The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf ·...

128
The Syntax of π : a study of subjects and ¯ A-movement constraints Philip Branigan Memorial University [email protected] March 30, 2004

Transcript of The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf ·...

Page 1: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

The Syntax ofπ: a study of subjects andA-movement constraints

Philip BraniganMemorial University

[email protected]

March 30, 2004

Page 2: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

Contents

1 Introduction 31.1 The issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Explaining subject/non-subject assymetries 82.1 The subject/non-subject asymmetry in root wh-questions . . . . . . . . . 82.2 The Pesetsky and Torrego (1999) analysis . . . . . . . . . . . . . . . . . 92.3 The difference between root and embedded questions . . . . . . . . . . . 172.4 Deriving the different landing sites for wh-movement . . . . . . . . . . . 212.5 Fixing the P&T analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 312.6 Topicalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3 The A-bar properties of subjects 623.1 Polarity licensingonly . . . . . . . . . . . . . . . . . . . . . . . . . . . 633.2 Locative inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643.3 ECM withwager-class verbs . . . . . . . . . . . . . . . . . . . . . . . . 713.4 Learnability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

4 That-trace effects and the Trace Deletion Constraint 834.1 The context ofthat-trace effects . . . . . . . . . . . . . . . . . . . . . . 834.2 Locality constraints on successive cyclic movement . . . . . . . . . . . . 874.3 A featural constraint on successive cyclic movement . . . . . . . . . . . 884.4 Thethat-trace effect derived . . . . . . . . . . . . . . . . . . . . . . . . 944.5 Avoiding thethat-trace effect . . . . . . . . . . . . . . . . . . . . . . . . 964.6 The ‘adjunct’ effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1064.7 Short wh-movement of the subject . . . . . . . . . . . . . . . . . . . . . 109

1

Page 3: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CONTENTS 2

4.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

Page 4: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

Chapter 1

Introduction

1.1 The issues

From the earliest stages, in Chomsky (1975), generative syntax has explained thesuperficial word order in grammatical sentences by positing abstract movementtransformations, which take grammatical elements from one position in a phrase markerand move them to another. As current understanding would have it, such transformationsreflect some of the computational resources made available to our minds by our geneticmake-up, in the form of a universal grammar, which serves as the basic foundation for theactual mental grammars used by individual speakers of their own languages.

Once a set of movement transformations is posited, the theory of grammar must thenaddress the question why certain expressions turn out to be ungrammatical which amovement-based theory would seem to predict will be good. Some plausible mechanismor set of mechanisms must be found which will ensure that movement does not‘overgenerate’ ungrammatical expressions. Hence, syntactic theory since the mid-60s,initiated by Ross (1967), has been devoted in large part to the search for explanatoryconstraintson movement transformations.

One particular set of ungrammatical expressions enjoyed particular notoriety in theliterature of the 80s and early 90s: sentences involving extraction of the subject of a finiteclause past a local complementiser or past a local wh-phrase in Spec-CP. Typicalexamples of these‘that-trace effects’(Pesetsky, 1982) are seen in (1).

(1) a. *What did Terri think that had happened?b. *I wonder who Claude will ask if called the contractors?c. *Which computer should we know how crashed?

The available data on this type of ungrammatical expression, although remarkably

3

Page 5: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 1. INTRODUCTION 4

complex, is robust and general, and is replicable to some extent across a wide range oflanguages. The data is especially intriguing because it is clearly of a variety which couldnot play a significant role in the linguistic experience of a child, so that thethat-traceeffect cannot be a ‘learned’ component of the grammar. The explanations proposed forthese facts in the literature are unfortunately also remarkably complex, and cannotgenerally be brought into the Minimalist fold in any obvious way. The most widelyaccepted treatments ofthat-trace phenomena are problematic even on their own terms,precisely because they must introduce special conditions to account for the precise limitsof thethat-trace effect. Thus, for example, Chomsky (1981) must distinguish betweengovernment of the subject position by a Case-assigner and ‘proper’ government of thesame position by a lexical head. Similarly, Chomsky (1986), Cinque (1990), and Rizzi(1990a) have to ensure somehow that traces in subject position are not‘antecedant-governed’ by intermediate traces in Spec-C, despite the lack of any apparentimpediment to such a relation in the structure. (The precise mechanisms invoked toensure the results required in each case are sufficiently well-known that I do not dwell onthem here. Some specific issues are examined in chapter 4.)

Our understanding of the notion ‘subject’ has been drastically revised in recentyears, under the impetus of the VP-internal subject hypothesis (Kitagawa, 1986;Koopman and Sportiche, 1991) and the exploded Infl hypothesis (Pollock, 1989; Belletti,1990). It is now clear that no single syntactic position can be associated with the entirecomplex of typical subject properties: nominative Case, agentive theta-role, clause-initialposition, etc. Instead, some subject properties belong to the Spec-vP position, and others,to the Spec-TP position. (cf. Baker (1988) and McCloskey (1997) for discussion of thispoint.) The upper Spec-TP position is the position where nominative Case is normallychecked (modulo the possibility ofin situchecking, as in Chomsky (pear), for example).It is also the position where the Extended Projection Principle (EPP) must be satisfied, byhaving some category appear there before Spell-Out. Finally, Spec-TP is sufficiently highin a clause to ensure that a phrase in this position will c-command most other elements,which presumably accounts for the binding and control properties of clausal subjects.

Besides the Spec-vP and Spec-TP positions, there may also exist intermediatepositions within the clause which can be occupied by subjects at some point in thederivation. Kayne’s 1989 PartP specifiers, which trigger participle agreement in Frenchpassives, would constitute one such case. McCloskey (1997) provides an overview of theevidence for this type of intermediate subject position. For the issues under discussion inthis monograph, however, the presence of an additional subject A-position belowSpec-TP is immaterial, so it will be left to the side for the most part.

Once we have identified the properties which can be naturally associated with theSpec-vP and Spec-TP positions, there still remain a handful of subject properties which

Page 6: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 1. INTRODUCTION 5

lack an explanation. Thethat-trace effect is one such subject property which has provedparticularly resistant to explanation in terms of the properties of Spec-TP or lower subjectpositions.

In the work presented here, I offer an account ofthat-trace effects which appears tome to be preferable on several counts. First, rather than stipulating a distinction betweendifferent types of licensing relations (government, proper government,antecedent-government,γ-marking, etc.) which apply to specific empty categories, Ishow that the difference between subjects and other XP elements follows from twoindependently motivated, principled innovations. First, I show that subjects are typicallyrequired to undergo a short step of A-bar movement in every finite clause. I argue thatevery full clause contains an A-bar head, for which I use the labelΠ, which has an EPPfeature normally checked by raising the subject to become the specifier ofΠ. Building onthe work of (Pesetsky and Torrego, 2000), I show that this movement takes place to allowΠ to check a particular feature which subjects share with the head of TP.

Secondly, I make and exploit the claim that A-bar chains must be maximally binary,with intermediate traces deleted subject to recoverability. When the notion ofrecoverability involved is made explicit, it turns out that subject trace deletion fromSpec-ΠP will be impossible unless special mechanisms are available to allow therecoverability condition to be satisfied. The special mechanisms required fall under thegeneral principle of Recoverability of Deletion, and become clear once an elementarytypology of A-bar checking features is developed.

Thus, in (1), the subject DP raises from its Spec-T position to an A-bar positionoutside of TP. It is precisely because subjects have raised to this particular A-bar positionthat they are normally constrained from undergoing further A-bar movement; theproperties of subjects which are determined by their argumental syntax (nominative Case,agreement with T, particular theta-roles and binding properties) are not involved.

The combination of these two claims is what will provide an account of thethat-trace effect. At the same time, the mechanisms which are developed in approachingan account of thethat-trace effect turn out to provide an account of some problematicaspects of other constructions, including Germanic residual and general verb-secondstructures, English locative inversion and some notoriously peculiar cases of ExceptionalCase-marking in French and English.

The analyses presented here rely crucially on several general principles governingsyntactic derivations: theMinimal Link Condition, thePhase Impenetrability Constraint,thePrinciple of Minimal Compliance, etc. I make no attempt to justify these beyond theexisting literature. If these particular principles are misguided, then so is my approach tothe general problem of subjects. Conversely, to the extent that the ideas presented hereare successful, they provide new confirmation that these general principles are sound.

Page 7: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 1. INTRODUCTION 6

The theoretical model which I assume as background is ‘minimalist’ syntax, assketched out in Chomsky (1995, pear, 2001b,a), and as developed by many others, somecited below. The fundamental assumptions I make are few, as befits a minimalist inquiry.I assume that phrase markers are built up by a succession of Merge operations, whichcombine two smaller objects (phrases or lexical units) into a larger phrase, with one ofthe components serving as the ‘label’ for the whole. The Merge operation can bebinary—‘external Merge’—with two separate objects combined into one, or it can beunary—‘internal Merge’—with a sub-phrase selected from within a larger phrase toundergo Merge at the root. The latter operation is syntactic movement of the familiarsort: raising, wh-movement, topicalisation, etc. Movement forms chains: complexlinguistic objects in which a single constituent occupies multiple positions. Movement isdriven by EPP features of the heads of phrases, and typically involves a simultaneousfeature-checking operation between the root and the internally selected phrase.

Checking and movement operations are subject to general locality constraints,including thePhase Impenetrability Constraintand theMinimal Link Condition. Theselocality constraints are evaluated subject to thePrinciple of Minimal Compliance(Richards, 1997).

1.2 Outline

As with any analytic task, the challenge in developing an account of thethat-traceaccount is to avoid restating the descriptivie generalizations in a less accessible technicaljargon. Ideally, an account of this—or any—construction should fall out from generalprinciples which have no particular assocation with the particular construction, i.e. withthat or traces or subject extraction.

The argument I present for my particular approach tothat-trace effects involvesseveral steps in which the relevance to my final goal is initially not obvious, because theyare intended to motivate or illustrate the general principles which will ultimately providean account of constraints on subject extraction as a side-effect. In chapter 2, I present theevidence for a particular view of the Spell-Out operation, and for the structure of theinvisible left periphery of root clauses. It is shown that theΠ head is a necessarycomponent of phase theory, and that this head has a specific set of checking propertieswhich interact with the subject of the clause. A general theory of Germanic verb-secondstructures is presented as evidence for the basic principles proposed there.

Chapter 3 then presents a serives of arguments which show that the Spec-ΠPposition is always an A-bar position. The evidence presented here involvesonly-focusphrases,wager-class ECM constructions in English and French ECM more generally, and

Page 8: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 1. INTRODUCTION 7

English locative inversion structures.Finally, the locality constraints on successive cyclic A-bar movement are examined

in chapter 4, and the account of thethat-trace is explicated. Various strategies forby-passing thethat-trace effect are then examined, including Frenchqui, Yiddishexpletive subjects and Scandinavian complementiser structures.

Why Π? Well, everybody likesΠ!

Page 9: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

Chapter 2

Explaining subject/non-subjectassymetries

2.1 The subject/non-subject asymmetry in rootwh-questions

The data in (1)–(2) will be familiar to everyone.

(1) a. Who pitched the tent?b. *Who did pitch the tent?

(2) a. *Which tent Sheila pitched?b. Which tent did Sheila pitch?c. *Where Sheila pitched the tent?d. Where did Sheila pitch the tent?e. *When Sheila pitched the tent?f. When did Sheila pitch the tent?

Takingdo-support as indicative of movement of T to C, what the data in (2)–(2)appear to show is that T raises to C when a non-subject undergoes root wh-movement,but not when a subject undergoes root wh-movement. There are two related issues raisedby this data: first, why does T-to-C movement take place in root wh-questions at all, andsecondly, why are subject wh-questions like (1-a) exempt from this movement?

The most compelling attempt to address these issues in the recent literature is foundin Pesetsky and Torrego (2000) (henceforth P&T). In my view, the P&T analysis includesa mixture of important insights and inaccurate generalisations, and it is in teasing theseapart that we can arrive at the first stage of a better analysis of the syntax of subject

8

Page 10: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 9

extraction. Such is the goal of this chapter, where a brief sketch of the P&T system ispresented, and then their ideas are tested and modified on the basis of comparative datafrom the Germanic language family.

2.2 The Pesetsky and Torrego (1999) analysis

Root questions

The foundation of the P&T analysis of the assymetry in (2)–(1) is that C in rootwh-questions has an uninterpretable feature to check which may be checked by either thesubject or T, where checking implies movement of the target. The feature in questionmust be something which is found in both T and a nominative DP, and P&T propose thatthis feature is [Tense], a feature which isrealisedas (uninterpretable) nominative Case inDP.1 This claim receives support from languages in which tense-marking appears on thesubject of a sentence, as P&T show.2

Consider now the structure of (2-b) at the point in the derivation after C merges withTP: (3).

1This idea is independantly proposed in Haeberli (1999).2By identifying T and Case features, P&T are able to put forward a strong thesis, their Relativized

Extreme Functionalism hypothesis, wherein the grammar does not include any purely uninterpretablefeature, but only interpretable features which may appear in syntactic slots where they cannot beinterpreted. This aspect of the P&T analysis is orthogonal to the questions I address here.

Page 11: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 10

(3) CP

�����

HHHHH

C[Tense][wh]

TP

�����

HHHHH

DP��� PPP

Sheila[Tense]

T′

����HHHH

T[Tense]

vP

����HHHH

t v′

���HHH

v�� HH

pitch v

VP��� HHH

V

t

DP���

PPP

which tent

The [Tense] feature is uninterpretable on both C and the subject DP, and interpretable onT. T checks Tense (=nominative Case) on the subject DP, as usual. (The Tense feature onthe subject is in fact already checked at this point in the derivation, but P&T assume,following Chomsky (2001b), that checked features are not ‘erased’ until the completionof the next phase. So the Tense feature of the subject is, crucially, still available forchecking operations within CP.) But the [Tense] feature on C still remains to be checkedat this point. It can do so by matching features with the closest target which bears thematching [Tense] feature in TP. Spec-TP and the head of TP appear to be equidistantfrom C—at least, this must be supposed in the P&T approach—so C has the option ofchecking the [Tense] feature on either of them.

Besides the uninterpretable Tense feature on C, C also must bear the uninterpretablefeatures which drive wh-movement: a [wh] feature, and an associated EPP feature (toforce wh-movement to be ‘overt’).

A successful derivation must ensure that each uninterpretable feature on C ischecked. The wh-feature and the EPP feature are checked in the usual way, whenwh-movement takes place. The Tense feature on C still needs to be checked. Like thewh-feature, the Tense feature is checked in a way which forces overt movement of thetarget of checking. P&T conclude from this that Tense in C is also associated with anEPP feature. Since the specifier position of CP will be occupied by the wh-phrase,movement of the Tense-bearing DP is impossible. However, [Tense] is also present in T,and movement of T to C does not require that a specifier position be available. Therefore,

Page 12: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 11

the derivation of (3) can be completed sucessfully if T raises to C, checking the [Tense]feature of C, and eliminating its second EPP feature at the same time.

The contrast between (2-b) and (2-c) reduces to a failure of feature checking in thelatter case, with the uninterpretable [Tense] feature of C resulting in a derivational crashat the LF interface.

Now consider the structure of (1-a) at the same point in the derivation.

(4) CP

�����

HHHHH

C[Tense][wh]

TP

�����

HHHHH

DP��� PPP

who[Tense]

T′

���HHH

T[Tense]

VP

���

HHH

t v′

����PPPP

pitch the tent

Again C must select a target in TP to check its [Tense] feature, and the target raisesto C when checked. C must also check and attract the subject wh-phrase into Spec-CP.Two distinct convergent derivations are available at this point. If C attracts T to check its[Tense] feature, then the Spec-CP position will be available for subsequent wh-movementof the subject. Alternatively, if C attracts the subject DP to Spec-CP to check [Tense],then it can check the [wh] feature of the subject at the same time. In either case, all of theuninterpretable features of C will be checked.

Since there are two convergent derivations possible, questions of derivationaleconomy become relevant. Where there is a choice between two derivations, the lesscostly derivation is presumably more highly valued by the computational component.And there is a clear winner in this case. Movement of the subject alone is moreeconomical than movement of both the subject and T, so the derivation in which C checks[Tense] on the subject should win.

In short, in the P&T approach, the subject/non-subject assymetry in root questionsreduces to a straightforward calculation of derivational economy, given the presence of a(‘strong’) [Tense] feature in C.

It is worth paying attention from the start to what is essential and what is not inworking out this analysis. The basic idea is that there is some checking relation betweenC and a feature found on both the subject and T. As C could in principle check either the

Page 13: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 12

subject or T, some other principles will have to determine which choice is made in anygiven context. And these other principles come from the syntax of wh-movement andfrom the precise implementation of the EPP.

Notice that the nature of the feature being checked in this analysis is entirelysecondary. Although P&T have interesting arguments to support their claim that therelevant feature is [Tense], nothing in the logic of the analysis depends on this choice. Ifthe feature were labelled as [Finite], or as [nominative], or as [martian], the analysis ofthe subject/nonsubject assymetry would be unaffected.3

The category label of the upper head in this account is also something which couldbe altered without affecting the essence of the analysis. Various authors have claimed thatthe subject position in root clauses can sometimes serve as a landing site forwh-movement (Rizzi, 1990b; Pesetsky, 1988). If so, then the P&T analysis could berestated with phrase structures in which the upper head is AgrS instead of C, for example.

(5)AgrP

����

HHHH

wh-subject Agr′

���

HHH

AgrS[Tense]

TP�� HHt T′

��HHT vP

vs. AgrP

������

HHHHHH

wh-objectetc.

Agr′

�����

HHHHH

AgrS�� HH

T AgrS[Tense]

TP��� HHH

subject T′

��HHT

t

vP

Alternatively, the upper head could be identified with Rizzi’s (1997) head, withShlonsky’s (1994) AgrC, with Branigan’s (1992)Π, with Culicover’s (1991) PolP, or withany number of others. The logic of the P&T analysis is largely independant of this choiceof labels, too, a fact which will be exploited as we proceed.

Embedded clauses

The assymetry in root questions is, for the most part, absent in embedded questions inEnglish, where auxiliary verbs remain in TP in all wh-questions: (6).

3In Holmberg and Platzack’s (2003) account of Scandinavian verb-second word order, P&T’s [Tense]feature is explicitly replaced by a [Finite] feature.

Page 14: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 13

(6) *John asked which tent did Sheila pitch.

There are several ways this root/embedded clause assymetry could be dealt withwhich are consistent with the P&T analysis (or at least the part of that analysis presentedabove). It could be stipulated that the [Tense] feature is present only in root questioncomplementisers. Obviously, this is the least interesting way to deal with the problem. Infact, accepting this stipulation would undermine the analysis of root questions, as thestipulation does little more than restate the general problem in other terms, unless aprincipled explanation for the limited distribution of the [Tense] feature can be found.

A better alternative would be to suppose that the [Tense] feature is found in both rootquestions and embedded questions, and that the root/embedded clause assymetry reflectssome other facet of the grammar. The best theory would maintain that the [Tense] featurefound in root questions is found in declarative root clauses too, as well as in declarativeand interrogative embedded clauses. In other words, if [Tense] is a property of root whcomplementisers, then it should be a property ofall complementisers. But then theproblem of the assymetry between root and embedded questions becomes even moregeneral, and more interesting.

P&T themselves propose an ambitious variant of this best theory, with the followingcharacteristics:

(7) The P&T ‘best’ theory

a. C always bears the [Tense] feature, which forces movement under checking.b. C always checks either the subject or T to check [Tense].c. T movement to C may be phoneticallyrealizedin different ways.d. In non-root declarative clauses, T-to-C is typically realized as placement of

that in the C position.e. In embedded questions, C checks [Tense] without movement of the target.

This is dialectally variable.

The flavor of this part of the P&T theory can be appreciated by examining a fewsentences.

(8) Peter said [CP that-∅ [TP the tent would be sturdy ]]

In the embedded CP in (8), the original head is an null C, which bears [Tense] and whichmust check [Tense] by attracting a matching [Tense]-bearing target. It selects the head ofits complement–the auxiliary verbwould—to check, and T then raises to C. But T-to-Cmovement in an embedded clause is subject to a special Spell-Out rule. Instead ofwouldbeing Spelled-Out as itself in the C position, a neutral Tense morphemethat appears in C

Page 15: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 14

instead, andwould is interpreted (aswould) in T. With the [Tense] feature of C nowchecked, the derivation can continue on by building matrix clause structure around CP.

Consider now a case of subject extraction:

(9) Which tent did Peter say [CPt ∅ [TP t would be sturdy ]]

In (9), as in (8), the head of the embedded clause is a null complementiser. Once again, Cmust check its [Tense] feature by attracting a [Tense]-bearing target from within TP. Thistime, however, the subject is a wh-phrase which must raise out of CP into the matrixclause with successive cyclic wh-movement. So the embedded C must attract the subjectto Spec-C to allow the derivation to converge, regardless of the need to check [Tense].And since the subject must raise to Spec-C, it may also, and economically, serve as thetarget for [Tense] checking by C. The auxiliary verb has no need to raise to C, and therewill consequently be no realization of T in C asthat.

The remaining case to examine is that of embedded questions:

(10) Peter asked [CP where∅ [TP the tent had stood ]]

Again, embedded CP is headed by a null complementiser. In this case, though, C does notrequire that its [Tense]-bearing target raise. C checks T (or the subject)in situ, and therest of the derivation takes place as usual.

The most exciting aspect of P&T’s analysis of embedded clauses is the elegantaccount of thethat-trace effect provided by the treatment of (9). This account comes atsome expense, however—principally in the idea thatthat is a morphological variant of Trather than a true complementiser. The problem is not that this claim is incoherent, oreven difficult to instantiate mechanically. Rather, it is difficult to see on what basisthat,and its equivalents in other languages, should be isolated from the other apparentcomplementisers.If, for example, appears to have as strong a claim on being a Spell-Outvariant of T asthat has. In conditional clauses, for example,if alternates with T-to-Cmovement: (11).

(11) a. If we had only shovelled the driveway earlier, we would have caught theplane.

b. Had we only shovelled the driveway earlier, we would have caught theplane.

In the absence of some sort of reliable diagnostic, it is hard to evaluate what factorsshould be relevant to treating a “complementiser” as C rather than T. And when this issueis problematic for the theoretician, it is likely to be far more so for the child faced withthe problem of categorising the words of her language.

Page 16: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 15

Suppose however that we grant that there is no principled reason whythat could notbe an instance of T. The question may still be raised whether we can find any independantevidence which can show whetherthat behaves like an element which has undergonehead-movement to C. In fact, German topicalisation structures provides a test for exactlythis question. (See Müller and Sternefeld (1993) and Staudacher (1990) for discussion.)And the German data shows that clause (7-d) of P&T’s theory cannot be maintained.

The argument is relatively simple. The first thing to understand is that longtopicalisation in German is sensitive to intervening structure in a way that longwh-movement is not.

In German, topicalisation may extract a topic from Spec-C in an embeddedverb-second CP, as in (12).

(12) Inin

zweitwo

wocheniweeks

glaubtbelieves

Anna,Anna

[CP ti hathas

MaxMax

gesagt,said

[CP ti werdewill

sieshe

ti

kommen.come

]]

Movement from one topic Spec-C position to another topic Spec-C position is subject toa special constraint: long topicalisation from a verb-second clause cannot use anintermediate specifier of an non-verb-second CP as an escape hatch in successive cyclicmovement.

Let us use the following labels to distinguish the verb-attracting C in topicalisationsfrom the independent C (daß, ob, etc.) in simple declaratives: topicalising C will be Ctop,and conventional C will be Cα. Movement from a non-verb-second clause can target aCtop clause, as in (12), but any movement from a Ctop clause to a Cα clause is impossible.Thus, (13) is ill-formed because movement from the lowest clause to the matrix has madeuse of the non-verb-second Spec-Cα in the intermediate clause.

(13) *Ankei

Ankesagtesaid

sieshe

[CP ei daßthat

erhe

glaubebelieves

[CP ti werdewill

ihmhim

seinehis

Arbeitwork

hierhere

bezahlen.pay

What is the significant difference between the Cα and the Ctop clauses? Evidently,Ctop attracts a verb while Cα does not. And this difference has consequences forsuccessive cyclic movement—consequences which we will pay close attention to inchapter 4. For the moment, however, it is enough to recognize that T-to-C movement orits absence has this effect.

Page 17: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 16

Now consider how the contrast between (12) and (13) will appear under a P&Tanalysis. Let us suppose thatdaßis equivalent to Englishthat, as a morphological variantof T. Then the structure of (13) will be (14).

(14) [CP *Ankei sagtej-∅ [TP sieej [CP ei daßk [TP er glaubek [CP ti werdel-∅ [TP ti el

ihm seine Arbeit hier bezahlen.]]]]]]

There are three T-to-C operations in the derivation of (14). The verbssagteandwerderaise to C in the usual manner; they are Spelled-Out in the C position, and their traces arenot pronounced. The verbglaubealso raises to C, but the result of this operation isrealized differently. The T in C is pronounced asdaß, and its trace is pronounced as theverb itself.

This pattern of constraints is not restricted to this language. In Belfast English,inversion is always possible in embedded yes-no questions. As Henry (1995) shows,inversion in such questions is allowed only if there is no overt complementiser, indicatingthat the verb raises to the argumental C.

(15) We couldn’t establish did he meet them.

‘We couldn’t establish whether he met them.

Inversion is possible with embedded wh-questions for some, but not all speakers ofthis dialect. (The examples which follow are all taken from Henry (1995).)

(16) He didn’t say why had they come.

In addition, and most clearly in contrast with other dialects of English, inversion canbe triggered by successive cyclic wh-movement, in a manner reminiscent of longwh-movement in German. And the long topicalisation in this dialect is subject to thesame TDC effect as long topicalisation in German, as shown by ((17-c)) (Alison Henry,personal communication).

(17) a. What did John hope would he see?b. Who did John say did Mary claim had John feared would Bill attack?c. *Who did John say did Mary claim that John feared would Bill attack?

The problem now is that there is no difference between the derivations whichproduce the grammatical long topicalisation and those like (16), which fail. If T-to-Ctakes place in all embedded clauses, then it is impossible even to describe the difference

Page 18: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 17

between the two types of sentences, let alone explain them.4

To sum up, the P&T analysis provides a compelling explanation of thesubject/non-subject assymetry in root questions, but the portions of this analysis whichhave to do with interrogative/declarative assymetries and root/embedded clauseassymetries cannot be maintained when a broader range of data is examined.

2.3 The difference between root and embedded questions

The landing site in root questions

Let us attempt a wider perspective on these issues. The English inversion found in rootquestions is also found in the other Germanic languages, where it appears to belong to alarger pattern of verb-second word order in clauses in which A-bar movement occurs. Itstands to reason that we may find a clearer picture of the particularities of the Englishcase if we understand the larger Germanic pattern better first.

Familiar reasoning about the root/embedded clause distinction goes like this: Weknow what wh-movement is—wh-movement is movement of a wh-phrase from withinTP to the Spec-CP position. This movement always does the same thing, because italways is driven by the same forces: C must check the [wh] feature on the wh-phrase, andC forces overt movement when it checks. But T-to-C movement accompanieswh-movement in (some) root clauses, but not in embedded clauses, so there must besome difference in the grammatical properties of root clauses.5

I want to call into question the first statement in this reasoning, the idea that weknow that wh-movement is always movement to the Spec-C position. In fact, it is likelythat this is not the case, and that wh-movement in root clauses is not the same operationas wh-movement in embedded clauses, at least in the Germanic languages. The positionto which a wh-phrase raises in root clauses is normally a different location from thelanding site for wh-phrases in embedded clauses.

4It would be possible to defend the P&T theory by claiming that Germandaßis not equivalent toEnglishthat. Germandaßwould be a true complementiser, in this view, rather than T raised to C. But theGerman constraint on long topicalisation also shows up in long wh-movement, as discussed below. AndBelfast English patterns with German. Since it is fundamental to the P&T argument that Belfast Englishthat be equivalent to standard Englishthat, the text argument cannot be avoided.

5This reasoning is hardly universal. Exceptions to this type of reasoning include Rögnvaldsson andThráinsson (1990) and Pesetsky (1988). (Pesetsky maintains that the subject position may serve as anA-bar position, so that wh-phrases may actually raise to Spec-IP under the right circumstances.)

Page 19: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 18

Long wh-movement in German

One type of evidence for distinct landing sites for wh-phrases in root clauses comes fromthe German constraint on successive cyclic movement discussed above. The generalpattern is that successive cyclic movement permitting long topicalisation may raise atopic from the specifier of a Cα to the specifier of a Ctop, but the reverse is not true. Andthe same pattern can be identified in some cases of long wh-movement in questions. Thusin (18), the wh-phrasewannraisesvia the Spec-Ctop of the complement clause into thematrix clause. In this respect, the landing site for root wh-movement is like that used intopicalisation.

(18) Wannwhen

hathas

FritzFritz

gesagt,said

t werdewill

sieshe

kommen?come

‘When did Fritz say she would come?’

In embedded questions, however, interrogative C behaves like a Cα complementiseras far as long movement from a verb-second clause is concerned. Successive cyclicwh-movement is possible in embedded clauses, but movement from Spec-Ctop to a higherembedded interrogative Spec-C is impossible.

(19) *AnnaAnna

fragteasked

mich,me

wanniwhen

ihnenyou

FritzFritz

gesagtsaid

hat,has

ti werdewill

sieshe

ti kommen.come

The patterning of interrogative C with Cα is also supported by the word order foundin sentences where both overt complementisers and wh-phrases appear. (German andDutch allow violations of the ‘doubly-filled Comp filter’.) (Examples are taken fromMüller (1995) and Zwart (1992), respectively.)

(20) a. IchI

weißknow

nichtnot

wenwhom

daßthat

sieshe

geküßtkissed

hathas

b. IkI

weetknow

nietnot

wiewho

ofdatif-that

MarieMarie

gekustkissed

heeft.has

The wh-phrase always precedes the complementiser, which means that the most likelyposition of the wh-phrase is the specifier for the argumental complementiser.6

6It is also possible to suppose that embedded questions of this type include two ‘argumental CPs’: theCP which hosts the wh-phrase and a second CP headed by the overt complementiser, as proposed by Zwart(1997). The structure of (20-a) would then be (i).

(i) Ich weiß nicht [CP wen [CP daß [CP sie geküßt hat ]]]

Page 20: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 19

In root questions, in contrast, wh-phrases can be raised freely from Spec-Ctop of anembedded clause: (21-b).

(21) a. (AnnaAnna

glaubt,)believes

inin

zweitwo

Wocheniweeks

hat/(habe)has

MaxMas

gesagt,said

ti werdewill

sieshe

ti

kommen.come

b. Wanniwhen

meinstthink

Duyou

denn,ti hathas

ihnenyou

FritzFritz

gesagt,said

ti werdewill

sieshe

ti kommen?come

Thus embedded interrogative C patterns with Cα and root interrogative C patterns withCtop.

The parallel behaviour of root wh-phrases and topics is also seen in (22), where thepresence of an intervening non-verb-second clause blocks long movement from Spec-Cof a complement clause.

(22) a. *Weriwho

sagtesaid

sie,she

daßthat

erhe

glaube,believes

ti werdewill

ihmhim

seinehis

Arbeitwork

hierhere

bezahlen?pay

b. *Ankei

Ankesagtesaid

sie,she

daßthat

erhe

glaube,thinks

ti werdewill

ihmhim

seinehis

Arbeitwork

hierpay

bezahlen.

The simplest account for the fact that root wh-phrases and topics follow the same patternwill start from the premise that the landing site of both phrases is the same position, aposition which is then distinct from the one used by wh-phrases in embedded questions.

Yiddish assymetries in short wh-movement

Yiddish offers a second, transparent argument that Germanic root wh-phrases raise to aposition with the properties of a Spec-Ctop position. In Yiddish, and marginally inEnglish, a wh-phrase in an embedded question can co-occur with a preposed topic(Diesing, 1990; Baltin, 1982; Rochemont, 1989).

(23) a. Zishe

isis

gekumencome

zento-see

verwho

frierearlier

vetwould

kontshen.finish

(Diesing, 1990)

b. the fellow to whom never had I written any letters

The wh-phrase here appears in the specifier of the interrogative CP (headed by awh-checking Cα, and the topic appears in a Spec-Ctop. Since the two preposed phrases

Nothing in the present work turns on the exact details of this type of question, so I adopt the less complextext analysis.

Page 21: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 20

evidently occupy different positions in these structures, the null hypothesis is that theyoccupy the same positions in simpler clauses, as well.

The happy coexistence of a wh-phrase and a topic in a stacked series of CPs isimpossible in matrix clause (Diesing, 1990).

(24) a. . . . vias

ikhI

veysknow

voswhat

bayby

mirme

tutdoes

zikhitself

(from Moorcroft (1995))

‘. . . as I know what goes on with me’b. *Ver

whohaynttoday

hothas

gegesneaten

dosthe

broyt?bread

c. *Voswhat

hothas

demto the

rovrabbi

MaxMax

gegebn?given

The ungrammaticality of (24-b) is to be expected if wh-movement in root clauses targetsSpec-Ctop, rather than Spec-Cα. Yiddish, like other Germanic languages, does not allowmultiple topics for a single clause. The root/embedded clause contrast remainsmysterious, however, if we insist that root questions involve movement of the wh-phraseto Spec-Cα.

The Yiddish pattern can be replicated in English, where monotone decreasing topicscan also co-occur with wh-phrases in embedded questions.

(25) a. It was with Mary at the helm that at no time was I concerned about mysafety, not with John.

b. ?The chief would like to know with whom at the helm at no time were youconcerned about your safety.

c. *With whom at the helm at no time were you concerned about your health?

Not only does the data discussed here indicate that root and embedded wh-phrasesoccupy different positions, but the position of overt complementisers in German and ofwh-phrases in Yiddish and English also allows us to identify the relative heights of thesetwo positions: (26).

Page 22: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 21

(26) CP

�����

HHHHH

embeddedwh-phrase

C′ 1

����

HHHH

Cα CP

����

HHHH

rootwh-phrase

C′

�� HH

Ctop clause

When both Cα and Ctop appear in the same clause, Cα evidently takes CtopP as itscomplement. Either type of complementiser can evidently permit a wh-phrase to serve asits specifier.

2.4 Deriving the different landing sites for wh-movement

Why should it be the case that root and embedded questions provide different landingsites for wh-phrases? The answer can be found, I believe, in examining the basicmechanisms of phrase construction. Following a long tradition, rooted in the generativesemantics literature of the late 1960s, I will argue that root clauses differ from embeddedclauses by virtue of the content which wedo not hear. In root clauses, the landing site ofa wh-phrase is lower down than in embedded clauses, because a root clause includesperipheral syntactic structure which is not interpreted at PF, and which is thereforeinaccessible as a target for movement.

This general approach has been developed in a number of ways in the recentliterature, too, including Branigan (1996a), Boškovic (2001), and Plunkett (1997). Thespecific implementation of this idea which I will pursue derives the presence of a silentroot C from the theory of phases and multiple Spell-Out.7

Cyclic Spell-Out and its implications for clause structure

Deferred phase Spell-Out

At some point in the derivation of any linguistic expression, the hierarchical ordering ofphrases must give rise to a linear ordering of phonetic units. In other words, we have to

7Fitzpatrick (2003) also derives the invisibility of some root content from phase theory, under slightlydifferent assumptions about the interpretation of root content at LF.

Page 23: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 22

be able to say our sentences, and any model of grammar should say something about howthis is possible. Non-minimalist transformational literature since Chomsky (1965)generally supposed that the pronunciation of phrase markers (Spell-Out) is determined onthe basis of the S-structure, a special level of representation which follows the overttransformations and precedes the covert transformations. However, if there is no suchprivileged level of representation (as I assume), then the point at which Spell-Out takesplace must be determined on other grounds, as Uriagareka (1999) observes.

In Chomsky’s (1993) model, Spell-Out is ordered after all operations which check“strong” features are concluded, with Spell-Out understood to affect an entire phrasemarker at once. But as Uriagareka (1999) points out, there is no reason to suppose thatthis is the case; given the fact that other operations take place before the entire root phrasemarker is constructed, Spell-Out would be expected to follow suit. In fact, a “multipleSpell-Out” model, in which Spell-Out affects relatively small parts of the tree in isolation,would appear to be the null hypothesis.

The question which now must be addressed is what parts of the tree will be subjectto Spell-Out, and when. Chomsky’s (2001b) proposal is that Spell-Out affects (strong)phases, the units of structure which are derived on the basis of individual numerations.(Phases as units of structure are independantly motivated in his analysis by localityconstraints (PIC) and by the distribution of expletives.) Although the full set of phasecategories is not yet clear, at least CP must count as a phase. (Chomsky includestransitive vP in the set of strong phase categories, too, but this is incidental to theargument I will make here.) Crucially, however, TP is not a phase. This distinction in thestatus of CP and TP is reflected in the distribution of expletives in embedded finiteclauses (which are always CP) and ECM infinitives.

(27) a. There was thought [TP t to have arrived an important message ]b. *There was thought [TP an important message[] to have arrivedt ]

(28) ?There were told several diplomats [CP that [TP a message had arrivedt ]]

In (27), the complement to the matrix verb is TP. In (27)[b], we see that thecomplement ofarrive cannot raise to Spec-TP, because there is an expletivethereavailable to fill the same position, and Merge is preferred to Move by the generaleconomy metric. However, in (28), the presence of atherein the matrix clause does nothave the same effect. Chomsky concludes from this that the Numeration used inconstructing an ECM complement is the same array as the matrix finite clause has accessto, while finite complement clauses have their own Numerations, which are exhausted bythe type CP is built up. In other words, TP is not a phase, while CP is one. In a complexsentence like (29), then, Spell-Out will not apply to the stringthe rain would fall in

Page 24: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 23

Spain; only to the stringthat the rain would fall in Spain.

(29) Higgins insisted [CP that [TP the rain would fall in Spain ]]

The identification of the CP phase as a constituent which is subject to Spell-Out iscomplicated by the fact that A-bar movement can extract material from CP into a higherclause, as occurs in successive cyclic wh-movement, for example. Such movement formsa chain in which only the head is normally pronounced. The problem can be seenimmediately with an example like (30).

(30) Whatdid they claim [CP whatthat [TP Claire had repairedwhat]]

The wh-phrasewhat raises from its base object position to become the specifier for CP inthe complement clause. If Spell-Out of the entire CP phase were to take place at thispoint,whatwould be pronounced at the left edge of the complement clause, rather than inits final position. In general, in fact, the idea that phases are subject to Spell-Out impliesthat movement will be possible only inside a phase, and that no successive cyclicmovement should occur.

Chomsky’s solution to this problem is attractive. He proposes that Spell-Out of aphase is deferred until the point in the derivation at which it can be determined whetheranything will be extracted from the phase in question. Since extraction typically raisesmaterial to the edge of the next higher phase, this means that Spell-Out of any one phasewill take place when the derivation of the next higher phase is completed. Chomsky’sformulation is (31).

(31) . . . where PH1 is strong and PH2 is the next highest strong phase:

(10) Interpretation/evaluation for PH1 is at PH2.

In (30), for example,whatwill be Merged into the specifier position of the root clause vPphase in order to eventually reach its final position in the root CP. At the point whereSpell-Out of the complement clause takes place, the structure will be (32).

(32) [vP what [ v′ they claim [CP whatthat [TP Claire had repairedwhat]]]]

With this much structure in place, thewhat trace in Spec-C of the complement clause canbe identified as something other than the head of its A-bar chain, so that it will not bepronounced.

Page 25: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 24

Root clause Spell-Out

Chomsky’s characterisation of how phases are interpreted carries its own set ofconsequences to be explored. One consequence in particular is pertinent to the questionswhich concern me here, and to the distinction between root and embedded clauses. TP isnot a phase, under Chomsky’s formulation. In fact, TP cannot be counted as a phaseunless the very basis for the distinction between phases and other categories is thrownout. As phases are the unit of structure which are subject to Spell-Out, the clearimplication is that TP cannot be subject to Spell-Out. But root clauses are certainlysubject to Spell-Out, since they are actually said. The only conclusion which can bedrawn is that root clauses are not TP categories—they must be something larger,something which contains a TP category, but which is itself a phase. Let us refer to thephase which undergoes Spell-Out in a root clause asΠP, and to its head asΠ (for theP(hase) at the top).8 Then the structure of a simple sentence must minimally be (33).

(33) ΠP�� HH

РTP�� HH

D T′

��HHT vP

The labelΠ is not to be understood as referring to a distinct syntactic category. It is anotational convenience, which I want to use to point to an existing syntactic category, thefeatures and properties of which we have not yet determined. We have established thatΠis situated outside TP—the question is then what categoryΠ belongs to.

There are three possibilities, given the two phase categories which Chomskyproposes: vP and CP.Π might bev. This seems unlikely, since v elsewhere requires a VPcomplement.Π might be C. OrΠ could be something else, which we have no idea about.This third alternative is clearly the worst case, so we should reject it until we haveexhausted the alternatives. The second choice is then the preferable one:Π shouldtentatively be identified as a variety of C. (In this respect,Π is quite similar to Culicover’s(1991) Pol head.) Root clauses are then simply CPs, which are subject to Spell-Out in thesame way as other CPs are. (It is worth noting that this conclusion is entirely consonantwith the widespread view, dating from Den Besten (1983), that Germanic rootverb-second clauses are CP structures. To my mind, there has been no sound explanationwhy root clauses in the verb-second languages should have to be CP, if other languages

8The labelΠ was introduced to refer to a similar head in Branigan (1992). My readers will no doubtappreciate how prescient that notation was in anticipating the later development of phase theory.

Page 26: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 25

allow TP root clauses. This mystery is resolved, of course, if root clauses can only beCPs, and the pan-Germanic pattern is then only to be expected.)9

But now a second problem arises. Phases are not interpreted immediately; phaseSpell-Out is deferred until the next strong phase is completed. Even if a root clause is aCP, then, it will not undergo Spell-Out unless it can be included in still another phase.What is more, the topmost phase cannot undergo Spell-Out at all, since it will never bepart of a higher phase. The consequence of Chomsky’s principle (31) is that any rootclause must include a silent head which triggers Spell-Out on the lower CP phrase, butwhich can never be interpreted itself.

As for the features of this necessarily silent head, the same reasoning as beforesuggests that we take it to be a complementiser rather thanv or some novel category.Root clauses then always make use of a CP-recursion structure, simply because there isno other way to have Spell-Out apply.

(34) CP

���HHH

C(silent)

ΠP�� HH

РTP�� HH

D T′

��HHT vP

The true root of a root clause is then an obligatory silent C. It may be that this C can beidentified with the Fregean ‘force markers’, as Chomsky (1995, chapter 4, p. xx) suggestsfor a similar postulated invisible root C.10

Consider now the structure of embedded clauses in the light of the precedingdiscussion. With the exception of ECM complements, embedded clauses are typicallyCPs. The conventional assumption has always been that—in the simplest case, withdeclarative force—C takes only a TP complement. Given the structure just deduced forroot clauses, an alternative structure now suggests itself for embedded clauses, parallel tothe CP-recursion structure in root clauses: (35).

9The subject/non-subject assymetries in verb-second root clauses documented by Travis (1984) andZwart (1993a) must then be derived from some other principles. See section 2.6 for discussion.

10Cf. also Boškovic (1997) for discussion of the checking properties of invisible root C.

Page 27: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 26

(35) VP�� HH

V CP�� HH

C ΠP�� HH

РTP�� HH

D T′

��HHT vP

The reasoning which leads to the CP-recursion structure in root clauses is withoutforce in the case of embedded clauses, which will be subject to Spell-Out whether or notthey have multiple CPs in place. On the other hand, the fact that recursive CP structure isinevitable in root clauses is itself informative. In the absence of evidence to the contrary,we assume that root clauses and embedded clauses will share the same structure. Thestructure (35) therefore appears to embody the null hypothesis for embedded clauses,because it makes no structural distinction between root and embedded CP structure.11

Wh-movement in root questions

We are now in a position to understand why root and embedded wh-questions mightprovide different landing sites for wh-movement. Chomsky’s Deferred Phase Spell-Outtheory ensures that a root C can never be subject to Spell-Out, because it never has ahigher phase to trigger the operation. The same will be true of any specifier for a root C.Therefore, in principle, a root C should not attract any category which needs to undergoSpell-Out in the course of the derivation. The structure (36) is then impossible in rootquestions.12

(36) CP

���HHH

wh-phrase C′

�� HH

C clause

11Iatridou and Kroch (1992) argue that CP recursion is permitted only in the special contexts whereverb-second clauses appear. This constraint is evidently not compatible with the approach developed here,where CP-recursion is wide-spread, and in fact the norm for embedded clauses.

12The corollary is that structures which appear to match the (36) pattern cannot be true root clauses, butmust be treated as subordinate to some invisible higher head. This will be necessary for exclamatoryclauses, amoung others.

Page 28: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 27

Nothing prevents the structure (36) from being used as an embedded question, however,as is typically the case.

But wh-movement evidently occurs in root clauses—at least, a wh-phrase alwaysraises to an A-bar position outside of TP. If this movement does not take place to providea specifier for the interrogative complementiser, then why should it take place at all? Aplausible answer again comes from phase theory, and from thePhase ImpenetrabilityConstraint, in particular.

ThePhase Impenetrability Constraintis Chomsky’s implementation of the‘subjacency’ principle in phase theory.

(37) Phase Impenetrability ConstraintFor [ZP . . . [HPα [ H YP ]]] , with ZP the least strong phase, the domain of H isnot accessible to operations at ZP, but only H and its edge.

This constraint enforces successive cyclic movement for long A-bar movement,among other things.

Consider the derivation of an ungrammatical root question like (38), in whichwh-movement does not occur.

(38) *[CP ∅ [ΠP Π [TP Beth [vP pitchedwhich tent]]]]

We understand whywhich tentin (38) cannot raise to Spec-CP. This movement wouldcause the derivation to crash at PF, because the specifier of the silent root C cannot beSpelled-Out. But why is it impossible to leave the wh-phrasein situ?

The answer is presumably the same for root clauses as for embedded clauses. Inboth cases, the interrogative C has a [wh] feature to check, and it cannot do so if thetarget containing a [wh] feature is too far away. C in (38) must still check its [wh] featureagainst the matching feature inwhich tent, but thePhase Impenetrability Constraintensures that C cannot do so.

Chomsky proposes a general derivational strategy to allow movement to take placewhen thePhase Impenetrability Constraintwill block successful derivations. Consider asentence like (39), with an interrogative embedded C which attracts a wh-phrase frominside a more deeply embedded declarative clause.

(39) We were wondering [CP which tent C [ Paul believed [CP that [ Beth pitchedt]]]]

In order to see how thePhase Impenetrability Constraintcan be satisfied in general,let us ignore all phase boundaries except the lowest CP. In this sentence,which tentoriginates inside the bottom CP. If nothing other than normal Merge operations occurs in

Page 29: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 28

the construction of the upper embedded clause, then at some point in the derivation of(39), the structure (40) will be formed.

(40) [CP C [ Paul believed [CP that [ Beth pitched which tent ]]]]

At this point in the derivation, root C must check its [wh] feature against an accessibletarget. But C cannot checkwhich tentin (40) because the complement clause is a phase,and the root C does not have access to any part of the structure past the headthat of thelower phase. The derivation must crash.

In order to allow for a successful derivation of (39) to be constructed, a rescuemechanism must be triggered at an earlier point in the derivation. So now consider thesituation when the bottom clause is first constructed: (41).

(41) [CP that [ Beth pitched which tent ]]

In this structure, there is no complementiser which needs to check a [wh] feature, butthere is a wh-phrase which ultimately needs to be checked. Even though there is no wayto permit the [wh] feature to be checked immediately, it is clear at this point in thederivation, that the derivation cannot converge unless the wh-phrase is somehow madeaccessible to a checking operation at a later point. Chomsky suggests that this situationcan be remedied by a rescue operation which introduces a new feature at the root of thephrase marker in order to force displacement of the wh-phrase to the edge of the phase.Some (defective) variant of the [wh] feature is added to the non-interrogativecomplementiserthat, and this feature then checkswhich tent, and forces it to raise toSpec-C at the same time.

Given this rescue operation, the situation is quite different when the next phase isconstructed. Rather than (40), the initial structure of the upper embedded clause will be(42).

(42) [CP C [ Paul believed [CP which tent that [ Beth pitchedt ]]

And now thePhase Impenetrability Constraintwill not interfere when the interrogativeroot C tries to check the [wh] feature onwhich tent.

If transitive vP is a phase, as well, as Chomsky suggests, then the derivation willsimply require more rescue operations to allow the wh-phrase to reach its ultimatelanding site. In that case, a [wh] feature must be added to all instances of v in the pathbetween the interrogative C and the wh-phrase at the points in the derivation where vPphases are constructed. And the wh-phrase will be attracted by these extra [wh] featuresto the edge of vP, as well as the bottom CP. In general, extra phases will simply mean thatextra derivational rescue operations are necessary to allow wh-movement to take place

Page 30: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 29

successfully.Now let us compare the case of (38) to the long wh-movement derivation in (39).

In (38), the silent root C must check its wh-feature, but it cannot do so because of thePhase Impenetrability Constraint. The first phase boundary abovewhich tentis the vP.Suppose that the general rescue operation which allows successive cyclic movementtakes place with vP. Then the structure of (38) will be altered to (43), because the [wh]feature added to v will attract the wh-phrase to the edge of vP.

(43) *[CP ∅ [ΠP Π [TP Beth [vP which tent[ v′ pitchedt ]]]]]

With this structure, vP no longer prevents C from checkingwhich tent. But theΠPphase is still in place, and thePhase Impenetrability Constraintwill again block C fromchecking its [wh] feature. In order to permit the derivation to converge, the same rescueoperation must be triggered at theΠP phase level. The (defective) [wh] feature willtherefore be added toΠ, and the wh-phrase will raise to Spec-Π, producing the structure:(44).

(44) *[CP ∅ [ΠP which tent[Π′ Π [TP Beth [vP t [ v′ pitchedt ]]]]]]

At this point, root C can check its [wh] feature, as necessary.For root subject wh questions, the logic will be similar, although the derivations will

be less involved. For a sentence like (45-a), for example, the derivation produces (45-b)as the first phase above TP.

(45) a. Which campers pitched that tent?b. [ΠP Π [ which campers T [vP t pitched that tent ]]]

As this phase includes the determinerwhich, it is clear at the point where (45-b) isformed that the derivation will fail unless the wh-phrasewhich campersis displaced tothe left edge ofΠP. A defective wh feature will therefore be added toΠ—the head of thephase—and this feature may then attract the subject to Spec-ΠP, producing (46).

(46) [ΠP which campersΠ [ t T [ vP t pitched that tent ]]]

The (46) structure is not subject to Spell-Out immediately, because this operation isdeferred until the next phase is formed. It is at this point that the silent interrogative Cmay be Merged with the (46) structure, forming (47).

(47) [CP [C ∅ ] [ ΠP which campersΠ [ t T [ vP t pitched that tent ]]]]

Page 31: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 30

The silent C carries the (non-defective) wh-feature which is to check the features ofwhich. As whichappears at the periphery of theΠP phase, nothing blocks the wh-featurechecking operation from taking place at this point. And then, with all uninterpretablefeatures disposed of, Spell-Out can apply toΠP, producing the sentence (45-a).

Notice that it cannot be the case that interrogative C must dispose of an EPP featureto allow the derivation to converge. This necessary component of the analysis could beimplemented in a number of ways. We could simply stipulate that silent root C lacks anEPP feature, whether or not it is interrogative. There is a lessad hocway to obtain theright result, however. Chomsky (1995) suggests that ‘strong’ features, i.e. EPP features,are properties of a head which must be eliminated before Spell-Out can interpret thathead. If we understand EPP features in this way, then even if the interrogative root Cbears an EPP feature, it will never trigger movement, simply because Spell-Out does notinterpret root C anyway. There appears to be no empirical evidence which militatesagainst this interpretation of the EPP effect, so I will adopt it in the following form:

(48) The EPP interpretation principle(EIP)EPP features force movement (re-Merge) of a checked target before Spell-Out.

Wh-movement in embedded questions

Suppose now that example (45-a) is to be used as an embedded question, as in (49):

(49) The park warden asked us which campers pitched that tent.

The derivation of the question itself will be identical to that of the rootquestion (45-a) until the embedded interrogative complementiser is Merged. Thecomplementiser in this case is subject to Spell-Out, together with any specifier it mayacquire. As such, an EPP feature on C will force movement of the phrase which Cchecks. Given the structure (50-a) , for example, where C must check the wh-feature ofthe subject determiner, wh-movement to Spec-CP will be necessary to satisfy the EIP.

(50) a. [CP [C ∅ ] [ ΠP which campersΠ [ t T [ vP t pitched that tent ]]]]b. [CP [C which campers∅ ] [ ΠP t Π [ t T [ vP t pitched that tent ]]]]

The CP phase will therefore end up as (50-b).13 And this CP will be interpreted in thisform by Spell-Out when the next higher phase is completed.

13In fact, in this particular case, the situation will turn out to be more complicated when thethat-traceeffect is re-examined in chapter 4.

Page 32: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 31

2.5 Fixing the P&T analysis

Let us return now to the subject/non-subject assymetry, and to the P&T approach to thisanalytic problem. Remember the nature of the problem for this approach which wasidentified in section (5). The P&T account of auxiliary-inversion in root questionsidentifies a [Tense] feature in C which must be checked by movement of either thesubject to Spec-C or of the T head itself to C. But auxiliary movement to C does notappear to take place in embedded clauses, and the idea thatthat is a variant of T wasshown to be untenable. And the word order in embedded clauses shows clearly thatnon-wh subjects do not normally raise to become specifiers for CP.

Consider now how that P&T approach can be accomodated with the more complexstructures for root and embedded questions which I have presented here. In rootquestions, both wh-movement and auxiliary inversion must take place withinΠP, and thetrue interrogative complementiser does not participate in the derivation, at least withrespect to any movement operations. The head which must bear the [Tense] feature toattract T in auxiliary inversion will therefore beΠ, rather than C. And, as in the P&Tanalysis, the [Tense] feature ofΠ is accompanied by the EPP property, so checking ofthis feature will force movement of the checking target.

Root wh-questions

Let us return to the (51) root questions once again, and see what impact the idea thatΠbears uninterpretable [Tense] has on the derivation.

(51) a. Which campers pitched the tent?b. Which tent did Sheila pitch?

The derivation of (51-a) will contine as before up to the point at whichΠ is Merged withTP, generating the (52) structure.

Page 33: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 32

(52) ΠP

������

HHHHHH

Π[wh]

TP

������

HHHHHH

DP

�����PPPPP

which[wh]

campers

[Tense]

T′

����

HHHH

T[Tense]

vP

�����PPPPP

t pitched the tent

At this point in the derivation, T has already checked the [Tense] feature on the subjectDP, but the feature has not yet been erased from the subject, and it will not be erased on Tat all. Both the subject and T are therefore accessible targets forΠ to check its own[Tense] feature against. What is more, the [wh] feature ofwhichremains unchecked, andwill need to be moved to the edge of the phase in order to be accessible to checkingoperations at a later point in the derivation.

So there are two problems to be solved at this point in the derivation: one involvingthe [Tense] feature, for which there are two potential solutions, and one involving a [wh]feature, for which there is only a single solution. Given this situation, it is clear what themost economical solution will be. IfΠ checks the [Tense] feature of the subject, andattracts it to Spec-ΠP, then both problems are solved at once. The uninterpretable featureof Π is checked, and the wh-phrasewhich campersis displaced to the edge of the phase.In contrast, ifΠ checks T and T raises toΠ, then further action will be required to raisethe wh-phrase to the edge of theΠP phase. The derivation in which the subject DP ischecked is clearly preferable on grounds of derivational economy.

Now consider the derivation for a non-subject root wh-question in (51-b), at thepoint where (53) is reached.

Page 34: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 33

(53) ΠP

����

��

HHHH

HH

Π[Tense]

TP

�����

HHHHH

DP��� PPP

Sheila[Tense]

T′

�����

HHHHH

T[Tense]

vP

����

HHHH

DP���

PPP

which[wh]

tent

vP���

PPP

t pitchedt

This time, the wh-phrase cannot be displaced as a by-product of [Tense] checking.Instead, a (defective) [wh] feature must be added toΠ to force the wh-movement. Thenthe [Tense] feature and the [wh] feature inΠ will both trigger movement of their targets.

The question of how checking operations are ordered when there multiple distinctuninterpretable features on a single head has not been examined much in the literature,but Bobaljik and Branigan (2004) have shown that in multiple Case-checkingconfigurations, there is a general principle which applies. Where extra, or marked, Casefeatures are added to a single head, it is always the extra Case feature which is checkedbefore the unmarked Case feature. Let us suppose that the same principle applies whenthere are multiple features present inΠ. Then we would expect that the ‘extra’ feature,the defective [wh] feature, would be checked before the original [Tense] feature ischecked. The derivation of (53) will then proceed as follows.

Page 35: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 34

(54) ΠP

����

��

HHHH

HH

Π[Tense]

TP

�����

HHHHH

DP��� PPP

Sheila[Tense]

T′

�����

HHHHH

T[Tense]

vP

����

HHHH

DP���

PPP

which[wh]

tent

vP���

PPP

t pitchedt

]

After the extra [wh] feature is supplied toΠ, Π starts to check its uninterpretablefeatures. It begins with [wh], and find the closest matching feature in its TP complement.When it checks the [wh] feature ofwhich, it forces a phrase containing this feature toraise to the root, where it re-Merges as the specifier ofΠP:

(55) ΠP

������

HHHHHH

DP���

PPP

which[wh]

tent

Π′

�����

HHHHH

Π[Tense]

TP

����

HHHH

DP��� PPP

Sheila[Tense]

T′

���HHH

T[Tense]

vP�

��H

HH

t vP���

PPP

t pitchedt

The [Tense] feature ofΠ must be checked next, soΠ must select either T or the subject tocheck. SinceΠ already has a specifier, and sinceΠP evidently does not accept multiplespecifiers, it must check T. Again, checking implies movement, so T raises toΠ to satisfythe EPP, and the phase is complete:

Page 36: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 35

(56) ΠP

�����

HHHHH

DP���

PPP

which[wh]

tent

Π′

����

HHHH

��HH

T Π

TP

���

HHH

DP�� PP

Sheila

T′

�� HH

T

t

vP��� HHH

t vP���

PPP

t pitchedt

The structure in (56) is not yet convergent, however, for two reasons. First, thisphase is not yet subject to Spell-Out, which applies only at the next phase level. Second,the [wh] feature ofwhichstill needs to be checked by a non-defective [wh] feature on atrue interrogative C. The derivation continues on by Merging the appropriate silent rootinterrogative C withΠP. C can then check its [wh] feature with the wh-phrase inSpec-ΠP, andΠP then undergoes Spell-Out. The derivation now converges:

(57) CP

�����

HHHHH

C ΠP

�����

HHHHH

DP���

PPP

which[wh]

tent

Π′

����

HHHH

��HH

T Π

TP

���HHH

DP�� PP

Sheila

T′

�� HH

T

t

vP��� HHH

t vP���

PPP

t pitchedt

For root wh-questions, subject wh-phrases and nonsubject wh-phrases both move tothe Spec-ΠP position, but they are attracted to this position by different features. Subjects

Page 37: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 36

are uniformly attracted by the [Tense] feature; nonsubjects, by the defective [wh] feature.This difference could be expected to have consequences. And in fact, this differenceprovides an immediate explanation for the subject/nonsubject assymetry in coordinatedquestions observed by Williams (1978).

The data which Williams finds are of the type in (58).

(58) a. Who should rake the lawn and will weed the garden?b. What had impressed Sean and should have impressed Shirley?c. Which horse is winning the race and may win the Triple Crown?

(59) a. What horse should Joanne ride and could Beth ride next?b. Which address was the package sent to and will the courier have to walk to?

(60) a. *Who should rake the lawn and will Maria pay for her services?b. *Which address is located on a hill and will the courier have to walk to?

These sentences are all coordination structures in which a wh-phrase is extracted“across-the-board” (ATB) from each half of a coordinated phrase no smaller than T′ .Williams shows that such ATB extraction is possible when the wh-phrase serves as asubject with respect to both conjuncts, as in (58), or when it serves as a nonsubject, asin (59), but extraction is blocked when the wh-phrase is the subject for one conjunct and anonsubject for the other, as in (60).

Amoung the possible structures to be considered for such sentences, the hardest toexplain is that in which the conjoined phrases are both larger than TP:

(61) a. What horse [C′ [C′ should [T Joannet ride t ]] and [C′ could [TP Betht ridet next ]]]

b. Who [C′ [TP t should rake the lawn ]] and [C′ [TP t will rake the garden ]]]c. *Which address [C′ [C′ [TP t is located on a hill and [C′ will [ TP the couriert

have to walk tot ]]]]

In the (61-a) sentence, for example, the coordinated phrase is clearly some variant of C′ ,which selects a TP complement. In both halves of the conjoined phrase, T raises to the Chead, and the wh-phrase is extracted to become the specifier of the conjoined C′ .Example (61-b) is structurally ambiguous, but one available structure should be similar tothe (61-a) case, with conjoined C′ phrases each of which have a TP complement fromwhich the subject is extracted ATB. Since these two sentences are grammatical, ATBextraction from conjoined C′ phrases must be allowed in principle. But then whataccounts for the ungrammaticality of (61-c), where a wh-phrase is again being extractedATB from two conjoined C′ phrases?

Page 38: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 37

In the model developed here, the ungrammaticality of (61-c) reflects a clash in thefeatures involved in the movement of the wh-phrasewhich addressto its surface position.As the sentences in (61) are root questions, the landing site for wh-movement must beSpec-ΠP and the conjoined phrases will all beΠ′ s. Thus, (61-a) should be understand as(62-a), and (61-c), as (62-b).

(62) a. What horse [Π′ [Π′ should [T Joannet ride t ]] and [Π′ could [TP Betht ridet next ]]]

b. *Which address [Π′ [Π′ [TP t is located on a hill and [Π′ will [ TP the courierthave to walk tot ]]]]

In each of the conjoinedΠ′ phrases in (62-a), the head must take on an extra [wh] featurein order to attract the wh-phrase in TP. Both of theΠ heads in the conjunction check their[Tense] feature by attracting T after they have already attracted the wh-phrase (ATB) toserve as their (shared) specifier. The ATB movement does exactly the same thing for bothconjuncts: it allows the attracting head to check its [wh] feature at the same time as itforces the target phrase to re-Merge at the root.

In (62-b), however, the movement of the two wh-phrases are driven by differentforces. In the first conjunct, the [Tense] feature ofΠ is checked on the subject, whichthereby raises. In the second conjunct, it is a deficient [wh] feature which attracts thewh-phrase. Since the wh-movement is driven by completely different features for the twoconjuncts, it cannot take place ATB.

To sum up, much as in the P&T approach, by postulating a [Tense] feature on C (andits Π variant), the subject/nonsubject assymetry in root questions can be effectivelyexplained. Let us then return to the issues which prove problematic for P&T.

If Π in root questions bears the [Tense] feature, then we should expect thatΠ willhave the same properties in other clausal contexts. (In the P&T theory, the same issuearises with respect to the feature content of C.) In fact, sinceΠ itself does not bear any[wh] feature inherently, and since it is not directly involved in the interpretation of [wh]questions,Π must have the same properties in questions as it has in declarative clauses.14

The question is actually harder to avoid than in the P&T model, where C is at leastdifferent in its interpretive function in questions and statements. What is more, since wehave no reason to suspect thatΠ functions differently in embedded clauses from rootclauses,Π should bear a [Tense] feature in embedded contexts, too.

How then should the derivation of a simple root declarative like (63) be treated?

(63) Those fellows pitched the tent.

14It would be possible to stipulate that only interrogative C selects aΠP complement in whichΠ bears a[Tense] feature.

Page 39: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 38

As in the question case, there are two possible ways forΠ to check [Tense]: byattracting the subject, or by attracting T. In either case, the EPP-linked property of[Tense] will be discharged, as long as overt movement takes place. But the absence ofdo-support and inverted word order indicates that it must be the subject which is checkedby Π in this situation. Thus, the structure of (63) is (64):

(64) ΠP

����

HHHH

DP����

PPPP

those fellows

Π′

��� HHH

Π TP

���HHH

t T′

�����PPPPP

T pitched the tent

(This is the point at which we part path with P&T, for whom the subject raises to check[Tense]only if it is an operator which must undergo A-bar movement for other reasons.)

Subject-oriented quatifiers offer an argument that it is, in fact, the subject whichraises to check the [Tense] feature ofΠ in simple declaratives. Sportiche (1988) showsthat floating quantifiers appear in a position from which the DP with which they areassociated has raised. VP-internalall or boththus marks the original position of thesubject DP in (65).

(65) a. The reviewers have [vP all panned this film ]b. Betty and Kate will [vP both attend the performance ]

In English, a quantifier of this type can also appear in a position preceding modal andauxiliary verbs, to the right of a subject DP.

(66) a. The reviewers all have panned this film.b. Betty and Kate both will attend the performance.

Although they are close to the subject DP, such quantifiers cannot belong to the subjectDP, because they may be separated from it by sentential adverbs.

(67) a. The reviewers probably all have panned this film.b. Betty and Kate certainly both will attend the performance.

Page 40: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 39

In fact, the floating quantifiers must follow sentential adverbs if they follow the subjectDP at all. (Parenthetical intonation on the sentential adverbs improves the status of suchexamples.)

(68) a. *The reviewers all probably have panned this film.b. ??Betty and Kate both certainly will attend this performance.

If the floating quantifiers are not located in the subject DP in these examples, they mustbe stranded in a lower position out of which the subject has raised. The lower position isevidently the specifier of a functional category headed by the auxiliary/modal verb. Onconservative assumptions, modals and auxiliarys appear in T in finite clauses like these,so the floating quantifier is in Spec-T. The subject must then have raised from Spec-I to ahigher position. Then the structure of (67-a) is (69).

(69) [ΠP the reviewersΠ probably [TP all t have panned the film ]]]

Since we already know that the wh-phrase is not destined to become the specifier forΠP, we understand why it ispossiblefor the subject to be selected in this case, but theremust be some reason why it ispreferable.

I suspect that the workings of the EPP are relevant. We see in examining thederivations of root wh-questions that a specifier forΠ is always found. What simplesentences appear to indicate is that this is not an accident. Although I have adopted thenotion from P&T (and Chomsky (1995)) that the EPP can sometimes be satisfied by headmovement, it seems still to be the case that movement to specifier position is thepreferred way to satisfy the EPP, and head movement is somehow secondary. Let us givethis premise a name:

(70) The EPP satisfaction contraintIf possible, EPP features in a single head must be satisfied by movement tospecifier position.

It may be that the (70) constraint follows from Richard’s (1998)Principle ofMinimal Compliance. If the grammar insists that the EPP be satisfiedonlyby movementto specifier position, then a single feature associated with EPP effects will always forcespecifier movement (as in English TP), but if the EPP is in force for several differentfeatures, then the first movement triggered could be said to pay the “specifier tax”,leaving subsequent EPP-triggered movements to behave more freely. However, I see noway to explore this issue more carefully within the confines of the present study, so I willsimply adopt (70) in the form given.

Page 41: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 40

There is one more question to consider before we turn to embedded clauses, and thatis the feature content of root C. I have suggested thatΠ is a variant of C, rather than anew syntactic category.Π moreover bears an uninterpretable [Tense] feature. It followsthat C, too, should bear an uninterpretable [Tense] feature, along with the [wh] feature. Ifso, then there must be some way for C to check [Tense].

Fortunately, no new mechanisms are necessary to permit this. The [Tense] feature ofΠ is checked whenΠP is complete, but it is not erased until the completion of the CPphase. Therefore C can check its own [Tense] feature against the [Tense] feature ofΠ.(The logic is entirely analogous to the P&T account of [Tense] checking in root subjectwh-questions, where the [Tense] feature on the subject is involved in a secondarychecking operation after it has itself been checked.)

The appeal of the P&T approach to auxiliary inversion structures is that it offers anexplanation of the subject/non-subject assymetry in which the special properties ofsubject wh-questions follow directly from their Case and positional properties. Theproblems in the P&T theory arise when the premise of a [Tense]-bearing C is examinedin other constructions. The “best” version of the P&T theory is one in which not onlyroot questions but all clauses bear the same [Tense] feature, with the same checkingrequirements, but the actual implementation of this “best” theory which they provide wasshown to fail. In the enriched theory of clause structure presented here, it is possible toincorporate the P&T approach, with all of its conceptual elegance. What is more, we nowsee that a “best” version of this revised theory may be tenable with no special stipulations.

(71) The distribution of the [Tense] feature

a. T bears interpretable [Tense], associated with an EPP feature.b. subject DP bears uninterpretable [Tense], realized morphologically as

nominative Case.c. Π (a variant of C) bears [Tense], associated with an EPP feature.d. Root C bears [Tense], (vacuously) associated with an EPP feature.

For both T andΠ, the [Tense] feature will normally be checked by attracting the subjectto specifier position. Special circumstances (in nonsubject wh-questions) will sometimesallow the Spec-Π position to be coopted for other purposes, forcing the [Tense] feature ofΠ to attract a head instead. And the silent root C, which must not attract anything, cancheck the [Tense] feature onΠ freely.

As is well known, auxiliary inversion in English takes place with displaced negativephrases, too, as in (72).

(72) a. Never have I seen a more elegant account of English word order.b. Few committee meetings had she deliberately attended.

Page 42: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 41

c. Only on the weekend will they consent to meet.d. In no finer pub could we hold this shindig.

The preposed ‘negative’ phrases in this construction are always interpreted as thesentence focus, and inversion is obligatory with preposed monotone-decreasing focii.

Previous analyses of auxiliary inversion in questions have dealt with the negativeinversion construction only with some difficulty. If auxiliary inversion is treated as aside-effect of movement to Spec-CP, then it should follow that negative focii raise to thesame Spec-CP category as wh-phrases do, and for the same reason—to be checked by C.But it is unclear in this view why only negative focii are treated in this way. If C may beara strong [Focus] feature to check, then it should be equally possible to form sentenceslike (73), with the preposed object interpreted as a sentential focus.

(73) *In this pub can we hold this shindig.

In the present model, however, a straightforward interpretation of these data ispossible, which avoids this problem, and others. For one thing, there is no reason tosuppose in the present model that C is involved in checking anything in negativeinversion sentences. Rizzi (1997) has proposed that preposed focii in other languages areassociated with a distinct Focus head within the CP ‘zone’, a head which is situated lowerdown in the tree than the true complementisers. Adapting this notion to the generalstructural notions argued for here, the structure of (72-d) will be (74) (prior to the finalMerger of the silent root C).

(74) FocusP

������

HHHHHH

Focus ΠP

����

��

HHHH

HH

PP����

PPPP

in no finer pub

Π′

����

HHHH

�� HH

T

could

Π

TP

����

HHHH

D

we

T′

�����

PPPPP

t hold this shindigt

Page 43: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 42

By analogy with the root wh-questions, we may suppose that the fronted negative PP in(74) raises only to Spec-ΠP, and not all the way to the higher FocusP. In fact, this must bethe case, unless the FocusP counts as a distinct phase in (74), because Spell-Out cannotprovide phonetic interpretation for anything beyond theΠP phase in (74). As for themotivation for movement of the PP in (74), the analogy with root questions againsuggests what is going on. The Focus head must have a [Focus] feature to check, and itmust be able to do so only if its target is close enough, under thePhase ImpenetrabilityConstraint. Thenin no finer pubraises to Spec-ΠP in (74) in order to get close enough toFocus to allow the [Focus] feature to be checked. Since the subject cannot then raise toSpec-ΠP,Π must attract T to check its [Tense] feature.

In this account, the preposed negative phrase is not attracted directly by any headwhich bears a [Focus] feature—instead, the negative phrase is attracted byΠ. Π, we mayassume, bears the same attracting feature in negative inversion sentences as it bears ininverting root wh-questions. This attracting feature is not the (non-defective) [wh]feature, as that resides in the silent root C. By natural extension, we can suppose that thisfeature is distinct from the (non-defective) [Focus] feature, too, which should reside inthe Focus head situated external toΠP.

Why now is the (73) sentence impossible with the structure (75)?

(75) [CP C [FocusPFoc [ΠP in this pub can-Π [TP we t hold this shindigt ]]]]

(C and Foc will both lack phonetic content in this structure, as they cannot undergoSpell-Out.)

We know that wh-phrases and monotone decreasing phrases can be attracted toSpec-ΠP by some feature, the content of which has been left unspecified. If this feature isrestricted to attracting only wh-phrases and monotone-decreasing phrases, then theungrammaticality of (75) follows directly, as the specifier could never raise to Spec-ΠP,even if the [Focus] feature of Foc cannot be checked without such movement. Let ussuppose, then, thatΠ checks the feature [MD] (for Monotone Decreasing), and that onlymonotone decreasing quantifiers (including wh-phrases) bear this feature, as interpretablesemantic content.15

To sum up, in root clauses,Π is always the topmost constituent which is subject toSpell-Out.Π always checks a [Tense] feature, which has an EPP-property which is oftensatisfied by attracting the subject. Sometimes, however, in order to bring wh-phrases andfocii to the left periphery of theΠP phase,Π takes on an additional [MD] feature, whichthen attracts monotone decreasing phrases to Spec-ΠP, in which case T raises toΠ in

15It is known that monotone decreasing quantifiers behave as a cohesive class with respect to otherconstraints, such as the licensing of negative polarity items (Ladusaw, 1979).

Page 44: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 43

order to satisfy the EPP-property of the [Tense] feature.

Embedded clauses

Now let us consider the case of simple embedded declaratives, such as (76).

(76) The warden discovered that those guys pitched the tent.

We can start with the structure ofΠP for the embedded clause. Following the Merger ofΠ with TP, the structure ofΠP will be the same as in the root declarative (64). As in theformer case, the [Tense] feature ofΠ will be checked by attracting the subject toSpec-ΠP. The [Tense] features ofΠ will remain accessible to the derivation until the nextphase is complete andΠP is subject to Spell-Out.

In the derivation of (76),ΠP will next be Merged with the complementiserthat: (77).

(77) CP

�����

HHHHH

C

that[Tense]

ΠP

�����

HHHHH

DP���

PPP

those guys

Π′

����

HHHH

Π[Tense]

TP

�����PPPPP

t pitched the tent

Like Π, that should bear an uninterpretable [Tense] feature to check, so it must checkΠ,just as the silent C does in root clauses. Thethat complementiser is subject to Spell-Out,though, so its EPP features cannot be ignored. But the subject does not raise to Spec-CP,as shown by (78).

(78) *The warden discovered [CP those guys that pitched the tent ].

I assume that movement of the subject to Spec-CP interferes with the interpretation of thecomplementiser, along lines suggested by Pesetsky (1998). An overt declarative C mustappear at the left edge of its clause in order to play its role as a clausal subordinator, orforce marker (Cheng, 1991). No specifier may raise to Spec-CP in declaratives for thisreason. But the [Tense] feature, with its EPP property, still requires movement.Π musttherefore raise to C in this case, producing the CP structure (79).

Page 45: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 44

(79) CP

����

��

HHHH

HH

C�� HH

Π C

that[Tense]

ΠP

����

HHHH

DP���

PPP

those guys

Π′

���

HHH

Π

t

TP

�����PPPPP

t pitched the tent

The structure of declarative complement clauses without an overt complementiser isslightly more difficult to resolve. On the one hand, the surface word order does not showthat there is no movement of the subject to Spec-CP. Thus, sentence (80-a) could beanalyzed as either (80-b) or (80-c).

(80) a. The warden discovered those guys pitched the tent.b. The warden discovered [C those guys∅ [ΠP t Π pitched the tent ]]c. The warden discovered [C ∅-Π [ΠP those guyst pitched the tent ]]

The (80-b) structure is a variant of what P&T propose forthat-less declaratives,which they treat as structures in which the subject checks the [Tense] feature of C. But itis not obvious why the distribution of such declaratives would be restricted to the domainof appropriate predicates, if (80-b) is allowed as a possible structure.16

With (80-c), of course, the same questions arise, but the model of Boškovic andLasnik (2003) provides good answers. They argue that null declarative complementisersmust undergo Morphological Merger with an adjacent predicate, which is possible onlyin the right syntactic environments.17 Thus in (80-c), the null complementiser islegitimate because it is joined at some post-Spell-Out level with the verbdiscovered.18

No problem arises in describing how the [Tense] features ofΠ and C can be checkedin embedded statements (in contrast to the P&T model). SinceΠ checks the subjectgenerally in statements, we need not invent a special morphological rule to make Tappear to have moved in embedded declaratives. But one problem arises in my account to

16P&T argue for this structure forthat-less embedded declaratives, but on theory-internal grounds whichinvolve the idea thatthat is a form of T raised to C. Since it has already been shown that this analysis ofthat is untenable, the arguments which P&T put forward lack force.

17The Boškovic and Lasnik analysis is a modification of earlier proposals by Pesetsky (1992).18Boškovic and Lasnik propose an account ofthat-trace effects as a part of their analysis which is less

convincing, as discussed in section (6).

Page 46: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 45

which P&T (and other accounts) are immune. Within the structure shown in (80), thereare two phase boundaries at the left periphery of TP. Under the PIC, movement fromwithin a phase to a higher position must always involve movement through the ‘escapehatch’ at the left edge of the phase. Given the structure in (80), A-bar movement from CPinto a higher matrix clause should then be possible only if the moving phrase first raisesto Spec-ΠP, and then to Spec-CP, before it can escape the complement clause. But theSpec-ΠP position here is already filled, so it cannot serve as an escape hatch. The clearprediction then is that no successive cyclic A-bar movement should be possible inEnglish. The prediction is false, as is evident from (81).

(81) Which tent did the warden discover that those guys pitched?

What is the significance of this false prediction? It is clear that there are languagewhich do not allow cross-clausal movement—many speakers of standard German do notaccept such movement except when special verb-second escape hatch structures are used;Tsez does not permit cross-clausal movement at all (Polinsky and Potsdam, 1999). It isequally clear that many languages, including English, do permit it. The German/Tsezfacts may now be explained by the structures I propose for embedded clauses, butsomething must be said about the other type of language.

I will argue that the reason long A-bar movement is possible in general—at least inlanguages with the general properties found in English—lies in the derivationalrelationship between C andΠ in these languages, where movement ofΠ to C results in anincorporation structure. Baker’s (1988)Government Transparency Corollary(GTC)constitutes an approximation of what appears to be an accurate descriptive generalizationfor incorporation structures in general. The generalization is that incorporation of thehead of a category into a higher head renders the contents of the first category moreaccessible to the higher head than they would otherwise be. For example, incorporationof a head noun in Mohawk allows the incorporating verb to treat the noun’s possessor asif it were an object. To my knowledge, the explanation for this descriptive generalizationis still to be found, but the empirical evidence on which it is based is overwhelming.

So for example, in Mohawk, the verb agrees with its subject, but not with thepossessor of its subject, as seen in (82-a).

(82) a. Ka-rakv3N-white

neDET

[ sawatisJohn

hrao-nuhs-a?3M-house-SUF

].

‘John’s house is white’b. Hrao-nuhs-rakv

3M-house-whiteneDET

[ sawatisJohn

t ].

‘John’s house is white’

Page 47: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 46

In (82-b), though, where the head of the subject NP has been incorporated by the verb,the verb now agrees withsawatis‘John’, which continues to be interpreted simply as thepossessor of the incorporated noun.

In Bobaljik and Branigan’s (2004) treatment of French causative structure, the samegeneral pattern is identified in structures like (83):

(83)Paul fera v′

�����

HHHHH

v

����

HHHH

C�� HH

acheter‘buy’

C

t v

VP

����

HHHH

t CP

����

HHHH

DP��� PPP

ce camion‘this truck’

C′

����

HHHH

C

t

TP

����

HHHH

DP���

PPP

à Joanne‘Joanne (DAT

T′

����PPPP

. . .t. . .t. . .t. . .

In this structure, incorporation of the head of CP into the matrix v allows v to accessobjects within TP for both Case-checking operations and to trigger A-bar movement toSpec-vP. In short, incorporation of the head of a phase allows the incorporating head tolook past the edge of that phase, despite the strictures of the PIC. I will refer to thispattern as thePhase Transparency Effect(PTE).

(84) Phase Transparency EffectGiven two headsα andβ, whereβ is the head of phase G, ifβ incorporates intoα, then features ofα may probe past the edge of G in AGREE.

This is, of course, not a complete explanation for anything. But the empirical basisfor GTC effects, of which the PTE appears to form a part, is robust enough that we maysafely take the generalization for granted in seeking explanations in other areas of thegrammar.

Now consider once again the structure of a normal declarative complement clausefrom which A-bar extraction is possible:

(85) What do you think [CP Π-that [ΠP Betht [TP t should [vP t [ v′ t do t ]]]]]

Page 48: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 47

Within ΠP in (85),Π checks its [Tense] feature onBeth. As Bethraises to Spec-ΠP, thereis no escape hatch at the edge ofΠP available to the wh-phrasewho. But whenΠincorporates intothat, theΠP becomes transparent to checking operations involving thecomplementiser (due to the PTE). If now the complementiser takes on a [wh] feature, itmay attractwho into its specifier position, making subsequent A-bar movement into thematrix clause possible. The PTE thereby obviates the PIC in this specific case.19

Now let us consider the case of embedded wh-questions, such as (86).

(86) Penny asked which tent Sheila pitched.

We can start with the structure ofΠP for the embedded clause. Following the Merger ofΠ with TP, the structure ofΠP will be the same as in a root wh-question.Π will bear a[Tense] feature, but not [wh] feature.

(87) ΠP

����

��

HHHH

HH

Π[Tense]

TP

�����

HHHHH

DP��� PPP

Sheila[Tense]

T′

�����

HHHHH

T[Tense]

vP

����

HHHH

DP���

PPP

which[wh]

tent

vP���

PPP

t pitchedt

It has already been established, in section 2.3, that the wh-phrase in an embeddedquestion does not raise to the Spec-ΠP position. This means thatΠ does not acquire theextra [MD] feature, and its [Tense] feature will then be checked by movement of thesubject to Spec-ΠP. WithΠP complete, an interrogative C may be merged at the root,with both a [Tense] feature and a [wh] feature with EPP properties. The resultingstructure will be (88).

19ThePhase Transparency Effectwill necessarily be quite limited in its empirical effects, since it willnormally occur only when the head of one phase incorporates the head of another phase.

Page 49: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 48

(88) CP

����

HHHH

C[wh][Tense]

ΠP

���HHH

DP�� PP

Sheila

Π′

��� HHH

Π[Tense]

TP

���HHH

t T′

����

HHHH

T[Tense]

vP

����

HHHH

DP���

PPP

which[wh]

tent

vP���

PPP

t pitchedt

If now C incorporatesΠ to check the [Tense] feature, theΠP phase boundary will betransparent to C for subsequent checking operations. C is then free to probe intoΠP,where it can find the wh-phrasewhich tentto attract to Spec-CP to check its [wh] feature.

2.6 Topicalisation

The discussion to this point has taken it for granted that subjects will be the first thing in aroot clause, unless auxiliary inversion takes place. In English, however, this is not quiteaccurate. Adverbial elements may freely appear in front of the subject, in both rootclauses and embedded clauses: (89).

(89) a. Karin thinks that sometimes aliens pursue her dachshund.b. Sometimes aliens pursue her dachshund.

Since such clause-initial adverbs are obviously interpreted by Spell-Out, they must be apart of the second highest phase in the sentence, i.e. they are part ofΠP. But they cannotbe in specifier position, since the subject already occupies Spec-ΠP. They must thereforebe adjoined toΠP, as in (90).

Page 50: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 49

(90) ΠP

�����

HHHHH

Adv

sometimes

ΠP

����

HHHH

DP�� PP

aliens

Π′

����

HHHH

Π TP

������

PPPPPP

t pursue her dachshund

In Chomsky’s (1995) terms, the adverbsometimesin (90) is includedbut notcontainedinΠP. It follows that inclusion in a phase is sufficient to allow a category to be interpretedwhen the entire phase undergoes Spell-Out.

For some speakers, non-adverbial phrases can also be preposed to clause-initialposition. Again, this may occur in both root clauses and embedded clauses: (91).

(91) a. ?Karin claims that dachshunds, aliens are fascinated by.b. ?Dachshunds, aliens are fascinated by.

For me, such preposed phrases must be understood as foci; for other speakers, they maybe understood as either a focus or as the sentence topic (?).

As with clause-initial adverbs, preposed foci or topics must be adjoined toΠP.Otherwise, they would not be subject to Spell-Out. Their semantic role can be explainedif this adjunction occurs in order to make them accessible to aΠP-external Focus orTopic head for checking. With (91-b), for example, the DPdachshundsmust be thesentence focus, which follows if the Focus head checks [Focus] on the DP in its adjoinedposition, where it is external to TP and therefore accessible under thePhaseImpenetrability Constraint. For those speakers for whomdachshundscan be understoodas a topic in (91-b), we may suppose that a Topic head checks a [Topic] feature on DP inthe same way.

Monotone decreasing foci, however, cannot be allowed to adjoin toΠP, given theungrammaticality of (92).

(92) *Few dogs, aliens are fascinated by.

The most likely explanation for this disparity is that adjunction preposing is a last resortoperation, which cannot take place if there are other ways to allow a sentence toconverge. With monotone decreasing foci, it is always possible to bring the focus close

Page 51: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 50

enough to the Focus head by using the auxiliary inversion strategy. Therefore, the more‘costly’ adjunction strategy is blocked.

The same reasoning provides an explanation for the fact that (93) is not anacceptable structure, with the wh-phrase adjoined toΠP.

(93) CP

����

��

HHHH

HH

C[wh]

ΠP

������

HHHHHH

DP���

PPP

which dogs

ΠP

����

HHHH

DP��� PPP

the aliens

Π′

����

HHHH

Π TP

������

PPPPPP

t were fascinated byt

Althoughwhich dogsin this structure is close enough to the silent root C to allow its [wh]features to be checked, the derivation is blocked by the preferred derivation involvingauxiliary inversion.20

A similar pattern arises in other Germanic languages, where the interaction betweeninversion and adjunction structures is slightly richer than in English. In Swedish, forexample, sentential adverbs can be found before the subject in embedded clauses, but notin root clauses.21 Vikner (1991) provides the data in (94)).

20Since T can attract only a [Tense]-bearing phrase to serve as its specifier, it follows that other types ofphrases might be allowed to adjoin to TP. This prediction is accurate for some adverbs which appear in aposition above T and lower than the subject (which is in Spec-ΠP).

(i) The reviewers probably will pan this film.

On the other hand, it is impossible to use TP as an adjunction site for fronted arguments: (ii).

(ii) a. *The reviewersthis film will pan t.b. *Yves in the pan will lay the filett.

However, since adjunction of arguments takes place only to allow them to have [Topic] features checked,this is expected. The Top head does not have access to anything adjoined to TP, since it is blind toeverything but the head and specifier of theΠP phase.

21Thanks to Anders Holmberg for pointing this out to me.

Page 52: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 51

(94) *(Hon(She

sasaid

att)that)

tyvärrunfortunately

JohanJohan

intenot

harhas

lästread

dennathis

boken.book

The root/embedded clause assymetry here makes sense if the adjunction site for theadverb is outside the domain of application of Spell-Out in a root clause, i.e. outsideΠP.Since the adverb is to the right of the complementiseratt in (94), the adjunction site mustbe a category between C andΠP. TopicP seems the most natural candidate in this case. Ifthe adverbtyvärr is adjoined to a Topic head, then it will not be accessible for Spell-Outin a root clause, but Spell-Out will interpret it when it interprets the complementiser inthe embedded clause.

Why should adjunction be restricted to Topic heads in Swedish, but allowed forΠPin English? One might suppose that this is a simple matter of parametric variation.Perhaps there could be a parameter associated with different functional heads whichdetermines whether each head permits something else to adjoin to it, or to its projections.It is likely, however, that such an account would be far too powerful, and would lead us toexpect much more language variation than we actually find. (The burden placed on achild learning her language in such a model would also be commensurately large.)

Fortunately, a less powerful, and more interesting account of the Swedish/Englishdifference is at hand. Topicalisation in Swedish follows the general pan-Germanicpattern, with topics raised to clause-initial position and verb-second word order anobligatory side-effect.

(95) Dennathis

bokenbook

harhas

hanhe

intenot

läst.read

As we have already seen, parallel sentences are impossible in English because whenΠ isinvolved in triggering A-bar movement, it can only do so by taking on the [MD] feature,which checks only monotone decreasing quantificational elements. Clearly, in Swedish,and elsewhere in Germanic, theΠ head must be somewhat more versatile, becauseΠmustbe able to attract other types of topics. Let us suppose, therefore, that there is afeature which is complementary to [MD], which can attract phrases which are notmonotone decreasing. I will refer to this other feature as [MD]. In Swedish,Π can takeon either [MD] or [MD] in order to attract topics and foci to the left edge ofΠP. (Positiveevidence must then be accessible in the linguistic environment of a Swedish child toallow her to learn this featural property ofΠ.)

Since topics can now always be raised into Spec-ΠP by using either [MD] or [MD]features, it follows that they never need to adjoin toΠP. In other words, the existence ofregular Swedish verb-second topicalisation structures is what blocks adjunction toΠP.

Page 53: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 52

As is well known, the non-English Germanic languages generally exhibitverb-second word order in root clauses, as well as in some embedded clause contexts.Given the analysis of English ‘residual verb-second’22 developed above, it should now bepossible to say something about the general Pan-Germanic pattern.

Consider the typical Dutch root clause paradigm in (96).

(96) a. Watwhat

heefthas

HansHans

gelezen?read

b. Datthis

boekbook

heefthas

HansHans

gelezen.read

c. Hans heeft dat boek gelezen.Hans has read this book

As in English root questions, the wh-phrase in (96-a) must raise to Spec-ΠP to reacha position where the silent root C can check its [wh] feature. The finite verbheeft, whichpresumably raises to T in all clauses, is then attracted toΠ to check its [Tense] feature.Root questions in English and elsewhere in Germanic appear to be essentially identical.

The topicalisation structure in (96-b) has the form it has becauseΠ in Dutch can takeon the [MD] feature which allows it to attract a topic to Spec-ΠP in order to make itaccessible for checking by a higher Topic head. And again, the finite verb must then raiseto Π in order forΠ to check its [Tense] feature.

Example (96-c), however, has no analog in English syntax. In this case, the [Tense]feature ofΠ is presumably checked by the subject, which raises to Spec-ΠP. The finiteverb is not therefore required to raise toΠ for reasons involving the [Tense] feature.Something else must be going on which triggers the verb movement in this case.

It is necessary to say that English and Dutch in some way simply in order to achievedescriptive adequacy with respect to sentences like (96-c). The difference should not,however, be a deep one. Basic principles governing derivations must be identical in bothlanguages. Optimally, in fact, the difference should be quite shallow, involving somesmall variable property of a single functional category.

Zwart (1993b) proposes that C in Dutch is affixal, in the sense that it must beattached to a head with lexical content. English, in contrast, has a complementiser whichcan stand alone. Adapting this proposal to the current model, we may say that EnglishΠmay stand alone, while DutchΠ must be attached to another head. In that case, the verbmovement in (96-c) may reflect a repair strategy in which the finite verb raises toΠsimply in order to satisfy the latter’s ‘affixal’ requirements. (And the same could then be

22The term belongs to Rizzi (1990b).

Page 54: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 53

true of T-to-Π movement in subject-initial root clauses in other Germanic languages.) Wemay even suppose that the verb movement in (96-c) is a stylistic operation, which takesplace after all core syntactic operations (within a phase) are finished. The differencebetween Dutch and English then is minimal, involving only the affixal status ofΠ.23

In German and Dutch, a particular conjunction structure is frequently found whichappears to be impossible in other Germanic languages. An example appears in (97).

(97) Alsif

nietsnot

meermore

helpthelps

namelijk, paktake

ikI

zo’nthen-a

blauweblue

koelbatterijcoldpack

uitfrom

dethe

diepvriesfreezer

enand

houdhold

diethis

tegento

huntheir

warmewarm

voetjes.little feet

‘If nothing else works, I get a coldpack from the freezer and hold it against theirwarm feet.’24

In this type of conjunction structure, the subject is shared by both conjuncts, andexpressed only once. But there is a topic which appears in first position relative to thefirst conjunct, and which is not syntactically related to the second conjunct.

Modifying the analysis in Branigan (1996c), we can account for this type ofstructure as follows. It is clear that some degree of CP-recursion is possible in general.The CP-ΠP shells which I am proposing in this work are nothing other than a particulartype of CP recursion. Let us say then that the set of legitimate CP recursion structuresmay include those in whichΠP is used recursively, when necessary. And onecircumstance in whichΠP recursion may be necessary is when the Spec-ΠP subject isshared by twoΠ ′ conjuncts.

The structure of (97) may then be analysed as (98).

23It is significant that the affixal requirements ofΠ are not sufficient to override the preference (statedin (70)) for a specifier over a raised head. In effect, an uneconomical derivation is preferred to theeconomical alternative. This peculiarity may be taken as evidence that the verb movement seen insubject-initial verb-second clauses truly is a stylistic operation, and therefore excluded from considerationsof derivational economy.

24Taken fromReflectie, by Theo Jansen, inde Volkskrant, September 15, 2001. The context is ahumorous one, involving a parent trying to get his children out of bed for school in the morning.

Page 55: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 54

(98)CP

�����

HHHHH

C(silent)

TopP

�����

HHHHH

Top ΠP

�����

HHHHH

CP���

PPP

als niets meerhelpt namelijk

Π′

����

HHHH

Π

pak

ΠP

������

HHHHHH

D

ik

Π′

����

���

HHHH

HHH

Π′

���PPP

t zo’n blauwekoelbatterijt

Conj

en

Π′

�����PPPPP

houd die tegenhun warme voetjest

In the derivation of (98), the verb-second word order in the second conjunct reflects thestylistic verb-raising operation which supports an affixalΠ when it does not raise to C.The same operation occurs in the first conjunct, and for the same reason, but thesupportedΠ is then raised further to support the secondΠ in the higherΠP. And thehigherΠP contains a topic in Spec-ΠP in a position where its features may be checked bya still higher Top head.

Of course, ifΠP-recursion is possible, then a simple root clause in Dutch or Germanwill automatically be structurally ambiguous—multiply so, in fact. A sentence like(99-a), for example, could in principle have the structure (100-a), (99-b), (99-c), or anylarger structure in the same sequence.

(99) a. ZitaZita

heefthas

HannikeHannike

gekusstkissed

‘Zita kissed Hannike’

(100) a. [ΠP Zita heeft [TP t Hannike gekusstt ]]b. [ΠP Zita heeft [ΠP t t [TP t Hannike gekusstt ]]]c. [ΠP Zita heeft [ΠP t t [ΠP t t [TP t Hannike gekusstt ]]]]

Page 56: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 55

We may assume, however, that this sort of structural ambiguity can be filtered by someprinciple of Economy of Representation (or its parsing equivalent), so that themulitiplicity of possible structures will not be problematic. (The same problem arisesmore generally anytime CP-recursion structures are permitted in the grammar, so ageneral solution must be available.)

In the case of (97), there is no non-recursive structure which can be considered,since the subject must be shared by the conjoinedΠ′ phrases, so the recursive structurein (98) is not excluded by Economy of Representation.

In embedded clauses, verb-second word order is severely constrained in Germanic,except for in Yiddish (Diesing, 1990) and Icelandic (Thrainsson 1987) . In Dutch, forexample, verb-second word order is possible only in the complement to bridge verbs (?).

Leaving aside for the moment the restricted distribution of embedded verb-secondclauses, consider how the affixal nature ofΠ plays a role in Dutch embedded clause wordorder in general. In normal embedded clauses, as already discussed for English, C mustcheck its [Tense] feature by attractingΠ. The same is presumably true in a Dutchsentence like (101).

(101) PietPiet

zeisaid

[CP dat-Πthat

[ΠP HansHans

t [ t dezethis

boekbook

gelezenread

heefthas

]]]

As a fortuitous side-effect of the attraction ofΠ by C,Π will be attached to a head withlexical content. In general, then, in non-verb-second embedded clauses, the lexicalsupport forΠ may come from the higher C which attracts it. No stylistic verb movementto Π is then motivated (and none occurs)

(In Dutch root clauses, of course, the complementiser is necessarily silent, andcannot attractΠ. In root clauses, the stylistic operation is the only way available tosupport affixalΠ.)

As with Zwart’s theory, this approach to embedded clause structure provides animmediate account of the embedded v/2 structure found in Dutch ‘assymetriccoordination’ (Hoekstra, 1994). (The data in (102) comes from Hoekstra, p. 288, 295.)

(102) a. Hetit

irriteertirritates

onsus

datthat

jeyou

tetoo

laatlate

thuiskomthome come

enand

jeyou

hebthave

geenno

sleutelmoney

bijwith

je.you

b. Alsif

jeyou

tetoo

laatlate

thuiskomthome come

enand

jeyou

hebthave

geenno

geldmoney

bijwith

je,you,

danthen

. . .

Page 57: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 56

In this construction, the first of two conjoined clauses is verb-final, while the second hasthe verb in second position. What makes this construction possible is the acceptability inDutch (and German) of an apparent violation of the Coordinate Structure Constraint inwhich the first of two clausal conjuncts has a constituent extracted from it. In thisconstruction, the dependentΠ head in the first clause is raised to the matrix C, leaving thesecondΠ inside its own clause, in a position where support can be found only if the verbraises toΠ.

(103) Het irriteert ons [CP dat-Π [ΠP je e [TP t te laat thuiskomt ]] en[ΠP je hebt-Π [TP t geen sleutel bij jee ]]]

What now can be said about the restricted embedded verb-second pattern inlanguages like Dutch, where the verb-second pattern is found once again.

(104) PietPete

zeisaid

datthat

datthat

boekbook

kendeknows

hijhe

niet.not

Given the structure proposed for root topicalisation structures, the embedded clausein (104) will have the form (105):

(105) Piet zei [CP dat [TopP Top [ΠP dat boek kende-Π [TP hij t t niet ]]]]

The Top head once again checks the [+topic] feature on the phrase in Spec-ΠP. The topicitself has been attracted to Spec-ΠP by an [MD] feature onΠ, so that the finite verb raisesto Π to check its [Tense]feature.

Since the finite verb raises toΠ, Π in (105) has no need for additional morphologicalsupport. As far as the requirements ofΠ are concerned, the structure of the embeddedclause in (105) is entirely satisfactory. But the [Tense] feature of C, and the associateEPP-effect, are not checked in the usual way in this structure. If neitherΠ or a[Tense]-bearing subject raise to C, then we would expect the derivation to crash. Whatthen makes this type of structure possible?

The general answer for Dutch embedded clauses is that this structure is not possible.Recall that embedded verb-second structures in this language are restricted tocomplements of bridge verbs. A reasonable conclusion to draw is that there is somethingspecial about bridge verbs which somehow exempts the complementiser in (105) fromsatisfying its checking requirements in the usual way. Either C does not need to check[Tense] in this context, or the [Teanse] feature of C lacks the EPP property, so thatΠ ischeckedin situ. (Since the form of complementisers in embedded verb-second clauses isessentially the same as in other contexts, the latter seems more likely.) But outside of thisspecial syntactic context, C will have no way to check its [Tense] feature in an embedded

Page 58: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 57

verb-second structure, and the derivation will then crash.Notice that the absence of movement ofΠ to C in embedded verb-second

complements makes an immediate prediction about extraction from within thecomplement clause. Recall that A-bar movement out of a complement clause is possiblein this model only because the PTE relaxes the effects of thePhase ImpenetrabilityConstraint. Π incorporation by C renders theΠP phase transparent for A-bar movement.If C does not incorporateΠ, then movement of anything from within TP into a highermatrix clause should be impossible. It is well known that embedded verb-second clausesare strong islands for A-bar movement, as shown in (106) (from Zwart (1997, p. 232)).The prediction is therefore corrent.

(106) a. Welkewhich

filmfilm

zeisaid

jeyou

datthat

JanJan

alalready

gezienseen

had?had

b. *Welkewhich

filmfilm

zeisaid

jeyou

datthat

JanJan

hadhad

alalready

gezien?seen

‘Which film did you say that Jan had already seen?’

In (106-a.), the finite verb does not raise toΠ, soΠ may raise todat. This movementobviates thePhase Impenetrability Constraintwith respect to theΠP phase, sodat canattract the wh-phrasewelke filmas its specifier, making it accessible for subsequentwh-movement into the matrix clause. In (106-b.), though, movement ofhad to Π ensuresthatΠ cannot raise further. TheΠP phase therefore remains active as a barrier toattraction ofwelke filmby the lower C.

As for the specific property of bridge verbs which allow C to check [Tense] with noEPP property, I can only speculate. Bridge verbs in general allow for complement clauseswith a range of ‘defective’ properties. ECM complements, for example, lack CP structureentirely, which is otherwise not permitted. The same is true of small clauses, which arenormally permitted only as complements to bridge verbs. As will be shown in chapters 3and 4, bridge verbs sometimes permit bareΠP complements. It may be that bridge verbsin Dutch, Swedish, Frisian, and similar languages are able to select complement clauseswith ‘bleached’ heads (where the [Tense] feature lacks the EPP property). This mightcoincide with some sort of abstract incorporation of the defective C by the bridge verb, asin Pesetsky (1992).25

In fact, there is clear evidence in mainland Scandinavian that the complementiserfound with embedded verb-second clauses has properties distinct from those of otherdeclarative complementisers, despite their phonetic identity. First, theatt complementiser

25See also Boškovic and Lasnik (2003) for an updated model of C incorporation which is largelycompatible with the analysis developed in the text.

Page 59: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 58

which appears in verb-second constructions does not behave like theatt we find withother types of embedded declaratives. When not associated with a v/2 clause,att may beomitted in complement clauses. But omission ofatt with verb-second clauses isimpossible (Holmberg, 1986).

(107) *PerPer

sasaid

hanhe

köptebought

intenot

boken.book-the

The second difference between theatt used in v/2 complements and the defaultatt iscofound in conjoined complement clauses, where it is peculiarly constrained. When twosuch clauses are conjoined, theatt complementiser appears only before the first clause.(Thanks to Anders Holmberg for the data.)

(108) JagI

trorbelieve

attthat

PerPer

ätereats

intenot

köttmeat

ochand

(*att)that

AnnaAnna

drickerdrinks

intenot

öl.ale

There is no obvious explanation for why the ‘bleached’ complementisers should havethese particular properties, but these facts do show the necessity for a distinction betweenthe two classes of declarative complementisers.

Whatever the precise mechanism which makes bridge verbs complements special is,it also appears to allow complementisers to be entirely absent in German. In thislanguage, embedded verb-second complements appear both as bare TopP, as in (109-a),and as bareΠP, as in (109-b).

(109) a. AnnaAnna

glaubt,believes

dasthe

Buchbook

habehas

Hans-PeterHans-Peter

gelesen.read

(German)

b. AnnaAnna

glaubt,believes

Hans-PeterHans-Peter

habehas

dasthe

Buchbook

gelesen.read

(German)

Given a Top head which checks the [Topic] feature ondas Buch, the structure of (109-a)will be:

(110) . . . glaubt [TopP Top [ΠP das Buch V-T-Π [TP Hans-Peter . . .t . . . ]]]

As in Dutch or Swedish,ΠP will attract the topic with its [MD] feature and the finite verbwith its [Tense] feature. Since there is no higher complementiser in the complementclause, the issue of C attractingΠ is moot. And the affixal requirements ofΠ are satisfiedby the position of the finite verb.

In (110-b), the structure of the complement clause is reduced still further: (111).

(111) . . . glaubt [ΠP Hans-Peter V-T-Π [TP t das Buch . . .t . . . ]]

Page 60: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 59

Here, the subject raises to check [Tense], and no TopP projection need be present at all.Notice that the verb raises toΠ, as in a root clause, to satisfyΠ’s affixal needs.

As in the Dutch case,Π in German embedded verb-second fails to incorporate into C(C being absent), so thePhase Impenetrability Constraintwill continue to recoganize thephasal status ofΠP. Again, we should expect embedded verb-second complements in thislanguage to serve as islands for movement, which is the case.

The situation is quite different for Yiddish and Icelandic. In both cases, embeddedverb-second clauses are found in a wider set of syntactic contexts, as illustrated in(112)–(113), taken from Vikner (1991).

(112) a. JohnJohn

tsveyfeltdoubts

azthat

morgentomorrow

vetwill

MiriamMary

friearly

oyfshteyn.get up

(Yiddish)

b. IkhI

veysknow

nitnot

farfor

voswhat

inin

tsimerkitchen

izis

dithe

kucow

geshtanen.stood

(113) a. JónJohn

efastdoubts

umthat

aðtomorrow

á morgunwill

fariMary

Maríaget up

snemmaearly

á fætur.

(Icelandic)

In both languages, embedded verb-second order is found in the complements tonon-bridge verbs. In Yiddish, as already discussed in section (22), verb-second order isalso sometimes possible inside an indirect question.

The same is true of English, to some extent. For some speakers, negative inversion ispossible both in complement clauses and in other embedded contexts, although theinterpretation of such structures will often be unfelicitous.

In these languages, the distribution of embedded verb-second structures cannot bebased on licensing conditions for a defective C, because defective C should not belicensed freely. Instead, we must suppose that C is able to check its [Tense] feature freelyin these languages, despite the movement of the finite verb up toΠ. There are two waysthat this could be possible. It could be the case that the [Tense] feature in C lacks an EPPfeature generally, so thatΠ would not raise up to C. Then movement of the finite verb toΠ would always take place because ofΠ’s affixal status. Alternatively, we could supposethatΠ raises to C in these languages even in embedded verb-second contexts, but that theeffects of this operation are not visible in the phonetic form. Surprisingly, the latterappears to be the right description of the facts, given the known transparency of Icelandicand Yiddish verb-second clauses for A-bar movement. Examples (from Vikner (1991)appear in (114).

Page 61: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 60

(114) a. Hvernighow

sagðisaid

húnshe

aðthat

íin

skólanumschool.the

hafðuhad

börninchildren.the

lærtlearned

söguhistory

t?

(Icelandic)

‘How did she say that the children had learned history in school?’ (withhow interpreted in the downstairs clause)

b. Voswhat

hothad

erhe

nitnot

gevoltwanted

azthat

inin

shulschool

zolnshould

dithe

kinderchildren

leyenenread

t

(Yiddish)

‘What did he not want the children to read in school?’

Given the lack of island effects, I suppose that complementisers in Icelandic and Yiddishdo incorporateΠ, albeit only abstractly (in Baker’s (1988) sense). In this respect, theyresemble the Romance restructuring verbs discussed by Roberts (1997), or theincorporatingfaire-causatives identified by Guasti (1993), which appear to incorporatelower verbs without forming a single complex verbal unit for the phonetic output. As CincorporatesΠ, its [Tense] feature is checked, and the full CP can in principle appear inany position a non-verb-second CP may appear in.

‘Abstract’ incorporation by C is appears to differ from over incorporation in severalrespects. For one thing, abstract incorporation is insufficient to support the affixalproperties ofΠ. For another thing, abstract incorporation is less constrained by localityconsiderations, so thatΠ appears to incorporate abstractly past a Top head, where overtΠ-to-C movement is blocked by the presence of Top. Therefore, in embeddedsubject-initial verb-second clauses in Yiddish and Icelandic, where the subject must beinterpreted as/checked by Top, the finite verb must still raise to supportΠ: (115). AsVikner (1991) shows, the expletive pronounses(Yiddish) andþað(Icelandic) must be‘topics’—meaning, in my terms, that they must be checked by the Top head. As a result,they always appear with verb-second word order in embedded clauses.

(115) a. IkhI

veysknow

azthat

es isis

gekumencome

aa

yingl.boy

(Yiddish)

b. ÉgI

veitknow

aðthat

það hefurhas

komiðcome

strákur.a boy

(Icelandic)

Page 62: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 2. EXPLAINING SUBJECT/NON-SUBJECT ASSYMETRIES 61

2.7 Summary

To conclude, we have seen that the principled strictures of phase theory, together withPesetsky and Torrego’s idea that C bears a ‘strong’, uninterpretable [Tense] feature,combine to provide a simple account of both residual verb-second structures in English,and of verb-second structures in general in the Germanic languages. The structures whichwe are led to involve CP-recursion for most clauses, both root and embedded types, withthe upper part of the ‘left periphery’ left without phonetic interpretation in all rootclauses. The functional heads Top and Foc may appear between he upper C and the lowerC (Π), when the interpretation of the clause requires their presence. However, Top andFoc lack EPP features (except in German and Dutch, possibly), so they check onlyphrases which are raised into accessible positions by theΠ head. All operations whichaffect the position of phrases and heads in the recursive CP structures are instances of thebasic AGREE operation, which operates on the [Tense] feature and on full and defectiveA-bar feature sets like [wh], [topic], [focus], [MD], and [MD].

Page 63: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

Chapter 3

The A-bar properties of subjects

In chapter 2, I argued that subjects raise to the specifier position of a particularcomplementiser phrase, one which appears immediately above TP, and which is normallythe bottom part of a CP-recursion structure. The argumentation is based in large part onthe idea that only certain categories are (strong) phases, such as vP and CP. Although thisidea is fairly well motivated by Chomsky (2001b), one could certainly question whetherthis motivation—largely drawn from theory-internal evidence and conceptualgrounds—is strong enough to support the conclusions which I have come to. Inparticular, it is a significant question whether the arguments which I have laid out arecompatible with other interpretations of thenatureof the lower phase head. In otherwords, would it be possible to identifyΠ as some other category without abandoning theexplanations provided in the last chapter for phenomena such as subject-aux inversion.

It seems to me that a re-labelled theory might well be possible to design. Suppose,for example, thatΠ were taken to be equivalent to a Mood head, or to the AgrS of Belletti(1992), or to Shonlsky’s (1994) AgrC. Then it would be necessary to say that MoodP orAgrS or AgrC is a phase head which has a [Tense] feature to check, but much of the restof the model would remain as before.

What significance is there, then, to my claim thatΠ is a token of the C category?The significance lies in the A-bar properties which we normally associate with theSpec-CP position. Unlike the specifier for TP, for small clauses, and in some cases for vP,the specifier for CP is in an A-bar position, with the properties which that entails. Thus,DP specifiers for CP are not typically involved in Case-checking relations, and theycannot serve as antecedents in A-binding relations, unlike DPs in A-positions. Phrases inA-bar position are also subject to a variety of reconstruction effects, which do not occurin A-positions. In (1), for example, the reflexive pronoun inside the specifier of CP isbound by the lower subject, because some of the content of the Spec-CP PP is

62

Page 64: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 63

reconstructed into a lower A-position for LF interpretation.

(1) They wanted to know [CP in which picture of himself [TP Renoir had painted afake moustachet ]]

In claiming thatΠ is a type of complementiser, I am affirming that the contents ofSpec-ΠP will pattern in such respects with the contents of Spec-CP, and not withcategories in an A-position.

The same issues have arisen in other studies. Thus for example, in Diesing (1990),the claim is put forward that topics can raise to Spec-IP in Yiddish, and that this positioncan function sometimes as an A-position and sometimes as an A-bar position. (Similarclaims are made for Icelandic and English by Rögnvaldsson and Thráinsson (1990) andPesetsky (1988), respectively.) And a similar claim could certainly be made aboutΠ. Onemight imagine, for example, that Spec-ΠP would count as an A-position if it wereattracted by the [Tense] feature and as an A-bar position if it were attracted by [MD] or[MD].

In order to arrive at a principled account of thethat-trace effect, though, it isnecessary to adopt a strong stance with respect to the A-bar properties of Spec-ΠP. In thischapter, therefore, my goal is to show that Spec-ΠP isalwaysan A-bar position.

The arguments for this claim must be fairly complex, because the usual diagnosticsfor the A/A-bar status of a position are not applicable. Cross-over tests rely on thepresence of a DP between a source location and the target location in question. But therewill never be a DP between Spec-ΠP and Spec-TP. Similar problems arise if we try toformulate example sentences to test involving A-binding relations.

Fortunately, there are a handful of structures which allow the issues to be addressed.

3.1 Polarity licensingonly

A first type of evidence that a subject occupies an A-bar position comes from certain factsabout polarity licensingonlyDPs. Progovac (1992) observes thatonlyphrases originatingin complement or adjunct positions can license polarity items only if they move toSpec-CP.1

(2) a. Only his girlfried does John give any flowers.b. ?*John gives only his girlfried any flowers.

The exception to this generalization is anonlyDP in subject position.

1It should be noted that individuals differ on the force of the judgments reported in (2). For somespeakers, (2)[b] appears to be fully acceptable. My own judgments accord with those Progovac reports.

Page 65: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 64

(3) Only Maryshowedanyrespect for the visitors.

Following Chomsky’s 1986 vacuous movement analysis, Progovac suggests thatsubjects are only exceptional in this respect in the surface syntax, and that at LF anonlyDP in subject position can license a polarity item if it moves to Spec-CP. Progovacprovides no reason why a non-subject should not be able to undergo the same sort of LFmovement, however.

If subjects already occupy an A-bar position—Spec-ΠP—then the properties ofonlysubjects are entirely unexceptional. Consider how anonlyphrase in the inversionstructure (2-a) is interpreted. As it occupies an A-bar position, it undergoes partialreconstruction to form an interpretable A-bar chain in which the head is an operator andthe foot contains a restricted variable.

(4) [ only x ] , John gives any flowers to [x: his-girlfried(x) ]

We may suppose that the content which remains in the operator then licenses thenegativelarity itemany.

The analysis of a subjectonlyphrase is entirely parallel. If the subject occupiesSpec-C, then it occupies an A-bar position, like the inverted phraseonly his girlfriendin (2-a). Then the same type of partial reconstruction may take place, producing thestructure (5).

(5) [ only x ], [ x: Mary(x) ] showed any respect for the visitors.

Once again, the content of the operator which remains in Spec-ΠP licenses the negativepolarity itemany.

3.2 Locative inversion

A second source for evidence about the A/A-bar status of Spec-ΠP can be found in thelocative inversion construction, and in particular, in considering the interactions betweenlocative inversion and the binding theory. In order to do so, it is necessary first to developan account of the basic structural properties of locative inversion sentences, such as (6).

(6) a. To Martin fell the task of introducing the first speaker.b. Into the lounge sauntered three gents in galoshes.

The fronted PP somehow replaces the subject at the start of the clause. The subject itselfoccupies a somewhat unexpected position at the right edge of the clause, where it istypically interpreted as the ‘narrative focus’ (Bresnan, 1994).

Page 66: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 65

The position of the preposed PP in locative inversion has been hotly debated. (See,for example, Postal (1974); Bresnan (1994); Coopmans (1989); Dikken and Næss (1993);Watanabe (1993); Collins (1996) to mention a few.) These debates have been primarilyconcerned with how the verb relates to the postposed subject, rather than with the PPproper. The general consensus in the literature appears to be that the fronted PP occupiesthe same position as subjects usually do.2 The reason to believe this to be the case is thatfronted locatives participate in several constructions/constraints which are otherwiserestricted to subjects.

Like true DP subjects, preposed locatives can undergo raising from an infinitivalcomplement.

(7) Inside this pillowslip appear [t to have been concealed the missing documents ].

In ECM contexts, a preposed locative can still appear in clause-initial position.

(8) I imagine beside this very river to have wandered Basho.

In addition, locative inversion is (marginally) possible inside afor-to infinitive.

(9) a. ?[CP For through these doors to saunter my kindergarten teacher] would shockme.

b. ?We would very much appreciate [CP for under the window to be hung threesmall ukiyoe prints. ]

(The marginal quality of such sentences appears to reflect, in part, the awkwardness offorbeing followed by a preposition. With smaller prepositions—to, by, in—the results areeven more awkward.)

Finally, wh-extraction of a preposed PP is subject to thethat-trace effect (Bresnan,1972), which otherwise affects only nominative subjects.

(10) *The lounge into which Claudia insisted (*that) had sauntered three gents ingaloshes.

If fronted PPs occupy the same position as normal subjects usually do, then theymust occupy the Spec-ΠP position. Then (6-b) has the structure in (11).

(11) [ΠP into the loungeΠ [TP sauntered the three gents in galoshes ]]

What needs to be explained then is how an argumental PP may be attracted byΠ in placeof the usual subject.

2Dissenting voices on this point include Dikken and Næss (1993) and Rochemont (1990).

Page 67: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 66

Clearly, unless the properties ofΠ are quite different in locative inversion sentences,Π must be able to check its [Tense] feature by attracting the locative PP. We must assumetherefore that argumental PPs may optionally bear a [Tense] feature. In this respect, theyresemble nominative DPs.3

The second assumption which we must make concerns the position from which thelocative PP is attracted byΠ. Π will have access only to the closest category which bearsthe [Tense] feature under theMinimal Link Condition. This means that the locative PPcannot be attracted directly from its base position within the verb phrase, because thesubject and T will always be closer toΠ than the contents of VP are. We must supposethat locative PPs in locative inversion raise to a position where they are at least as close toΠ as the other [Tense] bearing categories are. The only position with these characteristicsis Spec-TP. Since the subject always raises to Spec-TP too, the necessary structure is onein which T attracts multiple specifiers, along the lines of Richards (1997):

(12) TP

���HHH

DPi T′

��� HHH

PPj T′

�� HH

T vP�� HH

t v′

�� HH

v VP��HH

V tj

Although this structure is never found with this form, there is at least evidence thatlocative inversion involves some degree of A-movement of the PP upwards towards T.4

First, as shown by Aissen (1978), locative inversion cannot make use of adjunct PPs:

(13) a. *On Tuesday will arrive the package.

3Although there is no obvious morphological evidence to support the idea that PP may bear a [Tense]feature, there may be evidence from other quarters. There is a long tradition that suggests that P and Cshare certain featural properties, and complementisers are often homophonous with certain prepositions(Kayne, 1972). If C always bears a [Tense] feature, as argued already, then it is more plausible that [Tense]may be found on P, too.

4Collins (1996) and van Gelderen (1997) defend a similar claim, although for them the relevantchecking operation is a [D] feature, checked by T. The basic idea is also anticipated by analyses of Hoekstraand Mulder (1990) and Dikken and Næss (1993). Similar in crucial respects is the analysis of Ura (2000).

Page 68: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 67

b. *With a microphone will lecture Ken.

Assuming that adjunct PPs occupy anA-position, they cannot raise to Spec-TP, anA-position. This would be a case of “improper movement”, which is generallyillegitimate.

Secondly, as discussed in detail by Bresnan (1994) and Levin and Hovav (1994),locative inversion is impossible with transitive or purely agentive verbs.5 The transitiveexamples in (14) are ill-formed, for example:

(14) a. *To Sharon tossed the football Karen.b. *In the garden planted tulips Wilhelmina.

If the locative PP must always undergo movement to Spec-TP, then the status of the (14)examples can be explained by theMinimal Link Condition. As the object DP is higherthan the locative, T cannot look past it to attract the PP to an A-position.

(The object is also unable to raise to Spec-TP. I assume this is related to theaccusative Case features of the object, which presumably clash with the Case features ofT.)

As for the agentivity constraint on locative inversion, Bresnan (1994) provides theminimal pair in (15):

(15) a. Through the hoop shot the arrow.b. *Through the hoop shot the archer.

In this case, the best solution appears to be an extension of the account of the (14)sentences. There is clearly an understood object in the (15-b) example. Let us supposethat this object has syntactic reality, i.e. that there is a null object, the arrow, in objectposition in this sentence. The movement of the locative PPthrough the hoopin (15-b)will then be blocked by theMinimal Link Condition, just like in the more transparentlytransitive cases in (14).

The issue becomes even clearer when locative inversion is found in a raising context.For many English speakers, the PP can raise into a matrix clause (Bresnan, 1994):

(16) To the dogs appear to have been tossed many of the leftovers.

Here, the fronted locative originates as the complement of the verb in the lower clause.Movement directly to Spec-ΠP in the matrix clause would certainly not satisfy theMinimal Link Conditionfor the EPP feature ofΠ.

5Levin and Hovav (1994) show that the construction is not restricted to unaccusative verbs, however.

Page 69: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 68

If PP can be attracted by T, though, the derivation of (16) poses no problems. Thehead of the lower TP may attract both the lower subject and PP, in turn, generating thestructure: (17-a).

(17) a. [TP [DP many of the leftovers ] [PP to the dogs ] to have been tossedt t ]b. [TP [DP many of the leftovers ] [PP to the dogs ] T appear [TP t t to have been

tossedt t ]]

When the matrix clause is then built up to the TP level, the matrix T can then attract boththe lower clause subject and then the locative PP, to make (17-b). At this point, Merge ofΠ will allow movement of the PP to Spec-ΠP if Heavy-NP shift dislocates the subject tothe right.

Since the matrix T must attract a PP to permit the raising sentence to be formed, weexpect that any other PP which is situateden routemight block the locative inversion.The expectation is confirmed:

(18) *To the dogs appear to us to have been tossed many of the leftovers.

The structure in (18) is not attested in the surface syntax of English; an example like (19)is ungrammatical.

(19) * [DP The leftovers ] [PP to the dogs ] were thrown.

On the face of it, the ungrammaticality of sentences like this could show that PP cannotraise to TP, in which case the derivation of locative inversion sentences is different fromwhat I have argued. It is not clear what the best explanation for this fact is, though. Onepossibility is, of course, that the structure in (18) is simply not possible in English,because T does not attract multiple specifiers, contrary to what I have claimed above.Another possibility is that only one specifier is subject to Spell-Out in TP. Pesetsky(2000) shows that English multiple-wh questions always have at least two wh-phrasesattracted to C, although only the highest one is ever given phonetic content in thatposition. It might well be that the same is true of T. T might then freely attract multiplespecifiers, with only the topmost one interpreted by Spell-Out in the Spec-T position. Yeta third type of explanation might make reference to Stowell’s (1981)Case ResistancePrinciple (CRP), which requires that phrases in which the head checks Case not appear ina position where Case is checked. As finite T checks nominative Case, and as P checksaccusative Case, the CRP would require that a PP specifier to finite T undergo some sortof movement to a different position. In any case, there are explanations available for theunacceptability of (19) which do not conflict with the arguments that locative inversioninvolves a degree of A-movement.

Page 70: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 69

The structure in (12) is not yet one which will allowΠ to attract the PP. The problemis that the subject is still closer toΠ than the locative PP is. Although both are specifiersfor TP, and hence as close toΠ as T is, in multiple specifier structures, the higher onealways counts as closer to an external attractor (Richards, 1997).

Pesetsky’s (1994) proposals suggest how this last difficulty may be resolved. Sincesubjects always appear at the right edge of the clause in locative inversion, they mustsomehow be displaced from their usual position. (The right word order is not ensured ifsubjects simply remain in their base position in Spec-vP or lower.) Pesetsky suggests thatDP subjects may undergo movement to the right edge of TP as long as they do not needto raise to a higher A-bar position (equivalent to Spec-ΠP). The same mechanism isinvolved in Romance stylistic inversion of subjects. Let us adopt this approach, includingthe notion that rightward movement of the subject is a stylistic operation. Then themovement of the subject from Spec-TP will leave the locative PP as the highest specifierin TP, so that the [Tense] feature ofΠ can attract the PP. The full structure of (6-b) willthen be (20):

(20) ΠP

������

HHHHHH

PPj

����PPPP

into the lounge

Π ′

������

HHHH

HH

Π TP

����

����

HHHH

HHHH

TP��� HHH

ti T′

��� HHH

tj T′

���

HHH

T vP����

PPPP

ti saunteredtj

DPi

����PPPP

the three gentsin galoshes

Page 71: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 70

Crucially, in this structure the trace left by the PP raised to Spec-ΠP is lower thanthe subject of the sentence, which is itself lower than the PP.6 In this respect, locativeinversion provides a structure analogous to wh-movement, which uses A-bar movementto raise a non-subject past the subject. And it is understood that in the wh-movementcase, it is always the position of the trace which is significant in binding relationsinvolving a subject and a displaced wh-phrase. The locative inversion structure thereforeis one in which we can test to see whether the PP behaves as if it were higher than thesubject or lower as far as the binding theory is concerned.

Now let us turn to some standard binding theoretic tests for height.Consider the following sentences, in which a PP has been preposed with the subject

postposed ((21)).

(21) a. Into each other’s arms raced Ted and Joan.b. Beside a picture of himself was standing the count.

The ((21)) sentences show that an anaphoric element contained in the clause-initial PPcan be bound by the nominative DP. In example ((21)a), the reciprocaleach otherisbound by the nominativeTed and Joan. In example ((21)b), the anaphoric element ishimself, which is bound by the nominative DPthe Count.

Anaphor binding requires that the binder c-command the bound element, so thenominative objects in ((21)) must c-command the anaphors contained in the clause-initialPP at LF. This is impossible in (21) if the PP is bound in its surface position, where itcannot be c-commanded by anything else.

Now consider the alternative. If the preposed PP heads an A-bar chain, rather thanan A-chain, then it may be reconstructed into the position that it originates in within theverb phrase. Since the PP raises from a lower specifier position in TP, reconstruction ofthe PP will place its content in a position lower than the nominative subject (and itstrace). The reciprocal phrase can then be bound, as it must.

Similar conclusions can be drawn based on other principles of the binding theory.Pronouns cannot be coreferent with a c-commanding DP in the same clause. In this light,consider examples ((22)).

(22) a. Beneath a photograph of her sat Madonna.b. Past a statue of them raced the Crown Prince and his horse.

6The idea that locative inversion involves both A-movement andA-movement steps originates withDikken and Næss (1993). The motivation for movement in their account is quite different, however, as isthe landing site for theA-movement step.

Page 72: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 71

The pronounsher andthemcannot be coreferent with the DPsMadonnaandthe CrownPrince and his horse, respectively. This constraint on the reference of the pronounscannot easily be explained in surface syntactic terms. It can, however, be explainedreadily by principle B if the PP is again reconstructed from its surface position back intothe verb phrase. In that case, the content of the PP will be c-commanded by the subject,so that the pronouns will be c-commanded at LF by the DPs with which coreference isblocked.

A similar argument comes from examples like ((23)), in which a name contained inthe clause-initial PP is coreferent with the nominative object. (The italics indicateintended coreference.)

(23) a. *BehindTed’sdesk satthe bastard.b. BehindTed’sdesk satthe bastard’smother.

When reconstruction replaces the content of PP in its base position, the resultingconfiguration is ((24)).

(24) [TP [TP e [PP behind Ted’s desk ] T [vP sat ]] the bastard ]

The backwards binding effects with anaphors show that reconstruction of the fronted PPis possible. The non-coreference effects with pronouns and names allow us to draw astronger conclusion: reconstruction of the fronted PP is obligatory.

To conclude this rather involved argument, given the reconstruction effects found inlocative inversion, the preposed PP does not occupy an A-position. And since theposition of the fronted PP must be Spec-ΠP, Spec-ΠP itself must always count as anA-bar position.

3.3 ECM with wager-class verbs

The third source for evidence concerning the A/A-bar status of the Spec-ΠP positioncomes from the peculiar ECM structures in (25). These data illustrate a pattern which isfound with verbs likewager, assure, say, insist, allege, which have an agentive subject andwhich refer to states of cognition. Pesetsky (1992) dubs this the ‘wager-class’ of verbs.

(25) a. *They say Jean to be enjoying her retirement.b. *We wagered Marcel to have completed the marathon.

(26) a. Jean, who they say to be enjoying her retirement, . . .b. Which accountant did you wager to have completed the marathon?

Page 73: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 72

As first observed by many authors (Ruwet, 1991), normal ECM is not possible withinfinitival complements to verbs of this class. However, the subject of the infinitival canhave Case features checked if it is extracted by wh-movement.

In French, even verbs likecroire ‘believe’ or imaginer‘think’ behave like Englishwager-class verbs.

(27) a. *Onwe

imaginethink

[ JoséeJosée

êtreto be

maladesick

]

b. Josée,Josée

onwe

l’imagineher-think

[ t êtreto

maladebe

] .

sickc. Qui

whoimaginent-t-ilsthink-they

[ t êtreto

maladebe

]?

sick

This contrast between clause-external and clause-internal subjects has never beensuccessfully explained. The data is almost paradoxical, especially in French where the(27) pattern is the general case. A child acquiring French must be able to deduce theappropriate grammar from positive data. Given the acceptability of (27-b) and (27-c), achild might well assume that ECM of the English variety will be generally available inFrench. So an account of the (27) data must explain not only why the judgments fall outas they do, but also why children do not overgeneralize on the basis of the (27-b,c) data.

As will be shown, an explanation for this data in both languages is readily availablein the present model, but only if Spec-ΠP is necessarily an A-bar position, on whichCase-features cannot be checked.

Let us consider the data in the context of the theory developed here, beginning witha comparison of the ECM syntax ofbelieve-class verbs in both languages. Let us supposethat in French, as in English, the complement of an ECM verb can be smaller than a fullCP. The two languages must differ in some respect, so suppose that the complement ofcroire is larger than the complement tobelieve. In other words, in French, the infinitivalcomplement tocroire, imaginer, etc. is aΠP. This difference in complement size issupported by a second difference between the two languages: Frenchcroire-class verbsmay have controlled infinitival complements; Englishbelieve-class verbs may not.

(28) a. PaulPaul

imagineimagines

[ PRO êtreto be

douégifted

àto

chantersing

]

‘Paul imagines himself to have the gift of singing.’b. *Paul imagines PRO to be gifted for singing.

Suppose as well, as is standard, that Case is checked only on the head of an A-chain.Then simple ECM complement clauses in French and English will have the contrasting

Page 74: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 73

structures in (29).

(29) a. *Onwe

imaginethink

[ΠP JoséeJosée

Π [TP t êtreto be

maladesick

]] .

b. We believe [TP Josie to be sick ].

In the English (29-b), the matrix verb can check Case directly on the Spec-TP subject,since no phase boundary intervenes. In the French (29-a), however, the presence of theΠP phase boundary blocks checking of Spec-TP under thePhase ImpenetrabilityConstraint7. The contrast in grammaticality judgments between the two languagesreflects a simple difference in the structure of complement clauses.

Now consider the problematic examples (27-b) and (27-c) in theΠP framework,where they will have the structures in (30).

(30) a. Quiiwho

imaginent-ilsthink-they

[ vP t . . . [ΠP t Π [TP ti êtreto be

maladesick

]] ?

b. Josée,Josée

onwe

li’[vPher

ti imagine [ΠP

thinkti Π [TP ti être malade

to]]]be sick

Successive cyclic wh-movement in (30-a) takes place with a defective [wh] feature to thematrix v, which then attracts the interrogative pronoun to its edge from within thecomplement clause, which is then followed by movement ofqui from Spec-v to the rootSpec-ΠP position. At each step, successive cyclic movement deletes intermediate A-bartraces, so that the final A-bar chain is binary, containing the operator at the head and thehead of an A-chain at the foot.8 We may suppose a similar process for the cliticplacement operation in (30-b), with an intermediate A-bar trace left at the vP phase edge.

If we restrict the focus to the context which matters for ECM in (30-a), the structureis the vP phase in (31):

(31) [vP quii imaginent [ΠP t Π [TP ti être . . . ]]]

The question is how it is possible for the matrix verb to check Case on the trace inSpec-TP. And close examination indicates that there is no principle which will actuallyblock this checking operation in this structure. ThePhase Impenetrability Constraint,which would normally ensure that the matrix verb does not see into TP has been satisfied

7In Chomsky’s 2001a model, Case-checking past a phase boundary is permitted as long as the nexthigher phase is not yet constructed. In (29-a), the Case-checking operation will not be possible until thematrix v is introduced, so that the matrix vP phase is complete. ThePhase Impenetrability Constrainttherefore constrains Case-checking even in this formulation.

8The trace deletion process is considered in much more detail in chapter 4.

Page 75: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 74

already by the movement ofqui from Spec-ΠP to the edge of vP, so under thePrincipleof Minimal Compliance, thePhase Impenetrability Constraintis no longer active inconstraining operations between v and the contents ofΠP. . Case-checking of theSpec-TP trace therefore occurs, and the derivation may proceed on to the next stages.9

And ceteris paribus, the same mechanisms will apply in the derivation of (31-b).The only innovation in this approach to French ECM is the idea that matrix verb

may look past the Spec-ΠP position in checking Case features on a trace in Spec-TP. Thatthis option is in any case available is readily seen in Englishfor-to infinitives.

(32) Peter hoped passionately [CP for-Π [ΠP his new pipest [TP t to have a strongsound ]]]

In (32), the subject Case features are checked by the complementiserfor, which appearson the other side of the subject in Spec-ΠP. (ThePhase Impenetrability Constraintisobviated for theΠP phase by incorporation ofΠ by C.) C cannot check the subject inSpec-ΠP, since this is an A-bar position, so it must be able to look into TP and check thefeatures on the trace in Spec-TP.

A similar process occurs in thefor-to infinitive in (33), where Spec-ΠP is occupiedby a preposed locative PP.

(33) ?[CP For [ΠP under the bed [TP [TP t t to hide ] the puppies ]]] would beinconvenient.

Here again, thefor complementiser must look past the phrase in Spec-ΠP to check thesubject trace in Spec-TP.

Let us compare this account to previous attempts to deal with French ECM. Kayne(1981) and Massam (1985) propose an account of sentences like (30) in which all Frenchinfinitival complements are CP categories and where the Case-checking relation betweenthe matrix verb and the complement clause subject is established in a Spec-CP position.Ura (1993) and Boškovic (1997) offer updated versions of this approach, but stick withthe basic idea that Case is checked on an A-bar position.10 The (slightly idealized)structure involved would be (34), whereimaginercannot check Case on an argumentwithin the complement clause TP, but where the trace in the Spec-CP position isaccessible.

9The order of checking operations follows the pattern observed by Bobaljik and Branigan (2004),because the extra [wh] feature added to the matrix v will be checked before the inherent [accusative] Casefeature.

10Boškovic (1997) adds to the basic Kayne/Massam theory an account of why null Case is not on thesubject in (33). His description of the TP-internal mechanics of Case-checking is consistent with myproposals.

Page 76: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 75

(34) Quii . . . [VP imagine [CP t C [TP t être malade ]]]

The first problem with this approach is the nature of the Spec-CP position, whichtypically counts as an A-bar position. In other contexts, Case is always checked on thehead of an A-chain; it would require a substantial complication of the theory to allowCase to appear on the head of an A-bar chain as well. Furthermore, it is not obvious inthis approach that the checking of Case features on the trace will ensure convergenceanyway. It is clear that the features checked on the head of an A-chain are shared by theother members of the chain, presumably because the different members of the A-chainare identical tokens of the same linguistic unit (Chomsky, pear). But the features checkedon the head of an A-bar chain are usually not shared with the A-chain at its foot; instead,the reconstruction operation factors the head and foot into two distinct components of thechain, in which the operator features are kept as part of the head and eliminated from thefoot. If this procedure applies in the usual fashion in a structure like (34), then the Casefeatures checked on the trace will never be checked on the A-chain inside TP, so that thederivation will crash anyway.

More complex structures indicate that this second problem must be taken seriously.

(35) *Quiwho

PaulPaul

a-t-ilhas-he

ditsaid

[CP t qu’ilthat-it

ahas

étébeen

imaginéthough

[ t êtreto

maladebe

]] ?

sick

The ungrammaticality of (35) must result from the absence of Case on the trace left in thelowermost clause. This trace is not checked internal to its clause, which is non-finite, andit cannot be checked by the passive verb in the second matrix clause either. But if Casecould be checked on an intermediate Spec-CP trace, with checked features shareddownwards in the chain, then the verbdire in the root clause should check the trace inSpec-CP of its complement. Since this option does not make the sentence grammatical,we must conclude that Case-checking cannot be initiated in an A-bar position.

The way to avoid these complications would be to treat Spec-CP as an A-position,exceptionally. This is clearlyad hoc, but might at least allow the data to be explained tosome extent. Unfortunately, such a step would eliminate the distinction between A andA-bar positions. What is worse, in doing so, the account of the distinction between (27-a)and (27)[b,c] is lost. If Spec-CP were an A-position, then there would be no reason forthe subjectJoséein (27-a) to remain in its Spec-TP position. With unchecked Casefeatures, the subject should be free to raise to a higher A-position to allow the derivationto converge.

The analysis of French ECM indicates howwager-class complementation may betreated in English. The relevant data is (36):

Page 77: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 76

(36) a. *Mary said Brian to be a fine singer.b. The guy who Mary said to be a fine singer.c. *Sam assured us Sylvie to be ready for the job.d. Who did Sam assure them to be ready for the job?

With these verbs, an embedded subject extracted under A-bar movement may have itsCase features checked, while a subject which remains in its clause cannot. Just as inFrench, then, these verbs must select aΠP complement, so that the structure of (36-a) is(37):

(37) Mary said [ΠP BrianΠ [TP t to be a fine singer ]]

The matrix verb cannot check past theΠP phase boundary, so it cannot check features onthe trace in Spec-TP. If the subject is extracted, however, as in (37-b), the structure willbe:

(38) [vP who said [ΠP t Π [TP t to be a fine singer ]]]

As the matrix v satisfies thePhase Impenetrability Constraintin attractingwho, it is freeunder thePrinciple of Minimal Complianceto look pastΠP to check the subject trace.

Postal (1974) observes thatwager-class verbs also participate in ECM when thesubject of an infinitival complement undergoes Heavy-NP shift. Boškovic (1997) showsthat the same is true with French, when other variables are controlled, as shown in (40).

(39) Pam wagered to have won the race the candidate who promised to keep the plantopen.

(40) PierrePierre

croitbelieves

êtreto be

doctoressesdoctors

lesthe

femmeswomen

qu’ilthat he

ahas

rencontréesmet

l’annéeyear

dernièrelast

àin

la NouvelleNew

Orléans.Orleans

These cases can be accounted for if Heavy-NP shift is an operation which involves astage of movement through Spec-vP. With respect to (39), the structure of the matrixclause vP must be (41) at a certain point in the derivation:

(41) [vP [DP the candidate who . . . ] wagered [VP e [ΠP t Π [TP t to have won the race]]]]

Again, attraction of the subject DP bywagersatisfies thePhase ImpenetrabilityConstraint, so that under thePrinciple of Minimal Compliance, wagercan look into TPin subsequent checking operations. The Case features of the trace in Spec-TP can

Page 78: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 77

therefore be checked by thewager. Subsequent rightward movement to whereeverHeavy-NP shift places phrases will then produce the surface word order.11 The French(40) example can be explained along similar lines.

The general pattern for ECM verbs is that in both French and English, they mayselect aΠP infinitival complement. In French, they can also select a full infinitival CP;with some English verbs, they can also select an infinitival TP. In French, null Case maybe checked in a bareΠP, while in English only full CP contains a null Case checkinghead.12 In neither language, however, can Case features be checked directly on a DP inSpec-ΠP.

3.4 Learnability

The difference between ECM in French and English largely reduces to a contrastinvolving the size of infinitival complement clauses with verbs of cognition. This contrastmust then be within the bounds of parametric variation allowed by Universal Grammar. Itmust also reflect states of language which can be acquired by children on the basis of dataavailable to them in the normal time period, presumably on the basis of positive evidence.A complete account of the cross-linguistic variation should include some insight into thelearning path which makes such variation possible.

One possible learning algorithm suggests itself if we consider the way in whichchildren might deduce the properties of the functional heads in the course of acquiring alanguage like English. Suppose that the deferred multiple Spell-Out theory sketched outabove is supplied to a child’s mind by Universal Grammar, as indeed it must be if it iscorrect at all. Then a child knows much of the structure of a simple clause before sheeven confronts the data in her parents’ language. She would know that every full clausemust include two phasal heads, with one at the root, because otherwise there would be nopossibility of applying Spell-Out to a simple sentence. And it is quite plausible that achild may know that sentences are headed by T, which requires a vP complement. Then achild may assume for any simple sentence a minimal structure like (42), with the root Cnecessarily lacking in phonetic content.

11Antisymmetry-theoretical variants of this idea are easy to imagine, as well. What would be minimallynecessary in an account which eschews rightward movement would be an operation which raises v′ pastthe DP in Spec-vP to a higher position in the clause. Takano’s (2002) arguments suggest that somethingalong these lines might be necessary.

12See Boškovic (1997) for discussion of the difference in null Case checking between French andEnglish.

Page 79: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 78

(42) CP��� HHH

C ΠP

���

HHH

Π TP

���HHH

T vP��� HHH

subject v′

v V

Among the properties of the adult language which the child now needs to learn arethe types of structures which may serve as embedded clauses.

(43) Emily said her mommy can talk to dogs.

Given a sentence like (44), the surface word order is compatible with several hypotheses:the complement clauseher mommy can talk to dogscould be a TP, aΠP, or a CP. (Let usexclude a bare vP structure from consideration, on the basis of the modal verbcanto theleft of the verb.) And the problem is that empirical evidence is probably not availablewhich will allow a child to choose between these possibilities. We can draw thisconclusion on the basis of the theoretical literature. All three possibilities are defended,by linguists with an extensive knowledge of what data could possibly bear on the issue.(TheΠP hypothesis is defended only here, of course.)

If the data is insufficient, and if we assume that the child’s grammar must eventuallyconverge on some solution, then the course of acquisition itself must select from amongthe options. And the optimal solution for acquisition should be one in which the childdoes not need to posit any properties in embedded clauses which are not present in rootclauses. In that case, the default will be that embedded clauses are CPs, simply becauseroot clauses are always CPs. (This hypothesis embodies ‘degree-0 learnability’, inLightfoot’s (1991) terminology.) Any departure from the default will require positiveevidence to encourage the child to adjust the relevant parameter.

Now consider the consequence of this reasoning for the acquisition of infinitivalclauses, which only appear in embedded contexts. In the absence of any evidence to thecontrary, the default assumption may still be that an embedded clause is a CP. It followsthat the unmarked structure for infinitives is a control structure, with a PRO subject.

Non-control infinitives must then be acquired on the basis of positive evidence. ForEnglish ECM structures, it is easy to see what type of positive evidence might beavailable: an ECM sentence itself will suffice.

Page 80: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 79

(44) Teddy believes the food to be yummy.

Given the thematic structure, a sentence like (44) is incompatible with a grammar inwhich all embedded clauses are CPs. Only with a TP complement can the subject have itsCase/φ features checked by the matrix verb (or by any head at all). Therefore, the child’sgrammar must revise itself to allow (44) to be correctly generated.

One way to revise the grammar would be by allowing an infinitival TP to appear inany embedded position. This would obviously produce the wrong results in adultlanguage, since TP infinitives do not appear as the complements to activewager-classverbs or to factives or implicatives.

(45) a. *Donna muttered her feet to be sore.b. *Daisy loves the lightening to flash.c. *Daisy realized the solution to be correct.

If a child is able generally to include TP infinitival complements in her grammar, thennegative evidence of some sort will be necessary to lead her towards an adult state, whereTP complements are more restricted in their distribution.

A more effective strategy for the child would be to entertain a more restricted changein the grammar, by allowing TP complements to appear only as the complements to aparticular set of heads. Given Pesetsky’s generalisation, we can identify the right set ofheads as the non-agentive verbs. The conclusion which a child should draw then is thatEnglish infinitives are TPs only when they appear as complements to non-agentive verbsof cognition. Elsewhere, larger complement clause structures will be required, in theabsence of evidence to the contrary.

As for the acquisition ofwager-class complements, a similar process must obtain,but with less extreme results. Given the grammaticality of a sentence like (46), which wemust assume resembles data available to a child, she will conclude that an infinitivalcomplement to verbs of cognition, agentive and non-agentive, are smaller than CP.

(46) Which school has Sean assured themt to be the most engaging.

The smallest reduction in clause size which is possible is a change from CP toΠP, so thisis the first hypothesis she should entertain. And the theory thatwager-class verbs takeΠPcomplements is entirely compatible with the data, so our child learning Englishcomplementation will never need to revise this hypothesis further.

The task facing a French child is simpler. Since pure ECM is never found in French,and since partial ECM is found, the evidence will be available to lead a French child tosuppose that infinitival complements may beΠP for verbs of cognition. No furtherdeduction will be necessary to make the child’s grammar compatible with the data to

Page 81: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 80

which she is exposed.The general learning strategy which works here is one in which the core assumption

is that clause structure is the maximal available until evidence indicates otherwise. Thisin turn suggests a mechanism by which the checking properties ofΠ might be acquiredby children learning languages like English, French, or another Germanic language.Since a child is born knowing that basic root clause structure must match the tree in (42),consider what analyses are available for a simple declarative root clause like (47).

(47) Daddy will help you.

Let us suppose that the child has identified the meanings and categories for the mainlexical items:Daddy, help, you. Then the verb phrase will automatically consist of theverbhelpand the objectyou. The subjectDaddymust appear either within vP or in ahigher position to which it has raised. But the presence of the wordwill between thesubject and the verb indicates that the subject is notin situ in this language. So basicword order indicates to the child that there is a feature with an EPP property whichattracts the subject in this language.

Since the child knows that there are at least two functional heads available outside ofvP, she must know determine whether it is T orΠ which attracts the subject in (47). Onceagain, the decision should come from the language acquisition procedure itself, ratherthan from an analytic choice made freely by the child. So one of the two heads must be apreferable choice as the attractor at this point. And since maximal structures seem to bethe default choice—based on what we have concluded in the analysis of infinitivalcomplement—it should beΠ which is the preferred head to attract the subject in simpledeclaratives. Then (47) will be given the analysis in (48) at an early stage of acquisitionof syntax.

(48) ΠP

���HHH

D

Daddy

Π ′

���

HHH

Π TP

���

HHH

T

(will)

vP��� HHH

v VP�� HH

V

help

D

you

Page 82: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 81

In this hypothetical account, it is unclear at this point what the nature of the featureis which should bear the EPP property inΠ. But this indeterminacy will be quicklyresolved on the basis of simple questions like (49).

(49) What did Daddy say?

Again, we must assume that the child has knowledge of the meanings of the main words.With this knowledge, however, and the knowledge that root wh-movement can only raisean interrogative pronoun as high as Spec-ΠP (since the root C is silent), a child mayunderstand that the wordwhatoccupies the Spec-ΠP position in (49). But since sheknows that subjects—non-wh-phrases—are otherwise attracted byΠ, she may thenconclude that some other category must be satisfying the EPP property of thesubject-attracting feature inΠ. And the only candidate for this other category is the worddid. If now the child is able to identify the relevant features of the auxiliary verb, [Tense]andφ and [Finite], it is a small step to look for a feature which the auxiliary verb has incommon with the subject of a sentence. The feature shared by subjects and T is [Tense],so our young learner may deduce directly thatΠ attracts subjects by virtue of a [Tense]feature with the EPP property.

Other results follow quickly. Since theMinimal Link Conditionis supplied by UG,and since the child now knows that the subject is attracted in place of T in normaldeclaratives, she may conclude that subjects are as close toΠ as T is at the point in thederivation whenΠ attracts a subject. Therefore, subjects must raise to Spec-TP, and theremust be an attractor feature in T which forces this to occur. The EPP feature of T thenfollows directly from the properties ofΠ. What is more, sinceΠ and C are the samecategory, it follows that C, too, must bear a [Tense] feature which must be checked byattractingΠ. Any departures from these deductions must involve assigning specialproperties to specific tokens of T or C, and must be motivated on the basis of new positiveevidence.

3.5 Summary

The main claim defended in this chapter is that Spec-ΠP is always an A-bar position. Thelicensing properties ofonly-phrase subjects show that declarative clauses subjects countas equivalent to phrases raised to an A-bar position, a fact which is most simply explainedif the subjects actually do raise in the same way. The reconstruction effects found inlocative inversion show that fronted locative PPs—which occupy Spec-ΠP—behave inthe same way as phrases which have undergone A-bar movement. And the fact that thespecifier ofΠP cannot have Case-features checked by matrix verbs, in French ECM and

Page 83: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 3. THE A-BAR PROPERTIES OF SUBJECTS 82

in Englishwager-class ECM, shows that the specifier ofΠP is never allowed to be anA-position.

To return to the question of terminology, the evidence presented here shows that theΠ head should not be equated with some of the other theoretical constructs with with itshares a family resemblance. AsΠ is necessarily an A-bar head, it is different from theT-type heads which appear in any ‘exploded IP’ theory of clause structure. It must also bedistinguished from Shonsky’s AgrC or Rizzi’s Fin heads, at least to the extent that thesetheoretical constructs belong in part or in whole to the A-syntax. Of course, these labelsthemselves are unimportant, and the entire discussion up to this point could be interpretedas an exploration of the properties of Fin, AgrC, Mood, or some other equivalent head.What matters is that the properties of this head are now becoming clearer.

Page 84: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

Chapter 4

That-trace effects and the TraceDeletion Constraint

4.1 The context ofthat-trace effects

The previous chapters have established the structure and the formal motivation for thenormal position of clausal subjects. It remains to be shown how these give rise to thethat-trace effect. This will be the goal of the present chapter.

The existing literature on thethat-trace effect is large and impressive, both in termsof the range of data examined, and the theoretical sophistication of the accountsproposed. Before tackling the problem directly, therefore, some time should be devotedto showing that the problems posed by this family of data have not already been solved.

The major proposals in this area share a family resemblance, at least the proposalswhich fall within the general framework which I am adopting, that ofprinciples-and-parameters syntax. The predominant approach within theGovernment-and-Binding theory involved various formulations of the concept ‘propergovernment’, together with theEmpty Category Principle(ECP) (Chomsky, 1981):

(1) Empty Category Principle(Chomsky, 1981)An empty category must be properly governed.

In this approach, the difference between the grammatical (2-a) , and theungrammatical (2-b) is that the subject trace in the embedded clause is properly governedonly in the first sentence.

(2) a. What did Peter claim [t had happened ]?

b. *What did Peter claim [CP that [ t had happened ]]?

83

Page 85: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 84

The presence ofthat in the embedded clause blocks the necessary relation of propergovernment between the trace and some proper governor in a higher position in thesentence, where different theories differ on the nature and position of the higher propergovernor.

There are various empirical problems which arise with ECP-based accounts of thethat-trace effect. Some of the more problematic of these arise in confronting therelationship between thethat-trace effect and movement of non-subjects. This is the casewith English locative inversion and with embedded v/2 topicalisations in Scandinavianand German (cf. section 4.3. To my mind, however, the most challenging empirical issuesinvolve the cross-linguistic variation in subject extraction constraints in the Scandinavianlanguages (cf. section (41), where subject extraction is sometimes less constrained thanextraction of non-subjects. But the empirical issues are less troubling than the moregeneral conceptual problems which the ECP commits us to.

Consider what is necessary in order to formulate a principle like the ECP. First, weneed the notion of an ‘empty category’ to distinguish positions which require propergovernment from positions which do not. But ‘empty category’ in this sense does notrefer to any natural class of categories. It cannot refer simply to categories which lackphonetic content, because many such categories must be excluded from the scope of theECP. Null pronominal subjects and PRO must not require proper government (unlessspecial provisions are invoked). Null subjects normally serve as specifiers in finiteclauses, which is where subject extraction is most clearly constrained. PRO appears as aspecifier for non-finite clauses, but PRO itself must be protected from a higher (‘proper’)governor in order not to have non-null Case features checked. So some categorieswithout phonetic content do not count as ‘empty’ in the appropriate sense.

Another way to define the set of ‘empty’ categories might rely on the idea that theseare traces left by movement of a category to a higher position. Under the copy theory ofmovement, such categories are ‘empty’ only in how they are interpreted at Spell-Out, butthere is still a concept of trace which might be exploited. Unfortunately, not all traces aretreated equally by the ECP. As shown in detail by askLasnik and Saito (1984),intermediate traces left by successive cyclic movement must be exempt from the ECP. Sothere is no natural class derived from the concept of movement, either. The best whichcan be done, apparently, is to stipulate that ‘empty categories’ are just those traces whichare produced by the first step in A-bar movement. It seems that ‘empty category’ canmean nothing more than the tail of an A-bar chain.1 And yet even this is too general,because some tails of A-bar chains seem to be exempt (depending on how ‘proper

1A further complication arises if we consider the case of resumptive pronouns, which are marginallyacceptable in English if they appear in a position where the ECP would normally be violated:

i ??Who did Peter ask whether she had been invited?

Page 86: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 85

government’ relations are identified). The trace left by local A-bar movement of thesubject seems not to be subject to the ECP:

(3) a. They want to know [CP what [TP t happened ]]?b. the house [CP (OP) that [TP t fell on Jack ]]

Unless this case is one in which the ECP is actually satisfied for the subject trace, thennoworking definition of an empty category appears to be in sight.

Within the model adopted here, it is not even clear how reference to an ‘empty’category might be made. In Minimalist syntax, grammatical constraints are eitherlegibilitiy conditions of the LF or PF interface, or they are constraints on the derivationproper. The phonetic content of a syntactic object can be relevant only at the PF interfaceor within the derivation, where phonetic content is being operated on. But there appearsto be nothing problematic in the PF interpretation of (3-b); the wordthat can certainly bepronounced next to the following finite verb. And wh-movement of the subject does notappear to operate on the phonetic content of the subject in any way which distinguishes itfrom wh-movement of adjuncts or other arguments. So the phonetic content of thesubject trace in (3) is simply irrelevant to the contrast between (3-a) and (3-b).

The notion of ‘proper government’ is, if anything, even more difficult to pin down.This relationship is something which must involve subjects of complement clauses, andwhich must be disrupted by the presence of a somewhat heterogenous group ofconstituents of CP, including thethat andif complementisers and any wh-phrase inSpec-CP. Other complementisers, notably Frenchqui, do not disrupt the propergovernment relation. Most research in this area has concluded that there are two distinctrelationships which count as proper government: government by a lexical head, or“lexical government” and government by a coreferential antecedent within a chain, or“antecedent government”. Despite the terminology, it is difficult to see a commonconceptual core to the two notions. Granting, however, that both notions might bepertinent, we must ask if one or both allows us to make sense of the conditions underwhich proper government is disrupted by the syntactic context.

And here the literature largely fails us. Consider the contrast between (4-a) and(4-b), where the ECP is presumably satisfied only in the first case.

(4) a. What did they say that Peter thinks [t fell on Jack ]?b. *What did they say that Peter thinks [CP t that [TP t fell on Jack ]]?

In (4-a), either the matrix verbthinksproperly governs the subject trace, or there is a local

If these are interpreted as the tail of an A-bar chain, bound by the operator, then they are almost equivalentto a trace left in this position. We might then expect resumptive pronouns to be subject to the ECP, as well.

Page 87: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 86

antecedent in Spec-CP of the same clause which properly governs the trace (dependingon the structure and the theory adopted). In (4-b), both possibilities are blocked by thepresence ofthat as the head of CP. And here is where the problem lies. We might acceptthat thethat head could block a government relationship involving a higher head, as a‘minimality’ (Rizzi, 1990a) orMinimal Link Conditioneffect. But there is no reason whythe presence of a specific head should disrupt any relationship between a trace and itsantecedent. Everywhere else in grammar, it is locality and c-command which counts inchain formation, and intervening lexical content is irrelevant. So ‘antecedentgovernment’ should not be blocked, and we are left with no explanation for theungrammaticality of (4-b), with or without the ECP.

The problems become even worse if the conclusions reached already in the presentstudy are incorporated. Suppose that the structures of the (4) sentences are betteranalysed as in (5).

(5) a. What did they say that Peter thinks [ΠP t Π [ t fell on Jack ]]?b. *What did they say that Peter thinks [CP t that [ΠP t Π [TP t fell on Jack ]]]?

If antecedent-government of the subject trace were sufficient to satisfy the EPP, then both(5-a) and (5-b) should be grammatical, since the Spec-ΠP trace is in the samerelationship to the Spec-TP trace in both sentences. Given these structures, it wouldnecessary to suppose that the ‘empty category’ which is subject to the ECP is the trace inSpec-ΠP. But this trace cannot be the tail of an A-bar chain; it occupies an A-barposition, and should only have the status of an intermediate trace in successive cyclicA-bar movement. So there is no trace in (5-b) which could be violate the ECP, and again,the contrast between (5-a) and (5-b) remains mysterious.

Even supposing that the notions of ‘empty category’ and ‘proper government’ couldbe satisfactorily defined, the constraint imposed by the ECP still remains almostincoherent, especially when minimalist premises are adopted. Supposing that theinterfaces (LF and PF) are the natural sites for grammatical constraints to be evaluated,since no natural notion of D-structure or S-structure is available, the ECP must beapplicable at either PF or LF or both. PF would appear to be the natural interface to find aconstraint which applies to empty categories, but it is clear that the PF interface cannot beconcerned with whether or not a particular type of trace—the tail of an A-barchain—enters into some special government relation. The PF interface is concerned withwhether its input is appropriate to be interpreted as phonetic content, but there is noreason why extraction of a subject adjacent tothat should be phonetically problematic.

Most theories in this family in fact adopt the remaining possibility: that the ECPapplies at the LF interface. But this makes sense only if there is some reason why asuccessful interpretion of a phrase marker to Logical-Conceptual representations would

Page 88: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 87

be impossible unless the ECP is first satisfied. Interface constraints should be primarilyconcerned with ‘legibility’, i.e. whether a syntactic object contains content which can bemapped to a semantic representation. But the tail of an A-bar chain should be legible inthis sense as long as it is bound by a c-commanding operator. Proper government appearsto add nothing to legibility.

We are left with the stipulation (6):

(6) ECP decomposedAt the LF interface, interpretation of the tail of an A-bar chain is illegitimateunless there is a lexical governor or an antecedent-governor .

There is little explanatory value to a statement like this. The facts themselves are of acomparable degree of complexity, so the ECP-based theories have do little more thanrestate the question, in terms which cannot withstand a minimalist critique.

More recent approaches to the problem fare little better. The P&T theory has alreadybeen examined, and problems identified. Boškovic and Lasnik (2003) treat thethat-traceeffect essentially as a subjacency violation, which is resolved in successful subjectextraction cases by using a special complementiser with an EPP feature to provide anescape hatch. The solution, which is actually secondary to their main proposals, isessentially stipulatory.

I take it that the problems posed by thethat-trace effect remain open.

4.2 Locality constraints on successive cyclic movement

Thethat-trace effect arises most evidently in structures in which successive cyclic A-barmovement affects the subject of a clause. In order to appreciate the nature of thephenomenon, it will help to first recapitulate what factors constrain successive cyclicmovement in general in the theory of clause structure which I now assume.

Successive cyclic A-bar movement takes place when a wh-phrase, topic, or focussedphrase raises from a position in an embedded clause to a CP projection in a higher clause.Successive cyclic movement ensures that the derivation includes no operations in whichtoo large a portion of the phrase marker is skipped over by a single movement. In theapproach taken here, successive cyclic movement satisfies the Phase Impenetrabilitycondition, by raising a phrase to the periphery of its phase before the phase is mergedwith additional lexical material. If both CP and transitive vP count as phases, thenmovement to both Spec-CP and Spec-vP will play a part in long wh-movement: (7).

(7) How fast did Claire [vP t [ .v′ say [CP t that Pam had been [vP t driving t ]]]]

Page 89: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 88

In the model presented here, where two stacked CP projections—CP andΠP—occurin a single embedded clause, successive cyclic movement is constrained in a second way.SinceΠP counts as a phase, it cannot be ignored in raising a phrase to Spec-CP unless thePhase Impenetrability condition is abbrogated under the PTE by incorporation ofΠ by C.

Consider the derivation of (8), where the potentialPhase Impenetrability Constraintproblem arises before the embedded clause is merged with a matrix verb: (9).

(8) Who had Tony said that Pam hired?.

(9) [CP that [ΠP PamΠ [TP t [ vP who [v′ hiredt ]]]]]

The wh-phrasewhooriginates as the direct object in the complex CP in (9). It canonly raise to its final position in the matrix clause if it first raises via Spec-vP to thespecifier position of the embedded CP, becoming accessible to operations external to CP.But ΠP is also a phase—being a variety of CP—so thatwhocannot raise out ofΠP toreach the Spec-CP position. Andwhocannot use the Spec-ΠP position as an escape hatchfrom the inner phase, because that position is already occupied by the clausal subjectPam. But that in (9) incorporatesΠ, and by doing so becomes able to look past theΠPphase boundary into TP. Ifthat takes on an attractor feature, a defective [wh] feature, thenit can attractwhofrom the edge of vP up to Spec-CP, and the derivation may then proceedas usual.

4.3 A featural constraint on successive cyclic movement

Constrained landing sites in successive cyclic movement

As discussed in chapter 2, A-bar movement in English root clauses can be driven by thepresence inΠ of a non-[wh] feature—the [MD] feature—which is used primarily foraltruistic purposes. The [MD] feature is added toΠ to permitΠ to attract a wh-phrase orfocus phrase to raise to the edge ofΠP so that it may be accessible to higher heads. Andthe same procedure is used more generally in other verb-second constructions throughoutGermanic, whereΠ may take on additional [MD] features to attract a wider array oftopics, focii, etc. to its edge.

As mentioned in chapter 2, in German, Yiddish, and the Belfast dialect of English,the altruistic use of [MD] or [MD] (in the former languages) takes place in a mannerwhich allows successive cyclic movement to take place from inside an embeddedverb-second complements. I will illustrate the general pattern with German data, wherethe facts are the most easily established. Successive cyclic long topicalisation ispermitted in (10).

Page 90: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 89

(10) (12) In zweiin

wochenitwo

glaubtweeks

Anna,believes

[CP

Annati hat Max

hasgesagt,Max

[CP

saidti werde

siewill

tishe

kommen. ]]come

For some speakers, long topicalisation may also occur with no subject-aux inversion in acomplement clauses out of which the topic raises.

(11) Inin

zweitwo

wocheniweeks

glaubtbelieves

Anna,Anna

[CP ti daßthat

MaxMax

gesagtsaid

hat,has

[CP ti daßthat

sieshe

ti

kommencome

werdewill

] ]

For others, (11) is ungrammatical, which we may attribute to a lack ofΠ-to-C movementin their dialects.

It is never possible, though, to have long topicalisation in which V-to-C is triggeredin some but not all intermediate clauses (Staudacher, 1990; Haider, 1993; Müller andSternefeld, 1993).

(12) *Ankei

Ankesagtesaid

sie,she

daßthat

erhe

glaube,believes

ti werdewill

ihmhim

seinehis

Arbeitwork

hierhere

bezahlen.pay

Given the acceptability (for some) of (11), it is clear that thePhase ImpenetrabilityConstraintis not what blocks (12). If the non-verb-second complement clause serves as abarrier to A-bar movement in (12), both non-verb-second complement clauses should bebarriers in (11). Since this conclusion is evidently false, we must suppose that theproblem in (12) has to do with something else.

The descriptive generalisation can be identified immediately. The initial phrase in averb-second clause occupies Spec-ΠP. The position of an intermediate trace infeature-driven successive cyclic movement is Spec-CP. What we see in the contrastbetween (12) and (12) falls under the following description:

(13) A phrase cannot move to Spec-ΠP and then move at a subsequent stage in thederivation to Spec-CP.

The (13) statement is a far cry from a principled explanation. To make it into one, weneed to first isolate the relevant properties of the two positions, and then use thisinformation to derive the constraint on movement.

So what do we know about the difference between these two positions. Onedifference is structural:ΠP is located immediately above TP; CP is situated higher in a

Page 91: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 90

clause. But there is no obvious reason why this should matter in the case of long A-barmovement. Both Spec-CP and Spec-ΠP are legitimate targets for A-bar movement inother contexts, as long as independant constraints on locality and economy of derivationsare respected.

Another difference between Spec-ΠP and Spec-CP lies in the types of featureswhich are checked in attracting specifiers to these positions. As established in chapter 2,Π checks its [Tense] feature on subjects or on T, andΠ checks an [MDF] feature onmonotone decreasing focii in auxiliary inversion contexts. In both cases,Π attracts itsspecifier with a feature which is relatively independant of that used in conventionalwh-movement. C, on the other heand, attracts its specifier with either a wh-feature or a‘defective’ wh-feature, depending on whether C is the head of an embedded question, orsimply a bystander in successive cyclic wh-movement.

Given this distinction in checking properties of C andΠ, the descriptivegeneralization in (13) may be reformulated as (14).

(14) A phrase which undergoes A-bar movement to check [Tense], [MD] or [MD]cannot subsequently undergo A-bar movement to check [wh] (defective or full).

Thus, in (19) (repeated here) wherewannis first attracted to check [MD] onΠ in thelowest complement clause, it cannot be attracted by C in the next clause up to check the[wh] feature.

(19) *AnnaAnna

fragteasked

mich,me

wanniwhen

ihnenyou

FritzFritz

gesagtsaid

hat,has

ti werdewill

sieshe

ti kommen.come

And in (12) (repeated here), whereAnkeraises to Spec-ΠP in the second complementclause to check [Tense], it may not subsequently be attracted bydaßto check defective[wh] features, even though the target position for movement into the root clause is aposition where [Topic] features will be checked, rather than a [wh] feature.

(12) *Ankei

Ankesagtesaid

sie,she

daßthat

erhe

glaube,believes

ti werdewill

ihmhim

seinehis

Arbeitwork

hierhere

bezahlen.pay

The constraint in (14) implies an assymetry in features which needs to beunderstood. While movement may take raise a phrase from a defective [wh] checkingposition to an [MD] checking position, the reverse is not true. (Movement will never takeplace out of a non-defective [wh] position, because a wh-phrase in such a position isalready in the place where it will be interpreted.) The possibilities are encapsulated in(15)– (17-c):

Page 92: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 91

(15) Possible A-bar movements

a. [MD] . . . [MD]b. [MD] . . . [MD]c. [wh](defective) . . . [wh](defective)

(16) Possible A-bar movements

a. [wh] . . . [wh](defective)b. [MD] . . . [wh](defective)c. [MD] . . . [wh](defective)

(17) Impossible A-bar movements

a. [wh](defective or full) . . . [MD]b. [wh](defective or full) . . . [Tense]c. [wh](defective or full) . . . [MD]

Movement is permissible from one position to another position of the same type.Movement is possible from a defective [wh] position to any other position. Andmovement is impossible otherwise.

Evidently, the defective [wh] feature is special in some way. Although both full [wh]and defective [wh] are checked in Spec-CP, they are otherwise not alike. In fact, all thatwe know about the defective [wh] feature is that it has a certain role to play in successivecyclic wh-movement which has no immediate impact on semantic interpretation. There isno particular reason to call it a [wh] feature at all.

Suppose we were to adopt a different nomenclature, and call the special ‘defective[wh]’ feature something more neutral, such as [NO] (for ‘nonspecific operator’). Thepattern would be the same. Movement from an [NO] position to another A-bar positionwould be allowed, as would movement between two matching positions. Other movementwould be impossible. But now it becomes easier to speculate on what the different mightbe between [NO] and other A-bar features. The difference is not that [OD] alone permitssuccessive cyclic movement, because [MD] and [MD] also permit this. Instead, thedifference is that [NO] is somehow sufficiently similar to all the other operator features toallow it to interact with them, in a way that [wh] or [MD] features themselves cannot.

This suggests that there might be a difference in the degree of featural contentbetween [NO] and the other A-bar features. Recent work on the internal structure ofmorphological features (Noyer, 1992; Harley and Ritter, 1999) suggest how thisdifference might be encoded. Suppose we treat [NO] as a less specified type of featurecompared to the [wh], [MD], or [MD] features. ([Tense] will be brought back into thediscussion in due course.) Then [NO] may actually refer to a feature structure in whichthe minimal ‘operator’ content is found. The remaining A-bar features, which appear to

Page 93: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 92

be more richly specified, may have extra content, together with the minimal operatorcontent of [NO]. This difference can be characterised structurally as in (18), where Op isthe minimal content which both [NO] and other A-bar feature matrices have in common.

(18) [NO] [wh] [MD] [ MD]Op Op Op Op

| | |wh MD MD

Given this preliminary picture of the content of A-bar features, it becomes possible toreformulate the general constraint (14) as (19):

(19) A phrase which undergoes A-bar movement to check a feature F cannotsubsequently undergo A-bar movement to check a feature G unless G is asuperset of F.

(G will be a superset of F if G is identical to F, or if G contains all the featural contentthat F contains.)

Movement from Spec-CP will generally be allowed under (19), while movementfrom Spec-Π to Spec-CP will generally be impossible.

Trace deletion in composite chains

Of course, the constraint in (14) is hardly more explanatory than the descriptivegeneralization in(13). But by reducing the issue to a relationship between attractorfeatures and A-bar movement, some of the remaining issues become easier to identify.

Consider, for example, the reference to A-bar movement in the formulation (14).Obviously, we would have a simpler statement—if not yet a theory—if the reference tomovement types were eliminated. But if the reference to movement types whereeliminated, then we would expect embedded questions like (20) to be impossible.

(20) I wonder [CP who Peter thinks [TPt will [vP t win the prize ]]].

Subjects are normally attracted to Spec-TP to check the [Tense] feature, so thatmovement of the subject wh-phrasewhoto check [wh] features on Spec-CP would beblocked by the simpler formulation.

If the distinction between A and A-bar movement is significant here, then anunderstanding of (14) requires that we understand what this distinction involves.

To a large extent, the A/A-bar distinction involves a difference in the nature of thechains formed by the different movement types. A movement produces A-chains:

Page 94: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 93

linguistic objects in which the head of the chain bears the semantic content for the wholechain. The head of an A-chain is therefore implicated in binding relations, both as thec-commanding antecedent for anaphoric relations, and as the locally bound DP in thesame type of A-binding. The head of an A-bar chain, in contrast, may not bind an anaphordirectly, and it is not normally bound itself (although it can contain bound anaphors).2

Now it is certainly not the case that all movement is constrained by this type ofuniformity constraint. A-movement is subject to different locality conditions than A-barmovement is, the former originating with the Case/φ feature checking and the latter(usually) with operator features. But A-bar movement can originate with the head of anA-chain, as in (21), for example.

(21) Who had Diane imagined [CP t Peter had thought [TP t′ to be likelyt′′ to [vP t′′′

befriend Callum ]]]?

The difference seems to involve the types of chains which are formed at each step. Thechains which are subject to the uniformity constraint (22) are intermediate steps insuccessive cyclic A-bar movement. A-bar chains are assymetric objects at the LFinterface, with a head typically containing a quantificational component and the tail,some restriction on the range of the variable bound by the head. As Chomsky 1995observes, such chains are necessarily binary, as there is no interpretation available for athird member of the chain. Yet the process of forming an A-bar chain can producenumerous intermediate positions, whenever successive cyclic A-bar movement occurs.The solution to this paradox is that intermediate traces may be deleted when non-binary,n-ary chains are formed. In order to put this solution into practice, we must be able todistinguish those traces which may be deleted from those traces which must not be.

In (21), the result should be thatt is deleted, whilet′, t′′ andt′′′ are not. This wouldproduce the binary A-bar chain (who,t′) and the ternary A-chain (t′, t′′, t′′′), each ofwhich can be interpreted at LF. Assuming that deletion must be recoverable, we mayformulate the constraint on chain formation as (22):

(22) Trace Deletion Constraint(TDC)If phraseα is attracted by feature F from positionβ, then the trace ofα may bedeleted iffβ was created to check feature G, and F is a superset of G.

The effect of (22) is that the uniformity constraint (13) is derived from a moregeneral recoverability constraint on trace deletion. No specific reference to chainuniformity is necessary.

The role ascribed to v in successive cyclic movement must be consistent with the

2See especially (Chomsky and Lasnik, 1991) for discussion.

Page 95: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 94

analysis. As vP is presumably a strong phase, at least sometimes, it must be possible for vto take on the right features to attract various types of operators to the edge of vP. Sincethe trace created by movement to Spec-vP will generally be deleted (under successivecyclic movement), it must be the case that the TDC can be satisfied by such movement. Iconclude that v may take on [NO], [MD], and [MD] features freely in order to attract aphrase to Spec-vP, and that the choice of the right attractor feature on v is sufficient topermit successive cyclic movement to proceed along the usual lines. This conclusiondoes not seem problematic. Indeed, something of the sort must be said in any account ofthe constraint on long topicalisation in German.

We can now derive the description (13) as a corrolary covering the general case.Consider once again the derivation of long topicalisation in (12):

(12) Inin

zweitwo

wocheniweeks

glaubt-Πbelieves

AnnaAnna

[ vP i . . . ,has

[ΠP

Maxti hat Maxj [ vP

saidti tj gesagt,

will[ΠP

sheti werde-Π sie

cometi kommen. ]]]

The topicin zwei wochenoriginates in VP in the bottom complement clause, and it mustraise through each lower Spec-vP and Spec-ΠP in the clause to reach its final position atthe left edge. To attract it to the left edge of vP in the lowest clause, v can take on any ofthe A-bar features; if v takes on a [MD] feature in this case, then the derivation maysucceed. So v attracts the topic phrase to the edge with an [MD] feature, and the trace inthe originalθ position is not deleted. ThenΠ in the lowest clause must attract the topicwith an appropriate feature. IfΠ takes on [MD] to attractin zwei wochen, then the tracein Spec-vP will be deleted under theTrace Deletion Constraint. And the derivation maythen continue in the same way, with a series of v andΠ heads taking on the [MD] featureand deleting the trace in the lower specifier postion from which the phrase is attracted. Atthe topΠP, the topic is in a position where its [Topic] feature may be checked by a silentTop head, so that the derivation ultimately converges at LF.

4.4 Thethat-trace effect derived

Consider now thethat-trace effect in (23).

(23) *Which car had Marian decided that would suit her?

It has been established that the normal subject position is Spec-ΠP, so the wh-phrasewhich carmust raise from Spec-ΠP in the complement clause. It must pass the embeddedCP phase boundary, and this movement will satisfy thePhase Impenetrability Constraint

Page 96: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 95

only if an escape hatch is provided by C attracted the wh-phrase to its edge. Minimally,then, the structure of (23) will be (24).

(24) [ΠP which carΠ [TP Marian had decided [CP t that [ΠP t Π [TP t would suit her]]]]]

In the complement clause, it is the [Tense] feature ofΠ which attracts the subject toSpec-Π and the [NO] feature of C which attracts it to Spec-C.

The wh-phrasewhich carbelongs to two chains at LF: the A-chain which bears theembedded clause subjectθ-role, and the operator-variable chain which is formed bywh-movement. The latter must be binary, which entails deletion of every intermediateA-bar trace. In this case, the Spec-CP trace and the Spec-ΠP trace must both be deleted.

Given the trace deletion algorithm (22), a trace is deleted as soon as it is left behindby movement to a higher position, if it is deleted at all. Then the trace in Spec-ΠP mustbe deleted when the wh-phrase moves to Spec-CP of the complement clause. This willonly take place under (22) if the attractor feature in C is a superset of the attractor featurein Π. But the [wh] feature matrix does not contain the [Tense] feature at all, so it iscertainly not a superset of [Tense]. Therefore, the trace in Spec-ΠP cannot be deleted,and the derivation will therefore crash at LF.

This, I claim, is the essence of thethat-trace effect. Given the properties of the A-barposition which subjects occupy before wh-movement occurs, the recoverabilitycondition (22) prevents an interpretable set of chains from being formed for the subjectwhen a CP-ΠP structure is present.

Other contexts in whichthat-trace effects arise in English include indirect questions,which are necessarily CP complements.

(25) a. *Who did they ask if had prepared the dinner?b. **Which printer do you know when is going on vacation?

In both cases, the embedded clause is a CP, so that extraction of the subject directly fromΠP violates thePhase Impenetrability Constraint. In (25-a), as in the simplerthat-tracecases, movement from Spec-Π to Spec-CP will run afoul of theTrace DeletionConstraint. In (25-b), where a greater degree of ungrammaticality is detectable, there isnot even an option of satisfying thePhase Impenetrability Constraintby using a Spec-CPescape hatch.3

3The contrast between (25-a) and (25-b) seems to show that aPhase Impenetrability Constraintviolation involving movement from an A-bar position is worse than an LF crash by virtue of theTraceDeletion Constraint. The more severe violation could reflect an additional constraint on trace deletion,which requires locality to be respected as well as feature superset relations.

Page 97: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 96

The same account covers thethat-trace effect in locative inversion.

(26) a. *Which painting has Tom decided that should be hung on this wall before theweekend?

b. *On which wall has Tom decided that should be hung this painting?

In (26-a), the DPwhich paintingis extracted from Spec-ΠP via a Spec-CP position where[NP] is checked. TheTrace Deletion Constrainttherefore blocks trace deletion fromSpec-ΠP. In (26-b), it is the PPon which wallwhich is extracted from Spec-ΠP viaSpec-CP, and which leaves an undeleted trace in Spec-ΠP. In both cases, the resultingchain structure includes an uninterpretable A-bar chain because of the non-deleted A-bartrace left behind.

How now does this approach compare with earlier accounts of thethat-trace effect,and with Minimalist premises? An explanatory account of constraints on subjectextraction should not have to make special reference to the notion of a subject, to theSpec-TP category, or (especially) to the notion of an ‘empty category’. Constraintsshould reflect either properties of the interfaces or properties of the derivation itself.

The account I propose here makes reference to the notion of an interpretable A-barchain, which is certainly a legitimate notion, and one which can be identified at the LFinterface. An interpretable A-bar chain must of necessity be binary, so it requires nospecial stipulation to require that the derivation must be able to produce binary A-barchains by deleting intermediate traces. And theTrace Deletion Constraintis nothingmore than a specification of the derivational procedure for deleting intermediate traces, aprocedure which must be specified in some form in any case.

In short, in this account of thethat-trace effect, there is nothing to stipulate. Thefacts follow from general properties of the LF interface, together with a clearer picture ofwhat goes on in the formation of A-bar chains under successive cyclic movement.

4.5 Avoiding the that-trace effect

French qui

The exceptions must prove the rule, so the structures in which subject extraction canoccur must now be explained. Following a long tradition, let us look to the Frenchque-quialternation for insight into one way thatthat-trace effects can be avoided. As iswell known, thequi complementiser appears whenever subjects are successfullyextracted from a finite complement clause (Kayne, 1972).

Page 98: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 97

(27) Quiwho

imaginent-ilsthink-they

qui/*quethat

t ahas

ritlaughed

?

Qui may appear only in case the subject is extracted, as can be seen from (28).

(28) *Ilsthey

imaginentthink

quithat

MarieMarie

ahas

rit.laughed

The challenge posed by these data again has two sides to it. On the one hand, wemust say why subject extraction is allowed at all in sentences like (27), given the accountof that-trace effects developed above. In addition, we would prefer to be able to saysomething about the contexts in whichqui may appear.

In chapter 2, I suggested that the presence of subjects in declarative Spec-CP isblocked because they interfered with the interpretation of the complementiser in variousways. Overt C must be at the left edge of the clause to be understood as a force marker;null C must be adjacent to a higher predicate to undergo Morphological Merger. Left tothe side at that point was the question why CP must be present in complement clauses atall. What would be wrong with a finite complement clause which consists ofΠP alone?(Recall from chapter 3 that infinitival clauses consist ofΠP in some contexts.) The mostobvious answer is that finite clauses must include a force marker, which is normally therole of C.

But what ifΠ were to function as a force marker in place of the usualcomplementiser? ThenΠP could function as a full embedded clause if it were not for thepresence of the subject in Spec-ΠP. In a Germanic language, given the [Tense] feature ofΠ, this will be possible only if something removes the Spec-ΠP phrase, to leaveΠ at theleft edge of its clause. As has been shown in some detail, this is possible in Germanembedded verb-second complements. In most other Germanic languages, however, Cmust still appear in embedded clauses. Evidently, the use ofΠ as a force marker is amarked option, which must be acquired on the basis of evidence available to the child. Inthe absence of such evidence,Π is always taken to lack a force marker interpretation.

French behaves much like a Germanic language in the syntax of root questions(Den Besten, 1983). In complex inversion structures like (29), it is fairly clear that anon-subject raises to Spec-ΠP in the same way as in English, and that T raises toΠ tocheck its [Tense] feature.

(29) Quandwhen

sont-ilsare-they

partis?left

‘When did they leave?’

FrenchΠ must therefore have a [Tense] feature with the EPP property, like English.

Page 99: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 98

Now consider the case of Frenchqui, which appears in questions only when thesubject is extracted. Since subject extraction is impossible if the subject raises fromSpec-ΠP directly to Spec-CP (because of theTrace Deletion Constraint), this type ofmovement must not occur in (27). Instead,ΠP must be a bare complement to the matrixverb inqui-extraction complements. The structure of (27) will then be (30). (Deletedtraces are left in place to show the path of movement.)

(30) Quii imaginent-ils [VP t [ΠP ti qui/*que [TP ti a rit ]]

In French, then,Π may sometimes have phonetic content and appear asqui. With thesubject absent from Spec-ΠP,qui is at the left edge of its clause, and can be interpreted asa force marker.

Godard (1985) (cited by Rizzi (1990a)) observes that (for some speakers)qui canappear only in the complement to epistemic verbs and verbs of saying. This environmentoverlaps with that in which embedded v/2 clauses are allowed in German and themainland Scandinavian languages. Presumably, these contexts are those in which asemantic property of the matrix verbs allows the complement clause to either do withoutan argumental complementiser (German), or to make do with a less effective one(Scandinavian). It is natural to suppose that the same semantic property allows Frenchverbs to take a bare primary CP complement, in whichquedoes not appear because thematrix verb itself selects aΠP complement.

The fact thatqui only appears in subject-extraction contexts now makes sense.Unless the subject is extracted,qui cannot be interpreted at LF. Thus both (31-a) and(31-b) will be ungrammatical.

(31) a. *Ilsthey

imaginentthink

quethat

elleshe

quiqui

ahas

rit.laughed

b. *Ilsthey

imaginentthink

elleshe

quiqui

ahas

rit.laughed

Extraction of non-subjects will be irrelevant, of course, since non-subjects will neverhave any relationship withΠ in an embedded clause anyway.

More must be said about the fact that subject extraction is possible at all. SinceFrenchΠ has a [Tense] feature which normally attracts the subject, theTrace DeletionConstraintshould apply in French just as it does in English. Wh-movement from Spec-Πwill give rise to a proper binary A-bar chain only if the trace is deleted, which will in turnbe allowed only if the next attractor head has a superset of the [Tense] feature. But thenext attractor head should be v, which never checks [Tense] features. And even if v couldcheck [Tense], the next higher head would beΠ, which in (30) must check [MD] and not

Page 100: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 99

[Tense]. So one way or another, the derivation of (30) should fail.Evidently, there must be a loophole. We must attribute toqui some additional special

property which distinguishes it from other instances ofΠ. SinceΠ functions as a forcemarker, the temptation is to suppose that it has its special properties because of this, andthe special property is one which adheres to other force markers too.

While regular declarative complementisers are not revealing, the ‘special’complementisers found in most Germanic languages in embedded verb-second clausesare. In chapter 2 (p. 57), this class of complementisers was shown to have distinct, relatedproperties. First, they can appear only as complements to bridge verbs, for reasons whichremain mysterious. Secondly, they check a [Tense] feature which lacks the EPP property,so that [Tense] checking does not force movement. Frenchqui patterns with the specialverb-second complementisers in the first respect, and it is reasonable that it should do soin the second respect, too.

In that case, the derivation of (30) can proceed as follows. When TP is complete,quiMerges with TP.Qui checks its [Tense] feature on either the subject or T, but withouttriggering movement, sinceΠ’s [Tense] lacks the EPP property. In order to ensure thatthe wh-phrase in TP may ultimately be checked by a [wh] C,Π then takes on a [NO]feature, just as C does in other long wh-movement contexts. The wh-phrasequi is thenattracted to Spec-ΠP by [NO]. When the next stage of successive cyclic movement is athand, v will take on a [NO] feature to attract the wh-phrase from Spec-ΠP. Since [NO] isa superset of [NO], the trace in Spec-ΠP will be deleted under theTrace DeletionConstraintby the second A-bar movement operation, and the derivation can continue asusual from then on.

English subject extraction

The crucial property of Frenchqui which make it compatible with subject extraction isthatqui bears the [Tense] feature without an accompanying EPP property. As aconsequence, the subject is not raised to Spec-ΠP for the usual reasons. This makes itpossible for subject extraction to avoid the usual side-effect imposed by theTraceDeletion Constraint. In English, subject extraction in declaratives is usually possible onlywhen there is nothat complementiser. Is it reasonable to extend the analysis of Frenchqui to this situation?

(32) Which car does Bill insistt was parked in his space?

Such an extension is a trivial technical affair. In (32), for example, it is possible tomaintain that the complement clause is a bareΠP, rather than a full CP. The head of thecomplement clause would then be a null token ofΠ, which is unproblematic sinceΠ

Page 101: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 100

appears normally to lack phonetic content in English. Likequi, the particularΠ in (32)does not raise to C, but remainsin situ. Like Frenchqui, Englishin situΠ must lack anEPP property for its [Tense] feature, and likequi, it must be interpreted as a force markerby virtue of the lack of a specifier which appears at the left edge of the clause.

Although relevant data is hard to find, there is some slight evidence that this is morethan a technical analogy. Consider the data in (33).

(33) a. Penny feels (*sincerely) Paul should be given another chance.b. Penny feels sincerely that Paul should be given another chance.c. How many chances does Penny feel (*sincerely) Paul should be given?

As observed in Stowell (1981), in the normal case, a null complementiser must beadjacent to a matrix verb. This follows in the present model from Boškovic and Lasnik’s(2003) idea that null declarative C must undergo Morphological Merger with the verb toits left. If the verb is not adjacent to C, then no Morphological Merger can take place.Notice that A-bar movement out of the complement clause has no effect on the adjacencyeffect with null C.

But when the subject is extracted, the adjacency effect disappears, as in (34).

(34) Who does Penny feel sincerely should be given another chance?

We may conclude that there is a significant structural difference between subjectextraction clauses like (34) and clauses with the normal null declarative complementiser.Suppose now that English subject extraction patterns with French subejct extraction.Then the structure of (34) will be (35):

(35) Who does Penny feel sincerely [ΠP Π [TP t should be given another chance ]]

(The subject trace in Spec-ΠP is deleted under theTrace Deletion Constraintwhen itraises to Spec-vP of the matrix clause.) In (35), the nullΠ is not required to undergoMorphological Merger, because that licensing condition holds only of the regular nulldeclarative C. EnglishΠ can stand on its own. No problem arises therefore when thematrix verb andΠP are not adjacent.

Boškovic and Lasnik (2003) observe as well that Right Node Raising examples like(36) are impossible with null complementisers.

(36) a. They believed, and Mary claimed, *(that) John would murder Peter.b. Who did they believe, and Mary claim, *(that) John would murder?

This pattern again reflects the need of a null complementiser to undergo MorphologicalMerger with a matrix head, which is impossible if the null complementiser is deleted in

Page 102: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 101

the first conjunct.With subject extraction in the complement clause, however, no overt complementiser

is required (or allowed) in parallel Right Node Raising structures.

(37) Who did they believe, and Mary claim, would murder Peter?

Again, the ability ofΠP to appear as a bare complement clause under the right conditionsis what explains the acceptability of (37). Bothbelieveandclaim in (37) take aΠPcomplement, whereΠ is a legitimate force marker since it is at the left edge of its clause.And the entire structure is legitimate becauseΠ is not subject to special licensingconditions which are disrupted by Right Node Raising, unlike null declarative C.

English and French differ in only one, quite superficial respect here. In bothlanguages,Π has no phonetic content when it heads the complement to an argumental C.In French, it takes the formqui when it appears in its base position. In (standard) English,Π has no phonetic content even when it does not raise to incorporate into C.

For those English dialects which do permit subject extraction with an overtthatcomplementiser (Sobin, 1987), we may now simply assume thatΠ may be realised asthat when it is not the complement to a higher C. The dialectal variation then reduces to adifference in the phonetic form of a single functional head (Branigan, 1996b).

Yiddish subject extraction strategies

The purest illustration of the approach to successful subject extraction taken here comesfrom Yiddish. The contrasts in (38) show whatthat-trace effects look like in thislanguage. (The data comes from Diesing (1990).)

(38) a. *Verwho

hothas

erhe

moyrefear

[CP azthat

[ΠP vetwill

kumencome

]] ?

b. Verwho

hothas

erhe

moyrefear

vetwill

[ΠP

comekumen ]?

c. ?Verwho

hothas

erhe

moyrefear

[CP azthat

[ΠP esit

vetwill

kumencome

]] ?

In Yiddish, as in English and French, a bareΠP may appear as a complement toappropriate predicates, in free alternation with a full CP-ΠP structure. Movement fromSpec-ΠP to Spec-C is the only way for the subject to escape the effects of thePhaseImpenetrability Constraint. But theTrace Deletion Constraintensures that the trace inSpec-ΠP will not be deleted under such movement. Example (38-a) is thereforeimpossible. In example (38-b), the subject is attracted by a null force markingΠ, which

Page 103: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 102

presumably may attract it specifier with a [NO] or [MD] feature, rather than with [Tense].The force markingΠ must appear at the left edge of its clause, so it cannot appear after C,and it cannot appear unless the subject trace will be deleted. In this case, the subject maythen raise to a higher A-bar position without uninterpretable results. In example (38-c),the whole problem is avoided by the use of an expletive pronounesto satisfy the EPPfeature ofΠ. Movement of the subject to Spec-CP is then an available option to escapethe CP phase, and the subject trace left behind in Spec-TP does not need to be deleted.

Yiddish is subject to thethat-trace effect in ways that English is not, too. Longextraction of topicalised phrases gives rise to the same type of results as long extractionof subjects.

(39) Voswhat

hothas

erhe

nitnot

gevoltwanted

(*az) [ΠP t zolnshould

[TP mirwe

leyenenread

t ] ]

(40) Venwhen

hostuhave-you

gezogtsaid

(*az)that

[ΠP t hothas

[TP MaxMax

geleyentread

dosthe

bukhbook

t ?

The same mechanism lies behind these phenomena. Topicalisation of the object intoSpec-ΠP in (40) is possible because the head ofΠP attracts its specifier with [MD]. ButthePhase Impenetrability Constraintensures that the object will escape the complementclause only by raising to Spec-CP. But this latter movement is driven only by the [NO]feature added to C, and as [NO] is not a superset of [MD], the trace in Spec-ΠP is notdeleted under theTrace Deletion Constraint. The derivation must therefore crash at LF.

(41) a. Vos hot er nit gevolt [CP t zoln [TP mir e [TP t leyenent ] ] ]b. Vos hot er nit gevolt [CP az-Π [CP t e [CP t zoln [CP mir e [TP t leyenent ] ] ]

] ]

As the Yiddish data shows clearly,that-trace effects are not properties of nominativeelements, or even of subjects of any sort. They appear whenever it becomes impossible todelete a trace in Spec-ΠP because of the normal behaviour ofΠ and C.

That-trace effects in Mainland Scandinavian

The mainland Scandinavian languages offer a particularly intricate array of data foranalysis ofthat-trace effects. These languages are grammatically uniform with respect tothat-trace effects in embedded v/2 structures, but they vary considerably with non-v/2complements.

I illustrate the pattern for embedded v/2 clauses with Danish data. In these clauses,the phrase in first position cannot be extracted via wh-movement. In this respect, Danish

Page 104: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 103

differs from German.

(42) a. *Hvilketwhich

æbleapple

sigersay

dethe

sagkundigeexperts

[CP atthat

[ΠP t smagertastes

ikkenot

bedstbest

]] ?

b. *Welcherwhich

Apfelapple

sagensay

diethe

Expertenexperts

[ΠP t schmeckttastes

am bestenthe best

]

Following Vikner (1991), I take the prohibition to be athat-trace effect, so that theDanish-German contrast follows from the presence in Danish of anat complementiser inembedded v/2 complements. Thethat-trace effect found in Danish differs from theEnglish case because it affects both topics and subjects in initial position.

TheTrace Deletion Constraintensures the ungrammaticality of (42-a) as follows.As at serves as the force marker for the complement clause, theΠ head in its complementhas no exceptional properties.Π attracts the subject with a [Tense] feature to place it atthe left periphery. The movement of the verb toΠ takes place in order to supportΠ,which does not in this case raise to C, possibly because of a Top head situated between CandΠ.4 The complementiser must now attract the subject to Spec-CP to allow longwh-movement to satisfy thePhase Impenetrability Constraint. But sinceat attracts thesubject with [NO], the trace in Spec-ΠP will not be deleted under theTrace DeletionConstraint. The chain formed by successive cyclic wh-movement in (42-a) will be anillegitimate LF object.

Turning now to the non-v/2 complements, the data are as follows. In standardSwedish and Danish,that-trace effects are found with theat(t) andsomcomplementisers.(Platzack, 1986; Hellan and Christensen, 1986). (The Swedish data in (43-a) comes fromHolmberg (1986); that in .)

(43) a. *Vemwho

sasaid

duyou

attthat

t hadehad

komit?arrived

b. *Vilkenwhich

filmfilm

kundecould

ingenno one

minnasremember

[ vemwho

(som)(that)

allaeveryone

troddethought

[ somthat

hadehad

regisserat?directed

(The problem in (43-b) is not the obvious wh-island effect, because Scandinavianlanguages do not exhibit such effects, as shown by Engdahl (1984).)

In Swedish, subject extraction is possible when the complementiser is omitted, as inEnglish. In Danish, there are two ways to permit the subject to escape: the

4In example (42-a), the verb movement itself will not be motivated, and the word order would thereforebe different, if there is nothing to block incorporation ofΠ by C.

Page 105: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 104

complementiser may be omitted, as in Swedish, or an expletiveder may be inserted to fillthe Spec-ΠP position. The latter strategy resembles the use of Yiddishesalreadydiscussed.

In Finnish-Swedish (Holmberg, 1986) and in Norwegian, though, sentences of thistype are grammatical, as seen in the Norwegian topicalisation in (44) (from Hellan andChristensen (1986)).

(44) a. PetterPeter

vetknow

jegI

atthat

skalwill

komme.come

The dialectal variation in mainland Scandinavian can be treated along the same lines as inEnglish. In standard Swedish, it appears, the complementiseratt can only appear as asecondary C, with nullΠ heading the complement. WhenΠ is not the complement of C,it may bear the set of features which allow a trace in Spec-ΠP to be deleted aftermovement to a higher position. But extraction of the subject in (43-a) then violates theTrace Deletion Constraintbecause the subject raises from a position in the internaldomain ofΠ to a position with different positional properties. In other dialects, however,att may appear as the primary complementiser in a bare primary CP complement clause,so that the (Finnish Swedish) structure in which wh-movement takes place is (45).

(45) [CP sa du [CP vem att [TP t hade komit ] ] ]

Whenvemraises to the matrix clause Spec-C, it lands in a position which is also in thechecking domain of a primary C, so the lower Spec-Π trace can be deleted. Thederivation thus succeeds.

Unlike theatt complementiser, theom ‘if’ complementiser blocks subject extractionin standard Swedish, but not in Danish, Finnish Swedish, or Norwegian. The contrastbetween Swedish and Norwegian is seen in (46).

(46) a. *Vemwho

undradewondered

duyou

omif

t hadehad

komit?arrived

(Swedish)

b. PetterPeter

vetknow

jegI

ikkenot

omif

t skalwill

komme.come

(Norwegian)

This contrast calls for a different type of cross-linguistic variation, one in which theabsence of a secondary C does not play a role. Theomcomplementiser can only be asecondary C, so the structure of the embedded clause in both (46) examples must includeboth an argumental C andΠ. The difference can be explained only by supposing thatomselects aΠ with the right featural content to permit subject extraction to occur. I concludethatomexceptionally selects a form ofΠ in which the [Tense] feature lacks the EPP

Page 106: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 105

property. (This must evidently be a marked situation cross-linguistically, given the rarityof structures like these in other languages, so we would expect to find positive evidencein the linguistic environment of Danish or Norwegian children.) With this specific featurecomplex in place, the mechanism for subject extraction will be the same as that for objectextraction, since both subjects and objects will be attracted directly from within TP.

Extraction from anif -clause is disallowed in English even for non-subjects, so thereis a secondary effect to explain in the Norwegian (46-b). Compare (46-b) withEnglish (47).

(47) *Who don’t you know [CP if [ ΠP t Π [TP t will come ]]]?

The explanation for the island effect in (47) should not be thatif cannot take a specifier,because the nature of Merge is such that a specifier should be possible ifif has the rightEPP feature. A better account would be thatif cannot have an EPPwh added to it, sincethis is a necessary step in successive cyclic movement. But even this adds a degree ofextra power to the model, which we would prefer to avoid. The minimal account of (47)is that Englishif simply does not incorporate check features onΠ, so that theΠP phaseboundary is not transparent. Then the ungrammaticality of (47) follows from thePhaseImpenetrability Constraint, becausewho is removed from the interior of theΠP phase.

Danish and Norwegian take this distinction one step further. In these languages,subject extraction is even allowed from within indirect wh-questions (Taraldsen, 1986b),as illustrated by the Norwegian (48).

(48) Detthat

eris

ena

mannman

somthat

viwe

ikkenot

skjønnerunderstand

hvawhat

sier.says

Once again, we find a complementiser—the null [wh] C—with the exceptional propertyof selecting aΠ complement with features permitting subject extraction. Just as with theomcomplementiser, the wh-complementiser selects aΠ in which [Tense] lacks the EPPproperty.

Maling and Zaenan (1978) show that Icelandic is immune tothat-trace effects, too.

(49) a. Hverwho

sagðirsaid

þúyou

aðthat

t hefðihad

borðaðeaten

þettathis

epli?apple

b. Þettathis

eris

maþurinn,the man

semthat

þeirthey

segjasay

aðthat

t hafihas

framiðcommitted

glæpinn.the crime

c. Þettathis

sverðsword

heldurthinks

konungurinnthe king

aðthat

t séis

galdrasverð.magic sword

Although Icelandic allows expletivepro in embedded questions, Maling and Zaenen

Page 107: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 106

show that expletives cannot be used to make subject extraction possible. Icelandic doesnot pattern with Yiddish in this respect. Nor does it behave like English and French,which allow a bare primary CP as a complement clause. Instead, Icelandic patterns withNorwegian, which makes use of anΠ optionally lacking in EPP features to allow subjectsto escape without crashing the derivation.

As IcelandicΠ is affixal, it must be supported by someX head. Like Norwegian,Πcan be supported by moving to the argumental C, as in (50-a). But Icelandic has a secondmechanism to provide support toΠ as well. WhenΠ is left in its base position, if thesubject is extracted from Spec-Π , then stylistic fronting can raise a non-finiteX head tosupportΠ in situ: (50-b) (from Rögnvaldsson and Thráinsson (1990, p. 32)).

(50) a. Hver sagðir þú [CP að-Π [CP t e [TP t hefði borðað þetta epli ] ] ] ?b. Þennan

thismannman

héltthought

égI

[CP aðthat

[CP t Π-fariðgone

[TP t hefðihad

veriðbeen

t medwith

t

áto

sjúkrahús.hospital

The Scandinavian data is important because it shows quite clearly how irrelevant anynotion of government—antecedent or lexical—is to thethat-trace effect. If we imaginethat the subject trace is in Spec-TP, then every form of government from outside TP willbe blocked in sentences like (50), since lexical government cannot access anything insideCP, and since antecedent-government must always be blocked by the wh-phrasehva. Thealternative proposed here relies only on the content of set of features borne byΠ, whichmay vary according to the context (and the language).

4.6 The ‘adjunct’ effect

Culicover (1991) notes that thethat-trace effect is cancelled out when certain adjunctsappear to the immediate right of the complementiser.

(51) Which car did Terry say that just yesterday had won the Indy?

Preposed arguments do not have the same effect (Culicover, 1993). In fact, ratherthan improving the status of a sentence from which the subject is extracted, argumentpreposing makes it even worse.

(52) a. *Which car did Terry say that the Indy, had won?b. *Which car did Terry say that to Tonya, had been sold?

Page 108: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 107

The effect of preposed arguments is unsurprising. In (52-b),to Tonyais topicalised andadjoined toΠP, where it can be checked by the Top head to its immediate left. Therefore,the structure of the complement clause prior to wh-movement must be (53).

(53) [CP that [TopP Top [ΠP to Tonya [ΠP which carΠ [TP t had be soldt ]]]]]

Movement ofwhich carfrom Spec-ΠP to Spec-CP will only take place if C takes on the[NO] feature, and then theTrace Deletion Constraintwill ensure that the trace inSpec-ΠP is not deleted, leading ultimately to a crash at LF.

In fact, many adjuncts behave like fronted arguments in this respect. Haegeman(2003) shows that adjuncts which are construed in a lower clause than the one fromwhich the subject is extracted are unable to cancel out thethat-trace effect.

(54) a. ?Beth said thatjust yesterdayPeter thought that this car had won the Indy.b. ?Who did Beth say thatjust yesterdaythought that this car had won the Indy?

In (54-a), the adjunctjust yesterdaycan marginally be construed as referring to the timeof the race, rather than the time of Peter’s thinking. But in (54-b), where the suject of themiddle clause is extracted, the adjunct can only be construed locally, as referring to thethinking time.

Haegeman’s treatment of this contrast is that adjuncts construed in a lower clause arefound in the higher clause only by virtue of movement, while those construed locally arebase-generated in the position where they appear. Thus, in (51), the adjunctjust yesterdayis Merged into the position immediately afterthat, while in (54-a), under the lowconstrual reading, the adjunct is introduced initially into the bottom clause, and thenraises into the left periphery of the matrix clause.

Haegeman’s account, which I accept, fits naturally into the present theory.Topicalized arguments and adjuncts may raise freely to adjoin toΠP in order to be closeenough to Top to allow [Topic] features to be checked. The resulting structure in bothcases will be (55):

(55) [CP (that) [TopP Top [ΠP topic-XP [ΠP DP (Π) [TP t . . .t . . . ]]]]]

Subject extraction from Spec-ΠP in (55) will inevitably run afoul of theTrace DeletionConstraint, so thethat-trace effect shoud be found here, just as it is in clauses with notopicalisation.

The disappearance ofthat-trace effects when an appropriate adjunct is Mergeddirectly into a position to the right ofthat is hardly more difficult to explain. If the adjunctis adjoined toΠP, then the structure will be equivalent to (55) in all relevant respects. Butif some adjuncts may be Merged directly into Spec-ΠP, i.e. Merged withΠP, withΠ

Page 109: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 108

projecting, then the situation is quite different. Then the structure of (51) will be (56).

(56) Which car did Terry say [CP that [ΠP just yesterday [TP t had won the Indy ]]]?

As the subject does not raise to Spec-ΠP in (56), there is no Spec-ΠP trace to be deleted.TheTrace Deletion Constraintwill then be irrelevant to subject extraction in thissentence.5

As for why some locally construed adjuncts may Merge withΠP, the situationappears to be largely analogous to that ofthere-expletives which appear in Spec-TP. InChomsky’s (2001b) model,theremay form a part of the Numeration which constructs theCP phase once a vP phase is complete. The EPP feature of T can be checked either bymovement of the nominative DP or by external Merge of a new element from thenumeration, i.e.there, with external Merge being preferred when it is possible.Thereisautomatically Merged with TP, therefore, whenever it appears in the CP numeration.(With CP understood asΠP, this description remains valid for my purposes.)

The description of expletivethereis otherwise restricted by the principle of Full

5As Culicover observes, subjects can be extracted even when a monotone-decreasing adjunct appears, ina verb-second structure.

i Which car did Tony say that at no earlier date had been entered in the race?

He assumes that the structure in this case would be one in which the auxiliary verb raises to C, by analogywith the ‘negative inversion’ structure required when the subject remains within the complement clause:

ii Tony said that at no earlier date had that car been entered in the race.

Notice however that the surface string gives no evidence of T-to-C movement in the complement clause,with the subject missing. In fact,do-support is not possible in this configuration, as Culicover’s own datashows.

iii ??Leslie is the person who I said that only in that election did run for public office.

iv Leslie is the person who I said that only in that election ran for public office.

And as wh-movement eliminates theMinimal Link Conditionviolation, there is in fact no motivation forthe T-to-C movement: no violation, no repair. So this case falls together with that of the non-negativeadjuncts.

Thethat-trace effect is not cancelled out entirely in the negative-inversion structures. Even though thesubject can be extracted in this context, the negative phrase itself cannot be moved to a higher position fromSpec-Π .

iii *At no earlier date would Peter say thatt had his car been driven faster.

This parallels the Yiddishthat-trace effect discussed already (p. 102). The negative phrase cannot undergowh-movement here because its Spec-Π trace cannot be deleted afterwards.

Page 110: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 109

Interpretation.Therecannot appear in the vP numeration, because it cannot be anargument.Therewill automatically Merge with the first vP-external head with an EPPfeature, which is T, so the fact thattherealways originates in an A-position follow as amatter of course.

Now suppose that theΠP Numeration may include an adverb or the contents of alarger adjunct phrase: AdvP or PP. Adjuncts cannot be interpreted in A-positions, so suchan adverb may not Merge with TP. It must therefore Merge at theΠP level, becoming aspecifier. IfΠ checks its [Tense] feature on the subject or T at the same point in thederivation, then the EPP property of [Tense] may be satisfied without movement of anycategory at all.6

It follows that sentences with initial locally-construed adjuncts will be structurallyambiguous, with both of the (57) structures being legitimate.

(57) a. [ΠP tomorrowΠ [TP the rain will begin at noon ]]b. [ΠP tomorrow [ΠP the rainΠ [TP t will begin at noont ]]]

In languages whereΠP adjunction is impossible, such as the other Germaniclanguages, the only position for initial adjuncts will be Spec-ΠP. This position will beused both for external Merge and as a landing site for A-bar movement. Again,adjunct-initial clauses will be structurally ambiguous, so the Danish example in (58) willbe subject to both of the analyses shown in (59):

(58) a. Igåryesterday

köptebought

LenaLena

ena

nynew

bok.book

(59) a. [ΠP igår köpte [TP Lena en ny bok ]]b. [TopP Top [ΠP igår köpte [TP Lena en ny bokt ]]]

4.7 Short wh-movement of the subject

Relative clauses

One of the persistent problems with government-based approaches tothat-trace effectshas been thatshortwh-movement of subjects is typically possible in conditions which

6Chomsky (2001a) supposes that adjuncts may originiate outside the ‘narrow’ phrase marker, in fact, sothat a three-dimensional structure is appropriate for representing how adjuncts are included into a structure.The demands of the PF interface require that linearization of the 3-D structure be possible, so a 3-Dadjunction structure must be collapsed into a simpler configuration, in which the adjunct appears in thesame ‘plane’ as other material. I set the conflation source of some adverbs to the side for the purposes ofthis discussion.

Page 111: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 110

appear to violate the ECP. The problem is most evident with English relative clauses, butit shows up in other contexts, too. Consider the following data.

(60) a. the house (that) Jack builtb. the house (that) Jack lives inc. the house whose roof Jack repairedd. the house in which Jack lives

English relative clauses in which a non-subject serves as the relative pronoun allow for arange of surface realizations of C and the relative pronoun. If the relative pronoun isrealised phonologically, then C must be silent, at least for most speakers. This patternpresumably represents the ‘doubly filled Comp effect’ (?). When the relative pronoun isnot realised, then the complementiser may appear, but it may also be silent.

Boškovic and Lasnik (2003) propose that the optional silent complementiser inrelative clauses is licensed in the same way as null complementisers in declarativecomplement clauses, that is, by Morphological Merger with an adjacent matrix head. Inrelative clauses, however, the null complementiser joins with a nominal head, rather thana verbal one. In that case, the data in (60) is unproblematic.

When the relative pronoun is the topmost subject of the relative clause, the datapatterns differently.

(61) a. the guy *(that) built this houseb. the guy who built this house

In such cases, the complementiser is omitted only when the relative pronoun is realisedovertly. Optional deletion of the complementiser is impossible.

The fact that the complementiser may appear at all is problematic forgovernment-based theories, simply because thethat in (61-a) should presumably blockproper goverment of a subject trace. The fact that the complementiser is obligatory isentirely inexplicable in such theories.7

To see how the data may be handled better, we once again turn to the distribution ofthe Frenchqui complementiser. Thequi complementiser also—and mostfrequently—appears in relative clauses in which the subject is a null operator.8

7For Rizzi (1990a), the overtthat in subject relatives is itself able to properly govern the subject trace,so that the ECP is satisfied, if only by stipulation. The fact that the complementiser may not be omittedremains without explanation in his account.

8Thequi complementiser also appears in the complement to perception verbs, in the ‘pseudo-relative’construction. Examples appear in (i).

Page 112: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 111

(62) l’hommethe man

qui t nousus

ahas

aidé.helped

In relative clauses,qui must have essentially the same properties as thequi which appearswhen long wh-movement of the subject occurs. In the analysis ofqui in complementclauses, it was established thatΠ is realised asqui only if Π is not incorporated by C.This reasoning implies that there is no higher C to incorporateΠ in relative clauses like(62). Evidently, no CP is necessary in subject relative clause, probably because the Dhead of DP supplies the semantic content which would otherwise be brought to the tableby the higher C. Relative modifiers are frequently ‘smaller’ than finite clauses, so that thestructure (63) may be appropriate for example (62).9

(63) DP

����HHHH

D

l’

NP

����HHHH

N

homme

ΠP

���

HHH

Oi Π ′

���

HHH

qui IP����

PPPP

ti nous a aidé

Relative clauses in French are always formed with A-bar movement of the relativepronoun. In (63), the A-bar movement is not driven by a [wh] feature, but rather by the[Tense] feature which is always present inΠ. There appears to be no need to use a

(i) a. Toutall

lethe

mondeworld

ahas

entenduheard

MarcMarc

quiqui

ronflait.snored

‘Everyone heard Mark snore.’b. Louise

Louiseahas

vuseen

SalomonSalomon

quiqui

courait.ran

‘Louise saw Salomon run.’

Cases like this also appear to involve a bareΠP as the complement. (cf. Guasti (1993) for a similarclaim.) No secondary C appears, soΠ need not raise to check the [Tense] feature, and can be realised asqui. The questions raised by such constructions include the nature of the doubly-filled Comp effect, and theproperties of perception verbs which allow them to accept ‘small’ complements. Nothing in the textappears to shed any light on these questions, especially as concerns the pseudo-relative constructions.

9Kayne (1995) maintains that the head noun in a restrictive relative clause remains within the relativeCP. The text analysis can be readily transposed to such a framework.

Page 113: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 112

[wh]-checking complementiser in order to find the right interpretation for the relativeclause.

When the relative pronoun is not the subject, it still must undergo A-bar movementout of TP. Since the Spec-ΠP position is unavailable—being already occupied by thesubject—the only way for A-bar movement to take place is if a complementiser isMerged withΠP to provide a second A-bar landing site. With non-subject relativeclauses, then, the structure will always involve a full CP.

Returning now to the English data in (61), the question whythat must not be omittedin subject relative clauses now looks quite different. If English is like French, then thestructure of (61) will be (64).

(64) DP

���

HHH

D

the

NP

���HHH

N

guy

ΠP

����

HHHH

OP Π ′

����HHHH

Π

that

TP

�����PPPPP

t built this house

The wordthat in this case must be an instance ofΠ, and not a higher complementiser atall. In other words, the wordthat must have multiple uses in standard English.Sometimes it serves as a force-marking C in declarative clauses; other times, it functionsas the head of a bareΠP in relative clauses. And in some dialects,that serves asΠ incomplement clauses whereΠP appears without a sheltering CP.

Notice that the absence of anythat-trace effect is expected in structures like (64).Since thethat-trace effect arises only when there is movement from Spec-ΠP which doesnot delete the trace, there should never be any effect when an operator just raises toSpec-ΠP and stays there, and that is the situation in subject relative clauses.

This treatment of subject relatives analysis offers a principled account of one varietyof so-called ‘vacuous movement’ effects. Chung and McCloskey (1983a) observe thatrelative clauses formed by subject extraction are weaker extraction islands than relativeclauses formed by extraction of any other type of NP. Examples (from Chung andMcCloskey (1983a)) appear in (65).

(65) a. That’s one trick that I’ve known a lot of people who’ve been taken in byt.

Page 114: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 113

b. Isn’t that the song that Paul and Stevie were the only ones who wanted torecordt?

Such sentences compare favorably with sentences in which the operator comes fromsomewhere else, as in (66).

(66) ??Isn’t that the song writer that ballads were the only things that Paul wouldwrite t for t?

In indirect questions, the contrast is absent.10

As Chung and McCloskey (1983b) argue, this contrast is best explained if subjectsand non-subjects appear in different positions. For them, the subject occupies a differentposition because it does not move from its TP-internal position at all. But a lessad hoctreatment is possible in the present model. Since subject relative clauses do not raisehigher than Spec-ΠP, it is possible to find a vacant Spec-CP escape hatch fornon-subjects in these cases. Suppose that early in the derivation, the structure of therelevant DP in (65)[a] is (67):

(67) [DP people [CP ___∅ [ΠP whoi Π [ ti have been taken in by which ]]]]

The CP in (67) is a legitimate relative clause because it contains a relative pronoun whichhas undergone A-bar movement. The relative pronounwhodoes not need to raise anyhigher than Spec-ΠP, because there is no need for it to check an interpretable [wh]feature on a higher C. The relative clause itself may be a full CP, however, as long as thecomplementiser lacks phonetic content (becausewho is realised overtly). Given thisstructure, there is no reason why C could not aquire a [NO] feature to attract the lowerrelative pronounwhichup to Spec-CP, leavingwhich in a position high enough toundergo subsequent movement into still higher positions in the sentence.

In contrast, the structure of the island DP in (66) can only be (68).

(68) [DP the only things [CPjOP that [ΠP Pauli [TP ti would writetj for whom ]]]]

Here there is no escape hatch available for the relative pronounwhom. Both Spec-CP andSpec-ΠP are already occupied, sowhomcan only escape from the relative clause byviolating thePhase Impenetrability Constraint.

The pan-Scandinavian relative clause complementisersomillustrates a furtherwrinkle in the analysis of relative clauses. As shown by Taraldsen (1986b),somisobligatory in subject relatives, but optional with non-subject relatives.

10I differ from Chomsky (1986), who finds some slight effect even here.

Page 115: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 114

(69) a. kvinnanwoman.the

somtalarspeaks

medwith

AndersAnders

‘the woman who is speaking with Anders’b. kvinnan

woman.the(som)Anders

Anderstalarspeaks

medwith

‘the woman who Anders is speaking with’

In the former, it suffices to suppose thatsomis an instance ofΠ. As subject relativeclauses are typically bareΠP, the appearance ofsomin (69-a) follows the same pattern aswe have seen in French and English.

The pre-subject position ofsomis new however. Ifsomis Π, and the subjectoccupies Spec-ΠP, then how can this word order be derived. The explanation lies in thechecking requirements of C in relative clauses like (69-b). Like most othercomplementisers, the one which appears in non-subject relative clauses must check a[Tense] feature with the EPP property. It does so in (69-b) in the same way as usual, byincorporatingΠ. The net result is thatΠ must raise to C in non-subject relatives, and notin subject relatives.

The optional appearance ofsomin the C position in (69-b) may now be taken toreflect a degree of optionality in the morphological realisation of structures in which anull relative C containsΠ in relative clauses, i.e. of structures like (70):

(70) C�� HH

C

Π′

som

Given such structures, it is apparently possible to realise either C orΠ, but perhaps notboth.

Embedded Questions

In embedded questions, the movement of the subject must also be consistent with theTrace Deletion Constraint. Consider (71).

(71) Bob enquired which desperado had pitched this tent.

For this case, the obvious, familiar derivations are excluded by theTrace DeletionConstraint. Suppose that the structure of the complement clause is (72) when theinterrogative C is Merged withΠP.

Page 116: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 115

(72) [CP C[wh]

[Tense]

[ΠP which desparadoΠ [TP t had pitched this tent ]]]

C must attract a wh-phrase and a phrase bearing [Tense]. If it checks [wh] by attractingthe subject, then it will later check [Tense] by incorporatingΠ. But if C checks [wh] onthe subject, the trace in Spec-ΠP will not be deleted under theTrace Deletion Constraint,giving rise to an ininterpretable set of A-bar chains at LF. On the other hand, if C checks[Tense] on the subject, raising the subject to Spec-CP, then there will be no way for it tocheck its [wh] feature, and the derivation will crash anyway.

In fact, the principles which give rise to thethat-trace effect ensure that shortwh-movement of the subject will be impossible unless some loophole can be identified.This result is probably the right one, because there are languages in which even shortwh-movement cannot take place without unexpected effects arising. Consider the Yiddishdata in (73) (from Diesing (1990), for example.

(73) IkhI

veysknow

nitnot

verwho

*(es) isis

gekumen.come

‘I know who came.’

Without the expletivees, short wh-movement of the subject is impossible in Yiddish.This follows from theTrace Deletion Constraint. If the subject is raised from Spec-ΠP toSpec-CP, the trace in Spec-ΠP is not deleted, and the derivation will crash at LF. But if a[Tense]-bearing expletive is used to fill the Spec-ΠP position, then the subject may raisedirectly from Spec-TP—an A position—to Spec-CP.11 Only a single binary A-bar chainis formed by this movement, so there is no problem with interpretation at LF.

Yiddish also employs adjuncts to fill the Spec-ΠP to permit short wh-movement ofthe subject, as in (24-a), repeated here.

(24-a) . . . vias

ikhI

veysknow

voswhat

bayby

mirme

tutdoes

zikhitself

‘. . . as I know what goes on with me’

The question then is what types of strategies are available in different languages topermit short wh-movement of the subject to occur. Obviously the Yiddish strategy is notthe only one—English cannot use expletive subjects in this way, for example.

Taraldson’s 1986b description of embedded questions in Norwegian indicates asecond strategy available to some languages. In Norwegian,somappears obligatorily

11Movement from Spec-TP to Spec-CP is possible in Yiddish even in verb-second contexts, becauseabstract incorporation ofΠ by C is not blocked by movement of the verb toΠ.

Page 117: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 116

with a local subject wh-phrase ((74)). (The Norwegian data in ((74)) is taken fromTaraldsen (1986a).)

(74)we

Viknow

vetwho

hvem *(som)talks

snakkerwith

medMary

Marit.

The appearance ofsomin such embedded questions is initially quite surprising, given thatNorwegian is otherwise sensitive to doubly-filled Comp effects, even in cases wheresomis involved, as in the relative clause in (75).

(75) Herhere

eris

mannenthe man

hviswhose

hesthorse

(*som) vantwon

løpet.the race

Recall as well thatsomin non-subject relative clauses may be omitted freely,because there is some optionality in the realisation of the C-Π structure formed when Cincorporatessom. But somcannot be omitted in (74).

These peculiar properties ofsomin embedded subject questions both indicate thatsomin (74) does not appear in the C position. IfsomoccupiesΠ, then the absence of adoubly-filled Comp effect follows, simply because there is no overt head in C to clashwith the overt wh-phrase specifier. And ifsomremains inΠ, then it must always be overt,just as it is in the subject relative clauses.

The structure of the embedded question in (74) then must be (76).

(76) [CP hvem [C ∅ ] [ ΠP t [Π som ] [TP t snakker med Marit ]]]

In (76), the absence ofΠ-to-C movement implies that the [Tense] feature of C must bechecked by the subject. This is possible only if the subject is attracted from Spec-ΠP bythe [Tense] feature and not by the [wh] feature. And if the subject raises to check the[Tense] feature, then the trace in Spec-ΠP will automatically be deleted under theTraceDeletion Constraint. The A-bar chain which is left will be binary, and hence interpretableat LF.

This structure, in which the wh-phrase andsomappear in different projections, issupported by the fact thatsomcan be further separated from its wh-phrase byright-dislocation in Swedish (Holmberg, 1986).

(77) JagI

vetknow

vilkawhich

fotbollslag,football teams

ochand

PeterPeter

vetknows

vilkawhich

hästarhorses

som kommerwill

att

vinnawin

denthis

här veckan.week

Page 118: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 117

The right-dislocated phrase isΠP, which containssom, and out of which the subjectwh-phrases have raised to become specifiers for CP.

It is impossible, in fact, to leavesombehind and dislocate a bare TP (Platzack,1986). In this respectsomis more closely bound to TP than an argumentalattcomplementiser is. (Examples are taken from Platzack (1986)).

(78) a. *JagI

vetknow

vilkawhich

fotbollslagsoccer team

som, ochand

PeterPeter

vetknows

vilkawhich

hästarhorses

som

kommercome

attto

vinnawin

denthis

här veckan.week

b. JagI

trorbelieve

att,that,

menbut

vetknow

intenot

säkertfor sure

om,whether

dinyour

teoritheory

äris

korrekt.correct

The [wh] feature on an interpretable interrogative complementiser normally has theEPP property which drives wh-movement. In Norwegian structures like (76), it must bethe case that [wh] in C may optionally do without the EPP property. In other words, the[wh] feature of C simply checks the subjectin situ in Spec-ΠP, before the subject raisesto Spec-CP to check [Tense].

Two distinct strategies can be identified to allow short wh-movement in Germanic,the Yiddish strategy with expletive subjects, and the Norwegian strategy with a weak[wh] feature in C. The latter appears to be the more common one, as other Germaniclanguages show little evidence of an expletive pronoun in embedded subject questions. InEnglish, for example, only the Norwegian strategy seems plausible. Thus, the derivationof the embedded question in (79) should involve movement of the subject wh-phrasefrom Spec-ΠP to Spec-CP to check the [Tense] feature, and no true wh-movement.

(79) Bobbie wondered [CP which guy C [ΠP t Π [TP t had washed the windows ]]]

If subject wh-phrases are not attracted by wh-movement, then there is an explanation forthe assymetries in (80).

(80) a. I wonder whot likes Jan andt impresses Sue.b. I wonder who Jan likest and Sue counts ont.c. *I wonder who Jan likest andt impresses Sue.d. *I wonder whot likes Jan and Sue impressest.

These data are the second half of the cases discussed by Williams (1978), who shows thatsubjects cannot be extracted ‘across the board’ together with non-subjects. (An accountfor this pattern in root clauses was presented in section (35) (page 36), but as root clausesare subject to different principles, that explanation does not extend to the data in (80).)

Page 119: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 118

Only local extraction of subjects is constrained in this manner. When subjects areextracted from an embedded clause, they may be extracted in parallel with a non-subject.

(81) a. I wonder who Jan likest and Sue believes/wantst to be dependable.b. ?I wonder who Jan likest and Sue thinkst is dependable.c. I wonder who Sue thinkst is dependable and Jan likest.

The acceptability of ((81-c)) shows that the issue is not one of Case conflict, sincewhocomes from a nominative source in one clause and an accusative source in the other.Instead, the problem has to do with the derivation of embedded questions with shortwh-movement of the subject. In (80-c), for example, the wh-phrasewhomust be attractedto its Spec-CP position by the [Tense] feature in order to satisfy the requirements of thefirst conjunct, but it must be attracted by [wh] for the second conjunct. Since evenacross-the-board movement must be consistent in the features which drive movement, thederivation cannot go through.

4.8 Conclusions

The minimalist program in syntax sets out to investigate how close language designapproaches a ‘perfect’ design, with stipulation in grammar reduced to ‘virtual conceptualnecessity’. These are, of course, terms of art, which to some extent we interpret in theways that our analyses require. But at least the formulation of locality principles in aminimalist syntax should not depart from the core notions of any configurational theory:heads, projections, sisters, chains, and features. The primary intent of the present workhas been to show that minimalism in this sense can provide an illuminating account ofsome constructions which had been thought problematic, and which have provedremarkably resistant to explanation in previous, government-based models of syntax. Asalways, it remains to be seen which parts of this account will turn out to be misguided. Itis also unclear to what extent the approach developed here is minimalist in any deepsense, but the minimalist program is a hunch and a gamble anyway, and valuable as suchonly to the extent that it leads us to explore avenues we might not otherwise have noticed.

Besides the descriptions of different constructions developed along the way, theavenues which appear interesting to me at the end of this work are some which could nothave been explored before. The nature of the [Tense] feature certainly needs further studyat this point. While I hope to have shown that this feature ofΠ is centrally engaged in thesyntactic properties of subjects and the A-bar system, it remains unclear how strong theevidence for this particular terminological choice is, and whether further progress mightresult from adopting a different label. Similarly, the elementary typology of A-bar

Page 120: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

CHAPTER 4. THAT-TRACE EFFECTS. . . 119

attractor features presented here does the job that it is designed to do, but there shouldultimately be a foundation for such a typology which does more than stipulate thesuperset/subset relations. Clearly, much more work remains to be done in this area.

Other, potentially more problematic questions arise concerning the nature of chainsformed with resumptive pronouns or expletive wh-operators. The extensive argumentspresented in McCloskey (2002) for complex chains with A-bar movement of resumptivepronouns is particularly compelling in this respect. I have proceeded on the assumptionthat A-bar chains must be binary for LF interpretation, and there is evidence to supportthis assumption. But these other phenomena exist as well, and the theory must at somepoint be made consistent. Again, there is work to do. But not now.

Page 121: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

Bibliography

Baker, M. (1988).Incorporation: A Theory of Grammatical Function Changing.Chicago: University of Chicago Press.

Baltin, M. (1982). A landing site theory of movement rules.Linguistic Inquiry 13, 1–38.

Belletti, A. (1990).Generalized Verb Movement: Aspects of Verb Syntax. Torino:Rosenberg and Sellier.

Belletti, A. (1992). Agreement and case in past participle clauses in Italian. In T. Stowelland E. Wehrli (Eds.),Syntax and Semantics, Volume 26: Syntax and the Lexicon, pp.21–45. Academic Press.

Bobaljik, J. and P. Branigan (2004). Eccentric agreement and multiple Case-checking. InA. Johns, D. Massam, and J. Ndayiragije (Eds.),Ergativity. Kluwer. Volume in review.

Boškovic, Željko. (1997).The Syntax of Nonfinite Complementation: An EconomyApproach. Cambridge, Mass.: MIT Press.

Boškovic, Željko. and H. Lasnik (2003). On the distribution of null complementisers.Linguistic Inquiry 34(4), 527–546.

Branigan, P. (1992).Subjects and Complementizers. Ph. D. thesis, MIT. Distributed byMIT Working Papers in Linguistics.

Branigan, P. (1996a, April). The difference between root and embedded wh-questions inGermanic. Paper presented at GLOW 11, Athens, Greece.

Branigan, P. (1996b). Treatingthat-trace variation. In J. Black and V. Motapanyane(Eds.),Microparametric Syntax: Dialect Variation in Syntax, pp. 25–39. JohnBenjamins.

Branigan, P. (1996c). Verb-second and the A-bar syntax of subjects.StudiaLinguistica 50(1), 51–79.

120

Page 122: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

BIBLIOGRAPHY 121

Bresnan, J. (1972).Theory of Complementation in English Syntax. Ph. D. thesis, MIT.

Bresnan, J. (1994). Locative inversion and the architecture of Universal Grammar.Language 70(1), 72–131.

Cheng, L. (1991).On the Typology of Wh-Questions. Ph. D. thesis, MIT.

Chomsky, N. (1955/1975).Logical Structure of Linguistic Theory. New York: Plenum.Excerpted from unpublished 1955/1956 manuscript.

Chomsky, N. (1965).Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.

Chomsky, N. (1981).Lectures on Government and Binding. Dordrecht: Foris.

Chomsky, N. (1986).Barriers. Cambridge, Mass.: MIT Press.

Chomsky, N. (1993). A minimalist program for linguistic theory. In K. Hale and S. J.Keyser (Eds.),The View from Building 20: Essays in Linguistics in honor of SylvainBromberger., pp. 1–52. Cambridge, Mass.: MIT Press.

Chomsky, N. (1995).The Minimalist Program. Cambridge, Mass.: MIT Press.

Chomsky, N. (2001a). Beyond explanatory accuracy. Ms. MIT.

Chomsky, N. (2001b). Derivation by phase. In M. Kenstowicz (Ed.),Ken Hale: A Life inLanguage. Cambridge, MA: MIT Press.

Chomsky, N. (to appear). Minimalist inquiries: the framework. In R. Martin,D. Micheals, and J. Uriagareka (Eds.),Step by Step: Essays on Minimalist Syntax inHonor of Howard Lasnik. MIT Press.

Chomsky, N. and H. Lasnik (1991). Principles and parameters theory. In W. S. J. Jacobs,A. van Stechow and T. Vennemann (Eds.),Syntax: An International Handbook ofContemporary Research. Berlin: Walter de Gruyter.

Chung, S. and J. McCloskey (1983a). On the interpretation of certain island facts inGPSG.Linguistic Inquiry 14, 704–713.

Chung, S. and J. McCloskey (1983b). On the interpretation of certain island facts inGPSG.Linguistic Inquiry 14, 704–713.

Cinque, G. (1990).Types of A-bar Dependencies. MIT Press.

Collins, C. (1996).Local Economy. Cambridge, MA: The MIT Press.

Page 123: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

BIBLIOGRAPHY 122

Coopmans, P. (1989). Where stylistic and syntactic processes meet: Locative inversion inEnglish.Language 65, 728–751.

Culicover, P. (1993). The adverb effect: Evidence against ECP accounts of that-t effects.NELS 24, 97–110.

Culicover, P. W. (1991). Topicalization, inversion, and complementizers in English. Ms.,Ohio State University.

Den Besten, H. (1983). On the interaction of root transformations and lexical deletiverules. In W. Abraham (Ed.),On the Formal Syntax of the Westgermania, pp. 47–131.John Benjamins.

Diesing, M. (1990). Verb movement and the subject position in Yiddish.NaturalLanguage and Linguistic Theory 8, 41–79.

Dikken, M. D. and A. Næss (1993). Case dependencies: The case of predicate inversion.The Linguistic Review 10, 303–336.

Engdahl, E. (1984).The Syntax and Semantics of Constituent Questions with SpecialReference to Swedish. Dordrecht: Reidel.

Fitzpatrick, J. (2003, April). Deletion and movement. Paper presented at GLOW 2003,Lund, Sweden.

Godard, D. (1985). Propositions relatives, relations anaphoriques et prédication. Thèsed’Etat, Université de Paris 7.

Guasti, T. (1993).Causative and Perception Verbs. Turin: Rosenberg and Selier.

Haeberli, E. (1999).Features, Categories and the Syntax of A-positions: Synchronic andDiachronic Variation in the Germanic Languages. Ph. D. thesis, University of Geneva.

Haegeman, L. (2003). Long adjunct movement.Linguistic Inquiry 34(4).

Haider, H. (1993). ECP-Etüden: Anmerkungen zur Extraktion aus eingebettetenVerb-Zweit-Satzen.Linguistische Berichte, 185–203.

Harley, H. and E. Ritter (1999). Meaning in morphology: Motivating a feature-geometricanalysis of person and number. Unpublished ms.

Hellan, L. and K. K. Christensen (Eds.) (1986).Topics in Scandinavian Syntax,Dordrecht. Reidel.

Page 124: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

BIBLIOGRAPHY 123

Henry, A. (1995).Belfast English and standard English. Oxford University Press.

Hoekstra, E. (1994). Expletive replacement, verb-second and coordination.TheLinguistic Review 11, 285–297.

Hoekstra, T. and R. Mulder (1990). Unergatives as copular verbs: Locational andexistential predication.The Linguistic Review 7, 1–79.

Hoge, K. (1997). That-t effects in English and Yiddish. In G. M. Alexandrovna andO. Arnaudova (Eds.),The Minimalist Parameter, pp. 233–248. Benjamins.

Holmberg, A. (1986).Word Order and Syntactic Features. Ph. D. thesis, University ofStockholm.

Holmberg, A. and C. Platzack (2003). The Scandinavian languages. To appear inGuglielmo Cinque and Richard Kayne (eds),The Comparative Syntax Handbook.

Iatridou, S. and A. Kroch (1992). The licensing of CP-recursion and its relevance to theGermanic verb-second phenomenon.Working Papers in Scandinavian Syntax 50,1–25.

Kayne, R. S. (1972). French relative ‘que’. In F. Hensey and M. Luján (Eds.),CurrentStudies in Romance Linguistics, pp. 255–299. Washington: Georgetown UniversityPress.

Kayne, R. S. (1981). On certain differences between French and English.LinguisticInquiry 12, 349–371.

Kayne, R. S. (1989). Facets of past participle agreement in Romance. In P. Beninca(Ed.),Dialect Variation in the Theory of Grammar. Dordrecht: Foris.

Kayne, R. S. (1995).The Antisymmetry of Syntax. MIT Press.

Kitagawa, Y. (1986).Subject in Japanese and English. Ph. D. thesis, University ofMassachusetts at Amherst.

Koopman, H. and D. Sportiche (1991). The position of subjects.Lingua 85, 211–258.

Ladusaw, W. (1979).Polarity Sensitivity as Inherent Scope Relations. Ph. D. thesis,University of Texas at Austin.

Lasnik, H. and M. Saito (1984). On the nature of proper government.LinguisticInquiry 15.

Page 125: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

BIBLIOGRAPHY 124

Levin, B. and M. R. Hovav (1994).Unaccusativity: At the Syntax-Lexical SemanticsInterface. Cambridge, MA: MIT Press.

Lightfoot, D. (1991).How To Set Parameters: arguments from language change.Cambridge, Mass.: MIT Press.

Maling, J. and A. Zaenan (1978). The nonuniversality of a surface filter.LinguisticInquiry 9(3), 475–498.

Massam, D. (1985).Case Theory and the Projection Principle. Ph. D. thesis, MIT.

McCloskey, J. (1997). Subjecthood and subject positions. In L. Haegeman (Ed.),Elements of Grammar. Dordrecht: Kluwer.

McCloskey, J. (2002). Resumption, successive cyclicity, and the locality of operations. InS. D. Epstein and T. D. Seely (Eds.),Derivation and Explanation in the MinimalistProgram, pp. 184–226. Malden, MA: Blackwell.

Moorcroft, R. (1995).Clause-Level Functional Categories in Germanic V2 Languages.Ph. D. thesis, University of Toronto.

Müller, G. (1995).A-Bar Syntax: a Study in Movement Types. Berlin: Mouton deGruyter.

Müller, G. and W. Sternefeld (1993). Improper movement and unambiguous binding.Linguistic Inquiry 24, 461–507.

Noyer, R. (1992).Features, Positions and Affixes in Autonomous MorphologicalStructure. Ph. D. thesis, MIT.

Pesetsky, D. (1982). Complementizer trace phenomena and the nominative islandcondition.The Linguistic Review 1, 297–343.

Pesetsky, D. (1988). Language-particular processes and the earliness principle. ms. MIT.

Pesetsky, D. (1992.). Zero syntax, vol. 2. Ms. MIT, Cambridge, Mass.

Pesetsky, D. (1994). Some long-lost relatives of Burzio’s generalization. Paper presentedat the conference on Burzio’s Generalization, Holland.

Pesetsky, D. (1998). Optimality principles of sentence pronounciation. In P. Barbosa,D. Fox, P. Hagstrom, M. McGinnis, and D. Pesetsky (Eds.),Is the Best Good Enough?:Optimality and Competition in Syntax. Dordrecht: MIT Press.

Page 126: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

BIBLIOGRAPHY 125

Pesetsky, D. (2000).Phrasal Movement and Its Kin. MIT Press.

Pesetsky, D. and E. Torrego (2000). I-to-C movement: Causes and consequences. InM. Kenstowicz (Ed.),Ken Hale: A Life in Language. Cambridge, Mass.: MIT Press.

Platzack, C. (1986). COMP, INFL, and Germanic word order. In L. Hellan and K. K.Christensen (Eds.),Topics in Scandinavian Syntax, Studies in Natural Language andLinguistic Theory, Dordrecht, pp. 185–234. Reidel.

Plunkett, B. (1997). Attract and covert merge: Predicting interrogative variation. InG. M. Alexandrovna and O. Arnaudova (Eds.),The Minimalist Parameter, pp.159–174. Benjamins.

Polinsky, M. and E. Potsdam (1999). The syntax of topic and long-distance agreement inTsez. Ms. UCSD and Yale.

Pollock, J.-Y. (1989). Verb movement, universal grammar and the structure of IP.Linguistic Inquiry 20, 365–424.

Postal, P. (1974).On Raising: One Rule of English Grammar and Its TheoreticalImplications. Cambridge, Mass.: MIT Press.

Progovac, L. (1992). Nonnegative polarity licensing must involve Comp.LinguisticInquiry 23, 341–347.

Richards, N. (1997).What Moves Where in Which Language. Ph. D. thesis, MIT.

Richards, N. (1998). The principle of Minimal Compliance.Linguistic Inquiry 29,599–629.

Rizzi, L. (1990a).Relativized Minimality. Cambridge: MIT Press.

Rizzi, L. (1990b). Speculations on verb second. In J. Mascaró and M. Nespor (Eds.),Grammar in Progress: Glow Essays for Henk van Riemsdijk, pp. 375–386. Foris.

Rizzi, L. (1997). The fine structure of the left periphery. In L. Haegeman (Ed.),Elementsof Grammar: Handbook of Generative Syntax. Dordrecht: Kluwer.

Roberts, I. (1997). Restructuring, head movement and locality.Linguistic Inquiry 28(3),423–460.

Rochemont, M. (1989). Topic islands and the subjacency parameter.Canadian Journalof Linguistics 34, 145–170.

Page 127: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

BIBLIOGRAPHY 126

Rögnvaldsson, E. and H. Thráinsson (1990). On Icelandic word order once more. InJ. Maling and A. Zaenen (Eds.),Modern Icelandic Syntax, Number 24 in Syntax andSemantics, pp. 3–40. Academic Press.

Ross, R. (1967).Island Constraints in Syntax. Ph. D. thesis, MIT.

Ruwet, N. (1991).Syntax and Human Experience. Chicago: University of Chicago Press.

Shlonsky, U. (1994). Agreement in Comp.The Linguistic Review, 351–375.

Sobin, N. (1987). The variable status of COMP–trace phenomena.Natural Languageand Linguistic Theory 5, 33–60.

Sportiche, D. (1988). A theory of floating quantifiers and its corollaries for constituentstructure.Linguistic Inquiry 19, 425–449.

Staudacher, P. (1990). Long movement from verb-second-complements in German. InG. Grewendorf and W. Sternefeld (Eds.),Scrambling and Barriers, pp. 319–339.Amsterdam: Benjamins.

Stowell, T. (1981).Origins of Phrase Structure. Ph. D. thesis, MIT.

Taraldsen, K. T. (1986a). On verb second and the functional content. In H. Haider andM. Prinzhorn (Eds.),Verb Second Phenomena in Germanic Languages, Dordrecht, pp.7–25. Foris.

Taraldsen, T. (1986b).Somand the binding theory. In L. Hellan and K. K. Christensen(Eds.),Topics in Scandinavian Syntax, Dordrecht, pp. 149–184. Reidel.

Travis, L. (1984).Parameters and Effects of Word Order Variation. Ph. D. thesis, MIT.

Ura, H. (1993). On feature-checking forwh-traces.MIT Working Papers in Linguistics18: Papers on Case and agreement I, 243–280.

Ura, H. (2000).Checking Theory and Grammatical Functions in Universal Grammar.Oxford Studies in Comparative Syntax. Oxford University Press.

Uriagareka, J. (1999). Multiple spell-out. In S. Epstein and N. Hornstein (Eds.),WorkingMinimalism. Cambridge, MA: MIT Press.

van Gelderen, E. (1997).Verbal Agreement and the Grammar behind itsBreakdown.Tübingen: Niemeyer.

Page 128: The Syntax of π: a study of subjects and A-movement ...branigan/papers/branigan-subjects.pdf · The Syntax of π: a study of subjects and A-movement constraints¯ Philip Branigan

BIBLIOGRAPHY 127

Vikner, S. (1991).Verb Movement and the Licensing of NP-Positions in the GermanicLanguages. Ph. D. thesis, Université de Genève.

Watanabe, A. (1993, November). Locative inversion: Where unaccusativity meetsminimality. Draft Ms., University of Tokyo.

Williams, E. (1978). Across-the-board extraction.Linguistic Inquiry 9, 31–42.

Zwart, C. J.-W. (1992). Dutch expletives and small clause predicate raising.NELS 22,477–491.

Zwart, C. J. W. (1993a).Dutch Syntax: A Minimalist Approach. Ph. D. thesis, Universityof Groningen.

Zwart, C. J. W. (1993b). Verb agreement and complementizer agreement.MIT WorkingPapers in Linguistics 18, 297–340.

Zwart, C. J.-W. (1997).Morphosyntax of Verb Movement: A Minimalist Approach to theSyntax of Dutch. Dordrecht: Kluwer.