The Non-Unity of VP-Preposing



The Non-Unity of VP-Preposing*

Mark Baltin

Department of Linguistics

New York University

New York, New York 10003

719 Broadway, Room #424

mark.baltin@nyu.edu

August 3, 2005

Abstract:

This paper shows that a VP in English is only a VP at the outset of a derivation, and that VP-preposing in English is in fact preposing of the internal arguments of the verb, followed by remnant movement of the original VP. Therefore, English looks much more like German (Muller (1998)), than it appears at first glance The evidence for the non-constituency of the verb and its original arguments in preposed position comes from its solution to what has been termed Pesetsky’s Paradox, in that an object of a preposed VP can bind into an adverbial at the end of a sentence. The paradox results from the incompatibility of the phenomenon with the conjunction of two assumptions: (i) binding requires c-command; (ii) only constituents move.. Assumption (i) requires the object to be higher than the adverbial, but the preposing of the verb and object to the exclusion of the adverbial would then require that a non-constituent (the verb and object) prepose. The paradoxical nature of the phenomenon rests on the two assumptions, and the paper presents additional evidence that binding requires c-command, showing the contrasts between topicalized VPs and topicalized PPs. The full set of binding phenomena can be accounted for with a c-command requirement on binding, but cannot be accounted for with a rival account of command that makes reference to grammatical functions, known as o-command within HPSG (Pollard and Sag (1992, 1994) or ranking (Bresnan (2002)) or f-command (Dalrymple (1999)).

* This paper is an expanded version of a paper that will appear in A Festschrift for Joseph Emonds, to be published by Mouton, edited by Wendy Wilkins, Vida Samiian, and Simin Karimi. That paper deals with the structure of the VP in its canonical, non-fronted position, and does not deal with the VP-preposing construction dealt with here. A version of it was presented at the Western Conference on Linguistics, University of Arizona, in September, 2003, and at the NYU Workshop on Remnant Movement, October, 2003. I would like to thank those audiences, and in particular Paul Postal, David Pesetsky, Guglielmo Cinque, Marcel den Dikken, Richard Kayne, Tom Leu, Laura Rimmell, Howard Lasnik, and Chris Collins, for helpful comments. As usual, they deserve all of the credit and none of the blame.

The Non-Unity of VP-Preposing

“We may look different, but we’re all the same”

Kerchak’s dying declaration, Tarzan, Walt Disney Productions

This paper has two goals. The first goal is to propose and defend an analysis of VP-preposing in English that denies the constituency of the verb plus its complements in fronted position; the second goal is to demonstrate that optional adverbials are generated outside of the verb phrase.

The literary quote given above is a widely held view among formal linguists about the grammars of natural language (see, for example, the papers in Cinque and Kayne (2005)), and I will illustrate this principle with respect to a common phenomenon in Germanic languages known as “remnant movement”, (den Besten and Webelhuth (1990), Muller (1998)), in which a part of a phrase is extracted, and then the rest of the phrase is moved. An example is given in (1 (Muller’s (1)(a)):

(1) [b Gelesen] hat [ [IP das Buch keiner t1]

read has the bookacc no-onenom

No-one has read the book.

Muller presents extensive evidence that b in (1) is a VP in which the object of gelesen ‘read’ ,das Buch, has first been scrambled out of the VP, with the VP minus its object then topicalizing. In English, this is impossible:

2) *Read he the book.

In English, topicalizing the VP would require taking the object along with the verb, as in (3):

3) Read the book he did__.

Following a tradition in Germanic syntax (Thiersch (1985), den Besten & Webelhuth (1987, 1990), Muller argues that the unavailability of remnant movement in English is due to English’s lack of scrambling. However, it seems that English does in fact have scrambling (Johnson (2001), Baltin (2003)), evidenced in the English pseudo-gapping construction as in (4):

4) Although he didn’t give BOOKS to Sally, he did___MAGAZINES .

Assuming that only constituents undergo processes such as movement or deletion,[1] and that deletion takes place only under structural identity, the objects (books in the subordinate clause, and magazines in the main clause), would have to move out of the VPs prior to the deletion, so that the sequence give to Sally is a constituent. Assuming the correctness of this analysis, the unavailability of remnant movement in English cannot be attributed to English’s’ lacking scrambling, and would remain mysterious.

In the remainder of this paper, I will show that English not only has scrambling, but is like German in that VP topicalization is really remnant movement, so that the verb plus object in (3) does not constitute a single VP, but that the structure of (3) is really (5)[2]:

5) CP

C’

C TopP

VP Top’

V’ Top AgrP

V DPi DPi Agr’

Read t the book Agr TP

DP T’

He T VP

Did t

My evidence for this structure will come from an analysis of binding into clause-final adverbials from within fronted VPs, first discussed in Pesetsky (1995):

6) a.Visit themi he did on each otheri’s birthdays.

b. Visit many prisonersi he did at theiri lawyers’ requests.

The binding requirements from a fronted VP into a clause-final adverbial have caused some discussion within what can roughly be called “Chomskyan” syntactic theories (Minimalist and environs). I will discuss the noteworthy aspects of this construction, and Minimalist attempts to account for its properties. After discussing the problems with previous accounts of these properties, I will present my own analysis, in which the verb plus object do not form a constituent in fronted position. One might reasonably ask whether my conclusions would hold in other frameworks, such as Head-Driven Phrase-Structure Grammar ( Pollard & Sag (1992), (1994)) or Lexical-Functional Grammar (Bresnan (2002), Dalrymple (1999)). I will show some drawbacks to these approaches to binding, in comparing binding from fronted VPs with binding from fronted PPs. It will therefore become clear that the analysis of binding presented here is not based on purely parochial assumptions, and has solid empirical support.

It is helpful to tirst discuss an influential view within Chomskyan linguistics of the English verb phrase, that of Larson (1988).

I. Binding and C-Command

Barss & Lasnik (1986) observed that when English verb phrases contain two nominal complements, the first complement can apparently bind into the second, but not vice versa.

7) a. I introduced themi to each otheri.

b. * I introduced each otheri to themi.

8) a. I introduced many studentsi to theiri teachers.

b. *I introduced theiri teachers to many studentsi.

Larson assumes, as do I , the definition of binding given in (9), the binding principles given in (10), and the constraint on relating pronouns to quantified antecedents in (11):

9) a binds b if and only if a c-commands b and a and b are co-indexed.

10) Binding Principles:

Principle A= An anaphor must be bound in its minimal domain.

Principle B= A pronoun must be free (i.e. not bound) in its minimal domain.

Principle C= R-expressions must be free.[3],[4]

11) A pronoun that is related to a quantified nominal must be bound by the quantified nominal.[5]

One final factor that is necessary, which I will consider in greater detail in Section VII, is the failure of certain prepositions to count for c-command, as in (12):

12) a. I talked [PP to themi][PP about each otheri].

b. I talked [ to many prisonersi][PP about theiri problems].

c. *It seemed [to himi] that Johni would lose.

In (12)(a), the object of to can apparently bind into the PP headed by about, as in (12)(b). In (12)(c), if to is ignored for binding, the sentence is ruled out by Principle C.

The apparent failure of certain prepositions to block c-command is extremely problematic, on the face of it, for a definition of binding that relies on c-command; as Paul Postal (personal communication) has emphasized, c-command is a formal principle, meant to be defined on the formal geometry of phrase-markers. It cannot be simply stipulated as being ignored. Indeed, this apparent failure of such PPs to block c-command has led to alternative notions of command in Head-Driven Phrase-Structure Grammar (Pollard & Sag (1992)) and Lexical-Functional Grammar (Bresnan (2002)). I will consider these alternatives to c-command later, and will point to a problem that they share which indicates that their notions of command are in fact too weak to account for the data.

II. The hierarchical structure of the English verb phrase

A. VP- Shells

Based on the above principles of binding, c-command, and presumed invisibility of PPs for the computation of the latter, an English verb phrase with a verb and two internal arguments cannot exhibit ternary branching, as in (13):

13) CP

C’

C TP

DP T’

I T VP

Past V’

V DP PP

introduce them P’

P DP

to each other

Given that PPs are not computed for c-command, ternary branching within the V’ would cause DP and PP to symmetrically c-command each other, and would incorrectly predict grammaticality for (7)(b) and (8)(b). We therefore need a structure in which the object asymmetrically c-commands the PP. Larson’s structure is given in (14):

14) CP

C’

C TP

DP T’

We T VP

Past V’

V VP

e DP V’

them V PP

introduce P’

P DP

to each other

This structure is known as the “VP-shell” hypothesis, in that the highest VP is headed by an empty verb, which takes the lower VP as a complement. The object is generated as the specifier of this lower VP, and the PP is the lexical verb’s complement. The lower verb then moves to the higher, empty verb, yielding the structure in (15):

15) CP

C’

C TP

DP T’

We T VP

Past V’

Vi VP

introduce DPj V’

them Vi PP

t P’

P DPj

to each other

B. Adverbials

The same considerations of binding lead Larson and, apparently, us, to place adverbials as complements of the lexical VP. For instance, the object can bind into a temporal adverbial in (16):

16) I visited them on each other’s birthdays.

The structure of (16) would therefore be (17):

17) CP

C’

C TP

DP T’

I T VP

Past V’

V VP

DP V’

them V PP

visit P’

P DP

on DP D’

Each other ‘s birthdays

What is noteworthy about this structure for adverbials is that it obliterates the distinction between arguments, such as objects and indirect objects, and modifiers, such as temporals, locatives, conditionals, etc. The structure is the same. If the distinction is made anywhere, it is in the lexical entry for the main predicate,with objects, for example, only being possible if explicitly mentioned in the lexical entry for the verb, and modifiers being generally possible. An argument that Larson makes for this structure for adverbials is that it eliminates what has been called non-constituent conjunction, as in (18):

18) I visited Sally on Tuesday and Martha on Wednesday..

The structure for (18) is given in (19):

19) CP

C’

C TP

DP T’

I T VP

Past V’

V VP

e VP and VP

DP V’ DP V’

Sally V PP Martha V PP

visit on visit on Wednesday

Tuesday

The lexical verb in each of the two conjoined VP complements of the empty higher V will move in across-the-board fashion to the empty V, yielding (18). In this way, non-constituent conjunction, sanctioned in other frameworks such as Categorial Grammar (Dowty (1985)), can be avoided, and we can keep to the requirement that only constituents can conjoin.[6]

Later in this paper, I will show that the traditional generative treatment of adverbials, in that they are generated outside of the subcategorization domain of the verb, V’, is correct, but I must first discuss a problem that this analysis poses for the assumption that only constituents can move.

III. Pesetsky’s Paradox

Again, assuming c-command, we appear to be forced into the structure (18), in which the object c-commands the adverbial, and hence the verb and the object never form a constituent, to the exclusion of the adverbial. The standard assumption in generative grammar is that only constituents undergo grammatrical processes such as movement. How, then, can we account for the possibility of VP-preposing in which the verb and object prepose, leaving an adverbial in clause-final position and containing material which must be bound by the object, as in (20)?

20) Visit many studentsi he did on theirj birthdays.

This, then, is Pesetsky’s Paradox, noted in Pesetsky (1995): the assumption that only constituents move forces the verb and object to form a constituent, but the assumption that binding requires c-command prohibits the verb from forming a constituent.

It must be emphasized, however, that the paradox rests on the assumption that binding requires c-command, an assumption that is by no means uncontroversial in formal syntax.[7] Other theories of grammar, such as Head-Driven Phrase-Structure Grammar (Pollard & Sag (1992,1994) and Lexical-Functional Grammar (Bresnan (20002), Dalrymple (1999)) do not share this assumption, and therefore would, as far as I can see, be able to accommodate a structure such as (21) for (17):[8]

21) CP

C’

C TP

DP T’

I T’ PP

T VP P’

V’ P

V DP P DP

visited them P DP D”

On each other ‘s birthdays.

For expository purporses, it will be helpful to refer to the empirical phenomenon itself, without any theoretical account of its cause. Let us therefore refer to Pesetsky’s Paradox effects as binding from a preposed consituent into a clause-final adverbial, keeping in mind that the paradox exists only under one set of assumptions about binding and movement dependencies.

In Section VII,, I will provide further evidence for c-command, and against the notions of c-command and f-command, showing that Pesetsky’s Paradox truly does imply a phrase-structure in which the object is higher than the adverbial. Before I proceed in motivating my analysis, however, I must digress and discuss two treatments of Pesetsky’s Paradox effects that essentially reconstruct the fronted VP into its original position that alter the structure within the preposed VP and its trace, allowing the moved element and its trace (construcd as a copy) to have distinct structures.

IV. Cascades and Flexible Constituency Solutions to Pesetsky’s Paradox.

A. Cascades

To solve Pesetsky’s Paradox, Pesetsky (1995) posits two types of structural representations for sentences:(i) cascade structures, in which rightward linear order within a projection implies right branching,; (ii) layered structures, in which rightward linear order is compatible with left branching, and two different modules of grammar might require different structures, so that binding would require cascade structures, and movement and ellipsis would require layered structures (see Steedman (1995) for a related idea). Pesetsky eschews Larson’s VP-shell structure, and instead proposes a structure in which the object is in fact a specifier of the temporal PP, so that a cascade structure for (16) would be (22):

(22) C”

C’

C T”

D” i T’

I T V”

past D”i V’

t V P”

visit D” j P’

them P D”

on D” j D’

each other D N”

‘s birthdays

The layered structure which would allow the movement of the verb and the direct object, stranding the temporal adverbial, would be (23):

(23) C”

C’

C T”

D”i T’

I T V”

Past D”i V’

t V’ P”

V D”j P’

visit them P D”

on D”j D’

each other D N”

‘s birthdays

Pesetsky posits principles that relate cascade structures to layered structures.

We will return to this account, examining its ability to handle similar phenomena, after examining the account of Phillips (1996, 2003).

B. Flexible Constituency

Phillips (1996, 2003) is essentially concerned with the unification of grammatical and parsing mechanisms. Given his (and my) assumption that parsing is by its very nature dynamic, constantly altering structure with the addition of more input(i.e. new lexical items being encountered in the speech or reading stream), the grammar must reflect this dynamism. The claim that the grammar and the parser are identical has implications for the treatment of extraction phenomena as well. Consider leftward extractions, which are handled in, e.g. minimalism as copy plus deletion in a bottom-to-top procedure (i.e. Merge (Chomsky (1995), with the copy created at a point after the original element is merged. The parser encounters a reversed sequence, with the “moved” element (i.e. the copy) encountered prior to the position in which it is licensed (i.e. its theta-position if an argument, or “base” position if an adjunct).

Therefore, Phillips’ grammar creates phrase-markers in a top-down fashion, with the displaced element generated in its displaced position, and is copied in its canonical position. Given that constituents can be altered in the course of the derivation, the copy’s internal structure can be altered in the course of adding additional material. Phillips’ example is given in (24).

(24)(Phillips’ (29))a….and [give the books to themi in the garden] he did_- on each otheri’s birthdays.

i. and [give the books to themi] he did__in the garden on each otheri’s birthdays.

The relevant portions of (24) are generated as follows. Phillips does not adopt Pesetsky’s cascade structures, but rather Larson’s structure for the internal VP, in which a series of empty V positions are generated, but in reverse. Assuming that VP-preposing is adjunction to IP, the initial portion of the second conjunct in (24)) will be (25). (Phillips’ (30)(a)):

(25) IP

VP IP

V VP NP I

give NP V’ he did

books V PP

give P NP

to them

The “trace” of the preposed VP (copy-traces in italics) is then copied as the complement of I, causing the I to project to I’:

26) (Phillips’ (30)(b))

IP

VP IP

V VP NP I’

give NP V’ he I VP

books V PP did V VP

give P NP give NP V’

to them books V PP

give P NP

to them

Finally, adding an adverbial to a trace alters the internal structure of the trace, in this case the dative PP:

(27)(Phillips’ (30) (c ))

IP

VP IP

V VP NP I’

give NP V’ he I VP

books V PP did V VP

give P NP give NP V’

to them books V PP

give P VP

to NP V’

them V PP

give on each

other’s

birthdays.

V. Evidence for a Higher Placement of Adverbials

Pesetsky’s and Phillips’ solutions to Pesetsky’s Paradox rely on reconstruction of the VP which alters the VP’s constituent structure in a way that forces the adverbial to be generated lower than the direct object. However, there is a great deal of evidence that the adverbial must be generated higher than the direct object, outside of the VP, evidence that I will now marshal.

With this in mind, consider sentences such as (28), in which a quantifier object must bind a variable inside of an adverbial.

28) Visit every prisoneri though I may after hisi lawyer visits himi, it won’t matter.

Phillips discusses such cases of variable-binding, which is standardly assumed to require c-command, as does anaphoric binding. However, consider (29), in which the VP in the temporal phrase deletes:

29) Visit every prisoneri though I may after hisi lawyer does___, it won’t matter.

If the temporal is merged low within the copied VP, this would be a case of antecedent-contained deletion. To see this, consider the structure that Phillips’ system would posit:

30) PP

VP P’

V NP P IP

Visit every prisoner though NP I’

I I VP

may V VP

visit NP PP

every prisoner P IP

after NP I’

his lawyer I VP

does visit him

In this case, the deleted VP, indicated as deleted by bolding, is contained within its antecedent. Unless some mechanism now gets the null VP out of its antecedent, this is a case of the much-discussed antecedent-contained deletion ( Bouton (1970), May (1985), Baltin (1987), Larson & May (1990), Fox (2002) and many others), and would lead to ungrammaticality if the null VP is allowed to remain within its antecedent, due to an infinite regress problem in determining its antecedent’s identity.

Hence, Phillips’ parser is recovering, and his grammar is generating, a structure that still needs to be altered in order to avoid antecedent-contained deletion.

One influential view as to how to resolve the problem of antecedent-contained deletion (henceforth ACD) has invoked quantifier-raising at LF out of the antecedent VP (May (1985)’s idea). One might reasonably take the subordinate clause ACD as being resolved by LF movement out of the antecedent, in which case the possibility of ACD here would tell us nothing about the overt position of the subordinate clause.

Most examples of ACD have focused on relative clauses that contain the null VP, as in (31):

(31) I ate everything you did___.

There have been two main approaches to ACD in the literature: the LF-evacuation approach, originally advocated in May (1985), which situates the null VP within its antecedent in the overt syntax, but removes it by an LF operation, Quantifier Raising, so that the null VP is not within its antecedent at LF; and the overt evacuation approach, advocated in Baltin (1987), in which the null VP is removed from its antecedent by an overt operation ( such as extraposition, as advocated in Baltin (1987)).

It is clear that at least some cases of ACD would be straightforwardly resolved by placing the null VP outside of its antecedent in the first place.

In this case, for instance, adverbial subordinate clauses are treated as non-quantificational modifiers in most standard semantics texts (Larson & Segal (1995), Heim & Kratzer (1998)). Hence, the QR approach is at least inconsistent with the semantic treatment of adverbial modifiers. We can still capture the fact that antecedent resolution in apparent ACD cases tracks scope by reading scope off of the overt syntactic structure. Fox and Nissenbaum, (2002), for example, note that ACD forces a de re reading of the subordinate clause when the matrix clause is included in the antecedent VP for the null VP in (32).

(32) (Fox and Nissenbaum’s (3):

a. Room 1 wants to have dinner before Room 2 does < want to have dinner>

b. Room 1 wants to have dinner before Room 2 does < have dinner>.

As they note, (32(a) has only the de re reading (the reading where Room 1 doesn’t know what time Room 2 wants to have dinner, but picks a time which happens to be before Room 2’s desired time) while (32)(b) has the de dicto reading (the reading where Room 1 says “I want to have dinner before Room 2 has dinner.”). However, we can simply posit one structure for the adverbial in which it is included within the matrix VP, forcing the de dicto reading, and another in which it is outside of it, yielding the de re reading. In this way, the adjunct is never contained within the VP to which it is adjoined, and its scope is read off of its adjunction position.

The same facts requiring a higher position for the adverbial obtain even when the VP is not preposed.

33) I visited every prisoneri after hisi lawyer did__.

Notice that, unlike (29), the object binds a variable in the subordinate clause in the non-preposed position. We will return to the mechanism by which the object c-commands the subordinate clause in both the preposed and non-preposed positions (Pesetsky’s Paradox) shortly. For now, however, we can see that the scope of the subordinate clause must be contained within the scope of the object. If the object undergoes QR as well, this is perfectly consistent with the LF scope account of QR.

However, notice that what is doing the work here is the movement of the object out of the VP, in this case by QR. If the object is moving out of the VP, this is consistent with the adverbial being higher than the VP.

The parallels between variable-binding by a quantifier into an adverbial and binding by an antecedent into an anaphor inside of the adverbial (i.e. Pesetsky’s Paradox effects) cry out, I would claim, for a unified treatment. Assuming that binding requires c-command, the object must c-command the adverbial. QR would accomplish the higher movement of the object in the case of variable-binding, but would have nothing to say about the cases of anaphoric binding, assuming that the antecedent objects for the latter case need not undergo QR (i.e., proper names, for example).

One view of LF movement of the objects, Hornstein’s (1995) view, does not tie LF movement to QR. Hornstein argues at length that the movement in question is not specifically quantificational, but is in fact LF-movement to the Spec of a higher Agr projection outside of the VP for Case-checking purposes. We will see below, however, when we look at the evidence from pseudo-gapping, that the movement is in fact overt rather than covert.[i]

In the next section, we will see evidence from the British English do that supports a structure in which the adverbial is outside of the VP at all stages of the derivation.

C..British English Do

British English contains a type of VP-anaphora which looks,to all intents and purposes, like a variant of VP-ellipsis, so that (34)(a) and (34)(b) are, as far as I can determine, equivalent:

(34)(a) John will read the book, and Fred will__, too.

(b) John will read the book, and Fred will do, too.

However, there is a crucial difference between VP-ellipsis and British English do-anaphora. A VP-ellipsis gap, as is well-known (see, e.g., Hankamer & Sag (1976) and, for a recent lucid discussion of this issue, Kennedy (to appear)), exhibits internal structure, in the sense that it must house elements that would have originated in an overt phrasal counterpart of the gap. For example, VP-internal wh-phrases can appear within the gap:

(35)Although I don’t know which book Fred will read, I do know which book Tom will[____t].

Also, an understood quantified object within the VP-ellipsis gap can take inverse scope over the subject, presumably by QR of the object over the subject:

36) Some man will read every book, and some woman will __too. (allows the understood every book to scope over some woman).

Lasnik (1995) also provides evidence that internal arguments extract from an elided VP

in the pseudo-gapping construction:

(37)a. Although he didn’t give books to Sally, he did__magazines.

a. Although he wouldn’t put books on the table, he would___on the mantlepiece.

When we turn our attention to British English do-anaphora, however, we find a striking contrast. British English do does not tolerate any of the diagnostics for internal structure, so that wh-traces are impossible within the VP covered up by do.

(38)* Although I don’t know which book Fred will read, I do know which book Tom will do.

Inverse scope is impossible in the British English do construction:

39) Some man will read every book, and some woman will do too. (can only be understood with the subject scoping over the object, including the understood object in the second conjunct.).

The pseudo-gapping construction is impossible in the British English do construction:

40) *Although I won’t put the book on the table, I will do__on the mantelpiece.

Given this, we would not want to derive British English do in the same way in which we derive the VP-ellipsis construction. The most natural account of the difference would be to derive VP-ellipsis by deletion of the VP after the relevant operations ( wh-movement, QR, A-movement in the pseudo-gapping construction) occur, while British English do would really be a Pro- VP. For concreteness, let us follow Postal’s (1966) view of pronouns as determiners, and view British English do as v (or perhaps a lexical instantiation of Kratzer’s (1994) category Voice). In short, the overt pro-form is really a functional head, but perhaps a syntactically intransitive one, lacking a lexical complement in the syntax but having the semantic features of its typical complement in its lexical representation. Because the complement (in this case, a VP, but for typical pronouns, an NP) is not present syntactically, there would be no source for elements that would have to be generated within the complement. Hence, the structure of a clause with British English do will be as in (41):

(41) C”

C’

C T”

D” T’

John T M”

Past M’

M v”

will D” v’

t v

do

Given the lack of a syntactic VP, we can now test the hypothesis that adverbials originate within the VP. We would predict ,if VP-internal generation were correct, that the adverbial would not be able to co-occur with British English do, just as other elements which are assumed to have a VP-internal origin cannot. However, VP-internal origination for these adverbials is disconfirmed. Such adverbials occur perfectly in this construction, as we can see for the locative and benefactive, two examples:

42) a.Although he wouldn’t visit Sally on her birthday, he would do on her anniversary.

b .Although he wouldn’t bake a cake for Sally, he would do for Mary.

We have seen, then, two pieces of evidence for a VP-external origin for the adverbials that are overtly c-commanded by the object. Before positing a solution, I would like to note one more case of constituency clash in which an adverbial must be c-commanded by material within the VP, but the c-commanding material must be the verb itself. This can be seen in the so-called applicative constructions in the Bantu languages, discussed by M. Baker (1988) and Pyllkanen(2002). An applicative construction is one in which an argument is apparently added to the argument structure of a verb, and the role of the argument is marked by a morpheme added to the verb (the applicative marker). Pyllkanen makes a distinction between high and low applicatives, depending on whether the argument is generated outside of the VP (high) or inside of the VP (low). She takes the benefactive to be a high applicative, a conclusion dove-tailing with my own arguments that the English benefactive must be generated outside of the VP. An example is given in (43)[9]:

(43) (Pyllkanen’s ((12)(b)) (Chaga)

N-a-i-zric-i-a mbuya.

FOC-1SG-PRES-eat-APPL-FV 9-friend.

He is running for a friend. (from Bresnan and Moshi (1993))

However, if we assume Baker’s (1988) analysis of the applicative marker as heading the argument phrase and then incorporating into the verb, we would, on the face of it, have movement to a non-c-commanding position, unless the verb itself moves still higher, before the applicative morpheme incorporates into it.

In a certain sense, then, the phenomenon of Pesetsky’s Paradox is quite general. An adverbial that must be generated high will show evidence of c-command by VP-internal material. The answer which was suggested in the previous paragraph, and which I will try to present in greater detail in the next section, is to move the VP-internal material to a higher position than the adverbial. Specifically, the object must move out of the VP to a position that is higher than the adverbial, followed by movement of the rest of the VP.

IV.The Proposed Analysis

To recap, we need a clausal structure that will have the following two properties:

1. the adverbial must be outside of the VP which it modifies;

2. (ii) VP-internal material must c-command the adverbial.

On the face of it, these two properties would seem to be contradictory. The contradiction is illusory, however. I will now try to posit a structure that satisfies these two requirements.

Let us first consider a simple sentence with a transitive verb, such as (44):

(44) John visited Sally.

A great deal of evidence has accumulated for the position of Johnson (1991) that objects move overtly out of the VP, so that the structure of (44) is at least (45):

(45) C”

C’

C T”

D” i T’

John T V”0

Past D” i V’

t Vk Agr”

Visit D” j Agr’

Sally Agr V”1

V’

Vk D”j

t t

Lasnik (1995) takes pseudo-gapping to provide evidence for the structure in (45).Pseudo-gapping, as exemplified in (46), is analyzed as VP-ellipsis of V”1, after the remnant has moved out.

46) Although he didn’t visit Martha, he did ___Sally.

Evidence for pseudo-gapping as phrasal deletion is seen in (47), in which sequences larger than a single word delete.

(47)Although he wouldn’t give books to Sally, he would___magazines.

Baltin (2003) shows that all internal arguments of a verb have to be allowed to vacate V”1, as can be seen in (48), allowing for multiple arguments to be stranded:

48) Although I wouldn’t give books to Martha, I would ___magazines to Sally.

In Lasnik’s analysis, if V”1 does not delete, V1, the head, moves to v (V0 here). Baltin (2002) argues, however, that the verbal movement is phrasal movement, remnant movement, rather than head movement, on the basis of examples such as (49):

49) Although I didn’t try to persuade Sally, I did____Martha.

Hence, the structure of (50) would have to be (51):

50) I tried to persuade Martha.

(51) C”

C’

C T”

D” T’

I T Z”

Past V” Z’

tried to persuade Z Agr”

D” Agr’

Sally Agr V”

t

A word about the movement of V” in (51) is in order. I analyze it as movement to the specifier position of a higher functional projection, which I label Z here for expository convenience. I suspect that Z is really Pred0, as in Bowers (1993), given that the same considerations that motivate verbal movement in the instances of pseudo-gapping in the literature hold for movement of complex adjectival constructions, as in (52), discussed in Baltin (2002), as well as of predicate nominals such as (53), originally noted by Chris Collins.

(52)Although he isn’t very fond of Sally, he is__of Martha. (understood: very fond).

(53) Although he isn’t a student of physics, he is__of chemistry.

In any event, Johnson (2001) identifies the movement of the argument out of V”1 in the pseudo-gapping construction as the same process of object scrambling in Dutch, which operates out of infinitives into matrix clauses and which Baltin (2003) shows to be A-movement. V” would then move to [Spec, Z”], in this case as remnant movement.

In short, there is a great deal of evidence for movement of verbal complements, followed by remnant movement of the VP, from the original positions of these elements.

With this in mind, we are now in a position to account for the crucial properties of adverbials, (i) and (ii), that were isolated at the beginning of this section.

A comprehensive analysis of adverbials has recently been suggested by Cinque (1999), which takes clause structure to be much more articulated than has previously been thought, with functional heads such as Temp (for temporal), Ben (for benefactive), etc. Cinque’s analysis is based on a large cross-linguistic study of the ordering of adverbials and the order of verbal affixes that express the semantic notions of those adverbials. I propose that the object shifts out of the VP, and the VP then preposes, in a position subordinate to the adverbial. The VP then moves higher than the adverbial, followed by the object shifting to a position higher than the adverbial.

One word about the structure that I will propose that will not be motivated until Section VIII, when I discuss movements involving PPs. I propose that the object shifts and the VP preposes twice in the middle field when an adverbial is generated-once below the adverbial, and once above it. Hence, the structure will be as in (54):

(54) C”

C’

C T”

D” T’

I T Z”

V” Z’

Past

V’ Z AgrP

V DP DP Agr’

Visit t them Agr Temp”

PP Temp’

On each Temp Z”

Other’s

birthdays V” Z’

t Z Agr P

t Agr’

Agr V”

t t

In (54), Temp would be generated above the VP, but lower than [Spec, Agr”], the home of the shifted object, and lower than [Spec, vP], the home of the shifted VP.

We are now in a position to solve Pesetsky’s paradox, in which fronted VPs contain objects that can bind adjuncts in final position.

A. Pesetsky’s Paradox Revisited

The analysis of object shift, VP-movement to [Spec, ZP] above the adverbial would seem to work fine for cases in which the VP is in its canonical position, but we still have not accounted for the hard cases, in which the VP is preposed as a constituent, but can strand an adjunct within which the object can bind, as in (6), repeated here:

(6) John said that he would visit the children, and visit them he did __on each other’s birthdays.

My proposal to handle the preposed case is quite simple. The object shifts out of the VP in the middle field, and the VP then moves leftward, perhaps to a topic position (Rizzi (1997)), by remnant movement. The object then moves immediately below the VP. In short, the structure of the second conjunct is really (55).

(55)

V.Correlations Between Pesetsky’s Paradox Effects and The Possibility of Pseudo-Gapping

The evidence for this derivation will come from an examination of the correlation between the restrictions on pseudo-gapping and the restrictions on Pesetsky’s Paradox effects. Simply put, Pesetsky’s Paradox effects will be predicted not to occur when the extraction that is evidenced by pseudo-gapping is not possible. Needless to say, these correlations will not be expected under either a flexible constituency approach or a cascades approach to Pesetsky’s Paradox.

Recall that object shift can take place out of infinitives, as seen in (49) for English pseudo-gapping, and in Dutch object-scrambling shown below. We would therefore predict Pesetsky’s Paradox to exhibit itself between the object of a preposed infinitive and a final adjunct. To my ear,and those of my informants, the facts bear out this prediction:

(56) Try to visit every prisoneri though I may after hisi lawyer does___, I’m not sure that I’ll be successful.

I will now demonstrate two restrictions on object shift: (i) inability to apply out of an infinitive when a DP intervenes; (ii) inability to apply out of an infinitive that is introduced by an overt complementizer. I will then show the same restrictions applying in the VP-topicalization construction with respect to Pesetsky’s Paradox effects.

A.The Minimal Link Condition

Pseudo-gapping can delete a sequence of verbs, with the condition that the sequence not be interrupted by an overt DP. Hence, (57) is not a possible pseudo-gapping sentence(deletion indicated by bolding):

(57)* Although I didn’t try to persuade Sally to visit Martha, I did < try to persuade Sally to visit> Susan.

Assuming that pseudo-gapping is A-movement of an argument out of the inner VP, followed by ellipsis of that VP, it seems plausible to locate this restriction on pseudo-gapping within the Minimal Link Condition. The object of persuade would intervene between the object of visit and a position outside of the matrix VP, and hence movement of the object of visit, Susan, to this higher position would violate the requirement of Shortest Move (Chomsky (1995), Rizzi (1990)). Presumably, the farthest that the object of the infinitive VP could move would be to a position outside of its own VP.

We can then test whether object shift is responsible for the Pesetsky’s Paradox effects by topicalizing a complex VP and seeing whether the embedded object can bind a variable in an adjunct at the end of the sentence. The object shift account predicts that this will be impossible, and the results seem to bear this out:

(58)* Persuade Sally to visit every studenti though I may on hisi graduation day, it won’t matter.

One might attribute the unacceptability of (58) to its complexity. However, (59) seems much improved to me and my informants:

(59) Persuade Sally to visit Tomi though I may on hisi graduation day, it won’t matter.

The contrast between (58) and (59) is accounted for straightforwardly in the present account. The relation between persuade’s object and the pronoun in (58) must be one of variable binding, which requires c-command. The analogous relation in (59) is simple coreference, without binding.

B. The No-Overt Complementizer Restriction on A-Movement

Johnson (2000) shows that arguments can scramble out of infinitives into the matrix clause in Dutch, as in (60) (Johnson’s (28)), but not when the infinitive is introduced by an overt complementizer, as in (61) (Johnson’s (29)):

(60)…dat Jan Marie1 heeft geprobeerd [ t1 te kussen].

….that Jan Marie has tried to kiss.

… that Jan has tried to kiss Marie.

(61)* …dat Jan Marie1 heeft geprobeerd [ om t1 te kussen].

….that Jan Marie has tried C0 to kiss.

… that Jan has tried to kiss Marie.

Recall that Johnson (2001) is taking object scrambling to be the process that removes the argument from the inner VP in pseudo-gapping before that VP deletes. We might then ask whether we can test the restriction on argument removal from an English infinitive that is introduced by an overt complementizer. English, of course, has the for-complementizer, but an infinitive that is introduced by for has a lexical subject. PRO is clearly not counted as an intervening subject for the Minimal Link Condition (as seen in (26))), a fact that is interesting in its own right. However, movement of an argument over a lexical subject will violate the MLC, as in (62):

(62)*Although I wouldn’t hope for Sally to visit Martha, I would Susan.

Hence, in order to test for the restriction on A-movement out of an infinitive that is introduced by an overt complementizer, we would need to find a complementizer which introduces non-finite clauses in English and which does not take a lexical subject.

One such case exists: the complementizer from, originally discussed in Postal (1974) and then by Baltin (1995) and Landau(2002), as in (63):

(63) He refrained [from PRO visiting Martha]. (position of PRO irrelevant).

Landau (2002) analyzes from as a complementizer. Assuming this to be correct, we note that it is impossible to pseudo-gap a sequence of verbs that includes from:

(64)*Although he didn’t refrain from visiting Martha, he did Susan.

Contrasting with (64) is (65):

(65)Although he didn’t stop visiting Martha, he did < stop visiting> Susan.

Kayne (1981), who noted the ban on raising out of infinitives that are introduced by overt complementizers in the Romance languages (de in French, and di in Italian) suggested that the restriction was due to the ECP. However, an ECP account would predict a subject-object asymmetry in the class of extractable elements from this position, and (38) would suggest that the ban must have another source. I would suggest that the ban is due to Chomsky’s (2001) Phase Impenetrability Condition, in which the phases are vP and CP; the Phase Impenetrability Condition limits extraction out of a phase to the Spec and head positions of the phase. Presumably, the object of a from-gerund extracts to a position within the gerund.

With this in mind, consider the following contrast:

(66)*Refrain from visiting every studenti though she may on hisi graduation day, it won’t matter.

(67) Refrain from visiting Tomi though she may on hisi graduation day, it won’t matter.

Assuming the contrast between (66) and (67),( 66) would be ruled out by the requirement that the object quantifier c-command a variable that it binds. Because the object cannot move out of the from-complement, the requisite c-command relation will not obtain. Given that the embedded object in (67) is not a quantifier, it can be related to the pronoun by simple coreference, which does not require c-command.

Needless to say, the flexible constituency approach would have nothing to say about the contrasts between (58) and (59, or (66) and (67). The cascade theory fares a bit better with respect to the contrast between (58) and (59), since “argument categories” DP and CP block cascade formation, and hence the assumption that from is a complementizer would block cascade formation within it of the temporal adjunct. However, the cascade approach should still treat (58) and (59) on a par given that the infinitive is presumably of the same categorical type in both instances. The flexible constituency approach should allow the copied VP to be altered at the point of encountering the adjunct at the end of the copied sequence in all of these cases. The present account, however, takes the possibility of seeing Pesetsky’s Paradox effects to correlate with the possibility of argument-shifting out of the embedded VP in the pseudo-gapping construction. Taken together with the possibility of eliding the VP in a clause-final subordinate clause which shows evidence of c-command by an object within a fronted VP, it seems that an overt movement, within a single type of representation of constituency, is preferable to either a dual constituency approach or a flexible constituency approach.

VI. The Necessity for C-Command in Binding

It would be helpful, at this point, to review the central empirical findings in this paper, and to ascertain how solid is the support for the constituent structures that this paper advocates. I have shown that :

i) an object can bind into an adverbial.

ii) The adverbial must be outside of the VP.

I have assumed that binding requires c-command, and it is the requirement of c-command that has provided the bulk of the motivation for constituent structures in which the object is outside of the VP, shifting to a position higher than the adverbial. Without this assumption, it might be possible to place the object within the VP and place the adverbial in some higher position. The claim that binding requires c-command is controversial ( Pollard & Sag (1992), (1994)); Bresnan (2002), Dalrymple (1999)). One alternative, when one leaves minimalism and travels to Head-Driven Phrase-Structure Grammar (Pollard & Sag), or Lexical-Functional Grammar (Bresnan, Dalrymple), is to posit a hierarchy of grammatical functions, as in (68):

(68) SUBJ ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download