal. Here, we have a clear delineation between the zero person, which is ungrammatical with by-phrases, and the impersonals, which have varying results depending on the agentivity and the language in question.

. Conclusions Given the agentive adverbial and control tests discussed in the preceding sections, we conclude that the implicit argument is present in the semantic representation of the impersonal and zero person constructions in both Finnish and Estonian. However, as the pronoun data show, the implicit argument is not available for definite pronominal reference with either the impersonal or the zero person constructions. From this we conclude that it does not project a salient discourse entity. In other words, we interpret the data as indicating that although the implicit argument is present on the level of semantics in both languages in both constructions, it is not

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.27 (1626-1694)

Effects of demotion in Estonian and Finnish 

fully realized on the discourse level – at least not ‘strongly’ enough to be picked up by a pronoun in a subsequent utterance. What about the syntactic level: is the implicit argument realized on the syntactic level in Finnish and Estonian impersonal and zero person constructions? Given the binding data, we conclude that the implicit argument is syntactically realized (as a non-overt subject) in the zero person construction in both languages (see also Holmberg 2004 for further evidence regarding Finnish zero persons). This conclusion is also supported by the fact that the object in zero person constructions consistently maintains its object case, in contrast to impersonal constructions. In addition, the incompatibility of by-phrases with the zero person construction may support our view that the implicit argument in this construction is syntactically present but semantically underspecified, in the sense that it does not pick out a particular referent. However, we also noted complications that arise in using the by-phrase as a test for syntactic presence of the agent. The by-phrase test cannot be imported wholesale to non-Indo-European languages without a critical analysis of what information this test actually gives us, and it appears that the data and judgments on agent-specifying phrases in Finnish and Estonian should be viewed with some skepticism. Turning now to the impersonal, what can we say about the syntactic status of the implicit argument? We do not have clear evidence for either full realization or full demotion of the implicit argument in either Finnish or Estonian impersonal constructions. In Finnish impersonals, the implicit argument cannot act as an antecedent for reflexives or possessives, and it seems that in the case of ‘doubled’ reciprocals, we may be dealing with a case of internal binding licensed by the presence of a doubled form. On the whole, it seems that the implicit argument is not fully present on the syntactic level in Finnish, at least not on a footing comparable to overt constituents. In Estonian, even though the implicit argument can act as the antecedent for reflexives, possessives and reciprocals, it can nevertheless be trumped by an overt nominative constituent. This is shown by examples such as (13), where the implicit argument is outranked by an overtly realized preverbal argument, and (14), which shows that the implicit argument and the overt argument can both provide an antecedent for the possessive pronoun oma. Thus, even though one could regard the implicit argument as being more syntactically active in Estonian than in Finnish, it is not comparable to an overt argument. In fact, the binding data, combined with the other diagnostics, seem to suggest that (i) the Estonian impersonal demotes the highest argument on the discourse level but leaves it at least partially intact on the syntactic level (i.e. the impersonalized argument slot is satisfied in the argument structure but referentially unspecified), and that (ii) the Finnish impersonal construction demotes the agent both on the syntactic level (at least from the perspective of anaphor binding) and on the discourse level (at least from the perspective of pronominal reference).

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.28 (1694-1728)

 Elsi Kaiser and Virve-Anneli Vihman

This analysis is also compatible with the disagreement in the literature regarding the internal argument in Estonian and Finnish impersonals. On one hand, the patient looks like a promoted subject, often taking a preverbal position48 and sometimes nominative case, but it also shows indications of not being fully promoted. Conditions such as negation and low telicity induce partitive case, and the binding data is open to various interpretations. In Estonian, the ambiguous status of the internal argument may be related to the ambiguous status of the partially demoted argument. In Finnish, the internal argument behaves slightly more like a promoted constituent, which may be related to the more clearly demotional nature of the impersonal. The status of the internal argument needs further investigation. In sum, the idea that demotion is not a monolithic phenomenon that applies equally on all linguistic levels seems to provide a good fit for the data presented here: more specifically, we hypothesize that Estonian impersonals demote the highest argument on the discourse level but leave it largely syntactically intact, although unspecified, whereas the Finnish impersonal constructions demote the agent syntactically as well as on the discourse level. This leads to a surprising and rather intriguing finding: these two constructions – the Finnish impersonal and the Estonian impersonal – that seem so similar, both historically and synchronically, are nevertheless subtly different with respect to what is arguably their most central function, namely the demotion of the agent. We hope to investigate the validity of this preliminary hypothesis more in future work, and also to look more closely at the potential effects of the anaphoric systems of Finnish and Estonian on the binding possibilities to see whether they could in fact be responsible for triggering the differences that arise for impersonal constructions in the two languages. Furthermore, it is clear that in order to understand the data presented in this paper more fully, a detailed theoretical analysis is necessary. Thus, the crucial next step will be an attempt to capture the data in a principled way within a theoretical framework. A more formal analysis will (i) enable us to spell out concretely what the data say about the status of the implicit argument in different linguistic domains, and (ii) will allow us to investigate more explicitly the implications – and the theoretical validity – of the ‘dissociation’ hypothesis, i.e. our tentative idea that dissociation between domains may be possible, in the sense that an argument can be demoted in one domain but not in another.

. Note, however, that word order has different implications for grammatical roles in Finnish and Estonian: in Finnish word order may be more of a predictive tool than in Estonian.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.29 (1728-1820)

Effects of demotion in Estonian and Finnish 

Taken as a whole, in addition to shedding light on the referential properties of what one might call ‘invisible’ arguments, the findings presented in this paper have important implications for our understanding of the notion of agent demotion in voice phenomena.

References Ackema, Peter, & Maaike Schoorlemmer (1995). Middles and non-movement. Linguistic Inquiry, 26(1), 173–197. Alonso Ovalle, Luis (2000). Is the ’arbitrary interpretation’ a semantic epiphenomenon? University of Massachusetts Occasional Papers, 23, 155–183. Baker, Mark, Kyle Johnson, & Ian Roberts (1989). Passive arguments raised. Linguistic Inquiry, 20(2), 219–252. Bhatt, Rajesh, & Roumyana Pancheva (2004). Implicit arguments. Ms., University of Texas, Austin and University of Southern California. URL: www-rcf.usc.edu/∼pancheva/ lingpapers.html Blevins, James P (2003). Passives and impersonals. Journal of Linguistics, 39(3), 473–520. Condoravdi, Cleo (1989). Indefinite and generic pronouns. In E. Jane Fee & K. Hunt (Eds.), Proceedings of the Eight West Coast Conference on Formal Linguistics (pp. 71–85). Stanford, CA: CSLI. Erelt, Mati, Reet Kasik, Helle Metslang, Henno Rajandi, Kristiina Ross, Henn Saari, Kaja Tael, & Silvi Vare (1993). Eesti keele grammatika II: Süntaks (Estonian Grammar II: Syntax). Tallinn: Eesti Teaduste Akadeemia Keele ja Kirjanduse Instituut. Hakulinen, Auli, Maria Vilkuna, Riitta Korhonen, Vesa Koivisto, Tarja-Riitta Heinonen, & Irja Alho (2004). Iso suomen kielioppi (The big grammar of Finnish). Helsinki: Suomalaisen Kirjallisuuden Seura. Hakulinen, Auli, & Lauri Karttunen (1973). Kadonneet henkilöt: Suomen kielen geneeristä lauseista (Vanished persons: On generic sentences in Finnish). In Lea Laitinen, Pirkko Nuolijärvi, Marja-Leena Sorjonen, & Maria Vilkuna (Eds.), Lukemisto. (Auli Hakulisen) Kirjoituksia kolmelta vuosikymmeneltä (A reader: Writings by Aulin Hakulinen from three decades) (pp. 13–28). Translated by Maria Vilkuna. Helsinki: SKS. Hakulinen, Auli, & Fred Karlsson (1988). Nykysuomen lauseoppia (Modern Finnish grammar). Helsinki: SKS. Heim, Irene, Howard Lasnik, & Robert May (1991). Reciprocity and plurality. Linguistic Inquiry, 22, 63–101. Hiietam, Katrin (2003). Definiteness and Grammatical Relations in Estonian. Ph.D. dissertation, University of Manchester. Hiietam, Katrin, & Satu Manninen (2005). Some thoughts on Balto-Finnic passives and impersonals: A reply to Blevins. Ms., University of Liverpool and University of Lund. Holmberg, Anders (2004). The empty generic subject pronoun in Finnish. Ms., University of Newcastle. Holmberg, Anders, & Urpo Nikanne (2002). Expletives, subjects, and topics in Finnish. In Peter Svenonius (Ed.), Subjects, Expletives, and the EPP (pp. 71–106). Oxford: Oxford University Press.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.30 (1820-1942)

 Elsi Kaiser and Virve-Anneli Vihman

Häkkinen, Kaisa (1994). Agricolasta nykykieleen: Suomen kirjakielen historia (From Agricola to present-day Finnish: The history of standard Finnish). Helsinki: WSOY. Keyser, Samuel Jay, & Thomas Roeper (1984). On the middle and ergative construction in English. Linguistic Inquiry, 15, 381–416. Koenig, Jean-Pierre (1999). On a tué le président! The nature of passives and ultra-indefinites. In B. Fox, D. Jurafsky, & L. Michaelis (Eds.), Cognition and Function in Language (pp. 256– 272). Stanford, CA: CSLI Publications. Koenig, Jean-Pierre, & Gail Mauner (2000). A-definites and the discourse status of implicit arguments. Journal of Semantics, 16, 207–236. Kratzer, Angela (1986). Conditionals. In Papers from the Regional Meetings (pp. 1–15). Chicago, IL: Chicago Linguistic Society. Laakso, Johanna (2001). The Finnic languages. In Ö. Dahl & M. Koptjevskaja-Tamm (Eds.), Circum-Baltic Languages, 1: Past and Present (pp. 179–212). Amsterdam: John Benjamins. Laanest, Arvo (1975). Sissejuhatus läänemeresoome keeltesse (An Introduction to Balto-Finnic Languages). Tallinn: Teaduste Akadeemia Keele ja Kirjanduse Instituut. Laitinen, Lea (1995). Nollapersoona (Zero person). Virittäjä, 3, 337–358. Länsimäki, Maija (1999). Poimitaanko marjat mummon toimesta? (Are the berries picked by the grandmother? / Does the grandmother have the berries picked?). An article published in the Finnish newspaper Helsingin Sanomat, 27.10.1999 Löflund, Juhani (1998). Suomen kirjoitetun yleiskielen passiivi (The passive in standard written Finnish). Åbo: Åbo Akademis Förlag. Manninen, Satu, & Diane Nelson (2004). What is a passive? The case of Finnish. Studia Linguistica, 58(3), 212–251. Nelson, Diane (1998). Grammatical Case Assignment in Finnish. New York, NY: Garland. Nelson, Diane, & Virve-Anneli Vihman (2004). The interpretation of implicit arguments in Estonian and Finnish. Paper presented at the Workshop on Passives, 20th Scandinavian Conference of Linguistics. University of Helsinki. Nemvalts, Peep (1998). Kas väliseesti keeles on märgata süntaktilist omapära? (Is there any syntactic idiosyncracy in Estonian abroad?). In Väliseesti keeles (Estonian abroad) (pp. 166– 178). Tartu: Tartu Ülikooli Kirjastus. Penttilä, Aarni (1963). Suomen kielioppi ( Finnish grammar). Porvoo: WSOY. Rajandi, Henno (1999 [1968]). Eesti impersonaali ja passiivi süntaks (The syntax of Estonian impersonals and passives). Tallinn: Eesti Keele Instituut. Rapoport, Tova R. (1999). The English middle and agentivity. Linguistic Inquiry, 30(1), 147–155. Roeper, Thomas (1987). Implicit arguments and the head-complement relation. Linguistic Inquiry, 18(2), 267–310. Shore, Susanna (1988). On the so-called Finnish passive. Word, 39, 151–176. Siewierska, Anna (1984). The Passive: A Comparative Linguistic Analysis. London: Croom Helm. Stalnaker, Robert (1968). A theory of conditionals. Philosophia, 5, 269–286. Sulkala, Helena, & Merja Karjalainen (1992). Finnish. London: Routledge. Torn, Reeli (2002). The status of the passive in English and Estonian. RCEAL Working Papers in English and Applied Linguistics, 7, 81–106. Trosterud, Trond (1993). Anaphors and binding domains in Finnish. In Anders Holmberg & Urpo Nikanne (Eds.), Case and Other Functional Categories in Finnish (pp. 225–243). Berlin: Mouton de Gruyter. Tuldava, Juhan (1994). Estonian Textbook [Indiana University Uralic and Altaic Series 159]. Bloomington, IN: Indiana University.

JB[v.20020404] Prn:19/10/2006; 13:57

F: LA9605.tex / p.31 (1942-1970)


Vainikka, Anne (1989). Deriving Syntactic Representations in Finnish. Ph.D. dissertation, University of Massachusetts, Amherst. Vainikka, Anne, & Levy, D. (1999). Empty subjects in Finnish and Hebrew. Natural Language and Linguistic Theory, 17, 613–671. Vihman, Virve-Anneli (2004). Valency Reduction in Estonian. Ph.D. dissertation, University of Edinburgh. Vihman, Virve-Anneli (Forthcoming). Construction-based interpretation of implicit arguments. In N. Adams, A. Cooper, F. Parrill, & T. Wier (Eds.), CLS 40, vol. II: The Panels. Chicago, IL: Chicago Linguistic Society. Vilkuna, Maria (1996). Suomen lauseopin perusteet (Foundations of Finnish syntax). Helsinki: Edita. Williams, Edwin (1985). PRO and the subject of NP. Natural Language and Linguistic Theory, 3(3), 297–315.



JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.1 (46-143)

Argument demotion as feature suppression* Dalina Kallulli The primary goal of this article is to define the role of the special morphology present in passives, anticausatives and other constructions involving so-called ‘argument demotion’. The main claim is that passives and anticausatives differ only with respect to the event type features of the verb but both arise through the same operation, namely syntactic suppression by special morphology of a feature in v. Passive constructions across languages can be made compatible by relegating the differences to simple combinatorial properties of verb and prepositional types and their interactions with other event functors, which are in turn encoded differently morphologically across languages. New arguments are brought forward for a causative analysis of anticausatives. Agentive adverbials are examined, and doubt is cast on the usefulness of by-phrases as a test for agentivity and argumenthood.

.

Introduction

It is well-known that in many languages, notably Indo-European, the intransitive/inchoative or anticausative alternant of a transitive-intransitive alternating pair systematically involves morphological marking that is shared by reflexive and/or passive predicates, in the form of a pronoun, a clitic or verbal inflection (Partee 1965; Nedyalkov & Silnitsky 1973; Grimshaw 1982; Marantz 1984; Rosen 1984; Everaert 1986; Haspelmath 1993; among many others). For instance, Albanian, much like Latin and Modern Greek, uses two distinct conjugational paradigms, namely active vs. non-active, corresponding (roughly) to the unerga-

* Research for this paper was funded by the Austrian Science Fund, grant T173-G03. I am grateful to Anna Asbury, Noam Chomsky, Joe Emonds, Jean-Pierre König and Peter Svenonius for discussions. I also thank two anonymous reviewers and the editors for helpful comments, as well as the audiences of the 2004 Oslo workshop “Demoting the Agent: Passive and Other Voice-related Phenomena”, MIT Ling-Lunch, CUNY Syntax-Supper, the DGFS 2005 workshop “Lexical Encoding of Implicit Information”, where parts of this material were presented.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.2 (143-181)

 Dalina Kallulli

tive vs. unaccusative distinction.1 Although these syncretism phenomena have been known for a long time, it remains unclear why a reflexive, a clitic or an affix should be used in the anticausative alternant. This paper attempts primarily to define the role of the special morphology in the anticausative formation and scrutinize once again the syntax and semantics of anticausatives, which as discussed below have been subject to various analyses. As part of a broader research agenda, I seek to explain the systematic syncretism between anticausatives and other unaccusatives, as well as passives. Unaccusatives (including anticausatives) and passives have generally been viewed as argument demotion phenomena, so ultimately this paper seeks to provide a unified analysis to several argument demotion phenomena. Under one influential proposal all unaccusatives have underlying causative semantics (Chierchia 1989, 2004; Reinhart 1996). This position has however been challenged by other researchers, such as Levin and Rappaport Hovav (1995), who argue that only alternating unaccusative verbs are basically causative, thus differing from Chierchia and Reinhart in not extending this analysis to non-alternating unaccusatives. In spite of this difference, common to both types of analyses is the idea that the intransitive form arises from a (reduction) operation in the lexicon, which in turn leads to what can be viewed as argument demotion. However, the idea that argument realization can be determined by operations in the lexicon is incompatible with non-lexicalist frameworks such as Distributed Morphology and/or other constructionist approaches, as these reject the notion of the traditional lexicon to start with.2 This paper provides evidence for Levin and Rappaport Hovav’s position, namely that only alternating unaccusatives have causative semantics. However, contrary to Levin and Rappaport Hovav, my main claim is that the intransitive variant is not derived through a lexical operation, but in the syntax proper. Hence the analysis that I develop here reformulates certain lexical semantic claims in a . This correspondence is rough by virtue of the fact that while unergatives are always active morphologically, some unaccusative verbs appear in this voice (i.e. are morphologically unmarked) too. For details, see Kallulli (1999) on Albanian, Alexiadou and Anagnostopoulou (2004) on Greek. Crucially, however, in both languages unergatives cannot be formally nonactive, just as passives and (lexical) reflexives cannot be formally active. . It is important to note that the conception of the role of morphology in argument expression might in fact have an impact on the architecture of grammar even for purely non-lexicalist models. For instance, it is not immediately clear whether and how the conception of morphology as an argument demotion or suppression operation in the syntax could be made compatible with a framework such as Distributed Morphology either, since here morphology simply reads off the output of syntactic derivations by (among other things) supplying phonological content to positions in a hierarchical structure.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.3 (181-236)

Argument demotion as feature suppression 

way that is coherent with non-lexicalist frameworks while capturing the empirical observations that lexicalists justly raise as seriously critical for constructionist approaches. However, the analysis put forward here is very different from aspectual approaches to unaccusativity (Borer 1994, 2004; van Hout 1996, 2004), according to which telicity is the determinant of unaccusativity. This paper is organized as follows. Section 2 introduces overt dyadic unaccusatives and discusses the ramifications of the data for theories on the anticausative formation. In Section 3, I undertake a detailed comparison of the Albanian and the English passives and anticausatives which sheds new light on several phenomena, such as the (in)significance of by-phrases, purpose clauses and so-called subject-oriented adverbs for theories of passive and anticausatives. In Section 4, summarizing and extending my earlier work on non-active morphology (Kallulli 2006a, b), I provide a formal analysis which uniformly derives dyadic unaccusatives with overt causers, passives and anticausatives.

. Dyadic unaccusatives and two ontological primitives Many languages share the construction in (1), in which a dative (or in some languages, a genitive) combines with an anticausative core yielding among other possible interpretations a reading that in previous work (Kallulli 2006a, b) I have referred to as ‘unintended causation’.3 That is, in the absence of the dative DPs the sentences in (1) would mean ‘The vase/window/car broke’.4 (1) a.

Ben-it i-u thye një vazo. Ben-theDAT himCL -NACT broke.AOR,3S a vaseNOM ‘Ben unintentionally broke a vase’ b. Dem Ben ist das Fenster zerbrochen. theDAT Ben is theNOM window broken ‘Ben unintentionally broke the window’

(Albanian)

(German)

. The other possible readings are a possessor reading (‘Ben’s vase broke’), and an affected (in the sense of benefactive/malefactive) reading (‘The vase broke on Ben’). I have shown in Kallulli (2006a, b) that the unintended causation reading is not due to pragmatic factors but is really part of the semantics of the verb, that is, the sentences in (2) are not vague but truly ambiguous. Therefore I will not dwell on this issue here specifically, though one argument for this view is presented further down in this section. . The following abbreviations are used in the glosses in the examples: nact (for non-active voice), aor (for aorist), cl (for clitic), dat (for dative case), imp (for imperfective), nom (for nominative case), p (for past tense), s (for singular).

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.4 (236-304)

 Dalina Kallulli

c.

A Pedro se le rompió el coche. To Pedro REFL itCL.DAT broke the car ‘Pedro unintentionally broke the car’

(Spanish)

The unintended causation reading does not obtain with non-alternating overtly dyadic non-active predicates.5 Consider the sentences in (2). (Albanian) Ben-it i-u hëngër një mollë. Ben-theDAT himCL.DAT -NACT eat.AOR,3S an appleNOM (i) ‘Ben felt like eating an apple’ or: ‘Ben was apple-hungry’ (ii)*‘Ben unintentionally ate an apple’ (Albanian) ndërtua një shtëpi. b. Benit i-u Ben-theDAT himCL.DAT -NACT build.AOR,3S a houseNOM (i) ‘Ben felt like building a house’ or: ‘Ben was in a house-building mood’ (ii)*‘Ben unintentionally built a house’

(2) a.

The Albanian sentences in (2) are formally identical with the sentence in (1a): both sets of sentences contain a verb that is morphologically non-active, and two overt arguments. And, as I stated in the previous section and have argued in detail in Kallulli (1999, 2006a, b), non-active verbs are always unaccusative in Albanian (see also Footnote 1). Yet, the unintended causation reading does not obtain. Instead, an involuntary state reading, rendered for lack of a better alternative through the ‘feel like’ gloss, obtains.6 The ‘core’ of the sentences in (2) is not anticausative, that is, in the absence of the dative DP and the clitic doubling it these sentences, unlike (1a), are not anticausative but passive, meaning ‘An apple was eaten’ and ‘A house was built’. However, the fact that both the anticausative core in (1a) and the passive core in (2) are morphologically identical, as well as the fact that, unlike in English in Albanian by- and from-phrases are always interchangeable (i.e. entail each other, or have identical distribution) as shown in (3), the distinction between passives and anticausatives is so far nothing more than an intuitive one.7 In fact, I will argue in Section 3 that the theoretical import that theories of passive in English ascribe to facts involving the distribution of by-phrases (and from-phrases) is simply not justified. . Here I am concerned with what could be referred to as morphologically unaccusative verbs, i.e., verbs that bear non-active morphology. See also Footnote 1. . As was the case also with the sentences in (1), the sentences in (2) may receive a possessor and/or affected interpretation, too. See Kallulli (2006a, b) for details. . Since by- and from-phrases in Albanian are full synonyms, that is, strictly speaking one cannot distinguish between them as shown for instance in (3a), in (3b) for simplicity I use only one option and gloss it ‘by/from’.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.5 (304-351)

Argument demotion as feature suppression 

(3) a.

Ben-it i-u nga ndërtua një shtëpi Ben-theDAT himCL.DAT -NACT build.AOR,3S a houseNOM from prindërit / prej prindërve. parents / by parents ‘A house was built to Ben by his parents’ b. Ben-it i-u thye një vazo nga i Ben-theDAT himCL -NACT broke.AOR,3S a vaseNOM from/by at-i / er-a. dad-the / wind-the ‘A vase was broken to Ben by his dad’ or ‘A vase broke to Ben from the wind’

So the question arises as to what exactly the anticausative/passive distinction is. I propose that this distinction is nothing more than a reflection of the different atoms that enter syntactic computation. I contend that both the anticausative and the passive construction are the outcome of one and the same phenomenon, namely the suppression by the special morphology of a feature in v that encodes the ontological event type of the verb.8 More specifically, I claim that anticausatives arise as a result of the non-active voice suppressing the feature [+cause] in v, whereas passives arise when the non-active voice suppresses the feature [+act] in v. In other words, I argue that [+cause] and [+act] are primitives. I now turn to the evidence for this view. As I have shown in previous work (Kallulli 2006a, b), while the unintended causation reading is missing in (2a), both the involuntary state reading and the unintended causation reading may obtain with one and the same predicate. This is illustrated through the Albanian examples (4a) and (4b), which differ in terms of their grammatical aspect only. As revealed in the glosses in (4a,b), Albanian has two forms for simple past tense (P) that differ in their aspectual value: aorist (Aor), which is perfective, and imperfective (Imp). Only the perfective sentence in (4a) but not the imperfective in (4b) can get an unintended causation reading. On the other hand, with imperfective aspect only the involuntary state reading but not the unintended causation reading obtains. (4) a.

Ben-it i-u thye një vazo. Ben-theDAT himCL -NACT break.AOR,3S a vaseNOM (i) ‘Ben unintentionally broke a vase’ (ii)*‘Ben felt like breaking a vase’

. Here by “special morphology” I mean what is traditionally referred to as “intransitivizing” or “detransitivizing” morphology. However, in view of the dyadic unaccusatives discussed here, the traditional terms do not seem fit. Since several languages have the voice distinction, from this point on I will refer to this special morphology as non-active voice throughout.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.6 (351-412)

 Dalina Kallulli

b. Ben-it i thy-hej një vazo. Ben-theDAT himCL break-NACT,P,IMP3S a vaseNOM (i) ‘Ben felt like breaking a vase’ (ii)*‘Ben unintentionally broke a vase’

Crucially, however, the semantic complementarity observed in (4) does not obtain with a non-external causation verb. The sentences in (5a) and (5b) differ morphologically exactly in the same way in which (4a) and (4b) differ. However, the unintended causation reading of (4a) does not obtain in (5b). (5) a.

Ben-it i ndërto-hej një shtëpi. Ben-theDAT himCL build-NACT,P,IMP,3S a houseNOM (i) ‘Ben felt like building a house’ (ii)*‘Ben unintentionally built a house’ b. Ben-it i-u ndërtua një shtëpi. Ben-theDAT himCL -NACT build.AOR,3S a houseNOM (i) ‘Ben felt like building a house’ (ii)*‘Ben unintentionally built a house’

So why does the pattern in (4) not replicate in (5)? The only possible explanation must be that non-active morphology interacts differently with different (feature) primitives. That is, the lexical (and consequently syntactic) feature composition make-up of eat is different from that of break.9 Thus, the claim that activity verbs (e.g. build) differ from causative verbs (e.g. break) lexically (and syntactically) in that the former project an [+act] feature and the latter a [+cause] feature in v makes sense.10 The fact that only alternating unaccusatives may get an unintended causation reading thus lends strong support to Levin and Rappaport Hovav’s (1995) claim that only alternating unaccusatives have causative semantics, contrary to Chierchia (1989, 2004) and Reinhart (1996). In the following section I provide more evidence for this view.

. The fact that the unintended causation reading does not obtain with non-alternating unaccusatives is at the same time an argument that the unintended causation reading is not due to pragmatic considerations, but is related to the semantics of the verb (see Footnote 3). Note also that it is not impossible to say ‘Ben unintentionally built a house (instead of a hut)’ and ‘Ben unintentionally ate an apple (e.g. instead of an orange) and that made him sick (because he has an allergy ...)’; yet, the unintended causation reading is impossible in the sentences (5a) and (5b) with non-active voice. . The ontological event types represented by [+act] and [+cause] can alternatively be represented in terms of binary features, i.e., [±act] or [±cause].

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.7 (412-466)

Argument demotion as feature suppression 

. Passives and anticausatives in English and Albanian . By-phrases and from-phrases: How different are they? Theories on the passive in English and similar languages capitalize on the fact that the logical-subject argument, not realized on an NP in passives, is realized in a byphrase (Baker, Johnson, & Roberts 1989; and references therein), which as Marantz (1984) shows, is independent of its theta-role. (6) a. b. c. d. e.

Hortense was pushed by Elmer. (AGENT) Elmer was seen by everyone who entered. (EXPERIENCER) The intersection was approached by five cars at once. (THEME) The porcupine crate was received by Elmer’s firm. (GOAL) The house is surrounded by trees. (LOCATION) (Marantz 1984: 129)

The fact that the logical subject of a passive can be realized in a by-phrase has been taken as evidence that the syntactically suppressed argument of a passive verb is present, nonetheless, in argument structure – its presence being manifested in the sanctioning of by-phrases, as shown in (7) (Roeper 1987; Grimshaw 1990). Likewise, Baker, Johnson and Roberts (1989) take the fact that passives may combine with a by-phrase as evidence for their alternative analysis, namely that the passive morpheme en is itself an argument, the idea being that the by-phrase refers (obliquely) to this argument. (7) The window was broken by Pat.

In contrast, as is also well-known, by-phrases are generally not licensed in anticausatives, as is shown in (8).11 (8) *The window broke by Pat.

In sum, the complete English paradigm can be depicted as in (9): a passive licenses a by-phrase irrespective of whether the logical subject is a participant capable of willful agency (such as ‘John’) or an extrinsic instigator (such as ‘the wind’). In contrast, anticausatives do not license by-phrases. The fact that anticausatives . However, as Levin and Rappaport Hovav (1995) note, anticausatives in English are perfect with the phrase by itself. This is shown in (i) and (ii). (i) The plate broke by itself. (ii) The door opened by itself. Levin and Rappaport Hovav (1995: 89) note that this “adverbial appears to be modifying a cause, which, given its anaphoric nature, it identifies as the theme argument itself ”. However, taken at face value, the fact that the phrase by itself can appear with anticausatives substantially weakens the generalisation that by-phrases are not sanctioned in anticausatives.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.8 (466-535)

 Dalina Kallulli

are bad with by-phrases has been taken to show that the binding of the external cause in an anticausative takes place in the mapping from the lexical semantic representation to argument structure (Levin & Rappaport Hovav 1995). (9) a. The window was broken by John / the wind. b. *The window broke by John / the wind.

However, anticausatives even in English can generally license from-phrases identifying the (external) cause of an event, as in (10), an example due to Joe Emonds (personal communication).12 (10) The window cracked from the pressure.

Levin and Rappaport Hovav (1995) do not discuss the fact that anticausatives in English may systematically combine with from-phrases, the implication being that from-phrases in anticausatives, unlike by-phrases in passives, do not make reference to an implicit argument in the argument structure of anticausatives. However, a closer look at the distribution of from-phrases in English is revealing of certain facts that have hitherto not been discussed. These facts motivate the ideas for an alternative analysis of both the passive and the anticausative formation developed later in this paper. First, note that even with respect to the licensing of from-phrases in anticausatives an interesting asymmetry obtains: while a from-phrase is generally fine, it is bad when the external causer is a participant capable of willful agency, as in (11). (11) *The window cracked from John.

Secondly, the difference in grammaticality between (7) and (8), effected by the distinction between an extrinsic instigator and a participant capable of willful agency, is also replicated with unaccusatives that do not participate in the transitiveinchoative alternation, such as ‘die’, as shown through the opposition of (12a) and (12b). (12) a. Mary died from cancer. b. *Mary died from John.

. J. Emonds (personal communication) points out that, though in general anticausatives with from-phrases are fine, often the acceptability seems a bit weak. Also, for the “weak” cases, there is undeniably some variation with respect to judgements: some speakers find even the “weak” cases fine, others don’t like them. Crucially, however, even for those speakers who have trouble accepting all anticausatives with from-phrases, there is a clear distinction between anticausatives with by-phrases – these are simply ungrammatical – and anticausatives with from-phrases – which relative to anticausatives with by-phrases are judged as much better.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.9 (535-593)

Argument demotion as feature suppression

The fact that also a non-alternating unaccusative like ‘die’ can combine with a from-phrase identifying the external cause of the event appears to constitute a counterexample to Levin and Rappaport Hovav’s (1995) claim that only alternating unaccusatives have underlying causative semantics, while it may be construed as evidence for Chierchia’s (1989, 2004) position. However, things are not as black and white. Though from-phrases identifying causers are licensed not only with anticausatives but also with other unaccusatives (that is, unaccusatives that do not participate in the transitive-inchoative alternation), they are not licensed with all unaccusatives. For instance, as (13) shows, unless the from-phrase is interpreted as a location, both a from-phrase and a by-phrase are bad with the unaccusative verb arrive.13 (13) *The refugees arrived from the invasion.

The fact that not all unaccusatives license from-phrases suggests that not all unaccusatives have underlying causative semantics, in line with Levin and Rappaport Hovav (1995) and contrary to Chierchia (1989, 2004). Finally, from-phrases are disallowed in passives. Moreover, this does not seem to be related to whether the causer introduced by the preposition is a participant capable of willful agency or a natural cause. Both (14a) in which the cause is a natural one and (14b) in which the cause is a human are ungrammatical. (14) a. *Mary was killed from cancer. b. *Mary was killed from John.

Thus, a look at the distribution of from-phrases identifying the causer of a certain event reveals the generalisation in (15). (15) Generalisation From-phrases identifying the external cause in anticausatives cannot combine with participants capable of willful agency (that is, with animate beings).

The generalisation in (15) is important because it emphasises the fact that the distribution of by- and from-phrases in English cannot be captured merely through a distinction between passives and anticausatives. Turning now to Albanian, probably the most striking property concerning passives is that, strictly speaking, it is not possible to distinguish between passives and non-active anticausatives. But first let me explain what the term “non-active anticausative” means. In Albanian, certain verbs participating in the transitiveinchoative alternation may appear either in active or in non-active form, as in

. The example in (13) is due to Peter Svenonius (personal communication).



JB[v.20020404] Prn:6/11/2006; 9:28



F: LA9606.tex / p.10 (593-642)

Dalina Kallulli

(16). However, in spite of distinct voice morphology on the verb, the sentences in (16a) and (16b) entail each other (that is, they are fully synonymous).14 (16) a.

Dritar-ja u kris nga presioni. window-the NACT crack.AOR,3S from/by pressure.the ‘The window cracked from the pressure’ b. Dritar-ja krisi nga presion-i. window-the crack.ACT,AOR,3S from/by pressure-the ‘The window cracked from the pressure’

Passives and non-active anticausatives in Albanian are impossible to distinguish formally for two reasons. First, as just stated, unlike in English, passives and anticausatives in Albanian (like Greek) are often homomorphic (i.e. they bear the same morphology, namely non-active). Second, Albanian (like Greek) collapses by- and from-phrases.15 Note in this context that while English disallows byphrases in middles (Baker, Johnson, & Roberts 1989), in Albanian middles are fine with a by/from-phrase, a fact also known for Greek (Condoravdi 1989; Tsimpli 1989; Lekakou 2005, this volume).16 Indeed, any formally non-active predicate can combine with a by/from-phrase in Albanian. As the preceding discussion would lead us to expect, the licensing of byphrases, which is taken to be one of the most salient properties of the passive in English (Baker, Johnson and Roberts (1989) and references therein) and one that distinguishes passives from anticausatives, cannot be applied to Albanian. To illustrate, the Albanian counterparts of the sentences in (7) and (8) are given in (17a) and (17b), respectively. As expected then, the grammaticality contrast between the English (7) and (8) is not replicated in Albanian. (17) a.

Dritar-ja u kris nga window-the NACT crack.AOR,3S from/by ‘The window cracked from the pressure’ b. Dritar-ja u kris nga window-the NACT crack.AOR,3S from/by ‘The window was cracked by John’

presion-i. pressure-the Xhon-i. John-the

. The pattern described in (16a) and (16b) is also found in Greek, as described in Alexiadou and Anagnostopoulou (2004). According to these authors, unlike in Albanian, in Greek there are meaning differences between the active and the non-active anticausative. . More specifically, while in Greek the same word is used both for by and from, Albanian has two distinct prepositions both meaning by- and from. Consequently, by- and from-phrases are indistinguishable. . See also Stroik (this volume) on for-phrases in middles.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.11 (642-704)


Interestingly, the formally active anticausative (recall the distinction between (16a) and (16b)) can still combine with a by/from-phrase only when the causer is a participant that is not capable of willful agency, such as an event nominal. This is illustrated through the opposition in (18a) vs. (18b). (18) a.

Dritar-ja krisi nga window-the crack.ACT,AOR,3S from/by ‘The window cracked from the pressure’ b. *Dritar-ja krisi nga window-the crack.ACT,AOR,3S from/by ‘The window cracked from/by John’

presion-i. pressure-the Xhon-i. John-the

So then, the contrast between (18a) and (18b) is a replication of the contrast between the English sentences (7) and (8): like in English, the non-specially marked anticausative can combine with a from(/by)-phrase only when the latter introduces a causer participant that is not capable of wilful agency. This fact combined with the fact that, unlike English, Albanian collapses by- and from-phrases lends support to the generalisation in (15). Taken together, the arguments presented in this section, and in particular the fact that the distribution of by- and from-phrases in English cannot be captured by appealing merely to the distinction between unaccusatives (whether anticausative or other) and passives, as well as the fact that there are languages that altogether collapse the distinction between by- and from-phrases, suggest that by- and fromphrases might be more closely related than has been assumed in discussions that focus on the licensing of by-phrases. It is clear that once we draw into the picture languages that do not make the distinction between by- and from-phrases, the ability to license a by-phrase irrespective of the ability to license a from-phrase cannot be granted such a theoretical status as it has in studies that focus on the English verbal passive. Interestingly, as Clark and Carpenter (1989) note, children commonly use from-phrases instead of by-phrases in English passives, too. I contend that by- and from-phrases do not differ as to their ability to identify arguments (either implicit or syntactically expressed, depending on the theory), but rather with respect to other features that distinguish the passive and the anticausative formations. Specifically, as stated in Section 2, here I will argue that the passive/anticausative distinction hinges on the nature of the feature in v encoding the ontological event type of the verb. But before doing this, I will address another argument that has often been evoked in discussions on the implicit or syntactically encoded (depending on the theory) argument in passives.



JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.12 (704-758)

 Dalina Kallulli

. The issue of purpose clauses and agent-oriented adverbs Another set of facts frequently brought forward in discussions on the presence of an understood logical-subject in passives versus their absence in middles or unaccusatives involves purpose clauses and so-called subject- and/or agent-oriented adverbs. For instance, citing Manzini (1983) and Roeper (1987), Baker, Johnson and Roberts (1989) claim that in passives but not in middles the understood subject of control clauses (or rationale clauses, in their terminology) may be “controlled”, and subject-oriented adverbs may find an argument to modify, as in (19) and (20). (19) a. This bureaucrat was bribed [PRO to avoid the draft]. b. *This bureaucrat bribes easily to avoid the draft. (20) a. This bureaucrat was bribed deliberately. b. *This bureaucrat bribes deliberately.

Baker et al. (1989) take this as evidence for their view that the passive morpheme -en in English is an argument, since purpose clauses and subject-oriented adverbs require the syntactic presence of an argument. It is well-known that, unlike in passives, in anticausatives purpose clauses are illicit, as is shown in (21). (21) a. The boat was sunk to collect the insurance. b. *The boat sank to collect the insurance.

(Roeper 1987: 268)

Levin and Rappaport Hovav (1995) take this fact as evidence for their view that the binding of the external cause in anticausatives takes place in the mapping from the lexical semantic representation to argument structure and since the binding of a position in the lexical semantic representation prevents the projection of that position to argument structure, no reference can be made to the bound argument. Likewise, unlike in passives, in anticausatives agent-oriented adverbs are illicit, as is shown in (22). (22) a. The ship was sunk deliberately. b. *The ship sank deliberately.

Two issues arise here. First, all that purpose clauses and so-called agent-oriented adverbs do is identify an intention-bearing (i.e. animate) event participant as the source or initiation of the event described by the verb. However, in no way does this entail that animate or intention-bearing participants have to act intentionally.17 Therefore the term ‘agent-oriented’ is not a good fit. For this reason, I will refer to such adverbs more neutrally as adverbs of animacy. . This point is relevant for the analysis developed in Section 4, especially Subsection 4.5.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.13 (758-814)


Crucially, it can be shown that so-called ‘agent-oriented’ adverbs are not incompatible with unaccusative syntax, a point also made in Harley and Folli (2004). Harley and Folli (2004) give the Italian examples in (23), which demonstrate that both the unaccusative verb cadere ‘fall’, and rotolare ‘roll’, continue to exhibit the characteristic essere ‘be’ (vs. avere ‘have’) selection, typical of unaccusatives, even when the subject is clearly performing the action on purpose. (23) a.

Gianni é caduto/*ha caduto apposta. John is fallen / has fallen on purpose. b. Gianni é rotolato/*ha rotolato giu apposta. John is rolled/has rolled down on purpose.18

(Harley & Folli 2004: 47)

Similarly, the example in (24) describes the same fact for German (note the selection of the auxiliary sein ‘be’ and not haben ‘have’). (24) Peter ist absichtlich eingeschlafen. Peter is deliberately fallen asleep ‘Peter fell asleep on purpose’

Likewise, while it is uncontroversial that the logical-subject of the matrix clause serves as an antecedent for the PRO subject of the purpose clause, this antecedent does not need to be an agent as is shown by the German examples in (25), in which the matrix verbs are unaccusative, as witnessed by the fact that they take the auxiliary sein ‘be’ and not haben ‘have’. (25) a.

Die Pflanzen wachsen nach oben, um ans Licht zu kommen. the plants grow upwards in order to the light to come ‘Plants grow upwards in order to reach the light’ b. Eva ist gekommen um mir zu helfen. Eva is come in order me to help ‘Eva came to help me’

In sum, purpose clauses and/or so-called agent-oriented adverbs only make reference to, or imply an animate participant. I contend, however, that the animate participant in a passive construction is not introduced by a non-oblique argument (implicit or syntactically present, depending on the theory), but by a by-phrase and, this may in turn be either overt or implicit. If, as established in the previous section, animate causers are disallowed with from-phrases and anticausatives only license from-phrases but not by-phrases, then the inability of anticausatives to . Harley and Folli (2004) also note that although rotolare is better with ha than cadere is, this is due to the fact that rotolare is optionally transitive, so the ha rotolato sequence, while ungrammatical in the structure in (23), is familiar from transitive constructions; it’s a type of garden-path effect.



JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.14 (814-866)

 Dalina Kallulli

combine with purpose clauses and adverbs of animacy follows straightforwardly without further stipulations. Further empirical evidence for this view involves the fact that whenever a purpose clause is licit, a by-phrase identifying an actor can be inserted overtly.

. A formal analysis of argument demotion in unaccusative contexts The analysis that I develop here builds on my previous work, especially Kallulli (2006a, b), which I summarize and extend. Two main ideas are introduced. The first concerns the distinction between agentive and non-agentive predications, and the second the definition of non-active voice. Various extensions follow. . The structure of causative predications Davis and Demirdache (1995) and Demirdache (1997) argue that agentive and causative predications are universally derived from distinct frames. The basic idea behind their analysis is that an event participant identifying the instigation (or initiation) of a causative event is an agent if and only if that participant can volitionally or intentionally bring about such initiation, that is, if the causing participant has control over the event. To illustrate, Demirdache provides the example in (26) and argues that Rosa in (26) is an agent iff “Rosa performs some action of melting which causes the ice to be melted. In contrast, Rosa is a causer (but not an agent) when there is no intrinsic relation between the causing event and the resulting change of state – e.g. Rosa accidentally turns off the refrigerator and the ice melts” (Demirdache 1997: 129). (26) Rosa melted the ice.

While what is meant by “intrinsic relation” may not be equally intuitive for everyone, it is clear what Demirdache means: Rosa is an agent iff she intentionally does something in order to obtain a certain result, namely have the ice melt.19 Intentionality (or lack of it) can be successfully captured through the feature intent. I contend that there are two types of causatives, agentive and non-agentive, which differ in their lexical feature composition make-up. Specifically, while agentive causatives can be defined as consisting of an ordered tuple in v containing the features [+intent] and [+cause], as in (27), non-agentive causatives lack the feature [+intent], as in (28). Accordingly, due to the tuple , an . The fact that the instigator of an event is not necessarily an agent (or an event participant capable of agency) does not make its relation to the (sub-)event less intrinsic.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.15 (866-894)


agent will be projected in the specifier position of vP, as in (27). In contrast, when no such feature is contained in the tuple in v, what will be projected in the higher specifier position will not be an agent argument, but a causer, as shown in (28). (27) The structure of agentive causatives vP v’

Spec:Agent

VP

break Spec

V’ V

Compl

(28) The structure of non-agentive causatives vP Spec:Causer

v’

break

VP

Spec

V’ V

Compl

. The structure of activity/process predications In Kallulli (2006a, b) I have argued that, like causatives, activity predications fall into two different types: agentive and non-agentive. Agentive activities differ from non-agentive activities in terms of their feature composition: agentive activity predicates contain an ordered tuple consisting of the features [+intent] and [+act] in v, as in (29), whereas non-agentive activities contain an ordered tuple consisting of the feature [+act] only in v, as in (30). Parallelling the discussion in the previous section, the tuple makes an agent in Spec of vP, as in (29). In contrast, the tuple makes an actor, not an agent, as in (30).20 . As an anonymous reviewer points out, although the notions actor and agent are switched around here, basically the same point is made in Doron (2003), who argues that the function



JB[v.20020404] Prn:6/11/2006; 9:28



F: LA9606.tex / p.16 (894-932)

Dalina Kallulli

(29) The structure of agentive activities vP v’

Spec:Agent

build

VP

Spec

V’ V

Compl

(30) The structure of non-agentive activities vP Spec:Actor Anna

v’ VP

Spec

V’ V

Compl

In other words, I claim that a sentence containing an activity predicate as in (31) is ambiguous between an agentive and a non-agentive reading. (31) Rosa screamed.

Specifically, Rosa in (31) is an agent if and only if she intends her action (e.g. she could stop screaming if she so willed). In contrast, Rosa in (31) is an actor but not an agent if she does not intend her screaming activity (for instance, if she has taken drugs that make her scream), whether or not she is aware of it. . Defining non-active voice Much research has maintained that certain morphological operations apply either in the lexicon, or in the syntax. To wit, passivization, and/or reflexivization have commonly been treated as operations that suppress either an argument position (external or internal), a theta role in the thematic grid of the verb, or some of intensification in Semitic languages is to transform an agent, who might act intentionally or not, into an actor, who necessarily acts intentionally.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.17 (932-972)

Argument demotion as feature suppression 

element in the lexical-semantic structure of a predicate (depending on the theory) (Grimshaw 1990; Woolford 1993; Levin & Rappaport Hovav 1995; Reinhart & Siloni 2004; among others). Similarly, non-active and/or reflexive morphology has been treated as an operation that suppresses either the initial subevent of a predicate or its name (Kallulli 1999, 2004). In this spirit, I will analyse non-active voice as a suppression operation here as well. However, unlike the types of suppression just cited, I believe that the non-active voice operates in the syntax solely and purely in a linear fashion fully ignoring the content of the element that it affects. Specifically, I define non-active voice as in (32).21 (32) Definition of non-active voice Non-active voice suppresses the first feature in a predicate structure.

. Deriving the unintended causation reading I contend that the unintended causation reading of the dyadic unaccusatives in (1) and (4a) is derived from (dyadic) agentive causative predications, the structure of which was depicted in (27). If the definition in (32) is applied to the structure in (27), the outcome is the representation in (33), since the first feature in (27) is [+intent]. (33)

vP Spec:[-Agent]

v’

v0

VP

Spec: Affected V break

V’ Compl

Due to the suppression of [+intent], no agent will be projected in Spec of vP. The feature [+cause] on the other hand cannot assign a theta-role since the tuple is not intact, i.e., its integrity has been affected due to suppression of a feature.22 On the other hand, for the derivation to converge, the feature [+cause] has to . On this view, in the presence of the morphological gaps discussed in Footnote 1, one could even argue that Albanian (or Greek) are becoming more like English as far as the morphological marking of unaccusatives go (see also (16a) vs. (16b)). . The core idea here is that theta-role assignments are tuples; see also Kallulli (in press).

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.18 (972-984)

 Dalina Kallulli

be saturated. The only way for this feature to be licensed is by another argument moving to the specifier of vP. I claim that the dative argument merged in the Spec of VP fulfils this role. Let us assume that the feature that licenses the projection of the dative argument in Spec, VP is a [+affected] feature of the verb, which is why the dative here will be interpreted as an affected participant. When non-active morphology suppresses the feature [+intent], the dative argument moves from Spec of VP to Spec of vP so that the [+cause] feature is licensed. Consequently, once in Spec of vP, the theta-role of the dative will be something like an affected causer, which is what an unintentional causer is. Precisely because of this syntactic (and semantic) composition, the dative argument in an unaccusative construction will always be ambiguous between an affected and a causer interpretation, unless pragmatic considerations (dis)favour one of these readings.23 . Deriving the anticausative and the passive In Kallulli (2006a, b) I have argued that the anticausative and the passive are derived when non-active morphology applies to non-agentive causative and nonagentive activity predications, defined as in (28) and (30), repeated for ease of reference under (34) and (35), respectively. (34) The structure of non-agentive causative predications vP v’

Spec: Actor build

VP

Spec

V’ V

Compl

. Note that the implication here is only one way: datives do not need to be interepreted as unintentional causers. They are only interpreted as such once they are in Spec of vP.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.19 (984-1045)


(35) The structure of non-agentive activity predications vP v’

Spec: Actor

VP

build Spec

V’ V

Compl

Non-active voice was defined in (32) as an operation that suppresses the first feature in a predicate structure. Note that in the structures in (34) and (35) the first feature is [+cause], and [+act], respectively, so when non-active morphology operates on these structures, it will suppress this feature. Consequently, no Causer or Actor argument can be merged in Spec of vP. In other words, the operation of non-active morphology on the structures in (34) and (35) yields basically monadic structures, as shown in (36) and (37). (36)

vP v’ break

VP

Spec

V’ Compl

V

(37)

vP v’ build

VP

Spec

V’ V

Compl



JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.20 (1045-1085)

 Dalina Kallulli

I claim that the structure in (36) is that of the anticausative, which is derived when non-active morphology operates on the structure of a non-agentive causative predication. On the other hand, the structure in (37) is that of the passive, which is derived when non-active morphology operates on the structure of a non-agentive activity predication.24 In other words, I contend that the passive and the anticausative formations differ only in that the former is derived from an activity predication, i.e. a structure that contains a [+act] feature in v (which is suppressed by non-active and/or other passivizing morphology), and the latter from a causative predication i.e., a structure that has a [+cause] feature in v (which also gets suppressed by intransitivizing morphology). I suggest that the difference between by- and from-phrases in English and other languages that make this distinction is simply a reflection of the ontological difference between activity and causative predications. That is, the by-phrase in the passive in English does not make reference to an implicit non-oblique argument, but to the [+act] feature in v that is suppressed by passive morphology (i.e., non-active voice) in the syntax. Likewise, a from-phrase identifying the external cause of an event in the anticausative in English makes reference to the syntactically suppressed [+cause] feature in v. Note that, unlike [+cause], the feature [+act] entails an actor, that is, animacy. Since in the passive this feature is suppressed by passive morphology in the syntax, no actor can be merged in the Spec of vP. However, the actor can be realized obliquely in a (by-) phrase, as can the causer in the anticausative (through a from-phrase). In other words, a by-phrase simply makes reference to the syntactically suppressed actor and a from-phrase to the syntactically suppressed causer. The question arises, if a by-phrase simply makes reference to the feature [+act] (i.e. to the syntactically suppressed actor), how can we account for the fact that external causation verbs passivize? That is, how can a sentence like (7) – repeated here under (38) – be derived? The sentence in (38) is a passive construction containing the verb ‘break’, i.e. a verb that is typically associated with causative semantics and not classified as an activity-denoting verb (Levin & Rappaport Hovav 1995; references therein). (38) The window was broken by Pat.

. Like in Greek (Lekakou 2005, this volume), middles in Albanian are also non-active morphologically. Here I assume that the distinction between passive/anticausative on the one hand and middle on the other is due to the difference between a dispositional (imperfective) operator and/or other aspectual morphology. Specifically, the middle construction is derived when the verb in the structures in (36) and (37) is under the scope of a dispositional operator (Lekakou 2005, this volume). In contrast, passive/anticausative obtains when the verb in (36) and (37) is not under the scope of such an operator.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.21 (1085-1128)

Argument demotion as feature suppression 

A tentative answer builds on an intuition from Emonds (2000), namely that due to the fact that English lacks a verbally finite synthetic passive, both verbal and adjectival passives are in a sense “more adjectival”, than say in Latin, or even Albanian (or Greek), which have at least a partially verbal finite synthetic passive. This intuition is fundamentally an aspectual one, it seems to me. Indeed, anticausatives are more eventive than passives in English, a point which cannot be made with respect to the Albanian passive, which as discussed in Section 3 simply collapses the morphological distinction between passives and non-active anticausatives. So then, one could stipulate that the passive in English in a sentence like (38) implies that the breaking of the window by Pat was more sustained or involved an activity (though not necessarily volitional) on her part, as compared to the breaking event in an anticausative, which happens spontaneously or all-at-once. That is, the English passive, whether or not due to its special (adjectival) morphology, induces an implicature of activity even for external causation verbs, which is all too obvious through a comparison with an anticausative like the one in (39).25 (39) The window broke.

Above, I stated that the feature [+act] entails an actor, that is, animacy. The question then arises how to account for sentences such as (40) where a natural force, namely the earthquake combines with the preposition by. I think it is not unreasonable to assume that nominals denoting natural and/or other forces in sentences like (40) are personifications.26 (40) The window was broken by the earthquake.

The analysis that I have outlined here entails among other things that neither the passive, nor the anticausative are derived from agentive predications (discussed . The idea here would be that the special morphology of the English passive is a kind of event type-shifting device, one which turns a causative event into an open-ended event type (such as, activity). Alternatively, as one anonymous reviewer points out, the adjectivalness and longer duration (non-punctuality) of the English passive may “alternatively come from its possible reading of resultativity; i.e. the lasting result state, not the event being more sustained in the passive than in its active counterpart”. Importantly, however, both activities and states are openended events. . As one anonymous reviewer points out, it would probably be enough to describe the earthquake as well as other acceptable natural forces as ‘animate’ rather than claiming personification. These forces are seen as animate as opposed to inanimate forces that can cause breakage such as a “construction fault”, which as the reviewer presumes, is indeed ungrammatical in a by-phrase. The reviewer also suggests that it would be interesting to see judgments on a causer like “pressure building up over time” in a by-phrase and predicts that they might vary, as this could be seen as a very slow but nevertheless animate force, or else as a more stationary force, in which case it would probably be unacceptable in the by-phrase. I leave these issues open for future research.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.22 (1128-1165)

 Dalina Kallulli

in Section 4.1 and 4.2). However, what we refer to as “passive” can still combine with purpose clauses and adverbs of intentionality because as already mentioned in Section 3, purpose clauses and adverbs of intentionality simply make reference to participants capable of intentionality. However, it would be illogical to conclude on this basis that the animate participant controlling the purpose clause or adverb of intentionality in a passive construction is a non-oblique argument either implicit (i.e. not expressed syntactically, as espoused by Grimshaw 1990) or syntactically present (Baker, Johnson, & Roberts 1989). Instead, as I already suggested in Section 3.2, there is the alternative that the animate participant in a passive construction is not introduced by a non-oblique argument (implicit or syntactically present, depending on the theory), but simply by a by-phrase, which may in turn be either overt or implicit.

. Conclusion In this article I have examined a range of phenomena that may be generally considered in terms of argument demotion and offered a uniform and formal account for them. I have provided new evidence that anticausatives have causative semantics. I have also discussed a variety of to my knowledge new empirical arguments, which show that the picture depicted for the passive in English is quite idiosyncratic, and that the properties that have attained the status of identificational criteria of the passive are simply not revealing or even maintainable when looking at other languages. In particular, unlike generally assumed, neither by-phrases nor purpose clauses or agent-oriented adverbs witness the presence of a non-oblique argument (either implicit or syntactically encoded, depending on the theory). In contrast, the analysis that I have laid out here derives the properties of the passive and anticausative both in Albanian and English uniformly. The main conclusion here is that universally anticausatives and passives differ only with respect to the ontological event type feature (in v) of the verb, which can be affected by morphological operations in the syntax. The distinction between by- and from-phrases in English is a simple reflection of this feature: a by-phrase introduces an oblique actor upon suppression of [+act] and a from-phrase an oblique causer. As an anonymous reviewer points out, this conclusion may be relevant for Finnish, too (Nelson 1999). I have shown that the English verbal passive can be made more compatible with its Albanian cousin by relegating the differences with respect to the licensing of by-phrases to simple combinatorial properties of verb and prepositional types.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.23 (1165-1272)

Argument demotion as feature suppression 

References Alexiadou, Artemis, Elena Anagnostopoulou, & Martin Everaert (Eds.). (2004). The Unaccusativity Puzzle. Oxford: Oxford University Press. Alexiadou, Artemis, & Elena Anagnostopoulou (2004). Voice morphology in the causativeinchoative alternation: Evidence for a non-unified structural analysis of unaccusatives. In Alexiadou et al. (2004), 114–136. Baker, Mark, Kyle Johnson, & Ian Roberts (1989). Passive arguments raised. Linguistic Inquiry, 20, 219–252. Borer, Hagit (1994). The projection of arguments. In Elena Benedicto & Jeffrey Runner (Eds.), Functional Projections (pp. 19–47). University of Massachusetts Ocassional Papers 17. Borer, Hagit (2004). The grammar machine. In Alexiadou et al. (2004), 288–331. Chierchia, Gennaro (1989). A semantics for unaccusatives and its syntactic consequences. Ms., Cornell University, Ithaca, New York. Chierchia, Gennaro (2004). A semantics for unaccusatives and its syntactic consequences. In Alexiadou et al. (2004), 22–59. Clark, Eve V., & Kathie L. Carpenter (1989). On children’s uses of from, by, and with in oblique noun phrases. Journal of Child Language, 16, 349–364. Condoravdi, Cleo (1989). The middle: Where semantics and morphology meet. MIT Working Papers in Linguistics, 11, 18–30. Davis, Henry, & Hamida Demirdache (1995). Agents and Events. Paper presented at the GLOW Colloquium, University of Tromsø, Norway. Demirdache, Hamida (1997). Out of control in St’at’imcets Salish and event (de)composition. In A. Mendikoetxea & M. Uribe-Etxebarria (Eds.), Theoretical Issues on the Morphology Syntax Interface (pp. 97–143). Supplement of the International Journal of Basque Linguistics and Philology. Doron, Edit (2003). Agency and voice: The semantics of the semitic templates. Natural Language Semantics, 11(1), 1–67. Emonds, Joseph (2000). Lexicon and Grammar: The English Syntacticon. Berlin: Mouton de Gruyter. Everaert, Martin (1986). The Syntax of Reflexivization. Dordrecht: Foris. Grimshaw, Jane (1982). On the lexical representation of Romance reflexive clitics. In Joan Bresnan (Ed.), The Mental Representation of Grammatical Relations (pp. 87–148). Cambridge, MA: The MIT Press. Grimshaw, Jane (1990). Argument Structure. Cambridge, MA: The MIT Press. Harley, Heidi, & Raffaella Folli (2004). Waltzing Matilda around and around: On the licensing of directed-motion resultatives. Ms., University of Cambridge and University of Arizona. Haspelmath, Martin (1993). More on the typology of inchoative/causative verb alternations. In Bernard Comrie & Maria Polinsky (Eds.), Causatives and Transitivity (pp. 87–120). Amsterdam: John Benjamins. van Hout, Angeliek (1996). Event Semantics of Verb Frame Alternations. A Case Study of Dutch and its Acquisition. Ph.D. dissertation, Tilburg University. van Hout, Angeliek (2004). Unaccusativity as telicity checking. In Alexiadou et al. (2004), 60–83. Kallulli, Dalina (1999). Non-active morphology in Albanian and event (de)composition. In István Kenesei (Ed.), Crossing Boundaries (pp. 263–292). Amsterdam: John Benjamins.

JB[v.20020404] Prn:6/11/2006; 9:28

F: LA9606.tex / p.24 (1272-1360)

 Dalina Kallulli

Kallulli, Dalina (2004). De-agentivised causers or non-active causative predications. In Torgrim Solstad, Benjamin Lyngfelt, & Maria F. Krave (Eds.), Pre-Conference Proceedings of the Workshop Demoting the Agent: Passive and Other Voice-Related Phenomena (pp. 59–66). Oslo: University of Oslo. Kallulli, Dalina (2006a). Unaccusatives with overt causers and experiencers: A unified account. In Daniel Hole, André Meinunger, & Werner Abraham (Eds.), Datives and Other Cases: Between Argument Structure and Event Structure (pp. 271–301). Amsterdam: John Benjamins. Kallulli, Dalina (2006b). Unaccusative morphology and argument realization. Ms., University of Vienna. Kallulli, Dalina (In press). The syntactic visibility of intentionality: Evidence from dyadic unaccusatives. Proceedings of WECOL 2004. University of Southern California. Lekakou, Marika (2005). In the Middle, Somewhat Elevated. The Semantics of Middles and its Crosslinguistic Realization. Ph.D. dissertation, University College London. Levin, Beth, & Malka Rappaport Hovav (1995). Unaccusativity. Cambridge, MA: The MIT Press. Manzini, Maria Rita (1983). On control and control theory. Linguistic Inquiry, 14, 421–46. Marantz, Alec (1984). On the Nature of Grammatical Relations. Cambrigde, MA: The MIT Press. Nedyalkov, Vladimir P., & G. G. Silnitsky (1973). The typology of morphological and lexical causatives. In Ferenc Kiefer (Ed.), Trends in Soviet Theoretical Linguistics (pp. 1–32). Dordrecht: Reidel. Nelson, Diane (1999). Events, arguments, and causative psych predicates in Finnish. In Paul Foulkes (Ed.), Leeds Working Papers in Linguistics and Phonetics, Vol. 7. URL: http://www.leeds.ac.uk/linguistics/WPL/WPL7.html Partee, Barbara (1965). Subject and Object in Modern English. Ph.D. dissertation, MIT. Reinhart, Tanya (1996). Syntactic Effects of Lexical Operations: Reflexives and Unaccusatives. OTS Working Papers in Linguistics. Reinhart, Tanya, & Tal Siloni (2004). Against an unaccusative analysis of reflexives. In Alexiadou et al. (2004), 159–180. Roeper, Tom (1987). Implicit arguments and the head-complement relation. Linguistic Inquiry, 18, 267–310. Rosen, Carol (1984). The interface between semantic roles and initial grammatical relations. In D. M. Perlmutter & C. Rosen (Eds.), Studies in Relational Grammar Vol. 2 (pp. 38–77). Chicago, IL: University of Chicago Press. Tsimpli, Ianthi (1989). On the properties of the passive in Modern Greek. UCL Working Papers in Linguistics, 1, 235–260. Woolford, Ellen (1993). Symmetric and asymmetric passives. Natural Language and Linguistic Theory, 11, 679–728.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.1 (56-166)

A comparative view of the requirement for adverbial modification in middles* Marika Lekakou In this paper I address the question why Germanic middles require adverbial modification. I argue that a semantic/pragmatic account of the phenomenon is not on the right track. Part of the argumentation is based on Greek and French middles, which can do without adverbial modification. The data point toward the direction of some version of the ‘structural’ account, according to which the adverb helps recover the implicit Agent of middles, in those languages where the latter is not syntactically represented. I maintain that an approach along these lines can explain the adverb facts and also make predictions about the set of appropriate middle-modifiers. It also fits in well with the more general pattern of cross-linguistic variation that characterizes the realization of the middle semantics.

.

Introduction

It has long been noted for middles in English – and the same applies to Dutch and German – that they require an adverb like easily or well (Keyser & Roeper 1984; Fellbaum 1986; Pitz 1988; Condoravdi 1989). In the absence of modification, middles in these languages are unacceptable or ungrammatical (the precise nature of the deviance is at the heart of the debate): (1) a. Bureaucrats bribe *(easily). b. Pine saws *(easily). c. This book reads *(easily).

* This paper is based on a section included in the third chapter of my PhD thesis (Lekakou 2005a). I would like to thank the following people for helpful discussions and comments: Peter Ackema, David Adger, Cleo Condoravdi, Sabine Iatridou, Ad Neeleman, Øystein Nilsen, Ianthi Tsimpli, as well as the editors of this volume and two anonymous reviewers. All errors are mine.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.2 (166-222)

 Marika Lekakou

(2) Dutch a. Zo’n stuk zingt *(gemakkelijk). such.a piece sings easily b. Dit boek leest *(gemakkelijk). this book reads easily (3) German a. Das the b. Der the c. Der the

(Ackema & Schoorlemmer 2005)

Buch liest sich *(leicht). book reads refl easily Stoff wäscht sich *(leicht). fabric washes refl easily Wagen fährt sich *(leicht). car drives refl easily

(Fagan 1992)

At the same time, it has been claimed by Roberts (1987), Condoravdi (1989) and Ackema & Schoorlemmer (1995, 2005) that such sentences can be rescued if there is a modal, focus or negation present. The data and judgments below are taken from Ackema & Schoorlemmer (2005). At least some of these data will be contested, since native speakers do not corroborate the judgments reported in the literature. (4) a. Bureaucrats may bribe, but you never know. b. Pine SAWS/DOES saw after all! c. This book doesn’t read. (5) Dutch a. Dit vlees snijdt niet. this meat cuts not ‘This meat won’t cut.’ b. Die aardappels ROOIEN, niet te geloven! those potatoes dip.up, not to believe ‘I can’t believe how easy to dig up those potatoes are.’

Finally, there are a few cases of adverbless middles in English, Dutch and German reported in the literature which do not require stress, negation or a modal to be acceptable, contrary to examples like (4) and (5) above: (6) This dress buttons. (7) Dutch Deze jurk ritst dicht. this dress zips shut ‘This dress zips up.’

(McConnell-Ginet 1994)

(Ackema & Schoorlemmer 2005)

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.3 (222-278)

Adverbial modification in middles 

(8) German Dieses Kleid knöpft sich zu. this dress buttons refl part ‘This dress buttons.’

(Fagan 1992)

Overall, it is fair to say that Germanic languages impose a requirement that their middles feature an adverb like easily (cf. Abraham 1995). This requirement has been argued to be semantic (Condoravdi 1989; McConnell-Ginet 1994), pragmatic (Ackema & Schoorlemmer 1995; Ackema & Schoorlemmer 2005; Goldberg & Ackerman 2001; Steinbach 2002; Marelj 2004), or ‘structural’/syntactic (Roberts 1987; Pitz 1988; Hoekstra & Roberts 1993). According to the semantic approach, the adverb is required in order to provide the scope for the generic operator that is present in the semantic representation of middles. On this view, adverbless middles are semantically ill-formed. On the pragmatic account, the adverb provides the core of the assertion, and middles without it are uninformative, and hence pragmatically deviant. Finally, the structural account has it that the adverb is required in order for the implicit Agent to be recoverable. In the absence of adverbial modification, the sentence is ungrammatical. In this paper I provide arguments against the semantic/pragmatic account, and in favour of the ‘structural’ account. Part of the argumentation relies on the intuitions that speakers of English, Dutch and German have vis-à-vis adverbless middles, which they deem ungrammatical. Another argument comes from languages like Greek and French. In these languages, adverbless middles are not deviant at all. Moreover, the omissibility of adverbial modification does not seem to hinge on the availability of emphatic stress, or negation, or a modal. (9) French (Fagan 1992) a. Ce papier se lave. this papier se wash-3sg ‘This paper is washable.’ b. Le papier se recycle. the paper se recycle-3sg ‘Paper is recyclable.’ c. Cette racine se mange. this root se eat-3sg ‘This root is edible.’ d. Cette solution se discute. this solution se discuss-3sg ‘This solution is debatable.’

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.4 (278-341)

 Marika Lekakou

(10) Greek a. To nero edo pinete. the water-nom here drink-3sg.nonact.imperf ‘The water here can be drunk.’ b. To yiali anakiklonete. the glass-nom recycle-3sg.nonact.imperf ‘Glass is recyclable.’ c. Afta ta manitaria trogonde. these the mushrooms eat3pl.nonact.imperf ‘These mushrooms are edible.’ d. Afto sizitiete. this discuss-3sg.nonact.imperf ‘This can be discussed.’

Although this difference between English-type and Greek-type middles has not gone unnoticed – both Fagan (1992) and Steinbach (2002) have pointed it out, and see also Ackema & Schoorlemmer (2005) – we have as yet no explanation for it. I would like to argue that this difference between Greek- and English-type middles can be understood, as long as we relate to it the more general pattern of crosslinguistic variation that middles attest. Moreover, I maintain that the difference between Greek/French middles and their Germanic counterparts that concerns the adverb provides an important clue as to the nature of its requirement in Germanic. To recapitulate, we have two sets of questions to answer. First, why is adverbial modification in middles obligatory in some languages and optional in others? What is the nature of the requirement? And why is it stronger in some cases than in others even within the same more ‘stringent’ languages? The second question concerns the sort of adverbs that are admissible as middle-modifiers. These two issues are in principle separate, although the answer to why adverbial modification is required in Germanic middles informs the issue of what sort of modifiers are appropriate. In the following section I discuss the middle semantics and the factors that determine its realization across languages. In Section 3 I present the semantic/pragmatic account and the problems it faces. In Section 4 I provide the arguments that, in conjunction with the findings of Section 3, compel us to conclude that a structural account of the adverb requirement is to be preferred. Finally, in Section 5 I show that the approach to the adverb advocated here can make correct predictions regarding the set of appropriate middle modifiers. Section 6 concludes.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.5 (341-386)

Adverbial modification in middles

. Variation in middles . Some of the syntactic facts Middles across languages behave differently with respect to a number of syntactic properties; the requirement for adverbial modification is only one of the differences that set Greek middles apart from their English counterparts. However, the differences cluster together, to the effect that there exists a clear pattern of variation, which allows us to refer to two ‘kinds’ of middle: the Greek-type middle and the English-type middle (Lekakou 2002, 2003, 2004, 2005a; Ackema & Schoorlemmer 2005; Marelj 2004). In a series of articles, Ackema & Schoorlemmer (1994, 1995, 2005) have argued that English and Dutch middles are syntactically unergative and lack a syntactically represented implicit Agent (contra Stroik 1992; Hoekstra & Roberts 1993). The same conclusion is drawn by Cabredo-Hofherr (1997) for German middles. Ackema and Schoorlemmer (see also Zribi-Hertz 1993) reject the arguments advanced by Stroik (1992) in favour of the syntactic activity of the middle Agent, and provide compelling arguments in support of the unergativity of middles in these languages. The arguments for the syntactic unergativity of the middle verb largely involve its behaviour with respect to the unaccusativity diagnostics.1 The evidence that Ackema and Schoorlemmer and Cabredo-Hofherr have brought forward strongly suggests that Germanic middles feature a base-generated Theme-subject, in violation of the Uniformity of Theta Assignment Hypothesis (Baker 1988), which is stated in (11). (11) Uniformity of Theta Assignment Hypothesis (UTAH) Identical thematic relationships between items are represented by identical structural relationships between those items at the level of D-structure.

According to the UTAH, Theme arguments can only be base-generated VP-internally, and therefore any Theme-subject has to be derived by movement. However, middles in a language like Dutch systematically fail the unaccusativity diagnostics, which they should pass if their subject were a derived one. Ackema and Schoorlemmer propose an analysis of middles in these languages which involves base-generating the understood object in subject position at a presyntactic level of representation, thus accounting for their unergativity.

. For reasons of space, I cannot repeat all of the relevant arguments here. See Ackema & Schoorlemmer (1994) for Dutch and English, and Lekakou (2005a) for a recapitulation of the arguments, including those that address German, French and Greek.



JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.6 (386-435)

 Marika Lekakou

The data that pertain to the syntactic inactivity of the middle Agent appear below. In contrast to passives (see (13)), middles (see (12)) disallow Agent-oriented adverbs like carefully, they do not admit by-phrases, and the implicit Agent cannot control into a purpose clause. German and Dutch middles behave in the same way. (12) a. *Walls paint easily on purpose/carefully. b. *Walls paint easily by Harry/by anyone. c. *Walls paint best to protect them against the rain. (13) a. The wall was painted carefully. b. The wall was painted by Harry. c. The wall was painted to protect it against the rain.

Things are different in French and Greek. In these languages, middles are syntactically indistinguishable from reflexive passives (Tsimpli 1989; Ackema & Schoorlemmer 2005; Lekakou 2002, 2003; Zribi-Hertz 2003; Dobrovie-Sorin 2005).2 Greek employs the suffix of nonactive voice, which like the French clitic se can be interpreted as a reflexive, in addition to giving rise to passive, anticausative and reciprocal interpretations. Whatever differences exist between middles and reflexive passives in these languages are semantic in nature. For example, middles are generic statements, whereas reflexive passives can be either episodic or generic – although in French there is a strong preference for copula passives to be episodic and reflexive passives to be habitual/generic (Zribi-Hertz 2003; Dobrovie-Sorin 2005). With respect to the syntactic manifestation of the implicit Agent, note that middles in Greek (Condoravdi 1989; Tsimpli 1989) and French admit by-phrases: (14) Afto to vivlio diavazete efxarista akomi ki apo this the book read-nonact.imperf.3sg with pleasure even and by megalus. grown-ups ‘This book reads with pleasure even by grown-ups.’ akomi ki apo anoitus. (15) Afto to provlima linete this the problem solve-nonact.imperf.3sg even and by fools ‘This problem can be solved even by fools.’ (16) Ces étoffes se repassent facilement par tout le monde. these fabrics se iron-3pl easily by all the world ‘These fabrics can be ironed easily by everybody.’

. I employ the term reflexive passive to refer to formally reflexive structures, whose interpretation can be passive (and also anticausative, reciprocal or reflexive). In addition to the se-passive, on which the middle is parasitic, French possesses a copula passive.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.7 (435-490)


(17) La Tour Eiffel se voit de loin par tout le monde (qui the Eiffel Tower se see-3sg from afar by all the world who veut bien la voir). wants well her see-infin ‘The Eiffel Tower can be seen from afar by anyone who really wants to see it.’ (18) Le francais ne s’acquiert pas par tout le monde. the french neg se acquire-3sg neg by all the world ‘French cannot be acquired by everybody.’

I should note in passing that some speakers of Continental French do not like byphrases with the reflexive passive at all, be it episodic or generic. Eric Mathieu, who provided the examples in a personal communication, informs me that prescriptive grammars of French ban the occurrence of by-phrases with reflexive passives, which helps explain the resistance with which sentences like the ones below are met among some native speakers. Canadian French is much more permissive in this respect, as discussed by Authier & Reed (1996). There is thus robust evidence in favour of a bipartition of languages with respect to how their middles behave: there is a group of languages in which middles are syntactically unergative and lack a syntactically represented Agent, and there is a group of languages in which middles are syntactically identical to passives, i.e. they exhibit a derived subject (unaccusative syntax) and project the Agent in the syntax. The adverb facts fit in this pattern of variation: in Greek and French, middles can appear unadorned, whereas their Germanic analogues require adverbial modification.3 On the basis of the attested crosslinguistic variation in the behaviour of middles, I have argued, much in the spirit of Condoravdi (1989), that there is no middle construction, since there is no crosslinguistically coherent way to syntactically define it. What ‘middles’ share crosslinguistically is a certain semantics. The question of the crosslinguistic variation reduces to the question of the relevant morphosyntactic property that unites French with Greek against English, Dutch and German. A theory of middles thus has to answer the following question: which property of a given language determines whether its middles will be parasitic on unergative . Both Fagan and Steinbach point out in connection to the lenience of Greek/French middles that in these languages, middles ‘need not be eventive’. In my terms, this means that Greek/French-type middles (which are always generic) are parasitic on (reflexive) passives, which themselves can be episodic. Fagan further points out that episodic copula passives generally do not require adverbial modification. But that does not amount to an explanation, because middles could still, for whatever reason, require adverbial modification, even if episodic passives do not.



JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.8 (490-546)

 Marika Lekakou

or passive structures? In what follows, I will summarize an approach that takes the relevant property to be grammatical aspect for genericity. . Middles as disposition ascriptions and the role of imperfective aspect In Lekakou (2004) and Lekakou (2005a), I have proposed a novel characterization of the middle semantics which capitalizes on their genericity. The crosslinguistic variation then relates to the means that different languages have at their disposal with respect to encoding genericity. In particular, I have argued that middles are best characterized as a disposition ascription (to the understood object). Disposition ascriptions are subject-oriented generic sentences whose truth depends on properties of the syntactic subject: they are true ‘in virtue of ’ some property of their subject. The distinction between ‘in virtue of ’ and descriptive generalizations was introduced by Greenberg (2002) in the realm of NP-genericity, to account for the differences between singular indefinite and bare plural generic NPs, as in (19). (19) a. A boy doesn’t cry. b. Boys don’t cry.

My proposal is based on the idea that the distinction is relevant for sentencelevel genericity as well, which is the genericity that middles exhibit. To this effect, consider the different interpretations that the sentences below receive: (20) John goes to school on foot. (21) This machine crushes oranges.

(20) is a habitual sentence with no modal overtones, which tells us that events of John going to school are generally/normally events in which he walks. In addition to this kind of habitual reading, (21) can also have a dispositional interpretation. On the latter, the truth of the sentence depends on properties inherent in the subject referent (the way the machine is designed and manufactured etc.). On the dispositional reading, the sentence can be true even in the absence of verifying instances, i.e. even if no one has ever used the machine. For the habitual generalization to be true, on the other hand, we obviously need the prior occurrence of (a number of) events of the type denoted by the VP. Other cases of sentences which can express an ‘in virtue of ’, dispositional generalization in addition to a habitual, descriptive generalization are the following (from Krifka et al. 1995): (22) a. Kim speaks German. b. John rides horses. c. This car goes 200 km/h. (23) Mary handles the mail from Antarctica.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.9 (546-593)


The differences between habituals and other kinds of characterizing (generic) sentences have been discussed by Laca (1990), Scheiner (2003) and Van Geenhoven (2003) among others. Adopting a number of insights from Brennan (1993), I have proposed to treat the generic operator as a necessity operator which can be VP-level as well as Sentence-level. This is in the spirit of Brennan’s analysis of modals, according to which dynamic modals (dispositional will and ability can) are VP-operators, and not S-operators, which is what epistemic modals are. A VP-modal first applies to the VP, giving us a modalized VP, which is then predicated of the subject. The intuition behind Brennan’s proposal is that VP-level modals relate properties and individuals, and do not modify propositions (as do S-operators). I believe the same intuition applies in the realm of genericity. The generic operator has been assimilated to a (quasi-) universal necessity modal. I propose that, in addition to what we may call a ‘Sentence-level’ generic operator (e.g. as defined by Krifka et al. 1995), which is suitable for habitual sentences, we need a VP-level generic operator, which is employed in dispositional generics. The semantics of this dispositional generic operator is very similar to the semantics that Brennan attributes to the dispositional necessity operator. For reasons of space, I cannot go into the semantics of dispositionality any further here. I will restrict myself to a brief mention of two kinds of evidence in favour of the dispositionality component in middles. For extensive discussion and for the formal implementation see Lekakou (2005a). Dowty (2000) observes that “the only factors that determine whether a middle construction sentence [sic] is true are properties inherent in the object acted on” (Dowty 2000: 16) (cf. also van Oosten’s Responsibility Condition). This is exactly what the dispositional semantics can derive for us, as it makes reference to properties of the syntactic subject in virtue of which the assertion made is true. Following Brennan, I employ the ‘in virtue of ’ paradigm to show that middles as dispositionals are subject oriented. According to the VP-analysis of the modal operator, ‘in virtue of ’ adverbials are obligatorily subject-controlled only when combined with VP-modals. This is what we see going on in the case of middles, cf. (24). In fact, something very similar has already been pointed out: Van Oosten (1977) first noted (25) and Dowty (2000) offers (26). (24) This bread cuts easily ... a. ... in virtue of its texture. b. *... in virtue of the sharpness of the knife. (25) The clothes wash with no trouble ... a. ... because they are machine-washable. b. *... because I have lots of time.



JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.10 (593-661)

 Marika Lekakou

(26) This car drives well... a. ... because the suspension is engineered well. b. ??... because we are driving on smooth pavement.4

Second, middles and nonepisodic passives in languages like English are not semantically identical. The semantic difference is the difference between dispositionals and habituals: (27) This book reads easily, but it isn’t easily read. (28) a. Poetry isn’t translated. b. Poetry doesn’t translate.

The fact that (27) is not a contradiction shows that middles and generic passives encode different interpretations. That only middles are true dispositionals, and that generic passives are habitual statements is evident from the different way in which the sentences in (28) are interpreted: (28a) asserts that there are no events of translating poetry, whereas (28b) expresses that translating poetry is extremely difficult. I will discuss the interpretation of (28b) in a subsequent section. In addition to semantic facts, some of which I just noted, we can also get syntactic effects from ascribing to middles dispositional semantics.5 Dispositional generics employ a VP-level generic operator and are subject-oriented. This means that what will be ascribed a dispositional property is the entity denoted by the DP which occupies the syntactic subject position. In other words, if the understood object is to be ascribed a dispositional property, it needs to be promoted to subject position. Concomitantly, the otherwise most eligible candidate for the subject position, the Agent, has to be demoted. This happens either in the syntax, as in Greek and French, or at a presyntactic level, as in Germanic. The factor that determines this is the morphosyntactic encoding, or lack thereof, of genericity. . Note that the sentences in (25) and (26) feature ‘because’ clauses, which for some speakers allow more leaway (as does preposing ‘in virtue of ’, for reasons that I do not understand). An anonymous reviewer finds (i) well-formed: (i)

This Latin book translates easily for me because I’m a Latin scholar.

However, the speakers I have consulted do not agree with this judgment. Nor do they like the sentence when an in virtue of -phrase replaces the because-clause. The acceptability of (i) seems to hinge on the presence of the for-phrase, to which I will return. . Additionally, the dispositionality proposal can derive the fact that indefinite subjects of middles can only be interpreted generically and not existentially. This point will come up again at the end of this section. To simply say that middles are generic statements cannot derive this fact, because it is possible for indefinite subjects of generic sentences to be bound existentially, and not generically. See Lekakou (2005a) for extensive dicussion.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.11 (661-724)

Adverbial modification in middles 

In connection to the dispositional semantics, I have shown that the Agent in middles is interpreted as an inherently generic indefinite, as is the pronoun one. I have dubbed this interpretation ONE*. In virtue of corresponding to a generic indefinite, ONE* is only licit in the context of genericity (for the notion of a generic polarity item, cf. Chierchia 1995). I have proposed a correlation between the nature of ONE* and the nature of the generic operator that licenses it, in the form of an assumption of immediate licensing for ONE*: (29) ONE* needs to be licensed at the level at which it is expressed.

According to (29), ONE* can be syntactically active (and in fact is obligatorily syntactically active) only if it is licensed by an operator which is realized in the morphosyntax. This is what happens in Greek and French, whose imperfective verbal forms encode genericity. In English, Dutch and German, the generic operator is not morphosyntactically expressed, that is, it is only present semantically. Syntactic licensing cannot obtain in this type of language, because genericity is not expressed morphosyntactically. The criterion by which languages are classified to have or to lack aspectual morphology for the generic operator is given in (30). On the basis of this line of reasoning, the typological generalization made for the syntactic behaviour of the verb in middles is stated in (31): (30) A language encodes Gen in imperfective morphology iff in at least one tense it has two distinct verb forms for generic and nongeneric uses, i.e. iff genericity ⇒ imperfectivity. (31) A language employs an unaccusative structure to convey the middle interpretation iff Gen is encoded in imperfective morphology. Otherwise, an unergative structure is used.

Before returning to the issue of the adverb, I would like to raise the following issues. It has been suggested to me by an anonymous reviewer that middles need not always be generic (much less dispositional), because of cases like (32): (32) The book translated quickly.

(32) employs a simple past tense form, which can be either episodic or generic. It is thus not clear whether we are dealing with a real episodic middle, or with a generic, but temporally restricted, middle. I do not believe we can use our intuitions here to see if the sentence is episodic or generic, but in fact, there is something which can show us what is going on. We know that generic sentences can have existentially or generically interpreted indefinite subjects. This is the ambiguity of examples like (33) from Krifka et al. (1995). (33) Typhoons arise in this part of the Pacific. a. Typhoons in general are such that they arise in this part of the Pacific. b. This part of the Pacific is such that there exist typhoons which arise in it.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.12 (724-764)

 Marika Lekakou

However, it is impossible for an episodic sentence to have a generically interpreted indefinite subject, instead of an existential one. This is our test for the existence of episodic middles. If in (32) we use an indefinite subject, can we get an existential interpretation of it? For the speakers I have consulted at least, this is not possible. At least for those speakers, since indefinite subjects of middles are always generic, middles themselves are always generic. (32) is not a counterexample to this.6 The same reviewer brings up the issue of for-phrases that denote specific individuals (cf. also Stroik, this volume). These cases constitute counterarguments to the claim that middles always feature a generically interpreted implicit Agent. However, the acceptability of for-phrases in middles is as yet not very well-understood (Rapoport 1999; Ackema & Schoorlemmer 2005), and there are speakers who resist them. For those who accept them, these PP’s seem to restrict the domain of generic quantification over Agents; their semantic contribution definitely requires closer inspection. However, the questions they raise are not particularly devastating. Stroik claims that the for-phrase is to the middle what the by-phrase is to the passive. If so, there is nothing to prevent us from falsely predicting that Greek and French middles admit specific by-phrases. On the other hand, there is no argument against treating the for-phrase as the overt expression not of the implicit Agent of the verb, but of the implicit Experiencer/Benefactor of the adverb. There is clear evidence that the two arguments are always identified (see below). It remains to be seen whether there is a way to decide between this option and Stroik’s analysis. . Back to the adverb In what follows, I will argue that the adverb is required in languages whose middles do not have a structurally represented Agent, viz. in English, Dutch and German, because the adverb is the means to recover the implicit argument. Languages with grammatical aspect for genericity, i.e. French and Greek, do not need any support in the form of adverbial modification, because in these languages the implicit Agent is syntactically represented. From this reasoning, it follows that only adverbs which can serve the purpose of recovering the implicit argument are licit in middles; such adverbs are those that have an implicit argument themselves, which is identified with the implicit Agent of the verb. But that is not all: the adverb must not clash with the dispositional semantics associated with middles, i.e. it cannot refer to properties of the suppressed Agent. The question of what sort of adver. There may exist speakers for whom existential subjects in middles are possible, and hence episodic middles may exist. Since even then the genericity of middles still constitutes the core, this for me would mean that we are witnessing an extension of ‘the middle’. Such a development of course merits investigation.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.13 (764-819)

Adverbial modification in middles 

bials are permissible in middles is addressed in Section 5. In the following sections I give evidence that the role of the adverb is to aid the recoverability of the implicit argument.

. The semantic/pragmatic approach and its problems As mentioned already, there have been three positions defended in the literature concerning the reason why adverbial modification is required in middles. On the one hand, there is a syntactic or structural account, defended by Roberts (1987), Pitz (1988) and Hoekstra & Roberts (1993). On this approach, the adverb is required in order to aid the recovery of the suppressed argument in middles. On the other hand, there are accounts which attribute the requirement for modification to conditions on the semantic well-formedness of sentences. This position has been advocated most notably by Condoravdi (1989) and McConnell-Ginet (1994). A third, closely related view is expressed in Ackema & Schoorlemmer (1994), Fagan (1992) and Steinbach (2002), where the claim is that it is pragmatic informativity that makes modification necessary in middles. For Condoravdi (1989) and McConnell-Ginet (1994) the adverb is required in middles in order to provide the scope for the generic operator. The generic operator is taken to induce a tripartite structure, consisting of restrictor and nuclear scope. The semantic structure proposed by Condoravdi for This book reads easily is given in (34), which says that in general, events of reading the book are easy. (34) Gen: e [read(e) & Book(Theme,e)] [easy(e)] Restrictor Scope

The adverb requirement is, on this account, tied to the fact that the scope of an operator cannot be left empty, nor can it be contextually specified. Thus, on this view, adverbless middles (e.g. the sentences in (1a)–(3) above) are ungrammatical because the scope of the generic operator has been left empty. However, middles can do without an adverb just in case something else contributes the scope for the operator. This is, according to this approach, what happens in examples like (35) and (36), which McConnell-Ginet (1994) discusses. In these cases the context makes it possible for the verb, which would normally form part of the restrictor of the generic operator, to end up in its scope, thus satisfying semantic wellformedness:7 (35) This silk washes. (36) This dress buttons. (37) This car HANDLES. . For most of my speakers, only (36) is acceptable.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.14 (819-866)

 Marika Lekakou

What McConnell-Ginet suggests for (35) and (36) is that “the main verb is not needed as the restrictor because context has provided restriction via implicit contrast among different modes of doing something, the main verb then being free to designate one such mode (serving then to give content to the scopal element)” (McConnell-Ginet 1994: 247). Put differently, the context refers to modes of cleaning silk, and fastening dresses. The verb contributes manner modification, in that it specifies the manner in which this silk is cleaned, namely by washing, and this dress is fastened, namely by buttoning. As for (37), things are not very different, as McConnell-Ginet suggests that handle is like behave, in that it subcategorizes for an adverb. If an adverb is not present in the sentence, something like well is nonetheless understood. The sentence, according to her, “seems to presuppose events of driving or something like that and asserts that such events are also events where the car is successfully “handled” ”(McConnell-Ginet 1994: ibid.). This sounds reasonable enough. But note that what we are saying is that there is adverbial modification in examples like (35) and (36), although not contributed by an adverb, but by the verb itself. (Note also that McConnell-Ginet says that in the case of (37), ‘well’ is implicitly present.) There remain the cases of adverbless middles where it is the presence of focus, negation or a modal that apparently makes the lack of an adverb tolerable. The semantic approach can fairly straightforwardly handle the facts reported in the literature concerning focus. It has been generally acknowledged that focus interacts with the partition of a clause into restrictor and scope. Since stress/focus on the verb can ensure that it is mapped onto the scope of the generic operator, it follows for the semantic account that verb focusing is one of the factors which can rescue an adverbless middles. A similar effect, however, would have to be attributed to negation and modals. One would need to show that negation and modals can rescue adverbless middles precisely because they partition the clause in such a way as to place the verb in the scope of the generic operator. To the best of my knowledge, this has not been achieved as yet. On the pragmatic account of adverbial modification in middles, adverbless middles are odd, because they are uninformative. Our knowledge of the world dictates that books are and can be read, clothes can be washed, doors can be closed etc. It is therefore hopelessly uninformative to utter middles which simply state this, without making reference to the manner in which such common actions can or are generally performed. This is the sort of account that Fellbaum (1986), Ackema & Schoorlemmer (1994), Steinbach (2002) and Marelj (2004) defend. The pragmatic account is close to the semantic account, because on both the adverb is the core of the assertion. The first problem with this conception of the role of the adverbial is that it does not actually correspond to native speakers’ intuitions. If we look beyond middles, it seems true that sentences like the ones in (38)–(40) are uninformative. In other words, habitual active and passive sentences without adverbs (with and without

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.15 (866-918)


a modal) require a suitable context, otherwise they are pragmatically odd, but of course not ungrammatical. (38) a. One reads/can read books. b. Books are (generally) read. c. Books can be read. (39) a. One washes/can wash clothes. b. Clothes are washed. c. Clothes can be washed. (40) a. One drives/can drive cars. b. Cars are driven. c. Cars can be driven.

The pragmatic approach predicts that the corresponding adverbless middles yield the same reaction: they are not ungrammatical, they are just uninformative and hence pragmatically odd. But in actual fact, native speakers do not treat adverbless middles as uninformative and pragmatically deviant, but as simply extremely difficult to process (as middles). The native speakers of English, Dutch and German that I have had access to systematically reject adverbless middles, no matter how much stress is placed on the verb, and irrespective of the presence of negation or a modal. A different problem which plagues both the pragmatic and the semantic approach is the fact that the requirement for an adverb will have to be relativized to English, Dutch and German, in view of the Greek and French data in (9) and (10). This means that we would need to parameterize the existence of a condition on semantic well-formedness or pragmatic informativity, which is clearly undesirable. The fact that Greek and French can afford to not satisfy the constraint imposed on their Germanic counterparts strongly suggests that the constraint in question is not a semantic or a pragmatic one, but a syntactic one. In what follows I will argue in favour of this conclusion, and I will sketch an account that relates the adverb discrepancy to the independently available syntactic discrepancies between the two types of middles. Before doing that, however, let me mention two alternatives to the conclusions I just drew, which would rescue the semantic account, as pointed out by a reviewer. First, one could argue that middles across languages are not semantically identical. On this view, different semantic representations for French and English middles are to be expected. The second alternative would seek to find a property of French and Greek that is responsible for systematically placing the verb in the scope of the generic operator, when there is no adverb. Although these are possible alternatives, there is no evidence to support them at this stage. Further comparative treatment



JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.16 (918-969)

 Marika Lekakou

of (adverbial modification in) middles may be able to explore and assess these avenues of research in the future.

. A ‘structural’ account In the previous section we saw that a serious problem for the semantic/pragmatic account is the ungrammaticality of adverbless middles. Instead of the middle interpretation, such sentences yield an object-deletion reading, i.e. a reading where the surface subject is the Agent of the action and the Patient is implicit. In some cases, such as (41) and (42), this makes no sense, since inanimate entities such as books and shirts cannot be Agents; the sentences are deemed ungrammatical. The object-deletion reading is possible for (43), which contains an animate subject. (41) *This book READS. (42) *The shirt WASHES. (43) *Bureaucrats BRIBE. (ok on the object-deletion reading)

This strongly suggests that the adverb has something to do with recovering the implicit agent, which in these languages is not syntactically represented. Such an approach to the adverb complies with the cross-linguistic variation that I outlined in Section 2: English, Dutch and German middles lack an Agent in the syntax. Moreover, in English, Dutch and German, no morphology appears on the verb that could signal that the agent has been suppressed.8 In these languages, the adverb is required in order for the middle interpretation to obtain. By contrast, languages like Greek and French, which due to the nature of their aspectual system resort to passives to encode the middle semantics, do not need any assistance in the form of an adverb, in order to convey the middle interpretation: since they have a structurally represented Agent, there is no need for an adverb to aid in recovering it. For these languages, the pragmatic account might be tenable. This would be fairly reasonable, as it is standardly pragmatic/contextual factors that disambiguate a multiply ambiguous form, such as the Greek (imperfective) nonactive forms, and the French se-construction. For English, Dutch and German middles, my claim is that there is always adverbial modification of sorts,

. The reflexive sich employed in German middles does not contradict these claims, as there are arguments to the effect that this element is not a passivizer and does not correspond to the Agent (Lekakou 2005a, b). Consequently, German middles too involve no morphology that corresponds to the external argument.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.17 (969-1038)

Adverbial modification in middles 

and that there are in fact no grammatical middles without manner modification, because the requirement for an adverb is a structural one.9 That the role of the adverb in middles is to provide the means to recover the implicit argument has been proposed by Roberts (1987). Middles, according to him, involve a chômeur Agent. Chômeur θ-roles are theta roles that are unassigned in the syntax, but associate with the verb throughout the derivation. A thematic role becomes chômeur when “some lexical rule changes its realization without deleting it” (Roberts 1987: 188). In a similar vein, in Lekakou (2005a) I analyze Germanic middles as deriving from a transitive entry whose Agent is bound by the generic operator at the level of lexical semantics and does not project any further (it does not map onto subsequent levels of conceptual structure, nor does it reach the syntax). Although it is not present syntactically, the Agent is present semantically, and in that sense it survives throughout the derivation. The idea is that what ensures the survival of the suppressed Agent from the level of lexical semantics to the level of LF is the adverb.10 How does the adverb do this? According to Hoekstra & Roberts (1993), adverbial modification is a means of recovering the implicit Agent, by supplying an implicit Benefactor/Experiencer argument which is identified with the implicit argument of the middle verb. I will follow these authors in claiming that the adverb effects an identification process between the two implicit arguments, and therefore its presence rescues the targeted middle interpretation, which will not arise otherwise. This is summarized below: (44) a.

In Germanic middles, the adverb is required in order for the implicit Agent to be recoverable. b. The implicit middle-Agent is recovered via identification with the implicit adverb-Experiencer/Benefactor.

. For the relation between manner adverbs and nonactive voice, see also Cinque (1999). . We can assume, following Roberts, that object-drop verbs also contain a chômeur theta role. Rizzi (1986) has argued that object drop in English does not involve a syntactically represented argument, but is the result of saturation at a presyntactic level. The operator which effects saturation in this case too is the generic operator. The difference between the two ‘chômeur’ roles is that only the one corresponding to the middle Agent needs to be identified in the syntax by an adverb. Roberts is aware of the problem which in his terms amounts to the fact that Tense is an adequate licensor in object drop cases, but not in middles. The discrepancy could be due to an asymmetry between external and internal thematic roles (see Williams 1981; Marantz 1984; and related work). In fact, there seems to exist such a hierarchical effect: as Rizzi (1986) noted, in English, indirect internal roles, i.e. dative arguments, can be dropped much more easily than direct internal roles, i.e. accusative objects. In middles, which involve suppression of an external argument, the latter cannot be left unprojected in the syntax unless something else guarantees its recoverability. See the discussion in the main text.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.18 (1038-1097)

 Marika Lekakou

The upshot is that only adverbs with an appropriate thematic structure are compatible with middles: admissible middle-modifiers are dyadic adverbs that take an Experiencer and an event argument. My addition to this is that, due to the dispositional semantics of middles, any dyadic adverbs that make reference to properties of the implicit Agent are illicit, because they clash with the disposition ascription, which targets the internal argument. I will elaborate on this in the following section. As an indication of the fact that the implicit Agent is identified with the implicit Experiencer, consider the interpretation of a typical middle like This book reads easily. It is impossible for this sentence to mean that it is easy for x for y to read the book. In other words, the implicit argument of the middle-adverb is obligatorily coreferential with the implicit argument of the verb. A reasonable way to achieve this is by having the two arguments identified at some level of representation. For the workings of this identification process, Hoekstra & Roberts (1993) follow Higginbotham (1985), who proposes theta-identification as another means of theta-role assignment. In particular, Higginbotham proposes identification in order to derive the thematic properties of modification, as in examples like big butterfly. According to him, in such cases the adjective takes an argument which is identified with an argument of the nominal. In (45), this is what [1] stands for. (45)

(N’,[1]) (A,[1])

(N,[1])

big

butterfly

Like Hoekstra and Roberts, I adopt Higginbotham’s proposal and apply it to the identification taking place between the verb and the adverb in middles. One question raised by such an approach is, why is identification obligatory? A reviewer poses this question in the form of the following minimal pair: (46) Das Fenster öffnet sich. the window opens refl (47) Dieses Fenster montiert sich *(leicht). this window adjusts refl easily

In the absence of easily, (46) cannot have the middle reading. Because the verb ‘open’ has an unaccusative variant, the sentence is grammatical, precisely on the unaccusative reading (which in this case can be episodic/progressive, or generic). However, (47) is out without an adverb, because the verb montieren does not form an unaccusative and can only give rise to the middle interpretation. Recall that middle formation in Germanic languages involves the presyntactic suppression of

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.19 (1097-1140)


an Agent argument. Reinhart (2000) discusses a robust generalization, to the effect essentially that Agent arguments cannot be removed from the thematic grid of a predicate. The question of the obligatory recovery of the Agent (i.e. the obligatory presence of an adverbial) may thus have to do with the lexical semantic properties of the predicates that form middles, and in particular with the constraint on deleting Agent roles.11 Any version of the structural approach predicts that there exist no middles which do not feature an adverb of the appropriate type. The bulk of the remaining discussion revolves around this aspect of the analysis, namely the presence of an (implicit or explicit) adverbial in middles. In particular, there are two empirical challenges that this view of the role of the adverb faces, and in the rest of this section I will deal with these in turn. First, we must answer the question of what happens with middles which do not have a manner adverb. There are two subclasses involved. One comprises the examples discussed above in connection to the semantic approach to middlemodifiers. I suggested there that the proposed treatment of cases like This dress buttons implies that there is a manner component in the sentence, albeit not in the form of an adverb, but in the form of a manner component in the meaning of the verb. So this kind of example does not threaten the proposed account. The second class of cases is more interesting. These are the cases where focus, negation or a modal seem to rescue a middle in the absence of the adverb. As Ackema & Schoorlemmer (1994) observe, even if easily and well could be said to have an implicit argument of their own, which helps recover the suppressed middle Agent, this surely cannot be true of negation or focus (disregarding modals). I will show that the structural approach can deal with these cases as well. I will argue that adverbless middles which appear to be rescued by focus or negation are really rescued (to the extent that they are) by an implicit adverb. The claim is that there is a manner adverbial, but it is implicit. This will explain why such sentences are extremely difficult for speakers to accept: not only has the Agent been suppressed, but the very element that would help recover it is also missing. On this view, if focus, negation and modals have any effect on the acceptability of adverbless middles at all, the effect relates to the recoverability of the implicit adverb.

. Note that in any event, the obligatoriness of the recovery of the Agent does not relate to the genericity of the argument in any way, but rather to its status as an Agent. This may prove relevant for a question posed by another reviewer, namely why the adverb can tolerate its Experiencer to remain implicit/unrecovered. One answer would appeal to the fact that only the verb’s implicit argument is an Agent. Alternatively, one could argue that the adverb’s Experiencer is not its external argument, and only suppressed external arguments require recovery via identification.



JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.20 (1140-1198)

 Marika Lekakou

Although defending a pragmatic account of the adverb, Steinbach (2002) provides the following quite interesting examples, which give a first indication that there is an implicit adverb and which argue in favour of the above approach. (The examples used are impersonal middles, but this is not relevant here.) The paraphrase in (48) is mine, on the basis of Steinbach’s comments about the meaning of the sentence. The second example appears unaltered from Steinbach (2002). (48) Hier tanzt sich’s. here dance refl it ‘Dancing here is good.’ (49) Hier lebt es sich, sagt der Zander. here lives it refl, says the pikeperch ‘The pikeperch says this is a nice place to be.’

(48) means, not that one can dance here, but that one can dance well here; and similarly for (49), as Steinbach’s gloss suggests. The same point applies to the Dutch example from Ackema & Schoorlemmer (2005), repeated from above. (50) Die aardappels ROOIEN, niet te geloven! those potatoes dip.up, not to believe ‘I can’t believe how easy to dig up those potatoes are.’

The authors’ gloss suggests that the sentence involves adverbial modification by an implicit gemakkelijk, ‘easily’. What seems to contribute the implicit adverb is the phrase niet te geloven, ‘not to believe’ (along with heavy stress on the verb). Note that this is something standardly associated with this expression, cf. its contribution in the following examples: (51) a.

Jan zingt ARIA’s, niet te geloven! Jan sings arias not to believe ‘I can’t believe how well Jan sings arias.’ b. Jan ZINGT, niet te geloven! Jan sings not to believe ‘I can’t believe how well Jan sings.’

More generally, the interpretation that adverbless middles receive indicates that there is an adverb involved, albeit an implicit one. One of the very few ‘adverbless’ middles that native speakers of English accept involves the verb ‘translate’. We have already encountered example (52): (52) This poem doesn’t translate.

(52) can only be used to attribute to a book the property of not translating well (or easily), but not the property that it is literally impossible to translate. We can see

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.21 (1198-1254)

Adverbial modification in middles 

that if we construct the following scenario. We have discovered a book of poems that is written in Martian, which we have unfortunately not managed to decipher yet. (52) cannot be used in this context. This applies to all of the examples from the literature of adverbless middles with focus or negation. To the extent that they are good, they always also have an implicit well or easily as part of their meaning. To give another example, the sentence in (53), which Condoravdi (1989) brings up, cannot be used about a rock which resists cutting altogether, but only about a rock which presents difficulties for anyone who tries to cut it. (Having said that, (53) is still out for most of my informants.) (53) This rock does not cut.

Consider finally the example in (54) from Steinbach (2002). The context involves the speaker walking into a bar and wondering which beer she should order. (54) does not mean that this beer (alone) is literally drinkable; as the context suggests, it means that it is the only beer, compared to the rest, which drinks well.12 (54) DIESES Bier trinkt sich. Die anderen schmecken furchtbar. this beer drinks refl The others taste awful ‘THIS beer drinks. The others taste awful.’

The second problem that the account of the middle adverbs defended here faces relates to an objection raised by both Condoravdi (1989) and Ackema & Schoorlemmer (2005). The problem is that middles fail to fully parallel tough-constructions when it comes to the adverb. The reason why tough-constructions in particular are relevant is that they seem like a natural paraphrase of middles, as in (55). Moreover, the view of the adverb as the means of recovering the implicit Agent by supplying an Experiencer/Benefactor argument leads, according to Condoravdi, to the prediction that “the class of adverbs appearing in the middle should be coextensive with the class of adverbs having a benefactive role” (Condoravdi 1989: 20). Condoravdi then points out that the paraphrase of (56a) in (56b) fails: (55) a. This bread cuts easily. b. This bread is easy to cut. (56) a. Das Buch liest sich gut. b. It would be good for anybody to read this book. . The relevance of these considerations was originally brought to my attention by Sabine Iatridou (personal communication) in reaction to the claim advanced in Sioupi (1998) for Greek, that adverbless middles involve an ability modal operator (instead of the generic operator, which appears whenever there is an adverb). The situation with the Greek equivalents of these sentences is not clear, due to substantial inter-speaker variation and uncertainty of judgments. So it is difficult to determine at this point whether there is an implicit adverb in Greek as well and whether Sioupi’s claim about an ability modal operator is correct.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.22 (1254-1310)

 Marika Lekakou

Ackema & Schoorlemmer (2005) also point out that the set of eligible middle modifiers is not coextensive with the set of adjectives that appear in toughconstructions. They give the following examples: (57) a. This book is impossible to read. b. *This book reads impossibly. (58) a. This meat cuts just like that. b. *This meat is just like that to cut.

However, it is unclear that this objection constitutes a real problem, and it is unclear that the prediction made by this approach to the adverb is the one identified by Condoravdi and Ackema & Schoorlemmer. The tough-construction frame requires an adjective, whereas the middle requires an adverb. A tough-construction therefore independently cannot admit phrases like just like that, (or the Dutch als een trein, ‘like a train’ in Dit boek leest als een trein, discussed by Ackema & Schoorlemmer), because the phrase does not fit the frame, since it is not an adjective. Moreover, since there independently exist differences between adjectives and related adverbs – good and well, impossible and impossibly – we expect the parallel to break down in precisely those cases. A similarity between the two constructions which failed to be noticed by Condoravdi and Ackema & Schoorlemmer (but cf. Hoekstra & Roberts 1993) is that the implicit argument of the adjective in tough-constructions and (as mentioned already) of the adverb in middles is necessarily construed as identical to the implicit argument of the verbal predicate. This has been known for toughconstructions since at least Berman (1973). The following examples illustrate this (Berman 1973: 265): (59) a. For his children to be worrying about money is unpleasant for Joe. b. It is unpleasant for Joe for his children to be worrying about money. c. *Money is unpleasant for Joe for his children to be worrying about. (60) a. For Sally to take German would be good for her. b. It would be good for Sally for her to take German. c. *German would be good for Sally for her to take. (61) a. For Max to publish that rebuttal will be tough for Joe. b. It will be tough for Joe for Max to publish that rebuttal. c. *That rebuttal will be tough for Joe for Max to publish.

Assuming that the differences that Condoravdi (1989) and Ackema & Schoorlemmer (2005) observe between the two constructions are independent, this similarity indicates that the account of middle-modifiers endorsed here is on the right track. To conclude this section, the structural approach (a) allows us to make sense of the intuitions that speakers of English, Dutch and German have for adverb-

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.23 (1310-1369)

Adverbial modification in middles 

less middles and (b) fits in very well with the crosslinguistic variation in the obligatoriness of the adverb. The variation cannot be accounted for within the semantic/pragmatic account. I have rebutted the objections to a structural approach that have been put forward in the literature. In the following section, I will sketch how the class of legitimate middle modifiers can be derived on the basis of the proposed account of their role.

. Delimiting the set of appropriate modifiers The question of which are the permissible middle modifiers has received very little attention in the literature, and is hence far from being resolved. In this section, I will show that there are predictions made by the approach defended in the previous section. I will pursue the hypothesis that adverbs in middles are required in order to enable the implicit argument to be identified, and hence what is required are event-modifiers with an implicit argument, which however do not clash with the dispositional semantics of middles. So in what follows I will concentrate mainly on the kinds of adverbs that this approach predicts to be disallowed. Toward the end of the section, I will review Condoravdi’s proposal about the set of eligible middle adverbs and I will show that her problems with approaches similar to the one pursued here are not as devastating as she suggested. To start, we need a basic classification of adverbial modifiers. In what follows, I will employ Parsons’ (1990) classification. Partially relying on Jackendoff (1972), Parsons lists the following five categories of adverbial expressions. (62) Classification of Adverbial Modification13 a. Speech-Act Modifiers, e.g. fortunately b. Sentence Modifiers, e.g. necessarily c. Subject-Oriented Modifiers, e.g. rudely d. VP Modifiers, e.g. gently e. Other: merely, just, only

Speech-Act modifiers, according to Parsons, have the following effects. In a sentence like Fortunately, Mary arrived on time, two things are asserted, that Mary arrived on time (so Speech-Act modifiers are factive), and that this fact is fortunate. The adverbs that are included in this category are further subcategorized into four classes, but this need not concern us here. (See Geuder 2000 for this point.) Sentence modifiers are proposition-taking adverbials: they “operate on structures that . In this system, temporal modifiers constitute a separate category, which comprises adverbials that cut across the following five categories, for example twice, never, soon, at midnight, etc.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.24 (1369-1416)

 Marika Lekakou

are already full-fledged formulas of English” (Parsons 1990: 63). Subject-oriented modifiers relate entities to propositions. In that sense, Parsons suggests, they are similar to sentence modifiers. John rudely departed on the subject-oriented reading means that it was rude of John to depart.14 VP modifiers are mainly adverbials of manner, instrument, location etc.15 The adverbs that appear in the category ‘other’ are not discussed in Parsons (1990). The account of the role of adverbial modification in middles that I have been defending excludes ‘higher’ adverbs, as they are not event-modifiers. Categories (a)–(c) above are either fact- or proposition-taking adverbials (see Parsons 1990 for the distinction between facts and propositions). Out of the categories listed above, therefore, only the VP modifiers involve candidates which will be considered for middle modification. This is correct: (63) *(Fortunately) This book reads (fortunately). (64) *This shirt (necessarily) washes (necessarily). (65) *Bureaucrats bribe rudely. (66) *These potatoes merely peel.

Not all ‘low’ adverbs can occur in middles. This is because the dispositional semantics that characterizes the middle interpretation dictates that it is properties of the ‘dispositional subject’ that are crucial with respect to the generalization that the middle expresses. Properties of the implicit argument are irrelevant. In fact, any adverb that is Agent-oriented in the general sense of invoking or ascribing properties to the implicit argument will not be tolerated. We can see this in the following examples from Fellbaum (1986), who deemed these adverbs ‘Agent-specific’: (67) *This little flashlight plugs in expertly. (68) *Red wine spots wash carefully. (69) *Cotton irons cautiously.

Adverbs like expertly, carefully, cautiously attribute a property to the Agent of the action denoted by the verb, namely they specify that the Agent has some sort of expertise, is (being) careful and cautious respectively. See Geuder (2000) for an analysis of the relation between manner and agentive readings of adverbs, according to which the manner reading of adverbs like stupidly is derived from the . The adverb has a ‘manner’ reading as well, which is preferred when the adverb appears sentence finally. On this latter reading, the adverb belongs to the family of VP modifiers. I will suggest below that this reading is also ‘subject-oriented’ (or Agent-oriented) in a sense. . Nilsen (2000) uses the term ‘circumstantial adverbials’, a category which comprises more kinds of adverbs than does Parsons’ category.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.25 (1416-1480)


agentive reading. This means that to modify a middle with stupidly will be incompatible with the interpretation that middles target, according to which the generalization made is not restricted by properties of the implicit Agent (more precisely, the modal base that restricts the generic operator does not comprise properties of the implicit Agent). Some further examples of this kind of clash follow. (70) *This soup eats hungrily. (71) *Complaint letters write angrily. (72) *Meat doesn’t cut elegantly. (73) *Old ladies don’t startle proudly.

A parenthetical comment is in order here. In Greek, agent-oriented adverbs are not disallowed, cf. (74), but the resulting interpretation is either a purely habitual, or a prescriptive generalization (see Fagan 1992 for French). Crucially, the middle interpretation is unavailable. (74) Afto to komati pezete me sinesthima/prosektika. this the piece-nom play-nonact.imperf.3sg with feeling/carefully ‘This piece is played with feeling/carefully.’

This is not surprising, given that in Greek the middle is just one of the interpretations available for imperfectively marked passives. In other words, for the same reasons that obtain in English, having to do with the dispositional semantics, agent-oriented adverbs preclude the middle interpretation in this language too: the middle as a disposition ascription cannot qualify properties of the generically understood implicit argument. However, different interpretations may be possible, depending on the language. Note that the same range of interpretations is available for English generic passives. The fact that middles in English, Dutch and German are never associated with prescriptive modality, but passives can be, suggests that there is a link between prescriptive modality and a syntactically represented implicit argument, the bearer of the obligation.16 Additional support for this comes from generic unaccusatives, which lack an Agent and also a prescriptive/deontic interpretation. (75) only has a habitual interpretation: (75) To kokino krasi katharizi me alati. the read wine-nom clean-act.imperf.3sg with salt ‘Red wine comes off with salt.’ . Middles are thus of potential relevance to the issue of the argument structure of (deontic) modals; perhaps the bearer of an obligation needs to be syntactically represented, which obtains in passives but not in unergative-type middles, though, see Bhatt (1998) and Wurmbrand & Bobaljik (1999) for arguments against such an analysis of deontic modals.



JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.26 (1480-1546)

 Marika Lekakou

Condoravdi (1989) criticizes the approach to the adverb in middles that Roberts (1987) and Pitz (1988) take and suggests instead that only rate adverbs are good middle-modifiers: “if the middle is about the way an object determines the progress of an event it participates in, the admissible adverbs must be those which specify something about the mapping of events onto time, or about the amount of change effected by an event over time” (Condoravdi 1989: 21). This fits in particularly well with the view that Condoravdi takes of the restriction on the input to middle formation as being a restriction to Incremental Themes. More in general, Condoravdi’s suggestion is the only serious alternative there is at this point to a view à la Hoekstra and Roberts, which is why it merits consideration. Despite its attractiveness, I find the appeal to rate adverbs problematic. For one thing, Condoravdi introduces a category of adverbials for which a definition is still to be offered: it is not clear which adverbs qualify as rate adverbs. What is more, there are cases of adverbs which are not rate modifiers in any intuitive sense, yet they are perfectly acceptable in middles: e.g. beautifully, well, nicely. Another example of an adverb which does not seem to qualify as a rate modifier is safely. Condoravdi in fact uses it as an argument against Roberts (1987) and Fellbaum (1986), by claiming that its incompatibility with middles is mysterious for these accounts. According to the speakers I have consulted, however, (76) is not ungrammatical, but merely requires some effort to contextualize. (76) This rock shatters safely.

Condoravdi’s major weapon seem to be time-span adverbials like in a jiffy and in three hours, which do not plausibly involve an Experiencer/Benefactor but which can be thought of as rate-modifiers. (In fact, these phrases have been analyzed by Krifka (1998) as denoting measure functions over events.) However, even this seemingly most devastating argument against the approach defended here is upon closer inspection not that strong. For one thing, in a jiffy does not really specify the time it takes for something to be done, but means something like ‘easily’ (as Sabine Iatridou (p.c.) has pointed out to me). More crucially, even adverbials which more clearly indicate a time span are involved in contrasts of the following kind: (77) (Uttered about a thin book of short stories) a. This book reads in a day. b. ??This book reads in a month. (78) (Uttered about a very uncomplicated tent) a. This tent assembles in seconds. b. ??This tent assembles in a week.

What would alleviate the problem in (77b) is to use it about an enormous book which nonetheless is an easy read due to the way it is written. Similarly (78b) is

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.27 (1546-1577)

Adverbial modification in middles 

better if it is uttered about a different kind of tent, one which is so elaborate that we expect setting it up to take normally more than a week. But this suggests that time-span adverbials are acceptable only to the extent that they can be construed as expressing the ease with which the action denoted by the verb is carried out. What is noteworthy about these examples is therefore that ease is apparently measurable in units of time. If this is on the right track, then Condoravdi’s major objection against the account of middle modifiers defended here disappears. More importantly, we are led to conclude that easily is always modifying middles, in one guise or the other. This follows from a structural approach much more readily than from any other currently available account.17

. Conclusion and opened questions In this paper, I have focused on the question of why adverbial modification is required in middles in English, Dutch, and German. Of the existing proposals, the semantic/pragmatic account faces some problems, which oblige us to reject it (at least for Germanic middles). Both the ungrammatical status of adverbless middles in Germanic and the fact that adverbial modification is not cross-linguistically an uncompromisable requirement argue against this approach. I have concluded that, for the languages that impose this requirement, we need some version of the ‘structural’ account, in the spirit of Roberts (1987), Pitz (1988), and Hoekstra & Roberts (1993). I have argued that the reason adverbial modification is necessary in Germanic, contrary to Greek and French, is structural. The admissible middle-modifiers are required in order for the middle-Agent to be recoverable via identification with the Experiencer argument contributed by the adverb. In conjunction with the dispositional semantics ascribed to middles, this proposal makes predictions about the set of admissible middle-modifiers. Needless to say, there remain several questions which have not been treated extensively here. For one thing, more needs to be said about the nature of the identification process that is effected by the adverb. More in general, further in. Examples like (i), brought up by a reviewer, are a problem for the generalization that easily is always present in middles (but not for the generalization that an adverb is required). (i)

This book reads quickly but not easily.

A similar complication is presented by cases such as (ii), which involves no implicit easily: (ii) This dress does not button because it has a zipper. Sentences like (ii) lack the modal overtones characteristic of the core cases of middles and thus require something additional to be said regardless of the lack of the adverb.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.28 (1577-1682)

 Marika Lekakou

vestigation is required in order to bring to the fore which property sets (79) apart from (80). (79) *This book doesn’t read. (80) This book doesn’t translate.

References Abraham, Werner (1995). Diathesis: The middle, particularly in West-Germanic. In Werner Abraham, Talmy Givón, & Sandra A. Thompson (Eds.), Discourse Grammar and Typology. Papers in Honour of John W. M. Verhaar (pp. 3–47). Amsterdam: John Benjamins. Ackema, Peter, & Maaike Schoorlemmer (1994). The middle construction and the syntaxsemantics interface. Lingua, 93, 59–90. Ackema, Peter, & Maaike Schoorlemmer (1995). Middles and non-movement. Linguistic Inquiry, 26(2), 173–197. Ackema, Peter, & Maaike Schoorlemmer (2005). Middles. In Martin Everaert & Henk van Riemsdijk (Eds.), The Blackwell Companion to Syntax, Vol. 3 (pp. 131–203). Oxford: Blackwell. Authier, Jean-Marc, & Lisa Reed (1996). On the Canadian French middle. Linguistic Inquiry, 27(3), 513–523. Baker, Mark (1988). Incorporation: A Theory of Grammatical Function Changing. Chicago, IL: University of Chicago Press. Berman, Arlene (1973). Adjectives and Adjective Complement Constructions in English. Ph.D. dissertation, Harvard University, Cambridge, MA. Bhatt, Rajesh (1998). Obligation and possession. MIT Working Papers in Linguistics, 32, 21–40. Brennan, Virginia (1993). Root and Epistemic Modal Auxiliary Verbs. Ph.D. dissertation, University of Massachusetts. Cabredo-Hofherr, Patricia (1997). The German Middle Construction. DEA thesis, Université Paris 7. Carlson, Gregory, & Francis Jeffrey Pelletier (1995). The Generic Book. Chicago, IL: Chicago University Press. Chierchia, Gennaro (1995). Individual level predicates as inherent generics. In Carlson & Pelletier (1995), 176–237. Cinque, Guglielmo (1999). Adverbs and Functional Heads: A Crosslinguistic Perspective. Oxford: Oxford University Press. Condoravdi, Cleo (1989). The middle: Where semantics and morphology meet. MIT Working Papers in Linguistics, 11, 18–30. Dobrovie-Sorin, Carmen (2005). The SE-anaphor and its role in argument realization. In Martin Everaert & Henk van Riemsdijk (Eds.), The Blackwell Companion to Syntax, Vol. 4 (pp. 118–179). Oxford: Blackwell. Dowty, David R. (2000). The semantic asymmetry of ‘argument alternations’ (and why it matters) Groninger Arbeiten zur germanistischen Linguistik 44. URL: http://www.ling.ohiostate.edu/∼dowty Fagan, Sarah (1992). The Syntax and Semantics of Middle Constructions. Cambridge, MA: Cambridge University Press.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.29 (1682-1803)

Adverbial modification in middles 

Fellbaum, Christiane (1986). On the Middle Construction in English. Bloomington, IN: Indiana University Linguistics Club. Geuder, Wilhelm (2000). Oriented Adverbs. Issues in the Lexical Semantics of Event Adverbs. Ph.D. dissertation, University of Tübingen. Goldberg, Adele, & Farrell Ackerman (2001). The pragmatics of obligatory adjuncts. Language, 77(4), 798–814. Greenberg, Yael (2002). Two types of quantificational modalized genericity, and the interpretation of bare plural and indefinite singular NPs. In Brendan Jackson (Ed.), Proceedings of SALT 12. Cornell University, Ithaca: CLC. Higginbotham, James (1985). On semantics. Linguistic Inquiry, 16, 547–593. Hoekstra, Teun, & Ian Roberts (1993). Middle constructions in Dutch and English. In Eric Reuland & Werner Abraham (Eds.), Knowledge and Language II. Lexical and Conceptual Structure (pp. 183–220). Dordrecht: Kluwer. Jackendoff, Ray (1972). Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press. Keyser, Samuel Jay, & Thomas Roeper (1984). On the middle and ergative constructions in English. Linguistic Inquiry, 15(3), 381–416. Krifka, Manfred (1998). The origins of telicity. In Susan Rothstein (Ed.), Events and Grammar (pp. 197–235). London: Kluwer. Krifka, Manfred, Pelletier, Francis Jeffrey, Gregory Carlson, Alice ter Meulen, Goedehard Link, & Gennaro Chierchia (1995). Genericity: An introduction. In Carlson & Pelletier (1995), 1–124. Laca, Brenda (1990). Generic objects. Lingua, 81, 25–46. Lekakou, Marika (2002). Middle semantics and its realization in English and Greek. In UCL Working Papers in Linguistics 14 (pp. 399–416). URL: http://www.phon.ucl.ac.uk/ publications/WPL/02papers/lekakou.pdf Lekakou, Marika (2003). Greek passives on the middle interpretation. In Georgia Catsimali, Alexis Kalokairinos, Elena Anagnostopoulou, & Ioanna Kappa (Eds.), Proceedings of the 6th International Conference on Greek Linguistics. Lekakou, Marika (2004). Middles as disposition ascriptions. In Cécile Meier & Matthias Weisgerber (Eds.), Proceedings of Sinn und Bedeutung 8 (pp. 181–196). Konstanz: Universität Konstanz, FB Linguistik. URL: http://www.ub.uni-konstanz.de/kops/volltexte/2004/1383/ Lekakou, Marika (2005a). In the Middle, Somewhat Elevated. The Semantics of Middles and its Crosslinguistic Realization. Ph.D. dissertation, University College London. URL: http://ling.auf.net/lingBuzz/000122 Lekakou, Marika (2005b). Reflexives in contexts of reduced valency: German vs. Dutch. In Marcel den Dikken & Christina Tortora (Eds.), The Function of Function Words and Functional Categories (pp. 155–185). Amsterdam: John Benjamins. Marantz, Alec (1984). On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. Marelj, Marijana (2004). Middles and Argument Structure across Languages. Ph.D. dissertation, Utrecht: Utrecht Institute of Linguistics OTS. McConnell-Ginet, Sally (1994). On the non-optionality of certain modifiers. In Mandy Harvey & Lynn Santelmann (Eds.), Proceedings of SALT 4 (pp. 230–250). Ithica: Cornell University. Nilsen, Øystein (2000). The Syntax of Circumstantial Adverbials. Oslo: Novus Press. Parsons, Terence (1990). Events in the Semantics of English: A Study of Subatomic Semantics. Cambridge, MA: MIT Press. Pitz, Anneliese (1988). Middle constructions in German. Trondheim Working Papers in Linguistics, 6, 1–30.

JB[v.20020404] Prn:6/11/2006; 9:31

F: LA9607.tex / p.30 (1803-1876)

 Marika Lekakou

Rapoport, Tova R. (1999). The English middle and agentivity. Linguistic Inquiry, 30, 147–155. Reinhart, Tanya (2000). The theta system: Syntactic realization of verbal concepts. UiL OTS Working Papers in Linguistics. Utrecht: Utrecht Institute of Linguistics. Rizzi, Luigi (1986). Null objects in Italian and the theory of pro. Linguistic Inquiry, 17, 501–557. Roberts, Ian (1987). The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris Publications. Scheiner, Magdalena (2003). Temporal anchoring of habituals. In Marjo van Koppen, Joanna Sio, & Mark de Vos (Eds.), Proceedings of ConSole XI . URL: http://www.sole.leidenuniv.nl Sioupi, Athina (1998). Domes mesis diathesis. Mia sigritiki meleti elinikis–germanikis (Middle constructions: A comparative study of Greek and German). Ph.D. dissertation, University of Athens. Steinbach, Markus (2002). Middle Voice, Amsterdam: John Benjamins. Stroik, Thomas (1992). Middles and movement. Linguistic Inquiry, 23, 127–137. Tsimpli, Ianthi (1989). On the properties of the passive affix in modern Greek. UCL Working Papers in Linguistics, 1, 235–260. Van Geenhoven, Veerle (2003). The semantic diversity of characterizing sentences. Talk given at the 14th Amsterdam Colloquium, Amsterdam December 2003. Van Oosten, Jeanne (1977). Subjects and agenthood in English. Chicago Linguistic Society, 13, 451–471. Williams, Edwin (1981). Argument structure and morphology. The Linguistic Review, 1(1), 81– 114. Wurmbrand, Susi, & Jonathan David Bobaljik (1999). Modals, raising and A-reconstruction. Handout of talk given in Leiden University, October 1999. URL: http://wurmbrand. uconn.edu/research/files/Leiden-1999.pdf Zribi-Hertz, Anne (1993). On Stroik’s analysis of English middle constructions. Linguistic Inquiry, 24, 583–589. Zribi-Hertz, Anne (2003). On the syntax and semantics of the French mediopassive SE: Some comparative update. Talk given in Amsterdam, January 2003.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.1 (46-133)

From passive to active* Syntactic change in progress in Icelandic Joan Maling This paper discusses the results of a nationwide study of a syntactic change underway in Iceland. The new construction appears to contain passive morphology, auxiliary vera (‘to be’) and a passive participle which can assign accusative case to a postverbal argument. The study was designed to test the hypothesis that the innovative construction involves the reanalysis of passive morphology as a syntactically active voice construction with a phonologically null impersonal subject. Such a reanalysis parallels the completed development of the -no/to construction in Polish and the autonomous form in Irish. I show that verbal morphology can be ambiguous between active and passive voice, and speculate about the reasons why this change is happening in Icelandic but in none of the other Germanic languages.

.

Introduction

In the last half century, a new construction has sprung up spontaneously in various parts of Iceland, giving us a rare opportunity to investigate a major syntactic change at a relatively early stage. I will refer to this innovative construction, illustrated in (1a), as the “new construction.” In this paper I discuss the significance of * Most of the research reported here was supported by grants from Vísindasjóður Rannsóknarráðs Íslands (RANNÍS), Rannsóknarsjóður Háskóla Íslands, and Lýðveldissjóður to Sigríður Sigurjónsdóttir; the pilot study was supported in part by NSF grant 9223725 to Brandeis University. The material is based in part on work done while I was serving as Director of the Linguistics Program at the U.S. National Science Foundation. Any opinion, findings, and conclusions expressed in this material are those of the author, and do not necessarily reflect the views of the U.S. National Science Foundation. I am grateful to Elisabet Engdahl, Helgi Skúli Kjartansson, Knud Lambrecht, Svein Lie, Kjartan Ottósson, Halldór Ármann Sigurðsson, Sigríður Sigurjónsdóttir, Torgrim Solstad, Øystein Vangsnes, and an anonymous reviewer for helpful comments on various points.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.2 (133-200)

 Joan Maling

the results of a nationwide study conducted together with Sigríður Sigurjónsdóttir (see Maling & Sigurjónsdóttir 2002; Sigurjónsdótttir & Maling 2002) for our understanding of grammatical voice, arguing that despite its historical origins as a morphological passive, the new construction is best analyzed as a syntactically active impersonal. New construction Það var beðið mig að vaska upp itEXPL was asked-neut me-acc to wash up “They asked me to do the dishes” Canonical Passive b. Ég var beðinn að vaska upp I-nom was asked-masc to wash up “I was asked to do the dishes” Impersonal Passive í kringum jólatréð c. Það var dansað itEXPL was danced-neut around the.Christmas.tree “People danced around the Christmas tree”

(1) a.

Sentence (1a) is widely accepted by adolescents around the country, but sharply ungrammatical in the standard language, as shown by the results of our study, where over 73% of the youngsters (n = 1695) judged the sentence to be acceptable, as compared to under 8% of the adults (n = 200). Sentence (1b) is the canonical passive, while (1c) exemplifies the so-called “impersonal passive.” . Descriptive background The “impersonal passive” standardly occurs only with intransitive verbs, including verbs taking PP or clausal complements; the new construction, on the other hand, occurs with transitive and ditransitive verbs as well. For all speakers, expletive það is not a grammatical subject; it serves only to satisfy the Verb-Second constraint, which is common to all Germanic languages other than English. Thus það does not appear in either yes-no questions or clauses with another constituent in clauseinitial position. (2) Var (*það) beðið þig að vaska upp? was (*it) asked you-acc to wash up? “Did they ask you to do the dishes?”

New Construction

The standard Icelandic passive has the same basic syntactic properties as its counterparts in other Germanic languages, but adds much richer agreement and a greater variety of morphological case-marking. If the main verb takes an accusative object, that argument will correspond to a nominative subject in the passive voice, which triggers agreement with both the finite verb and the participle. If the verb governs lexical case on its object, e.g. dative, as in (3), that case is preserved in the passive; for definite NPs, movement to subject position is still obligatory (3b),

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.3 (200-267)

From passive to active 

giving rise to one class of the oblique subjects for which Icelandic is famous (see Zaenen et al. 1985; Sigurðsson 1989; inter alia): (3) a.

Það var sagt mér að vaska upp. itEXPL was told me-dat to wash up b. Mér var sagt að vaska upp. I-dat was told to wash up “I was told to do the dishes”

New Construction Canonical Passive

In general, in both passive and active voice sentences, subject NPs can occur in postverbal position only if they are indefinite: (4) a.

Það voru seldir margir bílar í gær. itEXPL were sold-masc.pl many cars-masc.pl yesterday “Many cars were sold yesterday” b. *Það voru seldir bílarnir í gær. itEXPL were sold-masc.pl the.cars-masc.pl yesterday

Thus the new construction has surface properties of both the standard passive and the active voice: It has the verb forms of the passive, but the accusative case and postverbal position of the pronoun mig ‘me’ in (1a) is like the active. The morphosyntactic innovations of the new construction are listed in (5): (5) morphosyntactic innovations in the new construction a. accusative case on underlying object b. lack of verb agreement with nominal argument c. postverbal position: no NP-movement to subject position d. lack of the definiteness effect

The auxiliary and past participle are invariant 3rd person singular, neuter. In standard Icelandic, NP-movement to subject position is obligatory in the passive; only an indefinite subject could occur after the participle. This syntactic innovation is a system-internal change, and cannot be attributed to borrowing. Although a similar syntactic change took place independently in Polish and Irish several centuries ago, none of the languages commonly spoken by Icelanders have the construction, so the usual scapegoats for linguistic contamination, namely, Danish and English, cannot be blamed. The new construction apparently dates back only a few decades. The oldest known “sighting” is from 1959. (6) Það var bólusett okkur. itEXPL was inoculated us-acc Intended: “They inoculated us”

(1959, girl born 1951)

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.4 (267-307)

 Joan Maling

A teacher recalls having heard her 8-year old niece from Akureyri say this sentence in 1959, but the girl’s mother doesn’t think the construction was common at the time (Haraldsdóttir 1997: 22). Twenty years later, however, the new construction was reported to be common in Akureyri, the “capital of the north”. In 1984, Helgi Hálfdanarson’s usage manual Gætum tungunnar [Let’s watch our language] included the new construction as language error #174, a usage to be corrected in children’s speech, but in general the new construction has gone either unnoticed or ignored by prescriptivists. Although elementary school teachers certainly recognize and correct this construction, the school system as a whole has not (yet) begun to marshal its forces against the “new impersonal” as they have against the so-called Dative sickness (þágufallssýki), which began around the turn of the 20th century (Svavarsdóttir 1982). Dative sickness refers to the substitution of dative case for standard accusative (or in a few instances nominative) on the experiencer subject argument of certain nonagentive verbs. The new construction is being observed in the usage of both children and some adult speakers, typically in the spoken language but also in writing. The adult speakers born in the 1940’s and 1950’s who have been observed using the new construction all seem to have grown up outside Reykjavík; this is consistent with our data, which show that the “new impersonal” is more widespread outside of the capital city. The construction is not limited to pronouns, and is also found in embedded clauses, both indicative and subjunctive. In the grammars of both children and adolescents, it co-exists with the canonical passive. Although it is likely that there are discourse differences between the “new impersonal” and the standard passive, they seem to be equivalent in terms of their truth conditions. Particularly interesting is the example in (7), from a sign posted at the Háskólabíó movie theater in Reykjavík in September, 2004. (7) Skoðað verður miða við innganginn. inspected-neut will.be tickets-acc on entrance “Tickets will be inspected on entering”

Whoever made the sign combined the innovative (substandard) accusative case on the postverbal patient argument with the (formal) “Stylistic fronting” of the past participle, which occurs only in clauses without a grammatical subject (Maling 1980). . How to analyze the new construction: Two hypotheses What are the syntactic properties of the new construction? How should it be analyzed? There are two basic classes of analyses, as sketched in (8).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.5 (307-367)

From passive to active 

(8) a.

Passive without NP-movement: [IP [e] Aux [VP Vppart (NP)]] b. Active Impersonal [IP proarb Aux [VP Vppart (NP)]]

Under the first analysis, the new construction is a variant of the standard passive in which NP-movement to subject position fails to apply, and the NP gets assigned accusative in situ, presumably by the verbal participle. This is essentially the analysis that Sobin (1985) gives to the accusative-assigning participial construction in Ukrainian. Under the second analysis, the new construction is a syntactically active construction with a null [+human] pronoun subject. Note that Icelandic is not a null-subject language; personal pronoun subjects are as obligatory as they are in English. However, Icelandic does have null subjects in clauses containing weather verbs and in various impersonal constructions.

. Syntactic change in the English auxiliary system Across languages, we find constructions that are clearly active voice, constructions that are clearly passive voice, and constructions where we cannot easily tell. So what is this new construction? The verb form looks like a passive, but the accusative case and the postverbal position of the “undergoer” argument resemble an active voice construction. If the surface properties of morphological case and agreement are not enough to distinguish between the two analyses, how do we decide which is the best analysis? Verbal morphology is often ambiguous. Let me illustrate this point with an example from the history of English. Consider the italicized verb forms in the following sentences: (9) a.

“The clock struck ten while the trunks were carrying down.” (1818, Jane Austen, Northanger Abbey, p. 155) b. “She only came on foot, to leave more room for the harp, which was bringing in the carriage.” (1818, Jane Austen, Persuasion, p. 50) c. “I met a dead corpse of the plague, just carrying down a little pair of stairs.” (Samuel Pepys, diarist, 1633–1703)

The lack of a passive auxiliary in the progressive examples in (9) renders them virtually undecipherable to the modern ear. While we eventually recognize that the italicized verb forms must be passive in meaning, it is clearly not the verbal morphology that provides the necessary clues. A sequence like were carrying was ambiguous between active and passive voice: (10) a. The trunks were carrying down the stairs. b. The men were carrying down the trunks.

(trunks=subject) (trunks=object)

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.6 (367-409)

 Joan Maling

The examples in (9) reflect a relatively recent change in the English auxiliary system; progressive passives didn’t appear until the very end of 18th century. The earliest known example is from 1795, cited by Visser (1973: §2158): (11) “a fellow whose uttermost upper grinder is being torn out by the roots by a mutton-fisted barber” (1795 Robert Southey, Life & Correspondence)

The two constructions co-existed for about a century, and an author might use both constructions in a given work.1 According to Visser, the grammatical subject of the older form was generally inanimate rather than animate, hence clearly not the agent. The change might be understood as simplification of the phrase structure rule for the auxiliary. Prior to the change, English had a templatic syntax which allowed for only one of the two distinct be-auxiliaries, one progressive, one passive, in a given verb phrase. If progressive aspect was to be conveyed, then the passive auxiliary was simply not allowed. The innovation consisted of allowing the two be-auxiliaries to co-occur. Gradually over the course of the 19th century, the new form took over, largely driving out the older form. What is relevant here is that during the century in which the two forms co-existed, the verbal morphology was ambiguous as to grammatical voice. If the verbal morphology is sometimes ambiguous even in a familiar language like English, how can we possibly tell whether a construction is active or passive just by looking at the verb form in an unfamiliar language like Northern Pomo, or even in Indo-European languages such as Irish or Polish/Ukrainian or Icelandic?

. Grammatical properties of passive vs. active voice . Discourse properties of the passive voice How do speakers choose between using active/passive voice? Grammarians have long noted two distinct discourse properties of the passive voice across languages:2 (12) Discourse properties of the passive voice a. to make the patient (if any) the discourse “theme” of the sentence; b. to avoid naming the agent . The construction is now so well-established that it is difficult to imagine the depth of feeling that this syntactic innovation engendered among language purists of the day. For example, R. Grant White, in Words and their Uses (1871: 336) declares that the new idiom “is the most incongruous usage of words and ideas that ever attained respectable usage in any civilized language” (cited by Visser 1973: 2427). See Visser for further discussion. . For discussion of the discourse properties of the Icelandic passive, see Kress (1982: 150) and Ottósson (1986: 97).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.7 (409-475)

From passive to active 

Since the patient remains in postverbal position, the Icelandic innovative construction cannot plausibly serve the purpose of making this argument the discourse “theme” or “topic”. Let us consider the second function of the passive voice. Kress (1982: 150) calls this function making the agent “anonymous”; Haspelmath (1990) describes it as backgrounding or desubjectivizing the agent. Using the standard passive is certainly one way of achieving this goal. For example, while a politician might concede that “Mistakes were made” using the passive voice, the same spokesperson would be very unlikely to say “We made mistakes” using the active voice unless talking about the opposition: “My opponent made many mistakes.” However, the same goal of making the agent anonymous can be achieved in other ways, for example, by using an active voice construction with an impersonal pronoun subject, e.g. nonspecific ‘they’ or ‘you’ in English:3 (13) a. English is spoken here. b. They speak English here, don’t they?

(passive voice) (active voice)

Space limitations preclude a more detailed discussion; it suffices to note that both the impersonal active and the passive can serve to background the agent (Haspelmath 1990: 49). If the line between the two is so nebulous from the point of view of discourse function, does it matter whether we call the new construction passive or not? My answer is that it matters very much. This terminological controversy is important because syntactically, there are sharp differences between the two grammatical voices. As Haspelmath observes, “The difference between passive and desubjective is of a syntactic rather than a semantic nature” (1990: 58). . Syntactic properties of active vs. passive voice clauses So what are these syntactic differences? Based on her study of the Polish and Ukraininan -no/to-construction and the Irish autonomous construction, Maling (1993) selected the four syntactic properties shown in (14) to use as diagnostics. (14) Syntactic properties of active construction with impersonal subject a. No agentive by-phrase is possible b. Binding of anaphors (reflexive and reciprocal) is possible c. Control of subject-oriented adjuncts is possible d. Nonagentive (“unaccusative”) verbs can occur in the construction.

A syntactically active impersonal construction with a grammatical subject, e.g. French on or German man, has all four of these properties; the standard passive . English sentences with one, e.g. “One shouldn’t tell a lie,” are quite formal in register, and therefore will not be cited here. See Huddleston & Payne (2002: 427).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.8 (475-554)

 Joan Maling

Table 1. Syntactic properties of active vs. passive voice clauses Syntactic property

Active

Passive

agentive by-phrase bound anaphors in object position control of subject-oriented adjuncts nonagentive (“unaccusative”) verbs

* ok ok ok

ok * * *

construction lacks all four properties.4 Thus we observe the sharply contrasting behavior shown in Table 1. The hypothesis of Maling & Sigurjónsdóttir (2002) was that the new construction in Icelandic is being reanalyzed as an active construction with an impersonal pronoun subject. The impersonal subject in this construction is null, but as a thematic subject, it is syntactically accessible to syntactic rules which refer to the grammatical subject of a clause. Most of these syntactic properties are well-known enough not to need further discussion. Furthermore, the grammaticality judgments are robust, except for subject-oriented adjuncts. Lack of an appropriate grammatical subject gives rise to the phenomenon known as the “dangling modifier”. Although generally considered ungrammatical by both prescriptivists and generative linguists, they do occur with some frequency. Nonetheless, native speakers recognize that there is something odd about sentences like those in (15). (15) a. ?*Laughing, the children were silenced by a glance from the teacher. b. *Laughing, there was a pillow fight.

Speakers find it difficult to say just who is supposed to be laughing. This intuition is robust enough that it was used as a diagnostic in Maling & Sigurjónsdóttir (1997, 2002).

. Cross-linguistic comparison: Passive morphology reanalyzed as active As mentioned earlier, a similar syntactic development has occurred independently in a number of languages. As discussed in Maling (1993), the so-called autonomous form in Irish and the -no/to construction in Polish each developed from a canonical morphological passive. In both these languages, the innovative construction has the syntactic properties listed in (14) in addition to the overt morphological properties of accusative case-marking and nonagreeing verb . As pointed out by Torgrim Solstad (personal communication), unaccusative verbs can form impersonal passives in German. A Google search turns up examples like Es wurde gestorben auf beiden Seiten “it was died on both sides.” Further investigation is needed.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.9 (554-624)

From passive to active 

(cf. Stenson 1989 for Irish; Dziwirek 1991; Lavine 2000; and various references cited in Billings & Maling 1995 for Polish). In Polish, the syntactic behavior of the canonical passive contrasts sharply with that of the accusative-assigning participial -no/to construction illustrated in (16a). The canonical passive has all the syntactic properties of a true passive expected if the underlying syntactic representation is something like (8a). An agentive by-phrase is allowed, but there’s no thematic subject to serve as a binder for bound anaphors, or as a controller for various subject-oriented adjuncts (nor can the underlying agent serve as the controller). The -no/to construction has the opposite properties. Finally, the Polish -no/to construction does not observe the 1-Advancement Exclusiveness Law (1AEX) of Relational Grammar (Perlmutter & Postal 1984), which rules out passives of unaccusative predicates; the only lexical restriction is a semantic one: The understood subject must be [+human]. To summarize, despite its historical origin as a morphological passive, the innovative -no/to construction in Polish now behaves syntactically like German man-sentences or French on-sentences, the only difference being that the impersonal pronoun subject is null. Particularly instructive is the contrast between the Polish -no/to construction and the superficially similar cognate construction in Ukrainian.5 (16) Two accusative-assigning -no/to participial constructions a. Polish (Maling & Sigurjónsdóttir 2002, ex. 8b) ´ atyni˛ Swi ˛ e zbudowano w 1640 roku. church-acc build-no in 1640 year b. Ukrainian (Sobin 1985: 653) Cerkvu bulo zbudovano v 1640 roc’i. church-f.acc was build-no in 1640 year ‘The church was built in 1640’

The Ukrainian -no/to construction discussed by Sobin (1985) differs from its Polish counterpart with respect to all four syntactic properties (cf. Billings & Maling 1995; Lavine 2000, 2001). The contrasting behaviors are summarized in Table 2; Polish and Ukrainian examples illustrating these properties can be found in Maling & Sigurjónsdóttir (2002). The systematic differences between the canonical passive and the -no/to construction in Polish would be unexplained if both constructions are analyzed as passive. The moral of this comparison is that we cannot tell what the syntactic be. The forms -n- and -t- in Polish and Ukrainian are allomorphs of the past passive morpheme; the -o, once the neuter singular inflection, is now invariant. See Lavine (2000: Ch. 3) for discussion of the morphological status of this ending. Lavine attributes the contrasting syntactic behavior to the presence (Ukrainian) vs. absence (Polish) of an auxiliary verb. The innovative Icelandic construction poses a challenge to this account.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.10 (624-702)

 Joan Maling

Table 2. Syntactic properties of various constructions in Polish and Ukrainian Syntactic property

Pol/Ukr passive

Polish -no/to

Ukrainian -no/to

agentive by-phrase bound anaphors in object position control of subject-oriented adjuncts nonagentive (“unaccusative”) verbs

ok * * *

* ok ok ok

ok * * *

havior of a construction is by looking at its superficial morphological properties (e.g. case, agreement). Despite their common historical origin and superficial similarity (i.e. the shared morphological properties of assigning accusative case and consequent lack of agreement), the Polish and Ukrainian constructions are polar opposites in terms of syntactic behavior. The obvious question, then, is: Which of the two polar opposites does the innovative Icelandic construction most resemble?

. The nationwide survey We developed a questionnaire to test the syntactic behavior of the innovative construction. Our questionnaire was a revised version of a pilot study conducted in the spring of 1996 (Maling & Sigurjónsdóttir 1997). An extensive nationwide study was conducted in the fall and winter of 1999–2000; the results are reported in Maling & Sigurjónsdóttir (2002) and Sigurjónsdóttir & Maling (2002).6 The population of Iceland is approximately 286,000; the population of Greater Reykjavík is approximately 178,000, more than half the population of the country. However, because a major goal of our study was to determine the geographical spread of the change, only about a third (583 of 1695) subjects in our study lived in Greater Reykjavík. The questionnaire was distributed to 1,731 tenth graders (age 15–16) in 65 schools throughout Iceland; this number represents 45% of those born in the country in 1984. Tenth grade is the last year of compulsory education in Iceland. The questionnaire was also given to 205 adult controls in various parts of the country. After excluding subjects who made more than one error on the ungrammatical control sentences, we had results from 1695 students, 845 males and 850 females, and from 200 adults. Table 3 shows the geographic variation in the acceptability rates for the “new impersonal” in the test sentences containing animate accusative and dative objects, simple examples of the innovative construction like “it was hit me” or “it was . See these articles for a more detailed discussion of the results and methods. The 68 test sentences, with glosses and English translations, can be found in the Appendix to Maling & Sigurjónsdóttir (2002).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.11 (702-728)

From passive to active 

Table 3. Geographical variation in acceptance of the new construction Geographical region

Accusative Objects

Dative Objects

7 regions outside Reykjavík Outer Reykjavík Inner Reykjavík

51–68% 53% 28%

57–75% 60% 35%

told me to do the dishes”. As expected, many students judged the new construction as something they might say. Our results revealed a statistically significant relationship between geographical region and the acceptability judgments. There was a clear difference between Reykjavík and the rest of the country; the difference was even more striking once we divided Reykjavík into the two parts, which we called Inner and Outer Reykjavík, based on the location of the schools. Subjects in Outer Reykjavík were nearly twice as likely to accept such examples of the “new impersonal” as subjects in Inner Reykjavík. This effect was highly significant; however, the difference between Outer Reykjavík and the rest of the country excluding Inner Reyjavík was not significant.7 This result justified combining the results for adolescents into two groups, Inner Reykjavík vs. the rest of the country, labeled “Elsewhere.” For adults, however, there was no significant effect between geographical region and acceptability judgments. In the tables below, the results for adults are reported in the right-most column; in the middle column are the results for adolescents in Inner Reykjavík, where the innovative construction is less advanced; and the results for all other adolescents are in the first column, labeled “Elsewhere.” Inner Reykjavík consists of the old downtown, or city center, and the newer western part of the city, including the independent municipality of Seltjarnarnes. Sociological data from 1999 indicate that the populace of Inner Reykjavík has the highest percentages of university education in the country (Morgunblaðið, 20 February, 2001, p. 13), and the students in these schools score the highest on the nationwide exams (Meðaleinkunnir skóla, Námsmatsstofnun). Our data showed a highly significant effect for the education levels of both mother and father at all levels (10 years of schooling (compulsory education), 14 years of school (through framhaldsskóli), and university level) and for all geographical regions.8 The higher the level of education, the lower the acceptance of the new construction. Multiple regression analysis, however, showed that it is geographical region that affects the acceptability the most, independently of parental education. The fact that the . As pointed out by an anonymous reviewer, Reykjavík has expanded enormously in the last 10–15 years, primarily the result of people moving from the countryside to the outskirts of Reykjavík. . Detailed documentation and references can be found in Sigurjónsdóttir & Maling (2002).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.12 (728-779)

 Joan Maling

syntactic change is most advanced outside of Reykjavík is interesting for at least two reasons. First, it shows that this change cannot be attributed to the corrupting influence of the big city, once a common theme in Icelandic literature. Second, it is unexpected given the widespread view among variationists that change is urban. The spread of this Icelandic construction runs counter to the view that “cities have always been at the center of linguistic innovation” (Labov 1994: 23). . Reliability of judgments It might be questioned whether 15–16 year old adolescents are capable of making reliable grammaticality judgments. Any such objection can be countered by noting that for the control sentences, both grammatical and ungrammatical, adolescents gave similar responses to the adults. For the eleven grammatical control sentences, the mean acceptance rate in the various geographical areas ranges from a low of 89% to high of 94% for adolescents, as compared to between 92% and 96% for adults. There was almost no difference among the different regions of the country, and no difference between Inner and Outer Reykjavík. Our data show that the innovative construction co-exists with canonical passive, just as in Polish, but unlike Irish where the innovative autonomous form eventually drove out the canonical passive. . Morphological case One of the well-known properties of Icelandic is that lexically case-marked NPs behave syntactically exactly like NPs bearing syntactic case; they differ only in that lexical case is preserved under NP-movement.9 Although we might expect the same to be true of the “new impersonal,” morphological case turns out to have a significant effect on acceptability. Kjartansson (1991: 18) speculated that the new construction was more common with verbs governing dative than with verbs governing accusative. Our results strongly support this observation. This result is consistent with the observations for Ukrainian/Polish that the change began with those forms where the morphological signs of nonagreement are least obvious. Recall that for verbs governing dative objects, as in (3a), only the violation of the definiteness effect marks a sentence as an example of the new construction; for verbs governing accusative objects, as in (1a), there are in addition the difference in morphological case and the consequent lack of agreement. . Case-preservation is usually attributed to the lexical/inherent case-marking associated with the relevant argument in the lexicon (Zaenen & Maling 1984; Zaenen et al. 1985). For a different account, see Svenonius (2005).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.13 (779-851)

From passive to active 

Table 4. Acceptance rates for the new construction as function of animacy vs. morphological case on the object

ACC DAT

Elsewhere Anim

Inanim

Inner Reykjavík Anim Inanim

74% 74%

54% 50%

43% 43%

19% 20%

Although subjects consistently liked dative objects more than accusative ones, morphological case is clearly not always the deciding factor. Examples of the innovative construction were judged more acceptable if the object was animate, or more precisely, [+human]. One might speculate that the higher acceptability of dative objects might be attributed to this preference for [+human] objects. As discussed by Barðdal (1993), many transitive verbs assign either dative or accusative, depending on the animacy of the object (see also Maling 2002). This is illustrated by the following examples taken from Barðdal (1993: 4, ex. 6a, b). (17) a.

Kristín Christine b. Kristín Christine

þvoði washed þvoði washed

handklæðið. the.towel-acc barninu. the.child-dat

However, our data show a strong preference for animate objects in the new construction regardless of morphological case (Table 4). It is striking that all of the examples of the innovative construction previously cited in the literature have human objects. It is not the case that inanimate objects are actually disallowed, since as noted earlier, such examples are attested (see example (7)). It may be that the pragmatics of the innovative construction favor affected objects of highly transitive verbs in the sense of Hopper & Thompson (1980). Further research will be needed to determine whether the relevant factor is animacy or a high degree of transitivity.10 . Testing individual predictions Our hypothesis was that the innovative construction is in the process of acquiring the syntactic properties listed in (14). First, we included a few sentences to test whether the postverbal NP in the new construction could possibly be analyzed as a grammatical subject. The data showed clearly that the accusative NPs cannot occur in subject position, namely between the finite verb and the participle, either in direct yes-no questions or in declaratives. . Engdahl (1999, this volume) reports that the choice between the s-passive and the bli-passive in Swedish is also sensitive to animacy.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.14 (851-929)

 Joan Maling

Table 5. Agentive by-phrase in grammatical control sentences Agentive by-phrase in grammatical control sentences a. b.

Honum var sagt upp af forstjóranum he-dat was fired PRT by the.director Það var samþykkt af öllum í bekknum að fara it was agreed by all in the.class to go í keilu bowling

Elsewhere Inner Rvík Adults 87%

93%

90%

94%

92%

94%

.. Agentive by-phrase Although overt agentive by-phrases are much less common in Icelandic than in English, they are grammatical in the canonical passive. To test whether subjects accept by-phrases, we included in the control sentences two canonical passives with an overt by-phrase, one sentence containing a transitive verb governing a dative object, the other a verb taking an infinitival complement. The results shown in Table 5 indicate that by and large adolescents accept such sentences as fully grammatical, just like adults. In a syntactically active sentence, on the other hand, co-occurrence of an agentive by-phrase with the thematic subject (either overt or phonologically null) would constitute a Theta-Criterion Violation. Recall that in Polish, agentive byphrases are fine in the canonical passive, but robustly ungrammatical in the -no/to construction, where native speakers report that a by-phrase is simply “redundant” (see Table 2). Thus if the innovative construction is syntactically active as hypothesized, the presence of a by-phrase is predicted be ungrammatical.11 The questionnaire contained two examples of the new construction designed to test this prediction. The results shown in Table 6 indicate that this prediction is largely confirmed; while not fully ungrammatical among speakers who accept the “new impersonal”, the by-phrase significantly lowers the acceptability of the new construction (compare these percentages with the percentages shown in Table 4 for sentences without a by-phrase). The contrast is shown clearly by the minimal pair in Table 7, where (a) is a canonical passive, and (b) is unambiguously the new impersonal construction; both contain overt by-phrases. Adolescents in Inner Reykjavík wouldn’t be expected to like (b) very much with or without the by-phrase, because these sentences are unambiguously instances of the new construction, but even here the . The distributional facts can be complicated during language change, especially in language contact situations. Kaiser & Vihman (this volume) compare and contrast the grammatical properties of the Estonian impersonal with the innovative personal passive.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.15 (929-1028)

From passive to active

Table 6. Agentive by-phrase in the innovative construction Agentive by-phrase a. b.

Það itEXPL Það itEXPL

var was var was

Elsewhere Inner Rvík Adults skoðað bílinn af bifvélavirkjanum inspected the.car by the.mechanic sagt honum upp af forstjóranum fired him PRT by the.director

33%

9%

0%

19%

9%

0%

Table 7. The acceptability of agentive by-phrases Agentive by-phrase a. b.

Honum var sagt upp af forstjóranum he-dat was fired PRT by the.director Það var sagt honum upp af forstjóranum PRT by the.director itEXPL was fired him


93%

90%

19%

9%

0%

presence of an agentive by-phrase clearly makes the sentence significantly less acceptable. .. Binding of anaphors If the subject position in the “new impersonal” construction is a theta-position, then binding of anaphors in nonsubject positions should be possible, since there is a thematic subject to bind such an anaphor. The questionnaire contained thirteen sentences designed to test this prediction, 4 with a plain reflexive sig, 2 with the compound self-anaphor sjálfan sig, 5 with a possessive reflexive, and 2 with a reciprocal. The results for sig-anaphors are given in Table 8. Our results indicated that simple reflexive objects in the “new impersonal” construction are judged highly acceptable, just as acceptable, in fact, as nonreflexive objects. This confirmed Sigurðsson’s speculation (1989/1992: 235) that the new construction is best with reflexive verbs. This result is not surprising, since in many languages, verbs with reflexive objects behave syntactically like intransitive verbs (Sells, Zaenen, & Zec 1987). As reported in Table 3, only 28% of adolescents in Inner Reykjavík accepted comparable sentences with nonreflexive accusative objects. Even more striking is that between 30–40% of adult subjects accepted the examples with a reflexive object. Adults were four times as likely to accept a sentence with a reflexive object sig than to accept a sentence with a nonreflexive accusative object, which would unambiguously be an instance of the new construction. Maling & Sigurjónsdóttir (2002) suggested that this represents the first step in the reanalysis of the past participle from passive to syntactically active. As discussed in more detail below, passives of intransitive verbs are syntactically



JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.16 (1028-1107)

 Joan Maling

Table 8. Binding of sig-anaphors sig-anaphor a. b.

c.

Svo var bara drifið sig á ball then was just hurried REFL to the dance innan dyra út af Það var haldið sig doors due to itEXPL was kept REFL in óveðrinu bad.weather um á svæðinu Það var skoðað sig itEXPL was looked REFL around in the.area “People looked around the area”


67%

40%

82%

65%

37%

72%

43%

31%

ambiguous between true passives and syntactically active impersonals. As observed by Haspelmath (1990: 35), “...intransitive desubjectives are indistinguishable from passives of intransitive verbs, so transitive desubjectives are the crucial case.” .. Subject-oriented adjuncts Many speakers find it difficult to interpret the understood agent of a passive as the controller of certain subject-oriented adjuncts. The question then is whether subject-oriented participial adjuncts12 can be used with passives of intransitive verbs, as illustrated in (18). (18) ?Það var dansað skellihlæjandi á skipinu. it was danced laughing.uproariously on the.ship

Sentence (18) was expected to be ungrammatical in the standard language because there is no thematic agent subject to serve as controller for the participial adjunct skellihlæjandi. This was expressed by the intuition of one native speaker that “someone is missing.” On the other hand, we predicted that speakers of the “new impersonal” should feel no such lack of an agent, since by hypothesis, the construction has a thematic subject to serve as syntactic controller for the participial adjunct. The questionnaire contained three sentences designed to test this prediction, two formed from intransitive verbs, one from a transitive verb. The results, shown in Table 9, largely confirmed the expectation. Note that even in Inner Reykjavík, participial adjuncts with intransitive impersonal passives were accepted by ap. Participial modifiers were used because they are invariant, unlike adjectival modifiers which would have to agree with the controller in gender, number (and case). Crucially these modifiers are subject-oriented, not agent-oriented. As is well known, adverbials of purpose (e.g. viljandi ‘on purpose’) can be controlled by the underlying agent in a passive (Zaenen & Maling 1984).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.17 (1107-1257)


Table 9. Subject-oriented Participial Adjuncts Participial Adjuncts a.

b.

Það var farið hágrátandi heim itEXPL was gone crying home “People went home crying” grátandi Það var lesið minningargreinina itEXPL was read the.memorial.article-acc crying “People read the memorial article crying”

Elsewhere Inner Rvík Adults

61%

50%

49%

62%

35%

4%

proximately half the subjects. Not surprisingly, the third sentence with a transitive verb was judged much less acceptable in Inner Reykjavík where the innovative construction is less widespread, and by adults. What is surprising is that approximately 50% of adults also accepted sentences like (a). Barðdal & Molnár consider such sentences grammatical, and suggest that such subject-oriented adjuncts “can be controlled by the underlying agent in impersonal passive sentences which do not contain a thematic subject” (2000:128). Our data suggest that there are actually two different groups of native speakers with different grammars. For both adolescents and adults, there is a highly significant correlation between the results for subject-oriented adjuncts and those for simple reflexives shown in Table 8. The more subject-oriented participles are accepted, the more simple reflexives are accepted. In other words, the similar acceptability rates cannot be explained as an averaging effect. Although it is certainly possible to allow statements in the grammar to the effect that the underlying agent can be a controller for subject-oriented adjuncts, such a constraint would not capture the correlation between this and the acceptance of reflexive objects as our analysis does. .. Unaccusative verbs Passives of unaccusative verbs are sharply ungrammatical in all the Germanic languages that allow intransitive verbs to form passives, including standard Icelandic. To test for changes in lexical restrictions, our questionnaire included five sentences containing unaccusative verbs with nonagent human subjects: detta (‘to fall’),13 koma (‘to arrive’), svitna (‘to sweat’), hverfa (‘to disappear’) and deyja (‘to die’). . Barðdal & Molnár (2000:129) argue that our detta-example in Table 10 (a) is acceptable because speakers interpret it as an agentive and actional verb, i.e. as an unergative. While this is a plausible account for some impersonal passives of unaccusatives (see next footnote), it is dubious for this example since the verb detta ‘to fall’ cannot be used in Icelandic to describe intentional falling, and the semantics of the example we used (slipping on ice) makes an agentive reading implausible.



JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.18 (1257-1269)

 Joan Maling

Table 10. Unaccusative verbs Lexical restrictions: unaccusative verbs a.

b. c.

d. e.

Það var dottið í hálkunni fyrir framan itEXPL was fallen on the.ice in front of blokkina the.building Í morgun var komið of seint í skólann this morning was arrived too late to school Í nótt var ekkert svitnað í last night was not.at.all sweated in svefnpokanum the.sleeping.bag sporlaust í stjörnustríðinu Það var horfið itEXPL was disappeared traceless in the.star.wars Það var dáið í bílslysinu itEXPL was died in the.car.accident

Else-where Inner Rvík Adults

55%

45%

25%

36%

38%

58%

31%

29%

31%

30%

23%

22%

14%

11%

2%

The range of acceptability rates for individual verbs was extremely wide, from a high of 55% with detta (‘to fall’) to a low of 14% for deyja (‘to die’)14 for adolescents outside of Inner Reykjavík. There must be some semantic difference among these unaccusative verbs that accounts for the variable behavior. Verbs denoting change of location may be more acceptable as “impersonal passives” than verbs denoting a (bodily) process or a change of state. Further research is needed to determine exactly what factors are relevant, but overall, the results suggest that the innovating construction is indeed beginning to extend its usage to nonagentive verbs which do not form passives in the standard language. One might speculate that changes in the lexical restrictions on a particular morphology are the last stage in the reanalysis from impersonal passive to a syntactically active construction. But in fact, adults liked many of the examples as much as the adolescents do, and crucially, they accept these examples of impersonal passives to a much greater extent than they accept definite postverbal objects. . As has frequently been pointed out, such unaccusative verbs may form impersonal passives when they can be interpreted as volitional actions, as in the following example from Ottósson (1988: Fn. 5): (i)

Enn er barist og dáið fyrir föðurlandið. still is fought and died for the.fatherland

Since such coerced readings do not bear on our hypothesis, we deliberately chose contexts which favor a nonvolitional interpretation, and not, e.g. the falling or dying of an actor on stage.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.19 (1269-1312)


. Conclusions Overall, the results of the nationwide survey support the hypothesis that what looks like a morphological passive is well along the way to being reanalyzed as a syntactically active construction with a phonologically null indefinite subject. Although the absolute numbers in the various geographical regions differ, the relative acceptability judgments are mostly the same for subjects in Inner Reykjavík and elsewhere in the country, as well as for adults. This indicates that we are tapping into psychologically real linguistic intuitions. The obvious questions are: Why is this innovation happening, and when did it start? . Why in Icelandic? This syntactic reanalysis is neither unnatural nor unique, since a similar diachronic development has occurred independently in both Polish and Irish. But why is it happening in Icelandic? As noted earlier, it is clearly not foreign influence, since none of the other languages commonly spoken by Icelanders has this construction. I know of no indications that a similar change lies on the horizon in any other Germanic language. Icelandic has a number of grammatical properties which “conspire” to make the reanalysis a natural development. A key factor, I suggest, has to do with how null subjects get interpreted: there’s a strong bias to interpret a null subject as referring to a human participant, even if the verb is strongly biased towards a nonhuman subject. Let me illustrate this in English (examples from Maling 1993): (19) a. It’s all too common to be warped in a lumberyard. b. It’s all too common for lumber to be warped.

This interpretative bias isn’t an absolute. As illustrated by (19b), an overt subject doesn’t have to be [+human], but when we hear “It’s common to...”, we expect that the predicate will denote something that a human might do. It is difficult to interpret (19a) as being about lumber; rather we imagine human beings with warped minds. Similarly for the verb migrate, the grammatical subject is expected to be birds or animals. Note the commonness of utterances like “It’s common to see migrating birds at this time of year” and the uncommonness of utterances like “It’s common to migrate at this time of year.” A Google search (January, 2003) for the string “common to migrate” yielded 9 hits – all with human subjects. Here are two: (20) a.

With a zoom move it is common to migrate from one type of camera shot to another, b. It has become common to migrate data directly ...

This interpretive preference is expressed in (21):



JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.20 (1312-1373)

 Joan Maling

(21) Interpretive Rule: In the absence of an overt controller, a null subject will be understood as [+human].

This universal preference for null subjects to refer to humans shows up in Icelandic in a number of constructions involving passive morphology. First, consider the so-called “impersonal passives” of intransitive verbs, which are found in standard Icelandic, as in all Germanic languages except English. (22) a.

Það var flautað. itEXPL was whistled “People whistled” b. Það var ekki talað um neitt annað itEXPL was not spoken about nothing other “People didn’t talk about anything else”

The understood agent of an impersonal passive can only be interpreted as a human. Even though the subject of the verb “whistle” can be many things, including teakettles or trains, the impersonal passive in (22a) can only be understood as describing human whistlers, not trains or teakettles.15 (Although English lacks impersonal passives, the same interpretive bias holds of the gerund whistling in “There was whistling”.) As noted earlier, the morphological passive of an intransitive verb is syntactically ambiguous: It can be analyzed either as a true passive, or as a syntactically active impersonal construction with a human subject. The latter analysis becomes all the more plausible given that by-phrases are generally “ungrammatical or infelicitous in impersonal passives” in Icelandic (Sigurðsson 1989: 322, Fn. 48). Recall that our data suggest that perhaps 50% of adult speakers of Icelandic analyze the so-called impersonal passives as impersonal actives with a thematic proarb subject, as shown by the fact that the construction occurs with bound anaphors and with subject-oriented adjuncts. Impersonal passives in all the Germanic languages show the same interpretive bias towards human subjects. A special property of Icelandic, which does not hold in the mainland Scandinavian languages, is that the same interpretive bias holds true of the personal passive. (23) a.

Snjóflóðið eyðilagði húsið the.avalanche destroyed the.house

(Active)

. An anonymous reviewer notes that the most salient reading of (22a) is the honking of a car; another speaker reports that for him, the most salient reading is that of a referee blowing a whistle. Under both readings there must be a human actor. Similarly, the impersonal passive Það var hringt “it was rung” cannot be used to refer to the ringing of an alarm clock (Helgi Skúli Kjartansson, personal communication).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.21 (1373-1446)

From passive to active 

b. Húsið the.house c. Húsið the.house

var eyðilagt (*af snjóflóðinu) was destroyed by the.avalanche eyðilagðist í snjóflóðinu. got.destroyed in the.avalanche

(Passive) (Middle)

The understood agent (or implicit argument) in a personal passive like (23b) must be a human, so (b) cannot mean that the avalanche destroyed the house, a meaning rendered by the morphological middle in (c); sentence (23b) can only mean that some human agent destroyed the house during the avalanche. Thus, as with passives of intransitive verbs, passive morphology on transitive verbs is correlated with human agents.16,17 A third Icelandic construction in which passive morphology is associated with human subjects is the class of aspectual modals, including byrja (‘begin’), fara (‘go’) and vera (‘be’) (Sigurðsson 1989: 64).18 These verbs take infinitival complements, and when used as aspectual modals, they take passive morphology, i.e., the verb occurs in the past participle and takes the passive auxiliary vera: . As pointed out by an anonymous reviewer, counterexamples can be found on the web. (i)

Bærinn var eyðilagður í bruna 1904 . . . the.town was destroyed in the.fire [of] 1904

The example in (i) is from a blog page dated August 1, 2005, written by an Icelander in London describing a visit to Ålesund. The three native speakers I consulted all found this example strange; for them, this example can only be interpreted as involving voluntary burning, which was obviously not the case in Ålesund. It may be that the generalization about human volitional agents is being blurred under external influence, mostly but not exclusively from English. . For many speakers, the agent can be a nonhuman animate NP, e.g. cows: (i)

Það voru étnir graskögglar í allan dag. it-expl were eaten grasspellets all day

. When used with an overt subject, the understood agent is not restricted to humans, as illustrated by the following example provided by an anonymous reviewer: (i)

Turninn var þegar byrjaður að skekkjast. the.tower was then begun to lean “By then the tower had begun to lean”

However, when used impersonally, the understood agent is restricted to humans as indicated by the translation provided for (24a). The exception is that weather verbs, which lack an actor altogether, are allowed, as noted by Sigurðson (1989) and the anonymous reviewer: (ii) Það var byrjað að rigna. it-expl was begun to rain “It had begun to rain”

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.22 (1446-1499)

 Joan Maling

(24) a.

Það var byrjað að flauta. it was begun to whistle “People began to whistle” = The teakettle(s) began to whistle. b. Það var verið að flauta. it was been to whistle “There was whistling”

Here there is passive morphology on the auxiliary verb, but no passive meaning whatsoever; the construction passes all the active voice diagnostics (no by-phrase is possible, bound anaphors are allowed, subject control of participial adjuncts is possible, and unaccusative verbs are allowed as long as the implicit subject is human). Many examples of the new construction seem to be functionally equivalent to the impersonal passive of aspectual vera (‘to be’), which is part of the standard language. The example in (25a), uttered by a Reykjavík girl, aged 4;4, might be rendered by an adult as (25b): (25) a.

Í gær þegar það var gefið mér lýsi, þá... yesterday when itEXPL was given me cod.liver.oil, then... b. Í gær þegar það var verið að gefa mér lýsi, þá ... yesterday when itEXPL was been to give me cod.liver.oil, then... “Yesterday when they were giving me cod liver oil, then....”

It seems likely that such constructions in standard Icelandic serve as models for the reanalysis of the impersonal morphological passive as a syntactically active impersonal construction. Such models are absent in the other Germanic languages. . Reanalysis as simplification of the grammar It is widely agreed that passive voice involves suppression of the external argument. External arguments include agents and causes (e.g. natural forces). As we have seen, in Icelandic, the things that can be the understood agent of the passive are a proper subset of the things that can be the subject of the corresponding active. The otherwise systematic relationship between active-passive pairs breaks down. The Icelandic restriction to [+human] agents requires an additional stipulation. Consider the two syntactic representations sketched in (8), repeated here for convenience:19 . Thus we have evidence for two kinds of null subjects in Icelandic: one which is not referred to by the rules of the grammar, in other words, an empty category which is really nothing, and another kind of null subject which can be accessed by the grammatical rules, and which is interpreted as referring to humans.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.23 (1499-1562)

From passive to active 

(8) a. [IP [e] Aux [VP Vpart NP]] b. [IP proarb Aux [VP Vpart NP]]

Canonical Passive (before NP-movement) Impersonal Active

The representation sketched in (b) is an active voice construction with a null impersonal pronoun subject. What kind of element is the pro subject in (8b)? Pronouns can vary for person, number, gender and animacy/human, etc., but thematic role (e.g. agent) is not a classificatory feature for pronouns (whether null or overt). We know that natural language tolerates a high degree of ambiguity. I have suggested that a sentence like Það var dansað (‘there was danced’) can get two very different syntactic analyses, either an impersonal active or a true passive. The results of our survey suggest that nearly 50% of Icelandic adults analyze the so-called impersonal passive construction as having a syntactically accessible null subject. But once a speaker makes this reanalysis for intransitive verbs, it seems natural to extend the construction to all classes of verbs, including transitive ones, provided of course that the verb selects for a [+human] subject. Maling & Sigurjónsdóttir (2002) suggest that the interpretive generalization in (21) sets the stage for the reanalysis of a thematically empty null subject as a fully thematic pro external argument. On the other hand, nothing forces this reanalysis. As illustrated in (26) for Norwegian, the impersonal passive in the mainland Scandinavian languages has all four syntactic properties of the canonical passive, and shows none of the signs of the innovative Icelandic construction.20 (26) Impersonal passives in Norwegian a. Det ble danset av alle og enhver i bygda.21 itEXPL was danced by all and everyone in the.village b. *Det ble låst seg (selv) inne i fabrikken. itEXPL was locked refl (self) inside in the factory c. *Det ble danset leende/gråtende/full/fulle/fullt. itEXPL was danced laughing/crying/drunk-sg/pl/neut

. Hestvik (1986) argued that in Norwegian impersonal passives, the indefinite postverbal NP receives accusative case. The arguments are theory-internal in that there is no overt morphological case. Note that the “transitive impersonal passive” in Norwegian is subject to the definiteness effect: the postverbal NP must be indefinite. . Elisabet Engdahl (personal communication) suggests that the av-phrase behaves more like an adverbial quantifier than as an identifier of the agent; the sentence means that there was general/widespread dancing. An agentive by-phrase can be used only in this adverbial sense in Swedish impersonal passives (although it can identify the agent in personal passives). In Norwegian, an agentive by-phrase is normally not possible with impersonal passives (Faarlund et al. 1997: 847), but judgments vary; Hovdhaugen cites some examples, but observes that the agent phrase is rarely expressed and often sounds unnatural (1977: 24).

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.24 (1562-1622)

 Joan Maling

d. *Under krigen ble det forsvunnet ofte uten spor. during the.war was itEXPL disappeared often without a.trace

What, then, makes Icelandic different from its Scandinavian cousins? The internal pressures are not the same as in the other Germanic languages. The Icelandic personal passive allows only the human-agent reading, whereas in Norwegian and Swedish the agent can also be a non-human agent, e.g. a force of nature.22 And only Icelandic has the aspectual modals – passive morphology associated with the human-subject reading – which can serve as a model. I cannot emphasize too strongly that this syntactic change is still very much in progress. The results reported here are just the first step in what we hope will be a longitudinal study designed to track the development and spread of this innovative construction. The results of our study suggest that the crucial first step in the reanalysis is the extension of the impersonal passive to inherently reflexive predicates; this then extends to non-inherent reflexives and other bound anaphors. Recall that even adult speakers accept bound anaphors in impersonal passives to a fairly high degree. Moreover, many adults also accept control of participial adjuncts, and many accept impersonal passives of presumably unaccusative verbs with unspecified human subjects. All of these factors indicate that even in the standard language, passive morphology is associated with a human agent reading, which makes possible the reanalysis as an unspecified human subject construction. Given this unexpected variation among adult speakers, it is not surprising that children have started to extend this analysis to all verb classes, including transitive verbs, which take human subjects. We speculate that the process has started with highly agentive verbs (verbs of high transitivity in the sense of Hopper & Thompson 1980), but this is an area for future research. We don’t yet know whether this change is proceeding lexical item by lexical item, or verb class by verb class. We are still left with the ultimate mystery of language change: How does the change get started in the first place? What has triggered the apparently spontaneous eruption of the new construction everywhere in the country except inner Reykjavík? We can understand what is happening only if we recognize that there are both structural factors and sociological factors at play. We haven’t even begun to investigate the discourse use of the new construction, or its acquisition by young children. How do children learn the constraints on the interpretation of the various constructions they hear? In particular, how do they learn that the null subject of the impersonal passive and the “new construction” must be interpreted as hu. Lavine (2000: 107f.), citing Wieczorek (1994: 53), notes that unlike the implied human-agent reading in the Polish -no/to construction, the Ukrainian construction is ambiguous between an implied human agent (=passive) and a spontaneous “inchoative” (=middle) reading.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.25 (1622-1703)

From passive to active 

man? A Chomskian would say that the bias in favor of the human interpretation must be part of Universal Grammar, i.e. a language universal which children assume holds unless presented with evidence to the contrary. In other words, it’s the default setting. I know of no better answer. Language is what makes us human. It is central to everything we do. Understanding language and language change gives us insight into ourselves. It is surely no surprise that humans are at the very center of our universe. What is perhaps more surprising is that this fact, our obsession with ourselves, seems to lie at the center of how syntactic structure is represented in the brain, at the center of mental representations of language.

References Barðdal, Jóhanna (1993). Accusative and dative case of objects of some transitive verbs in Icelandic and the semantic distinction between them. Flyktförsök: Kalasbok till Christer Platzack på femtioårs-dagen 18 November 1993, från doktorander och dylika, 1–13. Lund University. Barðdal, Jóhanna, & Valéria Molnár (2000). Passive in Icelandic – compared to Mainland Scandinavian. Working Papers in Scandinavian Syntax, 65, 109–146. Billings, Loren, & Joan Maling (1995). Accusative-assigning participial -no/to constructions in Ukrainian, Polish and neighboring languages: An annotated bibliography. Journal of Slavic Linguistics, 3(1), 177–217; 3(2), 396–430. On-line version downloadable at: http://www.indiana.edu/∼slavconf/linguistics/download.html Dziwirek, Katarzyna (1991). Aspects of Polish Syntax. Unpublished Ph.D. dissertation, University of California, San Diego. Engdahl, Elisabet (1999). The choice between bli-passive and s-passive in Danish, Norwegian and Swedish. Ms., Gothenburg University. URL: http://www.ling.gu.se/∼engdahl/ Faarlund, Jan Terje, Svein Lie, & Kjell Ivar Vannebo (1997). Norsk referansegrammatikk. Oslo: Universitetsforlaget. Hálfdanarson, Helgi (Ed.). (1984). Gætum tungunnar (Let’s watch our language). Reykjavík: Hið íslenska bókmenntafélag. Haraldsdóttir, Aðalheiður Þ. (1997). Það var sagt mér að skrifa ritgerð. Þolmynd eða ekki þolmynd? (There was told me to write a thesis. Passive or not passive?). Unpublished B.A. thesis, University of Iceland. Haspelmath, Martin (1990). The grammaticalization of passive morphology. Studies in Language, 14, 25–72. Hestvik, Arild (1986). Case theory and Norwegian impersonal constructions. Nordic Journal of Linguistics, 9, 181–197. Hopper, Paul, & Sandra Thompson (1980). Transitivity in grammar and discourse. Language, 56, 251–299. Hovdhaugen, Even (1977). Om og omkring passiv i norsk. In Thorstein Fretheim (Ed.), Sentrale problemer i norsk syntaks (pp. 15–47). Oslo: Universitetsforlaget. Huddleston, Rodney, & John Payne (2002). Nouns and noun phrases. In Rodney Huddleston & Geoffrey K. Pullum et al. (Eds.), The Cambridge Grammar of the English Language, Chapter 5. Cambridge: Cambridge University Press.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.26 (1703-1823)

 Joan Maling

Kjartansson, Helgi Skúli (1991). Nýstárleg þolmynd í barnamáli (A new passive in child language). Skíma, 14, 18–22. Kress, Bruno (1982). Isländische Grammatik. Leipzig: Enzyklopädie Leipzig. Labov, William (1994). Principles of Linguistic Change, Vol. 1: Internal Factors. Oxford: Blackwell. Lavine, James E. (2000). Topics in the Syntax of Non-Agreeing Predicates in Slavic. Unpublished Ph.D. dissertation, Princeton University. Lavine, James E. (2001). On the site of morphology and cross-linguistic variation. In Mary Antonia Andronis et al. (Eds.), Papers from the 37th Meeting of the Chicago Linguistic Society, Vol. 2 (pp. 437–451). Chicago, IL: Chicago Linguistics Society. Maling, Joan (1980). Inversion in embedded clauses in Modern Icelandic. Íslenskt mál 2, 175– 193. Reprinted in Maling & Zaenen (1990), 71–91. Maling, Joan (1993). Unpassives of unaccusatives. Unpublished paper presented at University of California at Irvine, University of Massaschusetts at Amherst, University of Helsinki, University of Iceland. Maling, Joan (2002). Það rignir þágufalli á Íslandi: Verbs with dative objects in Icelandic. Íslenskt mál, 24, 31–105. Maling, Joan, & Sigríður Sigurjónsdóttir (1997). The ‘new passive’ in Icelandic. In Elizabeth Hughes, Mary Hughes, & Annabel Greenhill (Eds.), Proceedings of the 21st Annual Boston University Conference on Language Development, Vol. 2 (pp. 378–389). Somerville, MA: Cascadilla Press. Maling, Joan, & Sigríður Sigurjónsdóttir (2002). The new impersonal construction in Icelandic. Journal of Comparative Germanic Linguistics, 5(1), 97–142. Maling, Joan, & Annie Zaenen (Eds.). (1990). Modern Icelandic Syntax [Syntax & Semantics 24]. San Diego, CA: Academic Press. Morgunblaðið (20 February, 2001, p. 13). Reykvíkingar eru umburðarlyndari en aðrir gagnvart lögbrotum. Þróunarsvið Reykjavíkurborgar kynnir skýrslur um lífsskoðanir landsmanna. Ottósson, Kjartan G. (1986). Mörk orðmyndunar og beygingar: Miðmynd í nútímaíslensku (The distinction between derivation and inflection: The middle voice in modern Icelandic). Íslenskt mál, 8, 63–119. Ottósson, Kjartan G. (1988). A feature-based approach to thematic roles. In Victoria Rosén (Ed.), Papers from the Tenth Scandinavian Conference of Linguistics (pp. 136–150). Bergen: University of Bergen, Dept. of Linguistics and Phonetics. Perlmutter, David (1978). Impersonal passives and the unaccusative hypothesis. In J. Jaeger et al. (Eds.), Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society (pp. 157– 189). Perlmutter, David, & Paul Postal (1984). The 1-advancement exclusiveness law. In David Perlmutter & Carol Rosen (Eds.), Studies in Relational Grammar, Vol. 2 (pp. 81–125). Chicago, IL: University of Chicago Press. Sells, Peter, Annie Zaenen, & Draga Zec (1987). Reflexivization variation: Relations between syntax, semantics and lexical structure. In Masayo Iida, Stephen Wechsler, & Draga Zec (Eds.), Working Papers in Grammatical Theory and Discourse Structure: Interactions of Morphology, Syntax and Discourse (pp. 169–238). Chicago, IL: University of Chicago Press. Shevelov, George (1963). The Syntax of Modern Literary Ukrainian. The Hague: Mouton. Sigurjónsdóttir, Sigríður, & Joan Maling (2002). Það var hrint mér á leiðinni í skólann: þolmynd eða ekki þolmynd? Íslenskt mál, 23, 123–180. Sigurðsson, Halldór Ármann (1989). Verbal Syntax and Case in Icelandic. Doctoral dissertation, University of Lund. Published 1992 by Linguistic Institute, University of Iceland.

JB[v.20020404] Prn:19/10/2006; 14:12

F: LA9608.tex / p.27 (1823-1860)

From passive to active 

Sobin, Nicholas (1985). Case assignment in the Ukrainian morphological passive construction. Linguistic Inquiry, 16, 649–662. Stenson, Nancy (1989). Irish autonomous impersonals. Natural Language & Linguistic Theory, 7, 379–406. Svavarsdóttir, Ásta (1982). ‘Þágufallssýki.’ Breytingar á fallanotkun í frumlagssæti ópersónulegrar setningar (‘Dative sickness.’ Changes in case usage in the subject position of impersonal sentences). Íslenskt mál, 4, 19–62. Svenonius, Peter (2005). Case alternations in the Icelandic passive. URL: http://ling.auf.net/ lingBuzz/000124. Visser, F. Th. (1973). An Historical Syntax of the English Language, III, 2nd half. Leiden: E. J. Brill. Zaenen, Annie, & Joan Maling (1984). Unaccusative, passive and quirky case. In Mark Cobler, Susannah MacKaye, & Michael T. Wescoat (Eds.), Proceedings of the Third West Coast Conference on Formal Linguistics (pp. 317–329). (Reprinted in Maling & Zaenen (1990), 137–152). Zaenen, Annie, Joan Maling, & Höskuldur Thráinsson (1985). Case and grammatical functions: The Icelandic passive. Natural Language & Linguistic Theory, 3, 441–483.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.1 (46-143)

The relation between information structure, syntactic structure and passive* Anneliese Pitz In this article I show how the interaction between the syntactic level and the pragmatic level leads to different choices with respect to voice even in closely related languages. My conclusions are based on translational evidence from German and Norwegian, and passive constructions in both languages and their translations are the object of study. Although passivization may be viewed as an information structuring device which allows for a reordering of constituents, its use depends on the language’s structural properties and stylistic preferences.

.

Introduction

It is widely accepted that voice is involved in information structuring (Leiss 1992; Lambrecht 1994; Doherty 1996, 2003). To shed light on the nature of this involvement, I investigated the occurrence of passive constructions in translation from German into Norwegian and vice versa. German and Norwegian are two closely related languages which share at least one type of passive, the periphrastic passive construction as illustrated in (1) below. However, they exhibit some structural differences which in the light of previous work make it promising to compare constructions like the passive. Searching in the Oslo Multilingual Translation Corpus reveals that there are twice as many passive constructions in the Norwegian translations as in the originals, while there are fewer passives in the German translations than in the German originals.1 These proportions may tempt us to assume that we * I am grateful to Torgrim Solstad, Benjamin Lyngfelt and two anonymous reviewers for suggestions which helped improve the content and the exposition of this paper. I would also like to thank Kjell Johan Sæbø for valuable comments. . The Oslo Multilingual Corpus (OMC) is a collection of text corpora comprising original texts and translations from several languages. For detailed information on the OMC Corpus see http://www.hf.uio.no/forskningsprosjekter/sprik/english/.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.2 (143-176)

 Anneliese Pitz

are being confronted with ‘shining through’ phenomena of the source language (Teich 2003), i.e. that the translation may have adopted translational patterns that would not have been chosen in the original texts, or at least not to such a great extent, but which are found in the source language. However, the number of passives in the Norwegian translations exceeds the number of passives in the German originals. Furthermore, not all passives in the German original are translated as passives, thus leaving us with even more passives in the Norwegian translation to account for. Obviously, voice is not just copied in the translation process although in the majority of the examples passive corresponds to passive. The example in (1) illustrates an analogue translation of a passive, i.e. the Norwegian bli-passive translated by the German werden-passive. (2) illustrates the discrepancy mentioned above:2 (1) a.

Resultatet av hele undersøkelsen ble oversendt result.the of whole examination.the pass send.ppart forsikringsselskapet, og de protesterte ikke. (GS1) insurance-company.the and they protest.past not ‘The results of the whole inquiry were sent to the company’s insurers and they didn’t protest.’ b. Das Resultat der ganzen Untersuchung wurde an die the.nom result the.gen whole examination pass to the Versicherungsgesellschaft geschickt, und dort protestierten sie insurance-company send.ppart and there protest.past they nicht. (GS1TD) not

(2) a.

Inzwischen wurden von der IAETA und der WHO in.the.meantime pass by the IAETA and the WHO Richtlinien zur Strahlenbehandlung von Nahrungsmitteln guidelines for.the irradiation-treatment of food-articles erarbeitet. (UR1) prt.work.ppart ‘Guidelines have been worked out by IAETA and WHO, and now these agencies of the UN are requesting all countries to alter their food legislation correspondingly.’

. The examples will always be presented in the order original > translation. In addition, the original is discernible from the translation in that it has not been annotated with T (N or D) (T for translation, N for Norwegian and D for Deutsch). The rest of the annotation (UR, GS1) refers to the actual text.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.3 (176-235)

Information structure, syntactic structure and passive 

b. I mellomtiden har den internasjonale atomenergieorganisasjonen in meantime.the has the international IAETA og WHO utarbeidet retningslinjer for strålebehandling and WHO out.work.ppart guidelines for irradiation-treatment av næringsmidler. of food-articles

Example (2) will be discussed more thoroughly in the next section. Here it suffices to notice that the German original is a periphrastic (werden-) passive clause which is translated as an active clause in Norwegian. The discrepancies between the number of occurrences of passive constructions in the originals and translations of the same language lead to the following question – what determines the choice of either active or passive in translation from German into Norwegian and vice versa? The following more general question is related to this – what is the effect of a change of voice? The answer to the latter question is anticipated by previous research and involves the interaction between different grammatical components, more concretely between a syntactic, semantic, pragmatic and a linear level. According to Leiss (1992) and Eroms (1974), this interaction can be depicted as follows – in German and other accusative languages, the agent is realized as the subject and coincides in the unmarked case with the theme and the topic function.3 If the agent is not the theme, another participant with the required thematic properties may occupy the topic position and be promoted to subject. These processes are accomplished by passivization.4 Passivization is thus a means to repair, at least partly, the unmarked mapping between the different components.5 However, according to Leiss (op. cit.), we may expect differences between languages with respect to the unmarked mapping between the different levels.6 For instance, languages may differ with respect to the element which preferably occurs in the first position. In fact, Leiss (op. cit.: 119) modifies the description for German by stating that the word order Topic > Verb > Agent > Patient is as unmarked . Topic in Leiss’ terminology refers to sentence-initial position. Theme is understood as an element at the discourse pragmatic level, containing the known entities as opposed to the rheme which presents new information. According to Leiss, thematicity correlates with definiteness. . Topicalization of the theme would seem to be another suitable mechanism. However, it does not involve the subject function and hence represents a more marked solution. . Thematicity is thus more important than agentivity in this mapping process (Leiss op.cit.:118). . Lambrecht (1994: 24) expresses a similar view. However, he assumes a prosodic level instead of a linear level.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.4 (235-311)

 Anneliese Pitz

as the order Agent > Verb > Patient.7 This mapping option represents a difference between German and English, as discussed in Doherty (1996, 2003). Doherty examines the effect of structural differences in translational contexts and shows that German favours adjuncts in topic position, “reframing” English subjects as prepositional phrases.8 Importantly, passivization is shown to play a major role in adjustments of this type. In the following we will investigate whether the discrepancies noted concerning the occurrence of passives in originals and translations in German and Norwegian can be explained by a different language specific interplay between the various grammatical components as Doherty proposes for the language pair English – German. Quite generally, it is to be expected that the syntactic level, in addition to the prosodic level, will have to accommodate pragmatic requirements (Leiss 1992: 118). Passivizing an active sentence in a translation might not seem the right mechanism for this type of accommodation since it maps a different participant to subject and possibly to theme and topic position. However, as we shall see, passivization offers the possibility to avoid certain structural and processing problems which adhering to the (original) active construction would lead to. Simultaneously, it preserves the information structure. Importantly, it is not only passivization which has this effect – the rendering of a passive sentence as an active construction also has this effect. This paper is organized as follows. In 2.1, I shall present the main structural differences between Norwegian and German and in Section 2.2, the passive constructions in German and Norwegian. In 2.3, I provide an overview of the number of occurrences of these constructions and their correspondences in the corpus investigated. In 2.4, I define the terminology for the description of information structuring. In Section 3, I shall discuss data from the OMC translation corpus, more specifically cases where the translation deviates from the original with respect to voice (3.1) and with respect to the elements in the theme or the rheme part (3.2) and analyze them against the background of structural and information structural differences. 3.3 provides some additional support to our assumptions and Section 4 concludes the paper.

. See Fabricius-Hansen and Solfjeld (1994) for a study of the difference concerning elements in sentence-initial position in German and Norwegian. It is shown that the subject occurs more often in topic position in Norwegian. . Doherty ascribes this preference to stronger selectional restrictions in German with respect to the semantic role of the subject, and to the attempt to avoid processing problems. Prepositional adjunct phrases cannot be mistaken for subjects as, for instance, preposed objects could be.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.5 (311-356)

Information structure, syntactic structure and passive 

. German and Norwegian . Structural contrasts While both languages are (finite) verb-second languages in main clauses, German exhibits SOV order in subordinate clauses. This property has implications for main clauses as well, as participles and other non-finite verbal parts are found in the rightmost position of the sentence. Another major difference concerns the position of the subject. In Norwegian, there is a fixed subject position after the finite verb (which has to be filled) while the subject in German may be found in different positions.9 The subject in German is identified by the nominative case and governs verb agreement. In other words, Norwegian exhibits structural coding of syntactic functions whereas German uses morphological means. When it comes to the ordering of internal arguments, both languages exhibit a word order which reflects the commonly accepted role hierarchy: (AG) > BEN > PATIENT (Primus 1999: 157). There is, however, a difference with respect to the position of adjuncts. In German, adjuncts precede the arguments and the verb. In Norwegian main clauses, they either precede the non-finite verb or follow the objects. In addition to the commonalities mentioned, both languages have a sentence-initial position for topicalized elements which I shall refer to as the Topic position.10 Topicalization of elements other than the subject is better exploited in German than in Norwegian.11 The basic word order patterns for main clauses can be represented in a simplified version as follows (Abraham 1986; Fanselow 1992; Faarlund et al. 1995):12 N Topic Vfin SUBJ adv Vnon-fiin OBJind OBJdir OBJprep ADV G Topic Vfin ADV (SUBJ) OBJind OBJdir / (SUBJ) OBJprep Vnon-fin

The parentheses around SUBJ are intended to indicate that the position of the subject is not fixed but depends rather on the thematic role assigned to it (Abraham . The subject NP in Norwegian may move from this position to the sentence-initial position but not to any other (Faarlund 1995: 872). Some descriptions would simply state that the subject appears either in a pre- or post-(finite) verbal position. However, since the sentence-initial position can be occupied by any constituent, it seems plausible to assume that the post-finite position is the basic position. . Topicalization refers only to the movement into the sentence-initial position, thus the position preceding the finite verb in declaratives. Topicalization is an option for any constituent. The Topic position corresponds to the traditional Vorfeld/Forfelt (Faarlund et al. 1995: 872). . As will be seen, this seems to be so even in the absence of distinctive case marking. . My intention is not to give a complete account of the various word order possibilities in German and Norwegian but rather to focus on the categories which are involved in the processes under investigation. This is the reason for settling with a flat structure as well.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.6 (356-376)

 Anneliese Pitz

1986; Primus 1999). In German, an agentive subject will precede the objects in the unmarked case while a subject with the proto-patient role will appear in the position of the direct object from where it may be scrambled or topicalized.13 In other words, the position of the subject is dependent on the rank of the semantic role while it is identified as the subject through the nominative case.14 Scrambling and topicalization are dependent on the information structural status of the subject with respect to other arguments, which factors such as pronominal form or definiteness are indicative of (Fanselow 1992; Lambrecht 1994: 79, 95). Examples (2) and (3) illustrate some of the contrasts (where (2) is familiar from the introduction): (2) a.

Inzwischen wurden von der IAETA und der WHO in.the.meantime pass by the IAETA and the WHO Richtlinien zur Strahlenbehandlung von Nahrungsmitteln guidelines for.the irradiation-treatment of food-articles erarbeitet. (UR1) prt.work.ppart ‘Guidelines have been worked out by IAETA and WHO, and now these agencies of the UN are requesting all countries to alter their food legislation correspondingly.’

. The notion of proto-patient role is understood in the sense of Dowty (1991). The fact that subjects with this type of role appear in the object position is confirmed by the existence of verbs which take their arguments in the following unmarked order: dat > nom. The example in (i) is taken from Fanselow (1992: 285): (i)

dass niemandem der Umstand aufgefallen war that nobody.dat the.nom circumstance strike.ppart be.past ‘that nobody had noticed the circumstance’

As far as scrambling is concerned, there is a difference between German and Norwegian. In German, the direct object may scramble across the indirect object producing marked constructions, while this is not an option in Norwegian. . According to Primus, basic word order in German is determined by the interaction of two ordering principles: the role hierarchy (ag > exp > pat) and the case hierarchy (nom > acc > dat). What may seem to be free word order in German is due to the fact that only one principle is followed (Primus 1999: 156). Pronominal arguments which appear in a different order than NP arguments only follow the case hierarchy. Note that in Norwegian, there is no special rule for the ordering of pronominal arguments (Faarlund 1995: 882).

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.7 (376-431)

Information structure, syntactic structure and passive 

b. I mellomtiden har den internasjonale atomenergieorganisasjonen in meantime.the has the international IAETA og WHO utarbeidet retningslinjer for strålebehandling and WHO out.work.ppart guidelines for irradiation-treatment av næringsmidler. of food-articles

(2a) is a passive construction where the promoted subject, i.e. the NP Richtlinien zur Strahlenbehandlung von Nahrungsmitteln, remains in its basic position, i.e. the object position close to the position of the non-finite verb form. Crucially, promotion to subject in German does not involve movement to a specific subject position (Haider 1984: 96). In the Norwegian translation, the (agent) von-phrase is converted into the subject of an active predicate as witnessed by its position after the finite verb. Note that if the passive version had been adhered to in the Norwegian translation, the promoted subject would have had to move from the object position into this post-finite position as well. Turning the German clause into an active clause would result in a construction parallel to (2b), i.e. the agentive subject occurring in a position next to the finite verb. The example in (3) illustrates the verb cluster as it surfaces in subordinate clauses. The complement most closely tied to the lexical verb, here the directional adjunct, immediately precedes it in German and follows it in Norwegian. In both languages, the instrumental adjunct occurs farther away.15 (3) a.

Publikum ropte frekke kommentarer, særlig til dem som onlookers shout.past rude comments especially at those who ble støtt ut av båten av motstanderens lanse. pass push.ppart out of boat.the by opponent.the.gen lance ‘The onlookers shouted rude comments, especially at those hurled off the boat by their opponents’ lance.’ b. Von den Zuschauern hörte man freche Kommentare, from the onlookers hear.past one rude comments, besonders über die, die von der gegnerischen Lanze aus dem especially on those who by the adversary lance out the Boot gestoßen wurden. (SL1TD) boat push.ppart pass

A contrast of a different type, actually more of a stylistic preference, is found in the main clause, where the topicalized subject in the Norwegian original is translated as an adjunct in topic position and the subject function is realized by the indefi. Directional adjuncts with the type of verb illustrated may be regarded as arguments. Cf. Doherty (2003) with reference to Pittner (1998). See Nilsen (2000) on the ordering of adjuncts in Norwegian.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.8 (431-508)

 Anneliese Pitz

nite pronoun man.16 The unmarked mapping of grammatical levels given in the introduction for German and other accusative languages, i.e. with topic, theme, agent and subject coinciding, thus seems more appropriate for Norwegian than for German. . Passive constructions in German and Norwegian For the purposes of this exposition, I have restricted my investigation to what I consider to be the core passive constructions in both languages, i.e. the werden-passive in German, as illustrated in (1b) and its closest correspondences in Norwegian, the bli-passive in (1a), and the s-passive, illustrated in (4b):17 (4) a.

Abweichungen müssen dem Therapeuten mitgeteilt deviations must the.dat therapist communicate.ppart werden, damit er orientiert ist und optimal helfen kann. (JKS1) pass so that he informed is and optimally help can ‘Deviations must be told (to) the therapist so that he is informed and can provide the best help.’ b. Avvik må meddeles terapeuten, så han er deviations must communicate.pass. therapist.the so he is orientert og kan hjelpe optimalt. (JKS1TN) informed and can help optimally

The properties shared by the constructions in (1) and (4) are that the Agent of the active construction is demoted, i.e. crucially not realized as the subject, and, . The translation of the subject as an adjunct exemplifies the process which Doherty (2003) terms “reframing”, as mentioned above. Importantly, as the example demonstrates, the differing realization of the NP in topic position does not necessarily involve a change of voice unless the PP corresponds to the passive by-phrase. . Norwegian has a third major type of passive, a stative passive with the auxiliary være, ‘be’, corresponding largely to the sein-passive in German. However, while the sein-passive is restricted to verbs with stative or resultative interpretations, constructions involving the auxiliary være and a past participle can represent an elliptical form of the bli-passive, and involve verbs which do not imply any change of state (Faarlund et al. 1995: 525): (i)

Hun var oppdratt på en av disse anstaltene . . . (EFH1)

(ii) Sie war in einer dieser Anstalten aufgezogen worden ... ‘She had been raised in one of those institutions. . .’ Crucially, in the German counterpart, the participle worden cannot be deleted. Thus, in Norwegian, a være-construction can correspond to the bli-passive. Due to this ambiguity, the værepassive and the ensuing difficulty involved in deciding their status, these forms have not been included in this investigation.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.9 (508-539)

Information structure, syntactic structure and passive 

subsequently, the direct object is promoted.18 However, it is not clear from the examples that the promotion to subject involves different processes in German and Norwegian. In German, in the absence of a special subject position, case marking and verb agreement are the indicators of subject status, while in Norwegian the direct object has to move to the post-finite position.19 The similarities between German and Norwegian passive constructions and the structural differences discussed in Section 2.1 lead us to ask the question whether and to what extent the contrasts interfere with the possibility of expressing the same structures. . The data My data are drawn from the Oslo Multilingual Corpus, more specifically from a subcorpus thereof, the German-Norwegian parallel corpus. At the time I collected the data, the corpus included 22 Norwegian texts and their authorized translations and 30 German texts and their translations, each text comprising between 10,000 and 15,000 words. The corpus contains fiction and non-fiction. To ascertain whether voice is kept constant in the translations, I examined the passives in the Norwegian and German originals and the respective translations. I have classified the deviating cases, i.e. cases where we see a change of voice in either direction, i.e. passives replacing non-passives (Table 4) and passives resulting in non-passives (Table 6).20 The number of occurrences in Table 1 refers only to the periphrastic werdenpassive in German, while for Norwegian, it includes both the bli-and s-passive.

. I assume that demotion is prior, since both German and Norwegian exhibit impersonal passives, i.e. passivization of intransitive verbs. Furthermore, I assume that the agent is still understood even when it is not expressed in a by-phrase/von-phrase/av-phrase. See Hellan (1988) for the notion of “degree of presenthood” of a role. According to this proposal, the agent is more present (degree 2) in a passive construction than in a middle construction (degree 1) or in an ergative construction (degree 0). . There is yet another difference between Norwegian and German. Norwegian offers other possibilities for promotion, a fact that might lead to a greater number of passives in the language. Apart from the patient argument, the beneficiary may be promoted. Furthermore, no argument at all may be promoted and an expletive may be inserted in subject position. However, as it turns out, these possibilities do not represent an important source of discrepancies concerning the passive constructions in the corpus. The same comment applies to the bekommen- or recipient-passive in German, a passive construction which promotes a beneficiary. The Norwegian counterpart, with the auxiliary få, is not investigated either. . Note that by ‘non-passives’ I refer to all constructions that are not werden-, bli- or s-passives.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.10 (539-636)

 Anneliese Pitz

Table 1. Occurrences of passive constructions per text (averages)21

Passive

Norwegian original

Norwegian translation

German original

German translation

51

98

70

40

Table 2. Occurrences of bli and s-passives in Norwegian originals (22 texts) Original

Total in NO/average per text

bli-passive 588/26.5 s-passive 542/24 1130

Translated by werden-passive Other translations Total % Rate

Rate

416/19 70% 216/10 40% 632

172/7.5 30% 326/14.5 60% 488

100 100

To those familiar with the languages at hand, some discrepancy between the number of passives in the original and translated texts would be expected.22 It is rather well known that German is characterized by a nominal style, making use of complex nominal structures containing nominalizations, extended prenominal attributes and postnominal satellites, and that these constructions receive a more sentential translation in Norwegian, often resulting in passive constructions (Solfjeld 1998). Furthermore, German has a range of constructions, termed passive alternatives, which result in a passive translation in English (Teich 2003). Although Norwegian has some of these constructions, it remains to be seen whether these constructions are used as extensively.23 Table 2 provides an overview of the occurrence of each of the two Norwegian passives in the Norwegian original texts and the distribution of (werden-) passive and other translations. It reveals that the two Norwegian passives are distributed equally in the original texts. However, most s-passives are translated by constructions other than the werden-passive, while the bli-passives mainly result . Importantly, the numbers of occurrences are given as an average per text since the number of texts varies as indicated. . As mentioned, the rather low number of passives in the Norwegian originals might be due to the fact that constructions involving the auxiliary være are not included. There are roughly as many være-passives as bli-passives, more precisely 598 instances of which 299 are translated by the stative passive with the auxiliary sein, 100 by werden-passive and 199 by other constructions. On the other hand, være-passive constructions are not included in the number of passives in the Norwegian translation corpus either. . They are passive alternatives in the sense that they leave the agent unexpressed or that they have a patient subject. These constructions have not been investigated separately but some of them will show up as sources for passive constructions in the Norwegian translations.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.11 (636-747)

Information structure, syntactic structure and passive 

Table 3. Total number of passive constructions in the Norwegian translation corpus (30 texts) Translation

Total in NT/per text

Werden-passive original

Other original

bli-passive s-passive

1557/51 1387/46 2934

915/30.5 530/18 1445

642/21.5 857/28.5 1499

58% 38%

42% 62%

100 100

Table 4. Sources of passives in Norwegian other than werden-passive bli-passive

s-passives

Total

642

100%

857

100%

Active Nominalization Participle* sein+Participle Reflexive Ergative Others**

152 133 50 47 52 56 172

22,0 18.0 8 7.0 8.0 8.5 27.0

127 89 43 33 131 77 357

15.0 10.5 4.8 4.0 15.3 9.0 41.2

*Includes participles occurring either prenominally or alone. **Others refers to a diversity of constructions which cannot be discussed here. Besides, it includes cases where the passive represents a free translation (69 for bli-passive, 91 for s-passive). For the s-passive, a range of constructions occurs combining passive and modal interpretations.24

in werden-passive translations. Table 3 presents the picture from the other side. There is a higher percentage of non-passive sources resulting in s-passives, i.e. 62%. Clearly, the bli-passive represents a closer equivalent of the werden-passive than the s-passive does. If Table 2 and Table 3 are compared, we notice that German keeps the voice of the original in 70 % of the cases of a bli-passive while Norwegian only translates 58% of the werden-passive originals as a bli-passive. Table 4 gives the main German source constructions resulting in passive constructions in the translation. There is one comment which applies to all source constructions except for the first, i.e. active, and the free translations: they convey a passive perspective in the sense of Doherty (1996), i.e. insofar as they have a sub. To just mention one major type: some adjectives receive a passive translation, such as those ending in the suffixes -bar or -lich, with the interpretation ‘can be Ved’: . . .das heißt die empirisch unerfüllbaren (DS1) . . . det vil si de som erfaringsmessig ikke kan oppfylles ‘... i.e. those who cannot be fulfilled empirically’

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.12 (747-799)

 Anneliese Pitz

ject, this subject carries a (proto-) patient role. If they do not, they nevertheless have a passive interpretation. For instance, nominalizations, more specifically the so-called process or complex event nominals, are readily transformed into passive predicates or clauses, their satellites being expressed as arguments of the verb (Grimshaw 1990; Solfjeld 1998). Moreover, they share with passives the property of realizing the agent only optionally. As far as participles are concerned, some accounts attribute the passive interpretation of the entire passive construction to these elements (Zifonun 1992; Åfarli 1992). Their prenominal use represents another stylistic preference in German. A more controversial source of passive translations is represented by reflexive and ergative constructions.25 Ergative constructions differ from passives in that they, by definition, lack an agent role, and the analysis of the reflexive constructions in question generally assumes the complete deletion of the agent (Steinbach 2002: 43).26 A passive translation of these constructions might not seem appropriate as it involves an understood agent.27 However, the selected passive translations share one important property with ergative and reflexive constructions; their subjects carry the same type of role, i.e. proto-patient. Examples (5)–(8) illustrate ergative and reflexive sources, translated in (5) and (6) as bli-passives, and as s-passives in (7) and (8). (5) a.

An einem für sich stehenden Haus oben auf dem in a.dat for itself standing house up on the Kapuzinerberg erloschen nacheinander die.nom Lichter. Kapuzinerberg went.out after another the lights ‘In an isolated house on the slope of the Kapuzinerberg the lights went out one by one.’ (PH1TE)

. The term ‘ergative’ is used here to refer to verbs which do not have agentive subjects and which express a change of state or location of the subject. In German, this class of verbs is characterized by selecting the auxiliary sein, their participial form can occur as a pronominal modifier of the subject and they cannot be passivized (Primus 1999). . But see Hellan (1988) for a different view. . This difference is also acknowledged by Doherty (1986: 603): The passive perspective is characterized by the projection of a lower theta-role, mostly the theme, into the subject. In this sense, verbs like hervorgehen (‘emerge’) could be said to have a passive-like perspective too, though their semantic form does not have the causative component of the syntactic passive with its mute argument. Doherty thus includes these types of verbs into the class of verbs conveying a passive perspective, although under the label “lexicalized” or “lexical passive”.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.13 (799-878)

Information structure, syntactic structure and passive 

b. I et ensomt hus oppe på Kapuzinerberg ble lysene In an isolated house up on Kapuzinerberg pass lights.the slukket, et for et. (PH1TN) put.out.ppart one for one (6) a.

In der Schaufensterscheibe sah ich, daß sich die Kapellentür in the exposition-window saw I that refl thenom chapel-door öffnete. (CH1) opened ‘In the show window I saw that the chapel door was opened.’ b. I utstillingsvinduet så jeg at døren til kapellet ble in show-window.the saw I that door.the to chapel.the pass åpnet. open.ppart

(7) a.

Von deutschem Boden in beiden Staaten sollen Friede und from German soil in both countries shall peace and gute Nachbarschaft mit allen Ländern ausgehen. (RVW) good neighborhood with all countries prt.go.inf ‘Peace and good neighbourly relations with all countries should radiate from the soil of both German states.’ (RVW1TE) b. Fra tysk jord i begge stater skal fred og naboskap from German soil in both states shall peace and neighborhood spres til alle land. (RVW1TN) disperse.pass to all countries.

(8) a.

Anschließend überträgt sich unsere Erwartung auf die in the following over.carries refl our.nom expectation on the Person oder auf das Objekt in der neuen Begegnung person or on the object in the new meeting ‘Correspondingly our expectation is transferred to the person or to the object in the new encounter.’ b. I tilknytning til dette overføres vår forventning, til in connection to this over.carry.pass our expectation to personen eller objektet i dette nye møtet. (JKS1TN) person.the or object.the in this new meeting.the

In (5)–(7), the Norwegian translation introduces an understood agent, which could have been avoided by choosing an analogue translation: in (5) the (ergative) verb slukke (or slokne), in (6) the reflexive verb åpne seg and in (7) the reflexive verb spre seg. The reflexive construction in (8) does not have an analogous counterpart in Norwegian. We will return to these cases in 3.3.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.14 (878-952)

 Anneliese Pitz

Table 5. Occurrences of werden-passives in the German original texts (30 texts) Original

Total in GO/average per text

bli-passive

s-passive

Other

werden %

2111/70 100

915 43

530 25

666 32

Table 6. Non-passive translations of werden-passive Other correspondences in N-TL

666

100

Active være + PPart få + PPart/N Ergative Others* Restructuring**

277 86 54 36 94 119

41.6 12.9 8.1 4.4 14.1 17.9

*Others refers to different types of constructions such as auxiliary + predicative or light verbs + PP/DP. **Restructuring refers to free translations or even deletion.

However, Table 4 does not reveal the fact that some passives in the German original texts are translated as active constructions.28 According to Table 5, 32 % of all werden-passives do not receive a passive translation. Table 6 presents some of the non-passive translation options for werdenpassives where ‘non-passive’ means any correspondence other than bli- and s-passives.29 Here, as in Table 4, the active correspondences constitute the interesting class. Why is the passive not adhered to? We have seen in (3) that the promoted subject of the passive construction in the German original may occur in object position though this is not an option in Norwegian. Clearly, this is not the whole story, since the passive voice of the original could be adhered to in the translation by moving the subject into the reserved position after the finite verb. . Information structure Of all the proposals concerning information structuring, I will follow Lambrecht (1994), Vallduví and Engdahl (1996) and Doherty (2003) in assuming that there is an unmarked information structuring sequence in most languages, quite generally . Interestingly, this change from passive to active occurs in the translations from Norwegian to German as well (cf. Table 2). . We notice that the være-passive shows up again as a translation possibility for the werdenpassive. As mentioned before, as the determination of their status is not straightforward, I will not deal with these constructions now although they are clear instances of agent demotion.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.15 (952-1011)

Information structure, syntactic structure and passive 

described as the ‘given information-new information’ partition, also called ‘focusground’ (Vallduví and Engdahl) and ‘presupposition-assertion’ (Lambrecht). For this partition, the terms ‘theme-rheme’ will be used, where ‘rheme’ carries new information and, citing Vallduví and Engdahl’s definition (of ‘focus’), is the “informative part that makes some contribution to the discourse or the hearer’s mental world”. ‘Theme’ is taken to correspond to Vallduví and Engdahl’s ‘ground’, defined as (op. cit.: 461) “the part that anchors the sentence to the previous discourse or the hearer’s mental world”. Furthermore, I will assume that both theme and rheme may be partitioned into background and focus elements. Inside theme, contrastive focus can be expressed; while rheme will include the presentational focus. As far as rheme focus is concerned, we can say that new elements will be the focus. Focus is associated with stress, the main stress falling on the verb-adjacent element, or, if there is none, on the verb itself (Abraham 1995). The element which carries the focus marking is called the ‘focus exponent’. If focus covers more than one element, the extension is called ‘focus projection’. Focus projection can cover everything except contextually given elements. Relevant to our discussion is the fact that in Norwegian and German, focus projection works in different directions due to the SOV-SVO contrast: rightwards from the verb and towards the end of the sentence in Norwegian and leftwards from the verb-adjacent element at the end of the sentence towards the finite verb in second position in German (Abraham 1995: 610). Following Doherty (2003), I shall assume moreover that an additional strategy of information structuring exists for more complex constructions, namely balanced information distribution, termed BID, which gives different results depending on the typological properties of the languages being considered. For English as an SVO language, BID will require there to be some post-verbal focus material, and it will thus prevent topicalization of certain adjuncts.30 Giving the most relevant constituent in a complex clause the value 1 and the constituent that has been mentioned most recently the lowest value 3 and the constituent which is . Doherty (2003) gives the following examples: (i)

The Julian calendar was introduced on 1 January, 45 BC.

(ii) On 1 January, 45 BC, the Julian calendar was introduced. According to Doherty, (ii) misleads the processor into identifying the verb as the focus since there is no verb adjacent focus candidate. This is avoided by leaving the adjunct in the postverbal position. In German, the problem does not arise since the subject is not bound to one special position: (iii) Am 1. Januar v.u.Z. wurde der Julianische Kalender eingeführt. On 1. january was the Julian calendar introduced.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.16 (1011-1048)

 Anneliese Pitz

mentioned somewhat farther away than 3 the intermediate value 2, the pattern for English will be 3-1-2. In other words, the informationally strongest element will occur surrounded by weaker elements in English. For German, where the new information and focus occur towards the end, Doherty proposes that “BID can only result in a pattern where the lowest value (the constituent which has been mentioned most recently) is surrounded by two higher ones, i.e. constituents which are more relevant informationally”. Hence the pattern for German will be 2-3-1. Considering the SOV-SVO contrast and the difference in focus placement in German and Norwegian, it is to be expected that the proposal for English applies to Norwegian too, i.e. new/more relevant information elements surrounding known/less relevant elements.

. What the translation pairs reveal Tables 1–6 in Section 2.3 have given some idea of the findings. Although most of the correspondences in Table 4 deserve a more thorough investigation, I will restrict the main discussion to active constructions as sources for passives and vice versa. In 3.1, I will discuss structural conditions making an analogous translation impossible or difficult to process. A change of voice will create the appropriate theme-rheme sequence. In 3.2, I shall discuss cases requiring additional adjustments. In 3.3, I briefly discuss the effect of passive translations of reflexive and ergative constructions before summing up in 3.4. . Change of voice First, topicalized objects in German are turned into subjects of passive constructions, as illustrated in (9) and (10): (9) a.

Der erzwungenen Wanderschaft von Millionen Deutschen the.dat forced wandering of millions Germans nach Westen folgten Millionen Polen und ihnen towards west follow.past millions Poles and them.dat wiederum Millionen von Russen. (RVW1) again, millions of Russians ‘The compulsory migration of millions of Germans to the West was followed by the migration of millions of Poles and, in their wake, millions of Russians’

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.17 (1048-1122)

Information structure, syntactic structure and passive 

b. Millioner av tyskeres tvungne vandring mot vest millions of Germans forced travel towards west fulgt av millioner polakker og de igjen av follow.ppart by millions Poles and they again by russere. (RVW1TN) Russians

ble igjen pas again millioner millions

In the German sentence, the theme is expressed as a dative phrase which is clearly identified as such by case marking. Its analogous translation, object preceding subject, illustrated in (9c), would result in a garden path sentence in Norwegian, i.e. the reader would start from the assumption that the NP millioner av tyskeres vandring is the subject and be corrected only somewhere after the conjunction og. The sequence subject – verb – object distorts the theme-rheme sequence (9d): (9) c.

d.

[Millioner av tyskeres tvungne vandring. . .] obj [millioner. . ..]31 subj

fulgte follow-past

[Millioner av polakker og russere] fulgte subj (Rheme) follow-past [millioner av tyskeres tvungne vandring] obj (Theme)

Passivizing the clause in the Norwegian translation, as in (9b), by promoting the object to subject, and moving it into the topic position, produces the appropriate information structure. A similar case is illustrated in (10), where an accusative object is topicalized in the original: (10) a.

Diese Forderung hatte soeben ein führender this.acc requirement have.past just a.nom leading Baukünstler aufgestellt,. . . (RM1) Building-artist prt.set.part ‘So claimed a leading architect of the moment’ b. Dette krav var nettopp blitt oppstilt av en ledende this requirement past just pass up.set.part by a leading arkitekt,. . . (RM1TN) architect

Note that the topicalized element in (10) could be interpreted as the subject in the German original as well, due to the ambiguous case marking. One possible ex. The square brackets indicate the phrases realizing the syntactic functions, not the themerheme partition.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.18 (1122-1184)

 Anneliese Pitz

planation is that a garden path sentence would be resolved at the latest when the reader gets to ein führender, the article in combination with the adjective ending revealing the nominative case marking, i.e. the subject status of the NP.32 In Norwegian, the closest alternative (10c) in terms of linearity would only be resolved after the entire NP en ledende arkitekt.33 The ordering subject > object would lead to a change in the theme-rheme sequence, as seen in (10d): (10) c.

[Dette kravet] har [en ledende arkitekt] nettopp stilt opp, ... obj subj d. [En ledende arkitekt] har nettopp stilt opp [dette kravet],. . . subj (part of Rheme) obj (Theme)

The next example involves a change from a German passive to an active translation where the topicalized spatial adjunct of the German (passive) original is turned into the subject of the active construction: (11) a.

Pflichtschuldig mahnte er bei Ministerpräsident Li Peng an, dutifully urge.past he to president Li Peng prt auch im Reich der Mitte müßten die Menschenrechte too in the land the Middle.gen must the people-rights eingehalten werden. (HPMHS1) prt.keep.ppart pass ‘Dutifully, he warned premier Li Peng that China too had to respect human rights.’ b. Pliktskyldigst formanet han ministerpresident Li Peng om at dutifully urge.past he minister-president Li Peng on that også Midtens rike måtte overholde too Middle.gen land must respect menneskerettighetene. (HPMHS1TN) people-rights.the

In (11a), the subordinate clause has (finite) verb-second order, the topic position being filled with a spatial adjunct. The analogue (11c) (except for exhibiting a subordinate clause word order) is more restricted when it comes to allowing a topicalized adverbial. Reordering the complements as in (11d) would distort the

. The German reader may be more prepared to have to wait for the lexical verb. . There appears to be a stronger tendency in Norwegian to avoid an indefinite subject not only in topic position. Considering our introductory remarks as to the mapping of components, again Norwegian is seen to be more generally conforming to such avoidance than German.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.19 (1184-1225)

Information structure, syntactic structure and passive 

theme-rheme sequence which makes (11b) the optimal translation, promoting the adjunct to subject.34 (11) c.

Pliktskyldigst formanet han ministerpresident Li Peng om at også [i Midtens rike] måtte [menneskerettighetene] overholdes adv subj keep.pass d. Pliktskyldigst formanet han ministerpresident Li Peng om at [menneskerettighetene] måtte overholdes [også i Midtens rike] subj adv

Clearly, the translator intends to adhere to the theme-rheme order of the original. In (11), this results in the choice of an active construction. This is similar to what happens in example (2) except for the topic position not being involved. I repeat the example, adding the alternatives: (2) a.

Inzwischen wurden von der IAETA und der WHO In the meantime pass by the IAETA and the WHO Richtlinien zur Strahlenbehandlung von Nahrungsmitteln guidelines for.the irradiation-treatment of food-articles erarbeitet. (UR1) prt-work.ppart b. I mellomtiden har IAETA og WHO utarbeidet in meantime.the has IAETA and WHO out.work.ppart retningslinjer for strålebehandling av næringsmidler. guidelines for irradiation-treatment of food c. I mellomtiden er retningslinjer for strålebehandling av In meantime have guidelines for irradiation-treatment of næringsmidler blitt utarbeidet av den internasjonale food.articles pass ppart by the international Atomenenergiorganisasjonen (IAETA) og WHO. nuclear-energy-organisation IAETA and WHO d. I mellomtiden er det blitt utarbeidet retningslinjer for strålebehandling av næringsmidler av IAETA og WHO.

Both (2c) and (2d) retain the voice of the original. However, as opposed to the German original, the promoted subject cannot remain in the object position. In (2c), it moves into the position after the finite verb and in (2d) an expletive, i.e. det, has been inserted in the subject position allowing the phrase corresponding to the subject in the original to remain in object position, followed by the agent phrase, i.e. the av-adjunct. Note that (2c) and (2d) are both acceptable, though . In a sense, this can be regarded as an example of reframing as described by Doherty; the PP adjunct in German is turned into the subject of an active clause in Norwegian.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.20 (1225-1287)

 Anneliese Pitz

due to the indefiniteness of the subject and the informational status, (2d) with the expletive subject is judged the better of the two. However, as in the previous examples, adhering to the voice of the original results in a change of the themerheme sequence which might explain why (2b) is preferred. . Change of topic So far, the choice of either active or passive in the translation has been explained by the translator’s intention to adhere to the information structure of the original and deviations are seen as devices to circumvent structural limitations. In the next example, we are confronted with a change in the information structure. The change of voice by itself does not result in an unambiguous and acceptable construction and there is an additional change in the ordering of thematic elements. (12) a.

b.

c.? d.? e.??

Wie wirkt die Ernährung mit Früchten auf den menschlichen how works the nutrition with fruits on the human Organismus? organism Der gesamte Stoffwechsel wird durch sie in Bewegung the.nom whole metabolism pass by her in movement gebracht, insbesondere wird der Eiweißaufbau in den bring.ppart especially pass the protein-up-build in the Bauchorganen angeregt. abdomen-organ stimulate.ppart ‘What do fruits do for the human organism when eaten. The whole of the metabolism is enlivened, and protein formation in the abdominal organs is especially stimulated.’ Hvordan virker ernæring med frukt på den menneskelige organisme? Den bringer bevegelse i hele stoffskiftet, og særlig it bring movement in whole metabolism.the, and especially proteinoppbyggingen i bukorganene stimuleres protein-up-build.the in abdomen-organs stimulate.pass [Hele stoffskiftet] bringes gjennom den i bevegelse. whole.metabolism.the pass through it in movement [Hele stoffskiftet] blir brakt i bevegelse gjennom den. bring.ppart through it [Hele stoffskiftet] bringer den i bevegelse. (subj?) (obj?)

As in the previous examples, the analogue (12c) does not give an optimally acceptable construction, the position after the finite verb not being the base position for adjuncts in Norwegian. In the German original, the agent phrase (or instrumental

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.21 (1287-1364)

Information structure, syntactic structure and passive 

phrase) durch sie occupies a canonical adjunct position. Moreover, according to the principle of balanced information distribution, BID, it is the position for the least specific or most accessible information (Doherty 2003). Movement of the PP gjennom den ‘through it’ to sentence-final position as in (12d), i.e. after the verbal complex consisting of the light verb bringe and the PP i bevegelse, does not improve the construction since the phrase constitutes background information. (12e), with the theme-rheme sequence of the original, has to be rejected as an alternative to (12b) since the initial DP will be interpreted as the subject. We can analyze the difference between (12a) and (12b) as an instance of language specific information balancing in the sense of Doherty (2003): the PP durch sie refers to the subject of the preceding interrogative clause and can be said to be the most easily accessible information unit. As such, it will be placed between two more relevant constituents: the verb-PP-complex hosting the new information and the constituent which is slightly farther away in that it is not the core of the preceding question. The subject in topic position can be said to stand in a partitive relation to the NP den menschlichen Organismus in the preceding clause. In Norwegian, there must be some material in focus position, i.e. after the verb, and the PP gjennom den, representing background information, clearly cannot fill this position. Consequently, in (12b), the most easily accessible unit is topicalized and what appears as the partitive theme in the original is placed in the post-verbal focus position. In Doherty’s terminology, the information structure in (12b) can be represented as the sequence 3-1-2, where 1 is the value assigned to the phrase with the highest degree of informativity and hence focus. The structure represented by (12c) would be 2-3-1 which is not appropriate in Norwegian. Example (13) largely illustrates the same point, i.e. the fact that change of voice might not be sufficient to accommodate the information structure of the original. (13a) has a sentential topic which is fully acceptable in German but which will be avoided in Norwegian. The way to achieve this is by reframing the durch-PP as the subject of an active construction. This, however, involves what Doherty (1996) describes as the conceptual extension or lexical flexibility of the verb with respect to the subject role. The verb erklären (‘explain’) preferably selects an agentive subject. The entity referred to by the durch-PP, i.e. diese Befunde, importantly is not a prototypical agent and hence not realized as the subject. In Norwegian, however, the PP is translated as the subject of the (agentive) verb forklare and then topicalized.35 (The context for the example sentence states that frogs get their territory by chasing other frogs with the noises they make.) . It is worth mentioning that in 98 of the 277 active translations of German passives, a PP is turned into the subject. In half of these cases, the PP corresponds to a by-phrase. Concerning the subject of the remaining active constructions, it can be noted that in 85 constructions the pronoun man is inserted and in 35 cases, we find another pronoun (vi ‘we’, de ‘they’).

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.22 (1364-1399)

 Anneliese Pitz

(13) a.

(Dies wird, wie amerikanische Forscher neuerdings herausgefunden haben, ganz einfach dadurch bewirkt, daß jedes Tier vor dem Quaken eines Artgenossen davonläuft.) Wie sich allerdings die Weibchen, die bei den meisten how refl the.nom females which by the most Fröschen stumm sind, über das Gebiet verteilen wird frogs dumb are over the territory distribute pass durch diese Befunde nicht geklärt. (KOL1) through these accounts not clarify.ppart ‘(As American scientists have recently discovered, this distribution is effected quite simply by the fact that every frog avoids the quacking sound of his own species.) This explanation, however, does not account for the distribution of the females, for these, in most frogs, are dumb.’ (KOL1TE) b. (Som amerikanske forskere ganske nylig har funnet ut, skjer dette helt enkelt ved at hvert dyr flykter bort fra det stedet der det hører kvekkingen fra en artsfelle.) Disse iakttakelsene kan likevel ikke forklare hvordan hunnene, these observations can however not explain how females.the som hos de fleste froskearter er stumme, fordeler seg over who by the most frog-types are dumb distribute refl over området. (KOL1TN) region.the

Norwegian thus chooses the same translation as English, where we expect this type of change. In German, however, this would not be an acceptable solution.36 In (13b), as in the previous example, the information structure is changed and the same explanation avails itself. However, in this case, it is not so much the risk of getting into a garden path which affects the choice of construction in German, but the heavy topic. . Additional support: Introducing an implicit agent Returning to the examples in (5)–(8) which illustrate the translation of ergative and reflexive constructions, there is no doubt that these correspondences can be taken as support for the description of voice changes given so far; they clearly conform to the theme-rheme sequence of the original and they do so at the expense of introducing an implicit agent. It might be argued that the latter need not be . Interestingly, not only are PPs turned into subjects in the translations from German to Norwegian, but Norwegian subjects are turned into PPs, similarly to subjects in the translations from English to German. Cf. Doherty (2003).

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.23 (1399-1427)

Information structure, syntactic structure and passive 

intrusive since all activities and processes necessarily involve some agent or causer and that this agent/causer generally can be inferred from the context. Doors, for instance, generally do not open completely by themselves although the situation is presented this way in example (6) above. Still, we could argue that it was the intention of the author to present the situation as if there were no agent/causer. A passive translation of these constructions, neglecting the choice of the author, shows even more clearly that identical theme-rheme order is more important than the argument structure of the verb.37 One may still wonder why Norwegian chooses passive when the analogue is available as in (5)–(7). We may have to resort to an explanation involving the notion of preference, as proposed by Doherty (1996), for some differences between German and English; Norwegian prefers passive constructions to reflexive and ergative constructions. We may alternatively assume that it is German which prefers reflexive and ergative constructions.

. Concluding remarks Assuming that the source language text exhibits the most appropriate information structure, we expect the translation to adapt it to the target language. The effort to adhere to the theme-rheme sequence may cause processing problems or result in marginal constructions due to structural differences between languages. For the languages investigated, these differences consist in the (A)(S)OV order in German as opposed to the SVO(A) order in Norwegian. In addition, Norwegian has a strong preference for subjects in Topic position. Change of voice turns out to be an essential device for avoiding interpretation problems and ungrammaticality by producing the expected word order. While not all deviations can be explained by syntactic constraints or reduced to structural differences, the study shows that the choice of voice is to a large extent determined by information structuring concerns, and more concretely, by the concern with adhering to the theme-rheme sequence of the source sentence.

. This is reminiscent of Leiss’ claim that thematicity outweighs agentivity in the structuring of a sentence. While Leiss’ description captures the mapping conditions in one language, the choice of voice in translation indicates that it applies more generally: information structure ranks above other properties, even the involvement of an agent.

JB[v.20020404] Prn:6/11/2006; 9:36

F: LA9609.tex / p.24 (1427-1546)

 Anneliese Pitz

References Abraham, Werner (1986). Word order in the middle field of the German sentence. In Werner Abraham & Sjaak de Meij (Eds.), Topic, Focus, and Configurationality (pp. 15–38). Amsterdam: John Benjamins. Abraham, Werner (1995). Deutsche Syntax im Sprachenvergleich [Studien zur deutschen Grammatik 41]. Tübingen: Gunter Narr. Åfarli, Tor (1992). The Syntax of Norwegian Passive Constructions. Amsterdam: John Benjamins. Doherty, Monika (1996). Passive perspectives: Different preferences in English and German – A result of parametrized processing. Linguistics, 34(3), 591–643. Doherty, Monika (2003). Parametrized beginnings of sentences in English and German. Across Languages and Cultures, 4(1), 19–51. Doherty, Monika (2004). Reorganizing dependencies. SPRIKreports 23. Oslo: University of Oslo. URL: http://www.hf.uio.no/forskningsprosjekter/sprik/publikasjoner/sprikreports.html Dowty, David (1991). Thematic proto-roles and argument selection. Language, 67, 547–619. Eroms, Hans-Werner (1974). Beobachtungen zur textuellen Funktion des Passivs. In ErnstJoachim Schmidt (Ed.), Kritische Bewahrung (pp. 162–184). Berlin. Faarlund, Jan Terje, Svein Lie, & Kjell Ivar Vannebo (1995). Norsk referansegrammatikk. Oslo: Universitetsforlaget. Fanselow, Gisbert (1992). “Ergative” Verben und die Struktur des deutschen Mittelfelds. In Hoffmann (1992), 276–303. Grimshaw, Jane (1990). Argument Structure. Cambridge, MA: MIT Press. Haider, Hubert (1984). The case of German. In Jindrich Toman (Ed.), Studies in German Grammar (pp. 65–101). Dordrecht: Foris. Hellan, Lars (1988). Anaphora in Norwegian and the Theory of Grammar. Dordrecht: Foris. Hoffmann, Ludger (Ed.) (1992). Deutsche Syntax. Berlin: de Gruyter. Lambrecht, Knud (1994). Information Structure and Sentence Form. Cambridge: Cambridge University Press. Leiss, Elisabeth (1992). Die Verbalkategorien des Deutschen. Berlin: de Gruyter. Nilsen, Øystein (2000). The Syntax of Circumstantial Adverbials. Oslo: Novus Press. Pittner, Karin (1999). Adverbiale im Deutschen: Untersuchungen zu ihrer Stellung und Interpretation. Tübingen: Stauffenburg. Primus, Beatrice (1999). Cases and Thematic Roles. Tübingen: Niemeyer. Solfjeld, Kåre (1998). Sententialität, Nominalität und Übersetzung. Frankfurt am Main: Peter Lang. Solfjeld, Kåre, & Cathrine Fabricius-Hansen (1994). Deutsche und norwegische Sachprosa im Vergleich. InfoDaf, 21, 499–510. Steinbach, Markus (2002). Middle Voice. Amsterdam: John Benjamins. Teich, Elke (2003). Cross-Linguistic Variation in System and Text. Berlin: Mouton de Gruyter. Vallduví, Eric, & Elisabet Engdahl (1996). The linguistic realization of information packaging. Linguistics, 34(3), 459–519. Zifonun, Gisela (1992). Das Passiv im Deutschen. In Hoffmann (1992), 250–275.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.1 (46-112)

Syntax and semantics of the deontic WANT-passive in Italo-Romance Eva-Maria Remberger The aim of this paper is to discuss the syntax and semantics of the deontic WANTpassive construction as manifested in some varieties of (Italo-)Romance. I will embed the data at issue in a general typology of various WANT-constructions with divergent semantic and syntactic subcategorization frames and parallel modal shifts. For the analysis of the syntactic structure of the WANT-passive, I am adopting a minimalist approach, based on the operation Merge and the mechanism of (semantic and syntactic) feature checking. In analyzing the semantics of WANT-constructions in general, I try to integrate volitionality and its semantic shifts in an appropriate formal representation, in order to show that WANT is a verb which is predestined to undergo grammaticalization processes.

.

Introduction: The data

In some varieties of (Italo-)Romance there are periphrastic passive constructions which use an auxiliary that is identical to a (matrix) verb meaning ‘want’. These passive constructions are characterized by a deontic modality of necessity.1 Consider the following examples from Calabrian (Cal., cf. (1)), Regional Southern Italian (S-it., cf. (2)), Sardinian (Sard., cf. (3)–(6)), Salentino (Sal., cf. (7) and (8)), Friulian (Friul., cf. (9)) and Venetan (Ven., cf. (10)): (1) Cal.

Tutti i figghjoli vonnu amati. All the children.m.pl.nom WANT.3.pl love.part.m.pl “All children must be loved.” (Informant from Bovalino, R.C.)

(2) S-it.

Questo lavoro vuole fatto con molta This work.m.sg.nom WANT.3.sg do.part.m.sg with a.lot.of precisione. precision. “This work should be done precisely.” (GGIC,II:152)

. In the following, deontic is used merely to denote deontic modality of necessity. Deontic possibility is not relevant in the context presented here.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.2 (112-185)

 Eva-Maria Remberger

(3) Sard. Cussa dzente keret tímita. These people.f.sg.nom WANT.3.sg fear.part.f.sg “These people must be feared.” (Jones 1993: 125) (4) Sard. Deu bollu agiudau po fai is iscalas. I WANT.1.sg help.part.m.sg for do.inf the steps. “I need help to climb up the steps.” (Sa-Limba 1999–2005) (5) Sard. Custa cícara de caffei bollit buffada. This cup.f.sg.nom of coffee WANT.3.sg drink.part.f.sg “This cup of coffee must be drunk.” (Blasco Ferrer 1986: 167) (6) Sard. Tui bolis arrestau. You WANT.2.sg arrest.part.sg.m “You should be arrested.”

(Sa-Limba 1999–2005)

(7) Sal.

(pro) Uliano cunsulate. 3.f.pl.nom WANT.3.pl.past comfort.part.f.pl “They [women] had to be comforted.” (Salvioni 1911: 379)

(8) Sal.

Lu pesce ulia mangiatu this.morning The fish.m.sg.nom WANT.3.sg.past eat.part.m.sg stammane. “The fish should have been eaten this morning.” (Salvioni 1911: 379)

(9) Friul. Al û savût. cl.m.sg.nom WANT.3.sg know.part.m.sg “It should be known.” (10) Ven.

El vole magná. cl.sg.m.nom WANT.3.sg eaten.part.sg.m “It should be eaten.”

(Ledgeway 2000: 258)

(Ledgeway 2000: 258)

This passive construction shows the typical raising of the internal argument to subject position; this results in the assignment of nominative case to the internal argument and its agreement with the WANT-auxiliary. The external argument is demoted, i.e. unexpressed. As will be shown later, the demoted argument can – at least in some Romance varieties – be reactivated by a prepositional adjunct. The aim of this article is to discuss the syntax and semantics of the deontic WANT-passive construction, laying out its relation to the syntax and semantics of other WANT constructions (mainly in Romance). In the remaining part of the introduction, I will describe some general properties of the WANT-passive construction (see 1.1), offer a short cross-linguistic excursus (see 1.2), and point out differences to other WANT constructions which might seem to be related (see 1.3). In the second section, I give a general characterization of passive constructions in generative grammar (see 2.1). I show that the WANT construction under discussion falls within this characterization of the passive (see 2.2). In Section 3, I propose a derivation of the passive, laying particular weight on the status of the external argument. In Section 4, I give a short introduction to some general

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.3 (185-246)

The deontic WANT-passive in Italo-Romance 

minimalist assumptions relevant to the WANT-passive (see 4.1) and I extend this derivation to the deontic WANT-passive (see Section 4.2). In Section 5, I sketch a general typology of various WANT-constructions; besides presenting a model whereby volitionality and its semantic shifts are integrated in an appropriate formal representation. In Section 6, a short summary concludes the paper. . Some general properties of WANT-passives The passive construction with WANT appears mainly in the 3rd person present tense, but it may also occur in other persons (see (4) and (6)).2 Irrespective of person, as soon as there is an animate subject which might have an interest in the event encoded by the predicate, as in (1), (4) and (7), a volitional3 interpretation is still not totally excluded.4 However, in the examples (2), (5), (8), (9) and (10), since the subject is clearly inanimate, there is no such interpretation. The Romance WANT-passive can also be found in other tenses than the present (see (7) and (8)).5 I assume that passive auxiliaries, like the prototypical copular auxiliary BE, are merged only late in the Romance languages treated here.6 This is because the higher tense projection TP (the IP in other frameworks) needs . Although it is natural for Sardinian speakers to produce the whole paradigm for the expression in (6), sometimes they have difficulties accepting it for the first person singular. This seems to be an issue of world knowledge: (i)

a. ?Deu bollu arrestau. WANT.1.sg Tui bolis arrestau. WANT.2.sg etc.

(Sa-Limba 1999–2004)

But, as (4) shows, constructions in the first person singular become more natural depending on the meaning of the main verb participle. . Sometimes volitive modality is subsumed under deontic modality. Palmer (1991) treats it not as deontic but as dynamic event modality. In this paper, I don’t use the term volitive but I use the notion volitional in the sense that there is an interested party as a participant in argument structure. . As a reviewer remarks, a volitional participant does not necessarily need to be identical to an agent or an external argument. . This (and fn. 2) is in clear contrast to Blasco Ferrer (1986: 167) who claims that, in Sardinian, the construction is allowed only in the third person and only in present tense. . In Italian, an exception has to be made for the auxiliary participle stato (a suppletive form from STARE ‘to stand’) which can appear in a copular construction but hosts some temporalaspectual information: therefore it is merged earlier. Unfortunately, I cannot go into syntactic details here (cf. Remberger 2002), but I would make note that there is no stato-counterpart for WANT constructions.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.4 (246-293)

 Eva-Maria Remberger

a purely verbal element (cf. Remberger 2002, 2006). Besides this, it can also be assumed that modality is mainly encoded in the tense projection (or even further up). Therefore it is no surprise that the deontic passive auxiliary WANT cannot appear in past perfect or present perfect tenses, i.e. in periphrastic tenses which encode perfective aspect, assumed to be located lower in syntax. WANT also cannot be combined with other modals, with the exception that it can appear in the conditional and the future in Sardinian (here both the conditional and the future are formed by a periphrastic construction involving a primarily modal auxiliary):7 (11) Sard. Sa lingua sarda iat a The language.f.sg.nom Sardinian [should.3.sg to].cond bolli imperada in sa comunicatzioni [...] WANT.inf learn.part.f.sg in the communication “The Sardinian language should be learnt in communication.” (Sa-Limba 1999–2005) (12) Sard. Cust’omine iat a bolli This man.sg.m.nom [should.3.sg to].cond WANT.inf arrestau. arrested.part.m.sg “ This man should be arrested.” (Sa-Limba 1999–2005) (13) Sard. Cuss’òmine at a bolli This man.sg.m.nom [have.3.sg to].fut WANT.inf mortu.8 killed.part.m.sg “This man will have to be killed.” (Sa-Limba 1999–2005)

I assume that the Sardinian future and conditional are encoded in the Tense Phrase, too (cf. Remberger 2002, 2006); therefore, both the periphrastic future / conditional auxiliary and the WANT-auxiliary can be considered to be merged late in syntax. . A cross-linguistic view In other languages too, WANT loses its “agent-oriented” volitional modality and gets a deontic meaning. Bybee, Perkins & Pagliuca (1994: 185–186) have the fol-

. To emphasize the difference between conditional and non-conditional constructions, the translation of (11) and (12) should rather be: “it WOULD be necessary that” (with conditional) vs. “it IS necessary that” (without conditional). . In Sardinian, mòrrere can be used as a transitive verb meaning ‘to kill’.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.5 (293-367)

The deontic WANT-passive in Italo-Romance 

lowing example from Tucano which shows a similar construction to the Romance WANT-passive, see (14): (14) Tuc.

ba‘a-ro ia-‘a eat.part WANT.imprs “One should eat.”

Bybee, Perkins & Pagliuca (1994: 185–186) also note that “because obligation is externally imposed, some of the source constructions for obligation are passive-like in structure.” That means that an unexpressed external argument as found in passive constructions facilitates an interpretation which involves external obligation; but it is the volitional modal context which seems to be particularly predestined for this shift towards external deontic necessity (see Section 5). In other languages too, when WANT itself is passivized and the passivized subject happens to be clearly unagentive, we can get a deontic reading; this is the case in the following German example:9 (15) Germ. Dieses Buch will gelesen werden. this book WANT.3.sg read.part become.inf “One should read this book.”

WANT here expresses deontic modality, and not volition. Still, here it is not WANT which has the function of a passive auxiliary, but the standard German passive auxiliary werden. A “true” deontic passive in German is only found with the marked passive auxiliary gehören ‘to belong to’ + participle.10 Indeed, a perfect translation into German of the Sardinian example (6), for example, would give: (16) Germ. Du gehörst eingesperrt. you BELONG.TO.2.sg arrest.part “You should be arrested.” . In German, WANT can also have an epistemic interpretation, preferably when its infinitival complement contains a compound time: See the following example in (i) where the subordinate proposition is put in doubt by the speaker: (i)

Germ. Er will noch nie betrogen worden sein. He WANT.3.sg yet never betrayed.part become.part be.3.sg “He claims never to have been betrayed.”

For a comparative study of German modals, cf. Heine (1995): He takes into account several “conceptual properties” (deontic force, controlling agent, dynamicity of the event, occurrence of the event related to reference time, degree of probability) which play a role in the interpretation of deontic vs. epistemic modality (Heine 1995: 29). . The semantic shift of verbs of possession (like HAVE or BELONG TO) to obligation is a well known phenomenon (cf. Heine & Kuteva 2002: 243–245) which cannot be treated here.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.6 (367-440)

 Eva-Maria Remberger

Yet in some English varieties, namely in colloquial Scottish English and in some spoken varieties of the American Midlands (which seem to be influenced by Scottish and Scots-Irish immigrants, cf. Murray & Simon 1999; Murray & Simon 2002), there are true deontic WANT-passive constructions, see the following examples:11 (17) S.-Ohio

Do you want smacked? (http://linguistilist.org/∼ask-ling/archive-1998.4/msg00126.html)

(18) Midland AE

The baby wants picked up.

(Murray & Simon 2002: 32)

(19) Midland AE

The cat wants fed.

(Murray & Simon 2002: 32)

(20) Midland AE

The car wants washed.

(Murray & Simon 1999)

The dialectal English WANT + participle construction is only allowed for speakers who admit the NEED + participle construction (like the car needs washed, cf. Frazer, Murray, & Simon 1996). The same is valid for the LIKE + participle construction discovered by Murray & Simon (2002): expressions like babies like cuddled seem to be allowed only for speakers who can use both the NEED and the WANT constructions (cf. Murray & Simon 2002: 59).12 Besides, as noted by Murray & Simon (1999: 160), not all speakers who use the WANT + participle construction would allow it with an inanimate subject, as in (20) (with the NEED construction they would). . Thanks to the participants of the “Demoting the agent”-workshop in Oslo, especially to Gillian Ramchand, who gave me the hint to the English varieties. The NEED-WANT-correlation needs further research, see also the WANT/NEED + gerund construction, cf. Quirk et al. (1997: 1176, 16.26): (i)

Your shoes need mending. / The door needs painting.

There it is said that “the above use of need is often replaced, in dialectally restricted usage, by an equivalent use of want.” Also in Latin American Spanish there seems to be WANT-passive constructions (Volker Gast, p.c.) as well as NEED-passive constructions (e.g. necesita lavado); since I have not gathered enough data yet I will not consider them here. . A brief internet survey (google) brought up the following examples for WANT + participle: (i)

[...] and I still have hair that wants washed.

(ii) I sincerely believe Mr. Reeve wants cured as badly as I do. (iii) I want to help everyone who WANTS helped. (iv) I think right now how she wants helped is by copying the tape [...] (v) Why is it OK for Kerry to own one but he wants helped to pass a law forbidding everyone else from owning one?

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.7 (440-530)

The deontic WANT-passive in Italo-Romance 

. Ledgeway’s WANT-passive in Romance In Romance, there are constructions similar to the ones in (1) to (10), see the sentences from Neapolitan in (21) to (23) below. They are all classified by Ledgeway (2002: 236–237, particularly Chapter 7) as different types of WANT-passives (subcategorized in Exceptional Case Marking, obligatory control, and subject raising, in his terms): (21) Neap. Mario vô mannata chella lettera. Mario.m.sg.nom WANT.3.sg send.part.f.sg this letter.sg.f.acc “Mario wants this letter to be sent (to him).” (Type 1: ECM) (22) Neap. Mario vô esse mannata chella Mario.m.sg.nom WANT.3.sg be.inf send.part.f.sg this lettera. letter.sg.f.acc “Mario wants to be sent this letter.” (Type 2: control) (23) Neap. Vô esse mannata chella lettera. WANT.3.sg be.inf send.part.f.sg this letter.f.sg.nom “This letter must be sent.” (Type 3: raising)

However, contrary to Ledgeway, I would not treat these examples as WANTpassives. The first type in (21) is not a true WANT-passive, but a construction where WANT subcategorizes a small clause (SC) (i.e. predication phrase PrP, cf. Bowers 1993, 200113 ). The subject of the SC is assigned accusative case by the matrix verb (hence ECM). Since the SC predicate is a participle, the result might be interpreted as a passive (especially because one of the argument positions of the ditransitive verb is not explicit14 ). Nonetheless WANT still has a clear volitional θ-marking, i.e. there is an external argument which represents the interested party in the event. Ledgeway’s second type in (22) represents a control structure which involves passivization, but here it is the indirect object PRO which is raised to subject position in a passive structure controlled by WANT, as is familiar from English

. The predication phrase according to Bowers does not necessarily need to be verbal: it is a generalized version of all kinds of predication, small clauses included, see fn. 26. . This implicit θ-role + “affectedness” of the volitional subject can trigger participle agreement with the subject (see the following example from Cosenza), which again, according to Ledgeway (2000: 251f.), is attributed to an underlying control structure: (i)

Cos.

zia Isa vo accattata ’u café aunt Isa WANT.3.sg buy.part.sg.f the coffee.sg.m “Auntie Isa wants the coffee bought (for her).”

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.8 (530-665)

 Eva-Maria Remberger

dative shift constructions.15 Although, in this sense, there is an embedded passive construction, WANT maintains its volitional semantics. Only in type 3 in (23) do we have a non-volitional, i.e. deontic, reading of WANT, since the subject of WANT is non-agentive and inanimate.16 Yet, this is not a true passive construction with WANT as an auxiliary, as the ones in (1)–(10). The passive construction itself lies in the infinitival phrase selected by WANT and is encoded by the auxiliary esse/BE + participle (it is therefore a parallel construction to the German example (15)). Hence the structures underlying (21) to (23) are the following: (24) [CP Mario [VP vô [SMALL CLAUSE mannata chella lettera]]] (25) [CP Marioi vô [CP PROi essePASSIVE AUXILIARY mannata chella lettera ti ] (26) [CP Vô [TP essePASSIVE AUXILIARY mannata chella lettera]]

WANT cannot be claimed to be a passive auxiliary in these constructions, as it can for constructions like (1) to (14) or also (17) to (20). In (24) and (25) there is a clear volitional external argument of WANT in subject position whereas in (26) we have another overt passive auxiliary (namely BE/esse) which means that WANT here is a mere modal marker.

. Ledgeway (2000: 304, fn. 4) refuses this interpretation: “Subjectivization of the indirect object with ditransitives embedded under VOLERE cannot be derived from a previous active structure in which the indirect object has been promoted to direct object, as in the Neapolitan example in (i): (i)

’o mannajeno ’na lettera him.acc send.past.3.pl a letter ‘they sent him a letter’

[...] we noted that the possibility of ditransitive structures like (i) proves the exception, rather than the norm, among Southern Italian dialects. Consequently, in most southern dialects [...] there are no active structures like (i) from which indirect object passives under VOLERE can be derived” (cf. also Ledgeway 2000: 239). Strangely enough, Ledgeway himself confirms that “far from being exceptional, the system of Case-marking described above for Neapolitan [i.e. that indirect objects may be referenced by an accusative clitic] appears to operate in other southern dialects” (Ledgeway 2000: 49). However, in Neapolitan at least, functional elements like the masculine singular accusative clitic ’o can refer to a specific, animate, i.e. personal indirect object. So why should this not be possible for the even more functional element PRO in a control structure? . Interestingly, WANT does not function as a control verb here, but it would, if the subject was human.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.9 (665-665)

The deontic WANT-passive in Italo-Romance 

. General remarks on the passive . Some proposals on passive constructions Haider (1986: 17) discusses several constructions in German which in the literature are analysed as passives;17 for him, however, a passive strictu sensu can only be a construction which is unambiguously encoded by a morphological process (e.g. the participle of the main verb in combination with a passive auxiliary) which reduces the designated (i.e. the external) argument.18 Also Jaeggli (1986) pursues the idea that passivization is a morphological process. For him, it is the participial verbal morphology which is responsible for the suppression of accusative assignment to the direct object and the absorption of the external θ-role (which is not restricted to agent19). But Jaeggli notes that, in some passive constructions, there might also be an overt direct object or no argument bearing structural Case at all; for example, when the indirect (instead of the direct) object is raised, there can be

. In German, besides the usual event passive construction with werden/BECOME, see (i), there is the so-called stative passive with sein/BE, see (ii), the related construction with bekommen/GET, see (iii), the above mentioned (cf. also (16)) deontic construction with gehören/BELONG TO, see (iv), the impersonal passive (which is also possible with intransitive verbs), see (v), and the impersonal reflexive construction (also termed ‘middle reflexive’, cf. Lekakou in this volume), see (vi): (i) (ii) (iii) (iv) (v) (vi)

Der Mann wird verhaftet. “The man is arrested.” [event] Der Mann ist verhaftet. “The man is arrested.” [state] Der Mann bekommt Geld geschenkt. “The man is given money.” Der Mann gehört verhaftet. “The man should be arrested.” Es wurde getanzt. “There was dancing.” Das Buch liest sich gut. “This book reads well.”

=> werden/BECOME => sein/BE => bekommen/GET => gehören/BELONG TO => impersonal passive (with expletive es) => impersonal reflexive construction

. Therefore, of course, not all constructions mentioned in fn. 17 are passives for Haider. . Cf. Jaeggli (1986: 599): (i) (ii) (iii) (iv)

Engl. Bill was killed by Mary. Engl. The package was sent by John. Engl. The letter was received by Bill. Engl. The professor is feared by all students.

(agent) (source) (goal) (experiencer)

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.10 (665-682)

 Eva-Maria Remberger

an overt direct object in a passive construction;20 or when an expletive is merged in the subject position of a passive construction, there can be no Case bearing argument at all.21 Burzio’s generalization (cf. Burzio 1986), which claims an interdependency between blocking structural case and missing an external θ-role is therefore not tenable for all passive constructions. Since impersonal passives can also occur with intransitive verbs (e.g. (v) in fn.17), Bowers’ definition that “it is not the Case assignment per se that determines passivization but the more fundamental property of transitivity” (Bowers 2002: 209) cannot be considered valid either. The main characteristic of passive constructions seems to be the fact that there is no overt external argument (and, of course, also no little pro since there is no overt counterpart of pro in the same position, hence no overt Case22 ) in an argument position. This is a view also expressed by Klaiman (1991) who calls the passive a marked structural voice configuration which shows “alternations in the configuration of nominal statuses with which verbs are in particular relationships”, i.e. omission and/or rearrangement; therefore the passive is mainly associated with voice “rules which affect the semantic role content of the subject position in structural configurations” (cf. Klaiman 1991: 1–43). In light of the considerations exposed above, I will assume the following operational definition for passive constructions: (27) A true (sentential) passive construction is given when an external θ-role is not realized as an overt argument, but if it can still be reactivated by an adjunct structure.

As shown later, the external θ-role is still implicitly present. Thus the notion of “reactivation” can be understood as ‘making something explicit which is unexpressed,

. See the English passive in dative shift constructions, such as in (i), or the GET-passive in several Germanic languages (see (ii) and fn.17, (iii)); there are also parallel structures with HAVE in Italian (here Neapolitan), see (iii) from Ledgeway (2000: 238–239); (i)

Mary was given the book by Peter.

(ii) Maria bekommt von Peter Blumen geschenkt. Maria GET.3.sg from Peter flowers give.part (iii) essa ha ’vuto rialata ’na bicicletta. she HAVE.3.sg HAVE.part give.part.f.sg a bycicle.f.sg “She was given a bycicle.” . See, for example, the impersonal passive in German (see fn. 17, (v)). . Sternefeld (1995) in fact assumes there to be a little pro in the specifier position of a functional Voice Phrase. Consequently he maintains that there is no case absorption in passive constructions. I will come back to this approach later (see Section 3).

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.11 (682-754)

The deontic WANT-passive in Italo-Romance 

but semantically present’. The Romance WANT-passives treated here comply with the conditions of this definition as will be shown in the following subsection. . The classification of the Romance WANT-passive The following examples from Italo-Romance show that the external argument θrole can be reactivated by a PP-adjunct:23 (28) Cal.

(pro) Voli ggiustata i cocchjunu. f.sg.nom WANT.3.sg adjust.part.f.sg by somebody “It should be adjusted by somebody.” (Speaker from Bovalino)

(29) Cal.

(pro) Voli ‘ncoraggiatu i cocchjunu. sg.m.nom WANT.3.sg encouraged.part.m.sg by somebody “He must be encouraged by somebody.” (Speaker from Bovalino)

(30) Sard. Sa mákkina keret accontzada dae unu the car.f.sg.nom WANT.3.sg adjust.part.f.sg by a meccánicu. mechanic “The car must be repaired by a mechanic.” (Jones 1993: 125) (31) Sal.

(pro) Vole di li mmia manu midicata. f.sg.nom WANT.3.sg by the my hands cured.part.f.sg “She should be cured by my hands/by me.” (Salvioni 1911: 379)

Summarizing the general characteristics of the Romance WANT-passive we can make note of the following properties: a. no overt / active / explicit external θ-role in argument position b. no accusative Case c. raising of the internal argument into subject position d. nominative Case assignment between TP/IP and the raised subject e. agreement between WANT and the raised subject f. possibility to reactivate the external θ-role, i.e. to make it explicit (in most of the varieties) g. morphosyntactically periphrastic form (auxiliary + participle) h. deontic modality . It is an open question whether really all of the varieties of Italo-Romance mentioned in (1) to (10) allow a reactivation of the demoted external θ-role. Probably different kinds of PPadjuncts should also be discussed for every variety. I leave these and other questions concerning parametric variation of the data to future research. In the case that some of the varieties do not admit reactivation (as it might be with Friulian or Venetan) we might have a parallel case to the German so-called stative passive (see fn. 17, (ii)), where the status as a passive construction is still discussed, among others, because of unclear reactivation facts (cf. the line of argument in Kratzer 1994, 2005; Rapp 1996; Maienborn 2005).

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.12 (754-817)

 Eva-Maria Remberger

. Passive constructions and the external argument in the MP In the following syntactic analysis, a minimalist approach (Chomsky 1995: Chapter 4) based on the operation Merge and the mechanism of (semantic and syntactic) feature checking shall be applied (the probing mechanisms of Chomsky 2000, 2001 are not yet considered here). From the outset, the status of the invisible external argument should be examined. Since this argument can be reactivated, it must be present somewhere in the structure, if only implicitly. Jaeggli (1986) assumes that a reactivated subject in a PP gets its θ-role because the morphology of the participle, which has absorbed it, is still able to transmit this θ-role to the preposition of the PP-adjunct. I would like to follow a proposal made by Baker, Johnson & Roberts (1989: 226), who draw a parallel between the invisible subject of a passive and the arbitrary (non specific) PRO. I will adopt this proposal relying on the following line of reasoning:24 If we compare unaccusative verbs and reflexive constructions with passive constructions, we find that the difference between them is that in passive constructions there is still an implicit external θ-role, while in constructions with unaccusative verbs this is not the case. In unaccusative verb constructions, the internal and external arguments are somehow the same (i.e. they have the same referent). Thus, one can never reactivate anything in a reflexive construction25 or in another construction with an unaccusative verb, see the following (Italian) examples: (32) It.

Mario è picchiato (da Gianni). Mario.m.sg.nom be.3.sg beat.part.m.sg (by Gianni) “Mario is beaten (by Gianni).”

(33) It.

Mario si è lavato (*da Gianni). Mario.m.sg.nom refl be.3.sg wash.part.m.sg (by G.) “Mario washed himself (*by Gianni).”

(34) It.

Mario è arrivato (*da Gianni) Mario.m.sg.nom be.3.sg arrive.part.m.sg (by Gianni) “Mario arrived (*by Gianni).”

. Also Kratzer (1994) suggests that there might be PRO involved. . I assume in line with Cocchi (1995, 1998) among others (cf. also Remberger 2002) that clitic reflexive constructions in Italian are clearly unaccusative, i.e. they involve a DP-raising of an internal object position to an external subject position. Approaches such as Reinhart & Siloni (2004) are untenable since they argue with different reflexive constructions cross-linguistically. There are fundamental differences between reflexive constructions in Romance and in Germanic: whereas German reflexive constructions clearly are not unaccusative, the Italian ones, which use the special clitic si, are (this can be convincingly shown by auxiliary selection, for instance).

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.13 (817-886)

The deontic WANT-passive in Italo-Romance 

Syntactic (positional) subjects of unaccusatives are internal and external arguments likewise: they have been merged in internal argument position but raised to external argument position. As has been noted (cf. Harley & Folli 2004; Kallulli 2006, this volume), agent-oriented adverbs are compatible with unaccusative verb constructions: (35) It.

Maria è arrivata apposta per questa Maria.f.sg.nom be.3.sg arrive.part.f.sg on purpose for this riunione. meeting “Mary arrived on purpose for this meeting.”

This means that there is an agent-oriented θ-role which can be referred to by the agent-oriented adverb in unaccusative verb constructions. The same is valid for passive constructions (as shown also by Sternefeld 1995, among others): (36) It.

La nave è stata affondata apposta. the ship.f.sg.nom has been.3.sg sink.part.f.sg on purpose “The ship was sunk on purpose.”

The difference between unaccusative verbs and passive constructions is that only in the latter can you reactivate the external argument using a prepositional adjunct. This is not possible in unaccusative verb constructions since there is no demoted external argument which could be reactivated: external and internal arguments are combined in the referent of the syntactic (positional) subject. The modifying adverb apposta, ‘on purpose’, in (35) clearly refers to a θ-role lying in the positional subject Mario. Therefore, it is not possible to introduce another agent-oriented referent (i.e. a reactivating PP-adjunct) in these constructions, since there is one already. If you take an ambiguous verb, i.e. one which could be used as a transitive as well as an unaccusative verb, like Italian affondare, ‘to sink’, the only reading you can get with a reactivated subject, is the passive one: (37) It.

La nave è affondata dall’ammiraglio. the ship.f.sg.nom be.3.sg sink.part.f.sg by the admiral “The ship was sunk by the admiral.”

Further arguments on similar lines dealing with the ‘visibility’ of the external argument have also been put forward by Sternefeld (1995). Using several tests on German and English passive constructions, he convincingly shows that there must be a structural position where the external θ-role is located. I agree with his line of argumentation but it doesn’t seem plausible to me that the unexpressed θ-role must be represented by a little pro. Within the set of empty (non overt) elements in syntax, little pro has Case, a θ-role and nominal features while big PRO is assumed to also have a θ-role and nominal features, yet no Case feature (or just a null Case

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.14 (886-938)

 Eva-Maria Remberger

feature, see below). Little pro is typical of null-subject-languages, like Italian or Spanish, where pro usually has nominative Case and is the bearer of the external θrole. Object pro (i.e. a pro which has other Cases than nominative) is not very common in Romance but can be found in other null-subject-languages, e.g. Basque. Pro can usually be paradigmatically substituted by an overt nominal phrase (i.e. a DP or an overt pronoun). I do not see why non-null-subject-languages, like German and English, should allow nominative, accusative, and dative pro in passive constructions (with no overt counterpart) when the same languages have no pro in canonical active clauses. Also, since there is no pro in German, in certain impersonal passive constructions (see fn. 17, example (v)) the expletive es is required to check nominative and nominal features. Thus, I will not follow Sternefeld’s approach with pro. PRO, on the contrary, is common to both null-subject- and non-null-subjectlanguages. It is typically the bearer of the external θ-role without the need of getting nominative Case. In a minimalist framework (cf. Chomsky 1995) it is assumed to have the so-called null Case. PRO can be arbitrary (as in uncontrolled infinitives) or controlled (e.g. by a referential antecedent in control infinitives). It is therefore an ideal candidate for the implicit, but somehow visible external argument stated by Sternefeld (among others). All unaccusative constructions have in common the raising of the internal argument to subject position (first to [Spec, vP/PrP26] then to [Spec, TP]27 ). In Remberger (2002, cf. also Remberger 2006), I assumed that in an unaccusative

. I will generalize the ‘little’ vP and interpret it as a predication phrase PrP as proposed by Bowers (1993, 2001). Bowers shows that a predication can be verbal, adjectival, prepositional or nominal; if there also is a Tense Phrase TP in the derivation, in the case of nominal predication we usually get copulative constructions. If there is no Tense Phrase, we find Small Clauses. A generalization of the little vP to a Predication Phrase has the vantage of capturing all kinds of predication. I assume that a PrP not only is responsible for the instantiation of the predication, but also for the encoding of the spatio-temporal relation between the Event Situation and Reference Situation (cf. Remberger 2002). In a minimalist approach, the assumption of a PrP is highly desirable since it reduces the number of functional categories. It then depends on the complement-feature hosted by the Pr-head (V, A, P, or N) if the predication is verbal, adjectival, prepositional or nominal. However, a detailed exposition concerning PrP would go far beyond the aims of this paper. . The TP, again, is the Tense Phrase. The Inflection Phrase, after the Split-Infl-Hypothesis (cf. Pollock 1989) has been split up into a TP and one or more Agreement Phrases, among others; minimalism according to Chomsky (1995) has reduced this structure again to a simple TP (therefore also abolishing AgrPs).

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.15 (938-1003)

The deontic WANT-passive in Italo-Romance 

predication there is a strong D-feature28 in Pr◦ which triggers this movement. We have seen that a passive Pr◦ can still implicitly maintain its capability to assign an external θ-role. Therefore I assume that a passive predication has its Specposition [Spec, PrP] occupied by PRO; that means that the difference between an unaccusative predication and a passive one is the presence of a null-Case-feature (cf. Chomsky 1995) in the predicational Pr-head. This also fits in the theory of control. If there is a controller, i.e. if there is the PP-adjunct typical for passive constructions, then we obligatorily have referential PRO (of course, there must be the compatibility of the θ-roles of controller and controlled PRO); if there is no PP-adjunct, then PRO remains uncontrolled and its reference is arbitrary.

. The WANT-passive in the MP . General assumptions In what follows, I will propose a minimalist derivation for the WANT-passive in Italo-Romance. I employ the general mechanisms of checking theory (mainly Merge and Move, cf. Chomsky 1995): Lexical and functional entries are merged into the derivation provided with featural content. These features have to be checked, overtly or covertly, in appropriate syntactic configurations. Categorial features can be weak or strong. Strong features must be checked overtly and therefore usually trigger overt movement. Syntactic parametrization lies in the featural composition of functional categories. Furthermore, I will assume that the Core Functional Categories (cf. Chomsky 2001) valid for verbal syntax are C, for the Complementizer Phrase, T, for the Tense Phrase, and v, the light verbal projection (the latter, as said before, generalized to Pr), and that there is the possibility to project multiple specifiers (cf. Chomsky 1995: 354–356). For the derivation of the WANT-passive, I propose that it works as other Romance passive constructions do (i.e. the ones with a modality-neutral interpretation), but that there is an additional modal (deontic) feature in (propositional) T. To summarize, I shall apply the following minimalist essentials: a. Every proposition has a predication phrase PrP (former little vP, cf. Bowers 1993, 2001); the Pr◦ is responsible for the argument structure of the predica-

. D here means ‘Determiner’, stemming from the nominal DP, i.e. the Determiner Phrase which has to be moved; a strong D-feature in a functional category might also be called EPPfeature, EPP meaning the realization of the Extended Projection Principle.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.16 (1003-1053)

 Eva-Maria Remberger

b.

c.

d. e.

tion;29 Pr is also responsible for the location (and, in case of aspect, also for the specification) of the event situation relative to reference time (cf. Remberger 2002, following a post-Reichenbachian approach, cf. Reichenbach 1947). PRO has null Case and Pr◦ can assign (check) null Case. In former theory, as mentioned above, PRO has no Case at all. Null Case is a technical improvement insofar as it can be checked by specific functional categories, e.g. by the ones relevant for control structures. There is no difference in feature composition between passive and perfective participles (cf. also Haider 1984; Sternefeld 1995), i.e. of relevance to these participles are the features [–finite] and [+perfective], and the fact that they might show nominal (and not verbal) agreement. Passive auxiliaries (at least in the Romance varieties treated here) are only merged late, i.e. under T◦ (cf. Remberger 2002 and what was said in 1.1). A deontic proposition has a deontic modal feature in T◦ .

. The derivation of the deontic passive with WANT The structure given in (38) (see the Sardinian example in (30)) represents the minimalist derivation of a WANT-passive construction:30 (38) Sard. Sa mákkina keret accontzada dae unu the car.f.sg.nom WANT.3.sg adjust.part.f.sg by a meccánicu. mechanic. “The car must be repaired by a mechanic.”

. E.g. there are several entries for verbal Pr◦ in the minimalist lexicon; these entries can be distinguished by their feature composition: a transitive predicational head has a featural content which differs from that of an intransitive or an unaccusative predicational head; a perfective predicational head will have a featural content which differs from an imperfective one. See the following overview for argument structure in Italian from Remberger (2002: 144): Table 1. Pr◦ in the minimalist lexicon Functional category Pr◦ Pr◦ Pr◦

Argument structure (di-)transitive intransitive unergative unaccusative

Features Pr, V-strong, D, accusative Pr, V-strong Pr, V-strong, D-strong

. This structure was produced using the GBX system, cf. Lalande (1997).

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.17 (1053-1086)

The deontic WANT-passive in Italo-Romance 

The passive Pr-head has a null Case feature, which can only be checked by PRO: The realization of the external θ-role therefore is automatically given by insertion of PRO in [Spec PrP]. The passive Pr-head also has a strong D-feature, which causes raising of the internal argument31 after the saturation of the θ-grid;32 this results in a multiple specifier construction: One specifier position is needed for the external argument given by PRO, another one is needed for the raised internal object. Since at some point we have an adjunct PP which is able to control PRO, PRO gets a referential interpretation. Further minimalist consequences are the movement of V◦ to Pr◦ (because of a strong V-feature in Pr◦ ), the movement of the Pr◦ -complex (called Vb = verbal complex in Chomsky 1995) to T◦ (because of a strong Pr-feature in T◦ , at least in Sardinian) and movement of the raised subject to [Spec TP] (because of a strong D-feature in T◦ ). Since there is still a finite verbal element missing, an auxiliary must be merged under (finite) T◦ . In the special case of the WANT-passive, it is not BE, the neutral passive auxiliary, which is merged, . The PRO cannot check this D-feature since checking of a strong feature is allowed only as a consequence of (overt or covert) movement (an exception being expletives). . I.e. assignment of the thematic θ-role to the direct object in [Spec VP], assignment of the (here agentive-eventive) subject-θ-role to [Spec PrP].

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.18 (1086-1140)

 Eva-Maria Remberger

but WANT: T◦ , which, besides being finite (hence with nominative), also has a deontic modal feature that must be checked. The only element compatible with this kind of T is auxiliary WANT.33 Finally, the Spec-head-configuration between the (double) raised subject and the auxiliary WANT allows φ-feature-checking (i.e. agreement).

. WANT: Subcategorization frames and modal shift . Syntactic typology of WANT-constructions The passive is obviously not the only construction in which WANT may appear. Considering a general typology of various WANT-constructions, WANT can be found in the following grammatical contexts:34 a. as a transitive verb with a DP complement: (39) It.

Voglio pace e libertà. WANT.1.sg peace and freedom “I want peace and freedom.”

b. with a finite clause as its complement (±subjunctive): (40) Sard. Cherzo chi istudies. WANT.1.sg that study.2.pl.subj “I want you to study.”

c. with a (±personal, ±inflected) infinitival complement (±control, ± obligatory restructuring, ±ECM, ±epistemic modality) (cf. Remberger 2005, for the whole series of examples): (41) Eng.

I want (him) to go.

d. with the so-called ‘balcanic’35 substitute for the infinitive (±subjunctive particle, ±subject obviation): . It is an open question why the Romance equivalent of need (e.g. Sard. bisonzat, It. bisogna) is not possible in this context. . Here in a mainly Romance perspective, but the parallels which can be drawn with other languages are obvious. The list is not exhaustive and gives only a few examples. For a first attempt to classify the data, cf. Remberger (2005). . The languages of the area of the Balkansprachbund are known to lack or avoid ‘true’ infinitival constructions, and to use a kind of subjunctive construction instead, cf. e.g. Calabrese (1992), Haase (1995), Tomić (2003).

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.19 (1140-1213)

The deontic WANT-passive in Italo-Romance 

(42) Cal.

Vogghiu i vaiu. WANT.1.sg that(particle) go.1.sg “I want to go.”

e. with a small clause complement (adjectival, participial, nominal, prepositional): (43) It.

f.

Suo padre lo vuole avvocato. his father him(clitic) WANT.3.sg lawyer “His father wants him to be a lawyer.”

as an impersonal verb (+locative/existential particle): (44) It.

Ci vogliono tre chili di mele. there(clitic) WANT.3.pl three kilogram of apples “One needs three kilogram of apples.”

g. as a deontic copula: (45) Cal.

’Sta pasta vo cu ru sugu. this pasta WANT.3.sg with the tomato sauce “This pasta goes with tomato sauce.” (Ledgeway 2000: 151)

h. as a future marker, as in Romanian: (46) Rom. O s˘a lucreze. WANT.invariable that(subj.particle) work.3.sg.subj “He will work.”

WANT can be said to be situated on a grammaticalization scale between full verb, modal verb, and auxiliary (cf. Heine 1993). It is found with divergent semantic and syntactic subcategorization frames and modal shifts. The status of WANT depends on argument structure on the one hand, and on the modal interpretation on the other. If there is an agentive-intentional subject (+animate), directly related to the argument structure of WANT, then the modality in a WANT-construction is volitional; if there is no such subject directly related to WANT, this results in a more or less impersonal (athematic) construction with deontic interpretation. But, as pointed out by Heine (1995: 30), the participants involved in modal expressions can be, at least, three: besides an agentive argument (usually the sentential subject36 ), there is the modal force “characterized by an ‘element of will’ ” (Heine . WANT as a matrix verb can itself be passivized. In this case, the external volitional argument is not a subject, but can be reactivated in the PP-adjunct: (i)

These children are wanted by God and loved by God.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.20 (1213-1269)

 Eva-Maria Remberger

1995: 29, where he uses an expression put forward by Jespersen, cf. Jespersen 1992: 320) and there is the speaker (or context of the speech situation). Modal force, in our case volitionality, can depend on any of the participants, as follows: the presence of a controlling agent as an external θ-role (and positional subject) directly related to WANT is decisive for a root volitional interpretation; a speakeroriented “element of will” might be the starting point of future, epistemic or even deontic modality; if the modal force lies in some non-human entity (such as the social context, normative behaviour, destiny, or superhuman forces), the original volitional modality easily shifts to a pure deontic interpretation: a. + agentive subject-θ-role: interested party = external argument => volitional interpretation AGENT-ORIENTED (cf. Bybee, Perkins & Pagliuca 1994)37 b. – agentive subject-θ-role: interested party = situative context => deontic interpretation SPEAKER-ORIENTED (cf. Bybee, Perkins & Pagliuca 1994) CONTEXT-ORIENTED In the passive constructions presented above, however, WANT has completely lost its status as a predicate. It has become a pure auxiliary, while a new predicate (encoded by the participle) is combined with its own external θ-role (even if demoted). Volitional modality, which is only present if WANT is used as a main verb,38 is completely lost here. Since deontic modality does not ask for an agentiveintentional subject, it is, however, still a constitutive feature of passive auxiliary WANT: the passive always lacks such an overt external subject-θ-role, at least in argument position.39

. Van Auwera & Plungian (1998) prefer the terms “participant-internal” and “participantexternal”; however, in the case of volitional modality, the ‘intentional agentivity’ of the interested party is a defining property of the modal meaning. . It is an open question whether WANT, like other verbal primitives, e.g. BE, GET, BECOME, HAVE etc., can be decomposed in purely functional elements, in line with Kayne (1993) and his modular hypothesis for HAVE as BE + an incorporated prepositional element (thanks to Cecilia Poletto, p.c. for pointing out this aspect of WANT to me). Harley (2003) offers such an analysis: For her, WANT + DP is no proof of the main verb status of WANT, since WANT always subcategorizes a functional prepositional element, and not directly a DP-complement. This results in the formula: WANT = MUST + P◦ ; therefore, the expression I want a beer, according to her proposal, is to be analysed as something like ‘A beer MUST BE P◦ + me’. However, I think, the volitional/agentive-intentional external θ-role still is essential. . A further development can be seen in the WANT-future, see (46), where this new external θ-role is active, so that WANT becomes a mere prospective marker without deontic modality.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.21 (1269-1305)

The deontic WANT-passive in Italo-Romance 

In all WANT-constructions, the modality of WANT and the semantic nature of the syntactic arguments involved are strongly interdependent. As a first approach to exploring this relationship, a semantic map of modality is proposed in what follows. . WANT and modality’s semantic map (cf. Van Auwera & Plungian 1998) Volitional modality is often subsumed under ‘desire’ (cf. also the overview in Verplaetse 2003). So, in Bybee, Perkins & Pagliuca (1994: 178) it is defined as “internal volitional conditions in the agent with respect to the predicate action”; they assume that “desire is future-projecting, but more internal” (cf. Bybee, Perkins & Pagliuca 1994: 185). However, WANT is also given as a possible lexical source for obligation (and later also possibly for future; cf. Bybee, Perkins & Pagliuca 1994: 181–186). Van Auwera & Plungian (1998) do not treat volitional modality as part of their necessity-possibility dichotomy. Yet, they do consider WANT as one possible “premodal meaning going to necessity” (Van Auwera & Plungian 1998: 111). The following figure (47) tries to combine the observations made by Bybee, Perkins & Pagliuca (1994), integrating these observations into a semantic map of modality as originally proposed by Van Auwera & Plungian (1998): (47) Van Auwera & Plungian (1998) + Bybee, Perkins & Pagliuca (1994) (modified for WANT) deontic necessity premodal meanings going towards necessity

speaker / context-oriented or participant external necessity

agent-oriented or participant internal necessity

postmodal meanings coming from necessity

epistemic necessity

My basic idea here is that volition can be seen as part of a modal path to necessity. To illustrate this, a chain of examples, this time from English, might suffice: A premodal meaning of want is found in its etymology (see (48)); the verb developed from meaning ‘to lack’ and then ‘to need’, to eventually, as late as the 18th century, mean ‘to desire’, an agent-oriented necessity (according to Bybee, Perkins & Pagliuca 1994: 178):

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.22 (1305-1376)

 Eva-Maria Remberger

(48) Engl. want < Mid. Engl. want ‘lacking, need’ < Old Norse vanta ‘lack’ (Bybee, Perkins & Pagliuca 1994: 178)

The internal ‘necessity to have something’ (= the need for something) is reinterpreted as agent-oriented necessity, i.e. volition. In the following example agentoriented necessity is expressed towards the event encoded in the infinitival clause (see (49)): (49) Stand. Engl. I want to wash the car.

From the agent-oriented necessity given in (49) there is a possible shift to deontic necessity, which might be interpreted as speaker- or context-oriented, if there is no syntactic argument in the expression which could be the interested party in the event expressed by the predication (see (50)): (50) Midland AE This car wants washed.

A possible further shift towards epistemic necessity can be found in (the not very common) example (51) (but see also the German epistemic use of WANT in fn. 9, (i)). Here, modal necessity is not oriented to the context or the speaker anymore. Deontic necessity is completely lost. Modality has shifted to an epistemic statement (the event encoded by the predication is described as being epistemically necessary): (51) Coll. Engl. “Do you have coolers?” “Coolers? They wanna be on one of the top shelves somewhere.” (Verplaetse 2003: 158)

A postmodal meaning stemming from a volitional verb can be observed in the English will-future (see (52)):40 (52) Stand. Engl. This car will be washed.

Future is classified by Van Auwera & Plungian (1998) as a typical postmodal meaning deriving from necessity. The most common example for this development (from deontic modality to future) is given by the Romance future stemming from HAVE + infinitive. . Here we find another volitional verb, i.e. will, which has lost its volitional modality to become a pure future marker (from Old English willan ‘to want’; cf. Heine & Kuteva 2002: 310– 313, also for several other languages). However, according to Palmer (2001: 78), English will still can be used “to express willingness”, see the following example: (i)

Why don’t you go and see if Martin will let you stay?

But usually, English will does not work as a volitional main verb anymore whereas German wollen and its Romance counterparts still do.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.23 (1376-1413)

The deontic WANT-passive in Italo-Romance 

The Romance WANT construction discussed previously in the main part of this paper fits well into this model: WANT originally expresses an agent-oriented necessity and has an intentional subject. However, in the context of a passive construction, non-intentional subjects are predicted. Nevertheless, the modal meaning of necessity is not lost, but shifts from the subject to a context-oriented, i.e. deontic, necessity. The Romance WANT construction is a perfect example for the interplay between syntactic structure (here, the passive), θ-role-compatibility (here, ±agentivity/intentionality) and modality (here, necessity). For a pan-Romance view on volitionality, its semantic structure and change further research wants (to be) carried out.

. Conclusion In this paper, I have discussed several constructions in Romance varieties which use the verb WANT as an auxiliary, together with some similar constructions in English and German. I mainly concentrated on passive constructions, i.e., constructions where the internal argument is promoted to the subject position whereas the external argument remains unexpressed. Since the demoted external argument is still implicitly present and since it can be reactivated in a prepositional adjunct structure, I assume it to be also structurally present. A covert caseless argument bearing the external θ-role can be represented by PRO which in its turn can be either arbitrary or controlled by the prepositional adjunct. As far as the auxiliary selection of WANT is concerned, I showed that volitional modality can be embedded in a more general semantic map of modality: the semantic path from volitional to deontic modality can be captured by the fact that the modality expressed by WANT can shift from agent-oriented to speaker-oriented and finally to purely context-oriented necessity. However, the results presented here leave several unanswered questions. An elaboration of the results achieved so far and further research on volitional modality as well as on WANT-constructions in the languages of the world are needed in order to provide further insights in the interaction of modal semantic features and syntactic mechanisms. The development of a formal model which coherently integrates both the syntactic and semantic features involved is the subject of an ongoing research project.

References Baker, Marc C., Kyle Johnson, & Ian Roberts (1989). Passive arguments raised. Linguistic Inquiry, 20, 219–251.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.24 (1413-1541)

 Eva-Maria Remberger

Blasco Ferrer, Eduardo (1986). La lingua sarda contemporanea. Grammatica del logudorese e del campidanese. Cagliari: Della Torre. Bowers, John (1993). The syntax of predication. Linguistic Inquiry, 24, 591–656. Bowers, John (2001). Predication. In Mark Baltin & Chris Collins (Eds.), The Handbook of Contemporary Syntactic Theory (pp. 299–333). Oxford: Blackwell. Bowers, John (2002). Transitivity. Linguistic Inquiry, 33, 183–224. Burzio, Luigi (1986). Italian Syntax. A Government and Binding Approach. Dordrecht: Reidel. Bybee, Joan, Revere Perkins, & William Pagliuca (1994). The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the World. Chigaco, IL: University of Chicago Press. Calabrese, Andrea (1992). The lack of infinitival clauses in Salentino: A synchronic analysis. In Christiane Laeuffer & Terrell A. Morgan (Eds.), Theoretical Analyses in Romance Linguistics (pp. 267–294). Amsterdam: John Benjamins. Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: The MIT Press. Chomsky, Noam (2000). Minimalist inquiries: The framework. In Roger Martin, David Michaels, & Juan Uriagereka (Eds.), Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik (pp. 89–151). Cambridge, MA: The MIT Press. Chomsky, Noam (2001). Derivation by phase. In Michael Kenstowicz (Ed.), Ken Hale: A Life in Language (pp. 1–52). Cambridge, MA: The MIT Press. Cocchi, Gloria (1995). La selezione dell’ausiliare. Padova: Uni Press. Cocchi, Gloria (1998). Ergativity in Romance languages. Studies on the Syntax of Central Romance Languages. In Olga Fullana & Francesc Roca (Eds.), Proceedings of the III Symposium on the Syntax of Central Romance Languages (pp. 83–99). Girona: Universitat de Girona. Frazer, Timothy C., Thomas E. Murray, & Beth Lee Simon (1996). Need + past participle in American English. American Speech, 71, 255–271. GGIC, II = Renzi, Lorenzo et al. (Eds.). (1991). Grande Grammatica Italiana di Consultazione. Vol. 2. Bologna: Il Mulino. Haase, Martin (1995). Finite Infinitive. Portugiesisch, Rumänisch und Süditalien. In Christian Schmitt & Wolfgang Schweickard (Eds.), Die Romanischen Sprachen im Vergleich (pp. 128– 149). Bonn: Romanistischer Verlag. Haider, Hubert (1986). Fehlende Argumente: Vom Passiv zu kohärenten Infinitiven. Linguistische Berichte, 101, 3–33. Harley, Heidi (2003). Wanting, having and getting: A note on Fodor and Lepore 1998. Linguistic Inquiry, 35(2), 255–267. Harley, Heidi, & Raffaella Folli (2004). Waltzing Matilda around and around: On the licensing of directed-motion resultatives. URL: http://dingo.sbs.arizona.edu/∼hharley/PDFs/ HarleyFolliMatilda2004.pdf Heine, Bernd (1993). Auxiliaries. Cognitive Forces and Grammaticalization. New York, NY: Oxford University Press. Heine, Bernd (1995). Agent-oriented vs. epistemic modality. Some observations on German modals. In Joan Bybee & Suzanne Fleischmann (Eds.), Modality in Grammar and Discourse (pp. 17–53). Amsterdam: John Benjamins. Heine, Bernd, & Tania Kuteva (2002). World Lexicon of Grammaticalization. Cambridge: Cambridge University Press. Jaeggli, Oswaldo (1986). Passive. Linguistic Inquiry, 17, 587–633. Jespersen, Otto (1992). The Philosophy of Grammar. Chicago, IL: University of Chicago Press [first edition 1924]. Jones, Michael Allan (1993). Sardinian Syntax. London: Routledge.

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.25 (1541-1652)

The deontic WANT-passive in Italo-Romance 

Kallulli, Dalina (2006). Passive as a feature-suppression operation. In Werner Abraham & Larisa Leisiö (Eds.), Passivization and Typology: Form and Function (pp. 442–460). Amsterdam: John Benjamins. Kayne, Richard S. (1993). Towards a modular theory of auxiliary selection. Studia Linguistica, 47, 3–31. Klaiman, M. H. (1991). Grammatical Voice. Cambridge: Cambridge University Press. Kratzer, Angelika (1994). The event argument and the semantics of voice. Ms., Amherst. URL: http://semanticsarchive.net/Archive/GU1NWM4Z/ Kratzer, Angelika (2005). Building resultatives. In Claudia Maienborn & Angelika Wöllstein (Eds.), Event Arguments: Foundations and Applications (pp. 177–212). Tübingen: Niemeyer. Lalande, Jean-Yves (1997). Verbstellung im Deutschen und im Französischen unter Anwendung eines CAD-basierten Expertensystems. Tübingen: Niemeyer. Ledgeway, Adam (2000). A Comparative Syntax of the Dialects of Southern Italy: A Minimalist Approach. Oxford: Blackwell. Maienborn, Claudia (2005). Das Zustandspassiv: Grammatische Einordnung – Bildungsbeschränkungen – Interpretationsspielraum. Ms., Humboldt-University Berlin. URL: http://amor.cms.hu-berlin.de/∼h0594bbb/pdf-files/2005-Zustandspassiv.pdf Murray, Thomas E., & Beth Lee Simon (1999). Want + past participle in American English. American Speech, 74, 140–164. Murray, Thomas E., & Beth Lee Simon (2002). At the intersection of regional and social dialects: The case of like + past participle in American English. American Speech, 77, 32–69. Palmer, Frank R. (2001). Mood and Modality (2nd edition). Cambridge: Cambridge University Press. Pollock, Jean-Yves (1989). Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry, 20, 365–424. Quirk, Randolph et al. (1997 [1985]). A Comprehensive Grammar of the English Language (14th edition). Essex: Longman. Reichenbach, Hans (1947). Elements of Symbolic Logic. New York, NY: MacMillan. Rapp, Irene (1996). Zustand? Passiv? – Überlegungen zum sogenannten ‘Zustandspassiv’. Zeitschrift für Sprachwissenschaft, 15(2), 231–265. Reinhart, Tanya, & Tal Siloni (2004). Against an unaccusative analysis of reflexives. In Artemis Alexiadou, Elena Anagnostopoulou, & Martin Everaert (Eds.), The Unaccusativity Puzzle. Explorations of the Syntax-Lexicon Interface (pp. 159–180). Oxford: Oxford University Press. Remberger, Eva-Maria (2002). Eigenschaften und Position von Hilfsverben. Eine minimalistische Analyse am Beispiel des Italienischen und Sardischen. Ph.D. dissertation. FU Berlin. Remberger, Eva-Maria (2005). WOLLEN im Romanischen. Eine minimalistische Analyse. In Georg Kaiser (Ed.), Romanische Syntax – generativ (pp. 145–159). Tübingen: Narr. Remberger, Eva-Maria (2006). Hilfsverben. Eine minimalistische Analyse am Beispiel des Italienischen und Sardischen. Tübingen: Niemeyer. Sa-Limba (1999–2005). Electronic Corpus of the Postings to the Mailing List ‘Sa-Limba’. Köln & Berlin. URL: http://www.lingrom.fu-berlin.de/sardu/sa-limba.html Salvioni, Carlo (1911). Note sintattiche II. Zeitschrift für Romanische Philologie, 35, 376–381. Sternefeld, Wolfgang (1995). Voice phrases and their specifiers. FAS Papers in Linguistics, 3, 48–85. URL: http://www2.sfs.nphil.uni-tuebingen.de/∼wolfgang/Downloads/FAS-2-TeXDrive.ps.gz

JB[v.20020404] Prn:6/11/2006; 9:38

F: LA9610.tex / p.26 (1652-1668)

 Eva-Maria Remberger

Tomić, Olga Mišeska (2003). The Balkan Sprachbund properties: An introduction to topics in Balkan syntax and semantics. Ms., Leiden: Center for Linguistics, University of Leiden. URL: http://wwwlot.let.uu.nl/GraduateProgram/LotSchools/Summerschool2003/ Tomic.pdf Van der Auwera, Johan, & Vladimir A. Plungian (1998). Modality’s semantic map. Linguistic Typology, 2, 79–124. Verplaetse, Heide (2003). What you and I want: A functional approach to verb complementation of modal WANT TO. In Roberta Facchinetti, Manfred Krug, & Frank Palmer (Eds.), Modality in Contemporary English (pp. 151–189). Berlin: Mouton de Gruyter.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.1 (47-110)

Agentivity and the virtual reflexive construction Nola M. Stephens The virtual reflexive construction, though it has received little attention in its own right, shares properties in common with a broad range of linguistic phenomena. This paper argues that virtual reflexives constitute an independent construction and addresses the relationship between virtual reflexives and other constructions, including lexical reflexives, patient-causer reflexives, and middles. The diverse thematic properties of each of these constructions are given as evidence for their uniqueness. For example, only virtual reflexives involve a distinct and volitional agent, while also allowing a partial transfer of agentivity to the patient-subject. These properties are reflected in the semantic nature of the verbs participating in the virtual reflexive alternation. The discussion presented here reveals that lexical semantic factors contribute meaningfully to an analysis of virtual reflexives.

.

Introduction

Patient-subject constructions figure prominently in the study of agentivity. To take the classic example, passives are generally regarded as constructions that demote the agent to an optional PP and realize the patient as the subject. Some alternations in argument structure that yield morphologically active patient-subject constructions are given in (1)–(4), where the small capital letters in (1b) indicate primary stress. (1) Virtual Reflexive Alternation a. Nichole solved the problem. b. This problem solves itself. (2) Lexical Reflexive Alternation a. Rania revealed her enthusiasm. b. Rania’s enthusiasm revealed itself. (3) Patient-Causer Reflexive Alternation a. Rachel turned the light on. b. The light turned itself on.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.2 (110-164)

 Nola M. Stephens

(4) Middle Alternation a. Branson tuned the cello. b. This cello tunes easily.

This paper addresses each of these patient-subject constructions, while focusing on the properties of sentences like (1b). Because the division outlined above has not been fully acknowledged in the literature, data of the type in (2b) and (3b) have been included in previous analyses of virtual reflexives (see Footnote 4). Moreover, virtual reflexives are often subsumed under the rubric of the middle construction and referenced as “reflexive middle constructions” or “middle reflexives” (e.g. Fiengo 1980; Williams 1981; Hale & Keyser 1987). I adopt the label virtual reflexive construction from Levin (1993). Although some discussion of the properties of virtual reflexives is available (see Lakoff 1977; Fiengo 1980; Hale & Keyser 1987; Fellbaum 1989; Levin 1993; Davidse 2002), these insights fail to present a coherent description of this construction and are insufficient for a thorough differentiation of virtual reflexives and the patient-subject constructions in (2)–(4). By expanding these analyses, this paper proposes to establish the uniqueness of virtual reflexives on the basis of semantic evidence. Section 2 compares virtual reflexives and reflexive constructions as in (2b) and (3b). Section 3 discusses the relationship between virtual reflexives and middles, and Sections 4–5 argue that lexical semantic factors influence the formation and differentiation of these constructions. Though I limit my analysis to English data, Section 6 points to the potential relevance of this study to cross-linguistic research. Ultimately, this investigation serves to generate a better understanding of the role of agentivity in virtual reflexives and to demonstrate that the primary semantic differences among the constructions in (1)–(4) are manifested in terms of thematic properties.

. Comparing reflexive patient-subject constructions The patient-subject constructions in (1)–(3) are alike in that each involves a reflexive marker. Though this list is likely non-exhaustive, comparing these constructions will offer general guidelines for distinguishing virtual reflexives from other reflexive patient-subject constructions. I argue in the ensuing paragraphs that constructions as in (1)–(3) exhibit the following semantic differences: those as in (2b) involve lexically reflexive verbs and represent non-causative/spontaneous events, while those as in (3b) represent events that are autonomously caused by the patient-subject; and virtual reflexives as in (1b) describe events that are caused by an external (and unnamed) agent. For convenience, I refer to constructions like (2b) as lexical reflexives and those like (3b) as patient-causer reflexives. On the

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.3 (164-222)

Agentivity and the virtual reflexive construction 

surface, the placement of stress on the reflexive and the acceptability of modifiers like practically are hallmarks that serve to differentiate virtual reflexives from other reflexive patient-subject constructions. Though some sentences seem ambiguous between a virtual reflexive and a lexical reflexive (e.g. the idea presented itself ) or patient-causer reflexive (e.g. the computer fixed itself ) reading, these properties of stress and modification should suffice to disambiguate the interpretations. As (5) demonstrates, virtual reflexives require the placement of primary stress on the reflexive pronoun (see Fiengo 1980; Hale & Keyser 1987; Fellbaum 1989). (5) a. These jokes write themselves. (A)1 b. *These jokes write themselves. (unmarked stress pattern)

Notice that the stress pattern in (5a) does not involve a contrastive reading as it does in standard reflexives such as (6). (6) a. Adam saw himself. (contrastive reading) b. Adam saw himself. (unmarked reading)

As Hale and Keyser (1987: 17) suggest, the stress on the reflexive pronoun in virtual reflexives indicates that the “literal reflexive reading” (e.g. that the jokes in (5a) actually compose themselves) is not intended. In contrast to this literal reflexive interpretation, virtual reflexives convey that the given activity (e.g. joke writing) is particularly easy for an unnamed agent to accomplish (see §3). Though further research is needed to explain why stress functions in this manner in virtual reflexives, this analysis is supported by the observation that (5b) receives a literal reflexive reading while the minimally different (5a) does not.2 Unlike virtual reflexives, lexical reflexives as in (7) prohibit the stress pattern in (5a). (7) a. An opportunity presented itself. b. *An opportunity presented itself.

. “(A)” indicates an attested example. These data were collected from various internet sites and checked by native speakers. Emphasis indicating primary stress was added. . Alternatively, Fiengo (1980) explains this stress pattern by arguing that virtual reflexives involve emphatic reflexive pronouns as in (i), which mean ‘without aid’ (cf. Stroik this volume). Virtual reflexives, however, are impossible when the reflexive is replaced by a true manner adverb with this meaning as shown in (ii) (see Fellbaum 1989: 126 for further critique of Fiengo’s position). (i)

Barry wrote the jokes himself/without aid/by himself/on his own.

(ii) *These jokes write without aid/by themselves/on their own.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.4 (222-282)

 Nola M. Stephens

Moreover, though patient-causer reflexives may in some cases allow stress on the reflexive in order to emphasize the autonomous nature of the event as in (8b), they are most natural with a neutral stress pattern. (8) a. The machine switched itself off. b. The machine switched itself off.

(Davidse 2002: 153)

Of these three constructions, only virtual reflexives characteristically place primary stress on the reflexive marker. Virtual reflexives (see (9)) also behave differently from lexical and patientcauser reflexives (see (10)–(11), respectively), in that only virtual reflexives categorically allow practically and similar modifiers (e.g. virtually, essentially, just about, pretty much, almost, more or less). (9) a. A recording like this recommends itself. b. A recording like this practically recommends itself. (A) (10) a. His impatience manifested itself. b. *His impatience practically manifested itself.

(Geniušien˙e 1987: 202)

(11) a. The engine caught itself on fire. b. *The engine virtually caught itself on fire.

Fellbaum (1989: 131) disregards virtual reflexives with practically, arguing that these adverbs increase acceptability such that they “obscure the conditions for reflexive middles formation.” Given the present discussion and the unacceptability of other examples with practically (see §§4–5), I disagree and maintain that these modifiers help expose the semantic properties of virtual reflexives. In particular, adverbs like practically reinforce the fact that virtual reflexives do not involve a literal reflexive reading. Taking (9b) as an example, the use of practically emphasizes that it is as if the recording recommends itself, not that it literally or even partially recommends itself. Note that the adverbs in virtual reflexives, thus, differ from measure adverbs, which are sometimes permitted in patient-causer reflexives (e.g. the computer practically (=partially) shut itself down). Because the stress pattern noted above and the use of modifiers like practically both have the function of signifying that the literal reflexive reading does not apply to virtual reflexives, the amount of stress on the reflexive may be reduced (at least for some speakers) when adverbs like practically are employed. In the absence of any such modifier, however, primary stress on the reflexive is obligatory. Importantly, virtual reflexives also differ from lexical and patient-causer reflexives in that only virtual reflexives involve an implied agent. As Geniušien˙e (1987) argues, lexical reflexives represent inchoative/non-causative events formed from lexically reflexive verbs, hence the label lexical reflexives. For example, suggest itself in (12b), forms a single lexical item and denotes a spontaneous emergence of

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.5 (282-334)

Agentivity and the virtual reflexive construction 

the idea, whereas suggest in (12a) means “bring (cause to come) [something] into the mind” (Geniušien˙e 1987: 201). (12) a. Layne suggested an idea. b. An idea suggested itself.

These sentences generally employ abstract nouns as their subjects and often convey a spontaneous coming into existence or appearance on a scene (see Levin 1993: 85). In each case, the reflexive serves not as an argument of the verb, but as a part of the verb’s lexical meaning that indicates an absence of causation (see Geniušien˙e 1987: 201). Note that the inability of these reflexives to tolerate primary stress (cf. (7b)) reinforces the claim that they do not constitute independent thematic arguments. In contrast to lexical reflexives, patient-causer reflexives denote causative events. As Davidse (2002) discusses, these constructions (“ergative reflexives” therein) take semantic potents as their subjects, and according to Geniušien˙e (1987: 206), they describe “autonomous processes brought about by the internal properties of the subject referent.” Unlike the non-argument reflexive in the lexical reflexive construction, the reflexive in the patient-causer reflexive is coreferential with the patient-subject, such that something inherent in the patient-subject is understood to have effected the change it undergoes. In Davidse’s (2002) terms, patient-causer reflexives express a ‘self-instigation’ of the event. Critically, though, in both lexical and patient-causer reflexives, no thematic role other than that of the patient-subject is represented (implicitly or overtly); the events in question are autonomous. By contrast, virtual reflexives are not semantically autonomous. They describe neither inchoative nor self-instigated events. Instead, virtual reflexives require an external cause for interpretation. For example, the construction in (13) does not mean that the focaccia emerged spontaneously or by any autonomous process. (13) Though you can’t skimp on the rising time, the mixing and shaping require so little work that the focaccia practically makes itself. (A)

If a focaccia is made, it must be the work of a distinct and volitional agent. In other words, the predicates in virtual reflexives require two distinct participants for interpretation, the patient-subject and an understood agent. Though there is, at least, a tacit agreement in the literature to the contrary, the agent in virtual reflexives is overtly expressible as evidenced by the constructions in (14) and (15). (14) a.

The commercial practically cut itself in the expert hands of Bill Dewald. (A)

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.6 (334-391)

 Nola M. Stephens

b. Coward is a playwright who writes brilliant dialogue that simply plays itself in the hands of competent actors. (A) (15) a.

For those who know what they’re doing, these movies practically direct themselves. b. Even for average students, problems like this practically solve themselves.

The objects of these in the hands of - and for-phrases are interpreted as agents, separate from the patient-subjects.3 For example, (14a) means that Bill Dewald cut the commercial, not that the commercial edited itself while he held it in his hands. Of course, agentive PPs are impossible in lexical and patient-causer reflexives as shown in (16) and (17), respectively. (16) a. *The idea suggested itself in the hands of the professor. b. *For those in attendance, the opportunity presented itself. (17) a. *The door opened itself in my uncle’s hands. b. *The class silenced itself for Ann. (where Ann = agent)

In essence, the critical semantic difference between virtual reflexives and the other two reflexive patient-subject constructions is that while virtual reflexives require the implication of a distinct agent, lexical and patient-causer reflexives involve only one participant, the patient-subject.4

. Comparing virtual reflexives and middles Middles as in (18a) and (19a) and virtual reflexives as in (18b) and (20a) differ clearly in that only virtual reflexives employ reflexive pronouns (cf. (19b)). Moreover, only virtual reflexives allow modifiers like practically (cf. (19c)), and only middles accept post-verbal modification with adverbs like easily (cf. (20b)). . Stroik (e.g., 1992, this volume) makes a similar argument for middles, maintaining that the object of a for-phrase as in (i) represents the agent. (i)

That book read quickly for Mary.

(Stroik 1992: 131)

As Stroik’s view is rather contentious, possible parallels between these PPs and those in virtual reflexives are not explored here. For arguments against projecting the agent of middles in the syntax see Fagan (1992), Ackema & Schoorlemmer (1994, 1995), Lekakou (2005), among others. . Fellbaum (1989) considers agentivity optional in virtual reflexives. For example, she analyzes “this door closes itself ” as a virtual reflexive (“reflexive middle” therein), claiming that it “precludes an interpretation with either an Agent or a Cause, stating instead that the Theme acted upon itself without external influence” (p. 129). This reading strongly suggests that this sentence, if acceptable, is a patient-causer reflexive (see Davidse 2002 on “ergative reflexives”).

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.7 (391-454)

Agentivity and the virtual reflexive construction 

(18) a. This problem solves easily. b. This problem practically solves itself. (A) (19) a. Carrot cake slices easily. b. *Carrot cake slices itself easily. c. *Carrot cake practically slices easily. (20) a. This trick just about performs itself. (A) b. *This trick just about performs itself easily.

Despite these distinguishing features, virtual reflexives are often regarded as a type of middle. This unified classification has been assumed because middles and virtual reflexives, beyond representing morphologically active patient-subject constructions, are semantically similar. Both constructions comment on the ‘doability’ of the activity (cf. Fellbaum 1989). For example, both of the constructions in (18) indicate that some property of the problem makes the solving activity easy.5 Fiengo (1980) and Williams (1981) propose two diverse strategies to account for virtual reflexives within the parameters of the middle construction (see Footnote 2 regarding Fiengo’s analysis). Williams (1981) argues that virtual reflexives are identical to middles that employ reflexive clitics as in the French middle below. (21) Ce livre se vend (facilement). this book ref sells (easily)

(cf. Williams 1981: 106)

According to Williams, middle formation produces a virtual reflexive only if the verb assigns case to the object position; otherwise, a standard middle results. Treating the reflexive pronoun in English as the functional equivalent of the reflexive clitic in French, however, does not explain why the English reflexive requires primary stress or why modifiers like easily are illicit in virtual reflexives. In French middles, the reflexive represents a reduced form and adverbs such as easily are freely allowed despite the presence of the reflexive morphology. The differences noted thus far between virtual reflexives and middles strongly suggest that virtual reflexives and middles are distinct constructions. As such, this paper adopts the view of Hale and Keyser (1987) and Fellbaum (1989) that virtual reflexives are structurally analogous to standard reflexives as in (22a), . Sections 4–5 focus on the semantic differences between virtual reflexives and middles. Note here, though, that a generic reading is often regarded as a hallmark of middles (see Keyser & Roeper 1984; Fagan 1992; Lekakou 2005, this volume; Stroik this volume, among many others). If this analysis is accurate, virtual reflexives and middles differ in this respect since virtual reflexives permit an eventive reading (cf. (i)). (i)

Then, a couple of years ago, I thought, ‘Why don’t I try this idea with the cast of American Empire’ – and then the screenplay practically wrote itself. (A)

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.8 (454-509)

 Nola M. Stephens

where the subject represents the agent and the reflexive pronoun represents the coreferential patient. (22) a. Davei hurt himselfi . b. This booki sold itselfi.

Critical to this approach is the understanding that, although the transitive syntax of the virtual reflexive suggests that the subject bears the agent role (i.e. the literal reflexive reading), this implication is not absolute. Instead, the virtual reflexive’s predicate requires two distinct participants for interpretation, the patient and an animate agent. For example, one would not take (22b) to mean that the book literally sold itself. As noted in Section 2, modifiers like practically and the unique stress pattern in virtual reflexives point to a non-literal reading. More explicitly, these features are useful in canceling the implication made by the syntactic structure that the subject and reflexive pronoun represent a coreferential agent and patient. The inanimate subject must, in fact, only represent the patient role (cf. Dowty’s 1991 discussion of proto-agents and patients). The distinct and animate agent required for interpretation is not overtly realized, but it is implicit in the meaning of the construction. Accordingly, Fellbaum (1989: 128) claims that the subject in virtual reflexives represents the agent only “in a metaphorical sense.” In particular, the transfer of agentivity to the patient as realized via the transitive reflexive syntax serves as a “metaphor” for the ease of the agentive activity (cf. Davidse 2002: 166). Though I leave the precise nature of the virtual reflexive’s “metaphorical” properties to future research, the follow paraphrase is appropriate for virtual reflexives: The event described by the verb was performed with such ease by the agent that it was as if no agentive involvement were necessary, as if the operation performed on the patient were brought about autonomously (cf. Lakoff 1977: 252). In essence, although agentivity has been transferred to the patient in virtual reflexives by means of the syntactic structure noted above, this transfer is only partial in the sense that the thematic role assignments of the virtual reflexive’s nonreflexive counterpart (e.g. John (Agent) sold the book (Patient)) are still present in the meaning of the construction. If this transfer were absolute as it is in standard reflexives like (22a), a distinct agent would be neither understood nor expressible in the form of a PP as in (14)–(15). Hence, the fact that the transfer of agentivity to the patient-subject is only partial in virtual reflexives licenses the presence of a distinct agent, be it implicit or overt. Taking the non-literal nature of virtual reflexives seriously may also contribute to an explanation of the restriction against post-verbal modifiers like easily (cf. (20b), (24b)). As shown in (23a), the semantic contribution made by practically (viz. the non-application of the proposition) is incompatible with the seman-

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.9 (509-576)

Agentivity and the virtual reflexive construction 

tics of easily even in standard transitive sentences (I thank Thomas Stroik for this example). (23) a. John (*practically) ate the whole pie easily. b. John actually/probably ate the whole pie easily.

Moreover, the grammaticality of (23b) reveals that this combination is not discarded simply because both adverbs end in -ly (see Nilsen 2003 regarding restrictions on adjacent -ly adverbs). Considering the awkwardness of (23a), it is not surprising that virtual reflexives with practically also prohibit middle-like modification. Nevertheless, the overt presence of practically cannot be the only source of the ungrammaticality of easily in virtual reflexives. As is evident in (24b), middle-like modifiers are illicit even in virtual reflexives without practically, though this is not true of standard transitives like (23a). (24) a. The tale (practically) spun itself. (A) b. *The tale (practically) spun itself easily/poorly/in a jiffy.

Unlike standard transitives, virtual reflexives necessarily involve the non-literal interpretation characteristic of adverbs such as practically. Presumably, then, it is this notion of the proposition’s non-application (i.e. the figurative nature of virtual reflexives) that is incompatible with the semantics of middle-like modification.6 To summarize, one difference between virtual reflexives and middles is that virtual reflexives are syntactically analogous to standard reflexives.7 Additionally, only virtual reflexives use a metaphorical device to convey the meaning that the . Alternatively, Fellbaum (1989: 125–126) implies that it is the transitivity of virtual reflexives that blocks the use of adverbs like easily. (23b) shows, however, that transitive sentences allow easily (as a manner adverb). . Nevertheless, virtual reflexives lack the syntactic flexibility of standard reflexives. Presumably the fact that agentivity is not fully transferred to the subject of virtual reflexives prohibits virtual reflexives from participating in the syntactic frames below. It seems that the (a) sentences must be interpreted such that the problem is the actual solver. (i)

a. *The problem solved itself to impress the math teacher. b. Nichole solved the problem to impress the math teacher.

(ii) a. ??The problem didn’t solve itself. (virtual reflexive reading) b. Nichole didn’t solve the problem. (iii) a. ??Did the problem solve itself? (virtual reflexive reading) b. Did Nichole solve the problem? As a reviewer noted, (ii)–(iii) reveal yet another difference between virtual reflexives and middles, since middles can be negated and questioned.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.10 (576-664)

 Nola M. Stephens

patient contributes to the action of the agent. The virtual reflexive equates the patient’s contribution with the dynamic carrying out of the activity, effectively exaggerating the patient’s conducive properties and the ease of the activity. By contrast, middles make a non-figurative statement regarding a property of the patient that affects the activity; the patient is not endowed with any degree of agentivity (cf. Fellbaum 1989: 128). Furthermore, the nature of the patient’s contribution in middles is not limited to that of ease/conduciveness as it is in virtual reflexives, but is dictated by the modifier (see (25)). (25) This pamphlet reads easily/quickly/poorly/in a jiffy/like Dr. Seuss/as well as Moby Dick.

The next sections explore a finer-grained differentiation of these constructions by addressing the lexical semantic properties of their verbs.

. A review of Fellbaum’s (1989) lexical semantic analysis Though there is general consensus in the literature concerning which semantic classes of verbs appear in middles, this is not the case for virtual reflexives. Contra Hale and Keyser (1987), Fellbaum (1989) claims that the set of verbs in virtual reflexives is not co-extensive with the set in middles. This observation is especially significant if the hypothesis explored by Hale and Keyser (1987), Levin (1993) and others is correct that verb meaning influences a verb’s syntactic behavior. In particular, isolating and comparing the different semantically coherent classes of verbs that participate in each construction should lend insight into the semantic properties that characterize and distinguish virtual reflexives and middles. Fellbaum’s lexical semantic assessment of virtual reflexives is summarized in Table 1. Fellbaum (1989) presents three lexical restrictions on the formation of virtual reflexives. One restriction is that the verb must select both an agent and a patient in the transitive use. This constraint bars unaccusative verbs that cannot assign an agent argument (Table 1, A) from forming virtual reflexives ((26) taken from Fellbaum 1989: 127).8 (26) a. *His mother/All that sugar rotted Johnnie’s teeth. b. Johnnie’s teeth rotted. c. *Johnnie’s teeth rotted themselves.

. Following other works on virtual reflexives (e.g. Lakoff 1977; Fellbaum 1989; Levin 1993), I use the (*) to indicate an ill-formed example. Nevertheless, as the present analysis is based primarily on the virtual reflexive’s semantic properties, these properties are taken to underlie the judgments reflected here.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.11 (664-690)

Agentivity and the virtual reflexive construction 

Table 1. Fellbaum’s (1989) assessment of the acceptability of verbs in virtual reflexives Example Unacceptable A. rot, mold, corrode Verbs B. frighten, scare, kill, bribe, photograph C. saw, slice, paint over, crush, pierce, shave

D. split, shatter, crumble, tear, rip, crush E. bleed, crumble, spill

Acceptable Verbs

F. digest, smoke G. compose, throw (clay pots), crochet H. sell, solve, suggest, (shift (bike gears), pay for)* I. open, shut**

Description/Classification (according to Fellbaum 1989) Unaccusative verbs that cannot assign an agent argument in transitive uses Psych verbs and other verbs that select an animate object in transitive uses Verbs of affect, which “imply a strong Agent involvement in the action as well as an instrument (either a tool or a hand) employed by the Agent” (p. 128) Verbs that “strongly require an Agent or Cause with a physical impact on the Theme” (p. 129) Verbs that “either select an Agent or a Cause in addition to the Theme” (p. 124) Verbs of ingestion Verbs of creation A small subset of the verbs that always assigns an agent in transitive uses A small subset of verbs that “can assign, in addition to the Theme, either an Agent or a Cause in the transitive constructions” (p. 124)

*Though shift and pay for were not specifically classified by Fellbaum, I assume they belong to this group. Even so, it may be a mistake to classify examples with pay for as virtual reflexives in that they seem to reflect a more literal meaning. **See Footnote 4.

(27) a. *This bread is perfect for our science project; it practically molds itself! b. *Daisies are great to have in the garden; they virtually sprout themselves!

As (27) illustrates, this class of verbs is impossible in virtual reflexives even under felicitous conditions, thus providing further evidence that virtual reflexives categorically require a distinct agent argument. Importantly, middles also prohibit verbs that cannot assign an agent argument. While the sentences in (28) are acceptable as generic unaccusatives, they do not have the middle reading. When used with an unaccusative verb that lacks a causative alternate, easily modifies the ‘probability’ that the event will occur, not the ‘doability’ of the activity for an agent (Fellbaum 1989: 125). (28) This bread molds easily/This bread easily molds.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.12 (690-751)

 Nola M. Stephens

Critically, non-alternating unaccusative verbs are excluded from participating in either the virtual reflexive or the middle construction. Another restriction given in Fellbaum (1989) is that verbs used with animate patients (Table 1, B) are unacceptable in virtual reflexives. This is because an animate patient may be misinterpreted as the actual agent, thus violating the semantic requirement for distinct agent and patient arguments ((29) taken from Fellbaum 1989: 128).9 (29) a. *Bureaucrats bribe themselves. b. Bureaucrats bribe easily.

As seen in (29b), the acceptability of middles is not directly affected by the subject’s animacy. In fact, middles also allow experiencer-subjects as in the sentence, Sam scares easily. Accordingly, the syntax of middles, unlike the transitive reflexive syntax of virtual reflexives, does not render the subject susceptible to an agentive interpretation. Fellbaum’s third restriction is that the formation of virtual reflexives is “constrained by the degree to which an Agent’s involvement is required for the action to be carried out” (1989: 128–129). Fellbaum claims that virtual reflexives, but not middles, disallow verbs that require a ‘strong’ agent. Purportedly, verbs considered acceptable in virtual reflexives have a ‘weaker’ agent than verbs deemed unacceptable (see Table 1). The notion of agentive strength seems to be motivated by whether or not the agent alters the physical integrity of the patient as in change of state and creation events. While Fellbaum’s (1989) first two constraints are well-supported by the data, the agentive strength requirement makes a false prediction about the range of verbs acceptable in virtual reflexives. For example, Fellbaum argues that creation verbs (Table 1, G) “by their nature have a very strong Agent (i.e. creator) requirement” (p. 130). One might, then, expect creation verbs to be worse in virtual reflexives than change of state (henceforth COS) verbs (Table 1, D). A preliminary internetbased search I conducted of the first 100 COS and creation verbs listed in Levin (1993), excepting 11 cross-listed verbs, casts serious doubt on this claim. This search found examples of virtual reflexives for approximately 57% of the creation verbs, but only about 7% of the COS verbs accessed (cf. (30)–(31)).

. Though this principle generally holds, exceptions may be possible to the extent that the subject can be clearly interpreted as the patient. For example, some speakers find (i) acceptable as a virtual reflexive. (i)

My kids were so easy to raise, they practically raised themselves!

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.13 (751-808)

Agentivity and the virtual reflexive construction 

(30) a.

I looked at and cast one of the rods that George built, an absolutely beautiful 3 pc. 7’3” for 5 wt. The rod practically cast itself. (A) b. This is my Batman Forever Batmobile model. This one practically built itself. I basically assembled it right from the box. (A) c. I had tons of great reference artwork for this sculpt, so it practically sculpted itself. (A)

(31) a.

One other quick note: this overmodified malt is really easy to mill! It practically crushes itself. (A) b. Actually, it seems that the lacing almost splits itself sometimes. When you find a spot to cut, the scissors do seem to follow a niche [. . .]. (A)

The events denoted by the creation verbs in (30) and the COS verbs in (31) all involve arguably ‘strong’ agents. Fellbaum’s agentive strength proposal cannot account for this type of data. The notion that virtual reflexives require a ‘weak’ agent is also conceptually problematic. If virtual reflexives emphasize the presence of a distinct agent, there is no reason to expect agentive involvement to be obligatorily restricted to a weak or minor role. Instead, a verb requiring ‘strong’ agentive involvement should felicitously underscore the presence of the unexpressed agent. Fellbaum (1989: 130) claims that, “It is difficult to interpret the created object as its own creator; therefore, the reflexive middle tends to be imcompatible [sic] with these verbs.” The point of the virtual reflexive, however, is not to claim that the patient-subject actually acted upon itself, but to comment on the ease of the activity for the agent (see §5). Just as virtual reflexives are generally incompatible with subjects susceptible to an agentive interpretation (e.g. animate subjects); they should be generally compatible with subjects unlikely to be interpreted agentively. Created entities are prime examples of the latter. Finally, Fellbaum’s three lexical semantic constraints are also insufficient to rule out other data as in (32). (32) a. *Silly Putty practically bounces itself. b. *This silk blanket virtually touches itself.

The examples above satisfy the conditions proposed by Fellbaum (1989) in that they involve (1) verbs capable of assigning an agent, (2) inanimate subjects, and (3) activities that do not require a ‘strong’ agent. . An alternative lexical semantic analysis Thus far, virtual reflexives have been shown to require a distinct agent, as supported by the following observations: virtual reflexives (1) categorically imply agentive involvement, (2) optionally express a distinct agent in a PP, (3) prohibit

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.14 (808-856)

 Nola M. Stephens

formation with non-alternating unaccusative verbs, and (4) generally ban patientsubjects vulnerable to an agentive interpretation. Nonetheless, as (32) illustrates, simply restricting virtual reflexives to verbs that allow an agent and an inanimate patient is insufficient to predict which verbs undergo virtual reflexive alternation. This section aims to provide a more accurate preliminary analysis of the lexical semantic properties that characterize virtual reflexives and distinguish them from middles. One lingering question is why agent-assigning COS verbs are less acceptable in virtual reflexives than in middles. Though examples of virtual reflexives with COS verbs have been attested (cf. (31)), these are relatively few in number, and speakers seem to generally disfavor virtual reflexives that denote COS events. Conversely, COS verbs typically form acceptable middles (see Fellbaum 1986; Hale & Keyser 1987; Levin 1993; Rapoport 1999). A well-established property of COS verbs is that they characteristically undergo causative alternation (see Smith 1970; Levin 1993; Levin & Rappaport Hovav 1995, among others). According to Levin and Rappaport Hovav (1995), causative alternation occurs when dyadic, causative verbs (i.e. verbs representing “externally” caused events) undergo a process of detransitivization that renders the cause argument unexpressed. This process is only possible “if the nature of the causing event is left completely unspecified” (p. 107). For example, the verb break does not require its subject to be an agent, instrument, or natural force but allows each of these possibilities as illustrated in (33). Because break does not lexically specify what type of cause argument it requires, it also undergoes causative alternation (cf. (33a–c) vs. (d)). (33) a. b. c. d.

Curtis broke the glass. (agent) The hammer broke the glass. (instrument) The hurricane broke the glass. (natural force) The glass broke. (fully unspecified cause)

As a point of comparison, note that creation verbs like write fully specify their cause as an agent and resist causative alternation (cf. (34a) vs. (d)). (34) a. b. c. d.

Carl Sandburg wrote the poem. (agent) *The pen wrote the poem. (instrument) *The summer breeze wrote the poem. (natural force) *The poem wrote. (fully unspecified cause)

It follows from Levin and Rappaport Hovav’s generalization that detransitivizing verbs “are those in which the eventuality can come about spontaneously without the volitional intervention of an agent” (p. 102). This is precisely the reading that virtual reflexives obligatorily avoid. Thus, in order to preserve their distinct-agent

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.15 (856-915)

Agentivity and the virtual reflexive construction 

interpretation, virtual reflexives tend to prohibit verbs that allow a non-specified cause reading. Given this observation, it may seem surprising that middles are often formed with COS verbs. After all, middles also imply agentive involvement (see §4). Nevertheless, unlike the unaccusatives with easily in (28), repeated here as (35), constructions like this glass breaks easily are ambiguous between agentive (see (36a)) and nonagentive (see (36b)) readings. (35) This bread molds easily/This bread easily molds. (36) a.

This glass breaks easily. (when someone breaks it, it breaks easily; see Stroik this volume) b. This glass breaks easily/This glass easily breaks. (a non-agentive entity/circumstance causes it to break, cf. (33b–c))

(37) This glass breaks with difficulty/quickly/in a jiffy.

The agentive reading in (36a) is relevant to middles, whereas the nonagentive reading in (36b) belongs to the class of generic unaccusatives (see Smith 1970; Hale & Keyser 1987; Fellbaum 1989; Davidse 2002; Lekakou 2005, among others). Furthermore, even this ambiguity disappears with the use of adverbs as in (37). In (37), only the middle reading (i.e. the agentive reading) applies. Though the source of the middle’s agentive interpretation has been a matter of controversy, Hoekstra & Roberts (1993), Lekakou (this volume), and others argue that the middle adverb is consequential in enabling this interpretation. Importantly, the lack of causal specification of COS verbs does not jeopardize the middle’s agentive reading. Securing the distinct-agent reading of COS verbs is more complex in virtual reflexives. Middle-like modification is illicit, and the transitive syntax makes virtual reflexives particularly vulnerable to interpretation as standard agent-subject reflexives. Though prosodic cues and modification with practically are helpful in suggesting the agentive reading (see §2), these features are relatively subtle and only indirectly indicate agentive involvement by highlighting the non-literal reading of the construction. Nevertheless, virtual reflexives sometimes permit COS verbs as in (31) and (38). In each of these examples, however, not only is the implication of an agent transparent, but this agent is also clearly volitional. As (39a) illustrates, virtual reflexives are generally unacceptable unless the agent is interpreted as volitional. By contrast, a volitional interpretation, though possible, is not obligatory for middles even though an unexpressed agent is implied under the middle reading (cf. (39b)). (38) A good chain breaker is a joy to have. This one’s okay, but a little out of alignment. But it [sic] you really dremel off the ends of the pins, the chain almost breaks itself. (A)

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.16 (915-967)

 Nola M. Stephens

(39) a. *My favorite glass ornaments practically break themselves. b. My favorite glass ornaments break easily.

The fact that virtual reflexives strongly prefer a volitional agent follows from the importance of securing an agentive interpretation, despite misleading syntactic properties (see, for example, Barker 1998 for discussion of the cross-linguistic relevance of volition).10 Similar to COS verbs, verbs of affect that imply the use of an instrument (Table 1, C) are also more acceptable in middles than in virtual reflexives (see Rapoport 1999 for discussion of these verbs in middles).11 Unlike COS verbs, however, verbs with an instrument component avoid causative alternation as shown by (40d) because they partially specify their cause argument. These verbs specify either an agent or an instrument as their cause argument but do not allow, for example, natural force causes (Levin & Rappaport Hovav 1995; (40) taken from pp. 102–103). (40) a. b. c. d.

The baker cut the bread. That knife cut the bread. *The lightning cut the clothesline. *The bread cut.

Because these verbs, classified by Levin (1993) as verbs of cutting, have more fully specified cause arguments than COS verbs, one might expect them to be more acceptable than COS verbs in virtual reflexives. Though further research is needed to thoroughly substantiate this claim, speakers do seem to more readily accept these constructions. In fact, these are the types of verbs Hale and Keyser (1987) deem suitable in virtual reflexives ((41) taken from Hale & Keyser 1987: 17).

. One reviewer noted that (39a) may be possible in certain contexts: (i)

My favorite ornaments break easily. They are so incredibly fragile that they practically break themselves.

Native speakers I consulted considered (i) acceptable only in the context that the party responsible for the breakage articulates (i) as an excuse. In any case, (i) requires the construal of an animate agent. The absence of such examples in my corpus search and necessity of extra context suggest that these sentences are relatively anomalous, and I leave a more detailed analysis of these examples to future investigation. . Rapoport (1999), in fact, argues that not all middles are agentive, but that the presence of an instrument or manner component in the meaning of some verbs yields the “agentivity effect.” I ascribe to the more conventional view that the agentive interpretation is categorically present in middles and that the nonagentive interpretation of sentences like (39b) belongs to generic unaccusatives.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.17 (967-1040)

Agentivity and the virtual reflexive construction 

(41) a. The toughest carrots virtually slice themselves with this handy tool. b. The hardest granite practically crushes itself with this new hammer.

Furthermore, even though the COS verbs in (31) and (38) do not inherently entail the use of an instrument (cf. Levin 1993: 156–157), context reveals that these events also involve an instrument. Importantly, while unaccusatives (cf. (35)) can be understood as denoting agentless events, predicates that include an instrument component imply the presence of an agent (see Hale & Keyser 1987; Levin & Rappaport Hovav 1995; Reinhart 2000, among others).12 Taking into account the amount of causal specificity (i.e. whether the verb leaves the cause argument fully unspecified or specifies the cause as an agent, etc.) also provides an explanation as to why some transitive motion verbs are more acceptable than others in virtual reflexives. For example, bounce and float (see Levin 1993: 264) are infelicitous in virtual reflexives, though they allow middle formation. (42) a. b. c. d.

*This ball practically bounces itself.13 This ball bounces easily. Cole bounced the ball. This ball bounced.

(43) a. b. c. d.

*This toy boat practically floats itself. This toy boat floats easily. Kyle floated the toy boat. The toy boat floated.

As shown by the (c) and (d) examples above, these verbs of motion typically permit causative alternation and must, therefore, be capable of leaving their cause argument completely unspecified. . Explanations for this well-acknowledged correlation vary. For example, Reinhart (2000: 7) claims that an instrument role is licensed “only if an agent role is present (overtly or implicitly).” Conversely, Hale and Keyser (1987: 17) argue that “the associated agency is inherent in the semantics of the instrumental construction itself.” In fact, Hale and Keyser give virtual reflexives with instrumental phrases as evidence for their position. Note, though, that virtual reflexives do not substantiate this view because they have an agentive reading regardless of the presence of an instrumental expression. . A reviewer noted that this judgment improves in the following context: (i)

This ball has been especially engineered to be super rubbery and bouncy. Heck, it is so well-engineered that it practically bounces itself.

Notice that the claim that motion verbs with a vehicle are better in virtual reflexives than those without is not invalidated by (i). Unlike (42a), the virtual reflexives in (44)–(46) are acceptable without additional context.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.18 (1040-1091)

 Nola M. Stephens

By contrast, verbs of motion using a vehicle (see Levin 1993: 267–268), typically resist causative alternation and appear to be more acceptable in virtual reflexives as illustrated below. (44) a. b. c. d.

In calm conditions paragliders practically fly themselves. (A) These paragliders fly well. Lynn flew the paraglider. ??The paraglider flew. (where paraglider = vehicle)

(45) a. b. c. d.

The bike practically pedals itself. (A) This bike pedals easily. Kaley pedaled the bike. *The bike pedaled.

(46) a. b. c. d.

This car practically drives itself. (A) This car drives easily. Lauren drove the car. ??The car drove.

Thus, verbs involving an instrument/vehicle more thoroughly specify their cause argument and are better in virtual reflexives than the verbs that leave their cause argument fully unspecified. The acceptability of middles with both types of motion verbs supports the conclusion that middles are not sensitive to the predicate’s amount of causal specification, especially in light of the fact that middles also allow verbs like sell that specify an agent cause argument. Given the analysis thus far, any verb that potentially selects an inanimate patient and specifies the other argument as a volitional agent should be acceptable in both virtual reflexives and middles (see below for discussion of unacceptable creation verbs in middles). Nevertheless, verbs of contact such as touch, caress, nudge, and stroke (see Levin 1993: 155) prove this restriction to be too general. In fact, these verbs resist both middle and virtual reflexive formation.14 (47) a. *That cashmere sweater touches easily. b. *That cashmere sweater practically touches itself. c. *That cashmere sweater touched.

The primary semantic conditions established thus far concern the nature of the agent’s role in the construction. Remember, however, that middles and virtual re. One temptation is to assume that these verbs are excluded because they do not describe causative events (cf. Tenny 2000: 301). This, however, seems to make the wrong prediction for verbs like read, which are acceptable in both constructions (cf. (i)–(ii)). (i)

The book is so well written it practically reads itself. (A)

(ii) This novel reads easily.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.19 (1091-1148)

Agentivity and the virtual reflexive construction 

flexives also indicate that some property of the patient contributes to the action of the agent. In (48a–b), something about the book is understood as contributing to the ease of the selling activity. (48) a. This book sells easily. b. This book practically sells itself. (A)

The examples in (47a–b) violate a semantic condition related to the role of the patient in virtual reflexives and middles. Specifically, the patient-subject must be capable of making a contribution to the relevant property (e.g. conduciveness) of the event denoted by the verb. For example, some books are easier to sell than others. On the other hand, though a certain artifact could be more or less tactilely pleasing, there is nothing inherent about a particular cashmere sweater that would make the touching activity especially easy (cf. Lakoff 1977: 252). The nature of the patient’s contribution is also relevant to explaining why creation verbs are more acceptable in virtual reflexives than in middles (cf. (49)– (50)).15 (49) a.

Our membership directory practically produces itself, thanks to Pete Campbell’s leadership and a terrific administrative team. (A) b. Once I saw these incredible paintings [. . .], and understood a little of the rich history of the Armenians, the stamp practically designed itself. (A)

(50) a. ??Our membership directory produces easily. b. ??The stamp designs easily.

Beyond specifying a volitional agent cause, creation verbs generally select an inanimate patient. Moreover, properties of this patient can conceivably contribute to the event. In (49b), the subject matter of the stamp is credited with having facilitated the design activity. Nevertheless, the link between the ‘doability’ of an activity and the (non-)conducive nature of the patient is indirect with creation events because the patient does not fully exist until the event is complete. Furthermore, the contexts given in (49) suggest that the conducive nature of the patient is not the primary focus of virtual reflexives. In (49a), the production activity is made easy chiefly by the efforts of competent collaborators. Even so, the patients seem to be most responsible for the ease of activity in the virtual reflexives in (51). . One reviewer correctly observed that the middles in (50) improve with a generic subject. Fellbaum (1986: 17), on the other hand, claims that middles categorically exclude creation verbs. While Fellbaum’s view is certainly too strong, her assessment does reflect a general tendency to disfavor creation verbs in middles (esp. in comparison to COS verbs). Offering insight into this tendency is the aim of the present argument.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.20 (1148-1245)

 Nola M. Stephens

(51) a. The route is so photogenic the picture takes itself. (A) b. Irene’s exquisite portfolio, a lifetime of work by a remarkable artist, practically organized itself. (A)

In principle, virtual reflexives do not obligatorily specify which factors are most responsible for the conducive nature of the activity.16 The main focus of virtual reflexives is the ease of the agentive activity, not the source thereof. This proposal is consonant with Lakoff ’s (1977) suggestion that virtual reflexives are agentfocused constructions. That is, they focus on how easy the activity is for the agent to perform. Conversely, middles focus primarily on the property of the patient-subject that contributes to the activity, not on the ease of the agentive activity or any property associated with the understood agent. Accordingly, Lakoff (1977) characterizes middles as having patient-focused properties (cf. (52a–b)). (52) a.

I’ve had ample experience selling Girl Scout cookies, so these thin mints practically sell themselves. ?? b. I’ve had ample experience selling Girl Scout cookies, so these thin mints sell easily.

The sentence in (52b) is strange because the experiences of an agent seem irrelevant to the statement made by the middle. In that middles are primarily concerned with the patient’s contribution, they disfavor contexts that distract from this contribution. Essentially, creation verbs are less acceptable in middles because the patient’s contribution is not sufficiently transparent to satisfy the middle’s patientfocused properties. In sum, this section has revealed that virtual reflexives and middles underline different aspects of their shared meaning. Virtual reflexives focus on the ease of the activity for the agent, while middles focus on the properties of the patient that influence the activity. Accordingly, creation verbs, which highlight the presence of a volitional agent and underemphasize the contribution of the patient, are more acceptable in virtual reflexives than in middles. The opposite pattern holds for COS verbs, which allow for a variety of cause arguments and often involve patients that are inherently more inclined to undergo the activity in question. Even so, both middles and virtual reflexives inevitably involve some contribution on the part of both participants as seen by the unacceptability of unaccusative verbs without agent arguments and verbs of contact without contributing patient arguments.

. This does not necessarily contradict Lakoff ’s (1977) claim that a characteristic feature of patient-subject constructions is that the patient bears primary responsibility (see also van Oosten 1977; but cf. Lekakou 2005). In a figurative sense, the subject in virtual reflexives is clearly

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.21 (1245-1268)

Agentivity and the virtual reflexive construction 

Table 2. General trends in the acceptability of verbs in virtual reflexives and middles Types of transitive verbs

Examples (see Levin 1993 for classifications)

Virtual Reflexives

Middles

Verbs that do not select an agent

Certain unaccusative verbs (mold, rot, corrode)

unacceptable

unacceptable

Verbs that leave the cause argument unspecified

Verbs of change of state/break verbs (break, crack, shatter), Verbs of motion/roll verbs (bounce, float, roll)

unacceptablemarginal

acceptable

Verbs that select an agent or an instrument

Verbs of cutting (saw, cut, slice), marginal Verbs of motion using a vehicle (fly, drive, row)

acceptable

Verbs that select an agent and allow for patient contribution

sell, solve, read

acceptable

acceptable

Verbs that select an agent but allow only marginal patient contribution

Verbs of creation (build, prepare, create)

acceptable

marginalunacceptable

Verbs that select an agent but allow no patient contribution

Verbs of contact/touch verbs (touch, nudge, stroke)

unacceptable

unacceptable

A final prediction is that transitive verbs that specify a volitional agent as well as an inanimate and contributing patient should form both constructions. Examples with sell and solve meet these criteria and are, indeed, some of the clearest cases of verbs that allow both alternations. The primary lexical semantic properties of the verbs allowing virtual reflexive and middle formation are summarized in Table 2 (see Appendix for additional examples of verbs attested in virtual reflexives).

. Conclusion In this paper, thematic properties have served to differentiate virtual reflexives from other patient-subject constructions. While agentivity is absent from lexical reflexives and patient-causer reflexives, virtual reflexives semantically require and optionally express a distinct agent. Furthermore, virtual reflexives, like middles, signify that a property of the patient contributes to the agent’s action. Unlike middles, however, virtual reflexives obtain this interpretation via a “metaphorical” endowed with this property. My argument here concerns the entity that is actually understood as being most responsible for the activity’s ease.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.22 (1268-1321)

 Nola M. Stephens

transfer of agentivity to the patient, whereby the nature of the “metaphor” limits the patient’s contribution to one of heightened conduciveness. Accordingly, virtual reflexives accentuate the ease of the agentive activity and are most acceptable with verbs that underscore the role of the agent. Middles, which focus more on the patient, tolerate a variety of patient contributions via adverbial modification, which may further serve to supply an agentive interpretation to causally vague or ambiguous predicates. It follows that middles are most acceptable with verbs that support a patient emphasis, while being capable of assigning an agent. Taken together, these arguments clarify the acceptability of certain semantic verb classes in the given alternations and lead to a more precise description of each of the constructions discussed. Importantly, virtual reflexives are shown to represent a truly distinct type of patient-subject construction in that they not only bear an agentive interpretation, but also focus primarily on the participation of an obligatory and volitional agent. The importance of these properties in differentiating among morphologically active patient-subject constructions is consistent with the position that lexical semantic properties of verbs play an important role in influencing the feasibility of alternations in argument expression. Finally, though a cross-linguistic analysis of virtual reflexives is beyond the scope of this paper, the virtual reflexive construction is certainly not unique to English. In fact, the (a) sentences in (53)–(56) may well represent this phenomenon in German, Norwegian, Hungarian, and Hebrew, respectively. Though Norwegian evidently lacks a middle construction, the middle counterparts of the German, Hungarian and Hebrew sentences are given in the (b) examples below (thanks to Helge Lødrup for (54), Tibor Laczkó for (55), and Itamar Francez for (56)). (53) a.

Das Problem löst sich (praktisch) von selbst. the problem solves ref (practically) by itself ‘The problem (practically) solves itself.’ b. Das Problem löst sich leicht. the problem solves ref easily ‘The problem solves easily.’

(54) a.

Klærne vasker seg (praktisk talt) selv. clothes-def wash ref (practically spoken) self ‘These clothes practically wash themselves.’

(55) a.

Ez-ek a történet-ek (gyakorlati-lag) mag-uk-at this-pl the story-pl.nom (practical-adv) self-3pl-acc ír-ják. write-pres.3.pl ‘These stories (practically) write themselves.’

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.23 (1321-1376)

Agentivity and the virtual reflexive construction 

b. Ez-ek a történet-ek könny-en/mag-uk-tól this-pl the story-pl.nom easy-adv/self-3pl-from ír-ód-nak. write-mid-pres.3.pl ‘These stories write easily.’ (56) a.

Im mevaSlim otan, ha-agvaniyot (ha-ele) mitkalfot if cook-pl them, the-tomatoes the-these peel-pl.fem me-acman. from-self-fem.pl ‘If you cook them, these tomatoes peel by themselves.’ (Implication: they become easy to peel) b. Ha-agvaniyot mitkalfot be-kalut. the-tomatoes peel.pl.fem in-easiness ‘These tomatoes peel easily.’

As these examples suggest, a careful comparison of virtual reflexives and other patient-subject constructions would likely yield important insights into the study of variation in argument realization in a range of languages. I leave this to future research.

Acknowledgements I am deeply grateful to Natsuko Tsujimura for her patient and insightful guidance, her challenging questions, and much more. I also thank two anonymous reviewers, Benjamin Lyngfelt, Torgrim Solstad, Barbara Vance, Beth Levin, Andrew KoontzGarboden, Daniel Dinnsen, and participants of the 2004 Oslo Workshop and the 2005 meeting of the LSA. I appreciate the financial support of the following Indiana University affiliates: The Wells Scholars Program, the Office of Research and the University Graduate School, the Department of Linguistics, the Department of East Asian Languages and Cultures, and the Honors College.

References Ackema, Peter, & Maaike Schoorlemmer (1994). The middle construction and the syntaxsemantics interface. Lingua, 93, 59–90. Ackema, Peter, & Maaike Schoorlemmer (1995). Middles and non-movement. Linguistic Inquiry, 26(2), 173–197. Barker, Chris (1998). Episodic -ee in English: A thematic role constraint on new word formation. Language, 74(4), 695–727.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.24 (1376-1480)

 Nola M. Stephens

Davidse, Kristin (2002). Nominative and oblique in English: Reflexive clauses as a test case for distinct agent-patient models. In Kristin Davidse & Beatrice Lamiroy (Eds.), The Nominative and Accusative and Their Counterparts (pp. 131–173). Amsterdam: John Benjamins. Dowty, David (1991). Thematic proto-roles and argument selection. Language, 67(3), 547–619. Fagan, Sarah (1992). The Syntax and Semantics of Middle Constructions: A Study with Special Reference to German. Cambridge: Cambridge University Press. Fellbaum, Christiane (1986). On the Middle Construction in English. Bloomington, IN: Indiana University Linguistics Club. Fellbaum, Christiane (1989). On the ‘reflexive middle’ in English. Papers from the Regional Meetings, Chicago Linguistic Society, 25(1), 123–132. Fiengo, Robert (1980). Surface Structure: The Interface of Autonomous Components. Cambridge, MA: Harvard University Press. Geniušien˙e, Emma (1987). The Typology of Reflexives. Berlin: Mouton de Gruyter. Hale, Ken, & Jay Keyser (1987). A view from the middle. Lexicon Project Working Papers, 10, 1–36. Hoekstra, Teun, & Ian Roberts (1993). Middle constructions in Dutch and English. In Eric Reuland & Werner Abraham (Eds.), Knowledge and Language II. Lexical and Conceptual Strecture (pp. 183–220). Dordrecht: Kluwer. Keyser, Samuel Jay, & Thomas Roeper (1984). On the middle and ergative constructions in English. Linguistic Inquiry, 15, 381–416. Lakoff, George (1977). Linguistic gestalts. Papers from the Regional Meetings, Chicago Linguistic Society, 13, 236–287. Lekakou, Marika (2005). In the Middle, Somewhat Elevated. The Semantics of Middles and its Crosslinguistic Realization. Ph.D. Dissertation, University College London. Levin, Beth (1993). English Verb Classes and Alternations: A Preliminary Investigation. Chicago, IL: University of Chicago Press. Levin, Beth, & Malka Rappaport Hovav (1995). Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge, MA: The MIT Press. Nilsen, Øystein (2003). Eliminating Positions. Ph.D. Dissertation, University of Utrecht. Rapoport, Tova (1999). The English middle and agentivity. Linguistic Inquiry, 30(1), 147–155. Reinhart, Tanya (2000). The theta system: Syntactic realizations of verbal concepts. UiL OTS Working Papers in Linguistics. Utrecht: Utrecht Institute of Linguistics. Smith, Carlota (1970). Jespersen’s ‘move and change’ class and causative verbs in English. In Mohammad Ali Jazayery, Edgar C. Polomé, & Werner Winter (Eds.), Linguistic and Literary Studies in Honor of Archibald A. Hill, Vol. II (pp. 101–109). The Hague: Mouton. Stroik, Thomas (1992). Middles and movement. Linguistic Inquiry, 23(1), 127–137. Tenny, Carol (2000). Core events and adverbial modification. In Carol Tenny & James Pustejovsky (Eds.), Events as Grammatical Objects: The Converging Perspectives of Lexical Semantics and Syntax (pp. 285–329). Stanford, CA: CSLI Publications. van Oosten, Jeanne (1977). Subjects and agenthood in English. Papers from the Regional Meetings, Chicago Linguistic Society, 13, 459–471. Williams, Edwin (1981). Argument structure and morphology. The Linguistic Review, 1, 81–114.

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.25 (1480-1599)

Agentivity and the virtual reflexive construction 

Appendix: Some additional attested examples of virtual reflexives Verb

Example (cf. Footnote 1)

assemble balance

This stand practically assembles itself! In classical styles, you can mic everything acoustically and it pretty much balances itself [. . .]. [. . .] hunting packs so advanced, they almost carry themselves. And it does a fantastic job. Look, the bag almost changes itself in no time. John Adams’s Fearful Symmetries is such a perfect score for dance that it virtually choreographs itself. You can get binary versions for windows and macintosh but have to compile it yourself on unix (no big deal – it practically compiles itself). The composer must immerse himself in the script completely [. . .]. Once this is accomplished the music practically composes itself. [. . .] in compiling the figures, this table virtually constructed itself. Actually, the recipe practically cooks itself, and the result is so unique and exquisite in taste [. . .] [. . .] artwork practically creates itself! Legally, the case almost defended itself [. . .]. Prize fighters were his favorite subjects because, “they are completely basic. Boxing action practically draws itself.” Through the efforts of literally thousands of persons over the past quarter of a century the Program has practically fashioned itself. [. . .] folding by this method makes the initial collapse of the rose very simple. It practically folds itself when you use good, stiff paper. I will now try to forge a link between modern physics’ conception of the world and the theory of the energon. This link forges itself. [. . .] this dichotomy is so glaring it virtually invalidates itself O2 is a mega-wide jersey ribbon. A merino wool and acrylic blend, it practically knits itself at 2 1/4 sts per inch on a #17 needle. My general recommendation is to shoot against fluorescent green if at all possible. It practically lights itself [. . .]. L-Glutamine doesn’t really have a taste and if you get something like AST’s micronized form the stuff practically mixes itself [...] A great sculpt of the big guy [Godzilla]. The kit is easy to build and practically paints itself (with a little help from Stan ‘The Man’ Hyde and MechaJohn Tucky.) A recording like this practically recommends itself – a feast for the ear and the soul The mix rolls very well indeed, in fact it almost rolls itself. Although the look is lavish, the table practically sets itself [. . .]. Showing the game in LA this year was quite easy as it is almost done and really shows itself. Optimism, innocence and exuberance: Janie Dee and Tim Flavin in My One and Only – ‘a Gershwin song almost sings itself ’

carry change choreograph compile compose construct cook create defend draw fashion fold forge invalidate knit light mix paint

recommend roll set show sing

JB[v.20020404] Prn:19/10/2006; 13:40

F: LA9611.tex / p.26 (1599-1599)

 Nola M. Stephens

Appendix (continued) Verb

Example (cf. Footnote 1)

spin

One fleece also went to Yolo Mill (see Links) who did a wonderful job on it – ‘Practically spins itself ’ says Bev. There is such a wonderful rhythm to the placement of the stitches that it almost stitches itself! Your hair can be cut and shaped to quickly style with a dryer, cut-set or combed in place for a youthful and stylish look that almost styles itself. The Hospitality Committee has secured a new coffeemaker system which uses less coffee and practically washes itself. Cotton is also easily woven, and a squared pattern is technically an easy pattern to produce: it almost weaves itself.

stitch style wash weave

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.1 (46-175)

Arguments in middles* Thomas Stroik Two of the most pressing questions about middle constructions are (i) how do the logical objects in middles become grammatical subjects and (ii) what happens to the logical subjects (external arguments) in middles. In this paper, I argue that middle formation involves the syntactic promotion of the logical object to subject position and the demotion of the external argument to an adjunct position. My arguments center on the for-PP phrase in English middles and on the mandatory presence of reflexives in German middles.

.

Introduction

Middle constructions such as those in (1) have stirred up significant controversy over the last twenty years in large measure because there is so little consensus about the grammatical and semantic properties of these constructions. (1) a. Bureaucrats bribe easily. b. Dit boek leest als een trein. this book reads as a train ‘This book reads like a train’ (Dutch, Ackema & Schoorlemmer 2003: 1)

Although linguists generally assume, in line with Ackema and Schoorlemmer’s (2003: 1) observations, that (i) “the external argument of the non-middle counterpart of the middle verb cannot be expressed as a regular DP-argument in the middle”; (ii) “in a middle, the logical subject argument of the underlying is semantically present” and (iii) “if the non-middle counterpart of the middle verb * I would like to thank Torgrim Solstad, Benjamin Lyngfelt, and Maria Krave, who organized the Oslo Conference on Demoting the Agent. Their efforts have done much to influence the shape of my paper. I would also like to thank the many participants at the Oslo Conference who commented on my paper, especially Marijana Marelj, Marika Lekakou, Nola Stephens, Joan Maling, and Elly van Gelderen. And finally, I would like to thank the anonymous reviewers of my manuscript for their many insightful comments.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.2 (175-183)

 Thomas Stroik

has a direct internal argument role, the subject of the middle sentence carries this role,” they don’t generally agree on how best to explain the suppression of the external argument in middles, or how to interpret the syntactically suppressed but semantically present external argument, or how to account for the subjecthood of the internal argument in middles. Some linguists – notably Fagan (1992), Zribi-Hertz (1993), and Ackema and Schoorlemmer (1994, 1995, 2003) – have proposed presyntactic analyses of middles. In these analyses, a “middle verb” is derived lexically from its non-middle counterpart by de-transitivizing its argument structure.1 This involves the lexical suppression of the non-middle verb’s external argument and the subsequent externalization of the next most prominent argument in the verb’s argument grid to the subject position. Other linguists – for example Hoekstra and Roberts (1993), den Dikken (1997), Bowers (2002), and Stroik (1992, 1995, 1999, 2000) – have advanced syntactic analyses of middles. These analyses derive middles syntactically, similarly to the way passives are formed. In particular, middle formation involves the suppression or demotion of the external argument of a middle verb, the syntactic projection of its internal argument into a canonical VP-internal position, and the syntactic promotion of the internal argument to the subject position. And still other linguists – such as Steinbach (2002) – have developed postsyntactic analyses of middles. In Steinbach’s analysis of German middles, the middle verb c-selects a non-argument reflexive object, which forces the internal argument to externalize and the external argument to be suppressed; as a result, a middle is a transitive reflexive construction that does not express its external argument syntactically. However, since middle verbs s-select external arguments, these arguments are visible semantically and are given semantic (generic) interpretations.2 . As Abraham (1994, 1995) points out, there is a difference between middle verbs (MVs) and middle constructions (MCs) – MCs require manner adverbs, while MVs do not. I don’t pursue this difference here. . In addition to the lexical, syntactic, and postsyntactic analyses of middle constructions mentioned, there are also analyses, such as Condoravdi (1989), Marelj (2004a), and Lekakou (2005), that propose that middle formation is not uniform across languages; rather, languages differ in the “parameterizable operations” (Marelj 2004a: 142) they use to form middles and, as an anonymous reviewer notes, they “employ different means to express the middle’s semantics.” Marelj, for example, claims that languages can form middles in different modules: some languages (e.g., English and Dutch) form middles in the lexicon and other languages (e.g., German and French) form middles in the syntax. Although these “parameterizable” analyses depart from the “pure” lexical and postsyntactic analyses in significant ways, they do share with the latter analyses the assumption that the external argument in middles is universally syntactically suppressed. Since my analysis focuses on this assumption and since the parameterizable analyses don’t offer perspectives on this assumption that differ instrumentally from those advanced by the “pure” analyses, I won’t integrate these parameterizable analyses much into my discussion of middles.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.3 (183-245)

Arguments in middles 

In this paper, I hope to shed some light on the theoretical viability of the above analyses of middle constructions by investigating how they handle the two facts about middle constructions that are most problematic to explain: (i) the optional presence of a for-PP in middles (2a); and (ii) the mandatory presence of a reflexive pronoun (RP) in German middle constructions (2b). (2) a. No Latin text translates easily (for me). b. Das Buch liest *(sich) leicht. the book reads RP easily ‘The book reads easily’ (German, from Steinbach 2002: 3)

I will show that the data in (2), which allow us to probe the essential argument structure of middles, are best treated by syntactic analyses.

. Genericity in middles Before I discuss the data in (2), I would like to make a few remarks about the genericity of middle constructions. I will follow Marelj (2004a: 113) in assuming that “the defining property of middles is that they are generic statements” – an assumption shared by most linguists including Condoravdi (1989), Steinbach (2002), Ackema and Schoorlemmer (2003), and Lekakou (2004). I will also follow Steinbach (2002), Marelj (2004a, 2004b), and Lekakou (2004) in assuming that the generic interpretation of middles results from the presence of a GEN(eric) operator at LF which binds any free variable in its sentential scope, including an event variable and, when present, a variable of the (implicit) Agent. Under the above assumptions, the middle construction This book reads easily will have the following interpretation: “it is a generic property of events of reading this book by any arbitrary person that they are events easily performed by that person.”3 The fact that the interpretation of middle constructions such as This book reads easily gives generic quantificational readings for both the reading-event and the arbitrary-agent of the reading, has led many linguists – including Fellbaum (1985), Fagan (1988), Zribi-Hertz (1993), Steinbach (2002), Ackema and Schoorlemmer (2003), and Marelj (2004a, b), among others – to assume that all middles have a syntactically-suppressed external argument that appears at LF, where it receives an ARB(itrary) or generic interpretation.4 I don’t subscribe to this assumption . This reading is patterned on Marelj’s (2004b) interpretation of middles. . In Stroik (1992, 1995), I propose an alternative way of explaining the generic interpretation of the “missing” Agent in (i). (i)

This book reads easily.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.4 (245-285)

 Thomas Stroik

because there are middle constructions that do not follow this pattern of interpretation. For me, middles such as This book reads easily for Lou must be interpreted as follows: “it is a generic property of events of Lou’s reading this book that they are events easily performed by Lou.” Under this interpretation, the middle has a generically interpreted event, but not a generically interpreted external argument. In fact, the interpretation of the external argument appears to be linked to the for-PP. I discuss this and other related issues in the next section.

. On for-PPs in middles The presence of the for-PP in (2a), as well as in sentences such as those given in (3), raises serious questions about the syntactic and semantic properties of middle constructions. Is this PP an experiencer argument of the adverb (as Hoekstra & Roberts 1993 argue), or the demoted external argument of the middle predicate (as Stroik 1992 proposes), or a beneficiary adjunct of the middle predicate (as Iwata 1999 maintains), or an experiencer adjunct of the middle verb (as Ackema & Schoorlemmer 2003 suggest)? (3) a.

The enemy battalion infiltrated suprising easily for the guerilla soldiers. (Tenny 1992: 10) b. Die stoel zit niet lekker voor opa. the chair sits not comfortably for grandpa ‘The chair doesn’t sit comfortably for grandpa’ (Dutch, Ackema & Schoorlemmer 2003: 54)

Let’s consider how presyntactic (lexical) analyses explain the properties of for-PPs in middles. These analyses, as previously noted, crucially assume that the external argument is completely suppressed in the syntax; hence these analyses insist that the for-PP in middles cannot be connected to the external argument (or the external argument won’t in fact be syntactically suppressed and the motivation for the externalization of internal argument will disappear – in other words the entire presyntactic analysis will unravel). This line of argument has compelled

Following Larson’s (1988) Principle of Argument Demotion, which allows an argument XP to be projected syntactically as an adjunct, I argue that the “missing” Agent argument in (i) is a non-overt DP (PRO) that has been demoted to a VP adjunct (see (ii)). (ii) This book [[reads easily] PRO]] This non-overt pronominal element will function as a variable that will be bound by the GEN operator at LF; as a result, the non-overt Agent (PRO) will receive an ARB or generic interpretation.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.5 (285-340)

Arguments in middles 

proponents of presyntactic analyses to propose that for-PPs must be interpreted as adjuncts unrelated to the external argument, typically as beneficiary adjuncts or experiencer adjuncts. However, under such an analysis, the interpretation of middles with for-PPs will have to include not only semantically-valued internal arguments and generically-interpreted external arguments, but also beneficiary or experiencer adjuncts. Are such interpretations really available for (2a) and (3a)? Let’s look at what happens when we try to give beneficiary readings to the for-PPs – we can see what such a reading might look like in (4). (4) a. They did it for Pat. b. It was for Pat (for Pat’s benefit) that they did it.

If the for-PPs in (2a) and (3a) can actually take benefactive readings similar to the one given to (4a) in (4b), then we should be able to paraphrase, under the assumptions about middle interpretation made by presyntactic analyses, (2a) as (5a) and (3a) as (5b). (5) a. It was for my benefit that people in general translated no Latin texts easily. b. It was for the guerilla soldiers (for the guerilla soldiers’ benefit) that people in general infiltrated the enemy battalion surprisingly easily.

Needless to say, the interpretations in (5) are not recognizable paraphrases of (2a) and (3a); in fact, the statements in (5) have little or no semantic connection to (2a) and (3a). A similar situation arises if we attempt to interpret the for-PPs in middles as experiencers. Now we know that for-PPs can be interpreted as experiencers – we can see this in (6). (6) a. It happened much too quickly for me. b. He makes mistakes much too often for his boss.

In these examples, the for-PPs can be experiencer adjuncts interpreted as “from X’s perspective.” Hence, (6a) can mean “from my perspective, it happened much too quickly” and (6b) can mean “from his boss’s perspective, he makes mistakes much too often.” Notice, though, that the for-PPs in (2a) and (3a) cannot be interpreted as experiencer adjuncts in similar fashion to the for-PP in (6a). That is, (2a) cannot be interpreted as (7a) and (3a) cannot be interpreted as (7b). (7) a. From my perspective, people in general translate no Latin text easily. b. From the guerilla soldiers’ perspective, people in general infiltrated the enemy battalion surprisingly easily.

As we can see from (7), for-PPs in middles cannot be interpreted as experiencers. Presyntactic analyses, as described above, cannot possibly get the interpretation of middles such as (2a) and (3a) right because the middle events in (2a) and

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.6 (340-408)

 Thomas Stroik

(3a) simply do not involve three participants – a generic external argument, an internal argument, and some sort of for-PP adjunct. Support for this conclusion comes from the fact that there is no way to paraphrase (2a) as (8), regardless of one’s interpretation of the for-PP. (2) a.

No Latin text translates easily for me.

(8) People in general translate no Latin text easily for me.

What (8) demonstrates, then, is that the interpretation of (2a) cannot include both a for-PP and a generically interpreted external argument. Consequently, since (2a) has a for-PP, it cannot also have a generically interpreted external argument. However, if, as is generally assumed, middles have semantically interpreted external arguments, and if (2a) has only an internal argument and a for-PP available for interpretation, then the for-PP must be a demoted external argument (as Stroik 1999, 2000 argues). In other words, (2a) should be interpreted not as (5a) or as (7a), but as (9). (9) I (generally) translate no Latin text easily.

It seems to me that (9) is indeed the translation of (2a) – a translation in which the genericity of (2a) applies to events in which I translate Latin texts and not to events that involve potential (generic) translators of Latin texts. If my analysis is correct, then data such as (2a) strongly suggest that middle construction involves argument demotion, not argument suppression (see Footnote 4 for a brief discussion of how middles without for-PPs still undergo argument demotion). Steinbach’s (2002) postsyntactic analysis fares no better in its ability to explain for-PPs in middle constructions than do presyntactic analyses. Since Steinbach follows presyntactic analyses in assuming that the external argument of a middle predicate is completely suppressed syntactically but that it is interpreted in the semantics generically, his analysis should have the same problem with sentences such as (2a), (3a), and (3b) that presyntactic analyses do – i.e., Steinbach would appear to be forced to assume, mistakenly, that these sentences have three participants. Surprisingly, Steinbach does not fall prey to this analysis because he does something presyntactic proponents don’t do: He grants that the for-PP can be linked to the external argument (therein he follows Stroik 1992, 1999). Doing so, however, leaves Steinbach with a logical problem. On the one hand, he must assume that “the first semantic [external] argument is not linked to a syntactic constituent” (2002: 2) to ensure that his Argument Linking Principle externalizes the internal argument of the middle predicate; on the other hand, he must admit that ‘examples [such as those in (2a)] . . . show that the first semantic argument can be linked to a syntactic constituent” (2002: 83) to ensure that middles with for-PPs are interpreted correctly semantically. This would appear to leave Stein-

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.7 (408-457)

Arguments in middles 

bach with a pernicious contradiction at the heart of his analysis. In an attempt to circumvent this contradiction, Steinbach maintains that the external argument of a middle predicate can be linked to a for-PP, but only if the for-PP does not refer to a specific entity or individual – see the examples in (10) taken from Steinbach (2002: 277).5 (10) a.

Ich finde, dass sich diese Bücher auch für kleine Kinder leicht I find that RP these books also for small children easily lesen. read ‘I find that these books also read easily for small children’ b. ??I finde, dass sich dieses Buch für Maria leicht liest. I find that RP this book for Maria easily reads ‘I find that the book reads easily for Maria’

In (10), the external argument can be linked to the for-PP only if the PP (and its linked external argument) can be interpreted generically. Needless to say, Steinbach’s caveat may mute the fact that he must both universally prohibit and circumstantially allow the external argument of a middle predicate to be linked to a syntactic constituent; however, the contradiction obviously continues to stand. Furthermore, even if we accept Steinbach’s caveat, we are still left with the problem of explaining the data in (2a) and in (3), where the for-PPs in these examples refer to specific individuals. Now it may be the situation that German differs from English and Dutch in not permitting middles to have specific for-PPs in middles. But regardless, it will still have to be the case that German allows external arguments to be linked to syntactic constitutents – the effect that features like genericity may have on this linkage in German, though relevant to the application or execution of the linking process, is irrelevant to the existence of the linkage. In other words, German does not differ from English or Dutch in permitting (some) external arguments of middles to show up syntactically in for-PPs as demoted adjunct arguments. The fact that the external argument of a middle predicate can be linked to a syntactic constituent seriously undermines the viability of presyntactic and post. A reviewer suggests, along the lines argued by Steinbach, that “though the for-PP [in English] restricts the range of possible candidates in the generic set, it [the range] cannot consist of a single member.” This does not seem to be correct because, as the examples in (i) and (ii) illustrate, the for-PP can in fact select a single member. (i)

This horse handles well only for his trainer.

(ii) This safe is supposed to open easily only for me.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.8 (457-495)

 Thomas Stroik

syntactic analyses of middle constructions. It does, however, square with the syntactic analyses posed by Hoekstra and Roberts (1993) and by Stroik (1992, 1999), which find structural parallels between the way that the external argument in a middle construction can be linked to a for-PP as a demoted argument and the way that the external argument in a passive construction can be linked to a byPP as an external argument. Some support for the above conclusion comes from the impossibility of having an argument show up twice in the syntax – once as a non-demoted argument and once as a demoted argument. We can observe this in (11). Notice that the prepositional object in the by-PP can be linked to the external argument (11a), but it can’t in (11b), where the external argument is projected in the syntax. (11) a. The book was read by me. b. John read the book by me.

In (11b), although the prepositional object can be interpreted in various ways, one interpretation not available to it is the one in which it is linked to the external argument. This interpretative restriction is imposed on the object because the external argument in (11b) is already linked to the DP John. A similar situation arises in (12). (12) a. The book translated quickly for me. b. John translated the book quickly for me.

Importantly, the prepositional objects in (12) appear to behave exactly as do the prepositional objects in (11). As with the objects in (11), the objects in (12) have quite different interpretations, which seem sensitive to the presence (or absence) of the external argument. That is, the prepositional object in (12a) can be linked to the external argument of the middle predicate; on the other hand, in (12b), the object is not allowed to be linked to the external argument when the external argument is linked to the DP subject John. The parallels between (11) and (12) are easily explainable if the prepositional objects in (11a) and (12a) have parallel relationships with external arguments.

. On German middles Not only do for-PPs reveal a great deal about the argument structure of middles, so do the reflexives in German middles. As Steinbach (2002) notes, German middle constructions are transitive reflexive constructions that must have reflexive objects, as (13) illustrates.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.9 (495-547)

Arguments in middles 

(13) Das Buch liest *(sich) leicht. The book reads (RP) easily ‘The book reads easily’

(Steinbach 2002: 3)

Amazingly, even impersonal middle constructions, which are formed from intransitive verbs, are transitive reflexive constructions – we can see this in (14). (14) Es tanzt *(sich) hier gut. it dances (RP) here good ‘Here you can dance well’

(Steinbach 2002: 29)

That German middles are transitive constructions and that German has impersonal transitives substantially undermines, according to Steinbach, both presyntactic and syntactic analyses of middle constructions. The reason that presyntactic analyses cannot account for the data in (13) and (14), as Steinbach points out, is that these analyses are essentially de-transitivizing analyses. They derive middle verbs by lexically suppressing the external argument, which then requires the next most prominent argument in the verb’s argument grid to externalize. Such analyses have no way to explain how a de-transitivized middle verb can become a transitive reflexive construcion, as in (13); and they are even more pressed to explain how the intransitive verb tanzen in (14) can undergo de-transitivization and yet become a transitive reflexive verb. Steinbach argues that since presyntactic analyses have no plausible explanation for the presence of the reflexive pronoun in German middle constructions and no way to account for impersonal middles, they are necessarily empirically inadequate approaches to middle formation. For Steinbach, syntactic analyses are as incapable of explaining German middles as are presyntactic analyses. The problem with syntactic analyses has to do with their assumption that middles are derived, in part, by promoting the object argument of a middle verb to satisfy the Case requirements of this argument. If middle formation does indeed promote an object argument to the subject position to satisfy the Case requirements, then German middles such as (13) leave us with a contradiction: On the one hand, the DP das Buch moves from its object position in the VP because it can’t have an accusative Case feature checked in the VP and on the other hand, as Steinbach observes, the reflexive sich in (13) does have an accusative Case feature that is checked in VP. If the DP cannot have an accusative Case feature checked, the reflexive shouldn’t be able to, either. The fact that the reflexive has its Case feature checked suggests that middles cannot be derived by the Case-driven syntactic promotion of object arguments. Furthermore, syntactic analyses have the logical problem of explaining how the syntactic promotion of an object argument applies to the derivation of impersonal middles (see (14)). Im-

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.10 (547-601)

 Thomas Stroik

personal middles have no object argument to promote. Hence, there should be no way to derive them along the lines required by syntactic analyses of middles. Although Steinbach’s (2002) arguments against presyntactic analyses are compelling, there are reasons to question his arguments against syntactic analyses. These reasons have to do with reflexive binding relations. Zwart (2002), following Kayne (2002), proposes that the coreferentiality that a locally bound anaphor has with its antecedent is a function of the Merge operation. According to Zwart, a locally bound anaphor is merged into the syntax together with its antecedent as a small clause XP (see (15)) and the antecedent is subsequently moved out of the XP to have its features checked.6 (15) [XP [antecedent] [anaphor]]

Now if Zwart is correct, then the German middle in (13) will have to be derived as follows (I make no comment here on the presence of the adverb). First, the verb will merge with the [XP [das Buch] [sich]] constituent, as in (16a); then the DP das Buch will move into the subject position to have its features checked, as in (16b). (16) a. [liest [[das Buch] [sich]]] b. [das Buch [liest [t [sich]]]] t is the trace, or copy, left by the DP movement.

In (16b), the DP das Buch will have its features checked in the subject position, and the reflexive sich will have its features checked in the VP. Notice that Zwart’s analysis of anaphoric coreference will not only support a syntactic analysis of middle formation, it will require a syntactic analysis. The viability of Zwart’s (2002) analysis of anaphors is obviously important to evaluating analyses of middles. Should Zwart’s analysis be correct, then both presyntactic and Steinbach’s (2002) postsyntactic analyses of German middles will be undermined because they won’t be able to explain the DP promotion in (16); however, if Zwart is wrong, then Steinbach’s previously cited arguments against syntactic analyses of middles will remain unchallenged. I won’t go into the particulars of Zwart’s analysis; however, there are two types of evidence involving middle constructions that supports a version of his analysis for middle constructions. First, the reflexive in German middles is, according to Steinbach, a special reflexive – it’s a non-argument reflexive (one that has no thematic/semantic interpretation). As a non-argument element, it has the same grammatical function

. Zwart (2002) doesn’t discuss non-argument reflexives in his analysis of anaphors. However, according to Steinbach (2002), the non-argument reflexives in German middles have exactly the same binding properties that argument reflexives have; hence, Zwart’s analysis should apply to these non-argument reflexives.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.11 (601-654)

Arguments in middles

of other non-argument elements, such as the expletives it and there in English, as in (17). (17) a. I resent it greatly that you fired my cousin. b. There are six trees in the orchard.

That is, the non-argument reflexive serves as a grammatical place-holder for a thematic argument to which it is linked (this is essentially Steinbach’s contention). It is important to underscore the fact that the non-argument reflexive (expletive) in German middles, the expletive it, and the expletive there are all linked to arguments. But what is the expletive-argument linkage? As Stroik (1996), Waller (1997), Yoon (2001), Kim (2002), and others argue, expletives are linked to arguments at Merge, akin to Zwart’s analysis of anaphors. In particular, the expletive it merges into the syntax together with is CP antecedent (see (18a)) and the expletive there merges with a NumP (see (18b)). (18) a. [CP it [that you fired my cousin]] b. [DP there [NumP six trees]]

The expletives will eventually have to move from their merged positions to have their various agreement features checked. If expletive-argument linkages are the function of the Merge operation, then it is also the case that the German reflexive expletive in middles is merged together with its link argument, as in (15) and (16a). On the meta-theoretical level this analysis of expletives is perfectly sensible because the Merge operation is, semantically, a compositionality operation that must merge only semantically non-vacuous elements. Hence, not only should it be impossible to merge an expletive into the syntax, but also the fact that an expletive and its linked argument share a single thematic role suggests that they must be merged as a single syntactic-semantic unit. Besides the forgoing conceptual argument for merging the reflexive in German middles together with its antecedent, there is also an emprical argument for doing so. Steinbach (2002) notes that non-argument reflexives have some special properties: They can’t be fronted (see (19a)); they can’t be dominated by a focus particle (see (19b)); they can’t be narrowly focused (see (19c)); they can’t be substituted for by a DP (see (19d)); they can’t be questioned (see (19e)); and they can’t be coordinated (see (19f)). (The examples in (19) are taken from Steinbach (2002: 140–141).) (19) a. *Sich RP b. *Das the

hat das Buch has the book Buch verkauft book sells

gut verkauft. well sold nur/sogar sich gut. only/even RP well



JB[v.20020404] Prn:6/11/2006; 9:40



F: LA9612.tex / p.12 (654-703)

Thomas Stroik

c. *Das Buch verkauft the book sells d. Das Buch verkauft the book sells e. *Wen hat das Buch who has the book f. *Das Buch verkauft the book sells

SICH gut. RP well sich/*seinen Autor gut. RP/its author well gut verkauft? Sich. well sold? RP sich und seinen Autor gut. RP and its author well

To account for data such as those in (19), Steinbach argues that non-arguments lack the semantic features necessary to participate in focus operations, question operations, argument-substitution operations, and coordination operations. The fact that the reflexive in German middles cannot undergo questioning (see (19e)) or argument-substitution (see (19d)) is unremarkable; afterall, the reflexive is an expletive and no expletives can undergo these operations – we can see this in (20), where the expletive it in (20a) can neither be questioned, as in (20b), nor replaced, as in (20c). (20) a. It seems that no one is waiting in the park. b. *Who/what seems that no one waiting is in the park? c. *Mary seems that no one is waiting in the park.

On the other hand, the inability of non-argument reflexives to be fronted and focused is surprising because expletives are not prohibited from participating in such operations. Notice, for example, that the expletive it can be fronted both to A-positions and to A-bar positions, as in (21a) and (21b). (21) a. It seems that Mary will win the election. b. It I firmly believe to be true that Mary will win the election.

In addition, this expletive can be dominated by a focus particle (as in (22)) and it can be narrowly focused, as (23) demonstrates. (22) a.

I believe not only it to be true that Mary will win the election, but also .... b. Not only is it true that Mary will win the election, but also . . . .

(23) I resent IT greatly that you fired my cousin.

The data in (21)–(23) show that expletives can be fronted and focused; hence, the fact that the reflexive in German middles fails to do so cannot simply be attributed to the non-argument (expletive) status of the reflexive. Although a purely semantic analysis can’t explain the data in (19a–b), a syntactic analysis can. If, as we have previously argued, the reflexive in German middles is merged together with its

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.13 (703-759)


antecedent, then (19a) and (19b) will have the derivations in (24a) and (24b), respectively. (24) a. *[t sich] hat das Buch gut verkauft. b. *Das Buch verkauft nur/sogar [t sich] gut.

The ungrammaticality of (19a) and (19b) can readily be explained from the derivations in (24): (24a) is ill-formed because the trace is not properly bound once the small clause is fronted; and (24b) is ungrammatical because it is not possible to extract constituents through a focus particle, as we can see in (25). (25) a. *Who did Mary hire only t b. *John was hired only t

And finally, of the conjunction data such as (19f), Steinbach (2002: 142) observes that “non-argument reflexives cannot be coordinated at all.” This remarkable property follows naturally under the analysis we are advancing here. If the nonargument reflexive in (19f) merges into the syntax as [das Buch [sich]], then it will be impossible, under any circumstances, to move the DP das Buch out of the structure [[das Buch [sich]] und [seinen Autor]] without violating Ross’s constraint on extraction out of coordinated structures. What the foregoing analysis suggests is that the special properties of the non-argument reflexives we see in (19a–f) cannot be explained in terms of the non-argument (or expletive) properties of the reflexives; however, they can be exhaustively explained if we follow Zwart’s (2002) analysis of anaphors – that is, if the reflexive is first merged into the syntax with its antecedent and then the antecedent is subsequently syntactically promoted.

. Towards an analysis of middle formation In the previous section, I have argued (i) that Steinbach’s analysis of the reflexive in German middles is essentially an antecedent-expletive analysis and (ii) that, as an expletive analysis, his analysis cannot explain the empirical data in (19a–c), despite his claims to the contrary. Furthermore, Steinbach’s analysis is beset with a conceptual problem. Since this analysis is an antecedent-expletive analysis, Steinbach must explain why, in (2b), the antecedent precedes its place-holder (expletive). (2) b. Das Buch liest *(sich) leicht. the book reads (RP) easily ‘The book reads easily’

(Steinbach 2002: 3)

The general relationship between antecedents and their place-holding expletives is that the expletives precede the antecedents (see (17)). This happens because



JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.14 (759-832)

 Thomas Stroik

expletives satisfy feature-agreement relations in structural positions too high for their antecedents to satisfy. Given these conditions, Steinbach must confront the daunting challenge of explaining why the antecedent in (2b) precedes its reflexive expletive (i.e., what agreement features the reflexive satisfies that could not be satisfied by the structurally superior antecedent), and of explaining how it’s possible to have expletives serve as the antecedents for other expletives, as happens in impersonal middle constructions (see (14)). (14) Es tanzt *(sich) hier gut. it dances RP here good ‘Here you can dance well’


What features are being checked by the two expletives es and sich in (14)? Do the reflexives in (2b) and (14) check the same features? And can an expletive (es in (14)) really serve as the antecedent for another expletive (sich)? Why should this happen? Needless to say, Steinbach’s analysis raises serious and important questions about the nature of expletives – questions that he does not address. The analysis of German middles I offer in the previous section is not an antecedent-expletive analysis; it’s an antecedent-reflexive analysis. As such, my analysis does not confront the conceptual problems surrounding the nature of expletives that Steinbach’s analysis does. Although my analysis side-steps the aforementioned problems, it raises other conceptual problems. In particular, if it is the case that the reflexives in German middles are not argument place-holders (as Steinbach suggests), then why do German and other languages (perhaps all other languages, according to Steinbach) exhibit reflexivity in their middles constructions? My analysis, then, must explain why reflexives show up at all in middles such as (2b); and it must explain the relationship between reflexivity in middles and both the syntactic demotion of the external argument (discussed in Section 3) and the syntactic promotion of the logical object (discussed in Section 4). What follows is a series of observations about the properties of middle constructions and about the syntactic derivation of middles. I attempt to show that there is a connection between the presence of reflexives in middle constructions, the event responsibility property assigned to the promoted object argument, the promotion of the logical object, the demotion of the external argument, and the absence of a vP in middles. My analysis at this point is highly speculative, but it seems promising because it offers a way to explain several inter-dependent properties of middles.7 . Although I discuss several of the salient properties of middle constructions, there are many properties that I will not address here, such as the modal interpretation of middles (see Fagan 1992 and Ackema & Schoorlemmer 2003), the obligatory presence of adverbs in middle con-

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.15 (832-874)


. Reflexives and event responsibility I will begin my analyis of middle formation by looking at the relationship that the presence of reflexives in middle constructions has with some of the salient characteristics of middle constructions. According to Steinbach, the most important characteristics of middle constructions are that they possess generic interpretations, that they have adjusted argument structures, and that they attribute special properties to the subject, including an event-responsibility property. In his analysis, Steinbach links the presence of reflexives in middles to the argument structure properties of middles.8 As we have already argued when we considered the sentences in (19a–f), there are good reasons to believe that the presence of reflexives in middles cannot be explained by (or reduced to) the argument properties of the reflexives; hence, the presence of reflexives is not directly related to the argument structure of middle constructions. But then reflexives do not seem related to the generic property of middles, either. After all, generic sentences needn’t have reflexives in them (see (26a)) and reflexives, in and of themselves, do not add any genericity to a sentence, as is the case in (26b), which does not have a generic interpretation. (26) a. Mammals urinate. b. John shaved himself yesterday.

What I would like to propose here is that the reflexives show up in middles not because they are linked to genericity or to argument structure, but because they are linked to event-responsibility.9 structions (see Abraham 1994), or the aspectual properties of middles (see Hulk & Cornips 2000), among other properties. . An alternative analysis of the presence of reflexives in middles has been advanced by Hulk and Cornips (2000), who link reflexives to the aspectual properties of middles. For Hulk and Cornips, reflexives convert a non-delimited event into a delimited one, thereby allowing a predicate to undergo middle formation. Ackema and Schoorlemmer (2003) show that this analysis of reflexives in middles runs into several empirical problems. . Linguists have long been aware of the importance of event-responsibility in middles – see, for example, Fagan (1992) and Steinbach (2002). Some linguists, including Fagan (1992), have attributed the grammaticality differences between (i) and (ii) to the fact that the subject in (i) contributes to the selling event, while the subject in (ii) doesn’t appear to bear any responsibility for the buying event. (i)

These books sell well.

(ii) *These books buy well. However, no one before me has proposed that event-responsibility is connected to the reflexivity expressed in middles (see my discussion of (33), (34), and (35)).



JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.16 (874-921)

 Thomas Stroik

Let me preface my discussion of the connection between reflexives and eventresponsibility with a few remarks about “event-responsibility.” Linguists including van Oosten (1977), Fagan (1992), and Steinbach (2002) have recognized that the subject of a middle construction has the property of being responsible for the event described by the verb (I call this property “event-responsibility”). This property, according to Ackema and Schoorlemmer (2003: 43), is essential to middle formation: “[it] appears to be needed next to whichever other conditions on middle formation might hold.” Despite the long-standing awarenss of the eventresponsibility property, no one has been able to explain how the grammatical subject (which is also the logical object) of a middle verb comes to have this property. My claim is that the subject in a middle construction (at least in some languages) can pick up this property from a non-argument reflexive to which it is linked because the reflexive has an event [RESP(onsibility)] feature. That non-argument reflexives do have an event [RESP(onsibility)] feature can be seen if we look at the non-argument, emphatic pronouns in (27).10 (27) a. The clerks will sell those books themselves b. *Old books amuse Mary themselves c. I amused Mary myself

Notice that the non-argument reflexive pronoun in (27a) is being used emphatically, but what it emphasizes is the clerks’ responsibility for the selling event. A similar situation arises in (27c): In this sentence, the emphatic reflexive emphasizes my responsibility for the amusing Mary event. Although (27a) and (27c) demonstrate that reflexives do have a [RESP] feature, (27b) shows that this feature must be linked to a DP that can be responsible for the event. In (27b), old books is a Theme argument that has no responsibility for the amusing Mary event; hence, the [RESP] feature of the reflexive cannot be linked to the DP subject old books. This leaves the [RESP] feature of the reflexive unlinked. Consequently, the [RESP] feature of the reflexive cannot be appropriately checked (or interpreted), which causes the sentence to be ungrammatical. We can find additional support for the connection between non-argument reflexives and event-responsibility if we examine the passive constructions and the middle constructions in (28) and in (29) respectively.11

. I also discuss the examples in (27) in Stroik (2004). . I have deliberately framed the sentences in (29) as questions and as negations to ensure that these sentences are read as middle constructions and not as virtual reflexive constructions (compare (29a) with (i), which is a virtual reflexive construction). (i)

My books (virtually) sell THEMSELVES. (The capital letters indicate primary stress.)

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.17 (921-972)


(28) a. *Will those books be sold themselves? b. *The dog wasn’t being chased itself. (29) a. Don’t my books always sell themselves? b. Memos to oneself generally don’t compose themselves very quickly.

Let’s consider the relationship between the non-argument reflexive and eventresponsibility in (28a). In this passive construction, the subject those books cannot carry a [RESP] feature because the implicit external argument continues to be the agent responsible for the selling event. As a result, the reflexive in (28a) cannot have its [RESP] linked to, and checked by, the subject DP. The [RESP] feature of the reflexive will then not be checked; and the sentence will be ungrammatical for the same reason that (27b) is. Notice that if the external argument is expressed syntactically in a passive, a non-argument reflexive can show up in the passive, but it must be linked to the DP Agent marked for event-responsibility. We can see this in (30), where a reflexive can be linked to a demoted Agent argument, but not to the sentential subject. (30) a. Those books will be sold by the store manager herself/*themselves. b. The dog was being chased by Mary herself/*itself.

(30) clearly illustrates that non-argument (emphatic) reflexives in passives must be linked to the event-responsible Agent arguments. Middles, however, behave differently than do the passives in (28) in their treatment of non-argument (emphatic) reflexives – we can see this in (29). As Steinbach notes, middles overtly express event-responsibility and they attribute it to their subject arguments (passives don’t have this property). Hence, in (29a), it is a property of my books that they are responsible for the selling event; more likely still, it is a property of the verb sell that it gives my books the property of being responsible for the selling event. In either case, the DP my books assumes a [RESP] feature. Given the above observations, the middles in (29a) will be well-formed, in part, because the [RESP] feature of the reflexive themselves will be checked against the [RESP] feature of the DP subject my books, thereby licensing the [RESP] feature for proper interpretation. The grammaticality of (29a), then, is a consequence of a proper linking between the reflexive and the subject DP, a linking that arises from sharing a [RESP] feature.

Stephens (in this volume) notes that although “virtual reflexives are often subsumed under the rubric of the middle constructions,” virtual reflexives have some interpretative properties that distinguish them from middles. However, as Stephens observes in Footnote 9, virtual reflexive readings do not appear possible in questions or in negation constructions. By framing the sentences in (29) as questions and negations, I circumvent the possibility that these sentences could be read as virtual reflexives.



JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.18 (972-1047)

 Thomas Stroik

Some indirect support for the role the [RESP] feature plays in middle formation can be seen in sentences (31a), taken from Ackema and Schoorlemmer (2003: 59), and (31b), taken from Ackema and Schoorlemmer (2003: 39). (31) a. *John believes t to be a fool easily. b. *Linguists sell such books well.

(Middle interpretation)

Example (31a) shows that middle formation cannot involve the raising of ECM subjects. This fact follows naturally under my analysis. If middle formation requires that the subject of the middle verb carry a [RESP] feature (possibly to check the event-responsibility of the middle verb), then (31a) will be ungrammatical because the ECM subject, as the logical subject of the embedded predicate be a fool, is not an argument of the verb believe and cannot have event-responsibility for the believing event. Consequently, the event-responsibility feature of the middle verb will not be checked in (31a) and, therefore, cannot be interpreted. The ungrammaticality of (31b) also involves the [RESP] feature of the subject. In (31b), the subject linguists is logically an indirect object and carries a Goal thematic role. As such, this subject does not participate causatively in the selling event and cannot be responsible for the event. Of note, even though ECM subjects and indirect objects will lack the appropriate [RESP] features to be the subjects of middle constructions, some adjuncts can be sufficiently responsible for events that they can be the subjects of middles, as is illustrated in the following Dutch examples taken from Ackema and Schoorlemmer (2003: 15).12 . English has a type of construction that looks quite similar to adjunct middle constructions (see (i) and (ii)). (i)

This pen writes very well.

(ii) This bed sleeps quite comfortably. Ackema and Schoorlemmer (2003) assert that the sentences in (i) and (ii) may appear to be adjunct middles, but they aren’t; instead, they are instrumental subject clauses. According to Ackema and Schoorlemmer, the reason that these sentences aren’t adjunct middles is because some of them can be transitive, as in (iii), while adjunct middles can never be transitives, as the example in (iv) illustrates, and because the sentences in (i) and (ii) can be eventive, while middles can’t be. (iii) This pen has written many a poem (iv) Deze theedoek droogt (*de borden) lekker this towel dries (the dishes) nicely ‘This towel is good to dry (the dishes) with’

(Ackema & Schoolemmer 2003: 17)

Given that Steinbach (2002) has demonstrated that middles are eventive constructions, Ackema and Schoorlemmer have only one way to discriminate adjunct middles from instrumental subject clauses, based on the potential transitivity of the verb. If we apply the transitivity test to the

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.19 (1047-1092)

Arguments in middles 

(32) a.

Deze naalden breien lekker these knitting-needles knit nicely ‘These knitting-needles are nice to knit with’ b. Deze stoel zit lekker this chair sits comfortably ‘This chair is comfortable to sit on’

As Ackema and Schoorlemmer argue, although the DP subjects in (32) have been generally analyzed as adjuncts to the events described by the verbs, these DP are actually arguments of the verbs in that they are necessary participants in the events. In (32b), for example, the subject deze stoel participates in the sitting event. The role this chair plays in the sitting event is so crucial to the event that not having the chair could prevent the event from occurring. In this way, the chair assumes a degree of responsibility for the event – enough responsibility to bear a [RESP] feature. A related explanation applies to (32a) and, according to Ackema and Schoorlemmer, to all “adjunct” middle constructions (for Ackema and Schoorlemmer, only argument adjuncts can be the subjects of adjunct middle constructions). The adjunct middles in (32), then, can be accounted for in a straightforward fashion if we assume that a [RESP] feature is at the center of middle formation. Given the forgoing remarks, let’s see how we can account for the derivation of middle constructions. My analysis, as outlined in the previous section and further developed in this section, will explain the German middle in (33) in the following way. (33) Das Buch liest sich leicht. ‘The book reads easily’

(Steinbach 2002: 3)

The middle verb liest, which has an event-responsibility feature, will select an internal object das Buch which must be marked with a [RESP] feature. Not having an intrinsic [RESP] feature, the object must pick up this feature from a [RESP] reflexive. The object will merge into the syntax, in accordance with Zwart’s analysis of anaphors, structurally linked to the reflexive sich that it binds in the [[das Buch] sich] constituent. The reflexive sich, which has a [RESP] feature, will share this feature with its antecedent das Buch via feature-agreement within the merge

sentence in (v), it would seem that this sentence is an adjunct middle because it has an “adjunct” subject but it does not permit a transitive counterpart, as (vi) shows. (v) This pool swims quite nicely. (vi) This pool swims (*the backstroke) quite nicely. cp. I can swim the backstroke in this pool quite nicely. The evidence in (v) and (vi) suggests that English might have adjunct middle constructions, in addition to having instrumental subject clauses.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.20 (1092-1148)

 Thomas Stroik

constituent. At this point in the derivation, the feature requirements of the various DPs will drive middle formation. The verb will check the Case and agreement features of the reflexive; however, it will not be able to check the features of the antecedent DP das Buch. Therefore, this DP will have to move into the subject position to have its morphological features checked. Notice that under this analysis of the German middle construction in (33), the DP object must be syntactically promoted to the subject position to ensure that its features are appropriately checked. I propose a similar explanation for all German middles, including impersonal middles – see (34). (34) Es tanzt sich hier gut. ‘Here you can dance well’


As a middle verb, the verb in (34) selects an object marked with an eventiveresponsibilty feature – which requires, as we argued above, that the verb take a [RESP] reflexive object. Since the verb tanzt does not take a thematic objectargument to link with the reflexive, the reflexive will have to be linked with an expletive-antecedent (to satisfy Zwart’s condition on anaphors). Hence, the verb tanzt will merge with [[es] sich]. The expletive es in (34) will subsequently move into the subject position for all the same reasons that the DP das Buch moves in (33). (We can see here that despite Steinbach’s claim to the contrary, it is in fact possible to give a syntactic analysis of German impersonal middles.) My syntactic analysis of middles also extends to English middles. If we follow Steinbach in assuming that English middles can have a null reflexive (Steinbach correlates middle constructions in Indo-European languages with weak reflexive markers and he suggests that the weak reflexive marker in English is the null reflexive). Under the above assumption, the middle verb in (35) will select an internal argument marked for event-responsibility. (35) This book reads well.

As with German middles, the internal object in English middles must inherit its [RESP] feature from a (null) reflexive marker to which it is linked in a Zwart’s type construction: [[this book] Reflexive]. After the DP+Reflexive constituent merges with the verb, the DP this book follows the same derivational path that the antecedent DPs take in (33) and (34). That is, the DP must move eventually into the subject position to have its morphological features checked. Importantly, given the presence of a null reflexive marker in English, we can syntactically derive English middles in a parallel fashion to the way we derive German middles – via the syntactic promotion of the object argument.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.21 (1148-1208)


. Middles and vPs In the previous section, I argue that middle formation crucially involves eventresponsibility: The middle verb selects an object marked for event-responsibility; the DP object must pick up an event-responsibility feature from a non-argument reflexive with which it forms a constituent [[DP] [Reflexive]]; and the DP must subsequently move from its merged object position to have its morphological features checked. I also argue that the process of middle formation described above applies to both English middles and German middles. Although I contend that should we derive English and German middles in parallel fashion, Marelj (2004a, 2004b) argues that we shouldn’t. She asserts that German middles should be derived syntactically, but English middles should be derived lexically. For her, the subjects of English middles are not syntactically promoted from an internal-argument (object) position to the subject position; rather, they are merged directly into the subject position as externalized arguments (this position is held by all presyntactic and postsyntactic analyses of middles). Despite her arguments, there is, I believe, one important reason for not accepting her analysis – that is, middle constructions, as I shall show presently, do not have vPs. If we follow Chomsky (1995) in assuming that the vP projection is the projection in which a verb’s external argument merges, then any construction which lacks a vP will not be able to have an external argument. In other words, should middle constructions not have vPs, they won’t be able to have external (subject) arguments and Marelj’s lexical analysis of English middles will be thrown into question, as will all the analyses that also externalize the object argument, such as those in Condoravdi (1989), Fagan (1992), Ackema and Schoorlemmer (1995), Steinbach (2002), and Lekakou (2005). My argument at this point, unfortunately, gets a little tentative because it’s rather difficult to isolate the vP projection; however, we appear able to target the vP projection in DPs that have reduced ing-relative clauses (formerly called Whiz Deletion constructions). If we look at the bracketed DPs in (36), we can observe that these DPs can include any reduced relative except for those with verbs known not to have vPs. (36) a. b. c. d. e. f. g.

I’ll talk to [anyone reading Chomsky’s latest book]. I’ll talk to [anyone arriving on time]. I’ll talk to [anyone resembling me]. I’ll talk to [anyone knowing the answer]. I’ll talk to [anyone wanting to run for president]. *I’ll talk to [anyone seeming to be leaving]. *I’ll talk to [anyone being tall].



JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.22 (1208-1251)

 Thomas Stroik

As the data in (36) show, the reduced ing-relatives can have stative verbs (36c,d), they can have unaccusative verbs (36b), they can have activity verbs (36a), they can have accomplishment verbs (36b), and they can have psychological verbs (36e). What these reduced ing-relatives, however, cannot have are verbs/predicates such as seem and be tall, i.e. verbs that lack a vP projection. It is important to emphasize here that we cannot explain the ungrammaticality of (36f) and (36g) in terms of the stativity of the predicates since, as (36c) and (36d) illustrate, stative predicates can show up in reduced ing-relative constructions. This strongly suggests that reduced ing-relatives are not sensitive to verb type; instead, they are sensitive to the presence or absence of a vP projection. Under the assumption that reduced ingrelatives are vP relatives, we can give a simple and natural explanation for (36f) and (36g). These sentences are ungrammatical, under the above assumption, because the reduced relative constituents must be vPs and the verbs in these two sentences cannot project vPs. Notice that the predicates seem and be tall are not prevented from participating in CP relatives, as is illustrated in (37); they are only prevented from being in vP relatives. (37) a. I’ll talk to [anyone that seems to be leaving]. b. I’ll talk to [anyone that is tall].

If the above analysis of the data in (36) is correct, reduced ing-relative constructions provide a good test of vP-hood. Using reduced ing-relatives as a test for vP-hood, we can now probe the internal structure of English middle constructions. To this end, consider the examples in (38). (38) a. *I’ll pay attention to [any Latin texts reading easily]. (cp. I’ll pay attention to [any Latin texts that read easily]) b. *I’ll buy [any silk dresses washing easily]. (cp. I’ll buy [any silk dresses that wash easily]) c. *I’ll spend hours in [any pool swimming comfortably]. (cp. I’ll spend hours in [any pool that swims comfortably]) d. *I’ll talk about [any Greek poem translating easily]. (cp. I’ll talk about [any Greek poem that translates easily])

We can see, from the data in (38), that middle verbs cannot show up in reduced ing-relative constructions. Given that the ability of a verb to co-occur within a reduced ing-relative correlates with the abilty of the verb to project a vP, we can conclude that middle verbs, which are prohibited in reduced ing-relatives, do not have vP projections. Importantly, positing that middles do not have vP projections allows us not only to account for the data in (38), but also to resolve the adverb modification problem noted by Marelj (2004a: 131–132). Building on Roberts’ (1986) obser-

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.23 (1251-1310)

Arguments in middles 

vation that middles in English resist certain adverbial modifiers – adverbs such as deliberately, voluntarily, and safely, as well as rationale clauses (see (39)) – because these modifiers are incompatible with the stativity of middles, Marelj finds it puzzling that middles also resist adverbials modifiers such as passionately and completely (see (40a)), which are compatible with stative predicates, as is shown in (40b). (39) *This book reads deliberately/voluntarily/safely/to impress Mary. (40) a. *This book reads passionately/completely. b. Mary loves John passionately/completely.

We can solve the problem posed by the data in (40) by looking at the way verbs that lack vPs treat the adverbs passionately and completely. The examples in (41) show that verbs that do not project vPs cannot occur with the above adverbs. (41) a. *Mary is tall passionately/completely. b. *Mary seems passionately/completely to be leaving.

The ungrammaticality of the middle construction in (40a), then, appears to be linked to the fact that middles do not have vPs – which nicely corroborates our analysis of the middles in (38). Interestingly, this line of thought invites the larger possibility that the constraint on modification in middles (see (39)) arises because middles lack vPs. That is, middles simply cannot have any vP modifiers since middles don’t have vPs. This possibility seems confirmed by the fact that predicates lacking vPs such as seem and be tall not only cannot co-occur with passionately-type adverbs, but they also can’t co-occur with deliberately-type adverbs (see (42)). (42) a. *Mary is tall deliberately/completely/safely. b. *Mary seems deliberately/completely/safely to have left.

It is worth noting that by correlating adverbial modification in middles with the absence of a vP projection, rather than with the stativity of the middle verb, we can explain not only the data in (40a) but also the data in (39). Let’s now try to synthesize the various observations and speculations we have been putting forth in this section. First, and perhaps foremost, middle verbs do not project vPs. This grammatical fact has several significant consequences: (i) a middle verb cannot syntactically merge with an external argument because it lacks the vP projection needed to receive/host this argument; (ii) the valency (argument) reduction identified by Steinbach (2002) as being one of the core properties of middles results from the fact that middles, lacking vPs, cannot project external arguments; (iii) the event-responsibility feature typically expressed in the vP cannot be expressed there in middles due to the absence of the vP – rather, this feature is absorbed by the verb and projected, if thematically possible, onto its complement;

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.24 (1310-1353)

 Thomas Stroik

and (iv) in the absence of vPs, adverbial modification in middles will not permit any adverbials that are vP modifiers. That middles lack vPs and cannot merge with external arguments limits the sorts of theories that can explain middle constructions. No theory that requires middles to externalize an argument – merging it in a subject position – should in principle be allowed because middles have no place in their structure to merge a thematic argument outside the VP. Since all currently formulated lexical, semantic, and hybrid (cross-linguistic) theories of middles assume that middles do externalize the object argument, they would seem built upon questionable premises. Second, without a vP projection to host a thematic argument, all the syntactically projected arguments of a middle verb must be VP-internal (this echoes Abraham’s (1994: 823) observation that “MCs [middle constructions] are VPs in terms of full (clausal) rhemata”). One of these arguments, however, must be able to carry a [RESP] feature, which it inherits from a [RESP] reflexive to which it is linked, in order to check the event-responsibility of the verb. Not able to have its morphological features checked in its VP-internal merged position, the [RESP] object argument must move for feature-checking purposes, ending up in the subject position. Because the argument movement is morphologically driven (and required), middles are necessarily syntactically derived constructions that involve object promotion operations. Last, but not least, the external argument of a middle predicate has not been syntactically obviated. This argument cannot be merged in its canonical argument position (the Spec vP position) because the middle lacks a vP; however, it can show up syntactically, though in a (demoted) for-PP adjoined to the VP (but also see my discussion of the “missing” external argument in Footnote 4).

. Conclusion In this paper, I have tried to answer two crucial questions about middles: Can an external argument show up syntactically in middles as a demoted argumentadjunct? And can middle formation involve argument promotion? If my analysis is correct, middle formation can involve both argument demotion and argument promotion. I have also speculated about the relationship between argument demotion and the (missing) vP projection in middles; the relationship between reflexivity and event-responsibility in middles; and some of the constraints placed on adverbial modification in middles. My speculations, tentative as they may be, will, I hope, lead to further investigation of middle constructions.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.25 (1353-1459)

Arguments in middles 

References Abraham, Werner (1994). Review of Fagan 1992 The Syntax and Semantics of Middle Constructions. Language, 69, 817–825. Abraham, Werner (1995). Diathesis: The middle, particularly in West Germanic. In Werner Abraham, Talmy Givón, & Sandra A. Thompson (Eds.), Discourse Grammar and Typology (pp. 3–48). Amsterdam: John Benjamins. Ackema, Peter, & Maaike Schoorlemmer (1994). The middle construction and the syntaxsemantics interface. Lingua, 93, 59–90. Ackema, Peter, & Maaike Schoorlemmer (1995). Middles and non-movement. Linguistic Inquiry, 26, 173–197. Ackema, Peter, & Maaike Schoorlemmer (2003). Middles. In Martin Everaert & Henk van Riemsdijk (Eds.), The Blackwell Companion to Syntax, Vol. III. CD-ROM. Malden: Blackwell. Bowers, John (2002). Transitivity. Linguistic Inquiry, 33, 183–224. Condoravdi, Cleo (1989). The middle: Where semantics and morphology meet. MIT Working Papers in Linguistics, 11, 16–31. Den Dikken, Marcel (1997). Take serials light up the middle. Paper given in Berlin, July, 1997. Fagan, Sarah (1988). The English middle. Linguistic Inquiry, 17, 501–557. Fagan, Sarah (1992). The Syntax and Semantics of Middle Constructions. Cambridge: Cambridge University Press. Fellbaum, Christiane (1985). Adverbs in agentless actives and passives. Chicago Linguistic Society, 21, 21–48. Hoekstra, Teun, & Ian Roberts (1993). Middle constructions in Dutch and English. In Eric Reuland & Werner Abraham (Eds.), Knowledge and Language: Lexical and Conceptual Structure (pp. 183–220). Dordrecht: Kluwer. Hulk, Aafke, & Leonie Cornips (2000). Reflexives in middles and the syntax-semantics interface. In Hans Bennis & Martin Everaert (Eds.), Interface Strategies (pp. 207–222). Amsterdam: Elsevier. Iwata, Seizi (1999). On the status of implicit arguments in middles. Journal of Linguistics, 35, 527–553. Kayne, Richard (2002). Pronouns and their antecedents. In Samuel Epstein & Daniel Seely (Eds.), Derivation and Explanation in the Minimalist Program (pp. 133–166). Oxford: Blackwell. Kim, Hye-Kyung (2002). Conjunction and agreement. Language and Information Society, 3, 63– 79. Larson, Richard (1988). On the double object construction. Linguistic Inquiry, 19, 335–391. Lekakou, Marika (2004). The crosslinguistic realization of the middle semantics: Imperfective aspect and reflexive paradigms. In Solstad, Lyngfelt and Krave (2004), 67–74. Lekakou, Marika (2005). In the Middle, Somewhat Elevated. The Semantics of Middles and its Crosslinguistic Realization. Ph.D. dissertation. University College London. Marelj, Marijana (2004a). Middles and Argument Structure across Languages. Utrecht: LOT. Marelj, Marijana (2004b). Uniformity and variation across languages: The case of middles. In Solstad, Lyngfelt and Krave (2004), 75–82. Roberts, Ian (1986). The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris.

JB[v.20020404] Prn:6/11/2006; 9:40

F: LA9612.tex / p.26 (1459-1524)

 Thomas Stroik

Solstad, Torgrim, Benjamin Lyngfelt, & Maria Krave (Eds.). (2004). Pre-Conference Proceedings of the Oslo Workshop ‘Demoting the Agent: Passive and Other Voice-Related Phenomena’. Oslo: University of Oslo. Steinbach, Markus (2002). The Middle Voice. Amsterdam: John Benjamins. Stroik, Thomas (1992). Middles and movement. Linguistic Inquiry, 23, 127–137. Stroik, Thomas (1995). On middle formation: A reply to Zribi-Hertz. Linguistic Inquiry, 26, 165–171. Stroik, Thomas (1996). Extraposition and expletive-movement. Lingua, 99, 237–251. Stroik, Thomas (1999). Middles and reflexivity. Linguistic Inquiry, 30, 119–131. Stroik, Thomas (2000). Syntactic Controversies. München: Lincom. Stroik, Thomas (2004). Review of Steinbach 2002 Middle Voice. Journal of Comparative Germanic Linguistics, 8, 225–241. Tenny, Carol (1992). The aspectual interface hypothesis. In Ivan Sag & Anna Szabolcsi (Eds.), Lexical Matters (pp. 1–27). Stanford: CSLI. Van Oosten, Jeanne (1977). Subjects and agenthood in English. Chicago Linguistic Society, 13, 459–471. Waller, Brad (1997). Towards a proper charcterization of the English there. Ms., MIT. Yoon, Hang-Jin (2001). Expletive it in English. Studies in Generative Grammar, 11, 543–562. Zribi-Hertz, Anne (1993). On Stroik’s analysis of English middle constructions. Linguistic Inquiry, 24, 583–589. Zwart, Jan-Wouter (2002). Issues relating to a derivational theory of binding. In Samuel Epstein & Daniel Seely (Eds.), Derivation and Explanation in the Minimalist Program (pp. 269–304). Oxford: Blackwell.

JB[v.20020404] Prn:10/11/2006; 12:41

F: LA96LI.tex / p.1 (46-157)

Language index

A Albanian , – passim C Calabrian , ,  Cosenza n Czech – D Danish , –, , –,  Dutch n, n, –, , n, , , – E English n, –, n, , –, , , , , , n, –, , –, n, –, –, –, , , –, , –, , , , , , –, –, , , n, –, , – Colloquial English  Midland American English ,  Scottish English  Estonian –, n F Finnish –,  French n, –, –, , , , 

Friulian –,  G German , n, , –, , , , , –, –, –, , –, , –, –, – Germanic , –, , , , , n, n Greek , , , , , , , – H Hebrew – Hungarian – I Icelandic , n, n, – Irish , , –,  Italian , – Southern Italian ,  K Khanty – Eastern Khanty – Vasyugan Khanty , – Yugan Khanty  N Neapolitan , n, n Norwegian n, –, , , –, n, , n, –, –, –, 

P Polish , , –, , , n

R Romance , , – Italo-Romance – Romanian  Russian , –, –

S Salentino –,  Sardininan –, – Slavic –, , ,  Spanish , , , n,  Latin American Spanish n Swedish , –, n, –, n, n, 

T Tucano 

U Ukrainian , , –, , n

V Venetan –, n

JB[v.20020404] Prn:13/11/2006; 11:32

F: LA96NI.tex / p.1 (46-190)

Name index

A Abraham, Werner n,  Ackema, Peter , , , , , , n, , , ,  Åfarli, Tor n,  Aissen, Judith – Alexiadou, Artemis n, n Anagnostopoulou, Elena n, n Andersson, Erik 

Doherty, Monika , n, n, , n, –, n, , n,  Dowty, David , n

B Baker, Mark , , ,  Barðdal, Jóhanna ,  Bhatt, Rajesh  Björnström, Ebba  Blasco Ferrer, Eduardo n Blevins, James  Bowers, John n, , n,  Brennan, Virginia  Burzio, Luigi  Bybee, Joan –, 

F Faarlund, Jan Terje ,  Fagan, Sarah n, n, , , , n, ,  Falster Jakobsen, Lisbeth ,  Fellbaum, Christiane , n, , n, –, n,  Fiengo, Robert n,  Folli, Raffaella  Frazer, Timothy  Fried, Mirjam , n

C Calabrese, Andrea n Carpenter, Kathie  Chierchia, Gennaro , ,  Chomsky, Noam n,  Clark, Eve  Cocchi, Gloria n Condoravdi, Cleo n, , –, , n, ,  Cornips, Leonie n D Davidse, Kristin  Davis, Henry  Demirdache, Hamida  den Dikken, Marcel  Dixon, Robert , n

E Emonds, Joseph n,  Engdahl, Elisabet , n, n, , n, n, – Enger, Hans-Olav ,  Erelt, Mati 

G Geniušien˙e, Emma n, – Givón, Talmy ,  Gronemeyer, Claire  H Haase, Martin n Haider, Hubert  Hakulinen, Auli ,  Hale, Ken , ,  Harley, Heidi , n Haspelmath, Martin , ,  Heine, Bernd n,  Hellberg, Staffan  Heltoft, Lars , , ,  Hoekstra, Teun , , , , , 

Holmberg, Anders n Hopper, Paul ,  I Israeli, Alina n,  J Jaeggli, Oswaldo ,  Janda, Laura n Jespersen, Otto  Johnson, Kyle , ,  K Kallulli, Dalina n, n, , –,  Kayne, Richard n,  Kemmer, Suzanne n Keyser, Jay , , , n Kim, Hye-Kyung  Kjartansson, Helgi Skúli  Klaiman, M. H.  Koenig, Jean-Pierre –,  Kratzer, Angelika n Kress, Bruno n,  Kulonen, Ulla-Maija , n L Laanemets, Anu –, n,  Lalande, Jean-Yves n Lambrecht, Knud n, – Larson, Richard n Lavine, James n, n Ledgeway, Adam , n, n Lekakou, Marika n, n, , n, ,  Levin, Beth –, , , n, –, , , , ,  Lødrup, Helge n, n

JB[v.20020404] Prn:13/11/2006; 11:32

F: LA96NI.tex / p.2 (190-272)

 Name index Löflund, Juhani ,  Lyngfelt, Benjamin n, n,  M Maling, Joan n, –, n, ,  Manzini, Maria  Marelj, Marijana n, , – Mauner, Gail –,  McConnell-Ginet, Sally – Mikkelsen, Kristian  N Nelson, Diane  Nemvalts, Peep  O Onishi, Masayoshi  Ottósson, Kjartan n, n P Pagliuca, William –,  Palmer, Frank n, n Pancheva, Roumyana n Perkins, Revere –,  Plungian, Vladimir n, – Poletto, Cecilia n Q Quirk, Randolph n

R Ramchand, Gillian n Rapoport, Tova n, n Rappaport Hovav, Malka –, , , n, –, , , , , – Reinhart, Tanya , , , n, n Remberger, Eva-Maria , n, , n, n Roberts, Ian , , , , –, , , , , ,  Roeper, Tom  S Schoorlemmer, Maaike , , , n, –, n, n, , –,  Shibatani, Masayoshi , , , ,  Siewierska, Anna , , n Sigurjónsdóttir, Sigríður , –, n, ,  Siloni, Tal n Simon, Beth  Sobin, Nicholas ,  Solstad, Torgrim n, n, n Steinbach, Markus n, , –, –, n, ,  Sternefeld, Wolfgang n, n, –

Stroik, Thomas , n, , n, ,  Sundman, Marketta  T Teleman, Ulf  Thompson, Sandra ,  Thorell, Olof  Tomić, Olga n V Vallduví, Eric – Van Auwera, Johan n, – Van Oosten, Jeanne ,  Vilkuna, Maria – W Waller, Brad  Western, August  Williams, Edwin ,  Y Yoon, Hang-Jin  Z Zaenen, Annie  Zribi-Hertz, Anne – Zwart, Jan-Wouter –, , –

JB[v.20020404] Prn:13/11/2006; 11:57

F: LA96SI.tex / p.1 (46-174)

Subject index

A absolute construction , n absolute vs. partial demotion  accessibility , –, ,  activation –, –, , –, ,  see also inactivisation; reactivation actor –, – vs. agent – adverb of animacy  see also agent-oriented adverb; subject-oriented adverb agent backgrounding – see also prominence: participant agent-oriented adverb –, , –, –, n,  see also subject-oriented adverb agent-oriented modality – agented passive –, –, t,  agentivity-neutral predicate  animacy –, –, ,  control – anonymous-agent reflexive – passim agent referent – communicative function , – event-reporting –,  human agent –,  imperfective verb , , – indirectness and politeness  patient referent –

anticausative –, –, , –,  see also unaccusative unintended causation –,  AR see anonymous-agent reflexive argument structure –, , , n,  aspect be-passive vs. anonymous-agent reflexive , –, – av-phrase –, n see also by-phrase

B background , –, t,  see also foreground balanced information distribution – be-passive –, –, –,  see also bli-passive communicative function  event endpoint , , – imperfective verb  perfective verb  syntactic transitivity  bli-passive –, –,  BID see balanced information distribution binding –, –, – see also control by-phrase –, –, –, –, –, – see also for-phrase; from-phrase middles –

C causative n, –, , – alternation –, , – dyadic  unaccusatives – passim cause –, –, , – see also causative vs. agent –, – external cause –, ,  feature – causer –, , , , –,  change of state , –, –, – COS see change of state Construction Grammar , , n constructional map , f, f control , –, –, , –, –,  subject control  control clause  see also purpose clause

D dative –, , – demotion vs. promotion –, –, , ,  deobjectification – direct vs. indirect agent – disposition n, n, n, – , – see also genericity

JB[v.20020404] Prn:13/11/2006; 11:57

F: LA96SI.tex / p.2 (174-303)

 Subject index

E episodic middle – see also genericity ergativity –, –, – ergative construction  see also Locative S/A construction event-centered – event focus –,  event-responsibility –, ,  event-type passive – expletive , –, , , –,  external θ-role –, ,  see also external argument external argument –, , , –, –, , , – see also external θ-role F for-phrase n, , n, – foreground , –, –, –,  see also background from-phrase –, –, ,  functional domain , , , – functional space , – G genericity imperfective aspect  middles , –, , n, , – reflexives  s-passive –,  virtual reflexive vs. middle n zero person construction ,  generic agent sentence – human agent  generic indefinite  see also genericity I imperfective aspect , –

impersonal construction –, –, , –, –, – see also impersonal passive; impersonal reflexive; zero person construction anaphor binding –, – definite pronoun referent  subject-oriented adjunct – impersonal passive –, , , , – impersonal reflexive – implicit argument binding –, –, t, –, –, –,  impersonal and zero person constructions – passim middles –, –, – passive –, , –, – reciprocals , –, – testing for –, –, –,  inactivisation  see also activation; reactivation indirect vs. direct agent – information structure –, –, , , – passim focus –,  rheme n, – theme –, –, , , – topic –, –, , , , , , – inherently agentive predicate  intentionality , ,  adverbs –,  intent feature – invisible argument see implicit argument L lexical reflexive – Locative S/A construction –, , , , –

M middle adverbial modification – passim,  presyntactic analysis , , – postsyntactic analysis , –,  syntactic analysis , –, –,  modality middle –, –, –, n, n WANT-passive –, , – morphological passive – N non-active anticausative – morphology –, – voice – non-agentive predication , – null subject , , –, n, – see also impersonal construction; zero person construction null-subject language  O oblique-marked agent –,  P passive agented –, –, t,  auxiliary –, –, ,  event-type – human agent ,  participle –, n, , , ,  pragmatic function , , – passive reflexive  see also anonymous-agent reflexive patient-causer reflexive – patient-subject construction – passim

JB[v.20020404] Prn:13/11/2006; 11:57

F: LA96SI.tex / p.3 (303-395)

Subject index  perfective aspect ,  periphrastic passive  see also be-passive; bli-passive possessive – pragmatic identifiability – presentational construction – PRO , , –, , – see also control; purpose clause little pro vs. big PRO – prominence participant ,  event endpoint  promotion vs. demotion –, –, , ,  Predication phrase , n, ,  argument structure , n PrP see Predication phrase purpose clause , , , , , –, , , , ,  see also control R rationale clause see purpose clause reactivation – see also activation; inactivisation reflexive passive n, – see also anonymous-agent reflexive; passive reflexive

RESP see event-responsibility result-centered – S s-passive –, – aktionsart – genericity –,  objective report vs. subjective prediction ,  tense – semantic map see also constructional map modality – subject-oriented adverb –, , , –, – see also agent-oriented adverb middle –,  vs. agent-oriented adverb n suppression argument , , –, , ,  feature , – syntactic unergativity , ,  see also ergativity T transitivity –, , , , –, – U unaccusative –, –, –, –, , –, n

active vs. passive n, , – agent-oriented adverb –,  alternation , ,  dyadic –,  genericity  impersonalisation  incompatibility with by-phrase –, – lexical restrictions on passive formation – virtual reflexives –, – Uniformity of Theta Assignment Hypothesis  V virtual reflexive – passim vP n, – W WANT-passive – passim word order , n, , , n,  Z zero person construction –, , , n, –, – see also impersonal construction anaphor binding – pronoun referent 

Linguistik Aktuell/Linguistics Today A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 101 Karimi, Simin, Vida Samiian and Wendy Wilkins (eds.): Phrasal and Clausal Architecture. Syntactic derivation and interpretation. In honor of Joseph E. Emonds. vi, 411 pp. + index. Expected February 2007 100 Schwabe, Kerstin and Susanne Winkler (eds.): On Information Structure, Meaning and Form. Generalizations across languages. x, 557 pp. + index. Expected February 2007 99 Martínez-Gil, Fernando and Sonia Colina (eds.): Optimality-Theoretic Studies in Spanish Phonology. vi, 548 pp. + index. Expected February 2007 98 Pires, Acrisio: The Minimalist Syntax of Defective Domains. Gerunds and infinitives. 2006. xiv, 188 pp. 97 Hartmann, Jutta M. and László Molnárfi (eds.): Comparative Studies in Germanic Syntax. From Afrikaans to Zurich German. 2006. vi, 332 pp. 96 Lyngfelt, Benjamin and Torgrim Solstad (eds.): Demoting the Agent. Passive, middle and other voice phenomena. 2006. x, 333 pp. 95 Vogeleer, Svetlana and Liliane Tasmowski (eds.): Non-definiteness and Plurality. 2006. vi, 358 pp. 94 Arche, María J.: Individuals in Time. Tense, aspect and the individual/stage distinction. 2006. xiv, 281 pp. 93 Progovac, Ljiljana, Kate Paesani, Eugenia Casielles and Ellen Barton (eds.): The Syntax of Nonsententials. Multidisciplinary perspectives. 2006. x, 372 pp. 92 Boeckx, Cedric (ed.): Agreement Systems. 2006. ix, 346 pp. 91 Boeckx, Cedric (ed.): Minimalist Essays. 2006. xvi, 399 pp. 90 Dalmi, Gréte: The Role of Agreement in Non-Finite Predication. 2005. xvi, 222 pp. 89 Velde, John R. te: Deriving Coordinate Symmetries. A phase-based approach integrating Select, Merge, Copy and Match. 2006. x, 385 pp. 88 Mohr, Sabine: Clausal Architecture and Subject Positions. Impersonal constructions in the Germanic languages. 2005. viii, 207 pp. 87 Julien, Marit: Nominal Phrases from a Scandinavian Perspective. 2005. xvi, 348 pp. 86 Costa, João and Maria Cristina Figueiredo Silva (eds.): Studies on Agreement. 2006. vi, 285 pp. 85 Mikkelsen, Line: Copular Clauses. Specification, predication and equation. 2005. viii, 210 pp. 84 Pafel, Jürgen: Quantifier Scope in German. 2006. xvi, 312 pp. 83 Schweikert, Walter: The Order of Prepositional Phrases in the Structure of the Clause. 2005. xii, 338 pp. 82 Quinn, Heidi: The Distribution of Pronoun Case Forms in English. 2005. xii, 409 pp. 81 Fuss, Eric: The Rise of Agreement. A formal approach to the syntax and grammaticalization of verbal inflection. 2005. xii, 336 pp. 80 Burkhardt, Petra: The Syntax–Discourse Interface. Representing and interpreting dependency. 2005. xii, 259 pp. 79 Schmid, Tanja: Infinitival Syntax. Infinitivus Pro Participio as a repair strategy. 2005. xiv, 251 pp. 78 Dikken, Marcel den and Christina M. Tortora (eds.): The Function of Function Words and Functional Categories. 2005. vii, 292 pp. 77 Öztürk, Balkız: Case, Referentiality and Phrase Structure. 2005. x, 268 pp. 76 Stavrou, Melita and Arhonto Terzi (eds.): Advances in Greek Generative Syntax. In honor of Dimitra Theophanopoulou-Kontou. 2005. viii, 366 pp. 75 Di Sciullo, Anna Maria (ed.): UG and External Systems. Language, brain and computation. 2005. xviii, 398 pp. 74 Heggie, Lorie and Francisco Ordóñez (eds.): Clitic and Affix Combinations. Theoretical perspectives. 2005. viii, 390 pp. 73 Carnie, Andrew, Heidi Harley and Sheila Ann Dooley (eds.): Verb First. On the syntax of verbinitial languages. 2005. xiv, 434 pp. 72 Fuss, Eric and Carola Trips (eds.): Diachronic Clues to Synchronic Grammar. 2004. viii, 228 pp. 71 Gelderen, Elly van: Grammaticalization as Economy. 2004. xvi, 320 pp. 70 Austin, Jennifer R., Stefan Engelberg and Gisa Rauh (eds.): Adverbials. The interplay between meaning, context, and syntactic structure. 2004. x, 346 pp. 69 Kiss, Katalin É. and Henk van Riemsdijk (eds.): Verb Clusters. A study of Hungarian, German and Dutch. 2004. vi, 514 pp.

68 Breul, Carsten: Focus Structure in Generative Grammar. An integrated syntactic, semantic and intonational approach. 2004. x, 432 pp. 67 Mišeska Tomić, Olga (ed.): Balkan Syntax and Semantics. 2004. xvi, 499 pp. 66 Grohmann, Kleanthes K.: Prolific Domains. On the Anti-Locality of movement dependencies. 2003. xvi, 372 pp. 65 Manninen, Satu Helena: Small Phrase Layers. A study of Finnish Manner Adverbials. 2003. xii, 275 pp. 64 Boeckx, Cedric and Kleanthes K. Grohmann (eds.): Multiple Wh-Fronting. 2003. x, 292 pp. 63 Boeckx, Cedric: Islands and Chains. Resumption as stranding. 2003. xii, 224 pp. 62 Carnie, Andrew, Heidi Harley and MaryAnn Willie (eds.): Formal Approaches to Function in Grammar. In honor of Eloise Jelinek. 2003. xii, 378 pp. 61 Schwabe, Kerstin and Susanne Winkler (eds.): The Interfaces. Deriving and interpreting omitted structures. 2003. vi, 403 pp. 60 Trips, Carola: From OV to VO in Early Middle English. 2002. xiv, 359 pp. 59 Dehé, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. xii, 305 pp. 58 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. 2003. vi, 309 pp. 57 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. 2003. vi, 405 pp. 56 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. 2003. x, 295 pp. 55 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 1: The syntax and semantics of noun phrases. 2003. vi, 362 pp. 54 Baptista, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2003. xxii, 294 pp. (incl. CD-rom). 53 Zwart, C. Jan-Wouter and Werner Abraham (eds.): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26–27, 2000). 2002. xiv, 407 pp. 52 Simon, Horst J. and Heike Wiese (eds.): Pronouns – Grammar and Representation. 2002. xii, 294 pp. 51 Gerlach, Birgit: Clitics between Syntax and Lexicon. 2002. xii, 282 pp. 50 Steinbach, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. xii, 340 pp. 49 Alexiadou, Artemis (ed.): Theoretical Approaches to Universals. 2002. viii, 319 pp. 48 Alexiadou, Artemis, Elena Anagnostopoulou, Sjef Barbiers and Hans-Martin Gärtner (eds.): Dimensions of Movement. From features to remnants. 2002. vi, 345 pp. 47 Barbiers, Sjef, Frits Beukema and Wim van der Wurff (eds.): Modality and its Interaction with the Verbal System. 2002. x, 290 pp. 46 Panagiotidis, Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. x, 214 pp. 45 Abraham, Werner and C. Jan-Wouter Zwart (eds.): Issues in Formal German(ic) Typology. 2002. xviii, 336 pp. 44 Taylan, Eser Erguvanlı (ed.): The Verb in Turkish. 2002. xviii, 267 pp. 43 Featherston, Sam: Empty Categories in Sentence Processing. 2001. xvi, 279 pp. 42 Alexiadou, Artemis: Functional Structure in Nominals. Nominalization and ergativity. 2001. x, 233 pp. 41 Zeller, Jochen: Particle Verbs and Local Domains. 2001. xii, 325 pp. 40 Hoeksema, Jack, Hotze Rullmann, Víctor Sánchez-Valencia and Ton van der Wouden (eds.): Perspectives on Negation and Polarity Items. 2001. xii, 368 pp. 39 Gelderen, Elly van: A History of English Reflexive Pronouns. Person, Self, and Interpretability. 2000. xiv, 279 pp. 38 Meinunger, André: Syntactic Aspects of Topic and Comment. 2000. xii, 247 pp. 37 Lutz, Uli, Gereon Müller and Arnim von Stechow (eds.): Wh-Scope Marking. 2000. vi, 483 pp. 36 Gerlach, Birgit and Janet Grijzenhout (eds.): Clitics in Phonology, Morphology and Syntax. 2001. xii, 441 pp. 35 Hróarsdóttir, Thorbjörg: Word Order Change in Icelandic. From OV to VO. 2001. xiv, 385 pp. 34 Reuland, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. xii, 255 pp.

33 Puskás, Genoveva: Word Order in Hungarian. The syntax of Ā-positions. 2000. xvi, 398 pp. 32 Alexiadou, Artemis, Paul Law, André Meinunger and Chris Wilder (eds.): The Syntax of Relative Clauses. 2000. vi, 397 pp. 31 Svenonius, Peter (ed.): The Derivation of VO and OV. 2000. vi, 372 pp. 30 Beukema, Frits and Marcel den Dikken (eds.): Clitic Phenomena in European Languages. 2000. x, 324 pp. 29 Miyamoto, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 2000. xiv, 232 pp. 28 Hermans, Ben and Marc van Oostendorp (eds.): The Derivational Residue in Phonological Optimality Theory. 2000. viii, 322 pp. 27 Růžička, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. x, 206 pp. 26 Ackema, Peter: Issues in Morphosyntax. 1999. viii, 310 pp. 25 Felser, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999. xiv, 278 pp. 24 Rebuschi, Georges and Laurice Tuller (eds.): The Grammar of Focus. 1999. vi, 366 pp. 23 Giannakidou, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. xvi, 282 pp. 22 Alexiadou, Artemis and Chris Wilder (eds.): Possessors, Predicates and Movement in the Determiner Phrase. 1998. vi, 388 pp. 21 Klein, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. x, 232 pp. 20 Laenzlinger, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. x, 371 pp. 19 Josefsson, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. ix, 199 pp. 18 Alexiadou, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. x, 256 pp. 17 Beermann, Dorothee A., David LeBlanc and Henk van Riemsdijk (eds.): Rightward Movement. 1997. vi, 410 pp. 16 Liu, Feng-Hsi: Scope and Specificity. 1997. viii, 187 pp. 15 Rohrbacher, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and prodrop. 1999. viii, 296 pp. 14 Anagnostopoulou, Elena, Henk van Riemsdijk and Frans Zwarts (eds.): Materials on Left Dislocation. 1997. viii, 349 pp. 13 Alexiadou, Artemis and T. Alan Hall (eds.): Studies on Universal Grammar and Typological Variation. 1997. viii, 252 pp. 12 Abraham, Werner, Samuel David Epstein, Höskuldur Thráinsson and C. Jan-Wouter Zwart (eds.): Minimal Ideas. Syntactic studies in the minimalist framework. 1996. xii, 364 pp. 11 Lutz, Uli and Jürgen Pafel (eds.): On Extraction and Extraposition in German. 1996. xii, 315 pp. 10 Cinque, Guglielmo and Giuliana Giusti (eds.): Advances in Roumanian Linguistics. 1995. xi, 172 pp. 9 Gelderen, Elly van: The Rise of Functional Categories. 1993. x, 224 pp. 8 Fanselow, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. xvii, 232 pp. 7 Åfarlí, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. xii, 177 pp. 6 Bhatt, Christa, Elisabeth Löbel and Claudia Maria Schmidt (eds.): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. ix, 187 pp. 5 Grewendorf, Günther and Wolfgang Sternefeld (eds.): Scrambling and Barriers. 1990. vi, 442 pp. 4 Abraham, Werner and Sjaak De Meij (eds.): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. v, 349 pp. 3 Abraham, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. vi, 242 pp. 2 Ehlich, Konrad and Jürgen Rehbein: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. viii, 150 pp. With many photographic ills. 1 Klappenbach, Ruth (1911–1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene MaligeKlappenbach. (Written in German). 1980. xxiii, 313 pp.

Demoting the Agent: Passive, middle and other voice phenomena

Demoting the Agent: Passive, middle and other voice phenomena (Linguistik Aktuell, V. 96)

Aharonov-Bohm and other cyclic phenomena

Plastic Bag Trees and Other Cyprus Phenomena

Aharonov-Bohm and Other Cyclic Phenomena

Widow's Mite and Other Psychic Phenomena

Agent of Vega and Other Stories

The Vision and the Voice with Commentary and Other Papers

Letters and Orations (The Other Voice in Early Modern Europe)

Letters and Orations (The Other Voice in Early Modern Europe)

Dialogues and Addresses (The Other Voice in Early Modern Europe)

Dialogues and Addresses (The Other Voice in Early Modern Europe)

Apology for the Woman Writing and Other Works (The Other Voice in Early Modern Europe)

Agent of Vega & Other Stories

Agent of Vega & Other Stories

Agent of Vega & Other Stories

Agent of Vega & Other Stories

The Voice

The Vision and the Voice

The Voice

Agent

Activating the Passive Church: Diagnosis and Treatment

The Voice

The Court Midwife (The Other Voice in Early Modern Europe)

The Court Midwife (The Other Voice in Early Modern Europe)

The Voice

The Voice

The Voice

The Voice

The Voice

Agent

Demoting the Agent: Passive, middle and other voice phenomena