the syntax-focus structure interface

THE SYNTAX-FOCUS STRUCTURE INTERFACE NOMI ERTESCHIK-SH/R Department of Foreign Literatures and Linguistics Ben-Gurion University of the Negev Beer Sheva, Israel

Focus structure (f-structure) theory is a model of informational structure. It provides a natural account of a wide range of linguistic phenomena, usually accounted for by syntactic constraints and rules of logical form (LF), including quantifier scope, interpretations of wh-questions, anaphora, and extraction. A model of grammar necessarily includes f-structure since it determines intonation. F-structure theory allows for a large part of syntax, including LF syntax, to be handled by f-structure principles. The model I propose is therefore maximally simple in that it enables interpretation off-structures directly without the mediation of LF. Constraints on f-structure take two forms: those that are only loosely determined by syntax rendering marked and unmarked sequences (quantifier scope), and others which pertain to a well-defined class of dependencies which require isomorphism between f-structure and syntactic structure. This chapter aims to illustrate these claims with a selection of such dependencies (wh-movement, negation). 1

1. INTRODUCTION TO FOCUS STRUCTURE F-structure is an annotated structural description in which Topic and Focus constituents are marked. F-structure feeds both phonetic form (PF) and semantics and is sensitive to lexical information. It feeds PF since this level provides the explicit Syntax and Semantics, Volume 29 The Limits (f Syntax

211

Copyright © 1998 by Academic Press All rights of reproduction in any form reserved. 0092-4563/98/$25.00

IV. UPDATE instructs the hearer to enter the predicate on the topic card and then to copy all entries to all cards activated by the focus rule. F-structure theory is a pragmatic theory which is concerned with felicity conditions on the relation between sentences and context. Thus TOPIC can be assigned only to constituents for which file cards are available on top of the file (i.e., cards which have been positioned there by the application of the f-structure rules to previous utterances). F-structure affects truth conditions. Rooth (1985) among others has shown that Focus assignment may determine truth conditions. Chierchia ( 1992) shows that Topic is what forms the restrictor on adverbs of quantification. Partee (1992) discusses the idea that the restriction of tripartite discourse representations is akin to Topic and that nuclear scope is the Focus. Following Reinhart (1981), I adopt the Strawsonian view that the topic is the pivot for assessment and show that f-structures involving both main and subordinate assignments of topic and focus are required for interpretation. In particular, I show in Erteschik-Shir (in press) that quantifier scope is determined by f-structure and that it makes unnecessary the need for an LF component. F-structure theory thus feeds semantics. It is a property of dynamic semantic theories that the borderline between pragmatics (involving circumstance of use) and semantics is blurred. F-structure theory blurs this line further: The constraints on f-structure (f-structure rules I and II) are pragmatic (i.e., context determines whether a particular f-structure can be assigned to a sentence). Yet, f-structure determines interpretation Cf-structure rule III) and therefore the view of grammar proposed here requires a semantic theory which takes f-structure, rather than LF, as its input. Since I also claim that certain syntactic constraints must be defined on f-structure, the model of grammar proposed here is distinct from most other proposed models in that it incorporates pragmatics (in the form off-structure) .

1.1. Illustration Rules I and II apply to the constituents marked TOP and FOC, respectively. Rules III and IV are interpretive rules which apply to all f-structures to which TOP and FOC have been freely assigned. The latter two rules are informally sketched here and clearly need semantic formalization. Assume the following interaction: A is speaking, B is listening. The cards for the speakers are available on top of the file: "I" is licensed as the topic of (2): A says: (2)

I [have a dog]. [It] is brown. FOC

TOP

In that case the f-structure of (2) would have only the object in focus as in: (4)

lroP have [a dog]poc

Here the focus and the predicate are not coextensive as in the previous case. The questions themselves trigger a manipulation of the file. They generate cards with variables in them. The card generated by the question in (3)b. roughly looks as follows: 3

Presuppositions are assumed to be existing entries on cards. It follows that the entry on the card is presupposed by the answer. 4 The answer simply replaces the variable with the entry a dog, resulting in a card identical to the one in card #1. This analysis accords with the traditional view that the constituent which answers a wh-question is the focus. If the object were a potential topic, the following f-structure could be assigned: (5)

TOPi [John has [the dogli] Foe

Here the topic is the object. Again, a card with the heading the dog must be available on the top of the file. And if the sentence is assessed as true with respect to the topic the dog, the new entry will be added:

The f-structure in (5) has an initial TOP followed by a focus constituent. The f-structures associated with syntactic topicalization structures are isomorphic to the parallel f-structures without topicalization. This f-structure thus exhibits a notational reflection of syntactic topicalization or can alternatively be viewed as an f-structure theoretical variant ofLF topic movement. 5 The following f-structure is ruled out by the premise that a focus must constitute a syntactic constituent: (6)

[John has]Foc [the doghoP

Specific indefinites can, however, be topics: (9) a. A friend of mine is intelligent. b. A certain student likes linguistics. c. A student who I know likes linguistics. Since topics must be either generic or specific (i.e., a card with the relevant heading must exist), it is not surprising that specific indefinites are allowed as topics. It is, however, rather mysterious what it is that makes these indefinites specific. In each case the indefinite is modified in some way. It is often recognized that specific indefinites are specific only to the speaker and that the hearer is not assumed to have an available referent in these cases. It follows from f-structure theory that complex constituents must be assigned an f-structure to get interpreted. F-structures assigned to subconstituents of the utterance are called subordinate f-structures and do not involve predication (assessment) since only full sentences are assessed. The subordinate f-structure assigned, however, is interpreted by subordinate update, the process that associates subordinate topics with their foci. Subordinate update involves the application of the Topic, Focus, and Update rules to a subordinate f-structure and differs from main f-structures only in that the rule of predication does not apply. Its role is to rearrange the cards according to the discourse rules and to make the relevant entries. The file manipulation for (9c ), for example, is equivalent to that of the following sequence of sentences: (10)

hoP [know a student]poc· HeToP [likes linguistics]poc·

In the first sentence of this sequence, 'a student' is contained in the focus and hence a new card is introduced (by focus rule I) allowing for the introduction on top of the file of a new indexed card for "a student":

Once this card is opened, the indefinite is (speaker) referential and qualifies for topic status. The fact that the relative clause introduces a subordinate application of the rules is also what explains the speaker-perspective of the reference. When a speaker introduces a new referent by means of a relative clause she or he instructs the hearer to open a new card, position it on the top of the file, and then go on to the sentence (without the relative clause). Only the speaker, not the hearer, is assumed to already have a card for student2 explaining speaker-specificity. Subordinate update must occur prior to the file manipulation triggered by the main

1.4. Restrictive Sets The focus constituent may define a set of cards, the students in my class, for example, forms such a set. Cards which contain a set of cards are called restrictive cards, the set they define is a restrictive set, and the focus which introduces them is a restrictive focus. A subordinate f-structure can be assigned to such a restrictive set in a partitive: (12)

[[Two]pocsub of [the students]rnP-subhoP [are intelligent]poc

The subordinate f-structure is formed around a card on top of the file which represents a discoursally available set. A constituent which defines a subset of this topic set is focused triggering the partitioning of this set. The new subset card is now available as the main topic. In sum, Topic and Focus are defined within a theory of discourse. Both individuals and stages provide topics. A rule of Predication is defined which assigns truth values. This rule is viewed as a relation between the topic of a sentence and its predicate. The truth value of a statement is determined by assessing it as putative information about its topic. The f-structure Topic and Focus rules apply to subordinate f-structures as well, but the rule of Predication applies only to the main f-structure. Since the topic is the pivot for the assignment of truth value, it follows that every sentence must have a topic.

2. R-DEPENDENCIES A distinction is made here between R-dependencies (relevant to the speaker's file) and I-dependencies (relevant to the hearer's file). R-dependencies (certain scopal relations, for example) are subject to the Topic Constraint, an f-structure theoretical constraint which determines a markedness hierarchy on f-structure. I-dependencies (e.g., anaphora, wh-movement) are sensitive to the Subject Constraint, a syntactic constraint on f-structure. In this section I briefly discuss R-dependencies, which I distinguish from I-dependencies to be discussed in more detail in section 3. I use the term R-dependence when reference is defined by f-structure update, as in the sentences in (13): (13) a. b. c. d.

John talked to a boy. He criticized her. Two girls arrested three boys. Everyone talked to someone.

In (13a) the indefinite object in the focus triggers the construction of a new card. The content of this card is determined solely by the entry made by update. All we

......... .., ..... J .............

......... - ......................... _,.. ___ .......... _____ _

which the object is interpreted as a topic, the unmarked interpretation is still the one in which the subject is taken to be the topic. This can be seen in the following interchange which fixes John as the topic: (15)

Tell me about John: a. He is in love with Mary. b. ??Mary is in love with him.

F-structures with stage topics are also unmarked. (16a), in which the object is interpreted as the topic, is less natural than (16b) with the stage topic: (16) a. TOPi [A girl talked to JohnJpoc (Answers: Tell me about John) b. sTOPt [A girl talked to John]poc (Answers: What happened next?) The Topic Constraint, which applies to R-dependencies, thus provides an explanation for the well-known fact that certain scopal interpretations are much harder to get than others. (13c), for example, illustrates an example with two weak quantifiers. Weak quantifiers can be interpreted as topics only if they are interpreted partitively. Again, the unmarked case will be the one in which the subject is taken as the topic, but if a set of boys is available in the context and a set of girls is not, the object and not the subject will function as the main topic. Since the topic has wide scope, context determines scopal relations in sentences. 9 Similarly for (13d). 10 The topic constraint is based on the intuition that processing is facilitated when the topic ('what we are talking about') precedes the predicate ('what we say about the topic'). It can be verified only empirically. The topic constraint is here viewed as a pragmatic constraint on f-structures.

3. I-DEPENDENCIES A distinction is made between R-dependencies and I-dependencies. I-dependencies are illustrated in the sentences in (17): (17) a. b. c. d. e. f. g.

John talked to himself. (anaphora) Everyone loves his mother. (bound anaphora) Who read what? (multiple wh-questions) Who did everyone talk to? (quantifier wh-interactions) Who did you talk tot? (wh- trace dependency) John didn't eat anything. (negation and focus of negation) John is the teacher. (copular sentences)

The dependent in an I-dependency characteristically does not trigger the construction of a new card, instead its 'identity' is fixed by the dependency. In the ex-

.L.1..1. ..... LIJ.1..1. ....... .llo. -.._ V ..... U..,, U' ....I. U ........ U.I. ..... .L.1..1. ....... .1..1.U""

The Subject Constraint also accounts for the restriction on f-structures for negated sentences as well as the licensing of negative polarity items. Finally, the Subject Constraint restricts reconstruction in cases of wh-movement accounting for constraints on wh-movement (island constraints), Crossover, multiple wh-questions, and wh-quantifier interactions. The Subject Constraint therefore plays a central role in the grammar of English, rendering a new view of the division of labor among the syntactic modules. Picture NPs illustrate the need for a unified account of binding, negative polarity, and extraction: (20) a. John saw pictures of himself. b. ?John saw the picture of himself. c. *John saw Susan's picture of himself. (21) a. John didn't see pictures of anyone. b. *John didn't see the picture of anyone. c. *John didn't see Susan's pictures of anyone. (22) a. Who did John see pictures of? b. *Who did John see the picture of? c. *Who did John see Susan's picture of? The unacceptable sentences in all three cases can be much improved if supplied with logophoric, contrastive, and restrictive readings. The following examples are quite good: 11 (23)

John said that he saw Susan's picture of himself.

(24)

John didn't see SUSAN's picture's of anyone. Those were HENRIETTA's pictures.

(25)

John was looking for Susan's pictures of Erhard, Marie, or Toby, but he couldn't find her pictures of anyone.

(26) ?If Susan doesn't have a picture of Toby, who did John see her picture of? (24) is logophoric. I argue in Erteschik-Shir (in press) that logophoric reflexives are coreferent rather than I-dependent. (24) shows that contrast changes acceptability. The account I offer below applies to noncontrastive focus structures. In Erteschik-Shir and Lappin (1983) and Erteschik-Shir (in press), I argue that contrastive or metalinguistic foci have quite different properties than other foci. Here I carefully exclude contrastive data from discussion. (25) exemplifies a restrictive context (i.e., one in which a contextually restrictive topic set is supplied). This again renders a different f-structure (anyone here ranges over this set and a subordinate f-structure is therefore added [see section 1.4]). In (26) the context supplies a presupposition to the effect that a picture was seen by John but that it couldn't have been a picture of Toby. The question then is which one of Susan's

4. WR-MOVEMENT It is well known that the predication rule, when applied to wh-questions, cannot involve assessment for truth values. Instead, it can be viewed as an instruction to fill in those individuals for which the predication would come out true. In other words, predication 'interprets' wh-questions as asking for the values of some variable x, such that those values will make for a true assertion. (28b) formalizes the application of the predication rule to the wh-question in (28a):

(28) a. What did Sue eat t? b. (?x) {xlPx (Sue1)} Let us briefly examine how the formalization in (28b) relates to the filing system model: (28b) "asks" for the values of x, such that P (of x) (of Sue) is true, where P x indicates that the predicate is parameterized by x. (25) assumes a context in which Sue is the topic of (28a). Px indicates that the predicate has a variable in it. The state of speaker's file as he asks (28a) is that on top of his file is the following card:

i

fl !l Ii Ii 111 IIii I! 11iI!11 lIlIll lllilli:

~~

The hearer is asked to fill in the value of x. The variable x is 'located' by reconstruction to wh-t position. 13 There exists an alternative interpretation of (28a) in which the wh-word is taken to be the topic off-structure: (29)

(?x) {x IP (x)}

(29) asks for the values of x, such that P of x is true. The topic is identified with the wh-phrase (i.e., x must range over a restrictive set). Under this interpretation the top of the speaker's card file has a set of cards, each headed with a different kind of food. The following card illustrates one such card:

The disjunctive entry on the card gives the two options in question: for the particular food in question, did or didn't Sue eat it. The question requires the hearer to choose one option for each food in the set. The answer will allow the speaker to eliminate one member of the disjunctive set on each card included in the set.

I-dependency. Note further that this type of I-dependency has the characteristic property associated with I-dependencies: the dependent does not have an existence of its own (i.e., it does not license the construction of separate card). If reconstruction to trace position is viewed as forming an I-dependency between the wh-phrase and its trace, the Subject Constraint will apply. 15 The Subject Constraint restricts wh-trace positions to positions contained in the main focus of the sentence in a canonical f-structure with either a subject topic or a stage topic. In the following sections I show that extraction islands can now be explained without adding any further machinery. Since Erteschik-Shir (1973), I have argued that islands are environments which cannot provide the main focus of the sentence. 16 Here, I show that this follows from the Subject Constraint which applies to an array of phenomena which I unify as belonging to the class of I-dependencies. A Subject Constraint violation causes a strong grammaticality infraction usually associated with ECP effects. Since the Subject Constraint covers many of the subject-object asymmetries considered to be the province of the ECP, this is not surprising. The idea that the nature of the extraction matters has also been around for a while. Ross (1971) suggests a hierarchy of extraction rules. In this hierarchy, topicalization is a "stronger" rule than question formation. In Erteschik-Shir (1973: 60) I demonstrate that wh-questions which presuppose a "list" license extraction where it would otherwise be blocked. More recently Cinque (1989, 1990) argues that D-linked or referential wh-phrases can enter a binding relation with their trace explaining their insensitivity to what he refers to as weak islands (negative and wh-islands). He adds that the relative wh-phrase is more easily interpreted as referential than the interrogative one. Although I believe referentiality is the wrong notion, Cinque's basic intuition prompted my understanding that these distinctions follow directly from the theory off-structure (i.e., that the fronted constituent must be interpretable as the main topic in order for extraction out of weak islands to occur felicitously). A wh-moved constituent is interpretable as a topic in a number of different circumstances: First, if it is topicalized, it functions as the main topic of the sentence. Second, an NP modified by a relative clause is necessarily specific or generic, qualifying it for topic status. Third, in D-linked wh-questions the fronted wh-phrase ranges over a topic set. Finally, metalinguistic focusing (as in clefting) again forms an f-structure in which the fronted constituent is the topic. In order for the fronted constituent to be interpretable as the main topic, it does not suffice for this constituent to qualify as a proper topic. It is also necessary that the rest of the sentence qualify as a predicate for this topic. Whether this is possible or not depends on context, the focus properties of the lexical items chosen, as well as the structure involved. It follows that when extraction is licensed in this way in sentences, they will vary in acceptability: The 'better' the topic extracted,

...... _

~.;;----·-

....

~--~

~-----

.....

--

.... ,.

. _ .... _.

....__._

c. Det har jeg set mange der har gjort. that have I seen many who have done e 'I have seen many (people) who have done that' In Erteschik-Shir (1982) I claimed that the property that distinguishes these relative clause constructions from the ones that do not allow extraction is that the matrix must serve merely to introduce the head of the relative clause into the discourse. An existential matrix does so by introducing an NP onto a stage. The other predicates which license extraction are predicates which can be used to select stages. The following f-structure is therefore available here: (36)

leg har set sTOP1 [mange der har gjort det]poc

The stage topic here takes the relative clause as its focus, making it the main focus of the sentence. Extraction is therefore possible since the following f-structure is well formed: (37)

Detrnp [har jeg set sTOP1_sub [mange der har gjort e]poc-sublFoc

In Erteschik-Shir (in press) I argue that implicit subordinate stage topics do not "interfere," (i.e., no gapped entry is made on the stage topic card and hence the f-structure is interpretable by the f-structure rules). Importantly, the Subject Constraint cannot be violated in Danish. The following examples show that only a "topic" can be extracted: 18 (38) a. *Hvad er der mange der kan lit?

What have you seen many that like t b. Hvad slags is er der mange der kan lit? What kind of ice cream are there many who like t

4.3. Negation and Wh-Islands Negation forms an I-dependency with a dependent focus. Reconstruction to trace position forms an I-dependency as well: (39)

wh-phrase [subject NEG ... t ... ]

I

L--J

Neither of the I-dependencies violates the Subject Constraint. Why is this structure then ruled out? I-dependencies enable the interpretation of the dependent element. Therefore, a constituent cannot be interpreted if it is a dependent in more than one I-dependency. This is the f-structure theoretical equivalent of the bijection principle. I formulate this principle as Principle I: A constituent cannot be a dependent in more than one I-dependency. I assume that I-dependencies are marked on f-structure. In view of the fact that a doubly identified constituent will result in an uninterpretable f-structure, it is

.L&&'-' UJ&&""A-.L"V'-'UIJI U ...& U'-'"U&'-' .L&& ... '-'&.&A'-''-'

Extraction again forms a double I-dependency: (44)

wh-phrase [subject V [wh-phrase [subject ... t ... ]

I

I

I

The same data as with negation is therefore predicted for multiple wh-questions and extraction out of embedded questions. There is one difference, however. The causal adjunct modifies the sentence with which it adjoins. Hence, it can only be inter:preted as modifying the matrix. Its inter:pretation inside the subordinate question requires reconstruction and a double I-dependency is again formed. Since adjuncts which are not possible stages cannot function as topics, such a sentence will have no possible derivation: (45)

Why /For what reason did you ask who went?

Here the only inter:pretation is the one in which the reason for asking is requested, not the reason for going. 20 It is pertinent here to consider the approach to weak islands in Szabolcsi and Zwarts (1993). They argue that although ct-linking may play a role in accounting for the type of wh-phrase which is immune to the weak island effect, the overriding factor is that they range over individuals. A strong argument for their view is that 'how many' extraction is possible in Romanian on an individual, but nond-linked reading. Since d-linking is marked by clitic doubling in Romanian, the data are clear. I cannot address this argument without a detailed study of clitic doubling in Romanian, Romanian f-structure properties, and in particular an understanding of what the canonical f-structure is in Romanian. This type of question must be answered in order to test f-structure theory in any language. It is, however, possible that ct-linking (which is responsible for clitic doubling in Romanian) and the notion of restrictive topic sets are not equivalent notions. In particular, I believe that restrictive sets necessarily range over individuals, and therefore the two approaches may be reconcilable. 21 I conclude that extraction out of negation and wh-islands is therefore governed by f-structure. The fronted phrase must be inter:pretable as a topic. Reconstruction in both cases results in a double I-dependency, which cannot be inter:preted.

4.4. That-Clauses That-complements introduced by "bridge" verbs allow the following three f-structures, depending on context and stress. 22 (46) a. IrnP [believe (that) Mary saw Susan]Foc b. I believe (that) MaryTOP [saw Susan]Foc c. TOPi [I believe [(that) Mary saw Susan];]poc

Manner-of-speaking verbs exemplify another class of nonbridge verbs: (49) a. ??Who/??Which girl did you mumble that you'd seen? b. *Who/*Which girl did you lisp that you'd seen? In Erteschik-Shir (1973) I argued that the predicates themselves are necessarily focused forcing the complement to be backgrounded (i.e., not part of the main f-structure): (50)

PeterTOP lispedpoc [that you'd seen Mary] 24

Since the complement clause is not part of the main focus extraction from within it will violate the Subject Constraint. A fronted topic will not be possible since the complement clause will not be entered on the relevant topic card blocking the interpretation of the gap inside it. Speakers vary as to which manner-of-speaking verbs require an f-structure such as (50). For some speakers and some of these predicates and in certain contexts, the f-structure in (46a) is also available.

4.5. Topic Islands Topicalization is generally possible only in main clauses. A few predicates which select a that-clause allow topicalization within it: (51) a. I said that this book Mary gave to Peter. b. I believe that on this shelf Peter put the radio. Topicalization in a that-complement forces an interpretation in which the complement forms the main f-structure as in (46b). This automatically limits the phenomenon to those predicates that allow the matrix to be backgrounded. With this f-structure extraction will be blocked as required: (52) a. *This is the person who I said that this book Mary gave to. b. *This is the radio that I said that on this shelf Peter put. This account leaves unexplained the following data from Culicover (1993) who shows that sentential adverbials in general do not give rise to such topic islands: (53) a. This is the tree that I said that just yesterday had resisted my shovel. b. This is the tree that I said that just yesterday I had tried to dig up with my shovel. Note that leaving out just reduces acceptability. Withoutjust, the reading in which the sentential adverbials must be interpreted as stage topics is prominent. I therefore assume that the sentential adverbials which do not cause topic islands do not function as topics in their f-structures. The Subject Constraint on I-dependencies accounts for certain subject-object

~---

...,._, aa--••

~

~--~

~-•

_...,._.,... • ...,

~

•• .,..., ..... _...,

structure marks the organization of sentences into focus and topic constituents. In Erteschik-Shir and Lappin (1987) we argue that all modes of perception are organized into foreground and background constituents. Focusing is viewed as a single task-specific mechanism which identifies the foregrounded constituent in representations of all modular systems. Focusing is therefore a nonmodular process which provides the interface between the modular system and the central cognitive mechanisms. It follows that although focusing plays a central role in grammar, it is not unique to the language faculty. The identification of a topic is, however, characteristic of linguistic structure. In vision, for example, foregrounding is necessary for the formation of a visual report, but this report is not "about" anything (i.e., it does not have a topic). Whereas the capacity to identify a topic is a distinctive property of the human brain and of the language faculty in particular, focusing is a basic property of all cognitive systems. The linguistic level of F-structure in which both topic and focus are identified is therefore a fundamental part of Universal Grammar (UG), which determines the class of possible languages. The rule of predication takes f-structures as its input, producing a well-formed discourse representation. F-structure can therefore be viewed as an interface level between the grammar and the C-I system replacing LF. F-structure also mediates grammar and PF, itself an interface level with the articulatory-perceptual system. This is necessary for the derivation of intonation. I have shown that certain conditions on structural descriptions (SDs) can be defined at F-structure. In Erteschik-Shir and Rapoport (in press, in progress) we show that certain F-structure properties are derivable directly from properties of verbal projections. It might therefore be feasible to hold that UG principles of the various modules of grammar all apply at F-structure. This would present a different perspective on the execution of the Minimalist Program (Chomsky, 1995). 25 Applying the basic tenets of this program to the f-structure theoretical model of grammar proposed here would clearly have very different results. Replacing LF with F-structure, as I suggest, is a natural consequence of its function as an interface to conceptual-intentional structure since focusing is an innate reflex, essentially involved in perception as a whole, and topic identification is an innate and essential part of C-1 structure itself. If grammar is f-structure driven it is important to explore how principles of f-structure are parameterized across languages, an issue I hope will be the subject of future research. A potentially interesting question to address is the parameters of canonical f-structure. In English, canonical f-structure takes the syntactic subject or alternatively a stage as its topic. Presumably, languages may differ as to what their canonical f-structure is. Since the Subject Constraint says that I-dependencies are restricted to canonical f-structures, this constraint will differ according to the canonical f-structures defined for a particular language. A language such as Hungarian, which has fixed positions for topic and focus, may differ in this respect from a language such as Japanese, which has morphological markers to do

~

~

Conference on Interfaces in Linguistics. Oporto, November 1995. APL/Edi96es Calibri, Lisboa. Erteschik-Shir, N., and T. R. Rapoport (in progress) Verbal Projections, Ben-Gurion University, Beer-sheva, Israel. Fiengo, R., and R. May (1994). Indices and identity. The MIT Press, Cambridge, MA. Frampton, J. (1991). Relativized Minimality: A Review. Ling. Rev. 8.1, 1-41. Frampton, J. ( 1992). The fine structure of wh-movement and the proper formulation of the ECP, in W. Chao, and G. Horrocks (eds.), Levels, principles and processes: The structure of Grammatical Representations. de Gruyter, Berlin. Grosu, A. (1982). Extragrammatical motivation for certain island phenomena. Theoretical Linguistics 9, 17-67. Kratzer, A. (1989). Stage-Level and Individual-Level Predicates, in Papers on Quantification. Department of Linguistics, University of Massachusetts, Amherst. Krifka, M. (1990). 4000 Ships Passed through the Lock: Object induced Measure Functions on Events. Linguistics and Philosophy 13. Kuna, S. (1976). Subject, Theme, and the Speaker's Empathy-a Reexamination of Relativization Phenomena, in C. Li (ed.), Subject and topic. Academic Press, New York. Krach, T. (1989). Amount quantification, referentiality, and long wh-movement. Unpublished manuscript, University of Pennsylvania, Philadelphia. Lambrecht, K. (1994 ). Information structure and sentence form. A theory of topic, focus, and the mental representations of discourse referents. Cambridge Studies in Linguistics, CUP, Cambridge, England. Landman, F. (in press). Events and plurality. The Jerusalem lectures. Unpublished manuscript, Tel Aviv University, Israel. Li, C., and S. Thompson (1976). Subject and topic: A new typology, in C. Li (ed.), Subject and topic. Academic Press, New York. Partee, B., H. (1992). Topic, focus and quantification, in Proceedings of the 199I SALT Conference. Department of Modern Languages and Literatures, Cornell University, Ithaca, New York. Partee, B., H. (1994). Focus, quantification, and semantics-pragmatics issues. Preliminary Version, in P. Bosch and R. van der Sandt (eds.), Working Papers of the Institute for Logic & Linguistics, IBM Deutschland Informationssysteme GMBH Scientific Centre, Heidelberg. Reinhart, T. (1981). Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27. Reinhart, T. (1983). Anaphora and semantic interpretation. Croom Helm, London. Reinhart, T. (1986). Center and periphery in the grammar of anaphora, in B. Lust (ed.), Studies in the acquisition of Anaphora. Vol. I, 123-150. Reidel, Dordrecht. Rizzi, L. 1990. Relativized minimality. M.I.T. Press, Cambridge, MA. Rooth, M. (1985). Association with focus. Ph.D. dissertation, Amherst, University of Massachusetts, reproduced by GSLA. Ross, J.R. ( 1971 ). Variable strength. Unpublished manuscript, M.I. T., Cambridge, MA. Schwartz, A. (1976). On the Universality of Subjects: The Ilocano Case, in Charles N. Li (ed.), Subject and topic. Academic Press, New York.

~

subject, than, say, about its object, since in the logical form, something is predicated directly of the subject's interpretation." More recently Lambrecht (1994) argues that the unmarked information-structure sequence is topic followed by focus and that the subject is the unmarked topic. I make no assumptions here with respect to f-structure markedness in a topic-prominent language. 11 The last three of these examples were supplied by Georgia Green (personal communication) 12 The distinction between restrictive and I-dependent questions is, however, pertinent and will be discussed below. 13 A reviewer correctly pointed out that the effect of a question on a context differs from that of an assertion in that the speaker's (rather than the hearer's) information state is examined. 14 I exclude from discussion rhetorical uses and test situations. Such uses, I argue in Erteschik-Shir (in press), are metalinguistic and different constraints apply to them. 15 Kuno (1976) and Grosu (1982) propose that extraction is conditioned by the accessibility of the extracted NP to interpretation as the topic of the domain of extraction. This constraint follows from the interpretation of the wh-phrase as topic and will not explain all the cases excluded by the Subject Constraint on I-dependencies. For discussion see also Erteschik-Shir and Lappin (1983). 16 Subordinate foci, including metalinguistic ones and their associated stress, can, of course, occur within islands. The Subject Constraint is, however, sensitive to main foci. 17 For arguments that no other f-structure is assignable to such sentences see ErteschikShir (1973) and Erteschik-Shir and Lappin (1979). 18 An issue addressed in Erteschik-Shir (1982) is the relative ease with which these and other constructions license extraction in Danish compared to English, for example. I related this to the prevalence of topicalization in Danish. This connects well with the claim that it is not the Subject Constraint which is violated in these cases in Danish. 19 Much discussion has been addressed to the extraction of adjuncts. See among others Cinque (1990), Kroch (1989), Rizzi (1990), and the references they cite. An initial version of the account given here also appeared in Erteschik-Shir ( 1992). 20 It is not clear whether wh-reasons can form an I-dependency with another wh-phrase at all: i.



*Who went why? (vs. Who went and why?)

ii. *What did John do why? (vs. What did John do and why?) iii. *Why did John do what?

"

21 The issues discussed in Szabolcsi and Zwarts (1993) only partially overlap those covered in Erteschik-Shir (in press). It will be a challenge for anyone to address the interesting and extensive coverage of data to be found in Szabolcsi and Zwarts (1993) from an f-structure perspective. 22 The term "bridge verbs" is derived from Erteschik-Shir (1973). There I argue that verbs that must be inherently focused block the interpretation of the complement as a focus in violation of the constraint on extraction (here rephrased as the Subject Constraint). 23 This is the class of predicates which licenses negative polarity items in their comple-