Concord and the Syntax Semantics Interface

Concord and the Syntax Semantics Interface In natural languages a functional operation can be manifested more than once in the morphosyntax of one sentence. Most notable are the phenomena of negative concord (where several negative elements contribute to one negation) and sequence of tense phenomena, where the same happens in the temporal domain. Similar observations can be made regarding the domain of mood, case-agreement, multiple Wh, and conditional sentences. On a general and intuitive level, the similarities are striking. These phenomena have each been studied in various domains, from both an empirical and a theoretical point of view. For this workshop we have solicited contributions which approach the phenomena from a general, cross-categorial perspective: across the various domains, theoretical, and typological as well. What should be the logical form of those constructions? To what extent (at what cost) does it allow for a compositional treatment? How similar and how general are the phenomena really? What do these phenomena tell us about the model of grammar? The general aim of this workshop is to bring together researchers of different disciplines (semantics, syntax, typology) in order to create new insights about concord phenomena and the syntax-semantics interface. The workshop provides a forum for advanced PhD students and researchers to present and discuss their work with colleagues and researchers who work in the broad subject areas represented at ESSLLI. This volume contains abstracts of the talks given by the invited speakers Pieter Muysken (Radboud University, Nijmegen) and Arnim von Stechow (Seminar f¨ ur Sprachwissenschaft, T¨ ubingen), the six papers which have been accepted for presentation, and the two alternates. All submissions (16 in total) have been reviewed by a program committee consisting of the invited speakers, Sjef Barbiers, Rajesh Bhatt, Regine Eckardt, Jack Hoeksema, Ed Keenan, Ad Neeleman, Henri¨ette de Swart, Raffaella Zanuttini, Alessandro Zucchi, Jan Wouter Zwart. We want to thank them wholeheartedly for the substantial comments they have given. Paul J.E. Dekker and Hedde H. Zeijlstra Amsterdam/T¨ ubingen, June 2006

1

1

Table of Contents Introduction • Concord and the Syntax Semantics Interface Paul J.E. Dekker and Hedde H. Zeijlstra . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Invited Speakers • Concord and functional categories: a South American perspective Pieter Muysken . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 • TBA Arnim von Stechow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Accepted Papers • Japanese wh-scope marking as left dislocation Yasuyuki Fukutomi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 • Modal concord Bart Geurts and Janneke Huitink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 • Irrealis and sequence of TAM Atle Grønn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 • Long-distance negative concord and restructuring in Palestinian Arabic Frederick Hoyt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 • Concord phenomena in first language acquisition Jacqueline Van Kampen and Arnold Evers . . . . . . . . . . . . . . . . . . . . . . . 33 • The syntax of Catalan and Spanish pre-verbal n-words Susagna Tubau . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Alternates • Event-argument homomorphism as a concord phenomenon Boban Arsenijevic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 • Cognitive explanation of case shift and verb concord in Georgian Rusudan Asatiani . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2

2

Concord and functional categories: A South American perspective Pieter Muysken, Radboud University Nijmegen A

I start out with the question whether functional categories have special semantic and syntactic status, and, time permitting briefly survey the features these categories are claimed to have in the literature:

Semantic features

Special abstract meaning. No modification No Theta-assignment No relation to ontology No sense relations No meaning coercion Flexibility Semantic projection Deletability and expression as a null form No selectional restrictions Provide a landing site for movement Syntactic features Heads Part of projection chains May be interpretable at interfaces or not Obligatorily transitive Take a single type of complement Do not head argument Inseparable from their complement Categorial specificity B

I investigate the phenomenon of concord as lying at the syntax/semantics interface and try to establish whether the set of elements undergoing concord is coexentisve with the set of functional categories (see also this conference), and conclude that it affects the middle layers, not so much the inner (lexiconcentred) or outer (utterance-centred) layers, at least in the verbal/clausal domain. Any concord outside the functional domain?

Gender Number Person Case Negation Tense Mood Modality Wh

la casa blanca les soldats marchent she eats de Bello Gallico Je ne regrette rien He said the car was rusting because of the humidity. Greek non-verdical contexts for subjunctive This substance may possibly be hazardous to your health. [CP Wasi glaubte Miró [CP welches Bildi Picasso ti gemalt hatte]?

3

Definiteness Control/free will Location Aspect Transitivity

het bruine paard / een bruin paard Georgian in der Garden / in den Garten ??? Event argument homomorphism Quechua clitic climbing

???

Outer layers: Evidentiality, focus Inner layers: Aktionsart, pluriactionality, voice

C

I look at some concord phenomena that have emerged in recent work on the languages of the Amazon region

(1)

Miraña (Seifart 2005: 288) classifier agreement

a:-nε tsájhtε-bέ=nέkU tε:-nε CON-GM.in take-GM.m.sg=REC PN-GM.in ‘he brought it, the hat’

gwatáhko-hı cover.NOM-SM.2D.round

á-hı-má=nέkU pε-:bε CON-SM.2D.round-SOC=REC go-GM.m.sg ‘and with it (round and flat, i.e. the hat), he left’ (2)

Movima (Haude 2006: 162, 165) nominal tense

a.

Ajlo:maj loy os no:no di’ pa:ko. narrate ITN art.n.p pet REL dog ‘I’ll you about my (former, deceased) dog.’

b.

jayna n-os imay-ni DSC obl-ART.n.p night-PRC ‘Then at night, my stomach hurt.’

(3)

Yuracare (Van Gijn 2006: 204) diminutive agreement

a.

todito anu all like.this

jayna tivij-ni DSC pain-PRC

shuyulë-nñu-0 beautiful-DIM-3

os chodo :wi ART.n.p stomach

a-itta 3sg.p-thing

a-alparatu-nñu bënama animal bëshë-nñu 3sg-adornment-DIM:PL be-like.this animal thing-DIM ‘It was all like this, very beautiful, with all kinds of little adornments, it was like this, all these little things and animals. b.

ana-ja-lë baja-nñu-0 DEM-MEA-AMP subside-DIM-3 ‘The water had subsided a little.’

D

I conclude with the question of why there is concord at all in natural languages.

4

v.Stechow

Draft:

THE AUXILIARY PRINCIPLE ARNIM VON STECHOW, TÜBINGEN The Auxiliary Principle (AP) is this: (1)

Each auxiliary is semantically empty but has a feature that makes visible a covert semantic operator under agreement.

Normally, the operator is in the specifier of the auxiliary. In subordinate constructions, the operator may be missing for type reasons. In this case the auxiliary agrees with a higher auxiliary giving rise to multiple agreement. The talks investigates the relation between agreement and variable binding. 1. Usually, the operator in the specifier semantically binds an argument in the complement of the auxiliary. Correspondence between agreement and semantic binding. 2. The operator may bind a variable in an adjunct to the complement. Misfit between agreement and semantic binding. 3. The relevant argument is not bound at all but left unsaturated. The morphology of the relevant operator agrees with a non local operator. Multiple agreement. • Here is an illustration of AP for the German sentence ( 2) Es ist heiß gewesen it is hot been ( 3) D-structure

The lines show the agreement relations between the features [pres] and [perf]. The semantically interpreted operators are Present and Perfect. • (4)

The example illustrates the normal case. Correspondence between agreement and semantic binding. Lexicon

1

5

v.Stechow

Draft:

[[ Present]] = !c.!s.tc [[ heißit]] = !c.!s.!t.It is hot in s at time t. [[ Perfect]] = !c.!s.!Pit.!t.("t’)[t’ < t & P(t)] Meanings are Kaplanian characters. c ranges of contexts. tc is the time of the context. s ranges over situations/worlds. Construction of the LF by ( 5)

Chomsky’s Principle of Full Interpretation (FI): Delete uninterpretable material and re-bracket the remnants according to logical types.

(6)

LF

= !c.!s.("t)[t < tc & it is hot in s at time t] Other VPs. ( 7)

John was going to the station [TP Past was [AspPPROG Asp [VP John going to the station]]] |____| |_______________| ing makes Dowty’s (1979) Progressive operator visible. (8) V pjat’ chasov ona pozvonila mame at 5 o’ clock she called mom Past T [AspPPerfective Asp [VP ona pozvonila mame]]] po makes visible Krifka’s (1989) Perfective-operator (inclusion of event time in reference time). la makes visible Past. • (9)

The next example from (von Stechow, 2002b) illustrates a misfit between agreement and semantic binding. Wolfgang spielte in jeder Woche an jedem Montag Tennis Wolfgang played in each week on each Monday tennis #t1[t1 is a week in Past $ #t2[t2 is a Monday in t1 $ "t3[t2 on t2 & Wolfgang plays tennis at t3]]]

The semantic Past is not an argument of the verb. It restricts the background of week. There is a syntacti agreement relation between Past and the finite Verb, but a semantic binding relation between Past and the time argument of week. ( 10)

semantic binding Past !i T [PP in every week(i) on every Monday] [VP Wolfgang played tennis] agreement

The following examples illustrate multiple agreement with SOT-phenomena. Standard data. Cf. e.g. (Abusch, 1994), (Ogihara, 1989). English: has relative PAST, but not relative PRES, has multiple Agreement. 2

6

v.Stechow

Draft:

(11)

Pres T John believes T Mary is sick (simultaneous) |_______________|________| Multiple Agree. No semantic tense in embedded clause. (12) Pres T John believes PAST T Mary was sick (anterior) |_______________| |____________| PAST is relative past, i.e. Perfect. (13)

Past T John believed T Mary was sick |________________|________|

(simultaneous)

Multiple Agree. No semantic tense in embedded clause. (14) Past T John believed T PAST Mary was sick (anterior) |_______________| |__________| Russian: Has relative PRES and relative PAST, has no multiple Agreement. (15)

(16) (17) (18) • (19)

(20)

Pres Ivan dumaet chto PRES Masha boleet (sim) |______________| |___________| Ivan thinks that Mary is sick Pres Ivan dumaet chto PAST Masha bolela (anterior; local agree) Ivan thinks that Mary was sick Pres Ivan dumal chto PRES Masha boleet (sim; local agree) Ivan thought that Mary is sick Pres Ivan dumal chto PAST Masha bolela (ant; local agree) Ivan thought that Mary was sick Person is not interpreted if the personal pronoun is semantically bound. For data see (Heim, 1994), (Heim, 2005). Nur ich habe meine Hausaufgaben gemacht. |_________| = I am the only one that has made his homework Mein Chef hat nur mich eingeladen my boss has only me invited = I am the only one who has been invited by my boss ! I am the only one who was been invited by his boss

References Abusch, Dorit. 1994. Sequence of Tense Revisited: Two Semantic Accounts of Tense in Intensional Contexts. In Ellipsis, Tense and Questions, ed. Hans Kamp. Stuttgart: Dyana-2 Esprit Basic research Project 6852. Dowty, David. 1979. Word Meaning and Montague Grammar: Synthese Language Library. Dordrecht: Reidel. Heim, Irene. 1994. Puzzling reflexive pronouns in de se reports: Handout from Bielefeld conference. Heim, Irene. 2005. Feature on bound pronouns. Ms. Cambridge/Mass. Kratzer, Angelika. 1998. More Structural Analogies Between Pronouns and Tenses. In SALT VIII, eds. D. Strolovitch and A. Lawson. Cambridge, Mass.: Ithaca: CLC3

7

v.Stechow

Draft:

Publications. Krifka, Manfred. 1989. Nominalreferenz und Zeitkonstitution: Studien zur Theoretischen Linguistik. München: Wilhelm Fink. Ogihara, Toshiyuki. 1989. Temporal Reference in English and Japanese: University of Texas at Austin. Paslawska, Alla, and von Stechow, Arnim. 2003. Perfect Readings in Russian. In Perfect Explorations, ed. Rathert Alexiadou, von Stechow. Berlin: Mouton de Guyter. von Stechow, Arnim. 1995. On the Proper Treatment of Tense. Paper presented at SALT V.Teresa Galloway and Mandy Simons.(ed.) von Stechow, Arnim. 2002a. Binding by Verbs: Tense, Person and Mood under Attitudes. Boston: NELS. von Stechow, Arnim. 2002b. Temporal Prepositional Phrases with Quantifiers: Some Additions to Pratt and Francez (2001). Linguistics and Philosophy 25:40 pages. Zeijlstra, Hedde. 2004. Sentential Negation and Negative Concord. Utrecht: LOT.

4

8

JAPANESE WH-SCOPE MARKING AS LEFT DISLOCATION Yasuyuki Fukutomi Fukushima University [email protected] Abstract This paper shows that Japanese has a so-called “wh-scope marking (WHM)” construction. WHM constructions can be considered to be one of the concord phenomena; that is, several whelements contribute to one question. We propose that the Japanese WHM construction should be analyzed as involving a “left dislocation” structure. In addition, scope properties of this construction lend some support to Kayne’s (1994) analysis of Japanese wh-questions according to which Japanese does not really have wh-in-situ, but overt wh-movement mediated by pied-piping of the larger clause.

1 Introduction In Japanese, the scope of a wh-phrase is determined by a c-commanding clause-final particle ka or no. The following examples illustrate the point: (1) a. Mary-wa [John-ga nani-o yonda ka] sitteiru. Mary-TOP John-NOM what-ACC read Q know ‘Mary knows what John read.’ b. Mary-wa [John-ga nani-o yonda to] omotteiru no? Mary-TOP John-NOM what-ACC read COMP think Q ‘What does Mary think that John read?’ c. *Mary-wa dare-ni [John-ga kono hon-o yonda ka] osieta. Mary-TOP who-DAT John-NOM this book-ACC read Q told Lit. ‘Mary told who John read this book.’ The wh-phrase nani ‘what’ appears in the embedded clause both in (1a) and in (1b). In (1a) the question particle ka appears on the embedded verb and yields an embedded question, while in (1b) the particle no attaches on the matrix verb and yields a matrix question. The ungrammaticality of (1c) results from the failure to fulfill the c-command requirement for the licensing of a wh-phrase. The wh-phrase dare ‘who’ appears in the matrix clause and cannot be c-commanded by the question particle ka attached on the embedded verb. Although scope marking by a particle has been well studied in the previous literature (Nishigauchi 1986, Hagstrom 1998, among others), it has hardly been acknowledged that Japanese possesses another way of expressing the scope of in-situ wh-phrases by adding an expletive-like wh-phrase doo to the matrix clause, as exemplified in (2a). To the best of our knowledge, the construction of this kind has received no attention in Japanese generative literature to date. (2) a. Mary-wa [John-ga nani-o yonda ka] doo omotteiru no? Mary-TOP John-NOM what-ACC read Q how think Q ‘What does Mary think that John read?’ b. LGB-o yonda to omotteiru yo. LGB-ACC read COMP think PRT ‘She thinks that he read LGB.’

9

The example in (2a) consists of two clauses each containing a wh-expression. In spite of the existence of two wh-phrases, this type of question does not have a multiple wh interpretation. Rather, specifying the value for the wh-phrase nani ‘what’ in the embedded clause is an appropriate way of answering the question. For instance, (2b) may be a felicitous answer to (2a). The dummy status of the additional wh-phrase doo indicates that this type of construction comes under the same class of questions known as wh-scope marking, or partial whmovement constructions. (3) Was glaubt Hans [mit wem Jakob jetzt spricht] WH believe Hans with whom Jakob now speak ‘With whom does Hans think that Jakob is now talking?’ The German example in (3) also involves the invariant wh-phrase was in the matrix clause, and expresses a question whose felicitous answers specify only the value for the embedded wh-phrase. The remainder of the paper is organized as follows. We begin in the following section with an introduction to the WHM construction in Japanese. Section 3 proposes a novel analysis of the WHM construction under which it is unified into “Left Dislocation” constructions. In section 4 we argue on the basis of scope interpretation that Japanese is underlyingly an SVO language, rather than SOV, with an overt large-scale leftward movement. Section 5 concludes the paper. 2 Basic Properties of Wh-Scope Marking Constructions In this section we present basic properties of the WHM construction in Japanese, which shows important similarities with the corresponding construction in other well-studied languages like German. Among its peculiar properties, we list a few below partly based on the discussion in Dayal (1994, 2000). A. There is no restriction on the type of wh-phrase: (4) a. anata-wa Mary-ga dare-to hanasita ka doo omotteiru no? you-TOP Mary-NOM who-with talked Q how think Q ‘Who do you think Mary talked to?’ b. anata-wa Mary-ga dokode/itu/naze odotta ka doo omotteiru no? you-TOP Mary-NOM where/when/why danced Q how thinks Q ‘Where/When/Why do you think Mary danced?’ B. Any number of embedded wh-phrases can be associated with the wh-scope marker: (5) anata-wa Hans-ga itu dono daigaku-de benkyoo-sita ka doo omotteiru no? you-TOP Hans-NOM when which university-at studied Q how thinks Q ‘When do you think Hans studied at which university?’ C. The wh-scope marker must be associated with an embedded wh-clause: (6) *anata-wa Mary-ga Hans-to hanasita to you-TOP Mary-NOM Hans-with talked COMP

doo omotteiru no? how think Q

Crucial for Japanese is whether the complementizer is [+wh] or [-wh], since replacing, for example, the embedded subject in (6) with a wh-phrase does not salvage the sentence. (7) *anata-wa dare-ga Hans-to you-TOP who-NOM Hans-with

hanasita to talked COMP 10

doo how

omotteiru no? think Q

D. The matrix verb must be one that selects a non-interrogative complement, despite the fact that the complement clause is formally an indirect wh-question. (8) a. anata-wa Mary-ga dare-to hanasita ka doo omotteiru no? you-TOP Mary-NOM who-with talked Q how think Q ‘Who do you think Mary talked to?’ b. *anata-wa Mary-ga dare-to hanasita ka doo tazuneta no? you-TOP Mary-NOM who-with talked Q how asked Q intended: ‘Who did you ask Mary talked to?’ E. The wh-scope marker and the associated wh-word must not be in the same clause. (9) *anata-wa nani-o doo yondeiru no? you-TOP what-ACC how be-reading Q intended: ‘What are you reading?’ This completes our basic description of the Japanese wh-scope marking construction. 3 Derivation of Japanese WHM Constructions In this section we propose that the structure of Japanese WHM constructions be captured within the Topic-Focus articulations of the clause originally proposed by Rizzi (1997). First of all, assuming that the Condition on Local Merge (Bhatt (2003)) is correct, we propose the following structure for embedded clauses of Japanese WHM constructions; that is, the scope marker and the embedded wh-question start out as a constituent that is complement to the matrix verb: (10)

[VP V [WH [doo] [CP … wh-phrase …]]]

The structure is basically equivalent to the one for a simple question ‘What do you think?’, where what asks the complement of a proposition-taking verb, i.e., WH. The only difference is that in the case of WHM, the category WH is more complex: another question serves to restrict the head of WH as its syntactic sister. Let us now turn to the derivation of Japanese WHM. We argue that the wh-scope marker is located at the specifier of FOCUS head. Supporting evidence comes from the fact that separating the scope marker from the matrix verb degrades the sentence as in (11): (11)

a. anata-wa Mary-ga dare-ni atta ka kinoo doo omotta no? you-TOP Mary-NOM who-DAT met Q yesterday how thought Q ‘Who did you think yesterday that Mary met?’ b. ??anata-wa Mary-ga dare-ni atta ka doo kinoo omotta no? you-TOP Mary-NOM who-DAT met Q how yesterday thought Q

As is well known, the focused element must be in the immediately preverbal position in Japanese like other SOV languages. This constitutes the evidence for Focus analysis of the wh-scope marker. After the raising of the scope-marker, the remnant embedded clause moves to the left of the matrix verb. Thus the exact derivation proceeds as follows: (12)

a. [wh doo [CP John-ga nani-o yonda ka]] → Merge a lexical verb and the complex wh-phrase b. [VP omotteiru [whdoo [CP John-ga nani-o yonda ka]]] → Raise doo into the specifier of FOCUS c. [FOCUSP dooi [VP omotteiru [wh ti [CP John-ga nani-o yonda ka]]]] → Raise the remnant of wh-phrase d. [XP [wh ti [CP John-ga nani-o yonda ka]j [FOCUSP dooi [VP omotteiru tj]]]]]

11

What triggers the raising of the remnant wh-phrase in Japanese? Here Mathieu’s (2002) observation concerning discourse properties of the wh-scope marking construction is suggestive. He comments as follows: (13)

“When partial WH movement takes place, the indefinite with which the WH operator is associated is defocalized element; it is no longer part of the focus structure, but is instead a topic. Partial-WH-movement construction asks question whereby a situation and its protagonists are familiar to the participants.” (Mathieu 2002: 209)

Assume following Mathieu’s (2002) observation that the embedded clause functions as a topic rather than a focus. Then we can take Japanese WHM question to be an instance of the “Left Dislocation” constructions. Moreover, the present analysis straightforwardly explains a peculiar property of the Japanese WHM construction that poses a recalcitrant problem with the so-called “direct dependency approach.” Consider the following contrast: (14)

a. anata-wa [Mary-ga nani-o yonda ka] doo omotteiru no? you-TOP Mary-NOM what-ACC read Q how think Q ‘What do you think Mary read?’ b. *anata-wa doo [Mary-ga nani-o yonda ka] omotteiru no? you-TOP how Mary-NOM what-ACC read Q think Q ‘What do you think Mary read?’

Descriptively, the embedded question must occur to the left side of the scope marker. In order for the contentful wh-phrase to replace the “wh-expletive” at LF, as assumed in the direct dependency approach, the former must be c-commanded by the latter, otherwise yielding the ill-formed downward movement. In this respect, Japanese contrast in (14) is surprising. If the wh-scope marker appears at the position that can potentially c-command the contentful whphrase, the sentence becomes unacceptable. On the contrary, if the contentful wh-phrase appears outside the c-command domain of the wh-scope marker, then the sentence is grammatical. Therefore the direct link between the scope marker and the contentful wh-phrase seems to be irrelevant for the acceptability of Japanese wh-scope marking construction. This constitutes the most recalcitrant problem with the direct dependency approach. The derivation proposed in (12), in contrast, circumvents the problem. Assuming the process of raising the remnant embedded clause is obligatory, the desirable word order automatically follows. Moreover, in (14b) the scope marker doo is separated from the main verb, which results in ill-formedness due to FOCUS nature of the wh-scope marker. To sum up, in this section we propose a new analysis of WHM in Japanese. One of the remaining problems is why Japanese has two strategies for scope marking. In the next section, we will consider one interesting aspect of scope marking in Japanese and speculate on this problem. 4 Universal SVO Base? Simpson and Bhattacharya (2003) examine Bangla data on the distribution of finite complement clauses containing a wh-phrase, and argue that to explain the availability of wide scope reading of a wh-phrase, the containing complement clause must be pied-piped to the left of the governing verb, deriving the SOV word order in Bangla. The same patterns occur in Japanese. Although Japanese has been taken to be a strict headfinal language and therefore object complement clauses cannot occur to the right of the matrix verb, we can actually find postverbal positioning of an object clause due to “right dislocation,” as illustrated in (15): 12

(15)

John-wa omotteiru yo, [CP Mary-ga LGB-o yonda to] John-TOP thinks PRT Mary-NOM LGB-ACC read COMP ‘John thinks that Mary read LGB.’

Like the case in Bangla, an embedded indirect question can be right dislocated (16a), while the postverbal positioning of a clause with a wh-phrase intended to have wide scope is not allowed as an option (16b): (16)

a. John-ga tazuneta yo, [CP Mary-ga nani-o yonda ka] John-NOM asked PRT Mary-NOM what-ACC read Q ‘John asked what Mary read.’ b. *John-ga omotteiru no, [CP Mary-ga nani-o yonda to] John-NOM thinks Q Mary-NOM what-ACC read COMP Intended: ‘What does John think Mary read?’

The standard account of this wh-patterning assumes that postverbal clauses are extraposed to their surface position from a preverbal complement position and that the extraposed constituent creates a barrier for LF movement of the wh-phrase. More recently, however, the assumption that postverbal CPs are extraposed has become the target of criticism and turns out to be untenable. As illustrated in (17), a pronoun in a postverbal clause can be construed as a bound variable: (17)

daremoi-ga omotteiru yo, soitui-no hahaoya-ga keeki-o tabeta everyone-NOM thinks PRT his-GEN mother-NOM cake-ACC COMP ‘Everyone thinks that his mother ate a cake.’

to. ate

To obtain such a bound variable interpretation, the pronoun must be c-commanded by the matrix subject, which immediately argues against the extraposition analysis of postverbal clauses. The availability of a bound variable interpretation in (17) above indicates that the right dislocated complement clause is located in a lower position than the matrix subject; that is, it is not adjoined to the matrix IP or CP. In addition, the contrast in (18) suggests that in order for the wh-phrase contained in the complement clause to obtain higher scope than the containing clause, that complement clause must be to the left of the matrix clause: (18)

a. John-ga [CP Mary-ga nani-o John-NOM Mary-NOM what-ACC ‘What does John think Mary read?’ b. *John-ga omotteiru no, [CP Mary-ga John-NOM thinks Q Mary-NOM

yonda to] read COMP

omotteiru no? thinks Q

nani-o yonda to] what-ACC read COMP

A simple explanation to account for the contrast in (18) is that the CP containing a wh-phrase is pied-piped to the matrix scope position from the postverbal object position. Summarizing, contrary to the standard assumption that Japanese is a strict SOV language, the language is actually an SVO with a large-scale overt movement. Interestingly, converting the sentence in (18b) into the wh-scope marking construction by adding a scope marker doo enables the embedded wh-phrase to take wide scope reading: (19)

John-wa doo omotteiru no, [CP Mary-ga nani-o yonda ka] John-TOP how thinks Q Mary-NOM what-ACC read Q ‘What does John think Mary read?’

13

Therefore there seem to be two ways of expressing wide scope reading of embedded whitems: one is the pied-piping of a containing clause, and the other is the movement of a whscope marker. An immediate question to be asked now is whether these two options can be unified or not? To conclude this section, we give an affirmative answer to this question and speculate on that. Recall that to account for peculiarities of wh-scope marking constructions, we assume that the wh-scope marker and the complement question form a constituent, and that the raising operation of the wh-scope marker into FOCUS extends the scope domain to the matrix clause. Now assume further that there is a null counterpart of the wh-scope marker in the case of a pied-piping operation. In other words, natural languages require the presence of a wh-scope marker to extend the scope domain. For instance, the derivation of (18a) above proceeds as follows (Ø represents a null scope marker): (20)

a. [wh Ø [CP Mary-ga nani-o yonda to]] → Merge a lexical verb and the complex wh-phrase b. [VP omotteiru [wh Ø [CP Mary-ga nani-o yonda to]]] → Raise Ø into the specifier of FOCUS c. [FOCUSP Øi [VP omotteiru [wh ti [CP Mary-ga nani-o yonda to]]]] → Raise the remnant of wh-phrase d. [TOPICP [wh ti [CP Mary-ga nani-o yonda to]]j [FOCUSP Øi [VP omotteiru tj]]]

In this derivation, the pied-piping can be taken to be a kind of “repair” or “follow-up” operation (Chomsky 1995). This concludes the discussion on the movement of a complement clause. In this section, we have demonstrated that there is persuasive evidence for the SVO word order analysis of Japanese. 5 Concluding Remarks In this paper we have shown that Japanese possesses the so-called wh-scope marking construction along with the well-known scope marking strategy by particles. By analyzing the construction as dislocation structure, we can provide a convincing explanation to a peculiar property of Japanese wh-scope marking constructions. Moreover, we have shown that some of scope properties can be straightforwardly accounted for by assuming that Japanese is an SVO language and that it has an overt leftward movement. If this consideration is on the right track, then it lends an additional support for Kayne’s Antisymmetry approach to syntax. References Bhatt, R.: 2003, Locality in Correlatives, Natural Language and Linguistic Theory 21, 485-541. Chomsky, N.: 1995, The Minimalist Program. Cambridge, Mass.: MIT Press. Dayal, S.: 1994, Scope marking as indirect wh-dependency, Natural Language Semantics 2, 137-170. Dayal, V.: 2000, Scope marking: Cross-linguistic variation in indirect dependency, in U. Lutz, G. Müller and A. von Stechow (eds), Wh-Scope Marking, John Benjamins, pp. 157-193. Hagstrom, P.: 1998, Decomposing Questions. Doctoral dissertation, MIT, Cambridge, Mass. Kayne, R.: 1994, The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Mathieu, E.: 2002, The Syntax of Non-Canonical Quantification: A Comparative Study. Doctoral dissertation, University College London. Nishigauchi, T.: 1986, Quantification in Syntax. Doctoral dissertation, University of Massachusetts, Amherst. Rizzi, L.: 1997, The fine structure of the left periphery, in L. Haegeman (ed), Elements of Grammar, Dordrecht: Kluwer, pp.281-337. Simpson, A. and Bhattacharya, T.: 2003, Obligatory overt wh-movement in a wh-in-situ language, Linguistic Inquiry 34, 127-142.

14

M ODAL CONCORD Bart Geurts & Janneke Huitink University of Nijmegen {bart.geurts/j.huitink}@phil.ru.nl

Abstract This paper discusses a phenomenon that we propose to call “modal concord”. We speak of modal concord if the combination of a modal adverb with a modal auxiliary seems to be interpreted as if just a single modal operator was expressed. It is shown that concord interpretations can only arise if both modals are of the same type (i.e. the same accessibility relation is involved in both cases) and have more or less the same quantificational force. Our analysis is based on the idea that, in modal concord, the meaning of the adverb is shifted to a functional meaning which checks whether its argument, the auxiliary, is of the right kind.

1 Introduction The most widely studied concord phenomenon is undoubtedly negative concord. But concord interpretations are not restricted to the domain of negation. In this paper we aim to show that concord readings also occur with certain combinations of modal adverbs and auxiliaries. Examples are:1 (1)

a. b.

You may possibly have read my little monograph upon the subject. Power carts must mandatorily be used on cart paths where provided.

The preferred interpretation of (1a) is the concord reading which says that the speaker considers it possible that you have read his monograph, not the cumulative one, according to which he thinks it is possible that it is possible that you have done so. Similarly, (1b) expresses that there is an obligation to use power carts, not that it is obligatory that there is an obligation to use power carts. Thus, even though (1a) and (1b) contain two modal expressions each, the sentences are interpreted as if they contained just a single modal operator. We propose to call this phenomenon “modal concord”.2 If modal concord was only available for epistemic modals, the phenomenon could readily be explained by appealing to elementary principles of epistemic logic. For epistemic necessity, the principle of veridicality (2a) ensures that the concord reading follows from the cumulative one: φ ⊢ φ. For epistemic possibility, the cumulative reading entails the concord reading if the principle of positive introspection is assumed. 1 Example (1a) is from Sir Arthur Conan Doyle’s, The hound of the Baskervilles. (1b) was found at www. glendalegolf.ca/tourneyregs.html using the Google search engine. 2 In all our examples the two modal operators are clause mates. It may be that modal concord sometimes occur across clause boundaries. Here is an example by von Fintel and Iatridou (2003, 183), who write that they “found it very difficult to make sure that the epistemic modal is indeed included in the composition of the interpretation”:

(i)

John thinks that Sarah must have played on every piano that we had predicted he would.

In this example, must is embedded in an attitude context. Unfortunately, in such examples, the difference between the cumulative and the concord reading is unclear, and therefore we are not convinced that this is really the same as what we observe in (1a,b). See also Hara (2006) on the interpretation of modals under attitude verbs.

15

(2)

a. b.

Veridicality (knowledge is factive): φ → φ Positive introspection (if an agent knows something, he knows that he knows it): φ → φ (≡ ♦♦φ → ♦φ)

Hence, if concord interpretations only occurred with epistemic modals, standard epistemic logic would suffice to explain the facts. However, as we saw in (1b), epistemic modals are not the only ones to engage in concord. The following Dutch sentence is another case in point: (3)

Alle deelnemers moeten zich verplicht registreren. all participants must self obligatorily register “All participants have to register.”

The example is entirely natural, and its concord reading is strongly preferred. The sentence may also have a cumulative interpretation, but this only becomes available in rather special contexts. Imagine that your department is organizing a conference. You don’t think that the participants should have to register, but the head of the department does. She might say: “It must not be the case that all participants register if they feel like it, but all participants must obligatorily register” — which would require a cumulative reading. The concord interpretation of (3) cannot be explained by deontic logic, because obviously veridicality doesn’t hold in such a logic: sadly, not everything that is desirable is actually the case. So we must find another way of explaining modal concord, at least in these cases. Some readers might be worried that modal concord is not really a grammatical phenomenon, but rather belongs to stylistics, and it is true that in English and Dutch modal concord is usually optional. It is not surprising, therefore, that (3) is felt to express a stronger obligation than either of its single-modal variants: (4)

a.

b.

Alle deelnemers moeten zich registreren. all participants must self register “All participants have to register.” Alle deelnemers zijn verplicht zich te registreren. all participants be obliged self to register “All participants have to register.”

However, not all occurrences of modal concord are optional. For instance, the Dutch expression wel eens only has a modal meaning when it co-occurs with other modals. In (5a) wel eens expresses epistemic possibility, and engages in concord with epistemic zou kunnen “may/might”, whereas in (5b) it lacks this modal interpretation. (5)

a.

b.

Dat zou wel eens de vrouw van Jan kunnen zijn. that would wel eens the wife of Jan can be “That could well be Jan’s wife.” Heb jij Jan’s vrouw wel eens gezien? have you Jan’s wife wel eens seen “Have you ever seen Jan’s wife?”

Further research is needed to establish whether there are languages in which modal concord is always obligatory, and whether there are languages where modal concord is not possible at all.3 3

Interestingly, in dialects of English spoken in the South of the United States that allow two auxiliaries to occur in the same clause, a concord reading of the two modals does not seem to be available. Thus, “I might could

16

Whatever the outcome of such investigations, the phenomenon of modal concord as it occurs in English and Dutch calls out for an explanation, since it is an obvious challenge to compositional semantics. Before we proceed, we pause to note that, although the phenomenon of modal concord has not gone unnoticed before, it has never been discussed in any depth, and ours seems to be the first attempt at analysis. The earliest sources, to the best of our knowledge, are Halliday (1970) and Lyons (1977). But despite the fact that it has not received much attention, modal concord plays a key role in several recent semantic theories on a variety of phenomena. It is an essential ingredient in Geurts’s (2005) account of free choice permission, in Geurts and Nouwen’s (2005) analysis of superlative quantifiers like at least/most three ducks, and in Huitink’s (2005) treatment of sufficiency modal constructions like you only need to VP.

2 Restrictions on modal concord There appear to be two main constraints on modal concord. First, two expressions can only participate in a concord construction if they are of the same modal type, i.e., if they are both deontic, epistemic, or whatever. Note that in (1a) both modals are epistemic, while in (1b) both are deontic. Of course, many modal expressions, like English must and have to, are ambiguous between construals of various types. Such expressions are disambiguated by the context of utterance (Kratzer 1981), which determines the set of worlds that the modal quantifies over. In many cases, the complement of the modal helps to make one of the readings more salient. Stative predicates tend to favour epistemic readings, as in (6a), whereas eventive predicates are biased towards deontic interpretations, as illustrated by (6b): (6)

a. b.

John must be at home. John must do the dishes.

Now consider sentence (7). The auxiliary might can only be understood as expressing epistemic possibility, but have to is in principle compatible with both an epistemic and a deontic reading. Since its complement describes an event, the deontic reading is favoured, so (7) contains one epistemic and one deontic modal. Consequently, the sentence doesn’t have a concord reading. The only available interpretation is the one according to which the epistemic modal takes scope over the deontic one: the speaker considers it possible that John has to work on Sunday.4 (7)

John might have to work on Sunday.

The second constraint on modal concord is that the modals involved have to have the same, or at least similar, quantificational force. This constraint entails that sentence (8) has no concord interpretation, but rather expresses the speaker’s certainty that there is a possibility that there were weapons of mass destruction in Iraq. One could utter such a sentence in a dispute about whether there may or may not have been such weapons. (8)

There may certainly have been weapons of mass destruction in Iraq.

For (8), a concord reading is impossible because the universal force of certainly is incompatible with the existential force of may. But, on the face of it at least, not all examples of modal do that” can only mean that I consider it possible that I am in the position to do that. See di Paolo (1989). 4 See Nauze (2006) and Nuyts (2004) for explanations why the reverse scope order deontic > epistemic isn’t possible.

17

concord involve modals with the same quantificational force. An example is (9): (9)

Pain in these diseases may probably influence the sleep process.

Perhaps, this sentence allows for a concord reading because the quantificational force of may is sufficiently close to that of probably. Combinations like may certainly, as in (8), on the other hand, are not close enough in quantificational force, apparently. Languages differ in which modals count as sufficiently similar. For instance, in English we find (9), but in Dutch waarschijnlijk “probably” combines with expressions of necessity rather than possibility. (10)

a.

b.

Amalia kan waarschijnlijk Spaans spreken. Amalia can probably Spanish speak “Amalia can probably speak Spanish” (cumulative only) Dat moet waarschijnlijk in 1943 zijn geweest. that must probably in 1943 be been “That was probably in 1943.”

Both in (9) and (10b) the speaker presents the embedded proposition as being probable, that is, whereas in (9) the adverb seems to strengthen the force of the modal verb, in (10b) the effect seems to go in the opposite direction. However, as we will see in a moment, there is another way of viewing what is going on in these examples.

3 Analysis We assume that modal adverbs are polysemous between their standard meaning and a functional meaning which, in effect, performs a type check on its argument. That is, we parse “It may possibly be raining” (on its concord reading) as [[possibly may] raining], where possibly checks whether may is of the same sort as itself. The two senses of possibly are related by a type shifting rule. In order to implement this idea, we adopt a partial two-sorted type theory with primitive types e for individuals, s for possible worlds, and t for truth values. The meaning of “Barney sneezes” can now be represented as λ jS jb, where i and j are variables of type s. Simplifying things somewhat, we analyze must and necessarily as propositional operators: [[must]] = [[necessarily]] = λpλi∀ j[Ri j → p j], where p is a variable of type st, ranging over propositions, and R is a constant denoting an accessibility relation. Assuming that “Barney must sneeze” is parsed as [must [Barney sneeze]], we obtain the following: (11)

[[Barney must sneeze]] = λpλi∀ j[Ri j → p j](λ jS jb) = λi∀ j[Ri j → S jb]

Thus, “Barney must sneeze” is true in world w iff Barney sneezes in all worlds w′ that are accessible from w. Our type shifting rule is to map [[necessarily]] into a function that yields [[necessarily]] if its argument denotes [[necessarily]], and is undefined otherwise. To define this, we introduce the following auxiliary construct: (12)

. [[P = Q]] = [[P]] if [[P]] = [[Q]], and undefined otherwise

(P and Q are variables of type (st)(st).) We now define our type shifting rule as follows: 18

(13)

. λQλP[P = Q]

c we analyse “Barney must Assuming for convenience that this operation is denoted at LF by , c necessarily sneeze” as [[ necessarily must] [Barney sneeze]], which gets the same truth conditions as “Barney must sneeze,” i.e. (11). c This is a strict way of analysing modal concord: on this version of the analysis, [[[ α]β]] = [[α]] if [[α]] = [[β]], and is undefined otherwise. Prima facie, this wrongly excludes examples like (9) and (10b). But on reflection it is not so clear that it does. It may be that one of the standard uses of English may is to express probability, or something close to it. (The preceding sentence might be a case in point.) If this is so, we can maintain after all that the adverb in (9) serves to select one of the senses of the modal auxiliary. And if it is the case that Dutch moeten “must” can express probability, too, we can account for examples like (10b), as well. For this line of defense to work, we would have to establish that there is a rather subtle difference between English and Dutch: it should be the case that, whereas Dutch moeten “must” may express probability, English must may not, while vice versa English may can but Dutch zou kunnen “may/might” cannot express probability. This is perhaps less implausible than it might seem at first, because the Dutch lexical field of epistemic modality doesn’t map one-to-one on its English counterpart, and it is well possible that the division of labour between modal lexemes is not exactly the same in the two languages. But it is evident that teasing out the relevant differences isn’t going to be easy. For example, if Dutch and English differed in the way just indicated, it might be the case that the Dutch sentence in (14) is more acceptable than its English gloss: (14)

Hij moet in Amsterdam zijn, maar hij zou ook in Berlijn kunnen zijn gebleven. he must in Amsterdam be but he could too in Berlin can be stayed “He must be in Amsterdam, but he may have stayed in Berlin, too.”

It is not obvious that this is so, but on the other hand it isn’t obviously false, either. This is an empirical issue, to be sure, but a hard one to decide. Rather than pursuing this possibility, let us a have a brief look at possible alternatives. Are there ways of changing the type shifting rule in (14), so that it produces functions that don’t enforce identity of meaning? As far as we can see, there is only one way that makes sense, and even that we don’t know how to formulate in the present framework (and we strongly suspect it cannot be done at all). The idea is to map the denotation m of an modal adverbial onto a function that checks if the denotation of its argument, m′ , is of the same “modal type” as m; i.e. it checks if m and m′ employ the same accessibility relation. The problem with this version of the analysis—if it could be implemented somehow—is that it would be too weak. It allows for combinations like may certainly, which don’t get a concord reading in any language we know of. We could try to pragmatically restrict this analysis by the constraint that the two expressions engaging in modal concord must be sufficiently close in terms of strength. Different languages can make different decisions as to which pairs are sufficiently close and which are too far from one another. But this raises empirical issues of the same sort as the one briefly discussed in the last paragraph. For we would need to argue on independent grounds that, whereas English probably is closer to may than to must, it is different in Dutch.

19

References von Fintel, K. and Iatridou, S.: 2003, Epistemic Containment, Linguistic Inquiry 34(2), 173– 198. Geurts, B.: 2005, Entertaining alternatives: disjunctions as modals, Natural Language Semantics 13, 383–410. Geurts, B. and Nouwen, R.: 2005, ‘At least et al’: the semantics of scalar modifiers, Ms, University of Nijmegen, http://www.ru.nl/filosofie/tfl/bart/papers/atleast.pdf. Halliday, M.: 1970, Functional diversity in language as seen from a consideration of modality and mood in English, Foundations of Language 6, 322–361. Hara, Y.: 2006, Non-Propositional modal meaning, To appear in the proceedings of Sinn und Bedeutung 10. http://people.umass.edu/yhara/download/darou_hara.pdf. Huitink, J.: 2005, Analyzing anankastic conditionals and sufficiency modals, in S. Blaho, L. Vicente and E. Schoorlemmer (eds), Proceedings of Console XIII, University of Leiden, pp. 135–156. Kratzer, A.: 1981, The Notional Category of Modality, in H. Eickmeyer and H. Rieser (eds), Words, worlds and contexts, Walter de Gruyter, Berlin, pp. 38–74. Lyons, J.: 1977, Semantics, Cambridge University Press. Nauze, F.: 2006, Multiple modals construction, To appear in the proceedings of Sinn und Bedeutung 10. Nuyts, J.: 2004, Over de beperkte combineerbaarheid van deontische, epistemische en evidenti¨ele uitdrukkingen in het Nederlands, Antwerp Papers in Linguistics. http://http: //webhost.ua.ac.be/apil/apil108/index.html. di Paolo, M.: 1989, Double modals as single lexical items, American Speech 64(3), 195–224.

20

Irrealis and Sequence of TAM Atle Grønn University of Oslo Abstract 1

In this paper , I will discuss an irrealis-construction in Russian, which poses non-trivial problems for a compositional analysis of tense, aspect and mood (the categories subsumed under the abbreviation TAM). The past tense morphology on the verb – in absence of a semantic PAST operator – is argued to be licensed by an IRREALIS operator. This concord phenomenon will be accounted for in terms agreement/checking as in Minimalist feature theory. The choice of imperfective aspect in this construction can possibly be explained from the perspective of competition between optimal form-meaning pairs.

1

In mood for chess

A linguistic puzzle frequently pops up in Russian chess annotations: (1) Posle 9.e4 belye vyigryvaliind,past,ipf peˇsku, ˇcego vpolne khvataloind,past,ipf dlja pobedy. (Internet) After [the hypothetical chess move] 9.e4 white would have won a pawn, which would have been more than sufficient for victory. This intriguing modal/counterfactual flavour in absence of overt markers of modality is, of course, not an invention of strong Russian chess players. On request, my Russian informants produce discourses like the following: (2) K sˇcast’ju ja ne provalilsjaind,past,pf na e˙ kzamene. Posle provala menja vygonjaliind,past,ipf iz universiteta. Luckily, I did not fail the exam. In case of failure [lit.: ‘After failure’] I would have been kicked out of the university. Compositional semantics alone cannot explain all there is to these data. But first, in order for the reader to appreciate the puzzle, a few words on the inventory of TAM-categories in Russian. 1I

would like to thank Kjell Johan Sæbø, Arne Martinus Lindstad, my Russian consultants Maria Filiouchkina Krave and Dmitrij Ostanin, and the anonymous referees of the workshop for valuable feedback at various stages of this work.

21

Following recent work by von Stechow, we distinguish between morphological and semantic TAM-categories. Grammatical aspect, that is perfectivity (Pf) and imperfectivity (Ipf), is morphologically encoded on the verb through prefixes and suffixes. A corresponding semantic operator PF or IPF transforms the verb stem – denoting a set of events – into a set of times. At the next stage at the (morpho-)syntax-semantics interface, a semantic TENSE, e.g. the semantic counterpart of the past tense suffix l, applies to this set of times, producing a truth value. Mood is not marked morphologically on the verb in Russian, but is expressed by a subjunctive particle by, cf. the examples below: (3) Keres vyigralpast,pf bysubj u Alekhina matˇc. (Internet) Keres would have won a match against Alechine. (4) Ja bylpast,ipf bysubj rad esli bysubj Kramnik dejstvitel’no bylpast,ipf silen, vyigryvalpast,ipf bysubj turniry. (Internet) I would have been happy if Kramnik really showed his class and won tournaments. Note that for each verb occurring in a non-indicative setting, there is a corresponding by.2 Furthermore, by cooccurs only with past tense morphology, even when the hypothetical situation concerns the utterance time. On the other hand, aspectual choice is independent of the presence of the modal particle.3

2

Modal Ipf, modal past, both, or neither?

Now, let’s return to the peculiar mixture of TAM morphology and semantics in our original example (1), characteristic of comments on hypothetical possibilities in a chess game. How can an indicative imperfective past trigger this counterfactual interpretation? These data have largely been overlooked in the literature on TAM in Russian, with the exception of (Restan 1989), who presents the phenomenon without any explicit analysis. Restan simply claims that Ipf has an additional modal meaning. Note in this respect that the use of Pf in contexts like (1) would completely alter the truth conditions of the construction, but not, as one might expect, with respect to the aspectual viewpoint distinction (in)complete events. Perfective aspect would rather produce a purely temporal and indicative reading: ”After 9.e4 (which was actually played), white won a pawn...”. 2 Cf. example (4) with 3 verbs and 3 occurrences of by. This is a case of Sequence of mood or multiple agree, since each token of by merely agrees with the highest SUBJUNCTIVE operator, without shifting the world-parameter relative to the preceding by. 3 In (3), the perfective sentence refers to a (hypothetical) single event described by a telic predicate (”to win a match”). This is to be contrasted with the use of Ipf in (4) reporting an iteration of hypothetical events (cf. the atelic VP: ”to win tournaments”). This is in accordance with the accepted view that Ipf encodes either progressivity or habitual-iterative readings. Importantly, however, Ipf is the unmarked/default aspect in Russian which, given the right circumstances, also may refer to singular complete events (Grønn 2004).

22

Restan’s ‘modal Ipf’ is not without support from cross-linguistic investigations such as (Iatridou 2000), where it is shown that imperfective aspect and past tense can have a modal interpretation in various languages. Still, I will argue that there is nothing inherently modal in the imperfective occurring in (1).The use of Ipf is due to its unmarked status in the aspectual opposition. It is well-known that Ipf in Russian can refer to singular complete events when focus is not on the temporal anchoring of the event (Grønn 2004). On the other hand, I will pursue the idea that the past tense marking in Russian is correlated with modality.

3

Checking of uninterpretable past

In order to capture the relationship between past tense morphology and irrealis, I turn to the feature theory of Minimalism advocated by von Stechow in his discussions of phenomena such as Sequence of tense in Germanic: (5) PAST[i-past] Mary said [u-past] that she was in the opera [u-past]. The embedded past tense carries the feature [u-past] which is checked by the PAST operator in the matrix [i-past]. The embedded past is pronounced at PF, but is deleted at LF. This is according to the feature system which requires that every uninterpretable feature be checked by an interpretable feature. Note also that interpretable features of overt material can check more than one uninterpretable feature (under c-command). My claim is that the past tense morphology in (1) carries the feature [u-past] which is licensed/checked by an IRREALIS operator associated with the PP. Before the derivation splits into PF and LF, our key example may be represented as in figure 1. Figure 1: Feature analysis of (1) S TP [ ]

PP [i-irr] [ ]

[i-irr]

[u-past]

After

9.e4

−l

AspP [ ] [i-ipf] IP F default

[ ] [u-ipf] −yva

23

VP

The checking of [u-past] by [i-irr] is presumably possible because of some deeper affinities between past tense and irrealis/counterfactuality, what has been referred to in the literature as distancing or remoteness. The idea can be made precise by an ordering relation on times or worlds, expressing distance from the evaluation time/world, see (Iatridou 2000) for references and some discussion.

4

Compositional semantics

At LF, we do not get a semantic PAST above [u-past], but there is a default IMPERFECTIVE operator which produces a complete event interpretation of the telic predicate and, furthermore, checks [u-ipf]. The logical form of (1) is given in figure 2.4 Figure 2: Compositional semantics for (1) [[(1)]]w0 ,t0 = 1 iff ! the closestw0 ,t0 w "= w0 and the max t! > t!! such that 9.e4(w! , t!! ) ∈ {< t, w > |∃e win(e)(w) ∧ e ⊆ t} P P the w! and t! ... such that 9.e4(w! , t!! )

AspP λwλt∃e[win(e)(w) ∧ e ⊆ t]









[[After]]

[[9.e4]]

[[IPFdefault ]]

[[VP]]

The preposition ‘posle – after’ clearly has a temporal meaning. It is normally considered to be factual as opposed to the possibly non-factual ‘before’ (Beaver and Condoravdi 2003). However, in our particular cases, ‘after’ combines with a hypothetical chess move and is thus counterfactual. At the same time, expressions such as ‘pri – by’, ‘v sluˇcae – in case of’, which are less clearly temporal, occur interchangeably in our contexts of hypothetical chess moves (posle 9.e4 ≈ pri 9.e4 ≈ v sluˇcae 9.e4 ). The compositional procedure outlined above allows us to manipulate both the temporal and modal dimensions of the PP by letting the complement of the preposition, the hypothetical chess move, denote sets of world-time pairs. 4 In addition to basic types for events , times , worlds and truth-values , I define a type as the basic type for world-time pairs, which is used when the latter is not considered as a function from worlds to times. The distinguished variables w0, t0 denote the world of evaluation and the time of evaluation, respectively, whenever free. Below, I give the semantics for the lexical entries in figure 2: [[After]] = λP. the closestw0 ,t0 w ! and the max t! s.t. t! > t!! s.t. P (w ! , t!! ) [[9.e4]] = λwλt[9.e4(w, t) ∧ w "= w0] (the irrealis condition should perhaps be formulated as a presupposition on hypothetical chess moves in chess annotations.) [[Ipfdefault ]] = λP λwλt∃e[P (e)(w) ∧ e ⊆ t] [[win]] = λwλe[win(e)(w)]

24

Consider for instance ‘v sluˇcae – in case of’, where a counterfactual interpretation comes as no surprise, but the temporal dimension is less obvious. In order to avoid a semantic representation according to which the winning event is simultaneous with the hypothetical move, we must either let ‘v sluˇcae – in case of’ itself carry the future meaning – as in figure 2 – or we would have to insert a covert FUTURE operator above AspP. In fact, I suggest that the schema in figure 2 can be applied more generally: (6) Esliif bysubj belaja lad’ja stojalapast,ipf na d1, belye srazu vyigryvaliind,past,ipf putem 40.Ld8. (Restan 1989, 204). If the white rook had been on d1, white would have won immediately with 40.Rd8. Note that ‘esli – if’, which combines with a sentence, cooccurs with the subjunctive particle ‘by’ and past tense on the verb ‘stojala – stood’. Still, the winning-event in the matrix has exactly the same form as in our previous examples.5 I propose to treat the if-clause in examples like (6) in analogy with the semantics accorded to the PP in figure 2. This implies that in the StalnakerLewis-Kratzer discussion on counterfactuals, our treatment of if-clauses comes closer to the original paper (Stalnaker 1968). The if-clause is not analysed as the restrictor of a universal quantifier over possible worlds, but as a definite description on possible worlds, cf. also (Schlenker 2004) for a recent implementation of this idea.

5

Competition

There is nothing in the account given above which explains why (1) must have a counterfactual/modal interpretation while the same sentence with perfective aspect would receive a factual/temporal interpretation. It appears that the tools available so far (feature checking and compositional semantics) must further be supplemented by something like bidirectional Optimality theory. Consider the following -pairs: a < { ind, past, pf }, {indicative, temporal, complete event interpretation}> Presumably, the perfective aspect is somehow specified for temporal anchoring (Grønn 2004). In absence of the modal marker by, the optimal interpretation is a purely temporal one where [u-past] is checked by a PAST operator. On the other hand, our initial example can be represented as follows: b < { ind, past, ipf }, {counterfactual, complete event interpretation}> The interpretation in b is suboptimal and arises at a later stage of optimisation. It emerges partly due to incompatibility of a purely temporal incomplete 5 This

can be seen as a case of multiple agree: both past tense forms in (6) agree with the semantic IRREALIS-operator, which here presumably is identified with the particle ‘by’.

25

event interpretation (≈ progressive) with the meaning of the temporal preposition. But, of course, the form-meaning pair in b also competes with the more straightforward one in c: c < { subj, past, pf }, {counterfactual, complete event interpretation}> From a competition perspective it is unlikely that b could arise in a system where c is available. This point has to be investigated further. Here, we can merely suggest some more fine-grained possible differences in the interpretation component of b and c: A subjunctive perfective sentence retains its ‘strong’ temporal meaning in the counterfactual setting, due to the temporal anchoring of Pf; the counterfactual interpretation in b is more specific and corresponds to a ‘would-conditional’ with a ‘false antecedent’ (i.e. the chess move under consideration always results in the consequent being true, but it is never instantiated in the actual position in the game.), while c is also compatible with a ‘could-conditional’ and a possibly true antecedent. Thus, it appears that both past tense and imperfective aspect mark counterfactuality in a rather indirect way. The former is licensed by a checking relation with an IRREALIS operator, while the latter is used as the default, unmarked aspect in competition with more marked form-meaning pairs.

References David Beaver and Cleo Condoravdi. A uniform analysis of before and after. In R. Young and Y. Zhou, editors, Proceedings of SALT XIII, pages 37–54, Cornell, 2003. CLC Publications. Atle Grønn. The Semantics and pragmatics of the Russian factual imperfective, volume 199 of Acta Humaniora. Unipub, dr.art thesis, Oslo, 2004. Sabine Iatridou. The grammatical ingredients of counterfactuality. Linguistic Inquiry, 31(2):231–270, 2000. Per Restan. Ne bylo, no moglo by byt’: O gipotetiˇceskoj modal’nosti. ScandoSlavica, 35:203–210, 1989. Philippe Schlenker. Conditionals as definite descriptions. Research on Language and Computation, 2(3):417–462, 2004. Robert Stalnaker. A theory of conditionals. In N. Rescher, editor, Studies in logical theory, pages 98–112. Blackwell, Oxford, England, 1968. Arnim von Stechow. Interpretiertes tempus: Temporale orientierung von modalen. Neue Beitrge zur germanistischen Linguistik, to appear.

26

Long-Distance Negative Concord and Restructuring in Palestinian Arabic Frederick M. Hoyt Linguistics Department University of Texas at Austin 1 University Station B5100 Austin, TX, USA 78712-0198 [email protected]

1 Negative Concord in Palestinian Arabic1 In Palestinian Arabic (PA), negative concord occurs with noun phrases headed by the determiner wEla “(not) even one”: (1)

Negative concord : The failure of a word or phrase that expresses negation in fragment answers to express negation in a sentence in which it co-occurs with another negation-expressing word or phrase (a.o. Watanabe 2004).

wEla-DPs are pronounced with strong focal stress, and are the most “emphatic” kind of NPI in PA. Less emphatic NPIs include hada “anyone,” iši ˙ “anything, or Paiy wa:had “anyone” or Paiy ši: ˙ “anything.” Both wEla-DP and Paiy-NPs are minimizers in the sense of Vallduví (1994): I refer to wEla-DPs as emphatic minimizers. “wEla-phrases” are interpreted as negative quantifiers (“NQ-wEla)” or as polarity-sensitive indefinites (“NPI-wEla”). The NQ-interpretation is available preceding the finite verb or verb complex in a clause (2-4) or in fragment answers (5-6): (2)

(3)

(4)

(5)

wEla hada fi:-hUm ˇsæ:f-ni. ˙ not.even one.ms in-them saw.3ms-me “Not even one of them saw me!” wEla yo:m Qaˇgabni l-Ekıl. not.even day pleased.3ms-me the-food “There wasn’t even one day the food pleased me!” wEla nıtfıt anu:Ta Qınd-ık. not.even bit femininity at-you(fs) “You don’t have the least bit of femininity!” Q: ˇsu kal-l-ak? A: wEla iˇsi. ˙ what said.3ms-to-you not.even thing “What did he say to you? Nothing at all.”

1 I thank Ghassan Hussein-Ali for his help with data. Thanks also to Peter Abboud, Jason Baldridge, Rajesh Bhatt, Lisa Green, Jeff Runner, Bernhard Schwarz, Junko Shimoyama, Osama Sultan, and Alexandra Teodorescu for comments at various points. Additional thanks to the Workshop abstract reviewers for their comments.

(6)

su:s ıbn yome:n. Q: mi:n ˇsUfti? A: wEla ˙ ˙ son two-days not.even chick who saw.2fs “Who did you see? Nary a two-day old chick!”

A preverbal wEla-phrase preceding a sentential negation marker causes the sentence to have a double-negation reading (7: compare with 3): (7)

wEla yo:m ma-Qaˇgabni l-Ekıl. not.even day not-pleased.3ms-me the-food “There wasn’t one day the food didn’t please me!”

NQ-wEla never occurs within the scope of negation but occurs in post-verbal positions which are not “thematically entailed” by the verb (8-9)2 : (8)

huwwa wEla iˇsi! he not.even thing “He is nothing!”

(9)

hiyya magru:ra ˙ Qala wEla iˇsi. she conceited.fs upon not.even thing “She is conceited for absolutely no reason!”

The NPI-interpretation is only available within the scope of antimorphic operators (Zwarts 1996) like sentential negation or bıdu:n “without” (10-13): (10)

tılıQti bıdu:n-ma tku:li wEla iˇsi. ˙ left.2fs without-that say.2fs even thing “You left without saying even one thing!”

(11)

ma:-XaDt-ıˇs maQ-i wEla iˇsi. not-took.1s-neg with-me even thing “I didn’t take a single thing with me.”

(12)

ma-Qind-hæ wEla nıtfıt Xaˇgal. not-at-her even bit shame “She doesn’t have the least bit of shame!”

(13)

la-s-sEnna ma-baQti:-hUm wEla lUkmi Ekl. ˙ to-the-year not-give.1s-them even bite˙ food “For the [first] year I don’t give them even a bite of [solid] food.”

2 Following (Herburger 2001), “thematically entailed” means that the meaning of the verb entails the existence of an entity filling the thematic role in question.

27

The NPI-interpretation is available with adverbial wEla-DPs as well as inside PP (14) and smallclause complements (15): (14)

ma-kaQatt [P P gˇænıb wEla hada fi:-hUm ]. ˙ ˙ not-sat.1s next-to even one in-them “I didn’t sit next to even one of them.”

(15)

QUmri ma-ˇsUft-hæ [læ:bısi wEla nıtfıt hari:r ]. ˙ ever-I not-saw1s-her wear.fs even bit silk “I have never seen her wearing even a bit of silk.”

More than one wEla-phrase can have the NPIinterpretation at a time: (16)

ma-kUlt wEla iˇsi wEla la-hada fi:-hUm. ˙ ˙ not-said.1s even thing even to-one in-them “I didn’t give anything at all to even one of them.”

NQ-wEla cannot license NPI-wEla (17): (17)

Negative Concord and Locality

PA negative concord is generally strictly local dependency: an NPI wEla-phrase must be contained within the smallest clause containing its licensor. It cannot be separated from its licensor by the boundary of either a finite complement (19) or a non-finite/irrealis complement (18): (18)

* ma-waQatt Ehki wEla maQ hada fi:-hUm. ˙ even with one ˙ not-promised.1s talk in-them

(19)

* batwakkaQ-ıˇs ınnhæ bıthıbb wEla hada. ˙˙ ˙ ˙ believe.1s-neg that.3fs likes.3fs even one

Similar sentences with weaker NPIs such as hada ˙ or Paiy hada “anyone” are acceptable: ˙ (20)

ma-waQatt Ehki maQ ( Paiy ) hada fi:-hUm. ˙ with any one ˙ not-promised.1s talk in-them “I didn’t promise to talk with any of them.”

(21)

batwakkaQ-ıˇs ınnhæ bıthıbb ( Paiy ) hada. ˙ ˙˙ ˙ believe.1s-neg that.3fs likes.3fs any one “I don’t think that she likes ANY one.”

Likewise, negative concord fails if a wEla-DP is embedded inside another DP, while similar examples with Paiy-DPs are acceptable: (22) (23)

(24)

ma-bıddna [ nXalli wEla zElami ]. not-want.1p leave.1p even fellow “We don’t want to leave even one man.”

(25)

ma-Qırıft [ Ektıb wEla kılmi ]. not-knew.1s write.1s even word “I couldn’t write even one word.” mıˇs ka:dır [ yıPatPıt wEla nıtfıt ıPtPu:ta ]. ˙ ˙ ˙ ˙ ˙ even bit bow bow.1s not able.ms “I can’t bow my head even a little bit.” wEla Xe:t ]. ma-rah tikdar [ taˇgmaQ ˙ ˙ not-fut˙ able.2ms gather.2ms even thread

(26)

* wEla hada kal-l-i wEla kılmi. ˙ ˙ not.even one said.3ms-to-me even word

It follows from the distributions of NQ- and NPIwEla that wEla-phrases are blocked from postverbal argument positions which are thematically entailed and which are not within the scope of an antimorphic operator. 1.1

(18-23) suggest that negative concord is a bounded dependency like agreement marking, thematic licensing, or reflexive binding. However, there are exceptions to this generalization. “Long-distance” negative concord occurs with wEla-DPs inside the complements of a small class of verbs including subject control verbs like bıdd- “want” (24), ha:wal “try” (29), kıdır “be ˙ able” (26), or Qırıf “know how to, be able˙ to” (25) and object-control verbs like Xalla “allow” (28):

* ma-hake:t maQ [ bınt wEla hada fi:-hUm ]. ˙ ˙ not-talked.1s with girl even one in-them ma-hake:t maQ [ bınt Paiy wa:had fi:-hUm ]. ˙ with girl any one ˙ in-them not-talked.1s “I didn’t talk to the daughter of any one of them.”

(27)

(28)

“You won’t be able to gather even a thread.” ma-Xallu:-ni:-ˇs [ ıˇstari wEla iˇsi ] not-let.3mp-me-neg buy.1s even thing “They wouldn’t let me buy even one thing!”

The embedding can be recursive, provided that only verbs in this class are used (29). (29)

bıddi:-ˇs aha:wıl Ehki wEla maQ hada. ˙ ˙ ˙ want.1s-neg try.1s speak.1s even with one “I don’t want try to talk with anyone at all.”

These verbs correspond to verbs found in many other languages which trigger a process often referred to as restructuring or clause union. I follow (Aissen & Perlmutter 1983) in calling them trigger verbs. Restructuring involves the “stretching” of the domain of locality for certain kinds of bounded dependencies from the complement of a trigger verb to include the clause that it heads. At present no other phenomena have been identified in PA which independently indicate restructuring. However, long-distance negative concord is identified as a restructuring phenomenon in several languages such as West Flemish (Haegeman & Zanuttini 1996, a.o.), Polish (Dziwirek 1998, a.o.), and Serbian (Progovac 2000, a.o.). As such, I hypothesize that long-distance negative concord in PA is a form of restructuring as well. All PA trigger verbs take non-finite complements headed by the “y-imperfect” stem of a verb agreeing with the controlled subject. In addition, some of the verbs in question allow their complements to optionally include complementizers, even in negative concord sentences (30-31):

28

(30)

(31)

ma-bakdar [ ( ın-ni ) akul-l-ak wEla ˇseiy ]. ˙ ˙ not-able.1s that-I say.1s-to-you even thing “I can’t say to you anything at all." ma-ha:walt-ıˇs [ ( ın-ni ) ahki wEla maQ hada ˙ ˙ ˙ not-tried.1s-neg that talk.1s even with one fi:-hUm ]. in-them “I didn’t try to talk with even one of them.”

The complementizer Pınn- “that” also appears in indicative complements (see 21 and 19 above). It hosts a pronoun clitic corresponding to the subject of the clause and precedes the negation marker: (32)

ha:walt ınn-i ma atkallam Qan nafs-i . ˙ try.perf.1s that-me not speak.1s about self-me

II Multiple wEla-DPs can be licensed at once (16). III Negative concord is generally clause-local (18-19) IV Exceptions to III occur in sentences in which the matrix verb is one of a small set of verbs that allow a “longdistance” negative concord between a matrix negation and an embedded wEla-DP (24-31). V Long-distance negative concord is licensed inside recursive embeddings, provided that the embedding verbs all belong to the verb class described in IV (29). VI The verbs which allow restructuring vary as to the size or category of the complements they take (30-34). VII wEla NPs in long-distance negative concord can be interpreted within the embedded clause (35-36).

2

Theoretical Implications

“I tried not to speak about myself.”

Other triggers verbs like bıdd- “want” and Xalla “let” exclude the complementizer: (33)

ma-bıddna ( *in-na ) nXalli wEla zElami. not-want.1p that-we leave.1p even man “We don’t want to leave even one person.”

(34)

( *ın-ni ) aku:l wEla kılmi. ma-Xallu-ni:-ˇs ˙ even word not-let.3mp-me-neg that-I say.1s “They didn’t let me say even one word!.”

Assuming that the presence of Pınn- and of verbal agreement marking indicate different functional categories, (30-34) show that trigger verbs vary as to the kinds of complements they take. Lastly, embedded wEla-DPs can (although need not) be interpreted in-situ. For example, in (35) a pronoun within the wEla-DP is bound by the NPI hada “anyone.” The NPI is interpreted within the ˙ scope of the complement clause, and therefore the wEla-phrase must be as well. (35)

b-akdar-ıˇs adfaQ hada wEla kırˇs mın ra:tıb-u. ˙ ˙ ˙ ˙ can.1s-neg pay.1s one.ms even cent from pay-his “I can’t pay anyonei even a penny of hisi salary."

Similarly, (36) can be said by a pauper with grand plans for getting rich and who is speaking about money that exists in his or her desire worlds: (36)

bıddi:-ˇs adfaQ wEla kırˇs Dara:yıb. ˙ ˙ want-1s-neg pay.1s even penny taxes “I don’t want to pay even a penny in taxes."

This shows that wEla kırš “even a penny” takes scope within the embedded clause. In sum, negative concord in PA has the following properties: I wEla-DPs within the scope of the verb and in thematically-entailed positions have only the NPIinterpretation and must be licensed by negation morphemes or bıdu:n “without” (10-13, 14-15).

The data raise two theoretical questions about negative concord in PA: (i) What mechanisms license it? (ii) Why do restructuring verbs allow long-distance licensing? I address these questions by looking long-distance negative concord, as this reveals the most about the properties of both negative concord and restructuring in PA. 2.1

Implications of the Data

Based on properties IV-VII several formal aspects of negative concord in PA can be inferred which narrow down the number of theoretical options available for analyzing the data. First, the NQ- and NPI-interpretations of wElaphrases arise from a lexical ambiguity between two homophonous morphemes. A theory which treated wEla-phrases as being uniformly negative quantifiers or negative polarity items would rely on global licensing mechanisms, such as a semantic construal mechanism (Haegeman & Zanuttini 1996, de Swart & Sag 2002) or insertion of an “abstract negation,” and would incorrectly predict a sentence like (17) to be acceptable. Second, negative concord is a purely syntactic phenomenon in PA. If it were a semantic process, long-distance negative concord would be predicted to be more generally available in embedded clauses. Instead, the availability of longdistance negative concord is a lexical ideosyncracy of an otherwise heterogeneous set of verbs. Third, negative concord licensing requires neither overt nor covert movement, as the wEla-DP is pronounced and can be interpreted in its base position. This rules out approaches to negative concord according to which wEla-DPs must raise to a local configuration their licensors (Haegeman & Zanuttini 1996, Watanabe 2004, Zeijlstra 2004).

29

If no movement is involved, then some other syntactic licensing mechanism must be. Fourth, the availability of long-distance negative concord is not a matter of complement size. This excludes an analysis based Wurmbrand (2001), according to whom restructuring complements are bare VPs. Rather, PA trigger verbs take complements which include functional structure. 2.2

Negative Concord as Feature Matching

These properties suggest a parallel between longdistance negative concord in PA and long-distance agreement in Hindi as analyzed by Bhatt (2005). Long-distance agreement in Hindi consists of an object of an embedded verb determining the agreement form of both the matrix and embedded verb in clauses headed by one of a small set of control verbs which correspond closely to trigger verbs in languages likes Spanish, Italian, etc: (37)

Vivek-ne [ kit¯ab parhn¯ı ] ˇc¯ah¯ı. ˙ Vivek-erg book.fs read.inf.f wanted.fs “Vivek wanted to read the book.”

Bhatt argues that long-distance agreement does not correlate with movement because the object can be interpreted with narrow scope, as in (38): (38)

Usha-ne [potluck keliye d¯al ban¯an¯ı] ˇc¯ah¯ı. Usha-erg potluck for daal.f make.f wanted.f “Usha wanted to make daal for the potluck.”

Likewise, Bhatt argues that the agreement marking on the embedded verb indicates functional structure in a restructuring complement and that long-distance agreement consists of a “parasitic” agreement relation in which the agreement form of the matrix verb is determined by the agreement form of the embedded verb. PA long-distance negative concord and Hindi long-distance agreement share the following:

2.3

Analysis

I assume that long-distance negative concord involves a “polarity” feature [ POL±]. To implement the interaction of uninterpretable and interpretable instances of the polarity feature, I assume Bhatt’s AGREE relation, but in order to emphasize that verb-argument agreement is not involved, I refer to it as ACCORD. Also, following Hiraiwa (2001), I assume that “multiple ACCORD” is possible, meaning that a Probe can simultaneously enter an ACCORD relation with multiple Goals with respect to a feature F provided that they have non-distinct values for F . This is essential for modeling examples like (16). Following standard assumptions after Chomsky (2000), ACCORD is constrained by the Phase Impenetrability Condition, which blocks that ACCORD relations across phase boundaries. I assume the following principles of grammar: (42)

Uninterpretable features are unvalued, and must be provided a value (Chomsky 2000);

(43)

Selectional features are sets of feature specifications including category, mood, and polarity;

(44)

Root clauses must be [ POL +] (the root clause polarity condition ) after Dowty (1994)); vP, CP, and DP are phases.

(45)

(44) is a stipulation, but can be related to proposals by Progovac (2000) and Przepiórkowski & Kup´sc´ (1999) according to which the semantic reflex of a negative concord is the specification of a negative event, an event which fails to meet a certain description. If (44) reflects a requirement that a root clause must be interpreted as asserting the existence of an eventuality, a negative clause would assert the existence of an eventuality that doesn’t meet the description provided by the predicate. PA has the following lexical properties:

(39)

They involve a morphological matching relation;

(46)

v 0 has an unvalued polarity feature [ POL x];

(40)

The relation is bounded except in restructuring;

(47)

wEla has an interpretable [ POL -] feature;

(41)

Long-distance licensing involves no movement;

(48)

The negation morpheme ma:- includes a [ POL -] feature among its selectional features, but projects a [ POL +] feature;

(49)

Trigger verbs (along with auxiliary verbs) do not specify a polarity feature for their complements;

(50)

Non-trigger control verbs include a [ POL +] feature among their selectional features.

Bhatt’s treats restructuring complements as lacking a PRO subject. The lack of a subject NP leaves the embedded T0 (Inf0 according to Bhatt) to enter an AGREE relation with the embedded object. The matrix T0 then enters an AGREE relation with the embedded T03 . As such, Hindi longdistance agreement does not actually involve a long-distance relation. Rather, it involves a chain of purely local AGREE relations. 3

Bhatt modifies Chomsky’s (2001) Agree to disassociate case feature checking from φ-feature checking.

The analysis for negative concord in a root clause is as follows: first, given the structure in (51), v 0 has an unvalued [ POL x] feature. It c-commands the wEla-DP and so enters into ACCORD with the it, with the result that its unvalued polarity feature is valuated as [ POL -]:

30

vP

(51)

v

VP šUft

v [ POL x]



DP wEla [ POL -]

NP hada fi:-hUm ˙

Further derivation builds an TP, which is then merged with ma:-, satisfying its selectional feature, and projecting an FP with a [ POL +] feature: (52)

FP [ POL +]

ma:

In the case of a restructuring complement lacking Pınn-, the vP is merged with the trigger verb. The matrix v 0 has an unvalued feature, and ccommands the embedded vP. The c-command relation does not cross a phase boundary, and so an ACCORD relation is established between the matrix v 0 and the embedded v 0 : vP

(55)

v0 v [ POL x]

VP bıddi



TP [ POL -]

Ehki wEla maQ hada fi:-hUm ˙ ˙

PRO

T

vP

T T šUft

wEla hada fi:-hUm ˙

A clause with an unlicensed wEla NP is ill-formed because it is rooted in a node with a [ POL -] feature, violating the root clause polarity condition. A clause with multiple wEla-DPs (as in 16 above) is derived as before, except that v 0 enters into ACCORD with all of them simultaneously: vP

(53)

v v [ POL x]

VP kUlt ˙

DP [ POL -]

V ˙

wEla iši

DP [ POL -]

wEla la-hada fi:-hUm ˙

A sentence with long-distance negative concord is derived as follows: as in (51) and (53), v 0 enters into ACCORD with the wEla-DP, so that its unvalued polarity feature is valuated as [ POL -]:

The derivation then proceeds as in (51). In examples like (30-31), I assume that Pınnoccupies T0 rather than C0 . Awad (1998) shows that Pınn- in indicative complements affects the pragmatic interpretation of the clause. This effect is absent in control complements containing Pınn, suggesting that the Pınn- in control complements (“nonfinite Pınn-”) is homophonous with indicative Pınn- but is a distinct morpheme. Mitchell & al Hassan (1994, p. 38) note that the use of Pınn- also indicates a shift a slightly more formal register of colloquial speech. Nonfinite Pınn- may be a calque from the Classical Arabic particle Pan which introduces subjective complements and is likely a T0 morpheme. In Palestinian and other dialects, Classical Pann- and Pan have fallen together, so PA Pınn- is the dialectal morpheme corresponding to Classical Pan. Therefore, I suggest that Pınn- spells out the head of T0 . This has important consequences for the analysis. Nonfinite Pınn- projects a TP. TP is not a phase, and therefore does not block an ACCORD relation with the matrix vP. Therefore, the derivation of an example like (31) proceeds just like the derivation in (52), modulo the presence of a TP projection in the complement: vP

(56)

vP

(54)

v0 v v [ POL x ]

vP [ POL -]

VP Ehki ˙

˙

VP

v ha:walt [ POL x] ˙ ˙

DP

TP Pınn-

wEla [ POL -]

PP maQ hada fi:-hUm ˙

vP [ POL -]

Ehki wEla maQ hada fi:-hUm ˙ ˙

31

Capturing the failure of long-distance negative concord with non-trigger vebs like waQad “promise” in (18) requires an additional stipulation: T0 has an unvalued [ POL -] feature as well, such that merging T0 with vP results in T0 having its [uPOL ] feature valuated by v 0 . Accord to (50), non-trigger verbs select complements with a [ POL +] feature. As such, long-distance negative concord in examples like (18) is blocked because this feature clashes with the [ POL -] value that the embedded T0 , blocking the derivation4 . Failure of negative concord with wEla-DPs in DP-internal positions (22) follows directly from the Phase Impenetrability Condition, as DPs are phases and block the ACCORD relation.

3 Conclusion I have presented an array of data describing longdistance negative concord in Palestinian Arabic. These data entail an analysis of negative concord and of restructuring which does not involve movement or reduced complements, and instead involves static feature matching. A typological implication of this is the term “negative concord” as applied to PA is to be taken literally, where concord is understood as a class of feature-matching relationships of which subjectverb agreement is just one instance. This means that formal devices used to express feature matching must be defined in a general way rather than just in terms of subject-verb agreement. The analysis captures the data, and may have interesting implications for how concord is modeled in the Minimalist Program. However, the analysis is largely a technical solution awaiting further phenomena to motivate it. Further research will consider additional factors, such as the roles that focus and prosody play in the locality restrictions on PA negative concord, and whether parallels can be drawn between negative concord and restructuring in PA on the one hand and comparable phenomena in other languages.

References Aissen, J. & Perlmutter, D. (1983), Clause reduction in spanish, in ‘Studies in Relational Grammar’, University of Chicago Press. 4 This analysis predicts that verbs which select for a [ POL ] complement should license negative concord. Whether such verbs exists in PA and therefore whether they do license negative concord has yet to be determined.

Awad, M. (1998), The syntax and semantics of complement clauses in arabic, in G. Hall, K. Homer, E. Lenell & L. Nicita, eds, ‘Colorado Research in Linguistics, v.16’, CRIL, University of Colorado (Boulder), pp. 1–29. Bhatt, R. (2005), ‘Long-distance agreement in hindi-urdu’, NLLT 23, 757–807. Chomsky, N. (2000), Minimalist inquiries: The framework, in R. Martin, D. Michaels & J. Uriagereka, eds, ‘Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik’, MIT Press (Cambridge), pp. 89–155. Chomsky, N. (2001), Derivation by phase, in M. Kenstowicz, ed., ‘Ken Hale: A Life in Language’, MIT Press (Cambridge), pp. 1–52. de Swart, H. & Sag, I. A. (2002), ‘Negation and negative concord in romance’, L&P 25, 373–417. Dowty, D. (1994), The role of negative polarity and concord marking in natural language reasoning, in M. Harvey & L. Santelmann, eds, ‘Proceedings of SALT IV’, Cornell University, Ithaca, New York, pp. 114–144. Dziwirek, K. (1998), ‘Reduced constructions in universal grammar: Evidence from the polish object control construction’, NLLT 16, 53–99. Haegeman, L. & Zanuttini, R. (1996), Negative concord in west flemish, in A. Belleti & L. Rizzi, eds, ‘Parameters and Functional Heads’, Oxford, pp. 117–179. Herburger, E. (2001), ‘Negative concord revisited’, NLS 9, 289–333. Hiraiwa, K. (2001), Multiple agree and the defective intervention constraint in japanese, in ‘Proceedings of the HUMIT 2000’, Vol. 40 of MITWPL, pp. 67–80. Mitchell, T. & al Hassan, S. (1994), Modality, Mood, and Aspect in Spoken Arabic, Keegan Paul International. Progovac, L. (2000), Coordination, c-command and ‘logophoric’ n-words, in L. Horn & Y. Kato, eds, ‘Negation and Polarity: Syntactic and Semantic Perspectives’, Oxford University Press (Oxford), pp. 88–114. Przepiórkowski, A. & Kup´sc´ , A. (1999), Eventuality negation and negative concord in Polish and Italian, in R. D. Borsley & A. Przepiórkowski, eds, ‘Slavic in HPSG’, CSLI Publications, Stanford, pp. 211–246. Vallduví, E. (1994), ‘Polarity items, n-words, and minimizers in catalan and spanish’, Probus 6, 263–294. Watanabe, A. (2004), ‘The genesis of negative concord: Syntax and morphology of negative doubling’, LI 4, 559–612. Wurmbrand, S. (2001), Infinitives, Restructuring and Clause Structure, Mouton de Gruyter. Zeijlstra, H. (2004), Sentential Negation and Negative Concord, PhD thesis, University of Amsterdam. Zwarts, F. (1996), Three types of polarity items, in F. Hamm & E. Hinrichs, eds, ‘Plural Quantification’, Kluwer (Dordrecht).

32

CONCORD PHENOMENA IN FIRST LANGUAGE ACQUISITION Jacqueline van Kampen and Arnold Evers UiL OTS, Utrecht University [email protected] Abstract Scope-bearing items for sentential negations and sentential questions appear in Dutch child language as ‘doubling’ constructions. A or element appears in sentence-initial position and is doubled in sentence-final position. The ‘doubling’ is not part of the adult system nor is it part of the child’s acquisition input. The temporary doublings suggest that the grammatical property of sentential scope is learned in a stepwise fashion by means of intermediate grammars that are temporarily simplified. The present view opposes to the idea that temporary constructions in child language betray an innate UG option (v.Kampen 1997, Crain et al. 2006).

1 Learnability background Children do not design a grammar reflexively, but they certainly develop over time the feel for grammatical consequences. A recent suggestion by Chomsky (2005) seems relevant in this respect. Chomsky suggests three separate factors that play a part in the set-up of grammar. With language acquisition in mind, we read these three factors as follows. (1) 1. General grammatical principles. These are revealed by comparative grammar. They enter the child’s acquisition procedure as a frame of grammar that is somehow genetically inherent in the human species (a faculty of language). 2. Primary input data. These enter the acquisition procedure as child directed speech. In the beginning, they will be not or only marginally interpretable. 3. General cognitive principles of the human mind. These are not necessarily specific to grammar. As for the third factor, we imagine things like getting the feel for musical structures, technical designs, card-, board- and field-games, various kinds of craftsmanship, all kind of physical abilities such as biking through city traffic, and so. You do not acquire this seemingly endless range of competences by much reasoning about them. Their internal logic somehow and mysteriously takes shape by practice. The same holds of course for mastering a language as well. This suggests to us something not directly advanced by Chomsky’s three points, but not excluded by them either. The first point, - the presumed biological frame for grammar -, could as well be something like an outcome, an optimal compromise when factor 2 and factor 3 interact, rather than being an independent source. Certain grammatical principles need not be innate. They may be invented and acquired. Their appearance in various further unrelated grammars is nevertheless reasonable, because they solve something. Not unlike techniques for swimming, fishing or building canoes, they are easily invented or reinvented when the circumstances are favorable. Possible solutions for grammatical problems may come up in child language but are given up later on, when the data from the specific grammar do not sufficiently support them.

33

2 Dutch acquisition data We will analyze two ‘concord’ phenomena, a double feature (2) and a double wh feature (3), along the line sketched above. The data come from Van Kampen (diary notes 1989-1994; CHILDES Van Kampen corpus), Coopmans (1995), Van der Wal (1996). (2)

a. b. c. d.

kwil nie(t) badje niet kan nie(t) vinden niet die is nie(t) ei niet hoort niet in kamer niet

((I) wanna not bath not) ((I) cannot find not) (that is not egg not) ((it) belongs not in room not)

(3)

a. b. c. d.

wa(t) (i)s dat nou ? (wa)t ga jij doen nou? waar (i)s die meneer nou? wa(t) (i)sse buiten nou?

(what are you going to do ‘now’?) (what wil you do ‘now’?) (where is that man ‘now’?) (what is there outside ‘now’ ?)

The first neg-element in (2) and the sentence-initial -element in (3) are most of the time cliticized to something that will later reveal itself to the child as ‘the finite verb’. The second neg-element niet in (2) and the second question marker nou in (3) take the sentencefinal position, may be as a ‘tag’. Both doubling phenomena are clearly manifest in some period of Dutch child language between 2-3 years, but they disappear later on. When the concord phenomena are not present in the primary data (and in adult Dutch they are not), it is not clear why the language acquisition device should mobilize such a device from options given by factor 1 in (1). By contrast, when grammatical innovations by the child can be motivated as rational reactions to perceived distributional problems, we get something like an explanation. Be it that we are now more in the free creative domain of factor 3 than in the preprogrammed options of factor 1. 2.1 Double Dutch children start their first negated utterances with a simple neg-element niet. In the twoword stage, niet appears in initial or final position. These forms are easily derived as reductions from maternal input sentences as indicated. (4)

Data from Sarah (Van Kampen corpus) age in years;months.days/weeks a. niet juie (= luier) (not diaper) (Sarah 1;10.13 / w.98) wil je [niet (je) luier] aan? (want you not your diaper on ? = don’t you want) b. boekje niet (book not) (Sarah 2;0.17 / w.107) ik wil dat [boek niet] (I want that book not = I do not want that book)

The early reduction leaves out articles and most grammatical morphology. When the sentences become longer and quasi-finite modals/auxiliaries appear as standard pragmatic markers of the utterance (De Haan 1987, v.Kampen 2005), the patterns in (5) arise. (5)

a. b. c. d.

doet Laura niet kan liedje niet vogeltje is niet deze hoefeniet oppe dak

(does Laura not) (can song not) (birdie is not) (this need not on roof)

(Sarah 2;0.17 / w.107) (Sarah 2;0.17 / w.107) (Sarah 2;2.18 / w. 116) (Sarah 2;2.18 / w. 116)

The neg-element continues to show up in sentence-final position (‘tag’-like) with a denotational (content) phrase or word as in (5)a,b but it may also appear sentence-initially (‘operator’-like), fused with a modal marker {kannie(t), moe(t)nie(t), hoe(f)nie(t), wilnie(t)}

34

(‘cannot’, ‘must not’, ‘need not’, ‘want not’) or with an auxiliary-like marker isnie(t), zijnnie(t), hebnie(t), as in (5)c,d (cf. Hoekstra & Jordens 1994). The doublings in (2), repeated in (6) for Sarah (Van Kampen corpus), come in when the operator phrase and the denotational phrase appear combined. (6)

a. [kwilnie(t)] badje niet b. [kannie(t)] zien helemaal niet c. die [isnie(t)] ei niet

((I) wanna not bath not) ((I) cannot see at all not) (that is not egg not)

(Sarah 2;4.1 / w.122) (Sarah 2;2.10 / w. 114) (Sarah 2;4.1 / w.122)

A denotational phrase or word [badje] (‘bath’), [zien helemaal] (‘see at all’), [ei] (‘egg’), [s(er)vetjes] (‘napkins’) is preceded by a negated operator-like modal/auxiliary. The negative tag is not a property of the adult input, but a neg-element in final position may appear easily in the adult V-2nd input, see (7). (7)

a. wil Laura ‘t popje niet tV ? (wants Laura doll not? = doesn’t Laura want the doll?)) b. Laura pakt ‘t popje niet tV

(Laura takes the doll not = Laura doesn’t take the doll))

A system that starts with binary expressions may develop into a double negation system when the initial modal phrase with fused negation is combined with a negated denotational phrase, see (8). (8)

a. Laura wilniet pap niet eten b. Laura hoe(f)nie pop niet pakken

Laura wil niet pap niet (eten) hoef niet niet - pakken

2.2 Double Dutch children start with wh-questions marked with a simple -element nou. In the twoword stage, the -element appears in initial or final position, see (9). These forms are easily derived as reductions from maternal input sentences as indicated. (9)

Data from Sarah (Van Kampen corpus) age in years;months.days/weeks a. dat nou ? (that ‘now’ ?) (Sarah 1;10.13 / w.98) wat is [dat nou] ? (what is the sweater ‘now’ ?) b. nou eend ? (‘now’ duck ?) (Sarah 2;0.17 / w.107) waar is [nou (een) eend] ? (where is ‘now’ a duck ?)

The early reduction leaves out articles and most grammatical morphology. When the sentences become longer and quasi-finite modals/auxiliaries appear as standard pragmatic markers of the utterance, the patterns in (10) arise. For some period nou appears as content-question marker on its own. (10) a. is dat nou ? b. ga doen nou ? c. is mij(n) stoel nou?

((what) is that ‘now’ ?) ((what goes do ‘now’ ?) ((where) my chair ‘now’ ?)

(Sarah 2;1.10 / w.110) (Sarah 2;2.10 / w.115) (Sarah 2;3.27 / w.120)

The adverb nou may also be added in adult questions. It has an emphatic function in questions and stresses that the speaker had already expected something like that, see (11)c. The Dutch

35

child does not perceive the emphatic function. For the child nou simply marks the question and she feels free to drop the question word in sentence-initial position, see (11)a. (11) a. early child Dutch: b. adult Dutch/later child Dutch: c. adult Dutch (emphatic):

doe jij nou ? wat doe jij ? wat doe jij nou ?

((what) are you doing ‘now’ ?) (what are you doing ?) (what are you doing ‘now’ ?)

When the wh-question words are added, the sentence-final nou is maintained for some time, but now a sentence-initial wh-element appears simultaneously and is fused with a modal marker {was (wat is), wasse (wat is er), waars (waar is), tga (wat ga)} (‘what is’, ‘where is’, what goes’). The doublings in (3), repeated in (12), come in when the operator phrase and the denotational phrase appear combined. (12) a. b. c. d.

was [=wat is] dat nou ? wasse [=wat is er] buiten nou? waars [=waar is] koning nou? tga [=wat ga] jij doen nou?

(what is that ‘now’?) (what is there outside now?) (where is the king ‘now’?) (what will you do ‘now’?)

(Sarah 2;2.10 / w.115) (Sarah 2;2.26 / w.117) (Sarah 2;3.17 / w.119) (Sarah 2;3.27 / w.120)

It seems arbitrary and not yet necessary to analyze the initial words in (12) as wh-pronouns combined with a copula. They rather are fused into a single question operator, just like the element nou in sentence-final position. 2.3 V-2nd and the disappearing of the early doublings in child language As we have seen above, both the wh-element and the neg-element niet appear in the same period (before week 122) as part of the sentence-initial group [wh+modal/auxiliary] and [modal/auxiliary+niet]. And both are repeated in the denotational part of the sentence, be it for a short period. Roughly, between week 110-123 Sarah is acquiring the V-2nd rule. A finite verb in second/first position becomes a standard grammatical property. See the graph in (13). (13) Dutch Sarah: The acquisition of V-2nd (marking) (from: Van Kampen 2004) 100

percentage

80

V-second

60 40 20 0 85

95

105

115

125

age in weeks

135

145

155

The rising percentage of finite verb placement in (13) is initially due to the operator-like verbs (auxiliaries, modals and copulas). Only later in the second half end of the graph in (13) (weeks 115-122) the finite form of denotational verbs comes in as well (see De Haan 1987, Van Kampen 1997, Evers & Van Kampen 2001). When the -variants of denotational verbs appear in the sentence-initial group, the operator elements (fixed forms of modals and auxiliaries) are reanalyzed as -elements of a verbal paradigm. The now empty place in the sentence-final group leads to a reanalysis of all sentence-final tags (nou? niet!). All these tags can now be reinterpreted as preverbal adverbs {niet + V and niet + tV} {nou + V and nou + tV}. The sentences continue to be intended as CP and CP. Scope implies that marking the foot of the verbal chain suffices for sentential scope. Sentential scope is thereby acquired.

36

2.4 Acquisition steps So far, we argued that initial child language derives two types of negation from reducing the parental input as binary constructions. There is a type of negation fused with the modal elements {kannie(t), moe(t)nie(t), hoe(f)nie(t), wilnie(t)} issenie(t)} (cf. Hoekstra & Jordens 1994). These neg-elements characterize the illocutive/pragmatic type of the utterance, see (5). There is also a neg-element added to a denotational word {niet bad/bad niet}, see (4). These constructions may be seen as reductions from more elaborate input, such that they fit into a binary construction. The combination of the negative modal elements and the negative denotational utterances lead to temporary doublings, not available in the input, see (6). The acquisition of the finite verb rule V-2nd changed the situation. The finite verb elements (denotational verbs as well as modals, auxiliaries and copulas) appear in the sentence-initial group, but are related to an ‘empty’ position in the final group. Initial and final group are from now on transformationally related. The fusions in the initial group are reanalyzed as the neg-elements already known as an adverb of the final group. The neg-doublings disappear. The neg-element related to the finite verb position in the final group obviously enjoys scope over the verbal chain and is thereby acquired as having scope over the CP as a whole. The neg-element niet, as well as the question element nou, do no longer appear in ‘tag’ positions. They precede the predicate denotational elements. The analysis of V-2nd acquisition is a notorious issue in Germanic syntax, but it is a matter of acquisition steps. The V-2nd construction does not follow from a sentence-reduction. See de Haan (1987), and in his wake Van Kampen (1997) and Blom (2003). See the graph in (12) where the two verticals show the period of and doublings around the acquisition point of the V-2nd rule. 3 Negation and quantifier scope The end of the doublings in (6) and (12) is not yet the end of doublings as a temporary way out for the acquisition procedure. A year or so after the acquisition of V-2nd and the extended scope of niet and nou, quantifier elements appear in child Dutch. Thereby, a problem arises for their scope in negative sentences. See the following attempts by the Dutch child. (14) a. child Dutch : niemand speelt niet met mama tV b. adult Dutch : niemand speelt − met mama tV

(nobody plays not with m.) (nobody plays with mummy)

(15) a. child Dutch : iedereen speelt niet met mama tV b. adult Dutch : niet iedereen speelt − met mama tV

(everybody plays not with m.) (not everybody plays with m.)

Obviously, the child understands the negative status of the CP in (14) and (15) and she maintains her hard won neg-element at the beginning of the final group. Although the position in front of the final group should give the neg-element CP scope, it does not establish in the adult grammar scope over the quantifiers niemand (‘nobody’) and iedereen (‘everybody’). Quantifiers define (in the adult grammar) a scope hierarchy over all elements c-commanded to the right. Hence, they would define (in the adult grammar) scope over negation contrary to the intention. For that reason, the adult input places the neg-elements more to the left. This step deviates from the earlier rule for negative placement. It turns out to be a very difficult property to acquire, often to the amazement of adult speakers, like the mother of Sarah. Yet, no matter how faultless the mother addresses her children, we believe they will all pass through an intermediate stage of a tentative neg-element in the final group that has scope over the universal quantifier on the left. One may even expect in the distant future a variant of Dutch that maintains the original solution by the children that final negation has automatically scope over the quantifiers on the left. Within that system, negative quantifiers are in for an

37

interpretation by negative concord, maybe emanating from an abstract operator left at the CPtop, as proposed for Czech by Zeijlstra (2004: 251), see (16). (16)

CP abstract neg-operator

CP neg-quantifier niemand

neg-adverb niet

tV

4 Perspective: The learnability of island constraints Negation-related markers and question-related markers are both scope-bearing items. For that reason, their acquisition story shows a few parallels in spite of their considerable differences. Both demonstrate a local referential point Do (argument) or Io (event) that is connected with a scope property CP or CP. Scope-bearing items suggest a ‘bi-local’ presence that shows up in doubling phenomena in child languages and often in the adult systems as well. They do so in Dutch child language in the short period that the V-2nd rule gets acquired and sentences are established as scope-bearing constructions. There are of course differences between the scope properties of and . The wh-marked phrases are subject to movement and sensitive to strong syntactic islands. The neg-marked phrases remain in situ and are sensitive to weak islands when they interact with a quantifier. Can strong and weak islands be acquired or are they mysteriously imposed by an idiosyncrasy of the human nervous system? In Evers and Van Kampen (to appear) we show that the learnability of islands is less forbidding than it seems. References Blom, E. (2003) From Root Infinitive to Finite Sentence PhD thesis Utrecht University. Chomsky, N. (2005) ‘Three factors in language design’, Linguistic Inquiry 36 (1), 1-22. Coopmans, P. (1995) ‘Taalontwikkeling Tim’, LINK 6-4. Crain, S., T. Goro & R. Thornton (2006) ‘Language acquisition is language change’, Journal of Psycholinguistic Research 35 (1). Evers, A. & J. van Kampen (2001) ‘E-language, I-language and the order of parameter setting’, UiL OTS Working Papers 00105-SS, Utrecht University. Evers, A. & J. van Kampen (to appear) ‘Parameter setting and input reduction’, to appear in: T. Biberauer (ed.) The Structure of Parametric Variation Amsterdam: John Benjamins. Haan, G. de (1987) ‘A theory-bound approach to the acquisition of verb placement in Dutch’, in: G.J. de Haan & W. Zonneveld (eds.) Formal Parameters of Generative Grammar III, 15-30. Hoekstra, T. & P. Jordens (1994) ‘From adjunct to head’, in T. Hoekstra & B.D. Schwartz (eds.) Language Acquisition Studies in Generative Grammar. Amsterdam: John Benjamins 119-149. Kampen, J. van (1997) First steps in Wh-movement Delft: Eburon. Kampen, J. van (2004) ‘Learnability order in the French pronominal system’, in: R. BokBennema, B. Hollebrandse, B. Kampers-Manhe & P. Sleeman (eds.) Selected Papers from Going Romance 2002, 163-183, Amsterdam: John Benjamins. Kampen, J. van (2005) ‘Language-specific bootstraps for UG categories’, International Journal of Bilingualism 9-2, 253-277. Wal, S. van der (1996) Negative Polarity Items and Negation. Tandem Acquisition. PhD thesis University of Groningen. Zeijlstra, H. (2004) Sentential Negation and Negative Concord PhD thesis UvA.

38

The syntax of Catalan and Spanish pre-verbal n-words Susagna Tubau Universitat Autònoma de Barcelona [email protected] 1. Introduction The present paper addresses the phenomenon of Negative Concord (NC) in Catalan and Spanish, with special emphasis on the licensing of n-words in pre-verbal position and their interaction with focus. As shown in section 2, Catalan is difficult to classify as either a Strict or a Non-Strict NC variety, since the sentential negative marker seems to be optional when Catalan n-words occur pre-verbally. Spanish, on the contrary, is clearly a Non-Strict NC variety. Following Roeper (1999), I assume that the optionality of the negative marker in Catalan NC constructions with pre-verbal n-words corresponds to the co-existence of two distinct varieties of Catalan: one which implements Strict NC, and one which is a Non-Strict variety. I refer to the former as Catalan A and to the latter as Catalan B. While Catalan B can be analysed on a par with Spanish, Catalan A displays some focus phenomena involving pre-verbal n-words that are not possible in Spanish/Catalan B. The paper is organised as follows. In section 2, the relevant data for Catalan and Spanish NC constructions are presented. Section 3 discusses the core assumptions the present analysis hinges on, whereas section 4 summarises the main conclusions. 2. The data As illustrated in the examples in (1), Catalan and Spanish n-words are licensed by the sentential negative marker when they occur post-verbally. Alternatively, a pre-verbal nword can license a post-verbal n-word (or more). The absence of the negative marker (or another suitable licensor) yields ungrammaticality. (1)

a.

*(No) comprarem res *(No) compraremos nada *(Not) will-buy.1.PL n-thing ‘We will not buy anything’

(Catalan) (Spanish)

b.

Ningú comprarà res Nadie comprará nada N-person will-buy.3.SG n-thing ‘Nobody will buy anything’

(Catalan) (Spanish)

Catalan and Spanish, however, differ in one crucial respect: while the negative marker can optionally co-occur with Catalan pre-verbal n-words, its occurrence with Spanish n-words results in a non-concordant result. This asymmetry is illustrated in (2). (2)

a.

Ningú (no) comprarà el llibre (Catalan) N-person (not) will-buy.3.SG the book ‘Nobody will buy the book’

39

b.

Nadie (*no) comprará el libro (Spanish) N-person (*not) will-buy.3.SG the book ‘Nobody will buy the book’

The data in (2a) make it very difficult to classify Catalan as a Strict or a Non-Strict NC language, especially considering the fact that for a number of speakers, the sentential negative marker with pre-verbal n-words is used indistinctively. For an everyday growing number of speakers, however, Catalan is already a Non-Strict NC language like Spanish. Roeper (1999) claims that optionality in a speaker’s grammar follows from the cooccurrence of various separate grammars. In line with this idea, I assume that those Catalan speakers who switch between (3a) and (3b) below use two different grammars, labelled as Catalan A and Catalan B respectively, to derive NC structures. (3)

a.

Ningú no llegirà el llibre (Catalan A) N-person not will-read.3.SG the book ‘Nobody will read the book’

b.

Ningú comprarà el llibre N-person will-read.3.SG the book ‘Nobody will read the book’

(Catalan B)

Catalan A and Catalan B/Spanish display a difference in the featural composition of the category T(ense). While in the former focused elements occupy a position in the left peripheral category Foc(us), T syncretically hosts focus features in the latter (Zubizarreta 1998: 103). Such a conclusion is reached on the basis of the fact that focused constituents are incompatible with pre-verbal subjects in Spanish, as shown in (4). According to Zubizarreta, this is because both occupy Spec, TP. (4)

a.

*Las ESPINACAS, Pedro trajo (y no las papas) (Spanish) the spinach Pedro brought (and not the potatoes)

b.

*Con MARÍA, Pedro habló (y no con Marta) with María Pedro spoke (and not with Marta)

(Spanish)

Zubizarreta (1998: 104) also shows that the same holds for focused n-words which undergo fronting to the left periphery. This is illustrated in (5). (5) a. b.

?*A NADIE, María le devolvió to nobody María DAT.CL returned

su manuscrito his manuscript

?*Con NADIE, María compartió su secreto with nobody María shared her secret

(Spanish) (Spanish)

For Catalan, on the contrary, Vallduví (1995) argues in favour of the availability of focus-fronting structures, which he calls focus-preposing. In line with Zubizarreta (1998), Vallduví (1995: 132) claims that “Spanish and Catalan contrast with respect to these phenomena. The Spanish equivalents of sentences like [6a] and [6b] are judged ungrammatical by many authors”. Interestingly, not every Catalan speaker accepts Vallduví’s (1995) examples, thus lending support to the hypothesis that the distinction

40

between Catalan A and Catalan B extends to other grammar domains beyond the licensing of pre-verbal n-words in NC structures. (6) a.

b.

Un ROLEX el Jordi s’ ha comprat a Rolex the Jordi self has bought ‘A ROLEX Jordi bought himself’

(Catalan)

El MERCEDES el Jordi no sap qui s’ ha comprat (Catalan) the MERCEDES the Jordi not knows who self has bought ‘The MERCEDES Jordi doesn’t know who bought’

3. Licensing of Catalan and Spanish pre-verbal n-words A recent proposal on the nature of NC put forward by Zeijlstra (2004) reduces the contrast between Strict and Non-Strict NC languages to the feature-interpretability of the sentential negative marker. In other words, in Strict NC languages, the negative marker is assumed to bear a [uNEG] feature which triggers the presence of an abstract negative operator (Op¬) which can check the [uNEG] feature of both the sentential negative marker and any post-verbal n-words present in the construction. The checking is done via Multiple Agree. In Non-Strict NC languages, on the contrary, the sentential negative marker bears an [iNEG] feature which can check the [uNEG] feature of post-verbal n-words. Pre-verbal n-words, conversely, are analysed as in Strict NC languages: since the sentential negative marker is incompatible with pre-verbal n-words (as it would yield a double negation reading), an Op¬ is needed to check the [uNEG] feature of n-words. When applied to the Catalan data in (3a) and (3b), Zeijlstra’s (2004) analysis would amount to saying that two separate sentential negative markers exist in Catalan –one with [uNEG] and another one with [iNEG], but with the same phonological realisation. This seems counterintuitive, since it advocates for an ambiguity account of the sentential negative marker for which there is no independent evidence outside the phenomenon of NC. My proposal, therefore, defends the idea that the sentential negative marker carries an [iNEG] feature in both varieties of Catalan. In addition, the present account also hinges on the assumption that Spanish and Catalan n-words bear an uninterpretable polarity feature, [uPol: ], which enters the derivation unvalued. Depending on which operator values this feature, the n-word will be interpreted as negative or as nonnegative. Another crucial assumption for the present piece of research is that n-words carry an uninterpretable focus feature, [uFoc], which is weak in the sense that it does not need to be checked in a Spec-Head configuration (Adger 2003) when focus is not phonologically prominent. The checking of the [uFoc] feature of n-words is claimed to take place at the distance by means of Agree (Chomsky 2000, 2001, 2005). The idea that n-words carry focus features is not new. Researchers such as Isac (2002) and Watanabe (2004) also endorse this position. For Isac (2002), n-words are indefinites that obtain their quantificational force from Focus. Pre-verbal n-words are assumed to raise to the left-periphery to check the focus features they carry. For Watanabe (2004), the [uFoc] feature of n-words is what sets the goal active for the Agree operation to take place between the sentential negative marker and the postverbal n-word.

41

In my account, an Op¬ with an [iNEG] feature values the [uPol: ] features of preverbal n-words in Spanish and Catalan B. T, which syncretically hosts focus features, checks [uFoc] at the distance by means of Agree. This is illustrated in (7c) and (7d). (7)

a. b.

Ningú comprarà res Nadie comprará nada N-person will-buy.3.SG n-thing ‘Nobody will buy anything’

(Catalan B) (Spanish)

c. d.

[TP Ningú [T[ifoc] [NegP Op¬[iNEG] [vP [uPol:Neg], [ufoc] [vP comprarà res[uPol:Neg], [ufoc] ]]]]] [TP Nadie [ T[ifoc] [NegP Op¬[iNEG] [vP [uPol:Neg], [ufoc] [vP comprará nada[uPol:Neg], [ufoc] ]]]]]

The most intriguing issue in relation to the data in (7) is why the sentential negative marker, if present, would lead to a double negation reading. Scholars such as Suñer (1995) simply state that no lexical material can intervene between the pre-verbal n-word and the verb for the NC reading to obtain and, thus, the sentential negative marker fails to lexicalize. However, it is not really the case that the sentential negative marker cannot occur in sentences of the type of (7c) and (7d), but the fact that if it does, the reading we obtain is non-concordant –i.e. the negative meaning of the negative marker is cancelled out– and the interpretation of the sentence is marked. Under Chomsky’s (2000, 2001 and 2005) latest work, the numeration is a one-time selection of all the lexical items that will participate in the derivation. This implies that if the sentential negative marker appears in the derivation, it must have been previously selected from the lexicon and then merged into the structure. In Spanish and Catalan B, it is observed that if no is part of the numeration, there is a blocking effect which prevents subject n-words to raise to a pre-verbal position if a concordant reading is intended. As shown in the examples in (8), this blocking effect is not found with regular subjects. (8)

a.

No comprará patatas nadie Not will-buy.3.SG potatoes n-person ‘Nobody will buy potatoes’

(Spanish)

b.

*Nadie no comprará patatas N-person not will-buy.3.SG potatoes

(Spanish)

c.

No comprará patatas Pedro Not will-buy.3.SG potatoes Pedro ‘Pedro will not buy potatoes’

(Spanish)

d.

Pedro no comprará patatas Pedro not will-buy.3.SG potatoes ‘Pedro will not buy potatoes’

(Spanish)

My intuition is that the focus features that n-words contain play a crucial role in the blocking. Actually, speakers of Catalan B seem to resort to the grammar of Catalan A when a subject n-word is focussed. That is, (9b) is preferred over (9a) when the subject n-word is heavily stressed.

42

(9)

a. b.

??RES N-thing

m’

ha passat has happened

(Catalan B)

RES no m’ ha passat N-thing not DAT.CL has happened ‘NOTHING happened to me’

(Catalan A)

DAT.CL

The example in (10) constitutes a further argument indicating that focus may be the key to explaining why the sentential negative marker does not co-occur with pre-verbal n-words in Spanish NC. As shown by the comma in (10b), nadie must receive stress with subsequent de-accenting of the prepositional complement a clase for the sentence to be acceptable. The exact characterisation of the nature of the observed blocking effects is left as future research, though. (10) a. b.

*No llevó el artículo Not brought. 3.SG the article

nadie a n-person to

clase class

(Spanish)

No llevó el artículo nadie, a Not brought. 3.SG the article n-person to ‘Nobody brought the article to class’

clase class

(Spanish)

Unlike Spanish/Catalan B, Focus constitutes a separate projection in Catalan A. The checking mechanism for pre-verbal n-words is similar to (7c), but with the difference that the sentential negative marker no values the [uPol: ] feature of both the subject1 and the object n-word, and the [uFoc] feature is checked by Focus rather than T. (11) a.

b.

Ningú no comprarà res N-person not will-buy.3.SG n-thing ‘Nobody bought anything’

(Catalan A)

[FocP [Foc [ifoc] [TP Ningú [T [NegP no[iNEG] [vP [uPol:Neg], [ufoc] [vP comprarà res[uPol:Neg], [ufoc]]]]]]]]

The fact that T does not have a focus feature and an extra position is needed can explain the facts in (12): for those speakers who use Catalan A and B interchangeably (i.e. for those speakers whose linguistic productions indistinctively display cooccurrence of pre-verbal n-words with the negative marker), (12b) is judged as more acceptable than (12a) when a non-subject n-word is focused. (12b) would arguably correspond to the grammar of Catalan A, which allows focused constituents to occur with subjects in Spec, TP, and where the sentential negative marker can co-occur with a pre-verbal n-word. (12) a. b.

?*NINGÚ, en Joan va N-person the John PAST.3.SG

veure see

NINGÚ, en Joan no va veure N-person the John not PAST.3.SG see ‘John saw NOBODY’

1

The issue of where subjects are generated in Romance is left aside. I assume that their original position is within the verbal phrase domain and that they further raise to higher positions for checking purposes.

43

In addition, it seems to be the case that for speakers who only use Catalan B, neither of the two examples are really acceptable, since both involve a doubly-filled Spec, TP. It is also predicted that speakers who only use Catalan A would totally reject (12a), since the [uPol: ] feature of n-words would be unvalued. Unfortunately, there are no speakers of the Catalan A variety only, since Catalan seems to be in a transitional stage moving from Strict NC to Non-Strict NC. Therefore, such a prediction remains untested. 4. Conclusion In this paper the apparent optionality of the sentential negative marker in NC constructions with pre-verbal n-words in Catalan has been dealt with. It has been assumed that there actually exist two different grammars of Catalan with respect to NC, labelled as Catalan A and Catalan B. These grammars mainly differ in the featural composition of the category T, which syncretically hosts focus features in the latter. In the former, by contrast, Focus has a separate functional projection. Several data related to the expression of focus and n-words have been presented. On the one hand, these data seem to lend support to the claim that Catalan A and B co-exist in the grammar of some speakers who indistinctively switch between Strict and NonStrict NC. On the other hand, some of the data also reveal very intriguing asymmetries between regular subjects and n-word subjects when the sentential negative marker is present in the derivation. Further research should confirm the intuition that the [ufoc] feature of n-words is responsible for subject n-words being unable to raise over the negative marker. References Adger, D. (2003). Core Syntax. A Minimalist Approach. Oxford: Oxford University Press. Chomsky, N. (2000). Minimalist Inquiries: The Framework. In R. Martin, D. Michaels and J. Uriagereka (eds.). Step by Step: Essays on Minimalist Syntax in honour of Howard Lasnik, 89-115. Cambridge, MA: MIT Press. Chosmky, N. (2001). Derivation by Phase. In M. Kenstowicz (ed.). Ken Hale: a Life in Language, 1-52. Cambridge, MA: MIT Press. Chomsky, N. (2005). On Phases. Ms. MIT, Cambridge, MA. Isac, D. (2002). Focus on Negative Concord. In R. Bok-Bennema, B. Hollebrandse, B. Kampers-Manhe and P. Sleeman (eds.). Romance Languages and Linguistic Theory 2002, 119-140. Amsterdam/Philadelphia: John Benjamins Publishing Company. Roeper, T. (1999). Universal Bilingualism. Bilingualism: Language and Cognition 2(3): 169-186. Suñer, M. (1995). Negative elements, island effects and resumptive no. The Linguistic Review 12: 233-273. Vallduví, E. (1995). Structural properties of information packaging in Catalan. In K. Kiss (ed.). Discourse configurational languages. Oxford: Oxford University Press. Watanabe, A. (2004). The Genesis of Negative Concord: Syntax and Morphology of Negative Doubling. Linguistic Inquiry 35(4): 559-612. Zeijlstra, H. (2004). Sentential Negation and Negative Concord. Ph.D. dissertation, University of Amsterdam. Zubizarreta, M.L. (1998). Prosody, Focus and Word Order. Cambridge, MA: MIT Press.

44

EVENT-ARGUMENT HOMOMORPHISM AS A CONCORD PHENOMENON Boban Arsenijevi!, LUCL, Leiden University [email protected] Abstract I argue that the Event-argument homomorphism (EAH) is very similar to the negative concord. After showing that EAH between quantified arguments and eventualities appears only with nonspecific arguments, and that the quantifier in such cases directly predicates only over the eventuality, I suggest that the quantifier is base-generated in the VP and not in the NP where it is lexicalized. The lexicalization is further argued to be an instance of agreement, parallel to n-words in the negative concord. In many languages, the quantifier is never lexicalized in the VP, because their lexicons have no units that realize quantificational predicates and may appear in the VP.

1 Introduction Event-argument homomorphism (EAH) presents one of the central topics of investigation in the field of aspect ever since its effects are observed (the first attempt to give it a formal account is Verkuyl 1972). EAH is illustrated in (1), where for the eventuality of eating, when the direct object is bare plural, the eventuality is attested a atelic (1a), and when the direct object is quantified, the eventuality is telic (1b). (1)

a. b.

John ate sandwiches for an hour/?in an hour. John ate five sandwiches in an hour/?for an hour.

The homomorphism is established with respect to boundedness. An unbounded quantity of the argument, denoted by the NP in bare plural, matches an unbounded reference of the atelic eventuality. A bounded quantity of the argument, denoted by a quantified NP, matches a bounded reference of the telic eventuality. Jackendoff (1996) and Ramchand (2002) argue that EAH is just another instance of distributive readings. According to them, the sentence in (1b) has two readings, one in which the quantity of five sandwiches represents the affected participant of one instance of the eventuality, and the other in which there are five instances of the eventuality of eating one sandwich. The former is just a regular singular reading of the eventuality, without any correlation between the direct object and the aspect of the eventuality, and telicity comes from event structure. The latter interpretation, however, involves distribution of a telic eventuality over its quantified participant. Both interpretations yield bounded reference for the eventuality, and hence the attested behavior referred to as telicity. The sentence in (1a) cannot form a singular reading, because singularity involves bounded reference, and bare plural is homogeneous and incompatible with bounded readings. There is only the distributive reading, between the eventuality and the bare plural NP in the object position. Due to the homogeneity of the bare plural, the eventuality is atelic. Assuming that the outlined approaches are on the right track, one important question emerges: What is the nature of the distributive interpretations, are they a syntactic, semantic, or pragmatic phenomenon? The aim of this paper is to present a technical account of the phenomenon of distribution between eventualities and their arguments, situated at the interface between syntax and semantics. 1

45

In section 2, I argue that in the distributive readings the quantifier that is lexicalized in NP is in fact base generated and interpreted in VP. This implies that EAH is a consequence of binding between VP and NP. In section 3, I present a parallel between the proposed analysis of EAH and the negative concord, arguing that the lexicalization of VP quantifiers on NPs comes from the same mechanism that lexicalizes negation on the arguments of a negated VP. 2 Quantifying eventualities Consider the distributive reading of (2a), in which there are five instances of the eventuality of the referee showing a red card. What is crucially shown by this example is that in the relevant reading the quantified participant does not have to appear in the respective quantity. There is only one red card – it only participates in five different eventualities. The reason why in (1b) there must be five sandwiches comes from the lexical semantics of the verb and of the object, to which Krifka (1998) refers as the uniqueness relation, and links it especially to verbs of creation, consumption and destruction. (2)

a. The referee showed five red cards. (five eventualities interpretation) b. ∃Eshow. [Eshow⊆E & |Eshow|=5 & ∀e∈E. [e∈Eshow ⇔ ∃x. show(e, John, x) & red_card(x)]].

Back to the moral of (2a), it looks like the participant on which the quantifier is lexicalized is not quantified by it. Only the eventuality is quantified, and quantification over the participant is indirect, through the eventuality. Each instance of the eventuality involves one token of the relevant participant, as formally represented in (2b), but it is not specified whether different eventualities have to involve different participants. This aspect is determined by the lexical semantics of the verb and the argument, and not directly by quantification. To avoid the suspicion that (2a) may be an exception, I give a list of equivalent examples in (3). Each is provided with an additional clause signaling that when the lexical semantics is not too strict, quantification over the participant comes only as an implicature. (3)

a. Seven pedestrians crossed this street today, and three times it was the same guy. b. John saw eighteen movies at that cinema, and he’s not sure if each time he saw a different one. c. This dealer sold over 2000 cars last year. Knowing that it covers a town with less than 1000 cars, the same car was sold, on average, at least twice.

Applying this view to (1a), repeated in (4a), we get the semantic representation (4b). There are five eventualities, each described as John eating a sandwich. That there are five different sandwiches involved is not entailed by the quantification alone, since it directly quantifies only over the eventuality. It comes from the fact that one sandwich may be completely eaten in only one eventuality (Krifka 1998’s uniqueness), which is a product of the lexical meaning of the verb eat (or by the pragmatic knowledge about eating) and the reference to an object of the NP sandwich (and not, for instance, to a sub-kind of sandwiches). (4)

a. b.

John ate five sandwiches. (five eventualities interpretation) ∃Eeat. [Eeat⊆E & |Eeat|=5 & ∀e∈E. [e∈Eeat ⇔ ∃x. eat(e, John, x) & sandwich(x)]].

This provides us with a precise technical account for the relevant distributive reading between eventualities and participants, referred to as EAH. It in fact involves a direct quantification over the eventuality, while the effect of distribution comes from the embedding of existential quantification over some arguments under the quantification over the eventuality. The representation in (4b) suggests that the quantifier is not interpreted where it is lexicalized, which is a problem for an analysis along the lines of (5a) below, in which the quantifier is 2

46

base generated in the NP and then moved higher. It is hard to explain the base generation of a quantifier in a position where it gets no interpretation whatsoever. The correct analysis seems to be to consider the quantifier base generated in the position where it is interpreted: immediately over the VP. Its lexical realization, in a position where the quantifier gets no interpretation, could be a consequence of some kind of grammatical relation, like binding and/or agreement, between the quantifier and an adequate variable in the direct object position, as represented in (5b). (5) Two possible strategies to structurally account for (4a) a. Base generation in the NP and movement b. Base generation in the VP and binding VP

five/NP

VP VP

five

John

John eat

movement

VP

eat

NP five

NP

binding

NP

sandwich

NP sandwich

On the first sight, there appear to be serious problems for an account like (5b) as well. Most prominently, the fact that the quantifier is lexicalized in a position where it does not appear at any stage of the derivation. It is also not obvious what makes certain arguments variables in the relevant sense, in order to be bound by the quantifier of the eventuality, and how this is marked on their NPs (i.e. why is John not bound by the quantifier?). In this section I only discuss the latter question, reserving a whole new section for the one related to the lexicalization of the quantifier. I link the variable behavior with the nonspecificity of NPs representing the participants of the eventuality. While specific, and especially definite, nominal expressions establish reference independently of the predicates in which they appear as arguments, nonspecific NPs are referentially dependent on the predicate in which they are introduced. The specific NP, as well as the definite one, in (6a) has a referent in the context which can be determined by means other than the eventuality of seeing introduced in the sentence. There is at least one other predicate that can serve as the definite description for this referent. However, the referent of the nonspecific NP in (6b) can only be determined based on the seeing eventuality introduced in this sentence. This eventuality provides the unique definite description for this referent, based on which the pronoun can be used to refer to it in the following sentence. In this way, nonspecific NPs are referentially bound by the predicate in which they are introduced. (6)

a. b.

John saw a certain movie/the movie. John saw some movie or other. He said it was interesting, but long.

From this perspective, it is natural that the quantifier in the predicate of an eventuality will bind the referents introduced by its nonspecific arguments, as long as they do not have their own quantification. The result is EAH, the effects of which are visible in tests of inner aspect. It is now clear what kind of participants enters the binding relation with the quantifier at the VP level: nonspecific arguments of the VP that are base generated as non-quantified. Nonspecificity therefore appears as the marking that signals that a certain argument NP may involve a lexicalization of a quantifier as a consequence of being bound from the position in which the quantifier is interpreted. 3

47

3 EAH and the negative concord The other problematic question concerned the mismatch of the base generation of a quantifier in the VP and its lexicalization in an argument of the VP. In this section, I argue that this lexicalization employs the same mechanism as the negative concord, involving a quantifier which binds all its c-commanded arguments that are unspecified for the relevant meaning. Let me first briefly present one of the most typical concord phenomena: the negative concord (NC). NC refers to the phenomenon present in some languages, that when negation appears at the level of VP or higher, one or more of the nonspecific arguments of the VP is marked with a morpheme that is closely related to negation (referred to by Laka 1990 as an n-word), as illustrated in (7) (NC is extensively discussed in, among others, Laka 1990, Zeijlstra 2004). (7)

Ni-ko ni-koga ni-kad ni-gde *(ne) NEG.who NEG.who.ACC NEG.when NEG.where not ‘No one invites anyone anywhere ever.’

zove. invite

S-C

All the NPs involving n-words in (7) are nonspecific: none of them can establish reference independently of the eventuality in which they appear. In addition, they represent participants in an eventuality that appears under a negative existential quantification. We expect, applying the reasoning from the preceding section, that the negated existential quantification over the eventuality also extends to its nonspecific participants. And indeed, (7) entails that there is no person to whom the subject or the object of the eventuality can refer, and there is no place and no time at which such an eventuality takes place. Strictly descriptively, a quantificational predicate, which is in this case a negated existential quantifier, appears on the predicate of the eventuality and triggers the lexicalization of a related morpheme (ni) on each of the c-commanded non-specific elements. I will assume for the ease of exposition, that this happens as an instance of agreement: the relation Agree (Chomsky 2000) is established, as sketched in (8), between the quantifier and all the c-commanded constituents that are not specified for a corresponding predicate. This relation triggers the appearance of the prefixed morpheme ni, a marker of agreement with the negation, on all the arguments entering the relevant agreement relation. (8)

Jovan ni-koga ne zove. S-C Jovan NEG.who.ACC not invite ‘Jovan doesn’t invite anyone.’ ∃Einv. [Einv⊆E & Einv={∅} & ∀e∈E. [e∈Einv⇔∃x. invite(e, Jovan, x) & person(x)]]. VP ¬∃ /ne/

VP Jovan /zove/

NP

binding/AGREE /ni-/

NP /koga/

Semantically, the relation established between the negated existential quantification and the nonspecific participant is that of binding: the quantifier binds an argument that is not quantified in the relevant way and therefore behaves as a variable. Syntactically, the relation is that of Agree, established between a probe and a goal (Chomsky 2000). 4

48

I do not elaborate on the Agree analysis of NC, since it is neither part of, nor crucial for, the aim of the paper (see Zeijlstra 2004 for an elaborate analysis along similar lines). It is essential that NC involves a relation between a c-commanding predicate, and a c-commanded element that is unspecified for the meaning of the predicate. This relation is lexically marked by an element that appears on the c-commanded element, while bearing a meaning related to the c-commanding predicate. However, not all languages show NC. There are two ways to account for this fact. One is to say that the (Agree) relation established in the described situation is parametric: it is present in some languages and absent from the others. The other is to say that this relation is always established, because it is necessary for the semantic effects which are universally present, but that it is parametric whether it will result in any lexical marking. The latter approach is more attractive, not only because it explains the semantic effects which are universal, but also because it allows for a more desirable theoretical explanation. Namely, it places the locus of parametric variation in the lexicon: if a language has agreement morphemes of the relevant kind, and these morphemes can be lexicalized within the category of the bound element – the language will show overt concord. Otherwise, the language lacks the marking of the relevant concord. But there is one more option: in a language that has lexical marking of a certain type of concord, when this marking is overtly realized, the lexicalization of the predicate that triggers concord becomes redundant. Hence, it may happen that it gets elided in languages that more easily drop informationally old or redundant constituents. This is what happens for instance in German: whenever a nonspecific participant is present in a VP, the negation at the VP level is obligatorily elided, and only the n-word on the nonspecific element is lexicalized (9). (9)

Unrecht (*nicht) begangen. Ich habe kein I have NEG wrong not commit ‘I haven’t done anything wrong.’

German

Now, the crucial parallel emerges. The same analysis can be applied to the lexicalization of the quantifier over the eventuality which binds a nonspecific participant. It appears that the agreement marking corresponding to the quantificational predicate is always realized on the nonspecific participant, and the quantifier itself is never overt in its base generated position. (10)

John ate five sandwiches. (for the five eventualities reading) VP

VP John

NP

binding/AGREE /five/

NP

While the negation may or may not be lexicalized in its base generated position in VP, especially when there are no nonspecific participants in the eventuality, this is not the case for quantification, or at least not in most of the well described languages. This means that if an eventuality involves no nonspecific participants – it is impossible to directly lexicalize the

5

49

quantification over the eventuality.1 In terms of the present analysis, it appears that most languages lack any lexical material that realize quantification and categorially combine with VP. Although this sounds a little odd, it is empirically confirmed. In spite of the fact that, strictly formally speaking, eventualities should be equally quantifiable as the nominal expressions, and that they, at the very least, do undergo existential quantification, most well described languages have no quantifiers that are conservative (in the sense of Keenan & Stavi 1986) with respect to sets of eventualities represented by VPs. Some languages, however, do provide exceptions from this tendency. Such are languages with event classifiers (also called verb classifiers), e.g. Mandarin (11). Classifiers like chang are reserved for eventualities realized by VPs, and when a numeral like liang ‘two’ combines with such a classifier, it will quantify over the eventuality that the classifier relates to. (11)

Ta qu-nian bing-le 3S last-year sick-PERF ‘He got sick two times last year.’

liang chang. two CL

Mandarin (Leo Wong, p.c.)

The second problem for the proposed analysis is now solved. The quantifier that is lexicalized on the nonspecific argument in the observed kind of reading is an instance of agreement, or concord, with the position in which the corresponding predicate is base generated. This higher position is in most languages never lexicalized, because their lexicons lack adequate units. EAH in this view dissolves in a binding of arguments of an eventuality by the components of the predicate of the eventuality for which the arguments are unspecified. 4 Conclusion I argued that EAH is underlyingly a phenomenon very similar to NC. I observed first that the so-called distributive readings between quantified arguments and eventualities appear only with nonspecific arguments, and that the quantifier really directly predicates only over the eventuality. This was explained by base-generating the quantifier in the VP and not in the NP where it is lexicalized. This lexicalization is further argued to be an instance of agreement, parallel to n-words in NC. In languages where the quantifier is never lexicalized in the VP, this is because the lexicon of the language has no units that realize quantificational predicates and may appear in the VP. References Keenan, E. & Stavi, J.: 1986, A semantic characterization of natural language determiners, Linguistics and Philosophy 9(3), pp. 253-326. Chomsky, N.: 2000, Minimalist inquiries: the framework, in R. Martin et al. (eds), Step by step: Essays on minimalist syntax in honour of Howard Lasnik. Cambridge: MIT press. Jackendoff, R.: 1996. The proper treatment of measuring out, telicity, and perhaps even quantification in English, Natural language and linguistic theory 14, pp. 305-354. Krifka, M.: 1998. The origins of telicity, in S. Rothstein (ed), Events and grammar, Dordrecht: Kluwer, pp. 197-235. Laka, I.: 1990. Negation in Syntax: On the Nature of Functional Categories and Projections, Ph.D. dissertation, MIT. Ramchand, G.: 2002. First phase syntax, Ms. Oxford University. Verkuyl, H.: 1972. On the compositional nature of the aspects. Dordrecht: Reidel. Zeijlstra, H.: 2004. Sentential negation and negative concord. PhD Dissertation, University of Amsterdam. 1

Quantification over the reference time (always, twice) is the most common strategy to make up for this.

6

50

Cognitive Explanation of Case Shift and Verb Concord in Georgian Rusudan Asatiani (Institute of Oriental Studies, Georgia) [email protected]

Abstract The Georgian Language has two types of verbal person affixes: the Vtype and the M-type. Traditionally they are qualified, consequently, as Subject and Object markers, but there are some “exceptions”: Sometimes S-markers (res. V-type prefixes) represent Object and, vice versa, O-markers (res. M-type prefixes) represent Subject. Based on a formal, functional and semantic analysis of the main models of case shift and verb concord in Georgian new interpretation of the verbal prefixes is suggested: It seems more adequate to analyze the verb markers and the cases without any functional qualification on the base of the semantic feature: “free will of arguments”. According to the feature some rules are generalized. The rules are hierarchically organized and reflect the process of linguistic structuring of situation. We suppose that such description and interpretation of the Georgian data mirror the cognitive background of grammatical structures. The presented in the paper attempt is not only descriptive and has more explanatory content.

I. Introduction The Georgian language is one of the Caucasian languages with the oldest literary traditions. The Georgian script was devised around 400 AD in order to facilitate the dissemination of Christian literature. It should therefore offer us the unique opportunity to study a history of 1.500 years of polysynthesis. The ‘Georgian’ population from the last Soviet census of 1989 is 3.787.393. Georgian and its related languages: Laz, Magrelian, and Svan are known as the South Caucasian or the Kartvelian languages. Georgian (as far as Laz, Megralian or Svan) personal verb-marking mirrors the system of the Kartvelian protolanguages and mostly has been stable over this time.

II. The Data of the Georgian Language The Georgian verb is polypersonal: the verb form shows the Subject as well as the Object: m - c’er - s (‘S/he writes me’) DO.1-write-S.3 The developed case system of Georgian distinguishes the arguments character and is corresponding to the verbal person markers. Semantically or functionally different arguments are represented by various case patterns and models of verb concord. 51

The patterns and models are different according to the tense forms. Formal analysis of the Georgian case patterns and models of verb concord gives us the following information: 1. The I/II personal pronouns me (I), shen (you-sing.), chven (we), tkven (you-pl) never add case markers, they are always unmarked. 2. All other nouns show three main case patterns: (a) NOM-(DAT) – I series forms: present, imperfect, present subjunctive, future, conditional, future subjunctive k’ac-i k’ac-i surat-s k’ac-i

cxovrobs xat’avs c’evs

(‘A man-nom. lives’) (‘A man-nom. paints a picture-dat.’) (‘A man-nom. lies’)

(b) ERG-(NOM) – II series forms: aorist, aorist subjunctive k’ac-ma k’ac-ma surat-i k’ac-i

icxovra daxat’a dac’va

(‘A man-erg. lived’) (‘A man-erg. painted a picture-nom.’) (‘A man-nom. lay’)

(c) DAT-(NOM) – III series forms: perfect, pluperfect, perfect subjunctive k’ac-s ucxovria k’ac-s surat-i dauxat’avs k’ac-i dac’olila

(‘A man-dat. has lived’) (‘A man-dat. has painted a picture-nom’) (‘A man-nom. has lain’)

This system prove the existence of the I/II - III formal dichotomy: the I/II person subsystem does not distinguish the roles of the nouns, since Ag and P (and Ad(dressee)) can all be unmarked. In the III person subsystem in the I series, S is always in nominative: -i (after consonants)// zero (after vowels), and O in Dative (-s). It is thus a nominative system, while II and III series are ergative systems: S.3 of the transitive (and active-intransitive) verbs in II series appears with a special case represented by the suffix -ma (after consonants)// -m (after vowels); in III series with the dative ending -s. All other nouns (S-intransitive and Otransitive) are in nominative. The nouns trigger definite models of verb concord. The Georgian Language has two types of verbal person affixes, the V-type and the M-type:

I II III

V-type sing. v-s,-a,-o

M-type sing. mgh-,s-,øø-

pl. v-t -t -n,-en,-an, -nen,-es

pl. gvg-t h-,s-,ø- (-t) ø-

Traditionally the V-type affixes are considered to be subject markers, while the M-type are object markers. Note, however, that this is not always the case: In the perfective-resultative tense forms and also with affective verbs the subject appears with the M-type and the object with the V-type: m-shia (‘I am hungry’), m-eshinia (‘I am afraid’), m-civa (I am cold’), medzineba (‘I want to sleep’), m-inda ( ‘I want’) and others. For that reason most Georgian scholars qualify these forms as inversive ones. Such ‘exceptional’ cases lead us to reanalyse 52

the data in order to found some semantic or functional feature according to which the description of the forms would be more adequate – in one to one correspondence. In general, it seems better to analyze these markers without any functional qualification, simply by their relation to cases: 1. Noun in the Dative always triggers the M-type affixes; 2. Noun in the Ergative always triggers the V-type affixes; 3. Noun in the Nominative triggers either (a) V-type (if there is no ergative linked with the verb), or (b) M-type (in case there is an ergative linked with the verb), or (c) Zero (if both ergative and dative appear in the construction). Based on the rules above, we can make the following generalizations: 1. The V-type affixes should be represented in the verb form in any case; 2. From the point of the triggering V-type affixes, there are the following hierarchical relations: ergative (always)>nominative (sometimes)>dative (never); 3. In case of several prefixes, a competitive situation arises, as there is no possibility of prefix clustering. When there is a competition among prefixes (both I/II persons are linked with the verb), the hierarchically weaker case wins: 1. I/II nominative and I/II dative ! I/II dative 2. I/II ergative and I/II dative ! I/II dative 3. I/II ergative and I/II nominative ! I/II nominative In case there is no competitive situation (this happens mostly when I/II meets III), both markers will be present and polypersonal verb forms arise.

III. Semantic and Functional Interpretation of the Data Semantically the Dative case and, consequently, the M-type affixes represent: Addressee (res. the indirect object), Experiencer (res. the subject of the affective verbs), Agent (res. the subject in the Perfect tense forms) and Patient (res. the direct object in the Present tense forms). The Ergative case and the V-type affixes show: Agent (res. subject of transitive and intransitive active verbs in the Aorist tense forms). The Nominative case represents the most complex situation –nominative together with the V-type affixes shows: Agent (res. the subject of intransitive verbs in any tense forms and the subject of transitive verbs in the Present tense forms), Patient (res. direct object of the transitive verbs in the Perfect tense forms); and nominative which triggers the M-type affixes corresponds to the Patient (res. direct object in the Aorist tense forms). Such complex and polyfunctional interpretations make clear that the syntactic functions or semantic roles are not sufficient for the simple analysis.

IV. Cognitive Explanation The cognitive explanation of the affective verbs gives an argument for the main semantic feature which is important for the distinguishing the prototypical subject from the subject of these verbs.

53

The constructions of affective verbs differ from the canonical ones of the same languages. As a rule, they are characteristic for the predicates with the specific semantics: (a) Possession/Existence; (b) Psychological state; (c) Physiological state; (d) Visual/auditory perception; (e) Modal state (may/must). The main and common feature for the subjects of these predicates (res. affective verbs) is that they do not act according to their “free will” and do not “control” their own action – feelings, emotions, perceptions. So, the S of affective verbs is far from the prototypical subjects, which control their actions and act according to their free will. In the majority of languages such deviation from the prototype is represented by the marked, non-canonical linguistic structures. In these structures S instead of the canonical form (nominative – for Nom./Acc. languages, or ergative – for Erg./Abs. languages) stands in marked, Dative (or any other oblique – Gen., Acc., Inst.) case. In Georgian alongside with this generalization affective verbs built non-canonical constructions: According to the universal tendency S stands in dative case and triggers the Mtype person markers in the verb form: me I.dat

m - i q’var - s 1.dat-love-3.nom

deda mother.nom

(‘I love the mother’)

Similar ‘inversive’ constructions arise in Perfect or Pluperfect forms. Georgian perfect demonstrates the additional semantic nuances: ‘apparently’, ‘it seems’, ‘probably’, ‘it turns out’. The Perfect represent the following aspectual situation: The speaker sees the result of the action, s/he does not pay any attention to Ag (or s/he is not sure; or s/he does not actually know; or s/he merely forgets, who was the Agent of the action), but because of the actually presented result (Patient), s/he says what ‘apparently’ happened; e.g. dauxat’avs (‘It seems that s/he has drown’), ucxovria (‘Apparently s/he has lived’), dac’olila (‘It turns out that s/he has lain’) and so on. Thus, Georgian perfect represents the situation when it is not known the Ag (res. Subject) acted according to its ‘free will’, or without of its ‘‘free will’.

V. Explanatory Model of Case shift and Verb Concord in Georgian It seems more adequate to analyze the verb markers and the cases without any functional qualification on the base of the above mentioned semantic feature: “free will of arguments”, which has double representation: the definite cases of the argument and definite models of verb concord. The Ergative case marks the argument whose free will is included in action; The Dative case marks the argument whose free will is not included in action or its will is not relevant for the situation; the Nominative case as the unmarked one shows the cognitive unmarkedness of the feature “free will” for the arguments. This feature plays an important role in person alignment and defines appearance of either the M or the V-type of markers: I.

II. III.

The argument whose free will is not included in the situation (or it is unknown whether its free will is included or not (this is the case when the verb is in the Perfect - the Georgian Perfect has an additional meaning of subject’s unknowness) triggers the M-type affixes. (Semantically such are: Addressee, Experiencer, an actually unknown Ag of perfective tense forms); The argument that acts according to its free will triggers the V-type affixes (such is Ag); The argument whose free will is not relevant for the situation (such is P), triggers 54

a. The V-type, if it is only argument linked with the verb (P); b. The V-type, if other argument’s free will is not included in the situation (There are different possibilities of roles combinations: P-Ad, or P-Exp, or P-unknAg); c. The M-type, if other argument’s free will is included in situation (It is the case: PAg); d. Zero, if all other (both) arguments (with +[fw] and with –[fw]) are linked with the verb. These rules are hierarchically organized: I>II>IIIa>IIIb>IIIc>IIId. As a result various morphological verb forms arise. E.g.: The form (‘I love s/he’) is a result of the following derivational processes: The Exp. ( I ) is the semantic role which does not control its feelings raised by the Stimulus ( s/he) and it acts without of its ‘free will’. Thus, according to I-rule it is marked in the verb form by the Mtype marker m- . The next argument can be qualified as P (the argument, whose ‘free will’ is not decisive) and, consequently, according to IIIb-rule it is represented by the V-type marker -s. Thus, the form m-i-q’var-s is constructed by the rules I>III.b: First of all M-type marker appears (according to I-rule) and, after, V-type marker appears -s (according to rule III.b). Forms – v-xat’av (‘I draw it’) or m-xatav (‘You draws me’) – are derived by the rules: II>IIIc; and so on: m - civ - a (‘I am cold’) – I>II I(Exp)-cold-it(St) m - xat’av - s ((S)he paints me’) – I>IIIc I(P) - paint-(S)he(Ag) m- i - qvar - s (‘I love her/him’) – I>IIIb I(Exp)-love-her/him (P) v - xat’av (‘I paint him’) – I>IIIc I(Ag)-paint-her/him(P) v-0-u-xat’av (‘I paint it to her/him’) – I>II>IIId I(Ag)-him/her(Ad)-paint cxovrob-s (‘S/he lives’) live-s/he(Ag) – II ixat’eb-a (‘It is drawn’) IIIa paint.Pass-it(P) The rules are free from tense (I-II-III series) or person (I/II – III) restrictions – there is no kind of split. They reflect the process of linguistic structuring of situation. The cognitive background of this process is simple: Arguments are distinguished according to their ‘free will’. This feature is decisive for the whole system of cases and verb concord. We suppose that such description and interpretation of the Georgian data has more explanatory content.

References Aronson, H. B. 1970. Towards a Semantic Analysis of Case and Subject in Georgian. Lingua, (25): 91-301.

55

Boeder, W. 1999. A Slot-filing Constraint in the Georgian Verb and its Syntactic Corollary. In Ergative Syntax and Language Change: The South Caucasian Languages. Sprachtypol. Univ.Berlin(52), 3/4:435-480. Breithweithe, K. 1973. Case Shift and Verb Concord in Georgian. PHD thesis, Department of Linguistics. The University of Texas at Austin. Bybee J.L. 1985. Morphology: A study of the Relation between Meaning and Form. Amsterdam, Philadelphia: Benjamins. Chikobava, A. 1968. mart’ivi c’inadadebis p’roblema kartulshi (The Problem of Simple Sentence in Georgian). Tbilisi:Mecniereba. Croft, W. 1991. Syntactic Categories and Grammatical Relations. The Cognitive organization of information. Chicago: University of Chicago Press. Croft, W. 2002. The non-existence of Syntactic Relations. Third Winter Typological School.Moscow: Russian State Humanitarian University Press. Givon, T. 1979.On Understanding Grammar. New York:Acad.Press. Harris, A. 1984. Inversion as a rule of Universal Grammar: Georgian Evidence. Perlmutter and Rosen (eds.). Kibrik, A. 1997. Beyond Subject and Object: Towards a Comprehensive Relational Typology. Linguistic Typology, I. Berlin-New York: Mouton de Gruyter. Shanidze, A. 1973. kartuli gramat’ik’is sapudzvlebi (The Foundations of the Georgian Landuage). Tbilisi: Mecniereba. Shibatani, M. 2000. Non-canonical Constructions: A Cognitive-typological Perspective. Second Winter Typological School. Moscow: Russian State Humanitarian University Press.

56