1999 Control In Grammar And Pragmatics _ A Cross-linguistic Study-j. Benjamins Pub (1999).pdf

CONTROL IN GRAMMAR AND PRAGMATICS

LINGUISTIK AKTUELL This series provides a platform for studies in the syntax, semantics, and pragmatics of the Germanic languages and their historical developments. The focus of the series is represented by its German title Linguistik Aktuell (Linguistics Today). Texts in the series are in English. Series Editor Werner Abraham Germanistisch Instituut Rijksuniversiteit Groningen Oude Kijk in ’t Jatstraat 26 9712 EK Groningen The Netherlands E-mail: [email protected] Advisory Editorial Board Guglielmo Cinque (University of Venice) Günther Grewendorf (University of Stuttgart) Liliane Haegeman (University of Geneva) Hubert Haider (University of Salzburg) Christer Platzack (University of Lund) Ian Roberts (University of Stuttgart) Ken Safir (Rutgers University, New Brunswick NJ) Höskuldur Thráinsson (University of Iceland, Reykjavik) Lisa deMena Travis (McGill University) Sten Vikner (University of Stuttgart) C. Jan-Wouter Zwart (University of Groningen)

Volume 27 ° zi†ka Rudolf Ru


Control in Grammar and Pragmatics. A cross-linguistic study.

CONTROL IN GRAMMAR AND PRAGMATICS A CROSS-LINGUISTIC STUDY

°
IC
KA RUDOLF RUZ University of Leipzig

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM / PHILADELPHIA

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data ° zi
Ru
c ka, Rudolf. ° zi
c ka. Control in grammar and pragmatics : a cross-linguistic study / Rudolf Ru
p. cm. -- (Linguistik aktuell / Linguistics today, ISSN 0166-0829; v. 27) Includes bibliographical references and index. 1. Control (Linguistics) 2. Grammar, Comparative and general. 3. Pragmatics. I. Title. II. Series: Linguistik aktuell ; Bd. 27. P299.C596R89 1999 415--dc21 99-22394 isbn 90 272 2748 9 (eur) / 1 55619 911 2 (us) (Hb; alk. paper) CIP

© 1999 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O.Box 75577 · 1070 an amsterdam · The Netherlands John Benjamins North America · P.O.Box 27519 · Philadelphia pa 19118-0519 · usa

Preface

This study has grown out of a series of papers starting with “Remarks on Control” (Linguistic Inquiry, 1983), and reflects a gradual process of clarifying my insights into the phenomena of Control and what is related to it. I have presented my views in lectures and courses at the Universities, academic institutions or Congresses in Brighton (Sussex), Leeds, Stockholm, Lund, Gothenburg, Prague, Kiew, Zagreb, Hamburg, Frankfurt, Berlin, Munich, Leipzig, and Jena. I wish to acknowledge the helpful comments and criticisms of a number of colleagues and friends, including Adriana Belletti, Manfred Bierwisch, Guglielmo Cinque, Milka Ivi´c, John Lyons, Jarmila Panevová, Barbara Partee, Luigi Rizzi, Inger Rosengren, Petr Sgall, Peter Suchsland, and Ljubov Vladimirovna Zlatoustova. I am particularly indebted to Werner Abraham for his valuable suggestions and comments on both content and exposition. Naturally, the final decisions have all been mine. Finally, I want to express my gratitude to María del Mar Rodríguez Fernández and Ralf R˚uzicka for much technical help.

Contents

Chapter 1. Introduction Chapter 2. Preliminaries 2.1. Reflections on Control and its domains 2.1.1. Control and economy conditions 2.1.2. On the array of Control – Its central domain – 2.1.3. Other major domains of Control 2.1.3.1. Controlled complement clauses with filled C(P) 2.1.3.2. Controlled clauses as adjuncts 2.1.3.2.1. Infinitival S-structures 2.1.3.2.2. Control in free participial and nominal adjuncts 2.2. Approaches to Control 2.2.1. Emergence of Control 2.2.2. The configurational approach 2.2.2.1. Manzini’s theory 2.2.2.2. Kayne’s similar approach 2.2.2.3. On the arbitrariness of PROarb 2.2.2.4. Configurational and binding approaches extended 2.2.3. Questions posed to Control 2.2.4. Clausal status of the controlled phrase or the propositional vs. the attributive account of Control 2.2.5. Θ-roles and Control 2.2.6. Specification of thematic properties and Control domains 2.3. Organization of the study and languages considered Chapter 3. The theory outlined 3.1. Object Control – a pilot demonstration 3.2. Constraints on “object” Control 3.2.1. Shape of the constraint

1

3 3 5 6 6 9 9 10 12 12 13 13 16 19 22 23 23 24 25 27

29 30 30

viii

Contents

3.2.2. Markedness as opposite feature values 3.2.3. Unmarked and marked Control constraint as a functional instantiation of lexical features 3.2.4.1. Interpreting and extending effects of the markedness distinction in constraint (B,b) 3.2.4.2. Violations of constraint (B,b). 3.2.5. Violating the m(arked) constraint and cross-linguistic variation. Multiple options 3.3. Grammar and recourse to pragmatics 3.4. Nominalized Russian prosit’(sja) 3.5. When pro takes over 3.6. A Comparison with an alternative analysis Chapter 4. The persuade subclass and cognate ones subject to the general constraint 4.1. persuade versus ask. Parameterization in pragmatics 4.2. On complement unaccusatives and unergatives 4.3. Neutralization of the null object parameter with respect to Control of persuade complements 4.4. Further subclasses obeying constraint (B,b) 4.4.1. The teach subclass 4.4.2. The case of signal as a Control verb 4.4.3. teach and ask. Larson’s comparison 4.4.4. A subclass of Control verbs subcategorizing for an oblique object 4.5. Ambiguous allow in Control and ECM-contexts 4.6. Lexically saturated theta-roles Chapter 5. The case of promise 5.1. Tenacity and untenability of subject Control 5.2. A classical violation 5.3. Violating the unmarked constraint 5.4. A structural-pragmatic conflict? 5.5. Larson’s division of labor 5.6. The analogous case of threaten 5.7. Considering the raising analysis of promise and threaten 5.8. Intervention of modal verbs

34 38 40 41 43 44 49 51 52

55 64 65 67 67 69 72 76 81 85

87 93 97 99 104 105 106 109

Contents Chapter 6. No Choice of Controller 6.1. Matrix active clauses 6.1.1. The case of try / attempt 6.1.2. try vs. intend 6.1.3. Effects on passive controlled clauses and unspecified PRO 6.1.4. Non-passive controlled clauses failing condition (D) 6.2. Matrix passive clauses 6.3. Raising contexts

ix

113 113 117 119 121 123 127

Chapter 7. Reflexive clitic impersonal clauses in Control structures 7.1. The situation in Slavic languages 129 7.2. Control and reflexive passive The status of pro in impersonal reflexive clauses as a prerequisite for controlling 130 Chapter 8. “Self-control” with evaluating and attitudinal predicates 8.1. Adjectival predicates 137 8.1.1. Two basic situations 137 8.1.1.1. Situation (a) and Epstein’s formula 138 8.1.1.2. Situation (b) 141 8.1.2. For “better” or “worse” 147 8.2. Verbal predicates of evaluational or attitudinal character 148 8.2.1. The oblique object controller 148 8.2.2. The subject controller 148 Chapter 9. Aspects of control in the “grain problem” and in VP-deletion 9.1. The “grain problem” and reciprocal constructions 9.2. Freedom of constraints as a precondition for strict identity in anaphoric contexts Chapter 10. Summary and conclusions, with some remaining questions 10.1. Essentials and main general proposal 10.2. The constraints and markedness 10.3. The classes of Control 10.3.1. Control class (B,b) (=II) and its main subclasses 10.3.1.1. The ask-type subclass (II)a

151 153

155 157 158 158 158

x

Contents

10.3.1.2. The persuade subclass (II)b 10.3.1.3. Subclass (II)c 10.3.1.4. Verbs of peculiar behavior in (II) 10.3.2. Control class subject to (B, a) (= I) 10.3.3. Self-regulating Control 10.3.3.1. The case of try and related verbs 10.3.3.2. A related Control process of evaluating and attitudinal Control elements 10.3.3.3. Verbal predicates indifferent to thematic feature constraints 10.4. Thematic and nonthematic Control 10.4.1. Two particular verbs 10.4.1.1. The German control verb vorschlagen 10.4.1.2. agree as control verb 10.4.2. Control in infinitival adjuncts introduced by ohne (without) 10.5. Closing remarks Chapter 11. Turning to the Minimalist Program 11.1. On the status of control 11.2. Derivational and inherent economy 11.3. Thematic features 11.4. Control treated with the PRO-theoreme 11.4.1. Conceptual considerations 11.4.2. Syntactic treatment 11.4.3. A different class of control verbs 11.4.4. Markedness 11.5. Can pragmatics be parametrized? 11.6. On approaches that give up postulating PRO 11.6.1. General conditions 11.6.2. Control in terms of a calculus of features 11.6.3. An illustration of control relations described without positing PRO 11.6.4. Dispensing with PRO and pro Notes References Subject index

160 162 163 163 165 165 166 167 168 168 168 169 169 171

175 176 176 177 177 178 179 180 180 181 181 182 184 186 187 197 205

Chapter 1

Introduction

This monograph is devoted to a study of cross-linguistic variation and accord in control conditions, and to a unifying account that explicitly relates to grammar the conceptual-pragmatic factors indispensable to the full understanding of control. A particular but pivotal area of control phenomena has been chosen, which I will claim is the best access to the empirical understanding and adequate methodological treatment of control in general. The area comprises infinitival complement clauses and, occasionally, relevant cognate fields. The result arrived at in this area will help to understand and distinguish more clearly the principled mechanisms of control in its different domains. The present study in a sense is complementary to the articles in the volume Control and Grammar (1992), which, in the editors’ introductory words, “… do not much address the vexed question of what determines whether a complement can be controlled and, if so, what syntactic and semantic properties of the clause determine the choice of controller. It seems that complex properties of the lexical items are involved.” (Larson et al. 1992: XVI) In particular, “… argumental infinitives in complement position get their understood subjects by the lexical control relation (whatever that turns out to be).” (Williams 1992: 314) In general, I adhere to the assumption that “… control is best treated in the context of lexical semantics of the verb or predicate in question …” (Joseph 1992: 200), an assumption which I started to elaborate in the ° context of thematic roles (Ruzicka 1983). Treatments of control in a similar spirit can be found in the works of Comrie (1985), Farkas (1988), Laduslaw and Dowty (1988), Abraham (1983), Wegener (1989). The present investigation goes a radical step further in its attempt to establish unifying constraints, which are shaped into conjunctions of feature specification values. The distinct pairs of values are associated with respective classes of control verbs or predicates. The constraints mediate between the richness of lexical information potentially relevant to control, and principled regulations for exercising licit control. They

2

Chapter 1

provide firm ground for relating language particular variations in control behaviour to other parameters, and, last but not least, for considering conceptual-pragmatic factors and backing them up grammatically. Fundamental points that will emerge in the investigation are: (a) Cross-linguistic variation in control conditions can be captured by constraints that generalize over the control behaviour as determined by the respective lexically semantic classes into which the control verbs and adjectives can be grouped. Such constraints interface the natural lexically bound control regulations and the range of language-particular options still available within lexical restrictions. In other words, the chances of accounting for cross-linguistic variation in control arise through the mediation of a formal apparatus of particular constraints that interlink lexical information with control conditions. Crucially, the format of the constraints allows for independently justified unmarked and marked interpretation. (b) Control conditions, however they are formulated, must contain, and refer to, theta-properties of the licit controller NP and the controllee. (c) Cross-linguistic variation in control, still squaring with lexical restrictions in (b), may be due to different settings of independent parameters, e.g., the null object parameter. Constraints, then, must be related to certain parameter values if they are to be instrumental in accounting for variation in control. (d) In describing and explaining control behaviour one cannot avoid crossing the boundaries of grammar proper, but the passage to conceptual-pragmatic or other areas can be justified and related to the grammatical “core”. Here again, the respective constraints mentioned above are involved in an empirically natural way. (e) The idea that pragmatics can be parameterized has been substantiated by putting forward interface parameters which mediate between pragmatics and grammar. (f) What comes to be at issue is the methodological justification for laying claim to control theory as a fully-fledged component or module of grammar, a “subsystem of principles” (Chomsky 1981: 5).

Chapter 2

Preliminaries

2.1. Reflections on Control and its domains 2.1.1. Control and economy conditions The empirical problems usually assembled under the notion of Control cannot be properly understood and accommodated in a chosen theoretical framework unless the linguistic devices are identified and surveyed from which the grammatical phenomena are isolated that have been given that half-metaphorical name. Theorizing on Control, we find ourselves committed to facts and principles of economy in the design of language, which empirically legitimize economy of derivation and explanation. Indeed, they manifest themselves consistently in the grammatical areas to be considered, demonstrating more conspicuously than elsewhere that they “… play a significant role in accounting for the phenomena of language and its use.” (Chomsky 1992) Why is Control a particular instantiation of economy? Speaking in terms of the external interface levels, that is, the articulatory-perceptual level PF and the level that constitutes the instructions for the conceptual-intentional system, LF (following Chomsky & Lasnik (1993) and Chomsky (1992)), we are concerned in principle with economy conditions displayed in pairings of PF and LF representations. Language(s) have to cope with a particular task of verbalization, which consists in rendering the intended conceptual relationship between two events, where event is understood in the abstract sense, as, e.g. in Davidson (1987), Higginbotham (1985), or in Bierwisch (1989), the latter interpreting event as the instantiation of a proposition. The key situation invoking Control as a phenomenon of “… understood reference for an argument position … if there is no expression in that position …” (Higginbotham

4

Chapter 2

1992: 79) arises when one and the same entity is involved in both interrelated events figuring as an NP-argument in both event descriptions. Setting aside conjunction or embedding as a finite clausal complement or adjunct, we are concerned with a different syntactic option, namely, infinitival complements or adjuncts, and participial adjuncts. This option is not taken on the chance of the reference of the “empty” element being reconstituted, but on condition that it readily can. The choice, e.g., of an infinitival S-complement creates the problem of reconstituting its empty subject. But we rather have to proceed in the opposite direction by claiming that the choice of the nonfinite complement (or adjunct) is determined by the presumed identifiability of the empty element. Syntactic organization is guided by economy principles in that phonetic absence of the recurring element in subject position is forced by the nonfinite verb. Note that if economy can dispense with assignment of phonetic shape, it cannot exclude the resulting empty category from the “overt” syntax, since it is pertinent to the LF-component, which channels instructions for cognitive “reality”. Which factors should naturally be expected to impose conditions or restrictions on the theoretical reconstruction of the elements left empty. In other words, how are the particular interface conditions realized in the relevant pairing of PF- and LF-representations possibly projected into conceptual-pragmatic mapping. The outward mechanism of control hinges upon the syntactic status of the nonfinite embedded clause. If it is an infinitival S-complement (and sentential argument) of the matrix verb, control is affected by the relationship set between the two events, more exactly the event descriptions. A second consequence that affects control directly: The controllee is an argument in the predicate-argument structure of the complement clause and carries a thematic characterization of its θ-grid. The complement clause in turn is a (sentential) argument of the matrix predicate and carries a θ-role assigned by the matrix verb. Since the matrix (control) verb determines thematic and, possibly, semantic properties of the potential controller NP as well as of the clausal complement whose empty subject NP argument (at S- or LF-structure) is the controllee, the identification of both NP’s is, presumably, sensitive to compatibility restrictions. For example, if the controlled NP is the subject of an active infinitival S-complement of promise, it can take only the subject of the matrix active clause as its controller. It is cross-clausal quasiselectional restrictions of this basic type which, developed into self-consistent devices, legitimize the economy of dispensing with the assignment of pho-

Preliminaries

5

netic shape to the respective pertinent NP-arguments. Since economy principles capitalize on such devices — on which we are going to enlarge — they cannot but produce, and admit to the language, delicate mechanisms which can be distinguished by at least three characteristics and their respective methodological consequences: (a) Compatibility restrictions can extend to less transparent thematic and semantic ground invoking subtler and possibly less elucidatory considerations and inferences. If it becomes imperative to leave grammar, the conceptual-pragmatic notions adopted must interrelate with grammar, and be referred to by it. (b) The linguistic and pragmatic issues sketched under (a) can be expected to be particularly sensitive to possibly parameterized cross-linguistic variation. An interesting aspect to be examined here turns on the hypothesis of parameterization extending into pragmatics. Corresponding parameter values fixed for certain phenomena may be distributed to grammar and pragmatics, respectively. Or one language allows for a certain structure pragmatically, and the other(s) set the parameter negatively. (c) There are exterior consequences linked to the characteristics given in (a), (b). Judgements of grammaticality and/or acceptability of the relevant structures tend to vacillate, to be subtle or to be hedged by pointing out “preferred” or “not preferred” readings, by indications or suggestions of degrees of grammaticality or typical or untypical instantiations of a structure. 2.1.2. On the array of Control — Its central domain What has been described in (a)-(c) may be regarded as a counterbalance to the effects of economy, that is, as the way that the grammatical organization reacts to principles of economy, “experimenting” with grammatical ones on suitable pairings of PF and LF. The interaction between deterministic stringency of principles or rules and the pursuit of economy principles is easily recognizable in the differentiated dependence of control on lexically determined properties of the matrix verb or adjective. Appropriate generalizing treatments of this dependence will require us to find the principles that mediate between thematic specifications and control conditions. They will have a common shape with distinct pairings of feature values determined by the lexical classes relevant for control behaviour. Most interesting and elucidatory in its ramify-

6

Chapter 2

ing effects on control are, for example, verbs like promise, ask, persuade, try, endeavour, signal, teach, threaten, help (and their cross-linguistic counterparts), which are distributed among lexical classes. It will become predictable from the character of the respective control conditions that the controlled element (PRO) is bound to be an argument: “Why is PRO always an argument?” (Brody 1993: 2) On the opposite side of the classificatory spectrum, we find verbs which do not seem to exercise any influence or impose selectional restrictions on thematic properties of the controlled subject of their infinitival S-complement. These are verbs like, e.g. hope, wish, be afraid, expect, hate, like, choose, want. (1)

hates    He is afraid  (PRO to likes 

kiss her ) be kissed   

It is this indifference which is responsible for, or at least involved in, what has been called “non-obligatory” control (Williams 1980: 208). (2) (3)

They expected (him to cheat her) They hated (him to do that)

The infinitival verb is often interchangeable with a gerund complement, which, possibly, is another consequence of specific properties of these verbs. See (4) and (5). (4) (5)

She was afraid of (PRO asking for help) He dislikes (PROi being spoken to ti like that)

If one can roughly speak of this group of verbs as attitudinal verbs, the indifference indicated above seems plausible. 2.1.3. Other major domains of Control 2.1.3.1. Controlled complement clauses with filled C(P) Controlled infinitival complements with non-zero C(P) furnish further evidence that control is sensitive to the interpretation of the matrix verb. Control reacts to distinct event structures even when they are associated with one and the same verb. Let us take for example ask, which subcategorizes for a wh- or question-operator.

Preliminaries (6) (7) (8)

7

Johni asked (CPhow (PROi to behave himself)) John asked (CPhow (PRO to behave oneself)) Johni was asked ti (CPhow (PRO to behave oneself)) (Manzini’s (1983) (adapted) examples (35)-(37))

(6)–(8) differ from each other in their respective control relations. Their control behaviour seems to be indifferent to the matrix (control) verb. However, the apparent indifference to the eligibility of an antecedent is induced by lexical properties of ask and, e.g. wonder, which seem to form a natural class relevant to control. Querying or wondering about something does not imply that the only available NP-argument of the matrix clause has the same intended referent as PRO. The event description of the matrix verb bears no or little relation to the internal structure of the event described in the controlled clause. In this respect, much similarity is observed between these matrix verbs and those of the attitudinal type of the preceding section. Consider (9) and (10). (9)

Johni asked the policeman (CPwhere (PRO(i) to go))

(9) is felicitous, e.g. if John is a guide who asks the policeman about the way he is going to lead others. (10) Ann was just wondering (CP howi (PRO to do it ti)) (10) is compatible with the interpretation that Ann was curious to know how jugglers did some magic trick. Thus, in (9), (10), e.g., John and Ann, respectively, can be controllers, but, in spite of their saliency, they do not exclude other “arbitrary”, better: discourse-related coreferents. Wellformed control cannot be computed here grammatically, that is, thematically or semantically. Control, which is implied by the choice of the independently licensed infinitival S-complement, is licensed in turn on the grounds that it warrants reconstitution of the intended referent of PRO to the extent of meeting reasonably relevant pragmatic requirements. Control is acceptable if compatible with discourse-related or situationally intended reference of PRO. In this lexical area of matrix verbs, grammatical wellformedness is determined before coindexing of PRO. There are, of course, natural independent restrictions to “arbitrarity” or options of control. Reflexive anaphors with PRO as their antecedent — as they are sometimes used heuristically — must rely on the overt occurrence of an intrasentential controller of PRO. The minimal contrast between (11) and

8

Chapter 2

(12) shows that in the former only John is available as a masculine gender controller of PRO, which cannot bind herself. In (12) John and Anne are available as not quite equally felicitous controllers of PRO. In (13), tell allows arbitrary, that is, context-related PRO and, thus, oneself. PRO, if controllable by John and binding himself, results in a bizarre interpretation.

*herself  himself   

(11) John asked how to behave

(12) John asked Anne how to behave

(13) John told Mary how to wash

himself  herself   

 herself   ??   himself   oneself 

The particularity of control in indirect infinitival questions is evidenced (a) by the fact that they are completely disallowed in German, at least with the particle zu, and (b), e.g. in Russian, by the optional occurrence of an overt dative subject. See (14a) vs. (14b). Abbreviations of languages p. 28. (14a)G

*Wir wußten nicht was zu sagen

In Austrian-Bavarian, infinitivals without zu is usual (Werner Abraham, personal communication). (14a′)G Ich habe nicht gewußt, was sagen I didn’t know what (*to) talk (14b)R

My ne

znali o

nam govorit’ [dat] [infinitive] we not knew about what for us to talk ‘We didn’t know about what to talk.’ cem

While the non-allowance of infinitives in German may be evidence of the negative setting of a configurational parameter, the admission to this structure of an overt dative subject replacing PRO in Russian is a variation induced by the possible occurrence of lexically overt dative subjects in infinitivals. Both these facts are indicative of the peculiar control behavior of ask or other verbs subcategorizing for indirect questions.

Preliminaries

9

2.1.3.2. Controlled clauses as adjuncts 2.1.3.2.1. Infinitival S-structures When the infinitival controlled clause is not an argument of the matrix verb, control is handled very differently: The controllee is not intrinsically related to its control verb by virtue of being the subject (at S-structure and LF) of its sentential argument. Crucially, it is not the membership of the matrix verb in a particular class which, when associated with pertinent principles, accounts for, or contributes to accounting for, control. Assuming that the relationship between the matrix clause and its controlled clause is given by a conjunction or complementizer, for example, in order to or by inherent modality of the bare infinitival as a purposive or rationale clause, identifying PRO needs and looks for cooperation in the whole expression and its discourse setting. Choice of the null subject infinitival relies on the reasonable certainty of its reconstitution, which is a combined grammatical and pragmatical effort to achieve a meaningful interpretation. See the following quote from Chomsky (1981). “In purposive infinitivals, reference of the subject PRO may be arbitrary as in (15) or determined by properties of the matrix verb as in (16): (15) The books were sold (PRO to help the refugees) (16) (i) (ii) (iii) (iv)

I sold the book [PRO to help the refugees] I bought Bill a book [PRO[PRO to give t to Mary]] I got a book from Bill [PRO[PRO to give t to Mary]] I bought a book from Bill [PRO[PRO to give t to Mary]]

In (16)(i), I is the controller. In (ii)-(iv), book controls the PRO in COMP which is the antecedent of the trace t, a variable in LF; PRO in the subject position is controlled by Bill in (ii) and by I in (iii),(iv), … again suggesting that semantic or pragmatic considerations are operative.” (Chomsky 1981: 77) From (i) of (16), the “arbitrary” interpretation of (15) would follow, with the unknown or discourse-related sellers of the books, the implicit Agent as controller. The distinctions between (i)-(iv) of (16) are of quite a different nature from those between (15), on the one hand, and (16) (i)-(iv), on the other. Properties of the embedded verb can contribute to determining control as well as the matrix verb. Consider (17), (18). (17) Henry gave Maryi (many of his books)j (Oj(PROi to read tj)) Oj is the moved null relative operator (or pronoun), tj its trace.

10

Chapter 2 (18) Henryi gave Mary many of his books (PROi to reduce his library)

Cases that cannot be disambiguated without resorting to discourse and situations are quite in order as the Russian example (19) shows. (19)R

On prines kniguj (Oj (PRO he brought a book

pocitat’ tj)) to read

It is open whether he brought a book to read it himself or to give it to somebody to read. The choice is clear in the Russian example (20). (20)R

On prines mnei (svoi stichi) (Oj (PROi procitat’ tj)) [dat] he brought me his poems to read

I will conclude the illustration by two straightforward instances of this type of control, in Russian and English, and a subtle one in Czech. See (21) – (23). (21)R

Jai narocno ne zakavycival eti I on purpose not put into quotes these stroki (ctoby (PROi zaintrigovat’ citatelja)) lines in order to arouse the curiosity of the reader ‘I purposely did not put these lines into quotes in order to arouse the curiosity of the reader.’

(22)

((PROi To tell you the truth) Ii have never really thought of them that way.) (ex. from Stump (1984))

(23)C

Zenui (pro) povezou do Budejovic (Oi (PRO spálit ti)) [acc] [3 ps pl perf.] the woman (they) will take to Budejovice to burn (her) (ex. from Svoboda (1962))

The pro-subject of povezou, which is [+plural], is not necessarily identical with PRO. ‘They will take the woman to B. for her to be burnt there.’ pro refers to a discourse-related domain of people no single person of whom is identified. 2.1.3.2.2. Control in free participial and nominal adjuncts In this short synopsis of the main types of control, it remains to deal with free participial and nominal adjuncts. Let us assume, following Chomsky’s (1992)

Preliminaries

11

suggestion, that conditions of economy in UG determine the notion of “optimality”, when linguistic expressions are considered to be optimal realizations of the interface conditions. “The derivation of a particular linguistic expression, then, involves a choice of items from the lexicon and a computation that derives the pair of interface representations (A-P, C-I) … A-P constitutes the instructions for the articulatory-perceptual system and C-I for the conceptual-intentional system. These are the interface levels.” Optimal pairings should meet the preconditions for warranting meaningful interpretation. If this must be set as the boundaries of the efficacy of economy conditions, a wide range of pragmatic considerations opens in judging fulfilment of those prerequisites. Consider (24), an example from Stump (1985). (24) Noticing that a crowd had gathered, Bill immediately called the fire department. “What relation is inferred to hold between the free adjunct and the superordinate clause in this sentence depends importantly on context. A causal relation might at first glance, seem most plausible … but in the context … a merely temporal or circumstantial relation is inferred for the free adjunct …” (Stump 1985: 21). Or take French examples (25), (26) in which the BE-auxiliary, as in other languages, is dropped: (25)F

Ils esperèrent qu’ ainsi plongés dans they hope that so driven to la misère les peuples chasseront leur dirigents misery the peoples will chase away their leaders

(26)F

Pousseés au désespoir, les victims driven to despair the victims du blocus auraient plutôt of the blockade have rather tendance à se rassembler derière leur leader the tendency to rally behind their leader

With étant (pouseés) dropped, the interpretation does not suffer. It is neither decidable nor subject to grammatical decision whether the intended relation is one of causality, of temporal precedence, condition or concession or some other. Some plausible if not definite conceptual value is required to be inferable which again tends to verge on critical situations in judgments of acceptability.1

12

Chapter 2

Unlike the preceding types of control structures, the relationship of the controlled participial clause to the matrix clause is less influenced by the thematic status of arguments and subject to strongly-marked variability. Its “logical” specification is not guided by grammatical indicators. While there are instances of a typified lexical equipment that can determine the respective ° relationships (cf. Stump (1985), Ruzicka (1982)), the dominant situation is the conceptual-pragmatic construal of (an approach to) the intended relation. What remains, then, for control of the PRO-subject in the participial clause to lean on? “The problems of control in free adjuncts are fascinating and not simple …” (Stump 1985: 8) The natural assumption suggests itself that saliency, a prominent position in the syntactic and thematic hierarchy or in information structure, presents itself as a consistent and most reliable pivot of support for identifying the antecedent of PRO. Not surprisingly, a varying understanding and divergent opinions of standards in correct language use have collided in this peripheral but, in particular genres, very lively mode of structuring expressions. Not only purists may misunderstand acceptability and lay claim to self-declared statutes of correctness of language use. Ideas of control grasping at the “subject” status of the potential controller, get stuck when, for example, in passive structures, the surface subject NP and the “subject” of the action differ. The judgments often run that control structures which cannot hold on to antecedents with a predetermined grammatical status are carried beyond the bounds of acceptability or literary norm (cf. e.g. Ickovic W.A. 1974). These subjective and accidental factors notwithstanding, we find ourselves in an important experimental field for observing how a good balance is achieved between derivations satisfying economy conditions and an outcome that warrants reasonably meaningful interpretation. Subtleties of judgments on grammaticality and acceptability seem to be inevitable concomitants of balancing optimal interface pairings against structural options that differ in interpretive accessibility.

2.2. Approaches to Control 2.2.1. Emergence of Control The concept and term control circumscribe phenomena that have figured in traditional grammar under different notions like, e.g. subject and object infini-

Preliminaries

13

tive, which betray the fact that (the interpretation of) the infinitival VP must have recourse to the higher subject or the object depending on the lexical class of its embedding verb (e.g. promise or persuade). If “the theory of control is concerned with the choice of antecedents of PRO” (Chomsky 1982: 7), PRO in turn is understood as a term coined for the theory of control. PRO is the symbol for a particular empty category which, at the relevant point in the derivation is present in the subject position of nonfinite VP’s at S-structure and/or LF, if no lexical subject can take this place (“obligatory” control). Being invariably an argument (Brody 1993: 2), PRO needs identification with an antecedent or binding by a quantifier. If neither an antecedent nor a quantifier is available, some other interpretation must be looked for, a case which has been treated under the general heading of arbitrary (arb) control. The theory of control is concerned, then, with the nonredundant predictability of identifying or, more generally, understanding what is the content and reference of each instance of representing PRO. Nonredundancy is important for laying claim to a theory of its own. Chomsky/Lasnik (1993: 73) suggest that “… there is considerable evidence for the existence of a distinct Control module in the theory of grammar.” Crucially, nonredundancy requires isolating issues of control from problems of identifying or binding empty categories of types different to PRO. On the general assumption of autonomous syntax, predictability in control has been thought to be achievable by an essentially configurational theory. The question has been whether solutions to the problems raised by the empty category PRO can be attained by subsuming them in theories already existing for other empty categories, or by taking or adapting concepts from them while upholding the claim to a distinct theory of control. I would like to sketch now the generalized main approaches to control with a view to appreciating how they probe into what are the substantive conditions of Control. 2.2.2. The configurational approach 2.2.2.1. Manzini’s theory Manzini’s (1983: 421) declared “essentially configurational theory … (of the) essentially configurational phenomenon (of control) is constructed on essentially the same notions on which binding theory is constructed in Chomsky (1981).” Her unification of “… control theory and binding theory as one theory” (1983: 421) anticipates crucial empirical problems configurational

14

Chapter 2

approaches have to face. She constructs a “geometrical” solution to notoriously critical cases which are left unaccounted for in later modifications of the configurational approach. Take, for example, Iwakura’s approach: “… core cases of antecedent-PRO relations fall under the binding theory , if the ‘binding theory’ version is adopted over the ‘governing category’ version …” (Iwakura 1990: 137) But then, “… It remains unclear how to account for examples like (i)” (Iwakura 1990: 130). See (27) (for Iwakura’s (i) in the above quote). (27) John was promised to be allowed to leave. Clearly, although the fact that “… antecedent-PRO relations and antecedentNP trace relations overlap in core cases” (Iwakura 1990: 121) has justified every endeavor to draw on proven theories, there remain some phenomena undoubtedly falling under Control that cannot be accommodated to principles of binding and/or Government. It seems questionable, then, whether even those that can should be treated by them considering their explanatory value in this field. Let us have a look first at cases of control which are hard to handle “geometrically”. Manzini (1983: 423) notes an obvious problem2 with her descriptive condition stated as the generalization (28), (her (20)), with respect to a PRO in an object sentence S; viz. (28). (28) A PRO in an object sentence of a sentence S is bound in S. Examples (29)–(32) (her (12)–(15)) and (33)–(36) (her (21)–(24)) illustrate her point. (29) John asked Bill [PRO to shave himself] (30) John asked Bill [PRO to be allowed to shave himself] (31) John promised Bill [PRO to shave himself] (32) John promised Bill [PRO to be allowed to shave himself] (33) Bill was asked [PRO to shave himself] (34) *Bill was asked [PRO to be allowed to shave himself] (35) *Bill was promised [PRO to shave himself] (36) Bill was promised [PRO to be allowed to shave himself] Manzini (1983: 423) decides to relegate the obvious problem somewhat vaguely to semantics: “However, with respect to this problem, it appears quite possible to assume that on syntactic grounds all of (29)–(32) and (33)–(36) are indeed well formed with both object and subject control; and that either

Preliminaries

15

subject or object control is excluded for other reasons — say, on semantic grounds.” This distribution of the weight of responsibility for wellformedness is the price to be paid for keeping up the validity of (28) (her (20)).3 (28) holds for PRO in object sentences, qualifying sentence types as (34), (35) — among the others — as syntactically wellformed. PRO has a domain-governing category and is bound in S, by a subject or an object. For PRO in a subject sentence Manzini (1983: 424) proposes the generalization (37), (her (30)). (37) A PRO in a subject sentence (co)refers freely.4 If subject sentences involve configuration (38) (= her (34)), “… where we assume that S’ and Agr are cosuperscripted, by (31) (=(i) of fn (3), RR) PRO does not have a domain-governing category; … (32 (=ii)) then does not apply, and PRO is correctly predicted to (co)refer freely.”5 (38)

S S′j

Inflj (Agr)

VP

S

PRO Manzini’s ingeniously constructed syntactic device accounts for the fundamental difference between control in object sentences and control in subject sentences with no more effort than drawing on the i-within-i condition and its effect in a newly introduced domain-governing category. Two generalized binding conditions, (28) and (37) (Manzini’s (20), (30)) take up notions of binding theory in Chomsky (1981). Under this approach, one must be prepared to judge, e.g. (29), (32) and (33)–(36) to be equally wellformed on syntactic grounds. Once the cost in either empirical adequacy or syntactic wellformedness is tolerated, Manzini’s configurational approach is important since she has laid the emphasis on the gulf in control conditions existing between object and subject clauses. The question is, whether she has put it in its true perspective, that is, whether substantive factors regulating Control must still be conjec-

16

Chapter 2

tured to exist behind her configurational distinction. The import of a fundamental bipartition of control conditions that hinges on the relation of the controlled clause (e.g. subject vs. object sentence) to its matrix clause strongly suggests the likelihood of an account with higher explanatory value. Manzini’s solution is compatible with the account I am going to propose in chapters (3) and (5) and may be construed to be incidental to it. 2.2.2.2. Kayne’s similar approach Kayne (1991: 677) considers his approach to PRO as “… significantly similar to that of Manzini (1983) … in having binding theory determine a governing category for PRO and hence delimit the possible positions for the antecedent of PRO … but (it) has the advantage that there is no need to add to binding theory any notion of domain-governing category.” Kayne’s approach differs from Manzini’s in that the deep division in control between object controlled clauses and subject ones does not come out prominently as it does in Manzini’s paper. Second, Manzini acknowledges syntactic wellformedness of both object and subject control with either subject or object control being excluded for unspecified other reasons. Kayne deals with this problem as a possible theory-internal barrier and seems resigned to not solving it by means of binding theory: “The question also arises whether the binding theory approach to PRO tells us anything directly about the difference between subject and object control … Although such cases are numerous, they are not typical …” (Kayne 1991: 678, fn. 77) What is at stake here, however, can hardly be dismissed as “untypical”, as we are going to show in this study. Kayne, trying to surmount the PRO-dilemma (“PRO is either a pronominal or an anaphor but never both at the same time.” (Bouchard 1985: 471) capitalizes on Chomsky’s (1986b: 169 ff.) slight revision of binding theory, which was meant to account for the contrast exhibited by (39) and (40): (39) the children like (each other’s friends) (40) the children like (their friends) “As the binding theory now stands, it is satisfied by (229ii) (= our (40)) but violated by (229i)” (= our (39)). (1986b: 170) Extending Chomsky’s revision of binding theory clad in Binding Theorycompatibility of indexing (I) and a licensing condition (Chomsky 1986b: 171 ff.) to PRO and control in a way clearly not considered in Chomsky’s (1986b)

Preliminaries

17

“Knowledge of Language”, Kayne (1991: 675 ff.) emphasizes that “the PRO theorem follows from the strict parallelism between Principles A and B of the binding theory. To the extent that strict parallelism fails to hold over some range of environments, the PRO theorem will fail to hold for that range. More specifically, it will fail to hold for any subject PRO governed by a lexical category found within the category of which PRO is the subject, since in such a case the governing category for PRO qua anaphor will not be identical to the governing category for PRO qua pronoun.” In other words, if the PRO theorem is violated only by violating the conjunction of Principles A and B, distinct (local) domains for A and B immunize against such violations. If PRO in particular is lexically governed by the infinitival verb, the IP of which it is the subject is not its governing category insofar as PRO is anaphor, since within this local domain it could not satisfy the role of an anaphor, that is, it could not satisfy the binding theory with some indexing. Thus, the term governing category would be senseless with respect to the anaphorical properties of PRO. The relevant governing category for an expression α has been identified with “… the least CFC (complete functional complex) containing a governor of α in which α could satisfy the binding theory with some indexing …” (Chomsky 1986b: 171) For PRO qua pronoun, however, the IP of which it is subject is a legitimate governing category, and if PRO is lexically bound by the infinitive, there is no violation of the PRO theorem, the IP being the smallest category that contains a subject position and the governor of the pronoun PRO. Kayne’s original construal seems to run the risk of partly voiding the PRO theorem or what remains of it by associating PRO qua anaphor with a governing category within which it could satisfy (contain a suitably accessible binder) as well as violate the PRO-theorem, whereas PRO qua pronoun is given a governing category in which it necessarily satisfies it. This asymmetry is reinforced in Kayne’s modified version of the PRO theorem by virtually excluding PRO qua pronoun from it, which might turn out to be the price of this approach: “… the PRO theorem continues to hold for all PROs other than those that are in subject position and governed by an internal governor.” (Kayne 1991: 679) If such PROs are not governed, they must be made to be governed: “All controlled PROs are governed at some level of representation.” (Kayne 1991: 679, numbered 92) In French, then, for cases like Kayne’s (91) Jean veut aller au cinema (Jean wants to go to the movies), the infinitive must move at LF, by analogy to S-structure movement in Italian, since at

18

Chapter 2

S-structure it does not govern PRO, which means, following Kayne, that PRO has an antecedent only at LF. Kayne gives as “… reason for the existence of (92) … that it is via government that PRO qua anaphor receives a governing category … that an antecedent for PRO must be within PRO’s governing category (that is, that an ungoverned PRO would be able to be associated with any antecedent at all).” (Kayne 1991: 679) These assumptions, while allowing Kayne to accomodate control to the binding theory, seem to beg the question whether control and the “reversed” PRO theorem can still be taken to follow from the conceptual rationale and range of binding. Kayne’s claim, however, about locality of the controller (antecedent), that is, about the local domain of control, is important, since it establishes what might be called an Adjacency principle of control: “In assigning to PRO qua anaphor the next IP up as governing category, the binding theory adopted here excludes the possibility that the antecedent of PRO … could be taken to be a subject NP two IPs up.” (Kayne 1991: 677) I have dwelt on Kayne’s approach because it may be considered to be the most intricate attempt to incorporate important aspects of control into binding theory and to derive variations in the existence of control structures (e.g. between Italian and French) from independent parametric differences of infinitive moving.6 But there are crucial aspects of control about which Kayne’s approach says nothing, in particular the eligibility of the controller, or as mentioned above, the fundamental question of subject and object Control. It is those aspects though not elucidated under this name, which will receive the main attention in this study, again with respect to the problematic claim on a distinct theory of control. Under this approach, we share Chomsky/Lasnik’s (1993: 72) suggestion that “… Control is different enough from anaphor binding that a separate mechanism for antecedent assignment is, in fact, justified.” Their arguments seem conclusive: The optionality concerning choice of binder does not regularly exist for choice of Controller: herselfi  himself  j  (42) Johnj told Maryi (PROi*j to leave) (Chomsky/Lasnik’s modified examples) (41) John j told Mary i about

The Controller is a specifically designated argument, which, in (42), not only can but must be the object-NP, given the membership of the matrix verb in a

Preliminaries

19

certain class. If it is easy to show that Control cannot be subsumed under anaphor binding and there is enough evidence to consider the existence of a distinct Control module or at least the specificity of PRO in the theory of grammar, the exact nature and status of control are still unclear. Some difficulties are particularly conspicuous in Chomsky/Lasnik’s discussion (1993: 81) of the Case problems of PRO ending up in assigning PRO “… a Case different from the familiar ones … it is the sole NP that can bear null Case (though it may have other Cases as well …, which makes it possible to maintain a simplified Visibility condition: (= Chomsky/Lasnik’s (315)) (43) “A chain is visible for θ-marking if it contains a Case-position.” This proposal resumes the thread of suggestions made in Chomsky (1981: 147, fn. 106; 188; 262) concerning idiosyncratic Case assignment to PRO (e.g. in Russian). 2.2.2.3. On the arbitrariness of PROarb “All PRO are controlled if apparent uncontrolled PRO actually has a hidden controller …” (Chomsky/Lasnik 1993: 20), and so-called “PROarb is really a subcase of controlled PRO …” (Kayne 1991: 680; Epstein 1984) The controversial discussion of PROarb is an appropriate occasion for an anticipatory comparison between the spirit of the configurational treatment and the prevailing idea of this approach. Consider an analysis of Brody (1993): “The contrast between (42a) and (42b) (my (44) and (45): RR) … reinforces the conclusion that antecedentless PRO with arbitrary interpretation in fact exists: without such a PRO the lack of a disjointness effect in (42b) would be difficult to explain. (44) *PRO to teach themx math is easy for the childrenx. (45) PRO to teach themx math is useful for the childrenx. In (44) them and the children cannot be coreferential since the latter NP controls PRO obligatorily and Principle B prevents the pronoun from being coreferential with PRO. Since (my emphasis: RR) no disjointness effect obtains in (45), we must conclude that the for-complement does not control PRO — but there is no other potential controller around either syntactically realized or implicit argument. Hence, PRO has no antecedent.” (Brody 1993: 19) I think the argument can be strengthened. The conclusion that antecedentless PRO with arbitrary interpretation exists presupposes that non-

20

Chapter 2

antecedentless PRO exists. The former kind of PRO in (45) is contrasted with the latter in (44). Both instances of PRO, the antecedentless arbitrary and the controlled one, are not independent of the distinct adjectivals, easy and useful, respectively. That is why (44) and (45) differ in their respective control relations, which are the basis of Brody’s argument. The principled common dependence of control behaviour on lexical properties seems to effect the control status of antecedent and antecendentless PRO. The disjointness effect in (44) and the lack of it in (45) are indeed by-products of control conditions. As to the basic motive for Brody’s argument, there can be no doubt that “antecedentless PRO with arbitrary interpretation in fact exists”. The real problems are whether and how “arbitrariness” could be qualified or specified, and more importantly, in which cases an antecedent must be assumed and represented although it has no phonetic shape. By way of a second preliminary illustration of differences between a configurational, in particular Case-theoretic approach and the approach pursued in this study, which does not exclude configurational restrictions, let us take a look at two examples juxtaposed by Chomsky/Lasnik (1993: 71). (46) *It is likely (PRO to solve the problem) (47) It is important [PRO to solve the problem] (Chomsky/Lasnik’s examples (255), (256)) “Since PRO in (255) is not in a configuration of Case assignment (a lexical NP is impossible here), that example might be expected to be grammatical, presumably with an “arbitrary” interpretation for PRO as in (256) … And (257) (our (48): RR) might be expected to be grammatical with an arbitrary interpretation, or possibly with PRO controlled by John, given the general lack (or at least amelioration) of condition A effects in clauses with expletive subjects, as illustrated in (258) (our (49): RR). (48) *John believes [it to be likely (PRO to solve the problem)] (49) Johni believes (it to be likely [that pictures of himselfi will be on display]) If the crucial factor determining the distribution of PRO is government and PRO must be ungoverned (Chomsky 1981) “… this will entail that PRO will not be Case-marked. But the requirement is now broader, since there is government without Case-marking. This is what we find in … (255), (257) … The distribution of PRO is thus correctly described.” (Chomsky/Lasnik 1993: 72)

Preliminaries

21

The ungrammaticality of (46) entails the ungrammaticality of (48). In both “… there is government without Case-marking.” (p.72) Thus, if the difference between (46) and (47) is taken to be accounted for, in (47), PRO is not governed and has no Case. Let us assume that this is a correct configurational and/or Case-theoretic account of the difference. We will try to ascertain whether we can rest assured that this explanation is sufficient and need not be substantiated. Leaving raising out of consideration, I suggest that in (46) the two-place relation of Control cannot be implemented , that is, the identification fails because no pairing of two arguments with identical interpretability is available. Why? Let us take a semantic path. Suppose we interpret arbitrary as discourse-related. Then, in (47), (the set of) individual(s) to whom something, namely the (incidence of) event described in the infinitival clause, is considered to be important can be identical (coindexed) with the PRO argument of this event description. For example, in (47), PRO can have a generic interpretation (For mankind it is important to solve the ecological problems). Or PRO may have an “understood” or implicit antecedent in the matrix clause. The burden of identifying the (domain of) persons to whom something is important rests with the matrix clause. PRO “inherits” its interpretation from the matrix clause. Of course, there is another interpretation of (47) in which the interpretation of PRO is independent of an optional prepositional argument in the matrix clause. Cf. (50). (50) It is important to the preservation of wildlife (PRO to solve this problem). Some such independent “arbitrary” interpretation might appear to be amenable to (46) as well. But it isn’t. This seems particularly surprising in view of the perfectly grammatical (51). (51) It is likely that the problems will be solved. What’s the difference between (46) and (47)? The potential prepositional θ-position in (47) is not available in (46). If we could insert a prepositional phrase in the latter (as in it seems to me), it would be the “psychological” subject of the epistemic attitude expressed by likely, that is, the attitude towards the existence or nonexistence of the state of affairs, the event identified by the infinitival complement. The assessment or evaluation predicate important does not concern the existence of a state of affair but takes it as feasible or given, relating its evaluation to somebody or something as de-

22

Chapter 2

scribed above. It is this lexically optional argument which by virtue of being a potential antecedent of PRO is a prerequisite for, or licenses, the control relation as the identification of two arguments. Thus, PRO is licensed in (47), even with its possible arbitrary interpretation, which hinges on the general precondition of relating PRO to a matrix argument. Differing from (47), no presumable PP in the matrix clause of (46) can control the PRO-argument in the event description of its infinitival complement. 2.2.2.4. Configurational and binding approaches extended The configurational approaches or pertinent aspects of it continue to be pursued. Recently, Larson (1991) has resumed attempts to give a configurational account of control with promise, relying on a Minimal Distance Principle analogous to that of Rosenbaum (1970). I will return to this account in due course (section 5.5) Grodzinsky/Reinhart (1993: 73) do not hesitate to treat PRO — “… ignoring here the issue of the precise interpretation of arbitrary PRO,” — as a bound variable: “All of the anaphoric elements in (8) (our (52): RR) also share a semantic property: they are all interpreted as bound variables. Although on the GB framework this has been explicitly stated only for reflexive anaphors and wh-traces (Chomsky 1982), there is no reason why it should not be extended to control PRO.” (52) a. b. c. d.

Luciei adores herselfi Whoi ti smiled? Felixi was fired ti. Alfredi promised PROi to cook well. (Grodzinsky/Reinhart’s examples)

Although we do not adopt this as a general approach, Grodzinsky/Reinhart encompass an important problem of control raised and discussed by Nishigauchi (1984: 233): “… thematically controlled PRO is essentially a bound variable (in a broader sense of the term than is used in the government and binding framework); but that pragmatically determined PRO shows properties of pronouns, and may serve either as a bound variable or as a deictic pronoun (for this important distinction, cf. Partee 1972, 1975, 1978).” I will return to this problem. Iwakura (1985, 1990) considers the hypothesis that PRO can be divided into two types in a similar spirit. She proposes “… a principled way to account for core cases of antecedent-PRO relations in terms of the binding theory that accounts for the distribution of other types of NPs.”

Preliminaries

23

(1990: 122) Following Bouchard (1984), she assumes that there are two types of PRO, anaphoric PRO (locally controlled PRO) and pronominal PRO (“long distance controlled PRO” and “arbitrary PRO”). On this assumption, “… antecedent-PRO relations fall under the binding theory if ‘governing category’ is replaced by ‘binding theory’”. (Iwakura 1990: 123) Iwakura argues in support of her approach by analyzing a series of control verbs. She adapts Chomsky’s (1980: 33) control rule to suit her analysis, modifying the i-within i condition in the attempt to maintain the configurational essence of Control. Her analyses observe important accompanying restrictions on Control, but I will claim do not concentrate on the substantive conditions of Control. 2.2.3. Questions posed to Control The empirical questions: What constitutes an eligible controller for a given controllee? Note that “eligible” has a double sense: (a) Which of two (or more) NP arguments is suitable to be chosen as the controller? (b) Is the single available NP suitable for control? (a) and (b) collapse to (c): Does an element in the set (possibly the unit set) of potential controllers qualify as the effective controller and how is it licensed?7 The problems posed by (b) are not considered in approaches restricted to the alternative of subject vs. object control, or to a “semantic version of Rosenbaum’s Minimal Distance Principle” (cf. Gazdar et al. 1985: 219). The controllee is “given” in that it is subject to (i)–(iii). (i) Only subjects (at S- or LF-Structure) are controlled. (ii) Only nonfinite clauses have controlled subjects (can be controlled clauses). (iii) The controlled element PRO is the head of a chain. PRO and its controller are not members (a link) of the same chain. They are assigned θ-specifications from distinct θ-grids. “PRO does not participate in Â-chains …” (Safir 1996: 331) 2.2.4. Clausal status of the controlled phrase or the propositional vs. the attributive account of Control It will be obvious that my approach to control implies the propositional account. I would rather endorse the arguments advanced by Higginbotham (1992) in his discussion with Chierchia, which elucidates subtle problems

24

Chapter 2

crucial for preferring the propositional account. I would like to provide some more arguments in favour of PRO, capitalizing first on evidence that acceptance of PRO is consequent on the assumption of traces for Raising. Thus, consider (53), which illustrates an interesting interaction between Control and Raising, when a Raising construction is embedded under a control verb (cf. Jacobson 1992: 188). See (53)–(53′) (53) John tries to seem to be nice. (53′) Johni tries ((PROi to seem (ti to be nice))). Assuming that Raising is involved in (53), the Raised subject must end up as PRO-subject of the clausal complement of try. PRO is the head of the A-chain (PRO, t) and should receive a θ-specification originating from try (see below, 6.1.1.). Although much is unclear, Control and Raising match with each other through the interdependence of the trace and PRO. Second, the appearance of PRO seems imperative if ambiguous long-distance and local binding of reflexives is to be represented in the most natural way. Consider Russian (54). (54) Komendanti prikazal ad-jutantuj ((PROj commandant ordered (his)aide-de-camp sojedinit’ sebjai,j s nacal’nikom štaba)) to connect him(self) with the chief of staff ° (cf. Ruzicka 1973: 444 – 481) Coindexing of the local binding variant requires PRO to be available. Otherwise, this kind of subject-oriented anaphora would not find its local antecedent represented. Most recently, E. Torrego (1996: 119) presents “… several types of evidence, drawn from the morphology, syntax and semantics of the floating unit, that confirm that control clauses have a PRO subject.” 2.2.5. Θ-roles and Control A fundamental result that has emerged in recent work is that control in at least a major class of relevant construction types is not independent of thematic relations meshing with semantic roles. Originating with Jackendoff (1972, 1974) and Gruber (1965), this basic approach was supplemented with, or diverted to, a search for configurational solutions (Chomsky 1980). But Chomsky followed up his ideas with “… the natural suggestion … that choice of controller is determined by θ-roles or other semantic properties of the verb,

Preliminaries

25

or perhaps pragmatic conditions of some sort.” (Chomsky 1981: 76). The convergence on acknowledging θ-roles as a crucial factor operative in major domains of control — an assumption compatible with a more complex network of determining factors — is well documented in the literature of the field. I would like to draw attention to Abraham (1983), Nishigauchi (1984), Koster (1984), Chierchia (1983, 1984), Melvold (1985), Culicover and Wilkins (1986), Rizzi (1986), Vanden Wyngaerd (1987), Williams (1985, 1987, 1989), Jackendoff (1987), and others. Notions basically cognate with the rationale of θ-roles have been advanced by Siebert-Ott (1985) and Wegener (1989), Köpcke/Panther (1991, 1993).8 A crucially distinct treatment of thematic relations in control theory was ° proposed in Ruzicka (1983a,b). Pared down to its basic rationale, the claim was advanced that it is the principled relation between the θ-role assigned to the potential controller argument and the θ-role assigned to the controllee that licenses control over wide areas. Farkas (1988: 54, 57) endorsing the essentially relational character of control, modifies its content in the spirit of situational semantics: “What is relevant to control is therefore not the thematic role or grammatical function of the controller but rather the semantic relation the participant linked to it bears to the participant linked to the controlled constituent.” Likewise, the relational concept underlies Wegener’s (1989) exclusively semantic analysis of control in German infinitival complements. 2.2.6. Specification of thematic properties and Control domains Under this approach, abstract thematic specifications related to a particular conceptual content are taken as the central notion and effective instrument to deal with control conditions over a family of construction types. They operate differently, constituting distinct subtypes that are correlated with lexical subclasses. They can do this, when their content is brought to bear on, and account for, control behaviour by being arrayed in the format of general constraints of a characteristic design, in particular as conjunctions of distinct combinations of feature values. The respective constraints predict the range and degree of interaction with independent linguistic or extralinguistic factors. If the lexical properties of a given class of (control) verbs/adjectives do not lend themselves to be associated with a general constraint of the presumed type, well-formed control is not a function of lexical properties, that is, a restrictive relationship between values of thematic specification.

26

Chapter 2

° Taking up the thread of previous work (Ruzicka 1983a,b, 1986, 1987), I will develop a strongly modified and extended framework, with the particular intention to relate general principles to cross-linguistic variation, considering interaction with independent factors as well as idiosyncratic developments. Control is assumed to be mainly a lexically driven syntactic process. And if “… θ-marking … amounts to establishing a correspondence between syntactic and conceptual arguments of a verb …” (Jackendoff 1987: 409), “… what grammatical relations (alternatively, syntactic positions, RR) a θ-role links to depends on the verb. However, for a particular verb, linking is largely predictable.” (Carrier-Duncan 1985: 6), (Ostler 1979). These ideas suggest chains as the target of control (Chomsky (1986b: 135). Roughly speaking, a mapping relation established between pertinent lexically semantic properties and functional chains can be the target of control conditions. CHAINS consisting of an expletive-argument pair at D-structure Chomsky (1986b: 135), cannot be a target of control conditions which are operative at LF-structure. I will assume that θ-marking includes or consists in transferring θ-specifications from the θgrids of lexical entries to functional chains. The operation of control principles must include LF, if only because of the fact that PRO can be the variable bound by a quantifier. Clearly, if θ-grids are “… devoid of semantic content … just an indexing device” (Jackendoff (1987: 379): Jackendoff refers to Higginbotham (1985: 555)), they can hardly be suitable for sustaining control relations. On the other hand, the traditional global characterizations of thematic roles merging in (names of) semantic roles, such as Agent, Goal, Theme etc., have not proven to be appropriate for a theoretical implementation of the relational concept of control. I deviate here from former work (1983a: 311 ff.). ° In Ruzicka (1987), I suggest that the relevant conditions, that is, the thematic Identity condition and the thematic Distinctness conditions should be replaced by thematic feature specifications. I will take up that suggestion and modify and elaborate it in the much broader framework of this comparative study. Thematic feature specifications in the spirit of Gazdar et al. (1985: 21), i.e., ordered pairs of the form 〈feature, feature value〉, will be the conceptual material out of which the generalizing constraints are constructed. As we said above, the lexical subclasses established on the grounds of properties determining control behaviour come to be definable by pertinent respective constraints associable with each of them. The format of the constraints is such that they can be applied to the pairs of DP’s potentially constituting the control relation, alternatively, to the corresponding chains. The conjunction of values

Preliminaries

27

of the relevant θ-specifications in the matrix and the controlled clause respectively, provides the appropriate format. The crucial relation between the θ-specifications of the potential controller and PRO is particularly effective iff the complement clause of which PRO is the subject is itself an internal argument of the matrix “control” verb, θgoverned and/or L-marked by it. Subject complement clauses of which PRO is the subject, which are external arguments and selected rather by VP, are less able to sustain the relation formulated in the constraints. Subject complement clauses are indeed a case in point: “A PRO in a subject sentence (co)refers freely.” (Manzini 1983: 424)). I will argue, then, that the substantive Control conditions can be expressed by pairings of specific feature values which, while basically grounded on conceptual thematic or semantic content, may function as relations of abstract F(eature) values. Crucially, distinct parameterized language particular properties may react differently to the general constraints of Control.

2.3. Organization of the study and languages considered The study is constructed as follows. Chapter 3 describes the basic theoretical approach demonstrating how it works on a characteristically problematic type of control. In Chapter 4 an illuminating contrast between “object” control verbs is explained on a principled basis. It deals with notorious and controversial control problems, discussing recent approaches and revealing distinctions that disintegrate “object control” but can be captured by an integrated approach. Chapter 5 is dedicated to an account of what is concealed under the summary notion of “subject control”. Chapter 6 deals with the particular control mechanism of try and cognate verbs. In Chapter 7 the special case of control with reflexive clitic impersonal clauses is discussed. Chapter 8 aims at supporting the general approach by discussing classes of control verbs and adjectives which produce “selfcontrol” and render constraints superfluous on the very basis that makes them necessary in other control classes. Chapter 9 deals with the “grain problem” and VP-deletion touching Control, and chapter 10 gives a summary of conclusions and empirical results. I have added a short chapter 11, in which I discuss some adjustments to my approach and to the relevant points of the minimalist program in view of accommodating the facts of Control.

28

Chapter 2

I would like to rely on cross-linguistic data of Russian (R), German (G), English, Czech (C), Italian (I), Polish (P), French (F), Dutch (D), SerboCroatian (S-C), Slovak (SL), Spanish (SP).

Chapter 3

The theory outlined

3.1. Object Control – a pilot demonstration “A recurrent problem in linguistic theory has been trying to provide a principled base for the distinction between subject and object control verbs …” (Comrie 1985: 74)). Under this approach, the substance of the problem is of a different character. I will illustrate the point with the following examples, confronting “expected” object control with subject control exercised by the same object control verb. Consider the Russian examples (55) vs. (56). (55)R

Ja poprosila ego ne byt’ zestokim [femin sg subj(agreem)] [acc] [instrum masc sg] I asked him not to be cruel

(56)

Ja poprosila ego ne byt’ [femin sg] I asked him not to be

iskljucennoj [part pass sg expelled

iz školy femin] from school

Neidle (1982: 410) comments on (55), (56) (her examples (45) and (46): “For prosit’, which seems fairly representative, there is no grammatical control of the infinitive. Consider (45) and (46) … In (45) the subject of zestokim is interpreted as the matrix object ego, while in (46) the subject of iskljucennoj is the matrix subject ja (which is also the semantic object of the verb ‘expel’). (The gender markings on the verb make these the only readings.) Since the interpretation of the subject of the embedded verb depends on the context, the subject is not grammatically controlled. Prosit’ can also occur without any overt object, as in (57) (Neidle’s example (47), RR).

30

Chapter 3 (57)R

on prosil [pro] [nom] [acc] he asked (pro = them, e.g.)

pomolcat’ to be quiet

The absence of a possible controller in (47) again shows that there is no grammatical control.” Neidle explains the issue away, that is, out of grammar, giving it over to some unspecified notion of context. As concerns (57), it is irrelevant to the current argument that prosit’ “… can also occur without any overt object”. What will deserve discussion is the fact that the covert zero object is syntactically active in Russian and supplies a licit controller (cf. section 3.3.). The difference between (55) and (56) can be accounted for within grammar, as we will show presently.

3.2. Constraints on “object” Control 3.2.1. Shape of the constraint As suggested above, I propose that general constraints can transform and transmit lexically determined control properties to their syntactic effects, establishing and defining lexical classes with respect to those inherent properties. The force of the respective constraints, then, should extend and generalize beyond language particular variation in control relations to cross-linguistic, possibly parameterized, distinctions subsisting within the respective lexical classes that are associated with control properties. This way of displaying the lexical-syntactic interface in control conditions, including, occasionally, pragmatic effects, will make transparent that neither a purely syntactic-configurational approach nor a purely semantic one can give an adequate account of control. The relevance of thematic properties to control is embodied as a feature specification whose (feature) value is expressed by the variable α ranging over {+, –}. The specific constraint associated, for example, with the lexical class containing ask, R. (po)prosit’, order, R. prikaz(yv)at’, C. rozkázat, persuade, R. ugovorit’, C. umluvit’ and other verbs has the format of a conjunction of two thematic feature specifications, one referring to (the chain of) the licit controller NP, the other to (the chain of) PRO-NP. Quasiindices CON and PRO mark the respective conjuncts as regards the respective DP-targets. Intuitively, constraints on control in the area of control verbs to be discussed are constraints on

The theory outlined

31

relations between θ-specifications of the controller argument and the controllee. They are part of the lexical information of the pertinent verb (class) and are best applied to functional chains at LF-structure. Assume that Int(entional) Act(ion) can be taken as the interpretation of the relevant feature specification and enters into a constraint as (A). (A)

〈intact, ~α〉CON & 〈intact, α〉PRO

Intentional action is a minimal characterization which I interprete in the spirit of D. Davidson (1980: 44 ff, 100, 101, 61): “To intend to perform an action is, on my account, to hold that it is desirable to perform an action of a certain sort in the light of what one believes is and will be the case … what is believed about the future course of affairs, …”9 Thus, e.g., if one believes that an act of sneezing can be performed intentionally, a sentence like (58) is not bizarre. (58) He asked himi [PROi to sneeze]. If α = +, (58) satisfies (A). Choosing the negative value of α will license, e.g., (59), Manzini’s (1983) example (13). (59) Johni asked Bill [PROi to be allowed ti [PROi to shave himselfi]]. “In (13) PRO is coreferential with the most distant controller, not with the nearest one … it appears quite possible to assume that on syntactic grounds (all of) (12) (my (60), RR) - (13) … (60) John asked Billi [PROi to shave himself] … are indeed well formed with both object and subject control; and that either subject or object control is excluded for other reasons — say, on semantic grounds.” (Manzini 1983: 423, 429)), (cf. above, 2.2.2.1.). Indeed, there is no syntactic way-out. The quotation from Manzini (1983) states the issue clearly, if indirectly: Can the relatedness between lexical properties and syntactic positions and domains be captured and systematically generalized beyond vague hints at semantic factors? Constraint (A) and the following ones, together with independent settings of parameters and clearly defined pragmatic factors, I would like to claim, can do this work. Applying (A) to (55), I assume that the singleton chain ego (him) carries the thematic specification satisfying 〈intact, ~α〉CON, and PRO is assigned 〈intact, α〉 (α = +).

32

Chapter 3

The former specification is projected from (the θ-grid of) (po)prosit’, the latter from zestokij, characterizing its intentional meaning, analogous to Polish (61) and Russian (62). (61) pro poprosiÑem Janai [PROi by´c [1 ps sg pret] [acc/gen] I asked John to be (Example from St. Franks)

szczerym ] [instr masc] sincere

(62) Maša

poprosila Vanjui [PROi spat’ golym] [acc] [instr masc] Masha asked Vanja to sleep naked

Example (56) obeys (A) with α = –. The licit controller (ja) carries intentional action (α = –, thus ~α = +). The chain consisting of the head PRO and its trace is assigned 〈intact, –〉 (α = –). I repeat (56). (63) Jai poprosila ego [PROi ne byt’ iskljucennoj ti [sg fem pret] [sg fem pass] I asked him not to be expelled školy] school

iz from

Clearly, constraints like (A) can be taken here as matching conditions requiring that designated pairings of values of a given feature specification are observed. With the variable α = {+, –} the shape of constraints analogous to (A) extends to, and exhausts, the relevant pairs of matching values. (B)

a. 〈F,α〉con & 〈F,α〉pro b. 〈F,-α〉con & 〈F,α〉pro

The Control feature F, interpreted as 〈intact, (α)〉, is available in all the hypothesized constraints. It has proven workable in a family of distinct lexical classes of Control verbs, where in every case both predicates, the Control verb and the complement verb discharge the respective contents of 〈F,α〉. Distinct Control behaviour of the lexical classes of Control verbs which are embraced by (B) is captured by the different values of the feature specifications as they are distributed among the potential Controller and PRO. I will claim that in this way intrinsic unified Control principles can be kept up which establish the interface between lexically semantic restrictions and their syntactic observance.

The theory outlined

33

Working in the traditional framework of θ theory, we would have to assume that the Agent role entails 〈+F〉, that is, 〈intact, +〉 and each of the rest entails 〈–F〉, 〈intact, –〉. Notice that no theta-theoretical redundancy is concealed here. I will anticipate three reasons: (a) Primary theta specifications are never given as their negative value. (b) Theta specifications contained in a theta grid cannot project into a clause whose theta roles are determined by the distinct theta grid of this clause (see chapter 6). (c) Theta specifications assigned to an NP from a given theta grid must tolerate to stand side by side with a contradictory thematic specification that licenses Control (see (G), in 4.1.). More motivations for the autonomous operation of the feature specifications used in the Control constraints will emerge in the following discussions. Spelling out the Control constraints (B) gives (C), with the matching effect produced by the distinct pairings of values of α. (C): a′ and a″ covers (B,a): a′ +Fcon & +Fpro a″ –Fcon & –Fpro b′ and b″ covers (B,b): b′ –Fcon & +Fpro b″ +Fcon & –Fpro The assignment of distinct values to α carries an important conceptual burden. The order of the indexed conjuncts (con, pro) in each conjunction is of course irrelevant. We are left with two constraints (B,a) and (B,b), which will be associated respectively with distinct nonintersecting sets of lexical classes of Control verbs. The application of the constraints consists in checking the potential Controller-DP’s and PRO. Importantly, differences of control behaviour within a lexical class, whether subject to parametric variation or not, and however related to other modules of grammar or pragmatics, are accounted for by intermediate construals which in turn are referred to by the respective constraints. Further, if control behaviour in infinitival complements (or adjuncts, cf. above 2.1.3.2.), is not subject or amenable to the constraints, it should be “computable” by independent syntactic, semantic or pragmatic considerations. Notice that I will occasionally continue rendering the constraints with the Control feature F represented as its interpretation, that is 〈intact, α〉con/pro.

34

Chapter 3

3.2.2. Markedness as opposite feature values At least two questions immediately arise, on which I will have to expand. (1) How is the apparent stipulation of reversing and using the respective values of α in (B) empirically justified? (2) More importantly still, can the parallelled value pairs (of α) in the conjuncts of (C), that is, in (a′), (a″) and (b′), (b″), be functionally implemented by being adjusted to the notion of markedness? Suppose constraints (B a,b) are considered unmarked if α = +, and marked if α has the negative value. Then: (55) satifies the unmarked version of (B,b), (56) and, e.g. Manzini’s (59), the marked one. Note that in (55) as well as in (56), the subject NP, by projection from the lexicon, is marked 〈intact,+〉 and the object-NP receives 〈intact, –〉. Thus, only in (55), with PRO being assigned 〈intact, +〉(PRO), the matrix object NP is a licit controller sanctioned by constraint (C,b′), (α = +). In other words, the “natural” conceptual-semantic identification of the person(s) to whom the request is directed with the person(s) intended to come to comply with it substantiates the control relation in (55). In (56) or (59), however, the controllee PRO does not refer to the person that is intended to comply with the request, and, consequently, the natural identification cannot sustain, or transfer to, the control relation. Any interpretive construal of (59), of course, includes an entity that represents the person designated to comply with the request. In (55), but not in (59), this entity is syntactically present as the PROsubject of the clausal complement. Thus, in (56) or (59) the direct transfer from lexically determined relational properties of ask, prosit’ to their syntactical control relations cannot materialize. But, as indicated, (59) displays a licit control relation, which should assert itself on other “natural” grounds in satisfying constraint (C,b″). To account for this and a wide array of facts, I would like to propose corollary (D). (D)

A non-singleton chain headed by PRO can carry the feature specification value 〈intact, –〉 (〈F,–〉) iff the theta grid (or – inventory) pertinent to the controlled clause domain contains 〈intact, +〉 (〈F,+〉).

(D) is a specification on the marked application of (B,a) and (B,b), that is, on all cases in which the chain headed by PRO does not receive 〈F,+〉 and, thus, cannot meet the unmarked version of (B,b). (D) requires 〈intact, +〉 to be

The theory outlined

35

contained in the theta-grid of the complement verb for the marked application of constraint (B,a) and (B,b) to license wellformed control. As we shall see later (D) extends its force to the positively set parameters (G) and (H). I will discuss now and motivate its particular application to (C,b″). If PRO is moved and heads an A-chain, it cannot transfer a theta-specification 〈intact, +〉pro, which is assumed to be linked to D-structure subject position. If the moved DP, that is PRO, does carry 〈intact, +〉, it must come from somewhere else, non-locally. Now consider (64). (64) Hei asked him ((PROi to be believed (ti to be his follower))) Assume that the chain headed by PRO is theta-marked by (be) follower of … with the coda of the chain being in the external argument position of the sentential argument theta-marked by believe. The theta-grid of believe contains the theta specification which is linked to the “subject” of the epistemic predicate, say Experiencer. In (64), the recipient of the putative specification is implicit. The matrix object him is very likely to corefer with, or to be included in the referent of, the implicit nominal argument of believe, depending on discourse-related factors. Note that if this is the intended coreference relation, it would embody the identification which in active embeddings of believe emerges as the unmarked control relation licensed by (C,b′). (65) He asked himi (PROi to believe that …) In (64) it manifests itself as an at least conceptually suggested coreference that is backed by the incidental work of the control constraint (C,b′): 〈intact, –〉con & 〈intact, +〉pro. I have tacitly assumed that the implicit believer-argument of believe receives 〈intact, +〉, which in conjunction with 〈intact, –〉 of him would satisfy — in the active configuration (65) — the constraint (C,b′). If we are not prepared to acknowledge the need for providing PRO in (65) and, consequently, the corresponding implicit argument in (64) with the specification 〈intact, +〉 (〈F,+〉), we might lose what can be considered a prerequisite for a rational interpretation of ask (ask somebody to resolve to believe). I will leave this case of (co)reference solution for the moment. Control in (64) is instantiated in its marked version. (C,b″) licenses he as controller. What is problematic is the thematic specification of PRO that is relevant to control. A-moved PRO carries along the assignment it receives from be follower, presumably 〈intact, +〉. Consider (66).

36

Chapter 3 (66) He asked himi (PROi to be his follower)

It does not seem relevant to control whether we assume 〈intact, +〉 or 〈intact, –〉 to be the correct assignment to PRO. If, as I claim, in (64), he controls PRO, satisfying the marked version of constraint (B,b), PRO must be assigned 〈intact, –〉pro. Then, we either have an apparent counterexample — if PRO continues carrying 〈intact, +〉 —, or an artificial assignment of 〈intact, –〉 as the correct specification of PRO simply begs the question. Such documentation as (64), however, can be interpreted as showing that movement to a particular landing site of a theta-marked DP, which must be a non-theta position, affects its thematic “understanding” and, consequently, its sensitivity to control. I would like to add (E) as a subsidiary to (D). (E)

Any PRO that is the non-thematic head of the A-chain produced by passive movement is thematically specified like a moved object-DP.

In (64), which is an instance of (E), PRO is marked 〈intact, –〉pro, which is generally transferred from (with) the moved object-DP, although, originally, it is rather marked 〈intact, +〉 by be the follower of. It seems natural that derived passive subjects coming under control conditions function homogeneously in so far as they can override transferred (non-local) thematic features — as 〈F,+〉 for PRO in (64) which would not obey marked (B,b) — turning them into 〈F,–〉 typical of their landing site. How is (D) met in (64). The specificatiom 〈intact, +〉, which PRO receives in its original position from (be) follower of …, does not count since it is not taken from the θ-grid of the predicate (verb) of the controlled clause domain, namely believe. As suggested above, believe, if assigning Experiencer, can arguably be assumed to discharge 〈intact, +〉 (〈F,+〉) as the specification sensitive to control.10 In other words, (control) verbs that are associated with constraints (B,b) may be assumed to describe events that are subject to potentially intentional action. Believe-events can reasonably be assumed to be not necessarily nonintentional. Intuitively, the critical boundary runs between believe and, e.g., sneeze, resemble.

to resemble his father  (67) He asked him i (PRO i  ) to sneeze  If somebody thinks that in some “world” resemble or sneeze may be subject to intentional action, the um constraint (B,b) can be referred to, licensing (67).

The theory outlined

37

Further, were it not for the conditon (D), (67) could meet the marked constraint (C,b″), with he licensed as controller. Marking the subject argument of resemble or sneeze as 〈F,+〉 is taken to be a decision exclusively at the pragmatic level, which would lead to the acceptability of (67), since the pragmatic marking in turn is referred to by the grammatical um constraint (B,b). Thus, (D) distinguishes, e.g. (64) from (67). In (64), 〈intact, +〉, as required by (D) and (E) is linked, possibly, to the argumental passive morpheme of the controlled clause as proposed by Jaeggli (1986) and Baker et al. (1989). (67), since there is no position for 〈F,+〉 to be projected, is ruled out grammatically, if, possibly, acceptable at a conceptual-pragmatic level. Not much of a serious problem seems to be left. Now consider (68). (68)

?*He

asked him (PRO to receive the whole sum)

PRO, if assigned GOAL, by (D) does not receive 〈intact, –〉, since the thetagrid of receive lacks 〈intact, +〉. Thus, (68) is excluded by (C,b″), since it cannot obey (D). Nor can it meet the um version (C,b′) of course. (68) does not warrant a well-formed interpretation, or it is semantically undefined. Corollary (D) applies to unaccusative verbs as well as to passive verbs in the controlled clause, which is not surprising if, following Belletti (1988: 6) “… verbs with passive morphology can be considered unaccusative in the same theoretical sense as lexical unaccusative verbs — namely, passive verbs do not have the capacity to assign structural accusative.” Assume that PRO is the head of the relevant A-chain that carries 〈intact, –〉. Then, again, condition (D), will prevent the marked application of (B,b) from licensing control, as the theta-grid of the unaccustive verb does not contain 〈intact, +〉. Recall that the thematic feature specification pertinent to control should be closely associated with the theta-grid of (classes of) verbs. Agent, e.g., could be assumed to entail 〈intact, +〉. (69) is excluded by the um(j,j) and, by virtue of (D), by the m(i,i) application of (B,b). (69)

*He i

asked himj (PROi,j to arrive home)

Consider now (70) versus (71). (70) We asked him (PRO to rest from his strict training for two days) Unmarked (B,b) is met with idiomatic rest assured if PRO is marked 〈F,+〉. Since, in (71) PRO cannot carry 〈F,–〉 by virtue of (D), (71) is ruled out by m (B,b) or rather already by um (B,b).

38

Chapter 3 (71)

?We

asked him (PRO to rest assured that we will do all we can).

Compare further (72) and (73) with (74). (72) *We asked them (PRO to come of age) (73) *We asked him (PRO to come of a good family) (74) We asked him (PRO to be patient) Clearly, if Agent entails 〈intact, +〉, the latter does not entail the former, as in (74). The control principles cannot be reduced to θ-roles (cf. above 3.2.1.). An hierarchical restriction is imposed on marked applications of (B,b). (F)

〈intact, +〉con is required to be carried by a DP in (D)-structure subject position.

The hierarchical condition could be expressed as m-commanding of PRO by the controller. (75) *The coachi was asked ti by Peterj (PROj to be included in the select team) The transparent linking of 〈intact, +〉con to the basic subject position must be maintained to be targeted on by the “weaker” marked applications of (B,b). As we shall see later (5.1.), a DP in adjunct-(argument) position that carries 〈F,+〉 may be accessible to the um application of constraint (B,a), which is associated with other lexical classes. Clearly, markedness distinctions related to the projection of thematic properties into syntax will prove to be sensitive to parameterized variation. 3.2.3. Unmarked and marked Control constraint as a functional instantiation of lexical features Lexical information, however represented, for a verb of the class associated with constraint (B,b) predicts that two pairs of “arguments” are typically involved in the ask- (and its ilk) event descriptions, with the arguments of either pair ocurring in the matrix clause and the complement clause, respectively. Importantly, the arguments of either pair have restrictive (co)referential properties for independent conceptual reasons, which are construed so as to function as control relations by virtue of their syntactic distribution in control configurations. The first pair consists of the argument associated with the

The theory outlined

39

addressee of the request and the argument associated with the person(s) expected to comply with the request. The former is linked to the direct objectDP or, in the passive, to the A-chain headed by the subject-DP binding its trace in object position. The latter is projected as the PRO-subject in nonpassive infinitival complements. The clear intended coreference of the two DParguments functions as control relation licensed by the um application of (B,b) if the first argument is rightly characterized as 〈intact, –〉 and the second (PRO) as 〈intact, +〉. The second pair consists of the argument referring to the person(s) who do(es) the asking, and of the argument presumed to represent the benefactive of the fulfilled request. The latter is very likely to be coreferential with the former if embodied as PRO in a passive infinitival complement clause. This coreference, again predictable from the lexical structure of askverbs, is employed as the control relation to be licensed in turn by the marked application of (B,b) together with corollaries (D) and (E): The matrix subject is assigned 〈intact, +〉, PRO receives 〈intact, –〉. With respect to the unmarked constraint (B.b) in passive complements, the lexically induced coreference pairing the addressee-DP and the Agent-DP, which backs the um application of (B,b), has no syntactic counterpart through which it could be channeled into the (unmarked) control relation: The relevant DP involved in the pairing cannot be PRO in passive complements. The principled distinction between unmarked and marked applications of the constraint (B,b) clearly correlates with lexical and syntactic facts including considerations of economy: The um version can be met by syntactically less complex structures, that is, active and passive matrix clauses and nonpassive complement clauses. It is substantiated by the prominent lexically induced coreference condition between arguments. The marked version is met by syntactically less transparent control relations, mainly occuring in passive controlled clauses. They are substantiated by the less conspicuous lexically induced coreference pairing of arguments. Recall that the marked application of (B,b) needs to be specified by corollaries (D) and (E). The correlation with independent thematically induced coreference provides strong factual and conceptual support for the chosen constraint type analysis and its markedness specification. We shall see that this approach maintains its strength in other domains of control.

40

Chapter 3

3.2.4.1. Interpreting and extending effects of the markedness distinction in constraint (B,b) Overall and parameterized effects on well-formed control relations correlate with, and are predictable by, the observance and violation of the unmarked and the marked version of constraint (B,b), respectively. Of course, violations of the unmarked version would formally coincide with violations of the marked one. Violations of unmarked (B,b). α β

〈intact, –〉CON & *〈intact, –〉PRO *〈intact, +〉CON & 〈intact, +〉PRO

Violations of marked (B,b), that is, (C,b″). α′ β′

〈intact, +〉CON & *〈intact, +〉PRO *〈intact, –〉CON & 〈intact, –〉PRO

The offending part is marked by an asterisk. Can the undesirable double be rationalized or should it be removed? Assuming that a subset of the set of control verbs associated with the general constraint (B,b) avails itself of both of its versions, e.g., the ask-type subset, it is natural to expect that the case of straightforward ungrammatically be linked with violations of its unmarked version. And it seems equally natural to expect that violations opening particular parametrized variations, possibly at the pragmatic level, rather result from disobeing the marked constraint. We shall see presently that not only ask-type verbs support this account by violations like α and α′, respectively. Thus, this system selects α and α′ as relevant violations, that is, those with the offending PRO-part. The theoretical decision is richly supported empirically. The contingency of violation is considered a function of the thematic status of the controller (〈intact, α〉CON) of PRO. β and β′ seem to be rendered superfluous. Note that they are related to um (B,b) and m (B,b) respectively. If (violations of) control is encoded in this way, that is, as the respective offending values of the PRO-conjunct in (C,b′)=α and (C,b″)=α′, a depth of empirical explanation is achieved to the extent that wellformedness of control can be systematically related to specified aspects of lexical semantics. Both versions of constraint (B,b), in connection with their respective violations α and α′ mediate between syntactic control conditions and lexically defined classes (and subclasses) of control verbs. As will be shown, transitional phenomena can be located within the boundaries set up by the constraints.

The theory outlined

41

3.2.4.2. Violations of constraint (B,b) As mentioned above, violations of the unmarked constraint (B,b), that is, instances of α can be expected to lead to ungrammaticality, while violations of type α′ still may not preclude control, given specific settings of independent parameters that conspire with semantic construals to license control relations. Consider (76) satisfying um constraint (B,b), and (77) violating it. (76)R

Odin iz nich vstaet i prosit sudejj [[PROi [gen] [acc-gen pl] One of them rises and asks the judges prigovorit’ drugogo k smertnoj kazni]] [acc-gen] [dat] to sentence the other to capital punishment.

(77)

*Odin iz nich vstaet i

prosit sudeji [[PROi byt’ [acc-gen pl] One of them rises and asks the judges to be prigovorennymii ti k smertnoj kazni]] [part pret pass instr(umental Case) pl] sentenced to capital punishment

PRO is assigned 〈intact, –〉 in (77) in accordance with corollary (D). (77) can be interpretable, if bizarre, with “subject”-control, that is with pro controlling PRO, when the participle has singular agreement. (78) [Odin iz nich]i vstaet i proi prosit sudej One of them rises and asks the judges [[PROi byt’ prigovorënnym ti k smertnoj kazni]] [part pret pass sing masc] to be sentenced to capital punishment Under the present theory, (78) satifies the marked version of constraint (B,b). Recall (2.2.2.1.) Manzini’s (1983: 423) judgment on her (12)′, (13)′ repeated here. (79) John asked Billi [PROi to shave himself] (80) Johni asked Bill [PROi to be allowed ti [PROi to shave himself]]11 The matrix passive counterpart of (79), with 〈intact, –〉 assigned to the chain (Billi ti), is wellformed according to um constraint (B,b).

42

Chapter 3 (81) Billi was asked ti [PROi to shave himself] (Manzini’s example (21): Manzini 1983: 423)

However, recalling condition (F), there is no well-formed passive corresponding to (80), a control structure that could obey the marked constraint (B,b). (82) *Billi was asked ti (by NPj) [PROj to be allowed to shave himself] (j ≠ i) The markedness characterization is supported by this gap. Constraint (B,b) could be met only by an “Agent” by-phrase, bearing 〈intact, +〉. But the “marked” control relation cannot materialize without syntactic “superiority” as observed above. Thus, the controller may be required, in addition, to m-command the controllee, which seems quite natural for a secondary control construal founded on a specification, say BENEFACTIVE, that ranks lower on the thematic hierarchy. Note, however, that an antecedent NP within a byphrase or its cross-linguistic counterparts is not generally excluded in specifications on control constraints (see below, 5.1.), which will lead to a broader generalization. The shift in control as accounted for by the um and the m version of constraint (B,b), with theta-specifications 〈intact,α〉 assigned to the respective chains, seems to be perfectly natural. See some more illustrations from German in (83) – (85). (83)G

Eri bat sie ((PROi schon 5 Uhr he asked her already at five o’clock geweckt zu werden)) woken to be ‘He asked her to be woken already at five o’clock.’

(84)G

Ai bittet B [PROi ti mitgenommen zu werden] A asks B to be given a lift

Bi wird (von Aj) ti gebeten [PROi ihnj B is (by A) asked him mitzunehmen] to give a lift (84) and (85) are from Wegener (1989: 206 – 208)) (85)G

(85) satisfies the um-constraint, namely C, b′ (see 3.2.1. and 3.2.2.), (83), (84) satisfies its marked version.

The theory outlined

43

3.2.5. Violating the m(arked) constraint and cross-linguistic variation. Multiple options Ask and the like evidence violations of marked type (B,b). 〈intact, +〉CON & *〈intact, +〉PRO (= α′ in 3.2.4.1.) They do not rule out wellformedness everywhere. Instances of the control verb ask and its cross-linguistic equivalents that violate m constraint (B,b) either emerge as perfectly wellformed, though marked to such an extent that they have been thought to exhibit a lexical homonym, or they are downright ungrammatical. The former situation is found in English and in Spanish. The latter obtains, e.g. in Russian and German. E.Bach (1979, 521) considers ambiguity in (86)a. and b. (his (36)a and (36)b). (86) a. They asked the teacher to go to the bathroom. (“ambiguous for some”) b. The teacher was asked to go to the bathroom. “Ambiguous verbs are disambiguated in the passive.” (Bach) And “… according to Radford (1981), the preferred interpretation of (i)b for many American speakers would be one in which John is understood as the antecedent of PRO … (i)b. John asked the teacher (PRO to leave early).” (Iwakura 1985: 48)) The disambiguation, e.g., in (86)b is accounted for by the condition (F) on m applications of (B,b). The problems posed by the foregoing instances of control can be systematically interrelated. In English, ambiguous (following Bach) (86)a either satisfies um constraint (B,b) with the teacher controlling as in the unambiguous (86)b, or, in the other correct reading, it violates m constraint (B,b), since the relevant feature value of the PRO-conjunct is positive. What appears to weaken m (B,b), in fact reveals its pertinence to the solution to a crosslinguistic puzzle of control: Ambiguity, as in (86)a, disappears not only in the passive but also in the absence of an overt object. (87) Johni asked (PROi to leave) ° (cf. Ruzicka 1983: 319)). Disambiguation here proceeds in the reverse order. The only reading of (86)b is excluded. The character of the null object will turn out to be a crucial factor. Exhibiting ambiguous control in (86)a, English seems to join with, e.g.,

44

Chapter 3

Spanish in contrast with German and Russian. The latter shun this ambiguity. Subject control that violates the m version of (B,b), as in one reading of (86)a and in (87), as opposed to subject control obeying it, as in (64), is completely ruled out. As to Spanish, numerous informants do not agree with Chomsky’s (1988) judgement that “… sentence (61b) (= (88)) in Spanish has a single interpretation: María must be understood as the subject of hablar. Here the subject of hablar in the mental representation is PRO, and it is a semantic property of the verb pedir that the PRO subject of its complement must be bound by the subject of pedir, rather as in the English constructions ‘María asked permission of Juan to speak to the boys’, where María must be the subject of speak. Note that the situation in English is different. Here, Juan would normally be taken as the subject of speak that is, the antecedent of the PRO subject of speak, in (61b). The verbs pedir and ask are thus slightly different in their syntactic-semantic properties. (88) María le pidió a Juan [hablar con los muchachos] María him asked to Juan [to speak to the boys] ‘María asked Juan to speak to the boys’.” (Chomsky 1988: 127, 128)). Ambiguity of (88), as in (86a), seems to be sufficiently supported by judgements, which can be systematically traced to the complex working device (B,b).

3.3. Grammar and recourse to pragmatics This somewhat bewildering picture raises the issue of selecting the appropriate theoretical tools that could capture the greatly diverging cross-linguistic facts. The conflicting effects of the violation of m (B,b) strongly suggest that to the extent that they significantly differ from each other they might provide criteria for deciding whether and at which point to cross the boundaries of grammar proper. Importantly, if violation of m (B,b) marks a crossroads at which different “solutions” can be chosen, one seems to feel justified in crossing from grammar to pragmatic considerations. A fair justification could arise if the specific setting of a syntactic parameter prevents a given language from developing a particular pragmatic pattern. Assume that in Spanish a pattern of conventionalized politeness has developed that allows the control

The theory outlined

45

relation as illustrated by Chomsky in (88). The verb pedir has “withdrawn” to a condensed conventional pattern of use which is essentially the equivalent to (89), in terms of which its control relation can be reconstructed (recall Chomsky’s paraphrase of (88): María asked permission of J. to speak …”). Thus, pedir may “include” ask to be allowed.

PERMIT  (89) X j asked SOME PERSON(S) Yi [PRO i TO   {ALLOW}   Xj [PROj to leave]] The first Xj and PROj display the control relation which ends up and is concealed in (88). English, in particular American English, giving way to the same option, maintains, if ambiguously, the standard um control, as shown in (86)a. But, importantly for the discussion to follow, English has to keep to the “Spanish” option if no overt direct object shows up. Why is English impervious to object control in this case, e.g. in (90). (90) *He asks ei [PROi to leave] Within the settings of the null object parameter, a sweeping argument could be that “… languages those like English … do not allow null objects of any type.” (Cole 1987: 603, 608)). Specifically, “… no structurally represented zero object is possible in English — that is, no structural position is available between the verb and the control clause … In object control structures the object NP must be structurally represented: a direct object controller cannot be omitted.” (Rizzi 1986: 503)). Rizzi’s version of “Bach’s generalization”12 can account for (90), in conjunction with the availability of the construal type (89) as in Spanish. In his comparative study, Rizzi observes that “… Italian allows null V-governed NP-positions carrying arbitrary interpretation or functioning as expletives, whereas English lacks both options.” (Rizzi 1986: 546)). The Italian parametric contrast to English is mirrored by Russian and German. I suggest that the English value of the null object parameter, as opposed to the respective German and Russian settings, is responsible for the incisive difference in control behaviour among these languages. Object control, including null objects, is the only option available in German and Russian. In English, following Rizzi’s assumptions (1986: 552), if the direct object theta-role is saturated in the lexicon and, thus, not syntactically visible, it cannot act as a controller, or, alternatively, if it is not saturated in the lexicon, it could control, but English has no appropriate null filler for the direct object position which is

46

Chapter 3

obligatorily projected by the Projection principle. In German and Russian, the theta-role of the null object is structurally projected, syntactically noninert, and representable as governed pro, its appropriate null filler. The upshot of these facts is that a zero direct object is never capable of controlling in English, whereas in Russian and German it is undistinguishable in control behaviour from its overt counterpart. It follows that there can be no German or Russian control structure correspondent with English (90), repeated here in its grammatical version. (91)

Hei asked [PROi to leave]

(92)G

Er bat [pro]i [PROi zu gehen] He asked prox [PROx to leave]

(93)R

On prosil [pro]i [PROi uchodit’] He asked prox [PROx to leave]

If we want to render (91) into German, the nearest equivalent is (95), a structure that joins in with (59) (repeated here) in satisfying m constraint (B,b); dürfen translates as be allowed/permitted to VP. (59)

Johni asked Bill [PROi to be allowed ti [PROi to shave himselfi]]

(94)G

Johni bat Bill [PROi sichi rasieren (PROi) zu dürfen … himself shave to be allowed

(95)G

Er bat, gehen zu dürfen

(95) has the simplified structure (96), in which the first PRO carries 〈intact, –〉, which is assigned from dürfen or via its decomposed lexical representation ((BE) ALLOWED __). (96) [Eri bat proj [PROi [VP [S’ PROi gehen] [ he asked to leave

zu dürfen]]]] to be allowed

The second PROi, the external argument of gehen, in turn matching um (B,b), is controlled by the S-structure subject of dürfen, which corresponds to the object-DP of allow. I cannot let it go unnoticed of course that informants can be found that would not reject the “English” interpretation of (92), that is, subject control, in addition to null object control. This contingency again sheds light on the conflict indicated with respect to (88) above and immediately below. Subject control is ruled out in (97), an example from Comrie (1985); his numbering is (20).

The theory outlined (97)G

47

Die Mutter bat proi [[PROi das Geschirr abzuräumen]]

“The German sentence (20) has an interpretation not available to its literal English translation, ‘Mother asked to carry away the dishes’” (Comrie 1985: 50)). Conversely, (97) does not share the only interpretation of its English word by word translation: “Mother asked for permission that she (Mother) be allowed to take away the dishes”. (97) is not ambiguous, I differ with Comrie on the judgment of (97).13 The only interpretation of (97) is (in Comrie’s paraphrase): ‘Mother requested of someone unspecified that that unspecified person carry away the dishes.’ If the equivalent of (97) in Italian is ambiguous, as A. Belletti and L. Rizzi think (p.c.), we can conjecture that the availability of pro reveals a further control parameter: pro behaves as nonempty DP’s do in obeying the um constraint (B,b), that is, it must control as in Russian or German. Or pro can give way, if not preferably, to the English patterns in which controlling pro is not available, or the Spanish pattern, in which a pro-object as well as an overt DP-object are compatible with subject control. These variations should not come as a surprise in strongly conventionalized utterance types that are sensitive to idiomatic moulding. Note that this varying control behaviour is not left unrelated to principled control constraints but is encompassed and in turn referred to by them. Control, which we try to bring as near as seems possible to grammatical principles, is a phenomenon which tends to yield to conceptual-pragmatic construal and reinterpretation, in particular in conventionalized utterance patterns of daily use. The control instance dürfen reveals the dominance of the relevant semantic theta-specification over configurational control conditions. Consider (98), which differs from (99) only in the modal verb. (98)G

Eri bittet ihn [PROi sich erholen zu dürfen] he asks him to take a rest to be permitted ‘He asks him to be permitted to take a rest.’

(99)G

Er bittet ihni [PROi sich erholen zu wollen] he asks him to take a rest to want ‘He asks him to want /be willing//choose/ to take a rest.’

° This juxtaposition goes back to Bech (1955, 1957) (cf. Ruzicka 1983a: 319). The Russian literal translation of (97) is equivalent to the German, not the English interpretation.

48

Chapter 3 (100)R

Mat’ poprosila proi [[PROi Mother asked (somebody)

ubrat’ posudu]] to carry away the dishes

Complicity of two pro’s is shown in (101). (101)R

Vikaj daval svoj apparat oficiantui i proj Vika gave his camera to the waiter and proi [[PROi nas sfotografirovat’]] (him) us to photograph ‘… to take a picture of us.’

prosil asked

Null subject and null object in the shape of proj,i are anaphorically referring pronouns in (101). For Russian and German, then, the only chance of licensing subject control, beyond the standard cases of obeying m constraint (B,b) (see example 83), could be provided by disposing of the syntactic position of the direct object. Russian can do it by affixing the reflexive clitic -sja/-s’ to (po)prosit’-sja. While saturating the object theta-role lexically, the affixed verb prosit’sja precludes the syntactic object position, exhibiting control behaviour like ask in, e.g., one reading of (86). The syntactically inert null object in English and the syntactically removed argument position (and thetaposition) in Russian lead to the same consequence: The subject is a licit controller under the interpretive construal as in (89). (102)R

Oni stal prosit’sja [[PROi He started to ask staryj dom]] at the old house

(103)R

Onii proposilis’ [[PROi u nego zit’]] they asked at/with him to live

posmotret’ to have a look

Interestingly, in the standard dictionaries (po)prosit’sja is defined as “(po)prosit’ dlja sebja razrešenija (pozvolenija) delat’ cto-libo” (‘ask for oneself for permission to do something’). In conclusion, Russian like Italian (cf. Rizzi 1986: 541)) has null Vgoverned and Case-marked pro, which can be an arbitrary pronoun as well as a referential definite pronoun (cf. ex. (101)). Drawing upon the classification in Rizzi (1986: 541), it appears that Russian manifests the type of pro which can function as a “nonargument, quasi argument, and referential argument”. I have shown that structural conditions, in particular positional restric-

The theory outlined

49

tions of chains combine with symmetry effects of two thematic pairings to sustain the unmarked and the marked version of the Control constraint that licenses control relations. In the marked version (B,b), a secondary layer of quasi-theta-specifications emerges strong enough to establish control. That is why ask-type verbs exhibit marked control behaviour, besides unmarked one, of course. Disobedience to the marked version can still yield acceptable control by pragmatic processing that interposes conceptual material (like PERMIT HIM/HER TO …). A hyphothetical parameter (H) (p. 62) is offered to capture the crosslinguistic facts. ask-type verbs are highly susceptible to such conventionally communicative patterning, being the commonly used cliché word in everyday conversation.14 I would still like to suggest that pragmatic control as in (87) and the “permission” option of (88) — construed in (89) — is systematically licensed and sustained by parameterized modality switching, which divides English and Spanish from German and Russian. In German counterparts, the modal verb dürfen is used (ex. 96, 98). The corresponding parameter (H) is paralleled by parameter (G). See section 4.1.

3.4. Nominalized Russian prosit’(sja) As is well known, in Russian, in contrast to Polish, nominalizations of reflexive verbs and their respective nonreflexive partners exhibit the same nonreflexive noun, mostly neutralizing the aspectual opposition. (104)R

Vozvrašcenie detej k roditeljam [gen sg] the return of the children to their parents relates to vozvratit’sja / vosvrašcat’sja (105)R

Vozvrašcenie knig [gen pl] return of the books relates to vozvratit’/vozvrašcat’. Control behaviour of pros’ba exhibits relationship to prosit’ and prosit’sja.

50

Chapter 3 (106)R

Pros’ba ne kurit’ (Notice in a room) [nom sg] request not to smoke (X’s request directed to Y that Y should not smoke) ‘You are requested not to smoke’.

(106) corresponds to standard um control cases according to (B,b), pros’ba is related to prosit’. Relationship of pros’ba to prosit’sja shows up in (107). Recall that prosit’sja “absorbs” object position, thus leaving only its subject as possible controller. (107)R

Posle vyslušannogo prigovora v pros’be [part pret pass gen sg] [gen sg] [loc sg] after (the) listened to sentence in (the) request sobrat’sja nami v odnu kameru, ctoby provesti [infinitive] [dat pl] [acc sg] to gather us in one cell so as to spend poslednie casy vmeste, otkazano ne bylo [acc pl] [part pret pass neuter] [neuter] (the) last hours together refused not (it) was ‘After we had listened to the sentence, (our) petition to be (gather) in one cell so as to spend the last hours together was not refused’ (from Ju. Trifonov’s “Starik”).

A puzzle arises here, because nam is the overt subject of the apparently controlled clause sobrat’sja … and is assigned default dative case by [+infinitive] under Inflection. Nam could hardly be the indirect dative object of otkazano, which assigns inherent dative, in the position it occupies in (107). The pronoun is presumably inserted to avert (empty) object control, in which case there would be no question of control. In reverse order, then, the subject of pros’ba is identified, preferably, by nam, or as a member (members) of the set denoted by nam. I conclude that control relations in complements of nominal abstracts of verbs, can be obviated in the marked but potentially ambiguous cases of subject control, which, in Russian, hinges on the reflexive affix (see p. 48). This affix, as mentioned above, is neutralized in verbal nouns. Dispensing with nam and positing a PRO-subject of sobrat’sja v odnu kameru in (107) would leave the latter without clarifying its identification. A functional motivation for dispensing with control and resorting to an overt subject seems justified.

The theory outlined

51

3.5. When pro takes over The analysis proposed for the control behaviour of ask, R. prosit’ receives strong support from the situation which obtains in “pro-drop”-languages in which finite complements can adjust to control relations. The empty pronominal category pro takes the place of PRO. Indeed what structural and functional properties pro can display in the role of the controllee presents enough evidence to prove that the constraints and principles advanced above are able to capture the empirical facts and explain them. Serbo-Croatian is a nullsubject-language suitable for demonstration. (a) If the um constraint (B,b) is satified, pro must appear, not the overt pronoun ona. (108)S-C

Petar je molio Marijui da proi (*onai) ostane Petar Aux asked Marija comp stay na veceri at dinner ‘Petar asked Maryi to stay for dinner.’

(The examples (108) – (110) are from Zec (1987) slightly modified.) (b) If m constraint (B,b) is violated, under the construal (89) and in the strong version demonstrated by the Spanish example (88), pro must appear, not the pronoun. Clearly, in finite complement control, this peculiar option, described above has a better chance to be realized. (109)S-C

Petari je molio Mariju da proi (*oni) ostane Petar Aux asked Marija comp (he) stay na veceri at dinner

It is exactly in the respective cases (a) and (b) that on(a) must be dispensed with. Independently of control, pro can of course occur iff agreement (or φ–) features of the finite verb transferred to pro identify it, which implies that they are different from those inherent to either the potential subject or object controller in the immediate matrix clause.

52

Chapter 3 (110)S-C

Petar me je molio da pro [+3 pers +sing] [+1 pers] Petar me asked that (they) ostanu na veceri [+3 pers +plural] stay for dinner

I should add — following Zec (1987: 154 f.) — that in cases (a) and (b) above on and ona can replace pro iff they are stressed.

3.6. A Comparison with an alternative analysis I would like to conclude this chapter by juxtaposing the analysis of two relevant cases of control that follows from my approach with the analysis put forward by Franks/Hornstein (1992: 41, fn. 15). The sentences instantiating the control phenomena analyzed are (111) – (113) (Frank/Hornstein’s (i) – (iii)) (111)

Johni told Fredj that PRO hanging each otheri,j’s pictures would be fun.

(112)

Johni said Billj asked where (PRO to hang each otheri,j’s pictures/pictures of each otheri,j.

(113)

*Johni said Billj asked PROi,j to hang each otheri,j’s pictures/ pictures of each otheri,j.

Frank/Hornstein’s account runs as follows: “In (i) and (ii), (111, 112: RR), PRO can be indexed by both Johni and Fredj. This allows each other to be bound by a plural antecedent, viz. PROi,j. In (iii), PRO is governed, hence anaphoric. Therefore, it does not permit split antecedents. Consequently, each other is not properly bound. The contrast between (ii) and (iii) (my (112), (113): RR) is of particular interest as the only apparent difference is whether there is a +WH COMP. If there is a WH in COMP, the embedded PRO cannot be anaphoric. If there is nothing filling the COMP, the PRO can be governed. It appears, then, that to a first approximation, with respect to split antecedents, the binding properties of governed PROs parallel those of anaphors while those of ungoverned PRO do not. This is what a theory of governed PRO leads us to expect.” (Franks/Hornstein 1992: 41, fn. 15) I will concentrate on (ii),

The theory outlined

53

(iii) (112, 113), which indeed are of particular interest, whereas (i), (111) is not directly pertinent to the contrast between them and their respective analyses. The (non)occurrence of +WH COMP is not the only difference. In (112), ask used as query or request information does not belong to the lexical control class subject to constraint (B,b). There is no inherent crossclausal thematic relationship between two arguments that could be captured by such a constraint. PRO is considerably less constrained in choosing an available and suitable antecedent, though of course not freely identifiable. Essentially, control may be logophoric and split up nonlocally, or be arbitrary. The fact that PRO is ungoverned if Spec(C) is filled, is incidental to the control situation as described under my approach. Indexing PRO in (112) by both Johni and Billi is induced by the reciprocal. Why is (113), (iii) ruled out? First, since in English the null object parameter is fixed negatively, in accordance with Bach’s Generalization no object controller is available for the “object” control verb ask. Second, Bill is the only possible and local controller, and it is licensed by marked application of (B,b) in virtue of construing ask (permission) as obeying m (B,b), which has been discussed in detail and can be understood as setting the pragmatic parameter (H) (4.1.) positively. Note that judgements may be subtle, as is natural when boundaries of grammar are crossed. John, in the casual embedding John said, is irrelevant to control: Being no argument of the control verb ask, it is not targeted by constraint (B,b). Third, since Bill remains as the only possible controller of PRO, each other is not bound by a plural antecedent. Replacing Bill in (113), for example, by Bill and Steven would make the sentence acceptable with the above proviso.

Chapter 4

The persuade subclass and cognate ones subject to the general constraint

4.1. Persuade versus ask. Parameterization in pragmatics. The control effects of lexical properties of ask-type verbs are easily distinguishable from those of persuade-type verbs. Members of the ask-subset are, e.g., beg, request, implore, conjure, beseech, entreat; Russian: uprosit’/uprašivat’, umolit’/umoljat’; German: ersuchen (request), beschwören (implore). For example, (114) obeys the m constraint exactly as, e.g. (87), or (89). (114)

The boyi begged [[PROi (to be allowed) to come with us]]

persuade-type control verbs, though constituting an opposite subset of direct object verbs in regard to control conditions, are subject to the generality of constraint (B,b) as well as ask-verbs. But the subsets mainly differ in exhibiting complementary control effects in the passive infinitival clause. (115)

Peterj persuaded Billi (PROi,*j to be examined by a doctor)

(116)

Peterj persuaded Billi (PROi,*j to be replaced at half-time)

(117)

Peterj asked Billi (PROj,*i to be replaced at half-time)

(118)

Peterj asked Billi (PROj,*i to be examined by a specialist)

At least three problems are posed by the control behaviour of persuade-type verbs: (I): First, (115) is perfectly acceptable though violating the um constraint (B,b). (II): the second problem is posed by the ungrammaticality of the lexical and syntactic counterparts of (115), (116) in German and Russian. (III): Last but not least, persuade complement control does not enjoy wellformedness in accord with the m version of contraint (B,b): “Subject” control is widely

56

Chapter 4

barred. The standard case, however, is accounted for by um (B,b), as in the ask-type subclass.15 Let us take a closer look at the three problems raised by persuade-type subclass. To start with, why are (115), (116) grammatical though they violate the um constraint (B,b)? Note first that the theta-grid of the complement verb meets (D) (3.2.2.), the specification on the marked application of (B,b), as can be read off of Dowty’s analysis given in fn 15. Though they satisfy (D), passive infinitival complements of the persuade-class verbs, in contrast to ask, resist the marked application of (B,b). Basically, (D) posits that 〈intact, –〉 in PRO is visible to the application of m (B,b), iff 〈intact, +〉 is available in the theta-grid of the complement verb. 16 Note that persuade type verbs select a lexically overt object. (cf. sect. 4.3.) Control in (115), (116), thus, must be prevented from being licensed by the m application of (B,b). Assume that they need access to the unmarked version of constraint (B,b). The only access to it is by providing PRO with 〈intact, +〉. I suggest that in English this feature value is projected by persuade onto PRO marking conceptual-pragmatic causative switch, analogous with modality switch in the ask-class (parameters (G) and (H)). Let us consider two perfectly natural aspects of this proposal. First, in complements of persuade, the referent of PRO gets involved in a decision to do something to bring about the event described in the infinitival complement. Second, PRO, which in (115) has been assigned 〈intact, –〉 by examine, can receive the “secondary” assignment of 〈intact, +〉 only on condition (D) that 〈intact, +〉 has been “internally” discharged by the complement verb, examine. In (115), e.g., it is carried by a doctor. The natural rationale of this condition hinges on the feasibility of influencing the (referent of the) designated Agent-DP, e.g., in the PP [by DP] to do what is described in the passive complement. It is (the referent of) PRO who influences, that is, is persuaded to do the influencing. Clearly, (D), which provides for availability of 〈intact,+〉, is a prerequisite for assigning it to PRO secondarily, but does not license it. This assignment, which salvages um (B,b) from being violated, is dependent on the positive setting of a parameter, as we shall see presently. Note that, as this subclass (persuade) evidences, violations of the unmarked (B,b) are restored to grammar without introducing quasi-lexical material that assigns “conceptual” theta specifications, as, e.g. be allowed in control of ask-type verbs, when marked (B,b) is violated. In this latter subclass, violations of the unmarked (B,b) are taken up by the marked application (B,b) (see ex. (118), (83)), since PRO

The persuade subclass and cognate ones

57

carries unchangeable 〈intact, –〉. Recall that violations of (B,b) are defined on the feature value of PRO. We feel justified in hypothesizing that a particular parameter is involved in the persuade subclass. Its values concern the interface between grammar and pragmatics. This leads to the second problem mentioned above. German, Russian and Czech, for example, fix this presumed parameter differently, that is, negatively. They grammatically resist the assignment of 〈intact, +〉 to PRO, which appears to be a natural option as well considering that the only thetatheoretical and functional interpretation of passive structures on which there is general consensus amounts to suppressing or defocusing the agent. This seems to do the reverse of what the presumed secondary assignment of 〈intact, +〉 to PRO does. Note that PRO is structurally prominent as the argument in the highest A-position of this clause. The negative parameter setting keeps the original thematic specification of this position as induced by passive immune to conceptual-pragmatic change. (119)R

ugovoril  *Petr   ubedi 

Peter persuaded osmotrennoj [part pass instr fem sing] examined (ex. from Comrie (1981)). (120)G

(121)C

Tanjui (PROi byt′

[acc] Tanja to be vracom. [instr case] by a doctor

überredete  *Peter   Tanjai (PROi von einem Artzt überzeugte Peter persuaded Tanja by a doctor untersucht zu werden examined to be

presvedcil  *Petr   otcei (PROi být osetren umluvil   [acc] Peter persuaded father to be examined v nemocnici) in a hospital

58

Chapter 4

The intended interface parameter leads to the prediction that, in rendering sentence types as literally as possible, those three languages and others that fix the parameter negatively must rely on PRO being solely and unchangeably provided with 〈intact, +〉. This is incompatible with controlled passive complements. An idiomatic lexical near equivalent is available. Czech and German use causative nechat and lassen, respectively.

(122)C

umluvil  Petr   otcei (PROi nechat  presvedcil  Peter persuaded se ošetrit himself examine ‘Peter persuaded hospital’

(123)G

(124)R

father to have (somebody) v nemocnici) in a hospital his father to have himself examined in a

überzeugte Peter   seinen Vateri (PROi sichi überredete  Peter persuaded his father himself von einem Facharzt untersuchen zu lassen by a specialist examine to have ‘Peter persuaded his father to have himself examined by a specialist’. ugovoril  Petr   ubedil 

otca (PRO podvergnut′

[acc] Peter persuaded father to submit sebja osmotru vracom) [dat] [instr] himself to an examination by a doctor ‘Peter persuaded his father to undergo medical examination.’ Thus, German, Russian and Czech rely on a radical syntactic and lexical solution to the conceptually inherent aspect of their respective counterparts of persuade. In all examples (122) – (124) PRO is assigned 〈intact, +〉 contained in the θ-grid of Czech nechat, German lassen and Russian podvergnut’ sebja, respectively. They are licensed by um (B,b). The parameter associated with the lexical subclass of persuade-type control verbs, which interfaces grammar and pragmatics, can be formulated as

The persuade subclass and cognate ones

59

in (G), with (D) again as its corollary. (G)

In the designated lexical subclass of the control verbs associated with constraint (B,b), its unmarked application can(not) be satisfied by assignment to derived PRO of feature specification 〈intact, +〉.17 licensing conceptual-pragmatic causative switch.

The parameter consists in the choice between allowing the assignment to PRO of the nonlocal or secondary thematic specification 〈F,+〉, or not allowing it. The relevant assignment is conceptual-pragmatically driven but referred to in turn by grammar, which by the um application of (B,b) can license the result if the parameter is set positively in the respective languages. 〈intact, +〉 is the only characterization that the chain headed by PRO can receive pragmatically. If the parameter is set negatively, the “original” syntactic assignment of 〈intact, –〉 to (the chain headed by) PRO resists being pragmatically overridden by the feature 〈intact, +〉. In other words, 〈intact, –〉 remains visible for checking by (B,b) and would be expected to satisfy its marked version in the same way as ask-type control sentences do, e.g., in (59), that is, result in correct “subject” control. But this expectation is not borne out, which means that the account of the grammaticality of control in (115), and the like, which excludes them from being licensed by m (B,b), is not sufficient (problem (III), mentioned 4.1.). It leaves open the question why, in English sentences as (115), (116), control can or even must evade licensing by the marked constraint, and why in German, Russian and Czech, which set the parameter negatively, that is, do not touch the feature value 〈intact, –〉 of PRO, the marked version of (B,b) does not license their counterparts either. The presumed interface parameter (G), thus, needs refinement with respect to its conceptual prerequisite conditions. Recall that the marked application of (B,b) is intrinsically bound up with pairings of semantic roles which can imply coreference of their respective two recipients. Coreference established by the interrelated thematic structures of the matrix clause and the embedded infinitive can be relied on to function as a control identification, if PRO is one of the paired DP’s (cf. the discussion above, 3.2.3.). Indirect implementation of control by conferring it on independently paired thematic roles that imply coreference of their respective DPrecipients is characteristic of marked control phenomena. It is exactly the weakness or conceptual complexity of this thematically driven coreference that is responsible for the variation of wellformed control in the cross-linguis-

60

Chapter 4

tic persuade-type subclass of control verbs, and for its general resistance to being licensed by the marked constraint (B,b). Persuade and its ilk lend themselves by far less easily to a construal in which the matrix subject can be seen as benefactive of the action described by the passive infinitiv. Thus, referential pairing of the interested “parts”, that is, identification of the matrix subject and moved PRO, is not normally amenable to a marked control relation. That is why (116) does not exhibit subject control as, for example, (117). Persuade-type events, as opposed to ask-events, characteristically contain arguing and reasoning, which rather neutralize aspects of the action directly related to its benefit for the person(s) doing the arguing. Thus, the derived PRO-DP that is assigned, say, theme or goal in the passive complement is not readily construed as referentially identical with the subject-DP. Still, if the chances for subject control in passive controlled clauses, in accord with m (B,b), are slim, persuade-contexts do not fully exclude it and may sporadically exhibit repercussions of the thematically driven coreference. Sensitivity to peculiar conceptual construal of thematic roles and their interaction as well as resulting fragility of control judgments have to be taken into consideration. Take for illustration an example from Rochette discussed by Vanden Wyngaerd (1987: 92). (125)F

Guillaume a persuadé Béatrice [PPde [NP[S’ PRO l’ aider à faire ses devoirs]]]

“In this example, control is ambigous, (25) (=(125), RR) being translated as follows: a. b.

Guillaume persuaded Béatrice to help him to do his homework. Guillaume persuaded Béatrice to be allowed to help her do her homework.

… we are dealing with a case of nonobligatory control … (the complementizer) de appears irrespective of the control properties of the matrix verb. Thus the exact nature of the structural contrast remains to be determined.” (Vanden Wyngaerd 1987: 92). As far as a structural contrast is concerned, it can be captured by the two versions of constraint (B,b): (125) satifies its um version in the a. interpretation. In the b. interpretation, if correct, we could wonder at the interpolation of be allowed, which, on this approach, would seem to induce “object” control as in (115), (116). A reading of (125), as paraphrased

The persuade subclass and cognate ones

61

by b., however seems to be licensed by the marked application of (B,b). This is quite unexpected considering that control matching marked (B,b) cannot materialize in seemingly similar cases like (115), since, in contrast to ask-type verbs, the basic pairing of feature values required by the m constraint (B,b) (〈intact, +〉con & 〈intact, –〉pro) does not square as easily with linking a conceptual-pragmatic specification like benefactive with PRO. In other words, PRO in (125), if assigned this specification, which sustains marked conditions, does not suggest that it be identifiable with the matrix subject-DP referring to the person(s) doing the persuading. Now, one might acquiesce in interpretation b. if (125) could be construed with the help of a conceptually interposed predicate that assigns benefactive to PRO. Interpretation b. is a literal paraphrase of a construal through which “subject” control in (125) can become acceptable. “Silent” be allowed functions as a conceptual intermediary providing PRO with a characterization which makes (125) satisfy marked (B,b) while, of course, violating it. (125), if interpretable as in b., thus might be considered to be analogous with the ask-type Spanish example (88) discussed above. Both of them violate the m constraint (B,b) and both are reconciled to it by the conceptual construal intuitively represented as be allowed (or be permitted). Their outward appearance is the result of conventionally motivated compression. Still, the parallelism between the supposed b.-reading of (125) and (88) is a spurious one with respect to the licensing conditions operative at the interface level. Phenomena that evidence violations of either um or m constraint (B,b) still tend to be receptive to construals that cope with the menacing “crash”. They are subject to parametric variations interfacing grammar and pragmatics. Upon my general contention that pragmatics can be parameterized, I posit another parameter (H), which, like (G), interfaces principled constraints of control and pragmatic interference in them. It consists in interpretive construals that can be referred to by the very constraints they seem to evade. (H), unlike (G), targets marked application of (B,b) and is meant to account for the contrast between the potential ambiguity of, e.g. (86a) or (88), and its absence in their literal German and Russian counterparts. Marked (B,b), by definition, is weaker than unmarked (B,b), being grounded on a thematic interrelationship between DP’s that is construed as licensing coreference. (see above, 3.2.3.) Clearly, construals that repair violations of m (B,b), as in (86a) or (88), tend to be obviated by an “ingrained” understanding that corresponds with

62

Chapter 4

unmarked (B,b), requiring 〈F,+〉PRO, that is, sets the parameter negatively. Ambiguity is excluded, then. (H)

Languages (do not) allow violations of marked (B,b) to be construed as matching it in the subclass ask, subject to the (B,b) constraint. Positive value is linked to modality, in particular deontic switch.

If they do not allow the construal under the conventionalized “pressure” as suggested above (3.3.), judgements are determined by the um application of (B,b) alone. (H) can be assumed to interact with the null object parameter as it is described by Rizzi (1986). Languages which set this parameter negatively, that is, in which there is no syntactically “active” pro capable of control are more susceptible to setting (H) positively, as described above. Let us shortly compare (G) with (H). (G) is a substantive interface parameter: Grammatical linking of thematic specifications gives way to conceptual-pragmatic “reconstruction” (English) or resists it (German, Russian). The parameter is associated with a subclass subject to constraint (B,b), covering persuade-type verbs. If it is set positively, repair is effected by conditions on, and recovering of, thematic specifications. Acceptability is attained by an allocation of feature specifications which is referred to and licensed by, the um constraint that was violated. The parameter (H) presumed to be amenable to the ask-type subclass of control verbs that is subject to (B,b) is of a different kind. It arises from violations of marked (B,b) and can be assumed to be purely pragmatic. Set positively, its work must end up in a thematic configuration that satisfies the very m constraint (B,b) it has violated. Restoration to interpretive wellformedness, that is, acceptability, consists in the conceptual arrangement of the thematic conditions on which matching the m constraint depends: PRO must be made thematically compatible with the subject of the matrix clause. Consequently and more precisely, it must be warranted to be identifiable with the higher subject. The clearest conceptual interpretation, namely be allowed, which effects identification and in some examples occurs overtly, is highly susceptible to conventional compression omitting it and conceptually restoring it. Positive setting of the pragmatic parameter associated with ask-type verbs (e.g. English, Spanish) consists in the conceptual interpolation of some equivalent of be allowed under highly conventional favorable auspices. Take

The persuade subclass and cognate ones

63

the critical cases potentially satisfying the um or violating the m constraint (B,b). I repeat (86) (from Bach) and (88) (from Chomsky). (86)

They asked the teacher to go to the bathroom.

(88)SP

María le pidió a Juan (hablar con los muchachos) ‘María asked Juan to speak with the boys’

Following Bach, (86) is “ambiguous for some”. Assume that (88) is ambiguous for some informants, too, though subject control seems dominant, following Chomsky (see above). Both languages set the parameter positively, that is, can “repair” violations of the marked constraint (B,b) if this interpretation is preferred to the one meeting its um version. The pragmatical arrangement restores a structure that can be checked by the grammatical constraint (B,b) straightforwardly. I repeat (89), in which a “distant” unmarked control relation meeting (B,b) is established by a conceptually constructed “bridge”, which in turn exhibits control. (89) Xj asked leave))

SOME

PERSON(S) Yi (PROi to

PERMIT/ALLOW

Xj (PROj to

In languages setting the presumed pragmatic parameter negatively, potential ambiguity of control is out of the question. Recall that the ambiguity cannot arise when the positive setting of this parameter combines with instances of the negative setting of the null object parameter, as in English. The unmarked application of (B,b) fails, since a syntactically inert null object cannot be a controller. In languages which set the null object parameter positively like German and Russian, only unmarked (“object”-) control meeting (B,b) is possible. The parameter value may be supposed to contribute to the negative setting of the pragmatic parameter related to the ask-type subclass. The spuriousness of the superficial structural analogy between (88) and (125) consists in the wrong step of transferring a systematic device for coping with violations of the marked constraint, which is parameterized at the pragmatic level, to a distinct subclass which is systematically associated with an interface parameter concerning violations of the unmarked constraint (B,b). Recall that the marked application of (B,b) is substantially restricted in the persuadesubclass. In addition to (125), I will give an illustration from German. In their incisive widely experiment-aided investigation of control in English and German (1991, 1993), Köpcke and Panther, though assuming that überreden, überzeugen, (persuade) convey Benefactive to their subject as well as bitten or

64

Chapter 4

beschwören (implore), seem to arrive at the conclusion that “subject” control (satisfying marked constraint (B,b) in this approach) of these verbs is rather bizarre or, in any case, considerably restricted as compared with the asksubclass. A comparison between (126) and (127), both interpreted as meeting m (B,b), shows a clear superiority in acceptability of the latter. (126)G

Renate überzeugt Heide, das Telefon benutzen Renate is persuading Heide the telephone to use zu dürfen to be allowed ‘Renate is persuading Heide to (that she (Renate)) be allowed/ permitted to use the telephone.’

(127)G

Renate bittet Heide, das Telefon benutzen zu dürfen Renate asks Heide to be allowed to use the telephone

As a consequence, violations of m (B,b) made acceptable by the positive option of the pragmatic parameter (H) require quite a particular interpretive situation in the persuade-subclass.

4.2. On complement unaccusatives and unergatives Neither class of intransitive verbs is left out of consideration, at least implicitly. We have contended that control behaviour of the lexical persuade- and ask-type subclasses can be captured by unmarked and marked versions of constraint (B,b) together with corollary (D) and subject to variation of parameters (G) and (H). Persuade and ask differ in their respective reactions to the marked application of (B,b), that is, in their peripheral and conceptually driven control behaviour. The differences have been accounted for on the grounds of matching effects and requirements between the identification of control and appropriate thematic specifications of the arguments to be coindexed. The overall condition contained in the um and m constraint (B,b) is the availability of 〈intact, +〉 in the θ-grid of the complement verb. If marked (B,b) is violated, its licensing conditions can only be restored by introducing conceptual notions which may be intuitively represented by meta-language lexical material that in turn induces secondary thematic assignment satisfying m (B,b). If the unmarked constraint (B,b) is violated, its conditions are reconsti-

The persuade subclass and cognate ones

65

tuted by the conceptual-pragmatic “super assignment” of 〈intact, +〉 to PRO. Unergatives in the infinitival complement do not seem to pose any problems for either version of constraint (B,b), if “… unergatives are verbs which take a single, truly agentive argument: a deep structure subject with no deep-structure object.” (Fowler/Yadroff 1993: 256). Beyond their purely structural properties, either class may have an effect on wellformed control. In unergative complements, PRO should be provided with 〈intact, +〉. Attempts to associate the two notions with thematic characterizations, however, have met certain reservations (Levin/Rappaport 1986; Grimshaw 1987). The only problem that could arise concerns some verbs which have been considered unergatives, e.g. cough, sneeze, laugh, and others. (cf. Levin/ Rappaport 1986: 654)

sneeze    (128) They persuaded him i (PRO i to cough ) laugh  These verbs can hardly be taken to denote volitional acts and, thus, to have 〈intact, +〉 in their θ-grids. Under this approach with its licensing conditions, such cases of control can no more be rescued than unaccusative complement verbs denoting “… states (and even acts) in which the patient is drawn into the action conveyed.” (Fowler/Yadroff 1993: 256). It should be clear that any lacunae in empirical and conceptual coverage and judgment have to be closed not by control theory but theta theory when the latter is associated with the bipartition of unergatives and unaccusatives. It is obvious again that refinement of control theory is also consequent on an enriched theta theory and in particular refined non amnestying pragmatics. (128) either violates um (B,b) and reaches no wellformed interpretation, or it is accepted on the understanding that sneeze etc. can be taken to be intentional actions (〈intact, +〉pro) in some “world”.

4.3. Neutralization of the null object parameter with respect to Control of persuade complements In particular cases of “strong” lexical government that require the overt appearance of the direct object, as, for example, persuade, g. überreden and überzeugen, differentiations within the null object parameter, set differently in

66

Chapter 4

Russian, German and English (cf. Cole 1987: 603)), play a minor role in control (cf. 3.3.). Thus, the strongly marked contrast between English and German arising in the control behaviour of ask and bitten insofar as it hinges on the null object disappears in the case of persuade and its German counterparts. The Russian counterparts, however, allow null objects, which are syntactically active and license control. (129)R

On ze

ugovoril

proi ((PROi otdat’ Rodju [acc sg] He (emphasis) (had) persuaded pro to send Rodja v korpus a Laru v gimnaziju)) [acc sg] to the military college and Lara to the gymnasium somebody ' He (had) persuaded  them / him / her

(130)R

 ...'   V DS (demokraticeskom sojuze) nikto i nikomu [nom] [dat] In (the) democratic alliance nobody and nobody nicego ne mozet prikazat’, mozet tol’ko ubedit’ proi [acc] nothing not can order can only persuade to ((PROi postupit’ tak ili inace)) act so or otherwise ‘In the Democratic Alliance nobody can order anybody anything, can only convince him to act somehow.’

A null object can combine with a null subject, both referring pro’s carrying ϕfeatures.

(131)R

proi ugovoril

mat′ k   pro  j  

PRO j,k

ostat’ sja

persuaded to stay doma)) at home ‘ He persuaded him / her / them (mother) to stay at home’. The counterpart to English one would be a plural pro in Russian.

The persuade subclass and cognate ones

(132)R

pro

djadjui    ((PRO j,i  pro j  [+plural][pret pl] [acc sg] ugovorili

one  uncle    persuaded   they him / her / them     ostat’sja)) to stay

67

k obedu

for dinner

Note that in all these examples the pro-object is not available in English and German, This fact, owing to lexical properties of (counterparts) of persuade, is independent of the null object parameter and their distinct values of the interface parameter (G), in contrast with (H). We take convince, urge, enjoin, incite, and some other verbs and their cross-linguistic counterparts to be in the persuade-subclass and follow this verb and their control behaviour most closely: They are subject to constraint (B,b), show a more or less recalcitrant attitude to its marked application and, crucially, their cross-linguistic differences in control are parameterized at the interface level with regard to violations of the unmarked (B,b). Dividing the class (B,b) up into further subclasses, separating them in particular from the persuade-type verbs above is motivated (1) by semantic selection, which allows some verbs to take a DP marked [–human] as their subject, and by dative case assignment to the controller argument DP (subclass under 4.4.4.).

4.4. Further subclasses obeying constraint (B,b) 4.4.1. The teach subclass Teach and its cross-linguistic equivalents pose no substantially new problems of control, but their treatment may prove instructive methodologically. As with persuade, intransitivization is impeded. Teach, g. lehren, r. (na)ucit’ obucat’, c. naucit’ obey constraint (B,b). These verbs do not denote intentional actions aiming to bring about that someone decides to do something (e.g. ask, persuade). Rather, in their main sense, they describe events, verbal and/or nonverbal intentional actions that (try to) enable someone to do some-

68

Chapter 4

thing. Imparting wisdom and/or ability to someone hardly lends itself to construals in which the person that acquires the ability, e.g. second language faculty, does not figure syntactically as the actor or, e.g. experiencer in subject position, that is, as PRO. Thus, teach, like persuade allows of violating the unmarked constraint (B,b) without loss of acceptability. Acceptability is restored as with analogous persuade cases, provided that corollary (D) is satisfied. (133)

The soldiersi have been taught ti (PROi not to be spotted ti), (PROi to camouflage themselvesi).

Correspondingly, in contrast to ask, PRO in passive complements and the matrix subject cannot be tied together and identified by bonds of something like Benefactive. Thus, these verbs have no access to the marked version (B,b). (134)

*Hei taught Anne (PROi to be allowed to ride the bicycle)

(133) shows that the account I have suggested for violations of um (B,b) in persuade complements (see 4.1.) is suitable here. However, the cross-level or interface parametrization (G) assumed above to account for persuade e.g. in (115), (116) and their counterparts (119), (120) does not include teach, sentence types (135), (136) are acceptable in Russian and German. ?Sie

(135)G

lehrten unsi ((PROi nicht gesehen zu werden)) they taught us not to be seen

(136)R

?Oni

naucili nasi (PROi ne byt’ zamecennymi)) they taught us not to be noticed

The particular control behavior of teach and its crosslinguistic counterparts is very revealing insofar as the major framework of the chosen lexical feature specifications turns out to receive straightforward empirical confirmation from the subtle interface parameter (G). The possibility of the contrast arising between the positive value set in English for (G) and the negative value fixed, e.g., in Russian and German, is contingent upon the actually present intention of the referent of PRO to act (〈intact,+〉), which is understood to be the result of persuading and the respective events expressed by cognate verbs subject to (G). The clausal complement of teach rather has a kind of generic character indicating that a particular ability how to do something is (being) acquired, which does not require an actual intention to exercise it. Thus, parameter

The persuade subclass and cognate ones

69

values of (G) may be neutralized with respect to teach in general. While events of persuading, if successful, bring about intentions to act (examples (131), (132), events of teaching need not (ex. (133), (135)). In the former case, German and Russian counterparts to persuade, in consequence of the (negative) setting of parameter (G), achieve assignment of 〈intact,+〉 to PRO through particular verbs (cf. (123), (124)) vs. ((119)–(120)), as opposed to English (115–116)). In the latter case, teach does not require lexical assignment of 〈intact,+〉 to PRO. The distinction between implicit causality switching, construed as positive value of (G), in descriptions of persuade-events, and the covert (or overt) interrogative manner adverbial, which is intrinsic to the complement of teach-event descriptions, surfaces syntactically in the contrast between (115), (116) and (119)–(124) on the one hand, and the conformity of (133) with (135), (136), on the other. 4.4.2. The case of signal as a Control verb A discussion of the control behaviour of signal may elucidate the operative range of principle (B,b) and its susceptibility to varying thematical instantiations exhibited by one and the same verb. To all appearances, signal belongs in the ask-subclass associated with (B,b). It clearly is amenable to its unmarked and marked version. The very evasiveness with which complements of signal comply with (B,b) throws light on the general mechanism of complement control. Signal meets the um constraint in (137), (138). (137)

John signaled to Maryi (PROi to behave herself) (ex. from Manzini)

(138)

Pat signaled Loui (PROi to leave) (ex. from Sag/Pollard 1991)

Marked application of (B,b) is illustrated in (139). (139)G

Der verletzte Torwarti

signalisierte (dem Trainer) [dat] the hurt goalkeeper signaled (to) the coach (PROi t ausgewechselt zu werden) substituted to be

How should we articulate the mechanism through which unmarked control matching (B,b) can be exercised by an empty controller if the latter seems to do that independently of its syntactic activity or inertness, that is, indepen-

70

Chapter 4

dently of the setting of the null object pro-parameter? (140)

Louise signaled to follow her (ex. from Bresnan)

(141)

Pat signaled to leave. (ex. from Pollard/Sag)

(142)R

Signaljat (ei) (PROi ostanovit’sja) [3 ps pl pres] [infinitive] (They) signal to stop

Do such structures contradict Bach’s generalization, as Bresnan (1982: 418) and Sag/Pollard (1991: 95) suggest? Bach’s descriptive constraint says that a direct object controller cannot be omitted. But signal, as I contend, has no inherent object. It neither subcategorizes for a structural direct object nor has a corresponding theta role in its theta grid. Signal and its counterparts is inherently subject-oriented. Signaling, one hopes to reach a receiving target and, of course, in many cases one may have a particular target in mind, but signaling does not hinge on reaching a recipient and its “success”, the adequate performance of signaling, is not dependent on finding a receiver of the signals. If a recipient is available or a particular “addressee” targeted, they need not end up as direct objects. They rather may be projected as indirect objects and/or prepositional phrases. (143)R

Signal’te na slupkiacc vernut’sja [imperat pl] Signal to the boats to return

But they need not be in the “numeration” to be computed. The interpretation of a presumable or intended recipient of the signal can be left to pragmatic inference. What can be signaled when the content of signaling is packaged in a bare infinitival complement? (S)he who signals either sends — in a conventional code of signs — a message (a) concerning his/her own intended behaviour supposed to be related or relevant to that of persons possibly involved or particular persons, or a message (b) to somebody “telling” him/her to do something. Ambiguity is a possible consequence. (144)

The car signaled to turn left.

(145)

Col. Jones signaled to land.

The persuade subclass and cognate ones (146)

71

Col. Jonesi signaled (the control tower) to position himselfi for a landing.

(147)

Col. Jones knew that the control tower had signaled (himi) to position himself for a landing. (examples (144)–(147) from Pollard/Sag (1991: 96 f.))

Ambiguity is independent of the phonetic appearance or syntactic “activity” of the potential “object”-controller. As mentioned above, availability of pro seems to play no role. In other terms, signal and its cross-linguistic counterparts do not assign a theta role linked to object position. Thus, PRO cannot be “coindexed with the designated θ-role.” (Rizzi 1986: 552). It is not unreasonable to assume, as suggested above, that directionality is an intrinsically lexical property of signal and, as a thematical feature, can be saturated in the lexicon (in the sense of Rizzi 1986: 553). Whether a (in)direct object-DP expressing the target person(s) of signaling is lexically present or not, being saturated in the lexicon and interpreted in relation to discourse, the signaler either directs the attention to her/his own ensuing or imminent behavior, warning the addressee in advance, or (s)he tries to guide the latter’s behavior. If at all, then only in this sense, “… the interpretation of the unexpressed subject (PRO, RR) is free.” (Pollard/Sag 1991: 99). Control in infinitival complements is not free from the constraint (B,b), which has to adjust to the lexical peculiarity of signal. Applications of the um and the m version of (B,b) have been illustrated by examples (137), (139) among others. Does (144), repeated here, provide empirical evidence supporting this approach? (144)

The car signaled to turn left.

The two possible readings of (144) should be subject to, or match the two versions of (B,b), respectively. “… the car in 108a (=(144), RR) could be a stopped police car signaling an oncoming motorist to take a detour.” (Pollard/ Sag 1991: 96). This meets the unmarked constraint (B,b) in the modified treatment of a lexically saturated abstract controller being assigned 〈intact, –〉. The identification may materialize at LF as coindexing PRO with a free variable which may be bound by relatedness to discourse. The account for “the more natural interpretations (which) involve the car signaling that it will make a left turn …” (Pollard/Sag 1991: 96) is less straightforward. This

72

Chapter 4

reading is “more natural” since it describes a perfectly usual and conventionalized event type. The car is “metonymically” assigned 〈intact, +〉, which finally matches marked (B,b) iff PRO can be pragmatically construed as a recipient of 〈intact, –〉. Clearly, the driver gives signs that he will turn left on the understanding that his/her decision be taken notice of for her/his benefit of unimpeded passage. Thus, the sentential argument of the infinitival complement is the theme of taking notice of and could be assigned 〈intact, –〉, meeting corollary (D). Percolating down to PRO, the feature specification could be referred to by the m (B,b). The addressee’s knowledge of (the referent of) PRO’s turning left is beneficial to the car, the metonymic antecedent of PRO. There is some analogy to ask-type control in (violations of) m (B,b) (cf. 3.2.5.) Note that being taken notice of is similar to construals resorting to be allowed. In both cases we have to do with strongly conventionalized and thus “natural” usage that obviates violations of the marked constraint since it can be construed to remedy them. Signal differs from ask in that its particular lexically semantic properties (see above) are responsible for a control behaviour which, obeying the principled constraint (B,b), evades the null object (pro) parameter and the pragmatic parameter (H). 4.4.3. teach and ask. Larson’s comparison Some recent ideas of Larson (1991: 131-135) about control properties of teach and ask suggest again that a purely configurational approach to control may fall wide of the mark. Larson observes that “The verb teach is similar to ask in allowing interrogative as well as nominal and infinitival complements … Furthermore, again as with ask, an intuitive interrogative semantics appears to assert itself with infinitives and certain nominals.” (1991: 133). These common configurational and distributional properties lead Larson to “… suggest that at least some examples with teach like (61b) (Mary never taught John to sing) might be analysed parallel to ask …” (p. 134). As we have shown above (4.4.1.), control properties of ask differ from those of teach with respect to the unmarked and marked application of constraint (B,b). Larson continues: “However, there is an important additional fact about teach that bears on control. Unlike ask, teach is a true dative-shifting verb …” (p. 134). Larson supposes this fact to indicate that, in addition to control conditions shared with ask, teach has control properties at least similar to those of the dative-shifting verb promise. He puts it even more strongly: “Since teach has not only an interrogative pattern like ask but also a dative pattern like promise, it follows

The persuade subclass and cognate ones

73

that we cannot rest with saying that teach shares the former’s control behaviour. Clearly, we must also say something about why it doesn’t seem to show the control behaviour of promise. Since teach exhibits both of the basic complementation patterns, why doesn’t it exhibit both control paradigms as well” (p. 134). Thus, Larson expects that teach combines the control behaviour of ask and promise, but it does not come up to his expectations with either verb. The basic reason is that those expectations are based on the wrong assumption that the configurational and distributational properties stated by Larson are essentially related to, or determine, control properties.18 While teach in fact shares at least the standard unmarked control relation with ask as with many others, in accord with um constraint (B,b) (cf. John taught Felix to sing), it has nothing in common with the control pattern of promise. Larson has some difficulty in accounting for these facts adverse to his approach. Not unexpectedly, he reverts to what is relevant for control in trying to explain the difference: “… teach does in fact realize both its ask-type complementation and its promise-type complementation with infinitives … the latter is represented by infinitives on their (injunctive) reading. This then suggests why the expected subject control behaviour of teach is missing. If the promise-type infinitives with teach always have an injunctive reading … then their understood subjects are always in a sense antecedently fixed. As a matter of semantics, they must always be construed as referring analogously to generic you or one in examples like You should always look both ways before crossing the street … The expected infinitival complementation pattern with teach is thus present, but promise-type subject control is excluded on independent grounds.” (Larson 1991: 135)). Thus, promise-type constructions do not exhibit promise-type subject control. Whatever syntactic similarity joins teach with promise, it has little effect upon control. Larson has got side-tracked here following consequently his bent for dealing with control syntactically.19 What Larson calls the “injunctive” reading of teach has some indirect effect on control in that it selectively permits a nonhuman subject, which of course is indifferent to 〈intact,α〉 and immune to m constraint (B,b). The range of candidates of controller narrows down to one. See (148) – (149). (148)R

Zizn’ naucila egoi ((PROi cenit’ ljudej life (has) taught him to judge people ne po vnešnosti not by (their) exteriors.

74

Chapter 4 (149)C

Az

tei zivot naucí ((PROi být opatrnejší)) [acc] When you life will teach to be more cautious pochopíš ze musíš být opatrnejší [2 ps sg] [2 ps sg] [compar] (you) will understand that (you) must be more cautious i v usudcích o lidech [loc pl] [locative pl] too in judgments about people ‘When life will have taught you to be more cautious, you will understand that you must be more cautious also in your judgments on people’ (From Václav Havel’s play “Asanace”)

Besides teach and signal (and counterparts), e.g. force, compel, impel, induce, prompt, tempt, encourage, embolden belong to the subclass of control verbs that fall under constraint (B,b) and permit subject-DP’s marked [–human]. (150)C

Taková reakce verejnosti donutila [fem sg] [gen sg] Such a reaction of the public prompted [príslušné instance]i PROi odlozit the corresponding authorities to postpone provedení reformy the realization of the reform

(151)R

Podobnaja reakcija obšcestvennosti pobudila [sootvetstvujušcie instancii]i PROi otsrocit provedenie reformy (gloss as (150))

(152)R

Mysl’ ob odnoj vcerašnej vstrece moej the thought on alone yesterday’s meeting my soblaznjala menjai (PROi zadat’ emu koj-kakie voprosy [dat sg] tempted me to put him some questions ‘The thoughts alone on my meeting yesterday tempted me to put him some questions.’

The persuade subclass and cognate ones

75

The latter group of control verbs subtly differs from teach in that they more readily adopt the control case analyzed as an instance of cross-level parametrized variation on verbs like persuade and considered to be a violation of the um version of constraint (B,b), that is, they instantiate the positive value of parameter (G) which presupposes satisfying (D). prompted  The collapse   him i (PRO i to be examined by a doctor). compelled   The unability of the subclass of control verbs under discussion to accept licensing by the marked constraint (B,b) is substantiated by the possible selection of DP-subjects with a [–human]-head noun. Such DP’s of course can not receive a thematic role as benefactive, the characterization considered to be a relevant precondition of the m application (B,b) to be successful (3.2.3.). Selecting a nonhuman DP-subject disqualifies control structures in the given subclass of verbs from taking part in correlations between “nonreflexive” and “reflexive” verbs. This situation emerges, for example in Russian, Czech and other Slavic languages. Cliticized “weak” pronominal reflexive elements regulate grammatical functions and corresponding linkage of thematic roles. Compare (154) with (155). (153)

(154)R

Ochotnik

naucil synai (PROi streljat’ iz [pret masc sing] [acc] The hunter taught (his) son to shoot (from) malen’kogo ruzja pocti bez promacha) [genitive] [genitive] a little gun nearly without missing

(155)R

Syni naucil-sja (PROi streljat’ iz malen’kogo [pret masc sing-refl] son learnt to shoot (from) a little ruz’ja) gun …

(156)C

Naucila se (PROi bruslit’) [fem sing pret] [refl clitic] she taught reflexiv to skate ‘(she) learned to skate’

76

Chapter 4

At PF, the clitic takes second position in Czech, but it is always affixed to the verb in Russian. (157)C

Nebo

se naucila správne [refl.] [pret fem sing] for pro (she) learnt correctly

mluvit to speak

“Subject”-control obtains after A-movement of the object-DP, which is the controller in (155)-(157), in accord with um (B,b). Since se (sja) appears to trigger the movement to nonthematic subject position, it can naturally be held responsible for, or at least involved in, the mechanism that prevents Case assignment. Since a DP receiving theme and describing what is taught to somebody, is Case-marked [+dative] in Russian and Czech, I take the corresponding clausal argument to occupy the same structural position in these languages. Note that the way the theta roles of the internal arguments of teach are linked to the direct and oblique object positions in English is the very opposite of Russian and Czech. The theta role assigned to the DP that is “dative shifted” in English, is borne by the direct object in the two languages. That is the reason why I adopt a ternary verbal structure here. (cf. Emonds 1993: 226, fn. 15). X’-licensing, that is, V’-licensing of the dative DP, or Larson’s (1988, 1990) “shell analysis” are alternatives. The choice of either of them does not bear on the issue discussed. 4.4.4. A subclass of Control verbs subcategorizing for an oblique object I turn to another subclass of control verbs within the domain of constraint (B,b): verbs that in languages considered here inherently assign dative to the NP which is theta-marked goal, benefactive or addressee. The control verb allow, promise and their respective cross-linguistic counterparts will be singled out in (sect. 4.5. and chapt. 5.). (158)P

Krystyna radzila

kolezankomi (PROi [dat pl] Krystyna advised (her) collegues nowy film) the new film

obejrze´c to see

The persuade subclass and cognate ones (159)R

Ostoroznyj Mazepa sovetoval proi (PRO i idti [dative] cautious Mazepa advised to walk beregom Dnepra) [gen sg] along the bank of (the river) Dnepr

(160)C

Redaktor

77

doporucil

autorovii (PROi provést [dative] (An) editor recommended the author to make nekolik zmen v textu several changes in the text (ex. from Panevová, personal communication)

Predictably, these verbs resist being licensed by the marked version of constraint (B,b). They do not lend themselves to the Benefactive-pairing “inadvertently” functioning as control of PRO by matrix subject in case of passive complements (see 3.2.3.).20 Recall that inaccessibility to the marked version of constraint (B,b), the version typically licensing control of ask-type verbs, amounts to a precondition for using the pragmatic chance of satisfying um (B,b) in case the corresponding interface parameter (G) is set favorably: the case of persuade-type verbs in English vs. the German and Russian values of the parameter (cf. 4.1.). Advise- and recommend-type verbs meet the condition, and their “metaactional” character with the dative NP taking on the accessory role of benefactive strongly suggests that they will accept license in the defined way if the respective parametric settings allow it. (161)

He recommended himi (PROi to be examined by a doctor)

Recall that German and Russian fix the cross-level parameter differently. Not too unexpectedly, we find the somewhat delicate construction accepted in Czech. (162)C

(pro) Doporucil Jirkovii (PROi být zarazen [3 ps pret masc] [dative] (he) recommended Jirka to be included in do vyšší kategorie) [gen sg] a higher category (Example from Panevová, personal communication).

78

Chapter 4

Judgments are delicate in languages cognate with languages that set the parameter negatively like Russian. Sensitivity to lexical environment and the cross-linguistic parameterized distribution give additional support to a pragmatic assessment of this type of control. Crucially, we have evidence that control here is maintained inspite of adverse parametric setting, e.g., in Russian, when the clausal complement is not passive, although the conceptual construal is clearly analogous to passive. This is the case in some reflexive structures that potentially receive a passive interpretation (cf. R˚uzicka 1992). Not being in a dethematizised position, their PRO subject is receptive to 〈intact,+〉, satifying um constraint (B,b) at D-structure.

(163)R

rekomendoval  On   sovetoval 

Borisui (PROi [dative]

recommended  he   advised  

Boris

posmotret’sja) [reflexive] to be examined (by a doctor) Typical representatives of this subclass of dative object control verbs within the grasp of constraint (B,b) are order and forbid, command, tell, instruct, direct, charge commission. Their cross-linguistic counterparts mostly have an oblique dative object. (164)C

(pro) rozkázal mui (PROi mlcet) [3 ps sing pret] [dative] he told him to be silent

(165)R

Major Berezkin prosnulsja, …, pozval avtomatcika [acc] major Berezkin woke up, …, called submachine-gunner Gluškova i velel prinesti sebe pomyt’sja [3 ps sing pret] [refl dat] Gluškov and ordered to bring himself to wash (himself) ‘Major Berezkin … ordered him (Gluškov) to bring him (major Berezkin) things to wash with’

The persuade subclass and cognate ones

79

An approximate simplified LF-structure of (165) is (166). (166)

[[major Berezkin]j … pozval [avtomatcika Gluškova]i] i [proj velel [DP pro]i [CP PROi prinesti sebej [DP ek [CPOk [IP PROj pomyt’sja tk]]]]]

The overt element corresponding to the null operator Ok would be cem (instrumental Case of the relative cto ‘that’). I assume that an empty operator that after movement leaves a trace in the S-structure binds a conceptually restricted category, possibly a pronominal (cf. Chomsky 1989), ranging over things necessary for washing oneself. The exact range of (Ok, tk) is of course discourse-related. Control of the second PROj depends on the content of the immediately superordinate predicate (prinesti …). Discourse-related identification of proj and proi, which carry case and theta-features, is ensured within (166). The dative reflexive sebe in (166) has no local antecedent. Long distance binding of reflexives, which on standard assumptions are instances of bound anaphora, is not uncommon in Slavic languages, in particular in Russian and Polish (cf. R˚uzicka 1973: 445–481)). The crucial issue is whether the apparent differences between local and unbound antecedents in Slavic languages can be reduced to precisely the same locality conditions that apply, e.g. in English, and “… follow from the interaction of the ECP with whether Infl is a lexical or a functional category.” (Cole/Hermon/Sung 1990: 2, 20)). “Infl” should be lexical, then, in Russian and Polish, and some other Slavic languages. Thus, (166) presents the combination of a free relative construction exhibiting the categorial matching effect without case matching (Ok, if overt, could be instrumental Case), with an empty operator binding a free pronominal or, perhaps, variable. We have to do a certain amount of work to impose an interpretation on (166). A Czech example licensed by um constraint (B,b). (167)C

Zakázal

jsem sii (PROi cokoli [aux 1ps sing] [refl dat] (I) forbad myself whatever ‘I forbade myself to want anything (I like)’ (Example from V. Havel’s play ‘Asanace’).

chtít) to want

One would expect that order (and counterparts) could not be licensed by the marked version of (B,b) since the pairing of the interested participants functioning as controlled identification does not seem to be easily realizable. Still,

80

Chapter 4

if it can be, as in the slightly bizarre situation of (168), we have evidence of licit control favoured possibly by an empty dative object, a discourse related pro possibly marked [+ dative]. (168)C

Šéfi prikázal

[proj] (PROi být jmenován [dative] [part pass] The principal ordered (someone) to be nominated clenem nové komise) [instrumental] [genit sg] a member of the new comission (example from Panevová)

Thus, unlike doporucit (see (162)) prikázat is less amenable to um (B,b) subject to (G), compare (169) and (170). (169)

He ordered himi (PROi to be examined by a doctor.)

(170)C

*pro Prikázal mui (PROi být ošetren lekárem)

If the marked application of (B,b) licenses Czech (168), Russian is recalcitrant in this respect. (171)R

??Nacal’nik

prikazal (PROi byt’ naznacennym [part pret pass masc instrum] The principal ordered to be nominated clenom novoj komissii [instrumental] a member of the new commission i

Russian of course may avail itself of control according to um constraint (B,b) to express the content of (171).

(172)R

Nacal’nik j

prikazal

 proi  nam  i 

PROi

naznacit’

[dative] [acc] The principal ordered (us) to nominate sebjaj clenom novoj komissii) [instr] him(self) a member of the new comission Sebja is long-distance bound by the nonlocal antecedent nacal’nik (R˚uzicka 1973).

The persuade subclass and cognate ones

81

What, for example, (168) illustrates again is the fact that though anchored within a firm framework of constraints control can give way even to particular pragmatical situations choosing the alternative controller. Thus, by meeting the marked constraint (B,b), (168) does not take the path predicted for the subclass, namely, violate the um constraint, a violation that is restored to acceptability by some kind of reconstruction if the corresponding parameter (G) is set positively. Still, all possible cases are in the grasp of the two versions of constraint (B,b) and the associated parameters.

4.5. Ambiguous allow in Control and ECM-context Allow-type verbs belong in the subclass that selects human and nonhuman subjects. Finer grained selectional features may help resolve the familiar but not clear-cut deontic versus epistemic ambiguity of allow/permit verbs. Larson (1991: 136ff.), following Mittwoch (1976), proposes to distinguish between the two interpretations by identifying a postulated implicit dative argument either with the Exceptional Case Marking subject (deontic), or by construing it as a generic or arbitrary pronoun (equivalent to one) (epistemic): “… allow/permit [α to VP] (to β) a. Deontic: β=α b. Epistemic:β=proarb” (Larson 1991: 137) This analysis is of course tailored to the assumed English ECM-structure of the sentential argument: “… ECM-structures with allow are actually oblique dative structures with the ECM complement corresponding to the theme object and with an implicit oblique dative argument whose identity determines the understanding of the complement as deontic versus epistemic.” (Larson 1991: 137) Control is not involved at all: “We see that despite superficial similarity, the pair in (74) (our (173), RR) involves control only in the first member; the second is an ECM structure. (173) a. John promised Mary to leave. b. John allowed Mary to leave.” (Larson 1991: 137) The ECM-analysis implies an overt complement subject that “absorbs” the overt appearance of a matrix oblique argument. Only one of them can “materi-

82

Chapter 4

alize”: Either the oblique argument in the shape of the ECM subject and identified with it by a. (deontic), or the independent ECM subject, with the oblique DP reduced to the obligatory proarb (epistemic). Compared to the Control analysis, the empty category is exchanged, the identification of PRO is reversed: The identification of PRO with the (overt) oblique object, effected by control, is replaced by the identification of the oblique part (to ß) with the ECM subject. The reversed “control” is made optional. It holds for, and “… determines the understanding of the complement as deontic …” (Larson 1991: 137) Larson seems to face the contingency that a given sentence may receive a twofold understanding, that is, its ambiguity cannot be resolved. More importantly: Does Larson’s analysis, or some elements of it, generalize crosslinguistically, in particular apply to languages in which there is no ECM-structure, a question which may bear on the correctness of the ECM analysis of complements of allow in English? First, in non-ECM-languages, the identity condition a. for deontic readings would have to be reinterpreted if possible. As indicated above, Larson’s deontic reading condition β=α emerges from the control relation and its reversed reflexion in the shape of an ECM complement in which the ECM subject takes the “controlling” part identifying the presumed “implicit” oblique argument pro. The ECM-infinitival-subject must be lexically filled and since no other position is available that could be lexically filled, the oblique argument must be “silenced” as pro. Second, how can the epistemic reading condition β=proarb with proarb in the fixed shape of the oblique dative argument in English ECM-structures, be related to its presumed counterpart in control structures, for example, in Russian? Epistemic proarb, of course, is not excluded from corefering with the ECM subject, but is referentially independent of it in principle. Under the corresponding control analysis, the dative oblique argument, lexically overt or covert (pro), implies identity with the PRO subject of the clausal complement by definition, and that in both deontic and epistemic variants. Consider (174), a case of possibly ambiguous reading. (174)

Congress has passed a law allowing [juveniles accused of violent crimes to be tried as adults.]

The presumed proarb does not corefer with the ECM-subject, but may be construed as corefering with the DP of an implicit by-DP phrase. The nearest counterpart of (174) in Russian would be (175).

The persuade subclass and cognate ones (175)R

83

Kongress prinjal

zakon pozvoljajušcij [acc sing] [part pres acc sing] Congress (has) passed a law allowing podvergat’ sudu junošej obvinjaemych v [infinitive] [dat sing] [acc pl] [part pres pass acc pl] to subject (to) court (to try) juveniles accused of nasil’stvennych prestuplenijach kak vzroslych [acc pl] violent crimes as adults

(175)′R … pozvoljajušcij proi (PROi podvergat’ sudu junošej …) (175) is a control structure with pro controlling PRO. The crosslinguistic problem, which may bear on the tenability of the ECM-analysis, seems to hinge on the status of the oblique dative argument DP, and the contrast between the passive complement in (174) and its Russian (active) complemental counterpart. The oblique position occupied by pro in (175′) can of course be filled with lexical elements, e.g. dative sudu (court) or jurisdikcii (jurisdiction), which controls PRO in accordance with um (B,b). Forming the counterpart of the passive complement in a Russian control structure would require that the ECM-subject juvenils … emerges as PRO controlled by the oblique dative argument junošam (juveniles). (176)

*… pozvoljajušcij junošami (PROi byt’ [dat pl] allowing juveniles to be privlecennymi k sudebnoj otvetstvennosti [part pret pass instr masc pl] called to legal account kak vzroslye as adults ‘… allowing juvenils to be tried as adults.’

(176) violates um (B,b) and is ruled out. We conclude that Larson’s ECMapproach and the control analysis of allow and its counterparts, e.g. for Russian, cannot be recoded in each other’s terms. In control structures, the identity fixed by a. for exclusively deontic readings holds in either case, deontic and epistemic: The obligatory pro-shape of the implicit dative object inversely corresponds to the likewise “implicit” complement PRO-subject,

84

Chapter 4

and the lexically overt ECM-subject can be revealed only in the oblique position of the matrix clause. In other words, the obligatory lexical appearance of the ECM-subject does the work done by control, e.g., in Russian. It follows that under the ECM approach there is no means of constructing or computing a syntactic direct counterpart of the twofold “empty” configuration of pro identifying PRO as in (175)′. If the controlling element in the matrix clause itself is empty as pro, its counterpart in English could turn up only out of the ECM-subject position. In (174), a position that takes an argument corresponding to the argument in the D-structure subject position is available as the implicit by-phrase which satisfies the wanted nonlexical status. Thus, the passive shape of the (ECM-) complement is the only syntactic device for coping with the pro to PRO relation as effected in (175)′. The control relation in turn has to pay for its privilege by ruling out just the passive complement clauses considered to be its counterpart to the English hypothesized ECMstructures as in (174). Before coming to conclusions that again relate to parameterization of a type already suggested, let us consider another example. (177)G

Der Bildungsgrad

gestattet ihmi nicht (PROi zu [dat] the degree of education permits him not to wissen, daß …) know that … ‘The degree of education does not permit him to know that …’

Clearly, (177) has an unambiguously epistemic reading. Replacing der Bildungsgrad with sein Vater (his father), and zu wissen with die Disko zu besuchen (to visit the disco) would result in a deontic reading. In summary, Larson’s proposal to account for both readings is on the wrong track. The crucial distinction between construing the presumed implicit dative with the ECM-subject (deontic reading) and construing it as a generic or arbitrary pronoun equivalent to one (epistemic reading) is empirically unsupported. In particular, the construal β=proarb, lifting the identification of β=α which is the counterpart to the control identification in non-ECM languages, appears to be an ad hoc adjustment to the ECM structure. The identity of the dative argument with the complement subject is unaffected by epistemic readings, and implemented by control, as shown in example (177). Thus, the potential nonidentity of the two arguments does not appear to be a possible distinctive property of epistemic readings. There are no conceptual arguments either for

The persuade subclass and cognate ones

85

the syntactic ECM-analysis of allow/permit complements. Abandoning control and resorting to the ECM-analysis in English is not a necessary answer to the heterogeneous empirical problems which have appeared to call for it. First, abiding by the control analysis in English poses the question of how to account for admissible passive complements, e.g., in (174) — juveniles … would have to be analyzed as the object DP controlling PRO in the complement — and their exclusion, e.g., in Russian, cf. ex. (176). On my approach, (175 satisfies unmarked (B,b), while (174) is a violation of it, which is remedied by the positive setting of the interface parameter (G) in association with corollary (D) described above (4.1.) for the persuade subclass. Second, what has given rise to the “epistemic” understanding, can naturally be understood as metaphorical use of allow/permit and their counterparts in other languages, which, similar to an attitudinal predicate, means something like “be compatible with …”. Although this slightly figurative use of allow, permit could be expected to weaken their sensitivity to the control constraint, namely, (B,b), they still obey the relevant constraint in its systematic, parametrically specified effects. In other words, violations of unmarked constraint (B,b) are dealt with depending on the respective settings of the interface parameter.

4.6. Lexically saturated theta-roles Consider Russian rasporjadit’sja (give instructions) and German anordnen (arrange, direct). They do not subcategorize for a direct or oblique object but conceptually implicate a goal- or addressee-argument, which does not show up. Nor can it taken to be a syntactically active null object. If the conceptually implicated argument could be materialize as an NP-object, it would qualify as a controller in a control relation licensed by unmarked (B,b), that is, 〈intact,–〉con, 〈intact,+〉PRO (see 3.2.1.). (178)R

Bolee vaznyj redaktor rasporjadilsja (PRO (a) more important editor gave instructions peredat’ skorbnuju novost’ v efir) [acc sg fem] to transmit the sad news into the ether

86

Chapter 4 (179)R

Slovom, generali vystavil Kosyginaj za in short the general turned Kosygin out of i rasporjadilsja (PROarb razoruzit and ordered/gave instructions to disarm

dver’ doors proj) (him)

Control verbs participating here, theta-govern (Chomsky 1986a: 70) neither an overt nor a syntactically noninert null NP-object. The external argument is the only NP argument, which is precluded from controlling by the um constraint (B,b), but in case of passive controlled clauses may be a licensed controller by virtue of its marked version. Thus, PRO stands alone, but it is identified in accord with constraint (B,b) by the conceptual-pragmatic construal of a domain of individuals that participate as “goal” or “addressee” in the event described by the matrix verb, for example R. rasporjadit’sja (see (178, 179)). In other words, the construal of the domain of individuals is discourse-related and the identification of (an) individual(s) referred to by PRO is dependent on this construal or “arbitrary”. Since no syntactic position for an object NPargument and thus no possible filler of it is available21 Rizzi’s notion of “lexically saturating a theta-role” could help: “… it is not unreasonable to assume that if a theta-role is saturated in the lexicon, it is not syntactically visible; hence, in particular, it cannot act as a controller.” (Rizzi 1986: 552)). The general constraint (B,b), however, extending its force beyond the syntactic processes, maintains the basic rationale of control. Importantly, the positive setting of the null object parameter (cf. 3.3.) seems to account for grammaticality, in contrast to English *this leads [PRO to conclude what follows] (Rizzi’s example), although no null object is involved. A German counterpart of Russian rasporjadit’sja is anordnen. See (180). (180)G

Peter hat angeordnet (PRO die Zimmer aufzuräumen) Peter has ordered the rooms to tidy up ‘Peter ordered the rooms to be tidied up’

Chapter 5

The case of promise

5.1. Tenacity and untenability of subject Control “As is well known, promise is one of a small number of verbs in English that select an object and an infinitive and show subject control.” (Larson 1991: 103)). Thus, subject control of promise, (g. versprechen, r. (po)obešcat’, c. slíbit, p. obiec(yw)a´c etc.) seems to be taken for granted. Admittedly, these verbs have raised problems which linger on in current discussion, but they cannot be plainly considered verbs of subject control, as has been shown in much previous work (e.g. R˚uzicka 1983, Farkas 1988, Wegener 1989, Köpcke/Panther 1993). Besides notorious examples like (181), there are many others which, on equal terms, would justify calling promise a verb of object control. I will provide some cross-linguistic facts first. (181)

Johni was promised ti (PROj to be allowed tj (PROk to leave)) (i = j = k)

For discussion see R˚uzicka (1983a: 310 ff.). (182)R

bez

nego (Stalina)

obešcano [part pret pass neuter sg] without him (Stalin) (it) promised bylo [NP nyne zivušcemu pokoleniju]i (PROi [pret neuter] (to) now [part pres dat sg] [dat sg] was living generation okazat’sja v Kommunizme) to find themselves in communism ‘… in the absence of Stalin (without his having had a hand in the matter) the now living generation was (had been) promised to find themselves in communism.’

88

Chapter 5 (183)R

Emui bylo obešcano (PROi byt’ [dative] [part pass pret neuter] (to) him was promised to be vkljucennym v sbornuju [part pass pret instr sg] included into the select team ‘He was promised to be included …’

(184)G

Ihmi war versprochen worden ((ins Kino) PROi gehen [dat sg] him had promised been to the cinema to go zu dürfen to be permitted ‘He had been promised to be permitted to go to the cinema’

German dürfen can clearly be paraphrased or “decomposed” as be allowed/ permitted, which requires an account analogous to (181). “Object” control is not restricted to passive matrix clauses, nor to passive controlled clauses. (185)F

L’entraineur a promis à Platinii the coach has promised (to) Platini admis à l’équipe nationale included into the national team (Example from Vanden Wyngaerd)

(186)G

Peter versprach Pauli (PROi Peter promised Paul mitgenommen zu werden taken to be ‘… to be given a lift’.

d’ PROi être to be

im Auto in the car

Both clauses are non-passive. (187)C

Petr slíbil Janovii (PROi stát se predsedou) [dat] [reflexive clitic] [instrum case] Peter promised John to become (the) chairman

(188)C

(pro) Slíbili vítezii (PROi dostat medaili) [3 ps pl pret] [dat] (they) promised the winner to get a medal (example from Panevová)

The case of promise

89

An overt or implicit agentive NP within a PP (by-phrase) may be marginally acceptable as controller in passive matrix clauses. (189)G

Sogar vom

Präsidenteni war

ihnen [dat pl] Even by the president it had them versprochen / zugesagt worden (auch den Verlierer PROi promised/(sent word back) been also the looser zum Bankett einzuladen) to the banquet to invite ‘They had been promised even by the president to invite also the looser to the banquet’ (… that he (the president) would invite …)’

In previous work (R˚uzicka 1983a,b), I posited a thematic Identity condition (TIC), which I claimed could account for all types of control illustrated by (181) - (189). The somewhat unclear relational notion of the identity of the theta-roles borne by a licit controller and PRO respectively raised some doubts ° about its range of validity. Jackendoff (1987: 370), referring to Ruzicka a (1983 ) writes: “… it (identity of thematic relations, RR) does not account for the evident object control in (5b) (The promise to Bill to be allowed to leave.)” I think it does account for it as well as it does for (181), assuming that the theta-role assigned to John by promise in (181) either identifies with the thetarole assigned to Bill by to in (5b), or, alternatively, that in both cases, as seems plausible, the same theta-role is taken from the theta-grid associated with promise [{+N-V}] and [{-N+V}]. This theta-role was taken to be identical with the the role assigned to object PRO by allow . Jackendoff (1987: 370) further brings up a notorious instance of control: “His (R˚uzicka’s, RR) solution is not restricted enough, in that it should allow *Bill was promised to receive the letter, where both subjects are goals.” Similarly, Vanden Wyngaerd (1987: 72) comments that “… whereas R˚uzicka can presumably explain (44) (=*? John promised (me) to get permission to leave RR) as a violation of the TIC, he has no explanation to propose for (43d).” (=*Bill was promised to get permission to go, RR). Even though these critical remarks can be answered within the framework of the thematic Identity condition (TIC) and the thematic Distinctness condition (TDC) (R˚uzicka 1983a), the claim that the global thematic roles could be matched so that they determine wellformed control relations may not be

90

Chapter 5

tenable: The rationale of Θ-theory, even in the uncontroversial part of its inventory of theta-roles, has not been concerned with the (nontrivial) identity of, or distinction between, theta-role assignments. Relational and comparative aspects between theta-assignments, in particular between the theta characterization of PRO and that of eligible controller, have gained in theoretical weight since their impact on considerable areas of control has been observed ° (Ruzicka 1983a; Farkas 1988; Pollard/Sag 1991). If it is not feasible, however, to capture this influence on control behaviour by relying on a substantive comparative analysis of global theta-roles, one cannot feel to be on independent firm ground. The crucial properties of well-matched or ill-matched pairs of theta characterizations that license or do not license control can be expressed more substantively and less “costly” in terms of designated abstract lexical features that are interpretable as thematic feature specification values, as I have done so far using constraints in the shape of (B,b) and, if necessary, adding corollaries like (D). Within this framework, the above critical instances of control put forward by Jackendoff and Vanden Wyngaerd — the last example had already been discussed by Chomsky (1980: 35, fn. 40) as one with the expected grammatical status (*John was promised to get permission to leave) — can be accounted for more convincingly. Control phenomena determined by the class of verbs of the promise-type are captured by the same machinery that captures control of the preceding class (ask, persuade etc.). We continue to use the format of constraint (B,b), changing the value of α in its first conjunct, that is, changing it to (B,a) (3.2.1.), repeated here. (B,a)

〈intact,α〉CON ^ 〈intact, α〉PRO

Recall that the conjoined thematic feature specifications express properties of argument-NP’s projected from the matrix verb and the verb (predicate) of the controlled clause, respectively. There is an analogy between subject-oriented behaviour of promise and subject-oriented adverbs like willingly, intentionally. Paraphrasing the promise-event in the spirit of the entailment (10) (see fn. 15), we would say that some x who is capable of acting intentionally, commits him/herself to act so as to bring about P, that is the event decribed in the controlled clause.22 The marked version of constraint (B,a) is arrived at exactly as the marked version of constraint (B,b), namely by reversing the values of α. The positive value of α in the PRO-conjunct is chosen as the

The case of promise

91

unmarked and the negative one as the marked version, in both (B,b) and (B,a). Thus, (B,a)′ is the unmarked, (B,a)″ the marked version of (B,a). (B,a)′ (B,a)″ (B,a)′ of (189). (190)

〈intact, +〉con^ 〈intact, +〉pro 〈intact, −〉con ^ 〈intact, −〉pro accounts for standard examples like (190) and the particular case Peteri promised John (PROi to leave).

(B,a)″ accounts for “unexpected” cases like (181), (183) – (186). (182) will be discussed in the context of broader evidence (below, 5.4.). Marked application needs a lexical controller: The head of the controlling A-chain must be “pronounced”. Crucially, the (un)marked versions of (B,a) and (B,b) respectively, are licensing instruments that generalize about varying empirical conditions of control which are associated with respective (sub)classes of control verbs. In the domain of (B,b), for ask-type verbs, a substantial justification for pursuing the markedness approach, namely the lucid availability of a second layer of semantic characterizations pairing, that is, identifying the “interested parts” (see above, 3.2.3.), subtly divides these verbs from persuade-type verbs, (cf. above, p. 59ff). Violations of the unmarked version of (B,b) again lead to different cross-linguistic fates within this event type (see above, 4.1.). In the domain of (B,a), for promise-type verbs or events, there is no analogous subdivision either with respect to marked or to unmarked control. In the former case, controller and PRO are identified through involvements in promise-type events that are complementary to the unmarked twofold agentive specification and are captured by 〈intact, –〉con/pro. The general symmetry of the proposed solution to promise-type control with the account given for ask/persuade-type control extends to the restriction that PRO can receive 〈intact, –〉 only from a predicate whose theta-grid contains 〈intact, +〉, that is, corollary (D) applies to this subclass as well. (D), thus, unifies two broad domains of complement control defined by the thematic constraints (B,a) and (B,b), respectively, singling out their common interthematic restriction as a generalization about marked control behaviour. Concerning promise, the clear intuitive basis for (D) is the fact that the property of the event described in the controlled clause may offer no hold for the promise to be fullfilled intentionally. If (B,a) and its corollary (D) are not

92

Chapter 5

satisfied in grammar, they still serve as an instrument of systematic, not amnestying, conceptual-pragmatic construal and explanation of control. The construal, if undertaken, tends to result in graded and individually differing acceptability. In the recalcitrant examples repeated here, (191)

*John was promised to get permission to leave (Chomsky’s example)

(192)

*Bill was promised to receive the letter (ex. from Jackendoff)

receive and get — the thematic properties of these verbs are not entirely clear (cf. in particular Dowty 1991: 581) — are not assumed to assign 〈intact, +〉, which , by (D), puts them beyond the reach of marked (B,a), resulting in ungrammaticality as “default” interim. The Czech examples (187), (188) violate um (B,a) , since PRO does not carry 〈intact, +〉. Nor do they meet corollary (D) and, consequently, they violate m (B,a): 〈intact, +〉 is neither in the theta grid of stát se nor in that of dostat. But the dative objects Janovi and vítezi, respectively are the exclusive controllers acknowledged by all informants. How do they come to satisfy pertinent control conditions? I would like to suggest that (187) and (188) are readily construed after patterns like (193) and (194), respectively, which could satisfy corollary (D) and the m contraint (B,a). (193)

Peter promised Jani (PROi to be elected tichairman)

(188) is obviously construed by analogy with (194). (194)

They promised the winneri (PROi to be awarded ti a medal)

In other words, 〈intact, –〉 is reconstituted for PRO through substituting a concept that “assigns” 〈intact, +〉 at the conceptual-pragmatic level: At this level, a representation is provided that can be referred to by the grammatical constraints, marked (B,a) in this case. In other words, (187) – (188) are not straightforwardly wellformed, as far as grammar reaches. They can prove acceptable since the lexical and syntactic skeleton suitable for control paves the way for a pragmatic “detour” that builds up a conceptual structure mirroring, and interpretable as, matched grammatical conditions.

The case of promise

93

5.2. A classical violation Consider (195), which seems to be a straightforward violation of marked constraint (B,a) (see 5.1.). But it needs further elucidation. (195)

*John was promised (PRO to leave)

(195) counts as the classical irreparable violation of syntactic constraints (cf. Chomsky 1980 and R˚uzicka 1983a), which, under this approach, is expressed by considering (195) a violation of the marked constraint: *〈intact, +〉PRO … Recall the directional interpretation of violating um and m constraint. Only the PRO-conjuncts can be the offending parts. Suppose, then, that the offending conjunct *〈intact, +〉PRO of m (B,a) is impervious to any conceptual construal overriding the original feature value and “changing” it into 〈intact, –〉. Still, constructions of type (195) may be expected to survive, if the um constraint (B,a) could be taken to apply to it, that is, if an element carrying 〈intact, +〉CON, which the theta-grid of promise contains, is understood to be present in the matrix clause. Indeed, some English informants have not downright rejected (195), apparently imagining an implicit agentive controller. The rationale of thus doing “violence” to English emerges more transparently in German, Russian and Czech examples. (196)G

Dem Trainer der Nationalmannschaft war [dat] to the coach of the selected team (it) had von [den Offiziellen des italienischen Klubs ]i versprochen by the officials of the Italian club promised worden (PROi van Basten für das Qualifikationsspiel been van Basten for the qualification match freizugeben) to release ‘The coach of the selected team had been promised by the officials of the Italian club that they would release van Basten for the qualification match.’

(197)R

Borisu bylo obešcano [dat] [neuter pret] [part neuter] Boris (it) was promised

94

Chapter 5 [direktorom školy]i ( PROi ne isklucit’ ego [instrum] by the headmaster of the school not to exlude him iz ekskursii) from the excursion (198)C

Bylo

mu slíbeno od [trenera]i (PROi zaradit [dat] [part pass neut] (it) was him promised by the coach to include ho do vybraného muzstva) [gen] him into the select team

German and some Slavic languages differ from English in that their exclusive impersonal passive in the matrix clause and in particular its overt (instrumental Case) or implicit agent phrase carrying 〈intact, +〉CON matches 〈intact, +〉PRO, thus satisfying um (B,a), as in (196) – (198). The division suggests a derivative parameterized variation in the range of syntactic projections associated with the respective counterparts to promise. The projection of an impersonal passive contrasts with, but functions like, the nonimpersonal passive in English after dative shift. Attributing the particularity of dative shift in English to its strong configurational type and excluding its pertinence to control for (morphological) Case-oriented languages like Russian, Czech and German, which preserve morphological dative in the passive, we trace back the control division to more general parameter settings. In English, agentive by-phrases, overt or not, seem to be unable to control, depending on the structure of the embedding control element.23 The problem reduces to the issue, why the NP in a by-phrase of a passive matrix clause cannot control PRO in an infinitival complement clause. NP’s in byphrases figuring as arguments of verbal nouns that derive from control verbs do control, as shown in (ii) b. c. of fn. (23). I assume that structures like (199) (199)

*The men were promised (by Frank) to leave,

are ruled out since the hierarchically dominant position of the derived subject does not yield control to the DP in the by-phrase, although the latter would satisfy unmarked (B,a), the subject itself being not eligible as controller by virtue of PRO violating marked (B,a). Corresponding passive structures, e.g., in German, Russian and Czech, in which the agent phrase is demoted — as in

The case of promise

95

English — but the argument carrying the θ-role Goal or Recipient is left in oblique (dative) object position — though excluded from controlling by PRO violating m (B,a) —, have no subject rivalling for control. The counterpart to the agent phrase is not hindered from controlling in accord with um (B,a) (see ex. (189), (196) – (198)). In English, when the hierarchical superiority of the (derived) subject disappears as in the nominals illustrated in footnote (23), the DP in the by-phrase can maintain control. The descriptive generalization (I) capturing these differences of control between English on the one hand, and, e.g., Russian, Czech, German, on the other, is a specification on the unmarked application of (B,a). (I)

The first controller-related conjunct in the um (B,a) can be satisfied by a demoted agent phrase (Case-marked as instrumental or PP), iff there is no raising to subject position, that is, the subject position is not filled by the head of an nonsingleton A-chain.

This is the case, for example, in Slavic languages and German (see ex. (189, (196), (197)), in which the argument-DP corresponding to the DP that is raised to subject position in English, is assigned morphological dative Case in the resulting impersonal passive. No syntactically dominant rival for control appears. Still, control by the demoted agent argument marked Instrumental Case in Slavonic languages or shaped as a PP as in German may be judged to be marginal. Note that (I) is not complementary to (F) (3.2.2.), which syntactically specifies the marked application of (B,b) to a pertinent subclass of control verbs. (F) cannot be cancelled under structural consequences of the distinctly parameterized morphosyntactical treatment of Case, which, e.g., results in the different phenomena of passive under discussion. Consider first (200). (200)

*Frankyj was asked by mei (PROi,j to be made his partner)

(200), in the i-indexed interpretation, is ruled out by (F) specifying m (B,b). The unmarked application of (B,b) fails in the offending PRO-conjunct (indexed j). From a syntactic point of view, we might expect that control verbs falling under constraint (B,b) which inherently assign dative Case retained in the impersonal passive (verbs under 4.4.4.), by analogy to (I), override (F) satisfying the first conjunct of m (B,b) by the demoted agent-DP. This prediction is

96

Chapter 5

not confirmed. Consider (201), an example from Russian. (201)R

*pro emu bylo poruceno [starym znakomym]i [dative] [neuter sing] [part pass neuter] [instr. Case] (it) him was charged (by) an old acquaintance (PROi byt’ vkljucennym v ekskursiju) [part pass instrum sing] to be included in the excursion ‘He was charged by an old acquaitance to be included in the excursion.’

The reason is that the given subclass, while subject to (B,b), cannot respond to its marked application since the relevant thematic pairing is not available to support it as it is, e.g., in ask-type control verbs (see above, 3.2.3.) We might speculate that unmarked licensing as in the case of (B,a) maintains its strength under syntactic conditions that change hierarchical relations, the more so when they are levelled by independent factors as in the impersonal passive in, for example, German and Slavonic languages. In general, the situation discussed corroborates my predictions of the strength of relations between lexically thematic features in determining control conditions. Considering structures like (199) that “crash” by virtue of m (B,a) and (I), evidence of hierarchical pressure opens the way to a natural explanation of the contrast between (202) and (203). (202)

The promise to Bill was to leave.

(203)

*The promise by Bill was to leave.

The contrast can hardly be elucidated by resorting to a general distinction between obligatory and nonobligatory control as proposed by Williams (1987a: 370). Let us first turn to the question why (203) is ungrammatical and the apparently very similar (204) (=(ii) b. of fn. 23) is judged grammatical. (204)

John expects a promise by Mary to remain faithful to him.

The relevant distinction between (204) and (203) consists in the structural positions of the controlled infinitival predicating of the DP that contains the N incorporating the control verb. I would like to suggest that if the control element, that is, a noun incorporating a control verb, is the head of the NP within a DP in subject position and the controlled infinitival clause is excluded from, and predicating over, the subject-DP, the agent (by-phrase) within this

The case of promise

97

maximal subject DP cannot contain a licit controller (ex. 203). In other words, a potential controller within the by-phrase cannot control beyond the maximal DP containing it, that is, into the predicating infinitival clause. Control in (204) is licit, since (204) sharply differs from (203) in that the controlled infinitival clause is part of the maximal DP, with promise heading all the rest. Assuming by analogy to the hierarchy in clausal structures like (199) that the DP in the agent by-phrase cannot control beyond its maximal DP by virtue of the “superiority” of the “subject”-phrase and in particular the Specifier position available in the latter, we should predict that a controller marked 〈intact, +〉 is licensed if within the Specifier position of the subject phrase. Thus we have the contrast between (205) (=(203)) and (206), (207). (205)

*The promise by Bill was (PROi to leave)

(206)

Bill’si promise was (PROi to leave)

(207)

Ouri promise was (PROi to work immediately)

It follows that a subject-DP containing an article or other Determiner as in (202) (repeated here as (208)) but lacking an agent phrase can provide an implicit controller. (208)

The promise to Bill was (PRO to leave).

(206), (207) satisfy um constraint (B,a), (208) requires the implicit controller to receive 〈intact, +〉 conceptually. We should expect now that the marked version of constraint (B,a) applies as well and, by analogy to (181), licenses experimental sentences like (209), at least marginally. This expectation seems to be fullfilled. (209)

?The

promise to Billi was (PROi to be allowed to leave).

The account given for the facts illustrated by the contrast between (203) and (204) and between (202) and (203) fits in with the suggested condition of immediate proximity (see above, 2.2.2.2.).

5.3. Violating the unmarked constraint Systematic violations of um constraint (B,a) should be expected to be uniform cross-linguistically: 〈intact, +〉CON & *〈intact, –〉PRO.

98

Chapter 5 (210)

*Hei promised them (PROi to be wronged by such comments).

Still, acceptability can be canalized with the support of lexical and contextual restrictions in discourse. While (211) illustrates the now familiar case of control licensed by m constraint (B,a), (211)G

Der Lehrer versprach dem Schüleri (PROi [dative] the teacher promised the pupil in die nächste Klasse versetzt zu werden) to the next class moved up to be

(212) exhibits a violation of um (B,a) which apparently maintains acceptability in the same way as violations of um (B,b) in persuade-type control, namely through according with corollary (D). Crucially, parameter (G) is not operative here. Languages do not seem to differ in acknowledging acceptance of structures like (212). Causality switch is evidenced by promise and its counterparts as well as by persuade. It is intrinsic, however, to (passive) complements of the former verb class by virtue of its commitment content. As to passive complements of persuade (see (115), (116)), causality switch is something to be achieved by persuading. It is “underspecified” and full interpretation is subject to parameterization. (212)G

Der Schüleri versprach seinen Eltern (PROi [dative] the pupil promised his parents in die nächste Klasse versetzt zu werden) to the next class moved up to be (examples from Wegener 1989)

Note that the basic syntactic and semantic structure of promise events on which constraint (B,a) rests is instrumental in interpreting (212): The pupil commits her/himself to do what is necessary to bring about her/his being moved up. With control vacillating between subject and object, and partly depending on their semantic content and relatedness to each other, the basic control relation embodied in um and m constraint (B,a) is unchangingly the pivot on which all pertinent instances of control turn. Corollary (D) enables us to regulate the operation of (B,a) in the neutral zones of control arising from the distribution and absence of 〈intact, +〉 in the theta grid marking the arguments of the controlled clause. Recall that corollary

The case of promise

99

(D) is interpreted as a condition which makes the assignment of 〈intact, –〉 to PRO dependent on the presence of 〈intact, +〉 in the same theta grid. A PRO which is not specifiable as 〈intact, +〉 or 〈intact, –〉 should be highly sensitive to controllers marked 〈intact, +〉 and 〈intact, –〉 respectively. Possible shifts of control are predictable, then, from the theta-grid of the controlled clause and the content of the lexical insertions. Taking the German verb kommen, at least in one reading, to lack 〈intact, +〉 in its theta grid, or more neutrally, as an ergative or unaccusative that lacks this thematic or semantic specification, we can predict the situation in (213): The relevant construals go beyond the computation of grammar proper. (213)G

Der Anwalti versprach dem Häftlingj PROi,j bald the lawyer promised the detainee soon aus dem Gefängnis )   j zu kommen  aus dem Urlaub ) i   from prison  to come from holidays   ‘... soon to be discharged from prison.’ ‘... soon to return from his holidays.’    (example from Wegener 1989)

The drastic shift of control would seem to follow from substituting the NP dem Gefängnis (prison) for dem Urlaub (his holidays) in the [PPX aus NP]. The choice dem Urlaub implies the construal of standard subject control, the choice (dem Gefängnis) inclines one to prefer object control. (213) is a diagnostic example. The standard control relation with der Anwalt controlling may withstand the bizarre interpretation connected with it in accord with um (B,a) and from prison. But subject control may be readily overcome by a simple construal which links PRO aus dem Gefängnis zu kommen (interpreted as be discharged from prison) to detainee and, thus, may be construed to satisfy m constraint (B,a), with PRO being conceptually marked 〈intact, –〉PRO.

5.4. A structural-pragmatic conflict? I will go on experimenting on the sensitivity of “neutral” PRO to independent semantic changes and the resulting trade-off between the principled con-

100

Chapter 5

straints and lexically effected construal. It would appear, under this approach, that if a certain amount of interpretive work can be done that achieves some conformity to the unmarked or marked constraint (B,a), considering their respective semantic and/or conceptual underpinnings, acceptability can be lent to deviating structures independently of further grammatical limitations. But I have not discussed so far that other syntactic or thematic conditions may prevent sound conceptual-pragmatic construals from being implemented, in particular that failing Corollary (D) may not be overcome by any pragmatic construal. Let us focus for a moment on the interpretation of (214) and its German version (215), where syntactic and pragmatic factors come into conflict with each other. (214)

Gladys promised Hubert to experience victory again.

(215)G

Gladys versprach Hubert, noch einmal einen Sieg zu erleben

Comrie (1985: 63) comments on (214): “The most coherent interpretation of (59) (= (214), RR) pragmatically, given that a promise is to the benefit of the addressee, would be that Gladys is promising Hubert that he, Hubert, will experience victory in the future, but English syntax excludes this interpretation … Abraham (1983: 221) cites German equivalents of (214) … as fully grammatical and acceptable in the pragmatically coherent interpretation.” Does English provide the alternative, but “incoherent” interpretation, or reject (214) altogether? Or, put differently, why does the possible “coherent” interpretation have the effect of acceptability in German, but not in English? What is it in English syntax that resists the coherent pragmatical “escape” — assuming that Comrie’s judgment of (214) is correct? In German, acceptability of (215) is not independent of lexical features arrayed as the marked constraint (B,a). But the construal by which it can be achieved requires a considerable amount of interpretive work, since corollary (D) is not met. Gladys must be assumed to have a hand in Hubert’s experiencing victory. Since PRO is neutralized, which is expressed by violating corollary (D), the conceptual effort to attain coherence does not seem to override the unmarked control pattern (B,a), though not complying with it. This is the English value of the syntactic-pragmatic interface parameter. The contrast, which may rank as a parameterized variation, in general should be ascribable to the more unyielding syntactic skeleton of a strongly marked configurational language like English compared with German.

The case of promise

101

In Russian and Czech, which rely on configurational-geometrical structuring to a considerably lesser degree, the counterparts to (214), (215) are unobjectionable. (216)R

Boris

poobešcal Ivanui (ešce raz PROi ispytat’ [dative] Boris promised Ivan once more to experience cuvstvo pobedy) [genit] the feeling of victory

(217)C

Boris

slíbil

Ivanovii (ješte jednou PROi [dative] promised Ivan once more vítezství

Boris zazit [acc] to experience victory

I would like now to experiment on taking Gladys (and Boris, respectively) as the controller in (214) – (217), construing observance of the um constraint (B,a). This construal again would have to manipulate PRO which carries neither 〈intact, +〉 nor, by corollary (D), 〈intact, –〉. Gladys must be held capable of calling forth her own feelings, an ability which might be represented as 〈intact, +〉PRO at the pragmatical level. As mentioned above, English grammar cannot be supposed to endorse this “incoherent” interpretation although it has no “remedy” for it, since it may find support in the standard um control relation. Russian (182) again is a case in point. PRO is unambiguously controlled by the dative DP. Though it violates marked (B,a) it does not resist construal in spite of failing corollary (D). Absence of 〈intact, +〉 in the thematic specifications assigned by okazat’sja excludes any construal in the direction of satisfying um (B,a) with the implicit Agent controlling as the explicit one does, e.g., in ((196)–(198)). If Comrie’s judgement of (214) is correct, configurationally prominent English Syntax thwarts construal of the “coherent” control reading although it does not turn subject-controlled (214) into an coherent one. Note that a construal that results in complying with m (B,a), via bypasssing corollary (D) — as in the German reading of (215) and Russian (216) — is much more feasible than construing compliance with um (B,a) by reconstituting 〈intact, +〉 for PRO.

102

Chapter 5

Such borderline cases of control demonstrate that rather subtle and even empirically controversial instances of control crystallize around the basic constraints and their respective m and um interpretations. Even in cases licensed by m (B,a) like (218), Comrie prefers the “syntactic” reading. (218)

Penelope promised her son to be allowed to leave.

“… for me the interpretation ‘Penelope promised her son that she (Penelope) would be allowed to leave’ is available, and it is in fact preferable to the interpretation ‘Penelope promised her son that he (her son) would be allowed to leave’ … Of course, on the preferred interpretation it is necessary to assume that Penelope has some influence over whether or not she is given permission to leave.” (Comrie 1985: 53). Note that in the less preferable (for Comrie) interpretation it is necessary to assume that Penelope has influence over whether or not her son is given permission to leave. In the preferable (for Comrie) control relation, um (B,a) is violated. Acceptability or interpretability may be restored in the way described above (ex. 212). On the interpretation, clearly preferable in German, Russian and Czech, in which her son is the controller, m (B,a) is satisfied. This discussion strongly suggests that confining oneself to a treatment of control phenomena that exclusively relies on structural notions of grammar proper simplifies, even distorts the picture. It seems to be typical of control phenomena that some questions raised in this section may be unanswerable if not meaningless on a sound grammatical basis. Grammar, of course, is responsible for fuzzy edges of interpretive capacity in the control design, which possibly arise because grammar provides and makes available wellformed bare structural skeletons that satisfy enough principles to leave the linguistic rest to varying judgments of convergence. I draw two conclusions from these considerations. First, we need a device that mediates between the fixed, the “hard” syntactic control structures and the “soft” areas of conceptual-pragmatic manipulation. Constraints (B,b) and (B,a), if specified by (D) - (I), constitute these intermediary principles. Second, restriction to configurational syntactic aspects must fail, since they do not reach and get down to some substantive or critical issues. We have to face situations which make it very difficult to intuitively distinguish between interpretability, acceptability and grammaticality of control expressions. I would like to speculate that control phenomena have conspired to create an experimental plot trying out and straining conceptual-pragmatic interpret-

The case of promise

103

ability beyond what can be restrained grammatically. But any such facts, being observed and grasped in nongrammatical terms, in turn are referred to, and finally elucidated by, appropriate grammatical principles, or principles that can be related to grammatical ones under a unifying approach. For example, if, in a given discourse situation, the choice of a controller in fact depends on the intersection of the interlocutors’ respective actual knowledge sets and in particular, on “… sets of (their) matters of current concern” (Yokoyama 1986: 141 ff.), such factors can be related to general constraints in the shape of theta-theoretical specifications. A case in point is (219) (example from Farkas). (219)

The mother promised the children to stay up.

The mother is a licit controller by um (B,a). The discourse situation is quite usual, of course, which suggests that the children figures as licit controller. The pragmatical construal by virtue of which the children qualifies as controller amounts to nothing else than interpreting the complement clause as something like to be allowed to stay up, which can be referred to by the m constraint (B,a) (cf. ex. (184), (218)). This is exactly the reason why the syntactic and lexical counterpart in German precludes object control, but of course not subject control. (220)

*Die Mutter versprach den Kinderni (PROi aufzubleiben)

In German, as already noted (3.3.), a lexical nonpassive equivalent for be permitted/allowed to is available. The construal of object control is “superfluous”, in German, whose only equivalent to (219), interpreted with object control, is (221). (221)

Die Mutter versprach den Kinderni (PROi

aufbleiben to stay up

zu dürfen, … to be permitted Here is an exact parallel to the preclusion of word for word German counterparts of (91), repeated here. (91)

Hei *Eri

asked PROi bat PROi

(91) can only be rendered by (222).

to leave zu gehen

104

Chapter 5 (222)

Eri bat (PROi gehen zu dürfen)

(219) and (91) are violations of the marked constraints (B,a) and (B,b), respectively. Conceptual construal, as noted above, makes up for them. The results are referred to by the respective m constraints and licensed. (221) satisfies m (B,a) directly.

5.5. Larson’s division of labor Larson (1991: 125 ff.) takes great pains to defend his configurational theory of control and in particular his version of Rosenbaums’s (1970) “Minimal Distance Principle”24 in the face of “… two well-known lines of criticism, one involving the specific verb promise and one involving other superficially similar ‘double object’ verbs like ask, teach, and allow … The structural relations among the complements of promise in (35a) (my (223): RR) do not appear different from those in the nonpassivized example (35b) (my (224): RR); hence, a shift in control is unexpected:” (223)

John promised Mary to be allowed to leave.

(224)

John promised Mary to leave.

Under our approach, (223) is clearly licensed by m constraint (B,a). Larson suggests “that the interpretation of the infinitival in (43a) (= (223): RR) is determined not by control, as in (43b) (= (224): RR), but rather through entailments, as in (43c) (= (225): RR). (225)

John promised Mary permission to leave.

“… On the suggested view, what governs object construal with promise is simply the standard set of entailments that this verb carries by virtue of its dative status. In general, then, construal of the understood subject of an infinitive has no specific principles here. Construal by control follows from the (independently motivated) shape of D-structure. And construal by entailment follows from the general entailments of the verb.” (Larson 1991: 126, 128, fn. 17) The “… general ‘division of labour’ suggested here between control and entailments seems plausible and seems moreover to answer to the general intuition that examples of ‘shifting control’ differ in status from their controlled counterparts — that although they can be made acceptable to a

The case of promise

105

degree, they are never as natural as the latter …” (op. cit. 129) Apart from the vague qualifications of “naturalness” and “status”, Larson’s endeavour to draw a neat line between independent D-structure shape and construal by lexical entailments fails to capture their intrinsic relatedness and their interaction and, contrary to his expectation, gives away the chance of drawing a line of demarcation that marks their interface. If examples like (223), (225) are held to be determined not by control, but rather through entailments specific to the lexical element promise, then, on perfectly equal terms, one can claim that (224), being determined by control, is determined through entailments of the lexical item promise. In (223) and (224), lexical properties are differently activated by, and related to, general principles such as (B,a), as a consequence of the distinct syntactic and functional structures of the embedded predicate. Larson has to admit that “One important question that I must leave unresolved here, but which must be addressed ultimately if this suggestion is to amount to an actual solution, is when precisely an infinitival complement is subject to control versus construal by entailments.” (op. cit. 129) This question is misleading. Its empirical rationale and problems can be dealt with in a distinct and unified system, as should be clear from our approach.

5.6. The analogous case of threaten Comrie (1985: 59) observes that “… indeed the main difference between a promise and a threat is whether the speaker believes the outcome to be to the advantage or to the disadvantage of the addressee.” S/he who threatens does not commit her/himself to fulfil the “promise” even if the addressee does not do or does what the threat tried to press or force her/him to do and not to do, respectively. Still, threaten and its cross-linguistic counterparts mirror the control behaviour of promise. In English, idiosyncratically, an addressee-NP does not go with a controlled infinitival clause: “English threaten is a control verb only in the absence of a main clause object …” (Comrie 1985: 52). Counterparts to threaten in other languages should be expected to be licensed by observing the same constraint as promise, that is (B,a). (226), e.g., an example from Fauconnier (1976), discussed also by Vanden Wyngaerd (1987: 82), meets m constraint (B,a).

106

Chapter 5 (226)F

Pierrei a été menacé par Paul Pierre was threatened by Paul subir les pires tortures) to undergo the worst tortures

(de PROi

The theta-grid of the “functional” verb subir coalesces with that of torture(s) in assigning 〈intact, –〉 to PRO.

5.7. Considering the raising analysis of promise and threaten In general, control behaviour of promise- and threaten-type verbs is more elastic, it yields more easily to metaphorical use and resulting flexibility. Metaphorization has gone so far that reanalysis as a distinct syntactic construction type has come to be acknowledged. Zubizarreta (1982) and Johnson (1985) point out “… that such verbs as threaten and promise have, in addition to their well-known control use, a raising use as well … Johnson notes that there are certain restrictions on the theta-role borne by the raised NP … One such restriction is that the raised NP may not have the role which Johnson calls ‘intentional agent’ … Thus, on the raising interpretation, (34) (= (227): RR) is ill-formed, and (35) (= (228): RR) is unambiguous, allowing only the theme interpretation for John and precluding the agent interpretation. (227)

*Mary threatens to slap Bill.

(228)

John promises to please Susan.”

I have quoted from Lasnik (1988: 6 ff, his numbering is (34), (35)). Under my approach, (227) is licensed by observance of um constraint (B,a), (228) is ruled out by it as a control construction: PRO does not carry 〈intact,+〉. In addition, it does not meet Corollary (D) (the theta-grid of please does not contain 〈intact,+〉) and, thus PRO cannot receive 〈intact,–〉, which, independently of the single potential controller renders impossible a pragmatical control construal that matches m (B,a). It is exactly this situation which may urge one to “withdraw” 〈intact, +〉 from the matrix external argument, a dethematization that must reduce it to the head of a chain linked to the empty clausal subject, which becomes its trace. The “resulting” Raising variant clearly is related to the Control construction: If the theta-specification of the (empty) complement subject-NP fits in with that of the external argument of

The case of promise

107

promise/threaten, interpretation as an instance of (subject) Raising is excluded (the case of (227) above). If it does not, that is, neither um nor m constraint (B,a) can be met, a metaphorization, which includes the removal of the matrix argument, is one of the possible construals, and a Raising analysis may be a syntactic account for it. The Raising situation also arises when the potential controller cannot be a recipient of 〈intact,+〉, which extends to PRO of course. (229)

The skeleton key promises to open the door.

(230)

The missile threatens to destroy the city. (Lasnik 1988: 6 f.).

A Raising verb “… imposes no ‘selectional restrictions’ and/or ‘thematic relation’ on this constituent …, one of the … three unusual properties (that a Raising verb manifests, RR) with respect to a constituent in one of its apparent (or, surface) argument position …” (Jacobson 1990: 428). If a Raising analysis tries to capture and interprete the relevant structures syntactically, it does not seem to provide a satisfactory elucidation for them. Syntactically, we have no evidence for structures corresponding to the stage of derivation before Raising. Compare (231), (232) with (233), (234). (231)

It seems that he is unfair

(233)

seems  to be unfair is likely    John promises to please Susan

(234)

*It promises that John pleases Susan

(232)

He

Some version of the Well-formedness constraint is not satisfied (cf. Partee 1979: 275 ff.). Substantial semantic support and justification of the Raising analysis might still be provided by considering Raising-type promise and threaten to be combinations of an epistemic and evaluational verb. A kind of propositional attitude (similar to seem) towards the event described in the clause embedded at D-structure, on the one hand, and its positive (promise) or negative (threaten) evaluation, on the other, have unified. Metaphorization consists in singling out a participant in the event or state under consideration and in attributing to it the ‘promise’ or ‘threat’ of its occurrence. Or the event as a whole, that is its full description, may figure as the entity promising, or

108

Chapter 5

threatening with, its realization in the absence of a subject [+human] as in (230) or (235). (235)R

[Bumaznoe okno]i

vzdulos’

(the) paper window swelled ugrozaja PROi lopnut’ threatening to burst

puzyrem, PROi [instrum case] like a bubble

Ugrozaja is a participle adjunct with PROi being controlled by the matrix subject. Importantly for a Raising analysis, in Russian, in contrast to Czech, Raising — apart from phrasal verbs — can never be combined with, or result in, an infinitival VP, which means that the result of Raising is exclusively a “small clause”. The Raising analysis, thus, is generally questionable at least in ° Russian (for detailed discussion, see Chvany (1975: 51, 226 ff.) and Ruzicka 1980: 30, 37 ff.). Let us try to approach a correct analysis by considering (236) – (242). (236)G

Es droht nicht (ei zu regnen) it does not threaten to rain

ei must be either PRO (Control) or ti (Raising). A third possibility is to assume a subjectless verb projection, that is, a bare infinitival verb phrase (Bierwisch 1992). The control interpretation would take es to be an argument, a quasiargument (Chomsky 1981: 335), which is not raised. (237)G

*Es drohte zu scheinen, daß … it threatend to seem that …

(238)R

*pro ugrozalo kazat’sja, cto … threatend to seem that … (237) suggests that the control interpretation should be preferred as the correct one for (236), since drohen and its counterpart, e.g., in Russian, do not seem capable of receiving a Raised expletive subject from scheinen (seem) as by “Superraising” in (239). The Raising verb seem, differently from drohen in (237), “accepts” the expletive subject of the Raising adjective likely. (239)

It seems (t to be very likely (that she’ll ring me tonight))

A clear instance of Raising (the quasi-argument) is of course (240), which, however, need not be syntactically equivalent to (236).

The case of promise (240)

109

It isn’t likely (t to rain).

Now consider (241). (241)G

Es it

scheint seems

zu drohen to threaten

Es is an argument DP raised from the subject θ-position discharged by drohen. (241) does not violate the θ-criterion. Es is the head of the argument chain (es,t): es scheint t zu drohen. I conclude nevertheless that a Raising analysis of threaten or drohen is feasible. Further empirical support for this assumption may be provided by (242), in so far as the threat of rain originates from the black clouds. (242)

The black clouds threatened rain.

5.8. Intervention of modal verbs Control processes are subtle in (243), (244), as shown by examples from Fauconnier (1976: 156) discussed also by Vanden Wyngaerd (1987: 83). (243)F

Mon père m’a menacé d’ aller My father me has threatened to go à l’ école du dimanche to Sunday school ‘My father threatened to [make me] go to Sunday school’

(244)F

On a menacé ma soeur d’accoucher sans docteur They threatened my sister to give birth without doctor ‘They threatened my sister with [having to] give birth without a doctor.’

The German counterparts of (243) and (244) are unacceptable. (245)G

*Man hat meiner Schwester gedroht, ohne Arzt [dat] One has my sister threatened without doctor zu entbinden to give birth

110

Chapter 5 (246)G

*Mein Vater hat miri gedroht (PROi [dat] My father has me threatened in die Sonntagsschule zu gehen) to the Sunday school to go

Such cross-linguistic situations should have become familiar by now. (243) vs. (246) is an instance of a further type of distinct cross-linguistic control behaviour that is contingent on the availability of particular modal and causative elements. Recall the differences described with respect to promise and ask, as illustrated by (219) vs. (221), (91) vs. (96), and the contrast arising in the persuade-type control regulations (see above, (115) vs. (123)). In the former type (promise, ask), the construal necessary to meet m constraints (B,a) and (B,b), respectively, is not required in German, as the modal verb dürfen that must be introduced to obtain grammatical renderings matching the meanings of (219) and (91) produces sentences directly meeting m constraints (B,a), (B,b). In the second type, (persuade), the same, that is, satisfying um (B,b) is done by the causative verb lassen. Third, the German counterparts of French (243), (244) resort to the modal müssen (must, have to, be forced to). In (243) and (244), deontic and causality switch seem to combine. The counterpart to (243) in German is (247). (247)G

Mein Vater drohte

miri (PROi die Sonntagsschule [dat] my father threatened (me) the Sunday school besuchen zu müssen) to go to to have to ‘I was threatened by my father with having to go to the Sunday school.’

What must be made explicit by using müssen, conceptualized as have to or be forced to, in rendering French (243), effects satisfaction of m constraint (B,a) quite analogously to the introduction of dürfen in (219), (91), the latter satisfying m (B,a), (B,b), respectively. The use of lassen, rendering English (115), (see 4.1.), results in meeting um constraint (B,b). Thus, what in German is necessarily elucidated by lexical material and results in more “regular” control behaviour, in English and French may be left to conceptual inferences induced by the respective control verbs and the semantic content of the clausal

The case of promise

111

complement, provided that the results of the pragmatic construal in turn can be referred to by the respective constraints which license them. Note that in either case, the quasi theta-specification of Benefactive supports identification in accord with the m constraint, and that on the “negative” interpretation: What is threatened is not to the benefit but to the disadvantage of the addressee (cf. 5.6.)25 (248) and (249) illustrate typical standard cases in Russian. (248)R

Vitalijaj proi ottuda iskljucili za [gen-acc] [acc sg] Vitali (they) from there excluded for nedisciplinirovannost’ i proi prigrozili (PROi [3 ps pl pret] undisciplined behavior and threatened otpravit’ proj s marševoj rotoj na front) [acc] to send (him) with a company on the march to the front

proj (j ≠ i) is a null pronoun Case-marked accusative, which is identified with the matrix object Vitali. proi is a discourse-related null subject pro, marked nominative and [+plural, 3ps] in Russian. It refers to a domain of persons no individual of whom is identified. (249)R

Oteci grozilsja (PROi menjaj vygnat’ Father threatened me to turn out lišit’ proj nasledstva [gen sg] deprive (me) of (my) inheritance

i and

The slightly non-standard form grozit’sja with the attached reflexive affix closes the position for an object-NP.

Chapter 6

No Choice of Controller

If only one NP-argument is available one has no choice but to take or reject it as the controller. The question whether a single argument is a suitable controller, has to be answered in any theory of control.

6.1. Matrix active clauses 6.1.1. The case of try / attempt Dowty proposed an entailment pattern for try that is analogous to (10) (see fn. 15). “… (250)

∀x ∀P [δ(P) (x*) ⇒ α(x*)] ∀x ∀P [δ(P) (x*) ⇒ β ([P(x*)] (x*)]

In the case of δ = try’, for example, an instance of the entailment α (x*) would be ‘x acts intentionally’, and an instance of ß ([P(x*)]) (x*) would be ‘x intends that x’s action will bring about P (x*)’” (Dowty 1985: 299). (“δ is an intransitive Equi verb, P a variable over VP-meanings …, x an individual variable, α a property of NP-meanings, and ß a relation between NP-meanings and propositions.”) Though stated in terms of a M(ontague)-analysis), Dowty’s entailments can be related to control conditions in the style of this approach. I will argue that verbs of the try-type (e. g. manage, dare; R umudrit’sja (manage, contrive), osmelit’sja (dare) assign the theta-specification 〈intact,+〉 to their respective external arguments and to the s-structure subject position (PRO) of their infinitival complement. Thus, they “generate” their own constraints on control26 .

114

Chapter 6

Assignment of one and the same theta-specification to two arguments, one of them associated with the distinct theta-grid of the complement verb, is empirically justified by the fact that ‘trying to do something’ and ‘doing something’ are inseparable, with respect to intentional action. “Thus if a man tries to hit a home run and succeeds, his try is his success …” (Davidson 1989: 72)) At LF, then, the subject position of the infinitival occupied by PRO, or e that is replaced by PRO (in the passive or with ergative verbs, for example), carries its internal theta-specification and the feature value relevant to control, which it receives from the matrix verb try, manage, and others of this type. The twofold assignment implies an inherent identification, that is, control matching unmarked (B,a). If it is intuitively arguable that a control verb like try must assign the lexical control feature value (〈intact,+〉) to its own matrix argument and to PRO, the formal and conceptual aspects of such an operation may appear dubious. It is tempting to wonder whether the concept of abstract verb incorporation developed by Baker (1985, 1989) and claimed (Grewendorf/Sabel 1994) to be independently viable as a means to derive the pattern of long Scrambling in German, might provide a natural solution to the control problems and double argument. Grewendorf and Sabel (1994: 264 ff.) have pointed out cases of long scrambling in control complements like (251) vs. (252), (Grewendorf/Sabel’s (2)a. and (2)b.) (251)

[C’ daß [IP[den Hund]i [IP keiner [CP PRO ti zu füttern] that the dogACC nobodyNOM to feed versuchte]]] tried

“The verb versuchen ‘try’ in (2a) but not the subject control verb zögern ‘hesitate’ in (2b) licenses long scrambling of the embedded object into the matrix clause … lexical properties of verbs play the decisive role in licensing long scrambling …” (Grewendorf/Sabel 1994: 265) (252)

*[C’

daß [IP den Hund]i [IP keiner [PRO ti zu füttern] that the dogACC nobodyNOM to feed zögerte]]] hesitated

Which is the lexical property that plays the decisive role in licensing long scrambling in (251) but not in (252), that is, which is the lexical property that

No Choice of Controller

115

is compatible with, or requires, abstract verb incorporation and constitutes the relevant distinction between versuchen and zögern?27 Zögern says that the enterprise has not yet been undertaken and may not be at all. The chance of success or failure is not (yet) taken. The difference can be assumed to be very near the characterization of try given by Davidson (see above), which applies to try and its ilk but not to hesitate. The specific lexical properties, however, on which control behaviour but not long scrambling hinges unify try and hesitate. The relevant common property is partly brought out in the pair manage versus fail, which clearly shares it. Manage may be loosely paraphrased as try something and accomplish it, fail as try something and not accomplish it.

managed  (253) He   to reach the semi - finals failed  What is common to manage and try, to fail and hesitate? Let us tentatively call the property metaactive. It will be represented as the lexical feature (value) 〈metaactive,α〉. Associated with a lexical class of control verbs, it will trigger the cross-clausal double assignment of 〈intact,+〉. Its intuitive understanding can be paraphrased as the relation of an agent designate to bringing about, or causing to happen, the event described by the embedded verb. For illustration, I will give a, of course, incomplete list of the verbs in this control class in English, Russian, Czech, German. (a) try, attempt, seek, endeavour, aim; (b) dare, venture; (c) manage, fail; (d) hesitate, hurry; (e) be able to; (f) arrange, plan; (g) be determined, be prepared to; (h) agree, refuse Russian: (a) (po)pytat’sja, probovat’, starat’sja, stremit’sja; (b) osmelit’sja, otvazit’sja; (c) udat’sja, (+dative-DP), ne uda(va)t’sja (fail); (d) lenit’sja (be too lazy to), spešit’; (e) umet’; (f) rešit’; (g) zatrudnjat’sja/zatrudnit’sja (find it difficult to); (h) soglašat’sja; (i) byt’ gotov (be ready) Czech: (a) pokusit se, hledet (seek), snazit se (endeavour); (b) odvázit se (dare); (c) dokázat, nedokázat (manage/fail); (d) váhat (hesitate), pospíšit si (hurry); (e) umet (know (how to VP)), být s to; (f) chystat se. German: (a) versuchen, sich bemühen (endeavour); (b) wagen, sich erkühnen; (c) (es ) (nicht) fertig bringen, (es) (nicht) schaffen (manage/(fail)); (d) zögern, sich beeilen; (e) fähig sein; (f) planen, vorhaben (g) entschlossen sein, bereit sein. This approach, which presupposes a propositional view of infinitives, helps cope with a spurious puzzle that has arisen with this view and has been used as

116

Chapter 6

an argument against it by Chierchia and Turner (1988: 292): “… if (8) [= try’ (m, read’ (m, Principia’))] is the meaning of (6d) [= Mary tries to read/reading Principia] and (7a) [= ∀x (try’ (m,x) ⇒ try’ (j,x))] the meaning of (6a) [= John tries everything that Mary tries], then what (6a) and (6d) should entail is something like (9) a. b.: (9)

a. try’ (j, read’ (m, Principia’)) b. John tries to bring about a situation where Mary’s reading Principia occurs.

But this is just not true. In conclusion, it would seem, prima facie, that on any variety of propositional analysis of infinitives and gerunds the validity of reasoning patterns such as those in (6) will be a total mystery.” The instance of sloppy identity can be cleared up by revising (7a). In (7a), x, by hypothesis, in particular in accord with (8), is a propositional variable. Its value must be constant, of course, in both occurrences. It is well known that “sloppy identity” phenomena involve variable binding (see, e.g. Reinhart 1983). The global propositional variable x in (7a), however, blurs the fact that the hidden subject argument contained in either occurrence of x is itself a bound variable: Within the first occurrence of x in (7a), it is bound by m, within the second by j, a legitimate variant of control. Under our approach, the “external”, cross-clausal assignment of 〈intact,+〉 to PRO — in addition to its “internal” assignment to the matrix subject — effected by try — is equivalent to or entails, the intrinsic referential identification of the two arguments, covered by the occurrences of x in the antecedent and consequent in (7a), with their respective overt arguments m and j. This would exclude (9a). The respective necessarily distinct values of the NP-argument variable hidden in the first and in the second occurrence of x clear up the puzzle of the two apparently identical x’s in (7a). The logical formula (7a) distorts the grammatical situation. I would rather propose (7a)′. (7a)′

∀P (try’ (m,P(m)) ⇒ try’ (j,P(j)))

Thus (9) a. b. is not entailed by (6a) and (6d). Zec (1987: 139-145) refutes Chierchia’s entailment (9) a. b. on the basis of control in finite clausal complements in Serbo-Croatian. A simpler illustration is (254). (254)

John tried to understand Kant and Peter tried it, too.

No Choice of Controller

117

Such instances of sloppy identity are not restricted to try. Analogous cases reappear in controlled complements of promise. (255)

Peter promised to take part and so did John.

Thus, cross-clausal assignment of the lexical feature 〈intact,+〉 is a sufficient, but not a necessary condition of “sloppy” or — in Higginbotham’s (1992: 90) terminology — covariant interpretation under it- or that-anaphora as the only available interpretation. Saying that “just because it is a verb of strong obligatory control, try is a verb for which, by virtue of its meaning, only covariant interpretation is possible …” (Higginbotham 1992: 106) could count as a terminological proposal. If the clause internal and the external assignment, the latter effected by try, coincide, that is, PRO receives 〈intact,+〉 two times, no problem is posed. This is the uninteresting standard case of control. The extension of assigning 〈intact,+〉 to the embedded subject position proves to be of import when it comes to less customary instances of control: “English speakers accept without hesitation sentences like (i) Karen tried to be elected … in the interpretation dictated by syntax … literal translations of (ii) Quintus versuchte bestraft zu werden into English Quintus tried to be punished are fully acceptable, even if the situations described especially in the case of (ii), are rather bizarre.” (Comrie 1985: 54, 63) Superficially, the presumed cross-clausal assignment, which presupposes the matrix intra-clausal one, creates a situation that matches um constraint (B,a). PRO in (ii) receives 〈intact,+〉 from the matrix verb as stated above. But in contrast to control of promise-type verbs, control of try need not be licensed by matching um (B,a) or, for that matter, some other constraint. This is what I have in mind talking of self-generating control. The case of promise is fundamentally different. PRO can only be specified by the theta-grid of the complement verb. Otherwise, the control mechanism of promise, determining cases like (211) – (213), would be evened out. In particular, the difference between (256) and (257) would get lost. (256)

Karen tried to be elected.

(257)

Karen promised him to be elected.

6.1.2. try vs. intend Consider (258), which is acceptable, and (259), which is not.

118

Chapter 6 (258)

I tried to be allowed to go.

(259)

*I intendend to be allowed to go

Jackendoff (1972: 233) takes (259) to be ungrammatical, Vanden Wyngaerd’s judgment is “unacceptable” but “semantically grammatical” (1987: 64) How does this theory reflect or deal with the situation, in particular, why is (259) not grammatical or unacceptable, but (258) is? With respect to control, try differs from intend, although the meaning of the former can be understood to imply that of the latter. As demonstrated above, (the process of) trying to do something is inseparable from doing something, but intend to do something clearly is separate from doing it. Thus, trying involves acting, but intending does not. I assume that intend, in contrast to try, does not assign 〈intact,+〉 to its matrix subject and complement subject position, which is a natural consequence of the separation of intend to do something from doing something. This accounts for the ungrammaticality of (259): PRO receives 〈intact,–〉 from allow as in (258), but not (across clause) 〈intact,+〉 from intend. Differing from try, the matrix subject of intend alone is assigned 〈intact,+〉. Thus, um constraint (B,a) is not matched “automatically”. (260) is grammatical, of course, satisfying um (B,a): the subject of intend receives 〈intact,+〉 and PRO is assigned 〈intact,+〉 from the theta-grid of go. (260)

He intended to go.

Recall that we distinguish between the stronger part of the control relation and its weaker part: 〈 〉PRO is the stronger part. (cf. above, 3.2.4.1.). Now, the particular constraint (J) on the application of (B,a) will capture the contrast between (261) and (262) left over for explanation. (J) Control relations in the designated class are not licensed iff the lexical meaning of the verb in the controlled clause is equivalent to a proper part of the thematic specification assigned by the control verb. The inverse does not hold. Alternatively: A control verb of the designated class (e.g. try) cannot control the PRO-subject of its proper part verb. (261)

He intends (PRO to try to take part in the competition)

(262)

*He tries (PRO to intend to take part in the competition).

In (261), PRO is characterized by 〈intact,+〉, he receives 〈intact,+〉 as well by intend, whose meaning is represented roughly as λpλx (x INTENDS p)

No Choice of Controller

119

considered a proper part of 〈intact,+〉. It is licensed by um (B,a), since the stronger conjunct 〈intact,+〉 satisfied by PRO contains intend predicated of he. In (262) the relation is reversed: the stronger conjunct of constraint (B,a) is determined by INTEND, which is included in the instantiation 〈intact,+〉 of the weaker one. In accordance with (J), this results in ungrammaticality. I will summarize the preceding discussion by formulating a condition on lexical information and control for the items try and cognate verbs like manage, r. umudrit’sja, c. být s to (be capable of) and verbs in the list above. (K) Specification 〈intact,+〉 of the theta grid of try and cognate verbs, beyond its projection onto an NP-argument or (PP-adjunct) in its own clause, is assigned to the S- (or LF-) structure PRO-subject of the clausal complement (in addition to the theta-specifications internal to, and distributed among, the arguments of the clausal complement). Recall that intend, whose meaning is considered equivalent to a proper part of 〈intact,+〉 that lacks -act(ion), does not obey (K). Still it assigns 〈intact,+〉 to its external argument, where the action intuitively relates to mental decision. Thus, (261) and (262) are captured straightforwardly. (K) holds and applies in active matrix clauses and in the (impersonal) passive ones, in which 〈intact,+〉 is assigned to crosslinguistic counterparts of the by NP-phrase (see sect. 6.2.)). 6.1.3. Effects on passive controlled clauses and unspecified PRO In passive controlled clauses, the derived PRO-subject carries a specification 〈F, −〉 that clashes with the specification that PRO is assigned by (K) from “outside”. From the viewpoint of grammatical control conditions, the internal theta-specification that PRO bears is not irrelevant to the external one it receives from the theta-grid of try by (K). Thus far grammar, in particular syntax, can go and has to go with respect to the passive controlled clauses on the assumption that the passive verb assigns 〈intact,+〉. Availability of 〈intact,+〉 satisfies condition (D), which, thus, also is a specification on the application of (K). Intuitively, if the content of the controlled complement clause is an intentional action, it provides a natural target for trying to affect it. Consider (263).

120

Chapter 6 (263)R

Oni pytalsja PROi byt’ osmotrennym [part/pret/pass/sg/masc/instr] he tried to be examined

specialistom [instr sg] by a specialist

Let’s see what it would be like to take an extreme position or accept a “null hypothesis”: “… in terms of control, anything goes with a verb like try … we suggest … to treat like on a par with try, which means that no restriction on control holds.” (Vanden Wyngaerd 1987: 75/76)) If try is treated on a par with like, the impact of passivization on control should be the same with try and like or hate. to praise  (264) John likes    to be praised 

(265) John hates   (266) John tries   (267) John tries  

to cheat at cards  to be cheated at cards  to be cheated at cards   to cheat at cards  to praise everybody  ?to be praised by everybody 

Clearly, (264), (265), with active and passive controlled clauses, are perfectly acceptable, whereas, to say the least, we have to do a certain amount of work to impose an interpretation on the passive variants of (266), (267), which has an effect on their acceptability. The conflict between 〈intact,–〉 and 〈intact,+〉 coexisting in PRO reflects this situation. Its settlement depends on pragmatic settings of discourse, situation and knowledge. There appears some apparently paradoxical evidence in favour of this approach on inverting the strategy: Passive controlled clauses may prove to be more readily acceptable than corresponding active ones. Compare (268) and (269).

(268)R

staralsja    (PRO zapodozrit’ ego  pytalsja  stremilsja  tried  Nobody (not)  to suspect him  endeavoured   Nikto

ne

v eresi) of heresy28

No Choice of Controller

(269)R

On

staralsja     pytalsja  (PROi stremilsja 

he

tried  endeavoured   

121

ne být’ zapodozrennym t i [part pret pass masc sing] not to be suspected

v eresi) of heresy (269) is less bizarre than (268). In (268), zapodozrit’ does not seem to assign 〈intact,+〉 to PRO. If the theta role experiencer is assigned to PRO it will clash with 〈intact,+〉 projected from the matrix verb. The minor or absent oddness of (269) can be attributed to the natural behaviour of trying to avoid suspicion. In other words, the specification 〈intact,–〉 assigned to the chain (PROi, ti) from the theta-grid of zapodozrit’ in (269) is easier to reconcile with 〈intact,+〉 than the internal specification of PRO is in (268). (cf footnote 26). 6.1.4. Non-passive controlled clauses failing condition (D) When 〈intact,+〉 is not available in the controlled clause, the clash between the feature value 〈intact,+〉 assigned to PRO by the matrix verb try or its ilk and the internal theta role it carries, cannot be handled in grammatical terms alone. The coexistence, compatibility and conceptual-pragmatic balance of the two characterizations modulate acceptability, more or less ready acceptance or rejection of the pertinent sentences. The problem is again whether the event or state described in the controlled clause can be construed as conceptually compatible with, or accessible to, trying it. Lasnik and Fiengo (1974: 553) mark (270) as ungrammatical. (270)

Billi tried (PROi to resemble Harry)

(271)R

Chotja jai i pytajus’ (PROi byt’ pochozim although I try to be like na Christa …) Christ … ‘… to resemble Christ’

122

Chapter 6

For Vanden Wyngaerd (1987: 75), (270) is “… good … if read with an appropriate context.” Judgments like “ungrammatical” can miss the point. By (K), PRO in (270), (271) is assigned 〈intact,+〉 cross-clausally, and, thus, they and their counterparts in other languages, are grammatical, as far as grammar goes. The conflict between 〈intact,+〉 and the theta-specification assigned to PRO by (the theta-grid of) resemble is settled at a postgrammatical conceptual-pragmatic level. Control in cases like (270) can be said “… to have obligatory pragmatic and semantic attachments.” (Fillmore et al. 1988: 504) Their ground is prepared by the conflicting specifications in PRO. Analogous situations recur in (272) – (273). (272)R

Petri pytalsja (PROi ispytat’ Peter tried

cuvstvo oblegcenija) [acc] [gen sg] to experience the feeling of relief

(273)C

(proi)

Snazilii se (PROi být veseli) [3ps pl] [lexical refl. clitic] (they) endeavoured to be cheerful ‘They were at pains …’

(274)C

Ve dne

proi se snazilai (PROi verit [refl. clitic] in the day-time (she) endeavoured to believe tomu, co Tomáš ríká …) [dat] that what Tomas was saying

In (272) – (274), there is again a conflict between 〈intact,+〉 and Experiencer both present in PRO. This is of course independent of the D-structure position of PRO. If PRO is not the external argument, the pair of internal and external specifications is assigned to the chain (PROi ti), e.g. in the case of monadic ergative verbs like arrive, escape, or verbs like Russian nravit’sja (to please). (275)

Hei tried (PROi to arrive ti (at ten)).

(276)R

Onai umudrilas’ (PROi nravit’sja) she managed to please ‘… to be attractive …’or ‘she managed to make people like her’29

No Choice of Controller

123

On this approach, it must be expected that “… adverbial modification often suffices to create a complement of the appropriate type as seen in the contrast between the two cases in (277) and (278). (277)

I tried to receive the good news *(with good grace)

(278)

I tried to suffer the insults *(good humoredly) (Lasnik’s (i), (ii))

(Lasnik 1992: 248, fn. 6). An anonymous reviewer of Lasnik’s paper observed “… that the ‘controllability’ phenomenon does not always reduce to theta role assignment in any straightforward way.” Clearly, under this approach, the conceptual conflict between the two assignments PRO receives from matrix try 〈intact,+〉 and receive, suffer, respectively, is considerably relieved with the adverbial implying intentional endeavour. Thus, (277), (278) are to be judged as grammatical with the adverbial, since the latter’s occurence implies an at least conceptual modification or extension of the theta-grid of suffer and receive that provides for 〈intact,+〉 being assigned to PRO. The adverbials do what condition (D) requires. PRO must end up with the local assignment 〈intact,+〉, independently of whether the respective adverbials effect it alone or in tandem with the verbs. The verbs alone cannot assign 〈intact,+〉. What they assign to PRO clashes with the external assignment by try so strongly that (277), (278), without the adverbial, are not interpretable, which amounts to ungrammatical at an interface level.

6.2. Matrix passive clauses The respective constraints on control including their concomitant role of guiding potential pragmatic construal of control extend to the passive forms of control verbs and, thus, “Visser’s Generalization”. To start with, I assume that (279) and (280) are ungrammatical on different grounds. (279)

*John was tried [e to go]

(280)

*it was

tried  attempted   

[e to see Bill]

124

Chapter 6

Koster (1984: 431, 433) argues that both are ruled out by the common labor of binding theory (A) and Control theory. Accounting for (279) and (280) distinctly, I assume that (279) is a violation of the theta-criterion, following Vanden Wyngaerd (1987: 76): “… because try does not subcategorize for an NP object, i. e. because the argument John receives no theta-role.” In other words, a wrong lexical property (of try) is projected. Similarly, Iwakura (1985: 31) argues against Koster’s analysis. As regards (280), its ungrammaticality revolves around two parameters. The first can be roughly characterized as lexical sensitivity to the matrix impersonal passive. I will sketch the problem first. “It is not possible at the moment to give necessary and sufficient conditions for such structures. As Williams (1980) points out, not all verbs with for-complements permit this construction. But if these constructions are possible at all, the complement is usually a for-complement.” (Koster 1984: 434)). Williams (1980: 215) illustrates the unclear situation: “Decide is another verb with an antecedentless passive: (53) It was decided to have dinner at 6. … But decide does not allow lexical NP: (54) *John decided (for) Bill to get the prize.” (Williams’ enumeration). (281) is an example in which the passive is not antecedentless. (281)

It was decided by John to behave himself.

(281) is judged grammatical by Manzini (1983: 427) and Vanden Wyngaerd (1987: 121), but a very similar example (282) is taken to be not acceptable by Chomsky (1980: 35). (282)

*It is preferred to arrive a few minutes late.

Possibly, the ergative verb is responsible for the judgement. (283), however, is considered grammatical by Williams (1980: 215) and Iwakura (1985: 47). (283)

It was preferred to have dinner at six.

German (284) is perfectly acceptable. (284)G

Es wurde vorgezogen, schon it was preferred already

7 Uhr zu frühstücken at 7 to have breakfast

The range of well-formed impersonal passive control structures of type (281), (283) does not seem to correlate with the occurrence of a complementizer (for). The complementizer effects are epiphenomenal. Cross-linguistically at

No Choice of Controller

125

least, distinct fixations of a pertinent parameter seem to interact with, or be dependent on, other basically lexical properties. Some parallelism between French and Slavic languages and German suggests itself. “… impersonal passives in French … are possible only if the verb takes an internal PP argument: (iii) Il a été parlé de vos frères hier soir.” (Kayne 1984) Another example is (i): Il sera procédé à la révision de ce procès. (It will be proceeded to the revision of this trial. ‘One will proceed to the revision of these proceedings.’) (Examples from Postal/Pullum (1988: 639)30 . This varying situation has effects on control when the passive verb is a control verb. (285)F

secourir  essayé de   visiter   help   it has been tried to   to visit  

* Il a

été

Paul. Paul.

The German counterparts are fully grammatical.

(286)G

besuchen Es ist versucht worden, Paul zu   helfen 

The implicit agent NP receives 〈intact,+〉, it is a controller licensed by (B,a). Try assigns the feature value to PRO independently of its voice of course. Russian offers no syntactic parallel. (287)R

*Bylo (it) was

pomogat’ Pavlu  poprobovano    posetit’ Pavla  [part pret pass neuter] tried to help / visit Paul

Thus, control could be well-formed, as in the active, wherever the structural type is grammatical, which depends on a parametrized variation whose factors are not completely lucid. (For details, cf. R˚uzicka 1986a, 1988, 1992). The parameter of impersonal passive constructions, if set positively, is still sensitive to lexical properties (and projected syntactical effects, e. g. unergative vs. ergative). The second parameter that can be set up with respect to impersonal passive constructions as well as to some associated control regularities concerns conditions of well-formed occurrence of (the respective counterparts) of by-phrases. Following Jaeggli (1986: 600) and Roeper (1987: 276 ff.), I assume that “… an NP in a by-phrase is assigned the ‘pure’ external theta-role

126

Chapter 6

of the passive predicate.” (Jaeggli) This NP, overt or implicit, receives 〈intact, +〉 from the theta-grid of pertinent predicates like try in the same way as the external argument in subject position at D-structure in the active. The hypothesized (cf. Dowty 1991: 605 ff; Burzio 1986); R˚uzicka 1992) parameter consists in observing this parallelism or not. Clearly, the second parameter is dependent on the first. But, as we will see immediately, they cannot be united. Importantly, the parameters are effective in particular lexical classes, e.g. for try, versuchen; dance, tanzen. English sets the first parameter and, consequently, the second negatively, differing, e.g. from German. If the parameter is fixed positively, (K) applies to the passive as to active matrix clauses, 〈intact,+〉 being assigned to the NP of the counterparts of by-NP and to PRO. (288)G

Es ist [PP von [vielen]i] it is (has) by many ([dieses Problem]j (PROi tj this problem

versucht worden tried been zu lösen)) to solve

This treatment extends to functional verb passive and verbal noun. (289)R

Byla sdelana popytka [NPe]i (PROi was made (an) attempt (by somebody) ugovorit’ ego) to persuade him

Under Safir’s (1987: 562, 584) approach, the implicit (empty) instrumental-NP is not mapped onto a syntactic position. The notion of “lexically satisfying a theta-role” (Rizzi 1986) might be invoked here. I leave this problem unresolved. The status of the second parameter is slightly complicated by crosslinguistic variations in the (im)possibility of overt appearance of (the counterpart to) a by-phrase with the NP carrying 〈intact,+〉. This is evidenced by modern Czech. Judgments on the grammaticality of the occurrence of the agent instrumental-NP vary. (290)C

… aby

bylo *úcastniky provozu) [pret neuter] [instrum pl] in order that (it) was/be by the participants of the traffic dbáno predpisu° [part pret pass neuter] [genit pl] observed (the) regulations

No Choice of Controller

127

We should hesitate to allow the assignment of a feature value, that is, 〈intact,+〉 to an implicit NP that can never appear overtly.

6.3. Raising contexts Instances of control that involve Raising, like (291) – (292), provide evidence and further support for this approach. (291)R

Borisi pytalsja (PROi pokazat’sja (ti Boris tried to seem [sportsmenom])INSTR (sportsman)

(292)R

Jai I i and

(293)

John tries to seem to be nice (example from P. Jacobson 1990)

molodym (to be) (a) young

riskuju (PROi pokazat’sja vam (ti suevernym risk to seem (to) youDAT superstitious nedalekim)INSTR limited

R. riskovat’ (riskuju) is among the verbal class which is subject to (K).31 Following Levin/Rappaport (1986: 646), I assume that “The XP assigns a theta-role to the NP of which it is predicated, so this NP may be called the external argument.” XP in (291), (292) is AP suevernym, NP molodym sportsmenom). The NP affected by Raising ends up as PRO in a non-theta-position, but receives a theta-characterization through the chain (PROi ti). PRO now is assigned 〈intact,+〉 by cross-clausal projection rule (K), e.g. in the derivation of (291), (292). The two theta-specifications, that is, 〈intact,+〉 assigned across clause boundaries and the internal specification of the chain (PRO, t) projected from the theta-grid of the predicative A(P) in (292) – (293), have to come to terms with each other. The arising conflict is rather one-sided with the theta-specification assigned by the AP- or NP-predicate to its external argument, the coda of the chain (PROi, ti), being quite a feeble characterization. Thus, 〈intact,+〉, which PRO has received by virtue of (K), prevails over it to its suppression, and the interpretation of, e.g., (291), (293), clearly, takes the attempt (pytalsja, tries) to be directed towards giving the impression

128

Chapter 6

(pokazat’sja, to seem, to appear to) that Boris is a young sportsman, and John is nice. This, indeed, is (part of) the well-formed meaning of (291) – (293). The cross-clausal assignment of 〈intact,+〉 to PRO in virtue of (K) bears the empirical burden of accounting for this meaning, capturing the special case of two theta-specifications competing with each other.

Chapter 7

Reflexive clitic impersonal clauses in Control structures

7.1. The situation in Slavic languages In Slavic languages, impersonal structures constituted by their respective clities se, sa, siøe, -sja, which are near equivalents to Italian si (cf. R˚uzicka 1992), cannot figure as infinitival clauses. They are exclusively finite. Thus, the problem discussed by Belletti (1982: 14 ff.) does not arise, namely, the “… incompatibility of control PRO with si … impersonal si and control PRO are mutually exclusive.” Violation of the Case Filter is answerable for it (Cinque 1988: 23). In the light of Cinque’s work, this incompatibility could still be accounted for by the obligatory argumental status which he assumes for si in nonfinite clauses. (Cinque 1988: 531) If si is an argument, the NP-subject is pleonastic pro (l.c. 534), which cannot identify with “chain breaking” argumental PRO. On the other hand, we do not expect a matrix pleonastic pro (subject) to be a controller. This is trivial, but may serve as a criterion which one could use when judging the argumental status of a pro subject. If, in the corresponding exclusively finite reflexive impersonal structures of Slavic languages, it is not the reflexive clitic but pro that has argumental status, as I have argued in R˚uzicka (1992), pro should qualify as a controller licensed by the relevant theta-specification. The following Czech examples substantiate this prediction. (294)C

[pro]i Zamyšlelo sei (PROi zmenit [3 ps sg pret neuter] refl intended reflexive to change organizacní strukturu podniku [gen sg] the organisational structure of the factory ‘One planned to change the organisational structure of the factory’.

130

Chapter 7 (295)C

[pro]i Rozhodlo sei (PROi neorganizovat [3 ps sg pret neuter] refl decided not to organise hromadné oslavy) mass celebrations ‘One had decided not to organize mass cebebrations.’

(296)C

[proi] Zkusilo sei (PROi to zvednout) [3 ps pret sg neuter] refl tried that to lift ‘One (has) tried to lift it.’

The Czech verb rozhodnout (of (295)) has a variant with the lexical clitic se: rozhodnout se, pro, the controller, is a discourse-related anaphor here. (297)C

[pro]i Rozhodl se (PROi být zamilován do Terezy) [pret masc sg] [part pret pass masc] (he) decided to be in love with Tereza

In (297), the pro-subject controlling the external argument PRO of the adjectival participle (phrase) carries [masc], [sg] by Spec-head agreement. PRO receives 〈intact,+〉 across clause boundaries by (K). In (294) – (296), the Czech reflexive clitic corresponds to Italian nominative si, which “… corresponds to man ‘one’ constructions in German”. (Dobrovie-Sorin 1998: 410) Dobrovie-Sorin (1998: 426, 427) points out a clear contrast between Spanish and Italian on the one hand and Romanian on the other hand. In Spanish and Italian, constructions equivalent to Czech (294) – (296) are available “… because se is a nominative clitic that is able to control the embedded subject. Since Romanian does not have nominative se, there is no available controller.” Besides c. zkoušet/zkusit (try), rozhodnout (se) (decide) is taken to be subject to (K), zamyšlet is treated like intend.

7.2. Control and reflexive passive The status of pro in impersonal reflexive clauses as a prerequisite for controlling In sentences like (294) – (296) the crucial issue is how the controller is “concealed” and how it is identified in the visible matrix clause structure. I

Reflexive clitic impersonal clauses in Control structures

131

have assumed so far that PRO is controlled by the pro-subject and its accomplice se whose interpretation is straightforward: It has either an “anonymous”, that is, indefinite specific, or a generic reading. The latter is illustrated in the Slovak example (298). (298)SL Najviac sa dnes [reflexive] mostly today srdcové chorobyACC heart diseases

(pro) umiera na [3 ps sg] (one) dies of

The assumption that in (294), (295) pro is the controller implies that it is an argument. How does it relate to the reflexive clitic se or sa, siøe (Polish)? The clitic is a lexical nominal element and I follow Cinque (1988: 536) in assuming that it “… is part of a CHAIN with [NP,IP] …” The concept of CHAIN is theoretically suitable to account for what the clitic does. Certain properties of the zero subject-DP are a function of the reflexive clitic. In the examples (296), (298), se, sa “conspires” with pro to furnish the (subject) position occupied by the latter with a designated content, that is, to reduce it to “anonymous” or generic meaning. This is nearly equivalent to saying that the reflexive clitic enables pro (and the rest of the clause) to satisfy F(ull) I(nterpretation) (see Chomsky 1986: 98 ff.): In an otherwise unchanged context without the reflexive clitic, one would either have to look for some quite different non-autonomous, e.g., discourse-anaphorical interpretation of the zero subject, or consider the sentence defective. Compare (299) with (300) (299)C

(pro) Zadalo se, ze S [neuter sing] refl ‘one demanded that S’

(300)C

(pro) Zadalo ze S it     something  demanded that somebody 

In (300) bare pro must provide its own discourse-related or deictic, possibly logophoric antecedent. It follows that se, sa (Polish sie) etc. carry or provide the feature specifications necessary for the self-sufficient interpretation and, being part of a CHAIN with [NP,IP], that is, with pro, and, by transitivity, with

132

Chapter 7

Agr, share relevant features and index with both. Besides [+person] (personal Agr), there is no other (number) feature specification required in the CHAIN. The reflexive structures of this type are incompatible with reciprocal anaphors (see R˚uzicka 1992: 152). But we don’t have to specify pro as [–plural], since in Slavic languages [3 ps sing (neuter)] is the default Phi-feature agreement specification if the s-structure subject position is occupied by a non-lexical (pro)noun lacking an inherent number feature [+pl] or by a (pro)noun carrying [–nominative].32 The nonargumental clitic does what Cinque (1988: 530) — with respect to Italian si — calls “… serve(s) as a syntactic means to supplement personal Agr with the features able to ‘identify’ … the content of pro as an unspecified (generic) person pronominal, an interpretation that would not be available to the ordinary person inflection paradigm.” Note that the feature [+human] is not a result of s-selection. The typical range of [+human]-external argument interpretation rather is the result of an interplay of principles identical with, or complementary to, those that result in passive. In other words, the particular type of structures “inserts” compatible verbs that in turn select [+human] subjects. Or, we might say, the reflexive formative licenses pro carrying [+human]. I will try to elucidate how the impersonal structures under discussion interrelate with the passive. The question arises how the reflexive clitic comes to function in this way. I would like to suggest that what ends up as the two structures just mentioned are consistent arrays resulting from the (co-)operation of identical and complementary principles, the latter being associated with distinct sets of lexical predicates (verbs), transitive vs. nonergative. In the reflexive passive, for example in Czech, the subject A-position is dethematizid by the clitic se, which, being argumental in these structures, binds or “withholds” (Cinque 1988: 535) the external theta-role, and becomes the target of object movement. With intransitives, in particular unergative verbs, the subject position is held open again by constituting a CHAIN with the clitic, but in this case no object is available to move in. So if the reflexive clitic need not dethematize the subject position and can be taken to be nonargumental, it seems to be obvious from the viewpoint of linguistic economy that what the reflexive clitic provides for by being part of a CHAIN with [NP, IP], namely the empty subject position be utilized as an argumental pro for Full Interpretation: pro is an argument to be interpreted so to speak according to its phonetic content, namely as “anonymous”, indefinite specific or generic. In both

Reflexive clitic impersonal clauses in Control structures

133

related structures, the theta role Agent is affected, with its respective DP-recipients being demoted and suppressed (in the passive) or deprived of referential readings (in the impersonal reflexive structures). As we shall see, the reflexive clitic se (or sa, siøe) on which these properties hinge must be marked [+person] in the reflexive passive as well. (301)C

Kongresi se peclive [nom masc sing] [+person] congress self carefully

pripravil ti [masc sing] prepared

has been  ‘The Congress   carefully prepared. was   While the pro-subject is available as recipient of the external θ-role in the intransitive structures like (294) – (296), the subject position in the reflexive passive is occupied and no recipient of the external role, withheld by the reflexive from subject position, is available at all, for example, in Czech, Slovak, Polish and Serbo-Croatian, as the counterpart to a by-phrase (see R˚uzicka 1992). In accordance with Cinque (1988: 530), I assume that in both relevant structures the most plausible position of the reflexive clitic is the adjunction site under Agr. Its effects on the shape of VP and its membership in the CHAIN with the subject-DP in the impersonal structures justify this location. Agr itself of course carries the “strong” agreement features corresponding with the subject pro. But pro does not transfer gender or number and the presumable intrinsic or conceptual plural of pro and se does not “come out” morphologically. Thus, the minimal and maximal set of agreement features of Agr consist of the default features [+neuter] as in (294) or [+3 ps, +sing] and the intrinsic feature [+person]. This is a case of Spec-head agreement with Agr dividing the agreement load between itself and its adjunct se. I neglect Case assignment. The reflexive clitic, a nominal element should receive [+nominative] in both structures. An alternative analysis has been suggested by Dobrovie-Sorin (1998). For its relevance to Control, see subsection 7.1. Summing up, in the passive, which may be considered to be a “transitive” nonimpersonal contextual variant of the main type of impersonal reflexive clitic, we assume that se is an argument and is assigned the external theta role, or, varying the metaphor, withholds it. As under Burzio’s Generalization, it deprives the verb of its capacity to assign accusative. Since the A-position and

134

Chapter 7

structural Case position of the direct object is closed, it seems natural that another argument position, the subject position, must open in compensation. Controlled clauses are incompatible with the reflexive passive in Slavic languages, a fact which conforms with Cinque’s (1988: 561) observation that “Infinitival control structures are generally taken to be incompatible with passive si …” Cinque continues “… passive si appears to be compatible with control … where … pure middle interpretation is possible …” But, for example, in Russian, even control cases in which an embedding noun “prompts” pure middle interpretation seem barred. (302)R

*Eti okna imejut svojstvo (PRO legko myt’sja) these windows have the property easily to wash-refl

(303) illustrates that reflexive non-middle passive cannot be realized in controlled clauses. (303)R

* Boris pytalsja (PRO Boris

tried

ne   not  

zamecat' sja  priglasat' sja  to be noticed  to be invited 

Of course, this is not due to the passive interpretation as such. (304)R

ne byt' Borisi pytalsja (PROi   Boris

tried

zamecennym  priglasennym  [ part pass pret instrum sg] not to be noticed   invited  

The participle (analytic) passive construction is compatible with control, as many examples besides (304) have evidenced. Cinque (1988: 558, 561) attributes the ungrammaticality of near Italian analogues like (305), (his (83)) to a violation of the Case filter, following Burzio (1986: 44) and others. (305)I

*Quei prigionierii vorrebbero [PROi essersi già liberati ei] ‘Those prisoners would want si to be already freed’.

“… the ungrammaticality of such cases as (83) does not follow unless we assume that si continues to require nominative Case as in (2b) (= *Sarebbe meglio lavorarsi un po’ di più ‘It would be better si to work a bit more’)”

Reflexive clitic impersonal clauses in Control structures

135

(Cinque 1998: 558). Cinque turns to “… a solution to this apparent paradox (“… si appears to absorb both accusative and nominative, with nominative being in fact shared by si and the NP determining agreement”) … suggested in Chomsky (1986b: 132 ff.).” He assumes si to be [+ arg] in examples like (306) (his (84)), (306)I

Gli spaghetti si sono già comprati ‘The spaghettis si are already bought (“have already been bought”)’

which means “… that si will have to be in a CHAIN assigned nominative Case and will withhold the external theta-role.” (loc. cit. 558) Since in (305) PRO has no Case, “… The unsatisfied need for nominative is, then, what causes the ungrammaticality of (305) …, where a pure middle interpretation of si is not available.” (p. 561)33 The grammaticality of Russian (303) need not founder as a result of a violation of Case-theory in the same way as (305), if the PRO-subject is correctly assumed to receive default dative Case. Rather than turning to Case theory, I would like to suggest that the potential contingency of analyzing pertinent reflexive structures as passive plays a significant part in the failure of control. This assumption is supported by the fact that the S-structure subject of reflexive passive is predominantly [–human], which often backs the understanding as passive and simultaneously may prevent infinitival sentential embeddings getting involved in Control. We generalize the conclusion that the reflexive passive, which may be contingent on the help of particular lexical factors and, thus, borders on construal, does not lend itself to the construal of Control. Approaching now an account for the contrast between “analytic” passive as in (304) and reflexive passive as in (301), (303), we may rely on a conceptually similar analysis of the former in the attempt to draw a distinctive parallel. “… like other clitics, -en forms a chain with a full NP (see, for example, Jaeggli (1982), Borer (1984)). The NP that forms the coda of the chain may be overtly realized as a by-phrase, giving rise to ‘long passives’ … If the NP is not overt, a ‘short passive’ is formed. The existence of a clitic chain in passives implies that -en has a referential index … in short passives, our claim is that there is an empty category linked to the argument -en.” (Baker et al. 1989: 223)) Thus, -en and se etc. are taken to be arguments in the respective passive constructions.

136

Chapter 7

Capturing the difference between the two passive constructions and their respective CHAINS34 hinges on the elements which are linked to the clitics. In Slavic, “transitive” reflexive passives have the as yet unfilled subject position e linked to, and dethematized by, the argumental reflexive clitic, an operation falling under the notion of functional CHAIN. In -en-passives, the by-phrase, its counterparts, or the corresponding implicit (IMP) is linked to the clitic -en. In these considerations, I assume that the Russian or other Slavic “analytic” passive is similar to the -en-passive in the relevant aspects. In the -en-passive (and its counterparts), the morpheme is an argumental affix tied to the Infl node which unequivocally “identifies” the argument and the thetarole, as the subject-NP explicitly does in the corresponding active. In contrast, the occurrence of counterparts to the by-phrase is strongly restricted and in some languages even excluded in the reflexive passive. Besides, object movement, common to both types of passive, is unique to passive structure only in the analytic type (cf. R˚uzicka 1992: 138). Crucially, the passivization process is not unequivocally determined by the reflexive element, that is, the surface subject of verbs to which the reflexive clitic is attached may retain external argument position, and it is privileged to contain a moved d-structure object of ergatives (cf. R˚uzicka 1992). Properties of the s-structure subject and other contingent lexical properties take essential part in determing a passive or non-passive interpretation of reflexive structures. In general, passive interpretation in reflexive structures is impeded by the presence of [+human] in the PRO-matrix. Control verbs, however, predominantly are accompanied by [+human] for PRO in their complement. As a consequence, pertinent Control constraints, in particular (K), often have no unequivocal theta-specification, namely 〈intact,–〉 to apply to in potential reflexive passive structures, in contrast with the analytic passive. Cf. (307). (307)R

*Otec

popytalsja (PROi

‘(Their) father tried detjam) [dative] to (his) children.’

vozvrašcat’sja [+refl.] to be given back

This seems to be the intuitive empirical underpinning of the preclusion of controlled reflexive passive, beyond its possible Case-theoretical account.

Chapter 8

“Self-control” with evaluating and attitudinal predicates

The controlled clause is the propositional (event) argument of predicates which are taken in a broad sense to include epistemic, deontic, emotional, moral, and psychological states, properties or attitudes. They are verbs or adjectives that constitute a straightforward case of self-regulating control. No particular constraints licensing control need be imposed either in terms of relations between theta-specifications or in other terms. Their inherent control mechanism is not any different than with verbs obeying (K), that is try, manage, etc., apart from even dispensing with (K) in the subset (a). Intuitively, it is quite natural that the evaluating attitude taken towards an event, be it epistemic, deontic, “psychological” or otherwise, is compatible with, or may result from, any specific part the referent of PRO takes in the event described in the controlled clause. Another intuition suggests that the identification of PRO poses few problems because either the only overt argument of the immediately superordinate clause is the antecedent of PRO, or if the single matrix NP argument is empty or implicit or there is no projected NP-argument at all, the problem of identification arises in the matrix clause, but not, originally, for PRO. Control, then, is determined indirectly by resolving identification in the matrix clause.

8.1. Adjectival predicates 8.1.1. Two basic situations There are two basically distinct situations that circumscribe the range of control and constitute two distinct subsets (a) and (b):

138

Chapter 8

(a) The participation or involvement of the referent of PRO in the event described by the controlled clause calls forth the — evaluated — state s/he experiences in taking part in this event. Thus, PRO effects being identified with the matrix controller, namely the experiencer-NP. (b) Any or a particular instance of the event described by the controlled clause is evaluated, e.g. morally, intellectually or otherwise, by a speaker or a person mentioned or understood, which implies evaluating the person(s) responsible for the event. In particular, the evaluation is expressed in relation to, and depending on, its specific responsible agent. If the latter, the only candidate for controller, is nonovert, the notion of arbitrary PROarb may prove practicable as in cases of propositions about an unlimited range of PRO, or of unequivocal pragmatic control. Importantly, only in (b) the thematic specification 〈intact,+〉 is involved in control,which can be considered a case of cross-clausal s-selection, implementable by (K). 8.1.1.1. Situation (a) and Epstein’s formula. “(∀x) if x plays baseball, it is fun for x …”, applied to It is fun to play baseball. “… the correct LF-representation of the sentence is (308)” (Epstein 1984: 502 ff., (8)). (308)

[S’ (∀x1) [S It is fun (for) x1 [S’ (for) [S x1 to play baseball]]]]

Syntactic evidence for the existence of these two NP positions (each occupied by a variable in (308)) is provided by, for example (309), (Epstein’s (9)). (309)

It is fun for Lucy for Joe to play baseball.

“With respect to the LF representation (308) … in the correct S-structure representation of the sentence, so-called PROarb is controlled by … a basegenerated quantificational empty category, namely, pro, occupying the governed complement NP position to the adjective.” (Epstein 1984: 502 ff.)) At S-structure, “quantificational” pro is the special case of the matrix NP which is subject to a “reverse control” process — participating in an event “precedes” the evaluated feeling arising from it — in the same way as an overt NP. This process is reflected in Epstein’s LF-representation (308), in which the Raised Quantifier (pro at S-structure) binds the two occurrences of the argument variable x1, identifying controller and controllee35 . Apart from basic differences in their respective approaches to control, Borer’s (1986) as well as Epstein’s analysis try to reflect the semantically

“Self-control” with evaluating and attitudinal predicates

139

driven and self-generating “reverse” control mechanism of (a). Epstein’s version, which more easily generalizes over overt controller NP’s, seems preferable, if the notion of universal quantifier is modified as restricted universal quantifier or as Borer’s set concept, which is more flexible and adaptable to the interpretive procedures, that is, accessing an antecedent or construing the content of the null element from the knowledge set of the addressee and/or the speaker, respectively. The problem comes out clearly in sentences like (310) (Epstein’s ex. (4)). (310)

Josh said it is fun to play baseball.

Epstein (l.c. 502) argues that “A representation of the coreferential interpretation of PRO is obtained only if pro is freely coindexed with the NP Josh. If pro is freely contraindexed (and is therefore antecedentless), a representation of the quantifier-bound variable interpretation of PRO is derived. The fact that there exist two possible interpretations in such sentences is thus explained.” Borer (1986) simply comments that: “… the null operator in control structures could optionally be coreferential with an NP in the clause.” Either account, however, records rather than explains what goes on here. Embedding the control structure in a higher superordinate clause, as in (310), is epiphenomenal to the identification of pro, the empty argument of the matrix predicate. If such an embedding clause happens to contain an NP referring to a person who says or thinks that, e.g., ‘it is fun, or easy, difficult or tiring for someone to VP’, the interpretation will naturally be prompted to include, or at least to consider including, that person into the (unit) set which it should be possible to access “cooperatively” in a given discourse situation. Accessing an antecedent of pro and, consequently, PRO, in particular choosing between an NP in the adjacent or higher clause, or in discourse, on the one hand, and “arbitrary” reference, on the other, when the former choice is not compelling in virtue, e. g., of gender agreement (ex. (311)), are contingencies that cannot be divided among areas definable by principles of grammar proper. Consider (311). (311)

thinks said 

  

thinks that Mary j   said   to feed oneself x     that it is easy pro i, j,x (PRO i, j,x herself  (to solve the problem 

Kate

140

Chapter 8

Desirable representations of these identification procedures must include specifiable parts of the “pragmatic universe of discourse” in particular mutual knowledge assessments (cf. Yokoyama 1986: 46 ff.). Consider now the difference between (312) and (313). (312)R

Legko lgat’, znaja cto otveta ne (it) (is) easy to lie knowing that (no) answer (not) posleduet will follow

A simplified S-structure of (312). (312)′R (( pro (ø legko (pro)i (PROi lgat’))) [pleonastic] [copula pres] [adj neuter] it (is) easy for X to lie (PRO’i znaja (cto (otveta ne posleduet)))) [part. adjunct] [gen. sg] knowing that answer not will follow The indexed proi, which carries [+human], [+pl], is the controller of the first PROi, which in turn controls the PRO’i-subject of the participle adjunct. pro is interpreted [+generic], alternatively as bound by a restricted universal quantifier. (313)R

Osudit’ bylo nelegko to convict (him) was not easy

(313)′R ((PROj

osudit proi)k ((bylo nelegko [acc]

proj) (tk)))

‘(For them) to convict (him (Jesus)) was not easy (for them)’ Both occurrences of pro are clearly discourse-related. The first, in direct object position, possible in a null object language like Russian, refers to Jesus, the second to a set of people concerned with convicting Jesus. (314)C

Bylo proi (it) was (for him)

nesnesitelné (PROi z˚ ustat v Curychu) [adj. neuter] unbearable to stayin Zürich

pro in (314) exhibits again a clear case of discourse anaphora, compare (315).

“Self-control” with evaluating and attitudinal predicates (315)C

Kolegovii bylo líto (PROi [dat sg] for the colleague (it) was grievous malé bastardy) the little dogs

141

zabíjet to kill

8.1.1.2. Situation (b) In situation type (a), participating or being involved in the event — in any “semantic role” — evokes the psychological or other, e. g., intellectual, state described (it is fun) or evaluated (it is easy) by the matrix predicate. Control is determined automatically in reversing the directionality as suggested above. Control constructions, while enforcing this identification, do not exhaust the referential pairings. For example, x may get his fun out of y doing something, as illustrated in (309) above. Such examples in turn show that two arguments belonging to different clausal structures are involved, which is a precondition to control. In situation type (b), the evaluating predicate (“mental property adjectives” in the terminology of Stowell (1991)) of the matrix clause is not related to the (mostly psychological) state that is engendered by participating in the event described in the complement clause. Rather, the way of intentional (responsible) acting that is chosen, as described in the infinitival complement, is evaluated, and, thus, the evaluating property must be attributed to the (human) agent as well. The consequence is “obligatory” control: The referent of PRO cannot be distinct from the target (person) of the evaluation given in the matrix clause. This is another distinction between (a) and (b). (316)

*It is foolish of Peter for Joe to neglect this problem.

(Compare (316) with (309)) The intrinsic relationship between the matrix predicate and the event argument, which distinguishes (a) from (b), that is, the difference between the event “engendering” a property (fun, for example in (310)) in type (a), and the event being characterized with respect to a property attributable to its responsible agent in type (b) may account for the contrast between (317) and (318), for which Stowell (1991) has “… no account to offer …” (Stowell’s (i), (ii) in fn. 9, p. 133) (317)

*It was stupid of John washing the car.

142

Chapter 8 (318)

It was fun washing the car.

A present participle adjunct (comp. washing the car was stupid of John) constitutes a relation to the main predicate restricting its merger with the latter to adverbial modification. A merger of the adjoined event phrase (washing the car) with predicating evaluations of its responsible agent (stupid) seems to be ruled out. The syntactic LF-construal and parsing would rather tend to direct the interpretation towards something like It was stupid of John to do x (while) washing the car. The experience of fun and the washing event, however, coincide or merge naturally, the latter “engendering” or accompanying the former. The distinction between the sets (a) and (b) finds support. Predicates (a), (b) otherwise induce control mechanisms which do not require an independent account. They both are stage-level predicates containing an event argument, following Kratzer (1989) and Diesing (1988). In (b), the event is an action denoting argument. Predicates of (a) and (b) have an obligatory, possibly nonovert, human argument. (319), (320) illustrate type (b): The question of a matrix antecedent will be taken up presently. (319)R

Byvajut

slucai,

kogda ø bolee celovecno (PRO [copula] (there)are situations when (it) is more humane solgat’) to lie

(320)R

On napominal nam o tom, cto naivno (PRO [pret] [dative] he reminded us about it that (it) is naive ob-jasnjat’ nacional’nuju katastrofu zlovešcej [instrum.] to explain a national catastrophy (by) the sinister rol’ju odnogo celoveka) [gen] role of one man ‘… by pointing out the sinister role played by one man’

If in type (a) the psychological state characterized by the matrix predicate is engendered in taking part or being involved in the (often “generic”) event described in the complement, in type (b) the property by which the (intentional) action is characterized in the matrix clause is exhibited in doing what is

“Self-control” with evaluating and attitudinal predicates

143

described in the complement. Importantly, whether somebody to whom the quality expressed in the matrix predicate is attributed shows up lexically or not, depends, to a certain degree, on the particular quality in question and the content of the complement. Thus, the assertion that it is naive, humane to VP’ in (319), (320) is not taken to be valid only with respect to particular individuals or groups of them. Now consider (321), in which the matrix predicate is related to an individual. (321)R

Bylo

glupo (PP s

tvoeji storony) (PROi [gen] (it) was stupid from your side vystupat’ v ego zašcitu [acc] to stand up in his defense ‘It was stupid of you …’

Tvoej, a possessive, genitive-like pronoun, “incorporates”, and functions as the antecedent of PRO (comp. Italian da parte di). Glupo is related to you and considered valid with respect to this individual. Incidentally, in many languages a prepositional phrase is the only syntactic means to express the restriction to individuals or sets of them. The relatedness of the matrix predicate to an NP that is syntactically expressed or understood raises the question whether predicates of type (b) project an NP argument. The corresponding adjectivals can take head nouns that are descriptions of persons as well as of actions or behaviour: a stupid boy (r. glupyj mal’cik), a stupid decision/ behaviour, action etc. Relatedness to individuals seems to be established indirectly via the event or action associated with their agents. If no overt NP is associated with the predicate, the default case is a generalized interpretation over human values, as in (319) or (320). Semantic differences dividing celovecno, in (319), and glupo, in (321), influence or may even determine the appearance of NP specifying individuals as in (321). This NP or D(P) will of course be the antecedent of PRO. But it is the construal of PRO that is transmitted to its antecedent and, thus, “precedes” it. In contrast to type (a) constructions, in which there is an inherent eventinternal identification of PRO with an argument in theta-position, in type (b), relatedness of the matrix predicate (adjective) to an argument is achieved by PRO, to whose referent the “mental property” predicate is attributed, exclusively by virtue of her/his decision to do what is described in the clausal

144

Chapter 8

complement. In other words, only one argument or at least only one theta role is available, basically (the chain headed by) PRO, which either is resumed in the matrix clause in overt shape or is left in the impliciteness or “arbitrariness” of PRO. Note that the evaluation associated with the matrix DP is inseparable from, or identical with, the evaluation of the intentional action of the complement clause and valid only with respect to it. Expressions like Peter is clever are understood as relating to his cleverness at something. Stowell (1991: 112) clearly recognizes that “… MP (mental property, RR) adjectives cannot attribute a property to an action without simultaneously attributing this property to the agent of that action … the human argument in question must be understood as the agent of the action-denoting argument.” I would like to put forward the suggestion that the matrix PP, e.g.. in it is stupid of John does not occupy a theta-position, and that its DP is raised from the embedded subject position leaving the chain (Johni, ti). By of-insertion, which adjectives permit (Chomsky 1986b: 89), the Case filter is met. The PP s tvoej storony is a clearly idiomatic variant. I will not pursue this radical proposal of a unified theta-role here, returning to a standard treatment, which can be transformed into the proposed one. Quite independently of whatever thematic specification PRO receives as an argument of the complement predicate (VP or AP), its intended referent is accessible to the matrix characterization only if s/he is understood as “a decision taking” individual, a fact which might be expressed by PRO being provided with 〈intact,+〉, via projection from the matrix predicate somehow inherently marked with it. But this proposal cannot be fully maintained. Consider (322), in which PRO receives 〈intact,+〉 originally from the theta grid of otsizivat’sja. (322)R

No edva

li

razumnee bylo (PROi otsizivat’sja [pret neuter] but hardly more sensible(it) was to take shelter odnimi bez oruzija sredi etoj pustyni) [gen sg] alone without arms in the middle of this desert

The dative plural of the semipredicative odnim, which is coindexed with PRO, refers to the discourse-related set of people to the exclusion of others. It must not be exactly this set of people whose potential decision is judged comparatively. But now consider (323), which must be ruled out since no decision

“Self-control” with evaluating and attitudinal predicates

145

taking individual or set of people are available. Only PRO could provide one saving (323) if it were supplied with 〈intact,+〉 on the strength of feature transfer or discharge. (323)R

Bylo glupo (PROi byt’ priglašennym ti na etu (it) was stupid to be invited to this konferenciju) conference

Compare: It was stupid to invite him to this conference. Thus, transclausal projection of 〈intact,+〉 to PRO from matrix predicates of type (b) cannot be admitted generally. This restriction will be stated as (L), which is similar to (K), a specification on the application of the latter with respect to predicates of type (b) (L) The presence of 〈intact,+〉 somewhere else in the complement clause prevents another instance of it from being assigned to PRO, (across clausal boundaries) As a third case, consider (324), which is perfectly acceptable.

(324)R

Bylo by

glupo  nerazumno  

(PRO predavat’ sja

stupid  insensible   

to indulge

[modal particle] (it) would be melancholii [dat sg] in melancholy 〈intact,+〉 is not discharged in the complement clause. The matrix predicate, however, requires that 〈intact,+〉 is present in PRO, if additionally, provided that, in view of (L) it is nowhere else in the complement. The across clauseassignment to PRO in cases like (322), though problematic, is induced by the necessity to identify a target person that exhibits the property expressed in the matrix clause in deciding to act as described in the complement. Note that the problem of cross-clausal assignment of 〈intact,+〉 to PRO disappears if the Raising analysis suggested above proves feasible, since the chain must be marked 〈intact,+〉. The control problem in (b) can further be elucidated by

146

Chapter 8

syntactic variants of the same conceptual content. (325)R

Oni postupil glupo he

acted

predavšis’ [participal adjunct] stupidly indulging having indulged

melancolii [dat] in melancholy (326)R

Oni postupil glupo / naivno / necelovecno, (PROi he acted stupidly / naively / inhumanely sdelav eto) [part. adjunct] doing that ‘Hei acted stupidly … in that hei did so.’ ‘It was stupid/naive of himi to do so.’

The Pro-VP postupil (3 ps sg pret) in (325) is explicit about what is construed in (324): In (324), the matrix adjective is a predication about the action- and decision-event of the complement clause. In (326), the adjective-adverb, an instance of a non-omisssible modifier, directly evaluates the action expressed by the Pro-VP, which functions as a dummy-antecedent of the discourse- or situation-related Pro-VP of the participial adjunct. The sole function of the matrix Pro-VP is to mediate between the evaluating modifier and the matrix subject, whose referent is characterized by it on the strength of doing or deciding to do so (sdelav eto in (326)). The two Pro-VP’s can be exchanged easily. But then postupit’ must receive an anaphorical marker, the manner Proadverbial tak, the contextual variant of eto, which combines with the Pro-VP (s)delat’. Tak and eto are in complementary (s-structure) distribution with the evaluating adverb, either of them an indispensable part of the Pro-VP. (327)R

Oni glupo sdelal (PROi postupiv tak) he stupidly did acting thus

The equivalence of the construction types exemplified by postupil glupo/ naivno (325), (326) and bylo glupo, naivno (in (321), (324)) as to their clausal adjunct and complement clause, respectively, clearly supports the necessary presence of the specification 〈intact,+〉 in the complement subject of the latter,

“Self-control” with evaluating and attitudinal predicates

147

which is assigned clause-internally, or across clausal boundaries, as in (322), if (L) is observed. 8.1.2. For “better” or “worse” Control adjectives like good or bad, more characteristically their respective comparatives better and worse, differ from the preceding ones in that any thematic specification or, for that matter, semantic role carried by PRO is compatible with a well-formed control relation, that is, provides a self-reliant identification. In the default case, it should be expected that PRO be identified with an overt matrix argument or an “implicit” one that directly or anaphorically refers to, or “arbitrarily” generalizes over, the set of people for whom the (non) occurrence of the event described in the complement clause is better or worse. The default case cannot emerge if the set of persons referred to, or an implicit “arbitrary” universal set, (those for whom it is better or worse) is not a participant in the complement event description, and, thus, cannot show up as PRO in it. One would expect them (her/him) to be specified in the matrix clause as those for whom the evaluation holds. This expectation is not quite fulfilled, as can be seen in (328). The interpretation may be left to inference. (328)P

Ani

poczøatku, ani ko´nca tym razem [gen sg] [gen sg] [instrum case] the beginning nor the end this time lepiej (PRO nie ujawnia´c

neither ø [zero copula] (it is) better

(not)to report

The interpretation of PRO is discourse-related. It refers to the author of the monthly review of films. The default interpretation would be it is better for me this time not to tell about either the end or the beginning (of the film). But the more natural one is: … it is better for the readers (the potential audience of the film) for me not to tell about … Both interpretations might even combine. Indeed, any combination of discourse-related and “arbitrary”, generic or generalizing interpretation is possible, which again would obviate special control regulations. I will add one example with an obvious default case of control.

148

Chapter 8 (329)R

I

mne kazetsja, cto ocen chorošo bylo by [dat] and to me (it) seems that very good (it) would be (novomu našemu agrarnomu komitetu)i (PROi [dative] for the new our agrarian committee zaimstvovat’ iz raboty russkogo zemstva to borrow from the work of the Russian zemstvo (before 1917 organs of agricultural self-administration)

8.2. Verbal predicates of evaluational or attitudinal character 8.2.1. The oblique object controller Such verbal predicates, which subcategorize for infinitival complements, pose no new problems. The other element they subcategorize for, an oblique objectNP with inherent dative in Slavic languages and German, provides the controller. The controller receives Experiencer with these verbs. No thematic restriction is imposed on PRO. (330)C

Líbí

se mii (PROi vyprávet) [lexical refl. [dat clitic] (it) pleases me to tell (stories)

(331)C

(PROi Cist) se to read ‘He likes to read.’

mui líbí [dat] him pleases

8.2.2. The subject controller Corresponding personal constructions present an analogous picture with the controller NP occupying the S-structure subject position. (332)C

Petri bál se (PROi ho urazit) [lexical refl.] [acc] Peter was afraid him to offend ‘Peter was afraid of offending him’

“Self-control” with evaluating and attitudinal predicates (333)R

boitsja [PROi komu-to ne ponravit’sja] [dative] he not is afraid somebody not to please ‘he is not afraid of displeasing somebody’

(334)C

A

149

Oni ne

ted’

dokonce [nekteri

lidé]i se domnívají (PROi [lexic. refl.] and now even some people believe být povoláni (PROi naplnovat deijnnou to be authorized to fulfil historical spravedlnost justice

Chapter 9

Aspects of control in the “grain problem” and in VP-deletion

This short chapter has been included in the investigation of Control since independent solutions to the grain problem as it emerges in Control, and to anaphoric contexts in VP-deletion, are perfectly consistent with the theory put forward in this study, providing instances of its formal indexing consequences.

9.1. The “grain problem” and reciprocal constructions The problem pointed out by Higginbotham (1985) and called “grain” problem “… reasserts itself in an even more pernicious form when simple control constructions, such as those in (45) and (46), are brought into the picture. (45) John and Mary persuaded each other to leave. (46) John and Mary promised each other to leave, … But the rub is that (45) and (46) are not ambiguous in the same way as (43). … (43) John and Mary told each other that they should leave. … this sentence is at least three ways ambiguous between what we termed the ‘I’, ‘you’, and ‘we’ readings … (45) has only a ‘you’ reading, (46) only an ‘I’ reading …” (Heim, Lasnik, May (1991: 79)) The authors continue that “It would seem then that unless we build the theory of control (redundantly) (my emphasis, RR)) into the semantics of these sentences, we are left without a satisfactory account of their interpretation.” Any description of control, of course, must account for the distinct control properties inherent to persuade and promise, respectively. Since the reciprocal each other occurs in object

152

Chapter 9

position but in fact conceals a subject-object relation, it must be taken apart for “subject”- and “object”-control to be theoretically reconstructed. But the complex reciprocal cannot be simply taken asunder, as each “instance” of other is determined by, and must be related to, its distinct “distributing” partner each. Heim, Lasnik and May (1991: 80) present a basic representation of reciprocal sentences which “… provides three loci of anaphoric binding: relative to the distribution index contributed by each, relative to the index of e other, the reciprocator, and relative to the range index of the antecedent subject NP.” Resolving the complex reciprocal into the subject-part each and the object-part e other, with the empty element coindexed with each, enables the authors to construe the respective LF-representations (50), (51) (their numbering) for (45), (46). (50) (John and Mary1 each2) (persuaded (e2 other)3 (PRO3 to leave)) (51) (John and Mary1 each2) (promised (e2 other)3 (PRO2 to leave)) PRO is coindexed with the subject-part each for promise, and, for persuade, with the object-part (e other) of the reciprocal each other. Either isolated element is to be considered an argument and recipient of a theta-specification, that corresponds to their respective subject and object positions. Thus, both of them are accessible to the generalized lexically-driven constraints (B,b) and (B,a), respectively. These constraints and the above indexing and binding device at LF-representations combine to give a full account for the control behaviour of the two classes of verbs in reciprocal constructions. An analogous analysis would be practicable, e. g. for the German, Czech, Russian counterparts of each other control constructions: g. ein-ander, c. jeden druhý, -ého, and r. drug druga. The Slavic languages exhibit the object function of the second part by morphological Case36 . I would like to stress that the success of the intriguing indexing system of Heim, Lasnik, May (1991) on its application to differences of scope interpretation and to control properties interacting with reciprocal constructions, substantially relies on “… inherent semantic properties of the controlling verb and properties of control PRO … narrow scope is excluded when the controlling NP must be under the scope of a distributor, as a function of the lexical meaning of the predicate of which it is an argument … the possibility of broad or narrow scope is a function of the semantic status of the controller of PRO.” (Heim, Lasnik, May 1991: 99). Thus, the main claim of this study that, in the

Aspects of control in the “grain problem” and in VP-deletion

153

circumscribed area of infinitival complement constructions, control properties are derivable from the lexically-semantic structure of specifiable classes of control verbs on which generalizing control constraints are imposed, is corroborated in this particular field. Interaction with scope problems is a special case of the interaction between selected semantic properties of the controlling verbs, which are channelled through the respective pertinent constraints and general principles of binding as well as distinct settings of relevant parameter values.

9.2. Freedom of constraints as a precondition for strict identity in anaphoric contexts PRO — in Higginbotham’s (1992: 90) terminology — “… covaries with focus, clefts, and various ellipses, notably VP-deletion; … admits invariant interpretations in some other anaphoric contexts.” His example (24) is from Tanya Reinhart. (24) John wants [PRO to be a doctor], but his mother doesn’t want that. Which are those contexts, in which PRO may be invariant, as in (24)? Invariant (strict identity) interpretations depends on freedom from both main constraints (B,a) and (B,b), that is, roughly, only the control predicates of the family assembled under chapter 8 are free from them. Thus, the constraints correlate with the exclusion or option of invariant interpretation. In other words, the respective thematic interrelationships fixed between licit controllers and PRO by (B,a), (B,b) and (K) are not consistent with strict identity in VP-deletion by conceptual necessity. If a control verb like, e.g., want imposes no restrictions on the mutual thematic relationship between the licit controller and PRO, anaphoric “recovery” of the reference of the PRO subject of the deleted VP doesn’t seem to be impeded. The fact of restricted identifiability correlating with the posited constraints provide further empirical evidence in support of them.

Chapter 10

Summary and conclusions With some remaining questions

10.1. Essentials and main general proposal What has received the slighty metaphoric name control is a linguistic syndrome rather than a module or subsystem of grammar that could be cast in a unified form. The complex of problems posed by the phenomena standardly subsumed under control nevertheless can be disentangled and organised into an array of constraints that determine control behavior and set aside factors which are epiphenomenal to, if necessarily accompanying, control. The natural prerequisite of attaining descriptive generalizations about control conditions is to specify the structural descriptions that must or can display control. The specification proposed divides the realm of control into relevant domains and selects for analysis those that are related and subject to principled constraints which can extend a certain homogeneity to the whole field. Crucially, the interplay between lexical semantics, thematic structure, syntax and explicated pragmatics hinges on the respective constraints and their format. In the major syntactic control domains, nonfinite adjunct clauses differ fundamentally from complement clauses. In the former domain, control properties of participial adjuncts differ from those of purpose or rationale infinitival clauses. Control in nonfinite adjuncts has not been dealt with here — for modern Standard Russian, a detailed description of control in participial ° adjuncts is Ruzicka (1982) with further literature — but the theoretical spirit in which it could be approached, as well as the methodological line to be taken have been suggested (2.1.3.). In the area of infinitival nonadjunct clauses, control in subject clauses differs considerably from control in object clauses. “Object position is pro-

156

Chapter 10

jected only from thematic structure …” (Rothstein 1995: 500). Subject position is projected syntactically and need not be thematically related to the predicate. Object clauses but not subject clauses exhibit strongly marked lexical determination of control, a fact which has to do with the assumption that the theta role of the subject — if it has one — is determined compositionally by the VP (Chomsky 1981: 105). Object complement clauses are internal sentential arguments of the control verb which theta-marks them. Their PROargument, which is the head of an A-chain, needs identification. It has been shown in this study that the identification of PRO ultimately hinges on specific properties and functions of the internal clause argument, the controlled clause. That is why there is a fundamental difference between control in object sentences and control in subject sentences, which, among other things, has contributed to the use of notions like arbitrary and logophoric control. The distinct types of control verbs, by thematically differentiating between their respective clausal complements, indicate how the content of the latter is to be incorporated in the sentence. In the literature, distinctions between thematic properties of clausal complements that arise from lexical characteristics of control verbs and in turn induce control conditions have not been given careful consideration. The clausal complements of, e.g. ask, promise, or try and their respective crosslinguistic counterparts have been treated with a certain disregard of such particular pairing of thematic specifications as could possibly attend or determine control conditions. Distinctions and influences that have this effect are taken to be a precondition of the main general thesis of this study: Thematical properties of clausal complements that are determined by their respective matrix control verbs are projected to, and distributed among, argument positions of the clausal complement and come out clearly to the effect that they impose distinct restrictions on control relations. Thus, conditions of coreference for the complement PRO are created by lexically thematic properties of the matrix control verb. PRO is required to have thematic specifications meeting those of a licit controller. How it must meet them depends on, and varies with, lexical classes of control verbs, which in turn are established by the respective types of referential interdependence they command.

Summary and conclusions

157

10.2. The constraints and markedness Implementing this relationship, each lexical class is associated with a particular constraint defining it as a control class. The constraints have the general format of a two-membered conjunction of feature specification values that are required of controller and controllee in the respective control classes. If F is the designated general control feature — which may be interpreted as intentional action (see section 3.2.1.) but can be taken to do the work independently of this particular interpretation — and α ranges over {+, –}, two constraints distinguished by a minimal difference suffice to cover two major clearly demarcated areas of control. They specify the respective pairs of feature values to be satisfied by controller and controllee. The conjuncts are distinguished by the superscripts CON(troller) and PRO, which mark where the required feature value belongs. Either constraint is associated in the lexicon with the verbs constituting the respective classes as regards control behavior. (B, a): or: (B, b): or:

〈F, α〉 CON & 〈F, α〉 PRO 〈intact, α〉 CON & 〈intact, α〉 PRO 〈F, -α〉 CON & 〈F, α〉PRO 〈intact, -α〉 CON & 〈intact, α〉 PRO

Both generalized constraints have clearly defined unmarked (um) and marked (m) application: α is valued “+” on unmarked applications, “−” on marked ones (section 3.2.2.). Only failure in meeting the value of the second conjunct 〈…〉 PRO counts as violation of 〈B, a〉 or 〈B, b〉 on unmarked as well as marked application (section 3.2.4.1.). Spelling out (B, a) gives (I) (I)

〈F,+〉CON & 〈F, +〉 PRO (unmarked) 〈F,−〉CON & 〈F, −〉 PRO (marked)

Spelling out (B,b) gives (II). (II)

〈F, −〉CON & 〈F, +〉PRO 〈F, +〉CON & 〈F, –〉PRO

(unmarked) (marked)

Unravelling the oversimplified notion of “object control”, we can say that the unmarked version of (II) consistently accounts for it, which means that the distinction between object and subject control can cut through one and the same control verb and, thus, it is at least inaccurate to speak of subject control

158

Chapter 10

verbs as against object control verbs. Ask. beg, for example, are verbs of “object” control if control is licensed on the unmarked application of (II), but they are verbs of “subject” control on the marked application of it, (section 3.2.4.2). Promise-type verbs behave in the opposite way with regard to (I) (section 5). I have explained why this is so. Importantly, if we rely on a markedness concept as specified here for complement control, various forms of more intricate and subtle control behavior can be included in, and captured by, the basic system of constraints. Specifications on the unmarked and marked application of (B, a) and (B, b) provide us with criteria substantiating a subclassification of the main control classes associated with the respective constraints.

10.3. The classes of Control I will present now a survey of the lexical control classes associated with the respective constraints and specifications on their application. More or less exemplary lists of members of the respective classes are added. The introduction of a systematic cross-linguistic perspective has justified the attempt to formulate interface (conceptual-pragmatic) and syntactic parameters, which suggests that pragmatics could be parameterized. The combined framework of the respective constraints, corollaries specifying their application, and parameters, namely (B) – (L), is taken to account for the complement control mechanism. (Order of succession as in the main body of the study) 10.3.1. Control class (B,b) (=II) and its main subclasses. 10.3.1.1. The ask-type subclass (II)a Members of the ask-type subclass (II)a: ask, beg, request, entreat, implore, beseech, conjure, pester, bother. German: bitten, ersuchen (request), anflehen (implore), beschwören (conjure). Russian: (po)prosit’(sja) (ask), umolit’/ umoljat’ (conjure), molit’ (implore), uprosit’/ uprašivat’ (entreat) Verbs in this subclass can satisfy unmarked and marked (II). (D), with the corollary (E), is a condition on the marked application of (II) and (I) and on

Summary and conclusions

159

parameters (G) and (H). Marked control under (II) is substantiated by the activation of the “rest” of available thematical specifications, which display complicity in control by inducing a less “conspicuous” referential pairing of arguments (section 3.2.3.). (1)G

Siei hatte gebeten (PROi im Kloster she had asked in a convent ti untergebracht zu werden) accomodated to be ‘She had asked to enter a convent.’

A natural restriction put on the “weaker” marked control in this subclass is expressed by (F): The more transparent linking of the relevant thematic feature value 〈F, +〉CON to the underived subject position must be maintained for it to be targeted by the marked application of (II) (3.2.2.). Violations of marked (II) (*…〈F, +〉PRO ) give rise to a parameterization in pragmatics which consists in rescinding them by means of the allowance construal — which is considered the positive setting of the parameter — and ends up in a thematic two-place configuration that satifies the very marked constraint (II) that has been violated. The presumed parameter ((H), section 4.1.) is contingent upon the strength of the effect the null object parameter has on the availability of (marked) subject and (unmarked) object control. Positive setting of that parameter, as in Russian, may exclude marked (subject) control — via the allowance construal as in the English example (91) — since the null object satisfies unmarked (II). (2)R

Ja ne prosil proi (PROi menja spasat’) I not asked me to save ‘I did not ask (anybody) to save me.’

English, for example, which has no syntactically overt pro excludes it from functioning as an unmarked controller in accordance with Bach’s Generalization. If in Russian or German, for example, control by the pro-object is mandatory — given that unmarked (II) is met —, English has to rely entirely on “derived” marked (subject) control, a situation which induces the allowance pragmatic-conventional construal (see example 91). Spanish evidences a strong instantiation of the conventionally pragmatic parameter (H) which is not linked with the null object parameter (section 3.3.). The presence of a lexical object argument does not exclude the interpretation licensed by the

160

Chapter 10

“remedied” violation of the marked constraint (II). Thus, ambiguity overrides a lexical object DP which is licensed by umarked (II) (example 88), which seems to imply the possibility of ambiguous unmarked and marked (through allowance construal) control in the presence of a pro object in null object languages like Italian (see comment on example 97). Thus, the situation is complex but transparent though judgements are subtle. Violations of unmarked (II) are illustrated by example (72, 73). Empirical evidence of ambiguity of control supports the assumption of predictable conflicts between the grammatical and the conceptual-pragmatic mechanism of control as well as their interplay. Note that control structures determined by lexical properties and affected by specified pragmatic factors are subject to a construal which in turn is referred to by principles of grammar, the constraints and specification on their application. 10.3.1.2. The persuade subclass (II)b The second major subclass (II)b associated with (II) (=B, b) includes these members: persuade, convince, prevail on, urge, advice, enjoin, invite; prompt, tempt, compel, encourage, incite, force; German: überreden (persuade), raten (advise), drängen (urge), veranlassen (prompt), auferlegen (enjoin), ermutigen (encourage); Russian: ugovorit’ / ugovarivat’, ubedit’ / ubezdat’ (persuade), sovetovat’ (advise), pobudit’ / pobuzdat’ (prompt); Czech: presvedcit, umluvit / umlouvat (persuade), pudit (urge). We have to distinguish subclass (II)b from (II)a. (II)b has to do without marked application of (II), which cannot license control because of the absence of complicity in fulfilling control by induced referential pairing of DP’s that carry corresponding relevant thematic specifications. Such pairs do not step in here to sustain coreference functioning as a control relation, as they do, for example, in the area of (II)a. Structures that would appear to be candidates for being licensed by marked (II) behave in accordance with, and are subject to, unmarked (II), provided that the interface parameter G (4.1.) is set positively in the respective languages. In this case, condition (D) must be met, as it must for “repairing” violations of marked (II) in the ask-subclass. Compare (3), which meets (D), with (4) and (5), which do not.

Summary and conclusions

161

(3) John convinced himi (PROi to be examined by a specialist) (4) *Johnj convinced Anni (PROi/ j to interest/ amuse/ bother Pat) (on the THEME reading of PRO) (5) *John forced mei (PROi to interest Sarah) examples (4), (5) from Thomas Stroik. In (6), which is acceptable, the straightforward reading which half metaphorically understands worry themselves as intentional meets unmarked (II), with PRO receiving 〈F, +〉. (6)

PROi Invited ti, on high authority, (PROi to worry themselvesi) Britonsi sensibly did just that.

Britons controls the chain headed by the first PRO which in turn controls the PRO subject of worry … . The parameter (G) is fixed positively, e.g., in English, and negatively, e.g., in Russian, German, Czech (section 4.1). The latter languages have resort to lexically causative verbs which yield direct compliance with unmarked (II) (example 122–124). In English, ask and persuade-type verbs evidence “complementary distribution” in control, with respect to the unmarked and marked application of (II) (= (B,b)). This variation is captured by (G) and (H). Control under ask-type verbs can be licensed by unmarked as well as marked (II), and by the positively set parameter (H), which through ‘deontic switch’ handles violations of marked (II), but it has no remedy against violations of unmarked (II). Control under persuade-type verbs can be licensed by unmarked (II) and by virtue of positive settings of parameter (G) handling through ‘causative switch’ violation of unmarked (II), but it can not be licensed by its marked application, much less by undoing violations of marked (II). The dominant role of relations between thematic specifications in control is again evident from the dependence of variations in unmarked and marked control on options of coreference induced by thematic specifications which suggest or determe that the DP’s bearing them in the respective clauses are referentially identical. Since verbs that select [± human] subject (prompt, tempt) are members of (II)b, the nearly complete exclusion of the external argument from controlling finds independent support.

162

Chapter 10

10.3.1.3. Subclass (II)c (II)c includes the following verbs: order, command, instruct, direct, tell, charge, commission, forbid, recommend, allow; Russian: prikaz(yv)at’, velet’(order), porucit’ / porucat’ (charge), zapretit’ / zaprešcat (forbid); Czech: rozkázat / rozkazovat, prikázat / prikazovat, porucit / poroucet (order); zakázat / zakazovat, zapovedet / zapovídat (forbid);ulozit / ukládat, narídit (charge); bránit (hinder); dovolit / dovolovat (allow); radit (advise); doporucit / doporucovat (recommend). German: befehlen (order); verbieten, untersagen (forbid); erlauben, gestatten (allow); raten (advise); auftragen (charge); empfehlen (recommend). Singling out subclass (II)c of verbs (section 4.4.4.) subject to (II) is not as neatly supported as it is in the preceding subclass, by their distinct behavior in regard to parameters (G), (H) and conditions (D), (F). Example (168) is evidence for licensing by marked (II). Control tends to vary with the lexical content of the infinitival complement, receiving license from marked (II) or from unmarked (II) through (G). Some cases have been taken to be amenable to ECM analysis. Judgements can be subtle. Compare (7) with (8). (7) (8)

The medical officer ordered himi (PROi to be examined ti by a specialist) The medical officer persuaded himi (PROi to be examined ti by a specialist)

Clearly, subclass (II)c ist not susceptible of following parameter (H). Separating (II)c from the other subclasses is primarily based on the fact that in many languages (e.g. German, Slavic languages), its members theta-mark the object indirectly, selecting dative Case. Thus, the verbs can serve for a comparison with English, where “… dative control contrasts with object control in that a dative controller can be omitted … an understood dative can control …” (Rizzi 1986: 550 ff.) (9)

a. John shouted/ said/ gave the order (to Bill ) to leave. b. John told/ ordered* (Bill) to leave. (Rizzi’s example (104) a. b., adapted from Roeper)

We can extent Rizzi’s summarizing statement b. to German and Slavic lan-

Summary and conclusions

163

guages. ″b. Both Italian and English have an understood null dative that is not restricted to generic contexts, can be pragmatically interpreted (can be discourse-related, RR), and can function as a controller, but not as a binder.” (Rizzi 1986: 551) In German and Slavic languages (morphological) dative is the counterpart of the direct object controller in English, a fact which simply means that for (II)c the internal dative argument carrying the indirect object theta role gains broader access to control. For the passive matrix sentences, it follows that derived subject control in English corresponds to unchanged dative control in the languages mentioned above. The result is impersonal sentences, since no object is moved. (10)R

Emui bylo poruceno (PROi napisat’ [dative] [partic. pass. neuter] (to) him (it) was directed to write neskol’ko bumag v St. Peterburg [genitve pl.] some papers to St.-Peterburg. ‘He was charged to write some papers to St.-Peterburg.’

The dative controller can occur as a pro-DP in the passive as well. I would like to emphasize that dative control is not affected by theories taking dative to be configurational, not inherent. 10.3.1.4. Verbs of peculiar behavior in (II) Some verbs that exhibit particular control behavior, which still is fully capturable within the range of (II), parameters (G), (H) and attendant corollaries, have been described individually, in particular signal (section 4.4.1.), teach (4.4.1.), allow (4.5.), with counterparts in other languages. 10.3.2. Control class subject to (B, a) (= I) A major class of control verbs somewhat erroneously called verbs of subject control is assembled around promise and its cross-linguistic counterparts. Their control conditions are captured by (B, a) (section 5) (= I), repeated here. (I) 〈F,α〉CON & 〈F, α〉 PRO α = (+) fixes unmarked application, α = (–) marked application of (I). The constraint differs minimally from (II).

164

Chapter 10

Whereas the ask-type and persuade-type subclasses of (II) are subtly set apart by their respective reactions to, and remedies against, violations of unmarked and marked application of (II), which imply cross-linguistic variation, there is no analogous division in the domain of (I) for promise-type verbs. The marked application of (I) is straightforwardly complementary to the unmarked one, reversing the value of α. Example (181–184) illustrate marked control under (I) with passive matrix clauses, (185–188) do the same with active ones. The particularly elucidating example (182) combines constraining grammar and conceptual construal by reinterpreting the complement verb in a way that matches marked (I). Passive is impersonal since the internal dative object is not moved to subject position. If no nonpleonastic subject-DP competes, as in German and Slavic languages, unmarked control by the demoted Agent-DP is facilitated (example 189, 196–198). In English the derived subject argument may be responsible for ruling out unmarked control under (I) by the demoted agent in the by-PP (cf. English glosses of (196) and (198)). The descriptive generalization (I) (in 5.2.) captures the cross-linguistic variation, that is, Case distinctions resulting in different control behavior. If marked (I) (B,a) is not met because (D) is not satisfied — recall that (D) is a condition or specification on the marked application of (I) and (II) as well as on the operation of (G), (H), acceptability can only be salvaged by a conceptual construal which amounts to the restitution of a structure that in turn meets (D) and can be referred to by the marked constraint (I): (187, 188) are handled by analogy to (193, 194) respectively. Another striking example is (219) as compared with (218). The former is clearly ambiguous, with the second reading which takes the children as controller being forced on the sentence by its pragmatic naturalness. The “natural” interpretation is not indifferent to grammar. To become acceptable, it must find its way to a licensing constraint, which is the marked application of (I). The conceptual construal can be made visible by the insertion of be allowed taken as a metalanguage “bridge” from pragmatics to grammar. (218) has be allowed in its structure and apparently meets marked (I) including (D). It should prove unproblematic. Still, there are judgements that prefer “subject” control. This situation may throw some light on the conceptual status and firmness of the grammatical notion of Control. Control may prove susceptible to a complex array of factors in which the syntactic ones appear more reliable but can lead astray. The available control structures may offer options of Full Interpreta-

Summary and conclusions

165

tion, a reasonable price for the inherent economy delivered to language by control phenomena. It is by relying on the very constraints that are violated that solutions to mismatches become available through processes which can be shared by (I) and (II). Take example (211), a clear case of matching marked (I). Slight lexical changes resulting in (212) lead to acceptability by a conceptualpragmatic construal salvaging (212) on the strength of unmarked (I) that is violated. The interpretation has no choice but to understand that der Schüler promises to endeavour to do all he can to be moved up, a conceptual structure which “returns” and conforms to unmarked (I). Abandoning license by marked (I) is enforced lexically and pragmatically. Functional exchange of the matrix DP’s in (212) would result in licensing by m (I). Such construals, as required for (212), however, are sensitive to lexical properties of the complement verb. (210) does not easily give way to them. Marked application of (I) may fail owing to the fact that the thematical underpinning of coreferential partners which in principle is pertinent to marked control (cf. end of section 5.1.) is not available. Bizarre, in particular ironical use , of course, is not inappropriate to (210) licensed by m (I). Grammatical evasiveness of control does not threaten to get out of hand if licensing shifts from unmarked (I) to marked (I) through slight lexical changes, as illustrated by example (213). Marked application of (I) is straightforward if aus dem Gefängnis zu kommen is naturally understood as be released from prison. As the discussion of (214) elucidates, the situation is particularly difficult and judgments are subtle when condition (D) is not matched. Convictions that syntactic configurations can override coherent pragmatically induced control relations vary with the typological status of the relevant language. Nevertheless, attempts to push back the boundaries of capturing the control device have to cross from grammar to pragmatics but can avoid amnestying pragmatic explanation if pragmatics is not released from grammar. 10.3.3. Self-regulating Control 10.3.3.1. The Case of try and related verbs A major class of control verbs centres round try as its prototypical member. Control is intrinsic to the thematic and semantic properties of these verbs, which do not pose a control problem. Principle (K), which licenses control (6.1.2.), is very similar to a lexical redundancy generalization. Its result is an

166

Chapter 10

instance of feature values that matches unmarked (I). An exemplary list of verbs in this class is in 6.1.1. The unique availability of try and V, and of try V-ing, the latter with an interpretation nearly as the sole V-ing-form, support the account stated as (K). The analysis is valid independently of whether we choose the VP-internal subject hypothesis as (11) (Baltin 1995: 240) or a traditional one. (11) I tried [C″[C′[C∅][I″[D″ PRO][I′[Ito][V″[D″ t][V′ be polite]]]]]] Idiomatically, in this class again, the controller-DP can be marked dative. (12)G

(Rechtzeitig PROi da zu sein ) gelang

(ihm i) nicht [dative] not late to be succeeded him not ‘He didn’t succeed in not being late.’ (Bierwisch’s example)

Without ihm, the understood object DP, syntactically overt as controlling pro, has a discourse-related interpretation. A particular, not fully conclusive discussion has dealt with impersonal passive main clauses and their overt or covert adjunct, prepositional or (in Slavic languages) Instrumental Case DP’s, as the source of the feature transfer (K). Independent, not fully clear parameterized variation influences licensing of control. Subtle judgments in English as in other languages discussed are in order (section 6.1.3., 6.1.4., 6.2.). 10.3.3.2. A related Control process of evaluating and attitudinal Control elements The control mechanism of the try-class is intrinsically akin to a family of adjectives and verbs exercising “self-control” as a lexically redundant process (section 8). Adjectival predicates can be divided into two subclasses according to their control behavior: (a) The participation or involvement of the referent of PRO in the event described in the controlled clause calls forth the — evaluated — state s/he experiences in taking part in the event. PRO effects being identified with the matrix “controller”, namely the Experiencer-DP. (13) It is fun (for mei) (PROi to play basket-ball) (14) It is frightening (PROi to be visited ti by a werewolf)

Summary and conclusions

167

(15) ((PROi to be visited ti by a werewolf) would be frightening) It would trouble himi (PROi to be interrupted ti by questioning) The experiencer-bound control relation in type (a) obliterates the distinction between control in object and subject sentences (section 2.2.2.1.). The thetamarked CP-argument can be moved to subject position. Independently of whether syntactic status is given to an implicit experiencer in the matrix clause (section 8.1.1.1.), the control relation is, as it were, reversed, since participation or involvement in something includes experiencing the psychological or other affects, if any, it has on the involved person(s). If other persons are affected, control structures of course do not occur (example 309). (b) In subclass (b) (8.1.1.2.), the evaluating predicate does not concern a state that is engendered by participating in the event described in the object or subject clause. Rather, the decision, the way of acting that is chosen, as described in the infinitival clause, is characterized with the evaluating predicate transferring to the (responsible) agent. The automatic control mechanism of (a) recurs in (b), though of course differently, but still in the “reverse” direction since the property attributed to the DP-referent in the main clause follows on associating the referent of PRO with the action that is characterized. Syntactically, however, the individual to whom the evaluation is related can be overtly accomodated only in the matrix clause, mostly by a prepositional phrase. (Examples 317, 321). The characterization by the main predicate itself may lexically imply a referential extension to a whole species. (17) (PRO to err ) is human. Or see example (313). Note that the application of (K) with respect to predicates of types (b) has to be restricted by (L) (8.1.1.2.), which accounts for the intrinsic interrelationship of the control device in different lexical control classes. 10.3.3.3 Verbal predicates indifferent to thematic feature constraints These are mostly “psychological” predicates as like, hate, please, fear and cross-linguistic counterparts. Control by the Experiencer-DP, be it in subject or object (direct or oblique) position, may still display subtle sensitiveness to thematic and semantic properties. Let us take want, for example, which

168

Chapter 10

reveals great freedom from thematic restrictions, belonging to the group of verbs under section 8.2.2. which are at once ECM-verbs and control verbs, a property for which a certain thematic indifference is a precondition. Such verbs are want, like, hate, expect, prefer, choose, and others. to kiss Mary  (18) John i wants PRO i   to be kissed by Mary   (19) Hei wants PROi to live to a great age (20) Hei wants PROi to arrive early interest  (21) *He i wants PRO i to   Tom irritate  Th. Stroik (1994) is right in ascribing unacceptable (21) to the THEME reading of PRO as an argument of PSYCH verbs. (22) and (23) are further examples from him.

(22) *(PRO to frighten Bill) would be a remarkable feat (23) *It is impossible (PRO to concern Sue) In (18)–(20) intentional behavior or action exerting influence upon attaining what is wanted may be inherent in the respective events described in the clausal complements, but it is not in (21)–(23). (22) borders on lexicalthematic contradiction, a sortal incongruousness subsisting between feat and frighten, again on the THEME-reading of PRO. (23) is an instance of evaluation in the spirit of (b) which is subject to a control mechanism presuming that there is someone to whom the impossibility is relativized and who is held responsible for what is done or happens. But nobody can be held responsible for, much less capable of, concerning somebody. Thematic incompatibility rules (23) out.

10.4. Thematic and nonthematic Control. 10.4.1. Two particular verbs. 10.4.1.1 The German control verb vorschlagen A well-known puzzle of complement control should not be omitted. I will discuss it in the light of the verb vorschlagen. Its infinitival complement contains the description of an event which, for varying reasons, is considered worth realizing. What is at issue is the range of the referents of PRO. Basi-

Summary and conclusions

169

cally, vorschlagen is a verb of control that belongs under control class (B, b) (= II). The application of the pertinent constraint, however, needs refinement. The oblique (dative) argument, the Addressee argument of the event of proposing, is (one of) the controller(s) whenever the corresponding DP is lexically overt. The strength of the probability that the proposing person will include herself into the participants of the proposed event or action is fully discourse-related depending on the pragmatic situation. Even third persons may be viewed as participant(s) of the proposed event, but still the addressee’s consent or agreement about what is proposed is requested. Other persons may come into question again depending on the discourse situation. (24)G

Ichi schlage vor (PROi,X I propose, suggest während du das Zimmer while you the room

in der Bibliothek zu arbeiten, in the library to work aufräumst) are tidying up

In German, the occurence of reflexive versus logophoric anaphors in the complement clause may clear or disambiguate the possibilities of coreference (Kamp/Roßdeutscher 1994). The issue of varying pronouns and their possible help in interpreting coreference is beyond the scope of this study. 10.4.1.2. agree as control verb The way of incorporating the content of the infinitival complement in the sentence as it is dictated by agree determines the range of its control properties. Agreeing with a decision to do what is described in the infinitival clause is independent of participating in the action under discussion. The person(s) expressing agreement may be involved, take part in it or not. Their status of course must be such that it is necessary or desirable to be in agreement with them. (25) Hei agreed (PROj to let me go) (i = j, or i ≠ j) (26) Wei agreed (PROi to leave at once.) 10.4.2. Control in infinitival adjuncts introduced by ohne (without) In his paper “Nachdenken über PRO” (1993), M. Bierwisch brought up the question of control in German infinitival adjuncts introduced by the preposition ohne. Such clausal modifiers correspond to English without V-ing except

170

Chapter 10

that in English lexically overt subjects are available. What conditions on control operate in the particular case of ohne-modifiers? The constraints that handle complement control are not applicable, requirements of thematic feature specifications would clearly be too strong. The event argument of the main clause can be related to the ohne-modifier by a (partial) matching function M. (cf. Rothstein 1995: 522). Since ohne is equivalent to NOT, negating the existence of an event e′ — as described in the clausal complement of the preposition — “within” the event e of the whole sentence would require a PRO argument or -quasiargument to be selectionally compatible with the main predicate. The predicates must have a semantic selectional intersection. Thus, Bierwisch’s (27) is acceptable, because baden and … kalt werden (become cold) share the selectional feature [+ human]. (27)G

Er badet ohne kalt zu werden anzufangen he is bathingwithout cold to become to begin ‘He is bathing without him beginning to get cold.’

Note that in (27) 〈intact, +〉 is valid for the controller only. (28)G

*Er badet ohne zu regnen anzufangen he is bathingwithout to rain to begin ‘He is bathing without it beginning to rain.’

The external argument of baden and the subject quasiargument of regnen do not share selectional features. The requirement of selectional intersection of the partners in the control relation is certainly too weak. First, I assume that the external argument or the moved argument-DP in subject position are clearly qualified as potential controllers. The demoted DP carrying the Agent role may function only as a more or less marginal controller. In (29), for example, the overt or implicit Agent-DP (Ag(e)) is a very questionable controller even when Principle B of binding excludes subject control. See (29). (29)G ?* Eri ist von [Spezialisten]j operiert worden he has by specialists operated upon been (ohne(PROi, j ihn(*i) zu anästhesieren without him to anaesthetize ‘… without anaesthetizing him.’ If we avoid violating Principle B, the situation does not much improve, neither in favor of subject nor towards agent control. Witness (30).

Summary and conclusions (30)G

171

Eri wurde (von Spezialisten)j befreit (ohne he was by specialists freed without (PRO(j),i jemanden [acc.] zu verletzen)) anybody to injure

(31), on the other hand, is perfectly acceptable. (31)G

Eri wurde operiert (ohne (PROi anästhesiert zu werden he was operated uponwithout anaesthetized to be. ‘…without being anaesthetized.’

The problem needs further investigation, but it seems obvious that contrary thematic specifications or a distance between them still to be analysed in detail are an impediment to control relations in ohne-modifiers as in other cases. Parallel syntactic hierarchy — as in (31) — may come into play in case of passive matrix clauses.

10.5. Closing remarks Subject to determining restrictive conditions, control goes to its limits and “experiments” with pushing them back. What we call control is an integral part of a most elaborate complex that provides for the inherent economy of linguistic structures in particular at the interface between morphosyntax and full interpretation. Control is not only a result of selecting infinitival, participial, small clause, gerund, and derived nominal structures, but an objective of selecting them as well. Control is naturally as local as possible. “… it appears that PRO searches for a possible antecedent within its own clause, and if it can’t find one there, looks outside.” (Chomsky 1981: 78) With respect to infinitival complements and subject sentences, where can it look? The search is on if no lexically visible local antecedent can be found. In this case, partly depending on the null object parameter value, a syntactically overt pro or a lexically satisfied (object) argument (e.g. German anordnen, Russian rasporjadit’sja (section 4.6.) may implement control. If, in the local environment, no antecedent is found and, looking beyond it, no discourse-related antecedent is obviously or logophorically (Williams 1992: 299) recoverable, a situation arises often dubbed “arbitrary control”, clearly rather an expedient term. Not surprisingly, Reinhart/Grodzinsky (1993: 73) state: “We ignore here the issue of the precise

172

Chapter 10

interpretation of arbitrary PRO.” A theory under the unifying notion of arbitrariness is not likely to materialize. PRO is able to search and choose its reference freely from the discourse, and PRO can (or must) inherit, or is involved in, the problem of interpreting the reference of an empty argument or implicit adjunct in the main clause without leaving any problem of its own. The economy-oriented general device of control relies on, and exploits, redundant lexical information, in particular argument structure, semantic incorporation of the infinitival complement, and theta grids. Heim, Lasnik, May (1991: 79) have concluded that “it would seem then that unless we build the theory of control (redundantly) into the semantics of these sentences (reciprocal structures, RR), we are left without a satisfactory account of their interpretation.” The conclusion fits the basic claims advanced in this description of control. The available lexical-thematic and projected semantic information has to be so arrayed that it can be put into effect for determining control. The constraints and partly parameterized principles which are meant to do this work could be considered efficient and wide-ranging enough only if they proved capable of capturing the interaction and trade-off between grammatical factors and conceptual-pragmatic construal which supports building up the inherent economy of linguistic structures. The apparatus of constraints, specifications on their application, and interface parameters expressing and realizing the idea that pragmatics could be parameterized, have proven capable of capturing (a) major regularities and conditions of control, (b) cross-linguistic variation in control and (c) the crosslinguistically distinct interaction between lexical-thematic properties and conceptual-pragmatic construal in control processes. In other words, the constraints and parameters are suitable and formally appropriate for being referred to in accounting for the interplay between grammar and conceptualpragmatic factors. Thus, entailments, as a straightforward case of lexical effects, display complicity in control. Construal by entailments is intrinsically linked with control. Trying to separate them, Larson (1991: 129) had to admit: “One important question that I must leave unresolved here, but which must be addressed ultimately … is when precisely an infinitival complement is subject to control versus construal by entailments.” I have argued (section 5.5.) that putting the question like this confuses the issue. My approach clearly assumes that PRO exists as a syntactic entity ° (Ruzicka 1994). Denying its syntactic status would complicate the implementation of the descriptive apparatus but not invalidate the theory and its substantiation.

Summary and conclusions

173

The Case of PRO has been under discussion (see in particular Chomsky/ Lasnik 1993). The results remain inconclusive: “The state of PRO with respect to Case marking and visibility is problematic … the problem arises in particular in Chomsky’s (1981, 1986b) theory with respect to what constitutes a visible head of a chain … Either PRO is directly stipulated to be visible, or it is stipulated to receive some suitably abstract type of Case, so that visibility follows on standard Case-marking grounds.” (Manzini 1992: 144). Pesetsky (1993: 558) leaves open the question “… whether quirky zero Case is the same as null Case for PRO recently discussed by Chomsky and Lasnik.” Avoiding a theory-internal stipulation, a solution based on empirical evidence might receive support, for example, from Icelandic and Slavic languages in which “… PRO may be not only Case-marked, but in fact marked with one of the standard Cases rather then a special ‘null Case’”. (Baltin 1995: 242) Baltin points out Sigurdsson’s (1991) work on Icelandic. In Russian, e.g., the subject of infinitivals is marked standard Dative Case. If PRO receives a special sort of Case, which is taken to be peculiar to subjects of infinitives, one theory-internal advantage of this account is — considering that disjunctive statements are not desirable — that “… a disjunctive requirement that A-chains must be headed by Case or PRO may be dispensed with for one that says that A-chains must be headed by Case.” (Safir 1996: 314) In a recent publication on Control, Petter (1998: 244) endorses Chomsky/Lasnik’s (1993) assumption that “… if PRO is syntactically represented it must also be assigned structural Case.”, but she specifies Case assignment quite differently: “… we may conclude that PRO is assigned structural nominative Case … in both non-finite and verbless clauses a(n abstract) Tense head assigns structural nominative Case to the subject in its specifier position.” (1998: 245). The conclusion is mainly supported by Icelandic data which “… show that predicate agreement only occurs when the predicate is related to a subject with structural nominative Case … this type of agreement also occurs when the subject is PRO.” (245) I perfectly agree with Pepper about drawing on the nominative as a likely candidate for PRO. Still, language particular and structural requirements seem to modify and restrict possible generalizations. Thus, as mentioned above, in Russian, infinitivals strongly suggest structural Dative Case for PRO, whereas for PRO in participle adjuncts and verbless constructions which are closely related to them structural nominative is the ° appropriate Case. (cf. Ruzicka 1982) Case-marking of PRO, which never is an expletive, does not affect the

174

Chapter 10

central aspects of control discussed in this study. Nor does the categorization of PRO. I would like to regard PRO as a pronoun like pro, which seems to come nearest to the truth considering the available theoretical and in particular conceptual tools. PRO can be an instance of coreference anaphora and it can be a case of bound variable (Heim, Lasnik, May 1991: 99). “It can enter into anaphoric relations without binding.” (Reinhart/Grodzinsky 1993: 73) PRO is replaced by pro, which takes over control in noninfinitival complements. The need for control principles and constraints, their interaction with syntax and pragmatics, does not depend on the final categorization of PRO.

Chapter 11

Turning to the Minimalist Program

11.1. On the status of control It is worth considering how the minimalist program may be assumed to accomodate the facts and theoretical needs of complement control as seen in this study. I presume that the actual elaboration of the minimalist program will continue attending to control problems, which would seem to belong to “… the huge mass of constructions for which no principled explanation exists …” (Chomsky 1995, fns. 22 and 93). What is subsumed under the notion of control hinges on the presence of a conceptual-thematical argument which has no phonetic content and corresponds to a D(P) ending up in subject position. Since it must be interpretable for a derivation to converge, it is identified with a D(P), at least syntactically overt in the immediately superordinate (matrix) clause, or with an antecedent someway discourse-related (“arbitrary control”). It has been considered theoretically expedient and suitable to posit in syntax the existence of a phonetically empty argumental D(P), named PRO, which is the controllee, controlled by identification. Resolving the substantive issues of control does not hinge on maintaining PRO or dispensing with it, That is not to say that the introduction or omission of PRO is a negligible problem. Obviously, considerations of economy would urge theorists to try and prefer derivations that can do without positing PRO. Indeed, syntactically well-founded proposals for parting with the use of PRO have been put forward by IJbema/Abraham (1998), Manzini/Roussou (1997), and, before the minimalist phase, by Bierwisch (1993). Advances in this direction, of course, will involve considering abandonment of PRO elsewhere, e.g. in free adjuncts, absolute constructions, versions of small clause and secondary predication (cf. Legendre (1997; 48–62)).

176

Chapter 11

11.2. Derivational and inherent economy In the area of control, as in some others, economy of derivation (or representation), which is laid down as a guideline of the minimalist program, meets inherent linguistic economy. The latter manifests itself in the relevant structures by economising on phonetic realization of the “controlled” D(P) argument. As regards economy of derivation, which I take to involve conceptual naturalness, nonredundancy and empirical adequacy, what is at issue is the necessity for the controlled entity to be available at the LF interface. Methodologically, it seems suitable to try to balance endeavours to achieve economy of derivation and the richness of the apparatus which is thought necessary to capture substantive conditions of control conceptually and empirically. Positing PRO — as I have done in this study — or not positing it, is a pertinent issue. Clearly, alternatives to using PRO would have to present analyses to which control conditions are amenable and which, desirably, are required independently.

11.3. Thematic features In this study, I capitalize on thematic feature specifications and values. Control arises from the thematical status which the control-related arguments have in their respective event structures. A designated controller is licensed if its thematic feature value matches that of the controllee in such a way that identification follows. Given that the control verb has a clausal or VPargument (VP-internal subject hypothesis) in which the entity subject to control in turn is an argument, it is to be expected that eligibility, or failure to be eligible, as controller is dependent on the thematical feature value of the controllee. On this approach, what is at the core of control at the LF-interface, is the predictability of the entity which enters into successful partnership with the controlled element, PRO on standard assumptions. How to cope with the problem of syntactically relating the entities is a theoretic sequel to the prior issue. Various theory internal solutions are available.

Turning to the Minimalist Program

177

11.4. Control treated with the PRO-theoreme 11.4.1. Conceptual considerations Chomsky (1982, 1995) takes the controllee to be represented as PRO, which is of category D(P), has semantic properties and, thus, “… may be the locus of what is loosely called ‘referentiality’” (Chomsky 1995:240). PRO is also posited as a consequence of the relevant head assigning null case to the subject (Chomsky 1995:345). (For some discussion of the notorious Case problem of PRO, see above, 10.5.) Though, with the minimalist program, D-structure is abandoned, arguments continue to be inserted into their canonical D-structure positions, now by merge. In these positions, they are assigned thematic characterizations. PRO is invariably an argument (Brody 1993) and head of a (singleton) chain in subject position. The controller, head of a chain as well, may occupy subject or (possibly oblique) object position. Constraints on control relations are amenable to analyses which use checking of feature values. The information necessary for licensing the controller must be available in the numeration. Assume that (meta)features which are instantiated on control verbs can be as simple as a and ~ a. a reads: the value of the relevant feature on the controller must be identical with that on PRO. ~ a reads: the feature value on the controller must be the negation of the feature value on PRO. How these tags work, I would like to illustrate on three examples. I repeat (256), (270). (1) (256) (270)

John tried to PRO run. Karin tried to PRO be elected. Bill tried to PRO resemble Harry.

Assume that John and PRO in (1) carry identical values of 〈F〉, which is a natural consequence of intrinsic lexical properties of try (cf. 6.1.1). I am setting aside the question how and when either subject is merged into the derivation. Note that the subject position of PRO is reached before spell-out, e.g. after movement in passive. Now let us look at (256): Since PRO receives 〈F,–〉 by virtue of the predicate structure of elect, and 〈F,+〉 by (K), contrary values of the same feature clash on PRO. But of course 〈F,+〉 is available also in the theta grid or, alternatively, in the predicate structure of elect. A crucial assumption is that the presence of 〈F,+〉 in the predicate structure of the complement, wherever it emerges syntactically, e.g. on the argumental pas-

178

Chapter 11

sive suffix (cf. Baker/Johnson/Roberts (1989)), has the effect of causing 〈F,+〉 to prevail over 〈F,–〉 on PRO, which results in a being satisfied. In other words, PRO receives 〈F,–〉 in object position and takes it to subject position by passive movement. Here the contrary value 〈F,+〉, for which try is responsible, is joined to it. PRO, on which two competing values meet, benefits from the internal presence of 〈F,+〉, e.g. in passive structures, to establish dominance of 〈F,+〉, which is a natural general characteristic of try-event structures. The relationship can be taken as a necessary condition: 〈F,–〉 is available on PRO only if 〈F,+〉 is present in the same predicate structure (see (D) in 3.2.2.). I assume that grammaticality (or acceptability) judgments are less straightforward (see, e.g. (267)) if a is satisfied via “strengthening” 〈F,+〉 on PRO. This will become clearer on discussing (270). PRO in (270) receives 〈F,+〉 on the strength of intrinsic lexical properties of try implemented by (K) via an appropriate checking configuration, but — in accordance with the above condition — does not receive 〈F,–〉. Whatever feature is discharged on PRO by the predicate resemble, the conceptual distance between it and the properly interpreted feature value 〈F,+〉 is relevant to judgments on (270), which have ranged from “anything goes” (Vanden Wyngaerd1987: 75) to “ungrammatical” (Lasnik/Fiengo 1974: 553). Still, we are not faced with a grammatical decision. Positing the cooccurrence of the two features on PRO is all grammar can do. Readiness to construe a conceptual-pragmatic interpretation and contextual elements (see ex. (277), (278)) decide on (degrees of) acceptability. 11.4.2. Syntactic treatment As an alternative to (K) (6.1.2), a grammatical procedure is feasible in which the feature that PRO has received from (the predicate structure of) resemble via an appropriate checking configuration is LF-attracted — without “pied piping” — by the matrix subject which of course carries the feature 〈F, +〉 assigned by try. Two distinct features between which the conceptual-pragmatic trade-off takes place, are assembled on the matrix subject. Clearly, the double presence of the respective features entails the identification of the two DP’s and control as its byproduct. The quasimodal contingency in which predicates of the try-class implicate the event described by its complement strongly suggests abstract verbal incorporation of the complement verb into try. The intrinsic semantic relationships between matrix verb and its complement would become conspicuous in syntax.

Turning to the Minimalist Program

179

11.4.3. A different class of control verbs Just as try and the like (6.1.1.) are not exactly verbs of subject control (cf. (284)–(289), so the subclass of ask, beg, etc. (cf. 10.3.1.1.) does not classify object control verbs either. Criteria for the latter are coherent with those of the former in the logic of control. Whatever judgement is made on (270) entails the same judgement on (2). (2)

?*John

asked Bill to PRO resemble Harry.

Nevertheless, the situation changes a lot with verbs of this subclass. Most importantly, they are tagged in the lexicon with the (meta)feature ~α. Recall that the feature values are distributed to PRO and to arguments in the immediately superordinate part of the sentence. Arguments of two distinct argument structures can enter a principled control relationship since the argument structure of the complement predicate constitutes a clausal argument of the matrix predicate. Metafeature ~α, with α ranging over values of a properly interpreted (intentional action or originator (Manzini/Roussou 1997)) feature F, projects lexical properties of the ask-class on basic conditions of (un)marked control. (2) is gibberish or made acceptable by pragmatic construal on equal terms with (270): constraint ~α would require that the controller NP have the reverse feature value to that on PRO, a requirement which no more can be satisfied than constraint a in (270). Neither John nor Bill in (2), can take a feature value reverse to that of PRO, since they have 〈F, +〉 and 〈F,–〉 respectively, and PRO is marked neither 〈F,+〉 nor 〈F,–〉. Recall that the latter is available depending on the former. Thus, PRO cannot have 〈F,–〉 and be controlled by virtue of 〈F, +〉 on [NP John]. PRO in derived subject position after passive movement carries feature value 〈F,–〉 with ~α of ask selecting the DP-subject in situ as proper controller. (3) The injured forwardi asked the coach to PROi be substituted at halftime. Recall the hierarchical restriction (F) on ~α (ex. (75), (82); 3.2.2) (4) *The coachi was asked ti by Peterj to PROj be included in the select team.

180

Chapter 11

11.4.4. Markedness Conditions a and ~α — which are abbreviations of (I) and (II) in (10.2.) respectively — are refined by distinguishing between unmarked and marked application. The former takes α with value 〈+〉, the latter takes α = 〈–〉. Thus, (3) is an instance of marked application of ~α: Since PRO carries 〈F,–〉, that is, α = 〈–〉, by ~α the controller must be marked ~α = ~~ = +, which [the injured forward]i satisfies. The distinction is no stipulation. The substantive thematical underpinning of marked control is less readily available (cf. 3.2.3). There is more motivation for the distinction (cf. next section).

11.5. Can pragmatics be parametrized? Marked application of both checking conditions, abbreviated as a and ~α, is set apart not only for capturing actual facts of control but also for indicating a borderline between “regular” control behaviour and control behaviour that is amenable to conceptual-pragmatic interference with the grammatical constraints. Marked application seems to be a precondition for conceptual-pragmatic handling of control configurations inasmuch as it construes them as instances of marked control. A particularly illustrative example is (88) (repeated here). (88)

María le pidió a Juan hablar con los muchachos María him asked to Juan to speak to the boys a. ‘María asked Juan to speak to the boys’ b. ‘María asked permission of Juan to speak to the boys’

Chomsky’s (1988) paraphrase (a) and (b) show that (88) is amenable to unmarked (a) and (b) marked application of ~α. As an instance of marked application, an intermediate conceptual structure is created which in its turn satisfies ~α, licensing the subject María to control. Since [NP María] is marked 〈F, +〉, PRO is required to have 〈F,–〉, which it can have received only with conceptual-pragmatic assistance. The proper structure accessible to, and satisfying, marked ~α can be intuitively elucidated — as Chomsky did — by “inserting” permission of ( b) as a bridge to the bare infinitival. The construal, which leads to a structure that can be referred to by grammatical notions, interfaces LF with cognitive modules by meeting interpretability conditions.

Turning to the Minimalist Program

181

When I resort to conceptual-pragmatic areas, I am not thinking of general processes of inference, situation- or discourse-related specification and refinements of interpretation. What is at stake here has a more general character. Concerning (88), I would like to call it modality switch, expressible by a deontic modality operator. Otherwise it is hard to explain why options, as in ( b) or in (86a), are not available, e.g., in German or Russian. It is arguable that we have to do with a parameterization of pragmatics: Conceptual-pragmatic interference is (not) strong enough to override what seems structurally “fixed” by grammar. Languages which set this conjectural parameter negatively have to use explicit lexical items, e.g., modal verbs like dürfen in German (cf. section 3.3.) to create interpretative counterparts of the pragmatically implemented modal switch. There is further evidence for assuming that pragmatics can be parametrized: The persuade-subclass (sections 4 and 10.3.1.2.) offers another instance of a general notion of grammar being conceptual-pragmatically implemented: languages fixing a negative value of the presumed parameter have to resort to explicit causal verbs like German lassen, Russian podvergnut’ (sebja), while English, which sets the parameter positively, allows clear interpretation of causality without them (cf. section 4.1.). Presumably, a “light verb” cause introduced at the relevant stage of conceptual enrichment of LF-interpretation may account for the implicit change to causal embedding (causality switch). Recall that the persuade-subclass is not amenable to marked application of ~α. Consequently, pragmatic interference targets its unmarked application.

11.6. On approaches that give up postulating PRO 11.6.1. General conditions The approach adopted in the main body of this study is not wedded to the view that numerations to be mapped to control structures which satisfy output conditions at the LF-interface must contain PRO. What I think is needed is an apparatus in which the entities are available on whose properties and interaction control hinges conceptually and empirically and without which an account of control cannot be a natural explanation. In the spirit of the minimalist program, positing as well as dispensing with an entity like PRO should not be the outcome of invoking construction-specific assumptions. If

182

Chapter 11

PRO is given up, it must be abandoned everywhere. IJbema/Abraham (1998) have argued that principled motivations for positing PRO put forward by Chomsky (1982, 34) and Chomsky/Lasnik (1993, passim) are no longer upheld under the minimalist program. More specifically, they propose an ingenious formal analysis in which the German and Dutch infinitival prepositions zu, te respectively raise to [Spec, vmax] and, thus, prevent an external argument from moving to this position. The behaviour of the infinitival preposition is shown to be quite analogous with that of the perfect passive participle morpheme ge-. Since the external theta role, if available, must be discharged, an element must be present to receive it, PRO being excluded. Notwithstanding that “… English to lacks the faculty to suppress the external argument of the infinitive” (Abraham 1997), and, e.g., Slavic languages lack the infinitival preposition, the universal PRO-theoreme is of course involved everywhere if it is weakened or disproved in one language or the other. Thus, whether we can “… take IPrep to sit in the position obtained by PRO …” (Abraham 1997) as in German or Dutch, or not, as in Slavic languages, the loss of PRO needs to be replaced with a new formal solution to the control relation. What is “left” of PRO must be located in the right place. The indispensable rest of PRO is its thematic specification: “the subject-θ raises to the matrix subject or object position.” (Abraham 1997). The crucial issue is which of the two positions in a given instance of control — excluding an adjoined PP– argument as a third possibility — proves to be the proper landing site. Abandoning PRO, we cannot abandon the thematic feature specifications PRO would have to meet if it could retain its place. The relevant feature (values) retrievable from the predicate structure and due to an invisible subject phrase, must be available for either selecting the proper controller phrase or licensing a single matrix DP to control. Identification realized by control is a function of cross-clausal thematic dependencies. We indeed move into “… a new realm of constraints to be stated with respect to the compatibility of matrix and embedded θ-roles …” (Abraham 1997).This is quite in the right ° spirit of earlier work by Abraham (1983), Ruzicka (1983a,b), Nishigauchi (1984), Farkas (1988), Wegener (1989), Köpcke/Panther (1991, 1993). 11.6.2. Control in terms of a calculus of features To investigate and specify the nature of the compatibility constraints has been the main concern of this study. In terms of selectional constraints, the matrix

Turning to the Minimalist Program

183

verb, straightforwardly selecting an infinitival complement, selects the controlling entity, since it determines which condition the respective feature values of the controlled and the controlling element must meet. The distinct constraints, given in the shape of conjunctions of feature values (cf. 10.2.), are associated with subclasses of control verbs. They encompass control behaviour restrictions at LF as well as grammatical-pragmatic interface phenomena. A control verb, when selected for the numeration, carries a (meta)feature marking its membership of a control subclass in the abbreviated form α and ~α. The main classes are divided into subclasses specified and referred to by lexical restrictions. “At the LF interface, it must be possible to access a lexical item LI and its nonphonological properties LF(LI): the semantic properties and the formal properties that are interpreted there.” (Chomsky 1995: 242) As indicated above, there are no other candidates for doing the work PRO has been taken to be responsible for than thematic features. Thus far, Manzini/ Roussou’s (1997) skeleton of an account of control in Minimalistic terms is perfectly consistent with the situation resulting from the loss of PRO. Their calculus of features, as presented in (1997), if highly formalized, does not predict distinct control behaviour and shifting control. It reflects given results of wellformed control relations. In other words, Manzini/Roussou account for structures which subsist after the choice between two potential controlling elements is made or a single DP is licensed to be a suitable controller. They start from the assumption that “… θ-roles are the by-product of the nesting of arguments within the aspectual structure of the predicate, which includes at least two Asp(ect) projections, AspM and AspO, O = Originator, M = Measurer.” For John runs, they suggest a derivation in which “First, the DP is merged in [Spec, IP] … D(P) then attracts Asp in the covert syntax, impelling movement of Asp to an I-adjoined position, so as to create a checking configuration between Asp and D(P). (3)

a. John I [AspO [run]] b. John [AspO -I][tAspO[run]]“ (3 a,b) in Manzini/Roussou’s numbering.

Now, since “… in general, overt DP-movement from θ- to Case position reduces to merger of DP into Case position, with subsequent abstract movement of Asp to the checking domain of D(P) …“, as shown in 3.a.b., Asp is assumed to be susceptible of being detached from association with a D(P) and DP movement. If Asp is ‚hypostatized‘, “… control can be construed simply

184

Chapter 11

as the movement of two distinct Asp’s to the same D(P).” (Manzini/ Roussou 1997: 42) As PRO is not available, the complement Asp, detached from a D(P), joins a matrix Asp in the common landing site. Multiple Aspmovement to I is allowed by Last Resort as a parametric option (Manzini/ Roussou 1997: 42). Control appears to be reduced to independent derivational requirements. Manzini/Roussou treat (5a), for example, as illustrated by their 5a.b.c. (5) a. b. c.

John tried PRO to run John I [AspO1 tried [to[AspO2 run]]] John [AspO2 [ AspO1 — I]] [ tAspO1 tried [to[ tAspO2run]]]

I would like to focus on two intrinsically connected questions which are not raised in Manzini/Rousson’s treatment of control. (a) Which AspX in the predicate structure of the complement is chosen for abstract movement to the pertinent D(P)? In other words, control is derivable only if abstract AspX-movement can be preceded by some equivalent to obligatory A-movement, for example DP-raising in the passive. Since PRO is subject to control as head of a chain, that is, in derived (final) subject position, an equivalent restriction to syntactic ranking must be projected on targeting the AspX to be moved. (b) How is the landing site of abstract AspX-movement predicted, given that there are more than one potential controller-DP’s? This is not the simple question of verbs of subject control and verbs of object control. 11.6.3. An illustration of control relations described without positing PRO I would like to illustrate the situation and eventualities of control behaviour by ex. (6), in which several controlled complements are recursively embedded. (6)

Peter asked John to try to persuade Bill to run again.

Following Manzini/Roussou, I will use Asp(ect) projections to account for the predicate structures of matrix and complement, respectively. Besides AspO (O = originator), at least a second notion AspX must be available in the apparatus to cope with control conditions on Manzini/Roussou’s method. AspX is meant to cover at least [theme, +], [oblique, +], both can be unified as [non [+O]]. I would like to mention before that both feature values of 〈O〉 are required for thematical characterization to serve the particular purpose of

Turning to the Minimalist Program

185

licensing identification under control. Assume that (6) is structured as (7), a simplification sufficient for what I want to demonstrate. (7)

Peteri I[ AspO1[ AspX1 asked Johnj [to1 [ AspO2 [try [to2 [ AspO3 [ AspX2 persuade Billk [to3 [ AspO4 run ]]]]]]]]]]

Proceeding cyclically, the first steps are covert movement of AspO4 to create a checking configuration with the DP Billk, the controller, and doing the same with AspX2 — presumably via a checking configuration in AgrOP — in view of the theme role Bill receives from persuade. Notice that contrary values AspO and AspX (=non [+O]) are checked on Bill. As concerns AspO3 , meant to reflect the external θ-role of persuade, it, apparently, cannot reach a checking configuration with a subject-DP, because there is none. We could consider adjoining AspO3 to to2, as absorbing or binding the external subject role in the spirit of Abraham (1997). to2 may serve as an intermediate position for AspO3 before it raises to be adjoined to AspO2. Note that the latter is responsible for controlling the external θ-role of persuade. Neither AspO3 nor AspO2 can get into a checking configuration with a DP-subject of try. Again we could consider moving [AspO3 [AspO2]] to to1, construing a checking configuration. The chances of doing this are better in German and Dutch than in English or Slavic languages, because “… English to lacks the faculty to supress the external argument of the infinitive” (Abraham 1997: 7) or, presumably, not bind it either. In Slavic languages, an infinitival preposition is not available. Assume that AspO3 and AspO2 are harboured somewhere in the [VP … try …] cycle, that is, are not left in the respective predicate structures. Collapsible to AspO2, they still have to do more control work, raising to create a checking configuration with D(P) Johnj, in which they will meet with feature (value) [non [+O]] = AspX1 contrary to its own. But why does AspO2 — collapsed from [AspO3 [AspO2]] — not raise to a checking configuration with the subject D(P) Peteri, joining with AspO1 instead of John. Peteri is to be placed in a checking configuration with AspO1, of course, as Johnj with AspX1. This is independent of control. What is relevant to control is the prediction that AspO2 must reach the landing site [DP John] to join with AspX1 and fix Johnj as controller of the external subject or external θ-role of try. The choice is induced by specific selectional properties of ask, which cross the complement boundary and must be available in the numeration into which ask has entered. The relevant information, carrried, for example, by the ask-subclass, can be conveyed, as proposed above, by the (meta)feature ~α, which reads as the

186

Chapter 11

thematic feature value of the controlling element is contrary to that of the controllee. The junction of AspO2 with AspX1 meets the demand in (7). Consider again example (3). (3)

The injured forwardi asked the coach to (PROi) be substituted at half-time.

The structure relevant to control on Manzini/Roussou’s approach could be (3′). (3′)

The injured forward I [ AspO [AspnonO [asked the coach[to [ AspnonO be substituted]]]]]

Matrix AspO and AspnonO are moved to create checking configurations with the injured forward and the coach respectively. It would be unknown whether AspnonO in the complement has to move to a checking configuration with The injured forward or with the coach if it were not for the metafeature ~α, which characterizes the control subclass to which ask belongs. 11.6.4. Dispensing with PRO and pro If, in addition to the PRO-theoreme, the pro-theoreme is abandoned, together with the null object and Pro-drop parameter, further provisions for a PRO-less control theory are needed. The distinction, for example, between syntactically overt null objects that do control, as in Russian or German, and syntactically “inert”null objects that do not control would have to be projected onto the control mechanism of Asp-projections. In general, it depends of course on the elaboration of the minimalist program and in particular of Manzini/Roussou’s approach to control whether it will accomodate control conditions in a natural, conceptually and empirically satisfying way.

Notes

1.

Stump (1985) has documented and interpreted grammatical (including semantical) restrictions on the variability of the logical relations between free adjuncts or absolutes and their respective matrix clauses.

2.

“An obvious problem with (20) (our (28), RR) as a sufficient descriptive condition is that, while (20) predicts that both subject and object control are possible in all of (12)–(15) (our (29)–(32)), only object control is possible in (12) and (15), and only subject control is possible in (13) and (14).”

3.

(20) (our (28)) is based on a series of formalizations. The c-domain is defined as follows: ″(1) γ is the c-domain of α iff γ is the minimal maximal category dominating α. (2) α c-commands β iff the minimal maximal category dominating α dominates β. (3) α governs β iff a. α is a lexical category, and b. α and β c-command each other.” Manzini defines a domain-governing-category for an element as in (i): ″(i) (her (31), RR) γ is a domain-governing category for α iff a. γ is a governing category for the c-domain of α, and b. γ contains a subject accessible to α. Further,” … PROs like NP-traces are pure anaphors … ″(ii) (her (32), RR) An anaphor without a governing category is bound in its domaingoverning category.” (p. 424) The examples with PRO in object sentences as in (our) (29)–(32) and (33)–(36), involve configuration (iii) (her (33)): (iii) S

NP

Infl

V

VP

S′ S

PRO

188

Notes “In (33) the c-domain of PRO is S′, since S′ is the minimal maximal category dominating PRO. Further, the governing category for the c-domain of PRO is S; for S is the minimal category containing S′, a governor for S′ (V), and a subject accessible to S′ (NP or Agr). Finally, S, the governing category for the c-domain of PRO, contains a subject accessible to PRO (NP or Agr again); for NP and Agr c-command PRO and coindexing of NP or Agr with PRO does not violate the i-within-i condition. Hence, by (31) (our (i), RR), S is the domain-governing category of PRO; and by (32) (our (ii)), PRO is correctly predicated to be bound in S.” (Manzini, l.c. pp. 421, 424, 425)

4.

Such cases are illustrated by (i)–(iv) (Manzini’s (26)–(29): (i) [PRO to behave oneself in public] would help Bill (ii) [PRO to behave himself in public] would help Bill (iii) Mary knows that [PRO to behave herself in public] would help Bill (iv) [PRO to behave himself in public] would help Bill’s development “The PRO can have arbitrary reference as in (i); or it can corefer into S, as in (ii), into a phrase superordinate to S, as in (iii), or into a phrase subordinate to S, as in (iv).” (Manzini, l.c.: 424)

5.

“In (34) (our (38)), as in (33) (= (iii) in fn. 3), the c-domain of PRO is S′ and the governing category for the c-domain of PRO, S′, is S, where the governor and accessible subject for S′ is Agr. But in (34), S, the governing category for the c-domain of PRO, does not contain a subject accessible to PRO; for S′ and Agr c-command PRO, but cosuperscripting of S′ or Agr and PRO would violate the i-within-i condition. Hence, by (31) (= (i) of fn. 3) PRO does not have a domain-governing category; (32) (= (ii) in fn. 3) then does not apply, and PRO is correctly predicted to (co)refer freely.” (l.c.: 425)

6.

The account of the Italian-French contrast with respect to control in the presence of se/si (Gianni non sa se andare el cinema, ‘Gianni NEG knows if to go to the movies’ vs French *Marie ne sait pas si aller au cinema) hinges on Kayne’s assumption “… that in the infinitive-clitic languages like Italian, the infinitive will in the general case move into a position that is hierarchically closer to PRO — left adjoined to the l′ — than the position it moves into in the clitic-infinitive languages. I would like to propose now, that in so doing the infinitive in (87) = (… se … [IP PRO … [l′ Vinf + [l’ … (Cl+) l …) blocks off government of PRO by C0 and thereby eliminates the potential PRO theorem violation induced by that C0.” (Kayne, 1991: 674)). Italian se, thus, does not govern PRO, because Vinf is a closer governor, with the notion of Minimality slightly modified. The PRO Theorem is not violated.

7.

(c) includes Farkas’s (1) and (4) of “… the main questions any theory of control has to address … (1) What are the principles which determine the set of possible controllers? (4) What determines the choice of the controller in case the set of possible controllers is of cardinality greater then 1?” (Farkas, 1988: 27, 28))

8.

Here is a collection of pertinent statements. Control theory “… crucially involves θ-grids … Control is a syntactic process, consisting of the coindexing of PRO with the designated θ-role; therefore, the designated θ -role must be visible in the syntax, when the control module applies.” (Rizzi, 1986: 552). “The system of thematic relations … forms an important basis for a significant class of control phenomena.” (Nishigauchi, 1984: 217). Bresnan (1982: 404) assumes “… thematic constraints on anaphoric control …”. A particular treatment of control in the spirit of thematic relations has been advanced by Jackendoff (1987: 371 ff.): “All of this is by way of justifying the introduction of

Notes

189

thematic relations as part of the account of control … The fundamental point, from which all else proceeds, is that thematic relations are part of a level of semantic/conceptual structure, not part of syntax.” Interestingly, Williams (1992: 316) releases PRO from being the target of identification in adjunct control: “The adjunct control cases stand apart as cases that do not involve PRO, but rather direct theta role assignment.” This radical step is very revealing. The reason seems to be that in adjunct control thematic properties of lexical items that determine or influence control cannot be systematized and be marshalled into generalized superordinate contraints. They rather must operate on their own considering characteristics of individual events, deictically or logophorically fixed understood reference. Recall the discussion of examples (15)–(16), or consider Higginbotham’s (1992: 105) example (80) (our (i)), which he adds to Bach’s example (ii): (i) Is that a book [PRO to read to each other]? (ii) Here’s a book to read to each other. Higginbotham comments on both examples: “We have seen that ‘pragmatic control’ of PRO is permitted as in Bach’s example (ii). But is it PRO itself that is pragmatically controlled, or rather an understood benefactive, which in turn serves as antecedent of PRO? The thrust of my discussion suggests the latter, but, it is not easy to test the question. Even an example such as (80) (our (i), RR) could, it seems, contain a benefactive … containing within the predicate nominal an understood ‘for NP’, with NP serving as antecedent of PRO.” Still, the basic problem of control remains the same. The antecedent of PRO is established by discourse-related construal. The thematic specification which is assigned to PRO by read is of course compatible with benefactive, it even predicts the latter for the understood or overt NP in the PP of the matrix clause. In this situation, we might say that control is reversed with PRO “controlling” the antecedent which it creates and which necessarily matches it thematically. A similar “reversal” of control will come up even in complement control (section 8). Naturally, not the whole community of investigators of control are in unison with a θ– role-driven approach to it. Emonds (1985: 106) tries to accomodate control to principles holding in other modules: “It can thus be concluded that all obligatory control can be reduced to subcategorization and the principles which govern it.” Koster (1984: 431) relies on control theory and on binding theory in accounting for his (31) *John was tried [e to go]: “This is, however, the point where the independent theory of control comes into play … it is this theory that explains (31) …” With respect to the same sentence, he draws on binding: “It is here that the independent binding theory comes into play.” 9.

The above interpretation of “Intentional action” implies the belief that the identified referents of PRO and its licit controller have control over the action described in the complement clause, a characteristic feature figuring in Farkas’ (1988) analysis. Davidson (1989: 46) stresses that “It is a mistake to suppose there is a class of intentional actions: if we took this tack, we should be compelled to say that one and the same action was both intentional and not intentional … Hamlet intentionally kills the man behind the arras, but he does not intentionally kill Polonius.”(p. 46).

10.

This assumption finds support in Døabrowska’s discussion of “Dative and nominative experiencers” (1994: 1029)

11.

Larson (1991: fn. 19) states that “… ask is normally (my emphasis, RR) object-controlled. Passive in its complement appears to license subject construal …” Since subject

190

Notes control in such cases does not fit in with Larson’s straightforward configurational account of control (in which c-commanding and subject control are harmonised by assuming a D-structure that places the inner object (of Double object construction) in a non-c-commanding position relative to the infinitve, and by fixing Controller choice at D-structure with the help of his Minimal Distance Principle, as in Larson’s (1991: 108-122) analysis of the verb promise), Larson removes it from the realm of control altogether: “Suppose then that examples like (ib) John asked Mary to be allowed to leave, like their counterparts with promise (see below 5.5., RR), do not in fact involve control but instead involve construal by ‘tranfer of possession’ entailments’. Then we expect the subject to be associated with the infinitive under (an appropriately modalized form of) the entailment: (iii) X – asks – Y – for – Z ⇒ X gets Z″ (Larson, 1991: 133, fn. 19). I will discuss Larson’s radical step, which he repeats for the recalcitrant behaviour of promise, below (5.5.).

12.

“A predicate phrase must agree with the object of a transitive verb phrase and the subject of an intransitive verb phrase in number, gender, and person.” (Bach, 1979: 520)).

13.

“The German sentence is ambiguous and on one interpretation, subject control, is synonymous with the English sentence …” (Comrie, 1985: 50).

14.

A peculiar but illuminating instance of control is presented by a Greek passage from the gospel to St. John (4, 9) and its respective translations. (i) Πωσ ~ σ υ` ´Ιουδα~ιοσ rν παρ´ ݵου~ πε~ιν αßτε~ισ … ? How is that you a Jew being from (of) me to drink (you) ask ‘How can you, being a Jew, ask me to drink’ (ii) Old Church Slavonic Kako ty ijudei sy ot mene piti prosiši …? (iii) Latin Quo modo tu ludaeus cum sis, bibere a me proscis ? The external argument (συ, you) is the controller. The request δüσ µοι πε~ιν give me to drink) precedes (i). The addressee of the request shows up in the PP παρ ݵου idiomatically linked to the verb αßτÝω (ask) and copied in the translations. Control in (i)–(iii) seems analogous to the English and Spanish constructions discussed above (3.2.5.), which retain full acceptability though violating the marked Constraint (B,b). In some languages, as in Spanish and English, pairing of the interested parts in the ASK-type events is construed by a conceptual operation, which in the given Greek example is modified, even strengthened, to something like (iv): (iv) [… ASK mei [PROi [to give youj [[ something]k [[operatork] [PROj to drink tk]]]]]]

15.

We can take advantage of Dowty’s (1985: 300) entailment patterns (his numbering): ″(10) ∀x∀y∀P [δ(P) (y*) (x*) → γ(y*)] ∀x∀y∀P [δ(P) (y*) (x*) → ζ ([P(y*)])(x*)] “P” is a variable over VP-meanings… For example , if δ=persuade, then an instance of γ(y*) would be “y is an agent capable of forming intentions to act”, and an instance of ζ([P(y*)]) (x*) would be “as a result of x’s action, y comes to intend to act so as to bring P(y*) about.” (10) can be illustrated by the control relation in (i): (i) persuade him i [[PROi to VP …]], Russian … ugovorit’ egoi ((PROi VP)) or by (ii): (ii) Johni was persuaded ti [[PROi to leave]]

Notes

191

〈intact,–〉 is assigned to the tail of the A-chain (Johni ti) and transferred to John. It follows from (10) (δ=persuade), if applied to (ii), that PRO carries a value α of the theta specification distinct from that of the internal argument John. If syntax is to satisfy (10), y must be mapped onto the direct object-NP of the superordinate clause at D-structure. (The referent of) this NP reproduces the relevant identity in (10) by being the target of x’s (implicit in (ii)) actions and the agent intending to act so as to bring about P. In other words, (ii) satisfies um (B,b). 16.

Recall that if the chain headed by PRO has received 〈F,+〉 (〈intact,+〉) itself, which must have been carried along with PRO moved from its departure site to its non-theta-position, it is assigned 〈intact,–〉 by (D) and (E) automatically (see above, the analysis of (64)).

17.

A theta-position is generally permitted to receive multiple theta-roles (cf. Chomsky, 1986b: 97), a license that need not be drawn upon in postgrammatical quasi-θ-assignments.

18.

In Russian, for example, the counterparts of teach have distributional properties different from obešcat’ (promise) (no dative-shift, inherent dative case of the theme-NP) without effect on control. (i)R Neuzeli ja vas dolzen knjaz’, ucit’ delikatnosti [acc pl] [dat sg] Really I you must, prince, teach tact ‘Do I really have to teach you, prince, tact’ Nor do we have in the German counterpart of teach the analogue to how to + infinitive construction. In German, embedded question-CP’s are not infinitival generally. (ii)G *Johannes lehrte ihn, wie zu spielen. John taught him how to play.

19.

Russian obucat’ (teach) has a reflexive partner obucat’sja which is a distinct lexical entry and has subject control. Its English counterpart is to learn: (i)R Obucali egoi ((PROi plavat’)) to swim (They) taught himi (ii)R Oni obucalsja ((PROi plavat’)) He learnt to swim Note that the attached reflexive clitic -sja is not a quasi-trace of movement.

20.

Köpcke/Panther (1991: 162) observe that the verbs empfehlen (recommend) and raten (advise) are not only unmarked with respect to the factor Benefactive for the matrix subject, but even exhibit a morphosyntactic marker (dative), which indicates the role of Benefactive of the matrix object. (“Die Verben empfehlen und raten sind nicht nur unmarkiert hinsichtlich des Faktors ‘Benefiziens’, für das Matrixsubjekt, sondern weisen sogar eine morphosyntaktische Markierung auf (Dativ), die auf die Benefizientenrolle des Matrixobjektes deutet.”)

21.

Discourse-related conceptual construal of a domain of individuals is a paraphrase of this domain being “understood”. A slight modification of the (interpretation of the) projection principle might be necessary: “A consequence of the projection principle is, to put it informally, that if some element is ‘understood’ in a particular position, then it is there in syntactic representation, either as an overt category that is phonetically realized or as an empty category assigned no phonetic form …” (Chomsky 1986b: 84).

192

Notes

22.

The translations (i) and (ii) of (i)’ and (ii)’ respectively, assumed in Gazdar et al. (1985: 203) do not seem to capture the whole spectrum of control of these two verbs and cognate ones. Still, the distinct hierarchical positions of the NP argument which may receive the theta-role goal reveals a basic intuition about the different status the objectNP has in the respective configurations projected by the two verbs. (i) persuade’ (leave’) (us*) (Kim*) (i)′ Kim persuaded us to leave. (ii) promise (us*) (leave) (Kim*) (ii)′ Kim promised us to leave. Control is thought to be captured by (iii): (iii) “The first NP argument to combine with a functor in which a VP occurs is the (semantic) controller of that VP” (Gazdar et al. 1985: 202).

23.

Bresnan (1982, 405) notes: “The first attempt in transformational grammar to explain the deviance of examples like (87) (= (i) b. RR) appears to be that of Jenkins (1972, 200 ff.), who proposed a constraint stating that the object of by cannot be coreferential with an implicit or expressed subject of a complement. (i) a. John promised Mary to be on time. b. *Mary was promised by John to be on time. If this were true, it would itself require explanation, but the following examples show that this by-phrase constraint does not express the correct generalization: (ii) (=(94)) a. John had been promised by Mary that she would meet him at the station. b. John expects a promise by Mary to remain faithful to him. c. An attempt by the gang of four to advance themselves now would be foolhardy. In all of these examples, the object of by is or can be understood as coreferential with the subject of the complement.”

24.

“Minimal Distance Principle (MDP): An infinitive complement of a predicate P selects as its controller the minimal c-commanding noun phrase in the functional complex of P.” (Larson, p. 115)

25.

Modals even do more, their interference in control behaviour is not restricted to a particular control type or class. For example, dürfen can intervene in control relations of counterparts to persuade as in counterparts to ask and promise as above, licensing control in cases in which it cannot be made acceptable pragmatically in English: (i) Scipio persuaded the Senate (PRO to have a free hand) Comrie (1985: 63) comments on his example (61) (= (i)): “… real-world knowledge of Scipio’s relations with the Senate make the only plausible interpretation that Scipio persuaded the Senate to allow him, Scipio, to have a free hand, but once again English syntax disallows this interpretation.” In German, the counterpart of (i) is made grammatical by “inserting” dürfen: (ii) Scipioi überredete den Senat (PROi frei handeln zu dürfen) ‘… to be permitted to have a free hand’. (ii) thus satisfies the m version of Constraint (B,b), but this type of modal interference seems to create the only possibility that persuade-type verbs meet m (B,b) (cf. above 4.1.) In English, which has to choose passive in paraphrasing dürfen, (ii) seems questionable, cf. (iii): (ii) ?*Scipioi persuaded the Senate (PROi to be permitted to have a free hand). (iii) Scipio persuaded the Senate i (PROi to be examined by a commission).

Notes

193

26.

What is implemented in this way compares with independent phenomena observed by Zubizarreta (1982), it is a kind of their directional reverse: “Zubizarreta explains these facts in terms of a projection … the rationale clause projects agency onto an argument in the matrix clause … the subject of the matrix clause receives an agent role from the rationale clause. (i) John was arrested by the police to impress his mother. (ii) Jesus died to save our souls. (iii) The boy fell to deceive his mother. (iv) The train derailed to save the child”. (Roeper 1987: 298), with the examples from Zubizarreta)

27.

This question has not been raised by Grewendorf/Sabel although they continue to claim “… that long scrambling in German is licensed only by members of a special class of verbs. In the following example with the matrix verb vorwerfen ‘accuse’, long-distance scrambling out of the complement clause into the matrix clause is not possible: auszukosten] (i) a. weil keiner diesem Manne [CP PRO den Triumph since nobody this manDAT the triumphACC to-savor vorgeworfen hat accused has ‘since nobody has accused this man of savoring his triumph’) (i) b. *weil [[den Triumph] i keiner diesem Manne [CP PRO ti since the triumphACC nobody this manDAT auszukosten] vorgeworfen hat]” to-savor accused has (i)a., (i)b. = Grewendorf/Sabel’s (37a), (37b) Vorwerfen does not belong, then, in the special lexical class presumed by Grewendorf/ Sabel. Under my approach it is a control verb of the persuade-type subclass that, in German, contains an oblique (dative) object.

28.

Sentence types like (268), (130), in which PRO is bound by a quantifier, support Nishigauchi’s (1984: 237) assumption that “… it seems clear that we need a mechanism to assure that PRO is represented as a variable bound by a quantifier or a lambda operator.” Incidentally, topicalization at S-structure and IP-adjunction at LF of (negated) quantifier phrases, with the subject pronominal (nikto) within the scope of the quantified time adverbial, may end up as conjoined phrases at the top of the sentence at PF, blurring their respective hierarchical and local origins. (i)R Nikto i nikogda ne stremitsja eticeski nobody and never (not) endeavours ethically ocenit’ javlenija prirody. to evaluate phenomena of nature ‘Nobody ever ethically evaluates phenomena of nature.

29.

The “oddness” of (i), (ii) noted by Hornstein and Lightfoot (1987: 33) is a case in point again. (i) John tried to receive the gift. (ii) John tried to know the answer. (iii) is the Russian counterpart of (ii): (iii) Dzon pytalsja znat’ otvet. Prefixing u to znat’ gives the perfective aspect uznat’ (to learn, to get to know), which in its result-oriented interpretation cancels the conceptual conflict. (iv) is readily accepted:

Notes

194 (iv)R Dzon pytalsja uznat’ otvet. John tried to get to know the answer 30.

In Czech and Slovak, grammaticality of impersonal passive structures correlates with the presence of complements of the (passivized) verbal head, quite independently of their morphological case (excluding of course accusative): (i)C Kartám musí být rozumeno. [dat pl] [3 ps sg pres] [infinitive] [part pret pass neuter] (it) cards must be understood ‘One must understand cards, know (how to play) cards.’ The relevant lexical properties may be interpreted as closer relationship and affinity to transitivity and subcategorizing complements, which seems to indicate the attraction and nearness to the typical passive of verbs governing a direct accusative object.

31.

I must note that in Russian the clausal complement of (po)kazat’sja (seem, appear) cannot take infinitival form at any level. Raising is incompatible with the appearance of a verb in the complement at all. Heads of complemental AP’s or NP’s necessarily exhibit inherently predicative morphological Instrumental case, adjectives in addition, may take exclusively their “short” form. “In Czech, infinitival VP’s may occur in the clausal complement of zdáti se (seem) (Svoboda, 1962: 109)). The situation in Russian has led some authors to deny or doubt Raising in Russian (cf. Chvany, 1975: 226 ff.): Our argumentation is not affected by this stance. (i)R *Oteci kazalsja (ti znat’ eto) Father seemed to know it (ii) Kazalos’ (cto otec znaet eto) (it) seemed that Father knows it But compare the Czech counterpart: se zdál (ti to vedet) (iii) Oteci [refl. clit] Father seemed it to know (být) nemocen) (iv)C Petri se zdá (ti [adj. “short” form] Peter seems (to be) sick

32.

This is similar to Icelandic verb agreement. “When the matrix subject in Icelandic is nonnominative, the matrix verb does not agree with the subject … the verb is in the third person singular form.” (Rothstein, 1995: 513, with further literature) Third person, singular, (neuter (in past tense)) are the minimal syntactic features constituting an unmarked, default grammatical form in Slavic languages. There is in Slavic languages a second nonanaphorical generic or “anonymous”, that is, indefinite specific pro-element which is inherently specified [+plural, 3 person, +hu° 1986: 243) man]. (see Ruzicka,

33.

Cinque’s analysis, which follows Chomsky (1986: 217, fn. 120) in assuming “… that si is a clitic in D-structure and that the relation of subject position to si is that of an expletiveargument CHAIN …”, may offer a chance of accounting for the radical difference between (305) and (306). In “transitive” si-structures like (306), the clitic is an argument that occupies the unique theta-position in the CHAIN, thus withholding the external theta-role, and the object moves to a non-theta-position. This position and si constitute the CHAIN assigned nominative Case.

Notes

195

34.

(“a chain is a CHAIN …” (following Chomsky, 1986: 132)). Baker et al. (1989: 225) use a notion of “chain” slightly different from Chomsky (1986: 132 ff.).

35.

Borer’s (1986, 395) analysis is similar to Epstein’s: “Though Epstein assumes that that operator (the operator that binds the infinitival [NP, S] position, RR) is specifically a small pro …, I will simply assume that it is an abstract operator, the precise nature of which requires further investigation. I will further deviate from Epstein’s analysis in assuming that the movement involved is syntactic (rather than LF) and that it adjoins the null arbitrary operator to S. Further adapting Epstein’s proposal to the system proposed here, I will assume that the abstract operator A-binds the anaphoric infinitival Agr and that its index is thus transmitted to the embedded PRO … This S-structure representation is thus given in (36) (Borer’s numbering, RR): (36) OPERATORi (it is not easy [e]i (S′ [e]i Agri to solve this problem))” Differing from Epstein, who treats pro as a universal quantifier, Borer associates a set interpretation with the null operator: given a set S, it is not easy for a member/members of S to solve this problem. (1986: 395)

36.

A particular analysis, which I will not undertake here, would be necessary for Czech and Polish variants of reciprocal expressions: (i)C Navzájem si pomáhají. [dat] refl clitic [3 ps pres] (they) each other help (ii)C Slíbili si navzájem navštívit prednášky toho druhého [dat refl. clitic] [gen] (they) promised mutually to visit the lectures (of) the other ‘They promised each other to visit each other’s lectures’. The adverb navzájem “reciprocally” marks the reciprocal meaning of the clitical dative reflexive si. (iii)P Obiecali sobie wzajemnie (PRO pójš´c do lekarza) [dat] (they) promised themselves reciprocally (mutually) to go to the doctor They promised each other …’

196

Notes

References

Abraham, Werner. 1982. Zur Kontrollbeziehung im Deutschen. Groninger Arbeiten zur germanistischen Linguistik 21: 112–167. Abraham, Werner. 1983. The Control Relation in German. In W. Abraham, ed., On the Formal Syntax of the Westgermania, 217–242, Amsterdam: Benjamins. Abraham, Werner. 1997. The Grammaticalization of the Infinitival Preposition. Toward a Theory of ‘Grammaticalizing Reanalysis’ (ms. Groningen) Adamec, Premysl 1988. K Vyjadrování a rozponávánií koreference v ruštine a v ceštine. Ceskoslovenská slavistika: 167–177. Bach, Emmon. 1979. Control in Montague Grammar. Linguistic Inquiry 10: 515–531. Bach, Emmon, and Barbara Partee. 1982. Anaphora and Semantics Structure. In Papers from the Parasession on Pronouns and Anaphora: 1–29. Chicago Linguistic Society, University of Chicago. Chicago, Ill. Baker, C. L. 1991. The Syntax of English Not: The Limits of Core Grammar. Linguistic Inquiry 22: 387–429. Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Baker, Mark, Kyle Johnson and Jan Roberts. 1989. Passive Arguments Raised. Linguistic Inquiry 20: 219-251. Baltin, Mark. 1995. Floating Quantifiers, PRO, and Predication. Linguistic Inquiry 26: 199–248. Bech, Gunnar. 1955, 1957. Studien über das deutsche Verbum infinitum. Det Kongelige Danske Videnskabernes Selskab. Historisk-filologiske, Meddelelser. 35, 2 (1955), 36, 6 (1957), Munksgaard, Copenhagen. Belletti, Adriana. 1988. The Case of Unaccusatives. Linguistic Inquiry 19: 1–34. Belletti, Adriana. 1982. The Impersonal Construction in Italian. Journal of Linguistic Research 2: 1–34. Bierwisch, Manfred. 1989. Event Nominalization. In W. Motsch, (ed.), Wortstruktur und Satzstruktur, Linguistische Studien des Zentralinstituts für Sprachwissenschaft der Akademie der Wissenschaften der DDR, Reihe A, H. 194, Berlin. Bierwisch, Manfred. 1993. Nachdenken über PRO. (unpublished ms. Berlin) Borer, Hagit. 1984. Parametric Syntax. Dordrecht: Foris. Borer, Hagit. 1986. I-Subjects. Linguistic Inquiry 17: 375–416. Bouchard, Denis. 1985 PRO, Pronominal or Anaphor. Linguistic Inquiry 16: 471–477. Bowers, John. 1993. The Syntax of Predication. Linguistic Inquiry 24: 591–656. Bresnan, Joan. 1982. Control and Complementation. Linguistic Inquiry 13: 343–434.

198

References

Brody, Michael. 1993. Θ-Theory and Arguments. Linguistic Inquiry 24: 1–23. Burzio, Luigi. 1986. Italian Syntax. Dordrecht: Reidel. Carrier-Duncan, Jill. 1985. Linking of Thematic Roles in Derivational Word Formation. Linguistic Inquiry 16: 1–34. Chierchia, Gennaro. 1984. Topics in the Syntax and Semantics of Infinitives and Gerunds. Doctoral Dissertation, University of Massachusetts, Amherst. Chierchia, Gennaro, and Richard Turner. 1988. Semantics and Property Theory. Linguistics and Philosophy 11: 261–302. Chomsky, Noam. 1980. On Binding. Linguistic Inquiry 11: 1–46. Chomksy, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomksy, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1986a. Barriers. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1986b. Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. Chomsky, Noam. 1989. Some Notes on Economy of Derivation and Representation. In I. Laka and A. Mahajan, (eds.), MIT Working Papers in Linguistics 10. Functional Heads and Clause Structure, 43–74. Chomsky, Noam. 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 Sylvian Bromberger, MIT, Cambridge, Massachusetts. Chomsky, Noam. 1995. The Minimalist Program. Cambridge: M.I.T. Press. Chomsky, Noam, and Howard Lasnik. 1993. Principles and Parameters Theory. In Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann, (eds.), Syntax: An International Handbook of Contemporary Research, 506–569, Berlin: Walter de Gruyter. Chvany, Catherine. V. 1975. On the Syntax of BE-Sentences in Russian. Cambridge, Mass.: Slavica Publishers. Cinque, Guglielmo. 1988. On Si Constructions and the Theory of Arb. Linguistic Inquiry 19: 521–581. Cole, Peter. 1987. Null Objects in Universal Grammar. Linguistic Inquiry 18: 597–612. Cole, Peter, Gabriella Hermon and Li-May Sung. 1990. Principles and Parameters of LongDistance Reflexives. Linguistic Inquiry 21: 1–22. Comrie, Bernard. 1981. Language Universals and Linguistic Typology. Oxford: B. Blackwell. Comrie, Bernard. 1985. Reflections on Subject and Object Control. Journal of Semantics 4: 47–65. Conrad, Rudi. 1969. Transformationsanalyse russischer Infinitivkonstruktionen. Linguistische Studien. Halle. Culicover, Peter, and Wendy Wilkins. 1986. Control, PRO and the Projection Principle. Language 62: 120–153. Dùabrowska Ewa. 1994. Dative and Nominative Experiencers: Two Folk Theories of the Mind. Linguistics. 32(6): 1029–1054. Davidson, Donald. 1989. Essays on Actions and Events. Oxford: Clarendon Press. Diesing, Molly. 1988. Bare Plural Subjects and the Stage Individual Contrast. Unpublished manuscript. University of Massachusetts, Amherst.

References

199

Dobrovie-Sorin, Carmen. 1998. Impersonal se Constructions in Romance and the Passivization of Unergatives. Linguistic Inquiry 29: 399–437 Dowty, David. 1985. On Recent Analyses of the Semantics of Control. Linguistics and Philosophy 8: 291–331. Dowty, David. 1991. Thematic Proto-Roles and Argument Selection. Language 67: 547– 618. Dowty, David, and William Laduslaw. 1988. Toward a Non-Grammatical Account of Thematic Roles. In Wendy Wilkins, (ed.), Syntax and Semantics 21: Thematic Relations, 61–73. Academic Press. Eisenberg, Peter. 1986. Zum Kontrollproblem im Deutschen: Infinitivkomplemente bei Wahrnehmungsverben. In A. Burkhardt and K.-H. Körner, (eds.), Pragmantax, 37–46 Tübingen. Emonds, Joseph. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris. Emonds, J. 1993. Projecting Indirect Objects. The Linguistic Review 10: 211–263. Epstein, Samuel David. 1984. Quantifier – pro and the LF Representation of PROarb. Linguistic Inquiry 15: 499–505. Farkas, Donka. 1988. On Obligatory Control. Linguistics and Philosophy 11: 27–58. Fauconnier, Gilbert. 1976. Complement Subject Deletion and the Analysis of menacer. In M. Lujan and F. Hensey, (eds.), Current Studies in Romance Linguistics, Washington: Georgetown University Press. Fillmore Charles, Paul Kay and M. C. O’Connor. 1988. Regularity and Idiomaticity in Grammatical Constructions. Language 64: 501–538. Fowler, George, and Michael Yadroff. 1993. The Argument Status of Measure Nominals in Russian. Journal of slavic linguistics 1(2): 251–279. Franks, Steven. 1983. Case and Control in Polish. In Cornell Working Papers in Linguistics 4, 23–49. Franks, Steven, and Norbert Hornstein. 1992. Secondary Predication in Russian and Proper Government of PRO. In Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 1–50. Dordrecht, Boston, London: Kluwer. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum and Ivan Sag. 1985. Generalized Phrase Structure Grammar. Oxford: Basil Blackwell. Grewendorf, Günther, and Joachim Sabel. 1994. Long Scrambling and Incorporation. Linguistic Inquiry 25: 263–308. Grimshaw, Jane. 1987. Unaccusative – an Overview. North Eastern Linguistic Society 17: 244-258. Grodzinsky, Yosef, and Tanya Reinhart. 1993. The Innateness of Binding and Coreference. Linguistic Inquiry 24: 69–101. Gruber, Jeffrey. 1965. Studies in Lexical Relations. Doctoral Dissertation, MIT, Cambridge, Mass. In Lexical Structures in Syntax and Semantics, Amsterdam: North Holland. 1976. Hale, Kenneth. 1992. Subject Obviation, Switch Reference, and Control. In Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 51–77. Dordrecht, Boston, London: Kluwer.

200

References

Hale, Kennth, and Samuel Jay Keyser. 1993. On Argument Structure and the Lexical Expression of Syntactic Relations. In Kenneth Hale ans Samuel Jay Keyser, (eds.), The View from Building 20. Essays in Linguistics in Honour of Sylvain Bromberger, 53– 109. MIT, Cambridge Massachusetts. Hajicová, Eva, Jarmila Panevová, and Petr Sgall. 1987. A Remark on Control. In Studies in Honour of Robert Busa S. J., Linguistica Computazionale IV, V, Pisa, 59–69. Heim, Irene, Howard Lasnik, and Robert May. 1991. Reciprocity and Plurality. Linguistic Inquiry 22: 63–101. Higginbotham, James. 1985. On Semantics. Linguistic Inquiry 16: 547–593. Higginbotham, James. 1992. Reference and Control. In Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 79–108. Dordrecht, Boston, London: Kluwer. Hornstein, Norbert, and David W. Lightfoot. 1987. Predication and PRO. Language 63: 23–51. Hust, J., and Michael Brame. 1976. Jackendoff on Interpretive Semantics. Linguistic Analysis 2: 243-277. Ickovic, Viktor A. 1970. Norma i ee kodifikacija. Moskva. Ijbema, Aniek and Werner Abraham. 1998. Die syntaktische Funktion des infinitivischen zu. Festschrift für Peter Eisenberg. Tübingen: M. Niemeyer. Ivi´c, M. 1972. Problematika srpskohrvatskog infinitiva. (Problems related to the SerboCroatian infinitive) Zbornik za filologiju i lingvistiku 15/2. Novi Sad. Iwakura, Kunihiro. 1985. The Binding Theory and PRO. Linguistic Analysis 15: 29–55. Iwakura, Kunihiro. 1990. Binding Theory and Two Types of PRO. Linguistic Analysis 20: 121–138. Jackendoff, Ray 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. Jackendoff, Ray. 1974. A Deep Structure Projection Rule. Linguistic Inquiry 5: 481–506. Jackendoff, Ray. 1987. The Status of Thematic Relations in Linguistic Theory. Linguistic Inquiry 18: 369–411. Jacobson, Pauline. 1990. Raising as Function Composition. Linguistics and Philosophy 13: 423–475. Jaeggli, Osvaldo. 1982. Topics in Romance Syntax. Dordrecht: Foris. Jaeggli, Osvaldo A. 1986. Passive. Linguistic Inquiry 17: 587–622. Jenkins, Lyle. 1972. Modality in English Syntax. Doctoral Dissertation, MIT, Cambridge, Mass. Johnson, Kyle. 1985. A Case for Movement. Unpublished Ph. D. dissertation. MIT. Joseph, Brian D. 1992. Diachronic Perspectives on Control. In Richard K. Larson, Sabine Jatridon, Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 195– 234. Dordrecht, Boston, London: Kluwer. Kamp, Hans, and Antje Roßdeutscher. 1995. Leider hatten wir vorgeschlagen, diesen Satz zu analysieren. (ms. Berlin). Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Kayne, Richard S. 1991. Romance Clitics, Verb Movement, and PRO. Linguistic Inquiry 22: 647–686.

References

201

Köpcke, Klaus-Michael, and Klaus-Uwe Panther. 1991. Kontrolle und Kontrollwechsel im Deutschen. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 44: 143–166, Berlin. Koster, Jan. 1984. On Binding and Control. Linguistic Inquiry 15: 417–459. Kratzer, Angelika. 1989. Stage-level and Individual-level Predicates. In E. Bach, A. Kratzer, and B. Partee, (eds.), Papers on Quantification, University of Massachusetts, Amhorst. Larson, Richard K. 1988. On the Double Object Construction. Linguistic Inquiry 19: 335– 391. Larson, Richard K. 1991. Promise and the Theory of Control. Linguistic Inquiry 22: 103– 134. Larson, Richard et al. (eds.). 1992 Control and Grammar. Dordrecht, Boston, London: Kluwer. Lasnik, Howard, and R. Fiengo. 1974. Complement Object Deletion. Linguistic Inquiry 5: 535–571. Lasnik, Howard. 1988. Subjects and the Theta-Criterion. Natural Language and Linguistic Theory 6: 1–17. Lasnik, Howard. 1992. Two Notes on Control and Binding. In Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 235– 251. Dordrecht, Boston, London: Kluwer. Legendre, Géraldine. 1997. Secondary Predication and Functional Projections in French. Natural Language and Linguistic Theory 15: 43–87. Levin, Beth, and Malka Rappaport. 1986. The Formation of Adjectival Passives. Linguistic Inquiry 17: 623–661. Manzini, Maria R. 1983, On Control and Control Theory. Linguistic Inquiry 14: 421–446. Manzini, Maria R. 1992. Locality. Cambridge, Mass.: MIT Press. Manzini, M. R., and A. Roussou. 1997. Interpretation as Feature Calculus: F-Movement and F-Control. GLOW Newsletter 38: 42–43. Melvold, J. 1985. Getting PRO under Control. ms., MIT Cambridge, Mass. Mittwoch, A. 1976. Raising or Equi or Both: Another Look at Root Modals and at Permissive Allow. Papers in Linguistics 10: 55–75. Mohanan, K. P. 1983. Functional and Anaphoric Control. Linguistic Inquiry 14: 641–674. Mohanan, K. P. 1985. Remarks on Control and Control Theory. Linguistic Inquiry 16: 637–648. Neidle, Carol. 1982. Case Agreement in Russian. In J. Bresnan (ed.), The Mental Representation of Grammatical Relations, 391–426, Cambridge, Mass.: MIT Press. Nishigauchi, Taisuke. 1984. Control and the Thematic Domain. Language 60: 215–250. Ostler, N. D. M. 1979. Case-Linking: A Theory of Case and Verb Diathesis Applied to Classical Sanskrit. Doctoral dissertation, MIT, Cambridge, Mass. Panevová, Jarmila. 1986. The Czech Infinitive in the Functions of Objective and the Rules of Coreference. In J. L. Mey, (ed.), Language and Discourse: Test and Protest. A. Festschrift for Petr Sgall, 123–142, Amsterdam: Benjamins. Panevová, Jarmila. 1996. More Remarks on Control. Prague Linguistic Circle Papers. NS (New Series), Vol. 2: 101–120.

202

References

Panther, Klaus-Uwe, and Klaus-Michael Köpcke. 1993. A Cognitive Approach to Obligatory Control Phenomena in English and German. Folia Linguistica XXVII/1–2: 57– 103. Berlin: Mouton/de Gruyter. Partee, Barbara. 1972. Opacity, Coreference, and Pronouns. In Donald Davidson and Gilbert Harman, (eds.), Semantics of Natural Language, 415–441. Dordrecht. Partee, Barbara. 1975. Deletion and Variable-Binding. In Edward Keenan, (ed.), Formal Semantics of Natural Language, 16–34. Cambridge University Press. Partee, Barbara. 1978. Bound Variables and Other Anaphors. In D. Waltz, (ed.), Theoretical Issues in Natural Language Processing 2, 79–85. University of Illinois. Partee, Barbara. 1979. Montague Grammar and the Well-Formedness Constraint. In F. Heny and H. Schnelle, (eds.), Syntax and Semantics, Vol. 10: 275–313, New York, San Francisco, London. Partee, Barbara H., and Emmon Bach. 1981. Quantification, Pronouns, and VP Anaphora. In J.A.G. Groenendijk et. al., (eds.), Formal Methods in the Study of Language, 445– 481. Amsterdam. Pesetsky, David. 1993. Topic … Comment. Natural Language and Linguistic Theory 11: 557–559. Petter, Marga (1998). Getting PRO under Control. A Syntactic Analysis of the Nature and Distribution of Unexpressed Subjects in Non-finite and Verbless Clauses. Holland Academic Graphics The Hague. Postal, Paul. 1970. On Coreferential Complement Subject Deletion. Linguistic Inquiry 1: 439–500. Postal, Paul, and Geoffrey K. Pullum. 1988. Expletive Noun Phrases in Subcategorized Positions. Linguistic Inquiry 19: 635–670. Reinhardt, Tanya. 1983. Anaphora and Semantic Interpretation. London: Croom Helm. Risch, Gerda. 1989. Kontrollverhalten in infiniten Komplementkonstruktionen. In W. Motsch, (ed.), Wortstruktur und Satz-struktur, 159-187, Berlin: Linguistische Studien, Reihe A. Rizzi, Luigi. 1986. Null Objects in Italian and the Theory of pro. Linguistic Inquiry 17: 501–557. Rochette, A. 1982. French Infinitival Complements. In Alex Marantz, Tim Stowell, (eds.), Papers in Syntax. MIT Working Papers in Linguistics IV. Roeper, Thomas. 1987. Implicit Arguments and the Head-Complement Relation. Linguistic Inquiry 18: 267–310. Rosenbaum, Peter. S. 1970. A Principle Governing Deletion in English Sentential Complementation. In R. Jacobs and P. Rosenbaum, (eds.), Readings in English Transformational Grammar, Waltham, Mass.: Ginn. Rothstein, Susan D. 1991. Syntactic Licensing and Subcategorization. In Susan D. Rothstein, (ed.), Syntax and Sematics 25. Perspectives on Phrase Structure: Heads and Licensing, 138–158. Academic Press, INC. Rothstein, Susan D. 1995. Pleonastics and the Interpretation of Pronouns. Linguistic Inquiry 26: 499–529. R˚uzicka, Rudolf. 1973. Reflexive versus Nonreflexive Pronominalization in Modern Russian and Other Slavic Languages. In F. Kiefer and N. Ruwet, (eds.), Generative Grammar in Europe, 445–481, Foundations of Language, Supplementary Series, Vol. 13, Dordrecht: Reidel.

References

203

R˚uzicka, Rudolf. 1980. Studien zum Verhältnis von Syntax und Semantik im modernen Russischen. Berlin: Akademie-Verlag. R˚uzicka, Rudolf. 1982. Kontrollprinzipien infiniter Satzformen: Infinitiv und Gerundium (deepricastie) im Russischen. Zeitschrift für Slawistik 27: 373–411. R˚uzicka, Rudolf. 1983a. Remarks on Control. Linguistic Inquiry 14: 309–324. R˚uzicka, Rudolf. 1983b. Autonomie und Interaktion von Syntax und Semantik. In R. R˚uzicka, W. Motsch, (eds.) Untersuchungen zur Semantik, studia grammatica XXII: 15–59. Berlin: Akademie-Verlag. R˚uzicka, Rudolf. 1984. Illokutionäre Kraft und Subjektkontrolle - Bitten und Fragen. Linguistische Arbeitsberichte 44: 7–10, Leipzig. R˚uzicka, Rudolf. 1986. Control in Competing Frameworks. In J. L. Mey, (ed.), Language and Discourse: Test and Protest, A Festschrift for Petr Sgall, 101–121, Amsterdam: Benjamins. R˚uzicka, Rudolf. 1986a. Typologie der Diathese slavischer Sprachen in parametrischen Variationen. Welt der Slaven XXXI, 2: 225–274. R˚uzicka, Rudolf. 1987. The Semantic-Pragmatic Interface and Interaction in Control Areas. In Albrecht Neubert and R. R˚uzicka, (eds.), Topics on the Semantic Borderline, Linguistische Studien, Reihe A, 166: 128–177, Berlin. R˚uzicka, Rudolf. 1988. On the Array of Arguments in Slavic Languages. Zeitschrift f. Phonetik, Sprachwissenschaft und Kommunikationswissenschaft 41, 2: 155–178. R˚uzicka, Rudolf. 1992. Slavic and Italian Impersonal Constructions with Reflexive Clitics. In I. Zimmermann, A. Strigin, (eds.), Fügungspotenzen, studia grammatica XXXIV: 133–161. Berlin: Akademie-Verlag. R˚uzicka, Rudolf. 1994. Gegen die Aufgabe von PRO. In Anita Steube and Gerhild Zybatow, (eds.), Zur Satzwertigkeit von Infinitiven und Small Clauses, 13–18. Tübingen. Safir, Ken. 1987. The Syntactic Projection of Lexical Thematic Structure. Natural Language and Linguistic Theory 5: 561–601. Safir, Ken. 1996. Derivation, Representation and Eesumption: The Domain of Weak Crossover. Linguistic Inquiry 27: 313–339. Sag, Ivan. A. and Carl. Pollard. 1991. An Integrated Theory of Complement Control. Language 67: 63–113. Sgall, Peter, Eva Hajicová, and Jarmila Panevová. 1986. The Meaning of the Sentence in its Semantic and Pragmatic Aspects. Academia Praha. Siebert-Ott, G. 1985. Bemerkungen zu den Elementen einer Theorie der Kontrolle. In Werner Abraham, (ed.), Erklärende Syntax des Deutschen. Studien zur deutschen Grammatik 25: 255–270, Tübingen. Sigurdsson, Halldór A. 1991. Icelandic Case-marked PRO and the Licensing of Lexical Arguments. Natural Language and Linguistic Theory 9: 327–363. Stowell, Tim. 1991. The Alignment of Arguments in Adjective Phrases. In Susan D. Rothstein, (ed.), Syntax and Semantics 25. Perspectives on Phrase Structure: Heads and Licensing, 105–138. Academic Press, INC. Stroik, Thomas. 1994. Kayne’s Universal Word Order and Psych-Verbs. (ms. Berlin). Stump, Gregory. T. 1985. The Semantic Variability of Absolute Constructions. Synthese Language Library 25, Dordrecht, Boston, Lancaster: Reidel.

204

References

Svoboda, Karel. 1962. Infinitiv v soucasné spisovné cestine. Praha: Nakladatelství Ceskoslovenské Akademie Ved. Torrego, Esther. 1996. On Quantifier Float in Control Clauses. Linguistic Inquiry 27: 111– 126. Wegener, Heide. 1989. ‘Kontrolle’ – Semantisch Gesehen. Deutsche Sprache 3/1989: 206– 228. Williams, Edwin. 1980. Predication. Linguistic Inquiry 11: 203–238. Williams, Edwin. 1985. PRO and Subject of NP. Natural Language and Linguistic Theory 3: 277–295. Williams, Edwin. 1987. Implicit Arguments, the Binding Theory, and Control. Natural Language and Linguistic Theory 5: 151–180. Williams, Edwin. 1987a English as an Ergative Language: The Theta Structure of Derived, Nouns. In B. Need, E. Schüller and A. Bosch, (eds.), The General Session, CLS 23, Part One. Chicago Linguistic Society. Wiiliams, Edwin. 1992. Adjunct Control. In Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 297–322. Dordrecht, Boston, London: Kluwer. Wyngaerd, Guido V. 1987. Control Theory. A Government and Binding Approach to Infinitival Complementation. Antwerp Papers in Linguistics 52. Yokoyama, Olga T. 1986. Discourse and Word Order. John Benjamins. Amsterdam, Philadelphia. Zec, Drago. 1987 On Obligatory Control in Clausal Complements. In M. Iida, St. Wechsler, and D. Zec, (eds.), Working Papers in Grammatical Theory and Discourse Structure. CLS/Numer 11. Zubizarreta, Maria-Luisa. 1982. On the Relationship of the Lexicon to the Syntax. Doctoral dissertation. MIT. Cambridge, Massachussets.

Subject index

A adjacency principle, 18 adjunct control, 9, 189 note 8 ambiguity, 43 f., 63, 81 arbitrariness, 13, 19 f., 22 B binding and control, 18, 22 C Case of PRO, 19, 173 causative switch, 59, 110,181 chains, 26 compatibility of theta specifications, 5, 176, 182 complementary distribution of control conditions, 55 f. constraints and lexical classes, 2, 32 control classes, 115, 158, 160, 162 coreference invoking selectional pairing of semantic roles, 38 f., 59 corollary (D), 34 D demoted agentives as controllers, 89, 93-97, 192 note 23 deontic switch, 62, 110, 180 discarding PRO, 181 f. discourse-related interpretation, 7, 191 note 21 division of labor, 104 E ECM or control, 81 entailment, 172, 190 note 11, 14

essence of control, 172 evaluating and attitudinal adjectives, 137-147 F feature calculus approach to control, 182 f. G grain problem, 151 f. H hierarchical restriction of controller positions, 38, 39, 93-97 I intentional action, 31 interface parameters, 59, 62 L lexical saturation of theta roles, 85 locality of control, 18, 171 M markedness, 34 f. 38 f. 59, 161, 180 minimal distance principle, 22, 23, 104, 192 note 24 modal verbs intervening, 109, 192 note 25 modality switch, 181 N null object parameter, 43-49, 52 f. interacting with interface control parameters, 58-62, 65

206 O object of by and counterparts as controllers, 93-97, 192 note 23 oblique object control, 76, 148 P parameterization of pragmatics, 2, 5, 55-63, 180 pragmatics, 44, 189 note 8 pro as controller and controllee, 51, 129-136 propositional versus attributive account, 23, 116 f. psychological state predicates, 141

Subject index S scrambling and control, 114 f. selectional cross-clausal underpinnings of control, 19, 38, 172 specification on the marked application of main constraints, 34 stage-level predicates, 142 subject complement clauses, 27 symmetry and distinction between main constraints, 91 f. T turn of approach to control in terms of thematical specifications, 25

Q questions of control, 23

U unaccusatives, 37, 64

R raising, 23 f.. 106 reciprocality, 151 f. reflexive impersonal clauses, 129-136 reflexive passive controlled clauses, 133 f.

V verbal noun control, 49 f., 96 f. violation of (marked) constraints, 40, 43 f., 165 VP-deletion, 152

In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titles have been published thus far, or are scheduled for publication: 1. KLAPPENBACH, Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene Malige-Klappenbach. 1980. 2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. 3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. 4. ABRAHAM, Werner & Sjaak De MEIJ (eds): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. 5. GREWENDORF, Günther and Wolfgang STERNEFELD (eds): Scrambling and Barriers. 1990. 6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. 7. ÅFARLI, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. 8. FANSELOW, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. 9. GELDEREN, Elly van: The Rise of Functional Categories. 1993. 10. CINQUE, Guglielmo and Guiliana GIUSTI (eds): Advances in Roumanian Linguistics. 1995. 11. LUTZ, Uli and Jürgen PAFEL (eds): On Extraction and Extraposition in German. 1995. 12. ABRAHAM, W., S. EPSTEIN, H. THRÁINSSON and C.J.W. ZWART (eds): Minimal Ideas. Linguistic studies in the minimalist framework. 1996. 13. ALEXIADOU Artemis and T. Alan HALL (eds): Studies on Universal Grammar and Typological Variation. 1997. 14. ANAGNOSTOPOULOU, Elena, Henk VAN RIEMSDIJK and Frans ZWARTS (eds): Materials on Left Dislocation. 1997. 15. ROHRBACHER, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and pro-drop. 1999. 16. LIU, FENG-HSI: Scope and Specificity. 1997. 17. BEERMAN, Dorothee, David LEBLANC and Henk van RIEMSDIJK (eds): Rightward Movement. 1997. 18. ALEXIADOU, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. 19. JOSEFSSON, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. 20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. 21. KLEIN, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. 22. ALEXIADOU, Artemis and Chris WILDER (eds): Possessors, Predicates and Movement in the Determiner Phrase. 1998. 23. GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. 24. REBUSCHI, Georges and Laurice TULLER (eds): The Grammar of Focus. 1999. 25. FELSER, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999. 26. ACKEMA, Peter: Issues in Morphosyntax. 1999.

° 27. RUZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. 28. HERMANS, Ben and Marc van OOSTENDORP (eds.): The Derivational Residue in Phonological Optimality Theory. 1999. 29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 1999. 30. BEUKEMA, Frits and Marcel den DIKKEN (eds.): Clitic Phenomena in European Languages. 2000. 31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000. 32. ALEXIADOU, Artemis, Paul LAW, André MEINUNGER and Chris WILDER (eds.): The Syntax of Relative Clauses. 2000. 33. PUSKÁS, Genoveva: Word Order in Hungarian. The syntax of È-positions. 2000. 34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. 35. HRÓARSDÓTTIR, Thorbjörg. Word Order Change in Icelandic. From OV to VO. 2000. 36. GERLACH, Birgit and Janet GRIJZENHOUT (eds.): Clitics in Phonology, Morphology and Syntax. 2000. 37. LUTZ, Uli, Gereon MÜLLER and Arnim von STECHOW (eds.): Wh-Scope Marking. 2000. 38. MEINUNGER, André: Syntactic Aspects of Topic and Comment. 2000. 39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, ‘‘Self’’, and Interpretability. 2000. 40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton van der WOUDEN (eds.): Perspectives on Negation and Polarity Items. 2001. 41. ZELLER, Jochen : Particle Verbs and Local Domains. n.y.p. 42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and ergativity. n.y.p. 43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001. 44. TAYLAN, Eser E. (ed.): The Verb in Turkish. n.y.p. 45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds.): Issues in Formal German(ic) Typology. n.y.p 46. PANAGIOTIDIS, Phoevos: Pronouns, clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. n.y.p.