Split Auxiliary Systems
Typological Studies in Language (TSL) A companion series to the journal Studies in Language...
53 downloads
666 Views
3MB 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
Split Auxiliary Systems
Typological Studies in Language (TSL) A companion series to the journal Studies in Language
General Editor
Michael Noonan
Assistant Editors
Spike Gildea, Suzanne Kemmer
Editorial Board Wallace Chafe (Santa Barbara) Bernard Comrie (Leipzig) R. M. W. Dixon (Melbourne) Matthew Dryer (Buffalo) John Haiman (St Paul) Bernd Heine (Köln) Paul Hopper (Pittsburgh) Andrej Kibrik (Moscow) Ronald Langacker (San Diego)
Charles Li (Santa Barbara) Edith Moravcsik (Milwaukee) Andrew Pawley (Canberra) Doris Payne (Eugene, OR) Frans Plank (Konstanz) Jerrold Sadock (Chicago) Dan Slobin (Berkeley) Sandra Thompson (Santa Barbara)
Volumes in this series will be functionally and typologically oriented, covering specific topics in language by collecting together data from a wide variety of languages and language typologies. The orientation of the volumes will be substantive rather than formal, with the aim of investigating universals of human language via as broadly defined a data base as possible, leaning toward cross-linguistic, diachronic, developmental and live-discourse data.
Volume 69 Split Auxiliary Systems: A cross-linguistic perspective Edited by Raúl Aranovich
Split Auxiliary Systems A cross-linguistic perspective
Edited by
Raúl Aranovich University of California, Davis
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 Split auxiliary systems : a cross-linguistic perspective / edited by Raúl Aranovich. p. cm. (Typological Studies in Language, issn 0167–7373 ; v. 69) Includes bibliographical references and indexes. 1. Grammar, Comparative and general--Auxiliaries--Congresses. 2. Language and languages--Variation--Congresses. P299.A95 S686 2006 415--dc22 isbn 978 90 272 2981 6 (Hb; alk. paper)
2006051684
© 2007 – 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
Foreword Split auxiliary selection from a cross-linguistic perspective Raúl Aranovich
vii 1
Transitivity parameters and auxiliary selection by L2 students of German Carlee Arnett
25
An irrealis BE auxiliary in Romanian Larisa Avram and Virginia Hill
47
Auxiliary selection and split intransitivity in Paduan: Variation and lexical-aspectual constraints Michela Cennamo and Antonella Sorace
65
The development of the HAVE perfect: Mutual influences of Greek and Latin Bridget Drinka
101
Agentivity versus auxiliary choice: Evidence from pronominal binding in German AcI-constructions Vera Lee-Schoenfeld
123
Optimizing auxiliary selection in Romance Géraldine Legendre
145
Auxiliary selection in Chinese Feng-hsi Liu
181
Parameterized auxiliary selection: A fine-grained interaction of features and linking rules Janet H. Randall Particle selection in Korean auxiliary formation Seongha Rhee
207 237
Table of contents
Language use and auxiliary selection in the perfect K. Aaron Smith
255
Language index
271
Name index
273
Subject index
275
Foreword
This collection of essays grew out of a selected set of contributions to a workshop on ‘Cross-linguistic Variation in Auxiliary Selection’. The workshop was held in 2003 at the University of California Davis, between May 31 and June 1. As one of the participants noticed, the meeting celebrated 25 years of David Perlmutter’s original proposal of the ‘unaccusativity hypothesis’ at a meeting of the Berkeley Linguistic Society. Since then, a growing body of research continues to uncover different phenomena related to split intransitivity, but as the organizer of the workshop I considered it was time to review the current state of research on split auxiliary systems, which constitute key empirical evidence in Perlmutter’s proposal. The original set of workshop presentations was expanded with solicited contributions, resulting in a cohesive overview of auxiliary selection across languages. Many people collaborated in making the workshop a success, and also in the review of the drafts of the submitted papers. My appreciation goes to my colleagues and students in the linguistics department at UC Davis, to the participants in the workshop, and to Delia Bentley, Giulia Centineo, Stanley Dubinsky, Raffaella Folli, Nunzio La Fauci, Namgeun Lee, Alessandro Lenci, Jean-Marie Marandin, Carol Rosen, Gillian Sankoff, Elly Van Gelderen, Martina Wiltschko, and Annie Zaenen.
Split auxiliary selection from a cross-linguistic perspective Raúl Aranovich University of California, Davis
.
Auxiliary selection and split intransitivity
The auxiliaries BE and HAVE are used to form the perfect tense in a variety of languages, as the Italian and Dutch examples below illustrate. The phenomenon has been extensively researched in French and German as well, and it can be observed in Icelandic, Danish, and (among non-Indo-European languages) Basque. (1) a.
Maria ha dormito molto ieri Maria has slept much yesterday ‘Maria slept a lot yesterday.’ b. Maria è caduta dal letto. Maria is fallen of-the bed ‘Maria fell off the bed.’
Italian
(2) a.
Dutch
De jongen heeft getelefoneerd. ‘The boy has phoned.’ b. In dat ziekenhuis zijn (er) veel patienten gestorven. In that hospital are (there) many patients died ‘In that hospital died many patients.’
In ‘split auxiliary’ systems, one of the auxiliaries, HAVE, occurs with transitive verbs, as in (3). The distribution of BE with a sub-set of the intransitive verbs is to a large extent lexically determined. Table 1 samples some Italian intransitive verbs used with essere ‘be’ and avere ‘have’. (3) Ugo ha difeso Gianni. Ugo has defended Gianni ‘Ugo defended Gianni.’
Raúl Aranovich
Table 1. Intransitives and aux. selection in Italian with essere è caduto è partito è sparito è rimasto è scoppiato è morto è arrossito
‘fell’ ‘left’ ‘disappeared’ ‘remained’ ‘exploded’ ‘died’ ‘blushed’
with avere ha sorriso ha mentito ha tremato ha sanguinato ha schiumato ha nuotato ha peccato
‘smiled’ ‘lied’ ‘trembled’ ‘bled’ ‘foamed’ ‘swam’ ‘sinned’
Accounts of split auxiliary systems can be roughly divided into those that consider the distinction between verbs that select HAVE (A verbs) or BE (E verbs) to be a matter of syntax, and those that see it as a matter of semantics. But regardless of theoretical orientation, any account of auxiliary selection has to address the question of whether the two verb classes can be characterized in a uniform way. Three empirical challenges to this notion are: a) other tests that divide intransitive verbs into two classes yield results that are only partially congruent with the classes of A verbs and E verbs; b) many verbs show ‘variable behavior’, appearing with BE or HAVE; c) verbs with similar meaning may be classified as A verbs in one language, but as E verbs in another. In this introduction, I will summarize the main theories about split intransitivity, and how they address these problems. Since Perlmutter (1978) first pointed out the importance of split auxiliary selection phenomena for modern theoretical linguistics, careful research has shown variation in auxiliary selection to be widespread, and has argued for new directions in the analysis of auxiliary systems. This is the common thread that runs through the contributions to this volume. They are summarized in the last section of this introduction.
. The Unaccusative Hypothesis and split auxiliary selection Syntactic analyses of split auxiliary selection are based on evidence that the subject of E verbs shares properties with direct objects of transitives, while the subject of A verbs shares them with the subject of transitives (see Section 4 for details). The formal expression of this observation is the Unaccusative Hypothesis (Perlmutter 1978, 1989; Hoekstra 1984; Burzio 1986), according to which unaccusative clauses have an underlying object but no underlying subject. Independent principles of the grammar insure that this underlying object is the final, or surface, subject of
Split auxiliary selection from a cross-linguistic perspective IP NP Maria
P P Sterven
2 1 Patienten
a. Relational network for unaccusative
I’
I é
VP
case
V’ V caduta
t
th-role
b. Unaccusative in GB
Figure 1.
the clause. The UH, originally formulated in Relational Grammar (RG),1 assigns an unaccusative clause like (2b) the representation in (1a). A syntactic analysis of an unaccusative clause in the Government and Binding framework is shown in (1b). Lexical properties of the verbs determine whether the subject is an underlying object. ‘Unergative’ verbs project their argument ‘externally’, i.e. directly onto the subject, while ‘unaccusative’ verbs project their single argument ‘internally’, inside the VP. The UH makes it possible to characterize the distribution of the auxiliaries in syntactic terms: BE is selected if the subject is also an object at some abstract level of representation. The fact that verbs with external theta-roles select HAVE is explained in Hoekstra (1994) as a result of the fact that HAVE differs from BE in its ability to assign accusative case. When HAVE combines with a transitive participle, the direct object receives its case not from the participle but from the auxiliary verb. BE lacks this case assigning property, so in conjunction with a transitive participle it can only yield a passive construction. If the participle is unaccusative, BE must be selected so that the internal argument rises to subject position, where it is assigned nominative case. With the auxiliary HAVE, on the other hand, the internal argument would receive accusative case, and it could not move to sub. Relational Grammar is a multi-stratal theory of syntax that takes grammatical relations as primitives. Each sentence is associated with a network of arcs and strata. At each stratum, arcs are labeled with the symbols P (for predicate), 1 (for subject), 2 (for object), 3 (for indirect object) or 4 (for oblique). Arcs that have different labels at different strata indicate changes in grammatical relations.
Raúl Aranovich
ject position. With unergative participles, finally, HAVE is selected simply as the default, since there is no internal argument to worry about.2 A question that emerges regarding unergative/unaccusative splits is whether the projection of a verb’s arguments onto the initial or underlying level of syntactic representation is arbitrary or semantically determined. Perlmutter (1978) argues for the latter option. According to the Universal Alignment Hypothesis (Perlmutter & Postal 1984), agents are always initial subjects (external arguments, in GB), while patients and themes are initial objects (internal arguments). The position that unaccusativity is semantically determined but syntactically encoded enjoys wide support (cf. Levin & Rappaport 1995; Sorace 2000), and is backed by arguments involving other ‘unaccusativity tests’ (see Section 4). But against the Universal Alignment Hypothesis, some claim that unaccusativity is arbitrary, without any semantic basis (Rosen 1984; Perlmutter 1989). Evidence for rejecting any universal alignment between semantic roles and grammatical relations comes from comparing the results of unaccusativity tests in several languages. With respect to auxiliary selection, Rosen notices that the claim that patients or themes are linked to initial objects is inconsistent with the behavior of verbs like cadere ‘fall’, which selects essere whether or not its subject is animate or inanimate, or fischiare ‘whistle’ which selects avere even if the subject is inanimate (v.g. a tea kettle). Many non-volitional predicates in Italian also select avere, including many verbs of emission (scintillare ‘sparkle’, cigolare ‘creak’, puzzare ‘stink’, etc.). Rosen’s position is also supported by the fact that auxiliary selection varies across languages, which I will discuss in Section 7.
. Semantics of auxiliary selection In spite of Rosen’s objections, the strength of the appeal of a semantic distinction between A verbs and E verbs has lead to a number of refinements and extensions of the UAH. A shortcoming of the UAH is that it relies on thematic role types to make distinctions in argument structure. Dowty (1991) argues that a predicate’s . Hoekstra’s analysis is based on Kayne’s (1993) observation that the use of BE and HAVE as main verbs in possessive constructions is related to their use as perfect auxiliaries. In Latin, for instance, the sentence ‘I have a book’ is Mihi liber est ‘I.dat book.nom is’. Kayne’s proposal is that the dative case of the Latin-type possessive construction is assigned by a covert preposition, which incorporates to the verb in English-type languages. The copula BE plus the incorporated preposition is spelled out as HAVE. Kayne suggests that the same is the case for the alternation between auxiliary BE and HAVE. However, Hoekstra (1994) notices that Kayne’s approach does not explain how the possessee receives accusative case in English. He argues that what distinguishes possessive HAVE from BE is precisely its ability to assign accusative, and this is the same property that characterizes auxiliary HAVE as opposed to auxiliary BE.
Split auxiliary selection from a cross-linguistic perspective
argument structure should be understood in terms of prototypical roles instead, which are defined as sets of entailments. The prototypical Agent (or Proto-Agent) is sentient, volitional, stationary, causally effective, etc. A prototypical Patient (or Proto-Patient) is inanimate, affected by the event, undergoes a change of state or location, and ‘measures out’ the event (i.e. is an incremental theme). These two protoypical roles correspond to the subjects of the E verbs and the A verbs. HAVE is selected if the subject has mostly proto-agent properties, and BE is selected otherwise. In Dutch, Dowty argues, zijn is selected if the subject has the the key proto-patient property of being an incremental theme. The idea that splits in auxiliary selection reflect the existence of two opposite semantic prototypes is also developed in Shannon’s (1990) functional-cognitive approach. Prototypical ‘transitive’ events involve two differentiated entities, in which one moves towards and makes contact with the other one causing it to change state or position. Prototypical ‘mutative’ events involve a single entity which changes state or position. Prototypical mutative events select zijn/sein, while prototypical transitive verbs select hebben/haven. Dowty’s proto-role analysis, as well as the semantic accounts based on the Universal Alignment Hypothesis (or versions thereof), provide ‘thematic’ criteria for auxiliary selection based on the semantic features of the arguments of A verbs and E verbs. A different approach is to specify ‘aspectual’ criteria for auxiliary selection, as in Van Valin (1990) Role and Reference Grammar (RRG) analysis of the essere/avere alternation in Italian. Van Valin decomposes verbal lexical meanings into primitive logical operators, predicates that are further classified into states (see’, broken’, be-at’, etc.) and activities (eat’, shout’, do’), and argument variables. States and Activities have simple structures, consisting of a state or activity predicate with its arguments. Achievements consist of a state as the argument of the operator BECOME, and accomplishments consist of an activity joined to an achievement by a CAUSE operator, as in (4). Essere is selected in Italian, Van Valin argues, when the subject is an argument of a state predicate. (4) andare ‘go’: [do’(x)] CAUSE [BECOME be-at’(x,y)]
Similar analyses based on lexical decomposition can be found in Levin and Rappaport (1995), and in Lieber and Baayen (1997). Lieber and Baayen assign semantic features to the logical predicates (or semantic functions, in their terms). Features like [+/– dynamic], for instance, determine whether an eventuality is an Event or a State. Lieber and Baayen argue that the distinction between E verbs and A verbs in Dutch is determined by the feature [+/– IEPS], for ‘Inferable Eventual Position or State’. This feature says that speakers must be able to know something about the resulting state or position of the predicate’s highest argument. Their rule for auxiliary selection is that only verbs that are [+dynamic, +IEPS] (change of position
Raúl Aranovich
verbs like komen ‘come’, and change of state verbs like groeien ‘grow’) can select zijn ‘be’.3
. Other unaccusative tests Semantic analyses of auxiliary selection are central to monostratal theories of grammar, such as RRG or the cognitive-functional approach, which deny that there exist abstract levels of syntactic representation.4 Proponents of the Unaccusative Hypothesis, on the other hand, contend that split auxiliary systems offer compelling evidence for a theory of syntax that recognizes initial or underlying grammatical relations, even if semantics plays a role in determining whether an argument is internal or external.5 One argument for a syntactic analysis of auxiliary selection is that it captures generalizations between the classes of A-verbs and E-verbs and other syntactic phenomena which, by their nature, call for a structural explanation. Three ‘unaccusativity tests’ that are often mentioned in connection with auxiliary selection are the distribution of partitive clitics in Italian (and French), impersonal passives in Dutch, and reflexive clitics in Romance. Italian has a partitive clitic ne that can cross-reference the subject of an E verb (5a), but not the subject of an A verb (5b). The fact that ne can cross-reference the object of a transitive verb, but not its subject, is evidence in favor of the claim that the subjects of E verbs are underlying objects (Burzio 1986; Perlmutter 1989; Rosen 1988),6 and that the conditions on the distribution of essere and avere have to be stated in syntactic terms. (5) a.
Ne arrivano molti. part arrive many Many of them arrive.
Italian
. In fact, Lieber and Baayen’s [+IEPS] feature is very similar in content to a traditional notion of ‘mutative’ verb, as pointed out in Hoekstra (1999). . This is not to say that all monostratal theories must adopt a semantic analysis of split intransitivity. Within Head-Driven Phrase Structure Grammar, for instance, it is frequent to find analyses of ‘unaccusative phenomena’ (such as auxiliary selection in German) that distinguish E verbs from A verbs by means of a feature such as [ergative]. See Kathol (1994) for a discussion of this approach. . Some researchers who espouse the idea of a semantic basis of split intransitivity, I should remark, remain uncommitted about the syntactic differences between E verbs and A verbs. This is the position adopted by Dowty (1991) and by Lieber and Baayen (1997). . See Guéron (1987) for a similar assessment of the French partitive clitic en.
Split auxiliary selection from a cross-linguistic perspective
b. *Ne studiano molti. part study many Many of them study.
Dutch impersonal passives offer similar evidence. In Dutch, A verbs like telefoneren ‘phone’ can form impersonal passives (6a), while E verbs like sterven ‘die’ can’t (6b) (Perlmutter 1978; Zaenen 1993). Assuming that only verbs with external arguments can form impersonal passives, the conclusion is that zijn is selected if a clause is unaccusative.7 (6) a.
Er werd (door een jongen) getelefoneerd. Dutch There was (by a boy) phoned. ‘A boy phoned.’ b. *In dat ziekenhuis werd er (door veel patienten) gestorven. ‘In that hospital was there (by many patients) died’ ‘Many patients died in that hospital.’
More evidence for a syntactic analysis of auxiliary selection comes from reflexive verbs in languages like Italian and French. In these languages reflexive verbs (i.e. languages with reflexive clitics) select essere/être as the perfect auxiliary. (7) a.
Ugo si è difeso. Ugo refl is defended ‘Ugo defended himself ’ b. De telles choses ne se sont jamais vues. of such things not refl are never seen ‘Things like those are never seen.’
Italian
French
The argument rests on the hypothesis that the subject of a reflexive verbs is an object at some abstract syntactic level. The verbs in (7a)–(7b), after all, are basically transitive, and the subject carries the thematic role that would normally project the complement. This is also apparent in the ‘anticausative’ use of the reflexive clitic, illustrated below: (8) a.
Ils se sont tus pour protéger son camarade. They refl are shut-up for protect their friend ‘They kept quiet to protect their friend.’ b. Ugo si è offeso. Ugo refl is offended ‘Ugo took offense.’
French
Italian
. In Relational Grammar, impersonal passives are the result of the advancement of a ‘dummy’ object to subject. Unaccusative clauses also have an object to subject advancement. Impersonal passivization of an unaccusative clause violates a law permitting only one such advancement (the 1AX law).
Raúl Aranovich
Grimshaw (1990) analyzes reflexive clitics as morpho-syntactic exponents of a lexical rule that binds the two theta-roles of a predicate, effectively ‘absorbing’ the external one. The internal theta-role is projected onto the object position, licensing an NP that must move to subject position to receive nominative case, as with other unaccusative verbs.8 (9) offendere + si:
Assuming this analysis, then, the UH allows for a general statement of the conditions on auxiliary selection in intransitives and reflexives: être/essere is selected if the subject of the clause is also an underlying object. Indirect reflexives pose a potential problem to the unaccusative hypothesis of auxiliary selection, as stated in the GB framework (Grimshaw 1990). In clauses like (10), the reflexive relation holds of the subject and an indirect object. (10) Gianni si è comprato un auto. Gianni refl is bought a car ‘Gianni bought himself a car.’
Italian
The presence of the reflexive clitic requires essere as the perfect auxiliary, even if the clause has an overt direct object. A clause like (10), then, is an exception to the general rule that the BE auxiliary only appears with intransitive predicates. Another exception to this claim is represented by some Dutch transitive verbs that select BE, such as volgen ‘follow’ and naderen ‘approach’, as discussed in Lieber & Baayen (1997). The trouble with indirect reflexives is that after the reflexive clitic absorbs the external theta-role, lexically binding it to the goal or benefactive, the direct object could climb to subject position. This, however, never happens. Likewise, suppose the Dutch verbs have two internal theta-roles, as suggested in Hoekstra (1994). Why doesn’t the direct object become the subject? A possible answer to this riddle, as argued in Cocchi (1994), is that these constructions include an ‘applicative’ structure, in which the indirect object is first promoted to direct object, and from there it becomes the subject. What appears to be a surface object is actually an oblique. In this analysis, then, sentences like (10) are not exceptions to the claim that BE appears only with intransitives.
. The RG analysis of reflexive clitics involves ‘multiattachment’: the surface subject of the reflexive clause bears two grammatical relations at once (i.e. subject and object) in some stratum (Rosen 1988; Perlmutter 1989). The reflexive clitic cancels the object grammatical relations, resulting in an intransitive stratum. Anticausatives involve ‘retroherent’ advancement: before multiattachment the clause has an unaccusative stratum. For other accounts in which the reflexive clitic absorbs an internal theta-role see Burzio (1986) and Grimshaw (1982).
Split auxiliary selection from a cross-linguistic perspective
. Mismatches among tests The bulk of evidence presented in the studies cited so far points out to the existence of two classes of intransitive verbs, syntactically defined. To account for some of these facts without assuming underlying grammatical relations, semantic theories of split intransitivity argue that various phenomena characterize two classes of intransitive verbs because of their meaning. Van Valin (1990) claims that Italian reflexives select essere not because of a syntactic representation in which the subject is an underlying object, but because of the lexical-semantic properties of reflexive verbs. In a reflexive verb, the subject is at once the argument of the causing event and the resulting state, as in (11). Essere is selected with reflexives, then, because the subject is an argument of a state predicate.9 (11) offendere + si ‘offend + self ’: [do’(x)] CAUSE [BECOME offended’ (x)].
Critics of syntactic, multi-stratal approaches to auxiliary selection often contend that the different phenomena used to distinguish unergatives from unaccusatives often fail to assign all verbs to the same classes. The ne-cliticization argument, for instance, is undermined by the existence of some A verbs that do allow their subjects to cliticize as a partitive. An example is given below, with the verb attechire ‘take root’. Other verbs with this property are funzionare ‘function, work’, oscillare ‘oscillate’, trillare ‘ring’, telefonare ‘telephone’, correre ‘run’, starnutiscere ‘sneeze’, camminare ‘walk’. (Lonzi 1986). (12) Ne attechirono pochi (di bulbi). part take-root few (of bulbs) ‘few of them (bulbs) took root.’
Italian
Likewise, in Dutch, the match between predicates that select hebben ‘have’ and the possibility to undergo impersonal passivization does not hold. There are some intransitive verbs – noncontrollable atelic events like stinken ‘stink’, for instance – that select hebben ‘have’ but cannot form impersonal passives. Moreover, telic (motional) predicates select zijn, even if the predicate is atelic (as a result of an adverbial phrase, for instance). But in that situation it is possible to have an impersonal passive There are exceptions to the claim that in languages with reflexive clitics the BE auxiliary always combines with reflexive verbs. In Old Spanish, for instance, a split . Centineo (1986 [1996]) argues that subjects of reflexives bear an UNDERGOER macro-role (the two macro-roles ACTOR and UNDERGOER in RRG mediate between argument structure and syntactic structure). Undergoers are normally projected as objects, so their realization as subjects is marked. For Centineo, essere is selected whenever such a marked subject occurs in an Italian clause.
Raúl Aranovich
auxiliary system is found up to the seventeenth century (Benzing 1931; Aranovich 2003). Old Spanish, like Modern Spanish, has reflexive clitics, but in the perfect some of them take ser ‘be’ (13a), and others take haber ‘have’ (13b). (13) a.
Estonçe Rruy Diaz apriessa se fue levantado. Old Sp. Then Ruy Diaz quickly refl was risen ‘And then Ruy Diaz got up in a hurry.’ b. el buen rey se habia apartado con voluntad de facer lo The good king refl have separated with will of doing the que a nadie es excusado. what to nobody is allowed ‘The good king went away to do what is not permitted of anybody.’
Mismatches between the results of unaccusativity tests and the classes of A verbs and E verbs pose a problem for syntactic theories of auxiliary selection. They seem to require that a verb be represented with an internal argument for some tests, but with an external argument for others. One strategy is to uphold the syntactic distinction between unergatives and unaccusatives, and take some tests as better indication of the split than others. Legendre (1989), for instance, takes this position when she argues that selection of être in French is a sufficient, not a necessary, condition for a verb to be classified as unaccusative. The alternative is to abandon the idea of a syntactic characterization of split intransitivity, and say that all distinctions are semantically constrained. From this point of view, all mismatches are semantic: each tests may be sensitive to different semantic factors. Impersonal passives in Dutch, for instance, are possible if the verb is volitionally controlled, while zijn is selected in the predicate is telic (Zaenen 1993; Dowty 1991). This is why the two tests fail to identify non-volitional atelic verbs like stinken ‘stink’ as unergative or unaccusative. This is a problem that syntactic theories of auxiliary selection (and split intransitivity in general) must address.
. Verbs of variable behavior Another issue over which syntactic and semantic analyses confront each other is the existence of verbs of variable behavior, i.e. verbs that select BE or HAVE. One type of alternation involves adverbial modifiers that change the aspectual class of the predicate. Verbs of non-directed motion, for instance, which normally select HAVE, are found with BE if modified by an endpoint adverbial phrase, as shown in the Dutch examples below. Proponents of semantic theories of auxiliary selection (Zaenen 1993; Dowty 1990) offer this as evidence that zijn is selected by telic predicates. Similar facts are reported for Icelandic (Yamaguchi & Pétursson 2003).
Split auxiliary selection from a cross-linguistic perspective
(14) a.
Hij heeft/*is gelopen He has/is run ‘He has run.’ b. Hij is/?heeft naar huis gelopen He has/is towards home run ‘He has run home.’
Dutch
In Italian as well, endpoint adverbials trigger selection of essere. Moreover, some verbs of emission (e.g. suonare ‘toll’), which normally select avere, can vary in their auxiliary selecting properties depending on the telicity of the sentence (Centineo 1986 [1996]). On the other hand, adverbial phrases can force verbs that normally select BE to combine with HAVE, when they turn the predicate into an atelic one. This can be seen with verbs of indefinite change of state, like Italian fiorire ‘bloom’ and salire ‘rise (temperature), or Dutch stijgen, ‘rise (temperature)’.10 (15) De temperatuur heeft 3 uurlang gestegen, maar is toen The temperature has three hour.long risen, but is then weer gezakt. dropped ‘The temperature rose for three hours, but then dropped again.’
Besides aspectual features, variation in auxiliary selection seems to be sensitive to thematic properties of the clause as well. Some verbs of change of location, which normally select HAVE (16a), occur with BE if the subject is inanimate (16b). On the other hand, verbs of continuation of state (e.g. durare ‘last’) go from BE to HAVE when the subject is animate (Sorace 2000). Il pilota ha/?è atterrato sulla pista di emergenza. The pilot has/is landed on-the runway of emergency ‘The pilot landed on the emergency runway.’ di emergenza. b. L’aereo è/?ha atterrato sulla pista The plane is/has landed on-the runway of emergency ‘The plane landed on the emergency runway.’
(16) a.
Italian
There is some controversy regarding the ability of thematic factors to trigger variation in auxiliary selection in the Germanic languages, but Shannon (1990) has argued with examples like (17) that this is indeed the case (einbrechen selects sein with inanimate subjects).
. Verbs of directed change of state or location do not change to HAVE in the same environment (Zaenen 1993). In German, verbs of change always take sein, regardless of telicity. So, it is not surprising that an example equivalent to (15) in German still selects sein (Keller & Sorace 2003: 66).
Raúl Aranovich
(17) Der Dieb hat eingebrochen. The thief has broken-into ‘The thief broke in.’
German
The existence of verbs of variable behavior throws into question the idea that auxiliary selection is lexically determined only. Aspectual features of the predicate (endpoint adverbials) and semantic features of the subject (volitionality, control) also influence the choice of BE or HAVE. To address this problem, theories of auxiliary selection in which the verb determines if a clause is unergative or unaccusative have to assign some verbs to more than one class, resulting in an unwelcome multiplication of lexical entries. Proponents of semantic theories, on the other hand, have concluded that the existence of verbs of variable behavior does away with the claim that a level of abstract syntactic representation is a factor in auxiliary selection. For Zaenen (1993) and Shannon (1990), auxiliaries are selected directly on the basis of the semantics of the predicate. Lexical decomposition analyses dispose of other means to deal with verbs of variable behavior. One solution, called by some ‘template augmentation’ (Levin & Rappaport 2005; Sorace 2000), is to introduce lexical rules that alter the semantic representation of a verb, as in (18), turning an activity into an accomplishment (see also Van Valin 1990). (18) run’(x) ==> [run’(x)] CAUSE [BECOME be-at’(y, x)]
Another solution is to leave some features ‘underspecified’ in the lexical semantic representation of the verb. In Lieber and Baayen’s (1997) analysis of Dutch, activities like lopen ‘walk’ are underspecified for the feature [IEPS]. The telicity of the predicate may fix the value of the feature as [+IEPS], resulting in selection of zijn. Directed change of state and change of direction verbs take zijn even if the predicate is atelic, because they are lexically specified as [+IEPS]. Lexical underspecification is also used in the ‘constructional’ approach to split intransitivity (Borer 1994). A verb’s arguments are not specified as internal or external. Rather, this task is left to functional projections in the syntax. This theory can easily explain the sensitivity of auxiliary selection to syntactic factors (i.e. endpoint adverbials), but, as argued in Sorace (2000), it cannot explain why some predicates select BE even if the syntax of the clause would dictate otherwise. In Dutch, for instance, verbs of directed change of state or location do not change to hebben when modified by an adverbial that makes the predicate atelic (Zaenen 1993). Sorace (2000) argues in favor of a combination of constructional and projectionist approaches using template augmentation, saying that it provides the right articulation between lexical and syntactic factors. This approach may also hold the key to solving some split intransitivity test mismatches. In a recent paper, Bentley (2004) argues that when a ne clitic cross-
Split auxiliary selection from a cross-linguistic perspective
references the subject of an A verb in Italian, as in (12), it is because the information structure of the clause licenses a state predicate that is not in the lexical semantic representation of the main verb. Sentence (12) can be glossed as ‘There exist few, of bulbs, which took root.’ Following Van Valin’s (1990) analysis, Bentley assumes that ne can cross reference arguments of state predicates. In the perfect, however, the presentational interpretation of the sentence is unavailable, and necliticization is possible only out of those verbs that have state predicates in their lexical-semantic representation.
. Mismatches across languages or dialects All theories of auxiliary selection must also address the issue of cross-linguistic variation in the distribution of BE and HAVE. Two verb classes are stable in their auxiliary selection properties across languages: verbs of directed change of location (like French arriver ‘arrive’, descendre ‘go down’), which select BE, and controlled volitional activities (like Italian lavorare ‘work’), which select HAVE. Aside from these two groups, the classes of E verbs and A verbs vary systematically from language to language. Verbs of definite or directed change of state (like Dutch sterven ‘die’) tend to select BE, but there are exceptions in French, where exploser ‘explode’ selects avoir. Verbs of indefinite change of location select sein in German (19), a feature that distinguishes it from the other major languages that have a split auxiliary system. (19) Die Kinder sind/*habben viel (herum-)gesprungen. The children are/have much (around-)jumped ‘The children jumped around (much).’
German
Verbs of non-directed change of state like ‘grow’ select sein in German as well, but in Dutch verbs like groeien ‘grow’ or springen ‘jump’ tend to select hebben, even though they display variable behavior. The same is true for Italian, but in French verbs like grandir ‘grow’, vieillir ‘age’ select être. Verbs of continuation of state like Dutch duren ‘last’ or French survivre ‘survive’ select HAVE in Dutch and German, and also in French. (20) Mes parents *sont survecus / ont survécu au tremblement de My parents are survived / have survived to-the shaking of terre. earth ‘My parents survived the earthquake.’
A curious exception in this class are the verbs ‘remain, stay’ (French rester, Dutch blijven) and the verb BE itself, which select BE, as they do in Italian. Italian con-
Raúl Aranovich
Table 2. Cross-linguistic variation in auxiliary selection
Dir. ch. of loc. Dir. ch. of state Indef. ch. of loc. Indef. ch. of state Cont. of state Existence and appearance Activity
Italian
German
Dutch
French
BE BE HAVE∼BE BE BE BE HAVE
BE BE BE BE HAVE/(BE) HAVE/(BE) HAVE
BE BE HAVE∼BE HAVE∼BE HAVE/(BE) HAVE/(BE) HAVE
BE HAVE/(BE) HAVE HAVE HAVE/(BE) HAVE/(BE) HAVE
trasts sharply with the other major languages in this class, since verbs of continuation of state normally select essere. In Swiss German, however, verbs like wohnen ‘dwell’ and schlafen ‘sleep’ select BE, breaking with the standard Germanic pattern. The remaining class of ‘stative’ verbs, as they are sometimes called, is that of verbs of existence and appearance. They occur mostly with essere in Italian (v.g. apparire ‘appeared’, emergere ‘emerged’, succedere ‘happened’), but in the other languages their behavior is inconsistent. In French they are mostly found with avoir (v.g. surgir ‘appear suddenly’, disparaître ‘disappear’), but apparaître ‘appear’ selects être. Likewise, in Dutch blijken ‘seem, appear’ selects zijn. Table 2 summarizes the main cross-linguistic differences. Another domain of variation in auxiliary selection across languages is that of reflexive verbs and inchoative verbs that alternate with a causative counterpart. As noticed before, reflexive verbs select BE in French and Italian. In the Germanic languages, however, this is not the case, as example (21) shows. This difference can be attributed to the fact that the Romance languages have reflexive clitics (Haider & Rindler-Schjerve 1987). (21) Dieser wagen hat sich gut verkauft. This car has refl well sold ‘This car has sold well.’
German
In the Romance languages, many transitive verbs have inchoative alternants that are formed with a reflexive clitic. In Italian and French these ‘pronominal verbs’ select BE. (22) a.
Luigi ha svegliato Arturo Luigi has woken-up Arturo ‘Luigi woke Arturo’ b. Arturo si è svegliato Arturo refl is woken-up ‘Arturo woke up’
Italian
Other examples include Italian accendersi ‘ignite’, aprirsi ‘open’, sporcarsi ‘get dirty’, confondersi ‘get confused’, French s’agrandir ‘get larger’, se noyer ‘drown’, s’ouvrir
Split auxiliary selection from a cross-linguistic perspective
‘open’, se casser ‘break’. In the Germanic languages, causatives and their inchoative counterparts are also morphologically different. In Dutch, many alternating verbs are marked by a prefix in the transitive counterpart (e.g. zinken ‘sink’ (intr.) ∼ afzinken ‘sink’ (tr.)). The intransitives also select BE, as expected.11 There are some inchoative verbs, however, that are not distinguished from their causative counterparts by a reflexive verb or a prefix (Dutch breken ‘break’, genezen ‘heal’, devalueren ‘devaluate’). In Italian and Dutch, these anticausatives select BE. French, however, is distinguished from the other languages with split auxiliary systems because these verbs select HAVE (e.g. sombrer ‘sink’, sécher ‘dry out’).12 (23) a.
Il prezzo è aumentato. The price is risen ‘The price increased.’ b. La neige a fondu rapidement. The snow has melted quickly ‘The snow melted quickly.’
Italian
French
Verbs of assume-position display interesting cross-linguistic features as well. These verbs normally take a reflexive pronoun, and select BE in the Romance languages: s’assoir ‘sit down’. Assume-position verbs also take a reflexive pronoun in German, but they select HAVE, presumably because reflexives do so in the Germanic languages. Southern German, however, allows for a non-reflexive variant of assume-position verbs (stehen ‘stand’, liegen ‘to lie’, sitzen ‘sit’) that selects BE. (24) a.
Das Kind hat/*ist sich auf den Boden gehockt. The child has/is refl on the floor squated ‘The child squatted down on the floor.’ b. Dä Albert isch uff dä Boddä glägge. The Albert is on the floor layed ‘Albert lay down on the floor.’
German
Southern G.
The existence of variation in slit auxiliary systems across languages is part of the evidence syntactic theories of auxiliary selection use to argue against a semantic basis for the selection of BE or HAVE (Rosen 1984; Perlmutter 1989). The BE auxiliary . In German, prefixes are also used to denote the ‘telic’ version of process verbs. Like other telic verbs, these prefixed versions of intransitives select BE. Compare die Blume hat geblüht ‘the flower blossomed’ to die Blume ist verblüht ‘the flower finished blossoming.’ . It has been argued that there is a semantic difference between intransitives with and without a reflexive. For instance, ramollir ‘soften’ denotes an indefinite change, but se ramollir indicates an end of process. The fact that the non-pronominal inchoatives select HAVE may be related to the fact that verbs of indefinite change of state do not select BE in French.
Raúl Aranovich state change of location
change of state
cont. of preexisting state
process existence of state
uncontrolled
BE
controlled controlled (motional) (non-motional) HAVE
Figure 2. Auxiliary Selection Hierarchy (Sorace 2000)
is selected by unaccusative verbs, they argue, and verbs with the same meaning can be classified as unergative in one language, but as unaccusative in another, because the distinction is arbitrary. An alternative is to uphold the universal semantic determination of unaccusativity, disregarding auxiliary selection as a good cross-linguistic test for unaccusativity, as Legendre (1989) does for French. None of these solutions, however, account for the generalization that there are some verbs that do select BE or HAVE universally, across those languages that have a split in auxiliary selection, and that these are the same verbs that show stable behavior within a language. These generalizations are captured in theories that recognize prototypical members of the E verb group and the A verb group, such as Dowty’s proto-role analysis or Shannon’s prototypical event theory. Variation in auxiliary selection is expected for those verbs that are assigned to the two ‘fuzzy’ categories by extension from the prototypes. Sorace (2000) proposes an Auxiliary Selection Hierarchy that ranges between change of location predicates and nonmotional controlled processes, which are the ‘core’ verbs that universally select BE or HAVE, respectively, in all languages that have a split auxiliary system. These verb groups, which correspond to the two prototypes of the functional-cognitive approach in Shannon, and to verbs whose arguments have the highest number of proto-patient and proto-agent properties, form the extremes of an Auxiliary Selection Hierarchy, presented in Figure 2. The ASH is based on a combination of thematic and aspectual criteria. Core E verbs are inherently telic. For atelic verbs, volitionality (or protagonist control) is a key property: atelic, controlled non-motional processes are the core A verbs. Predicates in the middle regions of the hierarchy may select BE or HAVE, depending on the language or other factors, but what Sorace suggests is that there is an implicational relation among these classes: If a verb class selects BE, then any other class higher up in the hierarchy will select BE as well. This ‘gradient’ approach not only accounts for systematicity in auxiliary selection differences across languages; it also predicts that the verbs that show variable behavior will be the ones in the middle regions of the hierarchy.
Split auxiliary selection from a cross-linguistic perspective
. Diachronic and sociolinguistic factors in auxiliary selection From a diachronic point of view, there is instability in split auxiliary systems as well. In many languages the distinction between the two auxiliaries has disappeared. In Spanish and English it has leveled to HAVE, but this is not always the case. In some Central Italian dialects, most notably Terracinese (Tuttle 1986), the perfect auxiliary is BE. The auxiliary BE already existed in Latin, as the perfect of the medio-passive conjugation. Vincent (1982) traces the path of ‘grammaticalization’ of the auxiliary HAVE, from its use as a verb that expresses a resulting state of an objects (see also Olbertz 1993). From there, historical changes in auxiliary selection follows a pattern that is determined by lexical semantics (Cennamo 2002; Aranovich 2003; Shannon 1990), in a way that is consistent with a ‘gradient’ approach. Aranovich argues that the verbs that resist the encroachment of HAVE the longest are the ones whose subjects have the highest number of proto-patient properties, while Shannon shows that in Germanic it is those verbs that are closer to the mutative prototype that appear with BE. As in many cases of diachronic language change, factors related to usage and social variation influence the process, in addition to grammatical principles. This can be seen in modern Canadian French, in which the use of HAVE has been extended, displacing BE even with verbs of motion and reflexive verbs (Sankoff & Thibault 1977; Russo & Roberts 1999; Canale, Mougeon, & Bélanger 1977). All the studies point to a tendency to keep être with the most frequent verbs (aller ‘go’, venir ‘come’, arriver ‘arrive’). In central Italo-Romance, the leveling of the system towards BE follows a path in which splits according to person replace splits according to verb class (Tuttle 1986; Zörner 1993). In many of these dialects, BE is used for the 1 and 2 persons, while HAVE is used for 3rd person, even with reflexives or transitives.
. Conclusions and contents of the volume The study of split auxiliary systems arose interest in theoretical linguistics mostly in the context of the Unaccusative Hypothesis. Syntactic analyses identify the subject of E verbs as an underlying object, an analysis that is supported by evidence from other constructions (ne cliticization, impersonal passivization, verbs with reflexive clitics, etc.). Semantic theories of auxiliary selection, on the other hand, often focus on the existence of mismatches among unaccusativity tests, and also on the observed variation in auxiliary selection within a language or across languages, to defend analyses of split intransitivity phenomena based on meaning. In this debate, analyses of the phenomena take many forms (lexical decomposition or cognitive prototypes, grammatical relations or structural positions, thematic
Raúl Aranovich
criteria or aspectual criteria), and some leave room for compromise, but more importantly, in recent years all theories have been fine-tuned to account for a growing richness of empirical details in the description of split auxiliary systems. Fresh data come from examining new languages, dialects, or historical stages of better studied languages, and also from new constructions and new methodologies (corpus analysis, experimental studies, etc.). In this process, split auxiliary systems have gained status as a topic of interest for linguistic theory in its own right, not just as another test for unaccusativity or for semantic theories of split intransitivity. The opposition between BE and HAVE, after all, goes beyond the identification of two classes of intransitive verbs, to present riddles of its own in its interaction with person hierarchies and diachronic change. It is in this context that the contributions to this volume have to be considered. In “Transitivity Parameters and Auxiliary Selection by L2 Students of German”, Carlee Arnett addresses the issue of the development of a split auxiliary system in learners of German as a second language. She assumes a gradient model of auxiliary selection based on Shannon’s cognitive approach, in which German sein ‘be’ is selected more consistently by verbs that are closer to the ‘mutative’ prototype. Testing English-speaking students enrolled at various levels of German courses for production of perfect forms in the description of two scenes involving motion (riding a bicycle, walking down the street), Arnett finds out that punctuality and telicity are relevant to capture generalizations about increased proficiency through the different levels, but only as additional cues in conjunction with number of participants, which is the main parameter students home in for selecting an auxiliary. Surprisingly, adding telicity to punctuality results in a less accurate selection rate. Arnett’s results, however, support the gradient model of auxiliary selection for the understanding of second language acquisition. In “An Irrealis BE Auxiliary in Romanian”, Larisa Avram and Virginia Hill argue that the Latin periphrastic perfect with ESSE is grammaticalized in Romanian as fi ‘be’, which is a marker of irrealis mood. The authors claim that Romanian has a split auxiliary system that differs from the one find in other Romance languages (i.e. Italian) in that the distribution of BE and HAVE is not tied to the distinction between unergatives and unaccusatives. Framing their analysis within Chomsky’s Minimalist Program, they assign to auxiliary fi the feature [irrealis] in addition to [tense/perfective], which it shares with HAVE. Both auxiliaries merge with Aspect and Tense for checking, but fi merges with a weak Tense head selected by Mood. Avram and Hill suggest that this analysis accounts for some grammatical properties of auxiliary fi, in particular its exclusion from the first position in the inflectional string and lack of inflection for subject agreement or for tense. In “Auxiliary Selection and Split Intransitivity in Paduan: Variation, and Lexical-aspectual constraints”, Michela Cennamo and Antonella Sorace examine auxiliary selection in Paduan, a North-Eastern Italian dialect. They show that
Split auxiliary selection from a cross-linguistic perspective
HAVE has a broader scope in Paduan than in Standard Italian, appearing with statives (durare ‘last’) and change-of-state verbs (cresere ‘grow’), as well as with some reflexive verbs. They also show that additional usual tests for unaccusativity in Standard Italian (participial agreement, ne cliticization, subject inversion) do not cluster around a well-defined class of intransitives or with each other (ne clitics, for instance, appear with HAVE on occasion). These facts lead the authors to weaken the case for a purely syntactic approach to split intransitivity, arguing for a gradient approach to auxiliary selection based on Sorace’s Auxiliary Selection Hierarchy. The threshold for selection of HAVE has moved further away from the core A verbs in Paduan, compared to Standard Italian. The ASH predicts that variation will be found in the intermediate regions of the hierarchy, and that variation in this domain will be affected by aspectual properties of the event and other lexical-semantic properties of the predicate (agentivity, internal/external causation, etc.). In “The Development of the HAVE Perfect: Mutual Influences of Greek and Latin”, Bridget Drinka challenges the received views on the grammaticalization of the Romance periphrastic perfect with HAVE with an external explanation. She argues that the Latin perfect with HAVE and a passive perfect participle (PPP) is a calque from Classical Greek, which had developed a periphrastic perfect with HAVE and an active aorist participle much earlier. The Greek perfect entered Latin through a learned channel, due to the prestige of Greek among the Roman educated elites, and its influence as a literary model on authors like Plautus, Cato the Elder, and Vitruvius. In a later development, however, Drinka argues that Latin influenced Hellenistic Greek. As the aorist and the perfect merged, Koiné Greek adopted the Latin HAVE perfect with a PPP. Through the influence of the Greek Christian writers and the Vulgata, this construction became stylistically associated with Christianity in the late Roman empire. In “Agentivity versus Auxiliary Choice: Evidence from Pronominal Binding in German AcI-constructions”, Vera Lee-Schoenfeld shows that there is a mismatch between auxiliary selection and another unaccusativity test in German, involving pronominal binding. In the Accusativum-cum-Infinitivum (AcI) construction, an embedded pronoun can be co-referential with a matrix subject only if the infinitive is an unergative verb. To account for the binding facts Lee-Schoenfeld adopts the constructionist analysis of unaccusativity. Unergative clauses have an additional structural layer associated with agentivity (‘little’ vP) which provides a domain in which bound pronominals can occur. She shows that some of the verbs that allow for pronominal binding in the AcI construction select sein ‘be’ in the perfect. From this she concludes that auxiliary selection is not a good diagnostic for unaccusativity in German, and that the HAVE/BE contrast is better explained in a gradient approach like the one Sorace proposes.
Raúl Aranovich
In “Optimizing Auxiliary Selection in Romance”, Géraldine Legendre develops an Optimality-Theoretic account of the gradient approach to auxiliary selection. She suggests that the fact that selection of the auxiliaries BE and HAVE is not categorically linked to a particular semantic feature requires an analysis in terms of ranked violable constraints. Variation in auxiliary selection emerges from the interaction of a markedness constraint against selecting the BE auxiliary (*E) and a hierarchy of constraints penalizing association of a scale of lexical aspectual and volitionality features with the HAVE auxiliary. The broader distribution of the HAVE auxiliary in French with respect to Italian results from the lower ranking of *E in Italian. Re-ranking of a constraint against associating HAVE with motion account for Romance/Germanic contrasts. The model of constraint interaction Legendre proposes makes the right predictions for the diachronic evolution of split auxiliary systems, and also for the synchronic range of free variation. Legendre also addresses the relevance of an OT approach to other unaccusativity diagnostics, drafting an implementation of a ‘Parameterized Linking’ analysis. In “Auxiliary Selection in Chinese”, Feng-hsi Liu examines the distribution of two aspectual auxiliaries, -le and -zhe, in Chinese locative inversion, a construction with existential and presentational functions. Liu argues that auxiliary selection in this construction is determined by the semantics of the event, in particular the interaction of lexical aspect (telicity and stativity) and agentivity. She proposes a gradient approach to the -le/-zhe alternation (similar in some ways to Sorace’s approach to auxiliary selection in Romance and Germanic) in which telic verbs at one end select -le, dynamic atelic verbs at the other end select -zhe, and verbs that fall in between may show variable behavior depending on other factors. Liu argues against a purely syntactic analysis of the -le/-zhe alternation based on the fact that auxiliary selection with ‘detransitivized’ verbs (analogous to passives) is sensitive to the same semantic parameters. In “Parameterized Auxiliary Selection: A Fine-grained Interaction of Features and Linking Rules”, Janet H. Randall supports the hypothesis that auxiliary selection is a reliable diagnostic for unaccusativity. Her ‘parameterized linking’ theory does so while at the same time accounting for the properties of verbs of variable behavior and for cross-linguistic differences in auxiliary selection. Linking rules map arguments from Conceptual Structure (CS) into internal (unaccusative) or external (unergative) positions in Argument Structure (AS), depending on the semantic features of CS. To account for the fact that locomotion determines selection of BE in German even for atelic predicates, she proposes a rule linking the argument of a locomotion verb to an internal position in AS, in addition to a Telic Rule that links the argument of telic predicates to internal positions in AS. The Locomotion rule is ‘parameterized’ to be inactive in Dutch, which selects HAVE for atelic locomotion verbs. Randall compares her model to the gradient approach in Sorace (2000) and Keller and Sorace (2003) arguing that it offers a better explanation for
Split auxiliary selection from a cross-linguistic perspective
unexpected experimental results concerning auxiliary selection by non-core verbs in German. In “Particle Selection in Korean Auxiliary Formation”, Seongha Rhee examines the patterns of grammaticalization of several verbal auxiliaries in Korean, which are all derived from serial verb constructions with the verb issta ‘exist’. Rhee argues that the differences in grammatical meaning between the auxiliaries are brought about by the non-finite particles that are part of the source constructions. The two coordinating particles ko and e, for instance, give rise to a progressive and a perfective auxiliary, respectively, because e entails a sequential relationship between the coordinated events, while ko allows for simultaneity. Rhee concludes that grammaticalization can progress along different channels in the same language even if the same source verb is involved, a process comparable to the grammaticalization of HAVE in Romance to express different grammatical categories (future, perfect). In “Language Use and Auxiliary Selection in the Perfect”, K. Aaron Smith traces the history of the BE perfect auxiliary in English, which existed in older varieties but was overtaken by HAVE in a process ending in the 19th century. Smith proposes a usage-based explanation, arguing on the basis of type/token frequency comparisons with Modern German. In German the HABE/BE split is very stable. German BE has a higher frequency than Old English BE, which may explain why the latter could not withstand the advance of HAVE. But Smith finds another reason for the entrenchment of the BE perfect in German: the periphrastic perfect is more frequent overall in German than in Old English, since in German it has simple past uses in addition to resultative/anterior uses. This extended use of the construction in German (compared to Old English) is enough, Smith claims, to stabilize the use of the BE perfect in spite of its low frequency relative to the HAVE perfect.
References Aranovich, R. (2003). The semantics of auxiliary selection in Old Spanish. Studies in Language, 27, 1–37. Bentley, D. (2004). Ne-cliticisation and split intransitivity. Journal of Linguistics, 40, 219–262. Bentley, D. & Eythórsson, T. (2001). Alternation according to person in Italo-Romance. In L. Brinton (Ed.), Historical Linguistics 1999 (pp. 63–74). Amsterdam: John Benjamins. Benzing, J. (1931). Zur Geschichte von ser als Hilfszeitwort bei den intransitiven Verben im Spanischen. Zeitschrift für romanische Philologie, LI, 385–460. Borer, H. (1994). The projection of arguments. In E. Benedicto & J. Runner (Eds.), Functional Projections [University of Masschusetts Occasional Papers 17] (pp. 19–48). Amherst, MA: GASLA. Burzio, L. (1986). Italian Syntax: A government-binding approach. Dordrecht: Reidel.
Raúl Aranovich
Canale, M., Mougeon, R., & Bélanger, M. (1978). Analogical leveling of the auxiliary être in Ontarian French. In M. Suñer (Ed.), Contemporary Studies in Romance Linguistics (pp. 41– 61). Washington, DC: Georgetown University Press. Cennamo, M. (2002). La selezione degli ausiliari perfettivi in napoletano antico: Fenomeno sintattico o sintattico-semantico? Archivio glottologico italiano, 87, 175–222. Centineo, G. (1986). A lexical theory of auxiliary selection in Italian. Davis Working Papers in Linguistics, 1, 1–35. (Revised version published as: Centineo, G. (1996). A lexical theory of auxiliary selection in Italian. Probus, 8, 223–271.) Chierchia, G. (2004). A semantics for unaccusatives and its syntactic consequences. In A. Alexiadou, E. Anagnostopoulou, & M. Everaert (Eds.), The Unaccusativity Puzzle (pp. 22– 59). Oxford: OUP. Cocchi, G. (1994). An explanation of the split in the choice of perfect auxiliaries. Probus, 6, 87–102. Dowty, D. (1991). Thematic proto-roles and argument selection. Language, 67, 547–619. Grimshaw, J. (1982). On the lexical representation of Romance reflexive clitics. In J. Bresnan (Ed.), The mental representation of grammatical relations (pp. 87–148). Cambridge, MA: The MIT Press. Grimshaw, J. (1990). Argument Structure. Cambridge, MA: The MIT Press. Guéron, J. (1987). L’hypothèse inaccusative. Travaux, 52, 79–90. (La transitivité, domaine anglais.) Université de Sain-Etienne, Saint-Etienne. Haider, H. & Rindler-Schjerve, R. (1987). The parameter of auxiliary selection: Italian-German contrasts. Linguistics, 25, 1029–1055. Hoekstra, T. (1984). Transitivity: Grammatical relations in government binding theory. Dordrecht: Foris. Hoekstra, T. (1994). HAVE as BE plus or minus. In G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi, & R. Zanuttini (Eds.), Paths towards Universal Grammar. Washington, DC: Georgetown University Press. Hoekstra, T. (1999). Auxiliary selection in Dutch. Natural Language and Linguistic Theory, 17, 67–84. Kathol, A. (1994). Passives without lexical rules. In J. Nerbonne, K. Netter, & C. J. Pollard (Eds.), German in Head-Driven Phrase Structure Grammar (pp. 237–272). Stanford, CA: CSLI. Kayne, R. (1993). Toward a modular theory of auxiliary selection. Studia Linguistica, 47, 3–31. Keller, F. & Sorace, A. (2003). Gradient auxiliary selection and impersonal passivization in German: An experimental investigation. Journal of Linguistics, 39, 57–108. Legendre, G. (1989). Unaccusativity in French. Lingua, 79, 95–164. Levin, B. & Rappaport-Hovav, M. (1995). Unaccusativity: At the syntax-lexical semantics interface. Cambridge, MA: The MIT Press. Lieber, R. & Baayen, H. (1997). A semantic principle of auxiliary selection in Dutch. Natural Language and Linguistic Theory, 15, 789–945. Levin, B. & Rappaport-Hovav, M. (2005). Argument realization. New York, NY: CUP. Lonzi, L. (1986). Pertinenza della strutura tema-rema per l’analisi sintattica. In H. Stammerjohann (Ed.), Theme-Rheme in Italian (pp. 99–120). Tübingen: Narr. Olbertz, H. (1993). The grammaticalization of Spanish haber plus participle. In J. van Marle (Ed.), Historical linguistics 1991: Papers from the 10th international conference on historical linguistics (pp. 243–263). Amsterdam: John Benjamins. Perlmutter, D. (1978). Impersonal passives and the unaccusative hypothesis. In Proceedings of the Fourth Annual Meeting of the Berkeley Linguistic Society (pp. 157–189). Berkeley, CA: Berkeley Linguistic Society.
Split auxiliary selection from a cross-linguistic perspective
Perlmutter, D. (1983). Personal vs. impersonal constructions. Natural Language and Linguistic Theory, 1, 141–200. Perlmutter, D. (1989). Multiattachment and the unaccusative hypothesis: The perfect auxiliary in Italian. Probus, 1, 63–119. Perlmutter, D. & Postal, P. (1984). The 1-advancement exclusiveness law. In D. Perlmutter & C. Rosen (Eds.), Studies in Relational Grammar 2. Chicago, IL: University of Chicago Press. Rosen, C. (1984). The interface between semantic roles and initial grammatical relations. In D. Perlmutter & C. Rosen (Eds.), Studies in Relational Grammar 2 (pp. 38–77). Chicago, IL: University of Chicago Press. Rosen, C. (1988). The Relational Structure of Reflexive Clauses: Evidence from Italian. New York, NY: Garland. Russo, M. & Roberts, J. (1999). Linguistic change in endangered dialects: The case of alternation between avoir and être in Vermont French. Language Variation and Change, 11, 67–85. Sankoff, G. & Thibault, P. (1986). The alternation Between the Auxiliaries avoir and être in Montrèal French. In G. Sankoff (Ed.), The Social Life of Language (pp. 311–345). Philadelphia, PA: University of Pennsylvania Press. Shannon, T. (1990). The unaccusative hypothesis and the history of the perfect auxiliary in Germanic and Romance. In H. Andersen & E. K. K. Koerner (Eds.), Historical Linguistics 1987 (pp. 461–488). Amsterdam: John Benjamins. Sorace, A. (2000). Gradients in auxiliary selection with intransitive verbs. Language, 76, 859– 890. Tuttle, E. F. (1986). The spread of ESSE as universal auxiliary in Central Italo-Romance. Medioevo Romanzo, 11, 229–287. Van Valin, R. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Vincent, N. (1982). The development of the auxiliaries habere and esse in Romance. In V. Nigel & M. Harris (Eds.), Studies in the Romance Verb (pp. 71–96). London: Croom Helm. Yamaguchi, T. & Pétursson, M. (2003). The speaker and the perfective auxiliaries hafa and vera in Icelandic. Language Sciences, 25, 331–352. Zaenen, A. (1993). Unaccusativity in Dutch: Integrating syntax and lexical semantics. In J. Pustejovsky (Ed.), Semantics and the Lexicon (pp. 129–161). Dordrecht: Kluwer. Zörner, L. (1993). I dialetti del’Oltrepò pavese tra il lombardo, lemiliano ed il ligure. Rivista italiana di dialettologia, 17, 55–98.
Transitivity parameters and auxiliary selection by L2 students of German Carlee Arnett University of California, Davis
Auxiliary selection in L1 German is sensitive to the degree of transitivity of a clause. This study focuses on L2 students of German and three transitivity parameters, number of participants, telicity and punctuality. This study shows that as students become more proficient in German, they are more accurate in auxiliary selection. The key parameter is the number of participants. All groups are most accurate with respect to this parameter. Three of the groups show high accuracy rates with punctual clauses as well. However, there are only a small number of these types of clauses. Intermediate students are more accurate when the clause is telic than first year students. None of the students named a transitivity parameter when asked how they made their decision.
Previous research in the field of auxiliary verb selection in the present perfect tense has focused primarily on the Romance languages (cf. Burzio 1986; Rosen 1988; Perlmutter 1989; Lois 1990; Cortes 1993; Centineo 1996), Dutch (cf. Lieber & Baayen 1997; Hoekstra & Mulder 1990; Shannon 1993), German (Shannon 1992; Klein 2000) and older Germanic languages (Shannon 1990, 1995). These studies describe the syntax and semantics of auxiliary selection for L1 speakers within a range of theoretical frameworks. With respect to L2 speakers, there are only two studies that address the issue of auxiliary selection. Sorace (1993, 2000) investigates the acquisition of correct auxiliary choice in L2 learners of Italian. This study assumes, as Sorace does, that auxiliary selection is a gradient phenomenon and investigates the factors that affect auxiliary selection for L2 speakers of German as well as their decision-making strategies. Auxiliary selection in German is sensitive to the degree of transitivity of a clause (cf. Shannon 1987, 1990, 1992, 1995; Eisenberg 1986; Zifonun, Hoffmann, & Strecker 1997; Klein 2000). The parameters of transitivity consist of two or more participants, aspect, punctuality, volitionality, affirmation, mode, agency, affectedness of the object and the individuation of the object (Hopper & Thompson 1980; Rice 1987; Shannon 1992). A common example of this is fahren ‘drive’ which takes
Carlee Arnett
haben or sein depending on presence of a direct object or other parameters of transitivity. This paper will address auxiliary selection in the present perfect in four groups of L2 students of German. This study is limited to the present perfect for the ease of elicitation and the fact that auxiliary selection in the pluperfect is not necessarily the same. University students at various levels of learning German were given a series of pictures and asked to write a short paragraph about the event. The pictures depicted events that needed to be described using verbs that can take either haben or sein depending on how the speaker chooses to represent the event. The last page of the survey asked the students to describe how they decided on haben or sein, state the rule they had learned for auxiliary selection, and answer some questions about their L1 and L2 backgrounds. I want to know if the students are sensitive to three of the parameters of transitivity (number of participants, telicity and punctuality) and, if they are, at what level of L2 development these become relevant.
.
Literature review
The Unaccusative Hypothesis proposed in both the Relational Grammar and Government-Binding frameworks claims that there are two discrete classes of verbs, unaccusative and unergative (Perlmutter 1978; Hoekstra 1984; Haider 1985; Burzio 1986). Unaccusative verbs have empty subject positions but filled object positions at deep structure. These underlying objects then obligatorily move to subject position in order to receive case. Unergative verbs have a filled subject position at deep structure, and thus no movement is required. Perfect auxiliary selection, passivization, as well as other syntactic tests, are used to distinguish unaccusative verbs from unergatives, but each test predicts a different set of unaccusative verbs (Fagan 1988, 1992, 1996; Shannon 1987, 1990). The binary classification scheme of unaccusative and unergative cannot account for the L1 data in German and Dutch (Zaenen 1993; Lieber & Baayen 1997; Sorace 2000 for other European languages) because no consistent list of verbs can be generated. Sorace (2000) claims that auxiliary selection depends on an aspectual/thematic hierarchy of verb type. Some verbs always take one auxiliary, while others allow both depending on their position in the hierarchy. The syntactic phenomenon of auxiliary selection is thus gradient and not binary. Sorace makes a number of interesting observations about L1 and L2 auxiliary selection. First, auxiliary selection is not consistent across languages/dialects. Some verbs are consistent in the auxiliary they select across languages/dialects and others are not. Second, within languages some verbs tend to select the same auxiliary regardless of aspectual or thematic elements, others switch in a particular context, and others allow both in the same context (Hoekstra & Mulder 1990; Levin & Rappaport Hovav 1995;
Transitivity parameters and auxiliary selection by L2 students of German
Table 1. Transitivity/Mutativity parameters
Participants Kinesis Aspect Punctuality Volitionality Affirmation Mode Agency O Affected O Individuated
High Transitivity
High Mutativity
Low Transitivity
2 or more action telic punctual volitional affirmative realis A high in potency O totally affected O highly individuated
1 participant action (event) telic punctual nonvolitional affirmative realis A low in potency O (=A) totally affected O (=A) highly individuated
1 participant nonaction atelic nonpunctual nonvolitional negative irrealis A low in potency O not affected O not affected
Pustejovsky & Busa 1995). Third, native speakers have clear intuitions about which auxiliary to use with some verbs. With other verbs, native speakers do not show clear preferences (Bard, Robertson, & Sorace 1996; Sorace 1992). Fourth, in L1 and L2 acquisition, auxiliary selection for some verbs occurs earlier than for others (Sorace 1993, 1995). Since a binary classification scheme cannot predict which verbs in German will take haben and which will take sein, another cognitive approach was developed by Shannon (1992, 1995), based on Hopper and Thompson’s (1980) transitivity parameters. Hopper and Thompson’s parameters are number of participants, kinesis, aspect, punctuality, volitionality, affirmation, mode, agency, affectedness of the object and separateness of the object. The underlying principle in Hopper & Thompson and Shannon is that it is not a verb that is transitive or intransitive, but a clause. Shannon (1993, 1995) claims that prototypically transitive and prototypically mutative events can be represented by expanding Hopper and Thompson’s parameters. Mutativity is a new term and it represents clauses that have one participant and are punctual. Shannon’s (1992: 103) classification scheme is given in Table 1. Although I have chosen a cognitive account for the purposes of this study, it does not mean that here aren’t other factors that influence auxiliary selection. Obviously, lexicalization plays a role in which types of clauses take haben and which take sein. Furthermore, frequency supports the lexicalization of haben as the most common auxiliary (Smith 2001; Smith, this volume). Prototypically transitive events (cf. Langacker 1982; Rice 1987) transpire in physical space, involving two or more distinct participants that interact asymmetrically. The first participant moves toward and contacts the second. The first participant acts volitionally and the second is maximally affected. Consider example (1). (1) The toddler squished the bug.
Carlee Arnett
The toddler wanted to do the action and the bug was changed by it. Prototypically transitive events are telic, punctual and occur in the real world. Prototypically mutative events also transpire in physical space but involve only one distinct participant that changes state or moves. (2) I went to the store.
Like prototypically transitive events, these events are telic, punctual and occur in the real world. In addition to prototypical transitivity and mutativity, Shannon posits a category for low transitivity. These events prototypically have one participant, are non-punctual, non-volitional, irrealis and atelic as shown in example (3). (3) The presentation lasted for hours.
In German, auxiliary selection corresponds to the transitivity/mutativity parameters. Clauses that are highly transitive select haben as their auxiliary in the present perfect. Clauses that are highly mutative select sein as their auxiliary. Consider the following examples with the auxiliary haben: (4) Die Krankenschwester hat den Physiker erwurgt. The.nom-nurse have.past the.acc-physicist strangle.pp The nurse strangled the physicist. (5) Die Wikinger haben in der Tat Vinland entdeckt. The Vikings-nom have-past in actuality Vinland-acc discovered-pp. The Vikings discovered Vinland.
Examples (4) and (5) show an event that is punctual and has two distinct participants in an asymmetrical relationship. The first participant acts volitionally and it is highly potent. The second participant is maximally distinct from the first and affected by the action of the first. The action is represented by the indicative and is presented as taking place in the real world. Clauses that are highly mutative select be as their auxiliary in the present perfect. Consider the following examples: (6) Der Zug ist vor 5 Minuten abgefahren. The.nom train be.past ago 5 minutes leave-pp. The train left 5 minutes ago. (7) Gestern ist sie in die Kirche gegangen. Yesterday be.past she.nom in the.acc church go-pp. Yesterday, she went to church.
Example (6) shows a one participant, punctual event in which the first participant changes state, but does not act volitionally. Example (7) is similar in all respects, except that the event is telic as well.
Transitivity parameters and auxiliary selection by L2 students of German
Clauses in the low transitivity category select haben as their auxiliary and exhibit properties of both the high transitivity and high mutativity categories. Clauses that are low in transitivity always have one participant and they are nonaction, atelic, nonpunctual and nonvolitional. Clauses that are low in transitivity might be negative and irrealis, but they need not be. Consider the following example: (8) Der Vortrag hat stundenlang gedauert. The.nom presentation have.past for hours last-pp. The presentation lasted for hours.
Example (8) is a prototypical example of low transitivity. There is one participant that is nonvolitional and low in potency. The event is not an action and it is nonpunctual. In the study described below, I will investigate if L2 German speakers are sensitive to the three most basic parameters of transitivity/mutativity: event structure, telicity and punctuality.
. Methodology In an anonymous survey (see Appendix A), students were asked to describe two pictures in the present perfect and then fill out a short questionnaire. The pictures showed events that are described in German by using verbs that can go with either haben or sein depending on how the speaker chooses to represent the event. The questionnaire asked the students to say how they chose which auxiliary to use and then write the grammar rule, if they remembered it. The questionnaire concluded with asking the students to provide information about their previous experience learning German. The questionnaire was first piloted on graduate students in the German Department at the university where the study was conducted. The survey was administered to three different groups: students in the second quarter of first-year German instruction, second year students, and students in upper-division courses for majors. After administering the survey, I divided them into four different groups. The first group was first-year German students who have just learned the present perfect tense. In the first year, not all students are true beginners. Some have had three or four years of high school German, but for whatever reason still place into first year German. The second group was second year students enrolled in the fifth and sixth quarters of German. The language requirement ends for most non-German majors after the fifth quarter, so there are fewer students in the sixth quarter. In the second year, there are students who have lived in a German-speaking country as children and placed into the second year. The third group was upper-division German students, i.e. students enrolled in third and fourth year courses, who had lived in a German-speaking country for
Carlee Arnett
less than six months or not at all. The fourth group was upper-division students who had lived in a German-speaking country for more than six months. A total of 45 first-year students took the survey in class in either the Fall or Winter quarter of the 2000–2001 academic year. By the second quarter of the first year, students have been introduced to the present perfect in the textbook. In Fall 2001, students used the textbook Deutsche Sprache und Landeskunde (Crean & Scott 1993). The grammar explanation in the book for auxiliary selection consists of two lists of participles that take either haben or sein. It offers no explanation about how to choose the auxiliary. In a note at the bottom of the list, the book states that fahren ‘drive’ occurs with haben when there is a direct object. Second quarter students in Winter 2002 used the textbook Neue Horizonte (Dollmeyer & Hansen 1999). This book states that most verbs use haben as their auxiliary, but that some use sein. In order to take sein, a verb must fulfill two conditions, 1) it must be intransitive, i.e. it cannot take a direct object and 2) it must indicate a change of location or condition (161ff.). The survey was administered to four groups and the information I gathered about the students is summarized in Table 2. The 45 first year students were a roughly equal mix of freshman, sophomores, juniors and seniors. There were 25 true beginners, 4 students with 1–2 years of high school German, 14 students with 3–4 years of high school German and 1 who had four years of high school German and a quarter at community college. None had lived in the German-speaking world for more than six months and none spoke German at home while growing up. There were 26 second-year students who took the survey. All 26 students were accurately placed in second year German, i.e. they had either taken first year at Table 2. The students First Year Second Year Upper-division: Upper-division: Less than six months More than six months Number of Students Men Women Average age True beginners 1–2 years HS 3–4 years HS Year abroad Residence abroad Spoke German at home Most comfortable using English
45 16 29 20 25 4 14 – 0 0 45
26 9 17 21 – 0 0 – 3 0 26
13 2 11 21 – 0 0 – 0 0 13
11 3 8 21 – 0 0 4 4 0 10
Transitivity parameters and auxiliary selection by L2 students of German
the university or they had taken the equivalent at community college or in high school. There were three students who had lived in Germany as children, one for 10 years, one for 8 and one for three. None of the students had spoken German in the home and all were most comfortable using English. In the upper-division, there were 13 students who had lived in a Germanspeaking country for less than six months. None of the students spoke German at home and all felt most comfortable using English. There were 11 students enrolled in upper-division courses who had lived in Germany for six months or more. Four had grown up primarily in Germany and come to the US for either high school or college. They had lived in Germany 17, 18, 13 and 9 years. There were no students who grew up in the US and spoke German at home. There were 4 students who had spent one year in Germany as exchange students. The remaining three had been on a short-term study or work abroad program for more than six months but less than a year. With the exception of one student who claimed to be equally comfortable using English and German, all students said they were most comfortable using English. The students were asked to write descriptions of two pictures in the present perfect. The first picture showed a boy riding his bicycle down the street without his hands on the handle bars. A woman is shouting something at the boy from the doorway of a house. The second picture shows two children walking down a street and then looking into the window of a toy store. The first picture is meant to get the students to use fahren either transitively, Er hat sein Fahrrad (ohne Hände) gefahren ‘He rode his bicycle (without hands)’ or intransitively Er ist (ohne Hände) gefahren ‘He rode without hands’. The second picture was expected to elicit verbs of motion, e.g. laufen ‘go’, that use sein.
. Accuracy of production The first year students produced a total of 308 clauses using the present perfect with an average of 9 clauses per student. One student did the task in the present tense in spite of the instruction to use the ‘ge-’ form. Of the 308 clauses, 215 occurred with haben, 80 with sein and 13 with either no participle or auxiliary. The students were 77% (166) correct with haben and 90% (72) with sein. There were a few students who made no mistakes in auxiliary selection, but most students would use sein in the first sentence and haben in the next sentence with the same participle. With three exceptions, the students were not aware about when to use haben or sein. Three students are able to do the task accurately and thus account for most of the right answers. Table 3 shows the accuracy rates of the first- and second-year students as well as the upper-division students.
Carlee Arnett
Table 3. Accuracy of production Haben Haben correct First year Second year Upperdivision: less than six months Upperdivision: more than 6 months
% Sein Sein correct correct
% Total clauses in % correct present perfect correct
215 137 82
166 119 70
77 86 85
80 78 28
72 64 28
90 82 100
308 215 110
77 85 89
57
57
100
42
42
100
99
100
The second year German students produced 268 clauses with an average of 10 clauses per student. One student did not use the present perfect because he didn’t know the forms. Of the 268 clauses, 215 were in the present perfect. 137 clauses occurred with haben and 78 with sein. Unlike the first year students, there were no students who omitted either the participle or the auxiliary. They did, however, conjugate the auxiliary incorrectly, have the wrong form of the participle and even write in the participle in English if they didn’t know the vocabulary word. The basic structure of the present perfect is there, even if in some cases not much else is. Students use haben correctly 86% of the time and sein correctly 82% of the time. Unlike the first year, where the students used haben correctly 77% of the time and sein correctly 90% of the time, when they used it, the accuracy rate is now roughly equal for the auxiliaries. Seven of the 26 students made no mistakes in auxiliary selection, but, like the first year students, most students used haben and sein interchangeably with the same participle. The third group was students who were enrolled in upper-division courses, but who have not lived in a German-speaking country for more than six months. These students produced 110 clauses with an average of 10 per student. The students were 85% correct with their use of haben and 100% correct with their use of sein. Students continue to become more accurate with respect to choosing the auxiliary. There are, however, still mistakes in the participles. Seven of the 13 students were completely accurate in both texts. The fourth group of students was enrolled in upper-division courses and had lived in a German-speaking country for more than six months. These students produce 134 clauses with an average of 12 per student. One student who was an exchange student for a year after high school wrote mostly in the present tense. The other students used haben 57 (100%) times correctly and sein 42 (100%) times correctly.
Transitivity parameters and auxiliary selection by L2 students of German
. Influence of parameters I am interested to see if students are sensitive to the transitivity parameters of event structure, telicity and punctuality. Event structure, i.e. a two or more participant, asymmetrical event, is the most basic parameter and it is also closest to the textbook rule. Telicity and punctuality might also be relevant and the students may indicate their awareness with adverbial phrases for goal or time. Furthermore, the students might point to these parameters in their answers about how they made their decisions if they are salient. Table 4 shows the accuracy rates for the four groups of students for each parameter. Table 4. Accuracy rates for parameters 100 90 80 70 60 50 40 30 20 10 0
First Year Second Year > 6 months < 6 months
# of participants
Telicity
Punctuality
If the student attempted to use the auxiliary + participle construction and failed in some way, either missing auxiliary or participle, I counted it as a present perfect. If a student did not use the present perfect at all, i.e. wrote in the present tense, I didn’t count it. There were a few students in the upper-division who wrote exclusively in the present tense. I did not include these students. . Number of participants The first year students produced 166 clauses that had two or more participants and the correct auxiliary haben, even if the chosen verb was fahren. Forty-nine clauses had one participant and the correct auxiliary should have been sein. Only two verbs that have one participant and take haben occurred in the corpus. There are two examples of gestanden ‘stand’ and both occur with haben. This is most likely the default use of haben as the auxiliary, rather than an awareness of low transitivity.
Carlee Arnett
There are 72 examples of sein used correctly with one participant. The errors by the students with sein include one each of ist . . . gestanden ‘stood’ (which is the correct auxiliary in the southern part of the German-speaking world), ist. . . geheist ‘is called’, ist. . . gespielen ‘was playing’, ist. . . ge[yelled]en ‘is yelling’, ist. . . gehälten ‘is stopping’. There is no way to know for sure why this type of error occurs, but it looks like a translation of the present progressive tense in English. Lastly, there is one example in which wollen ‘want’ is used as an auxiliary with a past participle The students use the auxiliary haben as the default auxiliary and do not seem to be sensitive to the number of participants in the clause (with one exception). A few students showed sensitivity to individuation of the object and could make the distinction between Er ist Rad gefahren ‘He was bicycling.’, Er ist Fahrrad gefahren ‘He rode a bicycle.’ and Er hat sein Fahrrad gefahren ‘He rode his bicycle.’. Most students, however, did not make this distinction at all and the few who got it right once had it wrong elsewhere in the text. In the second year, 119 clauses had two or more participants or were one participant verbs that take haben like stehen ‘stand’ or halten ‘stop’. There were a few students who used gestanden ‘stood’ and all used haben as the auxiliary. There were four students who used halten and three used it correctly with haben and one used sein. Like the first year students, the use of haben is probably the default rather than an awareness of low transitivity. The remaining clauses had one participant and the correct auxiliary should have been sein. There were eight students who used haben with gelaufen, three with gefahren, two with geritten, one with gegangen and one with Radgefahren. There are 64 examples of sein used correctly with one participant. The errors with sein include one each of ist. . . gestanden, ist gehießt ‘is called’ and seven instances in which Fahrad ‘bicycle’ is used as a direct object with the verb fahren, e.g. Er ist sein Fahrad gefahren. As mentioned above, ist..gestanden is a correct form in Southern Germany and the student may have heard this form from an instructor. Students also tend to use haben as the default and do not use sein unless they think they should. The ist. . . gehießt form looks like a translation of the English present progressive. In the second year, students have developed some awareness that the number of participants matters as an indication of transitivity. But there are as many instances of the incorrect ‘sein + direct object + fahren’ as there are the correct ‘haben + direct object + fahren’. There are many students who distinguish between fahren with a verb particle or direct object and maintain the accuracy on that point throughout the text, but there are just as many who do not. Two students account for six of the 11 errors where sein should be used instead of haben. These students use haben as the default auxiliary and never use sein. The other students use haben and sein with gerannt, gefahren, gekommen and gegangen interchangeably within the same text. The number of participants does not seem
Transitivity parameters and auxiliary selection by L2 students of German
to be a trigger for auxiliary selection for the two students who are still making errors at this level. In the third group, ten of the eleven students surveyed are completely accurate with respect to auxiliary selection. Like the nine students in the other upperdivision group, these students have constructed a model for auxiliary selection that works. When there are two participants in a clause with a verb of motion, the auxiliary is haben and when there is one participant, the auxiliary is sein. The students have learned that this parameter of transitivity is relevant for auxiliary selection. . Telicity The first year students produced 118 telic clauses. 68 occur with the correct auxiliary and 39 do not. Thirty-four of the thirty-nine incorrect clauses occur with haben, when sein should be used. Five telic clauses occur with sein when haben should be used. 45 of the 68 correct telic clauses occur with sein; 23 correct telic clauses occur with haben. Telicity seems to make no difference to the accuracy rates of auxiliary selection at this level. This is not entirely unexpected, since the most important parameter of number of participants is also not yet making a critical difference. In second-year German, 54 clauses were telic and 44 that occurred with sein were correct. The ten incorrect clauses can be described as follows. In one case the student wrote denn sie haben nach Hause gegangen mussen ‘then they had to go home’. In two cases, the students used sein with gestanden which is a correct form in parts of Germany. In five cases, the students wrote something like Das Mädchen hat auf die Straße gelaufen ‘the girl was walking on/down the street’. The last two cases have to do with the number of participants. One student writes Gunter ist. . . auf die Straße sein Fahrrad gefahren ‘Gunter was riding his bicycle down the street’ and misses that he needs to use haben when there are two participants. Another student writes Das Mädchen hat die Straße geläuft ‘The girl walked the street’. I can’t read minds so it is impossible to know if the second student thought that ‘street’ was a second participant and therefore used haben. To sum this up, the telic clauses were overwhelmingly correct with respect to auxiliary selection. The 13 upper-division students who had been in a German-speaking environment for six months or less produced a total of 30 telic clauses. Twenty-one clauses occurred with sein and nine clauses occurred with haben. Of the clauses with sein, all but one used the correct auxiliary. This accuracy rate shows the students’ overall tendency to be more accurate when using sein. The incorrect use of sein occurs, as in the second year, when there is a second participant, e.g. Ein Jung ist sein Rad ohne Hände gefahren ‘A boy was riding his bicycle without his hands’. Of the nine clauses that occur with haben, one is correct and eight are incorrect. Seven of the
Carlee Arnett
eight incorrect clauses use haben with a verb of motion. These clauses are produced by the same two students who always use haben in the present perfect. One of the incorrect clauses, Ein Kind hat mit einem Rad ohne die Hände gefährt ‘A child rode with a bicycle without hands’, may show haben because the student mistook mit einem Rad for a second participant. This error was not produced by one of the two students who always use haben. The fourth group of students produced 35 clauses with telicity markers. Although telicity is a property of high transitivity, its presence alone does not trigger the use of sein. The most important property remains the number of participants. There are 27 clauses that correctly occur with sein and eight clauses that correctly occur with haben. As with the other groups, telicity and the occurrence of sein tend to overlap, but telicity does not override the need to use haben if there is a second participant. . Punctuality The first year learners produce seven punctual clauses, a rate which is almost equal to the second year students and double that of both groups of upper-division students. Six of the clauses are correct and one is incorrect. Three clauses correctly occur with sein and three clauses correctly occur with haben. The one incorrect auxiliary is haben when sein should have been used. The first year students have a higher accuracy rate (85%) on punctual clauses than on the clauses that are both telic and punctual(57%). It is possible that the extra elements in the clause are confusing the students, but it is impossible to determine for certain what is going on. The model for auxiliary selection that the students have is still developing and they are still unsure where the number of participants, telicity and punctuality fit in. In the second year, there are eight clauses that are punctual. Six of the clauses occur with sein and two of the clauses occur with haben. The two clauses with haben show the correct auxiliary and five of the six clauses with sein are correct. The incorrect clause, Gunter ist sehr schnell sein Fahrrad gefahren ‘Gunther rode his bicycle very quickly’, should occur with haben because of the second participant. There are too few examples of punctuality to say what role it might play in the mind of the learner. Punctuality does correspond to accuracy in these examples, but non-marking does not correspond with inaccuracy. Further, the presence of a punctuality marker is not enough to trigger the correct auxiliary for the students. There are three punctual clauses from the third set of learners. Two of the clauses are correct and occur with sein and one is incorrect. The incorrect clause should occur with sein, but the student uses haben instead. As expected, this clause is produced by one of the students who consistently uses haben. Eleven of the thirteen students are very accurate with auxiliary selection, but two are not. Eleven
Transitivity parameters and auxiliary selection by L2 students of German
students have built a model for auxiliary selection that works. The two students that still use haben as a default are not responding to any of the linguistic cues offered by telicity, punctuality or the number of participants. Like the other learners, the fourth group produced far fewer punctual clauses than telic ones. There are three punctual clauses and all of them occur correctly with sein. Again, punctuality and the correct use of sein overlap. However, punctuality does not alone trigger the use of sein for the students and the number of participants is the most important parameter. Like the two students from the other upper-division group, there is one student who has no awareness of a system in the present perfect at all. It seems that in both upper division groups, auxiliary selection is either learned and the students are completely accurate or it is not learned at all to any degree. Thus these three students have not learned the principles of auxiliary selection, unlike the other second year students who are accurate in the 80 percent range. The second year students are actively creating a model and applying with a high degree of success. The three upper division students are unaware that there is a model at all. These three students have not understood that there is a conceptual distinction to be made and that it is obligatory. They have not learned this in the classroom from explicit teaching or from implicit learning while abroad. These students are out of step with other learners of German at the same level and those learners somewhat less advanced.
. Decision making After writing descriptions of the pictures, I asked the students to describe how they made the decision which auxiliary to use. I wanted to know if the students would a) provide the textbook rule and/or b) articulate any of the three transitivity parameters under consideration. In the first year, 24 students said that they used sein for motion verbs and haben for everything else or they had memorized the lists in the book. Thus roughly half of the first year students can recite the rule from the textbook. I can see from the data, however, that these students did not apply the rule or there would have been many more correct uses of sein. A summary of all students’ explanations and their knowledge of the rules is given in Table 5. Another common answer about the decision making process was the honest admission that they just used haben all the time. Some students added that they didn’t realize that they needed to use sein. A few students said that they used haben all the time and had forgotten about sein, but that it was used for verbs of motion. These students could recite the rule when prompted. A couple of students said that they made the decision by feeling or what sounded correct. This did not correlate with correct usage.
Carlee Arnett
Table 5. Explanations and rules first year
First year Second year > 6 months < 6 Months
Claimed to Claimed know rule not to know rule
Knew rule
Did not know rule
Decision like the textbook rule
Sounds Other right
30 17 8 9
25 14 5 6
15 13 3 3
24 19 7 5
0 4 3 3
13 10 5 1
10 3 3 2
It is not surprising that students are unable to critically evaluate their decision making process and I didn’t necessarily expect that. I wanted to see if students thought they were using the rule to make the decision (they did) and if that actually correlated with applying the rule (it didn’t). When asked specifically if they knew the rule, 30 students claimed knowledge. Thirteen students admitted that they didn’t know the rule, one did not respond and another wrote “kinda”. When asked to write out the rule, five pointed out that the use of sein was related to transitivity, i.e. the presence of a direct object. Twenty students wrote that sein was used for motion verbs and haben was for everything else. Thirteen students wrote nothing and two students wrote responses like “I get mixed up sometimes”. Of the five students who had a pretty accurate rule for German, none used it consistently or accurately when describing the two sets of cartoons. In the second year, we use two grammar books, German in Review (Sparks & Vail 2000) and Kaleidoskop (Moeller, Mabee, & Liedloff 1998). Both German in Review (2000: 40ff.) and Kaleidoskop (1998: 253) state that verbs using sein must meet two conditions, 1) the verb is intransitive and 2) the verb expresses motion or a change of condition. Most of the students who claimed to know the rule actually did know the rule, but this knowledge did not translate into accuracy in the text. Four students said that they used “what sounded right” and three relied on memorization. These two strategies also did not produce accurate results with consistency. The remaining answers were similar to the first year students’ with comments like “I didn’t think of using sein”, “haben is for possession and sein is for state of being”, “haben is used with walked or ran and sein with a change of state”, “some take haben and some take sein”, and “I wasn’t sure what counts as motion, i.e. writing”. These last statements indicate confusion on the part of the student and this needs to be uncovered and rectified. When asked in the third group if they knew the rule, eight claimed to and five admitted they did not. These students have all learned German in the classroom, so I would hope that they had access to some kind of rule. Of the eight that claimed to know the rule, five actually did and three did not.
Transitivity parameters and auxiliary selection by L2 students of German
When asked to explain how they made their decision between haben and sein, seven wrote an explanation that resembled the rule. The seven students who were able to explain their decision in a way that resembled the rule were not the seven students who were able to use the auxiliaries accurately. Two of the seven accurate students claimed not to know the rule and one wrote as an explanation that she used what sounded right and the other claimed that she did not know what the present perfect was, so she did not use it at all. Two additional students stated that they used what sounded right and the remaining three answers were unique. In the fourth group, the students had learned German primarily outside the classroom. Thus I expected that some would not know the rule. Nine claimed to know the rule, one claimed not to and one left it blank. Of the nine who claimed to know the rule, six were correct and three were not. When asked to explain how they made their decision, five gave an explanation that resembled the rule and three said they used what sounded right. Overall, there were a variety of interesting responses. One student claimed not to have made a distinction between the two auxiliaries, even though in the two texts he did. Another student gave an insightful response. She wrote that “I tried to think about transitive and intransitive verbs – when there was physical movement and when not. Of course, the examples poorly reflected movement, so I guessed frequently”. She is trying to apply the rule that verbs of motion take sein, but she is unable to determine what exactly constitutes motion. The inability to determine what constitutes motion was also a difficulty pointed out by a second year student. In spite of being unable to articulate the decision-making process about auxiliary selection, all students but one in the fourth group were able to do it accurately.
Conclusion As students learn German, they get more accurate at selecting the correct auxiliary. Students who have spent more than six months in a German-speaking country are completely accurate with the exception of the one student who used the present tense. Students who are upper-division but lived in a German-speaking country for less than six months are accurate to a high degree with the exception of two students who always used haben. Except for three students all others at this level have acquired auxiliary selection either from classroom experience or time abroad. The second year students, who only have experienced German in the classroom, have also acquired auxiliary selection, if accuracy above 80% is considered the benchmark for acquisition. This high accuracy rate for the second year shows that students do learn in the classroom either by explicit instruction or processing input. The first year students have not acquired auxiliary selection yet, given that their overall accuracy rate is 66%.
Carlee Arnett
The most relevant transitivity parameter for students is the number of participants. Their accuracy rates rise as they are able to determine if there are one or two (or more) participants in a clause. Given that all clauses will have one or two (or more) participants, the parameters of telicity and punctuality are additional cues for the students. First year students seem to be confused when telicity is added as an element. Perhaps they think it is a participant, if telicity is indicated by a prepositional phrase. Second year students and upper-division students are as accurate with telicity as with the number of participants. Punctuality leads to greater accuracy for all students except those who have lived in Germany for less than 6 months. The small number of examples makes it impossible to generalize about why this is the case. In order to see if the students would name any of the transitivity parameters, I asked them to state how they made their decision about which auxiliary to use. Although the students provided many interesting answers, none referred to the transitivity parameters. This does not mean that students are unaware of them, but they are unable to express this awareness on a meta-linguistic level.
Acknowledgements I would like to thank Robert Blake for his comments on an earlier draft of this paper, three anonymous reviewers and Susannah Martin for help collecting and coding the data.
Transitivity parameters and auxiliary selection by L2 students of German
Appendix A. Questionnaire Part A
In German and using the present perfect tense (aka the ge- form), please describe what happened in the cartoons above. [space for students to answer]
Carlee Arnett
In German and using the present perfect tense (aka the ge- form), please describe what happened in the cartoons above. [space for students to answer]
Student Questionnaire It would be greatly appreciated if you would please take the time to complete this questionnaire. By answering the questions and returning this questionnaire, it is assumed that you have given your consent for the information you provide here to be used in this study and that you are over the age of 18. All participants will remain completely anonymous. Thank you very much for your participation and cooperation!
Transitivity parameters and auxiliary selection by L2 students of German
Part B 1. 2.
In what level of German are you currently enrolled? Did you have any German classes prior to your enrollment here at UC Davis? ____________ No. Please proceed to Question 3. ____________ Yes. Please write the approximate period of time (1 year, 4 months, etc.) on the line preceding the situation in which your previous classes took place. __________________ Elementary School __________________ Middle School __________________ High School __________________ Other University __________________ Other
3.
Have you had any non-classroom German experiences? ___________ No. Please proceed to Question 4. ___________ Yes. Please write the period of time on the line preceding the type of experience you have had. __________________ Short term exchange program. __________________ Contact with family and/or friends __________________ Semester or year abroad __________________ Travel abroad on vacation __________________ Pen-pal in a German-speaking country. Other:__________________________________________________________________ ________________________________________________________________________
4. 5.
How much German do you plan on taking at the University level? Please indicate if you intend on minoring or majoring in German. What is your class standing? ___________ Freshman ___________ Sophomore ___________ Junior ___________ Senior ___________ Graduate Student ___________ Non-Degree Seeking
6. 7.
How old are you? _______________ What is your gender? ___________ Female ___________ Male
8.
What did you do the year before you began course work at this university? ___________ Attended high school ___________ Attended another college or university ___________ Worked ___________ Military service ___________ Lived or traveled abroad
Carlee Arnett
Other:__________________________________________________________________ _________________________________________________________________________ 9. 10. 11. 12.
What is your major?_______________________________________________________ What is your minor?_______________________________________________________ What is your intended profession?____________________________________________ In what language are you most proficient and what language do you feel the most comfortable using? ________________________________________________________________________
13. If you have not lived in the United States most of your life, how long have you been here? Where have you spent most of your life so far? ________________________________________________________________________ Now please tell us a bit about your written responses! It is imperative that you DO NOT refer back to the previous pages and that you DO NOT change any of your answers. 1.
Please explain how you decided when to use haben and when to use sein in the present perfect tense. ________________________________________________________________________ ________________________________________________________________________ ________________________________________________________________________ ________________________________________________________________________ ________________________________________________________________________
2.
Do you remember the grammatical rule regarding the use of haben or sein in the present perfect tense? Please circle your answer: NO YES
3.
If you answered YES in the previous question, please explain the rule: ________________________________________________________________________ ________________________________________________________________________ ________________________________________________________________________ ________________________________________________________________________ ________________________________________________________________________
References Bard, E. G., Robertson, D., & Sorace, A. (1996). A magnitude estimation of linguistic acceptability. Language, 72, 32–68. Burzio, L. (1986). Italian Syntax: A government-binding approach. Dordrecht: Reidel. Centineo, G. (1996). A lexical theory of auxiliary selection in Italian. Probus, 8, 223–271. Cortés, C. (1993). Catalan participle agreement, auxiliary selection and the government transparency corollary. Probus, 5, 193–240. Crean, J. & Scott, M. (1993). Deutsche Sprache und Landeskunde (4th ed.). New York, NY: McGraw Hill. Dollmeyer, D. & Hansen, T. (1999). Neue Horizonte (5th ed.). Boston, MA: Houghton Mifflin.
Transitivity parameters and auxiliary selection by L2 students of German
Eisenberg, P. (1986). Grundriß der deutschen Grammatik. Stuttgart: J. B. Metzlersche Verlagsbuchhandlung. Fagan, S. M. B. (1988). The unaccusative hypothesis and a reflexive construction in German and Dutch. In E. H. Antonsen & H. H. Hock (Eds.), Germanic Linguistics II (pp. 21–34). Bloomington, IN: Indiana University Linguistics Club. Fagan, S. M. B. (1992). The Syntax and Semantics of Middle Constructions. A study with special reference to German. Cambridge: CUP. Fagan, S. M. B. (1996). Tests for unaccusativity revisited. Paper presented at the Germanic Linguistics Annual Conference II, Madison WI. Haider, H. (1985). Von sein oder sein: Zur Grammatik des Pronomens sich. In W. Abraham (Ed.), Erklärende Syntax des Deutschen (pp. 223–254). Tübingen: Narr. Hoekstra, T. (1984). Transitivity: Grammatical relations in government-binding theory. Dordrecht: Foris. Hoekstra, T. & Mulder, R. (1990). Unergatives as copular verbs: Locational and existential predication. The Linguistic Review, 7, 1–79. Hopper, P. & Thompson, S. (1980). Transitivity in grammar and discourse. Language, 56, 251– 299. Klein, W. (2000). An analysis of the German perfekt. Language, 76 (2), 358–382. Langacker, R. W. (1982). Space grammar, analysability, and the English passive. Language, 58 (1), 22–80. Lieber, R. & Baayen, H. (1997). A semantic principle of auxiliary selection in Dutch. Natural Language and Linguistic Theory, 15, 789–845. Levin, B. & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax-semantics interface. Cambridge, MA: The MIT Press. Lois, X. (1990). Auxiliary selection and past participle agreement in Romance. Probus, 2 (2), 233–255. Moeller, J., Adolph, W. R., Mabee, B., & Liedloff, H. (1998). Kaleidoskop. Kultur, Literatur und Grammatik (5th ed.). Boston, MA: Houghton Mifflin. Perlmutter, D. M. (1978). Impersonal passives and the unaccusative hypothesis. Berkeley Linguistic Society, 4, 157–189. Perlmutter, D. M. (1989). Multiattachment and the unaccusative hypothesis: The perfect auxiliary in Italian. Probus, 1, 63–199. Pustejovsky, J. & Busa, F. (1995). Unaccusativity and event composition. In M. Bertinetto, V. Bianchi, J. Higgenbotham, & M. Squartini (Eds.), Temporal reference, aspect and actuality (pp. 159–178). Turin: Rosenburg and Selier. Rice, S. A. (1987). Towards a Cognitive Model of Transitivity. PhDdissertation, University of California at San Diego. Rosen, C. (1988). The Relational Structure of Reflexive Clauses: Evidence from Italian. New York, NY: Garland. Shannon, T. F. (1987). On some recent claims of relational grammar. Berkeley Linguistics Society, 13, 247–262. Shannon, T. F. (1990). The unaccusative hypothesis and the history of the perfect auxiliary in Germanic and Romance. In H. Andersen & E. F. K. K. Koerner, (Eds.), Current Issues in Linguistic Theory (pp. 461–499). Amsterdam: John Benjamins. Shannon, T. F. (1992). Split intransitivity in German and Dutch: Semantic and pragmatic parameters. In R. Lippi-Green (Ed.), Recent Developments in Germanic Linguistics (pp. 97– 114). Amsterdam: John Benjamins.
Carlee Arnett
Shannon, T. F. (1993). To be or not to be in Dutch: A cognitive account of some puzzling perfect auxiliary phenomena. In R. Kirsner (Ed.), The Low Countries and Beyond (pp. 85–96). Landam, MD: University Press of America. Shannon, T. F. (1995). Explaining perfect auxiliary variation: Some modal and aspectual effects in the history of Germanic. American Journal of Germanic Linguistics and Literatures, 7 (2), 129–164. Smith, A. K. (2001). The role of frequency in the specialization of the English anterior In J. Bybee & P. Hopper (Eds.), Frequency and Emergent Grammar (pp. 361–382). Amsterdam: John Benjamins. Sorace, A. (1992). Lexical Conditions on Syntactic Knowledge: Auxiliary selection in native and non-native grammars of Italian. PhD dissertation, Edinburgh: University of Edinburgh. Sorace, A. (1993). Incomplete vs. divergent representations of unaccusativity in native and nonnative grammars of Italian. Second Language Research, 9, 22–47. Sorace, A. (1995). Acquiring argument structures in a second language: The unaccusative/ unergative distinction. In L. Eubank, L. Selinker, & M. Sharwood Smith (Eds.), The Current State of Interlanguage Amsterdam: John Benjamins. Sorace, A. (2000). Gradients in auxiliary selection with intransitive verbs. Language, 76 (4), 859– 890. Sparks, K. & Vail, V. H. (2000). German in Review (3rd ed.). Forth Worth, TX: Holt, Rinehart and Winston. Zaenen, A. (1993). Unaccusativity in Dutch: Integrating syntax and lexical semantics. In J. Pustejovsky (Ed.), Semantics and the Lexicon (pp. 129–161). Dodrecht: Kluwer. Zifonun, G., Hoffmann, L., & Strecker, B. (1997). Grammatik der deutschen Sprache. Band 3. Berlin: Walter de Gruyter.
An irrealis BE auxiliary in Romanian* Larisa Avram and Virginia Hill University of Bucharest / University of New Brunswick, Saint John
Tests of distribution, word order and interpretation of clauses with the tense/ perfective auxiliary fi ‘be’ in Romanian show that: (i) there is a morpho-syntactic distinction between fi ‘be’ as auxiliary and fi ‘be’ as a copula for inflecting unaccusatives and passives; and (ii) fi ‘be’ occurs in contexts with non-specific time frame (modal clauses) and irrealis interpretation, and so it is in complementary distribution with am, ai, a ‘have’, which appears in contexts with definite time setting (indicative clauses) and realis interpretation. Crucially, the analysis situates the dichotomy of the ‘be’/‘have’ auxiliaries in Romanian on the axis irrealis vs. realis, which differs from the ergativity axis in Romance.
.
Introduction
In some Romance languages (e.g., French, Italian) the Latin esse ‘be’+ past participle yield periphrastic perfect paradigms that alternate with ‘have’ + past participle. This alternation depends on the semantic properties of the lexical verbs; for example, unaccusatives and/or unergatives may select ‘be’ (Burzio 1986; Perlmutter 1978). Romanian has lost this morphological procedure at an early stage, and has generalized the ‘have’ + past participle formula to all the verbs in the periphrastic past tense, as shown in (1).1 * We thank all the reviewers for their helpful comments. This paper has been partly supported by the SSHRC of Canada grant 410-2003-0608 (Hill). . The infinitive form for ‘have’ is avea. However, the inflectional forms of the lexical verb are generally different from the auxiliary forms, as in (i). (i)
Ion are o carte.// Ion a plecat. Ion has a book // Ion has-aux left
In order to avoid ambiguities, we refer to the auxiliary as am, ai, a as equivalent to the auxiliary ‘have’ in Romance.
Larisa Avram and Virginia Hill
(1) a.
Maria a c˘azut pe scar˘a. Maria has fallen on stairs ‘Maria has fallen down the stairs.’ b. Maria a râs toat˘a seara. Maria has laughed all evening ‘Maria laughed the whole evening.’
Both unaccusatives (1a) and unergatives (1b) take the auxiliary am, ai, a ‘have’ in Romanian. Although the generalization of ‘have’ to the periphrastic past paradigm is quite common in the Romance group (e.g., Canadian dialects of French), Romanian stands out through the use of fi as in (2) and (3). In (2) the morpheme fi seems to occur in free alternation with a ‘have’.2 (2) Maria a plecat.// Maria e plecat˘a. Maria has left// Maria is left-fem.sg. ‘Maria has left.’//‘Maria has left.’
Contrasting with the apparent randomness of (2), the examples in (3) have fi as the only option in non-indicative contexts, irrespective of their [+/– perfective] reading.3 (3) a.
Nu cred s˘a fi plecat not believe-1sg sa be left ‘I don’t think s/he/they may have left.’ b. Ar fi plecat. would-3sg be left ‘S/he would have left.’ c. Va fi plecat. will-3sg be left ‘S/he will have left.’ d. O fi plecat. may-3sg be left ‘S/he may have left.’ e. O fi dormind. may-3sg be sleeping ‘S/he may be sleeping.’ f. Vei fi vrând tu s˘a pleci. . . will-2sg be wanting you sa leave ‘You may well want to leave. . .’
. Gramatica Academiei treats the constructions in (2) as basically the same, with two auxiliary options. . The morpheme s˘a is a subjunctive mood marker and appears as SA in glosses.
An irrealis BE auxiliary in Romanian
This paper investigates the syntactic conditions which lead to the distribution and the interpretive impact of fi in (2) and (3). So far, the formal studies on fi focused on the morphological status of this element, which is defined either as an invariable affix (Avram 1999) or a clitic (Abbeillé & Godard 2000; Pîrvulescu 2002). The formal syntactic aspect has been less explored (cf. Avram 1994, 1999). For example, no study has accounted for the following: (i) clauses may not begin with fi-V, although they may begin with am-V; (ii) fi combines with gerunds as well as past participles; (iii) fi in matrix clauses has variable scope; (iv) fi is the only perfective auxiliary in contexts with an irrealis interpretation. These four properties of fi are at the core of the analysis developed in this paper. Crucially, our proposal is that there are two distinct grammaticalizations of fi in Romanian: one is the copula fi that occurs in constructions such as (2); the other one is the auxiliary fi that occurs in constructions such as (3). Only the auxiliary fi yields a systematic alternation with am, ai, a (i.e., mirroring the be/have alternation in Romance). However, this alternation is contingent on the context in which the auxiliary occurs (basically, on the contrast between a realis versus an irrealis interpretation), and not on the lexical properties of the supporting verb. At a theoretical level, the grammaticalization of Lat. esse ‘be’ into the Romanian auxiliary fi reveals a morpho-syntactic mechanism for encoding the expression of irrealis, an interpretational property belonging to the discourse/sentence semantics. This use of esse ‘be’ differs from its role in the other Romance grammars, where it encodes information about the sub-categorization properties of the selectors and their argument structure.
. Fi and unaccusatives This section will focus on constructions as in (2), resumed for convenience, where fi seems to alternate with am, ai, a in the same inflectional context. (2) Maria a plecat.// Maria e plecat˘a. Maria has left// Maria is left-fem.sg. ‘Maria has left.’//‘Maria has left.’
A series of tests compare fi in contexts similar to (2) with the passive fi (that we label “copula ‘be”’) and with the auxiliary am, ai, a. The systematic trend is that fi in (2) patterns with passive fi, not with the auxiliary am, ai, a.
Larisa Avram and Virginia Hill
. Morphologic status All Romanian auxiliaries are bound morphemes (affixes or clitics), so that no constituents may intervene between an auxiliary and the verb (Dobrovie-Sorin 1994). The auxiliary am, ai, a observes this rule, as shown in (4a), while the copula ‘be’, such as the passive, does not, as in (4b). Fi with unaccusative verbs as in (2) patterns with passive fi, as shown in (4c). (4) a.
A (*Maria) plecat (Maria) de mult. has Maria left Maria of long ‘Maria has left long ago.’ b. E (Maria) invitat˘a (Maria) peste tot. is Maria invited-fem.sg Maria all over ‘Maria is invited all over.’ c. E (Maria) plecat˘a (Maria) de mult. is Maria gone-fem.sg Maria of long ‘Maria has left long ago’/‘Maria has been gone for awhile.’
The test in (4) shows that the post-verbal subject may intervene between fi and the past participle, but not between the auxiliary a and the past participle. Since fi in (4c) behaves as a free morpheme, it is excluded from the class of auxiliaries, on a par with passive fi.4 . Tense morphology The examples in (2) show am, ai, a alternating with fi in the same context, as far as both morphemes display inflection for present tense. However, fi can take other inflectional forms (e.g., to express perfectivity in the past, present and future), while am, ai, a cannot: . The clitic status of am, ai, a versus the unbound fi in (2) is further confirmed by tests with past participle coordination. In such situation, the auxiliary must be repeated (i), while fi (ii) and the passive copula (iii) are not repeated. (i) *Am întâlnit-o s¸ i întrebat-o have met-her and asked-her (ii) E mâncat˘a s¸ i dormit˘a. is eaten-fem.sg. and slept-fem.sg. ‘She has eaten and slept.’ (iii) E invitat˘a s¸ i consultat˘a de mult lume. is invited-fem.sg. and consulted-fem.sg. by many people ‘She is invited and consulted by many people.’
An irrealis BE auxiliary in Romanian
(5) a.
Am plecat.//*Aveam plecat. have-1sg/pl left//had-1sg/pl left ‘I/We have left.’//Intended: ‘I/We had left.’ b. E plecat.// Era plecat. is left // was left ‘S/he has left.’// ‘S/he had left.’ c. E chemat.// Era chemat. is called // was called ‘S/he is called.’// ‘S/he was called.’
Again, fi in (2) patterns with passive fi (that can be inflected for various tenses), and not with the auxiliary am, ai, a. . Past participle agreement Further proof that fi in (2) is not the counterpart of the tense/aspectual auxiliary am, ai, a comes from the agreement contrast in (6). Left movement of objects does not trigger object agreement on the past participle of active verbs, as seen in (6a). Such agreement appears only in passive forms, as in (6b). The agreement is also obligatory in constructions like (2), as further shown in (6c). (6) a.
Fetele au plecat/*plecate de mult. girls-the have left-masc.sg/left-fem.pl of long ‘The girls have left long ago.’ b. Fetele sunt invitate/*invitat peste tot. girls-the are invited-fem.pl./invited-masc.sg. all over ‘The girls are invited all over.’ c. Fetele sunt plecate/*plecat demult. girls-the are left-fem.pl./left-masc.sg of long ‘The girls have left long ago’./’The girls have been gone for a long time.’
In (6) the verb pleca ‘leave’ is unaccusative. Thus, the argument in subject position originates as the object of the verb. Object to subject movement triggers object agreement in (6b) with passive fi, and in (6c), with fi, but not in (6a), with the tense/aspectual auxiliary au. Therefore, (6) indicates that fi in (2) is compatible with a structural pattern similar to passive constructions, and different from active constructions with tense/aspectual auxiliaries. . Past participle and adjective coordination The test in (7) shows that past participles of passive constructions may enter in coordination with adjectives (7a). Such coordination is ruled out when unaccusative
Larisa Avram and Virginia Hill
verbs carry auxiliaries (7b), but not when they carry fi (7c). Note that both the participial form and the adjective may take the adjectival modifier foarte ‘very’.5 (7) a.
E (foarte) felicitat˘a s¸ i (foarte) fericit˘a c˘a s-a terminat. is very congratulated-fem.sg. and very happy-fem.sg. that refl-3sg has ended ‘She is much congratulated and very happy that all is over.’ b. *A (foarte) plecat s¸ i (foarte) fericit dup˘a aceea has very left and very happy after that Intended: ‘S/he left and was happy after that.’ c. E (foarte) plecat˘a cu sorcova s¸ i (foarte) z˘ap˘acit˘a. is very gone-fem.sg with branch-the and very confused-fem.sg ‘She’s very scattered and very confused.’
The past participle and adjective coordination in (7) clearly indicates that fi in constructions as in (2) belongs to a passive structure compatible with a participial/adjectival pattern, different from the structure of constructions with tense/aspectual auxiliaries. The modifiers indicate that both the participial form and the adjective share categorial features that do not extend to the past participle in the present perfect form with the auxiliary am, ai, a. To sum up the tests in this section, fi in constructions such as (2) is not a morpho-syntactic counterpart of the auxiliary am, ai, a. On the contrary, fi in (2) systematically patterns with the passive copula fi. Therefore, what looks like a free alternation in (2) is actually an option between two different structures, and not an option between equivalent elements in the same structure. Since the objective of this paper is the morpheme fi with auxiliary properties (i.e., the morpho-syntactic equivalent to am, ai, a), the copula fi in constructions with (2) is irrelevant to our analysis, and will not be dealt with any further.
. The auxiliary fi This section offers the core analysis of fi with auxiliary function. It is argued that the auxiliary fi, as a bound morpheme, is inserted in the clitic cluster on verbs inflected for [mood]. The clitic property and the distribution in past/perfective contexts qualifies fi as the counterpart of am, ai, a. The distribution of the auxiliary fi helps to identify the feature cluster it carries, that is, [tense/perfective] and
. The modifier foarte ‘very’ is restricted to adjectives and does not occur with verbs; e.g., foarte frumos ‘very beautiful’ vs. *foarte plec ‘very go-1sg’. We thank Raul Aranovich and an anonymous reviewer for suggesting this test.
An irrealis BE auxiliary in Romanian
[irrealis]. The range of syntactic structures that ensure checking for these features may yield three interpretations: irrealis in the past; irrealis; anteriority. . The auxiliary status This section shows that the morpheme fi in constructions as in (3) qualifies as an inflectional auxiliary. More precisely, the tests in (8) indicate that in examples such as in (3) fi patterns with the tense/aspectual auxiliary am, ai, a. The definition of fi as a bound morpheme with auxiliary function stems from a series of tests proposed in recent formal studies (Avram 1999; Dobrovie-Sorin 1994, a.o.). In sum, a construction like (3a), resumed below, displays an auxiliary fi that cannot be separated from the verb (8a), has to be repeated under coordination (8b), and is not compatible with past participle agreement (8c). (3) a.
Nu cred s˘a fi plecat. not believe-1sg. sa be left ‘I don’t believe that s/he/they has/have left.’
(8) a.
Nu cred s˘a (*Maria) fi (*Maria) plecat (Maria). not believe-1sg sa Maria be Maria left Maria ‘I don’t believe Maria to have left.’ b. Nu cred s˘a fi plecat s¸ i *(s˘a) *(fi) uitat de tine. not believe-1sg sa be left and sa be forgotten about you ‘I don’t believe that I/we/s/he/they has/have left and forgot about you.’ c. Fetele nu cred s˘a fi plecat/*plecate. girls-the not believe-1sg sa be left-masc.sg/left-fem.pl. ‘I don’t believe the girls to have left.’
The example in (8a) shows that fi cannot be separated from the verb by phrasal constituents. A similar restriction has been mentioned for the auxiliary a in (4a). Obligatory repetition of fi in (8b) places it in the same class with other bound morphemes in the inflectional paradigm (e.g., the subjunctive mood marker s˘a) that cannot appear by themselves. Finally, lack of past participle agreement in the presence of movement from object position, as in (8c), indicates that the structure containing fi is similar to a past tense structure with the auxiliary au in (6a). Therefore, the tests in (8) show that fi in (3) and am, ai, a belong to the same class of auxiliaries. Hence, why can’t they alternate, in the context of (3), in the way they do in other Romance languages? . The distribution of the auxiliary fi This section discusses further constructions as in (3), where fi is the only possible aspectual auxiliary. Two contexts may be distinguished in (3): one in which fi com-
Larisa Avram and Virginia Hill
bines with a past participle (3a–d), and one in which fi combines with a gerund6 (3e–f). In both cases fi ‘be’ appears embedded under a M(ood) head and it is the only selectional choice for a periphrastic form, to the exclusion of am, ai, a ‘have’. The examples in (9) list the verbal moods that require fi for their anteriority and/or perfective paradigm: (9) – Subjunctive: a. θsi dore¸ste [s˘a fi cump˘arat o cas˘a.] refl-dat wishes sa be bought a house ‘S/he wishes s/he had bought a house.’ – Conditional: b. [Ar fi cump˘arat casa] [dac˘a avea bani.] would-3sg.cond be bought house-the if had-3sg money ‘S/he would have bought the house if s/he had had money.’ – Presumptive: c. [O fi cump˘arat casa] [dac˘a spune a¸sa.] will-3sg be bought house-the if says so ‘S/he may have bought the house if s/he says so.’ d. [Vei fi vrând tu [s˘a pleci. . .]] will-2sg be wanting you sa leave-2sg.subj ‘You may well want to leave. . .’ – Future: e. [Va fi cump˘arat casa] [când ajungi tu acolo.] will-3sg be bought house-the when get-2sg you there ‘S/he will have bought the house by the time you get there.’ – Infinitive: f. A murit [înainte de a-¸si fi cump˘arat o cas˘a.] has died before of to-inf refl-dat be bought a house ‘He died before having bought himself a house.’
In (9a) and (9b) fi feeds an irrealis interpretation: the act of buying has not taken place. In (9c–f) the truth value of the fi-clause is left unspecified at the time of utterance, and there is no prediction on whether the event denoted by the predicate is true or false. From this point of view, (9c–f) are comparable to the irrealis complements in Boškovi´c (1997), of the type John tried to talk to Peter (did he or did he not talk to Peter?). Along these lines, the presence of fi in (9a–f) coin-
. We use the term gerund, though the status of the -ind verbal form has been subject to debate in Romanian linguistics. Etymologically related to the Latin gerund, the Romanian -ind form actually behaves like a present participle, having exclusively [+verbal] features (Edelstein 1972).
An irrealis BE auxiliary in Romanian
cides with an irrealis interpretation, where the term irrealis covers possibility, doubt, non-factuality. In sum, the examples in (9) indicate that the distribution of the auxiliary fi is mapped according to the semantic/pragmatic properties of the structure. That is, only inflectional forms allowing for an irrealis interpretation may form their past/perfective paradigm with fi. . The inflection of the auxiliary fi This section compares the morpho-syntax of am, ai, a with the morpho-syntax of the auxiliary fi. While both auxiliaries carry past/perfective features and merge with perfective forms, they contrast in all the other features and display complementary distribution. Subject agreement is a first contrastive feature: while am, ai, a appears always inflected, fi is invariable: (10) a.
A plecat.//Au plecat. has left//have-3pl. left ‘S/he has left.’//‘They have left.’ b. Voiam s˘a fi plecat. wanted-1sg/pl.impf sa be left ‘I would have liked to have left.’/‘I would have liked for you/s/he/them to have left.’
A second contrast appears in the temporal value: the event of fi-constructions as in (10b) has a dependent, non-specified time reference, whereas am-constructions as in (10a) are temporally independent and specified for a particular tense value. Thus, only fi may merge with a gerundive form, as in (10c), that has a nonperfective value (i.e., the event time encompasses the reference time). (10) c.
Oi fi vrând tu s˘a mergi dar nu se poate. would-2sg.fut be want-ger you sa go-2sg.subj but not refl-se can-3sg. ‘You might want to go there but it is not possible.’
The third contrastive property concerns the auxiliary placement in the inflectional string. In this respect, fi but not am, ai, a appears obligatorily under mood markers. In (11) the only bound morpheme that can appear first or alone in the clitic cluster on the verb is a (11a), not fi (11b). (11) a.
A spus c˘a vine// spunea c˘a a venit. has said that comes/ said-3sg.impf that has come ‘S/he said s/he will come.// s/he said s/he has come.’
Larisa Avram and Virginia Hill
b. *fi spus c˘a vine//*spunea c˘a fi venit. be said that comes/ was saying that be come Intended: ‘s/he would have said s/he’s coming’/’s/he said that s/he would have come.’
The list in Table 1 sums up the inflectional properties of the auxiliaries fi and am/ai/a. Table 1. Property
fi
avea
merge with a perfective form merge with a non-perfective form subject agreement tense inflectional embedding
+ + – – +
+ – + + –
. A syntactic structure for the auxiliary fi The contrastive features and distribution of the auxiliaries fi and am, ai, a as listed in Table 1 are indicative of the structural configuration in which they occur. This section proposes a structural representation that yields the realis vs. the irrealis interpretation in past/perfective contexts. Irrespective of perfectivity, the main ingredient for a realis vs. irrealis interpretation is the choice of inflectional mood. Thus, indicative is conducive to a realis reading, while non-indicative moods (e.g., subjunctive, conditional, infinitive, gerundive) may trigger an irrealis interpretation. In Romanian, the interpretive impact of non-indicative moods are reinforced by the presence of the specialized perfective auxiliary fi. This polarization of forms according to the interpretation applies systematically, as can be seen from (9) to (11). In examples like (10a, b) the past/perfective background is uniform, but the auxiliary responsible for this value is different, and it occupies different positions in the clitic cluster: fi is obligatorily embedded under a mood marker, while am, ai, a is cluster initial, in indicative verbs. Hence, fi and am, ai, a are intrinsically equivalent auxiliaries, but the free riding features they allow during the syntactic checking (e.g., agreement, tense, modality) are in complementary distribution. The functional complementarity of the two auxiliaries takes the configurational pattern in (12).
An irrealis BE auxiliary in Romanian
(12)
a. Indicative clause = realis TP T+Agr
AspP
am
Asp
PastPart
b. Modal clause = irrealis MP M(+Agr) sa/as æ ,
T/AspP T/Asp fi - PastPart
The configurations in (12) show that both auxiliaries encode tense compatible with perfectivity and merge under T, selecting an Asp(ect) Phrase for checking. Their final position in the inflectional string follows from the nature of the higher functional heads with features that need checking against these auxiliaries. T/Agr in (12a) is non-lexical and attracts the auxiliary-V string, so am appears first in linear order. On the other hand, M/Agr in (12b) is always lexical in Romanian, containing an affix mood marker that attracts the auxiliary-V string for proclisis. Thus, fi may only be second in the clitic cluster. Implicit in the analysis in (12) is the parasitic nature of agreement features, as proposed since Chomsky (1995). Generally, agreement is considered to cluster with tense, because tense has compatible nominal features (D-features). However, M(ood) in languages with mood markers also presents the properties of a head with D-features. There are two arguments for assigning M a D-feature, based on the data in (13). (13) a.
Voia [ca Maria s˘a fi plecat mai repede.]// E mare [ca Maria.] wanted-3sg.impf that Maria sa be left earlier// is big as Mary ‘S/he wished Maria had left earlier.// S/he is as big as Mary.’ b. Ar pleca imediat. would-3sg.cond go immediately ‘S/he would go immediately.’
Larisa Avram and Virginia Hill
First, (13a) shows that an element with nominal features such as ca ‘that’ may equally select a subjunctive MP-complement or a DP, and gets its D-feature checked in both configurations. Second, some M-heads, such as the conditional in (13b), display morphological markers for agreement.7 Therefore, examples like (13b) show that [agr] features may cluster not only with [tense], but also with [mood], presumably when the tense feature is not strong enough to ensure free riding and checking for [agr] features. This is the morpho-syntactic justification for the redistribution of [agr] features as in (12), on heads with D-features and strong inflectional features that trigger a Spec-head configuration. If we adopt the distribution of features as in (12), then fi is not involved in the checking of [agr] features, and its morphological form is invariable.8 On the other hand, am, ai, a moves to the head with [agr] features, and it will always display subject agreement. The syntactic contrast between (12a) and (12b) reflects the interpretive contrast between realis and irrealis. A realis event can be analyzed as syntactically marked for Tense/Agr because the event argument is fully identified and/or specified, as in the finite indicative clauses represented in (12a). Conversely, whenever the event denoted by the lexical verb is intended as irrealis (i.e. as indefinite and non-specific), T has a weak [–tense] feature that cannot ensure free riding and checking for [agr]. Therefore, fi appears as a marker of temporal non-specificity, marking the event variable as [–definite] and [–specific], in contrast to am, ai, a, which is a marker of temporal specificity. So, fi takes temporal interpretation from its environment (just like indefinite NPs), whereas am, ai, a carries all the features necessary for its definite temporal interpretation. As a result, fi-constructions do not have an inherent ‘no choice’ temporal location, as the am-periphrastic construction does. To sum up this section, fi is an auxiliary merged in weak T heads selected by M. Since all M heads host overt material, fi occurs only in second position in the inflectional string. In these configurations, the M head, not fi, checks the [agr] features. Thus, fi carries [tense/perfective] features, on a par with am, ai, a. The fact that these features are checked under M [mood] selection indicates that fi
. For a definition of conditional avea ‘would’ and presumptive vrea ‘will’ as M-heads see Avram (1999). . In Old Romanian fi had [agr] features. Frîncu (1970) brings historical evidence for the replacement of [agr] with [irrealis] on fi when the selection criteria have changed for the subjunctive MP.
An irrealis BE auxiliary in Romanian
has also an [irrealis] feature that matches the [irrealis] feature generally associated with [mood]. The next section looks for further evidence for the [irrealis] feature of fi. . Fi in other checking configurations The auxiliary fi occurs in both matrix and subordinate clauses, where it is related to the following interpretations: (i) tense/perfectivity and irrealis in matrix and sentential complements to verbs; (ii) perfectivity in sentential complements to lexical categories other than verbs (e.g., N, A, P); (iii) irrealis in matrix and sentential complements with presumptive periphrastic forms. First, we consider constructions as in (14a), where an irrealis interpretation does not materialize. (14) a.
Faptul de a fi min¸tit. . . fact-the of to-inf have lied. . . ‘The fact that s/he has lied. . .’
Considering (14a) in light of the representation in (12b), we see that the [tense/perfective] features of fi check the past participle (for perfectivity) and M (for temporal non-specificity). However, the [irrealis] feature remains unchecked. This analysis assigns (14a) the representation in (14b). DP
(14) b.
CP
D
MP
C de
M
T/AspP
a [mood] [tense/perf]
fi mintit ,
The exclusive anteriority (i.e., a fact has taken place in the past) interpretation in (14) must be related to the absence of a higher modal selector for the MP. Such constructions show that the [irrealis] feature on fi may remain unchecked, and therefore, invisible to the interpretive module. The analysis in (14b) can be verified by extending it to other lexical heads (e.g., P, A) that do not involve modal features, as in (15).
Larisa Avram and Virginia Hill
(15) a.
[Înainte [s˘a fi spus asta]] trebuia s˘a te gânde¸sti la consecin¸te. before sa be said that ought sa refl-you think-2sg.subj at consequences ‘You should have thought about the consequences before having said that.’ b. Maria se declara [fericit˘a [de a-l fi convins.]] Maria refl-se declared-3sg.impf happy of to-inf him be persuaded ‘Maria declared herself happy to have persuaded him.’
In (15) the selectors of the perfective MP are a preposition and an adjective without modal features. As predicted, all these constructions receive an anteriority reading, following from the syntactic checking of the [tense/perfective] features of fi (against M and Asp, respectively), while the [irrealis] feature remains unchecked and, therefore, invisible to interpretation. In contrast to environments as in (14), (15), where MP is selected by N, P, A, root clauses with “unselected” MP, as in (16) and (17), involve an irrealis dimension. (16) a.
O fi dormind. will be sleeping ‘S/he may be sleeping.’
In (16a), the MP is a gerundive that lacks perfectivity. The M morpheme o ‘will’ has intrinsic tense features, because it appears as the future tense auxiliary in other contexts. Therefore, the selection of fi in the context of (16a) is not due to its [tense/perfective] features but to its [irrealis] feature. It is the [irrealis] feature of fi that changes the temporal specificity of o into non-specificity, and adds the irrealis dimension to the interpretation, in a configuration as in (16b). MP
(16) b. M/Agr
o [mood], [irrealis]
T/AspP
fi [irrealis] dormind [–perf.]
Further proof for the impact of the [irrealis] feature of fi on presumptive mood, as in (16a), comes from the fact that the mood marker o may not freely alternate with the future tense auxiliary va ‘will’ in this context (16c), as it does elsewhere (16d). (16) c.
O/*va fi dormind. will/will be sleeping-ger ‘S/he may be sleeping.’
An irrealis BE auxiliary in Romanian
d. O/va dormi.//O/va s˘a vin˘a. will/will sleep-inf//will/will sa comes-subj ‘S/he will sleep.’//’S/he will come.’
The auxiliary va ‘will’ represents a more recent grammaticalization for future tense than the old auxiliary o ‘will’. The strong [tense] features of va ‘will’ cannot be suppressed or weakened (to non-specific temporality) in the presence of fi (see also example (9e)). However, o may change its featural make-up to suit the discourse semantics matching the [irrealis] feature of fi. In sum, fi in (16a) cannot be excluded or replaced. That restriction cannot be related to lack of [tense] checking, since o may select other auxiliaries or functional formation compatible with future tense checking, as in (16d). Also, the restriction cannot be related to the [perfective] feature of fi that remains invisible to checking in the presence of a gerund. Therefore, the only function of fi in (16a) is to check the [irrealis] feature required by the discourse semantics and the [mood] value of o. It may also be the case that the conversion of o into a mood head with temporal non-specificity is the side effect of the [irrealis] checking of fi. So far, the examples discussed in this section show a functional dissociation between the two sets of features on fi, that is [tense/perfective] and [irrealis]. It appears that visibility of fi is ensured by checking one or another set of features, the interpretation being adjusted to the option (i.e., anteriority or irrealis). However, the most frequent use of fi involves checking of both sets of features, as seen in (17). (17) a.
Ar fi plecat would-3sg.cond be gone ‘S/he would have gone.’
b.
MP M/Agr
T/AspP
[mood]
Tense/Asp
[tense/perf], [irrealis] fi plecat
A conditional root clause as (17a) has the same configuration as (14b) and (16b), only that the two sets of features concur on fi. That is, the requirements of discourse semantics (i.e., optative interpretation) are met through the checking of the [irrealis] feature of fi, in conjunction with [mood], whereas the features of the mood marker ar (i.e., temporal non-specificity and perfectivity) are met through the checking of the [tense/perfective] set on fi.
Larisa Avram and Virginia Hill
Now we turn to embedded configurations where an irrealis interpretation follows from the insertion of fi. So far, we have seen that the irrealis interpretation is available in root clauses with discourse semantics effect, but not in embedded clauses, where the MP must strictly satisfy the checking requirements of the selecting heads (i.e., N, A, P). Hence, what allows for an irrealis interpretation in constructions as in (18a), where fi occurs in embedded context? (18) a.
θsi dorea s˘a fi plecat. refl-dat wished-3sg.impf sa be gone ‘S/he wished to have gone.’
The crucial difference in (18a) versus (14)–(15) is that the selecting matrix head is V, located in its extended inflectional domain, TP. It has been established in the literature (Kayne 1989 a.o.) that sentential complementation with non-indicative clauses as in (18a) involves a modal chain between matrix T and embedded T. The idea is that matrix T carries operator-like modal features that select compatible modal inflection for syntactic checking. An [irrealis] feature fits this derivational pattern. A sentence such as (18a), that contains a verb with semantic properties for an irrealis interpretation, must provide transparency conditions for T-to-T checking (i.e., M/T/Asp/V checking matrix T/V) to ensure the syntactic checking of the [modal],[irrealis] features, as in (18b).9 TP
(18) b.
VP
T [modal],[irrealis] V
MP M
T/AspP
[mood]
[tense/perf], [irrealis]
fi plecat
The analysis in (18b) finds confirmation in constructions as (19), where fi may have ambiguous scope.
. T-to-T checking does not necessarily involve T-to-T movement. However, transparency at clausal border is obligatory, and this is ensured in MP complements either through movement of the verbal string to CP-heads or through the exclusion of a CP.
An irrealis BE auxiliary in Romanian
(19) Ion ar fi dorit s˘a cumpere casa. Ion would-3sg.cond be wished sa buy-3sg/pl.subj house-the ‘Ion would have liked to buy the house.’
The example in (19) has two possible readings. On one reading, the wanting is doubtful; on the other reading, the wanting is real, so fi affects only the interpretation of the embedded clause, which has a counterfactual reading (he did not buy the house). Therefore, fi in the matrix may take scope over the embedded clause. This type of wide scope is possible because (19) has the configuration in (18b), where the transparency at clause border allows for T-to-T checking. To sum up this section, the auxiliary fi, with [tense/perfective] and [irrealis] features, is always selected by Mood and merges in an MP as in (12b). However, the selector of MP determines which features of fi will be activated. Thus, heads with [–V], [–mode] features select MPs where fi computes only its [tense/perfective] feature. Conversely, discourse semantics and heads with [+V], [+mode] features select MPs where fi computes its [irrealis] feature (alone, or in conjunction with the [tense/perfectivity] set).
. Conclusions This paper looked at the grammaticalization of Lat. esse in Romanian, where it took the form of the morpheme fi. Tests of distribution, word order and interpretation led to the following analysis: 1. One grammaticalization of fi ‘be’ takes the form of copula ‘be’ for inflecting unaccusatives and passives. The copula fi ‘be’ is morphologically and syntactically different from perfective auxiliaries such as am, ai, a ‘have’. 2. Another grammaticalization of fi ‘be’ takes the form of an auxiliary that qualifies as the counterpart of the perfective auxiliary am, ai, a ‘have’. 3. The two auxiliaries occur in complementary distribution due to contrastive association with specific [tense] and [mood] features. Thus, am, ai, a ‘have’ occurs in contexts with definite time setting and realis interpretation (i.e., TP), whereas fi ‘be’ occurs in contexts with non-specific time frame and/or irrealis interpretation (i.e., MP). 4. The auxiliary fi ‘be’ has two sets of features: [tense/perfective] and [irrealis]. Checking of one set of features against Mood is sufficient to ensure visibility. The other feature of fi ‘be’ may be checked or not, depending on the properties of the lexical or the functional selector of MP. This analysis shows that the dichotomy of the ‘be’/‘have’ auxiliaries in Romanian is different from their equivalent in Romance insofar as the Romanian auxiliaries
Larisa Avram and Virginia Hill
do not reflect the sub-categorization features of the verb but the semantic features that determine sentence interpretation. That is, the complementary distribution of fi and am, ai, a displays a dichotomy between realis and irrealis. The morpho-syntactic conditions responsible for the ‘be’/‘have’ dichotomy in Romanian vs. Romance are also reflected in the verbal paradigms. More precisely, both auxiliaries ‘be’ and ‘have’ are compatible with any mood in Romance, while in Romanian ‘have’ is limited to indicative verbs, while ‘be’ is restricted to nonindicative forms. The differences in the grammaticalization and use of ‘be’ and ‘have’ must be related to more general morphological (e.g., the presence of mood markers in Romanian) and syntactic patterns (e.g., the organization of the information structure) that account for language variation between the two linguistic areas.
References Abeillé, A. & Godard, D. (2000). Varieties of ESSE in Romance Languages. Http://cslipublications.stanford.edu/HPSG/1/ Avram, L. (1994). A contrastive analysis of modals in English and Romanian. Revue roumaine de linguistique, 39 (3–4), 267–284. Avram, L. (1999). Auxiliaries and the Structure of Language. Bucharest: EUB. Boškovi´c, Ž. (1997). The Syntax of Nonfinite Complementation. Cambridge, MA: The MIT Press. Burzio, L. (1986). Italian Syntax: A government-binding approach. Dordrecht: Reidel. Chomsky, N. (1995). The Minimalist Program. Cambridge MA: The MIT Press. Cornilescu, A. (2000). The double subject construction in Romanian. In V. Motapanyane (Ed.), Comparative Studies in Romanian Syntax (pp. 83–134). Oxford: Elsevier. Dobrovie-Sorin, C. (1994). The Syntax of Romanian. Berlin: Mouton de Gruyter. Edelstein, F. (1972). Sintaxa gerunziului românesc. Bucharest: Editura Academiei. Frîncu, C. (1970). Conjunctivul perfect românesc. Privire diacronica (Romanian perfect subjunctive. A diachronic view). SCL, 21 (2), 205–227. — — (1963). Gramatica Academiei Romane. Bucharest: Editura Academiei. Kayne, R. (1989). Null subject and clitic climbing. In O. Jaeggli & K. Safir (Eds.), The Null Subject Parameter (pp. 239–261). Dordrecht: Reidel. Perlmutter, D. (1978). Impersonal passives and the unaccusative hypothesis. In Proceedings of the 4th Annual Meeting of the Berkeley Linguistics Society. Berkeley, CA: Berkeley Linguistics Society. Pîrvulescu, M. (2002). Le concept de paradigme et la morphologie verbale. PhD dissertation, University of Toronto.
Auxiliary selection and split intransitivity in Paduan Variation and lexical-aspectual constraints* Michela Cennamo and Antonella Sorace University of Naples / University of Edinburgh
This paper examines the system of auxiliary selection and split intransitivity in Paduan, an Italian dialect spoken in north-eastern Italy. It provides further empirical evidence in support of the gradient model of split intransitivity put forward by Sorace (2000, 2003) and instantiated by the Auxiliary Selection Hierarchy (ASH). This hierarchy has been found to underpin variation in the choice of auxiliaries in several languages (and varieties of languages) and to be at work also in other manifestations of split intransitivity. In particular, it is shown that auxiliary selection in Paduan is accountable in terms of the ASH constraints. It indicates that the choice of auxiliaries in variable verbs is systematically correlated with specific aspectual interpretations, and that anticausatives, unlike in Italian, also present auxiliary alternations captured by some of the parameters represented by the ASH. The Paduan data, consistently with other languages, indicate the existence of gradient lexical-semantic effects on the syntactic realization of arguments and confirm that an account of these phenomena is thus firmly rooted at the lexicon-syntax interface.
* We wish to thank Paola Benincà for lengthy and informative discussions of the data presented in this study and for commenting on an earlier draft of the paper. We are also grateful to Delia Bentley, Adam Ledgeway, Michele Loporcaro, Mair Parry, Cecilia Poletto, Laura Vanelli, Maria Teresa Vigolo, Nigel Vincent, and Raffaella Zanuttini for discussing with us the data and some of the theoretical arguments developed in this paper. The usual disclaimers apply. Furthermore, we are indebted to Salvatore Glorioso and Antonella Mazzi for helping us to recruit the participants for this study, and of course to the participants themselves for generously volunteering their time. The collection and the analysis of the Paduan data were made possible by two grants (SGAN4054/APN5834 and Joint International Activity Grant) awarded to the second author by the British Academy, whose support is hereby acknowledged.This paper has been written jointly by the two authors. For academic purposes, Antonella Sorace bears responsibility for Sections 1, 2 and 5, and Michela Cennamo for Sections 3 and 4.
Michela Cennamo and Antonella Sorace
.
Introduction
Recent synchronic, diachronic and variational studies on different manifestations of split intransitivity, namely auxiliary selection (Aranovich 2003; Cennamo 1998, 2001, 2002; Keller & Sorace 2003; Legendre & Sorace 2003; Sorace 1995, 2000, 2003), ne-cliticization (Bentley 2004; Bentley & Eyrthórsson 2004; Sorace 1995) impersonal passives (Keller & Sorace 2003), pleonastic reflexives (Cennamo 1999, 2000) and inversion (Parry 2000), have shown that a gradient model of split intransitivity – based on the interplay of a number of aspectual and thematic parameters, such as the degree of telicity of the verb/predicate and the degree of control/affectedness of the subject, along the lines proposed by Sorace (1992, 1995, 2000, 2003) – may capture systematic variation in the morphosyntactic behaviour of verbs that cannot be explained within purely syntactic or semantic accounts of split intransitivity, or within existent models of the lexicon-syntax interface. The present paper contributes further empirical evidence in favor of a scalar notion of split intransitivity by analysing auxiliary selection, past participle agreement, ne-cliticization and inversion with monadic intransitives and anticausatives in Paduan, an Italo-Romance dialect from southern Veneto. The discussion is organized as follows. Section 2 illustrates the gradient model of split intransitivity, providing the theoretical framework for the analysis of the Paduan data discussed in Section 3. Section 4 demonstrates how this model allows one to capture systematicity within variation. Finally, Section 5 summarizes the conclusions and points to future lines of research in this area.
. Split intransitivity: A multidimensional, gradient view According to Levin & Rappaport Hovav’s (1995) much-quoted definition, unaccusativity is “semantically determined and syntactically encoded”. The definition underlies a range of approaches to the study of split intransitivity motivated by the need to account for the growing empirical database on split intransitivity and the consequent need to refine the Unaccusative Hypothesis (henceforth UH; Burzio 1986; Perlmutter 1978). The UH had the advantage of making testable predictions. Its main claim was that the class of intransitive verbs divides into two types, unaccusative and unergative, with distinct syntactic properties: the subject of unaccusatives is an underlying direct object promoted to subject position, whereas the subject of unergatives is base-generated. Unaccusatives and unergatives were therefore predicted to exhibit consistent syntactic behavior as subclasses: with respect to the selection of perfective auxiliaries, all unergative verbs were expected to select HAVE and all unaccusative to select BE. The semantic characteristics of subjects were assumed to be aligned with their syntactic status: agentivity with unerga-
Auxiliary selection and split intransitivity in Paduan
tive ‘real’ subjects, patienthood with unaccusative ‘apparent’ subjects. However, verbs are not always well-behaved and often do not conform to the predictions. On the one hand, there are verbs that do not respond to unaccusativity diagnostics in consistent ways, both within and across languages: so blush is unaccusative in Italian (arrossire, selecting BE) but unergative in Dutch (blozen, selecting HAVE); fiorire ‘blossom’ can take either HAVE or BE. On the other hand, there are verbs that can display either unaccusative or unergative syntax depending on the characteristics of the predicate: all verbs of manner of motion select HAVE in Dutch and German when they denote a process but take BE in the presence of a telic prepositional phrase; verbs of emission (e.g. rumble; see Levin & Rappaport Hovav 1995 for extensive discussion) are unergative in their default case but in some languages may exhibit unaccusative behaviour when they receive a telic interpretation. One of the main challenges opened up by the UH is therefore how to account for the variable behaviour of verbs. In the present discussion we do not address this issue from the theoretical perspective of which current model of the syntax-semantics interface, whether ‘projectionist’ or ‘constructional’ (Levin & Rappaport Hovav 2005) can better account for cross-linguistic variation in the mappings between a lexical-semantic level of representation and the level of syntactic structure (on which see Rappaport Hovav & Levin 1998: 127–129; Levin & Rappaport Hovav 2005: 189–193; Sorace 2003 for a discussion of the limitations of both models). We focus instead on the empirical aspects of the issue: the type and degree of variation that can be observed and the factors explaining them. In particular, we follow Sorace’s model, according to which intransitive verbs may be organized along a hierarchy (of auxiliary selection, ASH) – defined primarily by the degree of telicity of the verb and secondarily by its the degree of agentivity – ranging from verbs denoting “telic dynamic change” that categorically select BE to verbs of “atelic non-motional activity” that categorically select HAVE in languages with auxiliary selection. Between the two extremes are verbs that display variable behavior. The ASH is therefore an empirical generalization that identifies the notion of “telic change” at the core of unaccusativity and that of “atelic non motional CHANGE OF LOCATION > CHANGE OF STATE > CONTINUATION OF STATE > EXISTENCE OF STATE > UNCONTROLLED PROCESS > MOTIONAL PROCESS > NON-MOTIONAL PROCESS Figure 1. The Auxiliary Selection Hierarchy (ASH)
categorical BE selection
categorical HAVE selection
Michela Cennamo and Antonella Sorace
activity” at the core of unergativity. The closer to the core a verb type is, the more determinate its syntactic status as either unaccusative or unergative. Verbs that are stative and non-agentive are the most indeterminate. Sensitivity to contextual or compositional factors correlates with the distance of a verb from the core. This generalization substantiates the intuition that, within their respective classes, some verbs are “more unaccusative” and “more unergative” than others (Legendre, Miyata, & Smolensky 1991). This does not mean that unaccusativity or unergativity are inherently gradient notions, but rather that some verbs allow only one type of syntactic projection/construction whereas other verbs are compatible with different projections/constructions to variable degrees. Since the ASH was proposed, it has been found to be relevant not only for various aspects of split intransitivity in a number of languages (although not in identical ways), but it has also been shown to correlate with the order and the degree of difficulty in the acquisition of its properties in various languages (beyond Italian and French, for which it was originally formulated) (Montrul 2003; Sorace 1993a, b; Sorace & Shomura 2001) and their vulnerability to attrition (Montrul 2003). Research on the ASH can be developed in three ways. First, the empirical database needs to be extended to include data from other languages that have auxiliary selection. Second, more data are needed on other manifestations of split intransitivity in order to evaluate the validity of the hierarchy outside the narrow domain of auxiliary selection. Third, the ASH needs to be accounted for within a formal model of the syntax-lexicon interface (see Legendre, this volume, Legendre & Sorace 2003; Mateu 2003; Sorace & van Hout, forthcoming, for recent proposals).
. Split intransitivity in Paduan The field of Romance dialectology represents an extraordinarily rich source of experimental data for linguistic theories. One of the advantages of working with variants of the same language type is the possibility of investigating the limits of variation for given syntactic phenomena while at the same time keeping other syntactic variables constant (Benincà 1994:8–9). This will be illustrated in detail for split intransitivity in Paduan,1 which manifests itself, albeit to different ex-
. The following abbreviations are used: clt=subject clitic; f=feminine; inf=infinitive; m=masculine; pl/pl=plural; pp=past participle; rfl=reflexive; sg/sg=singular; 1=first person; 3=third person.
Auxiliary selection and split intransitivity in Paduan
tents, in four functional domains: auxiliary selection, past participle agreement, ne-cliticization and inversion.2 As is well-known and widely discussed in the literature since the late seventies (Burzio 1986; Perlmutter 1978, 1989; Rosen 1984, 1988; Centineo 1996; Van Valin 1990, among others), these domains also characterize split intransitivity in standard Italian (with variation that is however captured by the ASH), where they tend to cluster together in differentiating between unaccusative and unergative verbs.3 The situation is much more varied, instead, in non-standard varieties, as pointed out in Sorace 2000, 2003 for auxiliary selection, and even more so in the different Italo-Romance dialects (Loporcaro 1998, 2001, forthc., for an overview; Cennamo 2001; Ledgeway 2000 with reference to some southern Italian varieties). Some dialects, in fact, only select HAVE with monadic intransitives (e.g., Sicilian), others only select BE (e.g. some southern Lazio dialects) (Tuttle 1986, among others), whereas still others have split systems governed by verb classes, tense and/or grammatical person, often presenting inconsistent evidence (e.g., selecting BE only in the 3rd person singular, but HAVE in the rest of the paradigm, as in some varieties from southern Marche (Cocchi 1995), or selecting HAVE in the third person only, . As we shall see in the course of discussion (Sections 3–4) inversion too appears to be sensitive to the ASH. More investigation is needed, however, in order to determine exactly how it interacts with split intransitivity in Paduan. We only consider inversion in main declarative clauses. (For a discussion of inversion in non-declarative clauses in North-Eastern Italian dialects see Poletto 1998). . Unaccusatives (e.g., scoppiare ‘burst’ ‘nascere, ‘be born’, partire ‘leave’), generally select the auxiliary BE in compound tenses, require past participle agreement with the subject (ia) and allow the cliticization of a quantified postverbal subject with the partitive ne (ib); unergatives (e.g. lavorare ‘work’, parlare ‘talk’, dormire ‘sleep’) generally select the auxiliary HAVE, lack past participle agreement (which occurs in the unmarked masculine singular form) (iia), and disallow ne-cliticization (iib). (i)
a.
b.
(ii) a.
Gli pneumatici sono scoppiati all’improvviso the tyres are burst.pp.pl.m at-the sudden ‘The tyres have suddenly burst’ Ne sono scoppiati due (< sono scoppiati due pneumatici) of-it are burst.pp.pl.m two (< are burst.pp.pl.m. two tyres) ‘Two of them have burst ( Patient) have been associated with the well-known relational hierarchy Subject > Object to express markedness relations (e.g. Aissen 2001; Keenan & Comrie 1977; Silverstein 1976). In similar vein, we may formulate event scales for the features relevant to the A/E auxiliary distinction. (In fact, the telicity scale (atelic > telic) is adopted by Grimshaw 1990.) (26) Aux scale: A > E (27) Event feature scales:7 a. displacement: b. homogeneity: c. telicity: d. directed change: e. state: f. inherent volitionality: g. internal motion:
–DIS > +HOM > –TE > –DIR > –ST > +VO > –MO >
+DIS –HOM +TE +DIR +ST –VO +MO
By aligning two scales at a time we come up with a set of relations which express the markedness of the mapping of a certain feature – say TELIC – to a certain auxiliary – say E. Such harmonic alignments, as they are defined in OT, formalize markedness relations for mappings between certain properties across scales (Prince & Smolensky [1993] 2004; Aissen 2001). Note the change of symbol from ‘>’ (higher on a scale) to ‘’ (more harmonic – less marked) in the sample harmonic alignments in (28).
on the number of entailments of each kind the verb gives it. Some clear differences separate the two approaches however. Dowty is primarily interested in the theoretical status and inventory of thematic roles which he argues are not primitives of any argument structure theory; the present analysis denies any status to thematic roles per se. The mapping goes directly from finer-grained lexico-semantic properties to auxiliary. Dowty’s set of entailments is not identical to the set of lexico-semantic features in the present analysis (see for example the discussion of the feature HOMOGENEITY, including Footnote 3). Finally, his proto-roles are fuzzy categories; there are no fuzzy categories in the OT analysis proposed here. . Nothing in principle prevents all scales to be uniform in having the form: – F > + F. This would simply require the replacement of features HOMOGENEITY and VOLITIONALITY with their opposites.
Géraldine Legendre
(28) Harmonic alignments: a. E/telic A/telic ‘The mapping of [+telic] onto auxiliary E is less marked than the mapping of [+telic] onto auxiliary A’ b. A/atelic E/atelic c. etc.
Such alignments correspond to a hierarchy of constraints with polarity reversed (note again the change in symbol from ‘’ (more harmonic) to » (more dominant): (29) Constraint alignments: a. *A/telic *E/telic ‘don’t map [+telic] onto auxiliary A’ outranks () ‘don’t map [+telic] onto auxiliary E’ b. *E/atelic *A/atelic c. etc.
For the present we drop the lower-ranked constraints in each of these alignments, which target the unmarked mappings. Focusing first on those remaining constraints pertaining to auxiliary A, we have {* A/telic ≡ * A/[+te], * A/[+dir], * A/[–vo], . . .}. We now encapsulate all these constraints. . Formulating a *A Power Hierarchy Consider again Table 2. Down through feature [–VOLITIONAL] Table 2 expresses implicational relations among feature values. We can thus define a set C of ‘Epreferring’ feature values as in (30) and state an encapsulated constraint *A/C which is violated whenever a constraint in {*A/f: f in C} is violated. C ≡ {+DIS, –HOM, +TE, +DIR, +ST, –VO} (‘E-preferring’ feature values) b. F = *A/C An event with a C-feature is not mapped to auxiliary A
(30) a.
The fact that a candidate violating F= *A/C six times is more marked than one violating it twice is implemented via a standard OT Power Hierarchy (Smolensky 1995; Legendre et al. 1998). The power hierarchy is given in (31a); the constraint Fk is violated whenever F is violated k (or more) times. (31) Universal mapping constraint hierarchy a. F6 · · · F2 F1 (auxiliary/event semantics mapping) b. *A/+DIS *A/–HOM *A/+TE *A/+DIR *A /+ST *A/–VO
Given the implicational relations among features in Figure 1, the Power Hierarchy can be written in the equivalent form (31b). For example, any candidate violating F at least 3 times must violate the constraints {*A/+DIR, *A/+ST, *A/-VO}: the
Optimizing auxiliary selection in Romance
Table 3. Markedness as determined by *A Power Hierarchy Power Hierarchy: Constraint ranking: A/arriver ‘arrive’ +DIS, –HOM, +TE, +DIR, +ST, –VO A/suer ‘sweat’ –DIS, +HOM, –TE, –DIR, –ST, –VO A/nager ‘swim’ –DIS, +HOM, –TE, –DIR, –ST, +VO
(*A/C)6 *A/+DIS
(*A/C)5 *A/–HOM
(*A/C)4 *A/+TE
(*A/C)3 *A/+DIR
(*A/C)2 *A/+ST
(*A/C) *A/–VO
*
*
*
*
*
* *
feature sets yielding at least 3 violations are exactly {±dis, ±hom, ±te; +dir, +st, –vo}. Indeed, any feature set violating * A/+dir will necessarily also violate {* A/+st, * A/–vo} so it will violate F at least 3 times. F3 can thus be equated with * A/+dir.8 As shown in Table 3 (where * indicates a violation of the constraint at the top of the relevant column), it is more marked for arriver ‘arrive’ to be assigned auxiliary A than for suer ‘sweat’ because arriver ‘arrive’ violates constraint F 6 times while suer violates it only once. Recall that an additional feature is needed to distinguish motional from nonmotional controlled processes: nager ‘swim’ vs. travailler ‘work’. The feature proposed is INTERNAL MOTION. Because all verb classes above controlled motional processes have the value –MO as does the lowest class on the ASH that includes travailler, the feature MO does not stand in an implicational relation with all other features. The constraint *A/+MO is therefore not part of the Power Hierarchy proper. In Section 4 I argue that German auxiliary selection provides independent empirical evidence for placing MO outside of the Power Hierarchy proper. In Romance the constraint *A/+MO must be ranked below the lowest constraint in the Power Hierarchy to express the relative markedness of assigning auxiliary E to the nager vs. travailler classes. Diachronic evidence of auxiliary change in Spanish (see Section 5) supports the conclusion that work is “more core” than swim.9 With auxiliary A, swim violates *A/+MO, while work violates no *A constraint. Combining the *A Power Hierarchy and *A/+MO gives the *A constraint hierarchy in (32). For now *A/+MO is somewhat arbitrarily placed at the bottom of the constraint hierarchy in (32) before being revisited in Section 4. . It is worth entertaining the idea that the constraint hierarchy in (31b) derives in fact from the alignment of the scale –vo > +st > +dir > +te > –hom > +dis with the auxiliary scale A > E yielding two harmonic alignments: (i) A/–vo A/+st A/+dir A/+te A/–hom A/+dis and (ii) E/+dis E/–hom E/+te E/+dir E/+st E/–vo. In (31b) we are referring to only one of the two constraint hierarchies this alignment yields, namely the *A/feature hierarchy. . See also the data on swim in Italian in (16). Selecting E is not completely rejected by native speakers.
Géraldine Legendre
(32) *A constraint hierarchy: A/+DIS *A/–HOM *A/+TE *A/+DIR *A/+ST *A/–VO *A/+MO
Obviously, if no constraint against mapping onto auxiliary E ever entered the picture, selection of E would always be optimal, and we would have no cross-linguistic mismatches and no verbs selecting A. While the harmonic alignments stated in (29b) result in a hierarchy of *E constraints anti-parallel to that of *A constraints (29a), for the purposes of this discussion it suffices to state the *E constraint in the encapsulated version given in (33). (It may turn out that we need to interpose all *A/ and *E/ constraints as the present analysis is extended to account for further patterns.) (33) *E: “don’t map onto auxiliary E”
. Establishing the language-particular ranking of *E It should be clear by now that the cross-linguistic choice of auxiliary for a given featural profile will be determined by the relative ranking of *E – just where *E is interposed into the fixed *A hierarchy. Thus our analysis rests on the claim that while the ranking of *A constraints is universally fixed that of *E is language-specific. We take the input to optimization to be the featural description of individual predicates or predicate subclasses, as specified in Table 2. The candidate set simply consists of two candidates corresponding to the two auxiliaries, E and A. The candidate auxiliary which wins in each optimization is the one which incurs the least number of violations of the lowest-ranked constraints. A sample of tableaux is provided below which highlights the crucial optimizations responsible for the language-particular rankings of *E in French and Italian. Despite the fact that it is inherently telic (see Footnote 3), change of state verb pourrir ‘rot’ selects A. This motivates the French sub-ranking *E *A/+TE, shown in (34). (34) Tableau 1: pourrir ‘rot’ Input: –DIS, +HOM, +TE, +DIR, +ST, –VO, –MO French a. E F b. A
*A/ +DIS
*A/ –HOM
*E
*A/ +TE
*A/ +DIR
*A/ +ST
*A/ –VO
*
*
*
*
*A/ +MO
*!
In a standard OT tableau like Tableau 1, the relative left-to right order of constraints conventionally reflects their relative importance. The leftmost constraint (e.g. *A/+DIS) dominates all other constraints to its right. The optimal candidate
Optimizing auxiliary selection in Romance
output is specified with the pointed finger ( F ). The input features listed above the tableau reflect lexico-semantic and aspectual properties of individual verbs and determine which constraints in the constraint hierarchy are activated in a given optimization. For example, the constraint *A/+DIS is not activated if the input contains the value ‘minus’ for that feature. Violations of constraints are recorded as * in the relevant cells, *! records the fatal violation incurred by any suboptimal candidate. Not all telic change of condition verbs select A in French. Mourir ‘die’ selects E. This is where the constraint on mapping the feature HOMOGENEITY comes in: *A/–HOM *E, as shown in (35). (35) Tableau 2: mourir ‘die’ Input: –DIS, –HOM, +TE, +DIR, +ST, –VO, –MO French
*A/ +DIS
*A/ –HOM
F a. E
*E
*A/ +TE
*A/ +DIR
*A/ +ST
*A/ –VO
*
*
*
*
*A/ +MO
*
b. A
*!
In Italian, verbs of state select E. This entails that *E must be outranked by *A/+ST in Italian: see (36). In contrast, [–ST] verbs select A. Thus *E *A/–VO: see (37). (36) Tableau 3: esistere ‘exist’ Input: –DIS, +HOM, –TE, –DIR, +ST, –VO, –MO Italian
*A/ +DIS
*A/ –HOM
*A/ +TE
*A/ +DIR
*A/ +ST
F a. E
*E
*A/ –VO
*A/ +MO
*
b. A
*!
*
(37) Tableau 4: sudare ‘sweat’ Input: –DIS, +HOM, –TE, –DIR, –ST, –VO, –MO Italian
*A/ +DIS
*A/ –HOM
a. E
*A/ +TE
*A/ +DIR
*A/ +ST
*E
*A/ –VO
*A/ +MO
*!
F b. A
*
These crucial cases result in the language-particular constraint rankings stated in (38). (38) a.
French:
*E ↓
*A/+DIS*A/–HOM *A/+TE*A/+DIR*A/+ST *A/–VO*A/+MO
b. Italian:
↑ *E
Géraldine Legendre
. Verifying predictions in French and Italian For Italian the position of *E in the *A hierarchy entails that verbs that have any of the feature values [–DIS, –HOM, +TE, +DIR] select E. For example, peggiorare ‘worsen’ selects E in Italian, as expected; see (39). Its French counterpart, however, is correctly predicted to select A because *E outranks *A/+DIR in French; see (40). (39) Tableau 5: peggiorare ‘worsen’ Input: –DIS, +HOM, –TE, +DIR, +ST, –VO, –MO Italian
*A/ +DIS
*A/ –HOM
*A/ +TE
*A/ +DIR
*A/ +ST
*!
*
*E
F a. E
*A/ –VO
*A/ +MO
*
b. A
*
(40) Tableau 6: empirer ‘worsen’ Input: –DIS, +HOM, –TE, +DIR, +ST, –VO, –MO French
*A/ +DIS
*A/ –HOM
a. E
*E
*A/ +TE
*A/ +DIR
*A/ +ST
*A/ –VO
*
*
*
*A/ +MO
*!
F b. A
In both languages, controlled processes (motional and non-motional) are predicted to select A. That is because only *E and *A/+MO are activated in the case of motional processes (41) and only *E in the case of non-motional processes (e.g. travailler/lavorare ‘work’). Given that *E *A/+MO in both languages, A is the auxiliary of choice. (41) Tableau 7: nager ‘swim’ Input: –DIS, +HOM, –TE, –DIR, –ST, +VO, +MO French a. E F b. A
*A/ +DIS
*A/ –HOM
*E
*A/ +TE
*A/ +DIR
*A/ +ST
*A/ –VO
*A/ +MO
*! *
In Italian (but not in French) run selects E when it connotes a telic event typically expressed with a directional phrase as in run to school. In our terms run and run to school have different input specifications and the latter (but not the former) activates the constraint *A/+TE. Because *A/+TE outranks *E in Italian it is less costly for run to school to violate *E and select E in Italian. In French *E outranks *A/+TE with the consequence that A is selected as the less costly option.
Optimizing auxiliary selection in Romance
. General predictions of the OT analysis Significantly, the proposed general OT analysis does not predict total, unconstrained variation in auxiliary selection. Rather, it predicts a very specific typology of languages including languages in which all verb classes select E, languages in which all verb classes select A, and languages which each display one of a tightly limited set of splits. If *E all *A constraints, all verb classes select A, regardless of their semantics. Among those are Catalan, Spanish, as well as Abruzzese dialects of Italian spoken in the Valle d’Orte (Tuttle 1986: 269). Conversely, If all *A constraints *E, all verb classes are predicted to select E. Such a pattern is also available in Romance, in particular it is found in central Italian dialects such as that of Terracina (Tuttle 1986). Remarkably, Terracinese verbs selecting E are not limited to intransitive and reflexive verbs, they also include transitive verbs like drink wine, sing a song (Tuttle 1986: 267).10 Further languages are predicted to have different cut-off points along the universal hierarchy, including a point intermediary between French and Italian. As a result, more verb classes select E than in French but fewer than in Italian. This is largely the case in German and Dutch. In Dutch change of location and change of state verb classes select E/zijn while the remaining verb classes (continuation of a pre-existing state, existence of state, uncontrolled processes, as well as controlled processes) select A/hebben.11 This supports the constraint ranking in (42). (42) Dutch ranking: A/+DIS *A/–HOM *A/+TE *A/+DIR *E *A/+ST *A/–VO *A/+MO. . Tuttle also discusses a range of “intermediary” Italian dialects in which the diachronic shift to either A or E is not completed and where the ‘minority’ auxiliary is limited to certain persons (Tuttle 1986: 270). For example, in dialects close to generalizing E (Cori, Roiate/Zagarolo, L’Aquila/Avezzano/Pescara) A is still used with third person singular and plural subjects. In dialects close to generalizing A (Introdacqua, Lanciano) E is retained with second person singular subjects. Another Italian variety, Altamurano, discussed in La Fauci & Loporcaro (1989) allows both auxiliaries in all persons except for third person plural where only A is allowed with socalled unergatives and only E with third person singular with unaccusatives. The present OT analysis can easily be supplemented with harmonic alignments involving a person scale and possible interposing of the resulting constraints with aspectually based ones. . Some verb classes allow for auxiliary variation in Dutch. Verbs of indefinite change in a particular direction (stijgen ‘rise’, dalen ‘descend’) favor E but allow A while verbs of existence of state (blijven ‘remain’ bestaan ‘exist’) and verbs of controlled motion (rond lopen ‘run around’, zwemmen ‘swim’) favor A but also allow E. In Section 6 I propose that this fuzzy boundary is best construed as resulting from partial constraint ranking. See Randall (this volume) for additional discussion of auxiliary selection in Germanic.
Géraldine Legendre
The position of *E determines the basic cut-off point between verbs connoting change (change of location and of state) and remaining verbs. For example, exist selects A in Dutch (and French) but E in Italian. (43) Tableau 8: bestaan ‘exist’ Input: –DIS, +HOM, –TE, –DIR, +ST, –VO, –MO Dutch
*A/ +DIS
*A/ –HOM
*A/ +TE
*A/ +DIR
a. E
*E
*A/ +ST
*A/ –VO
*
*
*A/ +MO
*!
F b. A
German displays the same basic pattern of auxiliary selection as Dutch – with one striking exception. Controlled motional processes (e.g. laufen ‘run’, schwimmen ‘swim’) select E/sein in German (vs. A in Dutch). The explanation rests with the constraint *A/+MO which was identified in Section 3.1 as not being part of the Power Hierarchy of *A constraints. That is, its ranking is not universally fixed and it may be interposed into the fixed *A hierarchy as in (44) for German. (44) German ranking: A/+DIS *A/–HOM *A/+TE *A/+DIR *A/+MO *E *A/+ST *A/–VO.
Ranking *A/+MO (anywhere) above *E in German has the effect of overriding the violation of *E and selecting E as the optimal auxiliary. Because there is only one class of verbs that has the + value for the feature MO (see Table 2), no other class is affected by the relative ranking of *A/+MO. (45) Tableau 9: schwimmen ‘swim’ Input: –DIS, +HOM, –TE, –DIR, –ST, +VO, +MO German
*A/ +DIS
*A/ –HOM
*A/ +TE
*A/ +DIR
*A/ +MO
F a. E
b. A
*E
*A/ +ST
*A/ –VO
* *!
The present analysis also predicts impossible languages. These are languages which, compared to Italian, reverse auxiliary choice, e.g. languages with change of location verbs selecting A while controlled processes select E, languages with existence of state verbs selecting E while change of state verbs select A, etc. To the best of my knowledge, this prediction is correct. In fact the OT analysis predicts exactly 14 possible languages (= 7 possible positions of *E in the *A-hierarchy times 2 because *A/+MO may be ranked above or below *E). Of these we have discussed 6 types in descending position of *E in the hierarchy of *A constraints: Spanish/Catalan/Abruzzese, French, Dutch,
Optimizing auxiliary selection in Romance
German, Italian, and Terracinese. Further empirical work is needed to confirm this typology. It is important to distinguish the number of predicted languages (14) from the number of possible rankings predicted by the OT analysis: 56 to be exact. Given 6 constraints with fixed ranking, there are 7 possibilities for interposing *A/+MO, resulting in 8 possibilities for interposing *E (or 7 times 8 = 56). Alternative rankings commonly result in the same optimal output. As discussed earlier, the ranking of *A/+MO in German anywhere above *E will produce one and the same result: auxiliary E. Similarly, the ranking of *A/+MO anywhere below *E will produce the alternative auxiliary choice: A. For convenience, *A/+MO is ranked lowest in the Italian, French, and Dutch rankings displayed above but this is not necessary. The same result obtains as long as *A/+MO is ranked below *E.
. Diachronic predictions Beyond predicting a specific typology of languages from a synchronic perspective, the present analysis predicts a specific pattern of change in auxiliary selection over time. In particular, change from a 2-auxiliary system to a 1-auxiliary system should proceed from peripheral to core verb classes and not vice-versa. In this section I examine the case of Spanish which once had the dual auxiliary system typical of Romance languages. The empirical facts are taken from Aranovich (2003). In Old Spanish, verbs denoting controlled processes (trabajar ‘work’, pecar ‘sin’, etc.) occurred with A/haber, in contrast to all other verb classes attested with E/ser, as shown in Table 4. This, in turn, provides evidence of a dual auxiliary system at an early point in its history. Change from E to A across the board was completed by the 17th century (Benzing 1931). Table 4 lists the date of the last attested occurrence with E/ser (Benzing 1931; Aranovich 2003). Overall, the switch to E is not abrupt but the trend is very clear. The first verbs to switch were stative verbs like rastar ‘remain’ by the 14th century. Next to change were verbs of appearance (aparecer ‘appear’, desaparecer ‘disappear’, etc.) by the 16th century. Change of condition and location verbs like morir ‘die’ and ir ‘go’ were the last ones to give up E in the 17th century. Note in passing that this last subclass is precisely the one that still uses E in Modern French. Interestingly enough, verbs of manner of motion like errar ‘wander’ corer ‘run’ took longer to switch than say verbs of appearance and continuation of state, though they eventually did by the 16th century. These are the very verbs that still use E in German and I argued earlier that this follows from the formal status of *A/+MO. The feature MO is not in a set-inclusive relation with the other features in Table 2. As a result, *A/+MO is not part of the Power Hierarchy of *A constraints and it is freely re-rankable cross-linguistically. In German, *A/+MO is
Géraldine Legendre
ranked above *E (vs. below *E in Italian, French, and Dutch). Based on Table 4, I conclude that (a) *A/+MO was not part of the fixed *A-hierarchy in Old Spanish either and (b) *A/+MO was ranked above *E until the 16th century. Thus, the history of auxiliary selection in Spanish supports the general analysis argued for on formal grounds. Table 4. Last attested occurrence of E/ser in Spanish Spanish
13th
14th
15th
16th
17th
Change of location
exir ‘leave’ viar ‘return’ desviar ‘change direction’
arribar ‘arrive’
descender ‘descend’ tornar ‘return’
pasar ‘go by’ ir ‘go’ partir ‘depart’
change of condition
cenar* ‘dine’ yantar* ‘eat’
transir ‘die’
fallir ‘fail, die’ despertar ‘wake up’
venir ‘come” llegar ‘arrive’ caer ‘fall’ entrar ‘enter’ salir ‘leave’ huir ‘run away’ escapar ‘escape’ volver ‘return’ subir ‘climb’ fallecer ‘die’ finar ‘die’ fenecer ‘die’ adormir ‘fall asleep’ adormecer ‘fall asleep’ amanecer ‘dawn’ anochecer ‘grow dark’ acabar ‘finish’
appearance
cuntir ‘happen’
continuation fincar of state ‘stay’ quedar ‘remain’ rastar ‘remain’ controlled motional processses
aparecer ‘appear’ acaecer ‘happen’ desaparecer ‘disappear’ holgar ‘rest’
errar ‘wander’
corer ‘run’
* = classified as change of state verbs in Aranovich (2003)
caminar ‘walk’
nacer ‘be born’ crecer ‘grow’ morir ‘die’
Optimizing auxiliary selection in Romance
Recast in terms of constraint re-ranking, the history of Spanish can be sketched as follows. (46) Re-ranking of *E and *A/+MO in Spanish 13th–14th century: a. *A/+DIS *A/–HOM *A/+TE *A/+DIR *A/+ST *A/+MO *E *A/–VO 15th century: b. *A/+DIS *A/–HOM *A/+TE *A/+DIR *A/+MO *E *A/+ST *A/–VO 16th century: c. *A/+DIS *A/–HOM *A/+TE *A/+MO *E *A/+DIR *A/+ST *A/–VO 17th century: d. *A/+DIS *A/–HOM *A/+TE *E *A/+MO *A/+DIR *A/+ST *A/–VO
Ranking a with *E below *A/+ST in the fixed *A hierarchy entails that stative verbs select E while verbs denoting controlled non-motional processes select A. The ranking of *A/+MO above *E entails that controlled motional verbs select E, as is the case for errar ‘wander’. Re-ranking in b results in *E raising above *A/+ST with the consequence that stative verbs stop appearing with E. The relative ranking of *A/+MO and *E is maintained through rankings b and c: controlled motional verbs still select E. But in ranking c, *E moves up the constraint hierarchy again with the result that of non-motional verbs, only change of condition and location still select E. Ranking d involves a crucial reversal of the relative ranking of *A/+MO and *E resulting in controlled motional verbs switching to A. Note that the 17th century data available is also consistent with raising *E one step above *A/+TE – this is the ranking I have posited for Modern French. There is simply not enough data to provide a more definitive analysis.
. Indeterminacy in auxiliary selection Recall that Table 2 contains Italian and French auxiliary choices annotated with *, indicating that both auxiliaries may in fact be selected. Consider first telic verbs of appearance in French and their value for the feature HOMOGENEITY. An event of appearing is not typically understood as made of homogeneous sub-events. Rather, it is made of two basic sub-events, before appearing and after. In other words apparaître ‘appear’ is –HOM, activating the constraint *A/–HOM whose
Géraldine Legendre
violation is fatal based on the French ranking established in (38). As a result, apparaître patterns like mourir ‘die’ and selects E. The situation is actually a bit more complicated because appearance need not be instantaneous and may take on the related meaning of ‘becoming visible’. For example, it takes time for the sun to appear, the star is partly visible for a while, and the result is that sentence (47a) is not really infelicitous. This is even clearer with disparaître ‘disappear’. The entailment in (47d) is felicitous with the meaning ‘becoming obstructed’. Construed as such, the process of disappearing becomes homogeneous in a manner reminiscent of the process of rotting. (47) a.
Le soleil est en train d’apparaître. ‘The sun is in the process of appearing’ b. ??⇒ il est apparu un peu. it is appeared a bit c. Le soleil est en train de disparaître. ‘The sun is in the process of disappearing’ d. ⇒ il a disparu un peu. it has disappeared a bit
Given the ranking in (38) – in particular the *A/-HOM >> *E >> *A/+TE stretch of the constraint hierarchy – *A/–HOM is vacuously satisfied, the violation of *E is fatal, and disparaître selects A. This is illustrated in (48). (48) Tableau 10: disparaître ‘disappear’ Input: –DIS, +HOM, +TE, +DIR, +ST, –VO, –MO French a. E F b. A
*A/ +DIS
*A/ –HOM
*E
*A/ +TE
*A/ +DIR
*A/ +ST
*A/ –VO
*
*
*
*
*A/ +MO
*!
When apparaître/disparaître connote fast events (for example when they are predicated of a single being suddenly appearing on stage, disappearing behind a bush, etc.) their basic event structure is restored, the resulting events are construed as –HOM and the expected auxiliary is E. (49) a.
Dieu est/a apparu aux fidèles. ‘God appeared to the faithful’ b. Le chat est/a disparu derrière un buisson. ‘The cat disappeared behind a bush’
However, auxiliary A is also felicitous in (49) and this option cannot in any obvious sense be traced back to a change of value of the feature HOMOGENEITY. Rather it is a consequence of a general process of instability at the cut-off point which can be documented cross-linguistically.
Optimizing auxiliary selection in Romance
As Table 2 reveals, it is not the case that all subclasses are equally subject to such fluctuation. The Italian and Dutch examples listed in (50) show that auxiliary indeterminacy is found only with non-core verbs on the ASH (see Sorace 2000 for more examples). Whatever general solution is offered must capture this unique asymmetry. (50) a. Italian verbs of state (durare ‘last’, sussistere ‘subsist’) select E but allow A. b. Italian uncontrolled processes and motional processes (tremare ‘tremble’, correre ‘run’) select A but allow E. c. Dutch verbs of indefinite change in a particular direction (stijgen ‘rise’, dalen ‘descend’) select E but allow A. d. Dutch verbs of state (blijven ‘remain’, bestaan ‘exist’) select A but allow E. e. Dutch verbs of controlled motion (zwemmen ‘swim’) select A but allow E.
It is important to distinguish two kinds of variation. One pertains to different input specifications such as the telicization of run when it appears with a directional phrase. In Italian this causes a change in auxiliary from A to E, as discussed in Section 3.5. Effects of this kind fall naturally under the OT analysis. To the extent that there is a residue of variation best characterized as free variation in individual speakers such residue, I propose, follows from partial constraint ranking, i.e. some indeterminacy in the relative ranking of *E and the *A constraints. By definition, a partial constraint ranking yields a set of rankings rather than a single ranking. A set of rankings yields potentially different optimal outputs, hence variation in outputs. Such ranking indeterminacy is a hallmark of unstable linguistic states and is well attested in dialectal variation (Anttila 1997; Nagy & Reynolds 1997), learning (Boersma 1997; Boersma & Hayes 2001), language acquisition (Legendre et al. 2002; Davidson & Legendre 2003; Legendre et al. 2004), and diachronic change (Slade 2003). Without working out a detailed analysis of the facts of French, Italian, and Dutch, we can see how the model works. If *E floats over four positions in the middle of the *A hierarchy, say over the constraint ranking stretch identified in (51) we obtain a set of four rankings for the class of [–HOM, +TE, +ST] verbs, as stated in (52): (51) (52) a. b. c. d.
*E *A/–HOM *A/+TE *A/+ST *A/–HOM *E *A/+TE *A/+ST *A/–HOM *A/+TE *E *A/+ST *A/–HOM *A/+TE *A/+ST *E
Géraldine Legendre
Different rankings yield different percentages of verbs selecting a particular auxiliary. Specifically, verbs that are –HOM select E in rankings (52b, c, d) 75% of the time. Verbs that are +TE select A for rankings (52c, d) 50%; verbs that are +ST select E only 25% of the time (ranking (52d)). So the more extreme features – HOM and +ST correlate more homogenously with E or A; the middling feature +TE waffles 50/50 between the two. Put another way, there is more indeterminacy in the middle of the range than at the extremes.
. The larger debate: Unaccusativity The main result of the present analysis is that the ultimate choice of auxiliary in compound tenses is the outcome of optimizing the interface mapping between lexico-semantic and aspectual features and the auxiliaries themselves. The analysis is fundamentally a semantic one and the question naturally arises as to how it bears on the larger debate that auxiliary selection in Romance and Germanic languages has been traditionally part of, i.e. the Unaccusativity debate. Considered in isolation of other unaccusativity diagnostics, auxiliary selection is clearly a semantically determined process and does not directly provide evidence for the syntactic distinction which is the hallmark of the Unaccusativity Hypothesis. Yet, the systematicity of the mapping and the existence of a subset relation between French and Italian verb sets argue against analyses which analyze auxiliary selection in French as an idiosyncratic property of individual verbs requiring a stipulation in the lexicon (Cummins 1996). Semantic determination of auxiliary choice is true of lexical verbs but it leaves open the question of auxiliary choice in passives. While selecting the same auxiliary to mark passive voice (E), passive be selects A in French but E in Italian compound tenses, without any exceptions. (53) F a. Trois batiments ont été reconstruits. three buildings have been rebuilt I b. Sono stati ricostruiti tre palazzo. are been rebuit three buildings ‘Three buildings have been rebuilt’
The absolute character of the auxiliary choice in passives is reminiscent of reflexives (except that passive be and all reflexives may not share the same auxiliary, as is the case in French).12 Of course, the general patterning of passives with reflexives . Space precludes a treatment of reflexives in Romance briefly presented elsewhere (Legendre & Sorace 2003).The facts are comparatively simple – all reflexive verbs, regardless of their semantic or aspectual properties – select E in both French and Italian. From an OT perspective
Optimizing auxiliary selection in Romance
and so-called unaccusatives has historically provided core evidence for analyzing unaccusative verbs as selecting an underlying direct object (Burzio 1986; Levin & Rappaport Hovav 1995; Perlmutter 1978, 1989; Rosen 1984). To the extent that passives (and reflexives) require a syntactic analysis we have an argument that auxiliary selection with lexical verbs is also a syntactic process. Evidence that the syntactic process is grounded in semantics is therefore not evidence against a syntactic encoding of the unaccusative-unergative distinction. See Levin & Rappaport Hovav (1995) on this very point. However the auxiliary appearing in perfect tenses of the passive auxiliary be is the very auxiliary that be selects regardless of its function as a lexical verb, a copula, or a passive auxiliary: A in French, E in Italian. The absence of mismatch in auxiliary selection between passive be and lexical be supports the conclusion that the semantics of be are sufficient to explain the patterning of passives in the two languages. Note in passing that this is inconsistent with the traditional view that auxiliaries and copula are semantically empty. Whether auxiliary selection should be construed as evidence for the Unaccusative Hypothesis depends also on the existence of multiple and overlapping diagnostic tests, all pointing to one and the same syntactic distinction. Obviously, auxiliary selection is irrelevant to unaccusativity in many Romance languages. In Italian, auxiliary selection patterns with ne cliticization and other diagnostics to a remarkable degree, leading most researchers to assume that auxiliary selection is sensitive to a syntactic distinction. In French, however, we face a somewhat complicated picture. Few non-reflexive lexical verbs select E in French but they are a subset of the Italian verbs which select E. This leads to the following hypotheses: (54) a.
Auxiliary selection is an unaccusativity diagnostic in Italian and in French (but something more needs to be said about French) b. Auxiliary selection is an unaccusativity diagnostic in Italian but not in French (on a par with Spanish) c. Auxiliary selection is not an unaccusativity diagnostic in either language
Hypothesis (54a) is theoretically interesting because it abstracts away from the comparatively small size of the set of verbs selecting E in French and thereby confronts the strong mismatch across diagnostic tests.13 As discussed in Legendre this means that there is a constraint specific to reflexives (construed either as a morphological or a syntactic constraint) that overrides all other *A and *E constraints. This approach entails that a subset of Romance reflexive verbs must be analyzed as unergative, a position only partially shared by Reinhart & Siloni (2004)). See also Aranovich (2000) and Sorace & Cennamo (this volume). . Labelle (1992) argues that selecting E is a diagnostic test for unaccusativity and selecting A a diagnostic test for unergativity in French, on a par with Italian. I disagree that verbs selecting
Géraldine Legendre
(1989) and Legendre & Sorace (2003), French intransitive verbs display a clear split in participial constructions, completely reminiscent of the corresponding one in Italian (Perlmutter 1989; Rosen 1988). In particular, a subclass of the lexical verbs selecting A patterns like lexical verbs selecting E, as shown in (55), provided the secondary event denoted by the participial clause is anterior to the event denoted by the main clause. (55) a.
Le chat parti, les souris se mirent à danser. (partir : E) ‘the cat (having) left, mice started to dance’ b. Le père mort, les enfants vendirent la propriété familiale. (mourir : E) ‘the father (having) died, the children sold the family home’ c. La neige fondue, toutes les stations ont fermé. (fondre : A) ‘the snow (having) melted, all the ski resorts have closed down’
What does this entail for the present analysis? As far as I can tell it entails that the mapping constraints should refer to an abstract property cutting across auxiliaries and participial constructions, i.e. a syntactic configuration not reducible to a (set of) semantic properties, say S and O for subject and direct object, independently of their implementation in a particular syntactic framework. The *A constraints become *S constraints and the *E constraint is turned into a set of *O constraints (as is available from the original alignments in (29)). An additional constraint pertaining to participial constructions – *1/PC for short – is added at the top of the constraint hierarchy. The analysis entails that a change of condition verb like fondre ‘melt’ behaves as an unergative verb σ-taking with respect to auxiliary selection (A) but as an unaccusative verb (σ-taking if necessary) with respect to participial constructions. Unaccusativity is context-dependent within a language. Hypothesis (54b) is straightforward and the present OT analysis is consistent with it. Reflexive verbs that are briefly discussed in Footnote 12 appear to offer an argument in its favor. An important consequence of the semantically based analysis of auxiliary selection is that some reflexive verbs are unergative despite selecting E. These are the verbs at the bottom of Table 5, i.e. controlled processes, both motional se pavaner ’ strut about’, se retourner ‘turn over’ and non-motional A can be classified as unergative once-and-for-all in French, for three main reasons. Firstly, the subclass which selects E is only a subset of the unaccusative class (if one assumes that selecting E is a diagnostic test for unaccusativity at all). Secondly, part of Labelle’s evidence relies on assuming that Impersonal Constructions are a reliable test for unaccusativity in French, which they are not (Legendre 1989; Legendre & Sorace 2003). Thirdly, Labelle’s claim entails that Spanish and Romanian with their single perfect auxiliary A do not have any unaccusative verbs. However, occurrence in participial constructions and bare NP subject constructions positively identifies unaccusative verbs in Spanish (Aranovich 2000; Mendikoetxea 1999; Torrego 1989). Romanian unaccusative (but not unergative) verbs also productively occur in participial absolute constructions (Dobrovie-Sorin 1994: 182).
Optimizing auxiliary selection in Romance
s’écrier ‘exclaim’, s’adonner à ‘take to’, etc. (See Legendre (1989) and Legendre & Sorace (2003) for further discussion.) They are unergative on the basis of their semantics and the ASH and they systematically fail reliable unaccusativity tests like participial constructions in French (Legendre 1989). There also exist Romance varieties in which only a subset of reflexive verbs select E. In the Nuorese variety of Sardianian E is selected when the reflexive clitic functions as a direct object (56a) but A when the reflexive clitic functions as an indirect object of the verb or a dative of interest (56b). (The other well-known variety called Campidanese selects E in all contexts, just like Italian.) (56) S a. Juanne s’est vistu in s’isprecu. John refl is seen in the mirror ‘John saw himself in the mirror’ S b. Maria e Lukia s’an mandutu paritzas litteras. Maria and Lucy refl have sent several letters ‘Mary and Lucy sent each other several letters’
auxiliary: E
auxiliary: A
This suggests that a harmonic constraint involving REFL has to have a syntactic component though it is its relative ranking in the hierarchy which determines the cut-off point. Clearly, more work is needed on Romance varieties with splits targeting reflexive verbs before the debate is resolved. The alternative is to adopt Hypothesis (54c): auxiliary selection is not an unaccusativity diagnostic in either language. But this fails to capture the comparatively neat facts of Italian, the absence of mismatch among diagnostic tests, and the overall need for recognizing some abstract property that cuts across all diagnostic tests regardless of the language. To the extent that those generalizations survive further scrutiny, nothing at present seems to be gained by adopting Hypothesis (54c). It is worth noting that once harmonic alignment constraints are restated in terms of grammatical relations (S and O) they become linking rules similar in some respects to the rules proposed in Levin & Rappaport Hovav (1995). See also Randall (this volume). (57) Linking rules (Levin & Rappaport Hovav 1995) i. Immediate Cause linking Rule: The argument of a verb that denotes the immediate cause of the eventuality described by that verb is its external argument (S). ii. Directed Change Linking Rule: The argument of a verb that corresponds to the entity undergoing the directed change described by that verb is its direct internal argument (O). iii. Existence Linking Rule: > internal argument (O). iv. Default Linking Rule: > internal argument (O).
Géraldine Legendre
In their projectionist approach the lexical semantics of a verb deterministically map onto positions at argument structure. Verbs with variable behavior have different meanings, and therefore different lexical semantic representations, each with its own regular argument structure realization. Confronted with the complexities of Romance auxiliary selection, the projectionist approach faces at least two challenges – accounting for cross-linguistic variation without resorting to systematic duplication in the lexicon and accounting for French for which it overwhelmingly makes incorrect predictions. In particular, Rule ii incorrectly predicts that verbs of change of state (externally caused brûler ‘burn’, geler ‘freeze’, fondre ‘melt’, pourrir ‘rot’, etc. and internally caused rougir ‘blush’, pâlir ‘become pale’, etc.) select E. Rule iii incorrectly predicts that verbs of existence (exister ‘exist’, durer ‘last’, être ‘be’, etc.) select E. Rule iv incorrectly predicts that non-agentive manner of motion verbs (rouler ‘roll’, rebondir ‘bounce’, tournoyer ‘whirl’, etc. ) select E. In contrast, the present harmonic alignments refer to lexico-semantic and aspectual features rather than verb classes, they are in conflict with one another and it is the relative ranking of the constraints that determines the outcome. As Randall (this volume) argues, one can reconcile the linking rules in (57) with the Germanic and Romance facts by adopting Parametrized Linking. Her solution is to claim that different languages ‘can choose different sets of linking rules, [. . .], or use linking rules constructed from distinct but closely related features’. While this may account for the data she considers, it is not clear what specific typology, if any, the Parametrized Linking approach predicts. To the extent that the analysis developed here is ultimately successful in characterizing the typology of auxiliary selecting systems (including diachronic ones) OT does provide a unique formal resolution of the cross-linguistic mapping problem that has resisted formalization until now. Such success in turn provides further support for the validity of OT as a formal theory of the interfaces.
References Aissen, J. (2001). Markedness and subject choice in Optimality Theory. In G. Legendre, J. Grimshaw, & S. Vikner (Eds.), Optimality-Theoretic Syntax (pp. 61–96). Cambridge, MA: The MIT Press. Antilla, A. (1997). Variation in Finnish Phonology and Morphology. PhD dissertation, Stanford University. Aranovich, R. (2000). Split intransitivity and reflexives in Spanish. Probus, 12, 165–186. Aranovich, R. (2003). The semantics of auxiliary selection in Old Spanish. Studies in Language, 27 (1), 1–37. Bard, E. G., Robertson, D., & Sorace, A. (1996). Magnitude estimation of linguistic acceptability. Language, 71, 32–68.
Optimizing auxiliary selection in Romance
Benzing, J. (1931). Zur Geschichte von ser als Hilfszeitwort bei den Intransitiven Verben im Spanischen. Zeitschrift für romanische Philologie, 51, 385–460. Boersma, P. (1997). How we learn variation, optionality, and probability. Proceedings of the Institute of Phonetic Sciences of the University of Amsterdam, 21, 43–58. Boersma, P. & Hayes, B. (2001). Empirical tests of the gradual learning algorithm. Linguistic Inquiry, 32 (1), 45–86. Burzio, L. (1986). Italian Syntax: A government-binding approach. Dordrecht: Foris. Croft, W. (1990). Typology and Universals. Cambridge: CUP. Cummins, S. (1996). Meaning and Mapping. PhD dissertation, University of Toronto. Davidson, L. & Legendre, G. (2003). Defaults and competition in the acquisition of functional categories in Catalan and French. In R. Nuñez-Cedeño, L. López, & R. Cameron (Eds.), A Romance Perspective on Language Knowledge and Use (pp. 273–290). Amsterdam: John Benjamins. Dobrovie-Sorin, C. (1994). The Syntax of Romanian. Berlin: Mouton de Gruyter. Dowty, D. (1979). Word Meaning and Montague Grammar. Dordrecht: Reidel. Dowty, D. (1991). Thematic proto-roles and argument selection. Language, 67, 547–619. Grévisse, M. (1980). Le Bon Usage. Editions Duculot, Gembloux. Grimshaw, J. (1990). Argument Structure. Cambridge, MA: The MIT Press. Jakobson, R. ([1965] 1995). Quest for the essence of language. In L. Waugh & M. MonvilleBurston (Eds.), On Language: Roman Jacobson (pp. 407–421). Cambridge, MA: Harvard University Press. Keenan, E. & Comrie, B. (1977). Noun phrase accessibility and universal grammar. Linguistic Inquiry, 8, 63–99. Labelle, M. (1992). Change of state and valency. Journal of Linguistics, 28, 375–414. La Fauci, N. & Loporcaro, M. (1989). Passifs, avancements de l’objet indirect et formes verbales périphrastiques dans le dialecte d’Altamura (Pouilles). Rivista di Linguistica, 1 (1), 161–196. Legendre, G. (1989). Unaccusativity in French. Lingua, 79, 95–164. Legendre, G., Smolensky, P., & Wilson, C. (1998). When is less more? Faithfulness and minimal links in wh-chains. In P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis, & D. Pesetsky (Eds.), Is the Best Good Enough? Optimality and competition in syntax (pp. 249–289). Cambridge, MA: The MIT Press. Legendre, G., Hagstrom, P., Vainikka, A., & Todorova, M. (2002). Partial constraint ordering in child French Syntax. Language Acquisition, 10, 189–227. Legendre, G. & Sorace, A. (2003). Auxiliaires et intransitivité en français et dans les langues romanes. In D. Godard (Ed.), Les langues romanes: Problèmes de la phrase simple. Paris: CNRS editions. Legendre, G., Hagstrom, P., Chen-Main, J. Tao, L., & Smolensky, P. (2004). Deriving output probabilities in child Mandarin from a dual-optimization gammar. Lingua, 114, 1147–1185. Levin, B. & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax-semantics interface. Cambridge, MA: The MIT Press. McClure, W. (1995). Syntactic Projections of the Semantics of Aspect. Tokyo: Hitsujishobo. Mendikoetxea, A. (1999). Construcciones inacusativas y pasivas. In I. Bosque & V. Demonte (Eds.), Gramática descriptiva de la lengua española, Vol. 2 (pp. 1574–1629). Madrid: Real Academia Española / Espasa Calpe. Perlmutter, D. (1978). Impersonal passives and the unaccusative hypothesis. In Proceedings of the Fourth Annual Meeting of the Berkeley Linguistic Society (pp. 157–189). Berkeley, CA: University of California.
Géraldine Legendre
Perlmutter, D. (1989). Multiattachment and the unaccusative hypothesis: The perfect auxiliary in Italian. Probus, 1, 63–119. Prince, A. & Smolensky, P. ([1993] 2004). Optimality Theory: Constraint interaction in generative grammar. Malden, MA: Blackwell. Nagy, N. & Reynolds, B. (1997). Optimality theory and variable word-final deletion in Faetar. Language Variation and Change, 9, 37–55. Reinhart, T. & Siloni, T. (2004). Against the unaccusative analysis of reflexives. In A. Alexiadou (Ed.), The Unaccusativity Puzzle: Explorations in the syntax-lexicon interface (pp. 159–179). Oxford: OUP. Rosen, C. (1984). The interface between semantic roles and initial grammatical relations. In D. Perlmutter & C. Rosen (Eds.), Studies in Relational Grammar 2 (pp. 38–77). Chicago, IL: Chicago University Press. Rosen, C. (1988). The Relational Structure of Reflexive Clauses: Evidence from Italian. New York, NY: Garland. Ruwet, N. (1988). Les verbes météorologiques et l’hypothèse inaccusative. In C. BlancheBenveniste, A. Chervel, & M. Gross (Eds.), Mélanges à la mémoire de Jean Stéfanini. Aixen-Provence: Publications de l’Université de Provence. Sankoff, G. & Thibault, P. (1977). L’ alternance entre les auxiliaries avoir et être en français parlé à Montréal. Langue Française, 34, 81–108. Silverstein, M. (1976). Hierarchy of features and ergativity. In R. M. W. Dixon (Ed.), Grammatical Categories in Australian Languages (pp. 112–171). Canberra: Australian Institute of Aboriginal Studies. Slade, B. (2003). How to rank constraints: Constraint conflict, grammatical competition, and the rise of periphrastic do. In D. Eric Holt (Ed.), Optimality Theory and Language Change (pp. 337–385). Dordrecht: Kluwer. Smith, C. (1991). The Parameter of Aspect. Dordrecht: Kluwer. Smolensky, P. (1995). On the internal structure of the constraint component Con of UG, paper presented at University of California, Los Angeles. Sorace, A. (2000). Gradients in auxiliary selection with intransitive verbs. Language, 76, 59–890. Torrego, E. (1989). Unergative-unaccusative alternations in Spanish. In I. Laka & A. Mahajan (Eds.), Functional Heads and Clause Structure [MIT Working Papers in Linguistics 10] (pp. 253–272). Cambridge, MA: MITWPL. Tuttle, E. (1986). The spread of esse as universal auxiliary in central Italo-Romance. Medioevo Romanzo, XI, 229–287. Van Valin, R. D. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Vendler, Z. (1967). Linguistics in Philosophy. Ithaca, NY: Cornell University Press. Verkuyl, H. (1989). Aspectual classes and aspectual composition. Linguistics and Philosophy, 12, 39–94. Zaenen, A. (1993). Unaccusativity in Dutch: Integrating syntax and lexical semantics. In J. Pustejovsky (Ed.), Semantics and the Lexicon (pp. 129–161). Dordrecht: Kluwer.
Auxiliary selection in Chinese* Feng-hsi Liu University of Arizona
This paper argues that the distribution of the two Chinese aspectual auxiliaries -le and -zhe in the locative inversion construction is determined by the semantics of the event, in particular the interaction of lexical aspect (telicity and stativity) and agentivity. The -le/-zhe alternation shows a gradient effect similar to the selection of perfective auxiliaries in Romance and Germanic (Sorace 2000), since telic verbs select -le, dynamic atelic verbs select -zhe, and verbs that fall in between may show variable behavior depending on other factors. The fact that auxiliary selection with ‘detransitivized’ verbs (analogous to passives) is sensitive to the same semantic parameters is offered as evidence against a purely syntactic analysis of the -le/-zhe alternation.
.
Introduction
Auxiliary selection is perhaps one of the most studied phenomena associated with split intransitivity, partly because it has been used to support two opposing approaches to split intransitivity. On the one hand, auxiliary selection has been considered a diagnostic of unaccusativity in Germanic and Romance languages (Burzio 1986; Hoekstra 1984, 1999; Legendre 1989, among others), which together with other phenomena, e.g. impersonal passivization (Perlmutter 1978), and necliticization (Burzio 1986), can be explained by the Unaccusativity Hypothesis (Perlmutter 1978). According to the hypothesis, intransitive verbs are classified into two groups: unaccusative verbs, which are assigned an underlying object, and unergative verbs, which are assigned an underlying subject. On this hypothesis, verbs that select ‘be’ are unaccusative, while verbs that select ‘have’ are unergative. This is a syntactic approach to split intransitivity. On the other hand, it has * For valuable comments, I am grateful to the editor Raúl Aranovich, two anonymous reviewers and audiences at the Workshop on Cross-linguistic Variation in Auxiliary Selection, Davis, California, 2003.
Feng-hsi Liu
also been proposed that the choice of ‘be’ vs. ‘have’ is determined by semantic factors, including aspectual properties of the predicate and thematic relations, e.g. telicity and agentivity (Sorace 2000), lexical aspect for Italian (Van Valin 1990), telicity for Dutch (Zaenen 1993), inferable position or state for Dutch (Lieber & Baayen 1997), and patient-like (affected) subject for Old Spanish (Aranovich 2004). In a semantic approach to split intransitivity, verbs that select ‘be’ are not necessarily assigned an underlying object; rather, these verbs share properties that are characterized in semantic terms. In addition, the choice of auxiliary is not always clear-cut, but is gradable (Sorace 2000; Aranovich 2004). To date, the debate between the two approaches has been based on European languages. In this paper I would like contribute to the debate by examining auxiliary selection in a typologically different language – Chinese. I will show that the Chinese data cannot be accommodated in a syntactic analysis; rather, the data requires a semantic analysis. This in turn supports a semantic approach to split intransitivity. Auxiliary selection in Chinese has not received much attention in the literature. To my knowledge, Yu (1995), Hu (1995) and Yang and Pan (2001) are the only studies that have examined this phenomenon. While Yu (1995) (reviewed in 5.2) takes a syntactic approach, linking it to the Unaccusativity Hypothesis, Hu (1995) and Yang and Pan (2001) (reviewed in 4.1) consider semantic factors that affect the choice. All three studies cover a limited variety of verb types. A much wider range of verb types will be examined in this study. Since the only environment in Chinese where the issue of auxiliary selection arises is the locative inversion construction, I will first discuss aspects of the construction that are relevant for my study; in particular, I will examine which types of verbs occur in the construction with respect to their argument structure.
. Locative inversion Locative inversion in Chinese has the structure Loc + V + Asp + NP. The locative phrase that normally occurs after the verb with a preposition, as in (1b), is in the initial position without the preposition, as in (1a): (1) a.
Qianmian zhan-le henduo ren in-front stand-le many people ‘In front are standing many people.’ b. Henduo ren zhan zai qianmian many people stand at in-front ‘Many people are standing in front.’
The construction is also referred to in Chinese studies as existential sentences (Huang 1987), presentative sentences (Hu 1995), and existential structure (Yang
Auxiliary selection in Chinese
& Pan 2001). Two characteristics about Chinese locative inversion stand out. First, unlike English, where only intransitive verbs enter the construction, in Chinese the verb can be either transitive or intransitive. (1a) contains an intransitive verb, and (2), a transitive verb: (2) Yuanzili zhong-zhe yixie guoshu yard-in plant-zhe some fruit-tree ‘In the yard are planted some fruit trees.’
Secondly, an aspect marker is required for most of the verbs that occur in locative inversion. In (1) zhan ‘stand’ is marked by the perfective marker -le, while zhong ‘plant’ in (2) is marked by the imperfect marker -zhe. These sentences are ungrammatical without an aspect marker, as in (3): (3) a. *Qianmian zhan henduo ren in-front stand many people ‘In front are standing many people’ b. *Yuanzili zhong yixie guoshu yard-in plant some fruit-tree ‘In the yard are planted some fruit trees.’
Aspect marking is optional with compound verbs of the form V+ directional, as in (4) (Hu 1995): (4) Waimian chui-lai (-le) yizhen feng outside blow-come- le one-gust wind ‘From outside blows a gust of wind.’
Among the non-compound verbs, only two verbs don’t require aspect marking: the copula shi ‘be’ and the existential verb you ‘exist’. Shi ‘be’ cannot be marked; on the other hand, you ‘exist’ can be marked by either -le or -zhe, with different meanings, as given in (5): (5) a.
Tade lianshang zhongyu you-le/*-zhe xiaorong his face-on finally have-le/-zhe smile ‘On his face finally appeared a smile.’ b. Ta duzili you *-le/(-zhe) shuo bu wan de gushi he stomach have -le/-zhe tell not finish de story ‘In his stomach are endless stories.’
In (5a), with -le, you is a verb of appearance; in (5b), with -zhe, you is a verb of existence. Thus the choice of -le or -zhe is not free in (5a–b). Another thing we can observe is that -zhe in (5b) is optional. The meaning of the sentence remains the same whether -zhe is present or not. For verbs other than shi and you, however, an aspect marker is required, and the question of which aspect marker is chosen for a given verb naturally arises. This question will be the focus of my study.
Feng-hsi Liu
There is some confusion in the literature, however, on whether a given sentence of the form Loc + V + Asp + NP is indeed a case of locative inversion. This is an important issue because it concerns what constitutes the data for our study. Below I will examine the transitive and intransitive verbs in the structure. I will show that while intransitive verbs in this construction are free to take either an argument or an adjunct locative phrase, transitive verbs must take an argument locative phrase. . Intransitive verbs As noted by Huang (1987) and Yang and Pan (2001), for certain verbs the position for the locative NP need not be filled, as seen in (6)–(7): (6) a.
Lai-le yizhi gou come-le one-cl dog ‘There came a dog.’ b. Yuanzili lai-le yizhi gou yard-in come-le one-cl dog ‘In the yard came a dog.’
(7) a.
Si-le yige ren die-le one-cl person ‘There died a person.’ b. Cunzili si-le yige ren village-in die-le one-cl person ‘In the village died a person.’
Verbs that take an optional locative NP are directional verbs or (dis)appearing verbs, and they are intransitive. Other intransitive verbs, however, require the locative NP position to be filled, as in (8): (8) a. *Zuo-zhe yige ren sit-zhe one-cl person ‘There is sitting a person.’ b. Menkou zuo-zhe yige ren doorway sit-zhe one-cl person ‘At the doorway is sitting a person.’
I will take the locative NP that is required, e.g. menkou in (8b), as an argument, subcategorized by the verb, while the locative NP that is not required, e.g. yuanzili in (6b) and cunzili (7b), as an adjunct, which is not subcategorized by the verb. I assume that the contrast between (6)–(7) and (8) is due to a constraint that an argument locative NP must be overtly expressed in the syntax, while an adjunct
Auxiliary selection in Chinese
locative NP need not be. For intransitive NPs, then, the locative NP can be either an argument or an adjunct. . Transitive verbs When the verb is transitive, however, the locative NP position must be lexically filled, as illustrated in (9)–(10): (9) a. *Fang-zhe yiben shu put-zhe one-cl book *‘There was put a book.’ b. Zhuoshang fang-zhe yiben shu table-on put-zhe one-cl book ‘On the table was put a book.’ (10) a. *Xie-le yige zi write-le one-cl character *‘There was written a character.’ b. Heibanshang xie-le yige zi blackboard-on write-le one-cl character ‘On the blackboard was written a character.’
Therefore, I assume the locative NP in sentences with transitive verbs is always an argument. This assumption not only rules out (9a), but also excludes (11), as the locative NP chufangli ‘in the kitchen’ is not subcategorized by the verb ca ‘wipe’: (11) *Chufangli ca-zhe chuangzi kitchen-in wipe-zhe window ‘In the kitchen was wiping the window.’
The argument requirement on the locative NP also affects how a sentence is to be interpreted. For example, (12) only has the interpretation that three characters are written on the car; it cannot be interpreted as someone wrote three characters while in the car: (12) Cheshang xie-le sange zi car-on write-le three-cl character ‘On the car were written three characters’
If cheshang ‘on the car’ is an argument, then the interpretation of (12) follows, since as an argument, the locative NP predicates on the theme object, not the agent. Sentences such as (12), therefore, support the argument requirement of the locative NP in transitive sentences.
Feng-hsi Liu
Pan (1996: 426) notes that the locative phrase in locative inversion sentences must predicate on the theme phrase rather than indicating the location of the event. He gives (13), (his (50)), as an example: (13) a.
John zai zhuozishang xie-le John at table-on write-le ‘John wrote a letter at the table’ ‘John wrote a letter on the table.’ b. Zhuozishang xie-zhe yifeng table-on write-zhe one-cl ‘On the table is written a letter.’
yifeng xin one-cl letter
xin letter
(13a) is ambiguous, with the table indicating either where John was or where the letter was, but (13b) only has the reading in which the table is where the letter was. Although Pan does not say so, his constraint in effect requires that the locative phrase be an argument rather than an adjunct. This constraint, when applied to transitive verbs, as in Pan’s study, yields the desirable result, but it is too strong when applied to intransitive verbs. It would eliminate (6b) and (7b), which, as suggested earlier, include an adjunct locative NP. Nonetheless, the constraint can be used as a criterion to evaluate sentences that contain transitive verbs and have the surface sequence of Loc + V + Asp + NP to determine if they are indeed cases of locative inversion. Consider (14): (14) Wuli tan-zhe gangqin room-in play-zhe piano ‘In the room is playing the piano.
(14), if acceptable, is not a case of locative inversion, because wuli ‘in the room’ is not an argument of the verb tan ‘play’, as it is not subcategorized by the verb. Consider also (15) and (16): (15) Taishang chang-zhe daxi platform-on sing-zhe opera ‘On the platform (people are) singing the opera.’ (16) Fangjianli kai-zhe hui room-in open-zhe meeting ‘In the room is having a meeting.’
In both sentences the locative phrase acts as an adjunct, not as an argument, as the verbs also do not subcategorize for a locative argument. Therefore, they are not cases of locative inversion either. (15) and (16) are included in Yang and Pan (2001) as examples of the existential construction, which according to them, has the syntactic structure of NP loc + V + NP. I will not attempt to compare locative inversion as discussed here with the existential construction discussed by Yang and
Auxiliary selection in Chinese
Pan. I only want to point out that while (15) and (16) may be existential sentences; they are not instances of locative inversion as discussed here. Now with transitive verbs, an issue of argument expression arises. On the basis of the discussion above, transitive verbs that occur in locative inversion are threeplace verbs, and yet the construction only provides two argument positions, the question naturally arises: what happens to the third argument? From the examples above, it is clear that the external argument is never realized. It seems plausible that in locative inversion an operation has applied such that the transitive verbs are detransitivized. Pan (1996) postulates an agent-deletion rule whereby the agent is deleted from the argument structure of a two-place transitive verb. However, the rule as postulated is subject to a number of conditions. Rather than applying in general to three-place transitive verbs, the rule is triggered by the presence of the imperfective marker -zhe only. In Yang and Pan (2001), this analysis is abandoned because the rule has ‘many exceptions’ (p. 194). I suggest that rather than deleting the agent from the argument structure, the detransitivization operation suppresses the external argument from a threeplace transitive verb. Any transitive verb that also takes a locative argument can have its external argument suppressed so that it is not represented in syntax. The derived argument structure, with the external argument suppressed, then feeds naturally into locative inversion, where the locative argument is realized as the locative phrase, while the other internal argument – the Theme argument – is realized as the single non-locative argument of the verb. This operation is similar to the operation proposed by Grimshaw (1990) that derives the passive form of a verb from the active form, where the external argument is also suppressed. Although the external argument of a detransitivized verb cannot be expressed in syntax, it is present in the argument structure. Grimshaw shows that the external argument of a passive verb licenses the by-phrase, and it is also the controller of purpose clauses. Similarly, the external argument of a detransitivized verb in locative inversion also controls the following purpose clauses, as shown in (17): (17) Qianyuan zhong-le xuduo shu, xiatian hao chengliang front-yard plant-le many tree summer in-order-to get-cool ‘In the front yard (are) planted many trees, so that (people) can get cool in the summer.’
In this way the relation between detransitivized verbs and transitive verbs in Chinese resembles that between the passive form of a verb and the active form of a verb in English. In summary, in this section I have considered what types of verbs enter locative inversion with respect to a verb’s argument structure. While intransitive verbs take either an argument or an adjunct locative phrase, transitive verbs must take an argument locative in order to occur in locative inversion. In addition, I have
Feng-hsi Liu
suggested that transitive verbs that enter locative inversion are detransitivized, by which the external argument is suppressed in the argument structure. In the next section, we will see that the detransitivized verbs actually display a wider distribution in terms of verb types in locative inversion. With respect to auxiliary selection, however, they behave the same as intransitive verbs.
. Auxiliary selection So far we have considered what types of verbs can enter locative inversion in terms of argument structure. We are now ready to consider these verbs with respect to semantic properties and find out which aspect marker these verbs select. We will examine three broad classes: verbs denoting change, verbs denoting states, and verbs denoting processes. In (18)–(24) and (26) below, the (a) sentences include intransitive verbs while the (b) sentences include detransitivized verbs. . Verbs of change First, we consider verbs whose meanings include change. These verbs indicate either a change of location, involving an entity going from one location to another, or a change of state, whereby an entity transforms to another state. For change of location, only -le marking is possible. This is shown in (18): (18) a.
Duimian lai -le/*-zhe yiliang che opposite-side come -le/-zhe one-cl car ‘From the opposite side came a car.’ b. Dishang reng -le/*-zhe xuduo guopi zhixie ground-on throw -le/-zhe many peels paper ‘On the ground many peels and paper were thrown.’
Similarly, for change of state, if the change is definite and has reached a resultant state, then -le is selected: (19) a.
Lanzi li lan- le/*-zhe yige pingguo basket in rot- le/-zhe one-cl apple ‘In the basket rotted an apple.’ b. Panli qie-le/*-zhe jipian huanggua plate-in cut-le/-zhe a-few-cl cucumber ‘On the plate cut a few pieces of cucumber.’
If the change is indefinite and has not reached a resultant state, however, either -le or -zhe is selected. (20) is indefinite change of location (directed motion), and (21) is indefinite change of state:
Auxiliary selection in Chinese
(20) a.
Nar mao-le/-zhe yigu yan over-there rise-le/-zhe one-cl smoke ‘Over there rises some smoke.’ b. Jiaoshipang sheng-le/-zhe yimian qi classroom-by raise-le/-zhe one-cl flag ‘By the classroom is raising a flag.’
(21) a.
Tianli zhang-le/-zhe xuduo daozi field-in grow-le/-zhe many rice ‘In the field is growing a lot of rice.’ b. Guoli zhu -le/-zhe fan pot-in cook -le/-zhe rice ‘In the pot is cooking some rice.’
The detransitivized verbs behave the same as intransitive verbs with respect to the selection of aspect markers. The only difference between the two is that in the former it is understood that the situation is brought about by an external force, whereas in the latter the situation arises by natural force; no external force is implied. Thus eight verb types are included under the category of change, which is classified in terms of three features: intransitive vs. detransitivized, definite vs. indefinite change, change of location vs. change of state. Of the eight types, only intransitive definite change of location, e.g. lai ‘come’ and detransitivized indefinite change of state, e.g. zhu ‘cook’ have been mentioned in previous literature. The former is referred to in Yang and Pan (2001) as (dis)appearance verbs, while the latter is called verbs of treatment. The other six types have not been studied before. . States Next, we consider verbs whose meanings include the concept of ‘no change’. These verbs are generally stative. Two subtypes are covered: verbs that describe continuation of a pre-existing condition and verbs that describe existence of state. Consider existence of state first. These verbs include intransitive verbs of simple position, such as zuo ‘sit’, zhan ‘stand’, and verbs of spatial configuration, such as wei ‘surround, gather’; they also include detransitivized verbs of spatial configuration, such as gua ‘hang’ and detransitivized verbs of putting, such as fang ‘put’.1 These verbs have been considered (Huang 1987; Yu 1995; Hu 1995; Pan 1996; Yang & Pan 2001) as some of the typical verbs that occur in locative inversion, and as noted by Yu (1995), these verbs show variable marking, as illustrated in (22)–(23): . Verbs of creating an entity at a location, e.g. xie ‘write’, hua ‘draw’ are also included in this category.
Feng-hsi Liu
(22) a.
Menkou zuo-le/-zhe yige jingwei doorway sit-le/-zhe one-cl guard ‘At the doorway sits a guard.’ b. Qiangshang gua -le/-zhe yifu hua wall-on hang -le/-zhe one-cl painting ‘On the wall hangs a painting.’
(23) a.
Jie kou wei -le/-zhe yiqun ren street-corner surround -le/-zhe one-group people ‘At the street corner gathered a group of people.’ b. Zhuoshang fang-le/-zhe yiben shu table-on put-le/-zhe one-cl book ‘On the table lies a book.’
However, further examination reveals that variability disappears in certain contexts. For example, if the argument is inanimate, then only -zhe marking is possible, as in (24). (24) a.
Menkou zuo *-le/-zhe yi dui shishi doorway sit -le/-zhe one pair stone-lion ‘At the doorway sits a pair of stone lions.’ b. Tianshang gua-*le/-zhe yi lun mingyue sky-on hang-le/-zhe one -cl bright-moon ‘In the sky hangs a bright moon.’
(24a) can be compared with (22a), which has the same verb zuo ‘sit’, but with an animate argument. This might suggest that animacy is a factor for existence of state in the choice of auxiliary. However, on this analysis there is a puzzle: detransitivized spatial configuration verbs, which so far have patterned the same as intransitive spatial configuration verbs, don’t seem to be affected by animacy. In (22b) the argument is also inanimate, but the verb gua ‘hang’, which is the same verb as in (24b), can be marked either way. On close examination, we see that although in both sentences the verb gua ‘hang’ describes a state of existence, there is a difference in how the two states come into existence. In (22b) the state of hanging exists as a result of a volitional act, whereas in (24b) the state of hanging exists as a result of natural force. In other words, in (22b) the hanging of a painting requires an agent, while in (24b) the hanging of the moon does not. Similarly, the contrast between (22a) and (24a) is also due to volitionality. A guard sitting is a volitional act, whereas a stone lion sitting is not. When volitionality or agentivity is absent, variation also disappears. As a result, only -zhe is selected. Hu (1995) also notes that (24b) cannot be marked by -le. He does not mention agentivity; rather, he suggests that -le implies a resultative state, and it cannot be used here because the position of the moon cannot be the result of some previous action of hanging.
Auxiliary selection in Chinese
This explanation is actually compatible with the explanation offered here, since the action of hanging an object is agentive. Therefore, it appears that the way a state comes about matters to the choice of auxiliary. If volitional control is involved in bringing about the existence of the state, then either marking is possible; if, on the other hand, no volition is involved, only -zhe marking is possible. Positional verbs with animate argument and detransitivized spatial configuration verbs imply a preceding volitional act, whereas positional verbs with inanimate verbs do not. This explains the paradigm exhibited by (22)–(24). Thus it is not animacy, but agentivity or volitionality, which is responsible for the behavior of these sentences. Simple position verbs are also subject to event shifting, whereby the same verbs change from simple position verbs to ‘assume position’ verbs. This happens when the context implies a change of state. In this circumstance, only -le marking is possible, even though the structure remains the same. This is shown in (25): (25) Gangcai dianli yige ren ye meiyou, zenme yixiazi just-now store-in one-cl person also not-exist how-come suddenly wei -le/*-zhe zheme duo ren gather -le/-zhe so many people ‘Just now there was nobody in the store; how come suddenly gathered so many people?’
(25) has the ‘assume position’ reading, not the simple position reading. In short, simple positions verbs can be marked either by -le or -zhe; however, the variability is subject to whether the existence is understood to be brought about by an agent. If no agent is involved, the verb selects -zhe only. On the other hand, implication of change of state, hence telicity, shifts an event from one of simple position to one of assume position, and limits the choice of auxiliary to -le only. In contrast to existence of state, verbs that denote continuation of a preexisting state have not been observed in previous studies. These verbs exhibit a pattern very similar to existence of state. Variability in the choice of auxiliary is seen in (26): (26) a.
Keting yijing kong le, fanting hai liu -le/-zhe living-room already empty prt, dining-room still remain -le/-zhe yizhang yizi one-cl chair ‘The living room is already empty; in the dining room still remains a chair.’ b. Yinhangli cun-le /-zhe yidian qian bank-in keep-le /-zhe a-little money ‘In the bank is kept a little money.’
Feng-hsi Liu
However, just like the simple position verbs, variability is not always possible, and its availability depends on whether agentivity is implied. When a state is maintained because of the involvement of an agent, either -le or -zhe is used; otherwise, only -zhe is selected. This explains the variability in (26a, b), as leaving a chair in a room and keeping money in a bank both involve volitionality. (27) further shows that the same verb liu ‘leave behind’ can be interpreted either volitionally or non-volitionally, depending on the internal argument: (27) a.
Menshang liu -le/-zhe yige zitiao door-on leave -le/-zhe one-cl note ‘On the door was left a note.’ b. Chengshili hai liu *-le/-zhe zhanzheng de yiji city-in still leave -le/-zhe war de relics ‘In the city are still left relics of the war.’
In (27a), a note on the door is a result of someone’s volitional act; the note cannot appear on the door by itself. In this case, either -le or -zhe is possible. In (27b), on the other hand, relics of the war were left behind without intervention of an agent. In such a case, only -zhe can be used. Thus the choice of auxiliary varies according to whether liu is agentive or non-agentive. . Processes Finally, we consider verbs of dynamic processes. These verbs include neither change nor a state in their meanings. Processes are either volitional or nonvolitional. For non-volitional processes, only -zhe is possible: (28) a.
Waitou chui *-le/-zhe xie weifeng outside blow -le/-zhe some breeze ‘Outside is blowing some breeze.’ b. Ta shenshang liu *-le/-zhe zhongguoren de xueye he body-on flow -le/-zhe Chinese de blood ‘In his body flows Chinese blood.’
When it comes to volitional processes, however, a mixed picture emerges. Detransitivized verbs select -zhe only, as in (29); intransitive verbs, however, generally don’t occur in locative inversion, as shown in (30): (29) Zuili jiao*-le/-zhe kouxiangtang mouth-in chew-le/-zhe chewing-gum ‘In the mouth is chewing some gum.’ (30) a. *Yaolanli ku-zhe yige xiao yinger crib-in cry-zhe one-cl small infant ‘In the crib is crying a small infant.’
Auxiliary selection in Chinese
b. *Chili you -zhe yige nianqing ren pool-in swim -zhe one-cl young man ‘In the pool is swimming a young man.’ c. *Caochangshang tiao-zhe yige xuesheng field-on jump-zhe one-cl student ‘In the field is jumping a student.’
However, two verbs – pao ‘run’ and zou ‘walk’ – are exceptions to this pattern, as in (31)–(32): (31) Qianmian zou -zhe yige ren front-side walk -zhe one-cl person ‘In front walks a person.’ (32) Lushang pao-zhe yige ren road-on run-zhe one-cl person ‘On the road runs a person.’
Both verbs can enter into locative inversion. Yu (1995) and Yang & Pan (2001) include pao ‘run’ in their examples, and Yu’s examples also include zou ‘walk’. This shows that verbs of controlled processes don’t behave uniformly with respect to whether they enter locative inversion. (30b) can be contrasted with (33), where the same verb you ‘swim’ is used, except that it denotes an uncontrolled process: (33) Chili you -zhe yixie yazi pool-in swim -zhe some ducks ‘In the pool are swimming some ducks.’
This contrast is also noted by Yang & Pan (2001: 202). According to them, the contrast between (30b) and (33) is a consequence of a pragmatic requirement, referred to as “compatibility requirement”, on existential sentences. For dynamic manner verbs, the verbs have to describe a typical kind of movement of an entity. Swimming is a typical movement of ducks, but not of human beings; on the other hand, running is a typical movement of human beings. This explanation is similar to Birner’s (1994) study regarding the discourse function of existential sentences in English. Birner says that in inversion, the information carried by the post-verbal NP is less familiar than the information carried by the pre-verbal PP; further, the verb does not carry new information – it is informationally-light. In this context, we can see that movements typical of an entity carry a lighter load of information than movements that are not typical of an entity. This view can be used to explain some of the restrictions observed in Chinese, e.g. why tiao ‘jump’ cannot occur in locative inversion, as in (31c). Jumping is not informationally light in this context, but rather carries new information. However, discourse function alone does not seem to be adequate in explaining the entire
Feng-hsi Liu
range of Chinese data. Under this analysis, it is difficult to explain why except for pao ‘run’ and zou ‘walk’, no other verbs of volitional processes can occur in locative inversion. The following table summarizes the pattern of auxiliary selection that we have observed so far: (34)
Verb Types Auxiliary change of location/state – definite -le – indefinite -le /-zhe continuation of state, existence of state – agentive -le /-zhe – non-agentive -zhe non-volitional process -zhe volitional process (detransitivized) -zhe volitional process (intransitive), except for pao ‘run’ and zou ‘walk’
does not enter locative inversion
. Semantic determinants In this section we will consider what semantic properties are relevant for the choice of auxiliary in locative inversion. Before I offer my analysis, however, I will say a few words about Hu (1995) and Yang and Pan (2001). As mentioned in Section 1, they also offer a semantic analysis of auxiliary selection. . Hu (1995) and Yang and Pan (2001) Hu’s main concern is to characterize the verbs that can occur in presentative sentences, which correspond to locative inversion discussed here. He argues that the aspectual nature of the verbs determines whether they can occur in presentative sentences; further, the choice of aspect markers is determined by the aspectual nature of the verbs. Three groups of verbs are identified to enter presentative sentences: verbs denoting change of state, verbs denoting state, and verbs denoting resultative state. Change of state and resultative state select -le, while state selects -zhe. However, Hu does not recognize variability of choice of auxiliary; instead, when either choice is possible, he says there is a subtle difference in meaning between the two forms. According to him, (35a) describes a state as a result of some previous action, while (35b) describes an on-going state: (35) a.
Qiangshang gua -le yifu hua wall-on hang -le one-cl painting ‘On the wall hangs a painting.’
Auxiliary selection in Chinese
b. Qiangshang gua-zhe yifu hua wall-on hang-zhe one-cl painting ‘On the wall hangs a painting.’
Yang and Pan (2001) follow Hu and also take the aspectual properties of verbs as determinants of choice of auxiliary, although the details are not identical: change of state selects -le, resultative states select -zhe, and activities select -zhe. For verbs that exhibit variability, Yang and Pan suggest that the choice has to do with what the speaker wishes to emphasize; an emphasis on the change will result in the choice of -le, while an emphasis of the resultative state will result in the use of -zhe. Thus for them (35a) implies that the state of picture hanging was not true in the past. While neither study gives a complete picture, due to the narrow range of verbs considered, they have identified a major factor for the choice between -le and -zhe by associating change of state with -le. However, both studies are reluctant to acknowledge the existence of variability. Hu mentions subtle differences between the choices, while Yang and Pan talk about the speaker’s emphasis. Actually, in a situation where a picture is hanging on the wall, when asked if (35a) and (35b) can be used to describe the situation, almost all speakers respond positively, and most of them do not perceive a difference in meaning between the two sentences. In the next section, I will offer three factors to account for both the choice of marking and the variability. . Three factors Three semantic factors can be identified which account for the pattern observed in (34). Telicity is the primary factor, which divides all of the verbs into two groups: telic verbs, including verbs of change, and atelic verbs, including verbs of state and verbs of process. Telic verbs select -le. Difference in degrees of telicity is reflected in whether -zhe can also be selected. Verbs of definite change select -le only, while verbs of indefinite change select -le or -zhe. In contrast, atelic verbs all select -zhe; two other factors determine whether they also select -le: stativity and agentivity. Stativity divides atelic verbs into two groups: stative verbs and dynamic verbs; the latter are verbs of process. Dynamic atelic verbs (verbs of process) select -zhe only; as for stative atelic verbs, they are yet divided into two groups in terms of agentivity. Stative verbs that are non-agentive select -zhe only, while agentive stative verbs select -zhe or -le. The overall picture is a hierarchy given in (36):
Feng-hsi Liu
(36) Hierarchy of auxiliary selection in Chinese Semantic Properties telic telic atelic, stative, agentive atelic, stative, non-agentive atelic, dynamic
Verb Types change (definite) change (indefinite) state state process
Auxiliary -le -le/-zhe -le/-zhe -zhe -zhe
The three factors are not equal in terms of importance. Telicity is the most dominant factor. If a verb is telic, the other two factors are irrelevant; it selects -le regardless whether it is stative or agentive. Thus telic agentive, e.g. lai ‘come’, has the same marking as telic non-agentive, e.g. lan ‘be rotten’; and telic stative, e.g. po ‘be broken’, is marked the same as telic dynamic, e.g. zou ‘leave’. The factor that is next in ranking is stativity. If a verb is atelic and dynamic, then it selects -zhe regardless of agentivity. That is, all types of process select -zhe, whether it is volitional, e.g. jiao ‘chew’, or non-volitional, e.g. piao ‘flow’. The last factor, agentivity, is relevant only when a verb is atelic and stative. That is, it is only relevant when a verb denotes a state. A state that is brought out by volition takes either -le or -zhe, whereas a state that exists without intervention of an agent takes -zhe only. Telicity and stativity are both properties of lexical aspect. In Dowty’s (1979) verb classification, telicity distinguishes accomplishments and achievements from states and activities, while stativity separates states from the other three classes. In Smith (1997), telicity and stativity are two of the three features that define situation types. Therefore, overall, auxiliary selection in Chinese is governed by lexical aspect and agentivity. (36) can be compared with the hierarchy proposed by Sorace (2000) for auxiliary selection in Germanic and Romance languages, given in (37): (37) Auxiliary selection hierarchy change of location change of state continuation of pre-existing state existence of state uncontrolled process controlled process (motional) controlled process (non-motional)
Sorace (2000) (least variation) ‘be’
(least variation) ‘have’
Sorace’s hierarchy is set up based on telicity and agentivity. She takes a gradient approach. Verbs are distinguished into core and intermediate verbs. Core verbs are at the ends of the hierarchy; they show consistent marking, while intermediate verbs are in the middle of the hierarchy, and they show variable marking.
Auxiliary selection in Chinese
As we compare the two hierarchies, we note that the types of verbs that occur in the Chinese hierarchy are similar to the types of verbs listed in (37), but differ from the latter in terms of classification. Different types of changes and processes in (37) are collapsed in (36); on the other hand, states are distinguished in terms of agentivity in (36), a distinction lacking in (37). However, there are some striking similarities between the two hierarchies. First, just like (37), in the Chinese hierarchy verbs at the ends of the hierarchy show consistent marking, while verbs in the middle show variable marking. Second, both hierarchies include telicity and agentivity as determinants; further, the degree of telicity determines the degree to which the choice of ‘be’ or -le is categorical. Thus the Chinese data supports Sorace’s gradient approach to auxiliary selection. There are also a number of differences between (36) and (37). First, since not all intransitive verbs occur in locative inversion, (36) includes only a subset of intransitives in Chinese, whereas (37) covers a full range of intransitives in Germanic and Romance languages. On the other hand, (36) also applies to detransitivized verbs, although the two classes of verbs differ in terms of the verb types included in the hierarchy. For intransitive verbs, agentive process is excluded except for pao ‘run’ and zou ‘walk’; for detransitivized verbs, the full range of verb types is included. Another difference between (36) and (37) has to do with the division of labor among the factors. In (37), telicity and agentivity are relevant, but stativity is not. Sorace does not specify whether the two are comparable or ranked in terms of importance. It seems telicity should be ranked higher because while telic verbs (definite change) select ‘be’ regardless of agentivity, agentive verbs do not select ‘have’ regardless of telicity. Rather, only atelic, agentive verbs select ‘have’. In (36), as mentioned earlier, telicity plays a much more prominent role, while agentivity is relevant only to states. Yet another difference concerns the nature of variation observed in these languages. Chinese verbs exhibit less variation than verbs in Romance or Germanic languages in terms of both types of verbs and the nature of variation. Two types of verbs in (36) show variation in Chinese – indefinite change and state, while in Romance and Germanic languages most of the eight types in (37) display some degree of variation, with the middle showing the greatest variation. As to the nature of variation, in Chinese it seems that not much variation exists among native speakers. Judgments converge more or less as to whether a verb selects -le, -zhe or either. There is also less individual differences among verbs of the same type in Chinese. For example, when agentivity is implied, all stative verbs show variation. By contrast, in Romance and Germanic languages, more variation is seen among speakers as well as among individual verbs.
Feng-hsi Liu
. Syntactic approaches We now turn to the issue whether the data presented in Section 3 can be characterized in a syntactic analysis. Since auxiliary selection classifies verbs into -le marking verbs and -zhe marking verbs, a natural question to ask is whether the distinction is the unaccusative/unergative distinction, explained by the Unaccusativity Hypothesis. As mentioned in Section 1, in Germanic and Romance languages auxiliary selection has been considered as an unaccusative diagnostics. Is auxiliary selection in Chinese also a syntactic manifestation of unaccusativity? Below I will present arguments from two perspectives and suggest that a syntactic analysis of the phenomenon is untenable. If auxiliary selection is a characteristic of unaccusativity, two things should follow: First, we would expect to see other syntactic manifestations whereby intransitive verbs are distinguished into two classes along the same line and the difference is explained by the Unaccusativity Hypothesis. Second, earlier in Section 2, I have suggested that detransitivized verbs in locative inversion suppress their external argument. Given this, in a syntactic analysis, we would expect all of the detransitivized verbs to select -le uniformly, in the same way that passive verbs and reflexive verbs in Germanic and Romance languages select ‘be’ (Perlmutter 1978; Burzio 1986). Neither prediction is borne out, however. As for the latter, it is contradicted by the data presented in Section 3. There we saw that detransitivized verbs also show variable behavior; just like intransitive verbs, they select -le, -zhe or either. Further, except for controlled processes, the selection follows the same broad classification, i.e. change, state, process for both intransitive and detransitivized verbs. As for the former prediction, so far two phenomena have been proposed (Yu 1995) as unaccusative diagnostics, but as I show in 5.2, under close scrutiny they are not really unaccusative diagnostics. In fact, in studies of Chinese there has been no consensus on what is an unaccusative diagnostic or which verb types (e.g. change, state, process) are unaccusative. Below I will review three previous studies: Huang (1990), Yu (1995) and Pan (1996). . Huang (1990) Huang (1990) suggests that in Chinese verbs of existence, (dis)appearance and location have an underlying object, but no subject; they are unaccusative. On the other hand, agentive verbs such as ku ‘cry’ and tiao ‘jump’ have an underlying subject; they are unergative. The underlying representation of (38a) and (39a) is (38b) and (39b), respectively:
Auxiliary selection in Chinese
(38) a.
Ren si le person die le ‘This person is dead.’ b. [ e ] si le ren le. die le person prt ‘Someone is dead.’
(39) a.
Yiwai zhongyu fasheng le accident finally happen le ‘The accident finally happened.’ b. [ e ] zhongyu fasheng yiwai le finally happen accident le ‘An accident finally happened.’
Huang says a strong piece of evidence for assigning these verbs an object, but not subject, is that the ‘inverted’ structures of (38b) and (39b) are also well-formed. In contrast, with unergative verbs, such inversion is not possible: (40) a. *Hen gaoxing yige ren very happy one-cl person ‘A person is very happy.’ b. *Zheli ku-le sange ren here cry-le three-cl person ‘Here three people cried.’
That is, unaccusative verbs allow their single argument to occur in the object position at the surface structure. Another piece of evidence for the unaccusative/ unergative distinction that Huang offers is that the subject of unergative verbs is usually the agent, which is the external argument, while the subject of unaccusative verbs is usually a non-agent, corresponding to an internal argument. In (38a) and (39a) the subject is a theme, not agent. Therefore, the two types of verbs also differ in that unergative verbs have the external argument, while unaccusative verbs have no external argument, but an internal argument. Huang seems to take inversion as an unaccusative diagnostic. This means all of the intransitive verbs that enter locative inversion would be unaccusative in his analysis. Auxiliary selection therefore plays no role in unaccusativity. This view is also held in Gu (1992) and Li (1990). Thus unaccusative verbs would include not only verbs of existence, (dis)appearance and location, but also verbs of processes, eg. piao ‘float’, chui ‘blow’ and pao ‘run’. This raises some doubt, as cross-linguistically the latter verbs are unergative. Further, on the assumption that locative inversion is an unaccusative diagnostic, the post-verbal NP is an object in the underlying representation. Yet no such evidence is provided. In fact, it is questionable whether all of the post-verbal NPs that occur in the construction originate
Feng-hsi Liu
as objects. Pao ‘run’, for example, is usually assumed to have an underlying subject, not an underlying object. . Yu (1995) Yu (1995) offers a different view of unaccusativity in Chinese. He examines three groups of verbs, each containing six verbs, in locative inversion.2 He suggests that verbs that select -le are unaccusative, verbs that select -zhe are unergative, and alternating verbs, ones that select either -le or -zhe, are unaccusative. For -le marking verbs, the postverbal NP in locative inversion is an underlying object, while for -zhe marking verbs, the postverbal NP is an inverted subject, adjoined to VP. For alternating verbs, the postverbal NP is an underlying object when -le is selected, and an inverted subject when -zhe is selected. He provides two tests that correlate with the -le/-zhe selection: the definiteness effect and sub-extraction. The definiteness effect can be observed in -le marking sentences. When the verb is marked by -le, the postverbal NP is either indefinite or, if the NP is bare, interpreted as an indefinite expression. In a -zhe marking sentence, however, the NP can be definite. This is shown in (41)–(42). (41) Duimian lai-le yiliang che /*Laowang de che opposite-side come-le one-cl car /Laowang de car ‘From the opposite side came a car/*Laowang’s car.’ (42) Caochangshang pao-zhe yige xuesheng /women de laoshi playground-on run-zhe one-cl student /we de teacher ‘On the playground is running a student/our teacher.’
(43)–(44) show that if a verb takes either -le or -zhe, the NP can be either definite or indefinite when it takes -zhe, but the NP must be indefinite when the verb takes -le. (43) Taishang zuo-zhe sange ren /zhuxi tuan platform-on sit-zhe three-cl person /chair committee ‘On the platform are sitting three people /the chair committee.’
. The three groups of verbs are as follows: G1: si ‘die’, dao ‘arrive’, lai ‘come’, qu ‘go’, zou ‘leavel’, pao ‘run away’ G2: zhan ‘stand’, zuo ‘sit’, dun ‘squat’, gui ‘kneel’, tang ‘lie’, pa ‘lie on stomach’ G3: zou ‘walk’, pao ‘run’, fei ‘fly’, pa ‘crawl’, tiao ‘jump’, gun ‘roll’ Verbs in G1 select -le; verbs in G2 select -le or -zhe, and verbs in G3 select -zhe.
Auxiliary selection in Chinese
(44) Taishang zuo-le sange ren /*zhuxi tuan platform-on sit-le three-cl person /chair committee ‘On the platform are sitting three people/*the chair committee.’
The second test, sub-extraction, has to do with extracting the head of the postverbal NP to a pre-verbal position. It is possible with a -le marking verb, as illustrated in (45): (45) Nage diqu ren si-le xuduo that area people die-le many ‘In that area people died many.’
In (45), the head ren ‘people’ is preposed to a pre-verbal position, leaving behind the quantifier phrase xuduo ‘many’: (46) shows that sub-extraction of the head NP is not possible with -zhe marking verbs. (46) *Caochangshang xuesheng pao-zhe xuduo playground-on student run-zhe many ‘On the playground students were running many.’
If a verb can take either -le or -zhe, then sub-extraction is possible when it is marked by -le, but not when it is marked by -zhe. This is illustrated in (47): (47) Menkou ren zhan-le /*-zhe jige doorway people stand-le /-zhe a-few ‘At the doorway people stood a few.’
In Yu’s analysis, auxiliary selection is one of the three syntactic manifestations of unaccusativity, together with the definiteness effect and sub-extraction. However, although the latter two phenomena correlate with auxiliary selection, neither the definiteness effect nor sub-extraction is an unaccusative diagnostic in the strong sense. Levin and Rappoport Hovav (1995: 4) point out that in order for a phenomenon to be considered as an unaccusative diagnostic, the difference between the two classes of verbs needs to be explained by different syntactic configurations. In the case of definiteness effect or sub-extraction, there is no independent evidence that links them to the distinction between an object and a non-object. That is, it is not shown independently that in locative inversion, NPs in the object position, but not in other positions, must be indefinite, nor that sub-extraction is only possible in the object position, but not in other positions. Rather, Yu simply characterizes the differences in terms of differences in syntactic configurations. In fact, neither the definiteness effect nor sub-extraction is necessarily a syntactic phenomenon. In recent literature there have been semantic (Keenan 2003) and pragmatic (Abbott 1993; Zucchi 1995) accounts of the definiteness effect. As for sub-extraction, it is subject to discourse constraints. Sub-extraction involves mov-
Feng-hsi Liu
ing a post-verbal NP to a pre-verbal position, where an NP typically carries old information. If a post-verbal NP carries new information, when it is extracted to a pre-verbal position, the result is often unacceptable, regardless of whether the verb is marked by -le or -zhe, as in (48): (48) *Cantingli xuesheng zuo-le/-zhe yige cafeteria student sit -le/-zhe one-cl (Lit: ‘In the cafeteria, as for students, there sits one.’)
Thus the two tests that Yu provides may be explained by semantic, pragmatic or discourse analyses. Data from detransitivized verbs further undermines a syntactic analysis. The definiteness effect and sub-extraction also divide detransitivized verbs into two groups in the same way as intransitive verbs do – -le marking verbs are subject to the definiteness effect but allow sub-extraction, while the reverse holds for -zhe marking verbs. (49) illustrates the definiteness effect, and (50) illustrates sub-extraction: (49) a.
Dishang reng-le xuduo /*zhexie guopi zhixie ground-on throw-le many /these peels paper ‘On the ground many/*these peels and paper were thrown.’ b. Zuili jiao-zhe yikuai/ nakuai kouxiangtang mouth-in chew-zhe one-cl that-cl chewing-gum ‘In the mouth is chewing a piece of/that piece of gum.’
(50) a.
Guopi zhixie reng -le henduo peels paper throw -le many ‘Many peels and paper were thrown.’ b. *Kouxiangtang jiao-zhe bushao chewing-gum chew-zhe many ?*‘Many chewing gums are chewed.’
However, as discussed in 2.2, the two types of verbs have the same argument structure – both are assigned an internal argument, with the external argument suppressed. This means syntactic configuration cannot be the source of difference observed in (49)–(50). In short, while Yu has identified two phenomena that correlate with auxiliary selection, neither is an unaccusative diagnostic; moreover, the pattern displayed by detransitivized verbs cannot be stated as differences in the argument structure or syntactic configuration.
Auxiliary selection in Chinese
. Pan (1996) Pan (1996) treats all of the intransitive verbs in locative inversion as unaccusative, just like Huang (1990), but among the detransitivized verbs, only ones that select -zhe are unaccusative. As mentioned in Section 2.2, he proposes a rule that operates on a verb’s argument structure and deletes the verb’s external argument when -zhe is present. The derived argument structure is the same as that of an intransitive verb with a Theme argument, and therefore the derived verb, marked by -zhe, is considered unaccusative. On Pan’s analysis, auxiliary selection plays a role in unaccusativity only for detransitivized verbs. For him, unaccusative verbs include all types, e.g. change, state, and process, of intransitive verbs; but among detransitivized verbs, only verbs of state (if they select -zhe) and process are unaccusative. Notably, detransitivized verbs that denote change are not unaccusative in his analysis. Cross-linguistically, verbs of process are unergative, while verbs of change are unaccusative, and yet Pan has assigned just the opposite status to detransitivized verbs. Since no diagnostics of unaccusativity is provided, it is not possible to evaluate why unaccusative verbs are grouped this way in his analysis. . Summary In short, the strongest evidence against a syntactic analysis of auxiliary selection comes from detransitivized verbs. Rather than behaving uniformly, as predicted by a syntactic analysis, these verbs show variability, parallel to intransitive verbs. As for evidence for a syntactic analysis, so far in the literature only Yu (1995) provides it, but his two tests on close scrutiny don’t stand up as real syntactic diagnostics. It is also found that of the three syntactic approaches to split intransitivity, none of them have provided a clear unaccusative diagnostic. As a result, many discrepancies are found among the three analyses. There is disagreement on many verb types, e.g. verbs of process, alternating verbs, and detransitivized verbs, with respect to whether they are unaccusative or unergative. Auxiliary selection has been assigned three different roles with respect to unaccusativity: not relevant (Huang), very relevant (Yu) and partially relevant (Pan). Thus not only is the evidence for auxiliary selection as a syntactic manifestation of unaccusativity not strong, the evidence that unaccusativity is syntactically represented in Chinese is also lacking.
. Conclusion In this paper I have examined auxiliary selection in Chinese, a phenomenon that is found in locative inversion. Chinese is unique in that both intransitive and detran-
Feng-hsi Liu
sitivized verbs occur in the construction and both participate in auxiliary selection. I have suggested that this fact is best accounted for in a semantic analysis. Three semantic factors have been identified that are responsible for the hierarchy of Chinese auxiliary selection: telicity, stativity and agentivity, in order of importance. There are thus two semantic dimensions that Chinese auxiliary selection is sensitive to – lexical aspect, including telicity and stativity, and agentivity. Van Valin (1990) proposes that lexical aspect and agentivity are the semantic basis for split intransitivity cross-linguistically. Data from Chinese auxiliary selection certainly reinforces this claim. I also considered whether auxiliary selection in Chinese is a syntactic manifestation of unaccusativity. This question turns out to be difficult to answer because there is little agreement in previous analyses as to what is an unaccusative diagnostic in Chinese. Two tests have been proposed that correlate with auxiliary selection; however, both lack independent evidence as unaccusative diagnostics. This, together with data from detransitivized verbs, suggests that a syntactic analysis of auxiliary selection is not the right way to go. In sum, the Chinese data supports a semantic analysis of auxiliary selection. This does not mean that a syntactic analysis of split intransitivity cannot be put forward for Chinese, but at least one phenomenon – auxiliary selection – argues for a semantic account of split intransitivity.
References Abbott, B. (1993). A pragmatic account of the definiteness effect in existential sentences. Journal of Pragmatics, 19, 33–55. Aranovich, R. (2004). The semantics of auxiliary selection in Old Spanish. Studies in Language, 27, 1–37. Birner, B. (1994). Information status and word order: An analysis of English inversion. Language, 70, 233–259. Burzio, L. (1986). Italian Syntax: A government-binding approach. Dordrecht: Reidel. Dowty, D. (1979). Word meaning and Montague grammar: The semantics of verbs and times in generative semantics and in Montague’s PTQ. Dordrecht: Reidel. Grimshaw, J. (1990). Argument Structure. Cambridge, MA: The MIT Press. Gu, Y. (1992). On the locative existential construction in Chinese. In D. Bates (Ed.), 10th West Coast Conference on Formal Linguistics (pp. 183–195). Stanford, CA: The Stanford Linguistics Association. Hoekstra, T. (1984). Transitivity: Grammatical relations in government-binding theory. Dordrecht: Foris. Hoekstra, T. (1999). Auxiliary selection in Dutch. Natural Language and Linguistic Theory, 17, 67–84. Hu, W. (1995). Verbal semantics of the presentative sentences. Yuyan Yanjiu (Language Studies) 29, 100–112.
Auxiliary selection in Chinese
Huang, C.-T. J. (1987). Existential sentence in Chinese and (in)definiteness. In E. Reuland & A. ter Meulen (Eds.), The Representation of (in)Definiteness. Cambridge, MA: The MIT Press. Huang, C.-T. J. (1990). Two types of transitive verbs and intransitive verbs in Chinese. In Tang, Ting-chi et al. (Eds.), Dierjie shijie huayuwen jiaoxue yantaohui lunwen ji (Proceedings of the second international conference on Chinese language teaching) (pp. 39–59). Taipei: World Chinese Press. Keenan, E. (2003). The definiteness effect: Semantics or pragmatics? Natural Language Semantics, 11, 187–216. Legendre, G. (1989). Unaccusativity in French. Lingua, 79, 95–164. Levin, B. & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax-lexical semantics interface. Cambridge, MA: The MIT Press. Li, Y.-H. A. (1990). Order and constituency in Mandarin Chinese. Dordrecht: Kluwer. Lieber, R. & Baayen, H. (1997). A semantic principle of auxiliary selection in Dutch. Natural Language and Linguistic Theory, 15, 789–845. Pan, H. (1996). Imperfective aspect zhe, agent deletion and locative inversion in Mandarin Chinese. Natural Language and Linguistic Theory, 14, 409–432. Perlmutter, D. (1978). Impersonal passives and the unaccusative hypothesis. Proceedings of the Fourth Annual Meeting of the Berkeley Linguistic Society (pp. 157–189). Berkeley, CA: Berkeley Linguistic Society. Smith, C. (1997). The Parameter of Aspect. Dordrecht: Kluwer. Sorace, A. (2000). Gradients in auxiliary selection with intransitive verbs. Language, 76, 859– 890. Van Valin, R. D. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Yang, S. & Pan, H. (2001). A constructional analysis of the existential structure. In H. Pan (Ed.), Studies in Chinese Linguistics II (pp. 189–208). Hong Kong: Linguistic Society of Hong Kong. Yu, N. (1995). Towards a definition of unaccusative verbs in Chinese. In J. Camacho & L. Choueiri (Eds.), Proceedings of the Sixth North American Conference on Chinese Linguistics (pp. 339–353). Los Angeles, CA: University of Southern California. Zaenen, A. (1993). Unaccusativity in Dutch: Integrating syntax and lexical semantics. In J. Pustejovsky (Ed.), Semantics and the Lexicon (pp. 129–161). Dordrecht: Kluwer. Zucchi, A. (1995). The ingredients of definiteness and the definiteness effect. Natural Language Semantics, 3, 33–78.
Parameterized auxiliary selection A fine-grained interaction of features and linking rules* Janet H. Randall Linguistics Program, Northeastern University
Auxiliary selection is paradoxical: categorical for some verb classes, indeterminate for others; inconsistent across languages; and misaligned with other diagnostics. After critiquing one solution to the paradox, the Auxiliary Selection Hierarchy (Sorace 2000), we propose an alternative: Parameterized Linking, which approaches the semantic underpinnings of the unaccusative/unergative distinction and the syntactic-semantic mapping in a parameterized way. Languages can differ either in the Conceptual Structure features on which their linking rules are built – telicity, agentivity, locomotion – or on the linking rules they choose, or both, leading to different intransitive mappings. Using sets of parameterized, interacting, but still deterministic linking rules, the model can allow two features to produce a single continuum of gradience, predicting a hierarchy in auxiliary selection both within and across languages.
* This research profited from my collaborations with Angeliek van Hout, Jürgen Weissenborn and Harald Baayen, members of the former Argument Structure research project at the Max Planck Institute for Psycholinguistics in Nijmegen, where the experiments described here were carried out, with the assistance and support of the Institute and its staff. The writing of this article was supported by a Northeastern University Research Scholarship Development Fund grant to the author. I am grateful to Jill Carrier, Paulette Levy, and Antonella Sorace for much helpful discussion; to Luc Aalmans, Ralf Bernard, Boudewijn DeJonge, Angeliek van Hout, Ruth Juttermann-Bernard, Jose Wendel and Katherina Zippel for judgements; and to Rodolpho Ramina and Jonathan Dovev for technical assistance.
Janet H. Randall
. Introduction Three challenges have been raised to the classic view that auxiliary selection, in languages that exhibit it, is a syntactic diagnostic of intransitive verb class: (a) auxiliary selection does not always make the same split as other syntactic unaccusativity diagnostics; (b) within one language, a verb can choose different auxiliaries, or both, depending on the context; and, most strikingly, (c) across different languages, a verb in the same context can choose different auxiliaries. This article will focus on (b) and to a greater extent (c), the variation in auxiliary selection within and between two closely related languages, Dutch and German. After Section 1 lays out some of the data that challenge auxiliary selection as a reliable verb class diagnostic, Section 2 introduces one proposal for resolving these apparent contradictions, Sorace’s Auxiliary Selection Hierarchy (ASH), and two experiments on German (Keller & Sorace 2003) that test it. Section 3 critiques the ASH proposal and experiments, then offers an alternative approach, Parameterized Linking, based on Conceptual Structure (CS) representations and linking rules. This approach incorporates semantic features similar to Sorace’s – but in a parameterized way – to account for the variation both within and across languages. Section 4 briefly reviews experimental evidence from Dutch and German for Parameterized Linking. The final section briefly considers how Parameterized Linking offers the beginnings of a solution to the challenges faced by pure Projectionist or pure Constructionist models of the syntax-semantics interface, by bringing together aspects of the two models in a compatible way.
.
A brief review: Is auxiliary verb selection a reliable verb class diagnostic?
Since the Unaccusativity Hypothesis (Perlmutter 1978; Burzio 1986) identified two syntactic classes of intransitive verbs, auxiliary selection has been used as one diagnostic for distinguishing them, as shown for Dutch and German in (1). (For an overview, see Grimshaw 1987). (1)
unergatives Dutch Ik heb gelachen. German Ich habe gelacht. ‘I have laughed’
unaccusatives Ik ben vertrokken. Ich bin weggegangen. ‘I am left’
As shown in (2): unergative verbs have an external argument at the Argument Structure (AS) level, which maps to subject position at d-structure (DS). Unac-
Parameterized auxiliary selection
cusative verbs have an internal argument, which maps to object position, then moves to subject position.1
Auxiliary selection joins two other unaccusativity diagnostics for these languages, impersonal passivization and the attributive use of the past participle (Hoekstra 1984; Wunderlich 1985; Everaert 1986; Zaenen 1988, 1993; Grewendorf 1989; Hoekstra & Mulder 1990; Van Valin 1990; Seibert 1993; van Hout 1998). Impersonal passivization can apply to unergatives (3a) but not unaccusatives (3b), because it suppresses an external argument, which unergatives have but unaccusatives lack. (3)
Dutch
a. Er werd hardop gelachen. there was aloud laughed b. *Er werd stiekem verdwenen. there was sneakily disappeared German a. Es wurde (von Peter) getanzt. it was (by Peter) danced b. *Es wurde (von Peter) gefallen. it was (by Peter) fallen
Conversely, the attributive use of the past participle applies to unaccusatives (4b) but not unergatives (4a), which lack a necessary underlying object. (4)
Dutch
a. *de gelachen leerlingen b. de gevallen vaas the laughed pupils the fallen vase German a. *der geblutete Mann b. der ertrunkene Mann the bled man the drowned man
But as is well documented, these diagnostics do not always align. German unexpectedly forms an impersonal passive from fall in (5a), which according to (3b), is unaccusative (Zaenen 1988; Perlmutter 1978): (5)
German Es wurde von dem Schauspieler im richtigen Moment gefallen. there was by the actor in the right moment fallen
. AS, though represented on a separate line, is not an independent “level”, but is rather a shorthand expressing which DS arguments are internal and which are external, Though here “external” is defined relative to the VP node, other nodes are possible, depending on one’s theory of the clausal nucleus. The idea is simply that external arguments are “higher” than internal arguments.
Janet H. Randall
The attributive use of the past participle, though ungrammatical for bleed in (4a) and predicted to be grammatical for fall, gives opposite results in (6) and mixed results in (7) (Seibert 1993). (6)
German Der zu Tode geblutete Mann lag auf dem Boden. the to death bled man lay on the ground
(7)
German a. *das gefallene Kind the fallen child b. das auf den Boden gefallene Kind the onto the ground fallen child c. das runtergefallene Kind the downfallen child
And auxiliary selection is also problematic. A verb can switch auxiliaries depending on the subject, (8) (Keller & Sorace 2003) or other arguments it appears with, (9) (van Hout et al. 1993; Randall et al. 2004): (8)
(9)
German a. Die Täterin hat/?ist betreten dagestanden. the offender has/?is sheepishly there-stood b. Der Korb ?hat/ist unbeachtet dagestanden. the basket ?has/is unnoticed there-stood. a. have/*be Marie heeft uren in het rond gerend. Mary has run around for hours German Paul & Rita haben in dem Saal getanzt. Paul & Rita have danced in the room Dutch
b. *have/be Marie is in 5 minuten naar huis gerent. Mary is run home in 5 minutes German Paul & Rita sind in den Saal getanzt. Paul & Rita are danced into the room Dutch
Some verbs allow either auxiliary, such as (10) (Seibert 1993)2 and (11) (van Hout et al. 1993). (10)
German a. Der Reis hat/ist gekocht. the rice aux boiled b. Der Mann hat/ist geschwommen. he aux swum c. Paul & Rita haben/sind getanzt. Paul & Rita aux danced
. Seibert (1993) also accepts joggen (jog) and fahren (drive), though one reviewer accepts only tanzen.
Parameterized auxiliary selection
(11)
Dutch De temperatuur heeft/is 3 uurlang gestegen, maar is toen weer gezakt. the temperature aux 3 hours risen, but is then again dropped
Finally, and most strikingly, the same sentence can choose opposite auxiliaries even in closely related languages (Brinkmann 1992; van Hout et al. 1993; Randall et al. 2004): (12)
Dutch: John heeft urenlang door de zaal rondgedanst. German: John ist stundenlang durch den Saal herumgetanzt. John aux been dancing around the room for hours
Given these “unaccusative mismatches”,3 how can these diagnostics be reliable indicators of verb class? The explanation, according to most approaches, lies in semantic and aspectual characteristics of the verb and its arguments. Beneath the syntactic distinction that links “unergative” and “unaccusative” to external or internal position of an argument,4 the split has been tied to a number of semantic factors: telicity (Dowty 1991; Hoekstra 1984; Van Valin 1990; Tenny 1994; van Hout et al. 1993; van Hout 1998), transitionality /affectedness (Seibert 1993), change (van Hout et al. 1993); change over locations (Brinkmann 1995), directed change (Levin & Rappaport Hovav 1995), inferrable eventual position or state (Lieber & Baayen 1997); locomotion (Randall et al. 2004); and/or to properties of the subject: internal control (Levin & Rappaport Hovav 1992), internal/external/immediate causation (Levin & Rappaport Hovav 1995, and others). Considering these deeper properties is the key to resolving the seemingly contradictory data above.
. The Auxiliary Selection Hierarchy: An approach to unaccusative mismatches . The theory: Two semantic dimensions in a hierarchy of verb types Sorace (2000) proposed that auxiliary selection (and other syntactic behaviors) is sensitive to two of these “aspectual and thematic dimensions”: (a) telic change, correlating with be, and (b) agentive unaffecting process, correlating with have. Based on these she formulates a hierarchy of verb types, in (13).
. A term coined by L. Levin (1986). . Some investigators have abandoned the syntactic distinction altogether, claiming that the split is due to semantic differences alone. (Van Valin 1990; Dowty 1991; Seibert 1993; Lieber & Baayen 1997).
Janet H. Randall
(13) Auxiliary Selection Hierarchy (Sorace 2000) BE | | | | | HAVE
Change of location Change of state Continuation of a pre-existing state Existence of state Uncontrolled process Controlled process (motion) Controlled process (non-motion)
According to Sorace, core change-of-location verbs denote telic change; core nonmotion controlled process verbs denote agentive activity in which the subject is unaffected. These categories tend to make categorical auxiliary choices. Intermediate classes “incorporate telicity and agentivity to lesser degrees,” and “tend to have a less specified (basically stative) event structure”. They exhibit variation and inconsistency in auxiliary selection, both within and across languages, the degree correlating with their position on the hierarchy. Sorace cites evidence for her model from core and intermediate verbs. First, verbs at the top and bottom of the hierarchy, such as arrive in (14), and work in (15), are consistent across languages. (14)
Italian a. Maria è arrivata in ritardo. Dutch b. Marie ist laat gekommen. German c. Marie ist spät angekommen. ‘Mary is arrived late’
(15)
Italian a. I poliziotti hanno lavorato tutta la notte. Dutch b. De politieagenten hebben de hele nacht gewerkt. German c. Die Polizei hat die ganze Nacht gearbeitet. the policemen have worked all night
Second, within languages, core verbs’ auxiliary choices are not affected by other sentential elements: the core unaccusative verb, fall, takes be even with an agentive subject: (16)
German
Der Apfel/Peter ist gefallen. the apple/Peter is fallen
Third, the intermediate categories vary both within and across languages. Continuation of state verbs like last and continue prefer have in Dutch and German but be in Italian, though they allow have with agentive subjects (Sorace 2004: 259; 2000: 867–868).
Parameterized auxiliary selection
a. Het concert heeft/??is een hele tijd geduurd. the concert has/??is a long time lasted German b. Die Äpfel haben/*?sein den ganzen Winter gehalten. the apples have/*?are the whole winter lasted
(17)
Dutch
(18)
Italian
a. Il dibattito è continuato. the debate is continued b. Il presidente ha/è durato in carica due anni. the president aux lasted in post two years c. Mario ha continuato. Mario has continued
Existence of state verbs require have in Dutch (19a) and German (b) but be in Italian (c). (19)
Dutch
a. Het magische zwaard heeft echt bestaan. the magic sword has really existed German b. Die Dinosaurier haben wirklich existiert. the dinosaurs have really existed Italian c. I vampiri non sono mai esistiti. the vampires not are never existed
Uncontrolled processes5 (20) select have in Dutch (a) and German (b) but vary in Italian (c): (20)
Dutch
a. De zon heeft geschenen. the sun has shone German b. Der Zug hat laut gerumpelt. (K&S 2003: 25) the train has noisily rumbled Italian c. Il tuono ha/è rimbombato. (from Sorace 2004) the thunder aux rumbled
Finally, controlled process motion verbs in (21) select be with a telic PP, (b), but select have when the motion is construed as not involving a transition, (a): (21)
a. have/*be Luigi ha corso velocemente Luigi has run fast Dutch Max heeft uren in het rond gerend. Max has run around for hours German Paul & Rita haben in dem Saal getanzt. Paul & Rita have danced in the room Italian
. These are internally caused but non-agentive, with no volitional “actor”.
Janet H. Randall
b. *have/be Luigi è corsa alla stazione Luigi is run to the station Dutch Max is in 5 minuten naar huis gerent. Max is run home in 5 minutes German Paul & Rita sind in den Saal getanzt. Paul & Rita are danced into the room Italian
. Experimental evidence: Keller & Sorace (2003) Sorace cites further evidence for the Auxiliary Selection Hierarchy (ASH) from experimental work. Two studies (Keller & Sorace 2003) focused on German. .. Experiment 1 In Experiment 1, native German speakers judged the relative acceptability of sentences across 8 verb classes, (22).6,7 (22) Auxiliary Selection Hierarchy (Keller & Sorace 2003) B (a) Change of location abreisen ‘depart’ | (b) Change of state versterben ‘die’ | (c) Continuation of a pre-existing state verweilen ‘stay’ | (d) Existence of state (position) schweben ‘hover’ | (e) Uncontrolled process (involuntary reaction) taumeln ‘stagger’/ | zittern ‘shiver’ | (f) Uncontrolled process (emission) klappern ‘rattle’ | (g) Controlled process (motion) rennen ‘run’ H (h) Controlled process (non-motion) reden ‘talk’
For each class, 16 test sentences were constructed using 8 German verbs and the two auxiliaries. Each subject saw 16, two from each class. Seven of the eight classes had animate subjects. (Class (f) had inanimate subjects.) (23) illustrates a test sentence of class (f), which was presented with one of the two auxiliaries. (23) (f) Der Zug hat/ist laut gerumpelt. the train aux noisily rumbled
. In a “magnitude estimation” task, 54 subjects (mean age, 28) assigned each stimulus a number representing acceptability relative to a previously presented “modulus item”. The study also tested impersonal passives and investigated dialectal variation, which I will not discuss here. . The 8 classes were formed from the 7 in (13) by splitting the uncontrolled process verbs into two types: (e) involuntary reaction and (f) emission. Class (e) contained both motion (’stagger’) and non-motion (’shiver’). Class (d) was limited to position verbs.
Parameterized auxiliary selection
According to Keller & Sorace, the ASH makes two predictions: (24) a.
Auxiliary selection will be categorical for core verbs and gradient for intermediate verbs. b. Subjects will prefer be for transitions and states and have for atelic processes.
(24a) follows from the Hierarchy: class (a) should elicit be and class (h), have. Classes (b)–(g) should elicit weak preferences, if any. However, prediction (24b) is problematic. States, and all transitions but change-of-locations, being intermediate, should elicit have, not be. We will return to this in Section 2.3.2, below. The results, in the left-most column of (25), showed that the two core classes, (a) and (h), behaved as predicted. However, for the intermediate classes, all but (d) significantly preferred one auxiliary over another. (The lower case “have” for class (e) indicates that the preference was weak.)8 (25) Results of Experiment 1 be B (a) Change of location be | (b) Change of state have | (c) Contin. of pre-existing state no prefer. | (d) Existence of state(position) have | (e) Uncontr. process (invol. react.) | have | (f) Uncontr. process (emission) be | (g) Controlled process (motion) have H (h) Controlled process(non-motion)
abreisen ‘depart’ versterben ‘die’ verweilen ‘stay’ schweben ‘hover’ taumeln ‘stagger / zittern ‘shiver’ klappern ‘rattle’ rennen ‘run’ reden ‘talk’
.. Experiment 2 To investigate the unexpected strong preferences among the intermediate verbs in Experiment 1, Keller & Sorace ran a follow-up, Experiment 2, on 66 subjects, using the same procedure, instructions, and design, but modified materials. Experiment 2 also tested two assumptions that they claim are inherent in the ASH: (26) Assumptions (a) For core verbs, auxiliary selection is exclusively determined by the verb’s inherent lexical-aspectual features. Moving from the core, verbs become increasingly sensitive to other characteristics of the verb and non-lexical features of the predicate. (b) While telicity is the main factor separating be and have verbs, agentivity further distinguishes among have verbs. . Overall, the results showed a significant interaction of auxiliary and verb class, indicating that the verb classes differ in their auxiliary selection behavior.
Janet H. Randall
It is possible that in Experiment 1, class (b), change-of-state verbs, may have strongly preferred be because it included telic (erscheinen ‘appear’) with atelic (wachsen ‘grow’) verbs. Experiment 2, whose predictions are shown in (27), split these classes into: (k), atelic verbs (rosten ‘rust’), which become telic when prefixed, (j) (verrosten ‘rust’). The prefixed (j) forms were predicted to invoke be more than the unprefixed (k) forms. Class (c), continuation-of-state verbs, may have clicited a preference for have due to animacy,9 which they expected was relevant for all the intermediate classes. To test this, Experiment 2 changed the subjects of class (c) and (d) to inanimate, (l) and (m), which should elicit weaker preferences for have. For (e), uncontrolled process (involuntary reaction) verbs, which also elicited have, they noted that the class collapsed two subtypes: motion, taumeln (‘stagger’), and non-motion, zittern (‘shiver’). Since for German, motion, not telicity, has been claimed to be the relevant factor for unaccusativity (van Hout et al. 1993; Randall et al. 2004), Experiment 2 used only non-motion verbs (o). If the motion verbs raised the be responses in Experiment 1, type (o) verbs should drop it. In addition to the six modified classes, the two core classes (i), (p), served as controls. (27) shows the 8 classes, (i)–(p), the classes of Experiment 1 that they correspond to, the predicted auxiliaries and, in the leftmost column, the results. Again, weaker preferences are in lower case:10 (27) Predictions and Results of Experiment 2 results predictions be be [i] [a] change of location be be [j] [b] change of state (prefix)
‘depart’ ‘grow up’
no pref. no pref.
[k] [b]
‘grow’
have no pref. no pref. have have
[l] [m] [n] [o] [p]
be be have have have
[c] [d] [d] [e] [h]
[+animate] [–animate/ +telic] change of state (no prefix) [–animate/ –telic] continuation of state [–animate] existence of state (position) [–animate] existence of state (position) [+animate] uncontrolled process (-motion) [+animate] controlled process (-motion) [+animate]
‘continue’ ‘hover’ ‘hover’ ‘shiver’ ‘talk’
. Recall that with animate subjects, position verbs mean “act (volitionally) to maintain a position”, while with inanimate subjects, they mean “be in a position” and, as shown in (8), prefer different auxiliaries. This intuition is confirmed in English: only (8a) passes the Actor test (Jackendoff 1990: 126): (i)
a. What the offender did was stand there sheepishly. b. *What the basket did was stand there unnoticed.
. Again, the results (from 66 subjects) showed a significant interaction of auxiliary and verb class.
Parameterized auxiliary selection
Considering the two sets of results together, consistent with (26a), the “core” classes yielded the clearest judgements11 and most indeterminate were the intermediate existence of state classes (d, m, n). However, counter to (26) were the be results for class (b/j), telic change-of-state verbs, and the lack of an animacy effect either for (c) vs. (l) continuation of state or (m) vs. (n) position verbs. Overall, these results point to telicity as a factor in auxiliary selection, but not animacy.12 An important but unexpected finding, which we will return to below, was that class (g) controlled process motion verbs in Experiment 1 selected be just as categorically as core change-of-location verbs (i/a). Keller & Sorace adopt the suggestion (van Hout et al. 1993; Randall et al. 2004) that in German “locomotion” (motion involving displacement) is what determines unaccusativity for change-of-location verbs. This is supported by uncontrolled process verbs, (e), whose weak preference for have strengthened once motion verbs were removed in (o). . Questions and issues A number of questions arise both from the ASH as a model of auxiliary selection and from Keller & Sorace’s experiments to test it. .. Two features – one hierarchy One issue is how to encode two features on a one-dimensional continuum. The ASH ranks verbs on telicity. But telicity and agentivity are not inversely related: verbs can be [+telic +agentive]. To address this, Sorace weights telicity more, claiming in (26b) that agentivity matters only for [–telic] verbs. But this weighting does not follow from the Hierarchy. (And it may not be true; core telic verbs were only tested with [+animate] subjects.) .. Contradictory assumptions? Intermediate verbs should have weak preferences or none. Yet (24b) claims that states (c, d) and change-of-states (b), being transitions, should elicit be. And the goal of Experiment 2 was to show preferences among the intermediate verbs based on agentivity. Moreover, if agentivity is a factor in the intermediate verbs, then why is class (f), which requires [–agentive] subjects lower on the hierarchy than . However, strikingly, in both experiments, both core verb classes elicited some opposite responses and in Experiment 1, change-of-location verbs (a) elicited more have responses than change-of-state (b) verbs. See Keller & Sorace (2003) for details. . However, the weak vs. strong have responses in (o) vs. (p) suggests that while [±animate] is not relevant, a feature like [±agent] or [±actor] is. But the lack of an effect in the other classes differing on this feature (l) vs (c), (m) vs. (d), and (j) vs. (b), is still not explained
Janet H. Randall
(e), which can take [+agentive] subjects? And why was no experimental evidence for agentivity found? .. More features? As mentioned in Section 2.2.2 above, the strong preference of class (g) [–telic] motion verbs for be, counter to its ranking, led Keller & Sorace to adopt our proposal that [±locomotion] is a relevant feature for distinguishing verb class in German. But how would a third feature fit into the Hierarchy? .. “Gradient” auxiliary selection? For non-core verbs, Keller & Sorace predict “gradient auxiliary selection”: “what is gradient are the lexical-aspectual representations of individual verbs, which make them compatible with unergative or unaccusative syntax or both.” (K&S Fn.3) Does each verb have two representations? An underspecified representation? Sorace (2004) says that we need a model that explains why gradience exists in the non-core verbs. The ASH describes the output of such a model, but it is not the model itself. .. “BE-verbs” and “HAVE-verbs”? The ASH classifies only the two core classes as true have or be verbs; only changeof-location verbs, (a), are core telic verbs. But class (b) change-of-state verbs are also telic and patterned just like class (a) verbs in Experiment 1. According to (26b), telicity separates be from have verbs; agentivity distinguishes among have verbs. In more recent work, Sorace (2004: 1) says: “languages may have different cut-off points along the hierarchy.” But the ASH, at this point, does not integrate this suggestion. .. Known verbs One final issue is methodological. The experiments testing the ASH used actual verbs. But German, one reviewer suggests, may have some frozen auxiliaries.13 If so, then the actual factors underlying unaccusativity for German speakers would be better revealed by experiments using made-up verbs.
. See Lee-Schoenfeld (this volume), who takes the position that auxiliaries in German are to a large degree frozen, and not indicators of verb class.
Parameterized auxiliary selection
. Parameterized Linking: An alternative account We now turn to an alternative proposal for resolving the auxiliary selection challenges mentioned above: (a) inconsistent diagnostics, (b) language-internal variation and indeterminacy, (c) cross-linguistic auxiliary differences and (d) frozen auxilaries. This approach, Parameterized Linking, shares some features of the ASH but avoids many of its shortcomings. . Linking rules, compositional linking, and language-internal variation Underlying the syntactic distinction between unaccusatives and unergatives is a difference in the positions of their arguments at the level of Argument Structure (AS), as shown in (2). But these arguments are mapped to their AS positions by linking rules, from the level of Conceptual Structure (CS), as shown in (28) (Jackendoff 1990; Carrier & Randall 1993; Randall forthcoming).
CS encodes semantic representations, consisting of predicates, symbolized by variables (P, R) and their arguments (x, y) and it is at this level where semantic/aspectual features are registered. The distinct AS linkings result from differences in these features, which the linking rules are tied to. We can see how linking works by beginning with linking rules for two features: [actor]14 and [telic].15 . This is Sorace’s [agent], and we will use the terms [actor] and [agent] interchangeably throughout, setting aside their differences. These features also roughly correlate with: [control], [internal cause], [no direct external cause], [volitional cause]. Space prevents us from discussing the differences among them, which are not relevant here (see, for example, Jackendoff 1990). . So far, because our discussion of telicity has been in the context of Sorace’s work, we have been using the term ‘telic’ as she has. However, it is used in different ways in the literature. From here on, we will use it in its classic sense: a telic event is one that has an intrinsic endpoint. Atelic events are simply processes without outcomes that can stop at any time; they have arbitrary final endpoints (see Comrie 1976; Smith 1997: 19). (van Hout (1998), in contrast, considers as telic not just events that reach endpoints but all event types with a moment of temporal transition, including inchoative and causative events.)
Janet H. Randall
The Actor Linking Rule, (29), applies in predicates that contain act. It links the CS argument of this predicate, x (an “actor”), unergatively, to the external AS position. CS actors pass the “actor test” (Jackendoff 1990): What John did was: sing, dance (on the table), *arrive, *disappear. (29) Actor Linking Rule
(laugh, sing, dance (on the table))
ACT
The Actor Linking Rule links (30) unergatively in both Dutch and German. (30)
Dutch: John heeft urenlang op de tafel gedanst. German: John hat stundenlang auf dem Tisch getanzt. John has been dancing on the table for hours
Telic predicates contain INChoative BE, which means “come to be” (Jackendoff 1990) and is sometimes expressed as BECOME. The Telicity Linking Rule, (31), links an argument, x, that comes to be at a new place or state unaccusatively, to an internal AS position. (31) Telicity Linking Rule AS: ( a ) | CS: INC BE ( x , AT ... )
(arrive, appear, dance into the room)
(31) applies to all verb CSs containing INC BE (arrive, appear). But it also applies when the INC BE is supplied by a PP, as in dance into the room, in (32). (32)
Dutch: John is in 2 sekonden de kamer in gedanst. German: John ist in 2 Sekunden ins Zimmer getanzt. John is danced into the room in 2 seconds
To form the CS of the VP, the CS of [–telic] dance in (33) combines with the CS of into in (34), a two-place predicate containing INC be (Carrier & Randall 1993). The [+telic] result, in (35), means: “go into the room by dancing” with the CS for dance subordinated under the CS for into: (33) ACT
(34) into
INC BE (x , AT (y) )
Parameterized auxiliary selection
(35)
ACT
Linking proceeds CS hierarchically from the highest clause. INC BE triggers the Telicity Linking Rule and the entire VP links unaccusatively. Notice that although the Actor Rule is also satisfied in (35) by the presence of act, the geometry of CS prevents it from applying. As a result, dance links differently depending on the arguments it appears with, resulting in the different auxiliaries in (30) and (32). We now have an explanation for the first challenge with which we began, how a single verb can choose opposite auxiliaries in the same language. Notice that there is no inherent relationship between the features [telic] and [agent] or the rules based on them. But a hierarchical weighting is imposed on them by the geometry of CS, with the result in (36). (36)
telic –
+ – BE
x
agent
x A
+ x
HAVE
B
All [+telic] predicates (e.g., dance into the room), to the left of A, link unaccusatively. [+agent] predicates, provided they are [–telic] (e.g., dance on the table), to the right of B, link unergatively. Predicates between the lines are either indeterminate or are linked with a default rule. We discuss these possibilities below. Notice that the inherent weighting in (36), interestingly, coincides with Sorace’s assumption (26b) that [±agent] affects auxiliary choice only for [–telic] predicates. But here it does not have to be stipulated. And her distinction in (26a) between fixed “core verb” outcomes (outside the lines) and variable “intermediate verb” outcomes (between the lines) also falls neatly out of this two rule system. . Cross-linguistic variation: Parameterized linking rule systems Interactions among linking rules can also explain seeming contradictions in auxiliary selection across languages. The Locomotion Linking Rule, (37), applies to CSs containing [+locomotion] to link an argument, x, internally.16 . This rule should be taken as a first approximation, since probably the [+location] feature in dance around the room is not affixed to the DO function of the verb but is in the higher clause
Janet H. Randall
(37) Locomotion Linking Rule
(dance into the room, dance around the room)
Locomotion predicates, like telic predicates, are derived compositionally, depending on the PP. They include both [+telic] predicates, dance into the room, and [–telic] predicates, dance around the room,17 but not dance on the table or wiggle, which involve motion, but not locomotion. Recall that Dutch and German, though matched on the auxiliaries for [+telic] sentences like (32) and [–telic] sentences like (30), are not completely aligned. (38) (=12) takes have in Dutch and be in German. (38)
Dutch: John heeft urenlang door de zaal rondgedanst. German: John ist stundenlang durch den Saal herumgetanzt. John aux been dancing around the room for hours
(38) is not telic like (32). So the Telicity Linking Rule does not apply. However it is [+locomotion], whereas (30) is [–locomotion]. The three cases are shown in (39): (39)
contributed by the PP. What is important here is that the rule is triggered by a [+locomotion] feature in the predicate. . These denote no endpoint or necessary transition: Mary can “dance around the room” and end up where she started.
Parameterized auxiliary selection
This difference would result if the languages used two different features – and two different linking rules – to split these manner-of-motion intransitives, Dutch, telicity; German, locomotion. The Locomotion Linking Rule does not replace the Telicity Linking Rule in German. For change-of-state predicates, (40), German, like Dutch, uses telicity. (40)
a. Dutch De klei heeft in de zon gedroogd. German Der Ton hat in der Sonne (die ganzen Tag) getrocknet. the clay has in the sun (the whole day) dried b. Dutch De klei is in de felle zon helemaal uitgedroogd. German Der Ton ist in der glühenden Sonne ausgetrocknet. the clay is in the blazing sun out-dried
The effect is clear in cases like (41), which change telicity with a prefix: (41)
a. Dutch De man heeft gebloed. German Der Mann hat geblutet. the man has bled b. Dutch De man is doodgebloed. German Der Mann ist verblutet. the man is bled to death
The sets of rules are parameterized across the two languages; German selects both, Dutch only one: (42) Telicity Linking Rule Locomotion Linking Rule
Dutch German x x x
So while Dutch can be characterized by the tree in (36), German requires the additional [locomotion] feature, as in (43).18
. One question arises about (38). The Locomotion Linking Rule clearly applies, not the Actor Linking Rule, though the predicate contains act and CS geometry does not appear to be embedding act under [+locomotion] to rank these rules. This raises the possibility that when the geometry of CS allows two linking rules to apply, they may be extrinsically ordered. We leave this to further investigation.
Janet H. Randall
(43)
locomotion +
–
–
telic – +
telic +
agent – + BE
x
x
x
x A
x
HAVE
B
Notice that this tree, with two cuts, essentially depicts Keller & Sorace’s data in (25) and (27): all motion verbs, whether [+telic] (a–i) or [–telic] (g), as well as [+telic] change-of-state verbs (b, j), all falling to the left of A, select be; non-motion controlled processes [–telic] [+agent], (h, p) strongly select have. But uncontrolled processes, (o), which fall between the lines, show a weaker preference. Though (43) appears to account for German, the complete picture may be even more complex. Linking rules could be built on finer semantic distinctions, which would make additional cuts in the motion verbs: directed change (Levin & Rappaport Hovav 1995) or inferrable eventual position or state “[ieps]” (Lieber & Baayen 1997). It so, auxiliary selection would vary cross-linguistically depending on which features a language used, as shown in (44). (44)
+telic –telic +ieps –ieps BE
HAVE
+directed –directed change change +loco motion dance into the room Dutch
dance towards the door
dance along the wall
dance around the wall
–loco motion dance on the table
German
Parameterized auxiliary selection
And while some languages might use additional rules, it is possible that the Telicity Linking Rule is universal. This would produce the invariant behavior of Sorace’s core telic verbs across languages.19 Linking rules in other semantic domains could also be built on a range of features that languages could choose from, as in (45).20 (45)
–internal +internal cause cause BE
HAVE
–agent/ +agent/ actor actor –volitional +volitional
IC
A
V
The difference in (46a) and (47a) (Sorace 2000), for example, suggests that Italian and German split uncontrolled process verbs along different lines, either as a result of using different features or of using linking rules in a different way. (46)
(47)
Italian
a. Il tuono ha/è rimbombato. the thunder aux rumbled German b. Der Zug hat laut gerumpelt. the train has noisily rumbled
(from Sorace 2004) (K&S 2003: 25)
Italian
a. Paolo ha riso fino a sentirsi male. Paolo has laughed until he felt sick German b. Paul hat gelacht. Paul has laughed
. It would also explain our experimental findings that in selecting auxiliaries for locomotion verbs, German 4-year olds behave like Dutch speakers, not like German adults, incorrectly using telicity instead of locomotion as the relevant feature (Randall et al. 2004). . We are not distinguishing between actor and agent here, though a thorough discussion of these features would. On the distinction between agents and causers (non-volitional initiators), see Arad (1998).
Janet H. Randall
And (45) could also account for indeterminate auxiliary choice as in (46a). If Italian uses an Actor Linking Rule linking [+act] predicates externally and an Internal Cause Linking Rule linking [–internal cause] predicates internally, then all other predicates, falling between the internal cause and actor splits, would be indeterminate. Another parameter emerges from how languages formulate linking rules. Stative existence verbs, (48), (from (19), above) are unaccusative in English and Italian but unergative in Dutch and German. (48)
English Italian
a. There exist three versions of the manuscript. b. I vampiri non sono mai esistiti. the vampires not be never existed ‘Vampires never existed.’ Dutch c. Het magische zwaard heeft echt bestaan. the magic sword have really existed German d. Die Dinosaurier haben wirklich existiert. the dinosaurs have really existed
The same basic Existence Linking Rule could be built on the semantic primitive be, set to link in two different ways, the Italian/English rule internally, the Dutch/German rule, externally. (49) Existence Linking Rules a. Italian, English
(exist, seem, belong) b. Dutch, German
. Where linking rules do not decide: Indeterminate linking, default linking, multiple CSs and competing rules We have shown how the interaction of linking rules in the Parameterized Linking approach produces the cross-linguistic Dutch/German contrast in (40). Where no rules apply, for cases that fall between the lines in (36) and (43), we have predicted indeterminate auxiliary selection. However, there is another possibility: linking could be assigned by default, as in the unaccusative “Default Linking Rule” proposed for English and Italian (Levin & Rappaport Hovav 1995). In (9), repeated in (50), where there is no possibility in (a) of a [+actor] feature, this could account
Parameterized auxiliary selection
for a preference for be in Italian. (The acceptability of have in (b,c) exists to the extent that a [+actor] reading is possible.21 (50)
Italian a. Il dibattito è continuato. The debate is continued b. Il presidente ha/è durato in carica due anni. the president aux lasted in post two years c. Mario ha continuato. Mario has continued
In contrast, Dutch and German, which select have for the same class of verbs, as shown in (18), above, could have an unergative Default Linking Rule. This would explain the clear choice of have in (51a), which is [–telic][–agent] (in contrast to [+telic] (51b), which selects be):22 (51)
Dutch a. Het water heeft urenlang langs de muren gelopen. the water has hours-long along the walls walked b. Het water is binnen een minuut de zee ingelopen. the water is within a minute the sea in-walked
A Default Linking Rule also offers a second possible explanation for the difference in (48). If some or all of these languages had no Existence Linking Rule, then differently set Default Linking Rules, unaccusative for English/Italian (a, b) and unergative for Dutch/German (c, d), would produce the same results. Cases that a Default Linking Rule cannot explain are (52) (=10) and (53) (=11), where both auxiliaries are allowed. If German and Dutch had an unergative (or even an unaccusative) Default Linking Rule, then these should unambiguously select one auxiliary. (52)
German a. Der Mann hat/ist geschwommen. the man aux swum b. Paul & Rita haben/sind getanzt Paul & Rita aux danced c. Der Reis hat/ist gekocht. the rice aux boiled
(53)
Dutch De temperatuur heeft/is 3 uurlang gestegen, maar is toen weer gezakt. the temperature aux 3 hours risen, but is then again dropped
. Durato in (b) may be ambiguous like English stay, which can have a volitional reading. . van Hout (1998), the source of these examples, claims that the choice of have here argues against the [+agent] feature. But as is clear, the result arises naturally from the interaction of ranked Telicity and Agent rules and a have default for [–agent] predicates.
Janet H. Randall
The explanation here comes from an ambiguity in whether or not a transition has occurred (Rosen 1984; Seibert 1993). The two interpretations for each verb correspond to two different CSs. The CS of cook, on its telic change-of-state interpretation, contains INC BE, and links unaccusatively with the Telicity Linking Rule. Swim and dance, if conceptualized as [+locomotion] predicates, will link unaccusatively with the Locomotion Linking Rule. On their process interpretations, these verbs contain neither INC BE nor [+locomotion], so act in their CSs will trigger unergative linking with the Actor Linking Rule. Notice that invoking multiple CSs for these verbs does not run into the problem cited for “Projectionist” theories of the syntax-lexicon interface that assign each verb a different CS for every syntactic context that it appears in.23 They would assign run one CS for run around the woods, another for run in the woods, a third for run into the woods and so on, and would do so for every manner-of-motion verb whose auxiliary shifts with its PP. We are invoking double listing only when the verb itself has two readings. Though we will not explore this issue here, one more source of indeterminacy language-internally may be linking conflicts. It is possible that features that trigger two opposite linking rules are both present at the same CS level. If there is no weighting between them, then both auxiliaries may surface. . Parameterized linking rules and inconsistent diagnostics A final challenge that Parameterized Linking can answer is conflicting diagnostics within a language. Recall that German auxiliary selection does not always line up with impersonal passivization (IP). Fall takes be, even with a [+agent] subject, because it is linked as an unaccusative, by either the Telicity or Locomotion Linking Rule. (54)
German Der Apfel/Peter ist gefallen. the apple/Peter is fallen
IP has been claimed to apply to unergative verbs, like (55), but not unaccusatives, because it requires an external argument. So fall should block IP, as in (56a). (55)
German a. Es wurde (von Peter) getanzt it was (by Peter) danced Dutch b. Er werd hardop gelachen There was aloud laughed
. For an illuminating discussion of Projectionist and Constructionist theories, see Sorace (2000, 2004).
Parameterized auxiliary selection
(56)
German a. *Es wurde (von der Apfel/von Peter) gefallen. it was (by the apple/by Peter) fallen Dutch b. *Er werd stiekem verdwenen. there was sneakily disappeared
But fall allows IP when the subject is a volitional agent: (57) Es wurde von dem Schauspieler im richtigen Moment gefallen. there was by the actor in the right moment fallen
This would be explained if Impersonal Passivization in German used a different feature from auxiliary selection. If what matters for IP is solely [+agent], it will apply to verbs with either linking pattern – provided they have act in their CS.24 The results of one test will cross-cut the other, but now no conflict arises. If this is the case, then verbs like dance, which as we saw, link unergatively with some PPs, (58a), but unaccusatively with others, (58b, c), should form impersonal passives across all PPs as long as they have an agentive meaning. This prediction is borne out, as the corresponding IP cases in (59) show. (58)
German a. Paul & Rita haben in dem Saal getanzt. ‘Paul & Rita have danced in the room.’ b. Paul & Rita sind in den Saal getanzt. ‘Paul & Rita are danced into the room.’ c. John ist stundenlang durch den Saal herumgetanzt. ‘John is been dancing around the room for hours.’
(59) a. Es wurde in dem Saal getanzt. b. Es wurde in den Saal hineingetanzt. c. Es wurde durch den Saal getanzt.
That [agent] is the relevant feature for IP has also been claimed for Dutch (Pollman 1970; cited in van Hout 1998). Whistle in (60), though compatible with both [+agent] and [–agent] subjects (a), is unambiguously [–agentive] in the impersonal passive, (b): (60)
Dutch a. Piet/De ketel floot. Piet/the kettle whistled b. Er werd gefloten. somebody/*something whistled
. Zaenen (1988) calls the relevant feature “protagonist control”, Wunderlich (1985) and Seibert (1993), “agent”. In support of this, in Keller & Sorace’s (2003) experiments, while auxiliary choice was not sensitive to animacy, impersonal passivization was.
Janet H. Randall
. Support for Parameterized Linking: Experimental data One feature of the Parameterized Linking approach that is missing from the ASH is its capacity to include different sets of linking rules for different languages. By positing an additional linking rule based on a [locomotion] feature in German but not in Dutch, this theory accounts for the striking difference between these two languages on [–telic] [+locomotion] predicates like dance around the room, which are unaccusative in German and unergative in Dutch. Interestingly, Keller & Sorace’s experiments found evidence for this feature, yet it was not predicted by the 2-feature ASH. Keller & Sorace’s studies also found evidence for telicity in auxiliary selection, but as mentioned above, they failed to find an effect of agentivity. And they had a methodological shortcoming: in using actual verbs, with possibly frozen auxiliaries, they may not have tapped into the features speakers productively use to make verb classifications. A way to get around this problem is to confront speakers with new, unfamiliar verbs (for new, unfamiliar, actions), that they must categorize for the first time. Using novel verbs also makes it possible to systematically vary their features, control exactly what speakers know about each verb, and ensure uniform input across speakers, even speakers of different languages. These issues were addressed by a series of experiments that tested the three features, [±actor], [±telic] and [±locomotion], experiment 1 on Dutch adults and children (van Hout et. al. 1993) and experiment 2 comparing Dutch and German adults and children (Randall et.al 2004). Though we will only briefly sketch the experiments and the results for the adults here, they provide strong evidence for Parameterized Linking. In both experiments, after watching an experimenter act out a scene showing a novel action (using balls, puppets, stones, boxes, etc.) and describe it with a novel verb (live in Experiment 1, on videotape in Experiment 2), subjects were taught the new verb and then asked to help a forgetful puppet complete a sentence of the form in (61), by producing the new verb in participle form and crucially, the auxiliary that he or she thinks goes with it: (61) I saw that Jan [PP ] ge-[VERB ] [AUX ]
The verb/scene pairs were constructed to systematically vary the three features. One [–telic –locomotion +actor] scene (from Experiment 2) shows an Ernie puppet sitting on a pile of books, continually wiggling his mouth in a back-and-forth wavy motion. The sentence eliciting the auxiliary for the novel verb plurgen, ‘plurging’ is in (62):
Parameterized auxiliary selection
(62)
Ernie:
Ik zag dat Ernie op de boeken ge-, uh... ge-, uh... I saw that Ernie on the books ge-, uh... ge-, uh... Subject: ge-plurgd is/heeft participle aux
As predicted, Experiment 1, on Dutch, showed [±telic] to be highly significant and [±locomotion] to have no effect. But [±actor] also did not appear to play a role. This was predicted for [+telic] verbs (because of the inherent ranking of the Telicity and Actor linking rules) but not for [–telic] verbs. However, it is possible that the Actor Linking Rule did assign have to the [+actor][–telic] verbs, but its effect was masked by an unergative Default Linking Rule, assigning have to [–actor][–telic] verbs, where neither the Telicity Linking Rule nor the Agent Linking Rule would be invoked. Experiment 2 confirmed the presence of a [telic] feature in Dutch. be was highly preferred for all verbs with clear [+telic] endpoints. But these [+telic] cases were assigned be less in the [+actor] cases than in the [–actor] cases, suggesting that an [actor] feature is in fact also used productively by Dutch adults.25 However, as in Experiment 1, the [±actor] difference was not visible in the [–telic] verbs, which overwhelmingly selected have. But this difference, in fact, lends more support to the suggestion that the effect of [±actor] was indeed masked by a Dutch unergative Default Linking Rule. In German, as in Dutch, [telic] and [actor] both appeared as productive features. But significantly, German speakers did not make the same split as Dutch speakers. They added [+locomotion] verbs to the unaccusative category along with all detectably [+telic] cases. This result confirms the suggestion that Keller & Sorace’s similar findings on motion verbs were attributable to a [locomotion] feature in German, and supports the claim of the Parameterized Linking model that German differs from Dutch in using an additional Locomotion Linking Rule. In sum, Randall et al. (2004)’s experiments found evidence for the parameterized set of linking rules in (63): (63)
Dutch German Telicity Linking Rule x x Locomotion Linking Rule – x Actor Linking Rule x x Default Linking Rule (unergative) x –26
. But see van Hout (1998) for another interpretation. . Though there may be an unergative Default Linking Rule in German, we found no evidence for one. In the [–locomotion][–telic][–actor] cases the results were at chance, with equal numbers of have and be.
Janet H. Randall
. Conclusions This article addressed three challenges that arise if auxuliary choice auxiliary selection is used as a diagnostic of syntactic class: a) within languages, is categorical for some verb classes but indeterminate for others; b) cross-linguistically, it can give opposite results for the same sentence; and c) it splits the intransitives in a different place than some other supposed unaccusativity diagnostics. Our approach to these challenges, Parameterized Linking, maintains a syntactic distinction between unergatives and unaccusatives, and accounts for the mismatches semantically and in the linking rules that map semantic representations to syntactic positions. Underlying the syntactic level distinguishing the two verb classes is the semantic level of Conceptual Structure (CS), where semantic features are registered and where verbs and their arguments are hierarchically composed with CS functions. The resulting CS representations are related to the syntax by sets of deterministic linking rules. It is these elements, interacting parametrically, that produce the range of syntactic outcomes within and across languages. Within one language, two sentences containing the same verb can trigger opposite linking patterns depending on the semantic features of the other elements in the sentence. If two features that trigger opposite linking rules are both present in the same sentence, the hierarchy in its CS representation can effectively weight one feature over another to determine the ultimate linking pattern. In CSs of sentences whose linking conflicts are not resolved by geometry or where no linking rules apply, judgements will be indeterminate. Indeterminacy can also come from one verb being conceptualized in two different ways, leading to opposite linking patterns, with two possible auxiliaries. Across languages, opposite linkings for the same sentence will arise when one language has a linking rule that another lacks, as we proposed for Dutch and German, or when two languages have a linking rule that links in opposite ways, a possibility we raised for an Existence Linking Rule and for default linking rules. The misalignment of auxiliary selection with other supposed unaccusativity diagnostics is also explained. Unaccusative/unergative syntax is determined compositionally by interacting features. While auxiliary selection is tied to this syntax, other “diagnostics”, like impersonal passivization, may be tied, instead, to the presence of just one of the interacting features. By hypothesizing that linking rules in a variety of semantic domains can be formulated on any of a spectrum of features, making cut-offs in different places in different languages, Parameterized Linking provides a way to reconcile competing claims in the literature about which features are relevant to the intransitivity split cross-linguistically. In our own experimental studies, perhaps the most striking finding was support for this prediction of cross-linguistic differences. Subjects’
Parameterized auxiliary selection
categorization of made-up verbs confirmed our hypothesis that two different sets of semantic features – and linking rules – operate in Dutch and German. We showed how Parameterized Linking solved problems left by another approach to the auxiliary selection puzzle, Sorace’s Auxiliary Selection Hierarchy (ASH), which aims to characterize the variance (and non-variance) in auxiliary choices with only two features. Parameterized Linking provides the necessary algorithm that can take two features and produce one continuum of gradience, out of interacting, but still deterministic, linking rules. And by incorporating additional features, which may operate in some languages and not others, it makes finer distinctions than the ASH. As such, it actually predicts Keller & Sorace’s unexpected experimental results showing more than two features at work in German. And it predicts, too, the weighting of telicity over agentivity in determining verb class, something that the ASH simply assumed. In discussing the ASH, Sorace (2004) calls for a model of the lexicon-syntax interface that explains why gradience exists in non-core verbs; while the ASH describes the output of such a model, she says, it is not the model itself. And the correct model must overcome the problems facing strict Projectionist and Constructionist interface approaches. The Projectionist view, which deterministically pairs each verb with a unique auxiliary, is incompatible with the indeterminacy in actual auxiliary selection. Constructionist accounts cannot explain the difference between categorical core verb behavior and the variable behavior of intermediate verbs. What is needed, Sorace claims, is a model that can integrate the detailed lexical semantics of verbs with the effects of the constructions they appear in, and explain the constraints on verbs’ lexical and structural meanings so as to prevent inappropriate lexicon-syntax mappings. Parameterized Linking is a theoretical account aimed in this direction. It takes into account sentence-level properties of verbs in construction with their arguments, linking each fixed output deterministically to a syntactic position by the linking rules that a given language parametrically draws from a universal set of linking options.27 Within this framework, further analyses of the semantic features of verbs and their arguments and of the specifics of linking rules will lead to a more detailed picture of the unaccusative/unergative split and, with it, a better understanding of auxiliary selection across languages.
. In fact, the specific individual linking rules proposed here, each one based on a semantic feature, we believe, will ultimately be derived from more general linking principles, determined by the geometry of Conceptual Structure representations. See Randall (forthcoming).
Janet H. Randall
References Arad, M. (1998). VP structure and the syntax-lexicon interface. MIT Occasional Papers in Linguistics, 16. Brinkmann, U. (1992). Choice of auxiliary for intransitive verbs of motion: An analysis of an unaccusative diagnostic. Ms, Max Planck Institute for Psycholinguistics, Nijmegen. Brinkmann, U. (1995). The Locative Alternation: Its structure and acquisition. PhD dissertation, Nijmegen University. Burzio, L. (1986). Italian Syntax: A government and binding approach. Dordrecht: Reidel. Carrier, J. & Randall, J. (1993). Lexical mapping. In E. Reuland & W. Abraham (Eds.), Knowledge and Language Vol. 2: Lexical and conceptual structure (pp. 119–142). Dordrecht: Kluwer. Comrie, B. (1976). Aspect: An introduction to the study of verbal aspect and related problems. Cambridge: CUP. Dowty, D. (1991). Thematic proto-roles and argument selection. Language, 67, 547–619. Everaert, M. (1986). The Syntax of Reflexivization. Dordrecht: Foris. Grewendorf, G. (1989). Ergativity in German. Dordrecht: Foris. Grimshaw, J. (1987). Unaccusatives: An overview. Proceedings of NELS, 17, 244–259. Hoekstra, T. (1984). Transitivity: Grammatical relations in government-binding theory. Dordrecht: Foris. Hoekstra, T. & Mulder, R. (1990). Unergatives as copular verbs; locational and existential predication. The Linguistic Review, 7, 1–79. van Hout, A. (1998). Event Semantics of Verb Frame Alternations: A case study of Dutch and its acquisition. New York, NY: Garland Publications. van Hout, A., Randall, J., & Weissenborn, J. (1993). Acquiring the unergative-unaccusative distinction. In M. Verrips & F. Wijnen (Eds.), The Acquisition of Dutch [Amsterdam Series in Child Language Development 1] (pp. 79–118). Amsterdam: Universiteit van Amsterdam. Jackendoff, R. (1990). Semantic Structures. Cambridge, MA: The MIT Press. Keller, F. & Sorace, A. (2003). Gradient auxiliary selection in German: An experimental investigation. Journal of Linguistics, 39, 57–108. Levin, B. & Rappaport Hovav, M. (1992). The lexical semantics of verbs of motion: The perspective from unaccusativity. In Roca, Iggy (Ed.), Thematic Structure: Its role in grammar. Berlin: Mouton de Gruyter. Levin, B. & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax-semantics interface. Cambridge, MA: The MIT Press. Levin, L. (1986). Operations on Lexical Forms: Unaccusative rules in Germanic languages. PhD dissertation, MIT. Lieber, R. & Baayen, H. (1997). A semantic principle of auxiliary selection in Dutch. Natural Language and Linguistic Theory, 15, 789–845. Perlmutter, D. (1978). Impersonal passives and the unaccusative hypothesis. Proceedings of the Berkeley Linguistics Society 4 (pp. 157–189). Berkeley, CA: University of California. Pollman, T. (1970). Passieve zinnen en het geïmpliceerd logisch subject. Studia Neerlandica, 2, 34–51. Randall, J. (Forthcoming). Linking: The geometry of argument structure. Dordrecht: Springer. Randall, J., van Hout, A., Weissenborn, J., & Baayen, H. (2004). Approaching unaccusativity. In A. Alexiadou, E. Anagnostopoulou, & M. Everaert (Eds.), The Unaccusativity Puzzle: Explorations of the syntax-lexicon interface (pp. 332–353). New York: OUP.
Parameterized auxiliary selection
Randall, J., van Hout, A., Weissenborn, J., & Baayen, H. (In progress). Learning linking: Acquiring unaccusativity in Dutch and German. Ms, Northeastern University. Rappaport Hovav, M. & Levin, B. (1992). The lexical semantics of verbs of motion. In I. Roca (Ed.), Thematic Structure: Its role in grammar. Berlin: Foris. Rosen, C. (1984). The interface between semantic roles and initial grammatical relations. In D. Perlmutter & C. Rosen (Eds.), Studies in Relational Grammar 2 (pp. 38–77). Chicago, IL: University of Chicago Press. Seibert, A. (1993). Intransitive Constructions in German and the Ergative Hypothesis [Working Papers in Linguistics 14]. Trondheim: University of Trondheim. Smith, C. (1997). The Parameter of Aspect (2nd ed.). Dordrecht: Kluwer. Sorace, A. (2000). Gradients in auxiliary selection with intransitive verbs. Language, 76, 859– 890. Sorace, A. (2004). Gradients at the lexicon-syntax interface: Evidence from auxiliary selection and implications for unaccusativity. In A. Alexiadou, E. Anagnostopoulou, & M. Everaert (Eds.), The Unaccusativity Puzzle: Explorations of the syntax-lexicon interface (pp. 243–268). New York, NY: OUP. Tenny, C. (1994). Aspectual Roles and the Syntax-semantics Interface. Dordrecht: Kluwer. Van Valin, R. D. (1990). Semantic parameters of split intransitivity. Language, 66, 221–260. Wunderlich, D. (1985). An investigation of lexical composition: The case of German be- verbs. Linguistics, 25, 283–331. Zaenen, A. (1988). Unaccusatives and the Syntax-semantics Interface [CSLI Report 123]. Stanford, CA: Stanford University. Zaenen, A. (1993). Unaccusativity in Dutch: Integrating syntax and lexical semantics. In J. Pustejovsky (Ed.), Semantics and the Lexicon (pp. 129–161). Dordrecht: Kluwer.
Particle selection in Korean auxiliary formation* Seongha Rhee Hankuk University of Foreign Studies, Stanford University
Emergence of grammatical markers often involves a construction that contains, in addition to the core lexical item, forms that are apparently of lesser importance, such as particles. Grammaticalization research has largely ignored such peripheral forms. This paper, by presenting three major paths of auxiliary development taken by the verb of existence, shows that in the grammaticalization of certain auxiliaries in Korean, particles play a crucial role in determining the paths the construction to take and the resultant grammatical functions. It is therefore necessary to reconsider the dominant trend in grammaticalization research which focuses primarily on the main element, and to pay more attention to the significance of the roles played by these seemingly less important linguistic forms.
.
Introduction
Grammaticalization studies have found numerous crosslinguistic commonalities in language change, a discovery which has engendered many hypotheses concerning the principles and mechanisms operative in grammaticalization processes. Among such commonalities is the fact that similar lexemes give rise to similar grammatical functions. For example, it was pointed out in early grammaticalization studies that demonstratives are a common source for the grammaticalization of gender markers (Greenberg 1978). A similar idea reverberates in Givón’s (1973) assertion that the core meaning of a grammatical marker is included in the lexical meaning of the source item. Likewise, Bybee, Perkins and Pagliuca (1994) use a large scale quantitative study to show that words denoting ‘finish’, ‘throw * This research was supported by the Research Fund of Hankuk University of Foreign Studies. I am thankful for the valuable comments by the three anonymous reviewers. All remaining errors are mine.
Seongha Rhee
away’, ‘pass by’, etc. are common sources of Past markers; and those denoting ‘want’, ‘desire’, etc. are likewise common sources of Future markers. Their impressive crosslinguistic data point to the impressive correlation between meanings of source lexemes and grammatical meanings and lead them to propose the source determination hypothesis (Bybee, Perkins, & Pagliuca 1994). In apparent contradiction to these findings, there have been other claims that make reference to the fact that a single source lexeme can give rise to multiple grammatical markers. This is well illustrated in Craig’s (1991) analysis of bang ‘go’ in Rama that developed into multiple domains of grammatical categories through separate paths. She presents the notion of polygrammaticalization to refer to such processes where a single source lexeme develops into diverse grammatical markers in different areas. Likewise, Heine, Claudi and Hünnemeyer (1991) present an excellent case that illustrates the development of multiple grammatical markers from a single source lexeme, i.e. multiple functions developing from the Allative case marker in multiple, partially intertwined chains. This paper begins with a simple question of why Korean developed so many grammatical markers from a single lexical source, i.e. issta ‘to exist’1 as shown in (1). (1)
a. b. c. d. e.
Resultative Past/Perfect Pluperfect Progressive/Imperfective Futurity
-eiss-ess-essess-koiss-keyss-
In (1), three different kinds of non-finite connective markers, i.e. -e, -ko, and -key, are involved. Thus, Resultative, Past/Perfect and Pluperfect evolved from a structure containing the particle -e; Progressive/Imperfective, the particle -ko; and Futurity, the particle -key. It is argued here that the selection from the multiple paths of the auxiliary formation is crucially dependent on the selection of the particles involved in the source construction. Even though such diversity of the grammaticalized functions from a single source lexeme may be extraordinary, it is by no means remarkable from at least one perspective: developments of all those listed functions from existence verbs have been attested in numerous other languages, as lexicons such as Heine et al. (1993) and Heine and Kuteva (2002) show. Its extraordinary nature may lie, rather, in the extensiveness of the development: few languages seem to exhibit such a high number of grammatical forms developed from a single source as Korean. . The ending -ta in issta is merely the infinitive ending and is not an inherent part of the verb. Therefore, the verb may appear as iss- in its stem form.
Particle selection in Korean auxiliary formation
The development examined here has to do with both claims presented at the beginning of this article: source determination and multiple paths. This paper aims to reconcile these two apparently opposing observations by showing that the existence verb in Korean, which grammaticalized into diverse yet crosslinguistically attested functions, did in fact develop through multiple paths; and that these paths are determined by the functions of the participating particles.2
. Preliminary issues . Grammaticalizing forms In most grammaticalization studies that trace grammaticalization paths, attention has been paid almost exclusively to the lexical item in the source construction. For example, one of the most commonly cited English examples is the development of Future be going to or be gonna from the full-fledged lexical verb go. Such studies typically focus on how the verb of motion became a Future marker, thus invoking metaphor as a major mechanism (Lakoff & Johnson 1980). However, strictly speaking, what grammaticalizes is not only the verb go, but the entire construction be going to. Therefore, exclusive attention to go in be going to minimizes the roles played by be, -ing, and to. Hopper and Traugott (2003 [1993]: 88–90) argue that the Purposive to played a significant role here by inviting the conversational inference of later time, which consequently became semanticized as the construction fully developed into a futurity marker. Similarly, Bybee, Perkins and Pagliuca (1994: 11–12) argue that it is the entire construction, and not simply the lexical meaning of the stem, that grammaticalizes. This is of particular importance for an understanding of grammaticalization of auxiliaries in Korean, where constructions with the same stem yield different grammatical markers. Therefore, due attention should be paid to the participating particles in the investigation of their roles in the emergence of grammatical meanings. . Serial verb constructions Serial verb constructions are a special feature of the Korean language. A serial verb construction is “the combination of two or more asyndetically juxtaposed verbs with one shared argument in order to express a complex, but unitary action” (Lehmann 1995 [1982]: 34). Unlike some languages where two or more verbs are . In the absence of detailed information on the source constructions we cannot determine if, and to what extent, other languages listed in the lexicons have analogous situations as the Korean existence verb issta.
Seongha Rhee
simply juxtaposed in bare forms without conjunctives, i.e. truly ‘asyndetically’, Korean uses non-finite connective particles to combine them in the configuration of (2a), as exemplified in (2b), where the connective particle is glossed as a non-finite (NF) marker. (2) a. Verb1 + Connective Particle + Verb2+Tense/Aspect/Modality b. ccic-e-peli-ess-ta tear-nf-discard-pst-dec ‘(I) tore (it) and threw (it) away.’
Serial verb constructions are so susceptible to grammaticalization that they are even regarded as the seed of grammaticalization (DeLancey 1991: 15). Indeed, all serial verb languages seem to develop diverse grammatical function verbs out of the serial verb constructions by way of reanalysis (Bynon 1985), or through pragmatic extension, i.e. through interaction between conceptualization of the world events and iconicity in the language (DeLancey 1991; Bruce 1988). Typically it is the finite verb, or more correctly, the construction including the finite verb, i.e. Verb2 in (2a) above, that develops into a certain template and becomes a grammatical marker. This characteristic is shared by many other languages in different language families. Since grammaticalization is a dynamic and inherently diachronic process, a synchronic analysis of grammaticalizing serial verb constructions shows differing degrees of grammaticalization. Since these verbs of hybrid nature display diverse characteristics according to their degree of grammaticalization, there are various terms used to describe them, e.g., serial verbs, auxiliary verbs, verbids (Ansre 1966), catenatives (Brinton 1988), explicators (Nayar 1983), light verbs, etc. In this paper the term ‘serial verb’ is broadly defined to refer to the finite verbs that form a construction with one or more verbs linked by non-finite markers. One notable aspect of verb grammaticalization from serial verb constructions is their tendency to develop into aspect markers. Such a phenomenon has been frequently observed. Korean, being a serial verb language, displays no exception to this tendency. Many serial verbs in fact became aspect markers and some even became erstwhile tense markers or modality markers, which is consistent with the general directionality of the development of these grammatical markers (Heine 1993). This paper reports such grammaticalization phenomena in which the existence verb issta ‘to exist’ is grammaticalized in serial verb constructions. . Particles There are diverse devices for clausal connection in Korean. In particular, three connective particles, -e, -ko, and -key have been the focus of substantial controversy with respect to their syntactico-semantic functions. They (or their homophonous morphemes) have also been used as sentence-final markers signaling the sentence
Particle selection in Korean auxiliary formation
type. They (or their homophonous morphemes) were also used as adverbializers transforming verbs and adjectives into adverbials, thus resembling converb markers (Haspelmath 1995a, 1995b). These particles, therefore, are sometimes classified as adverbializers (Choe 1989 [1929]). On the other hand, Kim (1981: 37) classifies them as connective particles. Evidently the status of these particles is still controversial. In this paper, without evaluating the validity of such nomenclature, those particles are called simply connective particles or non-finite markers (NF) because they connect two (or more) verbs and make the verb to which they are attached appear normally in non-finite forms. The particles often participate in the grammaticalization of verbs in Korean and I will look at these particles more closely and provide a semantic characterization of each of them. .. Semantics of -e and -ko Of numerous particles in Korean, -e and -ko are among the most frequently used particles, and for clear contrast I shall discuss them together. Both particles share the connective function roughly resembling that of English and. When these particles are used as connecting particles they tend to combine non-finite verbs.3 Such a common connecting function is illustrated in the following examples where these non-finite particles are glossed as NF.4 (3) a.
pom-i ka-ko yelum-i o-ass-ta spring-nom go-nf summer-nom come-pst-dec ‘The spring has gone and the summer {has come, came}.’ b. ppalli o-e pap mek-ela quickly come-nf meal eat-imp ‘Come quick and eat the meal.’
Despite such superficial similarities, however, the two connective particles have distinct semantics. These semantic distinctions are so notoriously subtle that they have defied attempts to sufficiently distinguish the two particles, a fact reminiscent of König’s (1995) characterization of the semantics of converbs. Since the gist of the claim here is that the semantics of the particles are of crucial importance, I will venture to advance an explanation of the differences between the two. In order to do so, I will look into their respective uses and then contrast them to identify the differences. . In fact, they show different behavior with respect to this: -e does not allow connection of finite clauses, whereas -ko does. This has to do with their consolidating vs. isolating functions, to be discussed later. . The connective particle -e has allomorphs -a, and -ø. However, for ease of exposition, -e is used throughout.
Seongha Rhee
Koo (1987) argues that -e unifies two domains of events denoted by the nonfinite verb and the finite verb, and -ko separates them. Even though this observation is based on contemporary synchronic data, this analysis seems to be based on their early syntactic functions in Middle Korean. In the Middle Korean Penyek Nokeltay text dating from circa 1517, the following kinds of verbal connections are attested. (4)
a. hyangha-e ka face(v)-nf go ‘go toward’
b. tol-e o turn-nf come ‘return’
c. kali-e mek choose-nf eat ‘eat only certain things’
As shown in the above examples, the clauses combined by the non-finite marker -e tend to form a unitary concept of an event. Therefore, in (4a) for example, the ‘facing’ event and the ‘going’ event form a single event of ‘going toward’. Likewise, in (4b), the ‘turning’ event and the ‘coming’ event form a single event of ‘returning’; and in (c), the ‘choosing’ event and ‘eating’ event form a single event of ’eating only certain things’, which became an expression for describing someone who has particular eating habits. Another important aspect of -e connective is that the combined events are sequence-sensitive. It is common for language users to present events in their temporal or logical sequence, even though such an assumed relationship may be countermanded by explicit statements. This is not the case with the expressions combined by the connective particle -e. For example, if the order of the verbs is reversed the resulting expressions become uninterpretable. The preceding discussions lead us to the conclusion that -e is a particle that connects two sequential events and brings forth an interpretation that the events form a single consolidated event. On the other hand, in the same historical source of Middle Korean, the following kinds of verbal connections by the non-finite marker -ko are also attested (presented in modern orthography, all taken from the Penyek Nokeltay text). (5) a.
ca-ko ka sleep-nf go ‘sleep and go’ b. ssu-ko cek spend-nf write ‘spend (money) and record it’ c. posalphi-ko ca secure-nf sleep ‘secure (the doors) and sleep’
As seen in the above examples, the two events denoted by the verbs combined by -ko tend to be possibly related but separate or independent of each other. In (5a),
Particle selection in Korean auxiliary formation
for example, the ‘eating’ event and the ‘going’ event may be adjacent in time but are conceptually too individuated or independent to form a single event. The combined events of ‘spending money’ and ‘recording the expenditure’, and of ‘securing the doors’ and ‘sleeping’ are similarly independent. The above examples seem to suggest that the two events must be in sequential relationship. However, other examples show that that is not always the case, as shown in the following examples: (6) a.
ku kang-un nelp-ko kiph-ta the river-top wide-nf deep-dec ‘The river is wide and deep.’ b. ku kang-un kiph-ko nelp-ta the river-top deep-nf wide-dec ‘The river is deep and wide.’
The above examples show that the two events or states are not in the relationship of temporal sequence. They are simply in syntactic juxtaposition, and the connective distributes the focus evenly onto each of the combined events or states. This is well illustrated in the following peculiar example. (7) mwun tat-ko tul-e o-a door close-nf enter-nf come-imp ‘Come in and close the door.’ (Lit. ‘Close the door and come in.’)
The above example should be pragmatically awkward, because in its literal interpretation preceding ‘door-closing’ should preclude the following ‘entering’ event. However, this is a commonly used expression requesting someone at the door to come in and close the door. The speaker is simply demanding two events to come about without reference to their relative sequence. Then it can be assumed that the sequential relationship invoked in the interpretation of (5) is pragmatically motivated because for information processing, enumerating events in the order of temporal sequence, if applicable, is preferred (cf. Gricean maxim of manner). Therefore, -ko can be said to be a connective particle that enumerates events in co-ordination without sequential relationship.5 The differences between the two connective particles can be better illustrated by substituting the particles for each other. If -e in the previous example (4),
. One of the reviewers suggested that -ko can be used both sequentially and nonsequentially (Cho 2004). However, for the reasons presented here and the fact that some expressions with -ko where sequentiality is intended are less than acceptable (which can be fixed by using a sequential marker -se, instead), the sequential use of -ko seems to be pragmatically motivated following the observance of the Gricean maxim of “Be orderly.”
Seongha Rhee
repeated here as (8) for comparison, is substituted for -ko, the results are as shown in (9). (8) a.
hyangha-e ka face(v)-nf go ‘go toward’ b. tol-e o turn-nf come ‘return’ c. kali-e mek choose-nf eat ‘eat only certain things’
(9) a.
hyangha-ko ka face(v)-nf go ‘orient oneself and go’ b. tol-ko o turn-nf come ‘turn and come’ c. kali-ko mek choose-nf eat ‘put aside and eat’
There are certain differences between these corresponding expressions. One thing in common is that when -ko is used instead of -e, the interpretation becomes one denoting two separate events rather than one. For example, in (8a), with the particle -e, the ‘facing’ event and ‘going’ event are so much fused that the two components are no longer perceived as separate. Interestingly, (9a) is largely similar to (8a) and in fact, it is also occasionally used. A subtle difference is that while (8a) suggests that facing toward a certain destination is successful and well-maintained throughout, (9a) seems to suggest that the ‘facing’ event is incidental, and its successfulness can be brought into question. Therefore, a directional mistake en route is more likely to be associated with (9a) than with (8a). This suggests that the two events denoted by (9a) are more compartmentalized. A similar situation appears in a comparison of (8b) with (9b). Example (8b) refers to the situation when the traveler’s route is maintaining the reverse direction of the previously taken journey. On the other hand, (9b) suggests that the traveler is simply coming without reference to his or her previous travel such as after going around something (e.g. a track, a park, etc.). In other words, the traveler in (8b) is required to have been at the destination point before, whereas this is not required in (9b). A comparison of (8c) with (9c) also shows a similar point: (8c) suggests that choosing and eating are interlaced in a holistic eating event, and (9c) suggests that
Particle selection in Korean auxiliary formation
choosing something is completed before the person begins to eat. Therefore, (8c) is more likely to describe a person’s eating habits, while (9c) is more likely to describe an incidental event, as in a situation when certain alien material is found in the food. Now, the two particles from the previously given examples in (5) can be substituted to see the result. If the connective particle -ko in (5) is substituted for -e, all resultant expressions become uninterpretable. In other words, combining events with sequential consolidating connective -e is not possible with ‘sleep’ and ’go’; ‘spend’ and ‘write’; and ‘secure’ and ‘sleep’ because the events combined cannot conceptually form a single event, making the use of -ko the only viable option. All these findings in the preceding discussion point to the following generalization: (1) -e consolidates the sequentially combined events; and (2) -ko isolates the combined non-sequential events distributing equal focus on each event. This characterization should be fruitful in discussing their roles in auxiliary formations. .. Semantics of -key The connective particle -key was rarely used prior to the twentieth century, and its historical use in Middle Korean was largely restricted to Ingressive and Causative construction (Huh 1992 [1975]; Ko 1987). Comparing historical data between the Middle Korean version and the Modern Korean version of an identical text, Rhee (1996) shows that the function of -key as an adverbializer grew remarkably during the period. This adverbializer function is still the most frequent one, marking ‘manner’, ‘mode’ or ‘purpose’, as shown in the following examples: (10) a.
ku candy-nun masiss-key poi-ess-ta the candy-top tasty-nf seem-pst-dec ‘The candy looked tasty.’ b. ku-ka na-lul aphu-key ttayli-ess-ta he-nom I-acc ache-nf hit-pst-dec ‘He hit me painfully.’
In example (10a) above, the literal meaning is ‘the candy appears in a manner of being tasty.’ Likewise, the literal meaning of example (10b) is ‘he hit me in such a manner that I would feel pain.’ In these examples, the connective particle -key designates the manner of appearance and the manner of hitting, respectively. When the manner is associated with an action it acquires a meaning of ‘on purpose’. Whether the purpose is realized, e.g. whether I felt pain indeed, is indeterminable by the particle, but its realization is pragmatically inferred. Based on this fact, I will characterize the particle -key as a mode/purpose connective.
Seongha Rhee
. Auxiliaries from the existence verb issta In the grammaticalization history of Korean it is almost a routine process for verbs to go through a stage of serial verb construction. From this verb serialization with issta ‘to exist’, such grammatical markers as Resultative, Past/Perfect, Progressive, and Future evolved. When grammaticalization occurs, the clausal status of the predicate containing the existence verb becomes lost and the typical biclausals are thereby turned into monoclausals. The existence verb acquired versatile functions along the grammaticalization paths. These paths can be grouped into three major categories: the Past path, the Progressive path, and the Future path. Grouping these three paths coincides with the use of the three connective particles I discussed above, i.e. -e, -ko, and -key, to which I now turn. . The past path Through a verb serialization with the particle -e, auxiliaries of Resultative and Perfect/Past were developed. This is consistent with the analysis of Bybee, Perkins and Pagliuca (1994) where such grammatical markers as Persistence of State, Resultative, Anterior, Perfective, and Simple Past form a developmental path in terms of degree of grammaticalization. Further, Korean recruits the Past marker to invent the Pluperfect marker. Some examples of constructions in the Past path are shown in (11). (11) a.
Resultative (-eiss-) ceki say-ka cwuk-eiss-ta there bird-nom die-resul-dec ‘There is a dead bird over there.’ b. Perfect/Past (-ess-) ku-ka cwuk-ess-ta he-nom die-pst-dec ‘He died.’ c. Pluperfect (-essess-) ku-ka yeki o-essess-ta he-nom here come-plup-dec ‘He had been here before.’
. Diachronic development The serial verb construction -e issta had a wide distribution in early Korean data. In the studies on Middle Korean (notably Huh 1992 [1975]; Ko 1987, inter alia, it has been observed that by the fifteenth century this -e issta serial verb construction
Particle selection in Korean auxiliary formation
had already developed to the point that it carried the aspectual meaning of Resultative with a wide range of verbal or adjectival predicates.6 Rhee (1996) shows that the cooccurrence pattern of adverbials in the serial verb construction in Middle Korean displays both the disjoint and conjoint nature of the construction, suggesting that the verbs participating in the serial verb constructions had different levels of consolidation. In other words, in disjoint configurations the two verbs are in simple juxtaposition, whereas in the conjoint configurations the two are more unitized. From the seventeenth century, the digraphemic Resultative aspect marker -e iss- developed into the monographemic Perfect/Past marker -ess-, initially starting as a variant of the former. Prior to the seventeenth century, Korean did not have a morphological Past tense marker.7 The grammatical notion of Past was often indicated by temporal adverbs such as ‘yesterday’, ‘already’, etc. Modern Korean does not morphologically distinguish between the Present Perfect and the Past, and -essis used to express both. Crosslinguistically these two grammatical markers tend to originate from the same sources and undergo the same grammaticalization path, where the Past marker is usually the more developed form. Modern Korean has -e iss- for Resultative/State-Persistence marking, and -ess- for Perfect/Past marking, both coexisting as the offspring of the same serial verb construction. The grammaticalization journey continued when the Past/Perfect morpheme was recruited to make a grammatical morpheme to mark Past Perfect or Pluperfect (Dahl 1985: 144) by way of reduplication, i.e. -essess- from -ess-. I cannot find any historical data previous to Modern Korean that show the use of Pluperfect. Its emergence seems to be attributable to the exposure to foreign languages with Pluperfect, which prompted the filling-in of the functional gap by translators in an effort to achieve closer one-to-one correspondence between the source language and Korean. The strategy involved in devising a new grammatical marker from the Past/Perfect seems to be metaphorization: making use of iconicity by increasing the temporal distance, which is done by repeating the use of the Past marker. .. Source characteristics Let us now consider how the change from the serial verb construction to a grammatical marker of Resultative occurred. The change must have involved many . Since adjectives were also allowed in the non-finite position, the term ‘serial verb construction’ may be a misnomer. However, this anomaly is due to the idiosyncrasy of Korean where adjectives behave in a similar way as verbs, and distinction between the two categories is often obscure. . Crosslinguistically, this Middle Korean situation is exceptional because zero-marked Past is a very rare phenomenon across languages (Bybee, Perkins, & Pagliuca 1994: 91).
Seongha Rhee
facets of grammar and language use in general, but I will mainly focus on the semantic characteristics attributable to the particle -e. The sequentiality sense of the particle -e is critical here. Generally speaking, connective particles used in serial verb constructions imply temporal sequence when the verbs denote actions or events. This holds crosslinguistically because of the iconicity in language use. Among Korean connective particles, -e is particularly sensitive to the iconic sequentiality of the combined predicates, as I have discussed in 2.3.1. Therefore, the construction -e issta implies a sequence of two events, i.e. one denoted by the non-finite verb and the other the event of existence. The basic sense of Resultative includes the sequentiality between an event and its resulting state. The ‘state’ sense is directly derivable from the existence verb issta, because its meaning is closer to ‘to be at’ rather than ‘to live’ or ‘to be alive’. This was the case with Middle Korean, too, in which the verb predominantly referred to spatial location. Generally speaking, when the predicate including this verb does not specify spatial location, the meaning of location is also generalized to simply mean a state of being, and the spatiality is generalized to temporality, which is an instance of conceptual expansion (Heine, Claudi, & Hünnemeyer 1991) that seems to be responsible for the emergence of tense- and aspect-marking functions. As noted previously, the precursor of the Past marker was a shortened variant (-eys-) of the Resultative marker (-e iss-). This functional change accompanied an interesting formal change whereby the connective particle (-e) became obscure. With the particle now formally obscure, the form brings forth noteworthy consequences. When a phonological contraction occurs, the effect in orthography is direct. Korean orthography is a combination of syllabic and alphabetical writing systems, where each character is a syllable and each character is composed of two or more phonemic letters. When this contraction occurred, neither the connective particle -e nor the existence verb became identifiable because the previous two characters were reduced into one. This change must be more than an orthographic change: it must also entail loss of inter-clausal pause and vowel quality change. It must be noted that the meaning of Resultative is nothing much more than the product of combining the meaning of the connective particle and the meaning of the verb of existence. When both of these forms, which used to provide the grammatical meaning, became opaque, the meaning of the Resultative was jeopardized. The meaning of ‘persistence’ (i.e. ‘be in the state of ’) and ‘sequence’ (i.e. ‘after [verb1 event]’) was affected. After the ‘be in the state after’ meaning is deducted from ‘be in the state after the completion of verb1 event’, what is left is ‘completion of verb1 event’, the essential semantics of Past/Perfect. In sum, the change from Resultative to Past/Perfect has to do, at least in part, with phonological reduction that led to a corresponding reduction of the combinatorial meaning into a single indivisible concept of ‘anteriority’. This new grammatical concept of general ‘anteriority’ is central to the Past. It seems that
Particle selection in Korean auxiliary formation
this change was either enabled or facilitated in part by the loss of the connective particle -e which formerly provided the sequentiality sense. With reference to the emergence of Pluperfect, it is again notable that such recruitment of the extant form -ess- to create a new grammatical marker -essesswas not barred by the sequentiality sense of the particle -e, which was originally in the source construction but became invisible through phonological reduction at the time of Past/Perfect formation. . The progressive path The second path taken by the existence verb issta is the Progressive path. Grammatical markers developed along this path are Progressive and Imperfective. These two markers are homophonous in Korean and, therefore, the interpretation of some sentences is ambiguous. The following is an example of such a case, where (12a) is the Progressive interpretation, while (12b) is the Imperfective interpretation. (12) ku-nun yak-ul mek-koiss-ta he-top medicine-acc eat-prog/imperf-dec a. ‘He is taking medicine (at this moment).’ b. ‘He is under medication (these days).’
.. Diachronic development The aspectual -koiss- is a very frequently-used grammatical marker in Modern Korean. However, this marker is rarely found in historical data. The development of -koiss- seems to have begun in earnest in the late 19th century and the early 20th century (Huh 1987; Rhee 1996). A data survey suggests that when it first emerged as the Progressive marker, it seems to have been used with eventive accomplishment predicates (e.g. ‘build a house’ and ‘run’), then spread to habitual and semelfactive predicates (e.g. ‘smoke’ and ‘knock’), further to eventive achievement predicates (e.g., ‘arrive’ and ‘melt’), and then finally resultative predicates (e.g. ‘wear’ and ‘know’). From this final stage, the Progressive marker assumes the role of Imperfective marker and, perhaps in a remedial effort, a new, more periphrastic Progressive marker -koissnuncwungi- ‘be in the middle of ’ is introduced to mark true Progressives. The development from the Progressive -koiss- to the Imperfective is a very recent development. The Progressive and Imperfective aspects are not mutually exclusive, but the latter has a wider scope of denotation and subsumes the former, and the developmental direction from Progressive to Imperfective is frequently at-
Seongha Rhee
tested across languages.8 The cognitive force behind this development seems to be a transfer of the notion of unboundedness from an event to a general state of affairs, including statives. Progressives often lend themselves to a Habitual interpretation. In the Habitual aspect, the sense of duration required to constitute the Progressive/Imperfective aspect is naturally obtained from the real-life temporal duration which is occupied by the habitual event concerned. .. Source characteristics The development of the Progressive makes use of the particle -ko together with the existence verb issta. Since -ko is an isolating connective, the use of this particle to express something like ‘someone is in the middle of doing something’, in which the agent’s existence and his or her on-going activity seem to be blended, is somewhat puzzling. For this reason, the use of consolidating connective -e would seem to be more appropriate. However, a closer look reveals that the choice is a fully justifiable one because the particle -e has an inherent semantics of sequentiality, and its use would be problematic in that the on-going activity is not in a temporally sequential relationship with the event/state of existence. Strictly speaking, the two events are separate events occupying a certain segment of a time-line at the same time. The particle -ko invokes distributed attention to the coordinated events, i.e. the on-going event and the event/state of existence, and this effect seems to be crucial in the formation of the Progressive sense. Therefore, in the development of this Progressive marker, the particle -ko seems to have been employed for three obvious reasons: its lack of sequentiality, its isolating effect, and its focus-distributing property. . The future path The last grammaticalization path taken by the existence verb is the Future path. The grammatical forms along this path are various subcategories of Futurity, such as Conjectural, Intentional, Hypothetical Willingness, Current State as Futuristic Mode, Ability, Possibility, Evaluative, and Predestination (Rhee 1996).9 The Korean Futurity marker -keyss- was developed from the existence verb issta in the serial verb construction in which the particle -key is used. The use of -keyss- is illustrated in the following examples. . The possibility of using English Progressive for Habitual led Bybee and Dahl (1989) to speculate that English Progressive represents an intermediate stage in the development of Progressive into Imperfective, a widely attested change across languages. . Some of the uses of -keyss-, such as Conjectural and Evaluative, do not make temporal reference. These are later developments from the Future marker, but their new functions became considerably remote from the original temporal sense.
Particle selection in Korean auxiliary formation
(13) a.
na-nun nayil yehayng-ul ka-keyss-ta I-top tomorrow travel-acc go-fut-dec ‘I will go on a trip tomorrow.’ b. nayil pi-ka o-keyss-ta tomorrow rain-nom come-fut-dec ‘It will rain tomorrow.’
.. Diachronic development There is a controversy as to the historical origin of the Future marker -keyss- depending on its source composition. However, historical data show, at least, that Modern Korean -keyss- appears from the late Early Modern Korean period, i.e. at the turn of the 19th century. The studies by Na (1953) and Huh (1982a, 1982b), based on four different versions of a book entitled Hantyunglok, dating from 1795, 1801, 1802, and 1805, show that certain earlier forms of -keyss- were used interchangeably in corresponding texts, and that the interchangeability ceased by the end of the 19th century, when the use of -keyss- became established. During the past one hundred years or so, the Future marker acquired various senses through diverse mechanisms. These developments are not only varied in kind but also interlaced in causal chains, and therefore, how those senses came into being cannot be presented with certainty. However, I can analogize, though over-simplified, the emergence patterns of each sense. The core semantics of the marker was ‘current state as a futuristic mode’ as a result of the combination of mode/purpose connective particle -key and the existence verb issta. This core sense branched out into two: ‘intention with respect to the future event’ and ‘evaluation of current situation with respect to the future event’. From these interpretations emerged various senses such as Intentional, Hypothetical Willingness, Conjectural, etc. .. Source characteristics There is a controversy as to the historical origin of Future marker -keyss-. Ramstedt (1949) and others consider it a development from the particle -key and the existence verb issta. Na (1953), Huh (1982a, 1982b) and others, however, consider the origin to be the particle -key combined with the light verb hata ‘to do’ and the existence verb issta. Both theories are largely compatible and acknowledge in common the participation of the particle -key and the existence verb issta. It has been shown in the above that the main function of the particle -key is to designate manner/purpose. The manner/purpose of a situation and the purpose of an action have direct implications for the future because purpose is an inherently future-oriented notion and manner/mode is the bridge between one situation and its future situation or between one action and its future resultant situation. There-
Seongha Rhee
fore, a construction composed of Verb1, connective -key, and the existence verb signifies that the sentential subject exists in such a manner that Verb1 would be realized. When this construction is construed as a serial verb construction, the focus would be on the finite verb, i.e. ‘to exist’ because it is the final verb in serialization. When the conversational implicature invites attention to the non-finite verb to which the manner/purpose marker -key is attached, by virtue of the futureindexing ability of the particle -key, this verb would be reanalyzed as the main verb, a process analogous to the development of Future be going to from one where go was the main verb to the one where the verb following the Purposive to became the main verb. This type of grammaticalization process in Korean should have been possible largely due to the semantics of the particle -key that contained an implication of futurity.
. Conclusion Grammaticalization is a multi-faceted phenomenon, where a linguistic form is subjected to diverse forces at various levels from as low and local as its sound, to as high and holistic as the discourse it is situated in. However, since all grammaticalization phenomena are reducible to a change involving linguistic forms, the question of what participates in the process should be of primary importance. In most studies, however, the research focus tends to be on the main lexical item, often regrettably ignoring other seemingly peripheral forms such as particles. It has been shown that in the formation of some auxiliaries such apparently insignificant particles play a decisive role in determining the final products of the process, by determining the path a construction will take. Also shown are the examples utilizing the verb of existence, issta ‘to exist’, which underwent three separate paths of polygrammaticalization: the Past path, the Progressive path, and the Future path. Selection of these paths was determined by the sequential, consolidating connective -e, the isolating, focus-distributing connective -ko, and the manner/purpose connective -key. Since Korean has a rich particle system and many particles, themselves the result of grammaticalization, participate in the grammaticalization of other markers, identifying the limits and the precise natures of the roles of the particles should contribute to a better understanding of grammaticalization processes in language.
Particle selection in Korean auxiliary formation
Abbreviations acc: accusative; dec: declarative; fut: future; imp: imperative; imperf: imperfective; nf: non-finite; nom: nominative; plup: pluperfect; prog: progressive; pst: past; resul: resultative; and top: topic.
References Ansre, G. (1966). The verbid – a caveat to serial verbs. The Journal of West African Languages, 3 (1), 29–32. Brinton, L. J. (1988). The Development of English Aspectual Systems. Aspectualizers and postverbal particles. Cambridge: CUP. Bruce, L. (1988). Serialization: From syntax to lexicon. Studies in Language, 12 (1), 19–49. Bybee, J. L. & Dahl, Ö. (1989). The creation of tense and aspect systems in the languages of the world. Studies in Language, 13, 51–103. Bybee, J. L., Perkins, R., & Pagliuca, W. (1994). The Evolution of Grammar: Tense, aspect, and modality in the languages of the world. Chicago, IL: The University of Chicago Press. Bynon, T. (1985). Serial verbs and grammaticalization. In U. Pieper & G. Stickel (Eds.), Studia Linguistica Diachronica et Synchronica (pp. 105–121). Berlin: Mouton de Gruyter. Cho, S.-Y. (2004). Sequentiality and non-tensed verbal coordination in Korean. Paper presented at the 30th Berkeley Linguistics Society Conference, February 13–16, 2004. Berkeley, CA. Choe, H.-B. (1989 [1929]). Wuli malpon (Korean Grammar) (15th ed.). Seoul: Jung-um Publishing. Craig, C. (1991). Ways to go in Rama: A case study in polygrammaticalization. In E. Closs Traugott & B. Heine (Eds.), Approaches to Grammaticalization, Vol. 2 (pp. 455–492). Amsterdam: John Benjamins. Dahl, Ö. (1985). Tense and Aspect Systems. Oxford: Blackwell. DeLancey, S. (1991). The origins of verb serialization in modern Tibetan. Studies in Language, 15 (1), 1–23. Givón, T. (1973). The time-axis phenomenon. Language, 49 (4), 890–925. Greenberg, J. H. (1978). How does a language acquire gender markers? In J. H. Greenberg, C. A. Ferguson, & E. Moravcsik (Eds.), Universals of Human Language, Vol. 3 (pp. 47–82). Stanford, CA: Stanford University Press. Haspelmath, M. (1995a). The converb as a cross-linguistically valid category. In M. Haspelmath & E. König (Eds.), Converbs in Cross-linguistic Perspective: Structure and meaning of adverbial verb forms – Adverbial participles, gerunds (pp. 1–55). Berlin: Mouton de Gruyter. Haspelmath, M. (1995b). Contextual and specialized converbs in Lezgian (NakhoDaghestanian). In M. Haspelmath & E. König (Eds.), Converbs in Cross-linguistic Perspective: Structure and meaning of adverbial verb forms – Adverbial participles, gerunds (pp. 415–440). Berlin: Mouton de Gruyter. Heine, B. (1993). Auxiliaries: Cognitive forces and grammaticalization. Oxford: OUP. Heine, B., Claudi, U., & Hünnemeyer, F. (1991). Grammaticalization: A conceptual framework. Chicago, IL: University of Chicago Press.
Seongha Rhee
Heine, B., Güldemann, T., Kilian-Hatz, C., Lessau, D. A., Roberg, H., Schladt, M., & Stolz, T. (1993). Conceptual Shift: A lexicon of grammaticalization processes in African languages. Köln: Universität zu Köln. Heine, B. & Kuteva, T. (2002). World Lexicon of Grammaticalization. Cambridge: CUP. Hopper, P. & Traugott, E. Closs (2003 [1993]). Grammaticalization (2nd ed.). Cambridge: CUP. Huh, W. (1982a). 19-seyki kwuke ttaymaykimpep yenkwu (A study of the tense system in the 19th century Korean). Hankul, 177, 3–38. Huh, W. (1982b). Hankwukmal ttaymaykimpepuy keleon palcachwi (The development of the tense system in Korean). Hankul, 178, 3–52. Huh, W. (1987). Kwuke ttaymaykimpepuy pyenchensa (A Developmental History of the Korean Tense System]. Seoul: Saem Publishing. Huh, W. (1992 [1975]). Wuli yeysmalpon (Old Korean grammar). Seoul: Saem Publishing Co. Kim, S.-G. (1981). Hankwuke yenkyelhyeng emiuy uymi pwunsek yenkwu I (Semantic analysis of some Korean connective endings I). Hankul, 173/174, 35–64. Ko, Y.-K. (1987). Cwungsey kwukeuy sisangkwa sepep (Tense-aspect and mood in Middle Korean) (2nd ed.). Seoul: Tower Press. König, E. (1995). The meaning of converb constructions. In M. Haspelmath & E. König (Eds.), Converbs in Cross-linguistic Perspective: Structure and meaning of adverbial verb forms – Adverbial participles, gerunds (pp. 57–95). Berlin: Mouton de Gruyter. Koo, H. J. (1987). Ssikkuth -a, -key, -ci, -ko-uy ssuimkwa uymi (The usage and semantics of suffixes, -a, -key, -ci, and -ko). Konkuk Language and Literature, 11–12, 167–188. Lakoff, G. & Johnson, M. (1980). Metaphors we Live by. Chicago, IL: The University of Chicago Press. Lehmann, C. (1995 [1982]). Thoughts on Grammaticalization [LINCOM Studies in Theoretical Linguistics 1]. München: Lincom. Na, C.-S. (1953). Milaysisangpocoekan li-wa keyss-uy kyochey (The changes of future tense/aspect auxiliary stems li and keys). Korean Language and Literature, 6, 296–306. Nayar, D. (1983). Verb Sequences in Malayalam. PhD dissertation, Jawaharlal Nehru University. Ramstedt, G. J. (1949). Studies in Korean Etymology. Helsinki: Suomalis-Ugrilainen Seura. Rhee, S. (1996). Semantics of Verbs and Grammaticalization: The development in Korean from a cross-linguistic perspective. PhD dissertation. The University of Texas at Austin. Seoul: Hankuk Publisher.
Language use and auxiliary selection in the perfect* K. Aaron Smith Illinois State University
Within a usage-based model of language storage and processing, I contrast selection of have and be auxiliaries in the German and English Perfect. In my analysis, I view the apparent irregularities in selectional criteria in each language to be the product of unique diachronic forces. In the case of English, I give evidence that the stronger mental representation of the have type promoted its spread throughout the Perfect system, replacing all instances of older auxiliary be (although vestigial cases involving very high frequency verbs such as gone still occur in the older pattern as a kind of fossilized use). In the case of German, however, I argue that standardization of the language has been quite important in the modern-day pattern of auxiliary selection and that the greater frequency of the Perfect in German overall has resulted in more stable selectional criteria.
.
Introduction
European languages from the Romance and Germanic groups share a formally similar verbal construction, traditionally referred to as the Perfect.1 This Perfect is a periphrastic construction, made up of an auxiliary verb plus the Past Participle (PP) of the main verb (e.g. has seen). In these languages, the Perfect involves, or involved in some past stage of the language, a selection of one of two auxiliaries, either a grammaticized variant of the possession verb have or the copular auxiliary
* I am grateful to Dawn Nordquist for comments on an earlier draft of this paper and Susan Kim for reading and making suggestions on later drafts. I would also like to thank the two anonymous reviewers and Raúl Aronovich for helpful comments on the paper. All errors remain my own. . I will follow the convention of using capital letters when referring to language specific categories and lower-case when denoting categories in a universal sense, e.g. Comrie (1976).
K. Aaron Smith
be.2 Traditional grammatical treatments concerning the selection of have or be are quite similar across languages; typically be is said to occur with intransitive verbs, particularly those that are mutative, i.e. those indicating a change of state, e.g. become, or physical location, in which the subject moves into a location different from its starting point. Have is said to occur among transitive and non-mutative intransitive verbs, i.e. in those contexts where be does not occur. The most cursory observation among languages with have and be auxiliary types, however, shows that these selectional criteria are inadequate. Even modern linguistic treatments of auxiliary selection, which tend to elaborate these traditional syntactic/semantic criteria in terms of unaccusativity (e.g. Burzio 1986), telicity (Lieber & Baayen 19973 ) or prototype features of transitivity and mutativity (Shannon 1990, 1995), show “irregularities”, in which certain verbs become associated with one or the other auxiliary type despite their syntactic/semantic status (Sorace 2000: 887). Furthermore, given that there exists quite a bit of crosslinguistic variation in the specific lexical verbs that occur with have or be, it is odd that many approaches attempt to unify accounts for auxiliary selection between have and be cross linguistically, often even universally (e.g. Hoekstra 1999). Crucially, my approach to explaining the selection of have and be rejects a priori synchronic universal principles; instead this approach views particular patterns of auxiliary usage as a diachronic phenomenon. The auxiliaries have and be have their origins in non-auxiliary, main verb uses, about which much has been written (see for instance the survey in Vincent 1982). Have is originally a verb of possession and it is transitive. Have as a transitive verb will necessarily take a Direct Object NP; a Past Participle can be used as an adjective of that Direct Object NP, which we can paraphrase for our purposes here as “to have an NP in the state of ADJ”. The verb be, however, is intransitive, serving to link an NP subject with either another NP (predicate nominative) or an adjective (predicate adjective). When be is followed by a PP, it is taken to be a
. Throughout this paper, an auxiliary in small caps (have/be) is used to denote the abstraction of that auxiliary type and not necessarily the English specific verbs, have or be. . Lieber and Baayen’s analysis is actually different from those involving telicity in that they introduce the semantic feature [IEPS = Inferable Eventual Position or State] to describe auxiliary selection in Dutch. A [+IEPS] context means that the eventual position or state of the highest argument of a verb is inferable, in which case the verb would select be. Thus, [+IEPS] is similar to telicitity, but Lieber and Baayen claim that it is distinct in that telicity only indicates the endpoint of a verb (e.g. yawn) while it does not necessarily place the highest argument in a specific place or state (1997: 800). Thus, while a telic verb modified by a durational adverb will receive an iterative interpretation, a [+IEPS] verb cannot be so modified, *He arrived in Amsterdam for hours.
Language use and auxiliary selection in the perfect
predicate adjective in the early stages of the auxiliarization process. This stage of the relationship between have or be and the past participle is schematized in (1). (1) [have + [DO + PPadj ]] [be + PPadj ]
In this way, we can see how the have type comes to be associated with transitive verbs and be with intransitives. Be only allows one argument and thus it would select intransitive participles of one argument where the SUBJ of be would also be taken as the SUBJ of the participle. On the other hand, have selects transitive verbs where the NP object is understood to be both the object of the verb have and the adjectival PP. The issue is actually more complex, however, since be is not used with all intransitive verbs. Vincent (1982: 88–89), working with data from Romance, argues that it is intransitive verbs of 1) location, 2) change of location or state (including deadjectival verbs indicating movement into the state expressed by the adjectival base), and 3) impersonal verbs that evolve with the be auxiliary since they all take semantically neuter subjects, i.e. semantically inert subjects whose case meaning is derived by the verb (Vincent 1982: 76). Since be is the prototypical locative verb in a language, it also has a neuter subject and thus aligns with such verbs in the developing Perfect periphrasis.4 From this point, a reanalysis takes place where the PP in each construction is reinterpreted as belonging to the VP at which time it becomes a periphrastic verbal construction, summarized in (2). (2) [have + PP] + DO ]] [be + PP]VP
Since this process is similar, or the even the same, in all languages where have and be have auxiliarized, it is not surprising that when we look at selectional criteria cross linguistically we find similar syntactic/semantic patterns.5 In all such languages, have is transitive and thus selects transitive verbs and be is intransitive, thus selecting intransitive verbs early on. Whether an identical semantic set of intransitive verbs is always involved with the be-type auxiliary at this very early stage is still in need of investigation. What we do know, however, is that in each language where this auxiliarization process has occurred, subsequent use of have + PP or be + PP constructions will lead to their developing in unique ways such that . Vincent’s Case Theory account (e.g. Fillmore 1968) is reminiscent of the unaccusativity hypothesis, where the class of verbs selecting be have an underlying object but no subject. . This phenomenon has been referred to as the “source determination hypothesis” (Bybee et al. 1994).
K. Aaron Smith
they defy analyses positing synchronic universals. Thus, the only sense in which universality might be invoked is in the diachronic realm based on the fact that the grammaticization of have and be into auxiliaries is a nearly identical process in all languages where it occurs. Invoking a usage-based model of language storage and processing, I show how frequency impacts the mental representation of the have and be auxiliaries, leading to unique language-specific patterns of auxiliary selection in English and in German. In the case of English, I show how the stronger mental representation of the have type has fueled its spread in the language so that it has replaced all erstwhile uses of be. In German, however, the data show a very stable stage of auxiliary selection between be and have. Here I argue that this stability is a result of the overall greater frequency of the Perfect in German, which has led to the storage and categorization of have and be in the minds of speakers in ways very much like the usage prescribed in the Standard Language.
. A usage-based model of language storage and processing For the past several decades, dominant theories of language have been those in which the lexicon and the grammar are separate, involved in different phases of processing. However, data concerning certain aspects of language use have challenged linguistic models based on this “rule/list” view (Bybee 2001: 20, and see also Bybee 1985). One such piece of evidence involves the lexical diffusion of specific diachronic changes in grammar (Hooper 1976; Bybee 1995; Philips 1984, 2001), in which a pattern extends through language, moderated by certain aspects of frequency. Hooper (1976), and subsequently Bybee (1995), shows the effects of frequency involved in the morphological leveling of Past Tense verb forms in the history of English, where there has been a tendency for Strong Verbs (i.e. those employing ablaut to signal past tense, speak∼spoke) to “regularize”, becoming Weak Verbs (i.e. those using a dental suffix to signal past tense, -ed). To understand clearly how frequency is involved in this change, it is necessary to distinguish two types of frequency, token and type frequency. Token frequency refers to the number of times a given item occurs in running text. Thus, if we were interested in knowing the token frequency of a given verb, we might count it as appearing 25 times, in which case, it would have a token frequency of 25. Type frequency on the other hand refers to the number of different lexical items that are involved in a specific linguistic construction or pattern. For instance, if our concern is the Weak Past Tense form, we might count that it occurs with 15 different verbs in a given text, in which case we would say that it has a type frequency of 15.
Language use and auxiliary selection in the perfect
These two aspects of frequency are important in the replacement of Strong Verbs by Weak Verbs because, as Hooper (1976) shows, this pattern of regularization has a tendency to affect Strong Verbs of lower token frequency at a faster rate than Strong Verbs of high token frequency. Bybee (2001) argues that her findings in Hooper (1976) support a model of language storage and processing which she refers to as the Network Model (first presented in Bybee 1985), in which linguistic items, at the level of word, are stored at a single site in the mind and share a number of associative links, based on phonetic similarity, semantic affinity, and/or the specific patterns of organization and categorization that these stored items enter into (Bybee 2001: 8). An important aspect of this model is that items are stored with redundant grammatical information; Past Tense forms are stored along with the type of Past Tense they occur in. In such a model, it is hypothesized that aspects of language use, such as frequency, directly impact the stored representation of words as well as the grammatical patterns in which they occur. Words and patterns of usage of higher frequency have a stronger representation and those of lower frequency have weaker representations (ibid 202:10ff.), which in turn affect how they are processed. Thus, in the case that a given Strong Past verb has a high token frequency, its use in that pattern is easily recalled in processing, in which case it can retain that form for a longer time; that is, it is able to resist the spread of the Weak Verb pattern. Likewise, if a given Strong Past Tense form is not very frequent, its weaker representation makes it more difficult to recall and thus leaves it susceptible to change. Since the Weak Past form has a greater type frequency, again due to the fact that it occurs with so many different verbs, it is easily recalled; in the case that a weakly represented Strong Past verb is not recalled with its Strong Past form, it will likely take the stronger Weak Past pattern. This explains why frequent Strong Past forms such as spoke are not regularized to *speaked, but more infrequent Strong Pasts often are, in so far as a verb such as dive might yield dived∼dove among native speakers. An extreme case of this is a situation where paradigmatic alternates come from different stems, such as with the verb go, which because of its very frequent use in the past supports a suppletive Past Tense form, went. Having presented the basic tenets of a usage-based model and the effects that frequency can have on storage and processing within the model, I now turn to an analysis of auxiliary selection in English and German where I will demonstrate that certain aspects concerning the frequency of the Perfect construction in the two languages have led to very different outcomes of auxiliary usage and selection.
K. Aaron Smith
. The role of frequency in the replacement of be in English In Modern English, there is essentially no choice in auxiliary selection; all verbs select have as their auxiliary (but see below on be gone). An example of the Perfect in Modern English is shown in (3). (3) They have already spoken to her.
Although have is the only auxiliary which appears in the Perfect Construction in Modern English, older varieties of the language showed selection between have and be; an example of the Perfect Construction in Old English involving be is given in (4). (4) Nu is se dæg cumen now be:3rd.sing.pres def.art day:nom.sing. come:past participle ‘Now the day has come.’ Beowulf 2646 (Klaeber 1950)
In Old English, the choice between have and be is very much like the auxiliary selection criteria mentioned earlier; be occurs mostly with intransitive, mutatives and have with transitive, non-mutative verbs (see e.g. Mitchell & Robinson 1992: 111). Over time, however, more and more verbs fitting the intransitive, mutative type began to appear with have. In Smith (2001), I studied the effects of frequency on the diachronic spread of auxiliary have in English, and I determined that frequency was involved in very much the same way that it is with the selection of the Weak and Strong verb types referred to in the last section. Table 1 shows the type and token frequency for the be and have Constructions in five periods of English. The period is listed in the left column, under which the total number of words from the texts used to collect data from that period is given.6 For each period, the number of be and have types/tokens is given as a percentage of the total Auxiliary + PP (Perfect) occurrences, with the raw numbers expressed in parentheses. The last column shows the total instances of Auxiliary + PP (Perfect) in the sample from each period. From the data in Table 1, we see that the have auxiliary always had a higher type frequency, which is not surprising given the syntactically and semantically more restrictive contexts of be. The higher type frequency of the have-type gave it a stronger representation in the mind and thus have was the auxiliary to spread in the history of the language. However, while this representational strength might account for the overall direction of the replacement, the interaction of have with the token frequency of specific verbs shows how such a force proceeds. . For details about the sample, see Smith (2001).
Language use and auxiliary selection in the perfect
Table 1. Type and Token Frequencies of have and be Auxiliary Constructions in the History of English (based on Smith 2001) TYPE Old English (33,900 words) Early Middle English (112,766 words) Later Middle English (21,679 words) Early Modern English (96,679 words) 19th century (118,262 words)
BE 16% (11)
HAVE 84% (57)
N 68
11% (12)
89% (92)
104
11% (9)
89% (70)
79
92% (115)
125
97% (311)
319
HAVE 79% (85) 76% (214) 89% (96) 96% (319) 96% (839)
N 103 283 108 332 877
8% (10) 3% (8) TOKEN
Old English Early Middle English Later Middle English Early Modern English 19th century
BE 21% (18) 24% (69) 11% (12) 4% (13) 4% (38)
By the time we get to the 19th century, very few verbs remain with the be auxiliary, in fact only 8 types out of 319. Nevertheless, from these 8 types there are 38 tokens of the be + PP Construction, meaning that certain of these types have multiple occurrences. Closer investigation of the types that occur with be in the 19th century shows that of the 38 tokens, 11 occur with come and 17 occur with go. The high token frequencies of come and go in the 19th century within the moribund be Construction show that as the have Construction spread, it first affected be verbs of a relatively lower token frequency.7 This aspect of the replace. However to characterize all verbs in the 19th century data as being of high frequency is not accurate and indeed there are certain lexical verbs in that period of a rather low frequency which also appear in that construction. One such instance is the verb adjourn, which occurs once with be in Emily Brontë’s Wuthering Heights: . . .whence after putting on dry clothes and pacing to and fro thirty minutes to replace the animal heat, I am adjourned to my study feeble as a kitten – almost too much to enjoy the cheerful fire or smoking coffee which the servant has prepared for my refreshment. [p. 31] Important here is that this passage is uttered by a character whom the author wants to portray as upper class, educated and slightly uptight; Brontë’s use of the by-then archaic auxiliary con-
K. Aaron Smith
ment is quite similar to the regularization of Strong Verbs discussed in the last section; in the case that a given verb is infrequent, it is not easily recalled in the be Construction and thus the more dominant have pattern, which is easily recalled due to its high type frequency, will be substituted. As the 19th century data show, however, in the case that the lexical verb is of relatively high frequency, its occurrence with be can be more easily recalled for a longer time, thus resisting the spread of have.8 This suggests that a usage-based model can be extended from phonology and morphology to include syntactic phenomena as well (Smith 2001, and see also Bybee & Thompson 2000) and in fact, Rydén (1991) suggests that have and be types comprise a sort of “syntactic paradigm”, underscoring the similarities between some morphological and syntactic phenomena. Further evidence that frequency plays an important role in the patterning of auxiliary selection comes from the semantic split of the verb go between its use with be and with have. The oldest uses of the Perfect in English signaled resultative meaning, i.e. an action, event or situation that occurs in the past and continues into the present time. From there, the Perfect extended into anterior contexts, where the action, event or situation does not physically extend into the present but its effects do (see Carey 1994 on this development in English). The diachronic relationship between resultative and anterior is widely attested in the languages of the world (Bybee et al. 1994) and we can understand the semantic development partially in terms of bleaching, whereby specific semantic features are lost over time. Thus, once the resultative loses the semantic feature of physical continuation in the present time and only its effects are felt, it becomes an anterior (Bybee et al. 1994: 69). The highly frequent verb go shows use with both the be and have auxiliaries in Modern English; however, its use in the older morphosyntactic pattern, i.e. with be, retains the older resultative meaning while its use with have shows the later anterior meaning. In other words, go (or perhaps more correctly gone) has become entrenched in the older be auxiliary pattern with the older semantics of the Perfect due to its high frequency, which has allowed it to develop a very strong and somewhat independent (i.e. idiomatic) representation. In this way, be gone is similar to suppletive morphological forms, which also involve high frequency
struction emphasizes this characterization. Thus, the reason we find instances of be with such low frequency verbs like adjourn in the 19th century data is not an indication of the productivity of that construction, but instead has to do with the fact that the data comes from literary sources. In such a case, the author is able to suspend the selection of have and fall back on the older semantic criteria of auxiliary selection. . Aronovich’s (2003) study of the diachronic spread of have in Spanish shows similar frequency effects in so far as the last verbs to hold out against the spread were verbs such as ir ‘go’ and venir ‘come’, i.e. frequently used verbs.
Language use and auxiliary selection in the perfect
material, thus underscoring even more how similar certain types of syntactic and morphological structuring can be.
. Auxiliary selection in German . The stability of auxiliary selection in German As Shannon (1995: 144) points out, in cases where selectional criteria for the have and be auxiliaries make the wrong prediction in language, the productive pattern is always in favor of have (but see Tuttle 1986 on the opposite development in Italian dialects). In the model offered in the last section, we can explain this by the fact that have in all languages with the two auxiliary types always has a higher type frequency because of its fewer co-occurrence constraints, thus making it the more productive of the two auxiliaries. Despite however, the effects of type frequency on the productivity of have, we find that it does not predict be replacement in all languages where such selection occurs. For instance, auxiliary selection between have and be among speakers of Modern Standard German is very stable. In order to see this stability clearly, consider first the list of selectional criteria for be offered in a traditional description of the Standard Language, such as that offered in (5); have is, then, in complementary distribution, i.e. occurring everywhere else that the criteria in (5) do not specify. (5) Selectional Criteria for the be auxiliary in German (Durrell 2002: 246) – intransitive verbs of motion – intransitive verbs expressing change of state (i.e. mutative) – most verbs meaning happen, succeed or fail – verbs bleiben and sein
To these criteria, Durrell, and other grammars of Standard German (see e.g. Drosdowski 1984), adds that certain verbs show variation between have and be. The variation is accounted for in syntactic/semantic terms. For instance, be verbs that are normally intransitive may take have in the case that they are used transitively, as the sentence in (6a) shows (see also Shannon 1995). Other cases involve motion verbs that indicate a change in location, as for instance, the verb segeln ‘sail’ which is said to take have when the action of sailing does not place the subject in a different position from the starting point as in (6b). When it does, however, be is used, as shown in (6c). (6) a.
Sie hat einen neuen Porsche gefahren. she has a new Porsch driven ‘She drove/has driven a new Porsche.’
K. Aaron Smith
b. Wir haben den ganzen Tag gesegelt. we have the entire day sailed ‘We sailed/have sailed/were sailing the entire day.’ c. Sie sind nach Island gesegelt. they are to Iceland sailed ‘They sailed/have sailed to Iceland.’
Also, despite the overall syntactic/semantic nature of the selectional criteria presented so far, there is a certain amount of lexical specificity necessary for a comprehensive description, e.g. the listing of the stative verbs bleiben and sein. Other intransitive stative verbs occur with have, such as dastehen ‘stand there’, particularly in the North of Germany. Again, outside of these more restricted conditions, all other verbs occur with have and thus have has a higher type frequency in German. Given the effects of high type frequency in English, and other languages, we might predict that there would be a change in progress in German, where we would find evidence of erstwhile be verbs occurring with have. In order to test this hypothesis, I collected tokens of both have + PP and be + PP Constructions from a number of text types. In the case of be + PP, care was taken to consider only those instances of that construction that represented the Perfect and to exclude all tokens of the Stative Passive, which has the same periphrastic form, e.g. Das Haus ist schlecht gebaut. ‘The house is poorly built’. Such instances of the Stative Passive are distinguished from the Perfect in that agency in the former is demoted from the subject position while in the Perfect the subject is canonically the agent as well. My database for this test was a 36,230-word collection of transcribed interviews and newspaper articles available on the World Wide Web.9 The search yielded a rather large number of tokens which were then organized and analyzed for variation in auxiliary use. Despite a database of 1,409 tokens, however, essentially no variation was found other than that predicted by standard descriptions of the language such as those listed above in (5). The lack of variation in the data suggests that whatever the synchronic description of auxiliary selection in German, be it semantic, syntactic, lexical, or a combination of all three, there is apparently no change in progress in this area of the grammar in Modern Standard German. Furthermore, given the lack of variation and the efficacy of prescribed selectional criteria like those given above to predict auxiliary selection, we can conclude that the Standard Language is being closely followed by German speakers.
. See Appendix.
Language use and auxiliary selection in the perfect
. The effects of frequency on the stability of auxiliary selection in German The stability in auxiliary selection in Modern German is especially interesting when we consider that varieties of German in earlier periods of the language showed considerable variation. For instance, Grimm (1889) shows variation between the be and have type auxiliaries in the Middle High German (1050–1350), as shown in (7) and (8), where we see the both auxiliaries used with the main verbs gewesen ‘been’ and gevarn ‘traveled’. (7) a.
wâren gewesen be: 3rd pl.past be: past participle ‘had been’ (quoted in Grimm 1889: 188) b. hât gewesen have: 3rd.sing.pres. be: past participle ‘had been’ (quoted in Grimm 1889:188)
(8) a.
sîn wir gevarn her be: 1st.pl. we travel:past participle here ‘we have traveled here’ (Nibelungenlied 400, quoted in Grimm 1889: 191) b. ich her gevarn hân I here travel:past participle have:1st.sing. ‘I have traveled here.’ (Nibelungenlied 401, quoted in Grimm 1889: 191)
While Grimm (1889: 190) tells us that there is a difference in the use of verbs like fahren in (8a–b) where the be auxiliary is used when the situation denotes a completed action and haben for static (anhaltende) situations, similar to semantic descriptions involving change of location given by modern traditional treatments of Standard German (see above), the examples in (8) both contain the adverbial her ‘here’ which gives a clear end point to the movement and thus both can be said to show arrival in a place different from the starting point. Furthermore, evidence of variation in older varieties of German comes from data in the Old Saxon Heliand (ca. 830). Arnett (1997) finds instances of verbs that would normally be used with be appearing instead with have. Of the 36 motion verbs found in Heliand, all fit to some degree the mutative class, placing the subject in a different location at the end of the movement, and almost all of those 36 occur with the predicted be auxiliary. The two verbs given in (9a–b) from Arnett, however, do not, occurring instead with have.10 . While on the surface of things, the spread of have to verbs such as gegangen and furfana might seem to discount my earlier claim that the spread of have affects low frequency items at a faster rate than low frequency ones, given that both verb stems, e.g. gehen and fahren, are quite frequent in Modern German, in the variety of German of Heliand, neither has an especially strong frequency. Compounds with -gang- in the perfect occur only 3 times in addition to the one instance with have and compounds with -far- only twice. The specific Past Partciples
K. Aaron Smith
(9) a.
sô thiu frî habdun gegangen to them gardon so the women had go:pp to the garden ‘so the women had gone into the garden’ b. hie habit sia iu furfarana he has them already travel:pp ‘he has already traveled to them’
Although in need of finer-grained study, this type of variation might indicate that there was a change in progress in some older varieties of German. Consider for instance that from the data in Arnett (1997), instances of the be auxiliary construction accounted for 14% of the Perfect Constructions, while haben accounted for 85%. Thus, in Old Saxon we find a type frequency for have which is even higher than that for Old English, where the spread of have was clearly underway and for which we know the outcome.11 While increasing literacy and education in a standard where auxiliary selection was more codified explains in part why earlier patterns of variation diminished, knowledge of the Standard Language alone is not enough to account for the stability of auxiliary selection.12 While speakers can sometimes reflect consciously on semantic, syntactic or lexical criteria for auxiliary selection in the act of writing or some types of speaking, most language use situations are not like this, and auxiliary selection most often occurs in the unconscious act of on-line language production. Instead, it is the unique conditions that frequency of the two auxiliary types sets up in German which account for the ability of speakers to use the two auxiliaries in keeping with the criteria prescribed through education in the Standard Language. To being with, in the data collected for this study, be has a higher type frequency than the analogous auxiliary in any stage of historical English. Thus, as shown in Table 2, we see that be has a proportional type frequency of 212 out of a total of 994 different verbs; in other words, 21% of all verbs used in the Perfect select the auxiliary be. gegangen and furfarana, however, only occur once, so that we may consider these to be verbs of relatively low frequency in that text. . It is relevant here to note that the Low Germanic dialects which gave rise to English are historically identified with the Low Germanic dialects on the Continent (esp. Saxon). Thus, given that the oldest texts in both Old English and Old Saxon on the Continent show data suggesting a change in progress, it seems like that the replacement of the be auxiliary by have was already underway at the time when the Low Germanic tribes from Europe came to the British Isles. . In contradistinction, those periods of English where have made the most significant spread was during a time of very low literacy and little access to education for most speakers of the language.
Language use and auxiliary selection in the perfect
Table 2. Type Frequency for be and have in Modern German TYPE SEIN 21% (212)
HABEN 79% (782)
N 994
In contrast, the highest type frequency of be in the history of English during the OE period only reaches 16%. And while a difference of 16% and 21% might not seem very great, two factors make the difference more appreciable. First, the data from historical English comes from literary sources in which we can assume that the use of auxiliary be was more conservative than in the spoken language of the time, particularly given the clear direction of replacement already apparent in the literary data. Thus, it is quite likely that the type frequency for be in spoken OE, and every period of English, was actually lower than the numbers show. Furthermore, and most importantly, in databases of roughly equal size for Old English and Modern German, i.e. 33,900 words for Old English and 36,230 words for Modern German, the be auxiliary occurs about 18 times more often in Modern German than in Old English. The very frequent occurrence of the construction in German is explained by its specific semantic developments. As we saw earlier, the Perfect in Old English was limited to resultative and a few emerging anterior uses (Carey 1994; Smith 2001), but in Modern German, the construction has developed further such that it includes not only resultative and anterior, but simple past uses as well (see e.g. Abraham & Conradie 2001); when the present time relevance of the anterior is lost, the form can begin to signal simple past meaning (Bybee et al. 1994). Thus, in Modern German the Perfect has occasion to occur across virtually all of the past time domain, making it a very frequent construction.13 The high frequency of the Perfect overall is important because it reinforces the distinct syntactic, semantic and lexical types with each auxiliary. In some sense, it “drills” and entrenches the criteria for be selection that speakers have learned through education and literacy. In this light, it is interesting to note that in Western European languages where selection between have and be is maintained, the Perfect has moved into an area of the tense/aspect system where it signals simple past meaning (French, Italian, Dutch, etc.). As a result the Perfect has a higher frequency compared to those languages where its semantic domain is more limited, i.e. to only resultative and anterior uses (e.g, English, Spanish, Swedish, etc.14 ). . Note further that the past time domain in German does not have a perfective/imperfective distinction so that the Perfect is used for virtually all past time reference. . Although in some varieties of Modern Peninsular Spanish the Perfect has moved into the domain of the “simple past” or “preterit”, this development has taken place after replacement of be.
K. Aaron Smith
How much of the past time domain is signaled by the Perfect is thus perhaps an important condition in the stability of auxiliary selection because of its effects on the frequency of the two constructions overall although more work on this hypothesis is necessary before any more can be said.
. Conclusion In this paper, I have shown the importance of frequency in the patterning of auxiliary types in two languages, although to be sure, the specific forces that shape auxiliary selection patterns in any language are myriad and there are other factors involved. Whatever aspects of auxiliary selection one focuses on, however, a priori universal and synchronic organizing principles alone are inadequate and thus I have argued that it is only by looking at the more dynamic aspects of language use and the diachronic effects of language use on linguistic form that we can arrive at an explanation of auxiliary selection for any one language and Language more generally.
Appendix Sources for the German Database 1.
2. 3. 4.
German News, available at http://www.mathematik.uni-ulm.de/germnews (Database includes: Do, 09.01.2003 22:00 MEZ/ Do, 02.01.2003 18:00 MEZ/ Fr, 03.01.2003 22:00 MEZ/ Mi, 01.01.2003 16:00 MEZ/ Sa, 04.01.2003 23:00 MEZ/ Di, 07.01.2003 22:00 MEZ/ Fr, 10.01.2003 23:00 MEZ/ Mi, 08.01.2003 18:00 MEZ/ Mo, 06.01.2003 18:00 MEZ) Im Gespräch mit David Witts available at http://www.gfo2003.de/interviews/ Interview mit Klaus Kinkel available at http://europa.eu.int/en/agenda/igc-home/msspeech/ state-de/kinkel3.htm Ein Gespräch mit Helga Goetze available at http://www.helgagoetze.de/rotelieb.htm\
References Abraham, W. & Conradie, C. J. (2001). Präteritumschwund und Diskursgrammatik: Präteritumschwund in gesamteuropäischen Bezügen – areale Ausbreitung, heterogene Entstehung, Parsing sowie diskursgrammatische Grundlagen un Zusammenhänge. Amsterdam: John Benjamins. Arnett, C. (1997). Perfect auxiliary selection in Old Saxon Heliand. American Journal of Germanic Linguistics and Literatures, 9 (1), 23–72. Aranovich, R. (2003). The semantics of auxiliary selection in Old Spanish. Studies in Language, 27 (1), 1–37.
Language use and auxiliary selection in the perfect
Bybee, J. L. (1985). Morphology: A study of the relationship between meaning and form. Amsterdam: John Benjamins. Bybee, J. L. (1995). Regular morphology and the lexicon. Language and Cognitive Processes, 10 (5), 425–455. Bybee, J. L. (2001). Phonology and Language Use. Cambridge: CUP. Bybee, J. L. & Thompson, S. (2000). Three frequency effects in syntax. Berkeley Linguistics Society, 23, 65–85. Bybee, J. L. Perkins, R., & Pagliuca, W. (1994). The Evolution of Grammar: Tense, aspect and modality in the languages of the world. Chicago, IL: University of Chicago Press. Burzio, L. (1986). Italian Syntax: A government-binding approach. Dordrecht: Reidel. Carey, K. (1994). The grammaticalization of the Perfect in Old English: An account based on pragmatics and metaphor. In Pagliuca, W. (Ed.), Perspectives on Grammaticalization (pp. 105–117). Amsterdam: John Benjamins. Comrie, B. (1976). Aspect. Cambridge: CUP. Drosdowski, G. (Ed.). (1984). Der Duden in 10 Bänden, Band 4: Die Grammatik. Mannheim: Dudenverlag. Durrell, M. (2002). Hammer’s German grammar and usage (4th ed.). Chicago, IL: McGraw-Hill. Grimm, J. (1889). Deutsche Grammatik. Gütersloh: Bertelsmann. Fillmore, C. J. (1968). The case for case. In E. Bach & R. Harms (Eds.), Universals in Linguistic Theory (pp. 1–88). New York, NY: Holt, Rinehart and Winston. Hoekstra, T. (1999). Auxiliary selection in Dutch. Natural Language and Linguistic Theory, 17, 67–84. Hooper, J. B. (1976). Word frequency in lexical diffusion and the source of morphophonological change. In W. Christie (Ed.), Current Progress in Historical Linguistics (pp. 96–105). Amsterdam: North Holland. Klaeber, F. (Ed.). (1950). Beowulf and the Fight at Finnsburg. Lexington, MA: D.C Heath and Company. Lieber, R. & Baayen, H. (1997). A semantic principle of auxiliary selection in Dutch. Natural Language and Linguistic Theory, 15, 789–945. Mitchell, B. & Robinson, F. C. (1992). A Guide to Old English. Oxford: Blackwell. Phillips, B. S. (1984). Word Frequency and the actuation of sound change. Language, 60, 320– 342. Philips, B. S. (2001). Lexical diffusion, lexical frequency and lexical analysis. In J. L. Bybee & P. Hopper (Eds.), Frequency and the Emergence of Linguistic Structure (pp. 123–136). Amsterdam: John Benjamins. Rydén, M. (1991). The be/have variation with intransitives in its crucial phases. In D. Kastovsky (Ed.), Historical English Syntax (pp. 343–354). Berlin: Mouton de Gruyter. Shannon, T. (1990). The unaccusativity hypothesis and the history of the perfect auxiliary in Germanic and Romance. In H. Andersen & E. F. K. Koerner (Eds.), Proceedings of the Eighth International Conference on Historical Linguistics (pp. 461–499). Amsterdam: John Benjamins. Shannon, T. (1995). Toward a cognitive explanation of perfect auxiliary variation: Perfect auxiliary variation: Some modal and and aspectual effects in the history of Germanic. American Journal of Germanic Languages and Linguistics, 7, 129–163. Smith, K. A. (2001). The role of frequency in the specialization of the English anterior. In J. L. Bybee & P. Hopper (Eds.), Frequency and the Emergence of Linguistics Structure (pp. 361– 382). Amsterdam: John Benjamins.
K. Aaron Smith
Sorace, A. (2000). Gradients in auxiliary selection with intransitive verbs. Language, 76 (4), 859– 890. Tuttle, E. F. (1986). The spread of esse as a universal auxiliary in central Italo-Romance. Medioevo romanzo, 11 (2), 229–287. Vincent, N. (1982). The development of the auxiliaries habere and esse in Romance. In N. Vincent & M. Harris (Eds.), Studies in the Romance verb (pp. 71–96). London: Croom Helm.
TSL[v.20020404] Prn:26/02/2007; 10:46
F: TSL69LI.tex / p.1 (271)
Language index
A Abruzzese ,
B Basque
C Catalan , , Chinese , –, , –, –, , ,
D Danish Dutch , –, , , , , , , , , –, , , –, , , , –, ,
E English , , , , –, , , , , , , , , , , , , –, , ,
F French , –, , –, , , , , , –, –, –, G German , , –, , , –, , –, –, , –, –, , , –, –, –, , –, , , , – Greek , – I Icelandic , Italian , , –, , –, , –, , , , –, , , , , , , –, , , –, –, , –, –, , K Koiné Korean , –, –
L Latin , , –, , , –, –, –, O Old Saxon , P Paduan , , , , –, , , , , , – Q Québec French R Romanian , –, , –, , , , S Spanish , , , , , , , –, , , , , T Terracinese , ,
TSL[v.20020404] Prn:26/02/2007; 10:47
F: TSL69NI.tex / p.1 (273)
Name index
A Aerts , , , Amenta – Aranovich , , , , , , , , Arnett , , Avram , , , , B Baayen , , , , , , , , , Benincà , , –, –, , , , , , – Bentley , , Borer , Burzio , , , , , , , , , , , , , , Bybee –, , , , –, , C Chomsky , , , , , , Cocchi , Coleman D Dobrovie-Sorin , , Dowty , , , , , , , , , , Dubuisson , F Frey –, , –, G Grimm Grimshaw , , , , H Haider , , –, , , Happ Heine , ,
Hoekstra –, , , , , , , , , Hooper , Hopper , , Horrocks , , , Hu , , , , , Huang , , , , , K Keller , , , , , , –, , , , , , , , –, Koo Kuteva L Legendre , , , , , , , , –, Lehman Levin , , , , , , , , , , , , , , , , Lieber , , , , , , , , , Loporcaro , M McClure , , Montrul P Pan –, , , , –, , Perlmutter , , –, , , , , , , , , , , , , , , , Pinkster , Postal Pouilloux
R Ramat , Rappaport Hovav , , , , , , , , , , , , , , , Renzi Rosen , , , , , , , , , , S Shannon , , , , , , –, , , Smith , , , , , , –, Sorace , –, , , –, –, –, , , , , , , –, –, , –, , , , , , , –, –, , , , , –, , , T Traugott Tuttle , , , V Van Hout , , , –, , , , , – Van Valin , , , , , , , , , Vanelli , –, Vendler , Verkuyl , Vincent , , , , Y Yang , , , , , – Yu , , , , – Z Zaenen , –, , , , , , , ,
TSL[v.20020404] Prn:26/02/2007; 10:55
F: TSL69SI.tex / p.1 (275)
Subject index
A Accusativus cum Infinitivo (AcI) , –, –, , – activity , , , , , , , , , , , , , , , , , , actor , , , , –, , , , – adjunct – affectedness , , , , , , , , affix , agentivity , , , , , , , , , , , –, –, , , , –, –, , , , , , , , agreement , , , , –, , –, –, –, –, , , , anaphor(s), anaphoric , , , , , anteriority , , –, anticausative , , , , , , , , , , , , aorist , –, –, , , argument external , , –, , , , , , , , , , , , , internal , , , , , , , , , Auxiliary Selection Hierarchy (ASH) , , , –, , , , , –, , , , –, –, , , , , , , , , , , –, , ,
aspect, aspectual , , , , , –, , , , –, , , , , , , , , , , , , –, , , , , , , –, , , , , , , , , , , , , , , ,
B BE –, –, –, , , , , –, , –, –, –, –, –, , –, , , –, –, –, –, –, –, , –, –, –, , , –, –, – binding , , , , –, , –, , , ,
C calque, calquing , case , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , –, – causation , , , , , causative , , , , , , clitic –, , , , , , –, , , , , , , , , compound , , , , ,
constructional/constructionist approach , , , D definiteness , – deletion deponent detransitivized direction , , , , , –, , , , , displacement , , , G gerund , Government and Binding , , grammaticalization , , , , , , , , , , , –, , , , H harmonic alignment , , HAVE –, , –, –, –, –, –, –, –, –, –, –, –, –, , –, , –, , , –, –, , –, –, –, –, –, , , , , , –, –, , –, – homogeneity , , , , , , I iconicity , ,
TSL[v.20020404] Prn:26/02/2007; 10:55
F: TSL69SI.tex / p.2 (276)
Subject index imperfective , , , , , impersonal , , , , , , , , , , , , , , , , inchoative , , , indefinite , , , , , , , , –, , , , , , , , , , –, , infinitive , , , , , , –, , intransitive , , –, , , , , , , , , –, , , , , , , , , , , , , , , , , , , inversion , , , –, , –, , –, –, , , –, –, –, –, irrealis , –, , , –
N ne-cliticization , , , , , , , –, , , , , , –
P Parameterized Linking , , , , , , , – participle , , , –, , –, , , –, –, –, –, –, –, , , , , , , , –, , passive , , , –, –, , –, , , , , , , , , past , , , , –, , , –, –, –, –, , , , , , , , , , , , , –, –, –, , , , perfect , , , , , , –, , , , –, , , , , , , , , –, –, –, , , , , , , –, , –, , – periphrasis, periphrastic , , , , , , , –, –, , , , pluperfect , , , , , , power hierarchy , , , , present , , , –, , , , , , , , , , –, , , , , , , , , , , , , , , , , , , , , , –, , –, , , , , , , , , , , progressive , , –, , , , , , , projectionist , , , , , punctual, punctuality , , –, , , , ,
O Optimality Theory ,
R reduplication
L linking , , , , , , – locative , –, –, –, , locomotion , , , , , –, , , , M mediopassive , , –, motion, motional , , , , , , , , , , , , , –, , , , , , , , , –, , mutative, mutativity , –, ,
reflexive –, , , , , , , , , , , , –, , , , –, Relational Grammar , , restructuring , , resultative , , , , –, , , , , , –, , , Role and Reference Grammar
S stative (verb), stativity , , , , , , , , , , , , , –, , , , sub-extraction – subcategorization , ,
T telic, telicity –, , , , , , , –, , , , , , , , , , , , , , , , , , , , , , , , , , –, , , , , –, , , , , , , template augmentation , tense , , , , –, , , , , , –, –, , , , , , , , , transitive, transitivity , , –, , , –, –, –, , , , , , , , , , , ,
U unaccusative –, –, , , , , , –, , , , , –, –, , , –, , , , –, , , , , , –, – unergative , , , , , , , , , , –, , , , –, ,
TSL[v.20020404] Prn:26/02/2007; 10:55
F: TSL69SI.tex / p.3 (277)
Subject index
–, –, , , , –, – Universal Alignment Hypothesis , , V Verb activity , detransitivized , , –, , , – intransitive , , , , , –, , , , , , , , , , , , –, , , , , , , ,
of direction of emision , , , , , , –, , , , , of motion , , , –, , , , , , , , –, , , , , , , , , , , , , , , , serial , , , –, , stative , , , , , , –, , , , , –, , transitive , , , , , , , , , –, ,
unaccusative , , , , , –, –, , –, , , , , , , , , , , , , , , unergative , , , , , , , –, , –, , , , , –, , –, , , , volitional, volitionality , , , , , , , , , , , , –, , , , , , vP , , , , , , , , , , , ,
Typological Studies in Language A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 71 Nedjalkov, Vladimir P. (ed.): Reciprocal Constructions. With the assistance of Emma Š. Geniušienė and Zlatka Guentchéva. ca. 2150 pp. (4 vols.). Expected June 2007 70 Zúñiga, Fernando: Deixis and Alignment. Inverse systems in indigenous languages of the Americas. 2006. xii, 309 pp. 69 Aranovich, Raúl (ed.): Split Auxiliary Systems. A cross-linguistic perspective. 2007. vii, 277 pp. 68 Abraham, Werner and Larisa Leisiö (eds.): Passivization and Typology. Form and function. 2006. x, 553 pp. 67 Veselinova, Ljuba N.: Suppletion in Verb Paradigms. Bits and pieces of the puzzle. 2006. xviii, 236 pp. 66 Hickmann, Maya and Stéphane Robert (eds.): Space in Languages. Linguistic Systems and Cognitive Categories. 2006. x, 362 pp. 65 Tsunoda, Tasaku and Taro Kageyama (eds.): Voice and Grammatical Relations. In Honor of Masayoshi Shibatani. 2006. xviii, 342 pp. 64 Voeltz, F. K. Erhard (ed.): Studies in African Linguistic Typology. 2006. xiv, 426 pp. 63 Filimonova, Elena (ed.): Clusivity. Typology and case studies of the inclusive–exclusive distinction. 2005. xii, 436 pp. 62 Couper-Kuhlen, Elizabeth and Cecilia E. Ford (eds.): Sound Patterns in Interaction. Crosslinguistic studies from conversation. 2004. viii, 406 pp. 61 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 2. 2004. xii, 319 pp. 60 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 1. 2004. xii, 325 pp. 59 Fischer, Olga, Muriel Norde and Harry Perridon (eds.): Up and down the Cline – The Nature of Grammaticalization. 2004. viii, 406 pp. 58 Haspelmath, Martin (ed.): Coordinating Constructions. 2004. xcv, 578 pp. 57 Mattissen, Johanna: Dependent-Head Synthesis in Nivkh. A contribution to a typology of polysynthesis. 2003. x, 350 pp. 56 Shay, Erin and Uwe Seibert (eds.): Motion, Direction and Location in Languages. In honor of Zygmunt Frajzyngier. 2003. xvi, 305 pp. 55 Frajzyngier, Zygmunt and Erin Shay: Explaining Language Structure through Systems Interaction. 2003. xviii, 309 pp. 54 Aikhenvald, Alexandra Y. and R.M.W. Dixon (eds.): Studies in Evidentiality. 2003. xiv, 349 pp. 53 Givón, T. and Bertram F. Malle (eds.): The Evolution of Language out of Pre-language. 2002. x, 394 pp. 52 Güldemann, Tom and Manfred von Roncador (eds.): Reported Discourse. A meeting ground for different linguistic domains. 2002. xii, 425 pp. 51 Newman, John (ed.): The Linguistics of Sitting, Standing and Lying. 2002. xii, 409 pp. 50 Feigenbaum, Susanne and Dennis Kurzon (eds.): Prepositions in their Syntactic, Semantic and Pragmatic Context. 2002. vi, 304 pp. 49 Wischer, Ilse and Gabriele Diewald (eds.): New Reflections on Grammaticalization. 2002. xiv, 437 pp. 48 Shibatani, Masayoshi (ed.): The Grammar of Causation and Interpersonal Manipulation. 2002. xviii, 551 pp. 47 Baron, Irène, Michael Herslund and Finn Sørensen (eds.): Dimensions of Possession. 2001. vi, 337 pp. 46 Aikhenvald, Alexandra Y., R.M.W. Dixon and Masayuki Onishi (eds.): Non-canonical Marking of Subjects and Objects. 2001. xii, 364 pp. 45 Bybee, Joan and Paul J. Hopper (eds.): Frequency and the Emergence of Linguistic Structure. 2001. vii, 492 pp. 44 Voeltz, F. K. Erhard and Christa Kilian-Hatz (eds.): Ideophones. 2001. x, 436 pp. 43 Gildea, Spike (ed.): Reconstructing Grammar. Comparative Linguistics and Grammaticalization. 2000. xiv, 269 pp. 42 Diessel, Holger: Demonstratives. Form, function and grammaticalization. 1999. xii, 205 pp. 41 Frajzyngier, Zygmunt and Traci S. Curl (eds.): Reciprocals. Forms and functions. Volume 2. 2000. xii, 201 pp.
40 Frajzyngier, Zygmunt and Traci S. Curl (eds.): Reflexives. Forms and functions. Volume 1. 2000. xiv, 286 pp. 39 Payne, Doris L. and Immanuel Barshi (eds.): External Possession. 1999. ix, 573 pp. 38 Siewierska, Anna and Jae Jung Song (eds.): Case, Typology and Grammar. In honor of Barry J. Blake. 1998. 395 pp. 37 Giacalone-Ramat, Anna and Paul J. Hopper (eds.): The Limits of Grammaticalization. 1998. vi, 307 pp. 36 Newman, John (ed.): The Linguistics of Giving. 1998. xv, 373 pp. 35 Givón, T. (ed.): Grammatical Relations. A functionalist perspective. 1997. viii, 350 pp. 34 Givón, T. (ed.): Conversation. Cognitive, communicative and social perspectives. 1997. viii, 302 pp. 33 Fox, Barbara A. (ed.): Studies in Anaphora. 1996. xii, 518 pp. 32 Bybee, Joan and Suzanne Fleischman (eds.): Modality in Grammar and Discourse. 1995. viii, 575 pp. 31 Gernsbacher, Morton Ann and T. Givón (eds.): Coherence in Spontaneous Text. 1995. x, 267 pp. 30 Downing, Pamela A. and Michael Noonan (eds.): Word Order in Discourse. 1995. x, 595 pp. 29 Kahrel (PJK), Peter and René van den Berg (eds.): Typological Studies in Negation. 1994. x, 385 pp. 28 Givón, T. (ed.): Voice and Inversion. 1994. viii, 402 pp. 27 Fox, Barbara A. and Paul J. Hopper (eds.): Voice: Form and Function. 1994. xiii, 377 pp. 26 Lord, Carol: Historical Change in Serial Verb Constructions. 1993. x, 273 pp. 25 Svorou, Soteria: The Grammar of Space. 1994. xiv, 290 pp. 24 Perkins, Revere D.: Deixis, Grammar, and Culture. 1992. x, 245 pp. 23 Kemmer, Suzanne: The Middle Voice. 1993. xii, 300 pp. 22 Payne, Doris L. (ed.): Pragmatics of Word Order Flexibility. 1992. viii, 320 pp. 21 Downing, Pamela A., Susan D. Lima and Michael Noonan (eds.): The Linguistics of Literacy. 1992. xx, 334 pp. 20 Croft, William, Suzanne Kemmer and Keith Denning (eds.): Studies in Typology and Diachrony. Papers presented to Joseph H. Greenberg on his 75th birthday. 1990. xxxiv, 243 pp. 19:2 Traugott, Elizabeth and Bernd Heine (eds.): Approaches to Grammaticalization. Volume 2: Types of grammatical markers. 1991. xii, 558 pp. 19:1 Traugott, Elizabeth and Bernd Heine (eds.): Approaches to Grammaticalization. Volume 1: Theoretical and methodological issues. 1991. xii, 360 pp. 18 Haiman, John and Sandra A. Thompson (eds.): Clause Combining in Grammar and Discourse. 1988. xiii, 428 pp. 17 Hammond, Michael, Edith Moravcsik and Jessica Wirth (eds.): Studies in Syntactic Typology. 1988. xiv, 380 pp. 16 Shibatani, Masayoshi (ed.): Passive and Voice. 1988. xi, 706 pp. 15 Austin, Peter (ed.): Complex Sentence Constructions in Australian Languages. 1988. vii, 289 pp. 14 Hinds, John, Shoichi Iwasaki and Senko K. Maynard (eds.): Perspectives on Topicalization. The case of Japanese WA. 1987. xi, 307 pp. 13 Never published. 12 Nedjalkov, Vladimir P. (ed.): Typology of Resultative Constructions. Translated from the original Russian edition (1983). Translation edited by Bernard Comrie. 1988. xx, 573 pp. 11 Tomlin, Russell S.: Coherence and Grounding in Discourse. Outcome of a Symposium, Eugene, Oregon, June 1984. 1987. viii, 512 pp. 10 Ransom, Evelyn N.: Complementation: its Meaning and Forms. 1986. xii, 226 pp. 9 Bybee, Joan: Morphology. A Study of the Relation between Meaning and Form. 1985. xii, 235 pp. 8 Slobin, Dan I. and Karl Zimmer (eds.): Studies in Turkish Linguistics. 1986. vi, 294 pp. 7 Craig, Colette G. (ed.): Noun Classes and Categorization. Proceedings of a symposium on categorization and noun classification, Eugene, Oregon, October 1983. 1986. vii, 481 pp. 6 Haiman, John (ed.): Iconicity in Syntax. Proceedings of a symposium on iconicity in syntax, Stanford, June 24–26, 1983. 1985. vi, 402 pp. 5 Rutherford, William E. (ed.): Language Universals and Second Language Acquisition. 1984. ix, 264 pp.
4
3 2 1
Chisholm, William, Louis T. Milic and John A.C. Greppin (eds.): Interrogativity. A colloquium on the grammar, typology and pragmatics of questions in seven diverse languages, Cleveland, Ohio, October 5th 1981-May 3rd 1982. 1984. v, 302 pp. Givón, T.: Topic Continuity in Discourse. A quantitative cross-language study. 1983. vi, 492 pp. Haiman, John and Pamela Munro (eds.): Switch Reference and Universal Grammar. Proceedings of a symposium on switch reference and universal grammar, Winnipeg, May 1981. 1983. xv, 337 pp. Hopper, Paul J. (ed.): Tense-Aspect. Between semantics & pragmatics. 1982. x, 350 pp.