No title

JOURNAL OF SEMANTICS AN INTERNATIONAL j OURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE M...

Author: Oxford University Press

8 downloads 395 Views 23MB Size Report

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

DOWNLOAD PDF

JOURNAL OF SEMANTICS

AN INTERNATIONAL j OURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE

MANAGING EDITOR: PETER BoscH (University of Osnabriick)

ASSOCIATE EDITORS: NICHOLAS AsHER (University of Texas, Austin)

Ros VAN DER SANDT (University of Nijmegen)

EDITORIAL BOA RD :

MANFRED BIERWISCH (MPG and Humboldt University Berlin) BRANIMIR BoGURAEV (IB M TJ Watson Research Center) KEITH BRowN (University of Essex) GENNARO CHIERCHIA (University of Milan) J\NN CoPESTAKE (University of Cambridge) OsTEN DAHL (University of Stockholm) KEES VAN DEEMTER (University of Brighton) PAuL DEKKER (University of Amsterdam) KuRT EBERLE (linguatec-es, H eidelberg) REGINE EcKARDT (University of Konstanz) CLAIRE GARDENT (CNRS, Nancy) BART GEURTS (University of Nijmegen) LAURENCE R. HoRN (Yale University) JoACHIM JAcoBs (University of Wuppertal)

PHILIP N. Jo HNSON-LAIRD (Princeton University) HANs KAMP (University of Stuttgart) GRAHAM KATZ (University of Tiibingen) SEBASTIAN L >

>

Figure

>

+

274 Optimal

Reflexivity

in Dutch

hearer will assume that the subject has the control over the described situation, as stated in default constraint C r . Cr.

The subject normally controls the action described by the predicate.

This constraint C I cuts to both sides, as the speaker uses it in economizing on his words and utterance effort in suppressing control information when the agent is in control, and the hearer derives from it the information that the controller is the agent, unless there is information to the contrary.This can be made explicit in the speaker constraint SCI and hearer constraint HC I below. If the agent is in control, no need to say so. If there is no control information, the controller is the agent.

The hearer would have to continue after (r7) with (2 5), if he wanted to find out whether Peter consented to being subjected to the action. (2 5) Vond Peter dat goed? [Found Peter that good?] Did Peter consent to it? To continue with (25) after (I9) would be markedly redundant, OF at best demonstrate ignorance ofthe meaning of let, for let already assigns the control role to its subject.The default constraint that the subject controls the action, SC I , is flouted by ( I 9), and it also adds expenses on the part of the speaker for having to use two predicates instead of just one.9 But (I9) saves the hearer from having to ask (25) and the speaker again from answering it.The semantic content of a light verb assigns another thematic role to its subject, besides the role it was assigned in situ by the non-fmite predicate. If a light verb is analysed as a relation between its external argument and an event interpreting the descriptive predicate in its complement, this inviolable constraint is captured as lexical meaning-postulate in (26). (26) If P is an element in the set of light verbs {HAVE, LET, ...}, then \fP \fQ \fx lf y lfe lfe'[ (P(x, e) & Q(e', ... , y, ...) & e = e' & y = x) => th (x) -I- th (y)]

MAKE,

GET,

The constraint SC2 that uttering fewer words is preferable for a speaker is another soft constraint, for it is meaningfully flouted in ( I 9). The hearer's 9

The constraint that fewer words are preferable to more is a very soft constraint on the economy

considerations of the producer (speaker), as effort in pronunciation is a relatively cheap commodiry. Of course, saying nothing would be most economical, but that limit is ordinarily eliminated by the overarching human desire to be understood by others.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

SCr. CCr.

Alice G. B. ter Meulen 275

corresponding constraint HC2 requires him to rely on default rules whenever different interpretations are possible, a meta-constraint that leads him to use CCr for the interpretation of (I 7)· SC2. HC2.

Using words costs effort (the more complex they are, the more they cost). Use default constraints when ambiguities arise.

(27) The barber used a razorblade. If ( I 8) is used with PP, the same information can be expressed by continuing with (28). (28) He used a razorblade. Perhaps simplistically counting words, (I 8) -PP + (27) is optimized for SC2, counting - I (drop 3 words, add 2), and (I 8) + PP + (28) is less optimal, counting +4 (add 3 words and add I ). But in the referential hierarchy (27) with a full referring expression is lower down or less efficient than (28) with only a pronoun, so (27) demands more mental processing effort of the hearer, i.e. making an inference and accommodating a presupposition and identifying the referents of both as the same. Referring to someone, i.e. updating the common ground with a new referent, lightens the hearer's load, when that referent is referred to again in subsequent discourse. For the hearer (I 8) + PP + (28) would be more optimal than ( r 8) - PP + (27), if the pronoun in (28) could not be interpreted otherwise. But, besides deictically, the pronoun can also be interpreted as coreferential with Peter, even though conflicting descriptive information would result.1 0 That prevents the hearer from chosing that interpretation, as the over1 0 The conflict arises as an incompatibility between the meaning of 'let' and the meaning of the descriptive predicates, whereas using a razorblade is a specification of the way in which Peter was cut.


Together with the referential hierarchy and the strong configurational Binding condition B (co-arguments of a predicate that are not reflexive marked must be interpreted as disjoint in reference), we derive that (I7) is optimized for a situation where the barber controls his cutting someone else, called Peter.In comparison, to describe the same situation (I8) without agentive PP gains over (I7) in SC2, and still supports SCr . The implicit indefinite argument controls the action, and requires a novel referent, so must be interpreted as disjoint. If ( I 8) without PP is continued with a definite description referring to the barber, whose presupposition is accommodated by resolving it with the inferred existence of the agent argument, the information that the cutter was a barber is recovered, as in (27).

276

Optimal Reflexivity in Dutch

arching constraint tells him to avoid contradictions, (HC3), assuming that the speaker wants to be consistent, (SC3). SC 3 HC3

Be consistent and coherent. Avoid contradictions.

6 CONSTRAIN IN G QUANT I F I CATION With a quantificational antecedent the economic advantages of SE reflexives become even more apparent in the Dutch data, when constrasted to the fully acceptable bound pronouns in their English translations. Negative quantifiers (left and right decreasing) in subject position cannot bind non-reflexive pronouns in non-finite complements of light verb constructions, cf (37). (29) a. Een barbier knipte iedere student. A barber cut every student. b. Iedere student werd geknipt (door een barbier). Every student was cut (by a barber).


Although (19) with PP may be most informative, for it entails both (17) and ( 1 8), in OT terms ( 1 8) is optimal, unless John controls the action. Pronouns with overtly asserted referential antecedents are optimal for the consumer to create coreference chains in discourse, in comparison to definite NPs with implicit arguments as antecedents. The latter are optimal for the speaker in economizing on words, i.e. someone who weighs SC2 more heavily. In section 6 below it will be argued, however, that quantified antecedents may reverse this OT ranking for creating discourse cohesion. In ranking ( 1 9) with (2o)-(24), (19) economizes in using SE instead of SELF in (2o), as morphologically words are more expensive for the producer (SC2), and ( 1 9) wins over (2o) in being higher in the referential hierarchy for the consumer. Also (19) entails (2o) and (21), and not vice versa, as in (2o) and (2 1) SELF or the accusative pronoun may be assigned the agentive role, but in (19) SE cannot support that role. Here OT assesses (19) as the winner, and any model theoretic account would consider ( I 9) to have the strongest truth functional content as well. Finally, in comparing the three focus constructions (22)-(24), (23) is optimal in using SELF, since SE is unacceptable here, with the agent argument in optional PP, referring disjointly as it is indefinite. In contrast, (22) and (24) use a pronoun, that may, but need not corefer with the subject, so they cannot entail (23).

Alice G. B. ter Meulen

277

(3o) Iedere student liet zich knippen (door een barbier). [Every student let SE cut (by a barber).] Every student let himself get cut (by a barber). (3 1) Iedere student liet een barbier hem knippen. [Every student let a barber him cut.] Every student let a barber cut him.

(32) De barbier gebruikte een scheermes. The barber used a razorblade. Of course, (32) also allows for a globally accommodated existential pre supposition, introducing one barber cutting all the students. But this would require considerably more processing effort from the consumer, as he would also have to force wide scope to the existentially quantified implicit argument to make it corefer with the accommodated presupposition introducing the barber as the one doing the cutting. From OT perspective the same information could much more economically be expressed using (33). (33) a. Een barbier (H*LL%) knipte iedere student. Hij gebruikte een scheermes. b. One barber cut every student. He used a razorblade.


The active (29a) and passive (29b) are parallel to (17) and ( 1 8), although the definite was replaced by an indefinite here to address scope issues. Scope interpretation is facilitated in (29) by intonation, where H*LL% (prosodic pitch contour H(igh) + stress L(ow) L(ow) boundary) on een barbier would change its meaning into the focus meaning the cardinal one barber. The OT advantage of (3o) is first of all seen in C2, when reassigning the control to the student argument with the light verb construction. Making the barber argument implicit in (3o), it is existentially quantified and must remain in the scope of the quantified subject, that now binds the SE reflexive. Accordingly, (3o) without PP is optimal, as it avoids the ambiguity between the asserted NPs, present in (29a) and in (29b) with PP. As no argument within the nuclear scope of the quantificational subject would bind pronouns in subsequent discourse, any NP anaphoric to it must accom modate its presupposition within that domain. If a definite description is interpreted as continuing reference to the implicit agent of the cutting, its presupposition may be locally accommodated within the nuclear scope of the non-finite predicate of the preceding sentence, as in (32). This is strongly supported by the descriptive content, as using a razorblade specifies the way in which the students were cut. The quantified subject of (3o) thus extends its subordination to the following sentence, even though ordinarily quantifiers have no dynamic effect in discourse and restrain their binding potential to sentence internal dependencies.

278

Optimal Reflexivity in Dutch

Asserting a referential NP optlmlZes discourse coreference, whereas accommodating presuppositions of anaphorically interpreted definite descriptions is relatively more costly. Since (33) is optimal for a wide scope barber, (3o) PP continued with (32) is optimal for a narrow scope interpretation of the barber. -

(34) * Ieclere student; liet een barbier hem; knippen. Every student let a barber him cut. Every student let a barber cut him.

(3 5) a. Een barbier knipte geen student. A barber cut no student. b. Geen student werd geknipt (door een barbier). No student was cut (by a barber). (36) Geen student liet zich knippen (door een barbier). [No student let SE cut (by a barber).] No student let himself get cut (by a barber). (37) *Geen student; liet een barbier hem; knippen. [No student let a barber him cut.] No student let a barber cut him. The referential hierarchy assigns (36) a higher value than (37), which matches the Dutch data better than the English ones, as the English translation of (36) takes a toll on word count (Cz) over (37). English, lacking SE reflexives, makes word count outweigh the referential hierarchy.


When a non-reflexive pronoun is used in Dutch as in (34), it is considerably harder, if not outright unacceptable, to interpret it as bound by the quantificational subject, than as coreferential with a proper name. This is in accordance with the predictions of the referential hierarchy. In (34) the pronoun is preferred to refer to a referent available in the situation of use or to one available from memory, as (3o) is already optimal in using a reflexive pronoun and making the agent of the predicate implicit, confining it to nuclear scope. In English translation of (34), however, the bound interpretation of the pronoun is certainly acceptable and even preferred over non-local interpretations. This is a sharp contrast with the economical advantages of the Dutch SE-reflexive. When the quantifier is negative, decreasing in both the denotation of the noun and the predicate, the reflexive is the only option in a light verb construction (36), as the negative quantifier in (37) does not bind the pronoun in Dutch, though the English translation is fine.

Alice G. B. ter Meulen 279

7

C O N CL U S I O N S

Acknowledgements This paper was first read at the conference 'The Optimization oflnterpretation', UiL-OTS, University of Utrecht, 5 January, 2000. I am very grateful to Helen de Hoop, Petra Hendriks, and Harriet de Swart for organizing this innovative program for natural language semanticists and to Helen especially for providing me afterwards with many helpful comments. Remaining misunderstandings and errors are entirely attributable to me. Received: 3 ! .05.2000 Final version received: o I .09.2ooo

ALICE G. B. TER MEULEN

Center for Language and Cognition, RUG P.O. Box 7 1 6 9700 AS Groningen The Netherlands [email protected]

REFERE NCES Burzio, Luigi ( 1 998), 'Anaphora and soft constraints', in P. Barbosa et a/. (eds),

Optimality and Competition in Syntax,

MIT Press, Cambridge, MA, 9 3 - 1 r 3· Dekker, Paul & Rooy, Robert van (2ooo), 'Optimality Theory and Game Theory:

some parallels', lecture notes OTS 5 / r /oo, ILLC, University of Amsterdam. Grimshaw, Jane (1 995), 'Projection, heads and optimality', Linguistic Inquiry 28: 3 7 3 -422. Hoop, Helen de & Swart, Harriet de (eds)


The results of this paper may still b e inconclusive, as it has only outlined what an OT approach to binding may have to offer as explanatory account of preferences in the description of a given situation. The primary goal has been limited to an analysis of some of the constraints at work for both producer and consumer of the information in encoding referential (in)dependencies, contrasting English to Dutch, with its pervasive use of light verbs and SE-reflexives. Such cognitive issues of information pro duction and processing have not been addressed within the more traditional natural language semantics based on model theoretic tools, even within the now current systems of dynamic semantic interpretation. The next step is to render these insights useful as constraints on information sharing in a wider semantic theory, where interests of language users play a role, sometimes a conflicting one, in determining how information should optimally be expressed in a given context. In such a theory of information sharing there is room for analysing how misinter:pretations and misunderstandings arise and may be resolved, a concern that has suffered long neglect due to the Fregeau adagio that they fall outside the scope of semantic theory.

280 Optimal Reflexivity in Dutch (I999), OTSz: Papers on Optimality The oretic Semantics, Utrecht Institute of Lin


guistics, University of Utrecht. Meulen, Alice ter (I 999 ), 'Binding by implicit arguments', Amsterdam Colloquium Pro ceedings, ILLC, University ofAmsterdam. Meulen, Alice ter (2oooa), 'How to tell events apart: light verbs, SE-reflexives and Dutch verbal morphology', to appear in Carol Tenny & James Pustejovski (eds), Events as Grammatical Objects, CSLI Publications, Cambridge University Press, Cambridge. Meulen, Alice ter (2ooob), 'On the economy of interpretation: semantic contraints on SE-reflexives in Dutch', in Eric Reuland et a/. (eds), Interface Strategies, Royal Acad emy of Sciences, Amsterdam (prepub lished as OTS working paper no. TL 98-004, UiL-OTS University of Utrecht). Philip, William ( I 999), 'Disjoint reference in the short passive', in B. Vaux & S.

Kuno (eds), Harvard Working Papers in Linguistics, Vol. 7, Harvard University, Cambridge, MA, 64-92. Reinhart, Tanya ( I 997), 'Strategies of ana phora resolution', MS, University of Utrecht. Reinhart, Tanya & Reuland, Eric (I 993), 'Reflexivity', Linguistic Inquiry, 24, 4, 6s7-72o. Rootyck, Johan & Wyngaerd, Guido Vanden (I 997), 'The self as other: a minimalist approach to zich and zichzelf in Dutch', NELS, 28, 3 S9-3 7 3 · Safir, Ken (I 992), 'Implied non-coreference and the pattern of anaphora', Linguistics and Philosophy I S , I , I - S 2. Salmon, Nathan (I 992), 'Reflections on reflexivity', Linguistics and Philosophy I S , I , S 3 -64. Zribi-Hertz, Anne ( I 989), 'Anaphor binding and narrative point of view: English reflexives in sentence and discourse', Language, 65, 69s-727.

journal ofSnnantics 17: 281-3 14

© Oxford Unive�ity Press 2000

The Partitive Constraint in Optimality Theory AR TO ANTTILA Boston University/National University of Singapore V I V I E NNE F O N G National University of Singapore

Abstract

INTR O D U C T I O N I.I

Preliminaries

Current work in Optimality Theoretic (OT) syntax and semantics has been concerned with two closely related questions. 1 ( I)

• •

OT SYNTAX. Given a semantic input, what is its optimal expression? OT SEMANTICS. Given a syntactic input, what is its optimal interpretation?

OT syntax takes the 'speaker's perspective': given a well-formed semantic input, the goal is to select the optimal syntactic expression for this input from among a set of competing candidate expressions (see e.g. Bresnan I 997; Aissen I 999; Bresnan, to appear). OT semantics takes the 'hearer's perspect ive': given a well-formed syntactic input, the goal is to select the optimal semantic interpretation for this input from among a set of competing candidate interpretations (see e.g. de Hoop & de Swart 1 999; Hendriks & de 1 The following abb reviations are used in the glosses: ELA: dative; INE: inessive; plural; 1 P: fJISt person.

PAR:

partitive;

Pt:


This paper discusses a case of syntax/semantics interaction of a characteristically optimality-theoretic kind. Finnish partitive constructions exhibit a case alternation that is partly semantically, partly syntactically driven. The crucial semantic condition that plays a role in case selection is quantitative determinacy, which replaces the definiteness condition familiar from the Partitive Constraint. The crucial syntactic condition is the Case-OCP, which prohibits the assignment of the same case to both the head and its sister. The syntactic and semantic constraints conflict, which leads to various kinds of outcomes, including free variation and ambiguity, as well as preferences in expression and preferences in interpretation. We develop an optimality-theoretic analysis of these facts based on partially ordered optimality-theoretic grammars. In such grammars, conflicts among semantic and syntactic constraints are resolved in terms of ranking. Partial ordering is crucial in deriving preferences in expression as well as interpretation, including blocking effects.

282

The Partitive Constraint in Optimality Theory

Hoop 1 999). A third alternative is bidirectional optimization (Blutner 1 999) where both types of optimization are carried out simultaneously. 2 We first recognize that the question both OT-syntax and OT-semantics attempt to answer is essentially nondirectional and can be stated as in (2) :

(2) What are the possible form/meaning relations?

(3) Variation and ambiguity

In this schematic example, the form F, is unambiguous whereas F2 has multiple meanings (i.e. is ambiguous). Conversely, the meaning M, has multiple expressions (i.e. is variable), whereas M2 is invariant.3 At a more subtle level, we find the phenomenon of PREFERRED INTERPRETATIONS and PREFERRED EXPRESSIONS which can be viewed as weighted versions of ambiguity and variation, respectively. Such preferences obviously presup pose a quantitative treatment of some kind. In this paper, we will propose an approach that extends to all four types of effects. Our goal is to explore a particular way of drawing nondirectional maps between forms and meanings. We will take the perspective of OT syntax as our point of departure and see how it extends to the following empirical phenomena that we take as the central problems of our inquiry: (4) The phenomena to be accounted for: o Variation, preferred forms o Ambiguity, preferred readings, semantic blocking 2 Implicit in all these approaches is the assumption that the input to OT-syntax is a well-formed semantic representation and the input to OT-semantics is a well-formed syntactic representation. In other words, there are two levels of representation: a purely syntactic level that checks the well formedness of syntactic representations and a purely semantic level that checks the well-formedness of semantic representations. The levels are independent of each other and come with their own well formedness conditions, perhaps stated as optimaliry-theoretic grammars. In this paper, the terms 'OT-syntax' and 'OT-semantics' are construed narrowly as referring to theories of form/meaning correspondence. 3 The fact that optionaliry is just the flip side of ambiguiry is also noted by Asudeh (1 999).


Let us briefly consider the possible types of mappings between meanings and forms. In addition to one-to-one mappings (one-meaning-one-form), we find one-to-many mappings from meaning to form, which we call VARIATION, and one-to-many mappings from form to meaning, which we call AMBIGUITY. This is presented schematically in (3):

Arto Anttila and Vivienne Fong • •

28 3

Uninterpretability: Forms that get no interpretation Ineffability: Meanings that cannot be expressed

Of course, any theory of syntax and semantics must account for form/ meaning relations in some way and our simple diagram is intended to be theory-neutral. However, Optimality Theory makes one very general claim about the nature of the form/meaning relation that distinguishes it from most alternative theories, in particular compositional theories of semantics: (s)

•

Thus, if Optimality Theory is correct, we would expect to find evidence for constraint conflict and conflict resolution of a particular kind and it is precisely the existence of such evidence that we will set out to demonstrate in this paper. I .2

The Partitive Constraint

The specific empirical issue that we will be concerned with is the well known semantic generalization dubbed the PARTITIVE CoNSTRAINT (Jackendoff 1 977; Selkirk 1 977; Barwise & Cooper 1 9 8 1 ; Ladusaw 1 982). The Partitive Constraint states that in partitive constructions such as some of the cats, the embedded NP must be definite.4 This point is made by the following examples: (6) a. some of the cats most of the cats three of my cats

b. *some of most cats *most of some cats *three of cats

Despite its sound core, the Partitive Constraint is beset with a number of well-known empirical problems that have led many to question its status as a semantic constraint. Among such problems are the unexpected well formedness of half of a cookie, one third of all rental apartments, and one of a 4 In our discussion, we follow Jackendoff (1977), Selkirk ( 1 977), Barwise & Cooper (198 1), and Ladusaw (1 982) in treating partitives as comprising an upstairs determiner and a downstairs/ embedded NP. For example, in a construction like the upstairs determiner is and the downstairs NP is This is different from the approach taken by Keenan & Stavi (1 986) and Chomsky (1 970), who treat as a determiner.

most,

most of the twenty liberal delegates, the twenty liberal delegates. most of the twenty


•

The constraints governing the form/meaning relation are conflicting and violable. The actually attested form/meaning relations are optimal solutions to these constraint conflicts resolved by means of ranking.

284 The Partitive Constraint in Optimality Theory

(7)

o o o

What are the constraints on the downstairs NP (the cats)? What are the constraints on the upstairs determiner (some)? How are the two related?

In this paper, we will address these questions by bringing in new evidence from a language where partitivity is expressed in terms of overt morphological case marking. In Finnish, the part-whole relation can be expressed by means of two distinct morphological cases on the downstairs NP: partitive (PAR) and elative (ELA). This case alternation occurs with a class of determiners that roughly correspond to 'entity determiners' in de Hoop's (I 997) sense, including kilo 'kilo' and kolmasosa 'one third'. There is another class of determiners, roughly de Hoop's (I 997) 'set determiners', which include muutama 'some', kaikki 'all', and cardinal numbers. These determi ners induce a different case alternation and will not be discussed here. The observation that de Hoop's semantic distinction between entity determiners versus set determiners roughly coincides with an overt morphosyntactic distinction in Finnish is striking, but cannot be pursued in this paper.6 At first blush, the difference between partitive and elative seems to be indeed (in)definiteness, as shown in (8) and (9). 5 According to Barwise & Cooper (I98 I), definite NPs are those that have a non-empty generator set. Among other things, this excludes universally quantified NPs such as (see also de Hoop I997). 6 Syntactically, entity determiners head the partitive construction (Hakulinen & Karlsson I 979; Vainikka I 99 3 ). This is evident from the fact that when the entire phrase is assigned a case, this case is morphologically realized on the determiner, while the sister NP retains its partitive case marking, e.g. 'kilo-INE apple-PL-PAR, in a kilo of apples'. In this paper, we will assume that both partitive NPs and elative NPs are sisters of the head and we continue to refer to them as 'downstairs NPs'. For further discussion, see Vainikka (I993) and Hakulinen & Karlsson (I 979).

all rental apartments

kilo-ssa omen-i-a


number of applicants where the embedded NPs are not definite.5 Such examples have given rise to various responses. For example, Reed (I 99 I) and Abbott (I 996) conclude that the Partitive Constraint is a pragmatic constraint at best, whereas de Hoop (I 997) maintains that there is a semantic generalization involved, but reformulates it as a kind of semantic agreement between the upstairs determiner and the downstairs NP: if the upstairs determiner quantifies over entities, the downstairs NP must denote an entity; if the upstairs determiner quantifies over a set of entities, the downstairs NP must denote a set of entities. Thus, for example, half of a cookie is correctly predicted to be well-formed given that half (of) quantifies over entities and a cookie denotes an entity. For a more detailed summary of the issues, see the articles in Hoeksema (I 996), and also de Hoop (I998) and Barker (I998). As this brief discussion has already shown, partitive constructions such as some of the cats present three closely related questions:

Arto Anttila and

Vivienne Fong 2 8 5

It might thus seem that the part1t1ve case would parallel the English PSEUDOPARTITIVES (Selkirk I 977) such as a number of cats and a kilo of butter, which allow embedded bare plurals and mass nouns prohibited in true partitives, and that the dative case would parallel the true partitives that are subject to the Partitive Constraint. This seems to be more or less what has been assumed in earlier work. For example, Chesterman (I 99 I) and Alho (I992) claim that the partitive is indefinite, and the elative definite. However, a corpus of actually occurring examples readily shows that the indefinite/definite distinction is a rather poor approximation of the facts. 7 Thus, we must first answer the simple descriptive question: what determines the choice of case in Finnish part-whole expressions? We make the following claims which will be discussed further in section 2.

( 1 0) A preview of the basic observations:

The choice of case depends on multiple constraints, some semantic, some syntactic. The most important semantic constraint will be referred to as QuANTITATIVE DETERMINACY, which differs from the definiteness condition in the original Partitive Constraint. The most important syntactic constraint is the CASE-OCP which bans adjacent identical morphological cases. These constraints interact in an optimality-theoretic fashion. • The meaning of the elative case is lexically fixed, signalling quantitative determinacy. The meaning of the partitive case is flexible and assigned by constraint competition. In a Jakobsonian (I 95 7) sense, elative is marked, partitive unmarked. Most of our examples are based on the Suomen Kuvalehti 1987 corpus which contains all the •

7

issues of this Finnish weekly magazine. The corpus is available at the Universiry of Helsinki Language Corpus Server at http:// . ling.helsinki.fi/uhlcs. When the authentic corpus examples were too long for quotation in full, we have taken the liberry of shonening and modifying them as appropriate.

1 987

www


(8) a. kilo voi-ta kilo butter-PAR 'a kilo of butter' b. kilo voi -sta kilo butter-ELA 'a kilo of the butter' (9) a. kilo munkke-j -a kilo donut-PL-PAR 'a kilo of donuts' b. kilo munke-i-sta kilo donut-PL-ELA 'a kilo of the donuts'

286

The Partitive Constraint in Optimality Theory o

The form/meaning relation is not always one-to-one; there is a substantial amount of ambiguity (one form, multiple meanings) and variation (one meaning, multiple forms), often with recognizable preferences.

2 THE PARTITIVE /ELAT IVE C H O I CE 2.1

The semantic facts

We will first examine the semantic conditions that play a role in the distribution of dative and partitive cases. Our goal is to show that the case distinction (dative/partitive) and the definiteness distinction are orthogonal. All four types of examples are found: indefinite partitives, definite partitives, indefinite datives, definite datives. First, we observe that sometimes dative and partitive do not seem to differ in meaning at all. In examples ( I I ) and ( 1 2), both partitive and dative may occur on a definite singular NP and the meaning difference is elusive. It seems to us that in contexts like ( I I ) and (12), where the determiner osa 'part' is combined with a definite singular NP, the partitive and the dative are simply interchangeable with no obvious difference in meaning. This is an instance of FREE VARIATION. (1 1) a. osa ta-ta kaupunki-a part this-PAR city-PAR 'part of this city'


To account for these observations, we construct an optimality-theoretic analysis that captures the Finnish pattern. These facts are theoretically important in several ways. First, they reveal the existence of conflicting and violable semantic and syntactic constraints whose interaction is amenable to ranking, thus providing evidence for Optimality Theory in the domain of semantics and syntax. Second, they reveal a Jakobsonian markedness opposition between the two cases, one marked (dative), the other unmarked (partitive), with the meaning of the unmarked case assigned by constraint competition. This confirms the status of the partitive as the unmarked case argued on completely independent grounds in the domain of the clause by Vainikka ( I 99 3) and Vainikka & Maling ( I 996). Finally, the Finnish facts show that patterns of ambiguity, preferred readings and semantic blocking lend themselves to an OT syntactic analysis, showing that this approach has the potential of providing answers to certain fundamental semantic questions.

Arto Anttila and Vivienne Fong 287

b. osa ta-sta kaupungi-sta part thiS-ELA city-ELA 'part of this city' (12) a. osa Eurooppa-a part Europe-PAR 'part of Europe' b. osa Euroopa-sta part Europe-ELA 'part of Europe'

( 1 3) litra viini -a litre wine-PAR 'a litre of wine' OR 'a litre of the wine' The example in ( I 3) is another demonstration that the partitive is perfectly compatible with both definite and indefinite readings. However, it must be duly noted that the default reading of ( I 3) is indefinite ('a litre of wine') and this too requires an explanation. This is an instance of PREFERRED INTERPRETATION.8 Third, we also find that the elative case may occur on both definite and indefinite NPs, as shown in (I4) and (I S),9 including universally quantified NPs like (r6). (I4) osa Euroopa-sta part Europe-ELA 'part of Europe' (rs) neljannes lehma-n ruho-sta fourth cow-GEN carcass-ELA 'one fourth of a cow's carcass' 8 Actual corpus examples of amb iguity are hard to find b ecause the context often disamb iguates. Our examples of amb iguity are constructed out of unamb iguous corpus examples. 9 The context of this example is unamb iguous: 'On the pampa, the ab undance of meat is such that if a piece as b ig as one fourth of a cow's carcass falls from a cartload of meat, no one pays any attention, least of all the driver.'


These examples are clearly problematic for the definiteness hypothesis. In addition, they also go against Leino's (1993) proposal that the partitive case denotes unlimited substance whereas the elative denotes a specific set/mass. Second, ( I 3) shows that the partitive sometimes yields two readings that can be translated as definite and indefinite. This is an instance of AMBIGUITY.


(I6) kolmannes kaiki-sta vuokra-asunno-i-sta all-ELA rental-apartment-PL-ELA third 'one third of all rental apartments'

( I 7) Kolmasosa munke-i-sta/ ??munkke-j-a on italialaisia. monk-PL-ELA/ ??monk-PL-PAR are Italians I /3 'One third of the monks are Italians.' (IS) 63.4% suomalais-i-sta/ ??suomalais-i-a lomailee heinakuussa. 63.4% Finn-PL-ELA/ ??Finn-PL-PAR make.holidays inJuly '63-4% of the Finns have their holidays in July.' ( I 9) Enemmisto suomalais-i -sta/ ??suomalais-i -a haluaa majority Finn-PL-ELA/ ??Finn-PL-PAR want Koiviston jatkavan. Koivisto to.continue 'The majority of Finns want Koivisto to continue (as a president).' While the stigmatized forms sound quite bad, the pattern does not seem totally ungrammatical. We have found a handful of authentic corpus examples, among them the following: (2o) Valtaosa-ssa valikoiv-i-a abortte-j -a on kysymys . . . majority-INE selective-PL-PAR abortion-PL-PAR is question . . . 'In the majority of selective abortions, the question is . . . '

Thus, partitive is at least marginally acceptable here. We take this to mean that both cases are allowed in this context, but dative is the better choice (for reasons to be discussed shortly), which is reflected in corpus frequencies. This is an instance of PREFERRED EXPRESSION. In sum, dative is preferred over partitive on plural NPs if the upstairs determiner is a fraction or a percentage. In contrast, if the downstairs NP is singular, both partitive and dative become possible again, with no obvious difference in meaning. Again, notice that the partitive case gets a definite interpretation here.


Partitive constructions where the downstairs NPs are not definite are among the well-known counterexamples to the definiteness hypothesis in English as well. For example, consider halfofa cookie and one third ofall rental apartments, which are perfectly well-formed in English. Finally, singular and plural NPs behave differently with respect to the choice of case. When the upstairs determiner is a fraction or a percentage and the downstairs NP is plural, dative is strongly preferred whereas the partitive is marginal and often ungrammatical, as shown in ( 1 7)-(I 9).


(2 1 ) Kolmasosa kaupungi-sta "' kaupunki-a paloi. city-PAR burned city-ELA "' 1 /3 'One third of the city burned.' (z z) Suurin osa valastietoude-sta "' valastietout-ta tulee greatest part whale.knowledge-ELA "' whale.knowledge-PAR comes Kanadasta. from.Canada 'Most of the knowledge concerning whales comes from Canada.' (2 3 ) Napoleon valloitti puolet Euroopa-sta "' Eurooppa-a. Napoleon conquered half Europe-ELA "' Europe-PAR 'Napoleon conquered half of Europe.'

DEFINITE

INDEFINITE PARTITIVE kilo voi-ta 'kilo of butter-PAR'

ELATIVE

osa Eurooppa-a 'part of Europe-PAR'

neljiinnes lehman ruho-sta osa Euroopa-sta 'one fourth of a cow's carcass-ELA' 'part of Europe-ELA'

We thus conclude that the definiteness versus indefiniteness distinction corresponds to the elative versus partitive choice only approximately, if at all. Despite earlier claims to the contrary, partitives can be definite, elatives indefinite. In addition, both ambiguity and free variation are found. However, it is equally clear that the choice is not random either, but that there is some semantic notion, perhaps related to definiteness, that plays a role in case selection. 2.2

The semantic constraint

We now proceed to make our semantic proposal. First, we suggest that the semantic constraint relevant for case selection in the partitive construction is not definiteness, but QuANTITATIVE DETERMINACY.10 Following a standard 1 0 The term 'quantitatively indeterminate' NPs is used by Kiparsky (1998), who notes that the partitive case is assigned to quantitatively indeterminate NPs, for example, indefinite b are plurals and mass nouns. Kiparsky's discussion focuses on NPs that are arguments of the verb (see also Krifka 1 992). In partitive constructions, however, our analysis b elow shows that the partitive case occurs with b oth quantitatively determinate and quantitatively indeterminate NPs.


The examples so far have shown that the case distinction (elative/ partitive) and the definiteness distinction (definite/indefinite) are ortho gonal. All four types of examples are found: indefinite partitives, definite partitives, indefinite elatives, definite elatives.


assumption in model-theoretic semantics, we assume that NPs denote families of sets (i.e. generalized quantifiers). We define a Q(uantitatively) D(eterminate) Noun Phrase as denoting a family of sets such that the intersection of each member of the family with the common noun denotation is of a fixed size, that is, {X I X � E 1\ I [N] n XI = n }. All other NPs are Q(uantitatively) l(ndeterminate ).1 1 Examples of QD NPs and Ql NPs are given in (z s): 1 2 (z s )

NPs

that are QI:

cats some cats most cats wine

In the examples discussed earlier, elative appears on NPs with interpreta tions like 'the butter', 'this city', 'a cow's carcass', 'all rental apartments', and 'the monks'. These NPs all fall within the class of QD NPs. On the other hand, partitive occurs on various types of NPs, definite (e.g. 'this city') as well as indefinite (e.g. 'wine'). Notice that while 'this city' is a QD NP, 'wine' is a Ql NP. Based on the distribution of elative and partitive cases discussed above, we now make the following descriptive generalization:

(z6) In Finnish partitive constructions, elative occurs on QD NPs whereas partitive occurs on both QD NPs and QI NPs.

Next, we note that determiners can also be partitioned into two groups using quantitative determinacy as a criterion: those that require QD downstairs NPs and those that do not. We make the standard assumption that determiners denote relations between sets of individuals. For example, most(A, B) is true if and only if lA n Bl > lA - Bl. We define QD-determiners as those determiners D for which the truth of D(A, B) crucially refers to the size of the entire set A, not only to the size of A n B. 1 1 Our definition of QD NPs looks similar to Verkuyl's Quantity of A):

(i) lA n Bl = m (where

m E N ) => [+SQA]

(Verkuyl

( 1 993) notion of [+ SQA] (Specified

1 993: 101)

For Verkuyl, [ + SQA] picks out a class o f NPs that give terminative aspect i n aspectual composition. We note that this class of NPs does not coincide with the QD NPs in our list in (2 s . For example, NPs like and give terminative aspect readings and so are [+ SQA]; however, by our definition, they are QI NPs. This is because Verkuyl makes the assumption that such NPs-which are conventionally defined in Generalized Quantifier Theoty as having an 'open end'-must have a contextually finite denotation when they occur in sentences that get interpreted as 'expressing temporal structure' (see Verkuyl 1 993: 1 03). We make no assumptions of this sort here. 1 2 As a reviewer points out, in the case of universally quantified NPs such as we must assume that is the cardinality of the common noun.

most cats

some cats

) (Most cats died/Some cats came in (#for

hours)),

all cats

n


that are QD: exactly three cats all cats a cat the cat NPs

Ano Anttila and Vivienne Fong 29 r

All other determiners are QI-determiners. For example, in order to evaluate 'one third of the cats are black' we need to compare the number of black cats to the total number of cats in the domain. This implies that the set of cats must have a Hxed size; in other words, the downstairs NP ('the cat') must be a QD NP. The determiner 'some' is different: to interpret 'some cats are black' it is enough to check whether lA n Bl > o and there is no need to know anything about the size of the set of cats. Thus, 'one third' is a QD-determiner whereas 'some' is a QI-determiner. More examples are given in (27).

(27) Types of determiners

Do not require QD NPs (D): some plenty kilo three

Given that some determiners require a quantitatively determinate downstairs NP, and given that dative implies quantitative determinacy, we might expect to see a dependency between the upstairs determiner and the case of the downstairs NP in Finnish. In particular, a QD determiner, which requires a QD downstairs NP, should occur with dative downstairs NPs. (28) shows the choice of case (partitive vs. dative) with I 4 determiners in a sample of I ,404 partitive constructions in the Suomen Kuvalehti 1987 corpus.

(28 ) Upstairs determiner and downstairs case (a) gramma hiukan litra (b) osa suun osa (c) suunn osa paaosa n:sosa valtaosa puolet prosentti kolmannes neljannes viidennes

'gram' 'a little' 'litre' part' 'great part' ' greatest part' ' main part' 'nth part' ' majority' 'half' 'per cent' 'one third' 'one fourth' 'one Hfth' '

PAR% I OO I OO I OO 4I IS II 7 6 4 2 I 0 0 0

ELA% 0 0 0 59 8s 89 93 94 96 98 99 I OO 1 00 1 00

# OF ToKENS IO IO I6 337 84 125 I4 69 48 I 87 418 s6 16 I4


Require QD NPs (Dqd): most fractions (one third) percentages (Jo%) superlatives (the tallest)


2. 3

The syntactic constraint

Up to this point, we have been pursuing a purely semantic explanation for the choice of case. We have proposed a semantic constraint that restricts the dative case to quantitatively determinate NPs. However, it turns out that this semantic constraint is violated under a particular syntactic condition stated in (29): (29) Elative is blocked from the downstairs NP if the entire partlttve construction bears the dative case. The same applies to partitive. 13 See also Kopgevskaja-Tamm (forthcoming) who presents a similar cline of quantifiers that take elative and/or partitive case-marked NPs. 14 The potentially interfering factors include, for example, whether the noun is mass or count, and, if the latter, singular or plural. Recall that singular NPs may take partitive even under a QD determiner, whereas plural NPs very rarely do.


Group (a) contains QI-determiners, group (c) QD-determiners. Group (b) contains the potentially ambiguous osa 'part' and suuri osa 'a great part'. While we will assume that they do not require a QD downstairs NP, it seems possible that these determiners have two readings: osa can mean either 'some' or 'proper part', and suuri osa either 'more than some great number' (the QI-reading) or 'more than half' (the QD-reading). 1 3 The distribution of determiners and cases is surprisingly clear-cut. The determiners gramma 'gram', hiukan 'a little', and litra 'litre' are all QI determiners. In the corpus, they occur with partitives only. At the other end of the scale, we find fractions that are QD-determiners. In the corpus, they occur with elatives only. While the raw numbers in (28) hide several potentially significant factors, they reveal that QI-determiners typically co occur with the partitive and QD-determiners typically co-occur with the elative.14 The numbers in (28) corroborate our expectation that QD determiners should occur with dative downstairs NPs. It is evident that the choice of case is a function of the semantics of the determiner. An alternative would be to stipulate that determiners like fractions and percentages subcategorize for the elative case, but this would not explain the semantic generalization. However, notice that the co-occurrence of QD determiners with elative NPs is only a strong tendency (see ( 17)-(2o) and (2 1)-(23)). An interesting question also arises with the variable cases in (c): why do QD-determiners sometimes take partitive NPs? In what follows, we will see that the responsibility lies at least partly with a syntactic constraint that overrides the semantic constraint on quantitative determinacy.

Arto Anttila and Vivienne Fong 29 3

We now illustrate this syntactic constraint. Like many other languages, Finnish has verbs that assign special semantic cases to their arguments. For example, the verb tulia 'become' assigns dative to its subject. (3o) Sointu-sta tuli munkki. Sointu-ELA became monk 'Sointu became a monk.'

(3 I) [Kolmasosa-sta [*mieh-i-sta/ mieh-i-a]) tuli munkkeja. One third-ELA *man-PL-ELA/ man-PL-PAR became monks 'One third of the men became monks.' (32) [Kolmasosa [mieh-i-sta/ ??mieh-i-a]] ryhtyi munkeiksi. One third man-PL-ELA/ ??man-PL-PAR chose.to.be monks 'One third of the men chose to be monks.' The same pattern arises with the partitive. The verb rakastaa 'love' takes a partitive object, as shown in (33). (3 3) Anders rakastaa Helsinki-a. Anders loves Helsinki-PAR 'Anders loves Helsinki.' In (3 4), the entire partitive construction 'this part of Helsinki' is assigned partitive case by the verb. Again, the partitive is blocked from occurring on the embedded NP 'Helsinki', even though it does emerge in free variation with the dative elsewhere, as shown by (3 5 ). (34) Anders rakastaa [ta-ta osa-a [*Helsinki-a/ Helsingi-sta]]. Anders loves this-PAR part-PAR *Helsinki-PAR/ Helsinki-ELA 'Anders loves this part of Helsinki.' Helsingi-sta]] rakennettiin I 8oo-luvulla. (3 s) [Tama osa [Helsinki-a This part Helsinki-PAR "' Helsinki-ELA was.built in. the. I 8oo's This part of Helsinki was built in the I 8oo's.' rv

The determiners kolmasosa 'one third' and osa 'part' are syntactically heads of the construction (see fn. 6). We suggest that these blocking facts follow from a general syntactic constraint we will call CASE-OCP, following T. Mohanan ( 1 994). The universal core of the OCP is the prohibition of adjacent identical elements. K. P. Mohanan (to appear) has proposed that the


Now, consider (3 1 ). Here, the entire partitive construction 'one third of the men' is assigned the dative case by the verb. Since this construction has a plural NP ('men') embedded under a QD-determiner ('one third'), we would expect the dative case to be strongly favoured. Strikingly, the dative case is categorically blocked. If we change the verb to one that does not assign dative case, the expected dative emerges, as shown by (32).


grammars of particular languages specify the participating elements and domains, yielding language-particular manifestations of this universal principle. We assume that in Finnish the language-particular manifestation of Case-OCP prohibits the same case from being assigned to both the head of an NP and its NP sister: (36) Case-OCP (T. Mohanan 1 994): *aa For Finnish: * (N casea (NP caseo ]] NP

2.4

Summary

The semantic notion of quantitative determinacy is crucial in determining the dative/partitive choice in Finnish partitive constructions. We have proposed that the dative case only occurs on quantitatively determinate NPs, whereas the partitive case may occur on both quantitatively determin ate and indeterminate NPs. Determiners that require their downstairs NPs to be QD prefer dative (see group (c) in (28)), especially if the NP is plural (see ( 1 7-20)). However, this semantic generalization is overriden by Case OCP, which is a syntactic constraint. In precisely this context, the partitive can fill in for the dative case (see (3 1 )).

3 AN OT ANALYSIS We have now identified two core factors that influence the choice of case in Finnish part-whole expressions. The question remains how exactly these factors interact. The basic intuition is that the two factors are not equally 15

A reviewer notes analogous examples such as and Whether Case-OCP effects are found in all languages, to what extent, and in what environments are questions beyond the scope of this paper.

lnstituts .

Peter's sister's house

das Haus des Direktors des


We note that Case-OCP effects only emerge in certain syntactic configurations, but not others. As we have defined it, Case-OCP applies to a head and its phrasal sister, but crucially not to a head and its determiners and premodifiers, e.g. ta-ta osa-a 'this-PAR part-PAR' in (34) or ta-ssa punaise-ssa talo-ssa 'this-INE red-rNE house-INE, in this red house', where the determiners and premodifiers agree in case with the noun.15 In our corpus, the Case-OCP is responsible for several occurrences of the unexpected partitive under a QD-determiner. The upshot is that the syntactic constraint against adjacent identical cases is stronger than the tendency to select dative under a QD-determiner.

Ano Anttila and Vivienne Fong 295

important, but differ in strength. This can be described as the following informal algorithm: (37) Rule r Use either PARtitive or ELATIVE. 'I/3 of the cat (sg)' ___, kolmasosa kissa-a"-'kissa-sta (PAR"-'ELA) Exception r.r . . . unless ELATIVE is required (fraction + plural NP). 'r/ 3 of the cats (pl)' ___, kolmasosa *kisso-j-a/ki.sso-i-sta (*PL-PAR/PL-ELA) Exception r.r.r . . . . unless ELATIVE is banned (the Case-OCP). 'out of I/ 3 of the cats (pl)' ___, kolmasosa-sta kisso -j-a/*kisso-i-sta (PL-PAR/ *pL-ELA)

3.1

Input, output, constraints

To get the analysis off the ground, we need some INPUT and some OUTPUT. Following Bresnan (to appear), we will assume that the input is the language-independent content expressed in terms of features drawn from the universal space of possible grammatical and lexical contrasts, and the output consists of language-specific lexical items that carry with them their own interpretation of that content. The relationship be�een the two is regulated by ranked and violable constraints. Thus, for example, the input [BE PRES 2 sG) can be expressed in English by the lexical item (art : BE PRES 2 sG) which expresses the input perfectly faithfully. However, since this lexical item is not present in many speakers' lexicons, the next best match (are : BE PRES ) fills in for it. This lexical item neutralizes number and person and for this reason is not perfectly faithful to the input. However, in most dialects, it is better than (am : BE PRES I sG) , another unfaithful candidate that does not realize input person and in addition introduces an unlicensed first person. In such dialects, we must rank our constraints in such a way that are will be selected over am as the optimal output. In addition to input and output, we need CONSTRAINTS. In Optimality Theory, there are two types of constraints that are in inherent conflict: (i) FAITHFULNESS constraints that govern the input-output relation and strive to


This sort of generalization is very naturally expressible in Optimality Theory which is made for generalizations of the type 'Do X only if Y' or 'Do X except when Y' (Prince & Smolensky 1 993). Put slightly differently, in OT it is easy to express regularities that are only approximately true provided that the violations arise from an attempt to satisfy a more important regularity, which itself may be only approximately true and violated under the pressure of an even more important regularity, and so on. This is precisely the notion of CONSTRAINT RANKING that lies at the heart of OT and this is what the informal algorithm in (37) is intended to express.


preserve all the input contrasts, and only those, in the output, and (ii) MARKEDNESS constraints that exert pressure towards unmarked output structures and tend to obliterate input contrasts. Based on the above discussion, we now propose the following constraints:

The two faithfulness constraints are in fact families of constraints. For our purposes, the important special cases will be the following: (4o)

o

o o o

MAx(N) Express input number. MAx(q) Express input quantitative determinacy. DEP(N) Do not express number not present in the input. DEP(Q) Do not express quantitative determinacy not present in the input.

We also need to specify the content of the lexical entries, in particular the case suffixes. This is done in (4 1). (4 1) Lexical entries: a. ELA = (qo] b. PAR = [ ] C. PL = [PL]

Case variable. No semantic content.

In the lexicon, the dative suffix is assigned the feature (qo]; the partitive suffix is left unspecified. This is instrumental in accounting for the generalization that dative occurs with QD NPs, whereas partitive occurs with both QD NPs and QI NPs. In our analysis, partitive is the unmarked case whose meaning arises out of constraint interaction (for example, with reference to syntax). Consequently, partitive can fill in for the dative case under circumstances where the dative is for some reason blocked or dispreferred. What we see here is a classically Jakobsonian markedness opposition where the marked element 'signals A', and the unmarked element is a 'non-signal of A', but in neutralization contexts the unmarked element may also be compatible with 'signals A' (Jakobson 1 957). Thus, the ultimately appropriate characterization of the partitive is as an unmarked (unspecified) category. Finally, the plural suffix is assigned the feature [PL]. Since Finnish has no overt singular morpheme, none will be assumed in the analysis.


(3 8) Markedness constraints (hold of the output): o CASE-OCP. The same case cannot be assigned to both the head N and its sister NP (* [N casea (NP casea ]] NP)· o Q-AGR. A QD determiner must co-occur with a QD downstairs NP. o *ExPRESS. Do not express input features. (39) Faithfulness constraints (hold of the relation between input and output): o MAx: Express input features. o DEP: Do not express features not present in the input.


3 .2

Ranking

We now turn to four examples in order to figure out the relative ranking of the constraints. First, consider an example of free variation. Both kolmasosa kissa-a and kolmasosa kissa-sta mean 'one third of the cat (sg)' (e.g. 'One third of the cat is black'). (42) Free variation: 'one third of the cat (sg)'

____, ____,

kolmasosa kissa-a (PAR) kolmasosa kissa-sta (ELA)

(43) 'one third of the cat' Dqa N(SG,QD] kolmasosa kissa-a[ ]

a. b.

=>

MAx(q) *!

*

kolmasosa kissa-sta[qn] *EXPR

a. b.

=>

MAx(q) *

kolmasosa kissa-a[ ] kolmasosa kissa-sta[qn]

*:EXPR

*!

Note that kolmasosa 'one third' requires a QD downstairs NP and kissa 'cat' is a quantitatively determinate singular NP. The constraint Q -AGR will be satisfied in both cases and is thus irrelevant. However, not all variation is completely free. The next example differs 16

In all our tableaux below, we give only the serious contenders-the partitive and dative candidates-so as to demonstrate the relative ranking of the crucial constraints. We assume that all other possible candidates are ruled out by independent considerations. For example, a downstairs NP with no case marking would violate a high-ranking constraint that requires NPs to have case.


In order to handle vanatwn in OT, we will assume that optimality theoretic grammars are not restricted to total orders, but may also be genuine PARTIAL ORDERS (Anttila 1 997b). In other words, we will assume that ranking is an irreflexive, asymmetric, and transitive relation, but not necessarily connected. Now, consider the violations incurred by the two expressions in (42). In terms of markedness (*ExPREss), partitive is always better than elative because it is simpler: it does not express any input features. However, in terms of faithfulness (MAx(q)), elative is better than partitive if the input contains a feature QD that needs to be expressed, as is the case here. We can now capture the observed optionality by leaving the two constraints mutually unranked. This yields two totally ordered tableaux, shown in (43): partitive (kolmasosa kissa-a) wins under one ranking, elative (kolmasosa kissa-sta) under the other. In other words, the grammar predicts optionality.16


(45) 'one third of the cats' a. b.

=}

a. b.

=}

a. b.

=}

a. b.

a. b. a. b.

=}

=}

=}

Dqd N(PL, QD] kolmasosa kisso-j[PL]-a[ ] kolmasosa kisso-i[PL]-sta[QD1 kolmasosa kisso-j [PL]-a[ ] kolmasosa kisso-i[PL]-sta[QD1 kolmasosa kisso-j[PL]-a[ 1 kolmasosa kisso-i[PL]-sta[QD1 kolmasosa kisso-j [PL]-a[ ] kolmasosa kisso-i[PL]-sta[QD] kolmasosa kisso-j[PL]-a[ 1 kolmasosa kisso-i[PL]-sta[QD1 kolmasosa kisso-j(PL] -a[ J kolmasosa kisso-i[PL]-sta[QD]

Q-AGR *!

MAx(q) *

*ExPR * **

Q-AGR *!

*EXPR * ** *Q-AGR *

MAx(Ql *

MAx(q) *!

*EXPR * **

Q-AGR *

*ExPR *

Q-AGR *

MAx(Q) *

** ! *EXPR * ** !

MAx(q) *

Q-AGR *

MAx(Q) *!

*ExPR * **


minimally from the previous one: the downstairs NP is plural instead of singular. In this case, both elative and partitive are found, but partitive is marginal. In our corpus, we have only found a handful of partitives in this environment, among them (zo), showing that partitive is possible, but dispreferred. In contrast, elative is extremely common. (44) Marginal variation, with a preferred expression: 'one third of the cats' ___., ??kolmasosa kisso-j-a (PL, PAR) ___., kolmasosa kisso-i-sta (PL,ELA) If we compare (42) and (44), the more marginal status of the partitive in (44) is evidently connected to the fact that the NP is plural. Recall that the determiner kolmasosa 'one third' requires a QD downstairs NP (Q-AGR). While the singular NP kissa 'cat' is by its very nature QD, this is not the case with the plural NP. The elative case contributes the feature [QD], making the plural NP quantitatively determinate. The question then arises: why is partitive plural allowed at all? It seems that Q-AGR gives rise to a strong tendency, but only a tendency. The solution is straightforward: we simply add the constraint Q-AGR into the partial order which already contains two other constraints, MAx(q) and *ExPRESS, but we still do not posit any rankings. This results in six totally ordered tableaux, shown in (45):


We assume the following QUANTITATIVE INTERPRETATION for partially ordered OT grammars {Anttila 1 997b): (46) QUANTITATIVE INTERPRETATION {VARIATION): (a) A candidate is predicted by the grammar iff it w1ns m some tableau in the partial order. (b) If a candidate wins in n tableaux and t is the total number of tableaux in the partial order, then the candidate's probability of occurrence is n /t.

{47) Invariant pattern: 'out of one third of the cats'

____, ____,

kolmasosa-sta {ELA) kisso-j-a {PL, PAR) *kolmasosa-sta {ELA) kisso-i-sta {PL, ELA)

This reversal in judgements is due to Case-OCP that strictly dominates the three constraints discussed so far, as shown in tableau (48). Since Q-AGR, MAx{q) and *ExPR are mutually unranked, (48) actually corresponds to six tableaux. However, the ranking of these three constraints is irrelevant to the outcome: any of the six rankings will yield the same result because the higher-ranking Case-OCP will always pick (48a) as the winner. 17 Once all the constraints are in place, we will find that the predicted probabilities are 1 /4 for the partitive and 3/4 for the dative in contexts where the OCP is irrelevant. A reviewer points out that this may still not be a strong enough bias given the heavily marked status of ??koltiUlSosa kisso -j-a 'one third of cats'. This may well be so and more work is clearly needed. There are at least two ways to put such quantitative predictions to a serious empirical test: large naturally occurring corpora and controlled elicitation experiments. For precedents in phonology, see e.g. Anttila ( 1 997a), Anttila & Cho ( 1998), Ringen & Heinamaki (1 999), Hayes (to appear) and Boersma & Hayes ( 1 999).


In this example, the grammar consists of three constraints, but so far no rankings. Thus, the total number of tableaux t is 3 ! = 6. We find that ELA wins in 4/6 (= 2/3) of the tableaux and PAR in 2/6 (= 1/3) of the tableaux. The present grammar thus correctly predicts that elative is preferred over partitive if the upstairs determiner requires a QD downstairs NP and the NP is plural (contrast (43) and (45)). 17 Note that if some constraints were ranked with respect to some other constraints, as is often the case, t would be smaller than 6. If all constraints were ranked with respect to all other constraints, t would be 1 . This is the familiar case where a grammar equals a single tableau {total order). The third example (47) is different from the second in one respect: here the entire NP bears the elative case. Consequently, the preferences are drastically reversed. While in (44) partitive was strongly disfavored, in (47) it is the only possible option. This is our first example of an invariant pattern.

300 The Partitive Constraint

in

Optimality Theory

(48) 'out of one third of the cats' a. b.

=}

D4J NfPL, QD] kolmasosa-staela kisso-j[PL]-af loar *kolmasosa-sta,/a kisso-ifPL] -sta[Qo],Ia

OCP

Q-AGR MAx{Q) *ExPR * * * **

*!

(49) Anders rakastaa [ta-ta osa-a [*Helsinki-a/ Helsingi-sta]]. Anders loves this-PAR part-PAR *Helsinki-PAR/ Helsinki-ELA Anders loves this part of Helsinki. (so) 'this part (par.) of Helsinki' a.

b.

=}

D NfSG, QD] *osa-a[ ]par Helsinki-a[ ]par osa-af loar Helsingi-sta[QD]ela

OCP *!

Q-AGR MAx{Q) *EXPR * *

However, surface violations of the Case-OCP are found. Consider the following example: (s r) Etsi-n kilo-a [voi-ta/ *voi-sta]. search- r P kilo-PAR butter-PAR/ *butter-ELA I'm looking for a kilo of butter. In the context of OT, this immediately brings up the question: under what circumstances are such violations allowed? Put slightly differently, what is the higher-ranking constraint that forces such violations? We find that Case-OCP-violations are allowed precisely when the embedded NP is QI. This immediately reveals the solution: since the embedded NP is QI, using dative would express a meaning that is not present in the input, in violation of the constraint DEP(Q), which must thus rank above the Case OCP (see (5 2)). 18 Another possible way of satisfying Case-OCP would be to suppress the case of the head instead of the downstairs NP. We are here assuming a high-ranking faithfulness constraint that maintains the case of the head at the expense of its complement. The possibility of not having case at all is yet another way of satisfying Case-OCP. Again, we assume that this is ruled out by another high-ranking constraint requiring that NPs have case.


This example demonstrates a simple point: not only can partitive occur under a QD-determiner on a plural NP; it must occur in this environment if the alternative is a Case-OCP violation. In other words, the partitive case can fill in for the dative in syntactically adverse circumstances, revealing its unmarked nature. 1 8 Our final example is again minimally different from the previous one. Recall that the Case-OCP not only applies to datives (*ELA-ELA), but also to partitives (*PAR-PAR), as (49) shows:


(52) 'a kilo (par.) of butter'

a. b.

=>

D N[QI] kilo-a[ ]par voi -ta[ ]par *kilo-a[ ]par voi-sta[Qo],/a

DEP(Q)

OCP

* *!

Q-AGR MAx(Q) *ExPR

* *

*

(s 3) Rankings for Finnish: a. DEP(Q) » OCP b. OCP » Q-AGR c. OCP » MAx(Q) d. OCP » *ExPR e. MAx(N) » *EXPR 4 C O N SE Q UENCES 4·I

Variation

So far, we have illustrated our analysis with four special cases. We will now explore its consequences in more general terms. The purpose of our analysis is to answer the question 'Given a semantic input, what is its optimal expression?' Our grammar is designed to answer this question by establish ing the correct meaning/form relations. Assuming that grammars are partial orders, the following possible situations arise:20 (54)

• •

o

A given meaning has one expression (one meaning, one form). A given meaning has several expressions, possibly with quantitative preferences (variation, preferred expressions). A given meaning has no expression (ineffability).

1 9 We have not discussed the last ranking in the list: MAx{N) » *ExPR. This ranking is needed to express the fact that, in general, number is not neutralized under markedness pressure. For example, ' I IJ of the cats' does not get realized as kolmasosa kissa-sta ' I /3 cat-EI.A', but kolmasosa kisso-i-sta ' I I J cat-PL-ELA', with a n overt plural marker, even though the first realization i s clearly better in terms of markedness. 20 We will discuss INEFFABILITY in section 4· 3


The analysis correctly predicts that OCP-violations should only be found with the unmarked partitive for the following reason. An OCP-violation of type *ELA-ELA can be repaired by deleting one of the elatives: this is guaranteed by the ranking OCP » MAx(Q). An OCP-violation of type *PAR-PAR cannot be repaired because this will necessarily involve inserting a marked elative that is categorically prohibited by the ranking DEP(Q) » OCP. We conclude by summarizing the rankings for Finnish established so far. 1 9


The diagram in (5 5) is an example of a form/meaning map generated by a partially ordered grammar. (5 5) Variation in a partially ordered grammar

(56) For all semantically distinct NP-types; for all semantically distinct determiner types; for all total rankings subsumed by (5 3); generate the optimal expression(s). Since the number of different NP-types, determiner types and total rankings is reasonably small, taking these steps is not too difficult. First, we will consider the following five types of noun phrases to be semantically well-formed inputs: 22 (57) PossiBLE INPUT NP I . N[QI) 2. N[QD] 3· N[SG,QD] 4· N[PL, QI] 5· N[PL, QD)

EXAMPLE 'milk', 'some milk' 'the milk' 'a cat', 'the cat', 'this cat' 'cats', 'some cats', 'most cats' 'the cats', 'exactly three cats', 'all cats'

Second, we need to consider two types of determiners: those that require 21 In a complete analysis, we must also consider the OCP effect which adds another three-way choice: (i) The head noun is elative; (ii) the head noun is partitive; (iii) the head noun is neither of the two. We leave this step out to simplify exposition. 22 We will not consider *N[SG,QI] and *N[SG] well-formed, but instead assume that singulariry implies quantitative determinacy. We also assume that NPs are exhaustively classified as either QD or QI, thus *N[PL] is not welt-formed.


Each individual tableau within the partial order is depicted as an arrow. For each Mi, the number of outbound arrows is constant; this is the total number t of tableaux in the partial order. If all t arrows point to one particular form, there is no variation. If the arrows are split among several different forms, there is variation. The number of arrows pointing at a given form is proportional to the form's probability of occurrence. In order to see that the analysis really works, we must ensure that the grammar works correctly in all cases, no matter how the input is chosen (RicHNESS OF THE BAsE). What we must do is find the optimal syntactic output for all semantically well-formed inputs. This means we must take the following steps: 2 1


(s 8) Predicted meaning/form mappings (OCP ignored): INPUT a.

D + N[QI] D + N[PL, QI] b. D + N[QD] D + N[SG,QD] D + N[PL, QD] Dqd + N[QD] Dqd + N[SG,QD] c. Dqd + N[PL, QD]

EXAMPLE 'a litre of milk' 'a kilo of apples' 'a litre of the milk' 'a part of a/the city' 'a kilo of the apples' 'one third of the milk' 'one third of a/the cat' 'one third of the cats'

OUTPUT PAR PAR PAR "-' ELA PAR "-' ELA PAR "-' ELA PAR "-' ELA PAR"-'ELA (PAR "-')ELA

PREFERENCES PAR categorical PAR categorical 5 / ! 2 "' 7/ I 2 s/r2 7/!2 7/I2 s/I2 5 / 1 2 "' 7/ I 2 5 / I 2 "' 7/ I 2 I /4 "' 3/4 rv rv

We make the following observations. First, 'a litre of milk' {D + N[QI]) and 'a kilo of apples' (D + N[PL, QI]) are predicted to exclusively take the partitive case, which is correct. Second, we predict that a plural noun under a QD-determiner (e.g. 'one third of the cats' (Dqd + N[PL, QD]) should strongly prefer elative, which is also correct. All other cases are predicted to fall somewhere in between: in such cases both dative and partitive are predicted to be possible, elative slightly preferred. The obvious next step would be to match the quantitative predictions of the model with the actual corpus frequencies. However, this seems premature for the following reason: while the predicted statistical tend encies do emerge in the corpus, we know for a fact that there are lexical differences among the statistical patterning of individual determiners; yet as it stands, the model simply treats determiners as two distinct groups: those that require QD downstairs NPs and those that do not. However, litra 'litre', gramma 'gram' and hiukan 'a little' are strongly biased towards the partitive, whereas suuri osa 'great part' and osa 'part' are much less so. This is consistent with the observation of Koptjevskaja-Tamm (forthcoming) that Finnish determiners seem to form a continuum with respect to case selection. Before matching the predictions with the actual frequencies, we must incorporate these subtler determiner-specific distinctions in the


a QD downstairs NP and those that do not. We will not consider combinations of a QD-deterrniner and a QI downstairs NP, e.g. #one third of water, which we take to be semantically ill-formed. This leaves us with eight types of possible determiner-noun phrase combinations. Finally, we must check how these eight semantically well-formed inputs fare under all the total rankings subsumed by the ranked pairs in (s 3). There are 1 68 such rankings. This amounts to checking 1 ,344 tableaux in all. The results are spelled out in (s 8). The main observation is that our partially ordered OT grammar predicts both categorical and quantitative patterns. Three types of outputs are predicted: (a) invariant partitive; (b) variation with a slight preference for the dative; (c) variation with strong preference for the dative.

304 The Partitive Constraint

in

Optimality Theory

model. The two obvious possibilities are: (i) to make the semantic/syntactic analysis more fine-grained by additional constraints; (ii) to assume that different determiners subscribe to slightly different grammars within the partial order. We will leave this for future work. We have now answered the question 'Given a meaning, what is its optimal expression?' for all the possible inputs in our domain. The result was sometimes one-meaning-one-form, sometimes one-meaning-multiple forms (i.e. variation), with certain quantitative preferences. 4.2

Ambiguity

(59) Ambiguity in a partially ordered grammar

Again, in order to make sure that our grammar yields the correct semantic interpretation for all possible expressions, we must take the following steps: (6o) For all possible NP expressions; for all possible determiner expressions; for all total rankings subsumed by (5 3); retrieve the semantic input(s) for which the combined expression is the optimal output. The total number of output expressions we need to consider is 16. This is the result of combining two cases (ELA vs. PAR), two numbers (PL vs. non plural), two types of determiners (Dqd vs. D) and two types of nouns (mass 23 Zeevat (1 999) entertains a similar proposal, and suggests that optimality syntax is already a sound proposal for the architecture for optimal semantics. Later in his paper, Zeevat revokes this proposal in favor of Blumerian bidirectionality (Blumer 1 999).


We now turn to the OT -semantic question: 'Given an expression, what is its optimal interpretation?' If we are able to predict variation and preferences in expression, the obvious question is whether we are also able to predict ambiguity and preferences in interpretation. In terms of our diagram, the answer seems simple enough: all we need to do is reverse the direction of the arrows. Instead of taking the OT -semantics perspective (see section I . I ), we simply retrace our steps through the OT -syntactic tableaux. Ambiguity is a situation where one form can be traced back to more than one semantic input.23


vs. count). The results are spelled out in (61)-(63). We will ignore the Case OCP for now. (61) Ambiguity (2 expressions): EXAMPLE

litra maito-a kilo omen-i-a

INPUT D + N[QI] D + N[QD) D + N[PL, QI) D + N[PL, QD)

GLOSS

'a 'a 'a 'a

litre of milk' "' litre of the milk' kilo of apples' "' kilo of the apples'

(62) One form, one meaning ( 10 expressions): EXAMPLE

INPUT

kolmasosa maito-a Dqd + N[QD) kolmasosa maido-sta Dqd + N[QD) Dqd + N[SG,QD) kolmasosa omena-a kolmasosa omena-sta Dqd + N(SG,QD) kolmasosa omen-i-a Dqd + N[PL, QD) kolmasosa omen-i-sta Dqd + N[PL, QD) litra maido-sta D + N[QD) D + N[SG,QD) hiukan omena-a hiukan omena-sta D + N(SG,QD) D + N[PL, QD) kilo omen-i-sta

Gwss ' rf3 of the milk' 'rf3 of the milk' ' r f3 of an/the apple' ' r f3 of an/the apple' '1 I 3 of the apples' ' rf3 of the apples' 'a litre of the milk' 'a bit of an/the apple' 'a bit of an/the apple' 'a kilo of the apples'

(63) Meaningless syntax (4 expressions): OUTPUT Dqd + Nm-PL-PAR Dqd + Nm-PL-ELA D + Nm-PL-PAR D + Nm-PL-ELA

EXAMPLE

kolmasosa maito-j-a kolmasosa maido-i-sta litra maito-j-a litra maido-i-sta

INPUT

GLOSS

(meaningless (meaningless (meaningless (meaningless

syntax) syntax) syntax) syntax)

By taking the converse of the meaning/form relation we have now obtained all the possible interpretations for all the possible forms in our domain. In some cases, one form is predicted to have several meanings (ambiguity), in other cases we have one-form-one-meaning, and finally there are four expressions that are assigned no meaning at all (uninterpret ability). This is because there is no input/ranking combination such that these forms would be selected as optimal. These expressions are indeed peculiar: they consist of mass nouns with number morphology. From the point of view of our analysis, they are just pieces of uninterpretable syntax.24 The final question is whether we can go beyond ambiguity and predict preferences in interpretation. In other words, in contexts where the 2'

In fact, these strings do get an interpretation, but crucially one that implies an input where the noun has been converted to a count noun. Thus, the most narural interpretation of koltniiSosa maido-i-sta is 'one third of the milk cartons', implying that the mass noun maito 'milk' is treated as a count noun. Similarly, the string hiukan omena-a 'a bit of an/the apple' gets an additional reading 'a bit of apple' implying that the count noun omena 'apple' is being treated as a mass noun. mass


OUTPUT Dqd + Nm-PAR Dqd + Nm-ELA Dqd + N,-PAR Dqd + N,-ELA Dqd + N,-PL-PAR Dqd + N,-PL-ELA D + Nm-ELA D + N,-PAR D + N,-ELA D + N,-PL-ELA

306

The Partitive Constraint

in

Optimality Theory

(64) Ambiguity, no blocking

'fish-stuff' �

)

'fish'

� fish (6s) Ambiguity, partial blocking

'tree-stuff' � wood

/

)

'tree'

� tree

The question now is how to capture such blocking facts, partial as well as total. Given the theory of variation outlined above, the obvious solution is to apply the quantitative interpretation of partially ordered grammars in (46) to ambiguity mutatis mutandis. This would imply something like the following definition: 25 From the Oxford English Dictionary (OEDz CDROM, 1 . 1 1 ), we find the following used to denote ' [T]he substance of the trunk and boughs of a tree', in other words, examples of 'tree-stufF: c 1 440 Partonope 407 A 1 896 Kipling 'Seven Seas, Sea-Wife iv', To [a ship].

on

tree

ride the horse of tree

bry�e ofstone and not of tree.

vmion


OT-syntactic grammar predicts ambiguity, can we infer the preferred reading from a map like (59)? To take a concrete example, the fact that litra viini-i:i is ambiguous and can mean either 'a litre of wine' or 'a litre of the wine' is predicted by our OT -syntactic grammar because this expression wins under two distinct semantic inputs. But can we also derive the rather strong preference for the interpretation 'a litre of wine'? The fact that preferred interpretation may be influenced by the existence of alternative expressions for the same meaning is well known. Such effects are traditionally subsumed under the general notion of BLOCKING (see Aronoff 1 976 and Kiparsky 1982b; for an OT-semantics perspective, see Blutner 1 999). Blocking may be either partial or total (see e.g. Briscoe et al. 1 995; Copestake & Briscoe 1 995). A case in point is the well-known phenomenon of 'conceptual grinding' (see e.g. Pelletier & Schubert 1 989) whereby a count noun acquires a mass noun reading, for example, This is a fish (count) versus We had fish (mass) for dinner. However, the existence of a specialized mass noun blocks the grinding mechanism, suppressing the potential mass noun interpretation of a count noun. Thus, for example, the oddity of ?This table is made of tree is due to the existence of the lexical item wood.25 We illustrate this situation in terms of the following hypothetical meaning/form maps:


(66) QuANTITATIVE INTERPRETATION (AMBIGUITY): (a) A form F can be interpreted as M iff there is at least one arrow from F to M. (b) If n is the number of arrows from F to M and t' is the total number of arrows out of F, then the probability that F is interpreted as M is n /t'.

l(

' milk[QI]

maito-a (PAR)

' milk[QD\'

�

=ido-> [: v u]': =

(8) a. [u: LKu, • [: Bu] ] b. [: • [LKu, Bu] ] Now the first reading is stronger than the second, and the IP makes the same prediction as the BP. Note that it is irrelevant to this point how the uniqueness implication arises; we only need to assume that it does. Naturally, it may be doubted that definite descriptions imply uniqueness, but there is no consensus in this matter as far as I know (although it is generally accepted that definite descriptions do not imply uniqueness in the same way as they imply existence). Nor is there agreement on whether presuppositions in general carry uniqueness implications or not, simply


both principles are 'soft' constraints, which only apply ceteris paribus. If a given reading is unacceptable for pragmatic reasons, for example because it is contradictory or just implausible, the IP and the BP only differ about how the next best reading is determined, and therefore the choice between the IP and the BP has no bearing on Blutner's and Zeevat's claim that presupposition projection should be treated within the framework of optimality theory. Largely for accidental reasons, the empirical differences between the BP and the IP are not as pronounced as one might think. To explain this, let us begin with presuppositions triggered within the scope of a negative expression, such as (3), for example. Adopting the DRT framework, and assuming that the current discourse representation lacks an entry for the king of France, this sentence gives rise to two possible readings:

322 Buoyancy and Strength

because this issue has not been addressed in the literature. But I do not see why someone who insists that definite descriptions imply uniqueness could not consistently extend his claim to presuppositions in general.5 So the upshot of the foregoing observations is that negated sentences will not help us arbitrate between the BP and the IP, although prima facie it seemed likely that they would. And this is a pattern that repeats itself with other scope-bearing expressions, as we will presently see. Consider modal contexts, for example: (9) Perhaps I'll have lunch with the king of France today.

(1o) a. [u: Ku, o[: Lu] ] b. [: o[u: Ku, Lu] ] Key: Kx: x is the king of France; Lx: the speaker will have lunch with x today

Again, the first reading is preferred, and the BP predicts this without further ado,6 while the IP does not: if we feed ( I Oa, b) into a run-of-the-mill semantics for modal logic, neither will be stronger than the other, so the IP will not decide between these two interpretations. If, however, a less standard but arguably more realistic semantics is adopted, the IP may be brought back in line with the BP. For example, if (1oa) is interpreted along the lines suggested by Stalnaker (1 968) and Lewis (1 973), then we obtain a construal that may be paraphrased as follows: 'There is a king of France, call him a, and there is at least one world which is maximally similar to the actual world and in which the speaker will have lunch with a today.' But if a is the king of France in the actual world, and a given world w is maximally similar to the actual world, then presumably a will be king of 'France in w, also. On such an account of modality, (1 0a) will be synony111ous with:

( n ) [u: Ku, o[u: Ku, Lu] ] 5 The idea that factive verbs like know and regret or transition verbs like begin and stop have uniqueness presuppositions may not strike one as manifestly true, but it is less weird than it seems. For instance, if 'Fred knows that I + I = 2" presupposes that I + I = 2, the uniqueness presupposi tion would be that there is only one proposition to the effect that I + I 2, which is correct, I take it {and not just because this happens to be a mathematical truth). 6 Well, almost; it must be assumed that the main DRS in (1 0a, b) is accessible from the embedded DRS. I will suppose without further argument that this is the case, and that the same holds for modal contexts in general, as well as for attitude reports {to be discussed below). See Geurts (1 999a) for extensive discussion of this matter. =


This sentence has two readings, depending on whether the king of France is accommodated inside or outside the scope of the modal operator:

Bart Geurts 3 2 3

This is stronger than (I ob), of course, and now the predictions made by the IP and the BP coincide. If modal sentences do not differentiate between the IP and the BP, one might hope that attitude reports will. But, surprisingly perhaps, they will not. ( 1 2) Fred believes that I will have dinner with the king of France tomorrow. Suppose again that this is uttered in a context in which it is not yet settled if France has a king. Then this sentence may be construed as (I 3a), which entails that there is a French king, or as (I 3 b), which lacks this entailment.

Key: Kx: x is

the king of France; Dx: the speaker will have dinner with x tomorrow; BelFcp : Fred believes that cp

These readings are clearly distinct, and in this case it seems quite unlikely that a realistic semantics will allow us to maintain that either one is stronger than the other. That being so, the BP and the IP make different predictions: according to the BP, (I 3a) is preferred, while the IP does not predict a preference for one or the other. Which prediction is the correct one? In my opinion, the first prediction is better than the second, because ( 1 2) does seem to imply, all things being equal, that France has a king.7 But there is a snag: (12) also seems to imply, all things being equal, that Fred believes that France has a king. That is to say, the normal way of reading the king ofFrance in (12) is, in a sense, de re and de dicto at the same time, and may be represented as follows: (I4) [u: Ku, BelF[u: Ku, Du] ] There is little agreement in the literature on how this type of reading may be accounted for. Zeevat (I 992) defines his projection mechanism in such a way that it directly produces structures like this. Heim (I 992) outlines a theory of presupposition projection which yields (I 3b), and suggests that this reading is eked out on the basis of pragmatic inferences, so as to obtain (I4). Finally, the theory presented in Geurts (I999a) is the mirror image of 7

Intuitions about presuppositions triggered in attitude contexts are a delicate business, because

they are unusually sensitive to background knowledge, for obvious reasons: if ( 1 2) is uttered in a

context in which it is taken for granted that France is a republic, it is entirely natural to attribute the

presupposed content to Fred, by way of local accommodation. Readers who disagree with the author's judgment about ( r z) are therefore advised to replace

the king ofFrance with a less obvious instance of

presupposition failure, like my lawyer. For further discussion of the empirical problems besetting presuppositions in attitude contexts, see Geurts

(1 999a: 1 3 3-40).


(I 3) a. [u: Ku, BelF[: Du] ] b. [: BelF[u: Ku, Du] ]


( I 5) All customers of the Lone Star saloon have to part with their guns at the entrance.

This will ordinarily be interpreted as saying that all customers who have a gun are required to leave it at the entrance. On this reading, the occasional cowboy or Indian who does not have a gun and enters the Lone Star saloon does not falsify the statement in ( I 5 ). [x: Cx] (all x) lli, �: Gvu, Lxv] ] [x, u: u x, Cx] (all x) [u, �: Gvu, Lxv] ] [x: Cx] (all x) [�: Gvx, Lxv] ] [x, v: Cx, Gvx] (all x) [: Lxv] ] Key: Cx: x is a customer at the Lone Star saloon; Gxy: x is y's gun; Lxy: x has to part with y at the entrance

(I6) a. b. c. d.

[: [: [: [:

=

( I 6a) is the semantic representation of ( r 5) in which only two

presuppositions remain to be processed: the definite NP their guns triggers the two-part presupposition that (i) there is an individual u such that (ii) v is u's gun. The first presupposition is bound to the discourse referent x in the domain of the quantifier, as shown in ( r 6b), which is equivalent to ( r 6c). The second presupposition cannot be bound and must therefore be accommodated. Accommodation in the main DRS is not possible because this presupposition contains a discourse referent, i.e. x, which is introduced in the domain of the quantifier, hence the BP predicts that accommodation in the restrictor is the next-preferred option, and we obtain the DRS in (r6d), which is the reading we wanted to account for.


Heim's: my projection mechanism yields (I J a) by default, courtesy of the BP, and this is elaborated on by pragmatic reasoning, the outcome of which is ( 14). Without delving into the details of this discussion, it will by now be clear where the problem lies: if (I 4) is the preferred reading of ( 1 2), and this example is representative of the interplay between presuppositions and attitude reports, then attitude reports do not drive an observational wedge between the BP and the IP, either. Blutner (this volume) discusses a class of data for which the BP and the IP yield clearly distinct predictions. These are sentences in which a pre supposition cp is triggered within the scope of a quantifying expression Q and t.p contains a discourse referent bound by Q. If we adopt the BP, the DRT account of presupposition projection predicts that in such an event global accommodation is excluded (because the resulting DRS would not be a proper one), so on the preferred reading cp restricts Q's domain. The following is a case in point:

Bart

Geurts

J2 S

(17) Wilma always drinks [gin and tonic]F· With focus on the object NP, this is more or less synonymous with 'Whenever Wilma drinks anything, it is gin and tonic', which is precisely the reading predicted by the BP if it is assumed that non-focused information is presupposed (see Geurts & van der Sandt 1 999 for discussion). So the BP offers a rather neat account of the various ways in which information appearing in the nuclear scope of a quantifier may end up restricting the quantifier's domain, an account that will have to be forsaken if we supplant the BP with the IP. The IP has two related but distinct problems with quantified sentences such as ( I S) and (I7). First, it sometimes makes incorrect predictions, as we have just seen. Secondly, the IP wrongly predicts that the preferred interpretations of these sentences will vary with the choice of quantifier. For example, whereas in a sentence with a universal quantifier there should be a preference for accommodation in the nuclear scope, in a sentence with most there should not be any preference, as far as the IP is concerned, because accommodation in the nuclear scope does not yield a reading that is stronger than accommodation in the quantifier's domain, or vice versa. Therefore (1 8a) should have two equally likely readings, viz. (1 8b) and (I 8c) ( (1 8) = (4) ). (I 8) a. Most teachers spoil their children. b. Most teachers who have children spoil them. c. Most teachers have children and spoil them. This prediction is incorrect, too, because the choice ofquantifier does not seem to affect our preferences for local as opposed to intermediate accommodation. In my opinion, this evidence clearly speaks in favour of the BP and against the IP. But I must concede that this argument is not quite as


If we adopt the BP, we predict that whenever (i) a presupposition x is triggered in the second half of a structure of the form 'cp(Q u)'l/J', and (ii) x contains the discourse referent u, then there will be a preference for accommodating X in cp . If we adopt the IP, on the other hand, we obtain different predictions when Q is a universal quantifier, for example. If a presupposition arises on the right-hand side of all or every, the prediction is that it will preferably be accommodated in the quantifier's nuclear scope, because this reading is stronger than the one predicted by the BP. Thus (I 5) is predicted to imply that all customers of the Lone Star saloon have guns (not to complicate matters even further I ignore the modal here). This prediction, it seems to me, is plainly incorrect. A particularly common subtype of the phenomenon illustrated in (I 5) is exemplified by the following:


8 This paragraph may strike some readers as less than fully transparent, but the issue is rather intricate, both observationally and conceptually, and has been discussed at length elsewhere (Geurts & van der Sandt 1999). 9 One of the referees for this journal notes that, in an optimaliry- theoretic framework. the IP might coexist with a constraint that selected the weakest reading. Technically speaking, this is correct, but apart from that this suggestion causes more problems than it solves. I take it that, whether or not we operate in an optimaliry-theoretic framework. the constraints we introduce call for some sort of


powerful as it may appear to be at first, because speakers' intuitions about the data are sometimes a bit fuzzy around the edges, and in some cases speakers seem to favour readings that are stronger than the BP would lead us to expect (c£ Beaver 1 994). Note, for example, that the statement in (r s) would be somewhat odd if it were part of the common ground that it is unusual for the Lone Star's clientele to carry weaponry of any sort. Observations like this have been taken to suggest that ( r s) conveys that a majority of the saloon's customers have guns, which is not unlike the reading Blutner derives by means of his version of the IP. Nevertheless, it bears emphasizing that Blutner's prediction is still stronger than what is intuitively perceived, and his predictions concerning sentences with most are not borne out by the empirical facts, at all. Furthermore, Geurts & van der Sandt (1 999) have tried to show how the quasi-universal inferences associated with some instances of domain restriction can be accounted for without repudiating the BP.8 The moral of the foregoing discussion is the following. Judging only from what the BP and the IP say, one should expect the two principles to have clearly distinct observational consequences. It turns out that this is not the case: there are very few data that straightforwardly falsify one principle while corroborating the other. The only data that make a clear distinction are related to domain restriction, and although I should say that these facts suggest rather strongly that the BP outperforms the IP, the evidence is weakened somewhat by the considerations adduced in the last paragraph. If we consider supplanting the BP with the IP we should not hope for any predictive gains; empirical adequacy is not at issue. What is at issue, I take it, is the notion that the IP offers a principled alternative to what is little more than a stipulated rule. Saying that stronger readings are preferred to weaker ones certainly seems less ad hoc than saying that presuppositions want to float up to the main DRS. However, this impression is quite misleading. True, the BP cries out for some sort of justification. But the same holds for the IP, too. Why should hearers prefer strong interpretations over weaker ones? Isn't that an extremely risky processing strategy? Would it not be wiser to opt for the weakest reading, ceteris paribus? It surely would be the safest strategy, and it has been suggested more than once that people actually adopt it, too (e.g. Crain & Steedman 1 98s).9

Bart Geurts 327

autohyponyms: (19} Fred picked a fight with a Yankee. 1. Fred picked a fight with an inhabitant of the Northern States of the us.

u.

Fred picked a fight with an inhabitant of the US.

syntactic ambiguity: (2o) a. Barney's social circle consists of inarticulate philosophers and literary critics. i. inarticulate [philosophers and literary critics] ii. [inarticulate philosophers] and [literary critics] b. The cover of Betty's latest novel is decorated with pink fruits and vegetables. i. pink [fruits and vegetables] ii. [pink fruits] and vegetables scope ambiguity: (2 1} Everybody in this room speaks two Romance languages. i. Two Romance languages are spoken by everybody in this room. ii. Everybody in this room speaks two Romance languages. justification. Constraints do not come for free; they must be motivated. My concern here is with the justification of one panicular constraint, viz. the IP, and adding a funher constraint will not solve this problem. What is more, it will immediately raise the rather hairy question how we could motivate a pair of constraints that flatly contradict each other. 10 The term 'autohyponym' is due to Horn (1984), who also has a Yankee example (c£ McCawley 198 1).


Blumer (this volume} implements his version of the IP in an optimality theoretic reconstruction of a neo-Gricean model of interpretation. More succinctly, he sees the IP as a special case of Grice's first quantity maxim. It should be noted, however, that the maxim's raison d'etre is lost in Blutner's reconstruction. Grice's quantity maxim exhorts the speaker to make his contribution 'as informative as is required for the current purposes of the exchange' (Grice 1 989: 26; emphasis added). This is eminently reasonable, of course, and not in need of further justification. But it is obviously not the same as demanding that an utterance be as informative as possible. Thus far I have pretended as if the IP were a general principle that applies not only to presupposition projection but across the board. If this turned out not to be the case, that is to say if the IP held only in some cases but not in others, the explanatory merit of the IP would dwindle even further. Unfortunately, I see little reason to believe that the IP is a general principle of interpretation. The following examples all have two readings, where one (i} entails the other (ii): 1 0


(22) Fred did not pick a fight with a Yankee. 1. Fred did not pick a fight with an inhabitant of the Northern States of the US. n. Fred did not pick a fight with an inhabitant of the US. But even if there is a distinct preference for the first reading of ( r 9), which I doubt very much, I am quite sure that the presence of a negative expression does not alter this preference. The most elaborate attempt at showing that the IP (or at least something very much like it) applies in a non-presuppositional domain is Dalrymple et a I.'s ( I 998) theory of reciprocals. It is a familiar observation that sentences containing reciprocal pronouns admit of various types of interpretations, which are conditioned by world knowledge. The following examples are all from Dalrymple et al.'s paper: (23) a. House of Commons etiquette requires legislators to address only the speaker of the House and refer to each other indirectly. b. fu the preposterous horde crowded around, waiting for the likes of Evans and Mike Greenwell, five Boston pitchers sat alongside each other: Larry Anderson, Jeff Reardon, Jeff Gray, Dennis Lamp and Tom Bolton. c . They climbed a drainpipe to enter the school through a high window and stacked tables on top of each other to get out again. Dalrymple et al. observe that the preferred interpretations of these sentences can be ordered in terms of strength. (23a) says that every pair of legislators


If the IP were a general principle of interpretation, we would predict that the i-readings are always preferred to the ii-readings, which is surely incorrect. There may be some room for dispute in one or two cases, but the least one can say that there is no strong preference for either reading in most cases, and especially when we compare these examples with the standard examples of presupposition projection, where preferences are generally quite pronounced, it becomes highly unlikely that anything like the IP is at work in the examples just cited. Of course, this is not a knockdown argument, because the IP is a violable constraint, and it is always possible to claim that, in each of (r9)-(2 r), it is outranked by other constraints, which, as it happens, do not enter the fray in presupposition projection. However, this objection remains destitute of force as long as the relevant constraints are not identified, and it is demonstrated that they take precedence over the IP. Note, furthermore, that the IP, when applied across the board, predicts that negation would reverse the preferences predicted for ( r 9)-(2 r). For example, (r9) is then predicted to contrast with:

Bart Geurts 329

are required to refer to each other indirectly; (23b) says that the Boston pitchers were sitting in a row; and (23c) says that every table either supports or is supported by at least one other table. To see that these readings can indeed be ordered by means of entailment, it may be helpful to abstract away from all irrelevant particulars, as follows (see Dalrymple et al.'s paper for a more precise formulation). Let A be the domain of quantification and R the relation in question. Then the preferred readings of (2 3a-c) make the following requirements:

Clearly, as long as the domain of quantification is not too small, (24c) is entailed by (24b), which in its turn is entailed by (24a). Dalrymple et al. go beyond these observations by distinguishing six possible (sentence) mean ings any token of each other may give rise to. (For reasons that I fail to grasp Dalrymple et al. also maintain that each other is not ambiguous.) These readings are partially ordered in terms of entailment, and this ordering determines which reading is preferred in a given situation, because the hearer will choose the strongest reading that is consistent with his beliefs about the world. This is what Dalrymple et al. call the 'Strongest Meaning Hypothesis' (SMH}; it is a special case of the IP. There are a number of problems with this analysis, but I will focus here on what I take to be the most fundamental objection, which is simply that the SMH is an almost idle wheel in Dalrymple et al.'s theory. To see why, consider (2 3c}, and its schematic interpretation, (24c). According to the SMH, this reading is the outcome of a selection process which first discards (24a) and (24b), in that order, because they are inconsistent with our knowledge about tables and what it means for objects to be stacked on top of each other. But if non-linguistic knowledge must be employed to weed out incorrect readings, why not use it to directly select the correct reading? The crucial factor in Dalrymple et al 's account of the examples in (23) is world knowledge. Informativeness only needs to be brought into play because it is assumed that world knowledge cannot select a reading but will only serve to eliminate readings. This assumption cannot simply be taken for granted. In order to show that strength is implicated in the interpretation of reciprocals, one should turn to examples where world knowledge is inert, because the SMH leads us to expect that in such cases speakers' intuitions .


(24) a. Every individual in A participates in R with every other individual in A. b. Every individual in A participates in R with every other individual in A, if not directly then indirectly. c. Every individual in A participates in R with at least one other individual in A.

3 30 Buoyancy and Strength

will be solely determined by strength. So let us add two new words to the English lexicon, but without saying what they mean, and consider the following statement:

(2 5) The yogs are zagging at each other.

(26) I have here a number of tennis balls, which I have arranged in such a way that they touch each other. Question: How many tennis balls do I have? If Dalrymple et al. were right, there should be a distinct preference for one unique answer, viz. 'Four'. This prediction is manifestly incorrect. It is wrap-up time. I have argued that the IP does not improve on the BP, be it empirically or conceptually: the IP's predictive success is somewhat less than the BP's, and it is a mistake to think that the IP is less ad hoc because (a) there is no solid evidence that the IP applies beyond the presuppositional domain, (b) even if it did, the IP would call for justification just as much as the BP does. If the IP is not a satisfactory solution to the problem sketched at the outset, what is? The shortest answer I can currently give to this question is that I do not know, but I have a longer answer, as well. It is that I see a number of tacks one might pursue, but have not yet made up my mind which is the right one. So, to end on a positive note, let me briefly outline the options I have in mind.

First tack Instead of stipulating that speakers have a proclivity for maxtmtzmg informativeness, as the IP does, one might suppose that they strive to maximize relevance. The major advantage of this move is that it hinges on a principle that can stand on its own, for nobody will deny, I take it, that speakers are relevance seekers in some sense. But there are disadvantages, as well. For one thing, in the absence of an adequate relevance metric this solution will be difficult to make explicit. For another, even if we cannot say exactly what relevance is, it is by no means intuitively clear that the BP


Since world knowledge about yogs and zagging cannot possibly play a role here, it follows from the SMH that the strongest interpretation of (25) is preferred, i.e. every pair of yogs are zagging at each other. A5 far as I can determine this prediction is incorrect: this sentence may have any number of interpretations, and in the absence of further information about yogs and zagging, it is no use asking which of these is preferred. Another way of driving home the same point is by means of the following riddle:

Ban Geurts 3 3 I

always reflects our relevance judgements. For example, I have argued that the BP correctly predicts that a presupposition triggered in the nuclear scope of a quantifier will generally be accommodated in the restrictor (if binding and global accommodation are ruled out, that is). But I do not see why the resulting readings should be deemed more relevant, ceteris paribus, than the ones we obtain if the presupposition remains in situ.

Second tack

Third tack In Geurts (1 999b) it is suggested that the BP has nothing to do with presupposition per se, but applies to backgrounded material in general. Crudely put, the idea is that the bulk of the information contained in a sentence will float up to the main context, because only the small part the speaker wants to focus on stays where it is. This is rather similar to what Horn has dubbed 'assertoric inertia': Semantically entailed material that is outside the scope of the asserted, and hence potentially controversial, aspect of utterance meaning {Stalnaker I 978) counts as asser torically inert and hence effectively transparent to NPI-Iicensing and related diagnostics of scalar orientation {Horn 2ooo).

Thus presuppositions tend to go up because they are not asserted. Sadly, however, this will hardly qualify as an explanation as long as it is not


We could deny the premise that the BP stands in need of reduction or justification. According to most latter-day accounts, presuppositions are bits of information that are presented as given in the context of discourse. This much is common sense, but our near-truism takes on an less trivial aspect in a dynamic theory of interpretation, like DRT or context-change semantics, which hold that there may be any number of contexts open at any given time. In DRT, these contexts are the DRSs that are accessible from the spot at which an expression is interpreted. So if the speaker signals that something is given in the current context, something like an ambiguity arises. But it is obvious how such ambiguities will be resolved: the main context (i.e. the principal DRS) will generally be preferred precisely because it is the main context. Period. I am somewhat ambivalent towards this solution. From time to time it strikes me as blatantly obvious and utterly satisfying; but most of the time it does not.

3 3 2 Buoyancy and Strength

explained why assertorically inert material is 'effectively transparent'. But at least this suggestion widens the problem, and that might a good thing. Acknowledgements This paper is based on a talk given at the Utrecht Conference on the Optimization of Interpretation, in January of this year. I should like to thank the audience at the conference for a lively discussion, and the two anonymous referees for the journal ofSemantics for their extensive comments on the first version of this paper. This work was supported by a grant from the Deutsche Forschungsgemeinschaft (DFG). BART GEURTS

RE FERENCES Beaver, D . ( I 994), 'Accommodating topics', in P. Bosch & R. van der Sandt (eds), Focus and Natural Language Processing, Vol. ;: Discourse, Working Papers of the

Institute for Logic and Linguistics Vol. 8, Heidelberg. Blutner R. (this volume), 'Some aspects of optimality in natural language inter pretation'. Crain, S. & Steedman, M. (I985), 'On not being led up the garden path: the use of context by the pyschological parser', in: D. R. Dowty, L. Karttunen, & A. Zwicky (eds), Natural Language Parsing: Psycho IBM,

logical, Computational, and Theoretical Perspectives, Cambridge University Press,

Cambridge. Dalrymple, M., Kanazawa, M., Kim, Y., Mchombo, S. & Peters, S. (I 998), 'Reci procal expressions and the concept of reciprocity', Linguistics and Philosophy, 2 1 , I S9-2 ! 0. Gazdar, G. (I 979), Pragmatics: Implicature, Presupposition, and Logical Form, Aca demic Press, New York.

Geurts, B. (I 999a), Presuppositions and Pro nouns, Elsevier, Oxford. Geurts, B. (I 999b), 'Specifics', in B. Geurts, M. Krifka, & R. van der Sandt (eds), Focus and Presupposition in Multi-Speaker Dis course, Workshop reader, ESSLLI 99,

Utrecht, 99- 1 29. Geurts, B. and Sandt, R. van der (I 999), 'Domain restriction', in P. Bosch & R. van der Sandt (eds), Focus: Linguistic, Cognitive, and Computational Perspectives,

Cambridge University Press, 268-92. Grice, H. P. (I 98 I), 'Presupposition and conversational implicature', in P. Cole (ed.), Radical Pragmatics, Academic Press, New York, I 8 3-98. Reprinted in Grice (I 989: 269-82). Grice, H. P. ( I 989), Studies in the Way of Words, Harvard University Press, Cambridge, MA. Haspelmath, M. (2ooo), 'Optimality and diachronic adaptation', Zeitschrifi filr Sprachwissenschajt I 8: I 80-205. Heim, I. ( I 982), 'The semantics of definite and indefinite noun phrases', Ph.D.


Received: 0).04.2000 Final version received: 2 3· I 0.2000

Department of Philosophy University of Nijmegen Postbox 9 1 03 6soo HD Nijmegen The Netherlands [email protected] http://www.p hil.kun.nl/tjl/bart/

Bart Geurts 3 3 3 language game', journal of Philosophical thesis, University of Massachusetts, Logic, 8, 3 39-59. Amherst. Heim, I. (I983), 'On the projection prob McCawley, J. D. (I98 I), Everything that Linguists have Always Wanted to Know lem for presuppositions', Proceedings of the West Coast Conftrence on Formal Linguistics, 2, I I4-26.

sis of Specific Indefinites: Common Grounds as Structured Information States, Garland,

New York. Zeevat, H. (1992), 'Presupposition and accommodation in update semantics', journal of Semantics, 9, 379-4 1 2. Zeevat, H. (1999), 'Old material', in: B. Geurts, M. Krifka, & R. van der Sandt (eds), Focus and Presupposition in Multi Speaker Discourse, Workshop reader, ESSLLI 99, Utrecht, r 6 r -8.


Heim, I. (I992), 'Presupposition projection and the semantics of attitude verbs', journal of Semantics, 9, I 8 3-22 r . Horn, L. R. (I984), 'Ambiguity, negation, and the London School of Parsimony', NELS, 14, 108- 3 1 . Horn, L. R. (2ooo), 'Assertoric inertia', paper presented at the LSA meeting, Chicago. Kamp, H. {I98 I), 'A theory of truth and semantic representation', in J. A G. Groenendijk, T. M. V. Janssen, & M. B. J. Stokhof (eds), Formal Methods in the Study of Language, Mathematical Centre Tracts I 3 5, Amsterdam, 277322. Kamp, H. & Reyle, U. (I993), From Discourse to Logic, Kluwer, Dordrecht. Karttunen, L. (1973), 'Presuppositions of compound sentences', Linguistic Inquiry, 4, I67-93· Karttunen, L. & Peters, S. (I 979), 'Conven tiona! implicature', in C.-K. Oh & D. Dinneen (eds), Syntax and Semantics, Vol. 1 1 : Presupposition, Academic Press, New York, I-56. Lewis, D. ( 1 973), Counterfactuals, Harvard University Press, Cambridge, MA Lewis, D. (1 979), Scorekeeping in a

about Logic but were Ashamed to Ask,

Blackwell, Oxford. Sandt, R. A Van der (I992), 'Presupposition projection as anaphora resolution', Journal of Semantics, 9, 3 3 3-77. Stalnaker, R. C. (I 968), 'A theory of con ditionals', in N. Rescher (ed.), Studies in Logical Theory, Blackwell, Oxford, 98- I I2. Stalnaker, R. C. (I974), 'Pragmatic presup positions', in M. K. Munitz & P. K. Unger (eds), Semantics and Philosophy, New York University Press, New York, I97-2 I 3· Stalnaker, R. C. (I978), 'Assertion', in P. Cole (ed.), Syntax and Semantics, Vol. 9: Pragmatics, Academic Press, New York. Strawson, P. F. (I964), 'Identifying reference and truth-values', Theoria, 30, 96- I I 8. Yeom, J.-I. ( 1 998 ), A Presuppositional Analy

© Oxford University Press 2000

journal ofSmumrics 1 7: 3 3 5-395

Valence Creation and the German Applicative: the Inherent Semantics of Linking Patterns LAURA A. M I C HA E L I S University of Colorado at Boulder J O S E F RUPPE N H O FER University of California, Berkeley

Abstract '

r I NTRO D U C T I O N Compositional theories of sentence semantics have been centrally con cerned with the relationship between the meanings of lexical items and the meanings of sentences that contain those lexical items. Verbal argument structure has been of great interest in recent theory building because of the transparent nature of the relationship between the verb's semantic require ments and the number and kind of thematic roles in the sentence. The majority of theories of verbal argument structure accord a central place to the concept of ALTERNATION, exploring the nature of the relationship between argument frames licensed by a given verb. The recognition that


We provide a unified account of semantic effects observable in attested examples of the German applicative ('be - ) construction, e.g. Rollstuhlfa hrer Poul Schacksen aus Kopenhagen will den 1997 erschienenen Wegweiser Handiguide Europa Jortfuhren und zusammen mit Movado Berlin berollen ('Wheelchair user Poul Schacksen from Copenhagen wants to continue the guide 'Handiguide Europe', which came out in 1 997, and roll Berlin together with Movado.'). We argue that these effects do not come from lexico-semantic operations on 'input' verbs, but are instead the products of a reconciliation procedure in which the meaning of the verb is integrated into the event-structure schema denoted by the applicative construction. We analyze the applicative pattern as an ARGUMENT-STRUCTURE CONSTRUcriON, in terms of Goldberg ( 1 995). We contrast this approach with that of Brinkmann (1997), in which properties associated with the applicative pattern (e.g. omissibility of the theme argument, holistic interpretation of the goal argument, and planar construal of the location argument) are attributed to general semantico-pragmatic principles. We undermine the generality of the principles as stated, and assert that these properties are instead construction-particular. We further argue that the constructional account provides an elegant model of the valence-creation and valence-augmentation functions of the prefix. We describe the constructional semantics as prototype-based: diverse implications of be-predications, including iteration, transfer, affectedness, intensity and saturation, derive via regular patterns of semantic extension from the topological concept of COVERAGE.

3 36 Valence Creation and the German Applicative


argument-structure alternations are licensed by narrow semantic classes of verbs (Levin 1 993; Gropen et al. 1 989) has led to lexically based accounts of alternations, most of which posit minimally specified verbal valence structures along with general principles ('linking rules') governing the interface between verbal thematic structure and surface syntax. While the general principles are typically based upon some version of Fillmore's ( 1 968 ) semantic-role hierarchy, the greatest attention has been paid to those cases in which linkings to the core grammatical functions do not follow the predictions of the hierarchy. Mappings of this type are described by two general approaches. In the first approach, semantic (e.g. Aktionsart) representation is held constant, and the marked patterns are viewed as violations of mapping constraints (Foley & van Valin 1 984). The existence of the marked patterns may be attributed to optimization elsewhere in the system, including the achievement of a match between a given thematic role and a given functional role (e.g. location and topic in Bresnan's 1 994 analysis of locative inversion). In the second approach, the mapping constraints are assumed to be inviolable, while the lexica-semantic representations which provide inputs to those rules are manipulated through semantic operations on decomposed lexical structure (Gropen et al. 1 989; Wunderlich 1 997; Rappaport Hovav & Levin 1 998 ). Each model has been applied to the problem of the locative alternation. In the marked member of this alternation, a goal or location argument, which would otherwise receive oblique coding, receives the coding prototypically associated with the thematic role of patient (Foley & van Valin 1 984; Dowty 1 99 1 ; van Valin & La Palla 1 997). This pattern, exemplified by the English sentence She smeared the canvas with paint, has been viewed as theoretically important because its interpretation, involving an attribution of 'affectedness' to the goal argument, suggests something noncompositional about the operation of the lexical rule. The lexical rule appears to be adding meaning. The facts of German, while superficially similar to those of English, force us to address an additional, more fundamental question: Is there a lexical rule at all? At first glance, the locative alternation identified in English finds a straightforward parallel in German, with two obvious differences. First, the oblique-promoting device in German, like the applicative pattern in Bantu languages (Alsina & Mchombo 1 990; Wunderlich 1 99 1 ), involves morpho logical marking on the verb-in the case of German, the inseparable prefix be. Thus, for example, the applicative counterpart of the German verb schmieren ('smear') is beschmieren, as in Sie beschmierte die Leinwand mit Farbe ('She smeared the canvas with paint'). Second, the German applicative linking pattern combines with both intransitive and transitive verbs. That is, it accepts not only trivalent transitive verbs denoting transfer, e.g. schmieren,

Laura A. Michaelis and Josef Ruppenhofer 3 3 7

( r ) Message ID ([email protected])

[H]abe ich mich von meinen Kollegen [ . . . ] auch mal Have I myself by my colleagues also occasionally eifrig mit Kaffee bekochen lassen [ . . . ]. eagerly with coffee be-cooked let 'At times, I also let my colleagues busy themselves with making me coffee.''

1

Our data come from six different sources and are identifiable in the following ways. Examples

taken from the on-line corpora at the

Institut fur Deutsche Sprache begin with a sequence of capital WK = Wendekorpus). Examples taken from the Deja news service on the worldwide web start out with the words Message Jd. Data from the Lex.is- Nex.is research service start out with the name of the journal, newspaper, or magazine the quote is taken from (e.g. Suddeutsche Zeitung). Examples taken from the Frankfurter letters and numbers coding the specific corpus they are taken from (e.g.


but also bivalent intransitives denoting location or locomotion, e.g. wandern ('wander'). Accordingly, the verb wandern has the bivalent transitive counterpart bewandern, as in Sie bewanderte den Schwarzwald ('She wandered the Black Forest'). In both the trivalent and bivalent conditions, the location argument, which would otherwise receive oblique (preposition phrase) coding, is expressed by a direct grammatical function (a direct object when the voice is active). In the latter case, the applicative has a TRANSITIVIZING function. As a number of theorists (including Marcus et al. 1 995 and Brinkmann 1 997) have observed, the German applicative pattern is both productive and constrained: a great many verbs have applicative alternates and yet these verbs appear to cluster into relatively narrow semantic classes. In this regard, of course, the German applicative behaves much like its English analog, as described by Pinker ( 1 98 9) and Levin ( 1 993 ), inter alia. For example, causative verbs of position like English lean (or German Iehnen) do not generally form acceptable applicative sentences, as in e.g. *She leaned the field with ladders. All of the foregoing observations are consistent with a model of the German applicative pattern in which a lexical rule mediates between two entries for a given verb. This general type of model works whether the alternating verbs are bivalent or trivalent and whether or not the two verb entries related by the rule are assumed to contain identical sets of semantic entailments. However, a more comprehensive look at the inventory of verbs which license the applicative pattern in German suggests that we need a different conception of the function of this pattern than that suggested by lexical-rule based approaches. This broader picture includes examples which cast doubt upon the general claim that the German applicative is a device for 'promoting' location arguments that would otherwise receive oblique coding. In certain of these examples, illustrated in (r), the 'input' lexical entry is (arguably) trivalent but does not license a goal argument:

3 3 8 Valence Creation and the German Applicative

In ( I), be-preflxation has an effect similar to that of the English ditransitive pattern in allowing the linking of a BENEFICIARY argument to a core grammatical function (cf He made them coffee). The base verb in (I), kochen ('cook'), does not license an oblique expression denoting a goal, as illustrated by the ill-formed permutation *Ich habe meine Kollegen Kaffee zu mir kochen lassen, whose English translation is the equally ill formed *I had my colleagues make coffee to me. Another problematic class of bivalent applicative verbs are those whose base verb is monovalent. This class is exemplified in (2 ) for the verb schummeln ('cheat'):

(2) Peter beschummelte mich beim Kartenspielen. Unlike its ostensible English counterpart cheat, schummeln does not accept an oblique argument expressing the party deceived. The sentence * Er schummelte mir beim Kartenspielen ('He cheated me [dative] in cards') is ill formed, as are the variants illustrated by * Er schummelte auj!zu/gegen mich beim Kartenspielen ('He cheated on/to/against me in cards'). Therefore, the applicative verb beschummeln in (2) could not be said to code as a direct argument what would otherwise be coded as an oblique (preposition phrase or dative). Instead, the applicative pattern itself appears to license the 'malefactee' argument. This licensing effect is not limited to that of merely augmenting verbal valency. In our flnal class of cases, exemplified by (3), the input form lacks valency entirely; it is a noun rather than a verb: ( 3) Message ID ( I 99 8 0909 I 3 3 74 5 ooJAA I 946 5 @ laddero I .news.aol.com) . Es mag ja lustig sein, zwei hartgekochte Eier wie Clownskopfe mit angekeimten Sojabohnen zu behaaren und sie auf Gurkenscheiben zu stellen, ihnen mit zwei Tomatenstreifen Miinder zu verpassen und Auglein aus Sojasprossen einzudriicken. 'OK, it might be funny to hair two hard-boiled eggs like clown's heads with germinating soy beans, to stand them up on cucumber slices, to give them mouths from tomato strips, and to impress soy shoots on them as little eyes.' In (3 ), a trivalent applicative predication, the base form is the noun Haar ('hair'). This word is inherently nonrelational, and has no verbal counterpart Rundschau Korpus are marked by a single number. Examples collected from websites are given in the standard URL format. Examples that were provided by native-speaker consultants carry no marking.

We include a narrow gloss in example ( 1) in order to demonstrate both the internal structure of the verb and its transitivity. We will provide only broad glosses hereafter, although for sentences containing applicative verbs we have attempted to construct glosses which reflect as accurately as

possible both the form class of the stem to which the prefix resulting combination.

be- attaches and the transitivity of the


'Peter cheated me in cards.'

Laura A Michaelis and Josef Ruppenhofer 3 39

2

One reviewer has commented that (3) is problematic as an example of an applicative predication

('hair') is not literally the theme of the denoting transfer: the type denoted by the nominal transfer event, since bean sprouts are not hair. We do not find this objection compelling, since it is

Haar

meaning relative to a metaphorical schema and not 'literal' meaning that is at stake here. In the image

mapping which the reader is asked to perform, food items map to parts of the human head, and while

this mapping is in force, one can truthfully refer to the bean sprouts as hair. To believe otherwise would be to subscribe to the notion that metaphor is inherent falsity-a notion that Lakoff ( 1 987) and others have refuted.


outside of the applicative construction; there is no transfer verb * haaren ('hairV The applicative predication in (3) denotes a transfer event of the type denoted by trivalent applicative verbs like laden ('load'), and yet the transfer implication cannot be attributed to the semantics of the base form, which in this case is not a verb, let alone a transfer verb. In all of the examples (1)-(3), the appropriate inputs are simply lacking. These examples therefore suggest that the lexical-rule based model of the applicative pattern is inadequate. These examples also disturb the neat picture of constrained productivity presented above. They suggest that the applicative pattern is not as selective about its inputs as previous analyses have implied, as it combines with verbs that do not denote either transfer or location. This fact makes it more difficult to describe the use conditions upon the applicative. One could claim that the tokens in (1)-(3) are idiomatic or marginal uses which do not bear upon the function of the applicative. However, the denominal applicative exemplified in (3) is novel, and in fact many of the novel tokens used to illustrate the productivity of the German applicative involve verbs which do not have base forms denoting transfer, locomotion or location (see Giinther 1 974 for an extensive listing of such examples). Brinkmann (1 997= u ), for example, cites as evidence of the productivity of be-prefixation relatively unconventional attested tokens, including bedudeln (roughly, 'drone someone'), whose base form is the intransitive verb dudeln ('play tunelessly'). These novel examples of be-prefixation have little relevance for the productivity of the locative alternation, since they do not illustrate it. If we assume that the productivity of a form is evidence of a specific function, and that the function of the applicative pattern cannot be 'locative promotion', then we face the challenge of discovering what function of be-prefixation accounts for denominal examples like (3) and the examples typically used to illustrate the locative alternation. Even were we to broaden the function of the applicative to that of promoting any argument otherwise expressible as a preposition phrase (and not merely a locative argument), we would encounter difficulty. This more general version of the oblique-promotion analysis appears at first to be valid: certain bivalent be-verbs which do not qualify as verbs of location nevertheless have bivalent intransitive counterparts which license a

340 Valence Creation and the German Applicative


prepositional phrase. Brinkmann ( 1 99T 84-5 ) points to several verb classes in which such alternations can be found, including verbs of active perception, verbs of speech, and verbs of emotional expression. Pairs exemplifying alternations in each class are, respectively, riechen an/beriechen 'sniff [e.g. a flower]/sniff thoroughly'; sprechen iiber/besprechen 'talk about/ discuss'; and weinen um/beweinen 'cry about/mourn [e.g. a death]'. However, the broader alternation-based model would not, for example, extend to bivalent applicative verbs with monovalent base forms (e.g. bedudeln). The latter class of verbs includes not only beschummeln in (2) but also applicatives formed from other verbs of deception, including mogeln ('cheat'), schwindeln ('fib'), Jlunkern ('lie') and lugen ('lie'). Moreover, the more general model of the applicative alternation does not encompass applicative verbs which lack verbal base forms entirely, e.g. the denominal verb behaaren in (3). Thus, even a very broad conception of the 'promotion' function of the applicative is too narrow. If we abandon the idea that the applicative pattern is an argument promotion device, we must then ask whether it is used to achieve semantico-pragmatic effects that are not derivative of the promotion function. It would be a challenge to isolate and describe such effects, if they exist. The examples given thus far demonstrate that the range of meanings associated with be-prefixed verbs is large, and includes implica tions related to location, transfer, and malefaction/benefaction. Do these meanings have anything in common? Ironically, the very characteristic of be-preftxation which makes it worthy of an in-depth synchronic study-its high type frequency-also appears to point toward a bleached rather than rich semantics (Eroms 1 980) . The number of be-prefixed verbs is consider able: there are several hundred tokens listed by Gunther ( 1 974) . When a common semantic denominator is recognized, it is generally highly schem atic, and not obviously attributable to the presence of the prefix. Wunderlich ( 1 987) , for example, proposes that be-predications express 'topological local proximity'. Others have proposed a general implication of 'affectedness' of the object-denotatum (Filip 1 994) . Both of these analyses seem plausible, and yet the implications in question can also be analysed simply as properties of the semantic prototype associated with transitive predications (Hopper & Thompson 1980; Hopper 1 98 5 ) . Accordingly, we reject the idea that there is a single abstract meaning associated with the applicative pattern. Instead, we propose to capture the commonalities among usages of the be-pattern through an associative network based on a single semantic schema (Goldberg 1 995; Lakoff 1987) . This schema is one in which a THEME physically covers a LOCATION (either over the course of time or at a given point in time). We will propose that this schema is the basis for certain metaphorical extensions. The

Laura

A.

Michaelis and Josef Ruppenhofer 341


relationship between the schema and these metaphorical extensions will be represented by links denoting metaphorical mappings (Goldberg I 995; Lakoff I987). These links include the following independently motivated metaphorical mappings (Reddy I979; Lakoff & Johnson I98o; Sweetser I 990; Goldberg I995): DISCOURSE IS TRAVEL OVER AN AREA, PERCEIVING IS COVERING OBJECTS WITH ONE'S GAZE, THE CONDUIT METAPHOR, EFFECTS ARE TRANSFERRED OBJECTS. We further propose that, through a mode of grammaticalization called PRAGMATIC STRENGTHENING (Hopper & Traugott I 993; Konig & Traugott I 98 8), the applicative pattern has also come to express TRANSFER, ITERATED ACTION, INTENSIFICATION of the action or state denoted by the verb, and EFFECTS achieved by means of an action. These inference-based extensions conventionalize prototypical components of applicative semantics, while canceling entailments related to coverage. An analysis of this type will allow us to provide a specific semantic analysis for the applicative pattern while acknowledging that a given pair of applicative predications may have few semantic commonalities. In our proposal, the semantic features shared by applicative verbs are contributed by the argument-structure pattern with which those verbs combine. This proposal counters an analytic trend. While proponents of lexical-rule based approaches appear to agree that semantic constraints determine whether a linking rule can APPLY, most appear unwilling to embrace the idea that a linking rule can CONTRIBUTE conceptual content not found in the input verb. Thus, Gropen et al. (1991) and Pinker (I98 9) argue that the 'affectedness' implication associated with oblique-promoting patterns is a general implication of direct-object coding rather than a meaning component contributed by the linking rule (see also Rice 1 989; Rappaport & Levin 1988; Tenny I987). This position makes sense against the backdrop of an alternation-based model, in which linking rules neither create nor destroy any aspect of thematic structure. The effect of a linking rule is thereby limited to that of altering the expression of participant roles. Since the examples in (I)-(3) call into question the principle of conservation of thematic structure, they also call into question its corollary-the proposition that linking rules do not contribute meaning to sentences. We will argue that the meanings of examples like (1)-(3) are products of a reconciliation procedure in which the meaning of the verb is brought into conformity with the meaning of the applicative pattern. On this model, the be-prefix is a morphological feature of the applicative pattern, rather than a device for deriving new verbs or verb entries. The applicative pattern is an ARGUMENT-STRUCTURE CONSTRUCTION, in the sense of Goldberg (1995). Such constructions are linking templates which denote basic-level event types (like transfer and caused motion). As Saussurean signs, these pairings are highly similar to verbs. Both verbs and argument-structure constructions



have (a) thematic and Aktionsart structure, (b) meanings which may be extended metaphorically, and (c) idiosyncratic use constraints. The concep tion of grammar as a hierarchically organized inventory of form-meaning pairs (with greater and lesser degrees of internal complexity) is central to CONSTRUCTION GRAMMAR (CG) (Zwicky 1 994; Kay & fillmore 1 999; Goldberg 1 995; Michaelis 1994; Michaelis & Lambrecht 1 996; Jackendoff 1 997b). We will argue that widely identified interpretive properties of the German applicative pattern are idiomatic properties of the applicative construction, and that attempts to attribute these properties to general semantico-pragmatic principles have failed. In the CG model of argument structure as proposed by Goldberg, the semantic effects observable in (r)-(3) do not result from 'derivations' in which a restricted set of 'input verbs' undergoes modification of semantics and syntax, whether in the lexicon or elsewhere. In the CG model, verbs do not have alternate semantic representations. Instead, verb meaning is constant across syntactic contexts. No additional lexical entry is created to represent the meaning and valency of verbs found in specialized patterns like the ditransitive. Verbs unify with verb-level linking constructions which denote event types. These linking constructions assign grammatical functions to participant roles contributed by the verb. In addition, since these constructions denote event types, each licenses the theta frame entailed by its particular event type. This set of thematic roles may PROPERLY INCLUDE the set of roles licensed by the verb. In such cases, verbs which combine with the construction undergo modulation of their theta frames. In the case of the applicative pattern in particular, as we will show, the construction not only AUGMENTS verbal valency but CREATES valence patterns for open -class items which are not inherently relational. We will argue that the function of valence building cannot be revealingly modeled by lexical-rule accounts, while this function is predicted by the constructional model. This paper will be structured in the following way. In section 2, we will discuss a recent account of the function of be-preftxation, presented by Brinkmann (1 997) as part of an acquisition study of the locative alternation in German. Although the locative alternation has been widely described, we chose to react to Brinkmann's account because it is comprehensive in its attention to previous literature, provides a clear and well articulated example of the derivational approach to argument structure, and represents a strong challenge to the view advanced here-that the semantic effects observable in (1)-(3) are attributable to a specific formal pattern rather than to more general principles of interpretation. In section 3, we will more fully motivate the construction-based approach to argument structure and discuss the advantages that this approach offers for the analysis of the

Laura A. Michaelis and Josef Ruppenhofer

343

applicative pattern. In section 4, we will discuss the semantic schema associated with the applicative pattern, and its metaphorical and pragmatic extensions. Section s contains concluding remarks.

2

A N ALTERNAT I O N - B A S E D A C C O U NT O F THE APPLI CATIVE PATTE R N 2.I

Overview of Brinkmann (1997)

2.2

Representing valence augmentation and creation

It stands to reason that a derivational account of argument structure, in which a lexical process changes the position in decompositional structure of argument roles, should require that the relevant argument roles be present in the input representation. However, as mentioned in section I , there are applicative verbs which manifestly violate this requirement. These verbs fall into two broad classes: denominal and deadjectival be-verbs, on the one hand, and adjunct-promoting and valence-augmenting deverbal be-verbs on the other. We will discuss each of these two classes in turn, pointing to


Brinkmann's account is based on Wunderlich's (1 987, 1 99 1 , 1 997) LEXICAL DECOMPOSITION GRAMMAR. In this model, inseparable-prefix verbs that take locations as direct objects are lexically derived from their unpreftxed base verbs through FUNCTIONAL COMPOSITION of a verbal predicate and a prepositional predicate. Semantico-pragmatic implications of the applica tive pattern are attributed to general interpretive principles or to the semantics of the prepositional predicate rather than to semantic effects contributed by the linking rule itself In the following section (2.2), we will argue that the preposition-incorporation model does not provide a principled account of the valence-building function of the applicative pattern, and thereby fails to capture a major source of its productivity. In section 2 . 3 , we will challenge Brinkmann's claim that certain widely noted properties of applicative predications can be accounted for straightfor wardly by reference to general interpretive principles and etymology. We will focus on three such properties: the holistic interpretation of the goal argument, the omissibility of the theme argument, and the interpretation of the goal argument as a two-dimensional region. For each of these three features, we will show that Brinkmann's putatively general account cannot in fact be interpreted coherently without reference to semantic properties associated directly with the applicative linking pattern.


(4) Message ID (4lqz8k$lsr@nzr z.rz.uni-karlsruhe.de) Also * m.E.* regelt 4 r , wie ein Radweg zu beschildern ist [ . . . ]. 'Well, * in my opinion* [paragraph] 41 regulates how a bike path needs to be equipped with traffic signs.' Similarly, the verb befreien in (s) can be paraphrased only by the resultative construction Jrei bekommen 'get free, released'. The simple verb Jreien means 'to woo' rather than 'to free, liberate':

(s) Die Polizei befreite die Geiseln. 'The police freed the hostages'.

The lexical decomposition account seeks to deal with applicative examples like (4)-(5) by making two crucial stipulations. First, it allows for phonologically empty morphemes. In the case of denominals, empty verbs with appropriate meanings, such as 'put', simultaneously host preposition- and noun-incorporation processes.3 In the case of deadjectivals, a phonetically unrealized causative morpheme (cAusE) and an unrealized inchoative morpheme (BECOME) combine with the appropriate adjectival predicates to derive the necessary input representation. Second, the input representations of deadjectival and denominal verbs are prohibited via stipulation from being lexicalized. One can take issue with this approach on three counts. First, it is ad hoc. Second, it captures the facts only by complicating the syntax-semantics interface. Third, it does not explain what be-prefixation contributes to the formation of deadjectival be-verbs, since 3

This is the Wunderlich ( 1 987) account It is not entirely clear from the paper, but Wunderlich

{ 1 997) may funher decompose verbs like

PUT

into CAUSE {BECOME (wcATEo) ).


the additional layers of abstract representation which the Wunderlich Brinkmann model requires for each class. One may be tempted to extend the functional-composition account of deverbal be-verbs to denominal and deadjectival be-verbs by assuming that a conversion mechanism derives simple verbs from the base adjectives and nouns before the regular preposition-incorporation mechanism applies. Even if we accept the necessity of rule ordering, such an account lacks sufficient empirical support. First, there may be no simple verb that could be analyzed as the result of noun-verb or adjective-verb conversion. Second, even when there is a homophonous candidate verb, it may not have the appropriate meaning. Consider, for example, the verb beschildern 'put up traffic signs' in (4). It can be paraphrased accurately only by the predicate aufitellen 'put up', taking the nominal argument Schilder 'traffic signs'. A simple verb schildern exists, but its meaning is 'describe', not 'put up traffic signs':

Laura

A.

Michaelis and Josef Ruppenhofer 345

the semantic representations of these verbs do not contain stative locative predicates. The incorporation account is problematic even for be-verbs whose meanings appear to be compositional, in the sense of having semantically aligned base verbs. Some of these be-verbs license arguments that do not belong to the subcategorization frame of the base verb. For example, the verb wachsen 'grow' does not subcategorize for a location argument. By contrast, its applicative counterpart bewachsen does require a location argument, as shown in (6):

Wunderlich ( 1 991: 6 1 4) concedes that '[t]o describe this [phenomenon] by functional application, we have to assume that the modifier4 turns into an argument first, and then this argument is incorporated.' In other words, it is necessary to stipulate an additional operation and to order it before the derivation of the be-verb. While this move may be undesirable only for reasons of parsimony, there is another class of be-verbs for which the Wunderlich account appears descriptively inadequate. These are be-verbs whose unpreftxed counterparts actually disallow, whether as argument or adjunct, the expression of the 'location' role licensed by the be-verb. For example, consider the contrast between mogeln 'cheat, swindle' and bemogeln, shown in (7)-(8), respectively:

(7) Peter hat (*mir) beim Kartenspielen gemogelt. 'Peter cheated (*me) in cards.' (8 ) Peter hat mich beim Kartenspielen bemogelt. 'Peter cheated me in cards.' While it is true that at the level of conceptual structure we may assume the existence of a victim of Peter's cheating in (7) , this is irrelevant to the derivation of the be-form. According to Wunderlich, argument shifting by morphological operation is only possible at the level of semantic form or at the level of thematic structure (Wunderlich 1 997= 52-3). At both of these levels, the location argument of mogeln would be absent. Moreover, there are deverbal be-verbs which license arguments that do not even belong to the conceptual structure of the unpreftxed base verb. Consider, for example, beregnen 'spray with water': •

'Modifier' is Wunderlich's term for an optional verbal argument


(6) Message ID ( 3 sfa4ec9.o©netnews.web.de) SELBSTKLIMMER = Kletterpflanzen, die mit speziellen Haftorganen Wande oder andere Flachen direkt bewachsen [ . . . ]. 'Self-climbers = climbing plants that directly grow walls or other surfaces with the help of special adhesive/sticky extremities.'


(9) Message ID ([email protected]) Damals ware um ein Haar ein Fhissiggastank in Mitleidenschaft gezogen worden [ . . . ] die Petershausner Feuerwehr mu6te diesen daher intensiv beregnen. 'Back then a tank full of liquid gas almost got damaged. The Petershausen fire department had to make a great effort to douse it.' The theta frame licensed by the unprefixed verb regnen 'rain' does not include an agent or cause of the precipitation, as shown in (10)-(I I):

Thus, we find a multitude of cases (including apparently straightforward cases) in which the input needed by the preposition-incorporation model is unavailable. This suggests strongly that applicative semantics does not arise from operations on verbs, but instead from the imposition of a particular argument-linking pattern on a wide range of input lexical items. 2. 3

Interpretative principles

Applicative predications appear to have several idiosyncratic semantico pragmatic properties. First, the goal or location argument is generally interpreted as being affected in a holistic manner by the action which the verb denotes (HoLisM) . Second, the theme arguments in transfer predications can always be omitted (NuLL COMPLEMENTATION). Third, the goal argument is always interpreted as a planar region rather than, say, a three-dimensional space (ExTERIOruTY). According to Brinkmann, these properties are not truly idiosyncratic. She argues that the first two are instead epiphenomena of more general syntactic and pragmatic principles and that the third follows from the meaning of the prefix be-, which she analyzes as a bound preposition meaning 'on, at'. In the next three subsections, we will give evidence against Brinkmann's analyses. 2. 3 . I Holism

The goal argument of a be-verb is construed as wholly affected by the action that the be-verb denotes. For instance, the sentence Die Kinder bemalten den Tisch 'The children be-painted/be-drew the table' evokes a scene in which there are drawings all over the table. This semantic


(Io) *peter regnete die Blumen mit der Gie6kanne. 'Peter rained the flowers with the watering-can.' ( I I) *Das Gewitter regnete drei Tage lang. 'The thunderstorm rained for three days.'

Laura A Michaelis and Josef Ruppenhofer 347

characteristic of be-verbs is not specific to the German locative alternation: in the English sentence John painted the table it is similarly inferred that John painted the entire table surface. Accordingly, Pinker (I989) speculates that direct-object encoding of the location argument universally serves to encode affectedness of the location and that this function makes the alternation learnable. However, Brinkmann shows convincingly that the affectedness implication is not common to all German be-predications. She points out, for instance, that it is not clear in what way the cake undergoes a change of state in ( I 2) :

( I 2) Donna bestreut den Kuchen mit Zucker.

7I

)

On the basis of such observations, Brinkmann concludes that affectedness understood as change of state cannot be a defining characteristic of German be-predications. What is it, then, that gives rise to speakers' intuitions of holism concerning be-predications? Brinkmann argues that it is Lohner's (I 990) PRESUPPOSITION OF INDIVISIBILITY.5 Lohner's definition of the pre supposition runs as follows: 'Whenever a predicate is applied to one of its arguments, it is true or false of the argument as a whole' (our translation). Applied to ( 1 2), the principle predicts that sprinkling with sugar must be true for all parts of the surface of the cake, since den Kuchen is the direct object of the verb bestreuen 'strew'. Nothing needs to be said about the final effect of the denoted event upon the cake. However, as Lohner observes, the indivisibility presupposition cannot be a general constraint on predication, if it does not also hold for sentences like Harry boxte Moe 'Harry punched Moe'. Under our current formulation of the Lohner model it is not clear what the right subregions of Moe would be. To account for these cases, Lohner suggests that we refine our under standing of possible partitionings of an argument. The refinement involves positing two groups of predicates. On the one hand, there are SUMMATIVE predicates. A summative predicate applies to an argument if it also applies to each of the argument's parts.6 To take Lohner's example, the sentence The children are playing could denote a situation in which each child is playing his or her own game, a situation in which groups of children play together, or a situation in which all children engage in a single game. On the other hand, there are INTEGRATIVE predicates. Unlike summative predicates, these do not apply to arbitrarily chosen parts of the argument; 5

Brinkmann ( I 997) uses the term

indivisibility to refer to what LOhner calls holism. We adopt the

newer terminology. 6

'Ein Pradikat P mit einem Anwendungshereich, in dem eine Teil-von-Relation definiert ist, ist

genau dann sumrnativ, wenn fiir aile i aus seiner Domane und fiir aile zulassigen Aufteilungen A von i gilt: P(i)

= I

gdw. (genau dann, wenn; JR&LAM] P(i 1

)

= I

fiir aile i 1 aus A' (LOhner

I 990:

25

).


'Donna sprinkles the cake with sugar.' (= Brinkmann (s p):


2. 3 .2 Theme omissibility

When a be-verb denotes a transfer scenario with an agent, theme, and goal, the theme can be omitted in surface syntax, when its referent is recoverable from context, as in ( q):

(1 3) Die Jugendlichen bespriihten die Wand (mit Farbe). 'The youths sprayed the wall (with paint).' Brinkmann argues that this fact is predicted by what she calls the NONINDIVIDUATION HYPOTHESIS. A theme is said tO be nonindividuated if it is an unbounded mass or plexity, and is thereby not INCREMENTAL, in the 7

'Integrative Pradikationen iibertragen sich nicht von dem Argument auf beliebige Teile davon,

in vielen Fallen sogar auf iiberhaupt keine echten Teile'

(Lohner

1 990:

25).


in many cases they do not apply to any parts at alF There is only one admissible partition of the argument, the zero-partition: the argument must always be viewed as an undivided whole. Now, based on the observation that the sentence The children are playing, which has a summative reading, also has a zero-partition reading under which all children engage in a single game, it is argued that integrative predicates are really j ust the limiting case of summative predicates. Understood in this way, integrative predicates fall under the indivisibility presupposition. If we assume, however, that all verbs fall into one of two classes of 'indivisible' verbs, we lack an explanation for the fact that the vast majority of be-verbs have summative readings with respect to their location arguments, even though they would also comply with the indivisibility presupposition if they had integrative readings. For instance, predications containing the verb beladen 'load' are summative: each part of, say, a truckbed must have a load on it. Yet there is no principled reason to presume that predications containing transfer verbs like beladen would not be integrative. Why couldn't putting anything anywhere on the back of a truck be an event that we could refer to by means of the verb beladen, since punching a person anywhere is punching that person? As LOhner himself points out, one cannot determine whether a predicate is summative or integrative by looking only at an argument: a person could be viewed holistically as the object of punch or as being composed of parts and regions as the object of bespritzen 'besplash'. Given that the semantics of the goal argument do not favor one kind of reading over the other, and given that both summative and integrative predications in principle comply with the indivisibility presupposition, we conclude that the applicative pattern itself must be what gives rise to the summative readings of be-verbs.

Laura A. Michaelis and Josef Ruppenhofer 349

( I 4) The direct object of a transitive locative verb may be omitted only when the quantificational properties of the corresponding argument are irrelevant; the argument may then be existentially bound. (Brinkmann I 997= I I 3 ) By 'direct object of a transitive locative verb' in ( I 4) we assume that Brinkmann intends that argument which would be the direct object of a transitive transfer verb if this verb were not subject to the applicative linking. That is, Brinkmann is referring to the theme argument. Examples like ( I 5) show that ( I 4) cannot be a sufficient condition upon omission of the theme argument. One cannot, for example, omit the theme object of a locative verb when the goal argument is linked to an oblique grammatical function.

( r 5) *Feter lud auf den Wagen.

'Peter loaded [something] onto the wagon.'


sense of Dowty ( I 99 I ). When an entity is an incremental theme, each part of that entity is mapped to a temporal subpart of an event. The theme argument in ( I J ), and other goal-object sentences, is nonincremental, i.e. nonindividuated. Properties of the theme argument do not determine the endpoint of the event described, since it is wall space and not paint whose gradual exhaustion defines the time course of the spraying event in ( I 3 ). Brinkmann provides a general motivation for the nonindividuation of the theme argument in examples like ( I 3) by assuming that such predications denote processes. Accomplishment verbs, like sing, which otherwise select for an incremental theme, yield processual (activity) readings when their theme arguments are deindividuated, as in I sang songs. By the same token, she argues, the applicative pattern renders the theme's quantificational properties irrelevant by yielding a processual reading of verbs like laden, which would not otherwise denote processes (p. 1 20). Examples of null object complementation, e.g. I read or She smoked, are given as support for the idea that 'there is a close relationship between an incremental theme's omission from object position and its construal as nonindividuated' (p. I I s ) . We concur with Brinkmann that nonspecific oblique themes of goal object constructions are nonincremental. We will, however, call into question two assumptions which she makes in her explanation of the nonindividuation hypothesis. The first assumption is that the omissibility of the theme is determined by semantics alone. The second assumption is that nonindividuation of the theme-and thereby its omissibility-comes about because applicative verbs denote processes. Let us now look at the way in which these two assumptions are expressed. Brinkmann states the non individuation hypothesis as follows:


2. 3 . 3 Exteriority

The only semantic fact about be-verbs which Brinkmann attributes directly to the prefix be- is the restriction that the goal location must denote the exterior of an object (pp. S r -2). This constraint is illustrated by the ill formedness of ( r 6') when intended as a paraphrase of (23). The only acceptable interpretation for example ( r 6) is that the seeds are thrown at the outside of the garbage can rather than inside it:


Thus, the literal interpretation of the nonindividuation hypothesis gen erates an incorrect prediction for unpreflxed transfer verbs. As we have seen, Brinkmann treats oblique status and omission as two mechanisms for nonsyntactic expression. However, it appears that oblique status and omissibility are not the same thing. Instead, omissibility of the theme appears to be dependent upon oblique status, and capturing this fact requires that one refer to a particular linking. There is nothing in lexical decompositional grammar which would obviously allow one to do so, yet Brinkmann's formulation of the hypothesis does just that by referring to direct object, a grammatical function, and quantiflcational properties, a semantic notion. Further, the deindividuation (and thereby omissibility) of the theme cannot plausibly be attributed to a processual reading of the verb. We concur with Brinkmann that accomplishment verbs like beladen entail preparatory processes. Since, however, nonapplicative transfer verbs like laden are accomplishment verbs as well, they should entail the very same process predicates. What then is the aspectual, basis for distinguishing applicative and nonapplicative transfer verbs? Further, as Herweg (1991) demonstrates, entailing a process and being a process are two different things. If this were not so, he argues, accomplishment predications would share with activity predications the entailment pattern which he refers to as the DISTRIBUTIVITY PROPERTY (i.e. the subinterval property as per Bennett & Partee 1 978). Accomplishment predicates demonstrably lack the distribu tivity property. In sum, we have seen that the quantiflcational properties of the theme argument do not provide a sufficient condition for the omission of that argument and that the omission of the theme argument in trivalent applicatives cannot plausibly be attributed to a processual interpretation. Insofar as this is the case, omissibility of the theme argument appears to come from a particular linking (to an oblique grammatical function), and not from interpretative principles.

Laura A Michaelis and Josef Ruppenhofer 3 s r

Be-, like other inseparable prefixes of German, lacks a corresponding free form that can be used as a preposition or particle. As Brinkmann points out, be-'s closest living relative is the preposition bei, which is related to English by and has a meaning of 'by, close, near, at'. In none of its uses does bei involve contact between a theme and a landmark. Therefore, be-verbs cannot be formed-synchronically at least-by incorporating bei. Brinkmann observes, however, that be-verbs can be used to paraphrase unprefixed verbs with goal complements that are encoded either by a uf 'on (a horizontal surface]' as shown in (17), or by an 'on (a vertical surface]', as shown in ( 1 8). Both of these prepositions have the needed meaning element 'contact with a surface'.

( 1 7) Ted schmierte Butter auf die Tischdecke. 'Ted smeared Butter onto the tablecloth.' (1 7') Ted beschmierte die Tischdecke mit Butter. 'Ted smeared the tablecloth with butter.' ( I 8) Petra hangte Sterne an den Christbaum. 'Petra hung stars onto the Christmas tree.' ( I 81) Petra behangte den Christbaum mit Stemen. 'Petra hung the Christmas tree with stars.' Accordingly, Brinkmann postulates a preposition be with the same predicate-argument structure as the prepositions an and auf 'on, onto' The only difference among these prepositions is that be is a bound morpheme which occurs only as a prefix. This analysis is historically plausible, and coheres with the assumptions of the preposition-incorporation account, but its attractiveness is diminished by its idiomaticity-a particular lexical entry is created solely for the purpose of preserving a compositional account.


(16) Message ID ([email protected]# 1 / 1 ) Bin auch fur Kernk:raft! Oberlegt einmal, wieviel Kerne wir taglich in den Miilleimer werfen, ausspucken oder verschlucken, ohne ihre Kraft zu nutzen. 'I am for nuclear power, too! Just think how many seeds (lit. nuclei) we throw into the garbage can, spit out or swallow every day without using their power.' (161) Bin auch fiir Kernk:raft! Oberlegt einmal, mit wievielen Kernen wir taglich den Miilleimer bewerfen, wieviele wir ausspucken oder verschlucken, ohne ihre Kraft zu nutzen. 'I am for nuclear power, too! Just think with how many seeds a day we throw the garbage can, how many we spit out or swallow without using their power.'

3 s 2 Valence Creation and the German Applicative

But even if we ignore the issue of idiomaticity and assess at face value Brinkmann's model of be- as a bound version of an or auf, we face difficulties. For if the meaning of the bound preposition be- corresponds to that of the prepositions an and auf, why then are be-verbs not synonymous with an- or aufverbs derived from the same base? Consider the following pair of sentences. While the be-sentence in (19) implies coverage of a large portion of the tablecloth, the sentence in (2o), containing the separable prefix an-, does not. ( 1 9) Ted beschmierte die Tischdecke mit Butter. (= ( 1 71) ) (2o) Ted schmierte die Tischdecke mit Butter an.

3

THE C O NSTRUCTIO NAL APPRO A C H T O T HE GERMAN APPL I C AT IVE

What would it mean to adopt a constructional rather than a lexical model of the be-pattern? We will address this question in steps. section 3 . 1 will describe the nature of the challenge that Construction Grammar has offered to the principle of lexical licensing, discuss the constructional model of argument structure and set forth the constructional account of the German applicative pattern. This account rests on the assumption that the


If one assumes with Brinkmann that both verbs are formed by the same process of preposition incorporation, that the prepositions involved are closely aligned semantically, and that syntactic principles such as indivisi bility hold, then one would predict that (2o) means the same thing as (2 1), or at least that the two sentences do not differ in ways that are controlled by grammatical principles. Yet this is exactly what we find: only the applicative sentence in ( I 9) has the expected holistic reading. This latest finding should come as no surprise. We have seen on several occasions that if one grants each of the premises of Brinkmann's analysis, the resulting account still does not explain all of the facts. The indivisibility principle, for instance, by its very generality fails to predict that be-verbs have summative readings rather than integrative readings. Similarly, the preposition-incorporation model fails to extend to be-verbs like beregnen, in which the 'output' arguments are not present in the 'input' representation. These facts suggest strongly that the semantic and syntactic features of be-predications do not come from general operations and principles, but rather from a specific formal pattern to which specific semantic constraints are attached. In the next section, we will describe a syntactic theory which accords a central place to such patterns.

Laura A. Michaelis and Josef Ruppenhofer 3 s 3

valency of verb and construction may differ. In sections 3.2-3.6, we will apply the constructional model to five features of the applicative pattern that proved troublesome for the Brinkmann account. In section 3·7· we will address two interrelated questions: to what extent does the constructional model adhere to Jackendoff's ( 198 3) Grammatical Constraint and to what extent can it be described as compositional? Finally, in section 3.8, we will discuss the formal representation of applicative sentences within Construction Grammar.

3.1

Theory overview Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The principle of lexical licensing holds that the basic scene denoted by a sentence (the set of participant roles expressed) derives from the argument structure of the head verb. Thus, for example, it appears clear that a sentence like We gave the account to her denotes a scene of transfer involving an agent, a theme and a goal because the semantic frame associated with the head verb give denotes a scene of transfer, and likewise requires the presence of these three participants. This principle is intrinsic to a compositional theory of semantics-a theory which has been seen as central to any account of syntax-semantics isomorphism, including that of Jackendoff, who states (1 990: 9): 'It is widely assumed, and I will take for granted, that the basic units out of which a sentential concept is constructed are the concepts expressed by the words in the sentence, that is, lexical concepts.' A more recent version of this principle is stated by Jackendoff as the principle of syntactically transparent composition: 'All elements of content in the meaning of a sentence are found in the lexical conceptual structures of the lexical items composing the sentence' ( 1997a: 48). The lexical-licensing principle has been central to the description of argument structure in most formal theories. Many such theories (e.g. Lexical Functional Grammar as described by Bresnan 1 994 and Role and Reference Grammar as described by van Valin & LaPolla 1 997) posit universal linking rules, which capture generalizations concerning the syntactic realization of thematic roles assigned by verbs or verb classes (e.g. the class of transfer verbs). Such theories are driven by the assumption that 'argument roles are lexically underspecified for the possible surface syntactic functions they can assume' (Bresnan 1 994: 91). Universal linking rules map these argument roles to grammatical and pragmatic functions, and these rules do not add to, subtract from or alter the array of thematic roles associated with the verb. For example, in Bresnan ( I 994), locative inversion in English and Chichewa is represented as one linking possibility for verbs of location like stand, which subcategorize for locative and theme


arguments. Such verbs are subject both to the linking rule which produces the configuration in (2 I) and to the linking rule which produces the configuration in (22):

(2 I) Two women stood in the plaza. (22) In the plaza stood two women. Examples of locative inversion like (23) are, however, problematic Bresnan's framework:

m

(2 3) Through the window on the second story was shooting a sniper.

(24) Most likely they were fellow visitors, just panting up to the sky-high altar out of curiosity. (Lindsey Davis, Last Act in Palmyra, p. 28) (2 5) As they had waved us along the raised causeway and into. the rocky ' cleft [ . . . ]. (op. cit., p. 3 I) (26) They can't just analyze away our data. Goldberg points out that on the assumption that argument structure is determined exclusively by head verbs, we would need to posit a new verb sense for each of the usages exemplified in (3 I )-(3 3). Sentence (24) would require a special sense of pant equivalent to the formulation 'move while panting'; sentence (25) would require a special sense of the verb wave whose definition would be 'signal permission to move to a place by waving'; and, finally, sentence (26) would require one to view analyze as a verb which denotes (metaphorical) caused motion. Such word senses, as Goldberg points out are not only ad hoc and unintuitive, but also compatible only with an assumption of radical and unconstrained polysemy. Crucially, as Goldberg and Fauconnier & Turner ( I 996) have demonstrated, examples like (24)-(26) cannot easily be viewed as marginal or special cases. Sentence


Sentence (2 3) is problematic because the verb shoot assigns neither a locative role or a theme role, and yet can appear in the locative-inversion configuration. In such examples, Bresnan argues, a locative-theme argu ment structure imposed by the pragmatic requirement of presentational focus is superimposed on the argument structure associated with the unergative verb shoot. The agent role of shoot will consequently be identified with the 'overlay theme' (p. 9I). The problem with this type of account is simply that it is not explicit. If argument structures are products of the linkings licensed by given verbs, and not independent form-meaning pairings, it is difficult to understand the source of the 'overlay theme'. Adherence to the lexical-licensing principle results not only in ad hoc devices like the 'overlay theme' invoked by Bresnan (I 994) in cases like (2 3), but also, as Goldberg points out ( I 995: 9ff), appeal to implausible verb senses. Goldberg discusses examples like the following:

Laura A Michaelis and Josef Ruppenhofer 3 5 5

(27) OvERRIDE PRINCIPLE. If lexical and structural meanings conflict, the semantic constraints of the lexical element conform to those of the grammatical structure with which it is combined. Zwicky (1989) proposes a similar universal interactional principle, which he relates to Panini's Law, since it involves the specific taking precedence over the general: 'Requirements in an evoking rule override those in an invoked rule' (p. 3 8). In acknowledging the applicability of (27) to cases like (24)-(26), we embrace the view that linking patterns are meaningful-that is, that they contribute schematic semantic structure distinct from that contributed by the verbs with which those patterns combine. As grammatical constructions, linking patterns are complexes of formal, semantic, and pragmatic features. On this view, we would expect linking patterns to exhibit idiosyncratic constraints. For example, we would expect that semantic constraints above and beyond those which restrict the theta frame of the input verb would be relevant for determining the applicability of a given linking pattern. These idiosyncratic constraints might include constraints on the pragmatic role or topological properties of certain arguments. A prominent topological restriction on the semantics of the applicative pattern was discussed in section 2.3: be-predications, as Brinkmann observes (1 997: 8 1) 'describe motion to the exterior of an object'.


(24), for example, exemplifies a lexicalization pattern-conflation of manner and motion-which Talmy ( 198 5) and Slobin ( 1 997) have shown to be strongly entrenched in Germanic languages. Further, the examples in (24)-(26) cannot be regarded merely as violations of selectional restrictions associated with the verbal heads-or even as violations which might trigger manner-based implicata. If, for example, (26) merely exemplified a violation of the selectional restrictions associated with the verb analyze, we would fail to predict its well-forrnedness-let alone the uniformity of its interpretation across speakers; (26) is necessarily interpreted as denoting metaphorical caused motion. Cases like (24)-(26) give strong evidence that the principle of lexical licensing, despite providing a parsimonious account of transparent cases like (2 1)-(22), is invalid. The alternative, construction-based model of argument structure outlined by Goldberg ( 1995) is founded on a body of work, of which Talmy (1988) is representative, which focuses on universal dif ferences in the inventory of concepts expressed by open- versus closed-class elements, and in particular on the nature of the semantic interaction between grammatical and lexical elements. Crucially, grammatical con structions are viewed as belonging to the general set of meaning-bearing grammatical elements, which includes prepositions and derivational markers. An essential tenet of these works is expressed in (27):


Among the linking patterns considered by Goldberg ( I 995) are the ditransitive pattern (whose core semantics she captures with the formula 'x ' CAUSES Y TO RECEIVE z ) , the caused-motion pattern ('x CAUSES Y TO MOVE WITH ' ' RESPECT TO z ) and the resultative pattern ('x CAUSES Y TO BECOME z ). Examples of each of these patterns are given in (28)-(30): (28) We gave her the account. (29) She put the checkbook on the counter. (3o) We painted the walls white.

(3 I) We painted them a landscape. (32) She blew the dust off the picture. (3 3) We cried our throats ragged. The verb paint, a verb of creation, denotes a two-place relation, involving the creator and a created item. However, sentence (3 I ), an instance of the ditransitive linking pattern, adds an additional participant to the creation scenario-a potential recipient. This recipient is not intrinsic to the creation scenario; it is instead intrinsic to the transfer scenario with which the ditransitive pattern is associated. Likewise, while the verb blow is a one place relation, involving an agent, (32) adds two additional participants-a theme and a goal. These participants are licensed by the caused-motion construction which the sentence instantiates. Finally, in (3 3), the verb cry appears with two more participants than it ordinarily has-a patient and a resultant state. The additional participants are contributed by the resultant state construction that licenses (3o). The examples in (3 I )-(3 3) strongly resemble the examples in (24)-(26), which were used to undermine the validity of the lexical-licensing principle. Both sets of examples involve the override principle given in


Goldberg uses the term sentence type to refer to these linking patterns. In accordance with Fillmore & Kay (I997= Ch. 8), however, we will regard linking patterns not as sentence structures but as verb-level constructions, which unify with the lexical entries of verbs. This unification has the effect of augmenting what Fillmore & Kay refer to as the MINIMAL VALENCE of the verb (the repertoire of semantic roles inherent to the meaning of the verb). The FULLY SPECIFIED verbal valence which results from unification of a verb's lexical entry with one or more linking constructions is one in which each semantic role is assigned a grammatical function. A crucial assumption of Goldberg's account, which is adopted here, is the idea that the repertoire of thematic roles assigned by the linking construction may PROPERLY INCLUDE the repertoire of thematic roles in the verb's minimal valence. In (31)-(3 3) we give examples of proper inclusion for each of the linking patterns exemplified in (28)-(30):

Laura A Michaelis and Josef Ruppenhofer

3 57


(27). We can regard linking patterns like the ditransitive and caused-motion patterns as CONCORD constructions. The theta frames associated with these patterns may, and indeed typically do, match those licensed by the particular verbal head. Examples of concord, given in (z8)-(3o), are those which provide the motivation for the lexical-licensing principle. Goldberg (1 997) refers to these kinds of examples as instances of INSTANTIATION, in which the verb codes a more specific instance of the scene designated by the construction. The verb may also code the MEANS by which the action designated by the construction occurs (Goldberg 1 995, 1 997). Examples of the means relation are given in (3 r )-(3 3 ), in which, respectively, blowing is described as the means by which the dust is moved from one location to another and crying is the means by which the hoarseness is effected. The means and instantiation relations are mutually exclusive. The means relation is operative only when the theta frame associated with the construction properly includes that of the verb. We view the be-pattern, whose formal representation will be given in section 3 .8, as a transitive linking pattern like those exemplified in (3 1)-(3 3). Full consideration of its semantic and linking constraints will be delayed until section 4; it is sufficient here to say, as in the introduction, that it denotes thorough coverage of a location by a theme. This general semantic scenario is compatible with two more specific scenarios, which are minimal variants of one another: a trivalent causative scenario, in which an agent is present along with locative and theme, and a bivalent scenario entailing only theme and locative. The two versions of the construction differ with regard to their valency: the trivalent licenses the theta frame (agent, theme, locative) whereas the bivalent licenses the theta frame (locative, theme). While the bivalent version instantiates the coverage scene, the trivalent version entails that scene, in that the latter scene includes an agent which effects coverage of the location by the theme. This semantic intersection is reflected in the sharing of single linking constraint: the locative must be encoded by a direct (nonoblique) grammatical function. This situation is parallel to that described by Michaelis ( I 99 3) for Latin, in which entailment relationships between situation types, e.g. those of removal and lacking, are reflected syntactically in a shared linking constraint: the theme is ablative or genitive. We will use the trivalent version of the be-pattern to exemplify the semantic interaction between construction and verb. Following Goldberg (I995: so), we will use the term FUSION to refer to the mechanism by which interpreters infer coreference relationships between arguments of the construction and participant roles assigned by the verb, where the latter are more specific instances of the former. We will assume, also in accordance with Goldberg ( 1 995: 6 5), that fusion is constrained by the

3 s8

Valence Creation and the German Applicative

Shared Participant Condition: at least one argument role of the construction must be fused with a participant role assigned by the verb. The verb may bear a means relation or an instantiation relation to the event type denoted by the construction. Sentence (34) is an example of instantiation, while (3 5) is an example of the means relation: (34) Sie belegte ihre Pizza mit Salami. 'She put salami on her pizza [as a topping].' ( 3 5) Die Stadt will das Gelande mit Einfamilienhausern bebauen. 'The city intends to build up the site with single-family homes.'

3 .2

Null complementation

As discussed in section 2.3, the omission of the theme argument requires certain grammatical conditions-not merely semantic ones. Under a construction-based account of argument structure, these conditions can be represented in a straightforward way. Each construction defines constraints on null complementation. In the case of the oblique-goal pattern exemplified by nonapplicative transfer verbs like laden, we follow Fillmore & Kay ( 1 997) in assuming that null complementation is licensed


Example (34) exemplifies the instantiation relationship: the theta frame licensed by the trivalent verb legen is identical to the constructional theta frame. Example (3 s ) exemplifies the means relationship between the bivalent verb bauen and the trivalent constructional theta frame. In this instance of the applicative construction, the agent role of the construction is fused with the builder role of bauen and the theme role of the construction is fused with the patient (factitive theme) role of bauen. The location role is unfused, since it is assigned by the construction alone. In the resulting predication, the verb, which denotes the creation of a structure, simul taneously denotes the means by which coverage (of the site) is effected. The reconciliation of verb and constructional semantics during interpretation requires the inference that multiple buildings have been built, since only on this understanding is coverage entailed. By assuming a constructional account of argument structure, we account not only for such reconciliation effects but also for all other features of be predications which are not proper to the input verbs. These are precisely the noncompositional features of be-predications that proved troublesome for the Brinkmann account. For each of these features, we will consider the alternative account offered by a construction-based model. The phenomena to be considered are: null complementation, valence augmentation, valence creation, the exteriority constraint, and the holistic-goal constraint.


3 59

by a trivalent linking construction which pairs a locative argument with an oblique grammatical function. In the case of the be-pattern, null instantia tion is licensed by a trivalent linking construction which pairs a theme argument with oblique. The constructional account of null complementa tion is actually implicit in Brinkmann's statement of the Nonindividuation Hypothesis, presented in section 2. Despite the fact that Brinkmann claims to have provided a general constraint governing omissibility of themes, this constraint is not general but particular-it concerns a particular pairing of syntax and semantics which we will describe in section 3.8 as the OBLIQUE THEME CONSTRUCTION, a construction which unifies with the applicative construction.

Valence augmentation

As described in section 2.2, an alternation-based account cannot plausibly represent be-predications for which no 'source' or 'input' verb exists. One case of this nature is that in which the be-pattern 'adds' arguments that are not part of the theta frame of the unprefixed verb. One such case is that of bemogeln 'cheat', discussed in section 2.2. An attested example illustrating both prefixed and unprefixed uses of mogeln in a single passage is given as (36):

(36)

WKD/bza.oo1 4 I ,

Berliner Zeitung/89.1 2.07/s:3. W. Schwanitz: Es ist betrachtlich. Prozentzahlen wiirden das verdeut lichen, die kann ich aber nicht nennen. Kein Geheimdienst dieser Erde tut das oder er mogelt. Ich will Sie nicht bemogeln. 'W. Schwanitz: It is a considerable number. Percentages would under score this but I cannot give them. No secret service on earth does that, unless they are cheating. I don't want to cheat you.'

In the last sentence of (36), Ich will Sie nicht bemogeln, a malefactee argument which is not otherwise expressible appears as the direct object. An additional case of valence augmentation was discussed in section 2.3 with respect to (6), repeated here as (3 7): (37) Message ID (3 [email protected]) SELBSTKLIMMER = Kletterpflanzen, die mit speziellen Haftorganen Wande oder andere Flachen direkt bewachsen [ . . . ]. 'Self-climbers=climbing plants that directly grow walls or other surfaces with the help of special adhesive/sticky extremities.' In (3 7) the object function is linked to an argument whose status is that of a locative adjunct within the valence frame of the unprefixed verb (wachsen).


3·3

3 60


3 ·4

Valence creation

The limiting case of valence augmentation is that in which the head of the be-predicate is not a valence-taking element in the lexicon. These are cases of valence creation, as exemplified by deadjectival and denominal be predications. While deadjectival and denominal be-verbs are highly proto typical instances of the be-pattern, the Brinkmann analysis, as we saw in section 2.2 above, is forced to posit phonologically null verbs in the input semantic representations of these applicatives-a move that requires 8

For fusion to take place, the relevant roles of verb and construction need not always be identical;

they may instead be merely compatible, as in the case of befa e ('be-drive'), illustrated in (41 ) The supplies an agent. bivalent version of the be-construction calls for a theme, while the verb

hr n

.

fohren

However, since this verb denotes directed motion, the agent is also a theme, and therefore fusion of the verb's agent argument and the construction's theme argument is straightforward-the subject denotatum can easily be construed as both agent and theme.


The final case of valence augmentation discussed in section 2.3 combines both aspects of valence augmentation exemplified by bemogeln and bewachsen: the direct object represents a participant role which would otherwise have adjunct status and the subject represents a participant role which is not licensed at all by the unprefixed verb. Sentences (9)-(I I ) were used to illustrate this case for the verbs regnen and beregnen in section 2.3. The cases of bemogeln, bewachsen, and beregnen all prove problematic for an alternation-based view of argument structure, in which linking rules effect changes in the syntactic expression of some set of argument roles, but otherwise conserve thematic structure. These cases receive a straightforward and motivated account under the constructional view. On the constructional account, as described above, the valence set licensed by a linking construction may properly include that licensed by the verb with which that construction combines. Where verb and construction assign identical argument roles, as in the case of the subject argument in (3 s ) , the two roles simply fuse.8 Where the theta frame of the construction contains a thematic role or roles NOT licensed by the theta frame of the verb, there is override as per the principle (27) : the argument roles of the construction are 'added' into the verb's valence set as verb meaning and construction meaning are combined. What this means in the case of a sentence like (37) is that the theta frame of the verb wachsen intersects with the theta frame of the construction via the means relation. The predicate-argument structure resulting from the integration of verb and construction meaning contains a participant which denotes the location covered by means of growth.

Laura

A.

Michaelis and Josef Ruppenhofer

361


recourse to abstract constructs of dubious validity and the stipulation that the input semantic representation cannot be lexicalized. On the construc tional account, valence creation comes about through the same highly general mechanism that underlies valence augmentation. Let us consider again the examples of valence creation (4)-(s) discussed in section 2.2. Valence creation differs from mere valence augmentation in two respects. First, cases of valence creation involve a form -class override: the nominal or adjectival syntactic feature of the input element is overridden by the verbal syntactic feature of the construction in accordance with (27), the override principle. Second, cases of valence creation do not involve fusion of participant roles in the input and constructional theta frames, quite simply because the open-class element which combines with the applicative construction in cases like (4)-(s) has no theta frame. Instead, the repertoire of roles contributed by the input item are participants in the larger semantic frame which constitutes our understanding of the socio-cultural context in which the property or entity plays a role. By semantic frames we have in mind the schemas Fillmore (1977, 1 982, 198 5) uses to represent lexical semantics and which underlie his contention that all linguistic meaning is 'relativized to scenes' (I97T 5 9). For example, in (4) the nominal Schild is meaningful only relative to a schema which includes streets and streetworkers. In ( s ) the adjective frei is meaningful only relative to a schema which includes captives and liberators. By combining with the applicative construction, the noun not only receives a valence structure, but also an event construal which is compatible with the semantics of the applicative construction. In other words, the event denoted by the predication is one involving coverage of a location by a theme. For example, beschildern in (4) denotes the activity of placing signs at regular intervals along the bike path. The question that arises here is precisely how we capture a crucial felicity condition identified by Clark & Clark ( 1 979) upon the use of denominal verbs: the source or 'parent' word must denote one thematic role in the situation, while the remaining surface arguments of the denominal denote other roles in that situation. How do we ensure, for example, that the parent noun of the verb beschildern, Schild, is taken to be the theme of the coverage event? Our solution, which will be implemented in section 3.8, is to reframe this question by rejecting the assumption that the parent nominal in fact fills the theme role. The parent nominal is not referential, and therefore could not be said to refer to any particular participant in any given event. Instead, we claim, the parent nominal renders the actual theme argument RECOVERABLE, and thereby omissible as per the oblique-theme construction, which allows null instantiation of theme arguments as discussed in section 3.2. Evidence for this claim comes from the fact that any theme argument whose identity the


interpreter might NOT readily recover from the category denoted by the parent noun may be syntactically present, as in (38): (3 8) 4809664: Die Berkersheimer Grundschule in der U ntergasser Hohl wird mit zwei Tafeln beschildert, die in der StraBenverkehrsordnung gar nicht vorgesehen sind. 'The Berkersheim Elementary School on Untergasser Hohl Street will be signed with two signs that are not even part of the traffic regulations at all.'

3·5

Exteriority

The constructional account of the exteriority constraint discussed in section 2.3.3 with respect to the construal of beweifen in (1 61) is straightforward. Rather than appealing to a constraint imposed by the prefix, whose meaning, as discussed in section 2.3.3 would be construction-specific anyway, we view exteriority as constraint on the configuration of the theme element: it must be planar. The claim that this constraint belongs to the construction is substantiated by override effects. Example ( I 6') provides an example of such an override, but perhaps the most cogent example of an exteriority-based override effect comes from attested uses of the verb bifiillen 'fill'. This applicative verb at first appears to violate the planar surface constraint in that it necessarily describes an effect upon the INTERIOR of a three-dimensional space. However, we notice that attested uses of bifiillen have iterative readings, as in (39)-(4o): (39) Frankfurter Allgemeine Zeitung, August 7, 1 990 AuBerdem miiBten Betriebe, die Mehrwegflaschen bef'tillen, eme plotzliche Erhohung ihrer Pfandriickstellungen bewaltigen.


These examples strongly suggest that it is the skeletal structure of the construction alone, and not any aspect of lexical meaning, that licenses the interpretation of these sentences as involving transfer and the particular reading of (3 8) as involving multiple transfer events. A plausible account of valence creation requires the recognition of both 'bottom up' contributions to meaning (lexical meaning) and 'top-down' contributions to meaning (constructional meaning). The interpretation of denominal verbs like beschildern involves not only the top-down imposition of valence structure from the construction but the bottom-up importation of the rich frame semantics of the nominal element with which the construction combines.


'Moreover companies that fill returnable bottles would have to cope with suddenly having to raise their reserve for deposits.' (4o) Siiddeutsche Zeitung, July 25, 1 994. Dariiber hinaus werden im Jemen 5 Milliarden US-Dollar fur eine Erdgas-Verfliissigungsanlage bei Aden oder Mukalla fallig, urn dort die Tanker nach Japan zu befiillen. 'Moreover 5 billion US dollars will become due for a natural gas liquefaction plant near Aden or Mukalla that serves to fill the tankers to Japan.'

3 .6

Holism

On the account offered here, the source of the holism constraint is the semantics of the be-construction, rather than any general constraint upon the application of predicates to their direct arguments. As discussed in section 2.J.I, we reject Brinkmann's claim that the holism effect can be attributed to Lohner's Presupposition of Indivisibility. There we observed that this principle alone cannot explain why applicative predications have summative rather than integrative readings. We also depart from Pinker's account of the locative alternation, since we do not attribute the affectedness implication of be-predications to any general principle govern ing the construal of direct objects. Instead, we view the holism effect as entailed by the situation type denoted by the be-construction: saturation of a surface. The semantics of the be-construction entail coverage of location by theme at a given point in time or over the course of time. 3 ·7

Concreteness and compositionality

The constructional model is based on the sign: constructions are form meaning pairs which differ from words only in internal complexity. Like


In (39), the iterative reading comes from a construal in which bottles are repeatedly returned and filled. In (4o), the iterative reading comes from a construal in which tankers are repeatedly emptied and filled. The iterative reading arises in each case via the override principle (27), which requires that the filling scene denoted by the verb be reconciled with the planar coverage scene denoted by the construction. The 'compromise construal' is one in which an exterior surface is affected: the iterated filling events, insofar as each occupies a different pair of coordinates, collectively define a planar region over which coverage is effected. Further discussion of the relation of the exteriority constraint to the various senses of the construction will take place in section 4·

364


(A] more constrained theory is only as good as the empirical evidence for it. If elevated to the level of dogma (or reduced to the level of presupposition) so that no empirical evidence can be brought to bear on it, then it is not being treated scientifically (r 997a: so).

Jackendoff's response is a reminder that empirical criteria must take precedence over the theory-internal criterion of parsimony. But we have an additional response, which Jackendoff, an advocate of enriched composition but not of a construction-based syntax, does not give: the constructional model is in fact compositional, although not in the standard sense. If the meaning of a sentence is the result of integration of verbal and construc tional semantics in accordance with the override principle of (27) then that meaning results from semantic composition. In other words, we do not abandon a constrained theory of sentence meaning by acknowledging the existence of 'top-down' or constructional meaning. The mechanism of unification ensures that sentence meaning is the result of constrained combination of symbolic structures.


Montague Grammar, Construction Grammar pairs surface structure with a semantic representation; no 'deep' level of semantic representation intervenes between these two levels. Thus, the constructional model does not rely on either lexico-semantic 'transformations' or multiple types of semantic representation (like those representations which are input to L-command and functional application in the Brinkmann-Wunderlich model). In addition, the constructional model does not rely upon abstract (phonologically null) elements; denomi nal verbs, for example, are not represented as arguments of phonologically unrealized verbs but as nominals which assume valence structures by virtue of unification with argument-structure constructions. The constructional model is concrete, and therefore conforms well to Jackendoff's Gramma tical Constraint-a constraint on semantics which, in Jackendoff's words, 'serve(s) to make semantic theory responsible to the facts of grammar' (r98 r r 8 ). Since syntax encodes propositions related to events and states of affairs, we must assume that it does this in an efficient and relatively transparent way. It therefore makes sense to base syntactic theory on the assumption that argument-structure patterns directly express basic-level scenes. But in attributing meaning to syntactic patterns in the interest of concreteness, we potentially undermine the fundamental purpose of syn tactic theory: to describe sentence meaning compositionally. If the con ceptual content of a sentence now comes not only from lexical conceptual structures but also from the syntactic patterns that contain those lexical items, we sacrifice a constrained model of semantic composition. Like Jackendoff, we question the assumption that this model must necessarily be preserved. As Jackendoff observes:


3.8

365

The formal representation of the applicative pattern


In this section, we will give only a brief sketch of the applicative construction and the manner in which it combines with other independ ently motivated constructions to license applicative sentences. A detailed treatment of the model being described here can be found in Michaelis & Ruppenhofer (in press: Ch. 4). This model, which is based upon unification, is sufficiently straightforward to permit an intuitive explanation here. Unification of constructions can best be described in terms of a metaphor involving the superimposition of slides. Any slide (construction) can be superimposed upon any other as long as the semantic and syntactic specifications on each slide 'show through'-that is, provided there is no conflict among the specifications on the slides in the stack. All linking constructions operate upon the valence set specified by a given lexical item. The number of valence elements specified by a given linking construction may be greater or less than the number of valence elements specified by the lexical verb. One can think of the linking constructions as being superimposed in sequence upon a given verb's lexical entry (although in reality the interacting constructions apply simultaneously). The lexical entry contains a minimal valence, i.e. an array of thematic roles, whose grammatical expression is determined by the linking construction or constructions applied. A minimal lexical entry which is unified with linking constructions is said to be a fully specified lexical entry: one in which every thematic role supplied by the lexical entry is linked with a grammatical function. We depart from the Fillmore & Kay account of argument-structure unification only in one regard: we allow the override of the syntactic category of an open-class lexical item for cases in which nouns and adjectives unify with the applicative construction, a verb-headed construction. This move seems to be the only way that we can capture the observed form-class fluidity without resorting to the assumption of unconstrained polysemy that was criticized in section 3 . 1 . Linking in the Fillmore & Kay account is incremental, in the sense that each linking construction determines the grammatical function of only a single theta role, although all unifying linking constructions will denote the same event type and thereby have identical theta grids (e.g. the trivalent theta grid associated with transfer). Thus, the applicative construction merely constrains the syntactic realization of the locative thematic role. As described by Michaelis & Ruppenhofer (in press: Ch. 4), the grammatical realization of the remaining thematic role or roles will be determined by those constructions with which the be-construction combines, including the passive, active, subject and oblique-theme constructions. Figure 1 shows the

3 66


[

cat syn orm f sem

V

Be -

prefix

l

[ sem planar region val [role theme] , [ g obi ]

[COVER < THEME, LOCATION >]

PRED < >

Figure

val 1

lI

The applicative construction, bivalent type

cat syn orm Be -pre ix f f sem V

8

l

[CAUSE - COVER ]

PRED < >

[role agent], (role theme],

Figure

[:: : ] I pl

g on b:o ; [ locatiOn] 8

2 The applicative construction, trivalent type

bivalent version of the be-construction. Figure 2 shows the trivalent vers10n.9 As shown in Figure 2, the trivalent pattern denotes a causative event involving a placeable object. This event may be telic (as in the case of an accomplishment construal) or atelic (as in the case of an activity construal). This analysis of caused change of location differs from those advocated by e.g. Rappaport Hovav & Levin (1 998) and Hale & Keyser (1 993). These authors assume that such causative events entail a change of state in which the theme remains at the location as the culmination of the event. As we will show in section 4. 1 , however, trivalent applicative predications may be used in situations in which the objects moved do not remain at the location state, and in which the location is accordingly unchanged. The bivalent applicative construction has an even greater degree of aspecutal neutrality than the trivalent version of the construction. As shown .

9

A more extensive representation of the constructions in Figures

1

and

2

would show the

unification requirement between elements in the event structure denoted by the construction and elements in the valence set of the construction. We have avoided showing unification indices in order to simplify the diagrams to the extent possible.


[

I

1

role flocati�On


367

1 0 A reviewer has questioned the necessity of positing two applicative constructions, suggesting

that the two may be combined into a single constructioiL While considerations of parsimony would

suppon such a move, we see a compelling reason to distinguish two versions of the applicative constructioiL Since the situation type denoted by a linking construction must be of determinate valency, we would have to propose either a bivalent or trivalent version of the applicative constructioiL This fixing of valency would give us an undesired result

If,

on the one hand, the

proposed applicative construction were trivalent, there would be no construction to license bivalent applicative predications, and we would incorrectly predict the existence of trivalent applicatives formed from bivalent verbs like

wandern, 'wander'. If, on the other hand, the proposed applicative laden ('load'), since the agent

construction were bivalent, it would not unify with trivalent verbs like argument would fail to receive syntactic expressioiL


in Figure 1 , the situation type denoted by the bivalent pattern is under specified with respect to the static-dynamic contrast, since bivalent applicative predications may either denote states or events. The under specification of aspectual information in these two constructions reflects the fact that, as we will see in section 4, the coverage schema expressed by each is compatible with stative, processual (activity), and accomplishment construals. Thus, the transitivity of the argument-structure pattern entails nothing about telicity or even perfectivity. This is a position which will be defended more thoroughly in section 4· 1. 1 0 We postulate that the two versions of the applicative construction are related via an INHERITANCE LINK. Inheritance networks, as per Lakoff (1 987), Goldberg (1995), Michaelis & Lambrecht ( 1996), and Jackendoff (1 997b) are used to capture relationships between linguistic signs when these relation ships are not sufficiently productive to be represented as rules, but are nevertheless entrenched connections within an associative memory. The inheritance model is similar to that described by Pinker & Prince (1991: 2 3 2) as the connectionist model of memory. This model is 'both associative and superpositional: individual [linguistic] items are dissolved into sets of features, and similar items . . . overlap in their physical representations, sharing representation real estate.' In accordance with the CG tradition as established by Lakoff and Goldberg, among others, we will represent these overlap relations in terms of links in a hierarchical network, where a dominated construction inherits all nonconflicting specifications from the dominating construction. In the case of the two versions of the applicative construction, we propose that the bivalent version is related to the trivalent version by means of what Goldberg (1995) and Michaelis & Lambrecht (1996) call a SUBPART LINK: the trivalent construction subsumes the semantic representation and syntactic (linking) constraints of the bivalent construc tion. The analysis of trivalent be-predications involves the OBLIQUE-THEME construction, given in Figure 3· The oblique-theme construction has as its semantic value an event type in which an agent causes a theme to cover a location (the goal). Although we have not shown the relevant unification

368


syn [cat V] sem

l

11

CAUSE - COVER < AGENT, THEME, LOCATION >

�role [ :ft::�e ] val [8 agt], [8 location], l syn [mit]/zero P

Figure

3 The oblique-theme construction


indices, the agent, theme and goal of this event type will unify with those same thematic roles in the valence set of the lexical verb. The event type denoted by the oblique-theme construction is identical to that denoted by the trivalent applicative construction, ensuring unification between the two constructions. The only linking constraint imposed by the oblique-theme construction is this: the theme argument links to the oblique grammatical function, which is realized either as a prepositional phrase headed by mit or as a null (pragmatically recoverable) complement. We have said nothing in the statement of the oblique-theme construction about the construal of the theme as nonindividuated. We have chosen to remain agnostic concerning the appropriate representation of the nonindividuated construal, since it is a potential rather than a requirement. As Brinkmann and others have observed, an individuated construal of the theme is compatible with applicative semantics, as in Erna bekochte zehn Manner mit einem einzigen Truthan ('Erna be-cooked ten men with a single turkey'). The oblique-theme construction unifies with denominal applicative verbs like beschildern, as in (4). Here, the identity of theme argument can be reconstructed on the basis of the semantics of the lexical form with which the various linking constructions unify. This lexical form is a nominal; it is only by virtue of unifying with an argument-structure construction-in this case the trivalent applicative construction-that the nominal receives a theta frame. In accordance with the Gricean principle of omission up to recoverability (modulo constructional constraints), a theme argument which is not recoverable from the type denoted by the verb, owing to the greater specificity of that theme argument, will be syntactic ally realized, as in the English example He soled my shoes with gum soles and the German example given in (3 8). Our model of the interaction between a verb and a given version of the applicative construction, bivalent or trivalent, relies upon the distinct manner in which verbs and constructions ensure realization of thematic structure. In short, verbs rely upon constructions in ways that constructions do not rely upon verbs. For the verb, realization of thematic structure

Laura A. Michaelis and Josef Ruppenhofer 3 69

11

How do we ensure that each verb unifies with the appropriate version of the applicative construction? As it stands, there is nothing to prevent a trivalent verb like 'load' from unifYing

laden

with the bivalent version of the applicative construction. Although the agent of

laden would not fuse

with any role in the theta frame of the applicative construction, it would be linked to the SUBJECT grammatical function by default Subject Principle, since the theme and location are each subject to more specific linkings-oblique Theme and applicative, respectively. We also have no obvious way to prevent a bivalent verb like

wandern 'wander' from unifYing with the causative

(trivalent) version of

the applicative construction. In such an instance, the applicative would merely contribute an agent to the verb's valence and require nonoblique expression of the location. The theme would be subject to the oblique Theme linking, and the agent would again receive subject coding by the default Subject Principle. By allowing such unification, we overgenerate, since

bewandern e.g.

could not be used to

denote causation of coverage. It appears therefore that we must stipulate an optimization principle whereby the verb unifies with that version of the applicative whose valency is closest to its own.


means ensuring syntactic expression of each subcategorized argument through fusion: each role licensed by the verb must be identified with a role licensed by a linking construction. For the construction, realization of thematic structure means ensuring syntactic expression of each argument, whether or not that argument fuses with an argument licensed by the verb. What this means is that the construction's arguments may be expressed by maximal categories (e.g. NP) which are not part of the subcategorization frame of the verb. For the bivalent applicative construction, this situation is exemplified by predications containing bemogeln ('cheat'): the 'malefactee' argument is contributed by the construction in Figure 1 . This construction also constrains the grammatical function which can be assigned to this role: it must be either subject or object (to be determined by which of the two voice constructions, passive or active, unifies with this applicative con struction). For the trivalent applicative construction, verbal-valence aug mentation is exemplified by the verb bebauen ('build up'). As a verb of creation, bauen licenses two arguments, agent and theme. The construction both adds a goal (location) argument to the verbal valence set and restricts the grammatical function to which the location argument can be linked, as described above. Whenever the valence set licensed by the construction properly includes that licensed by the verb, the verb bears a means relation to the event type denoted by the construction. 1 1 The PRED variable in Figures r-2 is used in accordance with Goldberg (1995) to represent the open-class element with which the construction unifies. The angled brackets to the right of the PRED variable represent the theta frame licensed by the open-class element. As described in section 3 .4, the open-class element with which the be-construction unifies may be an inherently nonrelational element which thereby lacks argument structure. For example, the open-class element may be a noun, as in the case of beschildern 'put up traffic signs' and besohlen 'sole, as of a shoe'. As we argued in that section, appropriate thematic elements can be found in the rich background knowledge with which the particular word is associated. This

370


semantic frame contains participant roles, and these are the roles which will be 'plugged into' the argument set of the PRED variable. In the case of the verb besohlen, for example, this set will contain the cobbler, the sole and the shoe. These frame-specific roles are required to fuse with the more schematic argument roles licensed by the applicative construction. For this reason, we assume that denominal verbs necessarily have an elaboration relationship to the applicative construction.

4

THE SE M A N T I C S O F THE A PPL I CATIVE P ATTERN Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Like Brinkmann, we argue that be-verbs form a coherent semantic category. However, under our account this semantic coherence does not arise from the interaction of syntactic and general semantico-pragmatic principles. Rather it is seen as a reflection of the polysemy structure associated with the applicative construction. In accordance with Goldberg's (1995) analysis of the English ditransitive and other argument-structure patterns, we postu late that the meanings of the be-pattern are related via independently motivated patterns of semantic extension, and represent a radial category of senses. As in Lakoff's ( 1 987) description of radial-category structure in classifier systems, the polysemy structure at issue here allows for the cancellation of implications associated with the central sense. We will see that certain senses invoke components of the coverage scenario, like transfer, without entailing coverage. We will also see an example of chaining within the network: one extended sense shares semantic content with another extended sense but not clearly with the central sense. The exposition of senses associated with the applicative pattern will focus upon specific (and in many cases partially overlapping) classes of verbs (e.g. verbs denoting iterated activity). A question that arises is how this mode of description can be reconciled with our central contention-that the semantic effects which distinguish be-predications from their paraphrases are attributable to the semantics of the applicative pattern rather than to the semantics of a particular set of verbs. In other words, if we view the meanings of be-predications as the products of a reconciliation procedure whereby the meaning of the verb is brought into conformity with the meaning of the construction, how can we also treat the be-verbs as 'stored', i.e. listed, elements? The answer to this question requires us to reject a principle which Langacker ( 1 987) has called the RULE-LIST FALLACY. Langacker applies this term to the (typically implicit) principle which holds that complex structures that can be modeled by an on-line process cannot also be viewed as stored. For example, with regard to morphology,

Laura A Michaelis and Josef Ruppenhofer 3 7 I

12

Within the class of be-verbs denoting thorough discussion, there are, however, relatively recent

additions. For example,

besprechen

'discuss' seems to have lacked the 'discuss' sense in Middle High


Bybee ( I 99S ) rejects the idea that the products of regular morphology are exclusively generated online. On the basis of experimental work by Losiewicz ( I 992), Bybee argues that high-frequency regular past-tense forms are stored in the lexicon. By the same token, we view high-frequency be-verbs as clustering in narrowly defined lexical classes (like those listed by Gunther I 97 4) that reflect the (conventionalized) patterns of semantic extension observed for the applicative construction. However, we simul taneously maintain that the interpretation of be-predications involves the integration of constructional meaning and verb meaning. Were this not so, productive uses of the applicative pattern could not plausibly be modeled (see section 3·4 for argumentation on this point). The discussion will proceed as follows. In section 4· I , the central sense will be described. In section 4.2, we will describe three classes of be predications which involve metaphorical extensions of this basic meaning. In section 4·3 · we will look at five classes of be-predications whose meanings relate to the central sense via various inductive inferences. In section 4-4, we will consider the relationship of this semantic analysis to a diachronic analysis of the be-pattern given by Ruppenhofer ( I 999). In particular, we will consider the relationship between the meaning of the be-pattern as we have described it and earlier uses of the be-prefix to denote ENCLOSURE and PROXIMITY. This exposition of the usages of the applicative pattern requires a disclaimer: the order of presentation of the verb classes (that is, the usages of the construction) is not intended to reflect any avenue of development of these senses. Instead, we view each sense extension as exploiting a semantic potential inherent in the semantic schema which represents the core sense; no extended sense is viewed as dependent on any other. While there is evidence (to be discussed in section 4-4) which suggests the historical primacy of the coverage sense of the pattern, we cannot on the basis of this evidence propose a relative chronology of the extended senses. Further, we do not intend to suggest that metaphorical extensions of the applicative pattern are recent innovations. Metaphorical extensions, like the use of the applicative pattern to describe thorough discussion of a topic, are old. This semantic extension seems to be at least as old as Middle High German, as suggested by a look at the entries for bereden 'discuss', besprechen 'discuss', beklagen 'complain, mourn', besehen 'look at, examine', beschauen 'look at, examine', beschreiben 'describe', besingen 'sing about, of', and bedenken 'think about, reflect on' in Lexer's (r872) Middle High German dictionary and in the available installments of the Early Modern High German dictionary (Gobel & Reichmann 1 999). 1 2 The radial model of

3 72


4 The meanings of the applicative pattern

senses which we will offer, in which each sense of the pattern overlaps semantically with the core sense, is consistent with the diachronic facts and provides a plausible model of what might make the various senses cohere synchronically. Figure 4 shows the associative network of meanings which will be discussed in the forthcoming subsections.

4. 1

The prototype

The conceptual archetype with which the applicative pattern is associated is a scene in which a theme occupies multiple points within two-dimensional space. 1 3 We have represented this schema by the predicate-argument structure cover' (theme, location) in the representations of the bivalent and trivalent applicative constructions in Figures 1 and 2. As shown in those German but meant only ' 1 . agree on reciprocal 13

sich]'.

2.

talk to, address 3· accuse of 4· consult, confer [used with

The analysis presented here does not extend to what we may call pseudo be-participles, i.e.

forms that appear to be participles of be-verbs but that lack extant base forms. For instance, in the case of beamtet 'being state appointed [e.g. as a teacher, judge, etc.]' the corresponding verb,

beamten,

has beco.:ne obsolete. Such ornative applicative participles are analogous to uses of the English past participle to convey attributes, e.g.

long-haired.


Figure


-

(41 ) MK1/MHE.ooooo, HEUSS, ERINNERUNGEN I 90S- I 9 3 3 · Memoiren Nun hatte ich wohl die Ostsee befahren und die Nordsee geschmeckt [ . . . ]. 'True, I had sailed around the Baltic Sea and I had had a taste of the North Sea.' (42) 367 1 3673 Jugendliche Straftater wiirden bewuBt Sitze zerstoren, Fenster heraus treten, Wandverkleidungen beschmieren oder von den Wanden reiBen. 'Youth offenders would purposely destroy seats, kick out windows, smear wall coverings or tear them off the walls.' Sentence (41) describes extensive travel over the Baltic Sea while (42) entails that significant portions of each wall hanging are smeared. However, the notion of saturation at stake here does not require that an entire surface is covered: some parts of each wall hanging may have been spared and areas of the Baltic Sea may not have been reached. Moreover, while the location may often be affected in the sense that it undergoes a noticeable change of state, this is not a necessary consequence of saturation: the Baltic Sea is not changed by the sailor's travels in (41). Finally, notice that saturation may be summed over time, since, for example, the sailor can only occupy a single location at any given time. The type of construal is equivalent to SUMMARY SCANNING, a mode of cognitive processing in which co activation of scanning events produces a coherent gestalt (Langacker


figures, this predicate-argument structure is the semantic value of the bivalent version of the applicative construction while it is entailed by the causative semantic representation of the trivalent applicative pattern. We intend this predicate-argument structure as a pointer to a semantic analysis, rather than as a semantic analysis per se. As we will argue in this section, the semantic structure of the applicative construction is an image schema, in the sense of Lakoff (1 987) and Langacker ( 1987), and as such is not highly amenable to a propositional representation. We have seen, for example, that, as observed by Brinkmann (1 997), the semantic structure of the applicative pattern is not appropriately represented by a formula involving universal quantification over subparts of the location argument. By the same token, as we will see in this section, applicative semantics cannot be captured by any particular Ak:tionsart-based representation. In what follows, we will elaborate upon the image schema which underlies the concept of coverage that we have in mind. We will being this exposition by looking at the bivalent and trivalent instantiations of this schema, illustrated in (4 1 ) (42 ), respectively:

374


(4 3 ) MMM/ I 02.3 700 I : Mannheimer Morgen,

Leserbriefe; Die alteingesessenen Altriper, die den Ortskern his jetzt noch bewohnen, werden sich eben mit dem noch starker werdenden Durchgangsverkehr abfinden miissen. 'The long-time Altriperians who up until now are still inhabiting the town center will just have to get used to the increasing through traffic.' 09.02. I 99 I ,

The coverage concept associated with bewohnen is also compatible with a habitual construal, in which a single inhabitant effects coverage of a given location over time (44a). As shown in (44b), a bewohnen predication is anomalous when the location (in this case, a city) is too large to allow an individual to effect coverage over time; the prepositional paraphrase in (44c) is, however acceptable, as the coverage constraint is not operative here: (44) a. Ekkehard bewohnt ein Apartment in Berlin. 'Ekkehard inhabits an apartment in Berlin.' b. *Ekkehard bewohnt Berlin. 'Ekkehard inhabits Berlin.' c. Ekkehard wohnt in Berlin. 'Ekkehard lives in Berlin.' Within the class of perfective applicatives, both accomplishment and processual (activity) readings are attested. These two readings are closely aligned semantically, since accomplishments entail processes. Nonpunctual


I 98 7 : I 44- 5 ) . In the case of the sailing activity the individual scanning events establish distinctness of sailor and water surface and track the change in relative position. If all the scanning events are overlaid at once (in the way transparencies can be overlaid), there will be only one water surface but many positions occupied by the sailor. This overlay config uration is identical to the coverage schema denoted by the applicative construction. Notice that while we have used the terms configuration, schema, scene, and archetype in this section to describe the saturation constraint, we have not referred to an event type. This may appear puzzling, since the constructional model of argument structure upon which we base this analysis is one in which argument-structure patterns denote basic-level event structures. We have avoided the term event structure or event type here because these terms suggest that the coverage schema can be characterized as belonging to a particular aspectual class. Coverage is a topological notion, and has no temporal dimension. Accordingly, the saturation constraint may be satisfied by both perfective and imperfective predications. In (4 3 ), for example, a be predication containing the stative verb bewohnen ('inhabit') denotes a situation of coverage which holds at a single point in time:

Laura A Michaelis and Josef Ruppenhofer

375

perfective verbs (e.g. walk) typically have both activity and accomplishment readings, and the same can be said of bivalent applicative sentences like (4I), which have both telic and atelic readings. If the interpreter views the sailing activity as having culminated in thorough coverage of the body of water by the sailor's craft, the predication denotes an accomplishment. If instead the interpreter construes the sailing activity as a set of subevents with no inherent point of culmination, the predication denotes an activity (as expressed by the processual gloss 'sail around'). Because these two readings are available, both types of durational adverbials are licensed, as shown in (44):

.

1

c

Since bivalent applicatives denote both processual and static situations, the coverage schema in our framework cannot be equated with the concept of an incrementally interpreted location argument. Trivalent applicative predications, which are necessarily perfective, generally have accomplish ment readings, as in (4 5 ):

{ in*dre1drei. StunStunden } den lang } ,She loaded the wagon { in*Crorthreethreehours hours

(4 5 ) Sie belud den Wagen

In such cases our analysis overlaps with Brinkmann's analysis (as described in section 2.3.2): the location argument is incremental in the sense of Dowty ( I 99 I), since its 'exhaustion' determines the time course of the event. However, trivalent applicative predications do not always require telic construals and do not always permit them. The telic construal appears to be required only when the base verb is telic (contingent upon a bounded construal of the object-denotatum), as in the case of laden. The trivalent applicative predication in (46), which contains the atelic base verb werfen ('throw'), does not have a telic interpretation:

{ d*inre1.dreiStunStunden } m1t. Stemen beworfen. den lang hours } , { *inror three . Secunty rorces were pel ted with stones three hours .

. . (46) S1eherhe1tskrafte wurden ,

.

c

c


{ indre1drei Monaten } . . Monate l ang { rorin three months } , . ,They sa1 ed the Carn. bean three months .

1 (44) S1e b esegel ten d'1e Kan'b'k


(47)

3 6 8 9 1 726

Allein die Gelnhauser Stadthalle habe das Trio zweimal beklaut, berichtete ein Kriposprecher am Donnerstag. 'A police spokesman reported on Thursday that the trio robbed (be robbed) the Gelnhausen City Hall twice.'

4.2

Metaphorical extensions of the central sense

The coverage schema associated with the basic usage of the applicative pattern is compatible not only with concrete physical situations but also abstract ones. Various metaphorical links allow the use of the coverage schema in the domains of speech, perception, and attention. In accordance with Lakoff's ( 1 990) lnvariance Hypothesis, by which ontological compo nents of a basic-level semantic schema are conserved by metaphorical extensions of that schema, we can observe that these metaphorical mappings


Numerous acts of transfer occur in the course of the event denoted by (46) , and yet there is no point at which transfer is complete. Thus, while the trivalent applicative construction denotes a transfer event, insofar as an agent causes a theme to move to a location, this event type has both telic and atelic instantiations. Accordingly, the term tranifer event, like the term coverage, should not be taken to entail that an endpoint is reached. The trivalent applicative construction is unmarked with respect to telicity, and only predicate-argument structure fixes aspectual class. In proposing that the applicative pattern is aspectually neutral, we counter the recent analytic trend toward viewing argument-structure variation as aspectually driven (Rappaport Hovav & Levin 1 998). Since numerous other linking construc tions are aspectually neutral, e.g. the transitive and passive constructions, we feel it is appropriate to regard aspectual variability and valence variability as reflecting related but orthogonal categories of event structure. As pointed out earlier, entailments shared by two situation types may be manifested in common morphosyntactic coding. Transfer has a cognate concept, removal, which is expressed by the trivalent applicative pattern as well. In the case of transfer and removal, the shared entailment is causation of change of location, and languages like Latin appear to neutralize the direction of transfer, coding the two event types identically. For example, trivalent verbs like compleo ('fill') and privo ('strip') license the same case frame, with the possibility of either ablative or genitive coding for the oblique theme argument (Michaelis 1 99 3 ). In German, removal verbs coded by the applicative pattern typically denote robbing, e.g. bemopsen, berauben, and beklauen. An example involving the verb beklauen is given in (47):


replicate the thematic and linking structure associated with the bivalent applicative construction. We treat these metaphorical usages as distinct (conventionalized) senses of the be-pattern because otherwise there would be no way of predicting which theoretically eligible metaphors would be expressed by the be-pattern and which would not. In the following three subsections, we describe three metaphorical extensions. 4.2. 1 Seeing is contact with the percept

(48) Sie konnte ihre Augen nicht von ihm nehmen. 'She wasn't able to take her eyes off him.' (49) Er richtete seinen Blick auf das Buch. 'He turned [directed] his eyes to the book.' (so) GRr /TL r . r2 1 50 Weyden, C.: Traume sind wie der Wind. Hamburg 'lch hoffe es.' Und wieder ging sein Blick zu dem anderen Tisch hiniiber. ' "I hope so". And another glance went over to the other table.' Verbs, prepositions, and particles that are used to designate a trajector's movement to a physical object can be extended metaphorically to designate a metaphorical trajector's movement across the field of vision (cast a stone at-cast a glance at; point a weapon at-point one's eyes at). According to this model, inspection of a percept is thorough coverage of the percept by the perceiver. In be-predications that denote vision, the eyes are construed as a theme moving over the percept. The following be-predications instantiate this particular conception: (s r) BZK/Ws9.00790, WE 07.09·59, S.o6, LESERBRIEFE Niemand sollte sich den Gang zum alten Museum ersparen. Man muB einmal Zeit genug haben, den Isenheimer Altar des Meisters Matthies zu beschauen, den ganzen Riesenaufbau in all seiner Wucht und Farbentiefe. 'Nobody should avoid the walk to the old museum. For once, one has to spend enough time to examine (lit. look) the Isenheimer altar of master Matthies, the whole gigantic body in all its massiveness and depth of color.' (52) MKr /LFH.ooooo, FRISCH, HOMO FABER, Roman. Suhrkamp Spater auf Deck auBerte Sabeth (ohne Drangen meinerseits) den Wunsch, einmal den Maschinenraum zu besichtigen, und zwar mit mir [ . . . ]


One's gaze goes from one's eyes to what one sees. One sees whatever one's gaze touches (Lakoff I 98T 437). These metaphors underlie German usages like the following:

378


'Later on deck Sabeth expressed the wish (without me urging her) to tour (lit. sight) the engine-room, and to do so with me [ . . . ]'

(5 3) Message ID ([email protected]) Als Nichtraucher behaupte ich, daB sie nicht "stinken". Immerhin hatte ich auch schon Gelegenheit, Raucherinnen aus ziemlicher Nahe beriechen zu konnen. Ergebnis: sie stinken nicht. 'As a [male) non-smoker I say that they [smokers) do not "stink". At least, I have had the opportunity to sniff female smokers from rather close distance. Conclusion: they don't stink.' 4.2.2 Attending to something is directing one's attention to it

Conventional examples of this metaphor are given in (54)-(56): (54) Er richtete seine Gedanken auf das Thema. 'He directed his thoughts to the topic.' ( 5 5) Wir kommen mit unseren Gedanken zum Ausgangspunkt zuriick. 'We are bringing our thoughts back to our starting point.' (56) Wohin gehen Ihre Dberlegungen? 'What are you thinking of (lit. Where are your thoughts going to)?' Via this metaphor, tracking or monitoring a percept can be viewed as maintaining contact with it across a set of space-time coordinates, where the emergent configuration involves coverage of the region defined by that cluster of points. This conception underlies bivalent denominal be predications like the following, in which, as in the vision case described in section 4.2.1, the cognizer and the cognizer's focus of attention are conflated and jointly fuse with the role of theme:


Although one might presume from the description of the metaphor that the agent maps to the perceiver and the theme to the gaze, the two roles are not obviously distinct, and we have found no be-predications in which the gaze/theme maps to an oblique expression. Insofar as the perceiver is difficult to separate from his or her perception, the agent and theme roles appear to conflate, and only the bivalent version of the applicative construction is involved. The verbs in (5 I )-(52) have an instance relation to the applicative construction, since the perceiver and percept arguments map to theme and location, respectively. The joint fusion of the perceiver and the means of perception with the theme argument is a general characteristic of examples involving perception, including those which involve coverage of the percept via other sensory modalities, e.g. olfaction, as in ( 5 3):


(57) Der Polizist beobachtete den Verdachtigen. 'The policeman watched (lit. observationed) the suspect.' (58) Peter muB seine kleine Schwester beaufsichtigen, wenn seine Eltern zur Arbeit sind. 'Peter has to look after (lit. supervisioned) his little sister when his parents are at work.' (59) Peter hat den Unfall verursacht, weil er die Vorfahrt nicht beachtete. 'Peter caused the accident because he didn't pay attention to (lit. attention) the right of way.'

Other be-predications express metaphorical mappings via the TRAVEL META PHOR: mental activity and conversation are both movement through some metaphorical space, the space being identified with the subject-matter of thought or speech (Sweetser 1 987, 1 990). We find this travel metaphor in German usages like (6o): (6o) Auf dieses Thema miissen wir noch einmal zuriickkommen. 'We will have to come back to this topic.' In these metaphorical usages we find verbs, particles, and prepostttons extended from their original domain of physical movement to the domain of movement in speech or thought. Such metaphorical situations may be expressed by be-predications when the saturation implication is prominent, i.e. when the theme (the conversant) covers the location (the topic) comprehensively. Compare sentence (6 I ), which describes a serious discus sion, to sentence (62), which describes the conversational efforts of two previously unacquainted people on a first date: (6 r) Rr /TL1 .09008 de Groot, B.: Dein Vater wird uns liebgewinnen. Nebenan befand sich Olga Gorenkamps Nahzimmer, die Tiir war nur angelehnt. Jetzt besprachen sie eingehend die neue Situation. Und sie war auBerst giinstig fiir das Weingut. 'Next door was Olga Glorenkamp's sewing room. The door was ajar. Now they were discussing (lit. speaking) the new situation in detail. And it was extremely favorable for the winery.' 14 1 4 We view the adverb

discussion rather than

as

eingehend

('in detail')

as

sympathetic to the implication o f thorough

inducing that implication, since this implication would be present whether

or not this adverb were present.


4.2. 3 Discourse is travel over a topic


(62) MK1 /L]A.ooooo, JOHNSON, DAS DRITIE BUCH UBER ACHIM Karsch sagte ja. Er war erstaunt, daB sie gebeten hatte. Sie verabredeten sich fi.ir einen Abend, an dem ein FuBballspiel viele Fahrzeuge aus der Umgebung heranholen und dann wieder auf die Autobahn schicken wiirde, sie trafen sich vor dem Theater, sprachen i.iber die Unterschiede der heiden deutschen StraBenbilder, fuhren los. 'Karsch said yes. He was surprised that she had asked. They made an appointment for an evening when a soccer game would attract a lot of cars from the surrounding area and send them back on the highway later. They met in front of the theater, spoke about how different the street looked in the two Germanies, and then drove off'

4· 3

Other extensions of the central sense

In addition to the metaphorical uses of the central senses discussed in section 4.2. there are extensions of the central senses in which the coverage semantics is either missing or of secondary importance. The majority of these extensions appear to be the result of pragmatic inferences like those described by Hopper & Traugott's ( 1993) as examples of PRAGMATIC STRENGTHENING, a metonymic inference mode by which a semantico pragmatic 'side effect' of some signification is elevated to the level of a distinct meaning, which may lack entailments of the source meaning (see also Konig & Traugott 1 988). An example of pragmatic strengthening is the development of concessive or adversative markers from markers of temporal persistence like still; the newly developed marker is usable in perfective predications, where no temporal continuation is implied, as in She still got angry (Konig & Traugott 1982). In the following four subsections, we will describe four meanings of the be-pattern which appear to involve this metonymic mode of inference: transfer, iteration, intensification, and affectedness. A fifth subsection will consider a class of denominal be-verbs expressing social roles, which appear to involve the affectedness meaning. The first class of cases, that of applicatives denoting metaphorical transfer (4.3. I), deserves comment here because it is subject to dual categorizations: it is both a metaphorically based extension and an inference-based one. Since this class involves a metaphorical extension, thematic structure is isomorphic to that of the applicative construction, as in the metaphorically


The relationship between thorough discussion and coverage of a surface is evident as well in the English contrast between talking something over and talking about something.

Laura A. Michaelis and Josef Ruppenhofer 3 8 r

4. 3 . 1 Communication and affecting as transfer

Applicative predications may describe situations in which metaphorical objects are transferred to a goal or recipient. This sense can be viewed as an extension of the meaning of the trivalent be-pattern, in which saturation comes about through transfer of a concrete theme onto a location, e.g. behiingen 'be-hang', beladen 'be-load', bedecken 'be-cover'. In its metaphorical transfer sense, however, the be-construction does not entail saturation; transfer is the sole entailment. One kind of metaphorical theme is an idea. Via the IDEAS ARE TRANSFERABLE OBJECTS metaphor and the CONDUIT metaphor of communication (Reddy 1 979), the be-construction can be used to describe communicative events. The agent is a person delivering the idea, the idea is the theme, and the recipient is the goal. The be-construction evokes this metaphor in the following denominal example: (63) Message ID (22D3 CsE8Hooooo2AFH©p-alv.wds.mcnet.de) Er hatte niemals daran gedacht, sie zu fragen oder sie auch nur zu benachrichtigen, daB er Clarisse fiir verschwunden hielt. 'He had never remembered to ask her or to even inform her (lit. news) her that he believed Clarisse to have disappeared.' The metaphor is made explicit in a paraphrase of (63), sentence (64): (64) [ . . . ] oder ihnen auch nur Nachricht dariiber zu geben, daB er [ . . . ]. '[ . . . )or to just give them news about the fact that he [ . . . ]'


based extensions discussed in section 4.2. As per the lnvariance Hypothesis, predications which denote metaphorical transfer preserve the thematic and linking constraints associated with the (trivalent) applicative construction: they entail an agent, a nonoblique goal and a (canonically null-instantiated) theme. However, this metaphorical extension is based upon a prototypical rather than necessary component of the coverage schema: transfer. For this reason, the coverage entailment which plays a role in other metaphorical uses of the applicative pattern plays no role here. The remaining prag matically based extensions discussed in this section are nonmetaphorical. They are based solely on marginal implications of the applicative pattern, and do not partake directly of the semantics of location or transfer. These extensions thereby lack the thematic and linking constraints associated with the applicative pattern. These extensions meet only the valency and morphological conditions on applicatives: they are bivalent and thereby compatible with the transitive linking construction described in Michaelis & Ruppenhofer (in press).


Another metaphorical link to the domain of transfer is provided by the EFFECTS ARE TRANSFERRED OBJECTS metaphor (Goldberg I 995: Ch. 6). Consider the following denominal example: (65)

In this example, the source nominal (Friede) denotes a transferable effect. Transferable effects include properties, as expressed by deadjectival applicative predications like (66) (= (s) ): (66) Die Polizei befreite die Geiseln. The police freed the hostages.' These denominal and deadjectival applicative verbs lack base forms. Therefore, it is plausible to assume that their meanings come from the integration of constructional and lexical semantics via the elaboration relation, in much the same way that a transfer implication attaches to denominal applicatives which express coverage of a surface, e.g. besohlen ('be-sole') and beschildern ('be-signpost'). In both the literal (coverage) cases and the metaphorical cases in (6s)-(66), the lexical item which unifies with the construction is construed as a transfer verb via the override principle, and the oblique theme is null instantiated owing to its recoverability. The examples in (67)-(68) show that the transfer of abstract effects is a special case of a general model in which effects of all kinds, including physical ones, are transferable from an agent onto a patient-goal: (67) Message ID ( 782aug$grs$ I @infosunz.rus.uni-stuttgart.de) Jeder will dem Gegner so schnell wie moglich eine tiefe blutende Wunde beibringen, ihn 'abstechen', wie es im Jargon der schlagenden Verbindungen heifk 'Each wants to bring the opponent a deep bleeding wound as fast he can, that is 'stick' him, as they say in the jargon of dueling fraternities.' (68) Message ID (4tskos$n4m©ra.ibr.cs.tu-bs.de) Schaden konnen diese Bucher einem nicht zufugen. These books can't do any harm (lit. add harm) to people.'


WKB /TZ1 .oos67, taz (Sonderheft I und 2), Nach Polen und Ungarn . . . Versuchte die Partei mit ihrem anfanglichen unverbindlichen Dialo gangebot vergeblich, den demonstrativen Unmut der Bevolkerung einzudammen, so hofft sie jetzt, die Konflikte am runden Tisch zu kanalisieren und das Volk zu befrieden. 'If the party tried in vain to contain the ostentatious ill humor of the population with their initially non-binding offer of a dialog, it is now hoping to channel the conflicts at the round table and to bring peace to (lit. peace) the people.'

Laura A Michaelis and Josef Ruppenhofer 3 8 3

4. 3 .2

Iteration

Iteration is a frequent concomitant of concrete uses of the central sense. We saw that predications involving saturation typically involve multiple instances of a given action. For example, predications involving beladen ('be-load') often express scenarios in which many items are loaded in succession. However, the notion of iteration can also be expressed by be verbs independently of the transfer or saturation implications. Example (69) illustrates this:

This examples illustrates the use of the verb ziehen ('pull, draw') in the applicative pattern to denote regular reception of goods or funds (as when one subscribes to a newspaper or receives retirement income). Example (7o) presents another case in which an applicative predication prominently expresses iteration: (7o) MMM/9 1 2.44101: Mannheimer Morgen, OJ. 12.1989, Sonstiges Puppen aus Porzellan mit echtem Lockenkopf und zarten Sommer sprosschen gehoren fiir 700 Mark in jede Schicky-Micky-Kinderstube, zumindest eine Kathe-Kruse, trotz des bauerlichen Gesichtsausdruck ein Prestige-Objekt fiir hohere Tochter muB her, obgleich vielleicht eine kitschige Barbie im iiberladenen Nylon-Abendkleid viel mehr geliebt und bespielt wird. 'For 700 marks porcelain dolls with real curly hair and delicate freckles should be part of every fancy-shmancy playroom; for daughters of the upper-class it has to be at least a Kaethe Kruse-an object of prestige in spite of the rustic physiognomy-although a kitschy Barbie doll in a pretentious nylon evening dress might be loved and played with (lit. played) much more.' In example (7o) the predication containing bespielen does not entail that a given doll might become worn out and ragged as a result of playing. It is also difficult to detect a coverage implication here, since a doll is not a surface which one can cover by means of playing, as in the predication containing bespielen in (76) below. Instead, what is relevant is the frequency with which the child is likely to play with the Barbie doll. Examples like (7o) illustrate the circumstances under which the iteration implication otherwise a happenstance concomitant of the coverage implication con-


(69) 8 805 296: Es kann nicht angehen, daB auch auf kommunaler Ebene Wahlbeamte schon mit vierzig eine Pension beziehen. 'It cannot be the case that even on the municipal level elected offlcals be-draw a pension already at the age of 40.'

384


ventionally associated with the applicative pattern-comes to be the sole implication expressed by the be-pattern. For example, the iteration implica tion is the sole feature responsible for contrast pairs like hindern ('stop, prevent') vs. behindern ('hinder'). By assuming an iteration use of the be-pattern, we account for a usage of the verb bifahren which otherwise appears to violate a robust constraint on the applicative pattern: the location must be two-dimensional (see section 3 -5)· Use of bifahren to describe car travel appears to violate the planar location constraint, because the locative argument (the roadway) is a one dimensional rather than two-dimensional location. Example (7 r) illustrates this usage:

The use in (7 r ) is unexpected because it invokes a construal distinct from that associated with the applicative predication involving the sailor and the Baltic Sea in (4r ) above: in (7 r) there is no two-dimensional location for the driver (and vehicle) to thoroughly cover. The applicative predication in this example certainly does not lead one to construct a scenario in which a driver is 'covering' the road by swerving her car to the left and right. Nor is there any notion that the driver is driving with particular intensity. Similarly, it is not implied that the road is affected more heavily by be-driving than by some other kind of driving. Rather what is crucial for the use of bifahren in example (7r) is that driving represents an iterated activity. 1 5 One may, for instance, be-drive a particular road to work every day. That iteration is crucial to the use of befahren to describe car travel is suggested by fact that the majority of German speakers whom we consulted appear to have a strong preference for (72) over (7 3): (72) ??Ich befahre heute die A3. 'I'll take/drive on highway A3 today.' (73) Ich fahre heute die A3. 'I'll take/drive on highway A3 today.' 1 5 There is, however, a use of

bifahren which lacks the iteration implication. It is found primarily

in reports on traffic accidents which specify the direction in which the driver was driving on the road traveled. An example of this usage is the following:

4J 2 1 0 ! 8

Einem Bad Nauheimer Aurofahrer, der am Freitag gegen 1 4.40 Uhr die Wetteraustrafk in

Dorheim in Richtung Wolfersheim befuhr, lief ein achtjahriges Madchen direkt vor den Wagen.

'An 8-year old girl ran right in front of the car of a driver from Bad Nauheim who was driving

Wetteraustreet in Dorheim towards Wolfersheim.'


(7 r ) Hans befahrt diese Strasse taglich. 'Hans drives this street every day.'


385

4· 3 . 3 Intensification

Numerous be-predications share the notion of intensive action but do not involve coverage of any surface. This intensification sense may be attributed to a pragmatic inference of the following kind. Many be-predications entail repetition of subevents. The repetition of subevents is evidence that the activity is carried out with greater intensity than in the case of comparable isolated events. While we lack diachronic evidence for a path of grammaticalization along these lines, it is plausible in the light of other findings. For instance, Regier ( 1 994), in a typological study of the semantics of reduplication, links the INTENSITY-sense of reduplication to the PLURALITY-sense, which in turn is linked to the claimed central sense, REPETITION. Applicative predications which denote intensive action are given in (74)-(75 ): (74) Man kann Schadlinge ohne Einsatz von chemischen Giften bekampfen. 'It is possible to fight pests without the use of chemical poisons.' (75) Die Demonstranten beschimpften die Polizisten. 'The protesters verbally abused the policemen.' The verb bekiimpfen, exemplified in (74), differs from kiimpfen 'fight' in that predications containing the former verb necessarily describe active combat. While the type of action that bekiimpfen expresses involves repeated engagements, the iteration notion is not by itself sufficient to warrant the use of this verb: an event involving a series of defensive battles could not be described by this verb. By the same token, one cannot relate the meaning of beschimpfen ('verbally abuse') in (7 5) to that of schimpfen ('scold') by adding an iteration implication to the latter verb, since an event of verbal abuse


The relationship between verb semantics and construction semantics in examples like (69)-(7o) appears to be that of elaboration, since the verb and the construction are each bivalent. In the case of (7o), for example, the player and the 'instrument' roles assigned by the verb map to the agent and theme roles assigned by (this sense of) the construction. The elaboration relation is identified when the verb is a more specific instance of the event type designated by the construction. Does this characterization apply to the iteration usage? The applicative construction designates a sequence of iterated events, whereas the verb spielen e.g. does not denote a sequence of iterated playing events. However, insofar as the verb denotes the event type which is replicated in the constructional semantics, we can identifY an elaboration relation in this case.


does not consist of repeated scoldings. The meanings of beschimpfen and schimpfen can, however, be related on a scale of intensity for actions. Thus, although intensity and iteration are frequently inseparable, an applicative predication may imply intensive action alone. The relation between the verb and (this usage of) the applicative construction is again that of elaboration: the verb is a subtype of the situation type denoted by the construction. The situation type denoted by the construction corresponds to an extreme point on a scale for eventualities. 4· 3 ·4 Affectedness

(76) MMM/507.07898: Mannheimer Morgen, 1 4.07. 1 995, Lokales Die Frage, ob der Mundenheimer Platz bespielbar ist, wird kiinftig der Platzwartentscheiden. 'The question of whether the Mundenheimer Field is playable will in future be decided by the groundsman.' Affectedness is a frequent concomitant of saturation: damage to the soccer field results from the players' sequential or summary coverage of the field. We contend that this implication is the basis for a pragmatically based extension of the applicative pattern in which only affecting and not coverage is entailed. Predications which exemplify this usage express the means by which the effect is achieved, as in the nonce formation in (77): (77) Siiddeutsche Zeitung, May 27, 1 995 [. . . ] und wahrend einem die heil3en Rhythmen der kubanischen Musik in die Glieder fahren, stromen lachelnde Madchen, zweifelsfrei Sendboten eines fernen Planeten der Freude, in die graue Welt, von vorne, von hinten, von den Seiten, angetan mit Flitter und Tand, mit Federbiischen und Riesenblumen, mit Durchsichtigem und Undurch sichtigem, und sie betanzen dich, behexen dich, verwirren dich, machen dir warm urns Herz [ . . . ]. 'And while the hot rhythms of Cuban music get to you, smiling girls undoubtedly messengers of a far away planet of joy-stream into the gray world from all sides, clad in tinsel and finery, plumes and giant flowers, in transparent and nontransparent clothes, and they dance you, bewitch you, confuse you, make you feel hot.'


One can define affectedness as a change of physical or mental state which is (potentially or actually) effected by some action. A location is potentially affected by the theme's movement across it, as in (76):


In addition, most speech-act verbs can be used in the applicative pattern to denote a means of annoying the recipient: (78) Siiddeutsche Zeitung , November 28, 1992 HEADLINE: Die Leute wollen sich nicht belabern lassen. N-tv Geschaftsfiihrer. Karl-Ulrich Kuhlo iiber die Chancen semes Nachrichtenprogramms [ . . . ]. 'HEADLINE: People don't want to be blathered. N-tv executive Karl-Ulrich Kuhlo about the prospects for his news channel.'

(79) Message ID ( 3 779D77E.EC8o2A73©fh-konstanz.de) Meistens muB man den selber einbauen. AuBer du bist sehr gut im Leute bequatschen. 'In most cases you have to install it yourself Except if you're very good at persuading (lit. talking) people.' Applicative verbs of domestic and culinary activity also frequently express the means by which a (beneficial) effect is achieved. Sentence (8o) (a response to an on-line personals ad) exemplifies this usage of the applicative for the verbs kochen ('cook') and putzen ('clean'): (8o) Message ID ( 3 566638o.I 266473 6©news.netway.at) Und wenn Du [ . . . ] arbeitest, mir das Geld ins Haus tragst, mach' ich die Kinder [ . . . ], mach' ich den Haushalt, bekoch' dich ( . . . ], beputz' Dich (waschen muBt' Dich selber), und halt' Dir die Kinder vom Leib. 'And if you work [ . . . ] and bring the money home, then I'll take care of the kids, I'll take care of the house, I'll cook [for] you, clean [for] you (you have to wash yourself yourself), and keep the kids out of your way.' The applicative construction denoting effects may be combined with the oblique-theme construction, resulting in valence augmentation, if the added theme is construable as an instrument, as in ( 8 I ): ( 8 1) Message ID ( 3 6969A4 7.ADA 3 3 5 [email protected]) Ich habe sogar vor, meine Wohnebene (Uni) am Wochenende als Versuchskaninchen zu benutzen. Ich habe schon Tage lang nach einem guten Rezept gesucht, mit dem ich meine Mitstudenten am W oche nende bekochen kann.


These uses of speech-act verbs are not based upon a metaphorical mapping whose source domain involves saturation or coverage, e.g. the SPEECH IS TRAVEL ACROSS A TOPIC metaphor. Instead, speech-act verbs in examples like (78) express the means by which an effect is achieved. While in (77) this effect is unintended, the effect may also be an intended one, as in (79):

3 8 8 Valence Creation and the German Applicative

'I am even planning on using the people on my floor (college dormitory) as guinea pigs. I have been looking for days now for a good recipe with which I can cook [for] my fellow students over the weekend.' The affectedness implication associated with applicative sentences like (8o)-(8 I ) is typically associated either with an iteration implication, as when habitual beneficial activity is denoted (So), or a coverage implication, as when a bounded set of individuals is serviced (8 I ). The affectedness, coverage and iteration implications may be present simultaneously, as in (82):

In (82), a coverage interpretation is possible because the group of bene ficiaries, via its cardinality, 'measures out' the baking event-this event is completed when all of the individuals in the group are provided for. Since the event (as described) is iterated daily, the iteration implication is present here as well. That numerous implications of the applicative pattern may be simultaneously present is to be expected under the present analysis, since all of these implications constitute components of the coverage schema. 4· 3 · 5 Act in a p articular capacity toward someone

This small group of denominal be-verbs includes bemuttern (< Mutter 'mother'), bewirten ( < Wirt 'host'), bespitzeln ( < Spitzel 'spy'). Examples are given in (83)-(84): (8 3) Message ID (slrn6sf8ab.6 I [email protected]) Manner mochten auch zuweilen dominiert (bemuttert) werden, dann brauchen sie namlich auch nur zu empfangen und sind jeglicher Verantwortung (und Konsequenzen) enthoben. 'At times, men want to be dominated (mothered) too. For then they only have to receive and they are relieved of any responsibility (and consequences).' (84) http: Ilwww.phil.uni-passau.deldlwglwso711 2- I-97·txt [D]er Fall machte urn so mehr Aufsehen, als die junge, offenbar ubernervose Musikerin von der Kaiserin Maria Theresia begonnert wurde.


(82) http: Ilwww.tagesspiegel.delressortslportraitiBIBONA4oo.HTM Taglich holen einige Obdachlose aus der Teestube frische Brotchen und Kuchen in der Backerei ab. 6o-8o Menschen zu bebacken, kostet Bonau und seine Mitarbeiter am Tag eine gute Viertelstunde Arbeit. 'Every day some homeless people from the tea room pick up fresh rolls and cake in the bakery. To bake [for] 6o-8o people takes Bonau and his employees fifteen minutes of work per day.'


3 89

The case drew all the more attention since the young, apparently hyper-nervous musician was patronized by the empress Maria Theresia.'

4·4 Historical evidence The most plausible source for the semantics of coverage associated with the be-pattern is the 'around' schema associated with the historic precursor of be, the preposition bi. The modern English and German prepositions meaning 'around'-around and um/herum, respectively-may be used to express both the concept of 'surrounding an enclosed space' and that of 'being distributed over a surface area'. Consider the following data:


These verbs differ from other denominal be-verbs in an important respect: the source nominal does not express the type of the theme argument, as it does in the case of applicative verbs with literal transfer semantics, e.g. beschildern ('be-sign') or in the case of applicative verbs which evoke the effects-as-transferred-objects metaphor, e.g. befrieden ('be-peace'). Instead, the nominal base in examples (8 3)-(84) expresses the type of role assumed by the AGENT of the event. The agent's assumption of this role is a precondition upon the agent's achieving a particular effect upon the beneficiary. Therefore, we view this usage as inductively related to the affectedness usage discussed in section 4·3·4: an agent may affect a beneficiary by assuming a particular role relative to that individual. This fairly weak inductive relationship is expressed by a semantic-extension link labeled means in Figure 4· This use of the term means is distinct from that in which the term refers to a particular verb-construction integration relation. The verb-construction integration relation which we assume for bemuttern and other applicatives of this class is the instance relation. As in the case of denominals like beschildern, we postulate a type-shifting effect in which the base nominal receives a valence set via unification with the applicative construction. The arguments in the valence set of the applicative construc tion fuse with the corresponding participant roles in the frame semantics of the particular noun. In the case of bemuttern, for example, these participant roles are the mother and the child. These frame-specific roles fuse with the corresponding roles licensed by the (bivalent) applicative construction. In accordance with our general treatment of denominal applicatives, we reject the view that the base noun, e.g. Mutter, 'denotes' a particular participant in the event expressed by the predication, in this case an agent. Instead, the source nominal denotes a type, permitting omission of the agent argument where type information is sufficient, as in (8 3 ).

390


(8s) Die Spieler versammelten sich urn den Trainer. 'The players gathered around the coach.' (86) Die Spieler standen auf dem Platz herum und warteten auf den SchluBpfiff 'The players were standing around the pitch waiting for the final whistle.' (87) Die Abwehrspieler liefen orientierungslos auf dem Platz herum. 'The defense were running around the pitch disoriented.'

(88) Das Museum beherbergt geschmackvolles Geschirr und Porzellan Zierat. 'The museum harbors tasteful dishes and porcelain ornaments.' The surrounding class might be regarded as a low-productivity verb cluster associated with the applicative pattern. However, we chose to omit the enclosure usage from our semantic analysis of the applicative pattern because the be-verbs which continue the semantics of enclosure do not share the linking properties of the applicative pattern. For example, the subject of(active voice) beherbergen is a location rather than a theme (as in (88) ). By delineating the various entailments of the enclosure sense, we can understand not only the development of the coverage sense of be- but also the development of a PROXIMITY meaning which continues in the non bound preposition bei and, like the surrounding sense, is preserved in a few be-verbs. Following Grimm & Grimm ( 1 8 5 4: I 203), we treat the proximity meaning as having arisen by metonymy from the 'surrounding' -sense: what is in the vicinity of an object is close to it. An explanation along these lines is plausible since, for instance, the English preposition around has uses with the same implication of proximity, e.g. John likes to be around his family. The same is true for the Modern German preposition urn, which also can be used


In (8s) we see that um/herum can denote a static surrounding configuration. In (86) we see that um/herum can denote a static configuration involving coverage of a surface by multiple points. In (87) we see that the coverage scene expressed by um/herum, like that expressed by English around, is consistent with a dynamic scene in which a traJector covers a surface over the course of time. The prefix be- had the set of senses that um/herum and around display in (8s)-(87) through Grimm's time; Grimm lists Latin circum as the first sense for the be-prefix. But, as Ruppenhofer ( 1999) showed, be-verbs exemplifying the surrounding usage seen in (8 s) were already being lost in Grimm's time and continued to be lost thereafter. Among these lost verbs were, for instance, bearmen 'embrace', beziiunen 'fence in', and behullen 'surround'. Only a small number of be-verbs which denote surrounding or containment remain. Among these verbs is beherbergen 'shelter, harbor', exemplified in (8 8):


391


in sentences like Er hat seine Familie gerne um sich ('He likes to have his family around him'), where closeness to the theme or location rather than surrounding or enclosure of the theme is involved. Be-verbs which might be viewed as preserving the proximity semantics of the prefix are bekommen ('get, obtain'), belangen ('sue, prosecute'), besteigen ('mount, climb, climb into'), and betreten ('enter'), all of which are attested with such meanings in Middle High German (Lexer 1 872). Literal glosses, which express the semantic contribution of the prefix, are, respectively: 'come upon' (as in the English verb come by, meaning 'to obtain'), 'reach toward', 'ascend to', and 'step onto'. Another class of verbs which appears to preserve the proximity-denoting sense of the prefix are statives which denote scenes in which a theme remains in place. Included in this class are: behalten, 'keep' (lit. 'hold near'); belassen, 'leave in place' (lit. 'leave by a place'), beruhen, 'be based on' (lit. 'rest on'); and bestehen, 'exist, insist' (lit. 'stand by a place'). By treating proximity, enclosure and coverage as related concepts, we can account for a layering effect observable within the class of be-verbs: while the majority ofbe-verbs (and the totality of coinages in our corpus data) invoke the coverage component, certain small classes continue meanings associated with other components of the schema. As shown by the examples involving English and German um/herum, the patterns of semantic extension proposed both here and by Grimm for the be-prefix are plausible. These patterns collectively define a continuum of idiomaticity in the modern language: some instances of the applicative pattern represent transparent combinations of constructional and verbal semantics whereas others cannot be related to the semantic schema which makes the construction productive, i.e. the coverage schema. Our proposal that the 'coverage' sense of the applicative pattern is the prototypical usage receives support not only from the fact that this class has the highest type frequency in our corpus (see Ruppenhofer & Michaelis in press: Appendix), but also from observations about the patterns of loss and innovation within the class of be-verbs over the last 200 years. For instance, a once sizeable class of verbs with removal semantics has lost all of its members, with the exception of the idiomatic verbs of theft discussed earlier. No new removal verbs have been innovated outside of the small theft subclass. By contrast, be-verbs with concrete coverage meanings have been innovated in great number. Examples include bespiken, put spikes onto [e.g. tires of motocross bikes]', bestrahlen 'illuminate; irradiate [e.g. food]', beampeln 'put up traffic lights [e.g. at intersections]'. Although the coverage class has lost members as well, these losses are piecemeal and appear to be due to lexical obsolescence (e.g. beleitern 'put ladders down mining shafts [for access]', bezetteln 'put little pieces of paper on [as labels]', belehnen 'invest with a fiefdom', befrohnen 'impose corvee on').

392


s

CONCLUSION

Acknowledgements We gratefully acknowledge the many helpful comments and suggestions provided by two anonymous reviewers. We have also benefited from discussions with Adele Goldberg, Paul Kay, Knud Lambrecht, George Lakoff, and Andreas Kathol. Finally, we thank Michael Knolker for his insightful and tireless help with the German data. This paper is dedicated with appreciation and affection to Ekkehard Konig on the occasion of his sixtieth birthday. LAURA A MlCHAELIS

Department of Linguistics-295 UCB University of Colorado Boulder, CO 80309 USA michae/[email protected] JOSEF RUPPENHOFER

Department of Linguistics University of California Berkeley, CA 94720 USA ruppenho@socra tes. berkeley.edu

Received: Final version received:

5. I 0.99 I 2.09.00

I


W e have argued that semantic licensing effects associated with the applicative pattern can be attributed to the event type denoted by this pattern. Since a syntactic pattern can denote in the way that a word denotes-via convention-it makes sense that an argument-structure pattern should, like a polysemous word, denote an associative network of senses of the type discussed here. Along with Bybee et al. (1 994), we adopt the view that the semantic substance associated with a given formative is the accretion of a series of diachronic developments. Like these authors, we see the diachronic dimension as greatly increasing the explanatory power of semantic theory. As they point out, one cannot explain the existence of a particular construction by showing that it has a particular function or functions; one must also explain how that construction developed its functions (p. 3). Describing the relevant patterns of semantic extension is a coherent enterprise only if one assumes a sign-based semantics for constructions. In accordance with Goldberg (1995), we maintain that to admit 'top-down' or syntactic meaning does not conflict with the goal of providing a compositional theory of sentence semantics: on the construc tional account, sentence interpretation involves the reconciliation of verb meaning and construction meaning-a process whose net effect is an increase in the semantic potentials of verbs.


RE FERE N CE S Alsina, Alex & Mchombo, Sam ( I 990), 'The syntax of applicatives in Chichewa: problems for a theta-theoretic asym metry'. Natural Language and Linguistic Theory, 8, 493-506. Bennet, Michael & Partee, Barbara ( I 978),

Toward the Logic of Tense and Aspect in English, Indiana University Linguistics

Alternation in German: Its Structure and Acquisition, John Benjamins, Amsterdam. Bybee, Joan ( I 995), 'Regular morphology and the lexicon', Language and Cognitive Processes, 10, 42 5 - 5 5 . Bybee, Joan, Perkins, R., & Pagliuca, William ( I 994), The Evolution of Grammar, University of Chicago Press, Chicago. Clark, Eve & Clark, Herbert ( I 979), 'When nouns surface as verbs', Language, 55, 767-8 I I . Dowty, David ( I 9 9 I ), 'Thematic proto roles and argument selection', Language, 67, 547-6 I 9. Eroms, Hans-Werner (I 98o), 'Be-Verb und prapositionalphrase', Carl Winter Universitatsverlag, Heidelberg. Fauconnier, Gilles ( I 98 5 ), Mental Spaces, University of Chicago Press, Chicago. Fauconnier, Gilles & Turner, Mark (I 996), 'Blending as a central process of gram mar', in Adele Goldberg (ed.), Conceptual Structure, Discourse and Language, CSLI Publications, Stanford, I I 3-30. Filip, Hana ( I 994), 'Aspect and the seman tics of noun phrases', in Carl Vetters (ed.), Tense and Aspect in Discourse, Mouton de Gruyter, Berlin, 227- 5 5.

Functional Syntax and Universal Grammar,

Cambridge University Press, Cambridge. Gobel, Ulrich & Reichmann, Oskar (eds) ( I 999), Fruhneuhochdeutsches Worterbuch, Walter de Gruyter, Berlin. Goldberg, Adele ( 1 995), Constructions: A

Construction Grammar Approach to Argu ment Structure, University of Chicago Press, Chicago. Goldberg, Adele (1 997), 'The relationships between verbs and constructions', in Rene Dirven & Eve Sweetser (eds),

Lexical and Syntactical Constructions and the Construction of Meaning, Benjamins,

Amsterdam, 3 8 3-98. Grimm, Jacob & Grimm, Wilhelm (eds) ( I 8 54), Deutsches Worterbuch, Verlag von S. Hirzel, Leipzig. Gropen, Jess, Pinker, Steven, Hollander, Michelle, Goldberg, Richard & Wilson, Ronald ( 1 989), 'The learnability and acquisition of the dative alternation in English', Language, 65, 203- 5 7. Gropen, Jess, Pinker, Steven, Hollander, Michelle, Goldberg, Richard & Wilson, Ronald ( 1 99 1 ), 'Affectedness and direct objects: the role of lexical semantics in


Club, Indianapolis. Bierwisch, Manfred ( I 988), 'On the gram mar of local prepositions', in Manfred Bierwisch et a/. (eds), Syntax, Semantik und Lexikon, Akademie-Verlag, Berlin, I -65. Bresnan, Joan (I 994), 'Locative inversion and the architecture of universal gram mar', Language, 70, 72-I 3 I . Brinkmann, Ursula ( I 997), The Locative

Fillmore, Charles ]. ( 1 968), 'The case for case', in Emmon Bach & Robert Harms (eds), Universals in Linguistic Theory, Holt, Rinehart & Winston, New York, I -8 8 . Fillmore, Charles ] . ( I 9 7 7), 'The case for case reopened', in Peter Cole & Jerrold M. Saddock (eds), Syntax and Semantics, Vol. 8: Grammatical Relations, Academic Press, New York, 3-26. Fillmore, Charles ]. ( I982), 'Frame seman tics', in Linguistic Society of Korea (eds), Linguistics in the Morning Calm, Hanshin, Seoul, I I I - 3 8. Fillmore, Charles ]. ( I 98 5 ), 'Frames and the semantics of understanding', Quaderni di Semantica, 6, 222- 5 3 . Fillmore, Charles ] . & Kay, Paul ( I997), 'Construction Grammar', MS, University of California, Berkeley. Foley, William & Valin, Robert van (1 984),

394 Valence Creation and the German Applicative the acquisition of argument structure', Cognition, . p , I 5 3-95· Gunther, Hartmut ( I 974), Das System der

Verben mit be- in der deutschen Sprache der Gegenwart, Niemeyer, Tiibingen. Hale, Kenneth & Keyser, Samuel ]. (I 993), 'On argument structure and the lexical expression of syntactic relations', in Kenneth Hale & Samuel J. Keyser (eds),

The View from Building zo: Essays in Honor of Sylvain Bromberger, MIT Press, Cambridge, MA, 5 3 - I 09.

Understanding the Lexicon: Meaning, Sense and World Knowledge in Lexical Semantics,

Niemeyer, Tiibingen, I 1 0-24. Lakoff, George ( I 987), Women, Fire, and

Chicago Press, Chicago. Lakoff, George ( I 990), 'The invariance hypothesis: is abstract reason based on image-schemas?' Cognitive Linguistics, I , 39-74 Lakoff, George & Johnson, Mark ( I 98o), Metaphors We Live By, University of Chicago Press, Chicago. Langacker, Ronald W. ( I 987), Foundations of

Cognitive Grammar, Vol. 1: Theoretical Prerequisites, Stanford University Press, Stanford. Levin, Beth (I 99 3 ), English Verb Classes

and Alternations: A Preliminary Investi gation, University of Chicago Press, Chicago. Lexer, Matthias (ed.) ( I 872), Mittelhoch deutsches Handwiirterbuch, Verlag von S. Hirzel, Leipzig. Lohner, Sebastian ( I 990). Wahr neben falsch:

Duale Operatoren als die Quantoren natiirlicher Sprache, Niemeyer, Tiibingen.

Losiewicz, Beth ( I 992 ), 'The effect of frequency on linguistic morphology', unpublished Ph.D. thesis, Department of Psychology, University of Texas at Austin. Marcus, Gary F.; Brinkmann, Ursula; Clahsen, Harald; Wiese, Richard & Pinker, Steven ( 1 995), 'German inflec tion: the exception that proves the rule', Cognitive Psychology, 29, I 89-256. Michaelis, Laura A. ( 1 993), 'On deviant case-marking in Latin', in Robert D. van Valin, Jr. (ed.), Advances in Role and Reference Grammar, Beqjamins, Amsterdam, 3 I I -73. Michaelis, Laura A ( I 994), 'A case of constructional polysemy in Latin', Studies in Language, 1 8, 4 5-70. Michaelis, Laura A & Lambrecht, Knud (I 996), 'Toward a construction-based model of language use: the case of nominal extraposition', Language, 7 2, 2 I 5-47. Michaelis, Laura A & Ruppenhofer, Josef (in press), Beyond Alternations: A


Herweg, Michael (I 99 I), 'A critical account of rwo classical approaches to aspect', journal of Semantics, 8, 3 62-403. Hopper, Paul ]. ( 1 98 5), 'Causes and effects', Chicago Linguistic Society, 2 1 , 67-88. Hopper, Paul J. & Thompson, Sandra A. ( I 980), 'Transitivity in grammar and discourse', Language, 56, 2 5 I -99· Hopper, Paul ]. & Traugott, Elizabeth C. ( I 99 3 ), Grammaticalization, Cambridge University Press, Cambridge. Jackendoff, Ray ( I 983), Semantics and Cognition, MIT Press, Cambridge, MA Jackendoff, Ray (I 997a), The Architecture of the Language Faculty, MIT Press, Cam bridge, MA Jackendoff, Ray ( I 997b), 'Twistin' the night away', Language, 7 3 , 5 34-70. Kay, Paul & Fillmore, Charles ]. ( 1 999), 'Grammatical constructions and lin guistic generalizations: the what's x doing y? construction', Language, 7 5 , 1 -3 3 Konig, Ekkehard & Traugott, Elizabeth (I982), 'Divergence and apparent con vergence in the development of yet and stilf, Berkeley Linguistics Society, 8, I 70-9. Konig, Ekkehard & Traugott, Elizabeth ( I 98 8 ), 'Pragmatic strengthening and semantic change: the conventionalizing m of conversational implicature', Werner Hullen & Rainer Schulze (eds),

Dangerous Things: What Categories Reveal about the Human Mind, University of


Constructional View of the German Appli cative Pattern, CSLI Publications, Stan ford. Pinker, Steven

( I989), Learnability and

Cognition: The Acquisition of Argument Structure, MIT Press, Cambridge, MA.

Pinker, Steven & Prince, Alan ( I99I ), 'Regular and irregular morphology and the psychological status of rules of grammar', Berkeley Linguistics Society, 1 7,

230- 5 1 .

historical directions and metaphorical mappings in the two domains', Berkeley

Linguistics Society,

I 3,

446-59-

Sweetser, Eve ( I990), From Etymology to Pragmatics, Cambridge University Press, Cambridge. Talmy, Leonard ( I 9 8 5 ), 'Lexicalization patterns: semantic structure in lexical forms', in Timothy Shopen (ed.), Lan

guage Typology and Syntactic Description, Vol. 3: Grammatical Categories and the Lexicon, Cambridge University Press, Cambridge, 57- I49· Talmy, Leonard ( I988 ), 'The relation of grammar to cognition', in B. Rudzka Ostyn (ed.), Topics in Cognitive Linguistics, Benjamins, Amsterdam, I65-205. Tenny, Carol ( I987), 'Grammaticalizing aspect and affectedness', unpublished Ph.D. thesis, Department of Linguistics and Philosophy, MIT. Valin, Robert D. van & LaPolla, Randy ]. ( I 997), Syntax: Structure, Meaning and Function, Cambridge University Press, Cambridge. Wunderlich, Dieter ( 1987), 'An investi gation of lexical composition: the case of German Be verbs', Linguistics, 25, -

283-3 3 I.

Wunderlich, Dieter ( 199I ) , 'How do pre positional phrases fit into compositional syntax and semantics?' Linguistics, 29,

591-62 I.

Wunderlich, Dieter ( 1 994), 'Towards a lexicon-based theory of agreement',

Theoretical Linguistics, 20, 1-35.

Wunderlich, Dieter ( 1997 ), 'Cause and the structure of verbs', Linguistic Inquiry, 28,

27-68.

Zwicky, Arnold ( 1989), 'What's become of derivations? Defaults and invoca tion', Berkeley Linguistics Society, I 5,

IOO- J 4.

Zwicky, Arnold ( 1 994), 'Dealing out mean ing: Fundamentals of syntactic con structions', Berkeley Linguistics Society,

20, 6I 1-25.


Rappaport, Maika & Levin, Beth ( I988 ), 'What to do with theta roles', in Wendy Wilkins (ed.), Syntax and Seman tics, Vol. 2 1 : Thematic Relations, Academic Press, New York, 7-36. Rappaport Hovav, Maika & Levin, Beth ( I998 ), 'Building verb meanings', in Miriam Butt & Wilhelm Geuder (eds), The Projection of Arguments, CSLI Publications, Stanford, 97- I 34· Reddy Michael ( 1 979), 'The conduit meta phor', in Andrew Ortony (ed.), Metaphor and Thought, Cambridge University Press, Cambridge, 284-324. Regier, Terry ( 1 994), A Preliminary Study of the Semantics of Reduplication, Technical Report TR-94-0 19, International Com puter Science Institute, Berkeley, CA. Rice, Sally ( 1 989), 'Towards a cognitive model of transitivity', unpublished Ph.D. thesis, Department of Lin guistics, University of California, San Diego. Ruppenhofer, Josef ( 1 999), 'Lexical change across a semantic class', paper presented at the Annual Meeting of the Linguistics Society of America, Los Angeles. Slobin, Dan I. ( 1 997), 'There's more than one way to talk about motion: conse quences of linguistic typology for narrative style', paper presented at the Third Conference on Conceptual Struc ture, Discourse and Language, Uni versity of Colorado at Boulder. Sweetser, Eve ( I987), 'Metaphorical models of thought and speech: a comparison of

395

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents