1 COMPLEMENTATION 1.1. Aim of the course. Topics covered A presentation of the English complementation system, within the general framework of Chomsky's Minimalist Program (Chomsky 1995, 1998, 1999). Range of description: the domain of complement clauses point of view Complement clauses (informal definition): subordinate clauses which function as arguments of predicates (subjects, objects). Complement clauses: a) that-clauses b) infinitive clauses c) ing-complements (gerunds, participial constructions). (1) a. He considered that it was a mistake. b. He considered it to be a mistake. c. He considered accepting their offer. 1. 2. Classification of subordinate clauses A) The structural criterion ( informally , the nature of the introductory element: a complementizer, a relative/interrogative pronoun, a subordinative conjunction). If (most) subordinate clauses are CPs, the structural criterion concerns the type of constituents that fills the CP projection). Three types o subordinates may be identified: 1) Complement clauses: the introducer is a complementizer (C0), an abstract element whose role is to partly nominalize a clause, turning it into an argument of a predicate. a. It is spring. CP b. ...that it is spring c. I can feel that it is string C IP d. Everybody is aware that it is spring. | DP I that I C0 that, for, whether, if C0 --[IP I0 [+finite] ] that C0 --[IP I0 [-finite] ] for C0 --[IP I0 [ finite] ] whether Complement clauses are clauses introduced by complementizers, which function as arguments of predicates. Predicates (verbs, adjectives, nouns, prepositions) c-select and s-select complements, and their subcategorial properties are listed in the lexicon. A) The structural criterion 2) Wh-complements are subordinate clauses introduced by relative or interrogative phrases (pronouns, determiners, adverbs) which move to Spec, C. a. relative clauses (nominal modifiers) Wh- complements b. interrogative complements c. cleft-sentences c1) pseudo-clefts or wh-clefts c2) it-clefts (3) a. the man on whom people have pinned their hopes/ Whoever will come will be well-received. (free relative clause) (2)

2 COMPLEMENTATION b. c1. c2. I wonder on whom they are pinning their hopes in this disaster. What she needs is a good job. It is him who brought about the whole disaster

3) Adverbial subordination The subordinate clause is introduced by a "subordinative conjunction", an introductory element which indicates the semantic interpretation of the clause (a time clause in (4 a) concessive clause in (4b), a comparative clause in (4c) etc. (4) a. He abandoned her before he could find out the truth. b. He abandoned her although he had found out the truth. c. He abandoned her, as if he had not found out the truth. B) The functional criterion This criterion concerns the syntactic function of the clause. It is relevant to distinguish between a. subject clauses b. object clauses c. adjunct clauses (adverbial and attributive clauses). Subject clauses and adjunct clauses pattern alike regarding certain phenomena, such as the possibility of extracting constituents out of them. Both subjects and adjunct clauses are islands for extraction, differing from object clauses, which are transparent for extraction. (5) Object clauses a. John thought that Pedro told him that the journal had published the article already. b. What did John think that Peter told him that the journal had published t already ?

Subject clauses: c. [That Mary was going out with him] bothered you. d. Who did [that Mary was going out with him] bother ? B) The functional criterion Adjunct clauses Mary was bothered because Peter discussed her past. *What was Mary bothered [ because Peter discussed t ]? From other points of view, subjects and objects, which are arguments of predicates, have characteristic properties, not true of adjuncts. The Extraposition +It Insertion construction is characteristic of subjects and objects, not of adjuncts. (6) a. It was suggested to them that they should sell the house as soon as possible. b. He owes it to his fathers influence that the committee appointed him to this position. C) The type of licensing involved This criterion concerns the semantic integration of the subordinate within the main clause. a) argumental clauses are -licensed (complements, adverbials) b) clauses licensed as predicates on an element of the main clause. This element functions as the subject of predication. (e.g., relative clauses are predicates on their antecedents). (7) The man who was wearing the straw hat looked exhausted. Goals: a) an explanatory account

3 COMPLEMENTATION b) a descriptively complete account, even if surely not exhaustive Goals a. an explanatory account b. a descriptively complete account, even if surely not exhaustive Practical skills a. identifying the complex sentence patterns of English b. a good command of the distribution of these patterns. c. understanding the relation between meaning and structure for the patterns studied d. skills a), b), c) are required in any form of manipulating these patterns: paraphrasing, translating, editing, etc. 2. 1 Plato's problem and the GB program Distinctive features of the GB model 1. GB is modular (Modules of GB: X' Theory, - theory, Binding and Control, Case Theory, Move ) 2. Through its Move module, GB, contains a very unconstrained transformational component, because, in principle, Move allows any category to move anywhere at any time. Possible problem: overgeneration, hence the need of filtering away incorrect representations 3. GB has four levels of representation at which various conditions are applied to filter out illicit structures: D-Structure (DS), S -Structures (SS), Logical Form (LF), and Phonological Form (PF). 4. The central grammatical relation in GB is government. This relation is what lends formal unity to otherwise rather diverse subcomponents. Aim of GB: finding a suitable answer to Plato's problem; its success deoends on proposing plausible accounts of language variation and language acquisition. New problem: Which of the conceivable PP models is best, and the issue is in part addressed, using conventional (not uniquely linguistic) criteria of theory evaluation. 2.2 General design of the Minimalist Program (MP) Chomsky currently considers the following questions: How well is FL designed ? How close does language come to optimal design ? More narrowly, the MP seeks to discover to what extent minimal conditions of adequacy (=success at the interfaces) suffice to determine the nature of the right theory.

4 COMPLEMENTATION The program addresses the question of what conditions are imposed on the linguistic system by virtue of its interaction with the performance systems. a) The Articulatory-Perceptual System (A-P) b) the Conceptual-Intentional System (C-I). In so far as we can discover the properties of these systems, we can ask how well the language organ satisfies the design specifications they impose, providing legible representation at the interface. (8) Strong Minimalist Thesis Language is an optimal answer to legibility conditions (cf. Chomsky 1998) 2.3. Design of the MP Bare output conditions An expression converges at an interface level, if it consists solely of elements that provide instructions to that external level, thus being legible for the respective external level. The presence of objects which are not interpretable at an interface causes a derivation to crash. There are two linguistic levels which interface with performance modules, i.e, A-P and C-I: these are PF and LF respectively. PF and LF can be conceived of as those parts of the linguistic system which provide instructions to the performance systems. (9)The organization of a GB Grammar Lexicon D-Structure Move (Affect ) S-Structure Logical Form Phonological Form (10)The organization of an MP Grammar Lexicon (Spell-out) Logical Form Phonological Form

Grammar still associates structural descriptions with each sentence /expression. But instead of associating a sentence/ expression with four representation (D-Structure, S-Structure, Logical Form, Phonological Form), the structural descriptions of a sentence / expression is

5 COMPLEMENTATION now a pair of representations ( , ). is a PF representation interpreted at the articulatory perceptual (A-P) interface. is an LF representation interpreted at the conceptual-intentional (C-I) interface. Conclusions

1. The MP seeks a maximally simple design for language. Given this view, the linguistic levels are taken to be only those conceptually necessary -namely PF and LF - meaning that that there are no (intermediate levels of D-Structure or S-Structure. 2. Each expression is associated with a structural representation, a pair ( , ), where is a PF representation interpreted at the articulatory perceptual (A-P) interface, and is an LF representation interpreted at the conceptual-intentional (C-I) interface. 3. Structure of an (I)-language: A lexicon and a computational procedure 3.1 The components of a language are a lexicon and a computational procedure for human languages CHL, that is, a procedure for constructing or generating linguistic expressions using the items in the lexicon. Linguistic items fall into two main categories: substantive ( N, V, A, P) and functional (C, T D, etc.). The following procedures are involved in building expressions: (12) (i) Select lexical items from the lexicon ( a Lexical Array, a Numeration) (ii) Map lexical items to expression.

The computational system (narrow syntax) consists of a few trivial operations Select, Merge, Move, and (more recently) Agree. Select is involved in the initial choice of the Numeration, as well as in providing pairs of objects that undergo Merge. Merge operates on pairs of elements chosen by Select and maps them from a pair into a single element with a more complex structure. Merge is the basic combinatorial device for obtaining complex objects out of simpler or basic ones. 4. Merge. From X'-Theory to Bare Phrase Structure Defining Merge Merge takes two syntactic objects ( , ) and forms the new object (K ( , )) from them. (14) Input Output , , K Since the possibility of Merge depends on the c-selectional/ s-selectional possibilities of the combining lexical items, Merge is obviously the analogue of X'-Theory. The principle of endocentricity is still present in as much as, of the two items that combine, one, the head, is that which projects and transmits its lable. The relations of head-complement ( sister) and head-specifier continue to be available.

6 COMPLEMENTATION Conclusions 1. The bare phrase structure theory adopted by the MP is represented by the operation Merge. Merge takes two syntactic objects ( , ) and forms the new object K(( , )) from them. 2. Endocentricity continues to function given that, of the two elements that merge, only one, namely, the head projects. 3. The relations of head-complement and head-specifier are available, as before. 5. Spell-Out Elements interpretable at the A-P interface (e.g. phonologic features) are not interpretable at the C-I interface, and vice versa. At some point in the derivation, the computational system must then split into two parts, one forming , and the other forming , which do not interact any further after the bifurcation. S-structure was the point of this split in pre-minimalist versions of the PP theory. From a minimalist perspective, the problem with there being a distinct level feeding PF and LF, such as S-structure, is that, since it does not interface with any performance system , it is not conceptually necessary. Thus every substantive property attributed to S-Structure should be restated within the minimalist framework in either LF or PF terms. The only thing required under minimalist assumptions is a rule which splits the computation to form the distinct objects and . Chomsky (1993:22) dubs this operation Spell-Out. After Spell-Out the lexicon will no longer be accessed, and the items in the Numeration have been used up. The computation from Spell-Out to PF is referred to as the Phonological Component. The computation from Spell-Out to LF is referred to as the covert component, and the computation that obtains before Spell-Out is referred to as overt syntax. PF contains, in addition to phonological rules proper, a morphological subcomponent and it also deals with linearization. 6.Types of features and feature checking Why check? The items combined by Merge group features of different types: phonological, semantic, but also formal (grammatical features) (Person, Number, Gender,(= -features), Case, Tense, etc.). The formal features of the lexical items must be checked during the derivation. Intuitively, one has to verify that each item is suitably placed in an expression. Thus *We goes to school is ungrammatical because the Number feature on the subject does not match the number feature of the verbal inflection. Feature checking is thus an essential aspect of a derivation. 6.1 Strong/Weak features (Chomsky, 1993, 1995) Strong features illegible at PF, so they must be eliminated in the overt component of Syntax by overt movement. A strong feature must enter into a checking relation as soon as possible, causing movement or insertion. The analysis of strong (formal) features, formalizes the intuition that a strong feature is checked immediately and that it has visible effect (displacement). 6.2 Interpretable/Uninterpretable features

7 COMPLEMENTATION

A feature is interpretable if it is legible at LF. A feature like Case is always uninterpretable Number or Gender are interpretable on Nouns, but uninterpretable on verbs, adjectives. Uninterpretable features are not legible at LF, therefore they must be erased before LF. Following the same intuition, uninterpretable features must be eliminated as soon as possible, therefore they induce strict cyclicity. If uninterpretable features are also "strong", they are checked by overt movement and erased after checking. (A slightly different description of strong uninterpretable features will be given in the next chapters). Interpretable features survive to LF and may be used several times in a derivation. 7. Move and Agree 7.1 Move While, Merge forms a new object by concatenating two objects that are separate phrase markers, Move forms a new object by concatenating two objects that are in a single phrase marker. Move is defined as follows. It is possible to decompose Move into the simple operations of Copy and (re)merge. (18) Move (from Kitahara (1997) Applied to the category and , Move forms ' by concatenating and . Input: containing . Concatenate and , forming ' Output: ' (19) a. ' t( ) 7.1 Move Chomsky (1993) incorporates the copy theory of movement. According to the copy theory, a trace is a copy of the moved element which is deleted in the phonological component, but is available for interpretation at LF. A chain thus becomes a set of occurrences of a constituent in a constructed syntactic object. Summarising, Move appears to be a complex operation comprised of copy, merge, chain formation, and finally, chain reduction. Chain reduction is the deletion at PF of all the copies in the chain, but the highest ( the head of the chain). Move observes the following two requirements: (i) Constituents always move ( to the left) to c-commanding positions, of the same type. Therefore heads move to head positions, phrases move to phrasal positions (A or A).

b.

8 COMPLEMENTATION (ii) Locality The closest constituent that has the appropriate checkable feature is the one that moves. Locality becomes a built-in condition, stated as the Minimal Link Condition or the Minimize Chain Links Condition, which specifies that a constituent always travels the shortest possible distance, or equivalently, that if two candidates could check the same feature, it is the closest that actually checks it.

7.2. Agree Agree is a relation between two items, the probe, which is the agreeing item and which is a head that possesses uninterpretable features and the goal a phrase or a head, possessed of a feature that matches the feature of the agreeing head. In the case of Agree, matching of the features of the probe under identity with features of the goal is sufficient to delete the strong uninterpretable features on the probe, rendering movement unnecessary. Agree allows the checking and erasure of an uninterpretable feature, by matching it with an identical feature of another item, in a sufficiently local domain. The conditions governing Agree are summarized below (cf. Carstens (2000:149)). According to this conception , Agree is driven by uninterpetable features of the probe, which must be deleted for legibility. Agree operates between a probe and a goal iff a. has uninterpretable features. b. has identical interpretable features. c. c-commands d. There is no closer potential goal such that -commands commands b (25) [-interpretable] (probe) [+interpretable] (goal) (24)

and c-

Conclusion Move and Agree are alternative mechanisms of deleting uninterpretable features, so as to meet the legibility conditions of LF. 8. Economy Principles Economy of representation is nothing other than the principle of Full Interpretation: every object at the interface must receive an "external" interpretation, i.e., must be interpretable by the external performance systems. Full Interpretation thus determines the set of convergent derivations for a language.

9 COMPLEMENTATION Economy of derivation requires fewer steps than in another permissible derivation.

(27) Shortest derivation Condition Minimize the number of operations necessary for convergence. Instead of Conclusions The novelty of the MP lies in its addressing the question of the optimal design of language, the answer to which is the Strong Minimalist Thesis stated in (28) (28) Language is an optimal answer to legibility conditions (cf. Chomsky (1998)) Adopting the strong thesis has proved to have the following consequences: 1. The only linguistically significant levels are the interface levels (PF, LF) 2. The interpretability condition: Linguistic items have no features other than those interpreted at the interface, properties of sound and meaning. 3. The inclusiveness condition: No new features or symbols are introduce by CHL. 4. Relations that enter into CHL either (i) are imposed by legibility conditions, (ii) or fall out in some natural way from the computational process THE STRUCTURE OF THE ENGLISH CLAUSE 1. The endocentricity of sentences 1.1 Inflection as the head of the sentence. (1) (2) S NP ^ VP a. I0 Tense [ Agr] ^( Mood). b. I' I0 ^ VP c. IP DP ^ VP

Inflection (I0) is considered the head of the sentence, since it c-selects the VP and agrees with the subject DP, thus entertaining formal relations with the predicate (the head - complement relation) and with the subject (the head-specifier relation). (1) Tense[+Agr] s/ed The Stranded Affix Filter Tense is an affix which must be supported by a verbal root. The present or past form of a verb is derivationally produced, by combining the verbal stem and Tense affix during the derivation. Modal auxiliaries In English, Inflection includes in addition to Tense and Agr , the modal verbs: can, may, shall, will, must, need, dare. Justification modals are defective, having only finite Tense froms, i.e., Tense plus agreement features. When modals are present, they support Tense, and the tensed modals may further raise to C0. A clear indication that modals move to C0 is that I0-to-C0 takes place only in root clauses, i.e., only when the C0 position is not filled by a complementizer, as is apparent in the complementary distribution in (3) below, where either the complementizer whether, or the modal auxiliary, occupies the position before the subject (C0):

10 COMPLEMENTATION

(3)

a. Could [IP he ta be a fool]? b. I asked you [CP whether [IP he could be a fool]] c. *I asked you [CP whether could [IP he be a fool]]

Aspectual auxiliaries (4) Aux Tense (Modal ) ( have -en) (be-ing)

Definition: An auxiliary is a verb that subcategorizes a VP, and cannot assign -roles. 1.2. The projection of auxiliaries Similarties between modals and auxiliaries (the NICE properties) a) Negation He should not go. He has not gone. b) Interrogation Should he read this? Has he read this? c) Contracted sentences (tag-questions, etc.) He should do it, shouldnt he? He has read this, hasnt he. d) Emphatic stress (in emphatic assertions) Bob should not go there. Yes, he should go there. Bob has not read this. Yes, he HAS read this, Im sure. Dissimilarties between modals and auxiliaries a) Auxiliaries have a complete paradigm, with finite and non-finite forms, modals are defective to be, having b) Auxiliaries still have lexical uses, unlike modals which are always functional. He has a nice house. He has bought a nice house. He has had a nice house There are several manners of projecting auxiliaries. a) One may treat them as lexical verbs, under the VP, which c-select a VP (5) VP V0 ^ VP (6) have [ V[EN]] be [V [ING] ] (7) IP I' I0 VP [+Past]V0 VP V' ed have V0+en b) Alternatively, one might stress their functional nature and project them as heads of suitably labelled functional projections: Auxiliary Phrase, Aspect P, etc. (8) IP I'

11 COMPLEMENTATION I0 [+Past] ed AuxP Aux' VP

Aux0 have The syntax of auxiliaries

a) V-to-I In sentences where there are no modals, the highest auxiliary raises to Tense, to support the Tense affix, and then it may further move to C0. Example (9b) shows that the auxiliary have has raised out of the VP to T0, past the adverb often, adjoined to the VP, as shown in (9c). (9) a. She often visited the city. b. She has often visited the city. c. IP DP She I' I0 V0 have I0 s VP AdvP often V0 ta VP VP visited the city

b) I-to-C An auxiliary that has moved to I0 can further continue to C0, as shown in (10): (10) a. Has she often visited the city ? b. CP C' C0 I0 V0 I0 have+s she ta Has IP DP I' I0 AdvP often V0 VP VP VP ta V0 visited Extended Projections An extended projection defines a domain of movement for the head (i.e., the verb). a. English, auxiliaries raise all the way up to C0, b. English lexical verbs remain in the VP. (11) French vs. English a. Il embrasse souvent Marie. b. *He kisses often Mary. c. He often kisses Mary. d. Embrasse-t-il souvent Marie ? V' DP him

12 COMPLEMENTATION e *Kisses he often Mary ? f. Does he often kiss Mary? Conclusion The English clause has the following functional structure: (12) CP> IP >Vaux0 > VP I0 Tense[+Agr]^(Modal) 2. Main verbs and auxiliary verbs again. 2.1. Verb Movement and Verbal Morphology The lexicalist analysis: Verbs enter the derivation fully inflected, and merely check their inflectional features against the functional heads. If the inflectional features are strong, the verb raises to Inflection (or the Inflectional heads) to check its features.(The case of French). If the inflectional features are weak, the verb does not raise overtly. There is covert movement at LF English. The derivational account (bare stems +affixes) (13) TP T0 ed TP VP V walk T0 V0 The lexicalist account (inflected forms) (14) TP T0 [+Past] TP VP V walked VP V0 ed

T0 VP V0 T0 walked [Past] The difference between English and French can be stated as follows (cf. Lasnik (1995)): (15) a. In French, the V-features, i.e., those that check features of V, are strong. b. In English the V-features are weak. (16) Strong features surviving at PF cause the derivation to crash. (17) Delay an operation until LF whenever possible, that is, whenever delaying would not cause the derivation to crash. The parametric difference between English and French is now expressed in a different manner, namely the strong/weak difference between inflectional V-features. Strong features trigger overt movement, weak features do not. Auxiliaries One problem for this analysis of English is that English auxiliaries do raise to Tense0 and then to C0, as is apparent in the following types of well-known contrasts. (18) a. He often goes to movies. b. He has often gone to movies. c. Does he often go to movies ?

13 COMPLEMENTATION d. Has he often gone to movies ? 2.2. A Hybrid Approach (Lasnik 1995, 1998) In the hybrid approach, the fundamental difference between English auxiliary and main verbs lies in the choice of the checking mechanism, a difference that correlates with different types of lexical representations. Lasnik re-states the difference between English auxiliaries and main verbs, and between English and French as follows: (20) a. Have and Be are fully inflected in the lexicon (possibly correlating with the fact that they are highly suppletive, allowing for person /number variation). b. All other English verbs are bare in the lexicon. (21) a. Inflection is freely either an affix or a set of abstract features. b. Finite featural Inflection is strong in both French and English.

The choice of Inflection type (featural, affixal) is predictable from the type of lexical representation. If the lexicon lists inflected forms separately, Inflection will be featural, if the lexicon contains the bare form of the verb, Inflection is affixal. The final necessary mechanism is Affix Hopping. AH is morphophonemic; it will be a PF rule, since from the point of view of semantic interpretation, it is desirable that Tense should c-command the VP on which it operates. (22) Afix Hopping : Affixal Inflection must merge with a V, a PF process (distinct from head movement) demanding adjacency. Possible configurations a. (23) ...Infl...V... +F +F This configuration is well-formed. V raises (overtly) to Infl, and all relevant features are checked. This is the situation of be/ have/ do/ (modals) and all French verbs. b. (24) ...Infl......V.... Af bare This is the case of a bare verb and an affixal Inflection (English main verbs). In this configuration PF merger takes place as long as adjacency obtains, and the PF affixal requirement of Inflection is satisfied. Two more configurations (25c, d) will arise, but will lead to a crash, as can be seen below: c. (25) c. ...Infl....V....*at LF. +F of Infl will not be checked; +F bare *at PF as well, since +F is strong d. Infl Af V *at LF. +F of V will not be checked +F *at PF also, if merger fails

In sum, the gist of Lasnik's analysis is that lexical representation determines the type of Inflection, and the strength of features then determines whether feature checking takes place overtly or covertly. 2.3. Evidence for the hybrid approach : Verb Phrase Deletion (VPD)

14 COMPLEMENTATION VPD is a rule which deletes the second of two presumably identical lexical VPs, leaving an auxiliary behind. (26) a. Peter should [buy the text book] and Mary should [e] too. b. Peter will go to London and Mary will [e] too.

Main Verbs VP ellipsis can ignore certain inflectional differences between the antecedent and the elided verb (cf. Quirk e.a. (1972), Warner (1986)): (27) a. John slept, and Mary will too. b. John slept and Mary will slept too. c. John slept, and Mary will sleep too.

In (27a) the past tense form slept serves as antecedent for the deletion of the bare form sleep. The present tense form can also antecede the bare form, as in (28a). (28) a. John sleeps every afternoon, and Mary should too. b. *John sleeps every afternoon, and Mary should sleeps too. c. John sleeps, and Mary should sleep too. Similarly the progressive and perfect forms can antecede the bare form. It appears that a sort of sloppy identity is at work here, permitting tense and aspectual differences to be ignored. (29) a. ?John was sleeping, and Mary will too. b. *John was sleeping, and Mary will sleeping too. c. John was sleeping, and Mary will sleep too. (30) a. John has slept, and Mary will too. b. *John has slept, and Mary will slept too. c. John has slept and Mary will sleep too. Auxiliary Verbs Ellipsis with auxiliaries is markedly different, requiring strict identity. Thus, (31a), though seemingly parallel to (27), is unacceptable, because was cannot antecede be; nor can is antecede be, as shown in (49): (31) (32) a. * John was here, and Mary will too. b. *John was here and Mary will was here too. c. John was here and Mary will be here too. *John is here, and Mary will too.

Similar effects obtain with the auxiliary have. Ellipsis is markedly better in (33) with identical forms of have than in (34) with distinct ones: (33) a. John should have left, but Mary shouldn't (have left). b. ?John should have left, but Mary shouldn't (34) a. * John has left, but Mary shouldn't-(have left). b. John has left, but Mary shouldn't have left. (35) Results on VPE The bare form of a verb V other than be or auxiliary have can be deleted under identity with any other form of V. Be or auxiliary have can only be deleted under identity with the very same form. As Warner (1986) observes, this difference does not follow directly from the degree of suppletion. The paradigm of go is highly suppletive, yet the verb patterns with all the other main verbs considered above, allowing deletion under sloppy identity (cf. (36))

15 COMPLEMENTATION

(36)

John went, and now Mary will go. John went and now Mary will.

Thus, the relevant differences is that between main verbs and auxiliaries. Sag (1976) notices that all these cases could be accounted for by ordering VP deletion before Affix Hopping, i.e., by allowing deletion to take place at a point in the derivation where the inflected form of the main verb has not been created, so that deletion actually operates on identical forms. On a strictly lexicalist view, such as that of Chomsky (1993), described in (14) above, there is no such point in a derivation. Sag's insight is, however, convergent with the hybrid approach, whereby English main verbs come from the lexicon as bare uninflected forms. Identical occurrences may be deleted in syntax, while inflected forms are produced at PF by Affix Hopping: Schematically, (some of) the examples above are analysed as follows: (37) (38) (39) John slept, and Mary will too. John Infl sleep, and Mary will sleep too a. John was sleeping, and Mary will. b. John was ing sleep, and now Mary will sleep. a. John has slept, and now Mary will. b. John has en sleep, and now Mary will sleep.

On the other hand, if auxiliaries come from the lexicon fully inflected, and if deletion requires strictly identical forms, was or is will never be identical to be, since they are not formed in syntax out of Infl + be. (40) a. *John was here and Mary will, too b. John was here and Mary will be here, too Summing up: (41) a. A form of a verb V can only be deleted under identity with the very same form. b. Forms of be and auxiliary have are introduced into syntactic structures already fully inflected. Forms of "main" verbs are created out of lexically introduced bare forms and independent affixes. VP Deletion facts provide strong empirical support for the hybrid approach to English verb morphology. We will adopt it, and use it in the analysis of negation in English. Conclusion. 1. English verbal morphology can best be described by assigning different lexical representations to main verb and to auxiliary verbs. 2. Main verbs are represented with one bare form. They come uninflected into the derivation, and will merge with inflectional affixes during the derivation (Affix Hopping at PF). 3. Auxiliary verbs are represented with all their inflected forms in the lexicon. They come fully inflected into the derivation, and will simply check their inflectional features during the derivation. (Overt movement to functional heads). 4. The lexical representation of the verbs determines the representation of Inflection, either as a bundle of abstract features or as an affix. 3. Negative sentences

16 COMPLEMENTATION 3.1 Negation may affect different types of constituents in a sentence, and it is useful to distinguish between the following types of scope of negation: a) word negation - realized by means of negative affixes, mostly prefixes: unhappy, infelicitous, dislike, displease. b) phrasal negation: the negation not may adjoin to any phrase, taking scope over it. (42) He came to the party not long ago, didn't he? Not far away, it was still raining, wasn't it ?

c) Sentence negation - cases where not has sentence scope. A sentence is negative when its predicate is negated, in other words, when its Inflection, which is the head of the sentence, is negative. 3.2 The concept of negative sentence. Types of negative sentences A sentences is negative, not only by virtue of its meaning, but also because of its syntactic properties. Negative sentences have particular distributional properties, which identify them as such. It is instructive to compare pairs made of a negative sentence, and a nearly synonymous sentence, where negation is expressed by means of a negative word. There are several tests, due to Klima (1964), which distinguish between negative sentences and sentences with negative constituents. a. Tag questions. Under falling intonation on the tag question, negative sentences take affirmative tags, and vice versa: (43) a. Mary is happy/unhappy about her job, isnt she/* is she? b. Mary is not happy/unhappy about her job, is she/* isnt she? b. Not -even tag sentences require a negative host sentence: (44) a. George doesnt like smart girls, not even pretty ones. b. George dislikes smart girls even pretty ones /*not even pretty ones. c. Either conjoining. Two co-ordinated sentences can have the form S1 and S2 only if the second is negative. (45) a. Jack stayed at home all day and Mary didnt go any place either. b *Jack didnt go anywhere all day and Mary stayed at home either. c. John isnt happy and Mary isnt happy either. d. *John is unhappy and Mary isnt happy either. d. Neither tags require negative hosts. Affirmative sntences are followed by so-tags (46) a. Jack doesn't like lingusitics and neither does Mary / *and so does Mary b. Jack dislikes linguistics and so does Mary/ and neither does Mary. 3.2 The concept of negative sentence. Types of negative sentences A sentences is negative, not only by virtue of its meaning, but also because of its syntactic properties. Negative sentences have particular distributional properties, which identify them as such. It is instructive to compare pairs made of a negative sentence, and a nearly synonymous sentence, where negation is expressed by means of a negative word. There are several tests, due to Klima (1964), which distinguish between negative sentences and sentences with negative constituents. a. Tag questions. Under falling intonation on the tag question, negative sentences take affirmative tags, and vice versa: a. Mary is happy/unhappy about her job, isnt she/* is she? b. Mary is not happy/unhappy about her job, is she/* isnt she?

(43)

17 COMPLEMENTATION b. Not -even tag sentences require a negative host sentence: (44) a. George doesnt like smart girls, not even pretty ones. b. George dislikes smart girls even pretty ones /*not even pretty ones. c. Either conjoining. Two co-ordinated sentences can have the form S1 and S2 only if the second is negative. (45) a. Jack stayed at home all day and Mary didnt go any place either. b *Jack didnt go anywhere all day and Mary stayed at home either. c. John isnt happy and Mary isnt happy either. d. *John is unhappy and Mary isnt happy either. d. Neither tags require negative hosts. Affirmative sntences are followed by so-tags (46) a. Jack doesn't like lingusitics and neither does Mary / *and so does Mary b. Jack dislikes linguistics and so does Mary/ and neither does Mary. Types of negative sentences. a) Sentences where negation is in the Auxiliary (47) a. Bob has lost my respect. b. Bob has not lost my respect. c. Bob abandoned his pet cat. d. Bob did not abandon his pet cat.

b) Sentences where negation is expressed by negative quantifiers, like nobody, never, nothing. Syntactically, these negative quantifiers are determiners (no), pronouns (nobody, nothing) or adverbs ( never, nowhere). (48) a. He saw no rose-bush in the garden. b. He saw nobody in the garden. c. He had never visited that city.

c) Emphatic negative sentences are sentences where the negative constituent appears to the left of the subject, triggering inversion. (49) Never before had he seen such pretty girls. Polarity items One other famous problem that relates to negation is that of polarity items (items sensitive to the polarity of the sentence). Affirmative polarity items require assertive, non-negative contexts (sentences). Negative polarity items require negative sentences. Here are a few examples. Positive Polarity Items (50) a. It is still raining. b. He has already arrived. c. Mary is here, too. d. Mary was looking for some old pair of shoes. Negative Polarity Items a'. It is not raining anymore. b'. He hasn't arrived yet. c.' Mary isn't here, either. d'. Mary wasn't looking for any old pair of shoes.

18 COMPLEMENTATION Remark Negative polarity items occur in several contexts related by their semantic properties. Klima (1964) labels them contexts that contain [+affective] triggers. Here is the list of contexts which license NPIs: a) Negative sentences Negation is the strongest [affective] trigger. Use of an API instead of a NPI may lead to ungrammaticality. Most of the other contexts permit both NPIs and APIs, but the interpretation associated with the sentences are critically different: (i) He didnt lift a finger to help. *He lifted a finger to help.

b) Questions are also sensitive to polarity. NPIs are used when a negative answer is expected. APIs are neutral or expect a positive answer. (ii) a. b. a b. Are you expecting anyone this afternoon Are you expecting someone in particular? Do you want any more beans, perhaps? Do you want some more beans, perhaps?

c) Comparative clauses allow both NPIs and APIs, but the interpretations are very different: (iii) a. b. a. b. She was more beautiful than any princess that he had seen. She was more beautiful than some princess that he head seen. He is smarter than any student I ever had. He is smarter than some student I once had.

d) Relative clauses headed by indefinite determiners like no, any, every, few, little etc., as opposed to the definite article, demonstratives, each, several, (iv) I know no politician who has ever done anything for this country. He had every reason to refuse any help they offered.

e) If-clauses are also NPIs triggers, but they license APIs as well, roughly, under the same circumstances as questions: (v) If anyone comes, tell them to wait. If someone comes, invite him in the office. 4. Negation in the Auxiliary. 4.1 The Negative Projection English sentential negation can show up in two different shapes: the contracted n't or the full form not. It is generally assumed that the two formatives spell out the content of a Negative Projection, NegP, one of the functional categories of the verb. The examination of sentences with negative operator will offer evidence for projecting NegP as an independent phrase. The Negative Parameter (Laka, 1990) (51) a. Mary is not in the kitchen

19 COMPLEMENTATION b. Maria nu este in bucatarie. (52) The Negative Parameter distinguishes between: a. languages where Negation is above Tense; (Romanian) b. languages where Negation is below Tense. (English) a.

(55)

Not to accept this proposal (seems foolish) Neg>TP b. He has not accepted this proposal. TP> NegP A more restrictive hypothesis regarding functional structure: (56) Hypothesis. 1) The hierarchy of functional categories is invariant. The only thing that varies is the properties of the functional nodes (Borer 1984). 2) Functional categories are projected as a last resort. 4.2 The Split Inflection Hypothesis. (Chomsky 1993) (57) AgrSP > TP > AgrOP > VP s ed ? The analysis may be more detailed and extended by detailing the verbal features of Inflection. (58) AgrSP> TP(M) > AspP > AspP> (AgrOP)> VP s ed, may have be ? The position of NegP Following Lopez (1995), we will assume that NegP is above TP in English as well as in UG: (59) AgrSP> NegP> TP > AspP AspP> (AgrOP) VP s nt ed have be ? It is necessary to analyse the two items that may fill the NegP: not, and n't. 4.3. n't and not. English sentential negation can show up in two different shapes: the contracted n't or the full form not. In this section we will pay attention to their syntactic distribution, particularly to the problem of how the order auxiliary verb + negation obtains. 4.3.1. Nt N't is an affix to the auxiliary; it is a bound morpheme, incorporated into a modal or an auxiliary. Forms, such as, can't, aren't are pulled from the lexicon as fully inflected, and they will have to check their features during the derivation: Hasn't for instance must check [+Present, 3d Person, +Negative]. The hypothesis that n't is incorporated into the auxiliary explains the following: a) N't and the auxiliary raise together as in (60). b) N't attaches to the highest verbal projection of the sentence, (61). c) There can't be two formatives nt, (62): (60) (61) (62) Couldn't you give me that book / a. He couldn't have been fooling around so much. b. *He could haven't been fooling around so much/ **He couldn't haven't been so careful

20 COMPLEMENTATION The sentences in (61) confirm the hypothesis that there is a functional category, NegP with an abstract head carrying a strong feature, Neg [+neg], against which n't checks its own feature. This hypothesis explains the fixed position of n't, which must show up on the highest auxiliary, the one that raises. If n't attached to the lower auxiliary verbs, as in (61b), the features of n't could not be checked. In the same way, there can't be two n'ts as in (62), because there is only one functional head against which the two n'ts could check features and, as a result, the features of the lower n't would go to PF unchecked, causing the derivation to crash. The assumption adopted here (following Lopez (1995), Haegeman (1996)) is that the inflected auxiliary is projected under Tense ( do and the modals) or under Aspect (have, be), therefore, under a category whose content it lexiclizes, and then successively raises to check its inflectional features, ultimately getting to the AgrS0 head where it checks its [Person] features. (63) a. Mary hasn't come. b. AgrSP AgrS' AgrS0 [+person] NegP Neg' Neg0 [+neg] TP T' T0 AspP Asp" Asp0 hasn't VP

DP V' +present V0 +3d pers +neg come Negated modals are subject to the same analysis, except that they are generated under Tense. (64) a. He shouldn't go. 4.3.2. Not Consider now the syntax of not. It differs from nt in the following ways: a. It is not cliticized or affixed to auxiliary verbs. b. When auxiliaries raise to C0 past the subject, not must be left behind (cf. (65)).This suggests that not is not a head that checks features through head to head movement the way n't does. c. In sharp contrast to n't, not can appear in lower positions, as in (66a-c), where not may be adjoined to any of the verbal functional projections. d. There can be two nots, as in (66d). e. Finally the two negatives not, n't co-occur, suggesting that they occupy different positions.

21 COMPLEMENTATION (65) (66) a. Could you not stay home tonight for a change? b. *Could not you stay home tonight for a change? a. He could not have been fooling around so much. b. He could have not been fooling around so much. c. He could have been not fooling around so much. d. He could not have not been fooling around so much. e. He couldn't not do his homework

The following result has been obtained: 1) N't is an affixal head that checks features with an abstract functional category. 2) Not does not have to check features and does not have to be associated to sentence negation. Actually, not can be adjoined to verbal as well as to non-verbal projections as well, so that an adjunction configuration like (67c) below is generally available. (67) a. Not everyone can swim. b. He came here not long ago. c. XP Neg not XP

In sentences which are negative and pass the tests for negativity above, there is a NegP whose strong [+neg] feature must be checked. It can be checked by head to head movement, as already shown, or it can be checked by specifier -head agreement with a negative specifier. We may analyse not as a specifier of the NegP. The presence of not checks the feature [+neg] of the negative head "making the sentence negative" (i.e., negation has scope above tense). Not is a functional element. An alternative that comes to mind is to regard not as a negative adverb, in the lexical class not, never, hardly, scarcely, etc. The analysis of not as an adverb is undermined by the fact that, not triggers do-support, while the other negative adverbs do not. (68) a .* I did hardly buy Nixon's book. b. I did not buy Nixon's book. c. I hardly bought Nixon's book. d. *I not bought Nixon's book. It is also likely that not should not be analysed as a head (contra Laka (1990), Chomsky (1993)). Thus examples like the ones below, show a clear difference between n't which is affected by head to head movement, and not, which is not. If n't is a head and not is a Spec, it is predictable that auxiliaries can skip not, but cannot skip n't. (69) a. He should not have done it. b. Should he not have done it ? c. He shouldn't have done it. d. Shouldn't he have done it ? Conclusions 1. Neg sentences contain a NegP headed by a strong negative feature [+neg].

22 COMPLEMENTATION 2. The NegP is uniformly projected above the TP. Tense and negation are conceptually related, since what sentence negation denies is that the event holds at a particular time interval. 3. The Auxiliary verb + negation word order is due to the existence of a higher AgrS phrase,where the Auxiliary verb checks its [Person , Number] features. 4. Sentential Neg is a functional head whose content is retrieved in two ways, by checking with the affix n't, or by specifier- head agreement with not. Move is involved in both checking operations. The derivation of a negative sentence relies on the mechanisms presented in (71), and (72) (71) (72) [AgrSP [NegP NEG [TP [ PAST ] [AuxP hasn't]]] [AgrSP [NegP not [Neg' NEG [TP [ PAST ] [AuxP has]]]

[AgrSP hasn't [NegP t [TP t [AuxP t]]] [AgrSP has [NegP not [TP t [AuxP t]]] 5. Do-Support (73) (74) He did not come. (a) NegP Neg not Neg' Neg0 [+neg] DPsubj T0 -ed DP tsubj V0 come 5. Do-Support (b) AgrSP AgrS' AgrS0 [+ 3d person] NegP Neg not Neg0 [+neg] Neg' TP T TP T' VP V' ...

23 COMPLEMENTATION T0 VP did/*ed [+past] [+ 3d person] V0 come (75) (76) He didn't come. AgrSP AgrS' AgrS0 [3d person] NegP Neg' Neg0 [+neg] TP T' T0 didn't VP V' [3d person] V0 [+past] come [+neg] 5.1. Extending the analysis. Emphatic assertion

V'

The analysis can be extended to other contexts where do appears, namely: questions, emphatic assertions, short answers and VP-ellipsis: (77) a. Do you know this man ? b. Of course, I DO know the truth. c. Of course, I do.

In all of the cases do supports an abstract morpheme that is not phonetically overt, and which is above T: the question morpheme in (77a), the emphatic assertion morpheme in (77b). Consider emphatic assertions first, by examining the following paradigm: (78) a. Mary left. b. Mary didn't leave. c. *Mary did leave. d. Mary DID leave AgrSP AgrS AgrS0 [3d person] AffP

(79) (80)

24 COMPLEMENTATION Aff' Aff0 [+aff] TP T' T0 VP DID [3rd person] [+past] [+aff] 5.2. Questions and short answers: (84) a. Did she go ? b. What did she sell ? c. Yes, she did.

V' V0 leave

Questions are CPs, containing a question feature and a wh feature in C0. The question feature carries the interrogative meaning, the wh feature is the syntactic marker of a family of related constructions all of which involve wh-Movement (questions, relative clauses, cleft sentences). The question feature is strong in root questions and must be checked by moving an auxiliary verb to C0. This is the familiar rule T/Agr0-to- C0 (I0 -to -C0). Since the question feature is checked by moving a verb, it has to be conceived as some sort of verbal feature, and since only finite auxiliaries undergo movement to C0, the Q feature may be viewed as an uninterpretable Tense feature, finite Tense being the common property of modals, have, be, do. We will accept that root questions contain a Tense feature in C0, a feature which must attract an appropriate verb. In (85a), the aspectual auxiliary have will raise all the way up to C0, finally checking the Tense and wh features. (85) a. Has she come ? b. Is she still working with that company ? c. Could he still go there ? (86) CP C' C0 uTense uwh T/AgrP DP she T/Agr0 [+Present] AspP Asp' Asp0 VP has come Licensing NPIs: NPIs are always in the command domain of overt negation: a. Bill didnt buy any books. T/Agr'

5.3 (87)

25 COMPLEMENTATION b. c. d. Bill is not sure that anyone will lift a finger to help. (*Anyone will lift a finger to help.) *Anyone didnt come. Didnt anyone come?

Conclusions 1.Do Support occurs in a variety of environments. In all of them do supports an abstract morpheme (e.g. +neg, +aff, +uTense, +Agr) which appears above Tense, therefore above the position of the affixes s/ed, and which would remain invisible, unchecked. 2. The presence of these abstract heads bearing strong features forces Inflection to be featural, and forces the use of an auxiliary which can successively raise to check all the features. 3. Given its morphology, do is inserted under Tense and must raise further at least as far as Agreement, possibly to C0. 4. The requirement that these abstract features should be supported by do is a PF not an LF requirement 5. To claim that there is only one negation in an English sentence is to claim that the abstract Neg head licenses only one negative constituent. 6. Other types of negative sentences 6.1. Sentences with negative quantifiers

Consider the following sets of examples, containing negative quantifiers. (90) either. d. Nobody likes him, neither do I. (91) a. They found nothing in the garden. did they? b. They found nothing in the attic, not even old coins. c. They found nothing in the first room and they didn't find much in the second room, either. These examples point out to two things: a) Sentences with negative quantifiers are syntactically negative and pass all the tests for sentence negation. b) These sentences must be "marked" as negative by Spell-Out, because they overtly show the behaviour of negative sentences. The standard analysis of examples like these relies on the insight that sentences with Neg quantifiers contain a NegP, headed by a [+neg] feature, and it is this Neg head which licenses the negative quantifier, if it has sentence scope. Such a view is strengthened by the existence of negative concord languages (e.g., Romanian), where the sentence negator must appears on the verb, in order to license the negative QPs. Thus in Romanian, nu always shows up in sentences with nimeni, nimic. (92) a. Nimeni nu a venit b. *Nimeni a venit c. N-au gasit nimic. d. *Au gasit nimic. a. Nobody came to the party, did they ? b. Nobody came to the party, not even her brother. c. Few people showed up for the lecture, and no one showed up for the party

26 COMPLEMENTATION As to the specific licensing strategy, a frequently invoked solution is the Neg Criterion: (93) a. A negative operator (QP) must be in a Spec head relation with an [+negative] X0 head. b. A negative head X0 must be in a spec-head agreement configuration with a negative operator. A negative operator is a negative phrase in a scope A' position. (94) a. No one has come yet. b. AgrsP DP No one AgrS0 has AgrS' NegP Neg Neg0 ta TP T' T0 ta AspP Asp Asp0 VP ta come yet

(96) a. Nobody came b. NegP DP Nobody [+neg] Neg' Neg0 [+neg] T/AgrP

DP T/Agr' tnobody T/Agr0 VP ed come (98) a. Mary bought nothing b. NegP Neg' Neg0 DP Mary T/Agr0 ed T/AgrP T/Agr' VP

Op [+neg]

27 COMPLEMENTATION V' V0 buy (99) a. Mary has heard nothing. b. AgrSP DP Mary Agrs0 has Op [+neg] Neg0 [+neg] Agrs' NegP Neg' TP T' T0 AspP Asp' Asp0 V0 [+neg] Conclusions 1. Neg QPs may have sentence scope, so that sentences containing them pass al the tests for sentence negation. 2. When they have sentence scope, negative QPs are licensed by vebal negation, therefore by the NegP. 3. Negative Quantifiers are licensed by the Neg Criterion. 6.2. Emphatic negative sentences. The last type of negative sentences considered are emphatic negative sentences. (100) a. Not often did he digress from the topic. b. Not until yesterday did he change his mind. c. Seldom do I see him nowadays. d. Never before had he seen such a crowd. (101) a. Not long ago it rained. b. Not unreasonably, one may expect results from him. c. In no small measure. it is his attitude that is blocking progress. VP V' DP nothing DP nothing [+neg]

28 COMPLEMENTATION d. Not far away, it was raining very hard. It is easy to prove that sentences in (100) exhibit sentence negation, while those in (101) exhibit constituent negation, using the familiar tests. Instances of sentence negation admit neither tags, but instances of constituent negation do not. (102) Not often does Jack attend parties and neither does Jill. *Not long ago, Jack attended a party and neither did Jill. Secondly, instances of sentence negation most naturally take affirmative tags, while instances of constituent negation take negative tags. (103) Not often does Jack attend parties, does he? Not long ago Jack attended a party, didn't he ? When there is sentence negation, negative polarity items ( any, ever, etc.) are licensed, while otherwise they are not : (104) Not often does Jack attend any party. *Not long ago, Jack attended any parties. According to Rudanko (1980), phrases which trigger inversion all "seem to be principally composed of adverbials with an overt or inherent quantifier and motivational adverbs"(1980:356): not often, not always, not until, not even then, not because, not for any reason, not under any circumstances, etc. The attempt to give a sharp semantic characterization of the inversion-triggering phrases is undermined by the fact that the same element may or may not cause inversion: (105) With no job, John would be happy. With no job would John be happy. (106) In no clothes, Mary looks attractive In no clothes does Mary look attractive. This shows that it is the syntax of the sentence rather than the semantics of the phrase which is essential in the description of the contrast between examples (100) and (101) The contrast (100), (101) can be accounted for assuming that the negative constituents which trigger inversion are operators, i.e., sentence negators which have moved to a scope position satisfying the Negative Criterion. Accordingly, they will be licensed in a configuration of specifier-head agreement with a negative head. When an ordinary negated constituent is preposed, which does not qualify as an operator, it does not trigger inversion since it will not require to be in a Spec-head relation with a negative head. Inversion signals the presence of the abstract negative head. The derivation of emphatic negative sentences (107) Seldom do I see him nowadays AdvP Seldom [+neg] CP C' C0 [+neg] DP [+Tense] T/AgrSP T/AgrS' I T/Agrs0 ta AdvP tseldom V0 see him VP VP V' DP AdvP nowadays

29 COMPLEMENTATION Conclusions In the following description of English complementation, the finite clause will be assumed to have (at least) the following structure: CP>AgrSP> NegP> TP> AspP1> AspP2 > VP THAT COMPLEMENTS SYNTACTIC PROPERTIES OF THAT COMPLEMENTS 1. Similarities and differences between DPs and CPs 1.1. DPs and CPs share several properties: a. Both DPs and CPs occur as arguments of predicates. Thus the Longman Grammar (1999) states that "Complement clauses are sometimes called nominal clauses, because they typically occupy a noun phrase slot, such as subject, object, or predicative." b. DPs and CPs merge in - positions and are - marked by predicates that c-select and sselect them. Predicates, (V, A, or N) which combine with that complements have characteristic s-selectional properties. They accept an abstract argument, a Proposition/Theme, and more often than not, they also s-select a human role, Experiencer, or Agent. These two -roles appear in various syntactic functions. (1) a. I thought that it looked good. b. It surprised me that he was right. c. He is aware that he is mistaken. d. It seems to me that he is right. e. It is important (for all of us) that he is still here. f. I claim that he is right. Similarities (continued) c. DPs and CPs accept (some of) the same pronominal substitutes: it, this, that. This is because clauses too have default -features, selecting a clause substitute which is [+Neuter, +Singular]. (2) a. I believe that God is good. b. I believe this/ that / it. c. [That he knows the truth] is not sure. Differences between DPs and CPs: DPs must be case-licensed, i.e, DPs have case features which must be checked during the derivation. The Case Filter bars the occurrence of DPs which lack Case. CPs do no have to be case-licensed. The absence of Case is the main syntactic difference between DPs and CPs, from which all the other differences between DP and CP syntax can be derived. The distribution of CPs is not determined by the Case Filter. As a result, the distribution of CPs is less constrained by syntactic factors and more dependent on discourse factors. Through their syntactic position, that-clauses often code discourse function, like focus or topic. 1.2 Introducing Extraposition In this pattern regardless of its syntactic role ((Su(bject), D(irect) O(bject), Prepositional O(bject)), the complement clause appears at the right periphery of the sentence, while the

30 COMPLEMENTATION pronoun it appears in the position which ought to have been occupied by the clause, thus indicating its syntactic function. (3) Subject a. That Pauline moved to Kansas surprised me indeed. b. It surprises me indeed [that Pauline moved to Kansas]. Direct object a. The engineer wrongly figured out [that the bridge would hold ]. b. The engineer wrongly figured it out [ that the bridge would hold ].

(4)

(5)

Prepositional object a. Can you swear [ that the accused man was at your house all Friday evening? ]. b. Can you swear to it [ that the accused man was at your house all Friday evening?]. The extraposed clause is adjoined to the VP, as in (6) . In (6), the pronoun it occupies the Nom case position, the transitive verb checks the Acc feature of the object, so the CP must be devoid of case. (6) DP I0 VP It V0 surprises DP me IP I' VP CP that he didn't come

The term extraposition is due to Jespersen, MEG. The pronoun it is the so-called introductory-anticipatory it, since it introduces and anticipates the real object of the sentence. The introductory-anticiptory it is regarded as a type of formal subject or object, a "meaningless" or expletive pronoun. 2. The Case Resistance Principle 2.1. The difference between DPs and CPs with respect to case has noticeable empirical consequences. (7) a. I am happy that he left. b* I am happy his leaving. c. I am happy about his leaving. (8) a. I insisted that Mary should depart in the morning. b. *I insisted Mary's departure. c. I insisted on Mary's departure.

The first attempt to precisely state this difference between CPs and DPs is Stowell's 1981 Case Resistance Principle, stated in (9).[ According to Stowell, the case difference between CPs and DPs follows from a dfference between categories which assign case and categories which are case-marked.

31 COMPLEMENTATION (9) The Case-Resistance Principle ( CRP) Case must not be assigned to a category bearing a case-assigning feature. According to Stowell, the case assigning feature that CPs bear is [+Tense], which is involved in the assignment of Nom case in finite clauses. CPs, unlike DPs, bear [+Tense], a verbal case assigning feature, and cannot be assigned case as a consequence. Assuming that there is a difference between -positions and case positions, the consequence of the CRP is that CPs will be banned from positions of case-checking. English supports the CRP to a considerable extent, since in English CPs are excluded from the following three basic (structural) case-checking positions: a) the position after prepositions; b) the structural Accusative position; c) the Nominative position. a) The prepositional context: In English, CPs cannot be sisters to prepositions. (10) a. I insisted that Mary should depart in the morning b.*I insited on that Mary should depart in the morning. b) Structural Acc: the Acc+ Inf construction. The Acc (in italics in (11)) is -marked by the infinitive verb, but gets case from the main verb, (consider). Since the Case source is not the -assigner, the Acc is structural. The example in (11b) is analogous. The CP is -marked by the subordinate infinitive predicate, and would get case from the main verb (consider). The CP is in a structural Acc position, this leading to ill-formedness. (11) a. I consider [ this statement] to be a big mistake]. b *I consider[ [CP that Mary left] to be a big mistake]. c) The Nominative position Sentence (13a), with the subject clause in preverbal position, may be taken to show that the subject clause is in SpecT, i.e. in a position where it has been assigned Nom. On the other hand, sentence (13e). where the Auxiliary has moved to Comp is ungrammatical. Given the ungrammaticality of (13e), it is likely that in the well-formed (13a) the clause is in topic, rather than subject position. As suggested by examples like (13b), there is more than one preverbal position in English. In (13b), last night and in London are topicalized phrases. (13) a. That John hates Mary could be true. a' [[ TP That John hates Mary] [TP tCP [T'could be true]]]. b. Last night, in London, the killer struck again. c. Could this be true ? d.*Did last night, in London, the killer strike again ? e. * Could [that he hates her] be true ? 2.2. CPs may have to pass through case-marked positions Stowell's insight that DPs and CPs differ in terms of case is correct. Nevertheless, as stated in (9), the CRP is too strong and there are empirical facts which disprove it, since they involve CPs that have moved through case-positions, even if they do not remain there. One example is that of operator-variable constructions, movement constructions, where what moves is the CP, acting as a syntactic operator. It is well known that an operator's trace, (=a variable), must be in a case-marked position. Such operator-variable constructions include relativization, question formation, tough-movement, topicalization, a.o. Simple examples with DP operators show that variables are case-marked:

32 COMPLEMENTATION (15) (16) a. What are you so happy about tDP b*What are you so happy tDP a. Who tDP wrote it? b. * Who was it written tDP ? c. Who was it written by tDP ?

Examples (15b, (16b) are ill-formed since the trace is not case marked. The adjective happy in (15b) cannot case-mark the DP-trace. The passive verb in (16b) cannot case-mark the DP trace either, so the preposition by is necessary to case-license the trace, as in (16c) Safir(1985) investigates the behaviour of clauses in operator-variable constructions, systematically comparing extraposed and unextraposed clauses. Extraposed clauses are in caseless position, and predictably, they cannot participate in operator-variable constructions. The essential observation is that only unextraposed clauses participate in operator-variable constructions. Therefore, they leave behind case-marked traces. The operator-variable construction considered below is Topicalization. This rule moves a DP/CP to the left periphery, leaving behind a case-marked trace. (17) a. That Susan would be late John didn't think [ tCP was very likely] b. *That Susan would be late John didn't think [ it was very likely tCP] (18) a. That he had solved the problem we didn't really find [tCP to be very surprising] b.* That he had solved the problem we didn't really find [it to be very surprising tCP] (19) a. That we won't abandon him you may definitely depend on tCP. b. *That we won't abandon him you may definitely depend on it tCP. In every pair, only the unextraposed clause can be topicalized, while the extraposed clause cannot. The chains in examples (17a)-(19a) are correctly formed, containing the operator, that is, the topicalized CP, which binds a variable, i.e., a trace in a case-marked position. In contrast the trace of the extraposed clause is not in a case-marked position. The operator does not bind a variable in (17b)-(19b), the chains are incorrectly formed, so severe ungrammaticality results. Conclusions 1. At least sometimes, CPs must pass through positions where case is licensed, against the CRP. 2. One might interpret this as a sign that a CP may be used to check the strong case feature of some head. Thus, one might claim that in (17), the CP moves from Spec VP to SpecTP, and perhaps further on in order to check the strong features of Tense in English. The principle at work is Lasnik's 1995 Enlightened self interest: a constituent, in this case the CP, moves to satisfy the needs of another constituent, in this case Tense (Inflection). Tense may attract the CP, because the latter possesses features. Thus Case may not be the right way of eliminating the ungramamtical sentences in (10-13) above. 3.Given the data in (17-19), the CRP cannot be maintained in the strong form initially proposed by Stowell.

33 COMPLEMENTATION 2.3. More recently, Stowell's CRP has been reinterpreted as a categorial filter. The proposal is that the CP category is categorially unsuited in certain configurations. This forces clauses to move out of these positions. This interpretation is based on the intuition that predicates/heads should be categorially distinct from their arguments or, more generally from the constituents they govern. Evidence for a categorial filter comes from the fact that sequences of type *N NP (*destruction the city vs destruction of the city), *I ^IP, etc are usually unacceptable. (20) A head and its complement must be distinct in terms of their categorial features.

This categorial filter is sufficient to eliminate sequences of type *P^CP. Categorially speaking, CPs are surely [-N], which is why they do not need case. They may also be viewed as [-V], this allows them the possibiity to be -marked arguments. IF CPs are [-N,-V], they are non-distinct from prepositions and subordinating conjunctions, which are also traditionally described as [-N, -V]. We derive the unacceptability of (22) (22) a. *They complained about [ that salaries were too low]. b. *Although [that she had done her work], the master was displeased. Conclusions 1.Unlike DPs, CPs do not have to be Case-licensed. 2. Nevertheless, CPs can be attracted to case positions, at least when they are antecedents in operator-variable cosntructions. 3. As they pass through positions of case checking, they will be case-marked, this allowing them to appear in operator variable constructions. This property is not available to clauses in extraposed position precisly because they do not acquire a case feature. 4. By virtue of their categorial properties CPs are filtered away from certain environments, such as the position of sister to a preposition. 3. The Extraposition Structure (26) Subject a. That Pauline moved to Kansas surprised me indeed. b. It surprises me indeed [ that Pauline moved to Kansas ]. Direct object a. The engineer wrongly figured out [ that the bridge would hold ]. b. The engineer wrongly figured it out [ that the bridge would hold ]. Prepositional object

34 COMPLEMENTATION a. Can you swear [ that the accused man was at your house all Friday evening? ]. b. Can you swear to it [ that the accused man was at your house all Friday evening? ]. 2.2 Establishing a link between it and the CP The it+CP configuration does not represent a chain of type expletive +associate, for the following reasons: a) CPs do not have to be in a case-marked -chain. b) CPs do not inherit case form it. The CP is caseless when extraposed. Let us turn to claims a) and b), under the standard assumption that Case is inherited along the members of a chain. For example, whom in SpecCP in (30) is Acc-marked and so is its trace in DO position. Consider then the examples in (31): (30) (31) Whom did you see t ? a. *It was bizarre Mary's departure. b. It was bizarre that Mary left. c. *It was noticed Mary's departure. d. It was noticed that Mary left. e.There seems to be a man under your bed. Examples (31a-d) prove that DPs cannot occur in the position of the extraposed clause, because that is a caseless position and DPs need case. Evidence that extraposed clauses do not inherit Case from the expletive it comes from operator-variable constructions. Evidence that extraposed clauses do not inherit Case from the expletive it comes from operator-variable constructions. a) Topicalization The topicalized clause moves to the CP field, and it should leave behind a trace in a case-assigned position. If it were true that extraposed clauses inherit case from the expletive it, it would not matter, in operator variable constructions, whether the (unextraposed )clause is itself in a case position or whether the (extraposed) clause merely inherits case from the expletive it. However, this expectation is no confirmed. Only the unextraposed clause can be topicalized, while the extraposed clause cannot. This is because the trace of the unextraposed clause is in a case marked position (a subject trace in (32), a direct object trace in (33), while the trace of the extraposed clause is in a non-case marked position. (32) (33) a. That Susan would be late John didn't think [ tCP was very likely] b. *That Susan would be late John didn't think [ it was very likely tCP] a. That he had solved the problem we didn't really find [tCP to be very surprising] b. *That he had solved the problem we didn't really find [it to be very surprising tCP]

b) Appositive relative clauses. A that-complement can serve as the antecedent of an appositive clause only if the trace it ultimately binds through the mediation of the relative pronoun is in a case position. Since the relative pronoun itself is a DP, rather than a CP, the requirement that the relative pronoun should check case is natural.

35 COMPLEMENTATION (34) a.[ That Mary was leaving]i , whichi ti was noticed at once, upset Joe. b.[That Mary was leaving]i , whichi iti was noticed at once ti upset Joe. Thus the evidence from operator-variable constructions shows that there is no case transmission between it and the CP, so that it +CP are not members of a chain. Conclusions 1. The evidence argues both against Case-transmission from it to the CP 2. Safir (1985) proposes that it and the CP are simply related as members of a configuration. The semantic relation between it and the CP is that the CP is an adjunct which specifies the content of the pronoun, very much like an appositive clause, which specifies the content of the antecedent (e.g., the fact that he has abandoned his former love). 4. The subject-object asymmetry in Extraposition constructions On the motivaton of extraposition A fundamental remark regarding Extraposition in English, is that this structure is extremely frequent if not quasi-obligatory for subject clauses and marginal for DO and PO clauses. This asymmetry is motivated by structural as well as by functional considerations. The examination of the motivation for extraposition will help us choose among the various proposals on how to analyse Extraposition syntactically. Cullicover and Rochemont (1990), in work on Focus constructions, include Extraposition in a large class of constructions which are motivated by functional considerations. Quirk e.a. (1972) mention the two discourse principles of End-Focus and End-Weight, which play a major role in determining word-order in English. According to these two principles, other things being equal, constituents which are focussed and constituents which are "long" and heavy tend to occur towards the end of the sentence. 4.1 More on End-focus and End-weight. Structural Focus English disposes of two syntactic structures specifically designed to place a constitiuent in focus. These are the cleft sentence, in (39)&(40), and the pseudo-cleft sentence, in (41). In both instances the constituent which occurs after be is focussed, while the rest of the sentence contains presuppostional information. (39) a. What did he purchase for his wife? b. It was [ a brand new fur coat ] that John purchased for his wife. c. Focus: A brand new fur coat. d. Presupposition: He purchased something for his wife. a. Who purchased a brand new fur coat for his wife? b. It was [ John] who purchased a brand new fur coat for his wife c. Focus. John (purchased a new fur coat for his wife). d. Presupposition: Someone bought a brand new fur coat for his wife. a. What does Mary want ? b. What Mary wants is a rich husband. c. Assertion: (Mary wants) a rich husband. d. Presupposition. Mary wants something.

(40)

(41)

36 COMPLEMENTATION

4.1 More on End-focus and End-weight. Topicalization is a syntactic rule designed to indicate the topic of discourse or a link, the constituent which bridges between the given sentence and the preceding discourse. The topic is thus an informationally given element. (42) A: They would like to offer you roses. B: Roses I heartily dislike t.

Topicalization is possible only if there is a case-marked trace in the initial position of the topic. 4.2. The functional perspective Extraposition of a Su clause is functional, since it enables a Su clause, which is a heavy constituent, often containing new information, to occur in final position. S clause V O => it V O S clause

(43) a. He kept complaining. It annoyed him that inflation was running so high. b. ?That inflation was running so high annoyed him. c. He kept complaining. He had found out that inflation was running high. Extraposition from object position is not motivated by the same considerations, since object clauses already satisfy the principles of End-Focus and End-Weight. In a simple declarative transitive SVO senetnce with neutral intonation, the O is the expected focus. Hence, DO/PO extraposition is functionally superfluous and therefore, infrequent. On the other hand, when DO/PO extraposition does occur, the resulting structure has characterstic semantic and pragmatic properties. (see below). SV O clause => SV it O (46) John regretted [that he had abandoned the race]. b. John regretted it that he had abandoned the race. 4.3. The structural perspective Structurally, the motivation for extraposition+it insertion comes from the Extended Projection Principle and from Case Theory (see next section). English is a SVO language that requires an overt subject in preverbal position. Reinterpreting the EPP in feature-checking terms, the fact that a subject in SpecT is always required in English means that the T head has strong features, features which require checking by moving an (appropriate) constituent to SpecT. When there are reasons for the semantic "real" subject to be post-verbal, as is the case with extraposed subject clauses, a formal, 'meaningless' it subject is needed tocheck the strong feature of T and satisfy the EPP. While for the reasons explained, the Nom Su position must be lexically filled , there is no requirement to lexically realize the Acc DO position. If an object clause does not appear in its -position, the latter may remain empty, as in (46a) or it may be filled by the introductory anticipatory pronoun it, as in (46b). (46) a. John has known it for a long time that Mary will leave him. b. John has known tCP for a long time that Mary will leave him. a.

37 COMPLEMENTATION

Conclusions. 1. Extraposition is a discourse-related rule, which places a clausal constituent in Focus position and at the right periphery, satisfying End Focus and End Weight. 2. Extraposition is quasi-obligatory for Su clauses, but infrequent for DO/PO clauses. 3. The subject /object asymmetry is important enough to be looked upon as a structural phenomenon, therefore as a matter of syntax, rather than a matter of stylistic preference. 5. On the English expletives A central claim about expletives is that they occur only in subject position. The Su receives its -role in SpecVP. It follows that the Su position, Spect TP, is projected for purely syntactic reasons, having to do with the strong features of Tense and must therefore be filled even when it has no semantic relevance. Consider the passives below. The passive is an ergative configuration which lacks a thematic subject, but where the syntactic subject SpecT position must be filled nevertheless. It may be filled by the clause itself, or it may be filled by an expletive pronoun: (47) a. That the earth was flat was widely believed in ancient times. b. It was widely believed that the earth was flat in ancient times.

Since the object position is projected only from thematic structure, therefore only if the verb assigns a role in object position, expletive pronouns would not be expected to occur as objects. However, it has been claimed (cf. Postal and Pullum (1988)) that there are many counterexamples to this claim, such as those in (48) and (49). In each case there seems to be an expletive pronoun in what should be a -position, against GB theory. (48) a. I consider it obvious that you should have done that. b. I prevented/ kept it from being obvious that we were late. (49) a. I regretted (it) that he was late. b. They never mentioned (it) to the candidate that the jog was poorly paid. c. I resent it every time you say that. d. I hate it when you are late. The following claims will be defended here, following Rothstein (1995): 1) Expletives occur only in subject position and this follows from the distinguished syntactic nature of the subject position. 2) The examples in (48) are not counterexamples to the theory, since the pronoun is projected as a subject and is (at most) a derived object. 3) When the neuter pronoun it is an object (the examples in (49)), it is not an expletive, but an ordinary pronoun, which receives a role. 4) This leads to a disunitary analysis of textraposition, since only in the case of extraposition from subject position will the clause be initially projected in a -position (SpecVP). For the other cases, the neuter pronoun will be projected in the (prepositional or direct) object position, while the clause will be projected as an adjunct or in some other position. If this analysis is adopted, it is important to define the semantic relation holding between the pronoun and the clause, when the pronoun is not an expletive. 5.1. Licensing subject expletives. The EPP feature of Tense The expletive it is a neuter pronoun, whose main property is that it does not receive any role. As a result it appears in contexts where lexical DPs, which must be thematic, are banned.

38 COMPLEMENTATION In (50), the only overt) -role of the passive verb goes to the CP, so the lexical DP in (50b) cannot be interpreted and violates the - Criterion. (50) a. It was widely believed that the world was flat. b. *The hypothesis was widely believed that the world was flat.

Because expletives fail to be -marked, they cannot be questioned. (51) a. That he came was a blessing for them. b. What was a blessing for them ? c. It was a blessing for them that he came. d.*What was a blessing for them that he came? Expletive as quasi-arguments Since, in principle, pleonastic elements are devoid of content, it was proposed (cf. Chomsky (1991)) that these elements are deleted at LF, because they simply satisfy formal features which have no interpretation. This analysis proved to be problematic for at least two reasons: a) Different expletive elements with the same role, say different formal subjects, do not contribute in the same way to the interpretation of the sentence (cf. (53)) b) Secondly, sentences with expletives are not semantically equivalent with sentences without expletives. Thus, in (54), the variant without there, with the phrase some ghosts in SpecT, presupposes the existence of ghosts, while the sentence witht there in SpecT does not presuppose the existence of ghosts. (53) a. It was a man. (Who was it ?) b. There was a man. (Was there anyone in the room?) (54) a. There were som ghosts in the pantry. b. Some ghosts were in the pantry. The position on expletives adopted here is that expletives are legitimate LF objects with 'null' reference, since they make no contribution to the truth conditions of the sentence. It then becomes necessary to specify for it / there how they are licensed (legitimacy) and what interpretative contribution ( if any) they have. Assuming the principle of Full Interpretation, a natural question is what principles of the grammar license pleonastics. Currently there are two (convergent) ways of stating the intuition that expletive pronoun occur to fill a synatctic subject position: a) the syntactic predication account; b) the EPP account. a) The syntactic predication account (Rothstein (1995) claims that subjects occur to satisfy the condition that syntactic predicates must have subjects. This idea is stated as a Predicate Condition: (59) Predicate Condition Every syntactic predicate must be syntactically saturated. A syntactic predicate is an open maximal projection that needs to be saturated by being linked to a syntactic argument, its subject. Crucially, predicates need not have a thematic relation with their subjects, though they must have a thematic relation with their objects. It follows that expletive elements are licensed only as subjects. A pleonastic subject denotes the null element, since it is has no -role and, when the predicate takes a pleonastic subject, the truth value of the proposition is fully determined by the content of the predicate. The ergative verb

39 COMPLEMENTATION + its object represents a complex syntactic predicate which needs a subject. The subject is licensed only syntactically, to satisfy the needs of the (non-lexical) syntactic predicate. (60) a. It was obvious that we would be late. b. That we'd be late was obvious. c. It was obvious. The expletive interpretation is one way of reading an otherwise ordinary pronoun like it, in thoses cases where the syntactic predicate is completely responsible for the semantic interpretation of the sentence; therefore, the pleonastic appears as a default reading, made available by the interaction of the principles of interpretation and the properties of pronominals. b) The EPP account The analysis in terms of syntactic predication does not, however, explain the difference between English and, say, Romanian, where the semantic process is similar to English, and there are also cases when an ergative verb with its object expresses a complete proposition, but no pleonastic element is overtly present. The fact that the Su is overt in English is related to the well-known fact that English is non-pro-drop a language, that is a language where the Su is obligatory. The presence of the Su is related to the EPP. The obligatory preverbal Su position in SpecT is the effect of the features of the functional head T. T is assumed to have a strong D/N feature which can only be satisfied by Merging or Moving a DP/NP in the (lowest) specifier of T. 5.2. On the English Expletives. There and it behave differently, at least with respect to agreement. (61) a. There is a boy in the room. b. There are boys in the room.

The different agreement pattern follows from the obvious morpho-syntactic difference between it and there. There is an adverbial expletive, so it lacks -features. This is why in there sentences agreement features are checked with the lexical subject, which possesses features It is a pronominal expletive which has -features: it is a [+3d person, +neuter, +singular] pronoun. Subject it always imposes singular agreement on the verb. This is apparent under co-ordination: (62) a. That the president will be re-elected and that he will be impeached are both likely at this point. b. It is /*are equally likely at this point [CP that the president will be re-elected and that he will be impeached] Let us turn to the expletive there. There may be analysed as a DP that checks the case and EPP features, but cannot value the -features of Tense. The simplest analysis is to assume that there originates as a small clause subject and agrees with the predicative inside the small clause. There is thus a defective DP, lacking the - features of person and number, but bearing a case feature checked by T. This analysis suggets that Case may be checked without simultneously checking Agreement.

40 COMPLEMENTATION

(62)

There are monsters IP DP I V DP X There are tbe 5.2. Object Expletives tThere I VP XP X NP monsters

5.2.1 A real expletive The only case of true expletive objects is that of derived objects, that is, consituents which start out as (expletive) subjects, but are case-marked by the verb above them. (63) I find it impossible to live under these circumstances.

5.2.2 It is a -marked pronoun in other cases: (64) a. I regretted (it) that he was late. b. They never mentioned (it) to the candidate that the job was poorly paid. c. I resent it every time you say that. d. I hate it when you are late. In such cases, the extraposed CP/XPs must be independently licensed. There are three types of licensing such a phrase. One of them does not involve an it+ CP structure, but it is instrumental in understanding the object it +CP construction. a. It as an event variable bound by a quantifier over times We refer to examples of type (64 c, d) or (65) below, the DO it is followed by a Time Adverbial Quantifier. In (65a-c) the adverbial is a quantified DP (every time I have dinner with John, etc), while in (65e), it is followed by a when(ever) time clause. It designates an event variable, quantified over by the adverbial (65) a. I regret it every time I have dinner with John. b. The children enjoy it every time you tell them a story. c. They announced it publicly every time they decided to move house. d. He used to like it when(ever) it thundered late. In such examples is that it desgnates an event-variable. Sentence (65a) means "for every event of having dinner with John, I regret that event".. It is a variable that ranges over events of having dinner with John (example a), or events of deciding to move house ( example c). The

41 COMPLEMENTATION pronoun it now has a semantic role, it is an e- variable bound by the quantifier of time. The semantic value of it can be appreciated by comparing (65a) with (66) where it is missing: (66) I regretted every time I had dinner with John. In (66), the every phrase is the object of the verb. In (65), where the every phrase is an operator binding the pronoun, there is a "matching relation" between events named by the every phrase and events named by the matrix verb. Sentence (65a) asserts that every event of my having dinner with John is matched with an event of my regretting having dinner with him. By contrast, (66) asserts that I regretted all the occasions of having dinner with John, but it makes no claim about how many regretting events there were. Thus (66), but not (65a) is appropriate in a situation when, after ten years of happy dinner occasions, something happens that makes me regret that I ever had dinner with John. Selectional restrictions Since it denotes an event, with this interpretation, it is allowed with just those verbs that s-select events. Regret is one such verb, but claim is not. Evidence for this comes from gerunds, which as known, can express events. Expectedly, regret appears with the gerund, claim does not: (67) He regretted doing it /that he had done it. *He claimed doing it/ that he had done it

If a verb does not allow its Theme to be an event, then the verb does not occur in bound time adverbial constructions, like (65). The verb claim, for instance, can only select a proposition for its object, so it does not appear in the it+ (quantified) adverbial construction. (68) He claimed it, but it wasn't true. He claimed it every time he saw you.

The important point is that in the it+ quantified time adverbial construction, the pronoun has semantic content, designating an event variable, bound by the adverbial quantifier. The pronoun it is not an expletive. b) It is a specific context known event. It is an ordinary anaphoric pronoun. (69) a. I regretted it that he was late. b. They confirmed it that you had passed the entrance exam. c. They announced it that she had passed her exams.

In the absence of any QP, the pronoun it is free and denotes a specific entity recoverable from the discourse. In examples (69), it is optional, but not meaningless. The neuter it is appropriate when the object of the matrix verb is a specific event. Bolinger (1977) claims that in these circumstances it "must refer to some fact already broached". It is anaphoric.