Super Equi

IDAN LANDAU

CONTROL AND EXTRAPOSITION: THE CASE OF SUPER-EQUI ?

ABSTRACT. Previous analyses of control in Super-Equi have failed to account for theentire paradigm of relevant cases. A new generalization is stated: Obligatory Control (OC)obtains in extraposition only under psychological predicates. It is argued that extrapositionis driven by the requirement that VP-internal clauses be peripheral at PF. This is satisfiedby a CAUSER infinitive which is projected below an EXPERIENCER DP, but not by oneprojected above a THEME/GOAL DP. Thus extraposition is blocked in the former case andlicensed in the latter. Crucially, only when the infinitive is extraposed to an adjunct position(or intraposed to a subject position) can it give rise to Non-Obligatory Control (NOC); thisis supported by a correlation between NOC and failure of extraction from the infinitive. Itis claimed that in OC an Agree relation is established between the matrix functional headthat licenses the controller and an anaphoric infinitival Agr, which raises to the embeddedC as a free-rider on T. Since Agree is sensitive to islands, the distributional distinctionbetween OC and NOC reduces to the CED. Failing syntactic identification, the infinitivalAgr is licensed as a logophor, explaining some well-known properties of NOC in Super-Equi. The proposed account unifies a wide range of phenomena unrelated under alternativeanalyses of control and Super-Equi.

INTRODUCTION

Among the various problems that syntacticians study, some have longerstanding than others. Among those, a few have been around since the veryinception of generative grammar; and among those, yet fewer remained un-solved. I think it would not be an overstatement to say that the phenomenonof control belongs to this last category of very old and very recalcitrantproblems. Despite the wealth of research into this topic, we still lack secureanswers to fundamental questions: What is the theory of control a theoryof ? Is it an autonomous module of grammar or the locus of intersectionbetween other modules? Is control a relation defined on thematic/semanticrepresentations, or syntactic ones? Is it established derivationally or at the? This paper has benefited from the illuminating comments of many people: Hagit

Borer, Noam Chomsky, Sabine Iatridou, Alec Marantz, David Pesetsky, the audiences atNELS 29 and the linguistics colloquia of Ben Gurion University and Tel Aviv University,and three anonymous NLLT reviewers. I am grateful to all those people and take fullresponsibility for any errors that survived to the final version.

Natural Language and Linguistic Theory 19: 109152, 2001. 2001 Kluwer Academic Publishers. Printed in the Netherlands.

110 IDAN LANDAU

interface? What is the relation between control and binding or movement?Can it be reduced to either of the two? What is the status of the emptycategory PRO? Is it superfluous, is it a pronoun or anaphor, or both in different environments? Should obligatory and non-obligatory controlbe subsumed under the same theory? These are just a few of the relevantquestions that arise.

It should be said at the outset that the present paper will probably leavemany of the above questions unanswered. This statement, in itself, alreadytakes a position on some of the issues; yes, control is a multidimensionalphenomenon, hence no account can hope to explain all of its facets at once.Here I will be mostly concerned with the syntax of control, as revealedby its interaction with extraposition (another topic with a rich history ingenerative grammar). The empirical domain will focus on a single con-struction type: The Super-Equi construction. As it turns out, an in-depthinvestigation of this construction sheds interesting light on the nature ofboth control and extraposition. At the very least, novel generalizations willbe formulated that were not captured before.

The Super-Equi construction, first studied (and named) by Grinder(1970), poses a non-trivial challenge to any syntactic account of control the challenge of locality. To appreciate the problem, consider the followingparadigm:

(1)a. Mary knew that it disturbed John [PRO to perjurehimself/herself].

b. Mary knew that it damaged John [PRO to perjurehimself/herself].

c. Mary knew that [PRO perjuring himself/herself] disturbedJohn.

d. Mary knew that [PRO perjuring himself/herself] damaged John.

In each sentence of (1), there are two potential controllers for PRO Johnor Mary the first of which is contained in the clause immediately dom-inating the infinitive, the second of which is higher up. A quick look atthe data reveals that the choice of controller is sensitive to two parameters:The kind of predicate that governs the infinitive/gerund, and the syntacticposition of the latter. Notice that in the structure standardly called extra-position, Mary cannot control PRO in (1a), but can do so in (1b) (thoughlocal control is more natural; see Section 2 for discussion). However, this

CONTROL AND EXTRAPOSITION 111

contrast is neutralized when the nonfinite clause is in subject position, asin (1c,d).

Surprising as it may seem, the paradigm in (1) has never been con-sidered in its entirety; existing accounts of Super-Equi fail to explain atleast one of its members. Thus, some theories are too restrictive, blockingLD-control not only in (1a) but also in (1b) (Grinder 1970; Williams 1980;Koster 1984; Lebeaux 1984; Chierchia and Jacobson 1986; Manzini andRoussou 2000). Other theories are too permissive, allowing LD-control notonly in (1bd) but also in (1a) (Bresnan 1982; Manzini 1983; Bouchard1984; Williams 1992; Hornstein 1999).

Explaining this paradigm is the main goal of this paper, that is, develop-ing an analysis of control that interacts with natural assumptions about theargument structure of different predicates and the syntax of extraposition,in a way which produces the peculiar pattern of locality in (1). By the endof the road, we will obtain a fuller picture of the variety of ways in whichcontrol interfaces with other grammatical processes.

This paper is organized as follows: Section 1.1 states the empiricalchallenge facing any adequate theory of Super-Equi, in terms as neutralas possible: Two generalizations are formulated concerning the interactionof control with the semantic class of the predicate governing the infinit-ive and the infinitives position. Some cross-linguistic evidence, presentedin Section 1.2, suggests that the above generalizations are not limited toEnglish, and may be universal. An analysis is developed in Section 2,incorporating the proposal that extraposition is driven by a PF constraint;LF, however, may interpret the copy at the base position (reconstruction).I further propose a characterization of obligatory control (OC) as a relationholding between an infinitive in situ and a local controller. This descript-ive generalization, yet to be derived, covers all the relevant cases of OC,considerably simplifying earlier statements.

In Section 3, I explore some surprising configurational predictionsthis analysis makes: Correlations between extraction and non-local (i.e.,long-distance or arbitrary) control support the idea that OC is linked toinfinitives in situ and NOC to displaced infinitives (subjects or extraposed).In Section 4, I discuss some apparent problems which turn out, upon closeranalysis, to be harmless to the present proposal. Section 4.1 discussesOC out of inalienably possessed noun phrases and Section 4.2 shows thatapparent OC into sentential subjects is a lexically governed phenomenonorthogonal to control.

Section 5.1 returns to the OC Generalization stated in Section 2 andderives it from a minimalist adaptation of Borers (1989) Anaphoric Agrtheory; the control relation, however, is argued to instantiate the general

112 IDAN LANDAU

operation Agree (Chomsky 1998) rather than Condition A of the bindingtheory. In Section 5.2, I argue that NOC is a case of logophoric anaphora inthe sense of Reinhart and Reuland (1993) that is, an anaphoric Agr thatfails to be syntactically licensed. Alternative theories, which take PRO inNOC to be pronominal, overgenerate structures that are correctly ruled outunder the present proposal.

1. SUPER-EQUI: THE DATA

1.1. The Problem

The first to note and analyse Super-Equi was Grinder (1970), who alsonamed the construction. Grinder observed that in examples like (2), theunderstood subject of the bracketed nonfinite clause is coreferential withan NP (italicized below), which bears no unique syntactic relation to it,and can be indefinitely remote:

(2)a. Eric insisted that it would be ridiculous [to call for help].

b. That [covering themselves with mud] disturbed Spiro amusedDick.

Grinder proposed a rule of Super Equi-NP Deletion which deletes thesubject of the nonfinite clause under identity with a commanding NP.He further suggested to collapse this rule with the local rule of Equi-NPDeletion (Rosenbaum 1967), applying in control of complement clauses.The empirical content of his proposal resides in the characterization ofthe restrictions on the application of the rule. These can be seen in thefollowing paradigm cited by Grinder:

(3)a. John said that making a fool of herself in public disturbed Sue.

b. John said that making a fool of himself in public disturbed Sue.

c. John said that it disturbed Sue to make a fool of herself inpublic.

d. John said that it disturbed Sue to make a fool of himself inpublic.

To facilitate discussion, let us refer to cases where the controller occurs inthe clause immediately dominating the PRO-containing infinitive/gerund


as local control, and cases where it does not long-distance (LD) control.Let us further refer to the nonfinite clause in (3a,b) as in intrapositionand in (3c,d) as in extraposition, taking these terms as strictly descriptivelabels at the moment. Then Grinders facts amount to the observation thatif a local controller is available then it is obligatory in extraposition butoptional in intraposition.1

For nearly thirty years now, this asymmetry has been taken to consti-tute the fundamental empirical challenge facing any theory of Super-Equi.That is, the challenge was to formalize a locality principle that would berestrictive enough to rule out LD-control in extraposition with a potentiallocal controller (3d) but not too stringent to rule it out in extrapositionwithout a potential local controller (2a) or in intraposition ((2b) and (3b)).As mentioned in the introduction, this proved to be a non-trivial challenge,and quite a few theories of control in generative grammar fail to meetit. However, the starting point of the present account is a critique on thegenerality of Grinders own observations.

It turns out that Grinders paradigm is misleadingly partial and that oncethe full paradigm is considered, the line between LD and local controlceases to correspond to the line between intraposition and extraposition.The relevant factor is the semantic class of the predicate governing theinfinitive. The following examples present the full paradigm:

(4)a. Mary knew that it was painful to John [PRO to perjurehimself/herself].

b. Mary knew that it was harmful to John [PRO to perjurehimself/herself].

c. Mary knew that [PRO to perjure himself/herself] would bepainful to John.

d. Mary knew that [PRO to perjure himself/herself] would beharmful to John.

(5)a. Mary thought that it pleased John [PRO to speak his/her mind].1 For some speakers (3d) improves if the matrix tense is modal (would, might, etc.).

Yet for other speakers the example remains bad (as reported by Kuno (1975, ex. 33b),Clements (1975, ex. 5b), and Chierchia Jacobson (1986, ex. 1b). Crosslinguistic data (seeSection 1.2) reflect a similar tendency in some but not all languages; hence, I will assumethat the modal effect is secondary. However, to avoid controversy, I will use only past tensein this kind of example. In Section 5.2 I return to a possible source for this effect.

114 IDAN LANDAU

b. Mary thought that it helped John [PRO to speak his/her mind].

c. Mary thought that [PRO to speak his/her mind] would pleaseJohn.

d. Mary thought that [PRO to speak his/her mind] would helpJohn.

(6)a. Mary thought that it was a relief to John [PRO to take care ofhimself/herself].

b. Mary thought that it was a help to John [PRO to take care ofhimself/herself].

c. Mary thought that [PRO to take care of himself/herself] wouldbe a relief to John.

d. Mary thought that [PRO to take care of himself / herself] wouldbe a help to John.

The property that painful/please/relief have in common but which harm-ful/help/a help lack is that of being psychological. By that we simply meanthat statements like X is painful to Y carry certain psychological entail-ments with respect to the mental state of Y, whereas statements like X isharmful to Y carry no such entailments.

Two comments are in order: The (b) cases above trigger a residualgarden-path effect, giving rise to some difficulty under the LD-controlreading. There is a principled parsing source for this effect, which in factsupports the analysis developed below; I discuss it in Section 3. Crucially,though, speakers easily recover from that garden-path effect, given theappropriate context, whereas LD-control remains fatally bad in the (a)cases. This suggests a real grammatical difference. Secondly, the contrastis obscured (that is, LD-control in the (a) cases improves) if an intonationalpause is inserted before the infinitive; this favors a right-dislocation ana-lysis, which is again expected to affect the status of the sentences, giventhe analysis below.

Bearing these comments in mind, consider the pattern above: As anobject of psychological predicates, the EXPERIENCER John triggers anintervention effect in the (a) examples above, which involve extraposition.No comparable effect is attested in the (c) examples, which involve intra-position. As an object of non-psychological predicates, the GOAL/PATIENT


John triggers no intervention effect either in the extraposition cases (b)or the intraposition ones (d). Notice that the pattern is identical forverbs, adjectives, and nouns in the position of the predicate, hence thegeneralization is category-neutral.

Grinders effect is therefore restricted to infinitives under psycholo-gical predicates. An adequate theory of Super-Equi must account for thefollowing generalizations:2

(7)a. In a structure [. . . X . . . [it Aux Pred Y [SPRO to VP]]], whereY and S are arguments of Pred:(i) If Pred is psychological, Y must control PRO.(ii) If Pred is non-psychological, either X or Y may control

PRO.

b. In a structure [. . . X . . . [S [S PRO to VP] Pred . . . Y]], either Xor Y may control PRO.

(7) amounts to the claim that LD-control is susceptible to intervention onlyin extraposition, and only when the intervener bears an EXPERIENCER -role.

1.2. Crosslinguistic ConfirmationThe generalizations in (7) refer to properties of predicates (psychologicalvs. non-psychological) and structure (extraposition vs. intraposition) thatare attested universally. The strongest (and null) hypothesis would be thatthese generalizations are not language-specific; the English facts in (4)(6)should represent a universal pattern.

This expectation seems to be on the right track, at least with respectto the five additional languages that we have checked. German, Italian,Spanish, Swedish, and Hebrew all conform to the generalizations in (7):

German (Cornelia Krause, Susi Wurmbrand, p.c.).8/a. Mary1

Maryistis

sicher,sure

dathat

es

itdemthe.DAT

HansHans

nichtnot

schadendamage

wrdewould

seinerhis.DAT

Freundingirlfriend.DAT

ihr1her.ACC

Herzheart.ACC

auszuschtten.out-to-throw

Mary is sure that it would not damage John to reveal her heartto his girlfriend.

2 I presently disregard further conditions on X as an LD-controller in (7). As notedabove, these largely reduce to logophoricity (see Section 5.2).

116 IDAN LANDAU

b. Mary1Mary

istis

sicher,sure

dathat

es

itdenthe.ACC

HansHans

nichtnot

strenannoy

wrdewould

seinerhis.DAT


ihr1her.ACC

Herzheart.ACC

auszuschtten.out-to-throw

(Mary is sure that it would not annoy John to reveal her heart tohis girlfriend.)

c. Mary1Mary

istis

sicher,sure,

dathat

seinerhis.DAT


ihr1her.ACC

Herzheart.ACC

auszuschttenout-to-throw

demthe.DAT

HansHans

nichtnot

schadendamage

wrde.would

Mary is sure that to reveal her heart to his girlfriend would notdamage John.

d. Mary1Mary

istis

sicher,sure,

dathat

seinerhis.DAT


ihr1her.ACC

Herzheart.ACC

auszuschttenout-to-throw

denthe.ACC

HansHans

nichtnot

strendamage

wrde.would

Mary is sure that to reveal her heart to his girlfriend would notannoy John.

Italian (Sveva Besana, p.c.)

.9/a. MariaMary

pensathinks

chethat

a Giannito-John

piacerebbewould-please

parlareto-talk

diof

se stesso/himself/

se stessa.herselfMary thinks that it would please John to talk abouthimself/herself.

b. MariaMary

pensathinks

chethat

aiuterebbewould-help

GianniJohn

parlareto-talk

diof

se stesso/himself/

se stessa.herselfMary thinks that it would help John to talk about him-self/herself.


c. MariaMary

pensathinks

chethat

parlareto-talk

diabout

se stessaherself

aiuterebbe/would-help/

piacerebbeplease

a

to

Gianni.John

Mary thinks that to talk about herself would help/please John.

Spanish (Karlos Arregui-Urbina)

.10/a. Ato

Juan1Juan

leCl

entristecerawould-sadden

contarle1to-tell-Cl

lasthe

malasbad

noticias.news

(It would sadden Juan1 to tell him1 the bad news.)

b. ?Ato

Juan1Juan

leCl

perjudicarawould-harm

contarle1to-tell-Cl

lasthe

malasbad

noticias.new

It would harm Juan1 to tell him1 the bad news.

c. Contarle1to-tell-Cl

lasthe

malasbad

noticiasnews

leCl

entristecera/perjudicarawould-sadden/harm

a

to

Juan1.JuanTo tell him1 the bad news would sadden/harm Juan1.

Swedish (Anders Holmberg, p.c.)

.11/a. Barnenthe-children

frstodunderstood

attthat

detit

hjlpte/gladdehelped/pleased

JohnJohn

attto

vara

besnlla.nice (pl.)The children understood that it helped/pleased John to be nice(pl.).

b. Barnenthe-children

frstodunderstood

attthat

a

to

vara

besnllenice (pl.)

gladdepleased

JohnJohn

The children understood that to be nice (pl.) pleased John.

118 IDAN LANDAU

Hebrew

.12/a. RinaRina

amra

saide-zethat-it

siyeahelped

le-/to-/

icbenannoyed

etacc.

Gil1Gil

lelamedto-teach

oto1him

nimusim.manners

Rina said that it helped/annoyed Gil1 to teach him1 manners.

b. RinaRina

amra

saide-lelamed etthat-to-teach

Gil1Gil

nimusimmanners

yesayeawould-help

lo1/to-him/

yeacbenwould-annoy

oto1.him.acc

Rina said that to teach Gil1 manners would help/annoy him1.

The theoretical problem we now face is the following: How does the con-figurational sensitivity of control interact with the thematic structure ofthe governing predicate and the position of the infinitive to produce thegeneralizations in (7)? The following section addresses this problem.

2. AN ANALYSIS

The analysis I propose incorporates assumptions about Obligatory Control,extraposition, the interpretation of chains, and the projection of argumentstructure. Let us start with the first component:

(13) The OC GeneralizationIn a configuration [. . . DP1 . . . Pred . . . [SPRO1 . . . ] . . . ],where DP controls PRO: If at LF, S occupies a comple-ment/specifier position in the VP-shell of Pred, then DP (or itstrace) also occupies a complement/specifier position in that VP-shell.

Some distinction between OC and NOC is necessary in any theory of con-trol. Thus, Bresnan (1982) attributes the distributional distinction betweenOC-clauses and NOC-clauses to the distinction between open and closedgrammatical functions (although the latter is itself stipulated, not derived).Manzini (1983) reduces it to the distinction between governed and un-governed positions. It is in fact not clear that anything more complicatedthan the simple containment condition in (13) is necessary; at least with


respect to the present discussion, (13) is a sufficiently accurate descriptivegeneralization.

Notice that (13) makes no claims as to the particular choice of con-troller within the domain of OC; rather, it merely fixes the domain withinwhich such a controller must be found. Under this view, the choice betweensubject and object control is a semantic/pragmatic one. This approach toOC is in line with Chomsky (1981), Manzini (1983), Koster (1984), Com-rie (1984), Melvold (1985), Farkas (1988), Sag and Pollard (1991) andothers, as opposed to the MDP tradition of Rosenbaum (1967), Larson(1991), Hornstein (1999) and Manzini and Roussou (2000). The phe-nomenon of control shift (e.g., John was promised to be allowed toleave, Tim1 asked his mother2 to join him1/her2) provides ample supportfor the former view and against the latter one (for extensive discussion, seeWurmbrand 1998; Petter 1998; Landau 1999).

(13) makes immediate predictions as to the classification of controlledclauses in three types of positions: VP-complement/specifier, extraposi-tion, and intraposition. An infinitival in a complement or specifier positionof a predicate X satisfies (13) with respect to XP; hence, it must be locallycontrolled, that is, within the minimal VP-shell containing XP. By con-trast, a preposed or extraposed infinitive is not contained in the maximalprojection of the predicate of which it is an argument; therefore, it willconstitute an instance of NOC, allowing either arbitrary or LD-control.All of these predictions are borne out, as will be shown presently.3 Noticethat the position of the infinitive relevant for containment in (13) is itsLF position, or more precisely, its position at the stage where control isestablished. This point will have immediate consequences for infinitivalsin extraposition.

3 There is a line of research maintaining that NOC, and specifically arbitrary PRO,always involves control by implicit (quantificational) arguments (Epstein 1984; Koster1984; Roeper 1987), or a default option, available only in the absence of a local NP(Williams 1980; Lebeaux 1984; Manzini and Roussou 2000). The present proposal, bycontrast, exclusively attributes the possibility of NOC to the syntactic position of the in-finitive. The following examples, where arbitrary PRO co-occurs with a local NP in thematrix clause, argue against the former views and in favor of the latter:

(i) It would help Bill [PRQarb to behave oneself in public]. (Manzini l983)

(ii) It is dangerous for babies [PROarb to smoke around them]. (Kawasaki 1993)

(iii) It is crucial for the success of this conference [PROarb to limit oneself to 20minutes].

120 IDAN LANDAU

Next, I assume that extraposition is driven by the following motivation:

(14) ExtrapositionVP-internal clauses must be peripheral at PF.

(14) corresponds to the cross-linguistic observation that embedded clausesare typically peripheral to the VP and seldom intervene between apredicate and other internal arguments. Where this does not obtain inbase-structure (because of thematic mapping), the embedded clause mustextrapose. I assume that extraposition in this case is adjunction to VP (dir-ectionality subject to parameterization); some configurational evidence forthis is given below. I also adopt the standard assumption that dominatedby XP means dominated by every segment of XP; thus, an adjunct toVP is not dominated by VP.

The third component concerns the interpretation of chains:

(15) Chain InterpretationAny link in a chain may be the LF-visible link.

(15) is the null hypothesis under the view that traces of movement arefull copies of the element moved (Chomsky 1995). Reconstruction effectsthat track the path of movement suggest that, in principle, any link alongthat path can be the one feeding the interpretive component (Fox 1999;Bobaljik 1999). This is uncontroversial for A-bar movement, the status ofA-reconstruction being more contentious.

The relevant consequence for us is the following: We assume that extra-position is genuine movement, and furthermore, an adjunction operation.As such it falls under the A-bar rubric. LF may then choose to interpreteither the clausal copy at the base position or the PF-visible copy at theextraposed position. Since the base position is VP-internal and the extra-posed one is VP-external, the former option will result in OC and the latterin NOC, assuming (13).

Lastly, we mention (fairly standard) assumptions about the projectionof argument structure:

(16) Argument Projectiona. EXPERIENCER is generated above CAUSER.

b. CAUSER is generated above GOAL/PATIENT/THEME.

(16a) has been invoked to explain several unaccusative properties of psych-verbs (Belletti and Rizzi 1988), as well as certain scope reversal properties


they display (Pesetsky 1995, Stroik 1996). (16b) is a standard assump-tion that needs no further comment. Notice that the two clauses of (16)never apply jointly; no predicate selects EXPERIENCER, CAUSER, andGOAL/PATIENT at the same time. In what follows I will assume that them-atic specifications such as (16) are neutral with respect to the categorialrealization of arguments. In particular, the CAUSER argument may be eitherDP, IP or CP, all of which are generated in accordance with (16).4

Armed with (13)(16), we can explain all the relevant facts. Considerfirst the structures for extraposition with psychological predicates (theindex 1 indicates local-control; 2 indicates non-local control):

(17)a. It would please John1 [S PRO1 . . . ]

4 Cinque (1990, p. 167, fn.5) argues that CAUSER infinitives are generated above theEXPERIENCER DP and then are extraposed to the right, since case assignment to the DPdemands adjacency to the verb. This does not explain how the same word order arises whenthe EXPERIENCER is inside a dative/oblique PP, which needs no case from the verb.

122 IDAN LANDAU

b. It would please John1 [S PRO2 . . . ]

In (17a) the infinitive S is in situ, below the direct object (by (16a)).By (13), the direct object is an obligatory controller. S is already VP-peripheral, hence extraposition, as in (17b), is unmotivated. The un-grammaticality of (17b), with non-local control, follows from economyconsiderations.

Next, consider extraposition with non-psychological predicates:5

5 I am abstracting away from the small v analysis of causative structures or any otherfunctional verbs inside the extended projection.


(18)a. It would help Johh1 [S PRO1 . . . ]

OC: Infinitive interpreted in situ

b. It would help John1 [S PRO1=2 . . . ]

NOC: Infinitive interpreted extraposed

124 IDAN LANDAU

The infinitive S is generated above the direct object (by (16b)), hence notperipheral to VP. Extraposition is forced, by (14), creating a chain in whicheach link may be interpreted, by (15). If the base position is interpreted,as in (18a), S is VP-internal at LF, and the direct object is an obligatorycontroller by (13). If the extraposed position is interpreted, as in (18b), Sis VP-external at LF, failing (13) and becoming an instance of NOC; hencethe possibility (though not necessity) of non-local control.6

Finally, consider the case of intraposition, with both types of predicates:

(19)a. [S PRO1=2 . . . ] would please John1

6 While OC forces a local controller, NOC allows an LD-controller; thus, NOC sub-sumes OC in the range of possible controllers. There is in fact subtle evidence for thecompatibility of local control with syntactic extraposition, as I show in Section 3.


b. [S PRO1=2 . . . ] would help John1

With either predicate, S moves to [Spec,IP] to satisfy EPP.7 This is aninstance of A-movement, and we may assume (following Chomsky 1995)that unless special circumstances demand otherwise, it is the higher link ofthe chain that gets interpreted. Again, the infinitive escapes the domainof OC, this time regardless of the semantic class of the matrix predic-ate. Notice that intraposition removes the infinitive from VP, making (14) the peripherality condition on VP-internal clauses inapplicable. Al-ternatively, we may rephrase (14) to require that subordinate clauses beperipheral to either VP or IP at PF. The former option will be taken whenthe numeration contains the expletive it, the latter otherwise.

Although A-movement normally does not reconstruct, it is well knownthat it can reconstruct to allow bound variable readings. This yields an in-teresting interaction between control and variable binding in intraposition:

(20)a.??I think that [PRO talking about him1 on TV] would please everycelebrity.

b. I think that [PRO talking about his1 work on TV] would pleaseevery celebrity1.

In (20a) the gerund must reconstruct to allow variable binding. It is there-fore interpreted VP-internally, and OC by every celebrity is forced (due to

7 It has been suggested by various authors that the accusative case on experiencer ob-jects is inherent (e.g., Belletti and Rizzi 1988; Arad 1998). If this is true, then raising tosubject over that position violates no minimality conditions, since the experiencer object isfrozen (unattractable by Tense).

126 IDAN LANDAU

(13)). The result violates Condition B. In (20b), where Condition B is notat stake, OC by the QP is compatible with the bound variable reading. No-tice that (20a) is grammatical under a non-bound-variable reading of him.Under that reading, PRO may (but need not) be controlled by the local QP,suggesting that local control in itself is not diagnostic of reconstruction;it is only when the infinitive is interpreted in its base position that localcontrol is a consequence of the OC Generalization (13). We return belowto additional evidence for the partial overlap between OC and NOC.8

The present account predicts that if extraposition is motivated forsome reason with psychological predicates, they too should allow LD-control. This should arise in a context where the infinitive is separatedfrom the edge of VP by other material, despite being the lowest argu-ment. The relevant examples are hard to construct since normal adverbialscan occupy specifier positions in a Larsonian shell and are thus notclear VP-edge markers. However, if some other element (e.g., a relativeclause) is extraposed to the left of the infinitive, the latter must be also inextraposition:

(21)a. Mary1 thought that it pleased John [PRO1 to talk about herself].

b. ?Mary1 thought that it pleased the man greatly who hired her forthe job [PRO1 to talk about herself].

Admittedly awkward, (21b) nevertheless displays a weaker resistance toLD-control than (21a). This contrast follows on the assumption that LD-control is contingent on extraposition.

We thus account for the full Super-Equi paradigm. Notice that most ofthe ingredients of the analysis are motivated independently of the data un-der consideration; this is true of the assumptions on reconstruction (15) andargument structure (16), as well as on the role of economy in constrainingmovement. The OC Generalization (13) will be derived in Section 5 fromindependent principles governing the application of the syntactic operationAgree. This leaves the PF-constraint (14) as the only special assumptioninvoked. As noted, this constraint seems descriptively adequate, and onemay speculate that it is to be ultimately explained by principles of clausalprosody.

Finally, a note on parsing. It was already observed that LD-control isharder to get when a local controller is available. The same difficulty is

8 I thank an anonymous reviewer for clarifying this issue and leading me to theprediction in (20).


found with arbitrary control, which, under the present analysis, is also avariant of NOC:

(22)a. It [VP helped John1 [PRO1 to buy himself1 a new computer]].

b. It [[VP helped John1] [PROarb to buy him1 new computer]].

Given the structural difference between the two examples, the garden-path effect in (22b) comes as no surprise. It is well known that real timeprocessing is guided by certain parsing strategies, of which Branch Right(Phillips 1996) is a major component. Consider now the effect of thisstrategy on the analysis a hearer assigns to the substring It helped Johnto buy . . . . Given no evidence to the contrary, the parser attaches the infin-itive as a right-branching sister to the lowest V-projection, in compliancewith Branch Right. But this immediately determines local control, for theintermediate structure falls under OC, by (13). It is only when him is heardin (22b) that the parser is forced to backtrack (to avoid a Condition Bviolation) and re-attach the infinitive outside VP. Seen that way, the pres-ence of the garden-path effect in these cases in fact supports the structuraldistinction we have drawn between OC and NOC.

3. CONFIGURATIONAL CONSEQUENCES

Striking confirmation for the proposed analysis comes from the interactionof extraction and c-command effects with LD-control. It is a theorem of theset of assumptions in (13)(16) that whenever the infinitive is (syntactic-ally) in situ, control is local; and whenever control is non-local (arbitraryor long-distance), the infinitive is extraposed to a position adjoined to VP.Assuming that adjuncts are islands to extraction, we expect a correlationbetween non-local control and failure of extraction.

A word of caution is needed before we turn to test this prediction. Itis well known that infinitives are inherently very weak islands; indeed,arguments can be extracted from rationale clauses and NP-complements both island environments with no severe violation:

(23)a. ?What did John go home to look for?

b. Some people, I have no desire to meet.

Although in some positions infinitives constitute strong islands (e.g.,sentential subjects), it is not obvious that the extraposition cases under dis-cussion pattern with subjects rather than with the cases in (23). In certain

128 IDAN LANDAU

cases it is possible to highlight the contrast by extraction of adjuncts, whichdiagnoses islands more sharply. In what follows, I will focus on contrastivejudgments, showing that in each case, extraction out of a locally con-trolled infinitive is significantly better than extraction out of an ARB-/LD-controlled one.

The following paradigm confirms our prediction with respect to arbit-rary control:

(24)a. I would help Bill1 [PRO1 to introduce himself to these profess-ors].

b. To whom2 would it help Bill1 [PRO1 to introduce himself t2?]]

(25)a. It would help Bill1 [PROarb to introduce him1 to these profess-ors].

b. To whom2 would it help Bill1 [PROarb to introduce him1 t2 ?]

The examples in (24) involve local control, and those in (25) involve arbit-rary control. Notice that the latter is available in principle as the predicatehelp is non-psychological. We observe that extraction is licit from a locallycontrolled infinitive (24b) but not from an ARB-controlled one (25b). Inboth cases the infinitive is phonologically extraposed; local control in (24b)indicates that the copy feeding extraction may be in the base position,whereas non-local control in (25b) requires it to be in an adjunct position.9

The following paradigm demonstrates the same facts with LD-control:

(26)a. Hillary thinks it damaged Bill1 [PRO1 to talk about himself onthe Dave Letterman show].

b. Thats the talk show2 that Hillary thinks it damaged Bill1 [PRO1to talk about himself on t2].

9 An anonymous reviewer notes that since the CAUSER argument is in [Spec,VP], ex-traction from the infinitive in (24b) violates the Left Branch Condition. However, specifiersin Larsonian shells are not subject to the LBC:

(i) Thats the artist that Trump bought every picture1 of from its1 owner.

Variable binding in (i) ensures that the from-PP is lower than the direct object, which mustthen be in [Spec,VP]; yet extraction is possible. Similarly, subjects that remain in VP (inlanguages like Italian or German) are transparent to extraction (Diesing 1992).


(27)a. Hillary1 thinks it damaged Bill [PRO1 to talk about herself onthe Dave Letterman show].

b. Thats the talk show2 that Hillary1 thinks it damaged Bill [PRO1to talk about herself on t2].

The asymmetry is perhaps sharper in cases where, for pragmatic reasons,no ambiguity is possible, not even one that is resolved later on. Suchcases arise when the argument preceding the infinitive is not a potentialcontroller:

(28)a. It would kill the workers1 [PRO1 to build this dam].

b. What2 would it kill the workers1 [PRO1 to build t2?]

c. It would kill the forest [PROarb to build this dam].

d. What2 would it kill the forest [PROarb to build t2?]

(29)a. It would benefit John1 [PRO1 to explain the program to thedean].

b. Who2 would it benefit John1 [PRO1 to explain the program tot2?]

c. It would benefit the program1 [PROarb to explain it1 to thedean].

d. Who2 would it benefit the program1 [PROarb to explain it1 tot2?]

(30)a. It would delay the negotiators [PRO1 to insist on technicalissues].

b. Thats the issue2 that it would delay the negotiators [PRO1 toinsist on t2].

c. It would delay the negotiations [PROarb to insist on technicalissues].

130 IDAN LANDAU

d. Thats the issue2 that it would delay the negotiations [PROarb toinsist on t2].

Adjunct extraction also brings out a clear contrast (in the following ex-amples, matrix readings of the extracted adjunct, if possible at all, shouldbe ignored):

(31)a. It would be useful to Bill1 [PRO1 to talk about himself1 moremodestly].

b. How2 would it be useful to Bill1 [PRO1 to talk about himself1t2?]

c. It would be useful to Bill1 [PROarb to talk to him1 more gently].

d. How2 would it be useful to Bill1 [PROarb to talk to him1 t2]?

(32)a. It would change Oscar1 completely [PRO1 to paint himself1green].

b. What color2 would it change Oscar1 completely [PRO1 to painthimself1 t2?]

c. It would change Oscar1 completely [PROarb to paint him1green].

d. What color2 would it change Oscar1 completely [PROarb topaint him1 t2?]

The extraction data strongly support the structural analysis proposed inSection 2. Although string identical, locally and non-locally controlledinfinitives occupy different positions in the syntax: The former are VP-internal, the latter are VP-external. This follows from the operation of OCwithin VP, and the complementary operation of NOC elsewhere. Control,which is sensitive to syntactic configurations, sees this contrast; pronun-ciation, which is sensitive to phonological configurations, does not see it,since the extraposed copy is the one sent off to PF in both cases.

It is interesting to note that a very similar correlation between con-trol and extraction was noted by Chomsky (1986b) in the context oftough-constructions. Chomsky points out that sentences like (33a) are am-biguous, depending on whether or not they control PRO (whether, say, they


are the organizers or the crowds). The ambiguity, however, is resolved infavor of local control if extraction takes place from the infinitive (33b):

(33)a. They were too angry [PRO to hold the meeting].

b. Which meeting were they too angry to hold?

c. [they were [too angry [CP PRO to hold the meeting]]]

d. [IP [they were too angry] [CP PRO to hold the meeting]]

Chomsky suggests that the two readings of (33a) correspond to the twodifferent structures (33c,d): Where they c-commands PRO it controls it(33c), and where it does not c-command PRO, we get arbitrary control(33d). Crucially, the infinitive in the latter is ungoverned, hence a barrierfor extraction; therefore, only (33c) is a possible source for (33b), and wederive the desired correlation. Notice that modulo the notion of govern-ment which is absent from the present proposal, the logic of the argument isthe same: A VP-external infinitive is the common source for both arbitrarycontrol and islandhood.

The island effects suggest that a non-locally-controlled infinitive is anadjunct; however, they leave open the question of the exact adjunctionsite. The constraint in (14) requires adjunction to VP; we now turn tohierarchical evidence to this effect.

Given standard conceptions of English clause structure, NEG c-commands VP-adjuncts but not IP-adjuncts, while direct objects c-command neither. The following pattern of NPI-licensing shows that bothinfinitives in situ and in extraposition are within the scope of NEG, butonly the former are in the scope of a direct object:

(34)a. It wouldnt help Bill1 [PRO1 to praise himself at all].

b. It wouldnt help Bill1 [PROarb to praise him1 at all].

(35)a. It would be useful to no one1 [PRO1 to ever say such things inpublic].

b.??It would be useful to no one1 [PROarb to ever tell him1 suchthings in public].

These contrasts indicate that the extraposed position is in the scope of NEGbut not of the direct object (namely, VP). The position of a VP-adjunct

132 IDAN LANDAU

satisfies both conditions. Furthermore, Condition C tests show that a directobject does not c-command an ARB-controlled infinitive (36a),10 but doesc-command it if extraction takes place (36b):

(36)a. It would help him1 [PROarb to introduce Bill1 to theseprofessors].

b. To whom2 would it help him1 [PROarb to introduce Bill1 t2?]

This follows again on the assumption that extraction is only possible froma VP-internal position, placing the infinitive within the control domain ofthe direct object.

Finally, observe the following Weak Crossover effect:

(37)a. It would ruin this tune1 [PROarb to play it1 on a flute].

b.??What tune1 would it ruin t1 [PROarb to play it1 on a flute]?

c. What tune1 would it ruin t1 [PROarb to play e1 on a flute]?

Since the matrix object does not c-command the infinitive when arbitrarycontrol obtains, the configuration in (37b) violates Weak Crossover. Pre-dictably, the sentence becomes acceptable if the pronoun is replaced by anempty category, as in (37c); this involves a parasitic gap strategy, which islicensed precisely because the anti c-command condition is satisfied.

There is some redundancy between OC and NOC, as characterizedabove. To recall, the criterion distinguishing the two categories was notthe locality of the controller, but rather the (syntactic) position of theinfinitive: VP-internal infinitives fall under OC and all the others underNOC. Nothing in our system rules out choice of a local controller for anextraposed infinitive. One might wonder whether there is actual evidencefor this situation, which should be properly called local NOC. This wouldindicate that the above redundancy is in fact an advantage of the presentproposal.

There is an obvious empirical difficulty in finding cases of local NOC:The configurational diagnostics test for c-command between the control-ler and some phrase embedded inside the infinitive; however, if control islocal, the controller is coindexed with PRO, which c-commands everythinginside its clause, masking the hierarchical relation between the controllerand that clause. To overcome this difficulty, one needs to separate the c-command test from the controller; in other words, one should look for

10 It is easier to get coreference in (36a) when the pronoun him is destressed.


structures where the matrix VP contains two NPs, one of which serves asa controller and the other as an anchor for the c-command test.

Examples that satisfy these conditions are not too common; however,the following case plausibly illustrates our point:

(38) It will remind Sue1 of him2 [PRO1 to read Richards2 poems toher daughters].

In this case control is local, yet the infinitive is interpreted in extraposi-tion, as witnessed by the lack of Condition C violation between the NPsindexed 2. This follows if local control is compatible with syntactic ex-traposition: Obviously, local control does not require the LF-copy of theinfinitive to be the extraposed one, but may exploit this option if inter-preting the base copy violates some independent principle (like ConditionC).11

We further predict that if syntactic extraposition is followed by ex-traction, the example in (38) will yield an ungrammatical sentence. Thisprediction is borne out:

(39)a.??[Which of her daughters]3 will it remind Sue1 of him2 [PRO1 toread Richards2 poems to t3]?

b. [Which of her daughters]3 will it remind Sue1 of Richard2[PRO1 to read his2 poems to t3]?

In (39a) extraction requires the infinitive to be in situ, but to escape aCondition C violation it must extrapose; these conflicting demands resultin ungrammaticality. (39b) shows that when Condition C is not at stake,syntactic extraposition is not forced and extraction is possible from thebase copy. The judgments are somewhat subtle, yet the contrast appears tobe real.

11 In accordance with (16b), I assume that the infinitive (CAUSER) in (38) is generatedabove the of-object (THEME), hence subject to extraposition.

134 IDAN LANDAU

4. SOME PROBLEMS AND SOLUTIONS

4.1. Logophoric Extension/Inalienable Possession

There are cases noted in the literature where the controller is not a directargument of the matrix predicate, but rather embedded in that argument.Consider the following examples:

(40)a. It would help Bills1 development [PRO1 to behave himself inpublic].

(Manzini 1983, ex. 42)

b. PRO1 finishing his work on time is important to Johns1development.

c. PRO1 finishing his work on time is important to Johns friends1.(Chomsky 1981, ch. 2, ex. 19)

(41)a. It would help Bills1 friends [PRO1 to behave himself in public].

b. It would help Bills1 confidence [PRO1 to plan his itinerary inadvance].

c. It would help Bills1 car [PRO1 to plan his itinerary in advance].

d. [PRO1 causing an uproar] is important for Johns1 career.

Manzini (1983) cites (40a) to show that extraposition falls under NOC,where the choice of controller is free of structural constraints. This view,however, fails to account for Grinders intervention effect. Furthermore,notice that the minimally different (41a) is ungrammatical, showing thatextraposition does not generally allow free controller choice. Regardingthe contrast in (40b,c), Chomsky (1981, p. 77) comments: While PROmay have a non c-commanding antecedent, the latter may not be containedwithin an NP that is a possible controller. This formulation needs some re-finement in view of (41b,c): Although the intended interpretation of (41c)is perfectly plausible (Bills planning his itinerary in advance would helphis car), control by Bill fails here. Yet the car is not a possible controllerof PRO in this context, for semantic reasons. Something must distinguishbetween Bills car and Bills confidence, to allow control from within thelatter but not the former. Finally, in (41d) (N. Chomsky, p.c.), John can


control PRO even though it is embedded in an NP which is a possiblecontroller (careers can cause uproars).

The class of nouns that may contain the controller of PRO in theseconfigurations is in fact quite small and coherent. Its members all denoteabstract notions that reflect the individuality of the controller, via actions,character traits, or social attributes. A number of such nouns are listedbelow:

(42) career, status, confidence, performance, development, image,reputation, behavior . . .

Let us call Xs NP, where X denotes an individual and NP is a member of(42), a logophoric extension of X. Then we may capture the transparencyof such nouns to control via the following convention:

(43) For the purpose of control, a logophoric extension [Xs NP] isnon-distinct from X: [Xs1 NP]) [Xs NP]1.

(43) expresses the intuition that a logophoric extension of an individual-denoting noun does not introduce a new individual in the discourse. Thus,there is a clear sense in which Bill and Bills car denote two distinct in-dividuals in a given discourse domain, however Bills development merelyextends or rather focuses attention on some aspect of the denotation ofBill alone. Admittedly sketchy, this distinction suffices for the presentpurposes.

It is important to note that the cases under discussion are genuine in-stances of OC, patterning with normal cases of OC in extraposition andnot with NOC in intraposition. Thus, they license extraction:

a. It is crucial for Bills1 success [PRO1 to teach himself Japan-ese].

b. Which language is it crucial for Bills1 success [PRO1 to teachhimself]?

Given the preceding discussion, this suggests that the interpreted copy ofthe infinitive in (44) is the base copy, and control is governed by the OCgeneralization (13), via the convention (43).

Looking at the sample in (42), it is tempting to assimilate the class oflogophoric extensions to the class of inalienably possessed nouns. Notice

136 IDAN LANDAU

that many members of the former class denote inalienable attributes. In-deed, both abstract and concrete inalienably possessed nouns constitutelogophoric extensions for control:

(45) It would ruin Steves figure/career [PRO to eat so much ice-cream].

Grammatically, they display distributional parallels:

(46)a. Johns hand, the hand of John

b. Johns confidence/career, the confidence/career of John

If these observations are on the right track, then the rationale of (43) isstraightforward: An inalienably possessed noun is referentially dependenton its possessor, hence inherits (or does not block) its index. Under theseassumptions, one can maintain the condition that the controller in OCoccupies a complement/specifier position in the VP-shell of the controlpredicate.

4.2. Apparent OC in Intraposition

Occasionally it has been claimed that (7b) is incorrect in that some cases ofintraposition resist LD-control just as their extraposition counterparts do.Chierchia and Jacobson (1986) cite (47a) and Brody and Manzini (1987)cite (47b) (where only Peter is a possible controller) as cases of OC:

(47)a. John told me that [making a fool of herself/himself] was rudeof Mary.

b. Mary thought that [PRO to behave well] would be easy forPeter.

Chierchia and Jacobson (1986) call such cases semi-obligatory control,which simply refers to OC where the controller may remain implicit.In contrast, they classify the classical intervention cases, e.g., (3d), asNOC, which displays locality effects because the controller minimallyf-commands PRO.

This way of cutting the pie is exactly the opposite of the way advocatedabove. I have argued that cases like (3d) are no different than any othervariety of OC (i.e., complement control), rather than a case of NOC towhich a designated locality condition should be tailored. More importantly,


the minimal f-command condition is too strong, as it incorrectly blocksLD-control across PATIENT/GOAL interveners (e.g., (4b)). Still, the ques-tion remains: Even if Chierchia and Jacobson are wrong about the statusof (3d), the failure of LD-control in (47) looks very much like a propertyof OC. Does that not undermine the present analysis?

I would like to suggest instead that the apparent OC in (47) is reallyquite different from genuine OC. In fact, I would argue that (47) does nottell us anything about the theory of control, but rather something about thesemantics of a special class of predicates.

Notice that the facts in (47) remain unchanged even when the null sub-ject of the infinitival is replaced by a overt pronoun, or indeed, when theinfinitive is replaced by a finite clause:

(48)a. John told me that [him/his making a fool of himself] was rudeof Mary.

b. Mary thought that [for her to behave well] would be easy forPeter.

c. That John made a fool of himself was rude of/easy for Mary.

This is by no means a general property of control in intraposition, evenwith psychological predicates, as demonstrated by (49):

(49)a. John told me that [him/his making a fool of himself] disturbedMary.

b. Mary thought that [for her to behave well] would be surprisingfor Peter.

It therefore seems that predicates like easy and rude of are lexically restric-ted in a way that reflexively links their two arguments; e.g., for an eventx to be rude of a person y, y has to be the agent of x. Agenthood seems tobe necessary; cf. Weighing 350 pounds was rude of Mary. For Chierchiaand Jacobson, OC in (47) is achieved like any other kind of OC in theirtheory namely, through a lexical entailment associated with rude of. Thisentailment is based on the assumption that the infinitive selected by rude ofunambiguously denotes a property. However, (48) shows that the relevantentailment holds even when rude of selects a propositional infinitive, henceit cannot be a consequence of the predicational analysis of OC.

138 IDAN LANDAU

That the phenomenon at hand is orthogonal to control is already sug-gested by (48), where no control is involved. Further evidence is providedby the following contrast:

(50)a. Those comments by John were rude of him.

b. Those comments by John were rude of Mary.

Notice that one cannot assume a PRO NP-subject here, for then (50a)would violate Condition C. I therefore agree with both Chierchia andJacobson (1986) and Brody and Manzini (1987) that there is somethingspecial about predicates like easy and rude of, setting them apart frompredicates like disturb and useful to; yet I maintain that the special featureis not to be encoded in the theory of control, as it shows up in non-controlcontexts as well. The distinction drawn in (7) between extraposition andintraposition remains valid; apparent cases of OC in intraposition in-volve predicates that are intrinsically reflexive in a way that rules outLD-control on interpretability grounds.

5. DERIVING THE DISTRIBUTION OF OC AND NOC

The present analysis heavily relies on a new generalization, given in (13)and repeated below, concerning the distribution of obligatory control:

(51) The OC GeneralizationIn a configuration [. . . DP1 . . . Pred . . . [S PRO1 . . . ] . . . ],where DP controls PRO:If, at LF, S occupies a complement/specifier position in theVP-shell of Pred, then DP (or its trace) also occupies a com-plement/specifier position in that VP-shell.

What (51) amounts to is the claim that an infinitive in situ must find aclausemate controller, whereas a displaced infinitive is not similarly re-stricted, where the relevant positions are those occupied at LF. This viewof OC is distinguished from various alternatives in being fairly theory-independent, appealing to no problematic notions (open grammaticalfunction, government, etc.). It was argued above that it is not only con-ceptually simpler but also empirically adequate to handle all the propertiesof Super-Equi (and, of course, standard complement control). Nonetheless,(51) is merely a generalization, and one wants to know whether it can bederived from deeper principles of the grammar.


The task of this section is to derive the OC generalization in this sense;that is, I will show that (51) can be deduced from the theory of Agree, out-lined in Chomsky (1998), coupled with ideas developed in Borer (1989).We will then see that the category of NOC corresponds to a remarkable de-gree to the category of logophors in the sense of Reinhart Reuland (1993);that is, anaphors that fail to be syntactically licensed.12

5.1. OC and AgreeI propose that OC is an instantiation of the general operation Agree ofChomsky (1998). Agree is the classic transformational dependency, minusphonological displacement: It is a relation between two lexical items,triggered by uninterpretable features. In particular, Agree(, ) is estab-

12 One construction that I will not discuss is control into gerundive temporal adjuncts.There are complicated factors involved in these constructions, including logophoricity,[+human] restrictions and topic/comment sensitivity; there is also a considerable amountof speaker variation concerning the status of various examples (see Williams (1992) andespecially Kawasaki (1993) for discussion).

Manzini (1986) and Kawasaki (1993) note that an implicit agent of a passive can controlinto a temporal adjunct only if it is human, and only if the surface subject is nonhuman:

(i) The rain washed the stairs before entering the basement.

(ii) The stairs were washed before entering the basement.

(iii) Mary was washed before entering the basement.

Those speakers who accept (ii) construe the implicit agent controller as human, eventhough nothing about the pragmatics of the situation requires that (cf. (i)). Still, a humansubject, as in (iii), overrides the implicit agent and must be the controller itself. Kawasaki(1993) further observes that the controller in these constructions must be familiar in thediscourse (topic), explaining the contrast in (iv); and if this requirement is met, thereneed not be any grammatical controller at all (v):

(iv) After collecting some money, a bank account was opened by the/a business-man.

(v) After pitching the tents, darkness fell quickly.

These considerations suggest that control into temporal adjuncts does not fall under ordin-ary OC, appearance to the contrary notwithstanding. The same goes for rationale clauses,which are controlled by an intentional causer, who may, but need not, coincide with thegrammatical subject (e.g., Williamss (1974) example Grass is green to promote photo-synthesis). For these and related reasons, we keep control into (non-extraposed) adjunctsoutside the present system.

140 IDAN LANDAU

lished between a probe and a goal in the domain (=sister) of , invirtue of some uninterpretable feature of (suicidal Greed) and somefeature of that makes it visible or active. The operation Move consistsof Agree(, ), generalized pied-piping that determines the size of thephrase P() to be moved, and Merge (P(), P)), where P() is (normally)merged as the specifier of .

In the case of normal A-movement, it is assumed that uninterpretableAgr on T is the probe, and structural case is what makes the goal DPactive; the same goes for expletive constructions, where Agree alone,not followed by Merge, is established. What would be the features thatdrive Agree in OC? What is the probe? I suggest that the probe is theuninterpretable phi-features of the matrix functional head that agrees withthe controller.

In the case of subject control, the probe is T; in the case of object con-trol, it is small v. For dative and oblique controllers, I assume that they, too,are licensed as specifiers of suitable functional heads. The latter could bethe applicative v of Marantz (1993), the prepositional cascade heads ofPesetsky (1995), the Asp heads of Borer (1998), or any other analogue thatsuits the readers taste. All we need is that every matrix DP be associatedwith some functional head that agrees with it in phi-features (and checksits case).

Turning to the goal of Agree in OC, I suggest it is the infinitival Agr,made active precisely because of its anaphoric nature. This is in fact areinterpretation of the proposal made in Borer (1989). In Borers system,the null subject of the infinitive must be identified by Agr with features.The infinitival Agr, however, being anaphoric, possesses no inherent phi-features. Once Agr raises to C, it may be bound by a matrix argument inaccordance with the standard Condition A of the binding theory. Since thenull subject is coindexed with Agr, it is properly identified and inherits thereference of Agrs antecedent. Thus control of PRO reduces to bindingof Agr.

Borers most compelling argument for anchoring control to Agr ratherthan to PRO was that it enables one to express some cross-linguistic gen-eralizations that are otherwise mysterious. These are: (i) If an anaphoricAgr occurs in a finite Infl, one expects to find control into finite clauses;(ii) If overt pronouns can occur as subjects of infinitives, one expects tofind control of overt pronouns. Both cases are cross-linguistically attested(see also Suer (1984), Farkas (1985) and Yang (1985)). However, thereare various problems with the idea that control reduces to binding (see


Mohanan 1985; Rizzi 1986; Lasnik 1992).13 We will pursue a differentimplementation that nevertheless preserves the positive results of Borerstheory.

The derivation of OC I propose is illustrated below:

(52) Obligatory Control

In the embedded clause, Agree1 matches (and coindexes) the phi-featuresof Agr (in T) and PRO, followed by raising of the latter to [Spec,TP]. Next,T-Agr raises to the embedded C. Control is now implemented as Agree2(F, T-Agr). A third Agree relation is then established between F and DP(possibly followed by raising of DP), valuing the phi-features of F, andby transitivity, those of T-Agr and PRO. PRO gets to be coindexed withthe matrix controller. Syntactically, this relation is reflected by identity ofphi-features; semantically, PRO is interpreted as a variable bound by thecontroller.

The first question that arises is what drives the movement of T-Agr to C.In Borers original proposal, there are two different sources of Agr-to-C:In some cases it is simply motivated by PROs need to be identified (by aset of phi-features); since Agr is anaphoric (and PRO itself unspecified forphi-features) it must acquire the requisite features from an external binder.In other cases (where the subject of the infinitive is a lexical pronoun), Agr-to-C is motivated by the ECP, on the assumption that an empty C violatesthis constraint.

One can question this account on various points. First, the notion of anoperation driven by look-ahead considerations is dubious under minim-alist assumptions. Second, it is unclear why an abstract Agr can satisfythe ECP any more than an abstract C can. Third, it seems that lexicalcomplementizers do not block OC in a manner expected under Borersanalysis (e.g., di/de in Romance, om in Dutch, etc.).

13 To mention just a few asymmetries: Implicit arguments can control PRO but not bindanaphors (Rizzi 1986); in certain languages, dative/oblique arguments can control but notbind (Wurmbrand 1998); argumental anaphora cannot have split antecedents, but PRO inOC can (Petter 1998); and subject anaphors (in ECM) allow strict readings under VP-ellipsis, whereas PRO forces sloppy readings.

142 IDAN LANDAU

I will assume instead that Agr raises to C as a free-rider on T, whereT-to-C itself is licensed independently of control, reflecting the fact that Cmarks the tense domain of its clause. This idea goes back to Stowell (1982)and den Besten (1982), and receives ample cross-linguistic corroboration(auxiliary inversion, V2 effects, and inversion in conditionals, subjunctivesand counterfactuals). Thus, complementizers in various languages encodemood distinctions, tense and (subject) agreement. Rizzi (1997) proposes torepresent these C-T correlations in terms of selection: The lowest head inthe C-system, Fin(iteness), selects for certain T-heads but not others. An-other way to capture this (compatible with Rizzis) is by head movement:T raises to C to check off an uninterpretable feature on the latter. Thisspecific proposal was recently made by Pesetsky and Torrego (2000), whoeven extend it to simple that-clauses to account for various ECP and Comp-deletion effects. Pesetsky and Torrego argue that the complementizer thatis really generated under T and raises to C, to satisfy an uninterpretableT-feature on C.

Returning to control infinitivals, we assume, following Stowell (1982)and Pesetsky (1991), that they, too, contain a contentful tense. T-to-C thenuniformly applies, just like in finite clauses. If Agr is anaphoric, we getthe effect of OC; if it is not (as in Aux-to-Comp infinitivals in Italianand Portuguese), no Agree relation is established with a matrix argument,and the embedded subject is referentially independent. The conceptual andempirical problems with Borers proposal disappear under this approach:T-to-C is uniformly driven by an uninterpretable T-feature on C, regardlessof the nature of Agr or the lexical content of C.14

14 In fact, the present analysis makes a novel prediction: Agr will not raise to C intenseless infinitives, since there is no Tense for it to ride on. As Landau (1999) shows,this explains why the effects of partial control are limited to tensed infinitives:

(i)a. Yesterday, John preferred to visit Mary tomorrow.

b. The chair preferred to gather at 6.

(ii)a. Yesterday, John managed to visit Mary tomorrow.

b. The chair managed to gather at 6.

The irrealis complement of prefer is specified for its own tense, whereas the implicativecomplement of manage is not; hence the contrast between (i-a) and (ii-a). The correlatedcontrast between (i-b) and (ii-b) shows that the possibility of interpreting PRO as properlyincluding the controller is linked to the presence of tense. Landau (1999) argues that controlin (i) is mediated by Agr in C, whereas control in (ii) directly applies to PRO (as Agrremains lower). The mediation of Agr allows a mismatch in semantic plurality, givingrise to partial control in (i-b) (see Landau (1999) for details). Notice that in Super-Equi


A second question that arises in connection with (52) is how F and T-Agr can enter a second checking relation. Chomskys (1998, 1999) notionof the phase (a current version of the classical cycle) provides a principledanswer to this question. Chomsky (1998) suggests that checked featuresdo not erase right away when they are checked, but are accessible until theend of the phase, where the phase is the smallest containing vP or CP. Thisformulation is designed to reconcile the tension between the need to makethose features accessible to spell-out after checking, and at the same timeguarantee that they are invisible to later cycles. Chomsky (1999) furtherrelaxes this condition, effectively postponing evaluation (for convergence)and interpretation of a phase to the next phase up.15 Assuming this to betrue, we understand how a single functional head can enter more than onechecking relation in its own phase or the one above it.

Concentrating on the infinitival T-Agr, a further question arises: Evenif it is still available in the matrix phase, is it visible? The answer is pos-itive, assuming another component of Chomsky (1998, 1999), namely thePhase Impenetrability Condition (PIC). The PIC regulates the interfacebetween successive phases, by stating that only the head and the edge(=Spec(s)) of a phase are visible to operations from outside. Thus, Agree2in (52) can see T-Agr, which is at the head of its phase; and Agree3 canre-use F even after Agree2 had used it because it is not erased until thematrix CP phase is over.16

The locality of OC, as stated in (51), now follows from the PIC: Thehead F (and its specifier, the controller) must be found within the phaseimmediately dominating the infinitival CP phase. The fact that the con-troller must be a complement or specifier in the matrix VP-shell followsfrom the fact that only those positions enter into Agree relations with thefunctional heads that head the layers of the shell. Those heads, to recall,are the potential probes for Agree with the infinitival Agr.17

the infinitive is always tensed, being a CAUSER argument temporally preceding the mainevent.

15 Chomsky argues for this revision by showing that it is necessary in order to eliminatethe countercyclic implication of Holmbergs Generalization. The present proposal suggeststhat this bi-phasal derivation has further positive results.

16 Alternatively, we may stipulate that only an interpretable goal may erase the probe.Since T-Agr is uninterpretable, F will not be erased after Agree2, and will be still availableto Agree3.

17 A technical question arises as to how the infinitive is to be placed in the domain of averbal head when it occupies the topmost specifier in the VP-shell. Consider the followingexample:

(i) It would help John1 [PRO1 to loosen up]

144 IDAN LANDAU

So far we have accounted for the locality of the controller with respectto the infinitive in OC. What is still missing is an explanation for why thelocality conditions only apply when the infinitive is VP-internal (comple-ment or specifier). In fact, we already have all the pieces of the answer all we need to do is to put them together.

Consider again the distributional consequences of the OC Generaliza-tion (51). The four major positions an infinitive may occupy fall under thefollowing two categories:

(53)a. VP-complement/VP-specifier) OC

b. IP-subject/adjunct) NOCWe have already seen that this partitioning is supported by extraction

asymmetries: OC infinitives are transparent to extraction, whereas NOCinfinitives like subjects and adjuncts in general are islands. The claim Inow make is that the same structural distinction is responsible for whetheran infinitive displays OC or NOC: It is because VP-complements andspecifiers are transparent that Agree can apply to the anaphoric Agr in-side them, and it is because (IP-)subjects and adjuncts are islands thatAgree fails to penetrate them, resulting in NOC. The latter option, as willbe demonstrated shortly, corresponds to logophors namely, anaphoricelements that are not syntactically licensed. It should be noted that theliterature offers numerous diagnostics for the OC/NOC distinction, butvery few explanations for why certain infinitival positions are subject toOC and others to NOC. The conception of control as an Agree relation is agenuine explanation in that it reduces the distinction to the well-establishedCED of Huang (1982).18

(ii) It would [vP [vP [PRO1 to loosen up]LF v [VP help John1]][PRO1 to loosen up]PF].

The infinitive, being a CAUSER co-occurring with a THEME, is generated in [Spec,vP].Extraposition attaches a phonological copy of the infinitive to the right edge of the vP, whilethe LF copy is the one occupying the base position (see Section 3 for empirical motivationfor this derivation). Notice that the latter is not in the domain of v, the functional head thatagrees with the object controller John (assuming covert object shift). A possible solution isto allow (or require) v-to-T raising. Standard definitions of c-command would then let theadjoined v c-command the infinitive (and the infinitival Agr in its head), allowing Agree toapply. Notice that the implication is that the controller and PRO in OC do not obey strictc-command, as opposed to binders and reflexives (Barss and Lasnik 1986; Larson 1988;and Pesetsky 1995). This is in fact a positive result, as it converges on several independentarguments (see fn. 13).

18 Superficially, the present proposal appears similar to Hornstein (1999), where OC isviewed as an A-chain relation. In fact, there are significant empirical differences, beyond


5.2. NOC and LogophoricityThe original insight of Borer (1989) was that defective Agr features comein two varieties: They may be either missing altogether (as in certain par-ticipial constructions) or anaphoric, that is, dependent on some antecedent(typically, the case of infinitivals). I argued above against the assimilationof control to binding. As a consequence, condition A does not apply toanaphoric Agr, but the operation Agree does instead. One may wonderat this point what is gained by viewing the infinitival Agr as anaphoric,beyond the general property of being referentially defective.

The answer to this question comes from the reflexivity theory of Rein-hart and Reuland (1993). Under this theory, reflexives are distinguished bya certain formal feature, [-R], which reflects their referential dependence.Reflexives that are syntactically licensed (by Condition A) are anaphors,and all the rest are logophors. The dividing line is purely syntactic: Onlyreflexives occupying positions that qualify as syntactic arguments fallunder Condition A. Logophors are licensed by discourse factors such asfocus, perspective, and center of consciousness or communication. Al-though these factors are quite diverse, the class itself is well-defined, beingthe complement to the class of reflexives occupying syntactic argumentpositions.

Suppose this division applies to all anaphoric elements, and in par-ticular, to anaphoric Agr as well. We then expect anaphoric Agr (andderivatively, PRO) to be logophoric precisely in those environments whereit is not syntactically licensed namely, where it cannot enter an Agreerelation with a matrix element. These environments should display LD-

the formal distinction between Move and Agree. For Hornstein, OC virtually reduces toraising, and PRO is an NP-trace. Thus, the sentence John1 was hoped [t1 to win] isderivable in his system, but not in the present one. Second, for Hornstein the controllerin OC (like any raised NP) must be syntactically overt, since it checks case; thus, he leavesno room for implicit controllers - a robust, well-documented phenomenon (Kimball 1971;Epstein 1984; Manzini 1986; Brody and Manzini 1987; Roeper 1987; Clark 1990). Underthe present proposal, it is a functional head, rather than the controller itself, which must berepresented in the syntax (the implicit controller can be filled-in at LF). Third, Hornsteinclassifies interrogative infinitivals - islands for NP-raising - under NOC. But it was alreadyobserved by Chomsky (l986a, p. 127) that these environments do not tolerate LD-control:

(i) I thought they wondered how PRO to feed themselves/myself.

In the present proposal, the fact that the infinitival Agr is in C renders it visible to thehigher phase (see Landau (1999) for more evidence that interrogative infinitivals fall underOC). Fourth, the phenomenon of partial control (see fn. 14) is inexplicable under Horn-steins analysis (there is no partial raising). I return below to additional problems withHornsteins view of NOC as pronominal coreference.

146 IDAN LANDAU

control, subject to logophoricity constraints. The prediction is strikinglyconfirmed across a wide array of constructions, to which we now turn.

Consider first the Super-Equi cases analysed in detail above:

(54)a. Mary1 thought that it [VP [VP helped John] [IP PRO1 to speakher mind]].

b. Mary1 thought that [IP [PRO1 to speak her mind] would helpJohn].

Whether in adjunct or in subject position, the infinitive is an island; Agris unable to be identified by agreement with an element external to theinfinitive. Consequently, it can only be licensed as a logophor.

That LD-control in Super-Equi has a lot in common with logophoricanaphora is an old insight in generative grammar (Grinder 1970; Kuno1975; Lebeaux 1985; Williams 1992; Manzini and Roussou 2000). Kunoobserves the following parallels:

(55)a. John said to Mary that it would be easy to prepare herself forthe exam.

b. John said to Mary that there was a picture of herself with aMafia figure in the newspaper.

(56)a. John said about Mary that it would be easy to prepare herselffor the exam.

b. John said about Mary that there was a picture of herself with aMafia figure in the newspaper.

(57) John sued Mary for divorce because it was no longer possibleto support her/him.

Kunos Direct Discourse analysis, by which the anaphoric element mustrefer to a discourse participant, is readily understood as the restriction thatlogophors refer to the center of consciousness or communication in the dis-course.19 Notice that the failure of LD-control in (57), despite the fact thatMary is an argument of the matrix verb, shows that the problem with (56)

19 According to Sells (1987, p. 445), an antecedent for a logophor can be the source ofthe report, the person with respect to whose consciousness (or self) the report is made,and the person from whose point of view the report is made.


is not the grammatical function of Mary. Under the present proposal, thelogophoric nature of picture-anaphora and the parallel nature of displacedinfinitives are the very same fact.

The idea that PRO in NOC (or, more precisely, the infinitival Agr)is logophoric should be contrasted with an alternative view that takes itto be pronominal (Bouchard 1984; Hornstein 1999). Whereas for us theanaphoric nature of Agr/PRO is stable the mechnism of licensing beingdifferent in OC and NOC for Bouchard and Hornstein PRO is an ana-phor in OC but a pronoun in NOC. Given that logophors are subject tomore stringent antecedence conditions than pronouns, the two views makedifferent empirical predictions.

For a (non-deictic) pronoun to be felicitous, its antecedent merely needsto be familiar in the discourse. A logophor, however, can only refer toantecedents that are sources or targets of mental/communicative reports.Kunos examples (56a) and (57) clearly follow the latter pattern; theytolerate an overt pronoun but not PRO as the subject of the infinitive:

(58)a. John said about Mary that it would be easy [(for her) to prepareherself for the exam].

b. John sued Mary for divorce because it was no longer possible[for her) to support him].

The following contrasts further illustrate the same point:

(59)a. Johns wife thought that [(for him) to indulge himself indrinking] would destroy their marriage.

b. Mary discussed with friends of John1 the possibility of[him1/PRO1 divorcing her].

(60) Johns1 friends think it is illegal [(for him) to feed himself].(Chomsky 1986a, p. 125)

(61)a. [(him) having been away for so long], nothing really matters toJohn.

b. [(him) having been away for so long], nothing really bears onJohn.

(62) [(his1) having shaved already] shows that Mary arrived morethan 5 minutes after John1 did. (Lebeaux 1984)

148 IDAN LANDAU

Thus, one cannot conclude from the mere possibility of LD-control thatPRO in NOC is a pronoun; a more careful analysis reveals that thereferential properties of PRO in these environments are constrained by lo-gophoricity. For Bouchard (1984), the pronominal nature of PRO in NOCfollows from the contextual definitions of empty categories definitionswhich make no room for logophors. Hence, the facts above cannot beaccommodated in his system. For Hornstein (1999), the pronominal natureof PRO in NOC in fact does not follow from anything; it is rather stipu-lated that base-generated empty categories are pronominal. Thus, it couldjust as well be stipulated that PRO in NOC is a silent logophor. However,this would still leave Hornsteins system with no obvious relation betweenPRO in OC (an NP-trace) and PRO in NOC (a silent logophor); whereasin the present analysis there is just one anaphoric element (nonfinite Agr)which falls under different licensing principles in different environments,as nominal anaphors do.

Finally, the linking of NOC to islandhood may provide an answer toa problem raised in footnote I. It was noted there that for some speakers,Grinders intervention effect goes away even with psychological verbs when the infinitive is embedded under a modal tense. These speakers find(63b) better than (63a):

(63)a. John said that it disturbed Sue to make a fool of himself inpublic.

b. John said that it would disturb Sue to make a fool of himself inpublic.

Suppose that for the speakers under consideration, the infinitive in (63a)occupies a VP-internal position but the one in (63b) is adjoined to VP.Then the control facts will follow from the CED. But why should the ad-junct option be available here? A possible answer is provided by Pesetskys(1991) observation that sentences like (63b) have a conditional semantics,roughly paraphrasable as John said that it would disturb Sue if he madea fool of himself in public. Suppose that the speakers who accept LD-control in (63b) actually project the infinitival as an if-clause. If-clausesbeing adjuncts, the infinitive will behave just like an extraposed clauseunder a non-psychological predicate; in particular, it will allow NOC.Notice that the conditional semantics alone does not necessitate such asyntax-semantics mapping; the infinitive could equally well be a genuineargument in the syntax, being copied later on in the semantics onto anabstract if-clause. Hence we expect to find speakers for whom (63b) is notbetter than (63a), as seems to be the case.


To sum up, we see that the various instantiations of NOC all displaylogophoricity. This property emerges in island environments, where theinfinitival Agr fails to enter an Agree relation with a controller and can-not be syntactically identified. This state of affairs mirrors the pattern ofnominal anaphora to a remarkable degree, suggesting that the conceptionof control as a manifestation of anaphoric Agr captures a deep property ofanaphora in natural language.

6. CONCLUSION

In this paper I have argued from hitherto unnoticed asymmetries for a par-ticular view of how control and extraposition interact. Specifically, it wasshown that infinitives embedded under psychological predicates requirelocal control, whereas those embedded under nonpsychological predicates,or intraposed to a subject position, allow non-local control. The presentanalysis takes this paradigm to constitute the fundamental empirical chal-lenge for theories of Super-Equi. As far as I know, existing proposals donot meet this challenge.

The empirical content of the proposal resides in the correlations ob-served between the possibility of LD-control and the structural positionof the infinitive, diagnosed by extraction. In certain contexts, infinitivesbehave as though they occupy one position at LF but a different one at PF.This result lends interesting support to current views on reconstruction,where different links (copies) in the same chain feed different interfaces.An open question that remains is whether and how one can derive fromdeeper principles the PF requirement on VP-internal clauses to extrapose.

It was argued that a fairly simple theory of OC one whose central no-tion is containment of the infinitive and the controller in the same VP-shell can account for all the relevant facts. Under the proper interpretation,this view is consistent with some recalcitrant data, like the transparency oflogophoric extensions and cases of apparent OC in intraposition.

Finally, I argued that OC is a manifestation of an Agree relationbetween the functional head that licenses the controller and the anaphoricinfinitival Agr. To be able to enter such a relation, Agr must raise to theembedded C head; independent considerations suggest that Agr gets tothat position as a free-rider on T-to-C movement. Due to the CED, theAgree relation can only be established when Agr is inside a transparentinfinitive that is, in non-island positions. Accordingly, Agr in subjectand adjunct infinitives must resort to a different licensing mechanism. Itwas argued that in these cases, Agr (and PRO) is licensed as a logophor,much like nominal anaphora that fails to be syntactically licensed. This

150 IDAN LANDAU

explains traditional observations concerning NOC that have failed to beproperly integrated with the theory of OC in the past. The emerging pic-ture incorporates a principled division of labor between syntax (whichfixes the domain of OC), semantics (which fixes the controller in OC) andpragmatics (which fixes the controller in NOC).

REFERENCES

Arad, Maya. 1998. VP-structure and the Syntax-Lexicon Interface, unpublished Ph.D.dissertation, University College London.

Barss, Andrew and Howard Lasnik. 1986. A Note on Anaphora and Double Objects,Linguistic Inquiry 17, 347354.

Belletti, Adriana and Luigi Rizzi. 1988. Psych-Verbs and 2-Theory, NLLT 6, 291352.Bobaijik, Jonathan D. 1999. Implications of Adjacency: On the Typology of Covert

Movement, Handout of a talk given at the University of Ottawa.Borer, Hagit. 1998. Passive Without Theta-Grids, in Diane Brentari and Patrick Farrell

(eds.), Morphology of Its Relations to Phonology and Syntax, CSLI. Stanford, pp. 6099.Borer, Hagit. 1989. Anaphoric AGR, Osvaldo Jaeggli and Kenneth J. Safir (eds.), The

Null Subject Parameter, Kluwer Academic Publishers, Dordrecht, pp. 69109.Bouchard, Denis. 1984. On the Content of Empty Categories, Foris, Dordrecht.Bresnan, Joan. 1982. Control and Complementation, Linguistic Inquiry 13, 343434.Brody, Michael and M. Rita Manzini. 1987. Implicit Arguments, in Ruth Kempson (ed.),

Mental Representations: The Interface Between Language and Reality, University Press,Cambridge, pp. 105130.

Chierchia, Gennaro and Pauline Jacobson. 1986. Local and Long Distance Control, NELS16, 5774.

Chomsky, Noam. 1981. Lectures on Government and Binding, Foris, Dordrecht.Chomsky, Noam. 1986a. Knowledge of Language: Its Nature, Origin and Use, Praeger,

Westport, Connecticut.Chomsky, Noam. 1986b. Barriers, Cambridge, MIT Press, Cambridge, Massachusetts.Chomsky, Noam. 1995. The Minimalist Program, MIT Press, Cambridge, Massachusetts.Chomsky, Noam. 1998. Minimalist Inquiries: The Framework, MIT Occasional Papers

in Linguistics 15, MITWPL, Cambridge, Massachusetts.Chomsky, Noam. 1999. Derivation by Phase, MIT Occasional Papers in Linguistics 18,

MITWPL, Cambridge, Massachusetts.Cinque, Guglielmo. 1990. Types of A-bar Dependencies, MIT Press, Cambridge, Mas-

sachusetts.Clark, Robin. 1990. Thematic Theory in Syntax and Interpretation, Routledge, London.Clements, George N. 1975. Super-Equi and the Intervention Constraint, NELS 5, 1328.Comrie, Bernard. 1984. Subject and Object Control: Syntax, Semantics and Pragmatics,

in Claudia Brugman, Monica MacCauley, Amy Dahlstrom, Michele Emanation, BirchMoonwoman and Catherine OConnor (eds.), Proceedings of the Tenth Annual Meetingof Berkeley Linguistic Society, University of California, Berkeley, pp. 450464.

den Besten, Hans. 1982. On the Interaction of Root Transformations and Lexical DeletiveRules, in Werner Abraham (ed.), On the Formal Syntax of the Westgermania, JohnBenjamins, Amsterdam, pp. 47131.

Diesing, Molly. 1992, Indefinites, MIT Press, Cambridge, Massachusetts.


Epstein, Samuel. 1984. Quantifier-pro and the LF Representation of PROarb, LinguisticInquiry 15, 499505.

Farkas, Donca. 1985. Obligatorily Controlled Subjects in Romanian, in William H. Eil-fort, Paul D. Krocher and Karen L. Peterson (eds.), Proceedings of the Twenty-firstAnnual Meeting of the Chicago Linguistic Society, University of Chicago, Chicago, pp.90100.

Farkas, Donca. 1988. On Obligatory Control, Linguistics and Philosophy 11, 2758.Fox, Danny. 1999. Reconstruction, Binding Theory and the Interpretation of Chains,

Linguistic Inquiry 30, 157196.Grinder, John. 1970. Super Equi-NP Deletion, in Proceedings of the Seventh Annual

Meeting of the Chicago Lingui

Super Equi

Documents

Transcript of Super Equi