:35»: Gil-r5 .— l. , 53.3.3. \-AD¢-:l7.:< M v. {I :3: 5:, g..5: $.17 C i; 51.. (I... a!) a! strip) PC: 31......ng III)... ill?! 1 .52....) irlo.:»€l;.o§ .- 9») 1.5.71 (pf A. 4. (7: .rr .5... L)., l). Fix '02... La . r. 3 ar....1.rv.l.rdu.frh} .( it}?! .4 ((41:11. 5215!.) n ((5.13. 4: . it]! 3:17.}: I! (Ii: II . 1.. i .7 I.Iq.1.;r . III! {17; (ff/{1; ..P.r..l..v.«.r(5.f¢¢.1! 2. 5:5}: (vb/.IIII. .. , y . Wfififlg lllllllllllllllll am --.-.~ -r~ -- -. This is to certify that the thesis entitled UNDEFINABILITY OF AGREEMENT AND THE FINITE/INFINITIVAL CLAUSE DIStINCTION presented by Paul Timothy Kershaw has been accepted towards fulfillment of the requirements for M. A. , Linguistics degree in [y Major professor Date gf/f 2% 4732/ 0-7639 MS U is an Affirmative Action/Equal Opportunity Institution LIBRARY Kiel-Icon State University PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. DATE DUE DATE DUE DATE DUE Fit: 0 o 299;; MSU Is An Affirmative Action/Equal Opportunity Institution _ c:\clrc\datedue.pm$p.1 UNDEFINABILITY OF AGREEMENT AND THE FINITE/INFINITIVAL CLAUSE DISTINCTION by Paul Timothy Kershaw A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Linguistics and German, Slavic, African, and Asian Languages 1 992 ABSTRACT UNDEFINABILITY OF AGREEMENT AND THE FINITE/INFINITIVAL CLAUSE DISTINCTION by Paul Timothy Kershaw It is generally assumed in the Government-Binding theory of syntax that finite and infinitival clauses have different structures in that finite clauses carry in their Infl (or Aux) node an abstract marker [+Tense], which as— signs nominative Case and co—occurs with agreement features, while infinitival clauses carry the abstract marker [-Tense], manifested by the morpheme to, which does not assign Case anui which does run: co—occur with agreement features. In this paper, I attempt to show that it is the agreement features, run: the Case assignment, which are responsible for the difference between the two clause types, demonstrating this by replacing the more traditional notion of agreement with a rudimentary feature unification system. This analysis requires Case verification to be optional, which requirement allows for a simplification of the lexicon. Table of Contents 0. Introduction . . . . . . . . 1 1. Agreement and clause distinction . 9 §0: Introduction . . . . . . . 9 §1: The structure of Infl . . . . 12 §2: Syntactic features . . . . . 16 §3: Agreement as a feature set . . 19 §§l. Agreement in parsing . . . 19 §§2. Generation: Finite clauses . 23 §§3. Generation: Infinitival clauses 27 §4: Tense and infinitival clauses . 32 §5: meinative Case assignment . . 39 §6: Conclusion. . . . . . . . 43 2. Matrix verb types . . . . . . 46 §0: Introduction . . . . . . . 46 §l: Clauses with non-PRO subjects . A8 §§1. believe. . . . . . . . 48 §§2. seem and think . . . . . 53 §2: Clauses with PRO as subject . . 58 §§1. £39 as subject . . . . . 58 §§2. PRO and control verbs . . . 62 §3: Clauses with either. . . . . 68 3. Case verification and the lexicon . 73 §0: Introduction . . . . . . . 73 §l: One- and two-argument verbs . . 75 §2: Three-argument verbs . . . . 79 §3: ECM verbs . . . . . . . . 81 §4: Conclusion. . . . . . . . 83 A. Related issues and problems . . . 85 §0: Introduction . . . . . . . 85 §1: The pleonastic construction . . 86 §2: The passive construction . . . 91 5. Conclusions . . . . . . . . 98 Bibliography . . . . . . . . . 102 iii 0. Introduction According to the current framework of Government-Bind— ing Theory, there are three Case—assigning categories (at least in English and related languages), namely Verb, Prepo- sition, and finite Infl. Hence, in (1), the noun phrase Lenin receives nominative Case from the tense/agreement feature complex in Infl, the NP ea book on Marx receives accusative Case from the verb read, and the NP Marx receives accusative Case from the preposition on. These assignments are shown in (2): (l) Lenin read a book on Marx. (2) IP 1 l— N r i—fi "U 1s Lenin +BSING read a book on Marx +PAST L_ L->[ACC] [NOM]<————J >[ ACC ] The Case filter (Chomsky 1981, 49; Lasnik and Uriagereka 1988: 10, 12) asserts that every overt NP requires a Case, and implies (as I discuss below) that every Case available for assignment must be assigned. Since there are three categories in (2) that assign Case, each of which assigns one Case, all three NP nodes in (2) must contain an overt NP entry (as opposed to a non-Case—receiving NP, such as Pro). The two assignments of Accusative Case in (2) seem to be somewhat uncontroversial in the literature. Not so obvious is the validity of the claim that finite Infl as— signs Case. This claim is motivated by data such as in (3). The embedded clause in (3a) is said to be finite, while the embedded clause in (3b) is said to be non-finite: (3) a. It seems John ip happy. b. John seems pp pg happy. There are two differences between the embedded clauses in (3). One is that in (3a) the verb of the embedded clause is inflected for agreement and tense, whereas in (3b) it isn't; the other is that in (3a) the clause has an overt subject NP, while in (3b) it does not. Chomsky (1981: 52) at- tributes tflua latter difference tx> the content (H? Infl, saying that only those Infl which contain the feature com— plex [[+Tense], AGR] assign nominative Case. Similarly, according to the Case filter, (4a) is ill-formed because the subject of the embedded clause is not assigned Case, either by Infl or by the matrix verb (cf. Chomsky 1981: 49), and (4b) is ill-formed because the nominative Case available from Infl cannot be assigned to an appropriate NP (i.e., an overt NP, a wh—trace, or, in some languages, pro). (4) a. *It seems John to be happy. b. *John seems twp is happy. What is it in Infl that assigns Case? Chomsky (1981: 52) suggests that it is Agr, but the precise nature of agreement is still controversial. Pollock (1989) argues the Infl should be divided into two maximal projections, TnsP and AgrP. However, Iatridou (1990: 569) refutes this claim, suggesting that Agr is not even appropriate as a separate syntactic element at all. If agreement is not a syntactic element (but is instead a relationship between two elements), then it cannot be the element in Infl responsible for assigning Case. Let us assume, on the other hand, that it is the tens— edness of the Infl that assigns nominative Case. It is not obvious why this should be, since (as I discuss below) Tense and Case have no obvious relation, while Agreement and Case do. A claim that is motivated only by empirical facts about the phenomenon under question and not by any other (unrelat— ed) facts is a weak one at best, and if possible should be replaced by a claim that both supports the immediate facts and is supported by some facet or version of the linguistic theory (which, ideally, has been generated from other facts about language). In this thesis, I will demonstrate that the facts in (3) and (4) reflect subject—verb agreement phenomena, and not (directly) Case assignment. That is to say, (3a) is well-formed because the subject and the verb of the embedded clause agree. Likewise, (3b) is well-formed because the subject and the verb (via Infl) both lack agreement fea- tures. On the other hand, (4a) and (4b) are ill-formed because in each, only one of the subject-verb pair carries agreement features (the subject in (4a), the verb in (4b)). This claim has several repercussions. First of all, the question arises as to whether this applies to all matrix verbs subcategorizing for embedded clauses, or just.aa spe- cific subset of these. One motivation for a distinction between finite and infinitival clauses is that some matrix verbs subcategorize for either type of clause (such as seem, as above, and believe, (5)) while others allow only infini- tival clauses (such as want (6) and persuade, (7)): (5) a. I believe (that) Mary is innocent. b. I believe Mary to be innocent. c. I believe *o/myself to be innocent. (6) a. *I want Mary is innocent. b. I want (for) Mary to be innocent. c. I want ¢/?myself to be innocent. (7) a. *I persuaded Mary does it. b. I persuaded Mary to do it. This has led to the exploration of the idea that there is a structural difference between finite enul infinitival clauses. However, there is a major difference between the clauses taken by believe (and related verbs) and those taken by want: the phonetic form of the subject of the embedded clause when it is the same as that of the matrix clause (as illustrated in (SC) and (6c)). The difference between these two verb types and verbs such as persuade is more subtle, and is dependent upon theta—role assignment, and not directly upon structure. Hence, I assert that there are two clause types that have typically been grouped together as infinitival. The distinction that this paper explores in detail is time one between infinitival clauses taken by Raising/ECM verbs (e.g., seem and believe) and those taken by other verbs, although I also discuss briefly whether the embedded clauses in (6b) and (7b) have the same structure or not. Secondly, the Case filter implies that each Case avail— able for assignment must be assigned. However, I assert that this implication is insufficiently motivated, and that the empirical facts which support it can be explained in other ways. Indeed, as I will show, both the lexical en- tries of relevant verbs and the theory itself can be simpli- fied if Case assignment is assumed to be optional (that is, if there may be some unassigned Cases at S-Structure), although I make no adjustment herein to the overt assertion made by the Case filter (i.e., *NP, where NP is overt and has no Case). (Since I am using a feature system, Case is verified, not assigned. As in this paragraph, I use the traditional Government-Binding terminology only when appro- priate; otherwise, I refer to Case as being "available for verification" on VP and PP, rather than being "assignable".) Finally, any change in one facet of a theory will inevitably affect the analyses of related phenomena. The Case filter is motivated primarily by three types of phenom— ena: subject behavior in embedded clauses (the phenomena discussed herein), the passive transformation, and expletive it insertion. The current GB treatment of passive clauses is shown in (8), where (8a) is the D-Structure and (8b) the S-Structure of the same clause. Movement of the D-Structure object, Hildebrandt, to subject position is forced because passive verb forms assign a theta—role but no Case. (8) a. PAST be smitten Hildebrandt (by Belinda). b. Hildebrandt was smitten (by Belinda). In the case of expletive insertion, it is inserted in the subject position of the clause in (9a), to form (9b), in which the nominative Case assigned by Infl to the subject position is realized on the expletive. (9) a. D-Structure: PRES seem that John PRES be happy. b. It seems that John is happy. I will show how each of these other phenomena can be ex- plained in light of the disambiguation of the Case Filter contained in this paper. This paper deals with each of these issues in a sepa- rate chapter. The first chapter deals specifically with the central thesis, showing that subject-verb agreement is the relevant issue 1J1 distinguishing finite anui infinitival clauses. I present a system of features for agreement, distinguishing between nodes which are underspecified for features and those which are undefined for features, arguing that finite clauses may represent underspecification (i.e., that the subject or Infl may be underspecified), and that all infinitival clauses represent undefinedness with respect to agreement. Since Case is not the central issue in dis- tinguishing types of clauses, I assert that nominative Case, like the agentive theta—role, is assigned/verified by the verb phrase instead of by Infl, resolving the problem of using [+V] to classify Case assigners, pointed out by Chom— sky (1981, 52). In the second chapter, I clarify the scope of the thesis, illustrating time similarities anui differences be- tween the various types of infinitival clauses taken by verbs. I suggest that only raising verbs, such as seem, and so—called Exceptional Case Marking (ECM) verbs, such as believe, subcategorize for infinitival clauses with non-PRO subjects, whereas control verbs, such as want, subcategorize for infinitival clauses with PRO subjects. The latter type of infinitival clauses generally pose time biggest problem for any theory of infinitival clause behavior, as shall be discussed in comparing the verbs manage, want, wait, except, hope, and convince. The third chapter discusses optionality (as opposed to obligatoriness) (H? Case verification, discussing lunv this simplifies the lexicon, allowing for an initial generaliza- tion about the pattern of omission of objects when these are implied. Fer example, the difference in the number of objects in the clauses in (9) is caused not by different lexical entries for give but by a difference between the semantics and the theta-role configuration selected by the verb, with non-overt semantic arguments having a pragmati- cally implied referent: (9) a. I gave Debbie the salt. b. I gave blood. C. I gave (at the office). These claims allow for simplification of the lexicon. The final chapter deals with other Case related trans- formations, namely the passive transformation and expletive insertion. The passive transformation is run: affected in simple cases, since in.21 sentence such an; in (10), the object NP Nary still lacks Case, as the Case filter demands: (10) *It was smitten Mary. However, three—argument verbs such an; give create problems if they are assumed to have only one lexical entry (as opposed t1) two). Furthermore, tjma treatment cu? expletive insertion is strongly affected by this thesis, since it must be inserted in (11a), according to current theory, because the subject position is assigned Case. lWy claims seem to predict the possibility of (11b), which is clearly ill— formed: (11) a. It/*¢ is raining. b. *To be raining. I argue that the phenomenon in (11) is not due to Case but is instead due to the nature of tense. Now, I begin by examining, in more detail, the issue of subject-verb agreement within a feature system. 1. Agreement and clause distinction §0: Introduction Before discussing features, I would like to briefly clarify the relevant clause distinctions being examined, and to illustrate the differences between finite and infinitival clauses. As stated in the introduction, finite and non—finite clauses are typically distinguished on the basis of tense/agreement inflection. There are three types of non- finite clauses, depending on the form of the verb, for which Radford (1988: 288) uses the following terms: infinitival clauses (the bare form of the verb, usually preceded by to, in (1a)), gerunds (in (1b)), and participles (either past or present, in (1c)). (1) a. Freud seems [pp pg preoccupied with sex.] b. Freud likes [talking about other people's dreams.) c. Jung had [his theories examined by Freud.] This paper deals primarily with infinitival clauses, and specificaly with infinitival clauses bearing the parti- cle to, contrasting such clauses to finite clauses. Gerund clauses are relevant in that they overlap somewhat in dis— tribution with infinitival clauses, as will be discussed in a later chapter. Finite clauses and infinitival clauses appear to differ in the following respects: 10 (A) Finite clauses require their subjects to be overt NPs (with two exceptions: wh-traces, in English, and pro in pro-drop languages, such as Basque (Mugarza 1990); in both cases, I would argue that the non-overt NP still carries agreement features), whereas infinitival clauses may lack overt NP subjects; this is illustrated in (2), where the finite embedded clauses in (2a-b) have overt subjects, while the infinitival embedded clauses in (2c-d) differ, with (2c) showing an overt subject, and (2d) showing none. (2) . Kravchuk believes [Yeltsin is lying.] . It seems [there's some problem.] . Kravchuk believes [Yeltsin to be lying.] . There seems [p to be some problem.] Q-OD‘D) (B) Finite clauses require their subjects to be marked with nominative Case, whereas infinitival clauses require overt subjects, when present, to be marked as accusative; this is illustrated in (3), where the pronoun he is nomina— tive and the pronoun him is accusative. (3) a. Yeltsin seems honest, but Kravchuk believes [pg is lying.] b. Yeltsin seems honest, but Kravchuk believes [pip to be lying.] (C) In finite clauses, at PF, the subject NP and the verb must agree in all features for which they are both specified, while infinitival clauses show no such agreement. That is, as I discuss below, the unification of agreement features on the subject with those on the verb in finite clauses must exist. Agreement errors account for the ill— formedness of the examples in (4). (4) a. *The boy are mistaken. (NP: singular; V: plural) b. *The boys is mistaken. (NP: plural; V: singular) I argue that this is not an actual difference between clause 11 types, despite time superficial appearance (H? the data, since, under my definition of agreement, it is vacuously true that infinitival subjects agree with their verbs. (D) Finite and infinitival clauses differ in time way semantic tense is manifested morphosyntactically. While this is typically treated as a syntactic difference, I argue that it is not. This chapter consists of six more sections. In the first section, I discuss the structure of Infl within cur- rent GB theory and Iatridou's (1990) suggestion of treating agreement as a feature. In the following section, I discuss syntactic features in general, especially in non—transforma— tional generative grammars. In section three, I discuss syntactic features as specifically applied to agreement, distinguishing between the notions undefinedness and under— specification. ffiua following section discusses issue (D) above, showing that tense is the result of the meaning of morphemes, a claim that is contrary to the common treatment of tense as a feature (or as an abstract morpheme that is inflected for agreement). In section five, I assert that nominative Case may be assigned (or, more accurately, veri- fied) by VP, rather than by Infl. Finally, I summarize the chapter by relating the points to the four differences ((A)— (D)) cited above. 12 §1: The structure of Infl The internal structure of what is typically called Infl in Government-Binding theory has been controversial and remains unsettled, presumably because (H? its abstract na— ture. At D—Structure, it cannot typically contain free morphemes, with the arguable exceptions of to and modals (Chomsky 1981: 140 n28; Radford 1988: 304—5), and so its existence as a distinct syntactic entity is evidenced pri- marily by do—support. Normally, any non-lexical (or, in my View, non—free morphemic) entry in lkdfl..is morphologically bound to the adjacent matrix verb of the clause, at some level of the generation. The existence of Infl em; a separate syntactic entity has been argued for on the basis of ckxsupport, which re- sults (arguably) from time separation (n? tense/agreement features from the main verb. There are two cases in which Infl has been argued to be separated from the verb. The first, involving question formation, causes the D-structure contents of Infl tx>rmwne to the Comp position, separating these contents from the verb. In (5), the verb must be inflected for tense and person, as in (5b), because it is adjacent to Infl, whose bound entry has not been moved; hence the ill—formedness of (5c), where the verb is unin— flected. In (6), the verb cannot be inflected (as it is in (6b)) because the contents of Infl have been moved (NB INFL in (6) represents the D—Structure contents of Infl, not the syntactic node). Do is inserted in Comp to give a stem to 13 the bound agreement and tense morphemes, as in (6c): (5) . D-Structure: Belinda INFL regress. . Belinda regresses. *Belinda regress. . D-Structure: William INFL conquer who. . *Who INFL William conquer? . Who did William conquer? (6) OUDDOU‘N This analysis requires a stipulation concerning where the morphological amalgamation of the bound contents of Infl and the matrix verb occurs. It cannot occur at D—Structure (as Chomsky 1986: 68-9 suggests), since then.za movement could not separate the two elements. It cannot occur at S— Structure (as Chomsky 1981: 256—7; Radford 1988: 402—403 suggest), since it would then presumably be the result of a transformation such as affix-hopping or V—to-I raising, the latter of which Chomsky (1986: 68-69) asserts occurs before I-to-C raising. Finally, it cannot occur at PF, since the verbs be and have (in its Aux usage in SAE, or in Aux and Main verb uses in British English) do not allow do—support. Assuming that these verbs originate in the V position, in interrogatives they move t1) the Comp. There are tn“) possibilities: the verb could move directly to Comp, a movement which is prob— lematic, or the verb could move to Comp through the Infl, a movement which is impossible, since if Infl already contains tense/agreement information, and ii? the amalgamation takes place at PF, movement of the verb past the Infl (i.e., past the point of morphological amalgamation) is blocked. An alternate analysis, in Chomsky (1986: 68—9), ac- 14 counts only for questions involving free morphemes in Infl or questions having tmecnr have, and practically eliminates the need for Infl at all. Less controversially, the negative adverb not occupies a position between Infl and VP, roughly as in the fragment in (7), which, by separating Infl and VP, also necessitates do—support, as illustrated in (8). (7) IP I——‘——‘l NP 1' I AvP Av VP i not (8) a. *William Infl not conquered Italy. b. William did not conquer Italy. Since the adverb placement occurs at D—Structure (as part of the clause generation), it does not matter where the amalga— mation of Infl and V occur, as long as it is not inserted, already amalgamated, from the lexicon. If it is difficult to establish the existence of Infl, it is even more difficult to determine its internal struc- tnnmx However, until recently, the issue received little rigorous attention, leaving textbook sources to provide aui hoc analyses (Haegemann 1991: 104, 147; Horrocks 1987: 104- 5; van Riemsdijk and Williams 1986: 274). In all of these, Infl somehow contains information on both tense and agree- ment. However, as I discuss in an earlier paper (Kershaw 1991), all of these violate X—Bar Theory in some way. Pollock (1989: 397) argues that Tense and Agreement 15 require separate maximal projections, giving the structure in (9) (Pollock (77)) for a clause: (9) CP 0.... ._].._. "U l NP TP' l T NegP l I I Neg AgrP l l 1 Agr VP 1 (Adv) i Pollock motivates (9) using the behavior of verbal inflec— tions in the presence of adverbs and negation in finite and infintival clauses. Iatridou (1990) points out, however, that Pollock's motivation for (9) is insufficient, and argues that only Tense is a distinct separate syntactic element with its own maximal projection, and that "Agr is not a structural posi— tion but a relationship, specifically a spec(ifier)-head relationship" (Iatridou 1990: 569). Indeed, if a universal syntax is 11) be developed, i1: is ultimately desireable to treat agreement as ae'non—positional phenomenon, since lan- guages vary greatly in the amount and type of agreement that takes place. For instance, Chinese has none at all (Iatri- dou 1990: 552), while Basque verbs agree with their sub— jects, their direct objects, and their indirect objects (Mugarza 1990: 16). If each instance of agreement in Basque required its own structural position, and if such structural positions are universally required, then there would be 16 corresponding agreement positions in Chinese which are available but always unfilled (Iatridou 1990: 552). Following Iatridou, I argue that agreement is a feature relationship. The next section deals with syntactic fea- tures in general, after which I relate a feature system to agreement and infinitival clauses. §2: Syntactic features In an early work, Chomsky (1965: 170-179) offers a rudimentary set of syntactic features for agreement on the noun, listing some of these features in the lexicon and introducing others through grammatical rules. In his German example, inherently lexical features include those of gender and declension class (DC), while grammatically determined features include those of Case and number (Chomsky 1965: 171). Unlike phonetic features, syntactic features may be specified by integers instead of by just a binary value (Chomsky 1965: 171); like phonetic features, rules specify— ing syntactic features can be vmitten (Chomsky 1965: 174- 176). For example, the German noun Brdder has the inherent lexical features {[1 Gen], [1 DC], ... }, and may receive the grammatically determined features {[2 Num], [2 Case], ... } via various syntactic rules. According to Chomsky‘s example (1965: 175), time determiner receives Gender, Number, and Case specifications from the noun it modifies, yielding the subcategorization in (10) for the NP Die Brdder (based on Chomsky's (30)): 17 (10) NP l 1 Art N I l [l Gén] [2 th] [2 CaSe] '1: [1 Gen] [2 Num] [2 Cése] [1 DC] - .I ..é‘ Die Bru er l J Chomsky‘s feature system has several weaknesses. Principal- ly, though, i1: is merely descriptive, offering IN) limita— tions on the types of features that can be included or the types of rules that can involve features. This has lead to a general lack of discussion of the mechanisms involving features in GB theory, with the result that features are treated with a "black box". For instance, Theta theory requires that every NP (arguably, other than expletives) have a theta—role. In (11), then, Fernando, his medical texts, and the library must all receive theta roles. Ac— cording to CB theory, the verb give (being ditransitive) has two theta roles to assign (one to his medical texts and one to the library), and the VP give X Y has one theta role to assign (to Fernando). (11) Fernando gave the library his medical texts. However, it is rarely specified what these theta—roles are; in fact, since GB theory is (Hi the whole unconcerned with semantics, what the theta—roles are seems irrelevant. Further, it is not clear how it is that the theta roles are structurally related to the NPs they are assigned to. Other generative theories of syntax have been more concerned VfiJfll syntactic features, including augmented Phrase Structure Grammars, Unification Categorial Grammar, 18 and Lexical-Functional Grammar, in each of which a central theme is some variant of the notion of a Feature Structure (Sag, Kaplan, Karttunen, Kay, Pollard, Shieber, and Zaenen 1986: 238). Int LFG, for instance, clause structure is determined by constituent structures (c-structures, typical- ly given in tree diagrams), while each node of a c—structure is associated with a functional structure (f—structure), which contains grammatical and semantic information about each node (Sag et al. 1986: 246—51). f—structures also include agreement features. The f-structure for the English pronoun she, for example, is given in (12) (Sells 1985: 149 (19); see also Kaplan and Bresnan 1982): (12) Q ~ - PERS 3 NUM SG GEN FEM CASE NOM PRED 'PRO' The f—structure in (12) can be associated with the NP node in the c-structure fragment in (13) (Sells 148 (17)). The information in fb is unified with the information in some f3 associated with the VP node (cf. Sells 150 (20)) to give a complete f—structure, II, to the S node. (13) 3:1 I I I (f1 Subj) = f2 f1 = f3 NPfi3 VP:3 In this way, features are passed from the terminal nodes at the bottom of the c-structure up to the top node of the clause. 19 Phrase Structure Grammars, likewise, are concerned heavily with features. Gazdar, Klein, Pullum, and Sag (1985: 21—2), speaking of Generalized Phrase Structure Grammar, define categories as sets of feature specifica- tions, 21 great elaboration (Hi the quad—partite [:N 1V] system of Chomsky (1970). As an example of a possible category, they give (14) (2 Gazdar et al (3)): (14) { } (14) represents a third person singular accusative noun phrase. For the purposes of this thesis, I combine and simplify the feature systems used by LFG and GPSG to account for syntactic phenomena that must be accounted for using fea- tures, specifically agreement and Case (which Pullum (1984: 81), among others, treats as a type of agreement), while retaining traditional GB—theory to account for the majority of syntactic behavior. The next section deals specifically with agreement. 53: Agreement as a feature set §§1. Agreement ip parsing Before discussing agreement features within Government- Binding Theory, I would first like to discuss agreement in the context of parsing, whidh GB seems only marginally 20 concerned with, but which is a central topic in more func- tionally based generative grammars. Pullum (1984: 83) cites six types of cross-category agreement evidenced in natural languages, depending on the categories which are in agreement. These are subject—verb, determiner-noun, adjective-noun, verb-DirObj, verb—IndirObj, and preposition-NP. In English, only the first two are evi- denced. The examples in (15) illustrate subject—verb agree— ment: (15a) is ill-formed because the subject NP is in the third person, while the verb (inflection) is marked for first person; (15b) is ill-formed because the subject NP is in the third person singular, while the verb (inflection) is plural (and/or in the second person): (15) a. *He am sorry. b. *He are sorry. The examples 1J1 (16) illustrate determiner—noun agreemnt: (16a) is ill—formed because the determiner is singular, while the noun is plural; (16b) is ill-formed because the determiner, many, indicates a count noun, while the noun, controversy, is a mass noun: (16) a. *A boys ran. b. *The claim caused many controversy. The former type of agreement is of interest to the present discussion; the latter type is important here only in that it indicates the presence of another agreement feature, Countability, which is necessary for determiner-noun agree- ment but not for subject-verb agreement, an issue I will not discuss here. 21 A typical GB analysis of agreement is to assume that Infl somehow contains complete information on agreement, which is derived from or checked against the complete agree- ment information of the subject at some syntactic level (see the previous section for general comments, and Chomsky (1965: 170—184)). Hence, in (17a), the subject NP and the Infl both contain the information [3—PERS SING], generating (17b): (17) a. The boy-35g INFL—[nPAST]-35g run. b. The boy runs. Assuming that both the subject and the Infl are fully specified is problematic in that it requires an extreme number of homophonic lexical entries that differ only with respect to agreement features (Sag et al. 1986). For in- stance, all verbs in English except be have two present— tense (indicative) forms and (nme past-tense form, even though there are six possible person/number combinations for each tense. If Infl must be specified for both agreement feature values, Sag et al. argue, a past-tense verb such as laughed would have six homOphonous parser entries, as in (18). In (18a), for instance, laughed is parsed as include ing an Infl marked as past tense and first person singular, while in (18d), laughed is parsed as including an Infl marked as past tense and second person plural. (18) a. laughedi = INFL-[+PAST]-1sg + laugh b. laughedé = lNFL-[+PAST]-lpl + laugh c. laugheda = INFL-[+PASTJ-25g + laugh d. laugheda = lNFL-[+PAST]-2p1 + laugh e. laugheds = INFL-[+PAST]-3sg + laugh f. laughedb = INFL-[+PAST]-3pl + laugh 22 (My notation trivializes morphological phenomena by assuming that in the parser a compound word is divided into its component parts. I follow Kay (1985: 254) in assuming a single lexicon with two syntactic mechanisms, a parser and a generator, but assume his "strong claim" that there is one grammar accessed by both mechanisms). The same redundancy is also a problem for subject noun phrases in English, since some nouns have the same singular and plural forms. This is true, for instance, of fish, so that in (19), the subjects are homophonous, even though the inflections on the verbs show the subject in (19a) to be singular and that in (19b) to be plural (cf. Sag et al. 1986: 244—5 (5)—(6)). (19) a. Ihg fish swims. b. Tpg fish swim. Taken out of context, the noun phrase the fish is ambiguous between a singular and a plural reading, just as laughed in (18) is six ways ambiguous. It follows, then, that the combination of the fish with a past tense verb (other than be) will be ambiguous as regards number. This is shown in (20), which could be the past tense equivalent of either (19a) or (19b). (20) The fish swam. Sag et al. account for the phenomena above by allowing the features of lexical entries and nodes to be underspeci- fied. (20) is collapsed to a single entry which contains no agreement information (cm? the type tummur consideration 23 here), while the present tense forms of laugh have two parser entries, given in (21). (21) a. laughs = INFL-[wPAST]-35g + laugh b. laugh = INFL-[nPAST]-n35g + laugh (21b) is unusual in that it does not directly specify any agreement features, but rather specifies what feature set it is not compatible with. This is based on the assumption that there are two present tense inflectional morphemes in English (three for be), one of which is never overt. This is hardly anr uncontroversial claim; see, eugu, Andrews (1990) for a different treatment. §§2. Generation: Finite clauses Let me now apply the above discussion specifically to GB-theory. The structural relation of a subject to an Infl in current GB—Theory is shown in (22) (Chomsky 1986: 3): (22) IP F ' fl NP 1' r—L—fi I"—"__'I Det N' 1 VP (Adj?) N Agreement in (22) takes place in English between the termi— nal nodes N and I (and between N and Det). For the purposes of this argument, I assume that neither VP nor Ade in English carry agreement features. Following the theories discussed in the previous section (LFG and GPSG), and other feature—based theories, I assume that a mother node carries the agreement features of all of her daughters (although in 24 some circumstances the head daughter receives special treat— ment: see below). This is a rough application, but appro- priate for my analysis, with one modification I make below; see Gazdar et al. (1985), among others, for a more detailed treatment. Mother nodes in this way will accumulate fea- tures, so that the highest node in the tree fragment (22), InflP, will be as specified as possible for a given clause. There are four states possible for InflP. It could be fully specified, in which case the clause is not ambiguous as to number or person; it could be either underspecified or undefined (an important distinction), in which case the clause is ambiguous (or possibly general); or it could con- tain contradictory features, in which case the clause is ill—formed. In the first state, InflP could be fully specified because both N and Infl are fully specified. An example of this is given in (23), a representation of The boy is happy. The only features which are lexically induced are those on N and Infl; the others are a result of feature passing. (In the following examples, as in the rest of this paper, I assume that a given Infl contains one of a set of morphemes which specify tense as part of their meaning. I qualify this in the next section, but do not explain the morphologi- cal processes involved.) 25 (23) IP [35:31 i i NP 1' [338] [35g] '__.|__' i—_-J——i Det N' I VP 3 The boy -s be happy But note that InflP can be fully specified even if either (or both: (25) below) of its daughters is underspecified. In The boys laughed, given in (24), the Infl (and, as a result, the Infl') is unspecified for either person or number. (24) IP [3P1] I 1 I NP 1’ [391] i ] Det N' i VP [3911 I ] rT_L_1 (Ade) N [3p1] The bois -ed laugh In The fish swim, given in (25), the NP (through N) is only specified for person, while the Infl' (through Infl) is specified as being any combination other than [3-PERS SING] (5 denotes the non-overt [n3sg] present tense agreement morpheme). 26 (25) IP prll I NP iv [3] 43ng I..__l____‘ r__.l___l Det N' 1 VP [3] -[35g] (Ade) N [3] The fish S sw1m InflP can also be underspecified, resulting in number ambiguity. Recall that (20), The fish swam, had two possi- ble readings. This is predicted by its structure, shown in (26), in which InflP is underspecified due to a lack of number specification on either Infl or N. (26) IP [3] l I l NP 1' [3] [ ] r‘——*———1 r‘—‘L——1 Det N' I VP [3] l 1 (Ade) N [3] The fish —ed swim In the next sub-section, I discuss the case in which InflP is undefined. Finally, the case in which InflP contains contradictory specification can be seen in (27), which represents the ill— formed The boys laughs, since the noun is specified for [PLUR] and the Infl is specified for [SING]. 27 (27) IP *[3sg-pl] r ' I NP 1' ‘ [3P1] I3sg] r—'—'i r——'——1 Det N' I VP [391] [38g] (Ade) N [3?11 The boys -5 laugh §§3. Generation: Infinitival clauses I now return to the issue of undefinedness. Underspec- ification reflects a potential for specification that is unfulfilled, whereas undefinedness reflects a lack of such a potential. For the sake of clarity, I denote undefinedness with the feature [NONE] on relevant undefined nodes. [NONE] is incompatible with any agreement feature specification. The simpler case is that of undefinedness on Infl. I mention above that Infl may contain any of a set of mor— phemes. This set is composed of the bound morphemes {-s, S, -m}, the last of which is used for the [l-PERS SING] form of be, as well as the modals ({will, would, can, could, shall, should, may, might, must...}), and to. It is my claim that to is different from.tjme other morphemes in In“) respects. First of all, it cannot carry tense (see the next section). Second of all, it is always undefined in terms of agreement, whereas the rest of the set is either underspecified or fully specified. More complicated is the case of undefinedness resulting 28 from noun phrases in English. If (as I claim) all and only NPs marked with nominative Case agree with verbs in English (at PF), and both daughters of InflP must be undefined for agreement if either is, then it should follow that all other NPs (i.e., those marked for accusative Case and those un- marked for Case) are undefined for agreement (since they can occur in the subject position only in infinitival clauses). The source of this undefinition, however, cannot be the noun in an accusative NP, since these nouns agree just as much with their determiners as the head nouns in nominative NPs do. I assert, instead, that all NP nodes carry fea— tures, but that a language—specific restriction prevents percolation of these features up from the DH3'U3 its mother except in the event that the NP inserted at D-Structure from the lexicon is marked with nominative Case (this limitation is partially explained in section 1.5). This allows the generation of (28), the dependent clause the woman to be happy. In all of the above cases (23—27), the NPs have carried nominative Case; 111 (28), however, the In) carries accusative Case, and therefore its features cannot percolate to the InflP. 29 (28) IP [NONE-NONE] I l I NP 1' [BSg-Acc] [NONE] Det N' I VP [35g] [NONE] (Ade) N [35g] The woman to be happy The claim that percolation is blocked except when the NP carries nominative Case is supported by the fact that neither VP rmnr PP nodes carry agreement in English, since their heads are undefined and they always take NPs bearing accusative Case or lacking Case (i.e., NP—trace). Hence, in The emperor of'ckumui is Akihito, the agreement features of Japan stop rising below (mienui the agreement features of Akihito stop rising below be, so that only the features of emperor percolate up to the IanP node, as in (29): (29) IP [38g] 1 I I NP 1' [3sg-Nom] [358] r——J———1 r———L——1 Det N‘ I VP [38g] [38g] | I ' I V' N PP F—‘L—_—1 [35g] é v NP ' [BSg-Acc] r-4-1 I P NP N' [3sg-Acc] [35g] The emperor of Japan -s be Akihito The advantage of this analysis is that it allows finite and infinitival clauses to be analyzed with the same struc- 30 ture, with the apparent difference in agreement coming from the incompatibility of undefined (or non—percolated) nodes and (under)specified nodes. On the one hand, the clause in (30), *he to be happy is ill—formed because the features on the nominative NP he percolate up to the InflP node, clash- ing with the undefinedness of Infl. (30) IP *[3sg-NONE] I ' I NP 1' [3sg-NOM] [NONE] I N' I VP [35g] [NONE] He to be happy On the other hand, the clause in (31), *him is happy is ill- formed because the features on the accusative NP cannot percolate up to the InflP node, the resultant undefinedness clashing with the specification of Infl. (31) IP *[ACC-3sg] I ' I NP 1' [3sg-Acc] [358] I I—‘—I N' I VP [35g] [358] iii —s be happy Thus, in infinitival clauses, the subject and the Infl agree in that they must both deny features to the InflP, just as in finite clauses the subject and the Infl agree in that they must both be able to give what features they have to InflP. 31 A final problem: this discussion suggests (as it should) that VP and PP are undefined for agreement features in English. Why, then, is the unification of Infl and VP in (31) not ill—formed, since cnme is specified and time other undefined? A rudimentary solution is that undefinedness is relevant only in two situations: when it passes up to a maximal projection from a head, and when it results from a non-nominative NP. Otherwise, the undefinition is ignored. That heads should receive "special treatment" is supported by, for example, the Head Feature Convention formulated by Gazdar et a1 (1985: 94—9). This discussion. has concentrated (n1 cross-categorial intraclausal agreement in English, specifically subject—verb agreement. It is worth noting that other types of agreement also occur in English, particularly between antecedents and anaphors. LaPointe (1988: III) gives ea more comprehensive list of agreement types in languages than Pullum (1984: see above), and provides discussion. Ultimately, the theory above should account for all types of agreement, but I limit the disCussion fill this paper ti) subject—verb agreement (discussed above) and Case verification (below, and in later chapters). In the next section, I assert that viewing tense as a feature is problematic, supporting my claim above that tense is the result of the meaning of the morphemes in Infl. 32 §4: Tense and infinitival clauses Tense is aa more complex issue than ii: first appears. There are two phenomena which, taken separately, pose no serious problem but which, considered together, cause an apparent anomaly in a system that considers tense to be a syntactic feature. One difference between the two types of clauses is that finite clauses must be inflected for simple tense, while infinitival clauses cannot be, as shown in (32). However, both types of clauses may include the per- fect tense auxiliary have -en (as in (33)). (32) a. It seems John is/was innocent. b. John seems pp be/*beed innocent. (33) a. It seems John was/has been innocent. b. John seems pp be/have been innocent. The first phenomenon relates to the behavior of perfect tense in infinitival clauses. McCawley (1988), as I discuss below, implicitly argues that perfect tense, typically said to be an aspect, is a variant of past tense. Aspect describes the status of the state or action. In (34a), the continuous form indicates that the action is currently taking place; in (34b), the simple form indicates that the action is taking place regularly, although not necessarily at the moment of utterance (judgments may be vague since aspect distinctions are less clearly delineated in English than in other langauges). (34) a. I'm sitting on the lawn. b. He studies mathematics at Harvard. However, as Comrie (1976: 52) asserts, perfect tense is 33 fundamentally different from aspect since "it tells us nothing directly about the situation in itself, but rather relates some state to a preceding situation." That is, perfect tense juxtaposes the timeframes of two states or actions, whereas simple tense makes an assertion only about the timeframe of a single state or action. Perfect tense in a finite matrix clause conversationally implies that the state resultant from the action still persists, whereas simple past does not. Hence, in (35a), it is assumed that the penknife is still missing, whereas in (35b) no assump— tion is made (exx from Comrie 1976: 52). (35) a. I have lost my penknife b. I lost my penknife. As McCawley shows, the perfect tense can be also used in place of simple past when the latter is not available, namely, in infinitival clauses (1988: 216 (17)). Hence, (36a) is ambiguous between (36b), on which reading the perfect tense in (36a) is the result of a lack of availabil- ity of the simple tense in the infinitival clause, and (36c), on which reading the perfect tense iJI'Ume embedded clause is carried over to the infinitival form in (363). This ambiguity can be resolved in favor of (36b) with a time phrase which can only be used with simple past tense (e.g., yesterday), or in favor of (36c) with one that can only be used with perfect tense (e.g., until now). McCaw- ley also claims that (36d), where the two perfect tenses (one as in (36b) and one as in (36c)) collapse into one, is 34 a possible reading of (36a). but clear data in favor of this is difficult to find. (36) a. Eustace seems to have been happy (yesterday/until now). b It seems Eustace was happy (yesterday/?until now). c. It seems Eustace has been happy (*yesterday/until now). d It seems Eustace had been happy. McCawley's analysis is consistent with Comrie's statements, since infinitival clauses are almost always embedded, and so there will always be two states/actions available to be juxtaposed (the state/action of the embedded clause and that of the matrix clause). Another way in which finite and infinitival clauses behave differently with respect to tense relates to the tense of the matrix clause. If the matrix clause is in the simple past, then an embedded finite clause must be in the past tense, but infinitival clauses are not affected by the tense of the matrix clause, so that (37a) is ambiguous between (37b), where the action (H? the embedded clause is simultaneous to the action of the matrix clause (and the past tense morphology of the embedded clause is absolute, i.e., relative to the moment of utterance), and (37c), where the action of the embedded clause occurs prior to the action of the matrix clause (and the past tense morphology of the embedded clause is relative, i.e., relative to the moment of the action in the matrix verb). (The distinction between relative and absolute tense is discussed at length by Comrie (1985).) (37) a. It seemed Eustace was happy. b. Eustace seemed to be happy. 35 c. Eustace seemed to have been happy. The phenomena illustrated in (36) and (37) reveal the most important point of this discussion, since together these data show that treating tense as a feature and sug— gesting that the phenomena are syntactically induced through feature passing leads to an incorrect prediction. McCawley (1988: 216-7) uses (36) to assert that perfect tense in infinitival clauses (with readings as iii (36b)) is the result of the feature [+Past] moving somehow out of the tense-agreement feature complex (i.e., Infl under my analy- sis, as in (38a)). However, if the past tense in the embed- ded clause of (37a) is the result of the feature [+Past] being copied somehow from the matrix clause to the embedded clause (as a syntactic analysis would suggest; see (38b)), then it follows that (37b) is ill-formed, since the Infl of the embedded clause should at some point in the derivation contain the feature [+Past], which it has received from the matrix clause, and this [+Past] should then be converted into a perfect tense (yielding (38c)). (38) a. Eustace PRES seem [(to PAST) be happy.) (cf. (6a)) L-—>have -en b. It PAST seem [Eustace PRES be happy.] (cf. (73)) I >PAST c. Eustace PAST seem [(to ) be happy.] (cf (7b)) 1 >PAST L——>have -en The analysis in (38) predicts, incorrectly, that an accept- able past tense paraphrase of It seems Eustace is happy is Eustace seemed to have been happy, while the valid para- phrase of the former, (37b), is predicted to be ill—formed. 36 The phenomenon revealed by (37b) and (38) could be taken as support for the Cycle, since the movement in (38a) could not take place after that in (38b) since the former occurs completely within the embedded clause and the latter does not. However, this leaves the problem of an Infl containing (to PAST), leaving the PAST element morphologi- cally unrealized at PF (as in (38c')). (38) c'. Eustace seemed to PAST be happy. A better solution is to assume that tense is induced from the semantic meaning of the morpheme in Infl, as I suggested in the previous section. This solution is not without flaws, but it does deal better with the phenomena illustrated in (36) and (37). Below, I give an account of both infinitical and finite embedded clauses. The simpler case is that of infinitival clauses. The morpheme to contains an implication of relative tense, and so the timeframe of its clause is delineated by the form and nature of the matrix verb. To grossly oversimplify and overgeneralize, this means that the state or action of the infinitival clause (without have ~en) is simultaneous to the state or action of the matrix verb. In (39a), it is in the present time (the time of the utterance) that President Bush is lying, since the present time is time time suggested by the matrix verb; in (39b), it is on Christmas that the President is lying, since that is the timeframe of the matrix clause. (39) a. I know President Bush to be lying. 37 b. On Christmas, I knew President Bush to be lying. Stowell (1982: 562-3) shows that certain verbs relativ— ize their embedded infinitival clauses to a future time— frame. In (40), John's locking of the door occurs after John's remembering to do so, although in each case, (40a—c), the time of the action is relevant to the remembering, not to the time of the utteranCe. (40) a. John remembered to lock the door. b. Every day, John remembers to lock the door. c. I'm sure John will remember to lock the door. The infinitival cannot be used in an absolute time sense except in the unusual case of wishful exclamations, as in (41). (41) a. 0 to be in London in the summertime, when the sun is high! b. To have lived in such times as those! Such uses are rare enough to treat them as peripheral cases. The examples in (41) notwithstanding, this analysis accounts for the fact that matrix clauses must always be finite, since matrix clauses must be absolute (with qualifi- cations I do not include here), and infinitival clauses must be relative. The behavior of tense with respect to embedded finite clauses under this analysis is more complicated and involved than is appropriate to discuss herein. To grossly oversim- plify, the semantics of embedded finite clauses are relativ— ized to the timeframe of the matrix clause, just as infini- tival clauses are, but the morphology of Infl of the embed- ded clause is determined absolutely, based on the timeframe of the utterance. Note that if the matrix clause is in the 38 past tense, the embedded clause will necessarily be in the past tense (morphologically), even if the action of an embedded clause which contains a future tense modal has yet to occur in relation to the timeframe of the utterance. This is because postulated worlds exist relative to the time of postulation, not to the time of utterance (assuming the future modals in English express a possible (or at least, an unrealized) rather than a real state). This is illustrated in (42), where in the embedded clause the past tense modal would must be used instead of the present tense modal will, since there is no assertion that the embedded clause is still a belief that Tom holds. (42) a. Tom believed last week pg would make g million next year. b. *Tom believed last week pg will make a million next year. A matrix verb that is in the past tense does not have an affect on infinitival clauses because to cannot be inflected for tense and have -en does not have a absolute past tense meaning (but rather a relative one). The point of this discussion is that the phenomena noted above (illustrated by (35) and (36)) should not be treated as syntactic but rather as a combination of morphol- ogy and semantics. This section has shown that tense cannot be treated as a feature. In the next section I show that (1) the undefin- edness of Infl does not apply to Case, but only to other agreement, and (2) Case may be verified by the VP, instead of by Infl. 39 §5: Neminative Case assignment Following my treatment of agreement above (and sources I have cited above), I assume that Case should also be treated as a feature relationship between a noun phrase and a [+V] element. Case is hence determined as part of lexical insertion (although an element may be underspecified for Case), and it is verified through unification at some level. The following discussion can also apply to theta roles, an issue I shall not discuss until the third chapter. The typical analysis of Case is that there are three terminal nodes which carry a Case feature which needs to be unified with a NP's Case feature (that is, following GB terminology, there are three Case assginers). These are Infl, V, and P. Under my analysis, it is not immediately problematic to account for finite clause phenomena by stipu— lating that Infl carries nominative Case, and that, like other features, this Case unifies with the rmmdnative (or underspecified) Case on the noun phrase. There are two problems with this. The first, noted by Chomsky (1981: 52), is that the three nodes above do not form a natural class, following the quadpartite feature system [iN iV], since both V and P are [+V] and Infl is [—V]. Making Infl a Case assigner was motivated by the distinction between finite and infinitival clause distinc— tions discussed above; by placing the burden for these distinctions on agreement eliminates the need for making 40 Infl a Case assigner. The second problem is that the issue of so-called Exceptional Case Marking, mentioned above, is still not uncontroversially resolved” Recall that infinival clauses can have overt accusative subjects. If Case is just another agreement feature, then the [Acc] feature on the NP should be incompatible with the undefined Infl in such clauses, and ill—formedness should result. Further, there would tme no way for the feature to percolate up and unify with the verb verifying the Case. Hence, on both counts, (42) is ill- formed, as illustrated by the tree in (43): (42) He believes her to be lying. (43) IP [3sg-Nom] I ' I NP 1' [3sg-Nom] [3sg-Nom] ”J“ r l I I VP [3sg—Nom] I VI *[Acc—NONE] I ' I V CP [Acc] [NONE] I CI [NONE] F‘—_L“'—I C IP *[NONE-Acc] F—L—I NP I' [3sg—Acc] [NONE] f—Ip [NONE] 5‘ I He —5 believe her to be ying A less obvious but related problem is lmnv to deal with 41 three-argument verbs, which carry two accusative Case fea— tures. There must be some way of satisfying these one by one, but in the framework offered in section 1.3, it is not obvious how this is to be done. The solution, apparently, is that Case does not behave like the agreement features discussed above. Instead, Case percolates upward until it is unified with another Case, with at least one restriction mentioned below, even if the other agreement features do not "come with it." One reason for this may be simply that agreement features originating on NPs are irrelevant in English unless the NP also carries nominative Case. This resolves (43), yielding (44): (44) IP [3sg—Nom] I ' I NP 1' [BSg-Nom] [3sg-Nom] HT I ' I 1 VP [3sg—Nom] [NONE] I VI [Acc] I 1 I V CP [Ace] [Acc] I CV [Acc] I"_J——I C IP [NONE-Ace] r—Jr—I NP 1' [3sg—Acc] [NONE] I-_'___I 1 VP [NONE] T _. | . He —s believe her to be ying A benefit of this is that it allows the stipulation that 42 nominative Case originates in V3 rmn: in Infl, since nodes carrying Case can unify with nodes that do not (unlike agreement features). This satisfies time natural class problem by eliminating one of the members, Infl, from the list of Case verifiers. This yields the final version of the structure of (42), in (45): (45) IP [BSg-Nom] I l u NP 1' [BSg-Nom] [3sg-Nom] J- L I I VP [3581 [Nom] I VI [Nom,Acc] I ' I V CP [Nom,Acc] [Acc] I CI [Ace] 6 IP [NONE-Acc] NP 1' [3sg-Acc] [NONE] [__i__——1 I VP [NONE] r I He -s believe her to be ying As I noted above, there is at least one restriction to the way in which percolation can occur. It is well-noted in the literature: accusative Case rmmH: be verified (H1 an element that is adjacent at PF (or, possibly, with an inter- vening accusative NP) (Chomsky 1981: 94; Authier 1991: 726). This condition does run: apply ti) nominative Case, Ex) that 43 the only acceptable placement of the adverb quickly in the sentence John gave Mary the book (except for the trivial cases of initial and final placement) is between the subject and the verb, between which nominative Case is unified (as (45a) shows); the other possible placements (H? the adverb (in (45b) and (45c)) violate the adjacency restriction. (45) a. John quickly gave Mary the book. b. *John gave quickly Mary the book. c. *John gave Mary quickly the book. Unfortunately, my analysis offers no simple reason for this restriction, working within the GB framework. This remains a matter for further study. §6: Conclusion lit the introduction ti) this chapter, 1: described the four major superficial differences between finite and infin- itival clauses. In summary, I would like to summarize my treatment of each. (A) Finite clauses require overt subjects. This is a result of time agreement relationship described 5J1 section 1.3, where agreement features necessarily percolate up from a nominative NP and cannot percolate up from any other NP. It is important to note that restricting this behavior to nominative FUR; is language—specific, since certainly there are languages where the verb agrees with its object (e.g., Basque). Section 1.5 provides partial explanation for this restriction. The unification 1J1 finite clauses of I“) and Infl' to InflP results in time simultaneous unification of 44 agreement features and verification of nominative Case. In infinitival clauses with overt subjects, the accusative Case is separated from the agreement features on the subject NP. This allows the stipulation under discussion.tm)'be reduced to a more general form, namely, agreement features can only be unified at a given node if Case is verified. (B) Finite clauses require nominative PM) subjects; infinitival clauses cannot have them. Again, this is a result of subject—verb agreement: agreement takes place if the subject is nominative and the Infl is finite, or if the subject is not nominative and the Infl is not finite. (C) In finite clauses, the subject and the verb agree; in infinitival clauses, they don't. This is not true under my analysis, since infinitival clauses Show vacuous agree- ment. (D) Finite clauses have tense; infinitival clauses don't (in the traditional sense). This is :1 result of morphology and semantics, and not (directly) a syntactic issue. Hence, one difference between finite and infintival clauses is that finite Clauses evidence agreement through (under)specification, while infinitival clauses evidence agreement through undefinedness. This replaces the tradi— tional distinction, that finite Infl assigns nominative Case, whereas non—finite Infl does not, which is ad hoc and unexplained. The other difference between finite and infinitival 45 clauses is timn: some infinitival clauses require overt accusative subject NPs, while others require PRO 2m; their subject. In the next chapter, I show how different matrix verbs behave toward the two types of infinitival clauses. 2. Matrix verb types §0: Introduction This chapter investigates nine verbs which subcatego— rize for clauses, and which behave differently towards these clauses, and shows that the different behavior can be ex— plained using several principles unrelated to the tradition- al distinction between finite and infinitival clauses. In the first section, I discuss the types of verbs on which this thesis is primarily focussed: those which take either infinitival or finite clauses, and which require their infinitival clause complements to have overt accusa- tive subjects or NP—traces (depending on the verb). The verbs discussed here are believe (an ECM verb), seem (a raising verb), and think (which neither carries accusative Case nor raises to subject position). In the second section, I discuss verbs which take infinitival complements only, and which require clauses that lack an overt subject. These are typically called control verbs, and the examples included here are manage, want, and wait. In the third section, I discuss verbs which show both of the above behaviors, and argue that this reflects a difference in senses. The clearest cases are those of expect, hope and convince. 46 47 Before discussing the facts of English, I think it is significant to note that, in German, there are three types of embedded clause, of which a given matrix verb may select only one. The German equivalents of believe and seem pat— tern together, allowing only finite clauses, as in (1) (ex. based on Dippmann 1987: A-l2). (1) Ich glaube, daB er jetzt zu Hause ist. I believe that he now at home is. I think he's at home now. The German equivalent of want, meanwhile, patterns as a modal, occuring with an infinitive form of a verb in the same clause (as other modals, such as must and should, do in English). An example of this is given in (2) (ex. from Dippmann 1987: 65): (2) Ich will reich werden. I want rich to-become. I want to be rich. Finally, the German equivalent of try, for example, takes a dependent clause closest in form to the English infinitival clause, wherein the verb of the embedded clause is of the form zu + infinitive, as shown in (3) (ex. from Dippmann 1987: A—12): (3) Ich versuche, mich auf meine Arbeit zu konzentrieren. I try me on my work to concentrate. I'm trying to concentrate on my work. Since German is so closely related to English, these facts would partially support a claim that there anme as many as three distinct infinitival clauses in English, parallel to those in German, whose structures differ in some as yet unidentified way. However, this is a very strong claim, and 48 one that I do not make here. Instead, I will show that there are two different infinitival clause types: those with non—PRO subjects and those with PRO subjects. The diachronic distinction represented by the difference between (2) and (3) is apparently not extant in English. §l: Clauses with non-PRO subjects §§l..p§;;§1§ The verbs typically called Exceptional Case Marking verbs are so called because they can verify Case on the sub- jects of their embedded clauses, crossing a supposed barrier to Case verification (namely, C') (van Riemdijk and Williams (1986: 235). As the name suggets, they are taken by GB- theorists to represent the atypical case Iinr'matrix verbal behavior in English (while control verbs, discussed in the next section, are the unmarked case) (Chomsky 1981: 66). The typical example of an ECM verb is believe. ECM verbs in their active form require their embedded clauses to have an overt subject, whether this subject be nominative (and agree with the Ian' of the embedded clause), as in (4a), or accusative (contained in an infinitival clause), as in (4b). (I discuss the passive form in the next chapter.) (4) a. I believe gpg is lying. b. I believe pg; to be lying. A related issue, which I do not discuss here, is that the subject of the embedded infinitival clause must be reflexive if it has the same LF referent as the subject, as (5) shows 49 (cf. Soames and Perlmutter 1979: 79—83). (5) Shei believes herselfi/*heri to be innocent. The earliest solution to the phenomenon in (4) was that the subject of the embedded clause was raised to the object position of the matrix clause (Soames and Perlmutter 1979: 82—83; Postal 1974: 25; Authier 1991: 728—30). The major problem with this analysis, considered within the GB frame— work, is that it violates the Projection Principle, since it requires the presence of an object NP position of believe at S—Structure which is not present at D—Structure. The generally accepted solution during the early 1980's was S'-Deletion (Chomsky 1981; called S'-Erasure in Soames and Perlmutter 1979 and S'-Transparency in van Riemsdijk and Williams 1986), which suggested that in certain situations, the 8' node (C' in Chomsky (1986)) of an embedded sentence could be deleted, thus removing the obstacle for Case veri— ficationi ZHowever, following Chomsky (1986), tiUJ; results in the deletion of a head (C, along with C') while its maximal projection (C") it; still present. That such a. structure could exist certainly does not follow from any- thing in the theory (quite the contrary). Under Chomsky (1981), the weakness of S'-deletion is "merely" that it is ad hoc, since nothing iii the theory supports (N? prohibits it; under Chomsky (1986), the weakness of C'—deletion is more severe, since it results in a headless maximal projec- tion. A related possibility, that the entire Comp fragment 50 (C, C', Spec, C") is absent at D—Structure, is presented in Chomsky (1986: 23). ILf the Comp fragment is absent at D- Structure, then verbs such as believe must subcategorize for (and therefore distinguish between) two different types of clauses, while verbs such as manage subcategorize for only one. If the Comp fragment is deleted when it is empty, there must be some process available to perform the dele- tion, but such a process seems unnatural (since a grand- daughter of C", Infl", is retained). On the other hand, a strength of this analysis is that it prohibits that in infinitival clauses (since there is no Spec node to house it), a prohibition borne out by the data. This prohibition, however, is already caused by other principles in the theo— ry, and whether this strength outweighs the problem of retaining the finite/infinitival clause distinction is uncertain. I treat embedded finite and infinitival clauses as both being CP; under my analysis, though, ECM verbs do not pose any particular problem (indeed, they are the unmarked case). The question that Chomsky (1981) implicitly asks is: why do ECM verbs allow overt subjects? II ask, instead, the con- verse: why don't control verbs? I return to the issue of control verbs in section 2.2. It is fairly easy to show that ECM verbs behave the same toward their complement clauses, regardless of the form (infinitival or finite) of this embedded CP clause. The difference between (4a) and (4b) (repeated below) is that in 51 (4a), the nominative Case on the subject of the embedded clause is verified through unification with InflP, which is specified for agreement features due to the inflectional morpheme //-s// in Infl and which is specified for nomina- tive Case (verification) due to the verb be (or, arguably, the VP be lying), as in (6a). In (4b), conversely, the accusative Case on the subject of the embedded clause is verified through unification with time V' mother (H? believe (as in (6b)). (4) a. I believe she is lying. b. I believe her to be lying. (6) a. ... I V! [NOM~ACC] F J I V CP [NOM—ACC] [NONE] I I I Spec C' [NONE] I J I C IP [35g] I I I NP 1' [BSg-NOM] [BSg-NOM] I J I 1 VP [35g] [NONI . believe e 9 she -s be lying 52 b. ... I VI [NOM] I F v Ip [NOM-ACC] [ACC] Spec 0' [ACC] l I; Ip [ACC-NONE] I ' I NP I' [3sg-ACC] [NONE] I ' I 1 VP [NONE] [NOM] . believe 9 e her to be[Iling (A notational point: Cases do not appear on nodes after which verification has taken place; that is, if a [+N] node and a [-N] node are sisters which have the same Case avail— able for verification, this Case does not appear on the mother node, as in the case of the IP in (6a).) (6) accounts for the apparent difference in structure between (4a) and (4b) using agreement features. Of the con- troversial details in (6), some I have already argued for, some I will argue for in coming chapters (especially the unverified Cases, such as NOM on the lower VP). Another detail involves time percolation (H? Case fea— tures. In (6a), the ACC Case continues to percolate upwards past the V' (ultimately to the embedding CP, where it re- mains unverified). This raises several questions, the most important of whidh is: why can't the same thing happen to Cases originating on NPs? That is, why does a Case on an NP 53 with phonetic content have to be verified (or, contrariwise, why doesn't a Case originating on a V or P have to be veri— fied)? This is a situation in which being a head daughter is relevant (see chapter one and sources there). Certain nodes (e.g., V' and P') only allow Cases to pass upwards through them if the Cases originate on a head daughter, while others (e.g., I') allow Cases originating on any daughter to pass upwards. Cases on all other daughters must be verified or else the tree is ill-formed. Hence, in (6a) both NOM and ACC can pass upward past V' because they come from the head daughter, V. On the other hand, replacing believe with seem in (6b) results in an unverified Case on a non-head daughter of V', predicting correctly the ill—formedness of (4b') (see also below): (4) b'. *It seems her to be lying. §§2. seem and think Raising verbs, like ECM verbs, allow either finite or infinitival complements. The difference is that raising verbs, such as seem, only allow AULtrace as the subject of the embedded infintival clause it takes. The trace is left by the movement of the subject NP into the subject position of the matrix verb. This is shown in the contrast between (7a), with a finite embedded clause, and (7b), with an infinitival embedded clause (complete trees given in (8)). (7) a. It seems President Bush is losing popularity.]] 54 b. President Bush seems twp to be losing popularity.]] (8) 3. IP [38g] L I Ni 1' [BSg-NOM] [3sg-NOM] 4‘ I L I I VP [35g] [NOM] | VI [NOM] I I v (IZP [NOM] [NONE] l I Spec l' [NONE] I ‘ ' IP [NONE] I l I NP 1' [3sg-NOM] [BSg—NOM] I J I I VP [35g] [NOM] It -S seem e e President -s be losing Bush popularity 55 (8) b. 1P [35.12,] I F I NP 1' [3sg—NOM] [BSg-NOM] I ' I I VP [35g] [NOM] I VI [NOM] V l I V CP [NOM] [NONE] I ' I Spec C' [NONE] I ' I C IP [38g] I I NP I' [NONE] [NONE] I ' I I VP [NONE] [NOM] President ~s seem e e tupl to be losing Bush1 popularity The reason why this movement is forced is straightfor- ward: raising verbs do not carry any accusative Cases for verification, while ECM verbs do. The reason why this movement is allowed is not so straightforward: on the cnme hand, raising verbs carry nominative Case for \miification (Hi the subject position, but do not carry any theta—roles for this position, so that the chain formed by the overt NP (in the subject position of the matrix raising verb) and its trace is not in violation of theta theory, since at LF it only has one theta—role. On the other hand, the trace in (7b) must be governed (accord- 56 ing to the Empty Category Principle (Chomsky 1981: 250)), but the pair of CP-IP typically act as a barrier to govern— ment (Chomsky 1986) when the IP is finite (Fukushima 1992). If, as this thesis has been suggesting, there is no real syntactic distinction between finite auui infinitival clauses, then it must be either that all CP—IP pairs acts as barriers to government (making (7b) ill-formed, or that there must be some other determination of government of barrierhood and government that prevents certain CP-IP pairs from acting like barriers (as in (7b)). Chomsky (1981: 250) suggests that agreement may also contribute to government and hence (indirectly) to barrierhood, but this is under an analysis of Agreement as a separate syntactic element (as a subnode of Infl) and not as a feature set, as this thesis asserts. I leave unfortunately unsolved the problem of govern- ment in regards to the trace of (7b). For my theory to stand, this problem must be resolved. It should further noted, though, that the whole issue of gevernment seems somewhat incompatible in a feature verification analysis, where features percolate upwards and are unified or not according ti) certain principles rather timni being assigned (through other types (H? government, such an; Case government). Perhaps the solution lies not in an adaption of my own theory but rather in the adaption of the theory of proper government. It is perhaps interesting to note that verbs which are 57 ECM verbs when active are raising verbs when passive, since the agent theta-role is absorbed by the passive form (or diverted to the byhphrase). For instance, the examples in (4) (repeated below) can be passivized as in (9), where be believed patterns just as seem does in (7). (4) a. I believe she is lying. b. I believe her to be lying. (9) a. It is believed that she is lying. b. She is believed to be lying. Theoretically, in addition.tm) the structures exempli- fied in (7), there are two other possibilities: the subject of the embedded clause is raised, while the clause is finite (as in (10a)); the subject of the embedded clause is not raised, and the clause is infinitival (as in (10b)). (10) a. *President Bush seems twp is losing popularity. b. *It seems President Bush to be losing popularity. (10a) is ill—formed because of an agreement mismatch (the NP~trace is undefined for agreement features, assuming that such features are not shared throughout A-chains), while the Ian' carries features). (10b) is ill-formed because of the Case filter, since the Case on the embedded subject, Presi- dent Bush, is never verified. Raising, then, only occurs if and only if the embedded clause is infinitival and the matrix verb carries no theta- role for its subject; according to my analysis, it is the raising that causes the embedded clause to be infinitival, so that the only difference in structure (although not in features) between (7a) and (7b) (and between (9a) and (9b)) 58 is that the subject NP of the matrix clause in (7a) is a pleonast, inserted ti) satisfy syntactic constraints (dis— cussed in chapter 4), while in (7b) it is the subject NP which has been raised from the embedded clause. There are also verbs, such as think, which allow nei— ther raising nor ECM, so that, while (lla) is acceptable, (11b) and (11c) are both ill-formed. (11) a. John thinks he won the lottery. b. *John pg thinks to won the lottery. c. *John thinks pip to win the lottery. In (11), both underlined NPs carry an unverified Case, and thus the examples are ill—formed. Hence, think does not directly prohibit an embedded infinitival clause. Instead, it fails to carry more Cases than theta—roles, and so an embedded infinitival clause is barred by the Case filter. §2: Clauses with PRO as subject §§1. BBQ gg subject Contrary to the current GB—theory, my analysis of infinitival clauses treats the verbs of the previous section as the "unmarked" cases, where the difference between finite and infinitival clauses is based on subject—verb agreement. Certain verbs, however, take infinitival clauses with non— overt subjects. An example is shown in (12), in which try does not allow an overt NP subject in the embedded clause. (12) a. John tried to do his work. b. *John tried him to do his work. c. *John tried he does his work. Since the subject of a finite clause is necessarily overt 59 (or wh-trace or pro, both of which behave as overt NPs in regards to agreement), and since control verbs embedded only clauses with PRO subjects (which are not overt), the embed— ded clause of a so-called control verb such as try must necessarily be infinitival. There are two relevant questions when considering control verbs: (1) why isn't the subject overt? and (2) why don't ECM and raising verbs behave in the same way? One answer to the both questions is that the phenomenon illustrated le (12) is time result of time Avoid Pronoun principle (Chomsky 1981: 65), which selects the non-overt NP PRO over an overt NP whenever possible (it should be noted that this principle is both grossly informal and ad hoc (Fukushima 1992)). The selection of PRO is possible in three instances, of which I discuss two in depth below: when the referent in an embedded clause is determined by the matrix verb (that is, when the subject NP is necessarily co-indexed at LF with some NP in the matrix clause), and when the referent of PRO is uncertain and unneeded to inter— pret the proposition. The third is related to the first, but occurs when control verbs have been nominalized, so that some NP, explicit or implicit, within the matix clause subject determines the subject of the embedded clause. For instance, in (13), since the verb promise coindexes the NP bearing the Agent role to the PRO of its embedded clause, as I discuss below, the noun promise coindexes its semantic agent (which could be implied) to the INK) of the embedded 60 subject. A related example is (14) (Fukushima 1992). (13) a. John promised to clean the car. b. John's promise was to clean the car. (14) John made a promise to everyone. It was to leave. I mention these examples only in passing: they are too complicated to discuss in detail herein. I return now to the first two cases. There is a nota- ble difference between the verbs being discussed in this section and those discussed in the previous one: the infin- itival clause "PRO to X" can often be replaced by the gerun— dive NP "(The act/state of) King" (with a difference of relative tense with certain matrix verbs: see chapter 1, section 4 and Stowell (1982)). For instance, the examples in (15), in which the subject of the embedded clause is PRO and the referent is uncertain, can be paraphrased as in (16): (15) . To shower fully dressed is odd. a b. It is unwise to eat broken glass. c. To know him is to love him. (16) a. (The act of) showering while fully dressed is odd. b. (The act of) eating broken glass is unwise. c. ??The state of knowing him is the state of loving him. c'. ?Knowing him is loving him. Likewise, examples containing certain control verbs, such as in (17), can be paraphrased as in (18) (with a difference of relative tense in the (b) examples): (17) a. That little twerp just tried to look up my dress! b. John remembered to lock the door. c. I don't want my son to watch that type of movie. (18) . That little twerp just tried looking up my dress! a b. John remembered locking the door. c I don't want my son watching that type of movie. On the other hand, ECM and raising verbs never allow such 61 paraphrasing. (19a) has no acceptable paraphrase parallel to those in (15-16) and (l7—18), as shown by the oddity of (19b). (19) a. The monkey seems to like the chocolate. b. *The monkey seems liking the chocolate. This is further evidence for my claim that there are two types of infinitival clauses, and I distinguish these by grouping them into those which take THU) as their subject, and those which don't. However, there is a semantic difference between the examples in (15) and those in (17). In the latter case, the referents of the subjects of the embedded clauses are deter- mined by the nmtrix clause. ID) the former, the referents are unknown (or generic). I return to the issue of control verbs in the following section. In each of the examples in (15), repeated here, the person responsible for the action is unknown, and irrele- vant, since it is the act referred to, not the state of the world in which the act is committed, which is relevant. (15) a. To shower fully dressed is odd. b. It is unwise to eat broken glass. c. To know him is to love him. (15a), for instance, means something like "for any entity capable of showering, it would be odd for that entity to shower while dressed." Some paraphrases (15a) are given in (20a-c). My intui- tions indicate only a slight difference in meaning between the sentences (20a—c), since in.emui1 case the referent of 62 the NP in the fer phrase is not specific. When the referent is specified, on the other hand, there is a very strong difference in meaning, as shown when contrasting (20c) to (20d). (20) a. It would be odd for anyone to shower fully dressed. It would be odd for someone to shower fully dressed. It would be odd for everyone to shower fully dressed. It would be odd for Mary to shower fully dressed. DuOCT The examples in (21a-c) (an ECM matrix verb) and (22a—c) (a control matrix verb), on the other hand, do not Show the same mutual paraphrasing as (20a-c). In each set ((21) and (22)), there is a strong meaning difference between all four examples. (21) a. John would believe anyone to be guilty. b. John would believe someone to be guilty. c. John would believe everyone to be guilty. d. John would believe Mary to be guilty. (22) a. John would want anyone to be guilty. b. John would want someone to be guilty. c. John would want everyone to be guilty. d. John would want Mary to be guilty. This is evidence for my claim that PRO is possible as a subject in (15) because the referent of the subject of the embedded clause is unneeded to interpret the clause. §§2. PRO and control verbs The other instance in which PRO is possible is when the subject of an embedded clause is necessarily co—indexed with some predictable NP of the matrix clause. Which NP in the embedding clause is coindexed to the subject of the embedded clause depends on the matrix verb, but is either the subject 63 or the object of the matrix clause. The simplest example of this is manage, which allows one NP argument, and so it is this argument which is neces— sarily co-indexed at LF with the PRO subject of the embedded clause. In (23), it cannot be anyone other than John who opened the door. (23) John managed to open the door. In contrast, the subject of an infinitival clause embedded by an ECM verb such as believe can either be the same as or different than the subject of the embedding clause, as shown in (24a) and (24b), respectively. (24) a. Johni believed himselfi/*j to be innocent. b. Johni believed him*i/j to be innocent. The embedded clause in cases such as (23) is not structural- ly different from those in cases such as (24). The subject of the embedded clause is necessarily PRO in (23). Further, PRO is either undefined for agreement or it isn't (i.e., it's underspecified for agreement). If PRO is undefined for agreement, control verbs cannot embed finite clauses because doing EK)‘WOU1d result in.ze subject-verb mismatch. On the other hand, if PRO is merely underspecified for agreement, then by my analysis all control verbs must embed only finite clauses. Since some control verbs (namely, that in (23)) embed infinitival clauses, it follows that PRO must be undefined for agreement, for which it follows that all control verbs must embed only infinitival clauses. This argumentation is circular: control verbs embed only infini- 64 tival clauses iff PRO is undefined for agreement, and PRO is undefined for agreement iff control verbs embed only infini— tival clauses. One problem with this analysis that has been raised is that of (25) (Fukushima 1992): (25) John managed PRO to embarass himself/*themselves. If PRO is undefined for agreement, then how can it determine the form of a reflexive with which it is coreferential? 0n the other lmnui, if PRO it; lexically underspecified for agreement, then it can inherit its features from the matrix subject anui verify' these on ti) the reflexive ii: governs. That is, all the NPs which are coindexed at LP in (25), as shown in (25'), share the same agreement features. (25') Johni managed PR01 to embarass himselfi/*themselve51. Hence, the assumption that PRO is underspecifed but defined for agreement is not problematic for (25), while the oppo— site assumption (that PRO is undefined for agreement) is. This is an issue which must ultimately be addressed, but it is not as problematic for my theory as it first appears, since the issue (H? agreement ii) (23) involves satisfying the syntax at S-Structure and the issue in (25) involves satisfying the coindexing at LF. It may be the case that these two types of agreement are different in some fundamental way, in which case (25) is irrelevant. But I will not elaborate on this point further, for any other comments that would not go far afield from the main thesis could only be ad hoc. 65 Another example of a control verb is want, which takes two NP arguments and which coindexes its object to the subject PRO of the embedded clause. This is shown in (26), in which it is Buchanan, not Bush, whose dropping out is desired. (26) Bush wants Buchanan 1p[PR0 to drop out of the race.] In (26), I assert, the NP Buchanan is outside of the embed- ded clause, whereas in (243), the DH’.himself is within the embedded clause. This is contrary to Chomsky (1981: 99-100; 1986: 86—87), who claims that both NPs are within the embed- ded InflP. His evidence for this claim is weak at best (as he himself admits, it is based on small distinctions that I feel can be explained in unrelated ways, and on intuitions of well-formedness with which I do not fully agree). Chomsky's claim also comes at the cost of maintaining an infinitival/finite clause distinction, since want must subcategorize only for infinitival clauses. I present my argument for this using his theory and terminology: Consid- er the data in (27). In (27a), PRO is the embedded subject, and does not receive Case (since PRO must be ungoverned). Therefore, want has a lexical entry which does not assign accusative Case. (26) has shown that the LF subject of the embedded clause need not be the same as that of the matrix clause. So why, then, is (27b) ill—formed? (27) a. Bill wants PRO to win. b. *Bill wants John wins. There are no unassigned Cases, and no NPs without Case. 66 There are two possibilities: there is some abstract marker that distinguishes finite and infinitival clauses (perhaps the commonly used [iTense]), or want subcategorizes for an InflP, and a finite InflP node is necessarily the daughter of C'. This thesis is explicitly arguing against the first possibility, and the second is unmotivated. My analysis, on the other hand, naturally explains the extreme cases such 2m; (27), while leaving unexplained the minor intuitive distinctions given in Chomsky (1981; 1986). Having explained (26), I now explain (27a) through a vari— ant, given as (28). In (28), as in (27a), there is no overt object NP in the matrix clause, making the example appear to be a counter—example to my analysis. (28) Buchanan doesn't want to drop out of the race. However, it is not the case that the object of the matrix verb in (28) is structurally absent. There are certain transitive verbs in English (shave, dress, wash) which take a non—overt reflexive, following the Avoid Pronoun principle (since time theme of time action is normally the agent, or some part of the agent). The reflex- ive is overt when it is conjoined with another NP. Hence, in (29a), the reflexive is not overt because it is recovera- ble: its non—overtness does not reduce the semantic content of the clause, and therefore the Avoid Pronoun principle predicts that the reflexive need not be overt (unless other syntactic restrictions force it to be overt). In (29b), it is overt because it is morphologically required to be so (a 67 non-overt NP cannot be conjoined with an overt one). (29) a. John bathed (??himself). b. John bathed himself/*6 and his children. Other evidence for the presence of this non-overt reflexive is provided by Dutch. In Dutch, there are two overt reflex- ives, of which zich patterns like the non-overt English reflexive, and zichzelf’patterns like the overt one (van der Leek 1991); examples of this are in (30). (30) a. Jan kleedde zich aan. John clothed himself on. John dressed (??himself). b. Jan kleedde eerst zichzelf en daarna the kinderen aan. John clothed first himself and then the children on. John dressed himself/*6 first, then the children. Just as the overt relexive is required in (29b), so it is in (31), causing the non-overt reflexive object of want overt in a conjoined NP. (31) John wants Mary and himself/*6 to carry the piano. Therefore, I assert, the subject of the embedded clause in (28) is co—indexed with the non-overt reflexive, which is in turn co—indexed with the subject of the matrix clause. A final example of a control verb is wait, which has two senses. In one of these, it behaves like manage, as illustrated in (32a). In the other, it behaves like want, with the difference that for is required to verify the Case on the object NP, as illustrated in (32b) (for is presumably required because wait carries two theta roles but only one Case, NOM). (32) a. I waited to close the door. b. I waited for/*6 John to close the door. Syntactically, (323) looks like an instance of a non—overt 68 reflexive, but it is not possible to combine (32) into one sentence with a conjoined object NP, as in (33), since (33) may be bad on semantic, not syntactic, grounds. (33) *I waited for John and myself/me/o to close the door. However, there is a difference in senses between (32a) and (32b) in that in (32a) there is some other action which is being awaited, whereas in (32b) the awaited action is the closing of the door. The awaited action in (32a) can be specified, as in (34a); in (34b), meanwhile, it is still John's closing of the door that is being awaited. (34) a. I waited until Mary arrived to close the door. b. ?I waited until Mary arrived for John to close the door. In both senses of wait, though, the referent of PRO of the embedded clause is always predictable (it is the subject of the matrix clause in (32a) and the object of the matrix clause in (32b)). As I show in the next section, there are other verbs which have two senses, of which one selects an embedded clause with PRO as the subject, and the other selects an embedded clause with a non—PRO subject. §3: Clauses with either The discussion in this chapter has suggested that a given verb on a given sense will either select an embedded clause with a non-overt PRO subject (which must be infiniti- val for reasons of agreement), or an embedded clause with a non-PRO subject (which can be either finite or infinitival). 69 There are numerous apparent counterexamples to this claim. However, I intend to show in this section that these are not actual counterexamples, but rather pairs of verbs with homophonous forms. The verb expect, for instance, can embed either a finite clause, (35a), an infinitival clause with an cwert subject, (35b), or an infinitival clause with a PRO subject, (35c). (35) a. I expect that he will finish by tomorrow. b. I expect him to finish by tomorrow. c. I expect to finish by tomorrow. However, there is a meaning difference between (35a) and (35b), in that (35a) is a) general epistemic prediction, whereas (35b) can also Ime used deontically, suggesting an obligation (for a discussion of deontic vs. epistemic modal— ity, see, e.g., Palmer 1986: 121—125). This is clear in (36), in which the speaker is not making a prediction, but rather indirectly ordering time addressee, ems indicated by the response. (36) A: I expect you to do your own homework from now on. B: Yes, sir. Hence, (35b) is structurally ambiguous between (373), which yields the same reading as (35a), and (37b), whose reading is parallel to (36). (37) a. I expect 1p[him to finish by tomorrow]. b. I expect him 1P[PR0 to finish by tomorrow]. This analysis asserts that (35c) is also an indirect order, a self-inflected one, even though it seems like a pmedic- tion. But it seems reasonable to assume that self-inflected 70 orders will indeed be predictions, since it should be safe to assume (from the standpoint of Grice (1975)) that a person will obey their own orders. Taken out of context, the statement I'll come in the front gate is a prediction, but iflta context such as ill (38), it patterns like the other, parallel sentences, which are indirect orders. (38) We have to be organized. Smith, you'll go in the second story window. Johnson, you'll take the back door. Walker, you'll watch the side door. Jones, you'll stay in the car and keep it running. And me, I'll come in the front gate. For this reason, I claim that (35c) does indeed represent the same sense of expect shown in (37b), yielding the struc— ture in (39). (39) I expect REFL 1p[PR0 to finish by tomorrow]. The reflexive which is not overt ii) (39) is overt (albeit awkwardly) when the object NP is conjoined, as in (36). (40) I expect you and ?me/?myself/*¢ to finish by tomorrow. This supports the analysis in (39). Another example of a verb with multiple senses is hope, which embeds either a finite clause, (41a), or an infiniti— val clause with a PRO subject, (41b). Since hope does not verify accusative Case, it does not allow an infinitival clause with an overt subject, as shown by the ill-formedness of (41c). (41) a. I hope I finish/he finishes this paper soon. b. I hope to finish this paper soon. c. *I hope him to finish this paper soon. In (41a), hope implies nothing about time speaker's actual expectations, but rather only indicates a desire; (41a) can 71 be paraphrased using want, as in (42), while (41b) cannot: (42) I want (him) to finish this paper soon. In (41b), hope implies that the speaker is confident that the outcome (the content of the embedded clause) will occur. (41b) can be paraphrased using expect, as in (43), while (41a) cannot: (43) I expect to finish this paper soon. Hence, hope, like expect has two senses. A final example is thexmni) convince, which behaves similarly to hope, as illustrated in (44). (44) a. I convinced Mary that she likes Bill. b. I convinced Mary to like Bill. 0. *I convinced Mary her to do it. Here, the meaning difference is that in (44a), the speaker is convincing Mary of a fact, and any fact can be in the embedded clause, as shown in (45). (45) convinced Mary that the earth was round. a: I convinced Mary that she was wrong. c. I convinced Mary that Bill was cheating on her. However, in (44b), the speaker has convinced Mary to perform some action, and so the subject of the embedded clause must necessarily be Mary, the performer of the action. None of the examples in (45) can be paraphrased using infinitivals; replacing the finite embedded clauses with infinitival ones yield ill-formed sentences, except in the case of (45b), which yields (46). However, (46) is pragmatically odd and, at any rate, semantically much different from (45b). (46) I convinced Mary to be wrong. In this section, I have demonstrated that apparent 72 counterexamples to my analysis are merely cases of homopho- nous verb forms with different senses. In this chapter, I have dealt with the major types of verbs that take embedded finite and infinitival clauses. In the next chapter, I deal with the claim that all Case as- signment is optional, a claim which has been implicit in the discussion up to this point. I show that this claim leads not only to my analysis of agreement and Case features already given, but to a simplification of the lexicon in general. 3. Case verification and the lexicon §0: Introduction The analysis presented in chapter 1, section 5 is dependent upon the stipulation that nominative Case origi- nating on a [-N] element need not verify a nominative Case originating on a [+N] element (since the nominative Case originating (M1 the \l of an) infinitival clause :is unverified). However, it has been a tacit assumption in GB theory that Cases available for verification are, at least generally, obligatorily verified (hence time obligatoriness of it—insertion as in Chomsky 1981: 67; see also Authier 1991: 726, where the assumption is stated but not defended). This assumption is not forced by the Case Filter, which only bars unverified Cases originating on [+N] nodes (Chom— sky (1981: 49) states the Case filter as: "*NP, where NP has phonetic content and has no Case"). If this assumption is to be overridden, then the data in support of it must be explained in other ways, a task which I begin in the next chapter. In this chapter, however, I show that allowing [—N] Case to remain unverified allows for a simplification of the lexicon, since forcing [-N] Case to lme verified causes redundancy. In the first section, I summarize theta theory and its relevance to Case verification. I illustrate with two- 73 74 argument verbs such am; eat that requiring both theta role verification and [—N] Case verification causes redundancy. In the second section, I discuss three-argument verbs, assuming that verbs such as give have two lexical entries. While it is ideal to assume that such verbs have one lexical entry (Larson 1988: 369), such an assumption causes diffi- culties for my analysis (which I discuss in chapter 4). In the third section, I return to the issue of believe and other ECM verbs, showing that believe need only have one lexical entry (instead of the two or three usually given to it). I will thus show that the examples in (1) can be reduced to one entry. (1) I believe that Mary is lying. I believe Mary to be lying. I believe Mary. I believe. QOO‘CD Throughout this chapter, I will assume for the sake of convenience that verbs select theta role configurations (examples below) from 21 subcategorized list (H? possibili- ties. In general, a clause has as many NP positions as it has theta roles, so that the theta role configuration of a verb (along with prepositions, which carry one theta role) will determine the clause structure as regards the number of noun phrases. I discuss some exceptions to this in the next chapter. However, it may be possible that the possibilities are predictable. For example, 2) passive verb nmur select a configuration which either contains an Agent (in a by- 75 phrase) or not. Intuitively, it seems inappropriate to give two lexical entries for passives, one of which specifies for an Agent and one of which does not. On the other hand, Dowty (1989) offers some examples and discussion which argue against the general predictablility of which arguments can be deleted based on theta roles. In any case, it is irrele— vant to the main thesis of this paper whether theta—role omission/deletion is the result of general linguistic prin— ciples or of lexical subcategorization, and so I mention the issue here only in passing. §l: One- and two—argument verbs There are three syntactic relationships that may hold between a [—N] and a [+N] element: agreement, Case, and theta roles. Agreement limits the morphological form of both elements, as discussed above. Case parmits the phonet— ic realization of a NP, licensing the presence of an overt NP. Theta roles determine the semantic role of an NP as an argument of a V or P predicate (although, as I argue, not all arguments of a predicate must receive theta roles). Of these three relationships, I have already discussed agreement, and the latter two are of interest to the present discussion. In a given sentence, the predicates must carry (in sum) at least the number of Cases and theta roles suffi— cient to verify the Case and theta roles of the overt NPs in that clause. If there are four NPs in a given clause, there must be (at least) four Cases and four theta roles available 76 on the [-N] elements in that clause. For instance, in (2a), sleep carries a nominative Case and an Agent theta role, both of which are ultimately unified with the Case and theta role on the NP John (at InflP). In (2b), the verb eat carries two Cases (nominative and accusative) and two theta roles (Agent and Theme), which are verified on the NPs John and spaghetti. (2) a. John slept. b. John eats spaghetti. For expository reasons, I use a simplified notation in given the relevant portions of lexical entries, listing a predicate in three lines. (h) the first line is time ortho— graphic form of the infinitive; on the second, a listing of Cases it carries for verification; on the third, a listing of theta roles it carries. The verbs in (2) are listed as in (3), respectively. (I make no claim that the theta roles given throughout this discussion are completely accurate.) (3) a. sleep C: Nom 0: Agent b. eat 0: Nom Acc 0: Agent Theme In current GB theory, both Case and theta role assignment (verification, under my analysis) are necessary, making (4a) ill—formed if (3b), rather than (4b), is given as the lexi- cal entry for eat. (4) a. John eats. b.eat' CzNOm 6: Agent This requires multiple homophonous entries almost as prolif- 77 ic as those caused by requiring agreement features to be fully specified (see above and Gazdar et al. 1985), as I Show in the next section. Is there a way to simplify (3b) and (4b)? Arguably, the argument structure of eat in (2b) and in (4a) is the same, since in (4a), it is assumed that John eats (an inde— terminate) something. This is not true of other intransi- tive verbs; surely it is nonsensical to claim that in (2a), John slept (at? about?) something, and in (5a), there is no implication that John was looking at anything specific, even though (5b) and (5c) are similar in meaning. (5) a. John looked, (but he didn't see anything.) b. John looked at the girl. c. John saw the girl. It seems reasonable, then, ti) assume that time semantic content of the entries of eat in (3b) and (4b) are the same. It can be easily (even trivially) demonstrated that the phonetic content of the entries are the same. The only difference is in the subcategorization frame. It is my goal to show that (3b) and (4b) can be reduced into one entry, and that the behavior exhibited by eat in (2b) and (4a) can be reduced to general principles. One set of principles relates to Case verification. On the assumption that Case verification is optional, the entry for intransitive eat in (4) can be given as (6): (6) eat' C: Nom Acc 6: Agent Now, as far as Case is concerned, the difference between 78 (2b) and (4a) is not in the number of Cases available for verification but rather in the number of Cases actually verified: that is, in the number of overt NPs. I will discuss the optionality of theta—role verifica— tion in the fourth section of this chapter. However, for now, I adopt the notation in (7) to collapse (3b) and (6) into one entry; the parentheses indicate that the theta—role may be selected or not, allowing eat to satisfy either of the configurations in (7) at D—Structure. (7) eat C: Nom Acc 6: Agent (Theme) (8) 3. IP I ' I NP 1' I VP 1 l V l b IP J I NP 1' a. I VP F'__L_—_1 V NP I. I I1) Regardless of which configuration eat selects, Case verifi- cation will be satisfied only if Case originating on [-N] elements is able to remain unverified. One superficial problem with (6) is that it appears to allow (9), with a pleonastic it, since pleonasts are assumed in the literature (see in the next chapter) to be inserted to absorb unassigned Cases (Fukushima 1992). 79 (9) *John eats it. However, (9) is ill—formed because there is simply no place to put the pleonast, since the theta role configuration only allows one NP position, the subject position (see further discussion in chapter four, section one). That is, when eat carries one theta role and two Cases, the structure in (Be), not (8b), is selected, and there is no possible location for the pleonast. This solution seems only slightly worthwhile in the case of two-argument verbs, such as eat, since it only reduces two entries to one. However, lexical entries pro- liferate with verbs taking more than two arguments, unless Case verification is optional. I show this in the next Section. §2: Three—argument verbs In this section, I illustrate that requiring a separate lexical entry tint each theta—structure specification (H? a verb requires an unneeded proliferation of entries for verbs with three or more arguments. The verb give requires three arguments: an entity (or entities) who gives an object, a recipient of the object, and an object being given. In active declarative sentences with Inn) overt objects, such 2m; (10), these arguments are represented by the theta roles Agent, Recipient, and Theme, respectively. (10) The Olympic Committee gave the winner a gold medal. 80 [Agent ] [Recipient] [Theme ] However, just as with eat, not all three arguments must be overt. For instance, in (11a), there is an implied object being given, while in (11b), there is an implied recipient. (11) a. I gave to charity. b. I gave blood. With eat, the implied object is necessarily something which is expected to be eaten (saying This man was eating without qualification when what he is eating is an automobile is pragmatically inappropriate). With give, the implied object is necessarily something which the recipient wants or appre— ciates having; the implied recipient is the entity that the object is given to (saying (11b) when the recipient is, for instance, my sister, is pragmatically odd. It may be sig— nificant to note that the examples in (11) seem somewhat idiomatic, whereas John eats is completely standard; howev— er, I will not elaborate on this. In the case of passives, the agent may be optionally omitted, with no idiomatic reading whatsoever (as in (12)); this is generally true for all passive constructions. (12) a. The book was given to Mary (by John) b. Mary was given the book (by John) Further, it is possible to have only one argument overt. For example, in (13a,b), only the agent is overt, while in (13c), only the object (theme) is overt. (13) a. My girlfriend and I have a give—and—take relationship. I give and she takes. b. (Mother to children) I give, and give, and give, and what do you ever give to me in return? c. Several gallons of blood were given at last week's drive. In the active form of the verb alone, there are five possi— 81 ble subcategroization frames, given in (14): (14) a. John gave Mary a book. (__ NP NP) D. John gave. (__ ) c. John gave blood. (__ NP) d. John gave a book to Mary. (__ NP PP) e. John gave to charity. (__ PP) However, if we assume that Case assignment is optional, (14) can be reduced to two lexical entries, given in (15). (15a) abbreviates (14a—c); (15b) abbreviates (l4c—e). (15) a. givei C: Nom Acc ACC 0: Agent ((Recipient) Theme) ? b. givez C: Nom Acc 6: Agent (Theme) (Recipient) It has been suggested (Larson 1988) that it is better to assign one lexical entry to double object verbs, such as give, and to account for the phenomena differently; however, doing so is problematic for my account, as I discuss in the next chapter. It should be noted, however, that Larson's account is typical in that it does not discuss cases such as (l4b,c,e), so that Larson accounts for the derivation of (14a) from (14d), but still requires at least three entries (one with no object, one with one object, and one with two objects). More complicated is the issue of ECM verbs, which are discussed in the next section. §3: ECM verbs Under the more traditional analysis in GB, ECM verbs require two entries, regardless of whether or not clauses 82 require Case. An ECM verb, such as believe, may be followed by a finite clause (as in (16a)), an infinitival clause (as in (16b)), or an object NP (as in (16c)). (16) a. I believe she's innocent. b. I believe her to be innocent. c. I believe her. Assuming that clauses do not require Case, the respective lexical entries of the matrix verbs in (16), without apply— ing my analysis, are roughly as in (17). (17) a. believe C: Nom 0: Agent Theme b. believe C: Nom Acc 6: Agent Theme c. believe C: Nom Ace 6: Agent Theme Notice that (17b) and (17c) are already the same. ILf Case is optionally unverified when originating (n1.a verb, then (17b) is also acceptable for (16a), thus reducing the en— tries in (17) to one. Assuming that clauses require Case, the repective entries for (16), without applying my analysis, are roughly as in (18). (18) a. believe C: Nom Acc 6: Agent Theme b. believe C: Nom Acc Acc 6: Agent Theme c. believe 0: Nom Acc 0: Agent Theme Here, (18a) and (180) are identical, but again (18b), under my assumption about Case verification, may serve as the 83 entry for all three examples in (16). Since my analysis accounts for both assumptions (that clauses require Case, and that they do not), I will not stipulate here which assumption is more appropriate, since this is far from an uncontroversial issue. On the one hand, since embedded clauses often behave like DUE; (for example, finite clauses require theta roles; see Chomsky (1986: 14)), and appear in the same positions as NPs (compare, for instance, (16a) and (16c)), it would seem that they should have Case. On the other hand, they may also appear in positions which do not allow NPs, especially in the object position of passive verbs. While the movement of the NP in (19) is required, the parallel movement is only optional in (20), since the object in (20a) is a clause. (19) a. *It was hit the man with the two burly arms. b. The man with the two burly arms was hit. (20) a. It is believed by everyone that John is happy. b. That John is happy is believed by everyone. I prefer the analysis in (17) on intuitive grounds, but can provide no theoretical defense of it, except for the data in (20). §4: Conclusion The discussion in this chapter has concerned the op— tionality of Case verification. There are several strengths to this claim, and one major theoretical weakness. One strength of this is that it allows for the removal (H? Case from time finite/infinitival clause: distinction 84 Furthermore, it simplifies the lexicon and explains the quandary mentioned by Chomsky (1981) concerning the natural class of Case "assigners". This natural class, [-Noun] (Verb and Preposition), is the same as the class that veri— fies theta—roles. The weakness of my analysis is that it differentiates Case from agreement features and theta roles in that both of the latter appear to be obligatorily verified (in the case of agreement features, to the extent that they are speci- fied). It seems ultimately desireable to formulate a theory in which all such relationships behave in a uniform manner. It is uncertain how important this latter issue is. In general, every simplification of one facet of a linguistic theory is apt to make another facet more complex (just as it is in language, where diachronic simplification of, for instance, the phonology of a language will make its morphol— ogy more complex). I assert that the benefits of my analysis outweigh the costs, at least those mentioned above (in this chapter and elsewhere). In the next chapter, however, I discuss some more problematic issues which have been accounted for some— what straightforwardly in standard GB theory but which I have difficulty accounting for with my analysis. 4. Related issues and problems §0: Introduction In this chapter, I discuss two important issues related to my analysis. The analyses presented in this chapter are not as developed as time assertions already discussed, and the purpose of this chapter is to discuss, but not neces— sarily provide definitive analyses, of the issues presented herein. The first section deals with.ze phenomenon related to the Case filter, namely pleonast insertion and passive movement. Pleonast insertion, following the current GB model, is caused by an NP position which receives a Case but no theta role. This analysis relies crucially on the asser- tion that the pleonast it in (1), absent at D-Structure, fills no other syntactic or semantic purpose. (1) It seems that John is happy. However, my analysis explains (1), since the pleonast is required to carry agreement features for unification with those on the Infl. What is not obvious is how to explain (2), since it is in object position and thus does not carry agreement features (see chapter 2, section 2—2). (2) I want it to seem like an accident. I argue that the pleonast is necessary in (2) for semantic/pragmatic reasons. 85 86 Another issue is that of the passive transformation when considering double—object verbs. The passive transfor— mation is normally caused by the Case Filter, because the NP Napoleon in (3a) carries an unverified Case in its D—Struc— ture position, and must therefore move into subject posi- tion, as in (3b). (3) a. *was beaten Napoleon at Waterloo. b. Napoleon was beaten tn? at Waterloo. However, if double object verbs such as give are given only one lexical entry, then (4) should be well-formed, since the verb has two accusative Cases available, one of which is absorbed by the passive, the other of which is verified on the NP a monkey. The pleonast is inserted for agreement reasons (Fukushima 1992). (4) *It was given a banana to the monkey. I offer three possible solutions, all of which unfortunately seem ad hoc. §1: The pleonastic construction The pleonastic it (as opposed to the homophonous refer- ential pronoun) appears to have no semantic function and therefore occurs in chains lacking a theta-role. Under GB- Theory, this entails that the purpose of the pleonast is to absorb an otherwise unassigned Case. For instance, in (5a), the D-Structure of (5b), there are two lnfls which assign Nominative Case, -s and —ed, since (under the GB—analysis) these are finite. Since there is only one theta—role as- 87 signed (by be wrong), a pleonast must be inserted to ab- sorbed the unassigned Case. (5) a. e -s seem [John —ed be wrong] b. It seems John was wrong. In (5b), then, it has nominative Case and no theta—role, while John has nominative Case and an Agent theta-role. In (6), by contrast, pleonast insertion is not possible because a lack of Case assigned by the infinitival Infl of the embedded clause forces the subject of this clause, John, to raise to the subject position of the matrix clause, which NP accepts the nominative Case assigned by —s. Since there are no unassigned Cases, pleonastic insertion is blocked. (6) a. e -5 seem [John to be wrong] b. John seems t to be wrong. This analysis produces the desired results, but it depends crucially on the assumption that Case need be assigned. Otherwise, why is the pleonast necessary in (5b)? A first hypothesis, which I amend below, is that my analysis forces the presence of the pleonast in (5) not by making it the carrier of Case, as the above suggests, but rather by making it the carrier of subject-verb agreement. Since the Infls of the matrix clauses of both (5) and (6) carry agreement (being —s), there must be an appropriate NP in subject position (appropriate in English including overt NPs and wh-traces). The question might be raised, why is (7) ill—formed? As regards agreement features, ii: is acceptabLe since both the subject NP position and time Infl are undefined for 88 agreement. (7) *To seem John is wrong. Recall, however, that this has already been ruled out above on semantic/pragmatic grounds, since to can only occur under relative tense, while a matrix clause must occur under absolute tense, except in the rare case of hopeful exclama- tions such as "O to be in London in the springtime, when the pussywillows are in bloom!" (see chapter 1, section 4). The implied subject of such exclamations is the speaker, and so (7) would be possible only if (7') were acceptable, but (7') is blocked because the subject of the matrix clause carries an unverified theta—role. (7') *I seem John is wrong. Note, further, that (8a) is pmesible, albeit odd, because (8b) is possible, although (8b') and, hence, (8a') are ill— formed, even though (8c) is well-formed, supporting my suggestion that enmii wishful exclamations have EH1 implied (or, possibly, deleted) first—person subject. The pleonast it is inserted, however, in positions where there is no such subject. (8) O to seem to be innocent! I a. b. seem to be innocent. (8') a. *0 to seem I am innocent! b. *I seem I an innocent. c. It seems I am innocent. One question remains, and that is, what forces the pleonast to occur when agreement is not necessary, i.e., when a clause with a main verb such as seem is infinitival 89 and embedded in another clause, as in (9) (where (9') is ill-formed only on the relevant reading): (9) I want it to seem that John likes pudding. (9') *I want to seem that John likes pudding. The standard.(H3 analysis has In) immediate difficulty with the data in (9), since there is an accusative Case on the matrix verb which must be assigned. However, this is not an acceptable solution under my analysis, since Cases may remain unverified. There are at least two other tacks to take. One is that it really does receive some sort of theta—role (in (9), from either the embedded or the matrix verb), but that this theta-role may tme absorbed iJHX) chains with. other theta— roles. That is, calling its theta—role WorldState, any NP chain may verify a WorldState theta—role even if it carries a different theta—role. In (10), the chain formed by John and the NP trace carries the Agent theta-role, and verifies the Agent theta—role on the embedded verb, like, and the WorldState theta-role on the matrix verb, seem. (10) John seems tNP to like pudding. Following this through, Control verbs such as want might carry the WorldState theta—role for their object NPs, which eventually inherit (in some way) the theta—roles of the PRO they are conidexed with at 1J3 For instance, in (9), the pleonast verifies the WorldState role carried by went, the PRO in subject position of the embedded clause verifies the WorldState role carried by seem, and the pleonast inherits 90 the WorldState role from PRO (a trivial step in this case). The related tack is that all NP positions in an S— structure must be filled with one of a set of NP types, which set does not include e. PRO and overt NPs are lexi— cally inserted; traces are causes by the various movements of these; finally, a pleonast is inserted into all NP posi- tions that have not yet been filled. This seems to be different from the analysis in the above paragraph, but it is similar in that it follows the assumption that the number of NPs required in a given clause is determined by the number of theta—roles carried by all time verbs and prepositions in the clause, with two possible exceptions in English: (1) there is always a subject NP position; (2) in double—object constructions the Goal theta—role on the verb may be verified by the same NP that verifies the Goal theta— role on the preposition to (see Larson 1989 for a related View). I argue against the second exception in the next sec— tion as overly complicating. The first exception has the complications that it requires an analysis of wishful excla- mations in which the subject NP is deleted (rather than im— plied). A stength of this latter tack is that it maintains the one—to-one verification of theta—roles between NPs and [—N] nodes, although it is not immediately obvious that this one— to—one relationship need hold. Nonetheless, I favor the second tack, and claim that pleonast insertion is the result 91 of filling NP nodes which are not occupied by a member of a chain at S-Structure. The second exception given above is related to a prob— lem involved with the passives of double object verbs under a theory in which Case verification is optional. I discuss this in the next section. §2: The passive construction Before dealing with the problem caused by the combina- tion of my analysis with the passives of double object verbs, I would first like to demonstrate the passives of single object verbs do not pose a particular problem for my theory. Under the current GB framework, the passive forms of verbs absorb one accusative Case, while the associated VP does not assign an Agent theta role. For instance, in (11), the verb like assigns a Source theta role and accusative Case to the object NP Mary, the VP like NPx assigns an Agent theta role ti) the subject In) John, and time finite Infl assigns nominative Case to the subject NP, giving a well— formed clause where each NP has Case (and a theta—role). (11) John likes Mary. In (12), on the other hand, the passive participle absorbs the accusative Case, and the VP was liked X does not assign any theta role, so that in D~structure position, the NP Mary does not have Case and, according to the Case filter, (123) is ill-formed” If the DH’ is moved ti) subject position, 92 however, it may receive the Case assigned by the finite Infl, allowing (12b) ti) be well-formed (on time further stipulation that the moved NP governs its trace). (12) a. *(It) was liked Mary. b. Mary is liked twp. The point is, nowhere does this analysis depend on all available Cases being assigned, but rather on all overt NPs receiving Case (i.e. the Case Filter). Since my analysis does not abandon the Case Filter, the passive transformation is not problematic, since the passive participle absorbs the only accusative Case available for verification (by my terminology), and the DH’.Mary is forced ti) move ti) verify its Case, just as in (12). I Should point out, nonetheless, that the passive transformation as it currently stands is hardly uncontrover— sial. It is derived on a more complex and context—specific transformation from early Transformational Grammar, roughly as in (13a), exemplified by (13b) (13) a. NP1 X V Y NP2 Z —-> NP2 X be V—ed Y Z by NPi b. John [3s-pres] like a Mary o —-> Mary [BS-pres] be like-ed ¢ p by John The PP by John may optionally be deleted (cf. Borsley 1991 113 for a simplified example). There are, of course, prob— lems with (13a), but these are not important since the analysis 1mm; been abandoned. The motivation for (13), though, is to allow verbs to have a single lexical entry for both active and passive forms, since all passives are de— rived from their active forms in the same way (Soames and 93 Perlmutter 1979: 30—34). However, the GB approach implies (but doesn't necessitate) that timnme are separate lexical entries for the active and the passive participle forms of a verb, and so the claim that active and passive forms assign the same Case and theta roles (except as absorbed in the passive) is hardly as well—motivated as it is under the earlier TG approach. That is, by lexicalizing the morpho- logical difference between time active and time passive, GB loses the strong argument for generalization. For instance, it is part of the TG analysis of (12) that the passive verb participle absorbs the Agent theta role available in (11). It might be argued that a verb phrase assigns the highest priority theta role (in the sense of Lexical—Functional Grammar, for instance) that is avail— able on the verb to the subject position. Normally, this is the Agent theta role, but if time; is not available, other theta roles may be used. Once verified against the subject, of course, these theta roles are unavailable for object NPs. Passivization of (11), then, immediately generates (14), with no trace in object position as in (12b) because the NP Mary naturally verifies its theta role againts the highest priority role available on the verb like (i.e., Theme, since Agent has been absorbed by the passive). (14) Mary was liked. Whether accusative Case is absorbed or not in (14) is irrel— evant because its absorption is no longer required to force the movement of the object NP (since, of course, no such 94 movement takes place). I present this alternative analysis because, while both analyses work within the main theory developed in this thesis in regards one object (i.e. two theta-role) verbs, problems arise when combining the GB approach to passives, my theory, and double object verbs. There are many verbs in English which may have either two object NPs or one object NP and one object PP (where the PP is headed by to or fer). Examples include: (15) a. John gave Mary the book. b. John gave the book to Mary. (16) a. %John opened Mary 3 can of soda. b. John opened a can of soda for Mary. (17) a. John threw Mary a ball. b. John threw a ball to Mary. There are numerous ways to deal with this data. First of all, it might be claimed that there are two lexical entries for each verb in (15) - (17), and that the similarity of meaning between the (a) and the (b) examples is the result of semantics, and not syntax. Conversely, it might be claimed that there is one entry for each verb in (15) — (17) (Larson 1988). Supporters of either possibility must con— tend with certain problems (see Larson and sources cited there). ‘However, while Larson offers a) fairly convincing account of double object verbs under the second assumption, a potential problem arises in regards my theory. Oversimplifying, assume that time sole difference be- tween (15a) and (15b) is in theta role verification, so that 95 is (15a), give carries Coal and Theme theta roles, while in (15b), the Goal theta role is carried by the preposition to. Since give carries two accusative Cases, one is left unveri— fied (Mary's Case is verifed by to). The passive form of (15b) differs from (15b), according to CB, only in that one accusative Case is absorbed and the Agent theta role is no longer verified» One accusative Case 515 still available, and, at this point in the derivation, we have (18): (18) e was given the book to Mary. This is well—formed except that there is no element of a chain in the subject position, but this can be filled by a pleonast, yielding (19), which the theories mentioned above, taken together, predicts to be well—formed, but which is obviously ill—formed: (19) *It was given the book to Mary. The solution to this problem is obvious: one of the afore— mentioned theories (i.e., my thoery, the GB treatment of passives, and time one—entry treatment (fl? double object verbs) is wrong. While this weakens the strength of my theory, it does not critically cripple it, because both of the other theories involved are controversial. Abandoning either of time other two theories resolves the problem of (19). For instance, on a two—entry account of (15) — (17), we could say that in the (3) sentences, two accusative Cases are available, while in the (b) sentences, only one is, making (19) ill—formed. Further, on the alter— nate analysis of passives that 12 briefly discussed above, 96 (19) is ill-formed because the subject position must verify a Theme theta role. Either way, (19) is explained while my theory remains intact. .A more subtle problem arises 5J1 the passivization of (20). (20) I gave blood. Here, give verifies two theta roles, one on I and one on blood. It has three Cases available (assuming the verb has the same entry as in (15a)), one nominative and two accusa- tive. Using the GB account of passives, one of the accuas- tive Cases is absorbed by the passive participle, but one is still available (as in (18)). Inserting the pleonast yields (21), which is as ill-formed as (19). (21) *It was given blood. Unless the treatment of double object verbs is much more complex than is suggested above, (21) suggests that the flaw signalled by (19) is either with my theory or with the GB treatment of passives. This would be a crippling problem if the GB account for passives vmnme universally accepted auml uncontested. I claim, however, that the flaw is indeed with the GB account, not with mine. It is my opinion that (21) is ill-formed because, if there are any theta roles at all available on a verb, the highest prioity of these must be verified on the subject NP (as I suggest above). Hence, in (21), the only theta role (that of Theme) is verifed on the pleonast (which is not possible) and the NP blood carries an unverified 97 theta role (or no theta role at all). Both this analysis and that of pleonasts is unfortu— nately hesitant and inconclusive. The point in presenting them, however, in addition to acknowledging possible criti— cisms to my approach, is to show that the data the my anayl— sis inaccurately predicts within the framework of GB may indicate flaws in GB theory as much as it indicates flaws in my own theory. This chapter completes my brief treatment of the issues involved in reducing the finite/infintival clause distinc— tion to a function of agreement features and related phenomena. 5. Conclusions In this paper, I have attempted to explain the appar- aent distinction between finite and infinitival clauses in English as a result of differences in agreement features carried by the lexical item in Infl. Since this account has been developed for English, the problem reamins to determine which of the facets of my theory are specific to English and which are universal mechanisms. Obviously, if my entire account is language—specific, then either the theory is weakened or the claim that there should be generalizations about syntax across languages irs‘weakened (and time latter claim is focal to the linguistics of the latter half of this century). One facet of my theory that I would claim is universal is that there is no distinction between two Infls which may appropriately tme called a) [+Tns]/[—Tns] distinction. In- deed, the distinction has grown to be something of a circu- lar one in the literature; Tajsner (1988: 70) suggests that the subjunctive iii Polish has :3 [—Tns] lidfl_'because it doesn't act like voices with [+Tns] Infls are supposed to act. If this is how the [+Tns]/[—Tns] distinction is to be made, then the distinction is not an actual semantic one but rather just a convenient syntactic notation that is a short— cut to a real analysis and is dangerously misleading. 98 99 Iatridou (1990) argues likewise that there is no real sig- nificant evidence that favors both an [+Agr]/[—Agr] distinc— tion and a [+Tns]/[—Tns], but provides no evidence that the latter distinction is any better than the former. If the latter is just being chosen as a syntactic notation, then labels should be chosen which are not misleading in favor of some semantic interpretation or other. Another example is the problem of Portuguese, presented in Raposo (1987), where it is shown that there are three types of Infl in Portuguese: those in embedded clauses with subjects and agreement, those in embedded clauses with subjects and IN) agreement, and those iii non—embedded clauses. This runs counter to any theory (including mine) which attempts ti) rid time syntactic theory cm? two Infl distinctions (Tense anml Agreement). However, Raposo gum»- vides no evidence of embedded clauses in which the form of the verb (and hence possibly of the Infl) is the same as that of the non—embedded clause. Note that even in English I assert that there is a difference in lexical items chosen for absolute and relative tense. It is not an impossibility that in some languages (such as Portugeuse and Polish) that there are lexical items that include relative tense meaning and information about the subject (i.e. agreement), and well as similar lexical items with absolute tense and subject information. This is crucially a semantic distinction, not a syntactic one, and does not justify a syntactic distinc— tion. 100 Such an analysis might also be applied to Japanese, where embedded clauses are typically headed by different verbs than non—embedded ones. This is the only apparent clause distinction because there is no subject-verb agree— ment in Japanese, but again it does not justify in and of itself a [+Tns]/{~Tns] distinction because it may be as easily semantic eumi syntactic, and ti) call ii: necessarily syntactic would be misleading. As for the other various mechanisms I have presented in this paper, I have noted in some cases which ones seem to be specific to English, but the universality (or the breadth of their typology) of the others is a matter of further re- search. Another point of this thesis was to try to develop a feature system.tmn? Government—Binding theory ti) reduce its dependency (M1 "assignment" (a: terribly' anthropomorphizing concept) and to make it more similar to the other, currently competing syntactic theories. Because time concept (H? as— signment is so deeply ingrained in Government—Binding theo— ry, however, it is difficult to present in so short a space as this thesis more than a rudimentary feature system (even despite Chomsky's own early attempt (1965) to present a feature system of syntax). I have also not touched on certain aspects of the optionality of Case assignment, especially as regards Case assignment under conjunction in non-standard dialects of English (Kershaw 1992, Schwartz 1985), 2M1 issue which has 101 received little attention. The full implications of the optionality of Case assignment have yet to be explored. While there are issues that are unresolved, however, I feel that the theory I present in this thesis accounts for the same breadth of data that current GB theory accounts for without the extreme reliance cui.ad hoc mechanisms that are not motivated by anything semanntic, pragmatic, or psycho— logically real. My alternative analysis, then, is motivated by something other than a coldly objective analysis of the syntax in isolation, and any such analysis is to be pre— ferred over a theory that is merely descriptively adequate. Bibliography Andrews, IX. 1990 "Unification anmi morphological blocking," Natural Language and Linguistic Theory 8: 507—557. Authier, J.M. 1991 "V—governed expletives, Case theory, and the Projection Principle." Linguistic Inquiry 22: 721- 740. Borsley, R. IL. 1991 syntactic Theory; 14 Unified Approach. Edward Arnold: London. Chomsky, N. 1986 Barriers. MIT Press: Cambridge, MA. Chomsky, N. 1981 Lectures on Government and Binding. MIT Press: Cambridge, MA. Chomsky, N. 1970 "Remarks on nominalisation." In R. Jacobs and It. S. Rosenbaum (EKL) English lTansformational Grammar." Ginn: Waltham, MA. 184-221. Chomsky, N. 1965 Aspects of the Theory of Syntax. MIT Press: Cambridge, MA. Comrie, B. 1976 Aspect. Cambrdige University Press: Cam— bridge, UK. Comrie, B. 1985 Tense. Cambridge University Press: Cam- bridge, UK. Dippmann, G. 1987 A Fiactical Review of German Grammar. MacMillan Publishing Company: New York. Dowty, D. R. 1989 "On the semantic content of the notion of 'thematic role.'" In G. Chierchia, B. H. Partee, and R. Turner (ed.) Properties, Types, and Meaning, vol. II: Semantic issues. Kluwer: Dordrecht. 69—130. Fukushima, K. 1992 Personal communications. Gazdar, G., E. Klein, C. Pullum, and I. Sag 1985 Generalized Phrase Structure Grammar. Harvard University Press: Cambridge, MA. Grice, H. P. 1975 "Logic and conversation." In D. Davidson and G. Harman (ed.) The Logic of Grammar. Dickenson: Encino, CA. 64—75. 102 103 Haegeman, I“ 1991 Introduction ti) Government anmi Binding Theory. Basil Blackwell: Oxford. Horrocks, G. 1987 Generative Grammar. Longman Linguistics Library: Essex. Iatridou, S. 1990 "About Agr(P)." Linguistic Inquiry 21(4): 551-577. Kaplan, R., and J. Bresnan 1982 "Lexical-Functional Grammar: A formal system for grammatical representation." In J. Bresnan (ed.) The Mental Representation of Grammatical Representations. MIT Press: Cambridge, MA. 173-281. Kay, PL. 1985 "Parsing 5J1 functional unification grammar." Natural Language Parsing. Cambridge University Press: Cambridge, MA. 251-278. Kershaw, P. 1992 "Conjunction and Case—assignment: A prap- meter." Presented at the Second Language Research Forum. Kershaw, P. 1991 "Tense and Agr in Infl", ms. LaPointe, S.G. 1988 "Toward a unified theory of agreement." In M. Barlow and C. Ferguson (ed.), Agreement in Natu- ral Languages: Approaches, Theories, Descriptions. CSLI: Stanford, CA. 67—87. Larson, R» I(. 1988 "On time double—object construction." Linguistic Inquiry 19: 335—392. Lasnik, H., and J. Uriagereka 1988 A Course in GB Syntax: Lectures on Binding and Empty Categories. MIT Press: Cambridge, MA. McCawley, J. D. 1988 The syntactic Phenomena of English. 2 vols. The University of Chicago Press: Chicago. Mugarza, M. I. L. 1990 Negation in Syntax: 0n the Nature of Functional Categories and Projections. MIT PhD Disser— tation. MIT Working Papers in Linguistics: Cambridge, MA. Palmer, F. R. 1986 Mood and Modality. Cambridge University Press: Cambridge, UK. Postal, Paul M. 1974 an Raising: One Rule of English Grammar and its Theoretical Implications. MIT Press: Cambridge, Mass. Pollock, J.—Y. 1989 "Verb movement, Universal Grammar, and the structure of lihfl' Linguistic Inquiry 20(3): 365— 104 424. Pullum G. K. 1984 "How complex could an agreement system be?" Proceedings of the first Eastern States Conference on Linguistics. The Ohio State University: Columbus, OH. Radford, A. 1988 Tiansformational Grammar: A First Cburse. Cambridge University Press: Cambridge. Raposo, Eduardo 1987 "Case Theory and Infl-to-Comp: The Inflected Infinitive in European Portuguese." Linguis- tic Inquiry 18(1): 85-109. Sag, I. A., R. Kaplan, L. Karttunen, M. Kay, C. Pollard, S. Shieber, A. Zaenen 1986 "Unification and Grammatical Theory." Proceedings of the West Coast Conference on Formal Linguistics. (WCCFL 5): 238-254. Schwartz, B. D. 1985 "Case and conjunction" Southern Cali- fornia Occasional Papers in Linguistics #10: 161-186. Sells, P. 1985 Lectures on Contemporary Syntactic Theories: An Introduction to Government-Binding Theory, General- ized Phrase Structure Grammar, and Lexical—Functional Grammar. CSLI (3): Stanford, CA. Soames, S., and D. M. Perlmutter 1979 syntactic Argumenta- tion and the Structure of English. University of Cali- fornia Press: Berkeley, CA. Stowell, T. 1982 "The tense of infinitives." Linguistic Inquiry (13): 561—570. Tajsner, P. 1988 "On the nature of some indicative subjunc— tive asymmetry in Polish." Studia Anglica Posnaniensia 20: 69-81. van der Leek, F. 1991 "Iconicity and two-form reflexive systems." Presented at the Chicago Linguistic Society. van Riemsdijk, H., and Edwin W. 1986 Introduction to the Theory of Grammar. Current Issues in Linguistics. MIT Press: Cambridge, Mass. " MM 1, I