Current English Linguistics in Japan [Reprint 2011 ed.] 9783110854213, 9783110117813

Citation preview

Current English Linguistics in Japan

Trends in Linguistics State-of-the-Art Reports 16

Editor

Werner Winter

Mouton de Gruyter Berlin · New York

Current English Linguistics in Japan

Edited by

Heizo Nakajima

Μ out on de Gruyter Berlin · New York

1991

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter & Co., Berlin.

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging in Publication

Data

Current English linguistics in Japan / edited by Heizo Nakajima. p. cm. — (Trends in linguistics. State-of-the-art reports ; 16) Includes bibliographical references and index. ISBN 0-89925-505-1 (alk. paper) 1. English language —Grammar —1950 — 2. Linguistics—Japan. I. Nakajima, Heizo, 1946— . II. Series: Trends in linguistics. State-of-the-art reports ; 16. PE1106.C87 1991 91-22180 425 —dc20 CIP

Die Deutsche Bibliothek

— Cataloging in Publication

Data

Current English Linguistics in Japan / ed. by Heizo Nakajima. — Berlin : New York : Mouton de Gruyter, 1991 (Trends in linguistics : State-of-the-art reports ; 16) ISBN 3-11-011781-9 NE: Nakajima, Heizo [Hrsg.]; Trends in linguistics / State-ofthe-art reports

© Copyright 1991 by Walter de Gruyter & Co., D-1000 Berlin 30. All rights reserved, including those of translation into foreign languages. N o part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Typesetting: Arthur Collignon GmbH, Berlin. — Printing: Gerike GmbH, Berlin. — Binding: Lüderitz & Bauer, Berlin. — Printed in Germany.

Contents Introduction Heizo Nakajima

1

Non-localizable contextual features: Present subjunctives in English Shuji Chiba

19

Syntactic localization phenomena in English Hajime Fukuchi

45

Strong and weak barriers: Remarks on the proper characterization of barriers Naoki Fukui

77

An analysis of English descriptive genitives in the dynamic theory of syntax Masayuki Ike-uchi

95

On government by noun Kunihiro Iwakura

139

Light verb constructions and the syntax-morphology interface Tar ο Kageyama

169

Binding under the internal subject hypothesis Yoshihisa Kitagawa

205

Remarks on quantifier scope Susumu Kuno

261

Binding path and dependent categories Heizo Nakajima

289

On "null operator" constructions Masaru Nakamura

345

The perfective have and the progressive be as spec verbs and the I N F L system in English Harumi Sawada 381 A functional approach to preposition stranding in English Ken-ichi Takami

413

vi

Contents

The comparative syntax of English and Japanese: Relating unrelated languages Shigeo Tonoike

455

Bibliography

507

Index

529

Introduction* Heiz ο Nakajima

1. Purpose of the volume This collection of papers is an attempt to inform non-Japanese linguists about the current state of English-linguistic study in Japan. The volume contains thirteen papers on English linguistics, specifically, on the topics of the syntax of modern English and the syntactic comparison of English and Japanese. These two areas are those most widely and intensively studied in Japanese English linguistics. The majority of the contributors are actively engaged in research on English linguistics in Japan. The few exceptions are those who work outside the country, but have considerable influence upon English-linguistic study in Japan. It seems fair to say that the study of English linguistics by Japanese has matured enough to be appreciated by the linguistic community outside Japan. Many students have gone abroad to study linguistics at universities in the United States and in other Western countries, and many of them have successfully completed doctoral programs. Their doctoral dissertations have been favorably received not only when they treat the Japanese language but also when they deal with the English language proper or with the comparison of English and Japanese. Some of them have been frequently referred to in papers written subsequently by other researchers, and have made a significant contribution, both directly and indirectly, to the development of linguistic theories. Moreover, it is not unusual these days for papers on English written by Japanese to be accepted by highquality journals in other countries, such as Linguistic Inquiry (MIT Press), Language (the Linguistic Society of America), The Linguistic Review (Foris), Lingua (North-Holland), and so on.

* I am grateful to many people and companies for providing me with much information, in particular to Mineo Moriya (Kenkyusha Publishing Company), Jun-ichi Yoneyama (Taishukan Publishing Company), Ken Kawada (Kaitakusha Publishing Company), Akira Ota, Minoru Yasui, Masatomo Ukaji, and Sanseido Book Store. My thanks also go to Renee Oatway and Michael Ullman, who carefully read several versions of this paper.

2

Heizo Nakajima

This contribution to English linguistics at an international level reflects the large amount of activity taking place in the field in Japan. This has mainly been promoted by the influence of the study of generative grammar since the late 1960s. It is without doubt that generative grammar has captured many people's interest, and is at present the main area of English linguistic study in Japan. In the past few years, however, a trend has emerged among Japanese linguists which seeks new approaches to the English language. This trend is noteworthy and has not been observed before. Some of the new theories have close ties to generative grammar, and others are founded on traditional Japanese approaches to language description while incorporating some of the ideas of generative grammar, and still others are clearly divergent from generative grammar in some fundamental respects. Unfortunately, most of these approaches have not been widely disseminated in other countries, mainly because most of the papers have been written in Japanese, but also because of the general tendency of established journals not to accept articles developing littleknown frameworks. The purpose of this collection of papers is to make these new approaches accessible to researchers outside Japan. Since the general situation of English-linguistic study in Japan is not well-known to people in other countries, the rest of the introduction provides a brief account of the present situation and background of English linguistics in Japan, with the aim of providing a little more information about the field.

2. Current popularity of English linguistics in Japan In Japan, the study of English linguistics has been more popular in the last few decades than at any other time in the past. It is no exaggeration to say that English linguistics is currently as common as, or even more common than, the study of modern Japanese. The popularity of English linguistics is mostly due to the intriguing nature of the subject matter and the theories in which it is studied, though it is also partly due to the fact that there are many people in universities who are concerned with the target of the research, the English language. All universities and colleges are required to offer foreign-language classes, of which English classes are the most common. Most universities therefore need many English teachers, whose speciality is in most cases either English literature, American literature, or English linguistics. The ratio

Introduction

3

of these three fields is roughly 50%, 30%, and 20%, respectively. Though the author does not have the exact number of English linguists or English teachers in Japanese universities, a clue is provided by the number of members of the English Linguistic Society of Japan (ELSJ), the organization for English linguists in Japan founded in 1983, which is said to be more than 1,200. The total number of English linguists in Japan probably amounts to between 1,800 and 2,000. This number is quite large considering that it includes only one academic field in a country where English is neither a native nor an official language. English linguistics is taught in departments of English literature rather than in departments of linguistics (of which there are very few in Japan). Departments of English literature are among the largest departments in Japanese universities. Students choose one of the three majors (English literature, American literature, or English linguistics), and write B. A. or M . A . theses on a topic in that major. Students who major in English linguistics have historically been fewer in number than those who major in English or American literature. However, the number of Englishlinguistics students, as well as the number of English-linguistics teachers, has steadily increased during the past few years. The expression "English linguistics" may sound somewhat strange to Westerners, who use the term "linguistics" to denote the study of human language in general, and not that of a particular language. However, the study of linguistics in the Western sense is very rare in Japan; most linguistic activities are concerned with particular languages. The term "linguistics" is sometimes even understood to be equivalent to "theoretical linguistics of English", because linguistic study of English is the most common type of linguistic work carried out, and the theoretical approach is that most common in English linguistics. The term "English linguistics" is used ambiguously in Japan to refer to two different types of study. In a wider sense, it refers to the study of English language in general; in a narrower sense, to the theoretical study of the modern English language. The term "English linguistics" in the latter sense is used in opposition to "English philology", which denotes the traditional study of Old or Middle English, or of the usage of certain words and constructions. Until the 1960s the term "English philology" was employed to refer to the study of English in general. The transition to using "linguistics" instead of "philology" to refer to the general study of the English language suggests that theoretical (linguistic) studies of English now predominate over traditional (philological) studies of English.

4

Heizo Nakajima

Of the areas of English linguistics, the study of syntax is the most popular in Japan, as opposed to phonetics, phonology, semantics, or pragmatics. This may be related to the fact that English education in Japanese junior and senior high schools concentrates on teaching prescriptive grammar. For students who have undergone this training, syntax is the easiest component of language to understand. Papers on syntax are submitted to journals in far greater numbers than papers on other disciplines.

3. Recent history of English linguistics in Japan The recent popularity of English linguistics in Japan is an outgrowth of the success of generative-transformational grammar since the mid-1950s. Before generative grammar, English-linguistic study was not very popular, nor were there as many English linguists. For further information on the state of English linguistics prior to generative grammar, see Ota (1967), "The study of English in Japan". The appearance of generative grammar changed the situation of Japanese English linguistics drastically, and contributed immensely to its progress in many respects. The advances in Japanese English linguistics can be measured in terms of the publication of books and journals, the organization of societies, and the contribution of articles to journals in other countries.

3.1. Publication of linguistics books and journals The publication of books is a good indicator of the advances in an academic field. An increase in the supply of books indicates an increase in the number of students who require them, and a rise in the quality of books implies a rise in the qualitative standard of the discipline. A brief history of the publication of English-linguistics books will illustrate the progress made in both quantity and quality in English-linguistics study. Generative-transformational grammar began to exert an influence upon the publication of books in the mid-1960s. This influence first manifested itself in the translation of some popular books on generative grammar into Japanese. These translations served to make generative grammar familiar both to young students and to established researchers who had been trained in the framework of traditional grammar and American

Introduction

5

structural linguistics. A few examples of the translations in the early days are the following (the items in parentheses are the year of the translation, the name of the translator, and the publisher): Noam Chomsky, Syntactic structures (1963, Yasuo Isami; Kenkyusha); Emmon Bach, An introduction to transformational grammar (1969, Kazuko Inoue; Taishukan); Noam Chomsky, Aspects of the theory of syntax (1970, Minoru Yasui; Kenkyusha); Noam Chomsky, Cartesian linguistics (1970, Shigeo Kawamoto; TEC); Ronald Langacker, Language and its structure, (1970, Seiichi Makino; Taishukan); Jerrold Katz, The philosophy of language (1971, Nobushige Sawada and Yuji Nishiyama; Taishukan); Roderick Jacobs — Peter Rosenbaum, Grammar I—IV: An introduction to transformational grammar (1971, Tamotsu Matsunami; Taishukan). The popularity of these translations is clear from the fact that they have been reprinted many times. For example, Chomsky's Syntactic structures was printed sixteen times (about 13,000 copies), Aspects of the theory of syntax, thirteen times (about 12,000 copies), Bach's book seven times in the eight years after its initial publication, and Jacobs — Rosenbaum's volumes nine times in less than ten years after the first translation. In addition to those mentioned above, most of Chomsky's major works published so far have already been translated into Japanese, including Current issues in linguistic theory (with Morris Halle), Language and mind, Studies on semantics in generative grammar, Reflections on language, Essays on form and interpretation, Lectures on government and binding, and Some concepts and consequences of the theory of government and binding. In the 1970s, Japanese linguists became dissatisfied with translating books written in English, and began to publish English-linguistics books of their own. They aimed to write books which would be more easily understood by Japanese than translations of Western books, and which could survive criticism by other people. Publishers also felt that the number of English linguists had increased enough to balance the demand of books with their supply. Some of the books were published individually, while others were included in a series, a common way of publishing research books in Japan. Of the series started in the 1970s, two are of particular importance by virtue of their contribution to the progress in the study of English linguistics. One is Eigogaku taikei [Outlines of English linguistics] (edited by Akira Ota, 1971 — 1990, published by Taishukan), consisting of 16 volumes. The other is Gendai no eibunpoo [Modern English Grammar] (edited by Kazuo Araki, Kinsuke Hasegawa and Minoru Yasui, 1976 —

6

Heizo Nakajima

[not completed], published by Kenkyusha), consisting of 12 volumes. These series are both contrastive and complementary to each other. The former dedicates each volume to a major sub-field of English linguistics, such as theoretical linguistics, traditional grammars, philology, Englishlanguage education, psycho- and socio-linguistics, and so on. The latter, on the other hand, assigns to each volume the analysis of particular constituent types of English sentence structures, such as adjective, auxiliary, noun phrase, verb phrase, major sentence classes, and so on. It attempts a comprehensive survey of the major constituents of English sentences in the framework of early transformational grammar. Despite these distinguishing features, the two series are alike in that they both show the far-reaching influences of early forms of generative grammar. These successful series were followed by similar but more compact ones which incorporated the results of subsequent studies of generative grammar. A few examples follow (the year given is the date of publication of the first issue): Eigogaku soosho [Series of English linguistics] (10 volumes, edited by Akira Ota and Masaru Kajita, 1980, Taishukan); Eigogaku koosu [English linguistic course] (4 volumes, edited by Tamotsu Matsunami, Kunihiko Imai and Yoshihiko Ikegami, 1985, Taishukan); Gendai no eigogaku [Modern English linguistics] (10 volumes, edited by Minoru Yasui, 1987, Kaitakusha); Eigogaku nyumon kooza [Introductory lectures on English linguistics] (12 volumes, edited by Kazuo Araki, 1987, Eichosha). The books in these series are generally read as reference or text books. Books on particular research topics have also been published. A notable collection of such books is included in the Dissertation series, published by Kaitakusha. Most of these are based on dissertations submitted to American universities in the 1960s and 1970s, and are thus written in English. Some of them are concerned with the Japanese language, while others deal with English or with the comparison of English and Japanese. Included are Minoru Nakau, Sentential complementation in Japanese (1973); Masatake Muraki, Presupposition and thematization (1974); John Hinds, Aspects of Japanese discourse structure (1976); Shosuke Haraguchi, The tone pattern of Japanese (1977); Masaaki Yamanashi, Generative semantics studies of the conceptual nature of predicates in English (1977); Masatomo Ukaji, Imperative sentences in early modern English (1978); and Seiichi Nakada, Aspects of interrogative structure (1980). Regrettably, this series has been discontinued. There is no way, at present, for outstanding dissertations to be published as books in a series. Since the number of dissertations submitted to Western and Japanese universities

Introduction

7

is increasing, some means must be devised to guarantee the publication of Japanese linguists' dissertations. Other research books have been sporadically published in English and in Japanese. Some examples of generative studies are: Masaru Kajita, A generative-transformational study of semi-auxiliaries in present-day American English (1968, Sanseido); Yoshihiko Ikegami, The semological structure of the English verbs of motion (1970, Sanseido); Kunihiro Iwakura, Nichi-ei-go no hitei no kenkyu [A generative-transformational study of negation] (1974, Kenkyusha); Taro Kageyama, Nichi-ei hikaku goi no kozo [Japanese-English comparison of lexical structure] (1980, Shohakusha); Akira Ota, Hitei no imi [The meaning of negation] (1980, Taishukan); Heizo Nakajima, Eigo no ido gensho kenkyu [A study of movement phenomena in English] (1984, Kenkyusha). Most of these are based on dissertations submitted to Western universities; some are revised versions, while others are translations. Journals compensate for the dearth of book-publishing opportunities for young researchers in Japan. Journals are open to everybody, and play a crucial role in the training and development of young linguists. Until the early 1980s, there were two major journals which fulfilled such a role. One was Eigogaku [English linguistics] (edited by Minoru Yasui, published by Kaitakusha), and the other Studies in English Linguistics (originally edited by Akira Ota, though later he was joined by Susumu Kuno, Kinsuke Hasegawa, and Masaru Kajita; published by Asahi Press). The former journal was published from 1969 to 1984 with a total of 27 volumes, and includes 147 articles written by 124 different authors. The latter was published between 1972 and 1983 with a total of 11 volumes, and contains 96 papers written in English by 60 different authors. Both of them were discontinued upon the creation of the journal of the English Linguistics Society of Japan (ELSJ), English Linguistics, in 1984. English Linguistics is published annually. It accepts papers on English and the comparison of English and Japanese in all areas of linguistics. All articles are required to be written in English, for the society expects that they will be widely read by people in other countries as well as in Japan, and encourages English linguists in Japan to direct their efforts toward an international readership. The criteria for the acceptance of articles are more severe than those of its predecessors (Eigogaku and Studies in English Linguistics)', the acceptance rate is less than 40%. Each issue usually contains between 15 and 20 articles. Another journal relevant to the study of English linguistics is Studies in English Literature, published by the English Literary Society of Japan.

8

Heizo Nakajima

As is evident from its title, the journal was originally devoted to the study of English literature, but it now covers the study of English linguistics as well. It is published tripartitely (two issues in Japanese and one issue in English per year). Each issue includes two or three papers on English linguistics, as well as reviews of books on English linguistics published in Japan and in other countries. Working papers are also regularly published by graduate students. Most of them are circulated among universities and institutes, and are rarely brought on to the market. However, some are available at book stores that stock foreign books, such as Sanseido Book Store (1-1, Kanda Jinbo-cho, Chiyoda-ku, Tokyo, 101) and Maruzen (2-3-10, Nihonbashi, Chuo-ku, Tokyo, 103). Some examples of working papers available at these stores are: Descriptive and Applied Linguistics (International Christian University), Vol. 1 (1961) —20 (1987); Sophia Linguistics (Sophia University), Vol. 1 (1975)-26 (1989); Metropolitan Linguistics (Tokyo Metropolitan University), Vol. 1 (1981)-9 (1989); Tsukuba English Studies (Tsukuba University), Vol. 1 (1982)-8 (1989). The publication of several English-linguistics dictionaries is also worthy of note. The compilation of dictionaries is very popular with Japanese linguists; this results from a particular climate in the Japanese academic world (which will be discussed later in some detail). Actually, some English-linguistics dictionaries have achieved considerable success. One successful dictionary is Shin-eigogaku jiten [The Kenkyusha dictionary of English linguistics and philology] (compiled by Takanobu Otsuka and Fumio Nakajima, 1982, 1582 pages, Kenkyusha). It lists a total of 1752 items alphabetically, and gives for them brief or detailed explanations depending upon their importance. Another dictionary with similar features is Gendai eigogaku jiten [Seibido's dictionary of English linguistics] (compiled by Kotaro Ishibashi et al., 1973, 1303 pages, Seibido), which lists 1867 items. A dictionary with different characteristics is Taishukan eigogaku jiten [The Taishukan encyclopaedia of English linguistics] (edited by Tamotsu Matsunami, Kunihiko Imai, and Yoshihiko Ikegami, 1983, 1421 pages, Taishukan). This dictionary contains one chapter for each major field of English linguistics, with each chapter divided into several sections and sub-sections in which important topics of the chapter are discussed in some detail. The topics it covers range from the history of linguistic studies, historical linguistics and lexicography to rhetorics and orthography. A more specialized dictionary is Chomsky sho-jiten (Chomsky dictionary, edited by Kunihiko Imai, 1986, 373 pages, Taishukan). This covers various topics related to the life and work of Noam

Introduction

Chomsky, his background, transition, his politics, the schools and other academic generative studies from the countries.

9

his recent linguistic theory, his theoretical influence of his theory on other linguistic fields. All of these dictionaries resulted from 1960s to the present in Japan and in other

3.2. Organization of English linguistics societies The progress of English linguistics is also evident from the organization and growth of linguistic societies. One of the societies most directly relevant to English linguistics has traditionally been the English Literary Society of Japan, established in 1917. The society, originally founded for the study of English literature, now also pursues the study of American literature and English linguistics. Its major activities are the publication of its journal, Eibungaku Kenkyu [Studies in English Literature] and its annual meeting. The journal accepts papers on English linguistics, and its annual meetings have a number of sessions dedicated to English linguistics. However, as is clear from the name of the society, English linguistics is only a subsidiary interest of the society. Linguists had long wished to organize a society where they could work more actively, and which would accomodate sophisticated levels of English linguistics. After a few attempts, a new society for English linguistics, the English Linguistic Society of Japan (ELSJ), was founded in 1983. The establishment of the society was a symbolic event which showed that English linguistics in Japan had reached the level necessary to be considered an independent academic field. English linguistic study had matured to the point where there were enough researchers to sustain a journal and regular well-attended meetings. The ELSJ began with about 800 members, and now numbers more than 1,200. The society publishes its journal, English Linguistics, annually (see section 3.1), and holds its annual meeting for 2 days in November. This annual meeting usually has about 600 participants; it normally offers five symposia and allows about 25 screened papers to be read in several sessions. The ELSJ welcomes papers from other countries which are read at the annual meetings and appear in English Linguistics. Correspondence should be addressed to The ELSJ, c/o Kihara Shoten, 44-5 Koenji-minami 2-chome, Suginami-ku, Tokyo, 166, Japan. Besides these nationwide societies, smaller linguistic societies and circles are organized in many districts and in various universities. They hold

10

Heizo Nakajima

regular annual or monthly meetings which are the grass roots of English linguistics in Japan. Of these organizations, the ICU (International Christian University) Summer Institute of Linguistics is particularly noteworthy. Regrettably, it ceased to exist after the 25th annual meeting in 1986. It was akin to the CLS (Chicago Linguistic Society) in its liberal atmosphere and non-exclusiveness, and played a significant role in the development of young linguists. The Institute has now been reorganized as the Tokyo Linguistic Forum. Other local and university organizations also avidly pursue their own activities, such as studying important books on current topics, reading original papers, and inviting guest scholars from other countries.

3.3. Contribution to foreign journals The advances in the study of English linguistics can also be seen in the increase in the number of articles which have been accepted by highquality journals in other countries. It is not possible to list all these articles here; the bibliography at the end of this volume contains some of these, in particular those by the contributors to this volume.

4. Climate and future directions of English linguistics in Japan It might appear that English linguistics in Japan is similar to linguistics in the United States, particularly because generative grammar has been the major framework in both countries. However, there are fundamental differences underlying the research on language between the two countries. Roughly speaking, and with many exceptions, it can be said that the main concern of American linguists lies in "theory construction", whereas Japanese linguists are mainly concerned with "data description". Americans tend to regard data only as the basis for theory construction, while Japanese are liable to consider theory only as an instrument for data description. Therefore, many Japanese linguists who study English have a great interest in collecting English data from written sources and describing them succinctly from particular perspectives. Linguistic theories are sim-

Introduction

11

ply used to provide such perspectives. Linguistic works, even if theoryoriented, usually concentrate on verifying or modifying previous work, while keeping the theories used as frameworks intact. Seldom do new works attempt to overthrow the theories themselves. The tendency toward data description may be natural to the study of English in Japan because the Japanese are not native speakers of English, and thus the collection of English data is essential to those who study the English language. However, this preference for data description is not specific to the study of foreign languages; it more or less holds true of other academic fields in Japan as well. Thus, this tendency does not seem to originate from the mere fact that English is not a native language for English linguists in Japan. It comes instead from a more fundamental source, the climate of the academic world in Japan. The Japanese academic world tends to place great value on the accumulation of results obtained from academic research, rather than on the discovery of new perspectives and the construction of new theories. The accumulation of results necessitates the maintenance and continuation of a given perspective or theory. It is generally considered valuable and ethically good to continue studying a specific topic within a given framework for a long period of time. Alternation is viewed as inconsistent or even as morally reprehensible. (This view probably stems from the conviction that the continuation of one line of work is good — as a popular Japanese proverb indicates, "Perseverance will win in the end.") Therefore, much value is placed on data-description oriented studies conducted within fixed frameworks. This atmosphere in Japan is clearly in contrast with that of the American academic world, which is disposed to place much more significance on innovation and the discovery of new theories, i. e., on theory-construction oriented studies. This inclination towards data-description oriented study in Japan will point to answers to the following questions about English linguistic study in Japan past and present: (i) why did Japanese scholars of English linguistics adopt the standard theory of generative grammar so enthusiastically in the 1960s and 1970s? and (ii) why has the tendency appeared since the mid-1980s for some Japanese linguists to depart from the rigid version of the present generative theory (so-called Government-andBinding theory)? The standard theory in the 1960s and 1970s was engaged in formulating transformational rules which were intended to account for various constructions of English sentences. This task had to be undertaken in order to show that transformation theory could cover a wide range of linguistic

12

Heizo Nakajima

(or at least English) phenomena, and therefore could take the place of the preceding major linguistic theories, such as American structuralism. The formulation of various transformational rules called for the discovery of new data to support the postulates. The accumulation of new data was readily amenable to the Japanese academic climate of data-description oriented study. Japanese linguists welcomed works based on the standard theory as a source of new data, and simultaneously, the theory itself as a means to describe them. Data-description oriented study, however, is not consistent with the recent tendency of generative grammar to try to proceed toward the formulation of Universal Grammar in an even more straightforward way. The pursuit of Universal Grammar requires the comparison of data pertaining to a particular range of constructions among various languages rather than the discovery of new data pertaining to various constructions in a particular language. Recent generative theory does not necessarily answer the need for providing new data for English, and points away from data-description oriented study. Those who are interested in datadescription oriented study, more specifically, in collecting English data, are consequently losing interest in the current version of generative grammar. Now that a discrepancy is developing between linguistics in the United States and English linguistics in Japan (as a consequence of the differences in focus and in the academic climates of the two countries), English linguists in Japan must choose their future direction. Several possibilities suggest themselves. One is to follow consistently the tradition of the Japanese academic world, and to continue describing data in frameworks which are considered appropriate for this purpose, for example, in the framework of the standard theory of generative grammar. This approach may have some advantages, at least domestically: it will contribute to the accumulation of data on particular topics and on various constructions. Such results may be used, for example, in compiling dictionaries of English linguistics like those described in section 3.1. These works, being rare in other countries, may be valuable internationally as well. Another possibility is to try to keep in step with recent theories of linguistics in the world by applying them to Japanese, which, having different characteristics from Indo-European languages, is therefore expected to provide some intriguing ideas and insights into linguistic analyses. Furthermore, Japanese linguists should try to expand these theories through proposals put forward to explain Japanese data; thereafter these

Introduction

13

proposals should be applied to English and other languages. The research strategy should be reversed from the previous one, the English-to-Japanese program, to the new one, the Japanese-to-English (and to other languages) scenario. This approach will be welcomed by the recent theory of generative grammar which aims for Universal Grammar. It will not only contribute to the development of Universal Grammar, but also shed new light on the study of English. This last effect will also be welcomed by scholars of English linguistics. Still another possibility is to attempt to develop new theories or approaches which fit the tradition of linguistics in Japan (i. e., the datadescription orientation), and simultaneously have potential as theories of Universal Grammar. The traditional inclination, which cannot be easily dismissed, may be of advantage for the construction of new approaches, because it provides fresh viewpoints which are apt to be overlooked by people who have been educated in the tradition of Western linguistics. New approaches seek, in conceptually and technically different fashions than other theories, systems in which the data of a particular language (e. g., English) are to be described comprehensively and adequately. These approaches may also have implications for the description of other languages and for Universal Grammar. These last two possibilities are considered worthwhile, and they are probed in the papers in this collection. Some of the articles compare Japanese and English within the framework of current generative theory, and attempt to make a theoretical contribution to Universal Grammar. Others seek new approaches to English, and propose analyses of English syntactic phenomena based on those approaches. This volume is, thus, an attempt to achieve two goals simultaneously. On the one hand, it presents the present state of English linguistics in Japan, while on the other, it suggests and explores a direction in which English linguistics in Japan should proceed in the future.

5. Summary of the contents The papers are presented in alphabetical order of the authors' name. The contents are summarized below, with the frameworks of the papers specified. Shuji Chiba discusses the topic of present subjunctives in a descriptive and GB-theoretical approach, and shows that contextual features of

14

Heizo Nakajima

lexical items cannot be narrowly localized. Whether complement clauses will be in the present-subjunctive mood or not depends, in most cases, on the intrinsic properties of the lexical heads which take the complement clauses. However, Chiba points out that there are cases where such a local determination is impossible, and that other elements cooccurring with lexical heads, such as their specifiers, complements, modifiers, and other heads, play a crucial role in deciding for or against the use of subjunctives. A mechanism is proposed which transfers the feature of subjunctive mood from other elements to heads. This feature transference, it is suggested, is generally possible between elements that are in agreement relations. Chiba's treatment of subjunctives will thus provide additional evidence for the agreement mechanism assumed in Government-andBinding theory. Hajime Fukuchi deals with phenomena which he calls "syntactic localization". By syntactic localization, the author means syntacticosemantic dissociation resulting from the preference for syntax over semantics. Semantic units are not expressed by corresponding full-fledged syntactic units, but by more abbreviated or compact syntactic units. These compact syntactic units are in most cases only representatives or local parts of full-fledged syntactic units; hence, syntactic "localization" of larger semantic units. Four instances of syntactic localization are examined: concealed-propositional relative constructions, concealednominalizational NP constructions, continuative restrictive relative clauses, and S-control constructions. Based upon data from authentic sentences, the author attempts to explicate mechanisms which produce various sorts of syntactic localization. This method of argumentation is an instantiation of traditional Japanese approaches to linguistic description. Naoki Fukui attempts to elaborate the notion of barrier within relativized X-bar theory proposed in Fukui (1986). Relativized X-bar theory makes crucial use of the distinction between lexical and functional categories, and, on the basis of this distinction, relativizes the maximal barlevel of projections and the possibility of their recursion. The author proposes a definition of Blocking Category (and barrier) which is also relativized in another sense. Instead of Chomsky's contextual definition of Blocking Category of a single type, he claims that there are two types of Blocking Category, "strong" and "weak" barriers: X"s (XPs) are strong barriers, which function as barriers independent of their syntactic context, and yield strong effects as barriers, whereas non-L-marked X's are weak barriers whose barrierhood depends upon the configuration in which they

Introduction

15

occur, and whose effects as barriers are relatively weak. The relativized definition of barrier, in conjunction with relativized X-bar theory, ensures the effect of Chomsky's (1986 a) adjunction condition that adjunction is possible only to non-arguments, and of his stipulation that IP, though being a non-argument, does not allow for adjunction. Kunihiro Iwakura discusses, in the framework of Government-andBinding theory, the dissimilarities between the head Ν and V/A in their capacity to take certain types of complements. V and/or A, but not N, for instance, may take towg/z-constructions, raising constructions, ECM constructions, and finite clauses without the complementizer that. To account for such discrepancies, Iwakura proposes two principles, the Maximal-Projection Principle, which requires non-null maximal projections to be governed, and the Nondistinct-Governor Principle, which essentially prohibits nondistinctness in the feature composition between governors and their governees. He also proposes to relativize the definition of the notion of government depending upon whether a governor is Ν or V/A. Iwakura argues in favor of the necessity of the notion of lexical government, which tends to be considered unnecessary in recent Government-and-Binding theory. Masayuki Ike-uchi assumes the Dynamic Theory of Syntax originally proposed by Kajita (1977). The dynamic theory is basically different from Chomskyan generative grammar in that it postulates a non-instantaneous model of language acquisition rather than an instantaneous model. Then, rules and structures are supposed to be "extended" in the process of acquisition in conformity with certain general laws of transition. (Some instances of the dynamic-theoretic treatment of English constructions are reviewed in James McCawley 1988, particularly chapter 22.) Assuming the framework of the dynamic theory of syntax, Ike-uchi presents an analysis of English descriptive genitives such as a women's college. He claims that in the child grammar descriptive genitives are derived from simple prenominal adjective structures, through a tree-grafting rule, modeled on attributive adnominal structures, but that in the adult grammar they are directly generated by a derivative phrase-structure rule. The author also supplies descriptive properties of these structures, and general ideas of the dynamic theory of syntax. Taro Kageyama's paper analyzes light verb constructions in Japanese and English with a view to providing additional support for the Modular Morphology theory which has been developed in Shibatani — Kageyama (1989) and Kageyama — Shibatani (1989). As opposed to both strong lexicalism and transformationalism, the Modular Morphology theory

16

Heizo Nakajima

claims that word-formation processes take place in syntax as well as in the lexicon, and that their outputs are globally constrained by a set of general morphological conditions that apply across different modules of grammar. To handle light verb constructions, Kageyama proposes two word-formation operations, incorporation and abstract incorporation, both of which make essential use of syntactic notions like Case and thetarole and are thus placed in syntax. The modular approach explains, among other things, the parallelisms between these syntactic word-formation rules and the familiar lexical compound formation. Yoshihisa Kitagawa challenges the standard GB-theoretical attempt to unify the two opacity conditions, the Specified-Subject Condition and the Nominative-Island Condition, in terms of the binding principles which refer to the notion "subject". Instead of assimilating the NominativeIsland Condition to the Specified-Subject Condition, he proposes to incorporate the Specified-Subject Condition into the Nominative-Island Condition by supposing that opacity in binding is created by lexical heads assigning Cases. The two island conditions are collapsed as the Case Island, which basically states that the maximal projection of the Caseassigner of a given element (anaphor or pronominal) constitutes its binding category. The enforcement of the Case Island crucially depends upon the VP-internal subject hypothesis, which the author has maintained throughout his work. The Case Island is further elaborated into the Lexical-Case Island by taking account of the dichotomy between lexical and non-lexical Case marking. On the assumption of a parametrized dichotomy of Cases in English and Japanese, the Lexical Case Island makes it possible to account for various interesting and traditionally recalcitrant phenomena of binding in English and Japanese. Susumu Kuno deals with a variety of multiple quantification sentences within his well-known framework, Functional Syntax. After criticizing purely syntactic approaches to sentences of this kind, such as May's (1985) theory, which depend upon such syntactic notions as c-command, Kuno introduces eight non-syntactic principles which make essential use of semantic, discourse-based, and pragmatic factors. These factors are, of course, not specific to multiple-quantifier sentences, but have been well-motivated in his previous work. They interact with each other to determine the relative ease of wide and narrow interpretations of quantifiers in a given sentence. The ease with which a quantifier obtains a wide interpretation is basically proportionate to the number of nonsyntactic principles by which it abides. Thus, the quantifier-scope phenomenon is regarded not as an all-or-nothing phenomenon, but as a

Introduction

17

graded one. Kuno illustrates many cases in which one quantifier takes a wide scope more easily than another when purely syntactic approaches predict that both of them equally, or only one of them exclusively, will have a wide scope. Heizo Nakajima assumes the Binding Path theory, which he has been developing in his recent work (Nakajima 1985, 1986 a, 1986 b). The basic idea of the Binding Path approach stems from his characterization of Government and Binding type principles as conditions on categories or "points" in a syntactic structure which are defined independently of each other, as opposed to "linear" rules in pre-GB theories. The Binding Path theory attempts to integrate those "point principles" in the theoretical construct of binding path, and to account for the grammaticality of sentences as a function of adherence to those principles on a binding path. Nakajima applies the approach to various types of dependent categories, such as wh-traces, traces left by rightward movements, parasitic gaps, w/z-in-situ, multiple w/z-questions involving WH-island violations, and Dutch w/z-traces. It is claimed that all these dependent categories are to be licensed by essentially the same principles, with basic notions germane to the principles parametrized within limited ranges. Masaru Nakamura treats so-called null-operator constructions, a major issue in Government-and-Binding theory, and claims that it is not correct to assume that all null-operator constructions involve movement of a null operator. He groups those constructions into several classes, and differentiates their treatment in several ways. While derivations of all the constructions involve movement of some elements, the types of elements to be moved are different among the classes. Tough constructions move base-generated null anaphors; too-to constructions and purpose clauses, PRO; topic constructions, topicalized elements; that-cleft sentences, focused elements; and wh-cleft sentences, wh-phrases. Parasitic gap sentences are the only construction which involves the movement of null operators. Nakamura elaborates the typology of empty categories definable in terms of the combinations of the features [ + anaphor] and [ + pronominal], taking account of their derivational history, namely, whether they are base-generated or transformationally derived. Harumi Sawada proposes an auxiliary system called Multi-stratal Specifier Analysis to deal with perplexing problems concerning auxiliaries. The Multi-stratal Specifier Analysis claims that a verb phrase consists of two types of V-projections, V' and V", both of which can be iterated. Aspectual auxiliaries (i. e., the perfective have and the progressive be) are affiliated with V" as its specifiers, and the passive be and the copula be

18

Heizo Nakajima

are allied to V' as its heads. After motivating the differentiation of the auxiliaries into the two groups in terms of several syntactic operations, the author proposes to generalize the multi-stratal structure and the syntactic operations used for its motivation cross-categorially. The Multistratal Specifier Analysis provides answers for such questions as (i) why the perfective have and the be verbs, but not ordinary verbs, can move into INFL, (ii) why English has the periphrastic do, and (iii) why the auxiliary do cannot cooccur with other auxiliaries. Sawada's paper sheds new light on the structure of the specifier. Ken-ichi Takami argues for Functional Syntax in the treatment of preposition stranding which results from WH-movement. After showing the inadequacies of purely syntactic treatments, such as those of Chomsky (1981, 1986 a), Hornstein and Weinberg (1981) and Baltin (1978), he proposes a functional constraint defined by the notion of "more/less important information". The constraint only allows the movement of the object of PP that bears more important information than other words or phrases in a sentence. An attempt is made to characterize the central notion of more/less importance precisely and explicitly, so as to overcome the criticism of functional notions being vague. Shigeo Tonoike tries to innovate, in the principle-and-parameter approach, the X-bar theory of English and Japanese clause and noun phrase structures, which he calls the extended DP analysis. The extended DP analysis claims that English and Japanese clause and noun phrase structures have parallel three-layer structures, with clauses composed of the three maximal projections, VP, IP, and CP, and noun phrases composed of NP, IP, and DP. Major features of this analysis are the incorporation of Japanese so-called case-markers of noun phrases into DP as the head D and the nominal head I, and the integration of such elements as also, even, only and their Japanese counterparts as specifiers of maximal projections. These analyses lead to the claim that Japanese does not have a subject defined as "noun phrase in the Spec of IP"; so-called subjects in Japanese are nothing more than adjuncts. This and other claims in the article provide explanations for apparent differences between English and Japanese, such as the possibility of multiple subjects, the overt presence or absence of articles, and so on. Tonoike argues that those apparent differences all follow from one substantial parametric distinction of word order.

Non-localizable contextual features: Present subjunctives in English* Shuji Chiba

1. Introduction In transformational grammar, the structures in which verbs can appear have been described, especially since Chomsky (1965), mainly by means of descriptive devices such as subcategorization features and contextual features, as well as phrase-structure rules. At present some interesting work is available on some general characteristics of the constructions in which verbs (and adjectives) can occur, which can be considered to show the general appropriateness of these devices. However, as for Chomsky's idea, developed in Chomsky (1965), that contextual features are all narrowly localizable, some counterexamples have been pointed out by such linguists as McCawley (1968,1973), Kajita (1968, 1976), and O t a - K a j i t a (1974: 163-647), regarding both strictsubcategorization features and selectional features. According to Chomsky's hypothesis of localizability, in the case of selectional features, the features [Masculine] and [Human], for example, should be assigned to lexical categories such as Ν and A, not to major categories such as NP and S (see Chomsky 1965: 7 5 - 1 0 6 , 120-123). In the case of strict-subcategorization features, on the other hand, only the category symbols that are dominated by VP should be relevant to the strict-subcategorization of verbs (see Chomsky 1965: 90 — 106). This means that no elements outside of VP, for example the subject NP, can be relevant to it. This hypothesis should also predict that we can explicitly describe whether verbs can take that S or whether S as their object, by assigning contextual features such as [ + that S] and [— whether S] to each verb as one of its lexical properties.

* This paper is based on a talk I gave at the meeting of the Tokyo English Linguistic Circle on October 15, 1988. I am especially grateful to Masaru Kajita and Takao Yagi for their invaluable comments. I also would like to thank Heizo Nakajima for reading an earlier version of this paper and making helpful theoretical suggestions. I alone am, of course, responsible for any remaining errors.

20

Shuji Chiba

However, as mentioned above, it has been shown that this hypothesis cannot be wholly maintained. For example, McCawley (1968: 133 — 134; 1973: 66 — 67) pointed out that selectional restrictions cannot be correctly formalized by referring only to the head of NPs, but that reference must also be made to the entire N P including the modifiers of the head, as is clearly shown by examples such as the following: (1)

a. My neighbor is the father of two. b. * My buxom neighbor is the father of two. c. *My sister is the father of two.

That is, as McCawley says, sentence (lb) violates the same selectional restriction as does sentence (lc), but the violation of the selectional restriction in (lb) has nothing to do with the head noun, since (la) contains no selectional violation. Similar examples can be found in Bach (1968: 116): (2)

a.

The one that seemed most likely to turn out to be a friend was anxious to go. b. *The one that seemed most likely to turn out to be a table was anxious to go.

In the above examples, the crucial difference between (2a) and (2b) which is responsible for the difference in grammaticality is that of the predicate nouns friend and table in the relative clauses modifying one, the head of the subject NPs. These examples, therefore, also show that selectional restrictions cannot be completely decided by intrinsic features of the head of NPs, such as [Masculine] and [Human], but by those of the whole of the NPs. 1 Concerning examples such as those in (1) and (2), however, Kajita (1976) points out that they may not constitute crucial counterexamples to the hypothesis about localizability of feature assignment. Thus, he states: It must be noticed that, in such examples as (1) and (2) [ = examples (1) and (2) above], the NP in question has a specific head. Namely, the head neighbor in (1) is unspecified as to the feature [Masculine], and similarly the head one in (2) is unspecified as to [Human]. When the head is unspecified as to a certain feature, the value of the feature is usually decided by that of the modifiers of the head. This kind of phenomenon may have to be treated by a mechanism which Weinreich (1966: 429 — 32) called 'transfer feature'. If so, [Masculine] and [Human] in (1) and (2), respectively, which are the properties of buxom and friend/table, respectively, in deep structure, can be taken to be transferred to the entire N P later by the

Non-localizable contextual features

21

application of semantic interpretation rules. Thus, those features can be localized in deep structure. [My translation, S.C.] (Kajita 1976: 255 — 254).

This is the reason why examples such as (1) and (2) need not be considered genuine counterexamples to the hypothesis in question. Kajita (1968: 97 — 110), however, also shows that there are some cases in which such an explanation cannot be resorted to. As one of those linguistic facts, let us take a case of subcategorization of nouns in terms of the feature [ + S]. Consider the following: (3)

C a shame Λ a surprise It would be < *a snake > for you to leave so early. * blood \*Chicago J

Given only these examples, one might think that such subcategorization can be lucidly made. For example, nouns such as shame and surprise have the feature [ + S], but nouns such as snake, blood, and Chicago do not. However, this is not true, as shown by such examples as the following: (4)

a. * It b. It c. It d. *It

would be a situation for freshmen to take five courses. would be a normal situation for freshmen to take five courses. would be normal for freshman to take five courses. would be a normal snake for freshmen to take five courses.

As (4a) shows, the noun situation does not usually permit a sentential subject. However, when it is modified by adjectives like normal, it can do so. This is closely related with the fact that the adjective normal itself can take a sentential subject, as shown by (4c). But this does not mean that we can obtain a grammatical sentence with a sentential subject, whenever the noun in question, whatever it is, is modified by adjectives such as normal, as (4d) shows. Therefore, contextual features such as [ + S] can be decided neither by only the head noun nor by only the modifiers; they must be decided by the entire NP containing both of them. This is one real counterexample to the localizability hypothesis of features. 2 However, there is a way in which the idea of a transfer feature, which Kajita suggested for examples like (1) and (2), can also be used to solve the problem of examples such as (3) and (4). That is, we can assume that nouns such as situation are unmarked for the feature [S], although such nouns as snake are negatively specified for the same feature. Let us further assume that when a head which is unmarked for a certain feature is modified by a lexical item which is positively marked for the same

22

Shuji Chiba

feature, that positive feature is transferred to the position of the head, turning the unmarked feature into a positive one, as suggested by Kajita. On the other hand, when the head has no modifiers to change the unspecified feature into a positive one, we assume that the unspecified feature is changed into the corresponding negative feature. Let us assume at the same time that, when the head has already been negatively marked for the feature in question, that feature's value cannot be changed from (—) to ( + ) , even if there is a proper modifier having the latter value of the feature, as in (4d). With this mechanism of an expanded version of the transfer feature, 3 we can explain such examples as (3) and (4), as well as (1) and (2), in a similar way. In the following sections, I shall point out that the problem of localizability of contextual features which has been mentioned above can also be found in the case of present subjunctives in English, and that the mechanism of feature transfer introduced above is also helpful in solving that problem.

2. Subjunctive-taking verbs, nouns, and adjectives The verbs emphasized in the following sentences are generally called "present subjunctive verbs": (5)

a. They maintain that she resign immediately. b. The regulation is that no candidate take a book into the examination room. c. It is important that he come.

As is well known, to get acceptable sentences containing presentsubjunctive verbs, we need, in the main clause, a proper verb, noun, or adjective which can trigger the subjunctive verb in the embedded clause, such as maintain, regulation, and important, as in (5a —c). (Henceforth we shall call such embedded clauses "subjunctive clauses", for convenience's sake.) 4 More examples of such lexical items can be found in the following lists:5 (6)

Subjunctive-taking verbs: advise, beg, command, demand, entreat, forbid, grant, hint, insist, mandate, necessitate, order, postulate, recommend, stipulate, urge, vote, warrant.

Non-localizable contextual features

(7)

(8)

23

Subjunctive-taking nouns: assumption, basis, claim, desire, expectation, hypothesis, intention, law, necessity, objective, prerequisite, recommendation, suggestion, trend, understanding, view, wish. Subjunctive-taking adjectives: anxious, best, critical, desirable, eager, fair, good, imperative, keen, logical, mandatory, necessary, obligatory, preferable, relevant, sufficient, urgent, vital, wise.

Looking over these items, we notice some common semantic characteristics shared by them. They can be represented as "will", "wish", "expectation", "concession", "imagination", "demand", etc. Although it is not easy to pick only one of these as a representative semantic property of subjunctive-taking lexical items, we can roughly denote it as "will", following Onions (1965) and Chiba (1987: 5).6 However, let us assume here instead, following the idea of David Pesetsky (see note 6), that subjunctive-taking lexical items have the feature "irrealis event" and that they semantically select the category "irrealis event". Let us further assume, as I did in Chiba (1987: 26), that subjunctive clauses have the feature [ + Subj] (for subjunctive mood). If we adopt the concepts of Canonical Structural Realization (CSR) and Context Principle, the close relationship between categorial selection (c-selection) and semantic selection (s-selection) (see Pesetsky 1982: 35, 181; Chomsky 1986 b: 86 ff.) concerning present subjunctives can be represented by the following generalization: 7 (9)

If a predicate s-selects the semantic category "irrealis event", then it c-selects (subcategorizes) CSR (irrealis event) = S .8 [ + Subj]

That is, those lexical items which semantically select the category "irrealis event" can (or should, according to the lexical items and dialectal differences) categorially select an embedded S with the feature [ + Subj], and the subjunctive clause is assigned the theta-role "irrealis event" by those lexical items. With this brief observation of semantic characteristics of present subjunctives as a background, let us proceed to the main theme of this paper.

24

3.

Shuji Chiba

Some cases of non-localizable contextual features for subjunctive-taking lexical items

3.1. The relevance of subject The verb say, as Stockwell (1977: 1 5 - 1 6 ) and Gazdar et al. (1985: 76) have pointed out, may be considered one of those verbs which cannot allow subjunctive clauses. Indeed, there would be very few people who would accept without reservation examples such as the following: (10)

Uohn says that the one who wears the ring be offered as a sacrifice.

However, interestingly enough, if we replace the subject of this sentence with other proper NPs, we can get well-formed sentences such as the following: (11)

(The Beatles are pursued by a mysterious Eastern religious sect because of the ring Ringo Star wears.) The law of the religion says that the one who wears the ring be offered as a sacrifice.

This shows that although the verb say is intrinsically one of those lexical items which do not allow subjunctive clauses, it can be turned into one which allows them, if a proper subject is chosen such as law {of the religion). The reason why this is possible is probably that say, combined with the subject law {of the religion), for example, can mean something like 'provide' or 'require', and therefore it will acquire the semantic feature [ 4- irrealis event] and thus be able to c-select a subjunctive clause. 9 That is to say, examples such as (10) and (11) show that the subcategorization of verbs as to their ability to c-select subjunctive clauses cannot be fully decided without taking into consideration the semantic properties of the subject NP. 10 Here let us propose a general mechanism by which verbs such as say are allowed to c-select subjunctive clauses in examples such as ( l l ) . n First, notice that the (head of the) subject N P of (11) consists of a special noun which itself can c-select a subjunctive clause. The fact that law, as well as nouns such as rule, really has the feature [ + S ] can be shown by such examples as the following: [+subj] (12)

a. The Federal law that tests be free of racial or sexual bias notwithstanding, test-making is an unregulated industry.

Non-localizable contextual features

25

b. It would be very desirable to pass a law that all high school students be educated so that they become computer literate to some extent before graduating. c. There is a standing rule in golf-clubs that every one replace the turf which he cuts up. d. Present-day English is very strict about the rule that each finite clause have its overt subject. Thus we can presume, following the case of examples such as (4) in Section 1, that in (11) the feature [ + S ] of the subject Ν Ρ is transferred [ + Subj]

to the predicate verb say, enabling the latter to c-select a subjunctive clause. 12 As a general mechanism which enables such feature transfer, we can point out "SPEC [ = specifier]-head agreement". As is well known, there is an agreement relation between the subject and the main verb of a tensed clause. Strictly speaking, the subject can bear an agreement relation to the verb through the intermediate category A G R in INFL. Thus, an agreement rule connects the A G R element of I N F L with the subject, and a subsequent syntactic rule (e.g., V-raising to I [ = INFL]; see Chomsky 1986 a) takes care of the agreement relation between A G R and the verb. The relation between the subject and A G R is referred to by Chomsky (1986 a: 24) as "SPEC-head agreement". (Notice that the syntactic structure of a sentence can generally be represented as something like the following: [ c - ... [c C [ Γ N P [ v I [VP V ...]]]]], where the subject N P is the specifier, and I is the head of I" [ = S].) Thus, we can assume that feature transfer from the subject to the verb as in (11), is an example of "feature sharing" between two categories standing in a relation of SPEC-head agreement. As another case of SPEC-head agreement, we can point out the agreement relation between an adjectival modifier and its nominal head, as in German, 1 3 although it is not so overt in the case of English. Here, remember that we have already introduced in Section 1 a case of feature transfer from a modifier to its head, concerning sentences such as (4b) [= It would be a normal situation for freshmen to take five courses]. Thus, if we consider modifier-head agreement a case of SPEC-head agreement, both (4b) and (11) could be explained by the same mechanism of feature-sharing based on the relation of SPEC-head agreement. Besides SPEC-head agreement, we have other cases of agreement, such as head-head agreement and head-complement agreement. Therefore, if

26

Shuji Chiba

our explanation above of examples such as (4b) and (11) is right, it would naturally predict that a similar phenomenon of feature transfer would also occur in those cases of agreement. This prediction can indeed be borne out, as will be shown in the following sections. The following are some more examples in which a proper subject makes it possible for say to c-select subjunctive clauses:14 (13)

a. Prisoners Union rule says that no member of an iron or steel workers union be permitted to repair a sawed-off bar without approval and participation of a representative of the cell occupant. b. A few weeks ago, I read in the Bulletin that there were to be given Chinese classes in Cranston. The article also said that a person had to be 18 years old or over, and that he or she not be going to high school to attend these classes. c. The saturation constraint then amounts to saying that no feature value be left undetermined.

3.2. The relevance of adverbial modifiers In this section, let us turn our attention to another case of non-localizable contextual features for subjunctive-taking verbs, that is, the case where the semantic properties of adverbial modifiers of verbs have important effects on the subcategorization of those verbs as to their ability to take subjunctive clauses. The main point of this section is to show how such a case can be accounted for by the mechanism suggested in the preceding section. Consider the following examples: (14)

a. IThe widow wrote that Ball be given part of her property. b. Ball is arrested for the murder of a rich widow. The widow wrote in her will that Ball be given part of her property.

As example (14a) shows, the verb write, when it governs a subjunctive clause, usually makes the sentence unnatural. However, when it is modified by a proper adverbial phrase such as in her will, as in (14b), we get a more natural sentence. Just as we examined in the previous section a case in which the verb say can be turned into one of those verbs which can take subjunctive clauses, if it occurs with a proper subject, we have here a case in which the verb write, modified by a proper adverbial phrase, turns into one type of verb of this kind. This is another case

Non-localizable contextual features

27

which shows non-localizability of contextual features for subjunctivetaking verbs. If we posit here that write is negatively specified for the contextual feature [ S ], the fact that (14b) is far better than (14a) would remain [ + Subj]

unexplained (unless we resort to a more drastic feature-changing mechanism by which we can change a negatively specified feature into the corresponding positive feature). Let us, therefore, suppose that write is unmarked for that feature, as in the case of say. The remaining problem is to explain how the feature transfer is conducted in this case. Following the expanded idea of agreement introduced in the previous section, let us suppose that the feature in question is transferred from the PP in her will, originally starting from the noun will, to its head wrote through the mother node VP. The agreement relation involved in this case can be considered to be that of head-complement agreement. As another case in which a similar mechanism is involved, let us consider the following examples: (15)

a. We add to this requirement that the selection procedure be psychologically plausible. b. We have the option of adding as a necessary condition for membership in C that an element have the feature composition [aV, PN, ...].

In (15a, b), the subjunctive clauses are syntactically the object of the verbs add and adding, and they are not the complement clauses of the nouns requirement and condition. On the other hand, the triggers of the subjunctive clauses are not the verbs but these nouns. This is clearly shown in examples such as the following: (16)

a. * We add that the selection procedure be psychologically plausible. b. * We add to this fact that the selection procedure be psychologically plausible.

The examples (16a, b) confirm our supposition that, in (15a, b), requirement and condition play a crucial role in enabling the present-subjunctive verb to occur. According to the mechanism of feature transfer mentioned above, the positively marked feature for the subjunctive clause, which is originally assigned to requirement and condition, is eventually transferred to the head position occupied by add(ing), which we assume to be unmarked for that feature.

28

Shuji Chiba

As an alternative to this explanation, one might suggest a more semantically oriented one, such as the following (see Chiba 1987: 33 — 34). That is, in (15a, b), the embedded clauses are the object of the verb add(ing), as mentioned above. Semantically, the contents of these object clauses in (15a) and (15b) are interpreted as being added, respectively, "to this requirement" and "as a necessary condition". The embedded clauses of these sentences, therefore, can be interpreted as a kind of "requirement" and "condition", respectively. This means that (15a) and (15b) have in effect the readings represented in (17a) and (17b), respectively: (17)

a. We add to this requirement the requirement that the selection procedure be psychologically plausible. b. We have the option of adding as a necessary condition for membership in C the condition that an element have the feature composition [aV, ßN, ...].

Different though these two explanations seem to be, we can regard them as being in effect the same. Namely, (17a, b) can be taken to be a result of the application of semantic interpretations to (15a, b) respectively, by the help of the mechanism of feature transfer. 15

3.3. The relevance of modals As another case in which not only the head of a VP but also its modifiers must be taken into consideration in attempts to subclassify verbs and adjectives concerning their ability to govern subjunctive clauses, we can take up examples in which modals play a crucial role, such as the following: (18)

Why is (13b) ungrammatical? Apparently, when 0 is itself composed of conjuncts 0] and 0 2 they must both be f actored by the SD "Χ ΑΧ Β X", and it must be true of both of them that T(Ah Bj) be logically equivalent to 0,. b. *It is true of both of them that T(Ah B,) be logically equivalent tO 0;. a.

In both of these examples, the subjunctive clause is governed by the same adjective, viz., true. Where, then, does the difference in acceptability of these sentences come from? The adjective true is unspecified for the contextual feature [ S ], and if it appears in a structure in which no [ + Subj]

Non-localizable contextual features

29

transfer of the relevant feature is possible, then it eventually gets the negative feature [— S ]. Thus, it cannot c-select a subjunctive clause, [ + Subjj

as in (18b). On the other hand, if it occurs with the root modal must, as in (18a), then the positively marked feature [ + S ], which can be [ + Subj]

assumed to be one of its inherent properties, is transferred from must to true, enabling the latter to c-select a subjunctive clause. Hence the grammaticality of (18a). The feature transfer in this case means in effect the semantic combination of the root modal must and the adjective true, resulting in a subjunctive-triggering complex predicate meaning something like "to require". 16 Thus, the examples in (18) clearly show that the subcategorization features for subjunctive-taking lexical items cannot be localized to the head of major categories such as NP, VP, and AP. The same is true of the following examples: (19) a. In considering BW defense, it must be recognized that a number of critical meteorological parameters be met for an aerosol to exhibit optimum effect.17 b. *It was recognized that a number of critical meteorological parameters be met for an aerosol to exhibit optimum effect. Notice incidentally that, in (18a) and (19a), there is a kind of headhead agreement relation between the modal must, on the one hand, and true in (18a) and recognized in (19a), on the other. This can be shown by a structure such as the following: (20)

... [ y [i must] [Vp be [ AP [A true] ...]]] where I is I N F L and Γ is its single-bar projection.

In the following section we shall look at another case of feature transfer between two lexical items connected with the relation of head-head agreement. Note also that the modal must itself cannot suffice as a subjunctivetaking lexical item, as shown by such examples as the following, in which the main verb (in 21a) and the adjective (in 21b) are negatively specified for the relevant feature: (21)

a. *You must think that she take a nap. b. *It must be clear that he come to the party.

The reason why (21a, b) are ungrammatical is that a negatively marked feature generally cannot be changed into the corresponding positive

30

Shuji Chiba

feature, as suggested in Section 1. The modal, therefore, must always find the proper partner to transfer its feature to. In order to make clearer the characteristics of the "proper partner", we need further empirical study.

3.4. Coalescence of two verbs In the last section we saw a case in which two lexical items, each of which cannot itself be realized as a subjunctive-taking predicate, coalesce to form a complex predicate which can govern subjunctive clauses. From this observation and our discussions in the preceding sections, one might guess that this coalescence of lexical items may not be restricted to those cases taken up above. For example, one might suspect that two (or more) verbs (or adjectives) can coalesce to form such a complex predicate, and this can indeed happen, as the following examples show: (22)

a. *Bill brought it about that Harry go or be allowed to go. b. *Bill will bring it about that Harry go or be allowed to go. c.

I \ am, I Bill to bring it about that Harry go or be [ordered j allowed to go.

As (22a, b) show, bring about itself usually does not allow a subjunctive clause to follow. However, when it is preceded by another proper verb such as ask or order as in (22c), it can c-select a subjunctive clause. The reason why this is possible is that the contextual feature [ + S ], [ + Subj]

originally assigned to the verb ask or order in this case, is transferred to bring about, which is unspecified for the feature in question, turning the latter predicate into a subjunctive-triggering one. In example (22c), the predicates which enable bring about to c-select a subjunctive clause are themselves predicates which can directly trigger it, as shown by the following examples: 18 (23) However, the presence of a predicate which can freely allow a subjunctive clause, as in the case of ask or order, is not a necessary condition for the feature transfer to occur in examples such as (22c). This can be shown by the following examples:

Non-localizable contextual features

31

'expect Bill (24)

a

"

^ ' tried hard '

^ about that Harry go or be allowed

want Bill to go. expect

I

1

tried hard f*

^wry

go or be allowed to go.

want J Although the main verbs in these examples cannot be grouped into predicates, such as ask, beg, demand, and insist, which take subjunctive clauses freely, 19 there is a good reason to consider them subjunctivetriggering predicates. Chomsky's (1981: 19) statement about the verb want is helpful here: Sentence (4iv) [ = "the students want that Bill visit Paris"] is not idiomatic English, but we m a y assume this to be an accidental gap reflecting properties that are not part o f core grammar; thus assume (4iv) to be fully grammatical at the relevant level o f abstraction, as in the analogous case of (2ii) [ = "the students prefer that Bill visit Paris"] and as in languages otherwise similar to English. 2 0

We can find a similar idea in Pesetsky's (1982: 674) suggestion of distinguishing "possible but not actual" selection from impossible selection. Thus he classifies verbs like want, yearn, would like, need, and mean, as well as verbs like prefer, desire, and wish, among the subjunctive-taking verbs. The examples in (22c) and those in (24a), although they differ, as we saw above, in the productivity of the main verbs concerning subjunctive clauses, show the same kind of feature transfer, i. e., feature transfer from the main verb in the matrix sentence to the embedded main verb, as schematically shown in the following structure: I } (25) ... V . . . [ s . . . V [ s . . . V . . . ] ] In (25), the arrow shows the direction of feature transfer, and this is an example of head-head agreement. Before closing this section, let us add an example sentence which is a realization of the structure in (25) in which both the second and the third Vs are subjunctive verbs: (26)

I am asking that Bill bring it about that Harry go or be allowed to go.

32

Shuji Chiba

3.5. The relevance of the whole of the NP As another case showing inadequacy of the localizability hypothesis about subjunctive-taking lexical items, let us consider next some examples in which the semantic content of the whole of the N P must be taken into consideration. Consider the following examples: (27)

a. It's [MP a good thing] that he recognize his faults. b. It is not [NP a necessary feature of X'-theory] that it be set up this way. c. [ NP The important point] is that both be satisfied with the adjustment. d. It is [Np a matter of prime importance] that these elements be defined relatively to the other elements and to the interrelations among all of them. e. Perhaps [NP the most controversial part of the claim made in rule Γ] is that there be a lexically-entered noun that corresponds to each of the kinds in the set denoted by the newly-created noun.

Notice first that the direct triggers of the subjunctive clauses in the examples above are not the head nouns of the NPs enclosed in brackets, but the unitalicized words, which are each one of the modifiers of these head nouns in each sentence. This can be confirmed by the fact that we get an ungrammatical sentence if we replace these words with some nouns or adjectives which cannot allow subjunctive clauses, as in the ungrammatical version of the following sentence: (28)

It is [NP a matter of some disappointment to me ] that still many of my own countrymen *be/are too shortsighted to ascribe any symbolic significance to the plight of a minority, such as artists, in any social order.

This suggests that, in the examples in (27), what semantically governs the subjunctive clauses are the words emphasized or, rather, the whole NPs containing them, although syntactically the subjunctive clauses are governed by the head nouns of the NPs. These examples are a subjunctive version of the examples in (4), which we repeat here as examples (29a —d): (29)

a. *It would be a situation for freshmen to take five courses. b. It would be a normal situation for freshmen to take five courses.

Non-localizable contextual features

33

c. It would be normal for freshmen to take five courses. d. *It would be a normal snake for freshmen to take five courses. Just as we suggested in Section 1 that the relevant feature in (29b) is transferred from the modifier to the head, so we can assume here that the feature [+ S ] is transferred from the emphasized noun or adjective [ + Subj]

in (27) to the head noun, which is considered unmarked for the feature [ S ]. The only difference is in the types of modifiers involved: in (27), [ + Subj]

besides examples (27a —c), in which the modifier in question constitutes a pre-nominal adjective as it is in (29b), other examples are included in which the crucial modifier is in the post-nominal PP, i.e., examples (27d —e). As further examples in which the trigger of a subjunctive clause is found in the position of the post-nominal modifiers, we can include sentences in which the modifier in question is a relative clause, as in the following: (30) a. They believe that [NP the only form of independence [s that is possible or desirable for a woman ]] is that she be dependent upon her husband, or if she is unmarried, on her nearest male relative. b. [ NP The minimum [s that can be expected from such a reclassification]] is that it be able to accomodate all of the types of Mätzner-Jespersen.21 The semantic mechanism responsible for the correct interpretation of these sentences, especially with respect to the correlation between the subjunctive clause and its trigger, would be similar to the one which we can posit for examples (2a, b), which we repeat below as examples (31a, b): (31)

a.

The one that seemed most likely to turn out to be a friend was anxious to go. b. *The one that seemed most likely to turn out to be a table was anxious to go.

In these sentences, the features [ +Human] and [ — Human] which are assigned to the predicate nouns of the most deeply embedded clauses-are raised one by one to the next upper clause, and finally form a part of the semantic properties of the subject NPs of the main clause, contributing to the difference in grammaticality of these two sentences, as pointed out by Ota and Kajita (1974: 398).22

34

Shuji Chiba

3.6. Syntactic properties and semantic properties In the preceding sections we have pointed out that there are some cases in which the subclassification of lexical items according to the possibility of their allowing subjunctive clauses must be done by taking into consideration a rather subtle correlation between syntactic and semantic properties of the constructions containing those lexical items. In other words, we have endorsed, with evidence concerning the present subjunctive, Ota — Kajita's (1974: 397) claim that, while the strict subcategorization of lexical items is clearly a syntactic problem, the restriction of occurrence of various types of complement clauses is more or less tinged with semantic characteristics. 23 Ota and Kajita's claim can also be supported by the following interesting fact. First consider the following example, which is a paraphrase of the sentence I bought the toaster for Mary : (32)

I acted to bring about a certain practical result, namely, that the toaster be in my realm, and so, capable of being subsequently given to Mary by me.

In (32), the subjunctive clause is governed by the noun result. However, we must note here that result itself cannot usually be followed by a subjunctive clause, as shown by the following example: (33)

Nobody predicted the result that the toaster *be/would be in my realm.

Notice also that, as pointed out in Section 3.4., bring about does not itself usually allow a subjunctive clause, either (cf. 22a, b). This leads us to assume that, in (32), the original trigger of the subjunctive clause is the main verb acted, although it cannot itself syntactically govern a subjunctive clause, or any i/iai-clause, for that matter, and that the subjunctive feature of that verb is transferred to the lower verb to bring about, forming in effect a subjunctive compound predicate acted to bring about. Direct evidence for this, as we have already shown in Section 3.4., can be found in examples such as the following: (34)

a. I acted to bring it about that the toaster be in my realm. b. / acted to bring it about that the man meet with an accident.

(Example 34a may sound a little unnatural, because of its content and style.)

Non-localizable contextual features

35

To explain the occurrence of the subjunctive verb in such examples as (32), we must further assume that the subjunctive feature is further transferred to the head of the object NP a certain practical result. The noun result, now having been turned into a lexical item positively specified for the subjunctive feature, causes the occurrence of the present subjunctive in the embedded clause. The mechanism of feature transfer suggested above can be represented in the following structure: (35)

I ...v...

y...

tf } V[NP... Ν

[s...

v...

]]]...

The categories connected by arrows are in head-head agreement relations. It thus becomes evident that a lexical item which seems not to be able to c-select a subjunctive clause becomes a full-fledged subjunctive lexical item, i.e., a lexical item which can govern a subjunctive clause both syntactically and semantically. This means, as we have suggested, that, if other conditions are satisfied, the lexical item which syntactically governs the embedded clause in question need not necessarily be one which can freely take a subjunctive clause. Even if it is not positively specified for the subjunctive feature, the mechanism of feature transfer can turn it into a subjunctive-governing lexical item. A similar mechanism could also be posited to explain the occurrence of present subjunctive verbs in such examples as the following: (36)

a. Lord John Russell moved a resolution to [NP the effect [s- that Mr. Salomons be ordered to withdraw]]. b. The first is a requirement that such and such a factor be a constituent of a certain type. The second is a requirement to [NP the effect [s- that such and such a factor be a certain terminal element]]. c. The filter must be stated in [NP such a way [s- that it not require that Ns be Cased which are contained in ...]].24 d. We then require that Ρ be related biuniquely to pr, in [NP the sense [s that Ρ be uniquely recoverable from pr, and pr be uniquely constructible from P]].

The interesting thing about these examples is again that the nouns which (syntactically) govern the subjunctive clauses, i.e., effect, way, and sense, cannot usually take them. (In 36c, strictly speaking, the subjunctive clause is first governed by such and then they together modify way.)

36

Shuji Chiba

Semantically speaking, the reason why the subjunctive verbs are possible in these sentences seems to be that the subjunctive clauses can in effect be interpreted as appositive clauses to resolution and requirement in (36a, b) respectively, and as the object of state and require in (36c, d) respectively. That is, the subjunctive clauses can be connected to their original triggers by the "bridge expressions" to the effect, in such a way, and in the sense, which syntactically appear to work as barriers to such a connection. Thus the apparent discrepancy between the syntactic and semantic characteristics of these sentences disappears. We suggest, again, that this can be possible through the mechanism of feature transfer, which transfers the contextual feature [ + S ] from the [ + Subj]

original trigger to the head of the NPs which syntactically governs the subjunctive clause in question.

4. Conclusion In the above discussions, we took up the case of present subjunctives as another example in which contextual features cannot be localized. In previous studies on present subjunctives, in both the frameworks of traditional grammar and transformational grammar, it has generally been supposed that the question of whether or not a certain verb, noun, or adjective can be followed by a subjunctive clause can always be ascribed to the intrinsic properties of that lexical item. This supposition, however, turns out to be untenable, if we look more closely at the characteristics of present subjunctives. For example, we find some examples in which subject NPs or verb modifiers in VPs play a crucial role in deciding whether the verbs in question can take subjunctive clauses or not. In some other cases, we notice that the whole of the NPs containing modifiers such as adjectives or relative clauses must be taken into consideration. Subjunctive clauses, since they are a kind of complement clause, must be governed by a head verb, noun, or adjective. In simple cases, the head verb, noun, or adjective is also a trigger of those subjunctive clauses. In some cases, however, the crucial trigger may not be the same as the head which syntactically governs the subjunctive clause; instead, it may be (part of) the modifiers of that head. In still other cases, the head may form a kind of compound predicate with other words, which as a whole allows or requires a subjunctive clause to follow.

Non-localizable contextual features

37

As an answer to the question of how these phenomena can occur, we proposed an extended version of feature transfer. We suggested that a positively marked contextual feature can be transferred from a modifier to its head, from the subject to the verb, and from a verb to another verb, if the goal of the feature-transferring is unmarked for the feature in question. We further suggested that the feature transfer is generally possible between those categories which stand in agreement relations, in the broader sense of the term. In this paper we assumed that feature transfer is only possible to the position of a lexical item which remains unspecified for the feature in question. Thus we, in effect, suggested a tripartite division of verbs, adjectives, and nouns, concerning the subjunctive feature (i.e., those which are positively marked for it, those which are negatively marked for it, and those which are unmarked for it), without being able to consider in detail the remaining problem of how the group of unspecified lexical items is defined. This problem, and another which is closely related to it, the problem of the extent to which the relationship between categorial selection and semantic selection holds in the case of the present subjunctive, still remains unsettled. In order to answer these questions, we would need a deeper semantic analysis of subjunctive lexical items. I hope that this paper will provide a basis for further study of present subjunctives and, in particular, of general properties of contextual features. Notes 1. A similar discussion is made by Jackendoff (1972: 18 — 19) concerning the following examples: (i)

a.

I ate something that was the result of what Bill acknowledged backing process.

b. *I ate something syntactic

to be a new

that was the result of what Bill acknowledged

to be a

transformation.

Note also that Newmeyer (1986: 1 1 3 - 1 1 4 n. 16) states: Chomsky has continued to maintain (in class lectures) that at least some selection is syntactic, citing as evidence sentences like *the boy who was turned by magic into a swarm of bees dispersed. 2. For other examples, see Kajita (1968: 9 4 - 1 1 0 ; 1976: 2 5 3 - 2 6 9 ) . Especially about the problem of localizability for contextual features, Kajita pointed out an interesting fact about such verbs as serve, help, and suffice, showing the following examples: (i)

a. The ice melted. b. * The ice served to melt.

38

Shuji Chiba (ii)

a. The ice chilled the beer. b. The ice served to chill the beer. These examples show that such "a verb as melt used intransitively ... cannot appear in the sentential complement of serve while such a transitive verb as chill in [(ii)] can" (Kajita 1968: 103). This means that "the constituents within the embedded sentence are, indeed, sometimes relevant to the strict subcategorization of verbs" (Kajita 1968: 104). This problem has also been taken up by Bresnan (1972: 49 n. 5), who introduces Chomsky's suggestion on this problem "that selectional restrictions may be a more appropriate means of describing these phenomena". Later, in the current linguistic study within the framework of the government-andbinding theory, Lasnik (1988: 5 — 6) reminds us of this unsettled problem and suggests that "[t]he difference between the well-formed and ill-formed examples with serve is describable in thematic terms: the subject of the complement of serve must be an instrument", giving further examples such as the following: (iii)

3. 4. 5. 6.

7.

8.

a. *John served to eat lunch. b. * Edison served to invent the light bulb. c. * Susan served to accept the offer. I owe this idea to Heizo Nakajima. For a comprehensive description of the present subjunctive in present-day English, see Chiba (1987). For fuller lists, see Chiba (1987: 3 - 4 ) . In Chiba (1987: 5) I also suggested the term "non-fact". Among other terms representing the subjunctive mood are "thought-mood", "non-committal mood", and "irrealis". For further details, see Jespersen (1924: 317), Kruisinga (1931: 27 ff.) and Zandvoort (1975: 88). According to Heizo Nakajima (personal communication), David Pesetsky called it "irrealis event" in his class lectures at M I T (fall of 1988). Canonical Structural Realization (CSR) was first proposed by Grimshaw (1981: 174); it is a function mapping semantic categories onto syntactic categories. For example, CSR (object) = Ν and CSR (action) = V. The context principle is also originally Grimshaw's idea. She states (1981: 178) that "if a predicate selects a semantic type, it is subcategorized for the CSR of that type. I will call this the Context principle." The generalization stated in (9) is directly based on Pesetsky's (1981: 185) version of the context principle: "If a predicate s-selects a semantic category C, then it c-selects (subcategorizes) CSR(C)." We must be careful not to forget the fact that the meaning of a verb does not completely determine its subcategorization, as shown by the following examples (from Postal 1974: 366): chose 1 selected} Larry to be my assistant, picked J

1 1

chose

1

*selected> that Larry be my assistant. *picked J For similar comments and related examples, see Williams (1974: 38), Wasow (1976: 282), L i n e b a r g e r - S c h w a r t s - S a f f r a n (1983: 381), Gazdar et al. (1985: 32), and Chiba (1987: 1 1 4 - 1 1 5 ) .

Non-localizable contextual features

39

9. This fact seems to be closely related to the presence of infinitive constructions in which the embedded subject refers to a contextually understood person or persons to whom the direction or instruction is addressed, such as the following: (i)

a. She said to meet her at the station. b. It says on the bottle to take a spoonful every four

hours.

(Both of these examples are from the Longman dictionary of Contemporary

English2

s. v. say v. 6.) For further details, see Chiba (1987: 192 n. 10) and Pesetsky (1982: 1 9 8 - 1 9 9 ) . As to the well-known semantic affinity between present subjunctive clauses and for-to clauses, see Emonds (1970: 198; 1985: 1 2 3 - 1 2 4 , 297), and Bonney (1976: 21 - 2 2 , 48 n. 23). 10. This is in contradiction with the generally supposed assumption that "verbs do not subcategorize for subjects" (Chomsky 1981: 26), if the difference between the indicative and subjunctive forms of the main verb of the embedded clause should really be taken care of by the subcategorization of verbs. Notice in this connection that Kajita (1968: 9 6 - 9 7 , 1 0 0 - 1 0 1 ) and O t a - K a j i t a (1974: 2 7 1 - 2 7 7 ) have also pointed out that the internal structures of subject NPs are sometimes relevant to the subcategorization of verbs, showing such examples as the following (Ota —Kajita 1974: 271, ex. 59): (i) a. His carelessness surprised me. b. His carelessness caused the accident. c. That he was careless surprised me. d. *That he was careless caused the accident. See also Kajita (1968: 169 n. 55). 11. I am indebted to Heizo Nakajima for suggesting the theoretical treatment of nonlocalizable subjunctives which is presented below, in particular for the idea of feature transfer in terms of various types of agreement. 12. It should be noticed here that even if the subject of say has the feature [ + S ] as in I 4 Subj]

the case of rule or law, feature transference does not occur, if say is not used with the meaning "to order or command", as in the following examples: (i) a. Lagrange's law says that its velocity is/*be equal to the square root of the product of the depth times the acceleration due to gravity. b. But, obviously, the laws do not always work. One in particular — a rule that says that inflation goes/*go up when unemployment goes down — seems to have broken down. Note also that in older English the verb say was positively marked for the feature in question (at least for one of its uses) as shown by the following example: (ii)

Say vnto hym that he drynke to you in the name of good peace. (OED s. v. say v. B. 5. α 1533 Ld. Berners Huon lxxxiii. 260)

For other examples, see Visser (1966: 837 — 838), and for a similar observation, see Traugott (1972: 150). 13. Note Chomsky's (1986 a: 24) assumption that "any category α agrees with itself and with its head". 14. According to my informants, (13a) is not completely grammatical for some reason which I do not yet know. Notice also that, in (13b), we seem to get an ungrammatical sentence if we replace the first i/wr-clause after the verb said with a subjunctive clause, as in the following:

40

Shuji Chiba (i)

The article also said that a person *{should) be 18 years old or over, and that he or she {should) not be going to high school to attend these classes. The feature transference in (13c) is conducted not only between the subject and the main verb, but also between the main verb and the lower verb, as will be shown in Section 3.4. 15. In Chiba (1987: 34—35), I cited examples such as the following as another case in which the trigger of the subjunctive clause is not in the position of the head but in the position of its complement: (i)

Something was stirring in the 1980's. It was as recently as that decade that John Stuart Mill started a movement on the subject of women's suffrage, thus dissenting from his famous father's opinion that it was consistent with good government that women be excluded from suffrage because their interests were the same as those of men.

T h a t is, I suggested that, in the second sentence of the example above, the subjunctive clause is semantically governed by the adjective good in the complement PP with good government, not by the adjective consistent, which is the head of the A P consistent with good government. T h a t the adjective good can be the trigger of a subjunctive clause is clear from examples such as the following: (ii)

a. It is good that women be excluded from suffrage. b. It's a good thing that he recognize his faults. (Leech 1971: 108) The suggestion that consistent is unspecified for the subjunctive feature and that it can be assigned the corresponding positive feature by its complement would be supported by the following examples: (iii)

a.

That women be excluded from suffrage is not consistent with what they demanded. b. That women be excluded from suffrage is not consistent with their intention. c. *That women be excluded from suffrage is not consistent with the description in the textbook. d. *That women be excluded from suffrage is not consistent with the political fact about the country that Stuart Mill told us before. T h a t is, one may reason that (iiia, b) are grammatical because the complements of consistent contain subjunctive-triggering lexical items, i.e., demanded and intention, respectively, while (iiic, d) are ungrammatical because no such lexical items are contained in them. However, on the other hand, the fact that the following example is grammatical may be taken to suggest that consistent itself is positively marked for the feature in question: (iv) Cf.: (v)

That women be excluded from suffrage is not consistent earlier. *He said earlier that women be excluded from

with what he said

suffrage.

C o n f r o n t e d with contradictory evidence, I must leave this problem unsettled. 16. As a p h e n o m e n o n similar to this, one can consider what Kajita (1977) calls "syntacticosemantic overlapping". F o r example, Kajita (1977: 67 — 69) explains the case of toughmovement. T h a t is, this transformational rule is applicable not only to such lexical items as easy, hard, tough, difficult, impossible, and dangerous, but also to complex predicates as in the following examples:

Non-localizable contextual features

41

a. It does not require specialized knowledge to read the book. b. a. (Ü) b. Cactus takes deep plowing to get rid of. (iii) a. It is far, beyond the scope of this study to examine the question. b. The question is far beyond the scope of this study to examine. (iv) a. It was over his capacity to bear the burden. b. The burden was over his capacity to bear. For other examples of syntactico-semantic overlapping and his idea of a "dynamic model" for explaining those phenomena, see Kajita (1977); see also McCawley (1988 b: 731 - 7 5 3 ) . (i)

17. In (19a) the modal must can be replaced by other modals such as should and would. However, we need more empirical study before we can say with certainty that must is not the only modal that can transfer the subjunctive contextual feature. 18. Note that, in the case of order, there seems to be dialectal variation concerning the acceptability of sentences such as (23). For further details, see Chiba (1987: 123 — 130), James (1986: 126), and Nichols (1987). 19. As examples of grammatical sentences in which the verb expect appears with a subjunctive clause, see the following: (i)

a. It was not expected, surely, that a man stand idly by when his property was in imminent danger of being blown to fragments. b. First, it would be unreasonable to expect that there be a foolproof test which enabled us to decide whether any given constraint is grammatical or perceptual. c. Never was it to be expected that one seek to make anything come to pass via the humdrum process of labor.

For the special relevance of the factor of style to the occurrence of present subjunctives, see Chiba (1987: 9 - 1 1 ) . 20. See also grammatical sentences containing want and subjunctive clauses, such as those in (ii) and (iv) in note 22. 21. See note 19. 22. The examples below might be considered another case in which the triggers of subjunctive clauses are in relative clauses: (i)

a. [ NP What matters] is that researchers be prepared to formulate and listen to coherent criticism of their ideas, and that there be a sufficiently sound shared vision of goals and ontology to permit progress. b. "[ NP What is important] is that there be a strategic Arab determination to retaliate against the Zionist enemy", he said. c. [ NP All that is needed] is that one of the intermediate stages constructed in the process of decoding be perceptually complex. d. [Np All we can suggest] is that a teaching programme be designed in such a way as to ...

In these sentences, however, the relation between the subjunctive clause and its trigger is not as indirect as it might seem. Once we grasp the semantic characteristic of the construction of these sentences, i.e., that the post-copular ί/ιαί-clause is to be understood as the real content of what is represented as what or all, it is not necessary to regard them as one of those cases in which the subjunctive clause and its trigger are

42

Shuji Chiba only indirectly connected. What is interesting about these constructions is that they seem to facilitate the occurrence of the present subjunctive verb, as suggested by the following examples: (ii)

(iii)

a. *I want that you be happy. b. What I want is that you be happy. c. All I want is that you be happy. a. *They discussed that John be given permission to leave the country. b. What we discussed was that John be given permission to leave the country. c. All that was discussed was that John be given permission to leave the country.

Notice that we also get an ungrammatical sentence if we replace the subjunctive verb be in (iiia) with such verb forms as would be or had been. Concerning discuss, Sells (1985: 32) states that it "looks like a verb that should take S' argument, but [that] it only takes NP (as in We discussed the problem but not * We discussed that there was a problem)". However, considering the fact that want can also govern the present subjunctive in such examples as those in (iv) below, one might think that the crucial factor is not the special constructions themselves, but the "syntactic distance" between the verb in question and the subjunctive clause: (iv)

a. / want only that you be happy. b. John wants very much that the fighting stop. c. / never wanted it that they be treated like that. d. John wants it of Bill that he clean the house.

The same seems to be true of the following examples: (v)

a. THe cared that those he had be clean. b. What he cared about was that those he had be clean. c. He cared only that those he had be clean.

For other similar examples, see McCawley (1988 b: 97). 23. Their claim can also be supported by the fact, pointed out by Kajita (1976: 258 — 261), that in English there is a group of verbs which can take as their object whether S as well as that S, such as the following: (i)

know, find out, ascertain, establish, testify, say.

These verbs, however, take whether S only in some specific contexts, one of them being a negative environment such as the following: (ii)

(iii)

a. b. c. d. e. a. b. c.

/ don't know whether S. It is impossible to know whether S. I had no way of knowing in advance whether S. Whether S, I can't say. Whether S, I am not competent to say. Before we know whether S, ... Whether S is more difficult to establish. It is too early to say whether S.

Another environment in which the verbs in (i) can be followed by whether S is the one in which whether S occurs with expressions which do not necessarily specify the realization of the content represented by these verbs; for example, contexts which show future, demand, trial, purpose, etc., as in the following:

Non-localizable contextual features (iv)

a. b. c. d. e. f. g. h. i.

43

Time will testify whether S. One will soon enough ascertain whether S. You'll need to know whether S. ..., demanding to know whether S. Try to find out whether S. / want to find out whether S. He asked someone to find out whether S. It is very important to know whether S. It would be necessary first to check Fred's blood (in order) to ascertain whether or not it was of the same type as Papa's.

On the other hand, if the verbs in (i) accompanied by whether S occur with verbal forms which show the realization of the content, such as the affirmative simple-past form and the affirmative present-perfect form, the sentences become ungrammatical, as shown by the following examples: (v)

a. * Bill testified whether John took a bribe. b. * Bill has said whether John took a bribe.

To describe the generalization shown above, Kajita (1976: 260 — 261) introduced the feature [Realized], stating that the verbs in (i) cannot take whether S if they are combined with elements which have the feature [+Realized], while they can if they are combined with elements which have the feature [ — Realized] or which are unspecified as to the feature [Realized]. Thus, the feature [whether S] cannot be considered an intrinsic feature of those verbs themselves; instead, it must be considered a feature to be assigned to larger elements containing elements which have the feature [Realized], In other words, the feature [whether S] is an unrealizable contextual feature. 24. Notice that must is not obligatory in this sentence. Cf. (18a) and (19a) in Section 3.3. See also the following examples: (i)

a. The rule must be formulated/stated/written/stipulated in such a way that the subject and the object be exchanged in the subordinate clause. b. The rule was formulated/stated/written/stipulated in sich a way that the subject and the object be exchanged in the subordinate clause.

Syntactic localization phenomena in English* Hajime Fukuchi

1. Introduction In currently prevailing views of linguistic analysis, syntactic structure and semantic structure are levels which can and should be independently represented. But in fact syntax and semantics are closely related. Those who believe in the autonomy of the two linguistic levels do not go so far as to claim that they are completely separate from each other. Instead, there seems to be an implicit agreement that syntactic rules have certain correlates in semantic formation rules. By saying that syntax correlates with semantics, I mean that some principled correspondences are observed between syntactic structure and semantic structure, or, in more traditional terms, that there is a firm connection between form and meaning. Indeed, the meaning of a sentence consists of a rich variety of semantic ingredients. But it is highly reasonable, and it has actually been a widespread linguistic practice, to assume that, aside from scopal factors such as quantifiers, negatives, etc., the fundamental semantic relation of a single sentence can be reduced to a proposition which is made up of a predicate and arguments. The internal structure of a proposition varies in accordance with the property of the predicate and the semantic roles the arguments bear, producing many types of conceptual structures (in the sense of Jackendoff 1983). And yet, regardless of its type, the essentials of this minimal logico-semantic relation are retained when it is mapped onto the syntactic structure: a single proposition matches with an S or S-like category and an argument is characteristically realized as an NP; a complex of propositions emerges syntactically as a complex sentence. In addition, the dependency relations (predicational or modificational ones) which hold in semantic structure can in large part be represented in parallel in syntactic structure. If this syntactico-semantic parallelism is one of the principles that determine the nature of language, phenomena which in some way lack it * In writing this article, the author has greatly benefited from suggestions and criticism from Heizo Nakajima, the volume editor, and Masayuki Ohishi.

46

Hajime Fukuchi

are expected to be of linguistic interest in their own right. It is an accepted fact that language is a rule system that tolerates certain syntacticosemantic discrepancies within the limits of possible semantic interpretation. Grammarians have been constantly intrigued by constructions which exhibit such discrepancies in some way or other. Logically speaking, syntactico-semantic discrepancies develop in two opposite directions: in one, semantics affects syntax to "loosen" wellestablished syntactic patterns, and in the other, syntax influences semantics by camouflaging semantic relations through syntactic structure. The former type of discrepancies are numerous, and have been extensively investigated from such viewpoints as "reanalysis", "blending", etc. A few examples will suffice to illustrate the point: (1) (2)

I guess (that) there is some discontent among the members. There is some discontent among the members, I guess.

In (1), I guess syntactically constitutes a matrix clause which takes a complement clause, and has corresponding semantic functions. But under some conditions this clausal expression ceases to be a full matrix clause and comes to function simply as a marker of the speaker's judgment concerning the certainty of the content of the subordinate clause, i.e., a "hedge" which weakens the assertion of the subordinate proposition. This semantic factor sometimes triggers a syntactic deformation of "downgrading" the matrix clause to a parenthetical expression as in (2).1 I guess in (2) might be clausal in structure, but functions like a sentence-modifying adverb such as possibly. As Kajita (1977) suggests, downgrading also works on NP structures: (3) (4) (5)

He handed me what he had produced out of his pocket. The man entered the cockpit carrying a gun and a can of what the crew took to be gasoline. He was behaving what I could only describe as strangely.

Ordinarily, a free relative behaves syntactically as an NP, as in (3). But as (4) shows, there are cases where it is possible to interpret a free relative as simply modifying the clause-final predicate nominal. The semantic factor which causes the downgrading is the same as that found in such idiomatic free relatives as what is called, what they call, etc., that is, a hedge that dilutes description, like perhaps. In (5), the apparent free relative has completely lost the syntactic status of NP, modifying the

Syntactic localization phenomena in English

47

adverb that follows. Lakoff (1974) reports an extreme case of syntactic loosening: (6)

John invited you'll never guess how many people to the party.

The emphasized part of this example, which is structurally a clausal remnant, is irrevocably amalgamated with the matrix clause and semantically functions to modify many. The semantic factor in this amalgamating process is presumably a sort of "intensification". The second type of discrepancy has not attracted much attention from grammarians. But instances do exist where a semantic relation fails to be duly expressed in syntactic structure. One such case that has been discussed by many linguists is what is widely known as "raising" constructions: (7)

John seems to have won the game.

The syntactic structure of (7) cannot be said to reflect its propositional structure, which can be approximately represented as (8): (8)

[John won the game ] seem

Although several versions have been proposed of (8), what is common to them is the point that the matrix predicate is analyzed as taking a clausal, rather than a nominal, argument. That is, seem, which agrees syntactically with an NP John, has invariably been thought to be combined with a clausal argument, from a semantic viewpoint. If this is correct, the propositional structure (8) is deformed to a certain extent when it is materialized in the syntactic structure (7). Raising constructions have provided diverse sources of interest for grammarians, but this particular syntactico-semantic discrepancy has drawn grammarians' attention throughout thirty years of transformational-generative research activities. From the pioneering study by Rosenbaum (1967) to the recent Government-and-Binding analyses, researchers have continuously assumed that (7) and (8) are to be derivationally related by a movement operation which, roughly speaking, "raises" the subordinate subject John to the matrix-subject position. Within the framework of Government-and-Binding theory, the plausibility of the raising operation, reduced to "move a", comes from the interaction of the principles of Case-assignment and Theta-criterion (Chomsky 1981). But from another point of view, this movement can be regarded as an instantiation of a general process which for some reason brings about fluctuations in the parallelism in question.

48

Hajime Fukuchi

Aside from technical issues, it seems to me that the syntactic operation of raising has the effect of producing syntactically "tight and compact" constructions for semantic contents that would otherwise be realized as syntactically loose configurations. Concerning (7), the matrix predicate seem, which semantically is the predicate of the propositional content of the subordinate clause, is syntactically associated with a part of the proposition, i.e., a nominal expression {John). This may be said to contribute to creating a syntactically tight and compact structure, in the sense that verbs in general prefer to follow a nominal, rather than a clausal, subject or, in other words, that NP + VP is considered a more basic syntactic concatenation than S + VP. If the semantic content of (8) is straightly mapped as in (9), the result would have to be syntactically loose and, indeed, ill-formed: (9)

* That John won the game seems.

Of course, there is a related well-formed construction like (10), in which seem follows a nominal subject and the intended semantic relation of (8) is syntactically guaranteed: (10)

It seems that John won the game.

But (10) is still of loose syntactic structure, because the matrix subject cannot be considered the full (argument) NP that John is in (7).2 If raising has the property that is suggested above, we may say that the operation is fundamentally rooted in a general linguistic process which works to squeeze semantic structures into well-established syntactic frames, a process in which syntax influences semantics to adapt semantic contents to syntactically oriented configurations. This process has, at the same time, the side effect of camouflaging the semantic relations by the syntactic structures, hence of creating syntactico-semantic discrepancies, yet preserving the logical relations which are necessary for practical semantic interpretation. In this study, we are concerned with another aspect of this general process which may be called "syntactic localization". Syntactic localization can be conceived of as a way of realizing a sort of metonymy, in which a subpart of a constituent is made to stand for the whole constituent. More specifically, when an element A is to be semantically associated with a constituent B, syntactic localization makes it possible to produce a syntactic concatenation in which A is immediately associated with a subpart C of the constituent B. Let us represent the semantic association and the syntactic association by j—j and i—i respectively:

Syntactic localization phenomena in English

(11)

... A j . . . [b ... c ...γ... AJ...

(12)

... A, ... [B ... C ...], ... As ...

49

If the semantic relation of (11) is straightforwardly mapped onto syntactic structure, a syntactico-semantic correspondence is fulfilled as in (13): (13)

... A \ . . C . . . I . . .

A\...

But syntactic localization produces a syntactic association as in (14) for the semantic relation of (11):

(14)

... Aj ... [B ... Q ...] ... A; ...

This brings about a mismatch of i and j in the function of B, from which some syntactico-semantic discrepancy may be assumed to arise:

(15)

... A j i... [ B . . . q ...y... A j i...

In what follows, I will examine four complex-sentential constructions which display some syntactico-semantic discrepancies, and try to clarify in detail the properties of the localization process that contribute to creating syntactically oriented structures at the price of semantic integrity.

2. Concealed propositions Both syntactically and semantically, the relative construction appears to behave as a kind of noun. This has lead grammarians to analyze it as a complex N P which is formed from a head N P and a modifying subordinate clause. The modifying function of the subordinate clause can be — performed under the condition of referential identity — the condition that the relativizer (relative pronoun or relative adverb), which plays a grammatical role in the subordinate clause, indicates the same thing or person as the head N P (antecedent). Take, for example, (16): (16)

John wants to meet [the novelist [who he admires]]

In (16), who functions as the object of admires and is understood to refer to the same person as the novelist, which, on the other hand, fulfills the role of object with respect to meet. In this sense, the relative clause is said to modify the head N P by describing the attribute of the antecedent.

50

Hajime Fukuchi

2.1. There are cases, however, where such a familiar analysis cannot work satisfactorily. A well-known example is found in the relativization of a part of an idiomatic phrase as in (17): (17)

Our advisor was pleased with the headway we made.

It might appear that the matrix predicate pleased with is followed by a complex NP which is composed of a head NP (the headway) and a subordinate clause which modifies it {we made). But this subordinate clause does not exhibit a characteristic that ordinary modifying expressions possess: modifiers are optional. If the relative clause of (17) is missing, the sentence will be anomalous, unlike (16), which can make perfect sense without the qualifying clause: (18) (19)

*Our advisor was pleased with the headway. John wants to meet the novelist (who he admires).

So, (17) cannot be treated as a regular relative construction for the reason that the apparent head does not act as a full argument NP, as is also seen from (20): (20)

* ( The) headway was satisfactory.

To deal with the peculiarity in question, Schachter (1973) proposed analyzing a relative construction such as (17) as "headless", at least in underlying structure: (21)

Our advisor was pleased with [the

[we made headway]]

In addition, he offered an idea for the derivation which raises an NP of the subordinate clause {headway) to the empty head position to shape a syntactically well-formed structure, a "promoting" analysis in contrast with the "matching" analysis that works for (16). From a semantic view, the point of the promoting analysis is to look upon the construction as not nominal, but propositional, in the sense that (17) is virtually equivalent to a complement construction like (22): (22)

Our advisor was pleased [that we made headway ]

and to regard the fake head-modifier relation as the surface manifestation of a propositional content. Let us call a relative construction like (17) a "concealed proposition". Schachter seems to be inclined to extend this promoting analysis to all cases of relativization, but this is untenable. If we consider (16) again, it will immediately be clear that the analysis runs into serious difficulties,

Syntactic localization phenomena in English

51

because (16) would have to be derived from an unjustifiable underlying structure as (23): (23)

John wants to meet [the

[he admires the novelist]]

Note that (23) violates a cooccurrence requirement that meet be a transitive verb which takes a nominal, not a clausal, object. This is corroborated by the fact that it is not possible to make up a corresponding complement construction: (24)

*John wants to meet [that he admires the novelist]

Therefore, we will have to resort to classifying relative constructions into at least two kinds: one is essentially nominal, amenable to the matching analysis, and the other is concealed-propositional, receptive of the promoting analysis. 2.2. Given the existence of concealed-propositional relatives, I will not discuss the matter simply as a subcase of English relativization. I will argue instead that concealed-propositional relative constructions are produced as a result of an effort to bring a separate type of construction, say, complement constructions, close to regular (nominal) relative constructions. Suppose that the intended meaning of (17) is represented as something like (25), a complex of propositions: (25)

Our advisor was pleased (with) [we made headway ]

This configuration perfectly warrants that the matrix predicate is semantically associated with the entire subordinate proposition: (26)

Our advisor was pleased (with)1 [we made headway^

But from the semantic fact that headway naturally represents the content of the subordinate clause (McCawley 1988 b: 736), and from the syntactic requirement that the preposition with of the matrix predicate be followed by a nominal, rather than a clausal, constituent, the matrix predicate comes to be directly associated with a nominal expression of the subordinate clause, i.e., the focus nominal headway, by syntactic localization: (27)

Our advisor was pleased withj [we made headway^

Relativization, perhaps through promoting, occurs as a means to realize the effect of the syntactic localization, as in (28): (28)

Our advisor was pleased with^ [the headways [we made]}

52

Hajime Fukuchi

A syntactico-semantic discrepancy is assumed to arise from the mismatch of i and j in the headway we made. Make headway is a verbal idiom that means "progress". This implies that when it appears in a sentence, the nominal part of it does not refer to any specific object. As for (28), it must be the case that the headway cannot have any referent in the real world. However, it is also true that the idiom is not so amalgamated an expression that its components have completely lost their inherent syntactic properties. Headway is still a noun, as is seen from the fact that it can be prenominally modified to form a full-fledged NP: (29)

They made little/steady headway.

If nouns in general are referring when used as NPs, it is not unreasonable to assume that headway, which is exploited as non-referential in the idiomatic phrase, has a potential ability to refer to some entity, and that this potential is actualized in certain syntactic situations. Pragmatically, the propositional content of someone having made headway is conveyed as a specific event. If the content is condensed and represented by a single NP the headway in a syntactic frame, as was suggested above, the NP will be in a position to refer to the entire event, with some connotation which features the way or the extent of making headway. 3 It seems that the same can be said about the following examples, which are provided in Schachter (1973: 3 1 - 3 2 ) : (30) (31)

The careful track that she's keeping of her expenses pleases me. (*[The] careful track pleases me.) I was offended by the lip service that was paid to civil liberties at the trial. (*/ was offended by [the] lip service.)

Notice that things are not different if the degree of idiomaticity is lessened: (32)

(33)

Then, as now, officials were shocked at the ease with which an unidentified craft was able to penetrate vital and heavily defensed airspace. The event marked the first time that Iran had struck a vessel traveling under the flag of a superpower.

In (32) and (33) the relevant phrases {with ease, for the first time) might be literal, rather than idiomatic, expressions. But the component NPs are clearly non-referential, being unable to work as arguments in propositions:

Syntactic localization phenomena in English

(34) (35)

53

* Officials were shocked at the ease. *The event marked the first time.

Therefore, (32) and (33) should also be analyzed as concealed-propositional, expressed in syntactically tight and compact constructions by using non-referential NPs referringly. 2.3. The phenomenon of concealed proposition is not restricted to the relativization of idiom parts. Hawkins (1980: 41 —42) says that (36) and (37) have approximately the same meanings as (38) and (39) respectively: (36) (37) (38) (39)

I recalled the sweet child that Harry used to be. The fool that he was then became apparent to us all. I recalled that Harry used to be a sweet little child. That he was a fool then became apparent to us all.

Hawkins's observation seems correct because the sentences (36) and (37) will become anomalous if we replace the matrix predicates with ones which exclusively cooccur with a true NP: (40) (41)

*/ met the sweet little child that Harry used to be. *The fool that he was then hit me.

Again, the preculiarity of the relative constructions (36) and (37) stems from the issue of referential identity. Note that the constructions in question are cases where predicate nominals are relativized. Predicate nominals, like predicate adjectives, fundamentally function as describing the properties of the subjects. This means that they are not able to refer to any object even though they assume the syntactic form of NP. In (42), for instance: (42)

He is a fool.

the NP a fool does not indicate a specific foolish person, but simply depicts the property of being foolish. However, since NPs are intrinsically referring, a predicate nominal can conceivably be used in environments where some identity condition between two NPs should be satisfied: (43)

He is not the fool that he was ten years ago.

Indeed, (43) is by no means considered a case of concealed proposition, but the structure differs slightly from regular (nominal) relative constructions: semantically, the antecedent NP the fool, instead of working as a matrix argument, serves as the predicate that describes the matrix subject. This description happens to be identical with what is denoted by that, a

54

Hajime Fukuchi

description of the subordinate subject. Thus, the relative construction of (43) can be said to meet an identity condition which is thought to be extended from the principle of referential identity. This point is, admittedly, related to the way syntactic localization works in concealed propositions. In a sentence like (44): (44)

Harry was a sweet little child.

the semantic content is naturally conveyed with the gravity center on the predicate nominal. It is, then, quite likely that when (44) is embedded in a complex sentence, as in (45): (45)

I remembered [Harry was a sweet little child]

a clause-external element, i. e., remembered, will directly link with the predicate nominal of the subordinate clause, to the effect that what was remembered was (Harry's) property of being a sweet little child. But there is no effective way to syntactically manifest this association, except by forcing the predicate nominal to appear as a full NP which can work as a direct argument of remembered. It may be said that this is done through syntactic localization, to bring about a nominally-oriented (hence, a referent-oriented, rather than a truth-value-oriented) construction. 2.4. In the two types of concealed-propositional relative constructions discussed above, the head NPs are definite. It is difficult to offer a satisfactory explanation, but we can say at least that this is a restriction which is imposed on the head NPs when non-referential NPs are relativized: (46) (47)

Our advisor was pleased with thej*a headway we made. I recalled the/*a sweet little child that Harry used to be.

The definiteness restriction does not arise, however, when referential NPs occur in the head position: (48)

This word has exactly the sound of a heavy object that strikes something with a loud wham.

(48) does make sense even if it ends with a heavy object, without the relative clause that follows it: there is no serious problem in reading (48) as an ordinary relative construction. But semantically, the sound of is better associated with the propositional content that a heavy object strikes something with a loud wham, rather than with a single NP a heavy object. Therefore, (48) is preferably read as (49):

Syntactic localization phenomena in English

(49)

55

This word has exactly the sound of [a heavy object strikes something with a loud wham ]

If this interpretation is correct, we can say that the semantic relation of (49) is realized in the structure of a relative construction by localizing the effect of the sound of to a single argument of the subordinate proposition. We are not in a position to raise decisive evidence to show that a construction like (48) must be analyzed as concealed-propositional. However, this analysis is not simply a matter of preference. Consider the following sentences: (50) (51)

The tragedy began with demonstrators who pelted police with rocks, bottles, and smoke bombs. The affair ended with a police officer who was sentenced to four years in prison.

Verbal expressions like begin with and end with are naturally followed by NPs which denote events or actions. When events or actions are not explicitly expressed, it is usually the case that they can be implied. The sentence the concert began with a piano solo, for instance, means that the concert began with the playing of a piano solo, and his story ended with a moral is intended to mean that his story ended by pointing to a moral. Such implications cannot be observed in (50) and (51) if the relative clauses are missing. That is to say, (50) and (51) do not fully satisfy the selectional restriction between with and the following NPs, unless the relative constructions are taken to be concealed propositions. In some cases, the analysis can be supported by the form of anaphoric expression: (52)

The problem was traced to a squirrel that is thought to have scampered onto a power line in Connecticut while carrying a piece of aluminum foil. That caused a power failure that cut off electricity to more than 2,000 area homes and businesses.

In the above passage, that in the second sentence is preferably to be interpreted as a pro-form which refers to the propositional content concealed in the relative construction, rather than to a single NP (a squirrel). With respect to the examples (48) — (52), one might say that the relative constructions do not satisfactorily observe the condition of metonymy that a propositional content be practically expressible by a part of it. Indeed, the head NPs cannot be said to fully represent the contents of

56

Hajime Fukuchi

the subordinate clauses, or, in other words, the degree of localization is far higher in these sentences, compared with the cases where idioms or predicate nominals are relativized. But this semantic condition can actually be relaxed to allow the content of an event to be conveyed with a participant in the event as a "pivot", within the limits of retaining the logical relation (e. g., causal relation) of propositional contents. Note that the first sentence of (52) can be rewritten as (53), preserving the propositional content intact: (53)

The problem was traced to a power line onto which a squirrel is thought to have scampered while carrying a piece of aluminum foil.

So, given the two types of relative constructions, nominal and concealedpropositional, there is no convincing reason to deny that (48) —(52) involve concealed propositions. It seems that the relativization of referential NPs in (40) — (43) are parallel to the relativization of non-referential NPs, and as Kajita (1977: 62) speculates, the latter can be taken as a special case of the more general process which is exemplified by the former. 4

3. Concealed nominalization It may be agreed that the internal structure of typical NPs, excluding complex or heavy NPs, can be represented approximately as: (54)

Determiner + Modifier + Noun.

The syntactic nature of NPs is determined by the nouns. That is, the noun occupies the head position in NP, and the category of NP inherits the essential characteristics of the head, behaving fundamentally the same as the noun. This is reflected in the fact that the noun is an obligatory constituent of NP, whereas the determiner and the modifier are, in principle, optional. In the terminology of Government and Binding, an NP is the maximal projection of a noun. Semantically, the noun usually plays a principal role in associating the NP as a whole with NP-external elements (governors). In (55), for example: (55)

The lion jerked its long tail.

Syntactic localization phenomena in English

57

the verb takes as object the entire NP its long tail, satisfying a certain selectional restriction. But the restriction actually concerns the verb and the noun tail. (55) meets the selectional restriction because a tail is something to which the action of jerking can be applied. Thus, in ordinary NPs, the syntactic head corresponds to the semantic head, observing a sort of syntactico-semantic parallelism. 3.1. As in relative constructions, however, there are cases where the parallelism vacillates in NPs of prenominal modification. In the sentence below: (56)

The council worked for improved schools and hospitals in those communities and distributed some welfare funds.

it is needless to say that the apparent (syntactic) head of the unitalicized NP is schools and hospitals. But the conjoined nouns are not able to act as semantic head. The intended meaning of (46) is in no way that the council worked for schools and hospitals which were improved. Rather, the sentence is meant to say: the council worked to improve schools and hospitals, or the council worked for the improving of schools and hospitals. In other words, the semantic head of the NP is understood to lie in the syntactically modifying part improved, and the selectional restriction which is to hold between work for and improved schools and hospitals substantially bears upon work for and improved. A similar observation can be made for the unitalicized NPs in the following sentences: (57) (58)

The man paid the Stammers a nominal rent and "mended fences, vaccinated cows, picked fruit". She took to sending him serious magazines, reviews, pamphlets and so on, with marked passages on which she would be dying to hear his views.

In (57), the NPs in question might be somewhat intentional expressions, seeing that they are presented as a direct quotation. Nonetheless, it is inconceivable from this sentence that the man paid the Stammers fences, cows, and fruit, because supposedly they were the Stammers' possessions. What the sentence intends to state is, instead, that the man paid his rent by mending fences, vaccinating cows, and picking fruit. Then, what was really paid to the landlord was the labor the tenant expended in mending, vaccinating, and picking. With respect to (58), it is unlikely that she sent him magazines, reviews, and so on, together with (additional) passages.

58

Hajime Fukuchi

It is clear that these passages are meant to be ones which appeared in the magazines, reviews, and pamphlets he sent her. This makes it necessary to read the sentence as: she took to sending him magazines, reviews, etc., marking the passages on which she wanted to hear his views. Examples (56) —(58) are obvious cases in which the syntactic head fails to correspond to the semantic head. If it is recognized that, in addition to typical NPs which conform to syntactico-semantic parallelism, another type of NPs exists which deviates from it, then it seems advisable to analyze the following unitalicized NPs as the latter type, although the demarcation is not so clear-cut: (59) (60) (61)

Nevertheless, the lack of deeper discussion amounted to a missed opportunity. A total ban on advertising was necessary to achieve the goal of reduced smoking. Party and government bureaucrats fear lost privileges and deviations from socialist ideology.

3.2. What is particularly interesting about the peculiar NPs at issue is the semantic relation which holds between the apparent modifier and the head noun. Let us consider (56) again. I have suggested that the intended meaning of improved schools and hospitals is substantially the same as that of (62): (62)

the improving of schools and hospitals

Note that this is exactly the structure of nominalization, and the semantic relation of (62) can be reduced to the one which is contained in a proposition like (63): (63)

PRO improve schools and hospitals.

We may say, then, that an NP like improved schools and hospitals is a means of compactly expressing a propositional content in the sense that a nominalization is a "packed" or "condensed" expression of a clausal content. I will call such an NP "concealed nominalization", on the analogy of the concealed propositions discussed in the preceding section. When we compare the syntactic and the semantic structures of improved schools and hospitals, we can immediately see that a nominal argument {schools and hospitals) of the condensed proposition is "upgraded" in a way to the status of NP head, to be directly associated with the external predicate. This is an instance of syntactic localization. Recall that a concealed-propositional relative is formed by picking up an argument

Syntactic localization phenomena in English

59

from a proposition and making it the head of a complex NP construction. In a similar way, a concealed nominalization is produced by choosing an argument of a "nominalized proposition" and setting it in the head position of an NP. Through this process, the effect of the predicate which is to be associated with the propositional content of the nominalization comes to be focused on a part of it, bringing about slight unsteadiness in the selectional relation between the predicate and the NP head, as is observed in (56) —(58). In short, the localization proceeds as in (64): (64)

a. ... work for j [the improving of schools and hospitals^ b. ... work forj [the improving of schools and hospitals^ c. ... work for j j [improved schools and hospitals^

If the semantic relation of the improving of schools and hospitals is to be read in the form of improved schools and hospitals, there will be the ensuing problem of how to justify the conversion of improving to improved, i.e., the conversion of a nominal to an adjectival within the domain of an NP. This categorial change is not completely groundless, considering the phenomenon in (65) —(67) concerning a queer disagreement between the indefinite article and the quantified nouns: (65) (66) (67)

Last week, an estimated 5,000 people gathered at the cemetery. An additional 3.6 million books were damaged by water. For an arduous 3 hrs. 54 min., the cyclist pedaled up a hill.

Needless to say, an is not compatible with the plurality of the head nouns. This will lead us to a reasonable assumption that the unitalicized NPs in (65) —(67) are related in some way or other to NP structures like (68)-(70): (68) (69) (70)

an estimate of 5,000 people an addition of 3.6 million books an ardor of 3 hrs. 54 min.

These heavy NPs are not necessarily nominalizations, and the corresponding NP structures in (65) —(67) can by no means be interpreted as concealed-nominalizational. But the point of the discussion is that the articles in (68) —(70) are, syntactically and semantically, the associates of the head nouns (estimate, addition, ardor), and the fact that they remain as they are in (65) —(67) endorses the plausibility of the conversion of nominals to adjectivals to shape another type of NP constructions. Another way of thinking is to assume that suffixes, and hence categorial differences, play no major role in semantic structure. That is to say that

60

Hajime Fukuchi

there is no problem in representing the correct semantic relation of improved schools and hospitals in a rather abstract configuration as (71): (71)

(the) IMPROVE

(of) schools and hospitals

And the realization of IMPROVE as improved is taken to be simply a syntactic adjustment. Like concealed propositions, concealed nominalizations are a product of the effort to express a clausal content in a syntactically tight and compact structure, or, more specifically, in a nominally-oriented construction, without altering its truth value. However, it is only under fairly strict conditions that concealed nominalizations can work as NPs. It seems that they can occur only in the positions which immediately follow verbs or prepositions. In (72) and (73), for instance: (72) (73)

Improved schools and hospitals are difficult to work for. Picked fruit was paid to him.

the possibility is hopelessly thin that the subject NPs have readings of concealed nominalization. This indicates that concealed nominalizations cannot undergo syntactic movement operations, being able to perform their functions only when they are positioned to receive the direct influence of the predicates that induce syntactic localization. 5 3.3. When an NP has a reading of concealed nominalization, we may expect that this is manifested in some syntactic or semantic facts. And indeed, peculiar kinds of phenomena do exist which may be supposed to emerge from propositional contents that are squeezed into modifier-noun expressions. First, when NPs — and any conjoinable categories — are coordinated, there is a tendency for the conjuncts to share certain syntactic and semantic properties with one another. This tendency is difficult to characterize with precision, but we cannot deny that it exists as a linguistically significant principle that works to constrain actual sentence production in some way or other. If the principle goes so far as to involve the semantic relations which can be expected to be parallel in conjoined NPs, we can readily see that there is something strange about the coordination in (74): (74)

In the next half century, scientists fear dramatically altered weather patterns, major shifts of deserts and fertile regions, intensification of tropical storms and rise of sea level.

Syntactic localization phenomena in English

61

Of the four heavy NPs that constitute the object of fear, the first one differs from the other three in at least one respect. The last three NPs are, roughly speaking, of derived-nominal (nominalization) structure, with deverbal nouns {shift, intensification, rise) as their heads. But in the first NP this is not the case: in dramatically altered weather patterns a deverbal meaning is expressed in the modifier part instead of in the head noun. So, a parallelism can be retained if we analyze the first NP as having the semantic relation of (75): (75)

dramatic alteration of weather patterns

The same can be said about (76) if the unitalicized NP is read as (77): (76)

(77)

The beginnings of language must have been slow as the making of the first tools was slow. Hundreds of thousands of years for mastering the use of stone, bone, andflint, and the first discovered metal, then a gradual quickening of tempo leading to the crescendo of mechanization in which we live. The first discovery of metal

Secondly, it is generally the case that when an NP contains a complement or modifying clause, or a PP expression which has the corresponding functions, the target of modification or complementation is the head noun: (78) (79)

John made a vain attempt to rescue the drowning girl. John's arrival in Tokyo before the typhoon attacked the Kanto area was very lucky.

In (78), for example, the infinitival clause describes the content of John's attempt, and the adverbial clause of (79) serves to restrict the time of John's arrival. In this sense, post-nominal clauses or phrases are immediately associated with the head nouns. But this does not always obtain. Compare (80) and (81): (80) (81)

Also there is scant information on a pelvic fracture Mengele is said to have suffered in a wartime motorcycle accident. Early on, they discovered signs of an injury in the pelvic area that might correspond to a broken hip Mengele reportedly suffered in a wartime accident.

Pelvic fracture and broken hip are expressions for the same injury, but while the former has a typical NP structure which is smoothly followed by the relative clause, the latter cannot be straightforwardly modified

62

Hajime Fukuchi

through the association of the head and the relative clause. The reason is that the selectional restriction between the head and the verb of the relative clause is satisfied in (80), but not in (81), as is seen from (82) and (83): (82) (83)

Mengele suffered a fracture. *Mengele suffered a hip.

This strongly suggests that the relative clause of (81) is associated substantially with broken, making the syntactic modifier the semantic head of the complex NP. Something similar may be said of cases where an infinitival clause works as a post-nominal modifier. Consider the following passage: (84)

Given all this, is West Side Story on opera? "I say it is not", Bernstein has asserted. "It's a work on its way toward being one." But one of the salutary developments in recent years is the expanding definition of opera to include musical theater, a category that now encompasses everything from La Boheme to Sondheim's Sunday in the Park with George.

If the infinitival clause to include musical theater is associated with the head noun definition, we will get an semantic relation as in (85): (85)

The definition of opera to include musical theater is expanding.

But this is not congruent with the theme of the discourse. A more natural interpretation is that the infinitival modifies expanding to convey a semantic relations as: (86)

The definition of opera is expanding to include musical theater.

Similarly, it is conspicuous that the since-phrase of the following example is associated with sinking, not with popularity: (87)

Marcos' opponents can ill afford to have any potential supporters to stay at home. Despite the President's sinking popularity since Aquino's assassination in August 1983, Marcos remains a formidable foe.

A third type of phenomena that substantiate the reading of concealed nominalization concerns the choice of anaphoric expressions. As with concealed propositions which in discourse can be referred to by pronominal expressions suitable for clausal contents (cf. 52), concealed nominal-

Syntactic localization phenomena in English

63

izations sometimes assert themselves by being anaphorically followed by forms which have propositional flavor: (88)

"Papandreou has pushed his former centrist supporters the arms of the N.D.P.", says a respected analysis for Eurocommunist Party. " We are speaking of a shifting represents 10% of the electorate, and this shift could the outcome of the election."

over into Greece's vote that determine

In this passage, the antecedent of this shift cannot be the complex N P of which vote acts as head. From the lexical relationship, the definite description should be interpreted as referring to a part of it, i.e., shifting. If so, the possible antecedent will be an N P which comprises the lexical import of this modifier as its head: (89)

the shifting of a vote that represents 10% of the electorate

The anaphoric relation between this shift and a shifting vote is able to hold because the latter has a semantic relation as in (89). Of course, we can conceive of a context which develops as in (90): (90)

We happened to see a bird with beautiful feathers. The beauty was comparable to nothing.

Although the beauty clearly refers to the modifier beautiful, it is completely impossible to read beautiful feathers as a concealed nominalization. But with (88), because the topic of the discourse is Papandreou's losing his former supporters and this is shown in each sentence by such key expressions as pushed ... over into, shifting, this shift, it is preferable to analyze the complex N P as a concealed nominalization. In the following example, the subject of the last sentence takes exactly the form that expresses the intended semantic relation of the antecedent NP: (91)

Though the dam halted the annual flooding of the Nile that temporarily inundated some temples, it has also had a detrimental effect. The year-round availability of water has allowed increased irrigation. ... But the increase in irrigation, coupled with poor drainage, has raised the water table in the area.

64

Hajime Fukuchi

4. Continuative restrictive relative clauses As I have discussed in the previous sections, concealed propositions and concealed nominalizations are NP constructions which, instead of consisting of a head and a modifier, are semantically considered to express (subordinate) propositional contents. Concealed propositions are propositional in the sense that they can be interpreted in the same way as complement clauses; concealed nominalizations are also propositional in that they have the same semantic relations as clausal nominalizations which can be regarded as "condensed" expressions for clausal contents. Syntactic localization works to select an argument NP from the subordinate proposition and associate it directly with the matrix predicate, despite the semantic requisite that the matrix predicate be associated with the entire subordinate proposition. This process, together with ensuing structural adjustments such as promoting (in concealed propositions) and upgrading (in concealed nominalizations), contributes to producing nominally-oriented, hence syntactically tight and compact constructions. In this and the following sections, we will turn our attention to syntactic localizations in the opposite direction: cases in which a subordinate predicate which is semantically associated with the clausal content of the matrix clause as a whole is made to have a direct linkage with a single argument NP of the matrix clause. 4.1. It is well known that a certain class of non-restrictive relative clauses are associated with a phrasal or clausal content of the matrix clause: (92) (93)

Joe debated in high school, which Chuck did too. She took the children to the zoo, which was very helpful.

Although we can conceive of a number of analyses of appositive relative constructions, what must be explicated by any analysis of (92) and (93) is the fact that the unitalicized clauses are associated with something other than the immediately preceding NPs {high school, the zoo). In (92), which refers to an action denoted by the VP expressions {debated in high school), while in (93) the antecedent of the relative pronoun is understood to be an event which is described by the matrix clause {she took the children to the zoo). If we put aside the technical issue of the node to which they are to be adjoined in the tree diagram, these comma-separated relative clauses are best interpreted as simply being juxtaposed to the matrix clauses; (92) and (93) can be paraphrased as (94) and (95) respectively:

Syntactic localization phenomena in English

(94) (95)

65

Joe debated in high school; Chuck did that too. She took the children to the zoo; that was very helpful.

This observation can hold for a certain type of restrictive relative constructions as well. Compare the two relativizations in the following example: (96)

In early December, students of the University of Science and Technology in Heifei launched a wave of pro-democracy demonstrations that spread to a dozen major cities across China. Last week the same students found a quieter way to express their sentiments; at least 1,000 reportedly signed their names in souvenir albums that paid homage to the university's ousted president, Guan Weiyan. The veteran educator and physicist had been fired by the government for not exercising more control over University Vice President Fang Lizhi, an outspoken defender of political liberalization.

The first sentence of (96), in which the unitalicized clause syntactically and semantically modifies a single NP, says that students launched a wave of demonstrations and the demonstrations spread to a dozen major cities. This way of continuation may be said to characterize regular restrictive relative constructions. But the second sentence cannot be read in the same way. What paid homage to the ousted university president is not merely souvenir albums, but the students' action of signing their names in them. In other words, the true meaning of the sentence is that the students signed their names in souvenir albums to pay homage to the ousted president, or that the students' having signed their names in souvenir albums resulted in paying homage to the ousted president. This clearly shows that the relative pronoun that is not linked with the immediately preceding NP as its antecedent. Instead, the genuine antecedent is embodied by the entire matrix clause in the same way as in the non-restrictive relative clauses of (92) and (93), despite the fact that that is the relativizer most resistant to non-restrictive use. Such an S-antecedent continuation is also observed in the following examples: (97)

(98)

Paddle Artist Ahn Jaehyung beat World Champion Jiang Jailing and then defeated 1984 Asian Singles Champion Hui Jun in a game that sent the sellout crowd into a frenzy. As the leading opponent of South Africa's system of racial apartheid, the African National Congress has become the embodiment

66

Hajime Fukuchi

of the hopes and aspirations of the country's blacks. Secretary of State George Shultz has conceded that the ANC has emerged as an important part of the South African political equation that must be acknowledged. The unitalicized clause of (97) does not simply work by modifying the apparent head a game. That is, the true meaning of the sentence is not the same as: in a game that sent the sellout crowd into a frenzy, Ahn Jaehyung beat Jian Jailing and then defeated Hui Jun. Rather, the sentence is meant to say that Ahn Jaehyung's winning a game sent the sellout crowd into a frenzy, a situation where the antecedent of the relative pronoun is provided by the matrix clause as a whole. The same is true of (98). Since acknowledge is a verb which takes a complement clause, the embedded clause of the second sentence should be read as: "we must acknowledge that the ANC has emerged as an important part of the South African political equation", or "we must acknowledge the ANC as an important part of the South African political equation". Notwithstanding the semantic similarity observed between the nonrestrictive relatives of (92) — (93) and the restrictive relatives of (97) — (98), there is a conspicuous structural difference: unlike the non-restrictive relatives, the S-antecedent restrictive relatives syntactically constitute a single complex NP together with the preceding nominal expressions. This seems to be an instance of syntactic localization, whose process can be described as (99): (99)

a. [... NPP [that j ...] b. [... NPJ [that, ...] c. [... NPi \that\ ...fl

Through this process, there occurs a syntactico-semantic discrepancy concerning the association of the relative pronoun and its antecedent, or the association of the subordinate predicate and a matrix argument. But this discrepancy is not serious enough to damage the truth value of the propositions, in the same way as in concealed propositions and concealed nominalizations. 4.2. Cases in which the syntactic head of a restrictive relative construction fails to coincide with the antecedent of the relativizer are not unusual. Consider the semantic relations which hold in the banal sentences of (100) and (101), which are cited from McCawley (1981 a: 124):

Syntactic localization phenomena in English

(100) (101)

67

Tom wrote a novel which is also a great corpus of Tennessee dialect. Sam is a linguist who has a very good background in sociology.

As for (100), McCawley says that the relative clause does not simply modify a novel. In view of the function of also, which, like too, either, nevertheless, etc., normally works to relate two clausal contents, he argues that the sentence must be semantically analyzed as "Tom wrote an χ which is a novel and is also a great corpus of Tennessee dialect." In (93), the relative clause does not just describe the property of a linguist. Since predicate nominals lack referentiality, which is required for ordinary relativization, the possible antecedent of who can be found in nothing but the subject NP Sam. That is, the sentence should be interpreted as: Sam is a linguist and (but) he has a very good background in sociology. The apparent head-modifier relation in these relative constructions may be said to arise from a surface-structural requirement. Clearly, the constructions we are now discussing cannot be dealt with in the same way as those of (100) —(101), because the antecedents in (100) —(101) are particular NPs, not clausal contents. But McCawley's argumentation sheds much light on the nature of S-antecedent relatives. Consider (102), for example: (102)

Only a year ago President Sarney was hailed for imposing the freeze on prices. But the artificial restraints generated an intense consumer demand that in turn put renewed pressure on the economy.

In turn, in the unitalicized part, works to effectively connect propositional contents in the same fashion as also, too, etc. This amounts to saying that the content of the relative clause is meaningful if it is related to the clausal content of the matrix clause, not simply modifying a single NP an intense consumer demand. If we look for material in the matrix clause which is related to the relative clause through the function of in turn, it is found in the part generated an intense consumer demand. Following McCawley's notation, the second sentence of (102) may be represented as: but the artificial restraints did χ which generated an intense consumer demand and in turn put renewed pressure on the economy. Notice that χ is an entity (action) which is to be verbally denoted. But a verbal is not a referential expression. So, if the true antecedent of the relativizer is x, syntactically there will be no candidate for the head. This is why the role of the antecedent is played by proxy by an available nominal

68

Hajime Fukuchi

expression an intense consumer demand, i.e., through syntactic localization. The fact that in (101) the relative pronoun, which is to be associated with the subject NP, is syntactically attached to the predicate nominal is a consequence of the same structural requirement. Like concealed-propositional relativization, S-antecedent relativization works on idiom chunks. Let us examine (103), which is also cited from McCawley (1981a: 137): (103)

Parky pulled the strings that got me my job.

The logic of discussion is the same as in Section 2. In (103), pull the strings is an idiom which means "to be a real activator". So, the strings cannot be referring by itself. But since it takes a form of NP, it has a potential ability of being referential. This makes the NP able to act as the syntactic head of the relative clause that follows. It is easily inferred that the materialization of referentiality is extended to work for the relativization of less idiomatic expressions, as in: (104)

John gave Mary a kiss that angered her.

and for the relativization of V —NP expressions of literal meaning as in (101). Here again, syntactic localization is a process to find a possible nominal expression, and associate a clausal content with it to convey a propositional complex.

5. S-control phenomena Control phenomena have been one of the main sources of interest to grammarians since the advent of transformational analyses. As is well known, basic control relations are exemplified by sentences like (105) —(107): (105) (106) (107)

John tried [PRO to win ] John persuaded Bill [PRO to win ] John promised Bill [PRO to win ]

The elements that can control PRO are, precisely speaking, determined by the lexical property of the matrix predicate (Jackendoff 1972). In the examples above, the controller of PRO is determined from the idiosyncratic properties of try, persuade, and promise. But this idiosyncrasy is grammatically limited. Controllers are, in fact, confined to an argument

Syntactic localization phenomena in English

69

NP (subject or object) of the matrix clause, precluding the possibility of non-argument NPs controlling the missing subject of the subordinate clause. In (108): (108)

* John's sister tried [PRO to win]

PRO cannot be controlled by John because this NP is not an argument that participates in the semantic relation of the matrix clause. It is not always true, however, that the controller of PRO is provided by a particular argument NP. There are instances where PRO should be interpreted as being controlled by the propositional content of the matrix clause, instead of by a single NP: (109)

Nasa announced that part of a newly designed booster rocket had failed during a test firing at a Morton Thiokol plant, PRO causing an undetermined delay in the faltering effort to resume U.S. manned space missions.

In (109), the subject of the unitalicized participial clause must be the event which is described by the matrix clause. That is, the sentence should be read as, "the event of a booster rocket failing in a test firing caused a delay in the program of resuming manned space missions". This is quite similar to the cases of S-antecedent non-restrictive relative constructions (cf. 92 and 93). More interesting, the phenomenon is not restricted to comma-separated participial clauses. Consider (110): (110)

The Orioles lost the first six games for second-year Manager Cal Ripken, who was abruptly fired. Frank Robinson replaced him, and they lost ten more PRO to shatter an 84-year-old majorleague record.

It might be possible to read the second sentence to the effect that PRO is controlled by they (the Orioles), in conformity with a general principle of control, but semantically a better reading is obtained by associating the PRO with the content of the matrix clause: the Orioles' having lost six more games was an event that shattered a major-league record. If this is correct, we can see that the control relation in (110) will be in parallel with the S-antecedent restrictive relative constructions (cf. 98 and 99). The issue that arises is how such non-argument control is to be dealt with in the light of the principle of control. 5.1. The non-ordinary control phenomena, as pointed out above, have recently been discussed under the heading of "implicit argument". Jaeggli

70

Hajime Fukuchi

(1986), Roeper (1987), Lasnik (1988), and others tackle the issue by paying attention to the difference observed between (111) and (112): (111) (112)

*The ship sank PRO to prove the point. The ship was sunk PRO to prove the point.

From the fact that the infinitival clause can follow a passive, but not an intransitive, matrix clause, the researchers conclude, details aside, that the controller of PRO is restricted to an argument NP which bears the role of "agent", and when such an agent NP is not explicitly represented there is assumed to be an "implicitly represented" (phonologically null) agentive argument. For example, the controller in (112) corresponds to the (agentive) subject of (113): (113)

They sank the ship PRO to prove the point.

It is implied in this assumption that an implicit argument has an inseparable relation with agentivity, ruling out any possibility of controlling in sentences which involve no agentivity. However, as Williams (1985) observes, there are sentences which do not have any agent and yet can appear with an infinitival clause: (114) (115) (116)

Grass is green PRO to promote photosynthesis. Jesus died PRO to save our souls. The train derailed PRO to save the child.

If PRO must always be controlled by an (explicit or implicit) agent argument, the sentences of (114) —(116) would have to be ill-formed, because none of the matrix predicates need any agentive argument. This suggests that the assumption of implicit agent for all cases where the controller does not explicitly appear is too strong, and that there must be cases of "S-control". In fact, the control relations of (114) —(116) can be interpreted as (117) —(119), respectively: (117) (118) (119)

The fact that grass is green contributes to promoting photosynthesis. Jesus' death was for saving our souls. The accident of the train derailing resulted in saving the child.

The semantic relations of the matrix and the subordinate clauses in (117) —(119) are fundamentally the same as those in (109) and (110). Of course, there is a certain constraint on S-control. S-control is only possible when the selectional restriction between the predicate of the

Syntactic localization phenomena in English

71

subordinate clause and the content of the matrix clause is satisfied. A sentence like (120) is not permitted, for example: (120)

*The boat was sunk PRO to become a hero.

The reason is that it is not possible to say that "a boat's sinking" becomes "a hero". Note that this is not the issue of agentivity. (121) below, which contains an explicitly represented agentive argument, is also anomalous under the interpretation that the argument controls PRO: (121)

*The boat was sunk by John PRO to become a hero. (Lasnik 1988: 12)

On the other hand, (122) is perfectly permissible: (122)

The ship was sunk PRO to impress the king.

because it is highly conceivable that a ship's sinking impresses a king. 7 5.2. If the above observation is correct, can the issue be explicated by simply assuming that control is of two kinds: argument-control and Scontrol? I do not think so. Apart from the control relation in commaseparated constructions as in (109), the semantically warranted S-control in purposive infinitival clauses still leaves open a possibility that the subject of the infinitival (PRO) has some syntactic association with the matrix subject. Consider again (122), repeated below: (122)

The ship was sunk [PRO to impress the king ]

As was suggested above, the logico-semantic controller of PRO is a clausal content that the ship was sunk. But this semantic relation is not seriously altered by saying that the ship impressed the king. Put differently, the logical meaning of this sentence is indeed that the king was impressed by the ship's sinking, but we may say that in its syntactic realization the "impresser" is expressed by the matrix subject NP as a thing which can virtually represent the content of the event or the purposiveness of the matrix clause. That is, there is no practical semantic difference between (123) and (124): (123) (124)

The ship's sinking impressed the king. The ship impressed the king by being sunk.

This point can be made clearer by the following examples, which are provided by Jaeggli (1986: 617):

72

(125) (126)

Hajime Fukuchi

The structure of DNA was investigated PRO to advance our knowledge of molecular biology. *The structure of DNA was investigated PRO to be awarded the Nobel Prize.

(125) expresses the fact that the investigation of the structure of DNA advanced our knowledge of molecular biology. But it amounts to essentially the same thing if we say that the structure of DNA advanced our knowledge of molecular biology, despite the fact that it was actually people who caused this to happen. In (126), on the other hand, the matrix subject cannot by any means be associated with PRO because it is impossible to imagine a situation where the structure of DNA is awarded the Nobel Prize. It is also inconceivable that the investigation of the structure of DNA was awarded the Nobel Prize, since a prize can only be awarded to people. Similarly, in (127)

The ship was sunk PRO to collect insurance.

the meaning can be taken to be that the ship's sinking helped to collect insurance, by regarding the ship itself as an instrument for the purpose. If an event brings about a certain situation, a participant in the event can be deduced to be directly responsible for the situation. 8 5.3. It seems to me that syntactic localization works here too. The idea is similar to those suggested in the preceding three sections. Let us assume that there are cases where, logico-semantically, PRO should be controlled by a propositional content, and that, syntactically, the controller is relegated to a particular NP of the matrix clause through a process like (128):

(128) a. [NP ...}" [PRO" ...] b. [NPj ...] [PROj ...] c. [NPj ... [PRO'j ...]} Recall that in typical control relations (argument-control relations), PRO is to be associated with an NP which is in argument (subject, object, etc.) positions. But the relations of S-control, if straight-forwardly realized in syntactic structure, would be unable to meet this basic requirement. Syntactic localization is taken to be an impediment to accommodate Scontrol relations to the frame which is suitable for argument-control relations, thereby producing nominally-oriented constructions. It is important to note that in the formulation of (128), PRO is specified to be localized to the matrix subject, not to any other argument NPs.

Syntactic localization phenomena in English

73

This means that an S-control relation is to be syntactically realized to the effect that the agency of the subordinate proposition is embodied by the surface matrix subject. It is not very clear why this should be so. But one may say, at least, that if an event (or situation) is to bring about something, the role of the agency is most pertinently played by the central participant in the event, which emerges syntactically as subject NP. Consider (129): (129)

John was arrested by the police to impress his mother.

If this sentence has the reading that the fact that John was arrested by the police impressed his mother, what produced an immediate effect on the mother's mental state is reduced to John, not the police. But if we say, with a similar S-control reading, that the police arrested John to impress his mother, the situation is reversed. (See again [121] and [122]). To give another example, Roeper (1987: 299) observes that (130) is only grammatical under the reading that John seeks to make a good impression: (130)

John received Mary to make a good

impression.

Semantically, a situation is conceivable where Mary might visit John to make a good impression, but the syntactic structure of (130) does not permit such a relation to hold. Then what happens if (130) is interpreted under an S-control reading, as "the fact that John received Mary left a good impression on someone"? It is probable that the impression that is made is of John, not of Mary. It seems that the issue is to be ascribed to the ubiquitous notion of (un)markedness that is utilized for linguistic analysis. It is true that argument-control relations can obtain between PRO and various argument NPs in accordance with the semantic property of matrix verbs. But actually non-subject NPs do not necessarily appear (e. g., in intransitive constructions), in contrast with subject NPs, which are always required irrespective of the type of constructions. This makes it possible to say, in a sense, that cases of subject-control are unmarked, compared with those of nonsubject-control. The syntactic localization of S-control works in conformity with the unmarked pattern of argument-control.

74

Hajime Fukuchi

6. Conclusion We have so far examined some phenomena of syntactic localization which are observed in the relation of matrix and subordinate clauses. What is common to all instances is a situation where a part (an NP) of a clause is selected to stand for the whole clause and is syntactically designed to have direct association with the other clause. In the cases of concealed propositions and concealed nominalizations, syntactic localization is accompanied by certain syntactic deformations (or adjustments), whereas in continuative restrictive relatives and sentences of S-control the localization is carried out without observable syntactic rearrangements. But this process, producing nominally-oriented configurations, has an invariable effect of "tightening" the configurations which would be suitable for intended semantic relations, with some unavoidable syntactico-semantic discrepancies. One might argue against our assumption by saying that the phenomena we have discussed are to be reduced to an issue of dual semantic interpretation, e. g., NPs, which are semantically nominal under normal conditions, can be interpreted as clausal in certain circumstances. But this will produce a mere description of the facts, unable to bear fruitful results in any attempt to reveal the fundamental properties of language. If language is a means to express meaning by form, an ideal state of affairs will be one in which a semantic content can be satisfactorily conveyed by an available syntactic mould, with the logico-semantic relation corresponding to the syntactic relation. But in fact, language is a flexible system of rules that condones a certain degree of fluctuation in the syntactico-semantic correspondence. Our standpoint is that the fluctuation typically occurs when syntactic structure is "loosened" by some semantic influence to fail to fulfill its syntactic function and, conversely, when syntactic structure is "tightened" by efforts to produce more canonical syntactic configurations, suppressing some part of semantic content. Syntactic localization is an instantiation of the latter general process, as I have argued. If the flexibility mentioned is a property that characterizes natural language, linguistic theory must take the loosening and tightening into consideration in some way or other as two linguistically significant reverse processes for shaping syntactic structures. This study has been an attempt, in moving toward a new view of language, to clarify the way in which one subcase of these processes works.

Syntactic localization phenomena in English

75

Notes 1. We can have the same result if we assume, as Ross (1973), that (2) is derived from (1) by "Slifting" the complement clause to be able to syntactically work as matrix clause. 2. This is also fundamentally the case in epistemic modal constructions (cf. Ross 1969) and a certain type of intransitive inchoative constructions (cf. Perlmutter 1970): (i)

John may come. [John come] may. (ii) John began to smoke when he was young. [John smoke] began when he was young. 3. McCawley (1988: 736) tries to clarify this point by paraphrasing (i) as (ii). (i) My advisor was pleased with the headway I had made. (ii) My advisor was pleased that I had made (such) headway. 4. In addition to concealed propositions, there are relative constructions of another type which are often called "concealed questions" (C. L. Baker 1968; Grimshaw 1979): (i) (ii)

He told me the street that he lived on. He told me which street he lived on. I don't know the kind of girl she was ten years ago. I don't know what kind of girl she was ten years ago.

In such cases, the effect of syntactic localization is straightforward. But I will not discuss the matter here because it involves a highly semantic factor of interrogativeness, and the issue is not restricted to relative constructions: (iii)

Fred tried to guess the amount of the stolen money. Fred tried to guess how much money had been stolen. 5. Similarly, concealed questions can occur only immediately after verbs and prepositions. Ross (1977) notes that it is possible to read the complex N P of (i) as a concealed question if see is taken to mean "find out" or "discover", but the reading cannot be obtained in (ii) —(v): (i) John saw the pigeon that had escaped. (ii) The pigeon that had escaped was seen by John. (iii) The pigeon that had escaped, John saw. (iv) John saw, despite his tiredness, the pigeon that had escaped. (v) The pigeon that had escaped was tough for John to see. 6. A subcase of controlling is provided by what is widely known as control of "arbitrary reference": (i) (ii) (iii)

PRO to do it right now is very difficult. P R O climbing the mountain is prohibited by the government. The president was elected without P R O considering his competence.

The understood subjects of (i) —(iii) cannot be determined from the matrix clauses. Nevertheless, such controls belong to the type known as argument-control, because the possible controllers are by no means clausal. 7. The question is left open of why (i) is inacceptable whereas its passive counterpart is permissible: (i) (ii)

*The boat sank PRO to impress the king. The boat was sunk PRO to impress the king.

Hajime Fukuchi

76

Following Williams' (1985: 311) suggestion, I tentatively assume that the point is whether the matrix clause is purposeful. In (i), the content is just that the boat sank, without the implication that someone sank it. On the other hand, (ii) is in some sense purposive, regardless of who did it. That is, unlike (i), (ii) can be read as "the boat sank with some purpose". 8. It is interesting, in this connection, to note what Williams (1985: 308 — 9) suggests for a sentence like (i): (i)

The game was played

nude.

He says that although the subject of the AP nude appears to be controlled by a phonologically unrealized agent, it is possible that the AP modifies the overt subject the game: a game may be called a nude game if it is played by nude people.

Strong and weak barriers: Remarks on the proper characterization of barriers* Naoki Fukui

Introduction This paper attempts to construct a system in which the notion of "barriers" in syntax is properly characterized. Specifically, it will be argued that the effects of some stipulative statements made within the system of barriers proposed by Chomsky (1986 a), in particular his constraints on adjunction, can be derived as consequences of interactions of general principles if we adopt a certain version of X-bar theory, namely, the "relativized" X-bar theory proposed in Fukui (1986). After outlining the relevant part of Chomsky's (1986 a) system of barriers in Section 1, we will, in Section 2, point out some problems left open in his system, and suggest how these problems should be addressed. Section 3 very briefly summarizes "relativized" X-bar theory and explains in what respects it differs from "standard" X-bar theory such as that formulated in Chomsky (1986 a). Assuming relativized X-bar theory as a module of phrase structure, we will, in Section 4, explore a way to derive the effects of the constraints on adjunction operation made within the Chomsky (1986 a) system while avoiding some conceptual and empirical problems raised by the original constraints. Section 5 summarizes our discussion and draws some conclusions.

* This paper is an interim report of ongoing research. A larger work which will incorporate the content of this paper is now in preparation. The ideas contained in this paper were presented orally in various forms at the special forum "Cognitive Revolution: Approaches from Language" (1988) held in honor of Noam Chomsky's receiving the Kyoto Prize, and at colloquia in 1989 at the University of Pennsylvania, the University of Massachusetts at Amherst, the University of Texas at Austin, and Rutgers University. 1 am indebted to the audiences of these colloquia for valuable comments and suggestions. I would also like to express my deepest appreciation to Beatrice Santorini for her very useful comments, and to Heizo Nakajima for his warm encouragement and incomparable patience. Any shortcomings are of course my own.

78

Naoki Fukui

1. The Chomsky (1986 a) system One of the central goals of the barriers system proposed in Chomsky (1986 a) is, among many others, to unify the locality constraint on movement and that on antecedent-government in such a way that essentially the same notion of barrier can capture the fundamental similarities of locality constraints in both movement and antecedent-government cases, incorporating, in particular, the classical cases of the subjacency condition, namely Complex NP Constraint (CNPC) and the w/2-island constraint, and the cases falling under Huang's (1982) Condition on Extraction Domain, i.e., subject condition and adjunct condition. In order to achieve this goal, the following concepts are introduced. The basic idea behind the barriers theory of Chomsky (1986 a) is that any maximal projection can be a potential barrier. However, a potential barrier may be exempted from barrierhood if it is L-marked by a lexical head: 1 (1)

α L-marks β if β is a complement of a, and α is lexical.

Given this notion of L-marking, we may now turn to the definition of barrier. Barrier is defined in two steps: We first define Blocking Category (BC) as in (2) and then barrier in terms of Blocking Category as in (3): (2) (3)

γ, an X max , is a Blocking Category (BC) for β iff γ is not Lmarked and γ dominates β. γ is a barrier for β if and only if: a. γ immediately dominates δ, δ a BC for β; or b. γ is a BC for β, γ φ IP (Chomsky 1986 a: 14)

We understand "immediately dominates" in (3) to be restricted to a relation between maximal projections, so that γ immediately dominates δ even if a non-maximal projection intervenes between them. Condition (3a) is a case of "inheritance"; i.e., the category γ inherits barrierhood from a Blocking Category that it immediately dominates; in case (3 b), γ is a barrier intrinsically, being a Blocking Category itself. Extending the analysis put forth in Huang (1982) and Pesetsky (1982), Chomsky (1986 a) proposes that the scope of X-bar theory be extended to all categories, including the non-lexical (functional) categories C(OMP) and I(NFL), which have hitherto been more or less exceptions to the general X-bar schema. Thus, basic clausal structure looks as follows, where XP stands for X" (e.g., CP = C", IP = I", etc.). (4)

[CP SPEC [ c C [IP SPEC [ r I [VP .·.]]]]]

Strong and weak barriers

79

Under this conception of clausal structure, the traditional categories S' and S are regarded as the maximal projections of C and I, respectively. Chomsky further proposes that a w/z-phrase is moved into the SPEC of CP position, rather than into the C position as has been widely assumed in the literature. Now the notion of subjacency is formulated as follows: (5)

β is «-subjacent to α iff there are fewer than η + 1 barriers for β that exclude a. (Chomsky 1986 a: 30)

The definition of exclusion is given as follows: (6)

α excludes β if no segment of α dominates β. (Chomsky 1986 a: 9)

Notice that in Chomsky's theory (cf. also May 1985), a category β consists of a sequence of nodes (segments) (β,, ..., β η ), where ßj immediately dominates ßi + 1. Though in most cases a category consists of only one segment, i.e., η = 1, a structure of the form (7), a typical adjunction structure in which α is adjoined to β, presents a crucially differentiating case: (7)

[ ßl oc [ß2 ... ]]

The distinction becomes most relevant when the notion "dominate" is considered. May (1985) proposes the definition of "dominate" in (8) in order to ensure that α is not dominated by β in an adjunction structure such as (7): (8)

α is dominated by β only if it is dominated by every segment of P-

Thus, in (7), α is not dominated by β, which consists of two segments ß t and ß2, since there is a segment of β, namely β 2 , that does not dominate a. The term "dominate" used in the definitions of barrier and exclusion in (3) and (6) should be understood in this sense. The subjacency condition is now formulated as in (9). (9)

If (oti, otj + 0 is a link of a chain, then ocj +, is 1-subjacent to oij.2

The subjacency condition (9), coupled with the system of barriers briefly outlined above, handles all of the major cases of the "classical" subjacency condition and the cases of Huang's (1982) Condition on Extraction Domain (see Chomsky 1986 a for exposition). More specifically, the

80

Naoki Fukui

subjacency condition (9), which is based on the notion of barrier defined in terms of L-marking, brings the Condition on Extraction Domain, which prohibits movement out of subject or adjunct, under the subjacency condition, thereby giving a unified account of island effects. However, in order for the whole system of barriers to work properly in all the relevant cases, certain additional assumptions need to be made. Consider first the following example: (10)

[CP who, did [iP John [Vp see tj]]]

(10) is a perfectly grammatical sentence. However, if the movement of who in (10) takes place in one step from the position of tj to its surface position, then it violates the subjacency condition, since it crosses two barriers, VP and IP. Chomsky (1986 a) assumes that VP is not L-marked. Thus, VP is a barrier for tj, since it is not L-marked and is not IP. Furthermore, IP is another barrier for tj because it immediately dominates VP, a Blocking Category for tj. Therefore, the movement of who in (10) may not be in one step, but, rather, it must take place in two steps. Chomsky (1986 a) proposes that a wÄ-phrase first adjoins to VP and then moves to the SPEC of CP, as shown in (11): (11)

[cp who, did [IP John [Vp t'j [Vp see tj]]]]

Here, the definition of "exclusion" given above becomes relevant. In (11), VP is a barrier for tj for the reasons stated above. But this is irrelevant for the subjacency relation between tj and t'j since the VP in (11) does not exclude t'j, given that its lower segment does not dominate t'j. As for the subjacency relation between whoand t'j in (11), VP is not a barrier for t'j since it does not dominate t'j, for the reason just given. IP is a BC for t'j, but it is not a barrier due to (3b). Thus, the movement of who in (11) does not cross any barriers, obeying the subjacency condition. The "VP-adjunction" operation employed in the derivation of (11) is a kind of null hypothesis under a theory of movement that allows both adjunction and substitution (Chomsky 1986 a) and therefore does not pose a conceptual problem. However, if we allow adjunction of a whphrase to any maximal projection, then the definition of the subjacency condition becomes too permissive. Take, for example, the case of subject condition, a subcase of the subjacency condition: (12)

a. *who, did your interest in tj surprise Bill? b. [CP who, did [ IP t"j [IP [ NP t'j [ NP your interest in tj]] surprise Bill]]]

Strong and weak barriers

81

(12a) involves extraction out of a subject phrase and needs to be excluded in terms of the subjacency condition. However, if a wh-phrase can freely adjoin to any maximal projection, then it should be possible to derive (12a) as in (12b), with who adjoining first to the subject NP, and then to IP, and finally moving into the SPEC of CP. This derivation does not violate the subjacency condition, since each maximal projection (NP, IP) neither dominates nor excludes the trace adjoined to it. Essentially, the same situation arises in all the other subcases of the subjacency condition (i.e., Complex N P Constraint, w/i-island constraint, etc.). In short, if adjunction creates an escape hatch and its application is unconstrained, then it will void virtually all the effects of the subjacency condition. Noting this problem, Chomsky proposes the following constraint on adjunction: (13)

Adjunction is possible only to a maximal projection that is a non-argument. (Chomsky 1986 a: 6)

The constraint (13) correctly allows adjunction to VP, which is necessary as we saw in (11), since VP, being a predicate, is clearly not an argument while it prohibits adjunction to a subject N P as in (12). However, adjunction to IP, which is another problematic operation in (12), cannot be excluded by (13), since IP is not an argument there. The following example shows that adjunction of a w/j-phrase to IP needs to be prohibited in other cases as well (only relevant portions are represented): (14)

a. V.whatx did John wonder wheres Bill put ts tj? b. [ CP whati did [ IP t'" [1P John [Vp t" [Vp wonder [CP wherej [IP t'j [IP Bill put t, tj] ... ]]]]]]]]

(14a) is a typical w/z-island case and therefore should be treated as involving a violation of the subjacency condition. However, if the movement of what can proceed via adjunction to IP (as well as via adjunction to VP) as shown in (14b), then it will cross no barrier in the derivation of (14a). Chomsky (1986 a) therefore proposes the following additional constraint on adjunction: (15)

vv/z-phrases cannot be adjoined to IP. (Cf. Chomsky 1986 a: 5)

Given the constraints on adjunction (13) and (15), Chomsky's (1986 a) system of barriers works quite nicely in almost all cases of the "classical" subjacency condition as well as of the Condition on Extraction Domain, providing a general and unified account of these island phenomena.

82

Naoki Fukui

2. Posing the problem The system of barriers outlined in the preceding section is very attractive in that it gives a unified account of both the "classical" subjacency cases and Huang's (1982) cases of Condition on Extraction Domain. There are, however, some problems still left open in the Chomsky (1986 a) system, to which we now turn. The problems we will discuss in the following have to do with the constraints on adjunction (13) and (15) discussed above. There are two kinds of problems concerning the two proposed constraints on adjunction. One is of a conceptual nature. The constraint (13), for example, still leaves us with the question: why is it that only non-arguments allow adjunction? An analogous problem arises for the constraint (15): it gives us no explanation as to why movement of a w/z-phrase should be constrained in such a way that it cannot adjoin to IP. In short, the constraints on adjunction as they are formulated in (13) and (15) bear a highly stipulative flavor and do not seem to qualify as principles of Universal Grammar. The other problem has to do with the empirical predictions made by the constraint (13). If we interpret the term "non-argument" used in (13) literally, the constraint would predict, as pointed out by many linguists including Chomsky, that adjunction to adjunct clauses is possible, thereby voiding the adjunct condition effect, since adjuncts are, by definition, non-arguments. The same problem arises in the case of relative clauses, since CP in relative clause structures is clearly a nonargument and hence should allow adjunction to it. We thus lose an explanation for half the cases of Complex NP Constraint. The relevant structures are schematically represented as follows (intermediate adjunctions that are not relevant here are omitted): (16)

a. adjunct:3 [CP wh{ [ [I< I [ . . . ] [ C P T'I [ •· ...]]]]] t It I b. relative clause: [CP wh{ [IP ... V [ NP ... [CP t'; [CP ... tj ...]]]]] t Jt I IP

V P

C P

How do we approach these problems? Let us consider the conceptual problem first. In order to solve the conceptual problem associated with the constraint (13), Chomsky (1986 a: 16) proposes, on the basis of a suggestion made by Kyle Johnson, that it be derived from the thetacriterion. In particular, he proposes that adjunction to a category makes the category "invisible" to theta-marking, changing the status of the original category as "head" into "non-head" with respect to theta-mark-

Strong and weak barriers

83

ing. Thus, adjunction to an argument will always yield a violation of the theta-criterion. As for the constraint (15), Chomsky (1986 a: 5) also suggests that the effect of this constraint is derivable from the general property of wA-phrases, which states that these elements have clausal scope. Thus, they must move to a pre-IP position and cannot be adjoined to IP. While these suggestions seem plausible and therefore are certainly worth pursuing, it is also true at this point that they can hardly be more than suggestions, and it is not at all clear precisely how these ideas can be worked out. Turning to the empirical problems, one possibility to solve the problems along the lines of Chomsky's suggestions discussed above is to assume that adjuncts and clauses in relative clauses are theta-recipients. The idea that adjuncts receive a certain kind of theta-role has already been suggested in the literature (Zubizarreta 1982 a), afid this idea can easily be extended to CP in relative clauses, i. e., clauses in relative-clause structure are assigned some type of theta-role, perhaps via predication. If this approach is tenable, then adjunction to adjuncts and CP in relative clauses will always yield a theta-criterion violation, and will thus be disallowed by theta-theory. All of the suggested solutions to the conceptual and empirical problems raised by the constraints (13) and (15) on adjunction are based on the fundamental assumption adopted in Chomsky (1986 a) that every maximal projection is equal with respect to its potential barrierhood, and that only contextual factors such as L-marking are relevant to the determination of barriers. In what follows, I will explore a different approach in which, in addition to contextual factors, some category-inherent factors are involved in characterizing barriers to movement. It is important to bear in mind when taking this approach that the class of potential barriers should not be simply stipulated in terms of category types such as NP or S. This would take us back to the "classical" subjacency condition, where just this is done with the class of "bounding nodes". Rather, it would be desirable to derive the categorial asymmetry with respect to barrierhood from some general principles of grammar. To start our exploration, let us first clarify the kind of problem we are to solve. We saw above that if we interpret the constraint (13) literally, some unwanted derivations will be permitted. From our present perspective, the problem lies in our interpretation of the concept of "arguments" used in (13). The term "argument" has two different meanings. Firstly, the term "argument" is used as a notion contrary to "adjunct". We have so far interpreted the term in this sense. However, the concept "argument"

84

Naoki Fukui

is also used as an opposing notion to the concept of "predicate", as is customary in predicate logic. Notice that "argument" in the first sense, i. e., "argument" as opposed to "adjunct", is a relational notion. The same category (CP, for instance) may or may not be an argument, depending on the syntactic configuration in which it appears. On the other hand, the term "argument" in the second sense, i.e., "argument" as opposed to predicate, is an absolute notion. A given category is an argument or a non-argument (predicate) inherently and independently of context. We have already seen that if the notion of "argument" is used in the first sense, then the constraint (13) will face certain empirical problems. This suggests that the concept of "argument" in the absolute (category-inherent) sense might be the relevant notion to the proper characterization of barriers. If this idea is correct, then the intuition behind Chomsky's constraint (13) on adjunction can be rephrased as follows: (17)

Only potential arguments can be barriers (or more or less equivalently, "predicates are not barriers").

In the remainder of this paper, we will explore a natural way of capturing the intuition expressed in (17) within the framework of the "relativized" X-bar theory proposed in Fukui (1986).

3. Relativized X-bar theory In this section, we will very briefly introduce the relevant part of relativized X-bar theory. The basic idea of this version of X-bar theory is that the fundamental difference between the two distinct lexical classes, viz., lexical categories (Ν, V, etc.) vs. functional (non-lexical) categories (C, I, etc.), should be reflected in the ways they project in accordance with Xbar theory. Lexical categories have "meaning", however this term is to be defined precisely, so that they play a role of being the core of any thought expression. Functional categories, on the other hand, do not have comparable "meaning", at least to a large extent. 4 Their function in syntactic structure is basically to associate syntactic categories via relations such as "agreement". This fundamental difference between lexical and functional categories is represented in the following scheme of projection (linear order is irrelevant):

Strong and weak barriers

(18)

85

Lexical projection: [L- X [L< ... [L< X [L< L° complement]] ...]] (L° = Ν, V, A, etc.)

I * selection

Functional projection: [F- SPEC [ F F° complement]] { fl * (F° = C, I, D, etc.) agreement selection Thus, a lexical head projects to the single-bar level, taking a complement as its lexical property, and at that level it allows free recursion. 5 A functional head also projects to the single-bar level in the same way as a lexical head. However, a functional projection differs from a lexical projection in that it can project up to the double-bar level with the help of its agreement features. The status of a "specifier" in the functional projection calls for some clarification. The basic role of a specifier in a functional projection is clearly that it "closes" the projection. This characterization of specifiers can be interpreted in two slightly different ways. One is to assume that the specifier position must be licensed by some syntactic relation, typically by an agreement relation with its head (F°), and that therefore a functional projection stops at the single-bar level when the head has no agreement features. The other way is to assume that the specifier position is freely generated with an empty category underneath it, but that the actual occurrence of some maximal projection in the specifier position must be licensed by some principle (such as the "agreement principle" of Fukui 1988 a), in which case a functional projection can optionally project up to the double-bar level with an empty specifier position even when its head bears no agreement features. Though the latter approach seems to have some desirable consequences, including a fairly straightforward account of successive cyclic movement into the specifier of CP, and thus would be worth pursuing, I will, in what follows (leaving the ultimate choice between the two approaches open), assume the first approach simply for the sake of exposition, and state the following condition: (19)

A functional category projects to XP (i.e., to the double-bar level) if and only if its SPEC is licensed.

The relativized X-bar theory briefly outlined above differs from "standard" versions of X-bar theory such as that formulated in Chomsky (1986 a) in a number of respects. One crucial difference which is relevant to our present concern is that while the notion of "maximal projection"

86

Naoki Fukui

is equivalent to a certain number of bars in the standard X-bar theory (in Chomsky's [1986 a] version of X-bar theory, the number is two), these two notions are dissociated in relativized X-bar theory, where the maximal projection is defined as being, roughly, the top node of a given projection. Thus, while the maximal projections of lexical categories are always X's, the maximal projections of functional categories are either X's or XPs, depending on whether or not SPEC is licensed. For a fuller exposition of relativized X-bar theory, the reader is referred to Fukui (1986, forthcoming).

4. Toward a solution of the problem Assuming the relativized X-bar theory briefly outlined in the preceding section, let us now explore a way of capturing the intuition stated in (17). Let us first consider the class of potential Blocking Categories (and barriers) in relation to the lexical vs. functional distinction. Vmax is a typical category that does not function as a Blocking Category, and is a lexical projection. CP and IP are typical Blocking Categories and are functional projections. The parallelism between Blocking Categories and functional categories breaks down when we consider "NP", which is a typical Blocking Categories but is assumed to be a lexical projection. This problem is resolved, however, if we adopt the "DP analysis", according to which noun phrases are analyzed as determiner-headed DPs, determiners being a functional category (cf. Brame 1981, 1982; Fukui — Speas 1986; Abney 1987 for relevant discussion). Thus, assuming the DP analysis, we have the following generalization: (20)

functional categories = potential Blocking Categories lexical categories = non-Blocking Categories

This generalization is clearly in accord with our earlier statement (17), since functional categories (CP, IP, DP, etc.) are typical arguments and lexical categories, particularly VPs and APs, generally function as predicates. One possible way to derive the distinction stated in (20) within relativized X-bar theory is to assume the following condition: (21)

Only an XP can be a Blocking Category (and hence a barrier).

Strong and weak barriers

87

This approach is taken by Fukui — Speas (1986) and is further explored, with considerations of cross-linguistic data, by Uriagereka (1988: Chapter 2). Under this approach, lexical categories are never barriers, because their projections never reach the double-bar (XP) level. Some of the consequences of this approach are summarized below: (22)

a. "VP-adjunction" is not necessary. b. No lexical projection (in particular, V') can be a barrier. c. A functional category that projects only to X' cannot be a barrier. d. A language that has no XPs would have no barriers (in D- to S-structure mapping).

Some problems arise with respect to (22b —d). First, (22b) would imply that the "zero-subjacency" approach to parasitic-gap construction proposed by Chomsky (1986 a) cannot be maintained. Consider the following contrast taken from Chomsky (1986 a: 64): (23)

a. what[ did you file t; [PP Ο; [PP before [t; [you read e;]]]]? b. *whox [t; [Vp spoke to you [PP Ολ [PP before [tj \you met ej]]]]]?

Putting aside the problem arising from adjunction of an operator to PP, 6 a plausible candidate for the explanation of the contrast under the "zerosubjacency" approach is the existence of an intervening VP between t; and Οj in (23b) and the lack thereof in (23a). That is, in (23a) the operator 0\ is "zero-subjacent" to the trace ti5 whereas in (23b) the intervening maximal projection VP breaks such a "zero-subjacency" relation, thereby making the sentence ungrammatical (see Chomsky 1986 a for discussion). Thus, if we want to maintain the "zero-subjacency" analysis of parasitic gaps, then Vmax should count as a barrier at least in the cases where no adjunction to Vmax is involved. Secondly, (22c) makes wrong predictions concerning some cases of the subject condition. Consider the following: (24)

a. *1who\ did a picture of t; please her? b. [CP who, [c- did [IP [D< a [N< picture of tj]] please her ]]]

(24a) is a typical case of the subject condition violation. Therefore, we would expect the movement of who from its original position indicated by tj to its surface position to cross more than one barrier. However, as (24b) shows, there is no barrier between the two positions. The crucial factor here is that the subject does not have a SPEC position due to the lack of agreement features, and hence projects only up to the single-bar level. Thus, D' in (24b) does not count as a Blocking Category, and hence

88

Naoki Fukui

not as a barrier, and consequently the IP dominating the D' does not inherit barrierhood. In short, (22c) would make it impossible to account for the subject condition effect when the subject phrase does not contain a specifier position. As for (22d), Japanese is a case in point. It has been argued in Fukui (1986, 1988 a) that Japanese lacks specifiers and therefore, given the statement (19), is a single-bar language. If this is correct, then (22d) would predict that the language does not exhibit any island effects. This prediction is not borne out, as the following scrambling facts show (Saito 1985):7 (25)

ΊΒϊΙΙ-ο, [s John-ga [NP [s Mary-ga t, sakete-iru to-yuu ] Acc Nom is avoiding that uwasa]-o kiita (koto) rumor heard that 'John heard the rumor (which says) that Mary is avoiding Bill' b. V.ano hon-o, [s John-ga [ NP [ s ej [tj katta] hit0j]-0 that book bought person sagashite-iru rashii is looking for seems 'It seems that John is looking for the person who bought that book' c. Isono-hon-oi John-ga [s- Mary-ga tj yomi-oete kara] that/the finish-reading after dekaketa (koto) went out 'John went out after Mary finished reading that/the book' a.

All of these examples involve scrambling out of alleged islands: (25a) involves scrambling out of a complex (non-relative) NP; (25b) scrambling out of a relative clause; and (25c) scrambling out of an adjunct clause. As indicated, (25a —c) are not perfect, and, in conjunction with other relevant phenomena, including naze 'why' extraction (Fukui 1988 b) and the comparative construction (Kikuchi 1989), suggest the following observational generalization regarding extraction phenomena in Japanese: (26)

Japanese does exhibit some island effects, but they are, in many cases, "milder" than those found in languages like English.

Obviously, this observational generalization poses some problems for the prediction made by (22d).

Strong and weak barriers

89

Based on these considerations, we would now like to explore an alternative way of capturing the intuition stated in (17) and its restatement in (20). The basic idea behind this approach can be stated as follows: (27)

The strength of a barrier depends on the "depth" of projection. An XP is a "strong" barrier, whereas an X', when it is not Lmarked, is a "weak" barrier.

Incorporating this idea into the characterization of Blocking Categories, we have the following definition of BC: (28)

γ, an X max , is a BC for β if and only if γ dominates β and: (i) γ is an XP (strong barrier); or (ii) γ is an X' that is not L-marked (weak barrier)

The terms "strong" and "weak" used in this definition have dual sense: An XP is a "strong" barrier in that it functions as a barrier in its own right, independently of the syntactic context, and also in that it is "strong" in its effect as a barrier. Likewise, an X' is a "weak" barrier both in the sense that its barrierhood depends on the configuration in which it appears, and that its effect as a barrier is relatively "weak". Notice that under this approach lexical projections can be (weak) barriers when they are not L-marked. However, in cases where movement is involved, their barrierhood will always be voided by adjoining a phrase to them. Such an adjunction operation is available, since relativized Xbar theory permits free recursion at the single-bar level of a lexical projection. When movement is not involved, on the other hand, lexical projections function as a barrier (when they are not L-marked). This distinguishes the case of the parasitic gap construction where Vmax functions as a barrier, from the cases of extraction out of Vmax where Vmax should not function as a barrier. In the case of functional projections, no such option is available, since relativized X-bar theory allows recursion neither at the single-bar nor at the double-bar level in this case. The desired result can be obtained without further stipulation by assuming relativized X-bar theory to hold at both D- and S-structures: (29)

Relativized X-bar theory holds at both D-structure and Sstructure. 8

If this approach is correct, then it might be that there is no "true" (i. e., structure-creating) adjunction operation in D- to S-structure mapping, and that processes such as extraposition and Heavy NP Shift should be reanalyzed in one of the following ways: (i) as belonging to PF-compo-

90

Naoki Fukui

nent, (ii) as substitution into SPEC of a functional category, (iii) as "adjunction" to a lexical category, or (iv) as being base-generated in one of the positions permitted by the X-bar theory. One potential problem for this approach is topicalization of the following kind: (30)

I believe that [this bookj [you should read tj]

If this type of topicalization involves adjunction to IP, as argued by Baltin (1982) and Lasnik and Saito (1987), then it poses a problem, since adjunction to a functional projection is prohibited under the approach we are exploring. One possibility (suggested to me by Noam Chomsky) is to assume that IP is "lexical" in the relevant sense, due to its close relation to a verb. Thus, IP, unlike other functional categories such as CP and DP, exceptionally allows adjunction to it. However, a problem with this solution is that it requires us to re-adopt the constraint (15) until we find satisfactorily general reasons for disallowing such adjunction. Another possibility is to analyze topicalization of the above kind as substitution into the specifier position of TP (Tensed Phrase), on the assumption that subject appears in the specifier of AGRP (Agreement Phrase), modifying slightly the analysis of clause structures proposed in Pollock (1989) and Chomsky (1989). This analysis would capture the relationship between topicalization (in the embedded context) and tensed elements. Specifically, the analysis accounts for the impossibility of topicalization in infinitival complements: (31)

a. * I believe [Marys [Bill to have met tj]] b. *John tried [the exam; [PRO to pass t,]]

To the extent that there are alternative accounts, however, this fact does not in itself constitute supporting evidence for the suggested analysis of topicalization. For example, (31a) is independently excluded as a violation of the adjacency requirement on Case-marking, with Bill being Caseless. The treatment of (31b) is not as straightforward as that of (31a), but, as Mamoru Saito has suggested to me, it might be possible to rule out (31b) as a violation of control theory. In the absence of conclusive evidence, we will leave open the choice between the two possible analyses of topicalization in the embedded context illustrated in (30). Let us now turn to some further consequences of our current approach. The definition of Blocking Category given in (29) predicts that in a language that has no XP projections, like Japanese, there are only weak barriers. And as we have already seen above, this prediction is fulfilled. Thus, we now have a principled way, though at a rather speculative level,

Strong and weak barriers

91

to explain our general intuition that island effects in Japanese are "milder" than those found in English, as far as movement between D-structure and S-structure is concerned. Another consequence of our approach is that it opens up a possibility of explaining the well-known difference between "syntactic" wA-movement and w/z-movement in LF. That is, as has been pointed out in the literature (cf. Huang 1982), it seems that wA-movement in LF is less constrained by the subjacency condition than "syntactic" wA-movement. If our approach is on the right track, then this difference between "syntactic" wA-movement and LF wA-movement could be an automatic consequence of the fact that X-bar theory holds at S-structure, but not at LF. That is, in LF a wA-phrase can freely escape from an island by adjunction without being constrained by relativized X-bar theory. Though there are various important questions regarding wA-movement in LF, particularly ones concerning the "pied-piping" analysis (Nishigauchi 1986; Pesetsky 1987), the possibility mentioned above appears worth exploring. A further consequence of the approach being explored here is that at least some aspects of the so-called "specificity" phenomena could be handled as falling under the theory of barriers. Consider the following contrast: (32) (33)

a. Whoi did you see [a picture of tj? b. *? Who, did you see [John's picture of tj? a. [Cp who; [did [IP you [v- t" [v- see [D- a [N- t'j [ N picture of tj] * It I* I b.

[cp whOj [did [ΪΡ you [v- t" [y see [DP John's [N< t It t'j [χ. pictures of t j ...]]]]]]]] _Jt I

Under our analysis, the examples (32a) and (32b) have the derivations (33a) and (33b), respectively. In (33a), the movement of who crosses no barrier. The likeliest candidate for a barrier, D', is L-marked by the verb see and thus is not a barrier. In (33b), on the other hand, the object phrase projects up to the double-bar level due to the presence of John's in its specifier position. Therefore, the DP in the object position functions as a strong barrier even though it is L-marked by see. We thus account

92

Naoki Fukui

for the difference in grammaticality between (32a) and (32b) in terms of the theory of barriers. There are of course various other factors involved in the "specificity" phenomena, and most of them perhaps fall outside the scope of the barriers theory. Our approach, however, suggests that at least some factors involved in the phenomena can be handled in terms of the theory of barriers. Incidentally, the approach to the "specificity" phenomena discussed above could possibly be extended to the well-known distinction between tensed clauses and infinitivals with regard to extraction, by utilizing the difference in the "depth" of projection between the two cases, i.e., tensed clauses are IPs and infinitivals are generally I's. We will not explore this possibility here, leaving it for future research.

5. Summary and conclusion In this paper, we have explored a general way of deriving the effects of the two constraints on adjunction operation proposed in Chomsky (1986 a), and have made, among others, the following claims. (i) In addition to contextual factors such as L-marking (Chomsky 1986 a), there is a category-inherent asymmetry with respect to barrierhood. (ii) Given relativized X-bar theory (Fukui 1986), with the assumption that it holds both at D- and S-structures, such an asymmetry does not have to be stipulated. Rather, it is a reflection of a fundamental categorial distinction: lexical vs. functional categories. (iii) Data from Japanese (and other data as well) suggest that the strength of a barrier depends on the "depth" of projection. Thus, there are two types of barriers: 1) strong barriers (XP's), which are independent of the context, and 2) weak barriers (X's), which are sensitive to the syntactic environment in which they appear. Notes 1. The characterization of L-marking given in (1) is slightly different from that given in Chomsky (1986 a), and is close to Chomsky's modified formulation presented in his class lectures (Fall, 1986). 2. Note that this formulation of the subjacency condition is neutral on the issue of whether the subjacency condition is a condition on movement or on the resulting representation. 3. Here, we take adjuncts to be CP. The category type of adjuncts is not relevant to our present discussion, however.

Strong and weak barriers

93

4. Certain functional categories, e. g., I and D (see below), may function as "operators" in LF, but this is essentially different from predicative relations in which lexical categories play a central role. 5. For arguments that lexical projections must allow recursion at the single-bar level, see Fukui (1986: Chapter 2) and the references cited there. 6. It might be that the problem has to do with the obscure status of Ρ with regard to lexical/functional distinction. Other problems also arise regarding the status of P. 7. We will rather freely use the traditional symbols S', S, and N P when their exact categorial status and internal (X-bar theoretical) structures are not directly relevant to our present discussion. 8. Independent evidence that (standard) X-bar theory holds at D- and S-structures is presented in Chomsky's series of lectures (Tokyo, 1987) and in van Riemsdijk (1987).

An analysis of English descriptive genitives in the dynamic theory of syntax* Masayuki

Ike-uchi

0. Introduction In this article we will first suggest that a descriptive genitive expression in English has a dual categorial status, and will argue that (an N P with) a descriptive genitive is assigned a "tree-grafted" structure. After giving a brief introduction to the dynamic theory of syntax proposed by Kajita (1977, 1983, 1984, 1985, 1986 a, 1986 b), upon which this article is based, we will develop a dynamic analysis of the descriptive genitive expression which elegantly explains its acquisition process. We will also demonstrate that this analysis based on the dynamic theory of syntax gives an interesting account of the question of how this expression is generated in the adult grammar of English. Before entering into the main discussion, however, we will as a preliminary examine several syntactic and semantic properties of (NPs with) descriptive genitives.

1. Syntactic and semantic properties of descriptive genitives Descriptive genitives are, roughly speaking, those genitives which act as modifiers rather than as determiners, expressing a quality or a characteristic. Expressions which are often cited as representative examples of English descriptive genitives are given in (1):

* I wish to thank Masaru Kajita, Takao Yagi, Masanori Suiko, Takayasu Namiki, and Noriko Imanishi for their invaluable comments and suggestions on earlier versions of this article. I would like to thank Yukio Otsu for stimulating discussion on language acquisition and for making "Wales Data" available to me. I also wish to express my gratitude to Jeffrey Jones, who kindly acted as an informant and corrected stylistic errors. I am especially grateful to Heizo Nakajima for giving me an opportunity to write this paper and for his detailed comments and suggestive remarks.

96

(1)

Masayuki Ike-uchi

a women's college, these boy's socks, a child's language, a doctor's degree

Notice that the determiner goes with a head noun of the entire N P (e. g. college), and not with a descriptive genitive noun (e. g. women). One might suggest that noun phrases with these descriptive genitives are fixed unproductive constructions. This, however, is not correct. There are a large number of NPs with descriptive genitives which have quite a variety of descriptive-genitive head combinations. Observe (2), (3), and (4) (see Woisetschlaeger 1983, Quirk et al. 1985: 328, and Shumaker 1975): (2)

(3)

(4)

a teacher's guide, a user's manual, a collector's greed, children's literature, a snail's pace, the open driver's window, the drivers' championship standings, a players' strike, that ship's steward, a poet's mouth, a reporter's curiosity, a businessman's blue suit, the logician's sense, the ¥ 2 million winner's check, a 5,000member citizen's group, a barmaid's dream of a real gentleman an elderly citizens' home, a well-known mathematician's proof (Woisetschlaeger 1983: 146), a first-year undergraduate's essay (Quirk et al. 1985: 328), an old men's home, a ten minutes' walk that John Bull's manner of yours {MEG II: 301), a Brink's truck

In these noun phrases, descriptive genitives play a modifying role, as mentioned above. These phrases ("a Y's X") can typically be paraphrased as "an X particularly/exclusively for Y " (e.g., a women's college), or as "a kind of X that is stereotyped with respect to Y " (Woisetschlaeger 1983: 147) or "an X characteristic of Y " (e.g., a child's language). To this extent, the meaning of the descriptive genitives above are transparent. Thus they are for the most part compositionally determined and predictable. To put it differently, they are not as opaque as those of the following descriptive genitives: (5)

bvXYs-eye, cat's (-)paw, sheep's eye

For instance, bull's-eye is not "an eye characteristic of a bull", but "a circular center of a target". It has a figurative/metaphorical meaning that is less predictable than those of the descriptive genitives in (1) —(4). We will not deal here with this type of (so-called) compound, but concentrate on those productive NPs with descriptive genitives that have transparent meanings exemplified above. We may include in the category of descriptive genitive in the sense of this article the following genitive expressions that mean something like

English descriptive genitives in the dynamic theory of syntax

97

"an X among Ys", "the most Y-like X", or "the most typical X", where X = Y: 1 (6)

a. What sort of man would the Heavyweight Champion of the World kiss? A man's man, of course. (G. Keillor, Lake Wobegon Days, p. 226) b. "It was his time", said Elmer at the time, reflecting the general Knute philosophy of death as a lottery, and yet they miss him so much every year, the Duck Hunter's Duck Hunter. A man who lived for the hunt, lived by the lake about a beer can's throw from the blind, and took his gun to bed with him in season ... (Lake Wobegon Days, p. 240)

Of course, not all genitives have a descriptive genitive interpretation defined here. For example, the genitives in (7) have a descriptive genitive meaning, but those in (8) do not: (7)

(8)

a. Language aside, Tovah's major at the university had been journalism. From early on she'd had a reporter's curiosity and a reporter's eye. (I. Wallace, The Seventh Secret, p. 54) b. Under the north window was a draughtsman's easel and a rickety swivel stool. a. He jerked awake at the sound of a man's voice calling, "Breaker, breaker ... Roadmaster to Silver Hawk." (J. Humphreys, Over the Top, p. 67) b. Public and press alike are legitimately interested in a candidate's children ... (K. Mills, The Los Angeles Times) c. Revenues of the inheritance tax, a tax levied on land and other properties transferred to a decedent's heirs, grew 56.0 percent ... (Jiji Press)

In (8a), for instance, the most natural interpretation will not be "a manly voice", but "the voice of a certain man". Recall that we would not take into consideration real compounds like (5). Thus we are claiming that NPs with descriptive genitives in our sense do not constitute true compounds. Note that they do not have the familiar properties of the true compound: (i) the opacity of meaning (see above), and (ii) the syntactic inflexibility — e. g., no elements can be inserted between the two words, and coordination and other syntactic processes cannot be applied. The following illustration will clarify these points. 2 First, the interpretation of our NPs with descriptive genitives is to an allowable extent transparent, as noted above. Secondly, they are syntac-

98

Masayuki Ike-uchi

tically flexible, which is not the case with compounds. We can insert an element between the descriptive genitive and the head noun: (9)

a businessman's blue suit, his boy's rosy cheeks (MEG VI: 280), the National Women's Political Caucus

One might claim that (10a) is grammatical but (10b) is not, and thus no element can stand between the descriptive genitive and the head noun: (10)

a. This is a blue children's bicycle. b. (*)This is a children's blue bicycle.

This is not entirely correct. If there is a thing which is a blue bicycle exclusively for children's use, then (10b) will be acceptable. Then the question is not a syntactic one (cf. Woisetschlaeger 1983: 146). Indeed, phrases like No. η and η th (where η = numeral) often precede the descriptive genitive and thus do not occur between it and the head noun, as in (11). Incidentally, they follow the true genitive, as, for example, in

(12): (11)

(12)

a. Connors, the No. 6 men's seed, beat... Lloyd, the No. 2 women's seed, continued ... (AP) b. The ninth women's seed, Lori McNeil, fell to Judith Wiesner of Austria ... (AP) Becker, the tournament's No. 2 seed seeking his first clay court title ... (AP)

But they can also occupy the position between the descriptive genitive and the head noun, as is seen in (13): (13)

a. Andre Agassi, the men's No. 4 seed, won his first match ever at the Open ... (AP) b. Martin Jaite, the men's 14th seed from Argentina, fell ... (AP)

Furthermore, descriptive genitives can be coordinated (but see Di Sciullo - Williams 1987: 105; Quirk et al. 1985: 971): (14)

a. Twelve men's and women's events will be held ... (AP) b. American Scott Davis and Barbara Potter capped a successful start to their preparations for Wimbledon by winning the men's and women's titles at the Beckenham Grass Court Tennis Championships Sunday. (Reuter-Kyodo)

And a certain kind of reduction process can be applied to NPs with descriptive genitives, as is seen from (15):

English descriptive genitives in the dynamic theory of syntax

(15)

99

a. No mere bankers' plan will meet the requirements, no matter how honestly conceived. It should be a merchants' and a farmers' plan as well. (Curme 1931: 68) b. Superior dress material for [0] ladies' and [0] children's wear. (Kruisinga 1932: 57)

The conclusion that seems quite well established on the basis of these considerations is that our NPs with descriptive genitives are not true compounds. Having surveyed several of the general syntactic and semantic properties of (NPs with) descriptive genitives that we would like to focus on here, we can now turn to a discussion of the structure of these phrases.

2.

On the structure of noun phrases with descriptive genitives

2.1. Descriptive genitives as adjectival phrases As mentioned in the preceding section, descriptive genitives semantically play a modifying or classifying role. They describe and specify the properties of the following head noun. Therefore, from a semantic point of view it can be said that they are quite similar to adjectives. There are also several pieces of syntactic evidence showing that descriptive genitives are adjectives or adjectival phrases. First, a descriptive genitive can be premodified by an adverb: (16)

a. Among predominantly black colleges, Fisk, Morehouse, the Tuskegee Institute, Howard, andSpelman are well-known. ... Among predominantly women's colleges, Mt. Holyoke (1837), Vassar (1861), Wellesley (1870), Smith (1871), and Bryn Mawr (1885) are quite famous. (P. McLean, America: Inside and Out, pp. 5 9 - 6 0 ) b. an all girls' school (Quirk et al. 1985: 328)

In (16a), the adverb predominantly modifies the adjective black in the first sentence, and it modifies the descriptive genitive women's in a parallel position in the second sentence. Thus the whole NP in question means "colleges predominantly for women". 3 In (16b), similarly, we have an interpretation "a school entirely for girls", taking all as an adverb. 4 These facts thus strongly suggest that a descriptive genitive has a categorial status of adjective (or adjectival phrase).

100

Masayuki Ike-uchi

Second, a descriptive genitive appears in a position parallel to an adjective in a prenominal coordinate structure: (17)

But the logicians of St. James and Versailles wisely chose to consider the matter in dispute as a European and not a Redman's question. (Poutsma 1914: 90)

Here the descriptive genitive Red-man's corresponds to the adjective European. Third, a descriptive genitive can conjoin with true adjectives in the prenominal position. 5 Conjoined adjectives and descriptive genitives do not form a hierarchical structure but are simply juxtaposed, and do not have any cumulative effect on meaning: (18)

a. It seemed that his pointed, boar's teeth had grown still longer. b. ... he hated his white, invalid's hand. {MEG VI: 2 7 9 - 2 8 0 )

Fourth, there is a phenomenon which can be regarded as the stacking of adjectives in the prenominal position: (19)

a. ... since he was wearing a white c h e f s cap and a white apron. {The Seventh Secret, p. 317) b. If it was a boy, a son, he would probably build the new children's wing that the hospital so badly needed. (J. Archer, Kane and Abel, p. 17) c. ... and as ever he was not in uniform but in a businessman's blue suit. {The Seventh Secret, p. 70) d. I could see no more of his boy's rosy cheeks than of his boy's trim little jacket. {MEG VI: 280) e. Look at those three quaint old Cornish fisherman's two-bedroomed terraced cottages. (Radford 1988: 224)

Stacked attributive adjectives form a hierarchical structure and have a cumulative interpretation. In (19a), for example, a white chefs cap means "a cap for a chef which is white". It is surely the cap which is white rather than the chef. It will have a structure like (20): (20)

[a [white [chef's [cap]]}]

In sum, we conclude that a descriptive genitive is an A(P) — i. e., it has a syntactic categorial status of A(P), and that an N P with a descriptive genitive has (at least) a structure like (21): 6

English descriptive genitives in the dynamic theory of syntax

(21)

101

NP N'

Det

a

AP

N'

A

Ν

women's

college

2.2. Descriptive genitives as NPs There is also some evidence which shows that the descriptive genitive must be an NP, though. The first piece of evidence concerns the internal structure of the descriptive genitive itself. At first sight, it may seem that the descriptive genitive constitutes a simple Ν (see the examples in [1] and [2]). This, however, is not correct. The examples in (3) show that the descriptive genitive must at least be an N ' (cf. Woisetschlaeger 1983: 146). And those in (22) indicate that it must constitute an NP — i.e., it must have an internal structure of NP: (22)

a. my this year's examination questions (Quirk et al. 1985: 328) b. a this year's loon (Curme 1931: 68)

The second argument is related to POSS ('s) insertion. It is well known (cf. Chomsky 1986 b) that the normal context for s-insertion is (23): (23)

NP

a, where a = N' or VP.

Thus if the descriptive genitive is an NP, then we need no special requirements for 5-insertion. This again indicates that the descriptive genitive has an NP status. The third possible piece of evidence comes from the ellipsis phenomena. It is known that the simple deletion of the repeated head noun is generally (or usually) prohibited, and that the substitution of one is necessary in the construction of Det -I- A(P) + N', as in (24): (24)

a. This previously unprotected species is now classified as *an endangered 0 jan endangered one. (Quirk et al. 1985: 901) b. We turned off the main road on to *a quieter 0 ja quieter one. (Quirk et al. 1985: 901) c. The book I bought yesterday is the little book, not *the big 0 / the big one.

102

Masayuki Ike-uchi

And acceptability can be the other way around in the construction of a true determinative genitive + N ' — i. e., ellipsis is possible and the onesubstitution is generally impossible: (25)

a. John's car runs faster than Harry's 0/* Harry's one. b. Don't touch those cards — they're my partner's 0 j*my partner's ones.

Notice that both the ellipsis and the P R O to make a movie]]

23. See Anderson (1979: 125), who writes that "nouns d o not allow bare infinitival complements". Abney (1987: 130) also writes that "nouns cannot take reduced clause complements, but only full C P complements". 24. This assumption is not novel here. For instance, Aoun — Sportiche (1983: 234 note 15) suggest that I of a gerund is marked [ + N] and Abney (1987: 224) also suggests that the gerundive -ing has the feature [ + N], See also Reuland (1983) and Stowell (1981). 25. Stowell (1981) suggests that not only N P s and gerunds but also S' complements (i.e., tensed and infinitival clauses) have the features [ + Ν, — V]. 26. It is not plausible to assume that modal auxiliaries appear within I (cf. Chomsky 1981: 140 note 28). If one assumes that a modal auxiliary appears within I, it follows that I may dominate two essentially different elements, [[+ Tense], [ + Ν, — V]] and a modal auxiliary, although other heads dominate only one element. This undesirable result can be avoided by assuming that a modal auxiliary, like the perfective have, the progressive be, and the passive be, is an auxiliary verb which selects VP (cf. Ross 1969; Emonds 1970, 1976; Iwakura 1977). M o r e specifically, a modal auxiliary may be subcategorized as + I [ + Tense] VP and then an example like (ia) is assigned a structure such as (ib) (after affix hopping):

On government by noun (i)

165

a. John may have been reading the book. b. . . . [ , [ [ + Tense], [ + N, - V]]] [ vp [v may] [vi> [v have] [Vp [v he-en] [Vp reading the book ]]]]

The feature + I [ + Tense] VP correctly ensures that a modal auxiliary precedes other auxiliary verbs and the main verb. 27. It should be noted that the Nondistinct-Governor Principle does not apply to cases such as (i) (from Chomsky 1970) since trace has no phonetic content: (i)

the city,'s destruction tj

With respect to the condition " a and β are not coindexed" in (41), it should be noted that A G R is coindexed with the N P it governs (Chomsky 1981). Thus A G R can govern the subject N P without violating the Nondistinct-Governor Principle since coindexing makes A G R a "distinct" governor of the subject NP. In this connection, it should be noted that an element in C coindexed with a trace can be a governor of the trace (Chomsky 1981; Lasnik - Saito 1984). 28. Williams (1983: 303 — 304), discussing a suggestion by Pesetsky, mentions a general "X XP" filter. With respect to the sequence "V VP", Heizo Nakajima points out to me that although Stowell (1983) holds that α = VP in examples like (i), the NondistinctGovernor Principle predicts that α cannot be VP: (i)

I saw [a John enter the room].

In this connection, it should be mentioned that Williams (1983) proposes a filter such as (ii) and argues against a small-clause analysis of examples like (i) since it is incompatible with the filter: (ii)

*V,prp

VP l P r P

Instead, he proposes to analyze perception and causative constructions as follows (Williams 1983: 303): (iii)

a. I [s/r saw John [vy> leave]] b. / [vp made John [Vp leave ]] 29. To permit these cases as well as the above cases of auxiliary verbs, the definition of "nondistinct governor" might be slightly modified as follows: (i)

α is a nondistinct governor of β if α and β have the same features [ ± N, + V], and (a) if α and β are not coindexed, or (b) if α is not subcategorized as + β.

30. Case (42e) is exemplified by examples such as the following (see note 18): (i)

a. *[..> J°hn decided to visit the country] delighted Mary. b. * It delighted Mary [IP John decided to visit the country]. In either case, the A G R in the IP clause is coindexed with the subject N P John (Chomsky 1981) and this index percolates up to the IP (cf. Williams 1982), and I assume that the indexed IP is not subject to coindexing with the A G R in the main clause. Hence the deviance of (ia) and (ib), as predicted by the Nondistinct-Governor Principle. 31. See Stowell (1981: 149) for a different approach to examples like (44a) and (44b). In this connection, Heizo Nakajima points out to me that conjunctions such as in that and except that are counter-examples to the restriction that Ρ cannot cooccur with a f/wf-clause complement. However, I agree with Emonds (1985: 249 — 250) who holds that they are compound subordinate conjunctions, each consisting of an atrophied Ρ

166

Kunihiro Iwakura

and that, which show "no syntactic behavior justifying anything beyond a completely amalgamated lexical unit." 32. Thus, the definition of "nondistinct governor" should be revised so that the features include [ + WH] as follows: (i)

α is a nondistinct governor of β if α and β have the same features [ ± N, ± V] and [ + WH], and (a) if α and β are not coindexed, or (b) if α is not subcategorized as + β.

33. With respect to the ungrammaticality of examples like (48), Chomsky states that undeletability of that in (i) below implies that "the position is not properly governed, if Stowell's theory is correct" (Chomsky 1986 a: 36): (i)

John expressed [NP the feeling [CP *(that)

the meeting should not be held]]

In fact, Stowell (1981) accounts for the difference in grammaticality between examples like (ii) and (iii) by maintaining that the noun claim, being in apposition with the tensed-clause complement, neither assigns a θ-role to the complement nor properly governs it and that the empty category in C violates the ECP (Stowell 1981: 397): . (ii) (iii)

*John 's claim [s· [e] [ s he would win ]] John's attempt [s- [e] [ s P R O to win]]

A possible problem for this account is noted by Stowell himself (1981: 430 note 30) with respect to examples like the following: (iv)

* Mary's happiness /certainty jawareness [s- [e] [ s Charles is leaving J] (Stowell 1981: 205)

According to Stowell, adjectives such as happy, certain, and aware, and their corresponding nouns have a special lexical feature, [ + R], and a [ + R] noun, just like a [ + R] adjective, assigns a θ-role to its tensed-clause complement (by a special rule of θ-role assignment) and hence properly governs the complement and its head (i.e., the empty category in C). Hence the ungrammaticality of examples like (iv) is left unexplained. Furthermore, Stowell (1982) notes that infinitival clauses may appear in positions that are not properly governed: (v)

a. [To ski down the slope] is thrilling. b. It is thrilling [to ski down the slope ].

To permit examples like (va) and (vb), he suggests that the infinitival clauses in these sentences lack a C position, although infinitival complements of control verbs are assumed to have an empty category in C. Finally, it is unclear how he will permit what Emonds (1970,1976) calls root sentences such as (vi) if tensed clauses always have an empty category in C: (vi)

[s< [e] [ s we climbed the mountain last

summer]].

34. If (49) is not base-generated as such, one possibility is that it is derived from a structure such as (i) by obligatory deletion of the empty C and the node CP (cf. Chomsky 1981): (i) John's [N- [N belief] [Cp [c e] [iP Bill is a romanticist]]] It should be noted that if a believe-type verb does not select an IP complement, it is necessary to assume that this type of verb obligatorily induces deletion of empty C and CP so as to account for the fact that this type of verb does not allow PRO to

On government by noun

167

appear in the subject position of its infinitival complement, as shown by the deviance of (ii) (from Chomsky 1980 a: 35): (ii) *I believe [PRO to VP]. Then it will not be unreasonable to assume that a noun corresponding to a believetype verb also obligatorily induces deletion of empty C and CP. 35. See, e.g., Haider (1986) for similar arguments. 36. Jackendoff credits this argument to Vergnaud (1974). 37. Pro-forms include anaphoric pronouns, onejones, thatjthose, do so, and so forth (see, e.g., Lakoff - Ross 1966; Wasow 1972; Palmatier 1972). Jackendoff (1977: 1 8 6 - 1 8 7 ) gives examples similar to (53a) and (53b) and writes that "the X"-complement theory of degree clauses has the same apparent difficulty of interpretation as the NP-complement theory of relatives: nonconstituent identity" (1977: 216). He notes that in the NPcomplement theory, it appears difficult to account for the semantic difference between the paired examples in (i) (Jackendoff 1977: 185): (i)

a. The man who hated lox who came to dinner. b. The man who came to dinner who hated lox. However, he argues that this is not a drawback of the NP-complement theory since he holds that the phenomenon of "stacking" should be accounted for in the system of presupposition and focus.

38. Jackendoff (1977) advances and develops an X'" system in which N'" = NP, A'" = AP, and V'" = S (cf. Emonds 1985). 39. This may not be a strong argument against the NP-complement theory since the pronoun it can refer to nonconstituents such as the unitalicized parts in the following examples from Wasow (1972: 109) (who credits this observation to Akmajian 1968 and Chomsky 1971): (i)

a. John punched Bill in the nose, but he wouldn 7 have done it to Pete. b. We may manage to eliminate water pollution, but we'll never do it with air pollution.

Evidently these are exceptional cases, and the NP-complement theory requires that examples like (53a) and (53b) also be treated as exceptional cases, whereas the Chomskyadjoined theory does not. 40. Although Jackendoff (1977) proposes to treat nonrestrictive relatives as N'" complements, there is not much motivation for this treatment except his desire to preserve his X'" system. His X'" system relies crucially upon the assumption that S is V'". If this assumption turns out to be untenable, there will be very little justification, if any, for his X'" system. See also Emonds (1979) for a critique of Jackendoffs approach to nonrestrictive relatives. 41. Jackendoff (1977: 186—187, 191) gives examples similar to (55) and (56) and discusses examples like (56) in connection with presupposition and focus.

Light verb constructions and the syntax-morphology interface* Tar ο Kageyama

1. Introduction Among the outstanding issues in generative grammar, the following two are of paramount importance from the viewpoint of morphology: (I) Where in the grammar does word formation take place? (II) How does word formation interact with syntax? The history of generative grammar has seen two opposing approaches to these questions, transformationalism and lexicalism, of which the latter is generally believed to have ousted the former. It is certainly true that unconstrained transformationalism is incompatible with the basic fact that words constitute special morphological units that are distinct from phrases and sentences. In a sentence She is a big housekeeper, for example, the adjective can only be interpreted as modifying the whole compound, and not just the noun house contained in it. This special property of "lexical integrity" or "syntactic atomicity" (DiSciullo — Williams 1987: 52) of words has led to the Strong Lexicalist Hypothesis, in which word formation is not scattered around in various components of the grammar (Jensen — Stong-Jensen 1984: 496) but is strictly confined to the lexical component, so that word formation processes have no chance of interacting with syntactic rules. The Strong Lexicalist Hypothesis thus provides straightforward answers to both question I and question II. Recently, however, evidence has been amassed, especially from agglutinative and polysynthetic languages, that demonstrates active interactions between syntax and morphology. In particular, Shibatani — Kageyama (1988) and Kageyama — Shibatani (1989) develop a modular theory of word formation, as represented in (1), on the basis of Japanese Ν + V compounding that operates directly on sentential and noun phrase

* I am grateful to Heizo Nakajima and Takao Gunji for helpful comments on earlier versions of this paper.

170

Taro Kageyama

structures. (See also Kageyama [1989]. An analogous model is proposed independently by Baker [1988] on the basis of syntactic incorporation in various languages of the world.) (1)

Modular morphology lexicon: lexical word formation Morphology ^ theory

D-strijcture

\

syntax: syntactic word formation This model offers significantly different answers from the Strong Lexicalist Hypothesis to questions I and II. As to question I, it allows word formation to take place in the syntactic component (and probably the phonological component as well) in addition to the lexical component. This, however, does not mean a return to the old transformationalism. On the contrary, we maintain that word formation rules in the syntactic component do not freely intermingle with transformational rules but are properly constrained by a set of general morphological conditions which checks the well-formedness of the outputs of word formation processes in various components in a global manner. These general constraints are thus viewed as a subtheory of the grammar, morphology theory, which functions in conjunction with other subtheories such as θ-theory. The functioning of morphology theory is readily illustrated with two types of passives in English. As shown by Wassow (1977) and Borer (1984 b), verbal (2a) and adjectival (2b) passives are best analyzed as obtaining in the syntax and in the lexicon, respectively: (2)

a. My boss was much interested by the new plan. b. My boss was very interested in the new plan.

The syntactic differences between the two types of passives are attributed to their different origins, while the morphological identity of regular (-ed) and irregular (e. g., broken) forms in both types is accounted for by common morphological conditions. To give another example, just as an ordinary (i.e., lexical) verbal compound consists of a verb and a noun, as in book-reading as opposed to * child-book-reading (cf. Selkirk 1982: 37), so the syntactic rule of Noun Incorporation can incorporate only one noun per verb (Shibatani — Kageyama 1988: 462). Thus, the model

Light verb constructions

171

in (1) makes interesting claims as to question II, allowing only limited (though as yet unexplicated) kinds of interactions between morphology and syntax. In this paper, we will provide additional support for the Modular Morphology model by analyzing light verb constructions in Japanese with occasional reference to the corresponding constructions in English (such as He took a look at the report). These constructions display interesting properties with respect to questions I and II. As regards question I, it is shown that light verb constructions, which are generated in syntax but behave in some ways like compound words, involve Abstract Incorporation, a kind of word formation that takes place in syntax. As regards question II, we will observe delicate interplay between syntax and morphology with respect to inheritance of θ-roles and Case properties.

2. Light verb constructions We will analyze periphrastic constructions like the following and an analogous construction in Japanese: (3) (4)

a. b. a. b.

John made a proposal of marriage to the ballerina. The new president will make an appearance at the party. The lieutenant took a close look at the body. Bill made an approach to the Japanese diplomat.

These constructions are characterized by a small set of "light verbs" (Jespersen 1961: 117) such as make, take, give, and have taking deverbal nouns as objects. Deverbal nouns may be derived with particular suffixes as in proposal and appearance in (3), or by conversion (zero suffixation) as in look and approach in (4). We will refer to the object NP headed by a deverbal noun as a "Predicational NP". The combination of a light verb and a Pred NP serves as a "complex predicate" (Cattell 1984). Semantically, these constructions are peculiar in that the deverbal nouns rather than the main verbs appear to θ-mark argument phrases (PPs). Thus the various PPs in (3) and (4) are linked to the predicational nouns (which in turn inherit these θ-roles from their base verbs) rather than to the verbs make and take, which normally do not select them. Even more puzzling is the fact that the semantic linkage between predicational nouns and PPs is contradicted by familiar tests of syntactic

172

Taro Kageyama

constituency. Passivization of (3a), for example, yields (5a) rather than (5b): (5)

a. A proposal of marriage was made to the ballerina. b. *A proposal of marriage to the ballerina was made.

This apparently indicates that the correct syntactic structure of (3a) is not (6a) but something like (6b), where the PP stands outside the Pred NP: (6)

a. NP John

VP NP a proposal of marriage to the ballerina

b.

a proposal of marriage

to the ballerina

In order to resolve this kind of mismatch between syntactic structure and semantic linking, Jayaseelan (1988) postulates a mechanism called "Argument Promotion" that operates on a syntactically motivated structure like (6b). Thus in (6b), make and proposal are assumed to have the θ-frames {Theme} and {Theme, Goal}, respectively (I omit Agent for the sake of simplicity). Within NP,, proposal assigns Theme to of marriage, while Goal is unassigned and hence is promoted. At the VP level, make assigns Theme to NP,, whereas the promoted Goal is assigned to the PP. Jayaseelan (1988: 98) goes on to propose that Argument Promotion is not specific to the complex predicate constructions but is an instantiation of the following general principles: (7)

A. θ-marking is strictly local. B. The θ-frame of a phrasal node is determined compositionally by its elements.

Jayaseelan's analysis, which aims to dispense with the special notion of "complex predicate", predicts that Argument Promotion takes place freely insofar as the conditions in (7) are met. But this prediction is obviously

Light verb constructions

173

wrong, since the possibility of Argument Promotion is limited to a particular set of "light verbs". Were made in (6b) replaced by accepted or declined, the sentence would become unacceptable, at least in the postulated syntactic structure: *John accepted/declined [NP a proposal of marriage] [PP to the ballerina]. It thus seems mandatory to specify what the main verbs are. In this respect, Grimshaw — Mester's (1988) analysis, dealing with an analogous construction in Japanese, seems more plausible. This Japanese construction takes the form of a Pred N P followed by the verb suru 'do'. The Pred N P is headed by a lexical category " V N " [Verbal Noun], which has a dual function of noun and verb: As a noun it can head NP, and as a verb it can Case-mark its object (if any) (cf. Kageyama 1982: 217).' Consider, then, the Japanese example (8), which closely corresponds to the English sentence (3a): (8)

John-wa [bareriina-ni] [kekkon-no mooshikomi-o] shita. J-TOP ballerina-DAT marriage-GEN proposal-ACC did 'John made a proposal of marriage to the ballerina.'

This sentence readily fits into the structure (6b), because in Japanese, case particles like ni 'to' can occur as such only in sentential structure, whereas arguments in N P structure must be accompanied by the genitive particle no. Compare the following: (9)

a. Amerika-ni jidoosha-o yushutsu-suru America-DAT car-ACC exportation-do 'to export cars to America' b. America-e-no jidoosha-no yushutsu America-DAT-GEN car-GEN exportation 'the exportation of cars to America'

Note that unlike the English preposition of, the Japanese no occurs on all NPs and PPs within a NP (except the head itself)· (In [9b], additional changes are noted: the dative ni is replaced by another dative marker -e and the accusative -o drops out.) The absence of the genitive marker thus provides prima facie evidence that bareriina-ni 'to the ballerina' in (8) is outside the Pred NP. Based on this structure, Grimshaw — Mester (1988: 211) postulate an impoverished lexical entry (10) for the "light verb" suru 'do': (10) suru

a. θ-marking: ( ) b. Case marking: (accusative)

174

Taro Kageyama

(10a) indicates that suru is devoid of its own argument structure, by virtue of which it can take over the unassigned argument roles of the VN and discharge them to PPs at the clausal level. To illustrate this with (8), the Theme role of 'marriage' is assigned within the Pred NP while the Goal role is first transferred to suru and then assigned to the 'to'-phrase. Grimshaw — Mester call this manner of θ-assignment "Split θ-marking". Now, although suru does not assign any θ-role to its "object", the Pred NP in (8) is nonetheless marked with accusative (-o). This accusative Case, Grimshaw — Mester hypothesize, is assigned by suru, as stipulated in (10b). The two assumptions in (10) provide the starting-point for our further investigation. Since the constructions under consideration comprise two predicates, a light verb and a predicational noun, it is convenient to divide our discussion into two parts: (i) external syntax or the syntax of light verbs (sections 3 and 4), and (ii) internal syntax or the syntax of predicational nouns (sections 5 and 6).

3. Case marking of predicational NPs Previous discussions on light verb constructions have largely focused on thematic structure, ignoring Case properties. As far as the English light verb constructions are concerned, Case will not pose any significant problem, since all Pred NPs can occur in the object position of a light verb, presumably bearing accusative Case. In Japanese, a language with overt case-markers, however, the situation is radically different. The discussion in this section reveals that Case plays no less important role in inheritance than 0-roles. We shall begin with the second part (10b) of Grimshaw — Mester's lexical entry — the stipulation that suru necessarily assigns accusative Case to Pred NPs. This stipulation, it turns out, is empirically false: Not all Pred NPs can be comfortably marked with the accusative. The possibilities of accusative-marking can be seen most clearly in intransitive VNs. Compare (11) with (12): (11)

iede-o suru 'to run away from home', shokuji-o suru 'to have a meal', jisatsu-o suru 'to commit suicide', undoo-o suru 'to take exercise', uwaki-o suru (lit.) 'to have a flirtation', toorui-o suru

Light verb constructions

175

'to steal base', rikon-o suru 'to get a divorce', dansu-o suru 'to do a dance', arubaito-o suru 'to do a part-time job' *shiboo-o suru 'to die', *tanjoo-o suru 'to be born', *seechoo-o suru 'to grow', *zetsuboo-o suru 'to be in despair', *(ziko-ga) tahatsu-o suru '(for accidents) to happen frequently', *(bukkaga) jooshoo-o suru '(for prices) to go up', * (pisutoru-ga) boohatsu-o suru '(for a pistol) to go off by accident', *(iro-ga) matchi-o suru '(for colors) to match', * (shinkyoku-ga) hitto-o suru '(for a new song) to be a hit'

(12)

The VNs in (11) designate volitional actions that can be controlled by the subjects, whereas those in (12) denote non-volitional events that are beyond the control of the subjects. Although the grammaticality judgment here is rather subtle, many native speakers of Japanese find only the former to be compatible with accusative-marking. The significance of volitionality or controllability will be further shown by the following observations: (13)

a.

Musuko-ga iede-o shita. son-NOM run-away-from-home-ACC did 'My son ran away from home.' b. * Musuko-ga joohatsu-o shita. son-NOM evaporate-ACC did 'My son is missing; he disappeared the way water evaporates.'

Although these two sentences describe virtually the same situation, their ways of presentation are different. (13a), compatible with accusativemarking, describes the subject's will to take the particular action, whereas (13b), incompatible with accusative, gives an objective depiction of the fact that the son disappeared like an inanimate substance, where the subject's volition is not involved. Examples like (14) corroborate the feasibility of subject volitionality: (14)

a. Densha-ga hassha-o shita. train-NOM start-ACC did 'The train left', b. Basu-ga kyuuteesha-o shita. bus-NOM sudden-stop-ACC did 'The bus made a sudden stop'.

These sentences are judged acceptable if we attribute volitionality to the subjects 'train' and 'bus' (or, more precisely, to their drivers). 2

176

Taro Kageyama

The presence or absence of accusative is reflected in the syntactic behavior of Pred NPs. The VNs that permit accusative-marking can undergo H^A-movement such as Topicalization, Relativization, and Pseudo-Cleft, whereas those that do not cannot: (15)

a.

Iede-wa, Taroo-ga shotchuu shite-iru. run-away-from-home-TOP T-NOM often doing-is 'Taro often runs away from home.' Daijin-ga shita totsuzen-no jisatsu-wa kokumin-o minister-NOM did sudden suicide-TOP nation-ACC odorok-ase-ta. surprise-CAUSE-PAST 'The sudden suicide of the minister surprised the nation.' Nichiyoobi-ni Ken-ga shita no wa saikuringu da. Sunday-on K-NOM did COMP TOP cycling Copula 'What Ken did on Sunday was to go cycling.' b. *Boohatsu-wa pisutoru-ga shotchuu shite-iru. go-off-by-accident-TOP pistol-NOM often doing-is 'Pistols often go off by accident.' * Daijin-ga shita totsuzen-no shiboo wa kokumin-o minister-NOM did sudden death TOP nation-ACC odorok-ase-ta. surprise-CAUSE-PAST 'The sudden death of the minister surprised the nation.' *Shin-kyoku-ga shita no wa hitto da. new-song-NOM did COMP TOP hit Copula 'What the new song was was a hit.'

The disparity between (15a) and (15b) follows from the assumption in Government-and-Binding theory that traces of ^ - m o v e m e n t must be Case-marked. Let us return to the contrast between (11) and (12). Such a distinction of intransitive verbals has already been discussed in Relational Grammar and Government-and-Binding theory under the rubric of unaccusativity. Particularly relevant is Burzio's (1986: 178) generalization: (16)

All and only the verbs that can assign θ-role to the subject can assign accusative Case to an object.

The "θ-role" mentioned in (16) typically refers to Agent. Since transitive verbs typically have Agentive subjects, they can assign accusative Case to their direct objects. Intransitive verbs, which by definition are not

Light verb constructions

177

subcategorized for by direct objects, are nevertheless capable of assigning accusative if their subjects are Agentive. Thus we have two groups of intransitives: "unergatives", which assign a θ-role to the subject, and "unaccusatives", which do not. The potential accusative Case of unergative verbs, Burzio claims, may be realized on expletive objects, as in (17a), but such expletive objects are impossible with unaccusative verbs (17b): (17)

a.

He talked my head off. They walked the hell out of those shoes, b. *They arrived the hell out of the bus terminal. *The troops were withdrawing the hell out of Vietnam.

It now appears that the Japanese data in (11) and (12) accord with the unergative-unaccusative distinction. Pred NPs with unergative VNs (11) serve as spurious objects of the light verb suru, comparable to the expletive objects in (17a), and hence are assigned accusative. Notice that Burzio's generalization does not identify the specific θ-role label of subject, such as Agent. In point of fact, accusative marking of Pred NPs seems possible not only with Agent subjects but also with Experiencer subjects, as in (18): (18)

a. Kodomo-ga kega-o shita. child-NOM injury-ACC did 'The child got injured.' b. Boku-wa i-no shujutsu-o shita. I-TOP stomach-GEN operation-ACC did Ί underwent an operation on my stomach.'

These observations indicate that the accusative Case on Pred NPs is not a lexical idiosyncrasy of suru as Grimshaw — Mester assume, but is generally derivable from the semantic properties — especially, volitionality — of the VNs themselves. Leaving the exact nature of the relevant semantic properties for future investigation, we will provisionally assume that transitive and unergative VNs have the feature [ + acc] while unaccusative VNs have the feature [— acc]. Then the above correlation between the Case-marking of Pred NPs and the types of VNs will be formally captured in terms of inheritance of those Case properties: the Case features are first transmitted (or perhaps, copied) onto suru and are then realized on the Pred NPs. What we obtain is, so to speak, "reflexive" Case-marking:

178

Taro Kageyama VP

(19)

NP (o)·*VN

V suru

[a acc] Why does such a strange phenomenon occur? An obvious answer resides in Case. As mentioned earlier, VNs have a dual function of verb and noun. As verbs they can assign Case, but as nouns they must receive Case. Thus Pred NPs, headed by VNs and placed in the subcategorized ("object") position of suru, must meet the Case Filter requirement, in the same way as ordinary NPs. The reflexive Case-marking stated above is a ready way to fulfill this purpose for Pred NPs with [ + acc] VNs, whereas those with [ — acc] VNs have no way of receiving Case and hence result in the breach of the Case Filter, as in (20a). (20)

a. *Pisutoru-ga boohatsu-0 (kyuuni) shita. Pistol-NOM go-off-0 (suddenly) did 'The pistol suddenly went off.' b. Pisutoru-ga (kyuuni) [boohatsu-shita ]v. Pistol-NOM (suddenly) [go-off-did]v

Japanese offers a special strategy, however, which serves to resolve the ungrammaticality of (20a): a syntactic rule of Incorporation which combines a VN and suru to yield a compound verb (Kageyama 1976 — 1977; 1982). Since incorporated nominals generally need not bear Case as argued by Baker (1988), application of VN Incorporation to (20a) gives rise to the well-formed (20b). More generally, VNs of the unaccusative type are licensed only in the incorporated structure, as in (21a). 3 Here we observe the interaction of a syntactic rule (Case assignment) and a word-formation rule (VN Incorporation): (21)

a. shiboo-suru 'to die', boohatsu-suru 'to go off by accident', hittosuru 'to be a hit' b. shokuji-suru 'to dine', jisatsu-suru 'to commit suicide', undoosuru 'to be exercise'

Incidentally, VN Incorporation is not limited to unaccusatives but freely applies to unergatives and transitives as well, as shown in (21b). This observation presents a new argument for the syntactic rule of VN Incorporation. In Kageyama (1982: 246), I gave (22) as evidence that VN -suru compounds are derived in syntax:

Light verb constructions

(22)

179

Gakkai-de, Amerikajin-wa yoku academic-conference-LOC Americans-TOP frequently [hatsugen-suru]v ga, Nihonjin-wa amari [remark r do] but Japanese-TOP much prOi [shi-nai]^. proi [do-not]v 'At academic conferences, Americans often speak out while Japanese seldom do.'

Note that the verb 'do' in the second conjunct is the light verb suru, and not an auxiliary verb due to "Do-support", which is simply lacking in Japanese. Now, if the whole compound verb hatsugen-suru 'to speak out' were derived in the lexicon and inserted as such at D-structure, (22) should be outlawed by the Anaphoric Island Constraint (Postal 1969), because then the coreferential rule that identifies pro with hatsugen 'remark' would have to penetrate into the compound verb. Such a problem does not arise in the syntactic Incorporation analysis, where (22) is derived as shown in (23): (23)

(D-structure) Amerikajin-wa yoku hatsugenro suru Americans-TOP frequently remark-ACC do ga, Nihonjin-wa amari prOj shi-nai. but Japanese-TOP much pro; do-not (Incorporation) —*· Amerikajin-wa yoku t; Americans-TOP frequently tj [hatsugen-suru]v ga, Nihonjin-wa amari prOj [shi-nai]^. [remarkj-do]v but Japanese-TOP much pro, [do-not]v

At D-structure the VN hatsugen 'remark' in the first conjunct is generated as the "object" of suru, and after Incorporation its trace is interpreted as the antecedent of the null pronoun pro in the second conjunct. Note that VN Incorporation does not apply to the second conjunct of (23) because empty categories in general do not participate in word formation. Since the coreferential relation is determined at the clausal level, no violation of the Anaphoric Island Constraint ensues. Miyagawa (1987: 37) argues against my syntactic analysis, providing (24) as a problematic case: (24)

*Taroo-wa itsumo seekoo-suru ga, Hanako-wa Taro-TOP always success r do but Hanako-TOP tokidoki shika prOj shi-nai. sometimes only pro, do-not 'Taro always succeeds, but Hanako only sometimes.'

180

Taro Kageyama

Since (24) is ungrammatical, Miyagawa contends that the argument I advanced for (22) does not hold. In fact, however, (24) can be viewed as supportive evidence of the Incorporation analysis. What is crucial is the fact that the VN used in (24), namely seekoo 'succeed', is unaccusative; as Miyagawa himself notes it is incompatible with accusative marking: *seekoo-o suru 'success-ACC do', and the patterns of (22) and (24) seem to hold generally for unergative/transitive VNs and unaccusative VNs, respectively. Postulating the same syntactic derivation for (24), we should expect something like (25) as its S-structure: (25)

*Taroo-wa Taro-TOP tokidoki sometimes

itsumo tj [seekoorsuru] always tj [success r do] shika prOj [shi-nai]y. only pro; do-not

ga, but

Hanako-wa Hanako-TOP

Provided that the coreference rule matches pro and its antecedent in the Case-assigning property as well as in other features, we find that the pro in (25), coreferential with (the trace of) the unaccusative seekoo 'success', is not assigned Case (by reflexive Case-marking), nor can it be incorporated. Thus this pro must remain Caseless. On the assumption that Japanese null pronouns need to be Case-marked in the same way as English personal pronouns, the Caseless pro impinges upon the Case Filter requirement. 4 In other words, we consider that (25) is ruled out for precisely the same reason as is (20a) above. In contrast, (22) is wellformed because the pro there, matching the unergative hatsugen 'remark', is assumed to bear accusative Case. Our syntactic Incorporation analysis can thus provide a principled account of the contrast between (22) and (24) in terms of the presence or absence of Case on Pred NPs. Returning to reflexive Case marking, we find striking confirmation in a special group of VNs that require nominative {ga) instead of accusative: 5 (26)

a. amamori-ga suru 'a rain-leak occurs', jihibiki-ga suru 'rumbling of the ground occurs', inabikari-ga suru 'lightning occurs', toonari-ga suru 'distant rumbles (of thunder) are heard', sokobiega suru 'to feel chilled to the bone' b. munasawagi-ga suru 'an uneasy feeling occurs', ikigire-ga suru 'breath runs short', muneyake-ga suru 'heartburn occurs', memai-ga suru 'to feel dizzy', kokorogawari-ga suru 'a change of heart occurs', yokan-ga suru 'to have a presentiment'

These VNs, typically denoting the occurrences of natural (26a) or physiological (26b) phenomena, are semantically categorized in the unaccu-

Light verb constructions

181

sative class. We may point out that most VNs in (26) have the morphological composition of subject + intransitive verb (e.g., ama- 'rain' + mori 'leak'), where it might appear as if the nominative Case assigned to the subject nouns inside these compound VNs were percolated up to the whole of the Pred NPs. Since, however, not all these VNs have such a compound structure, we must assume that the nominative here is inherently endowed with these VNs as a whole. Postulation of such an inherent nominative Case by no means contradicts Burzio's generalization, because it only precludes structural accusative Case, where the possibility remains of having an inherent Case other than the accusative (see also Belletti [1988], who argues for an inherent partitive Case with Finnish unaccusative verbs). The inherent nature of the nominative in (26) can be seen from the fact that the VNs of this group resist incorporation, in conformity with the general constraint that PPs or inherently Case-marked NPs are barriers for Noun Incorporation (cf. Baker 1988): (27)

*[inabikari-suru]„ lightning-do

*[muneyake-suru]v heartburn-do

In this subsection, we have shown that the case-marking of Pred NPs in the Japanese light verb construction is generally determined by the lexicosemantic properties of the VNs themselves. This observation suggests that unaccusativity is not so much a matter of syntactic configuration as is usually assumed in Government-and-Binding syntax, as one of semantics such as the Agentivity (or in some cases, the Experiencerhood) of a subject entity, as discussed by Van Valin (1987). It will not be unreasonable to assume that the assignment of accusative to VNs, which do not seem to have any argument relation with suru, is induced by the general AgentPatient pattern in which Agent prototypically bears nominative and Patient accusative.

4. Theta-marking of Pred NPs In the foregoing section, we proposed that the Case features of VNs are percolated (or copied) to suru to provide this "light" verb with a Caseassigning property. Now, is it only the Case features that are subject to percolation? In this section we will observe that certain information on the thematic structure of VNs is also susceptible to percolation.

182

Taro Kageyama

In (15) we saw a disparity in the applicability of W^-movement between unergative and unaccusative VNs, where it was found that only Casemarked Pred NPs may undergo W^A-movement. We shall now turn our attention to NP movement, in particular to Passivization. The question we address is whether VN(-o) suru 'do VN' can be passivized into VNga s-areru Ύ Ν is done'. First of all, unaccusative VNs are not qualified for Passivization, since they lack the accusative Case which is necessary for Passivization, in the first place: (28)

a. *Shiboo-ga s-areta. die-NOM do-PASS lit. 'To die was done.' b. *Seechoo-ga s-areta. grow-NOM do-PASS lit. 'To grow was done.'

The examples in (29), however, reveal that the presence of accusative Case is a necessary, but not sufficient, condition on Passivization: (29)

a. *Mai-asa sanpo-ga s-areru. every-morning walk-NOM done-is Ά walk is taken every morning.' b. *Kinmonkyoo-de yoku jisatsu-ga s-areru. Golden-Gate-Bridge-at often suicide-NOM done-is 'Suicide is often committed from the Golden Gate Bridge.'

The VNs in (29) are unergative intransitives, and despite their ability to assign accusative, they fail to undergo Passivization. These should be contrasted with the VNs in (30), which are transitive and amenable to Passivization: (30)

a. Kekkon-no mooshikomi-ga s-areta. marriage-GEN propose-NOM do-PASS Ά proposal of marriage was made.' b. Enjo-no yoosee-ga s-aretajnas-areta. aid-GEN request-NOM do-PASS Ά request for aid was made.' c. Kuruma-no tenken-wa mada s-arete-inai. car-GEN inspection-TOP yet do-PASS-not 'An inspection of the car has not been carried out yet.'

One might be tempted to postulate two lexical entries for suru: intransitive suru in (28) and (29), and transitive suru in (30). Such a proposal, however, completely misses the correlation between the passivizability and the

Light verb constructions

183

transitivity of embedded VNs as observed above. It is more reasonable to assume that all these examples have one and the same suru, and that its seemingly disparate behavior is ascribed to VNs. Specifically, the generalization can be captured by assuming that the transitivity property of a VN is copied onto suru, with the result that suru takes on the same transitivity value as the VN. If the VN is intransitive, then suru is also intransitive, so that there is no chance of Passivization; if the VN is transitive, suru also becomes transitive, whereby the Pred NP is looked upon as the formal "object" of suru and undergoes Passivization. In the terminology of Government-and-Binding theory, this correlation can be formally captured in terms of Case and θ-role. Below are the Case and θ-role properties of the three classes of verbals: (31) a. unaccusative b. unergative c. transitive

accusative

"object" θ-role

no yes yes

yes no yes

Assuming with Grimshaw — Mester (cf. [10a] above) that suru has no thematic structure of its own, we can generalize that not only the Case feature but also the θ-role property of a VN is copied onto suru and then projected back to the Pred NP in "reflexive" fashion. This provides us with the proper delimitation of the range of passivizable VNs: Passivization applies only to those Pred NPs that bear both Case and θ-role, namely transitive VNs. The reflexive Case and θ-role assignment will in turn be necessitated by two independent assumptions in Governmentand-Binding theory: (i) Passivization absorbs accusative Case, and (ii) the terminal of an Α-chain must have a θ-role. 6 If this analysis is correct, it follows that Pred NPs headed by unergative VNs are "defective" in that they have only Case without a θ-role. Such a situation, however, is generally believed not to occur, as stipulated, for example, by Jones' (1988: 105) Revised Case Filter: (32)

*NP α θ-role -α Case

In this respect, unergative Pred NPs are like expletive NPs in object position (Postal - Pullum 1988). Our analysis of the Japanese suru construction will provide a clue to the understanding of English light verbs: make, give, and take. Here a

184

Taro Kageyama

distinction seems necessary between two classes of predicational nouns: those with overt suffixes (e. g., recommendation, amendment, approval, comparison) and those with zero-suffix (e.g., offer, attack, look, laugh). The former readily passivize with make, give, and take (Live 1973: 38). This means that they are no different from ordinary NPs used as objects of transitive verbs and receive both Case and θ-role from the light (or rather, "heavy") verbs. The matter is more complicated with zero-suffixed deverbals. Among make, take, and give, make is least restricted, allowing Passivization regardless of the transitivity of a predicational noun: (33)

a. An exchange of opinions was made, [transitive] b. A leap was made, [intransitive]

As Live (1973: 37) aptly notes, Passivization with make is feasible even for intransitive predicational nouns whose corresponding simple verbs cannot be passivized (e.g., exit, advance, appeal, bow, comment). This is accounted for by assuming that make assigns Pred N P a θ-role on its own (perhaps Chomsky's [1981] "quasi-argument role" # , as proposed by Cattell [1984: 52]). In contrast, the passivizability of take and give appears to be contingent upon the transitivity of predicational nouns. Thus Live (1973: 37 — 38) observes that these two verbs passivize readily with virtually all predicational nouns of transitive origin (with some exceptions, such as taste, smell, and fancy), but with only about half of the intransitives she collected. According to Live's judgment, the possibilities of Passivization with intransitive nouns are as follows (I give only those deverbals that take an indefinite article): (34)

take passivizable: bow, dive, escape, fall, journey, nap, pause, profit, rest, ride, shower, smoke, stand, step, swim, tour, walk unpassivizable: exit, glance, halt, jump, leap, look, lunge, peek, plunge, spill, swerve, turn, whirl give passivizable: account, bid, comment, dance, lecture, pantomime, remark, reply, retort, signal, talk unpassivizable: bound, bow, curtsey, echo, glance, laugh, leap, look, lunge, move, slip, splash, start, step, swerve, turn, whirl

As far as the unpassivizable intransitives and passivizable transitives are concerned, we would apply the same analysis that we postulated for

Light verb constructions

185

Japanese. That is, the object θ-role (if any) of a predicational noun (more precisely, of its base verb) is copied to take and give, with the result that only transitive predicational nouns that take objects can undergo Passivization. The passivizable intransitives pose a problem, however. We can only speculate that they have attained the full status of nouns and thus are assigned a θ-role by the verbs (as "heavy" verbs). Furthermore, as is usually the case with word formation phenomena, there are indeterminacies due to lexical idiosyncrasies and dialectal/idiolectal variations. For example, it is difficult to give a principled reason why take a bow passivizes but give a bow does not. For the present I must content myself with the simple conclusion that the English make, take, and give, which assign Case and on some occasions a θ-role as well, are not so "light" as the Japanese suru, which entirely lacks θ-role and Case on its own.

5. Conditions on the internal structure of Pred Ν Ρ Thus far we have been concerned with the external syntax of light verb constructions, namely the Case and θ-marking of Pred NPs in relation to light verbs. We shall now explore the internal structure of Pred NPs. Grimshaw — Mester (1988: 215) argue that three generalizations hold for the positioning of various arguments of VNs inside and outside Pred NPs in the Japanese construction: (35)

i. The subject argument must always be outside the NP. ii. At least one argument apart from the subject must be outside the NP. iii. For VNs that take a Theme and a Goal, if the Theme argument is realized outside NP, the Goal must also be realized outside NP.

According to Grimshaw — Mester, generalization (i) rules out examples like (36b), and generalization (ii) examples like (36c): (36)

a.

Ken-ga Naomi-ni puropoozu-o shita. K.-NOM N.-DAT proposal-ACC did 'Ken made a proposal of marriage to Naomi.' b. *[Ken-no Naomi-e-no puropoozu-o]w shita. K.-GEN N.-DAT-GEN proposal-ACC did c. * Ken-ga [Naomi-e-no puropoozu-o ]NP shita. K.-NOM N.-DAT-GEN proposal-ACC did

186

Taro Kageyama

I take it that generalization (i) is correct except for the special group of nominative VNs discussed earlier in (26), and that it need not be stipulated but follows from Predication theory (Rothstein 1983). I also concur in generalization (iii), whose effect is illustrated below: (37)

a.

Shoonen-ga [murabito-ni] [ookami-ga kuru to] boy-NOM villagers-DAT wolf-NOM come COMP keekoku-o shita. warning-ACC did 'The boy warned the villagers that the wolf was coming.' b. Shoonen-ga [murabito-ni] [[ookami-ga kuru to]-no boy-NOM villagers-DAT wolf-NOM come COMP GEN keekoku]-o]w shita. warning-ACC did c. * Shoonen-ga [ookami-ga kuru to] [murabito-e-no boy-NOM wolf-NOM come COMP villagers-DAT-GEN keekoku-o]^ P shita. warning-ACC did

The grammatical sentence (37a) has all three arguments (Agent, Goal, Theme) outside the Pred NP. The contrast between (37b) and (37c) is attributed to generalization (iii), which Grimshaw — Mester (1988: 225) argue is explained by postulating a hierarchically organized argument structure in lieu of an unordered list of θ-role labels as is customarily assumed (e.g. Williams 1981): (38)

keekoku 'warning':