Categorization and Category Change [Unabridged] 144385140X, 9781443851404

Citation preview

Categorization and Category Change

Categorization and Category Change

Edited by

Gianina Iordăchioaia, Isabelle Roy and Kaori Takamine

Categorization and Category Change, Edited by Gianina Iordăchioaia, Isabelle Roy and Kaori Takamine This book first published 2013 Cambridge Scholars Publishing 12 Back Chapman Street, Newcastle upon Tyne, NE6 2XX, UK British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Copyright © 2013 by Gianina Iordăchioaia, Isabelle Roy, Kaori Takamine and contributors All rights for this book reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner. ISBN (10): 1-4438-5140-X, ISBN (13): 978-1-4438-5140-4

TABLE OF CONTENTS

Preface ....................................................................................................... vii PART I: CATEGORIES AND CATEGORIZATION Chapter One ................................................................................................. 3 Categories and Categorization: Introduction Gianina Iordăchioaia, Isabelle Roy and Kaori Takamine Chapter Two .............................................................................................. 11 The Category of Participles Björn Lundquist Chapter Three ............................................................................................ 33 DP-decomposition Analysis of Japanese Demonstrative so Kaori Takamine Chapter Four .............................................................................................. 59 Is Russian a Verb Classifier Language? Laura A. Janda PART II: ISSUES IN CATEGORY CHANGE Chapter Five .............................................................................................. 87 Issues in Category Change: Introduction Gianina Iordăchioaia, Isabelle Roy and Kaori Takamine Chapter Six ................................................................................................ 95 Towards a Syntactic Account of Affix Combinations: From Nouns to Adjectives and Vice Versa Antonio Fábregas Chapter Seven.......................................................................................... 123 Event Related Nominalizations Isabelle Roy and Elena Soare

vi

Table of Contents

Chapter Eight ........................................................................................... 153 ‘Direct Participation’ and ‘Agent Exclusivity’ Effects in Derived Nominals and Beyond Artemis Alexiadou, Gianina Iordăchioaia, Fabienne Martin, Florian Schäfer and Mariángeles Cano Contributors ............................................................................................. 181 Index ........................................................................................................ 183

PREFACE

This book is a collection of selected papers that were presented at the Workshop "Categorization and category change in morphology", which was held at the University of Tromsø in December 2011. The workshop was organized by Antonio Fábregas and Kaori Takamine with the support of the project “Nominalizations: explorations at the syntax-lexicon and the syntax-semantic interfaces (NOMEXPL)” which was funded by the Research Council of Norway together with the German Academic Exchange Service and EGIDE, the French national agency for the promotion of higher education, international student services, and international mobility, via three joint mobility programs: the Aurora program between Norway and France, the German-Norwegian collaborative research support scheme and the PROCOPE program between France and Germany. The book addresses theoretical and empirical issues related to categorization and category change in syntax and morphology. Linguistic descriptions have always made widespread use of lexical categories, i.e., divisions of words into distinct “parts of speech”. Nouns, verbs and adjectives (and sometimes adpositions) are generally considered the three (sometimes four) major lexical classes. Despite this prevalence in linguistics, lexical categories remain a relatively under-developed area of formal linguistic theory and many open questions remain to be addressed. Common questions that generally arise concern the proper definition of the classes with their specific properties, the inventory of categories across languages, and the link between categories and formal linguistic theory. The book is structured in two thematic parts. The first part, Categories and categorization, consists of papers that are concerned with means to distinguish among categories in the lexicon and in the syntax, whether they fall within the well-defined categories or pose a challenge to the traditional definition of categorial classes. The second part, Issues in category change, deals with the specific syntactic and morphological derivational processes that are at play when words shift category. It is concerned with the formation of complex words, in particular, how properties of the source category are preserved or modified in the output. The individual contributions in the volume are in the areas of formal syntax, morphology, the syntax-semantics and the syntax-morphology interfaces. The relevant issues are explored within various theoretical and corpus-based frameworks and within a wide range of languages including

viii

Preface

English, French, German, Greek, Japanese, Romanian, Russian, Spanish, and Swedish. The editors thank the organizers and the audience of the workshop in Tromsø. We also thank the reviewers of the papers and the authors for contributing their works and for participating in the reviewing process.

PART I: CATEGORIES AND CATEGORIZATION

CHAPTER ONE CATEGORIES AND CATEGORIZATION: INTRODUCTION GIANINA IORDĂCHIOAIA UNIVERSITÄT STUTTGART

ISABELLE ROY UNIVERSITÉ PARIS 8–CNRS

KAORI TAKAMINE NORWEGIAN UNIVERSITY OF SCIENCE AND TECHNOLOGY

The classification of words into categories is found from the earliest days of the history of linguistics. Most of the classifying strategies of the old grammars are still used nowadays to identify categories: morphological patterns (e.g., patterns of inflection and/or derivation), distributional patterns (i.e., combination with other lexical items; e.g., determiners for Nouns) and interpretational properties of lexical items. PƗnini (4th century BC) in his treatise of Sanskrit grammar Ashtadhyayi, categorizes nouns by gender and inflections for case and number. Dionysius Thrax (100 BC) in one of the earliest Greek grammars, the Art of Grammar (ȉȑȤȞȘ īȡĮȝȝĮIJȚțȒ) distinguishes, on the one hand, a class of words (called ónoma i.e., nouns), which inflect for case and express “concrete and abstract entities”, and on the other hand, another class of words (called rhema i.e., verbs), which inflect for tense and person and express “an activity or process performed or undergone”. The morpho-syntactic identification of lexical classes immediately raises two issues regarding the universality of categories. First, to a certain extent, classificatory criteria are language specific and must be defined for each language (or language family) separately. For instance, both French and Russian have an identifiable class of nouns, but French does not inflect its nouns for case and Russian does not introduce them with determiners. So there is not a universal definition of the morpho-syntactic properties of nouns, for instance, in that sense. Second, there are

4

Chapter One

significant variations cross-linguistically as to what counts as a relevant distinction between categories. Japanese, for instance, is known to distinguish two classes of words which function semantically like adjectives (i.e., they denote properties) but they do not exhibit typical morpho-syntactic properties of adjectives. One class (the i-adjectives) are often called verbal-adjectives or adjectival-verbs because they inflect for tense, and they are morphologically similar to stative verbs. Another class (the na-adjectives) are often called nominal-adjectives or adjectival-nouns because, as nouns, they need to combine with the copula da in order to receive tense and have a subject (Backhouse 2004). Verbal-adjectives and verbs are morphologically similar but also show differences, and so do nominal-adjectives and nouns; are these differences sufficient to motivate a separate lexical class? A positive answer would force a multiplicity of lexical categories (Japanese, in this example, would have two distinct lexical classes of adjectives; also distinct from nouns and verbs). If the answer is negative, and depending on what is assumed to be the major criterion for classification, a proper description will lead to a multiplicity of sub-classifications inside a given class of words (‘type1’/‘type2’; ‘pure’-X/‘quasi’-X; class X/class Y/class XY, etc.). The definition and the identification of lexical classes with their properties lead to the rampant issue of mixed-categories more generally. Mixed-categories are lexical items that share properties of more than one lexical class. Japanese (verbal/nominal-)adjectives can be considered mixed-categories (cf., above); but so can participles and infinitives in Romance and Germanic languages, for instance. To take an example, participles in Dutch can exhibit in the same utterance properties of As (i.e., their use as prenominal modifiers) and have typically verbal complements (Sleeman 2011): de met zijn handen etende man (lit. the with his hands eating man) ‘the man eating with his hands’. Similarly, Spanish infinitives can be constructed with a determiner and take a subject at the same time: El cantar yo la Traviata (lit. the sing-inf I-nom the Traviata) ‘Me singing la Traviata’. Mixed-categories are a problem for feature-based approaches to lexical categories, since these specify discrete classes. The introduction of X-bar theory (Chomsky 1970) offered to identify and distinguish lexical categories on the basis of a restricted set of internal features. The major categories were defined on the basis of two Boolean category features, +/íN, +/íV, leading to a typology that recognizes four lexical categories. (Chomsky 1970 does not mention prepositions, which we call here Adpositions as to encompass pre- and post-positions; these were later added by Jackendoff 1977.)

Categories and Categorization: Introduction

(1)

Nouns: Verbs: Adjectives: Adpositions

5

+N; íV íN; +V +N; +V íN; íV

Other attempts at formalizing the major lexical categories within a binary distinctive feature-based system are in terms of functional features (i.e., the function that the part of speech operates in syntax) rather than lexical features (i.e., N and V). For instance, Jackendoff (1977) proposed a system based on whether lexical items can have subjects and/or objects (+/ísubj; +/íobj): both verbs and nouns can have subjects, while adjectives and adpositions cannot; verbs and adpositions can take complements whereas nouns and adjectives cannot. The typology (now outdated) has been shown not to stand cross-linguistic examination. Déchaine (1993) argues for a rather similar partition, but with a mixed system of lexical and functional features that makes use of referentiality instead of subjecthood: both nouns and verbs are referential (the latter referring to events rather than individuals). Baker (2003) notes that all these systems are rather arbitrary and that the features and the assignment of values do not always naturally explain some of the syntactic properties of lexical categories. He argues that a proper understanding of the three lexical categories N, V, and A must make use of two distinct and orthogonal dimensions and cannot be achieved by working with purely lexical or purely functional features alone. Instead, he proposes that both their internal syntactic properties and their meaning be taken into consideration, yielding a typology based on referentiality (formalized by a referential indexing) and projection of a specifier. Verbs are defined as lexical categories that take a specifier (and hence a subject), and nouns as bearers of a referential index. The third lexical category, adjectives, is distinguished negatively, having neither of these properties. The features are claimed to be universal. Cross-linguistic variation is in part dealt with by functional projections that can be overt or covert, partly blurring the categorical distinctions. (For Baker adpositions are functional rather than lexical items and thus escape the classification.) (2)

Verbs: Nouns: Adjectives:

+specifier +referential index íspecifier; íreferential index

An entirely different generative approach to lexical categories is taken in Distributed Morphology (Marantz 1997, Harley and Noyer 2000, Arad

6

Chapter One

2003) and in the Exo-skeletal approach developed in Borer (1999, 2005). In these frameworks, both representative of syntactic approaches to word formation, the smaller lexical pieces are roots, which are category neutral. Categories are determined by the syntactic environment in which the root is inserted. In Distributed Morphology, categories are specified via dedicated functional heads (little v, little n). For Borer no such dedicated position is assumed, and instead lexical categorization is the result of a “top-down” combination with functional projections that are nominal or verbal. A mixture of the two syntactic models is assumed in Acquaviva (2008). For a recent discussion and more background on lexical categories and categorization in syntactic approaches see Panagiotidis (2011) and references therein. Feature-based approaches and syntactic approaches have in common to assume discrete classes. There are no fuzzy boundaries between the lexical categories, because they are categorized in terms of features values (‘+’ or ‘í’) or functional projections (either absent or present). Going back to the issue of mixed-categories, there are no means, thus, to express fuzziness in the existing systems. One solution that has been proposed for the ‘fuzziness’/’mixed-category’ issue is to assume a continuum of categories rather than discrete classes. Ross (1972) argues for such a continuum from verb to noun: verb > present participle > passive participle > adjective > preposition > adjectival noun (e.g., run, snap) > noun. For Ross the distinction between N, V and A is one of degree rather than kind. Featurebased approaches do not offer an alternative to the continuum of categories and mostly fail to account for mixed categories. Syntax-based approaches do offer a way to treat fuzziness/mixed-categories, however, as they allow for (complex) words associated with a particular lexical category to have properties of other lexical categories as well depending on the type of (functional) projection they embed. This kind of approach can easily explain mixed-categories in a language like English, which most generally concern derived words, i.e., words obtained through morphological derivation from other (categorized) words or (lexical) roots. Evidently, in these cases, mixed lexical properties must be the result of mixed structure inside the derived word: derived words retain some properties of their bases and the variation among attested forms is related to how much structure is actually retained. Participles, for instance, can have adjectival properties because they project an A(djective)P or a DegreeP, and therefore are lexically adjectives; and yet retain properties of verbs because they are built upon a vP layer. However, this kind of approach cannot be applied to the Japanese adjectival-verbs, for instance,

Categories and Categorization: Introduction

7

in absence of evidence that they involve complex internal structure or are structurally derived, and there the issue remains. Beyond the major lexical categories N, V and A, there is still a great deal of variability regarding the inventory. Items that don’t pertain to the main classes are either treated as mixed categories (e.g., gerunds, participles, infinitives), or as functional categories. Baker (2003) takes this latter position for prepositions, for instance. As defined by Abney (1987) functional categories are closed classes, they are phonologically and/or morphologically dependent, they take an obligatory complement from which they cannot be separated. Their meaning is grammatical rather than lexical and it must be computed within a full phrase/sentence in order to be obtained (e.g., prepositions). Functional categories can also contribute grammatical information about tense, aspect, determination, negation, and so on. The lexical/functional distinction has also been claimed to be orthogonal to word classes (see, e.g., prepositions, which are lexical and/or functional). Independently of whether they form a word class or not, functional items play a crucial role in the identification of the lexical classes, as words are often classified on the basis of the functional material they can combine with (D with nouns, Tense with verbs, etc.) The papers that constitute the first part of this volume broach two of the major issues introduced above: the problem of mixed-categories and its reflex on the inventory are dealt with by Lundquist and Takamine, while the contribution of functional items in defining the properties of a lexical class is tackled in Janda’s and Takamine’s studies. Lundquist addresses participles in Swedish and the mixed category features of adjectival/verbal derived words from a syntactic perspective. He argues for a structural account where the differences between (socalled) adjectival and verbal participles are related to the internal structure of these expressions, while both are reduced to the same lexical class, that of adjectives. The paper discusses valency/voice changing and category changing operations with verbal and adjectival passives, and so-called middles, including anti-causatives. Lundquist argues that verbal/eventdenoting passives crucially differ from adjectival passives and middles/anti-causatives in that they do not alter the event description. Adjectival passives, on the other hand, always omit event-related projections. He analyzes periphrastic passives in Swedish (the so-called bli-passives) as also involving omission of the highest event-related projection in the VP. He further argues that the English get-passive is identical to the Swedish periphrastic passive, at least in some uses. On the other hand, he takes middles/anti-causatives to involve re-coding of the referential indices within the VP, making them qualitatively different from

8

Chapter One

both verbal and adjectival passives (including English get-passives). He argues that “verbal” and “adjectival” participles are both of the category Adjective. Their distributional differences are derived from restrictions on degree modifiers, making eventive participles pattern with non-gradable adjectives, and adjectival participles with gradable ones. Takamine investigates the contribution of functional layers to define lexical categories. The paper is concerned with the projection D(eterminer) in Japanese and the status of demonstratives as adjectives or Ds. Since Abney’s (1987) DP theory, Japanese nominals have been argued to project a DP layer, despite the apparent lack of “articles” in the language. More recently, several decompositional analyses of Japanese nominals have been introduced. In particular, Watanabe (2006) argues that the Japanese nominal phrase projects at least five different extended projections: NP, NumberP, CaseP, QP and DP, each of which is responsible for introducing morphological and semantic pieces of the nominal system such as numeral classifiers, case markers, and quantificational elements. However, the recent theories of Japanese nominal system have not discussed structural properties of demonstratives. Takamine’s paper explores the structural position of demonstratives in Japanese in the decompositional nominal system. Focusing on the demonstrative so, Takamine distinguishes the use of so that involves no definiteness (‘non-definite so’) from the definite so according to their different interpretational properties in NP-ellipsis. She, then, reports that the two uses of so exhibit different behavior in terms of extraction from the noun phrase, i.e., the definite so, and not the nondefinite so, induces island effects, a fact that strongly indicates that the two uses of the demonstrative so must be assigned different structures. Adopting Watanabe’s layered DP structure, Takamine proposes that the demonstrative so is generated in the intermediate functional phrase CaseP which is the locus of reference, yielding the non-definite use, and they may further rise to DP for checking definiteness features, yielding the definite uses. In her analysis, demonstratives in Japanese are argued to be functional categories and the different interpretations and syntactic properties observed between the two uses of the demonstrative so are attributed to the feature properties of the functional heads demonstratives are associated with and the checking mechanism. Janda provides new grounds for defining the V category in addressing verbs and verb prefixes in Russian. She investigates the status of such prefixes and broaches on the functional/lexical issue. She puts forth a new hypothesis that Russian verbal prefixes are a verb classifier system similar to those found in Australian and East Asian languages (McGregor 2002). Sixteen Russian prefixes have a “purely aspectual” use where they do not

Categories and Categorization: Introduction

9

change the meaning of the verb, as in s-varit’, which means ‘cook’ and is merely the perfective partner verb of varit’ ‘cook’. She argues that the “purely aspectual prefixes” constitute a system of aspectual classifiers akin to numeral classifiers. She presents tests for this hypothesis that include comparison of distributional data with definitions for classifier systems, plus five statistical studies proving that the behavior of each prefix is unique and explainable by recourse to its meaning. Recognizing Russian as a verb classifier language brings numerous advantages, facilitating cross-linguistic comparisons and improving both description and theoretical understanding of classifier systems.

References Abney, S. 1987. The English noun phrase in its sentential aspect. Doctoral dissertation, MIT. Arad, M. 2003. Locality Constraints on the Interpretation of Roots: The Case of Hebrew Denominal Verbs. Natural Language and Linguistic Theory 21.4: 737–78. Acquaviva, P. 2008. Lexical Plurals: A Morphosemantic Approach. Oxford: Oxford University Press. Backhouse, A. 2004. Inflected and uninflected adjectives in Japanese. In Adjective Classes: A Cross-Linguistic Typology, ed. by R. Dixon, and A. Aikhenvald, 50-73. Oxford University Press. Baker, M. 2003. Lexical Categories: Verbs, Nouns and Adjectives. Cambridge Studies in Linguistics. Cambridge University Press. Borer, H. 2005. Structuring sense. Vol 1: In Name Only. Oxford University Press. —. 1999. Endo-Skeletal vs. Exo-Skeletal Explanations in Linguistics. Ms., USC. Chomsky, N. 1970. Remarks on Nominalizations. In Readings in English Transformational Grammar, ed. by R. Jacobs, and P. Rosenbaum. Ginn, Waltham, MA. Déchaine, R. 1993. Predicates Across Categories. Doctoral dissertation, University of Massachusets, Amherst, Massachusetts. Harley, H. and R. Noyer. 2000. Formal versus Encyclopedic Properties of Vocabulary: Evidence from Nominalizations. In The Lexicon/ Encyclopedia Interface, ed. by B. Peeters, 349-374. Amsterdam: Elsevier Press. Jackendoff, R. 1977. X-bar Syntax: a study of phrase structure. MIT Press, Cambridge, MA. McGregor, W. 2002. Verb Classification in Australian Languages. (=

10

Chapter One

Empirical Approaches to Language Typology 25). Berlin/New York: Mouton de Gruyter. Marantz, A. 1997. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In University of Pennsylvania Working Papers in Linguistics Vol. 4.2, ed. by A. Dimitriadis, L. Siegel, C. Surek-Clark, and A. Williams, 201-225. Philadelphia. Panagiotidis, P. 2011. Categorial features and categorizers. The Linguistic Review 28: 365-386. Ross, J. R. 1972. The category squish: Endstation Hauptwort. In Papers of the eighth regional meeting of the Chicago Linguistics Society, ed. by P. M. Peranteau, J. N. Levi, G. C. Phares, et al., 316-328. Chicago Linguistic Society 8. Sleeman, P. 2011.Verbal and adjectival participles: internal structure and position. Lingua 121: 1569-1587. Watanabe, A. 2006. Functional Projections of Nominals in Japanese: Syntax of Classifiers. Natural Language and Linguistic Theory: 241306.

CHAPTER TWO THE CATEGORY OF PARTICIPLES BJÖRN LUNDQUIST

UNIVERSITY OF TROMSØ, CASTL/NORDIACORP1

1. Introduction In an influential paper, Wasow (1977) argued that some passive participles are derived in the lexicon while others are derived in the syntax. One of Wasow's main motivations for a syntax-lexicon split was that some participles behave just like adjectives, so called adjectival participles, while others had at least some verbal properties, so called verbal participles. Wasow assumed that category changes could only take place in the lexicon, and since participles are formed from verbs, adjectival participles must be derived in the lexicon (at least if we take them to literally be adjectives). Verbal participles on the other hand, were assumed to be true verbs, and could thus be assumed to be formed in the syntax. In non-lexicalist frameworks, like DM and Nanosyntax, several attempts have been made to give a syntactic account of both adjectival and verbal participles. In addition, a more fine-grained typology of participles has been argued for, see e.g. Kratzer (2000), Embick (2004), Taraldsen and Medova (2006), Lundquist (2008). All these accounts agree that adjectival and verbal participles differ in terms of syntactic size of the constituent that the participial ending attaches to (or spells out), in ways similar to Abney (1987). Furthermore, in these accounts, the typical semantics of adjectival participles (e.g., the stativity) is not provided by the participial morphology, but rather originates either within the (verbal) root itself (e.g. a Davidsonian state-argument, Kratzer 2000, or a result/state projection inside a decomposed VP) or some aspectual material, either attaching inside or outside the participial phrase. These accounts neatly capture the differences in the internal syntax of adjectival and verbal participle phrases, but they fail to account for the differences in (external) distribution between the different types of participles. For

12

Chapter Two

example, Wasow's observation that only adjectival participles can appear in the complement of a raising verb like seem, in the complement of remain and as a prenominal modifier is hard to explain if we assume that different types of passive participle only differ in the attachment site of the participial morpheme: both types of participle are after all headed by the same morpheme, and we expect external distribution to be determined by the head and not the internal structure of the participle phrase. In this article, I will take this problem as a starting point and discuss the relation between lexical categories and "derived" categories, like participles and nominalizations. The central question is how the internal structure of a constituent affects its distribution. I will argue that all participles have the distribution of adjectives, though the presence of event-structure in some participles makes them illicit in certain typical adjectival contexts where either certain scalar properties or stativity is required. I will further argue that the theory of lexical categories argued for by Baker (2003) gives us a good tool to deal with participles of different sizes. In the last two sections of the paper, I will compare the behavior of the different types of participle to the different types of nominalization, and I will argue that just as nominalizations that contain event structure are still "nouns", participles with event structure are still "adjectives". On analogy with the terminology used for different types of nominalizations, I will propose that the terms event structure participles and result/stative participles should be used instead of the theoretically more loaded terms verbal and adjectival participles.2 I will however use adjectival and verbal participle in this article in the discussion of previous literature on the topic.

2. Different takes on participles Participles are traditionally defined as adjectives derived from verbs. The following definition is from Crystal (1991): (1)

Participle: "a word derived from a verb and used as an adjective"

Participles can however differ in how much verbal structure they contain. For passive participles, a distinction has been made between verbal passive participles and adjectival passive participles (e.g. Wasow 1977) (see section 5 for discussion of active past participles). A verbal passive participle differs from an active verbal clause only in the syntactic realization of the arguments of the verb, and has the same event structure and argument structure as an active verb. Both the active (2a) and the

The Category of Participles

13

passive (2b) verb phrase could thus describe the same event. They are in other words semantically equivalent. (2)

a. John broke the window yesterday. b. The window was broken by John yesterday.

Both sentences above refer to a breaking-event taking place yesterday, of which John is the agent and the window is the theme. In so-called adjectival passives on the other hand, the argument structure and the event structure are somehow reduced, or possibly absent, as illustrated in (3): (3)

The window was still broken (*by John) yesterday.

The adverb still forces a stative reading of the participle, and the participle can thus no longer refer to a breaking event. Once the event-component of the predicate is removed, an agent adverbial can no longer be licensed. The verbal participle can be said to have event structure, equal to that of the active verb, while the adjectival participle refers to a state or a property, just like an adjective.3 The central question of this article is whether the internal structure of the participle phrases has any relevance for the category issue. The definition of participle in (1) says basically that participles are adjectives, or at least that they have the same function as adjectives. The internal structure of participle phrases, at least event denoting participle phrases, is presumably quite different from the internal structure of non-derived adjectives, but that should be irrelevant for the category issue: just as most linguists acknowledge that there are event-denoting nouns with verbal substructure, it should be straightforward to acknowledge that there are event-denoting adjectives with verbal substructure. The fact that event structure participles contain more verbal substructure than stative/resultative participles should thus not lead us to conclude that they differ categorically. If we choose to define categories from the morphological properties of the words, we find that participles, both verbal and adjectival, have the typical morpho-syntactic properties of adjectives. For example, participles inflect for number and gender (and possibly case) in languages where adjectives inflect for these categories, but not for person, unlike verbs, as illustrated in (4):4

14

(4)

Chapter Two

a. El hombre es querido por Def.M.Sg. man is loved.M.Sg. by alto. tall.M.Sg. "The man is loved by his parents/tall." b. Las chicas son queridas por Def.F.Pl. girls are loved.F.Pl. by altas. tall.f.pl. "The girls are loved by their parents/tall."

sus padres/ his parents/

sus padres/ their parents/

The examples above compare verbal participles to adjectives, but the same of course holds of adjectival participles. In other words, there is no correlation between the presence of event-structure and adjectival inflection.5 Further, as mentioned before, participle phrases have the same core distribution as adjectival phrases: they can appear in the complement of a copula, and they can appear as adnominal modifiers (more on this in section 2.1). However, there are some substantial differences in distribution between verbal and adjectival participles: adjectival participles can appear in the complement of the raising verb seem, just like adjectives but unlike verbal participles. As shown below, a participle in the complement of seem cannot license an agentive by-phrase: (5)

a. John seems happy. b. The window seems broken (*by John). c. The window seems to be broken by John.

If we were to take this to be a conclusive argument for treating adjectival participles as adjectives and verbal participles as verbs, we would have to give up the definition in (1). I will however argue that passive participles are always (derived) adjectives. Following Matushansky (2002), I will argue that seem can only take gradable complements, and that eventstructure participles crucially are not gradable. Wasow (1977) gives two more distributional differences between adjectival and verbal participles: adjectival participles can appear as prenominal modifiers, and they can appear in the complement of remain. Below I will look more closely at the differences in distribution between verbal and adjectival participles, and show that these differences do not impose a categorial split among participles. Rather, all participles are externally adjectives (just like all nominalizations externally are nouns).6

The Category of Participles

15

2.1 Wasow (1977) and the distribution of participles According to Wasow, one of the main reasons to assume that adjectival participles are adjectives is that adjectival participles have the distribution of adjectives. Most notably, they can appear as ad-nominal modifiers, and they can appear in the complement of a number of raising verbs: (6)

a. the broken cup b. The cup seems broken.

However, the restriction on attributive participles is much less strict than the distribution of participles following e.g. seem, as shown in (7): (7)

a. the recently made headway—all that headway was/??seems made in a day. b. the most recently taken photos—these photos were/??seem taken recently. c. the kicked out guests—they were/??seem/??seemed kicked out.

As shown in (7a), even idiom chunks can appear in prenominal participle phrases, which we can take as evidence that the participle has a phrasal source. and can thus not have been derived in the lexicon (see Kratzer 2000 for discussion). It is not obvious that there is any restriction at all on so-called "verbal" participles to be used as prenominal attributes. It is, however, clear that participles with agentive by-phrases are illicit as prenominal attributes (8a); but on the other hand, even regular adjectives with PP modifiers/arguments are illicit prenominally (8b): (8)

a. the (*by John) broken window/the broken (*by John) window b. the (*of John) jealous man/the jealous (*of John) man

Rather, adjectival and participial phrases with PP-modifiers/arguments need to surface postnominally, as in (9): (9)

a. the window broken by John b. the man jealous of John

The fact that participles with agentive by-phrases cannot surface prenominally thus does not tell us anything about the category of the participle phrase. It just tells us that pre-nominal attributes cannot take PP modifiers/arguments.7 We also know that in languages where PP-modifiers

16

Chapter Two

of attributes are licit, agentive by-phrases are also licit in this context, as in the following example from Rapp (2000): (10)

der vom Kellner eingeschenkte Wein the by waiter served wine "the wine served by the waiter"

In other words, there is no reason to assume that only "adjectival" participles, i.e. participles that are event/argument structurally reduced, can be used as pre-nominal attributes, at least not in languages like English and German (though it could of course be the case that some languages only allow adjectives with certain scalar properties in pre-nominal position).8 The restriction on event-structure participles in the complement of seem cannot however be deduced from the syntactic shape of the participle phrase (i.e. from the presence of PP-arguments/modifiers), since PP's are licit in the complement of adjectival complements of seem: (11)

a. He seems fond of the situation. b. He seems very interested in this type of problems.

However, as argued by Matushansky (2002), seem selects for gradable adjectives (or an IP/CP). As shown in (12), non-gradable adjectives are not licit in the complement of seem: (12)

a. This music seems nice/*choral. b. This problem seems insoluble/*mathematical (from Matushansky 2002)

However, most non-gradable adjectives can undergo "scalarity coercion", and it is thus hard to find adjectives that are strictly ungrammatical under seem, as illustrated in (13) (from Matushansky 2002): (13)

a. This music seems almost choral. b. This problem seems pretty much mathematical.

The same is true for participles too: it is possible for most participles to undergo scalarity coercion. However, when they do, they lose their argument/event structure, as shown in (14 b), where the agent phrase is no longer available:9

The Category of Participles

(14)

17

a. That book was/*seems written (by Hamsun). b. This book seems very well-written (*by Hamsun).

The gradability sensitivity is also very clearly seen with present participles. Both gradable and non-gradable participles are licit in a prenominal position, and in the complement of be, while only gradable present participles are licit under seem: (15)

a. the (very) fascinating/moving/boring movie b. the (*very) running, laughing, dancing man

(16)

a. This movie seems (very) fascinating/moving/boring. b. *John seems (very) running/laughing/dancing.

As will be returned to below, it is not obvious what it means for event structure participles, or even active verbs, not to be gradable. In fact, they can be, it is just that an adverb has to be added to introduce some kind of scale. Sometimes, an adverb like much is sufficient, but other times a more semantically rich adverb like e.g. badly or poorly is needed (see Kennedy and Levin 2002 for more discussion on this issue): (17)

a. They injured him so much/so badly that he could hardly walk. b. He broke the stereo set so badly/*so much that it could not be fixed again.

The adverb is still required in event-structure (ES) passives, but only optional in stative passives: (18)

a. He was injured so much/so badly by the gangsters that he could hardly walk afterwards. (ES passive) b. He was so injured (*by the Gangsters) that he could hardly walk. (stative passive)

(19)

a. The stereo set was broken so badly/*so much by John that it could not be fixed again. (ES passive) b. The stereo set was so broken (*by John) that it could not be fixed. (stative passive)

I will not be able to answer here exactly what the adverb provides: it could either be a new scale altogether, or just a specific value of a scale already present in the verb. Either way, degree modifiers like so and very cannot

18

Chapter Two

directly access a scale provided by an event denoting head like e.g. "v" (see below on "v"). Whatever the reason is that makes so and very unable to access a scale provided by an event denoting head, we can assume that seem is unable to do so for the very same reason. Remain on the other hand seems to take only stative complements, irrespective of their gradability. As shown in (20), the ambiguous participle broken cannot take a by-phrase when appearing in the complement of remain, indicating that only a stative/adjectival participle is licensed under remain. However, as shown in (21), a participle formed from a stative verb, with maintained event structure, can surface under remain, though not seem (it is at least highly marked), indicating that stativity really is the crucial feature involved: (20)

The window remained broken (*by John) for many days.

(21) a. London Lite, like its free sister morning newspaper, Metro, remained owned by Associated Newspapers, the same media group that owns the Daily Mail. b. ??London Lite and Metro seem owned by the same company. It is possible that other adjectival positions are sensitive to the state–event distinction as well, for example secondary predicates (see Embick 2004 for discussion). To summarize, above I have shown with the help of morphological and distributional diagnostics that so called "verbal" and "adjectival" participles should not be treated as two different categories. Rather, both have the morphological and distributional characteristics of adjectives. The most obvious adjectival characteristic of "verbal" participles is their need of a copula to express tense etc. in regular passive clauses (see e.g. 2b above). Even though "verbal" participles have a slightly more restricted distribution compared to prototypical adjectives, there are no positions where participles but not adjectives can appear. I have suggested that participles with event-structure, i.e. participles that contain some eventdenoting projection, are illicit in the complement of seem due to the fact that they lack the relevant scalar properties that seem selects for. However, scalar structure is not a definitional characteristic of adjectives, since there are non-gradable adjectives. Furthermore, remain can only take a stative complement, which explains the ungrammaticality of event-structure participles formed form non-stative verbs in the complement of remain. However, being stative is not a sufficient criterion for being an adjective, since verbs (and nouns) can be stative as well. In general, there is no

The Category of Participles

19

reason to assume that passive participles formed from stative verbs are more adjectival than passive participles formed from non-stative verbs, just as we can't assume that stative verbs are more adjectival than nonstative verbs. There is presumably no difference in the relation between the active and the passive members in (22a) compared to (22b): in both cases, a verb has been turned into a participle, but neither the event structure nor the argument structure has been changed: (22)

a. John broke the stick - the stick was broken by John. b. John owned the company - the company was owned by John.

In short, both stative/adjectival and verbal/event structure participles have the distribution of adjectives. They differ in their internal structure, but there is no reason to assume that e.g. adjectival participles have an additional "adjectival" projection that is absent in verbal participles (see e.g. Lieber 1980 for an analysis in that direction).10

3. Adjective as a default category Classifying participles as adjectives is quite pointless unless we have a theory about (lexical) categories. Focusing on adjectives and verbs, we have seen above that adjectives can be accessed by certain degree modifiers, while verbs cannot. However, not all adjectives are gradable, which at least suggests that a word can be of the category adjective, without having the relevant scalar properties. That is, being gradable is not a necessary condition for being an adjective. Verbs tend to denote events, in contrast to adjectives, which tend to denote properties or states. However, not all verbs denote events, and the difference between a verbal predication (23a) and an adjectival predication (23b) can often not be stated in terms of eventivity (or stativity) (see Baker 2003, for discussion): (23)

a. The square root of four equals two. b. The square root of four is even.

Baker (2003) argues that there is a structural, rather than semantic, difference between adjectives and verbs. He gives the following definitions of the three lexical categories:11 (24)

a. Noun: "has a referential index" b. Verb : "has a specifier" c. Adjective: "has neither referential index, nor specifier"

Chapter Two

20

I will return to nouns in section 4. Adjective is for Baker just a default category. According to Baker, a verb always has a specifier where a subject (or external argument) can be introduced. Adjectives (and nouns) require an additional functional projection to introduce a subject, which Baker labels Pred(ication), following Bowers (1993). For Baker, Pred and V are different in that V is a lexical category, while Pred is a functional category.12 However, an adjective can undergo head movement to a Pred position, thereby filling Pred with lexical material, which changes the label Pred to V. Baker suggests that the adjectival predicates in (25-a) and the verbal predicates in (25b) have the same underlying structure, and differ only in the timing of the vocabulary insertion: (25)

a. Fred is hungry/ Fred is fond of spinach. b. Fred hungers /Fred likes spinach.

In the adjectival cases, vocabulary insertion takes place before the merging of Pred (derivations below from Baker 2003, p. 87): (26)

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

A [AP A (PP)] Merge [AP hungry/fond (NP)] Vocabulary insertion [Pred [AP hungry/fond (NP)]] Merge [PredP NP Pred [AP hungry/fond (NP)]]] Merge [PredP NP Ø [AP hungry/fond (NP)]]] Vocab. Insert [NPi bej + Tense [AuxP ti tj [PredP ti ; [AP hungry/fond (NP)]]]]]

In the verbal cases, vocabulary insertion takes place after the merging of Pred. The adjectival stem moves to the Pred head, which turns Pred into a normal V: (27)

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

A [AP A (NP)] Merge Pred [AP A (NP)] Merge Ai + Pred [AP ti (NP) ] Move like/hunger [AP ti (NP) ] Vocab. insertion [VP NP like/hunger [AP ti (NP) ]] Merge [NPj Tense [VP tj like/hunger [AP ti (NP) ]]]

The difference between a participle and a full verb can presumably be described in the same way. A participle could just lexicalize a verbal

The Category of Participles

21

structure where no Pred is present yet, especially if we take a Pred associated with a verbal root to introduce an external argument. For adjectival participles, a derivation similar to that in (26) is presumably more or less straightforwardly applicable. If we are to apply the derivation of (26) to event-structure participles as well, we need to show that they do not contain external arguments (in the form of PRO or pro). There is a big debate about whether eventive verbal participles contain an external argument or not, in the form of a pro or PRO. Below I give three arguments against the presence of an external argument in regular passives, which makes it possible to apply a derivation like the one in (26) to event structure participles as well, though with a event-denoting, specifierless, projection added before the merging of Pred.13 1. Anaphoric binding: In contrast to an overt subject (28-a) or a PRO subject (28-b), the implicit external argument of a passive participle cannot bind an anaphor (28-c). Examples below are from Swedish, since the contrast is most clearly seen with possessive anaphors (and English lacks a special set of possessive anaphors): (28)

a. Han åt upp hela tårtan på sin födelsedag. he ate up whole cake on Refl.Poss birthday "He ate the whole cake on his birthday." b. Att äta tårta på sin födelsedag är högst normalt to eat cake on Refl.Poss birthday is high.Sup normal "Eating a cake on one's (own) birthday is highly normal." c. *Hela tårtan blev uppäten/ åts upp på sin whole cake.Def was up.eaten/ate.Pass up on Refl.Poss födelsedag. birthday int. "The whole cake was eaten up on his birthday." (i.e., he ate the whole cake on his birthday)

2. Principle B/C violations: Certain types of referential expressions that occur in the complement of a passive participle can be interpreted as coreferent with the implicit external argument (29a). This is impossible when the subject is overt in a finite clause (29b) or PRO in a control infinitive (29c): (29)

a. Van Goghi usually painted out in the fields, but this painting was painted in the artist'si own garden.

22

Chapter Two

b. *Van Gogh usually painted out in the fields, but hei painted this painting in the artist'si own garden. c. *To PROi paint a painting in the artist'si own garden didn't seem like a good idea (to Van Gogh). If an external argument were syntactically present in the passive in (29a), a Principle B (or possibly Principle C) violation would have been expected. 3. Control of adjectival agreement: The implicit external argument cannot control number and gender agreement on a depictive adjective, as shown in (30a) (and we assume that depictive adjectives require agreement with some argument, and don't allow "default" agreement). However, a depictive predicate in the shape of a PP (which shows no agreement) is licit. A PRO-subject easily can license agreement on depictive adjectives, making (30-c) grammatical:14 (30)

a. *Middagen åts alltid naken /naket /nakna dinner.Def ate.Pass always nude.CG /nude.Nt /nude.Pl under sommaren. under summer.Def "The dinner was always eaten nude during the summer" b. ?Middagen åts alltid utan kläder under dinner.Def ate.Pass always without clothes under sommaren. summer.Def "The dinner was always eaten nude during the summer" c. Att äta middag naken sågs som helt normalt. to eat dinner nude.CG.Sg see.Pst.Pass as fully normal "Eating dinner nude was regarded as completely normal."

Had there been a syntactically present external argument in the eventdenoting passive, we would expect this argument to be able to trigger agreement on the depictive adjective, just as PRO can trigger agreement. There is however no doubt that there is some event-denoting category contained in eventive passives, for example a PROC projection in the terms of Ramchand (2008), or a little v of a certain flavor, as in various DM accounts (see e.g. Embick 2004 and Harley 2005). We can see this in the fact that a depictive PP modifying the external argument is licit in eventive passives, as well as purpose clauses. What is important though, is that there are no signs of the presence of a true external argument, i.e. no Voice or Pred (or whatever you take to be the relevant projection that

The Category of Participles

23

introduces the external argument). We can thus assume that an eventstructure participle has the structure of (31):15 (31)

-ed [v [Root ]]

After v has been merged, there could in principle be at least three options for the next merge: either a VoiceP/PredP (or simply a specifier) is merged, and you get a verb, or a referential index is added to v, resulting in an event-structure nominalization (see more on this below), or participial morphology is added, resulting in a specifier-less structure, i.e. an adjective. The question is why you need to merge participial morphology, given that the specifier-less vP already is structurally an adjective (at least according to Baker). There is no obvious semantic or syntactic function of the participial morphology: tense and aspect are in general not expressed by the participial ending (at least not in the languages discussed in this article), but rather by copulas and auxiliaries, and the passive voice of the participle is presumably not encoded in the participial ending either: given that the specifier hosting the external argument is missing, the structure that the participial ending attaches to is already "passive". It is probably better to think of the participial ending as a host for adjectival inflection, or as a slot for phi-features, see Lundquist 2008 for an analysis of the participial ending as uninterpretable Gender features (see also Kratzer's 2000 similar idea that participial morphology is needed to license the absence of verbal morphology). Note that I have used the label "v" in the structure in (31), which is of course contradictory if we believe that a projection only can carry the label v if it has a specifier. The label "v" should preferably be replaced by a more semantically transparent label, like "process", "event" or even "state". For the purpose of this article, I will simply assume that "v" (or whatever its correct name is) introduces an event variable, but not an external argument, which rather is provided by Voice (as in Kratzer 1996) or Pred. Once a lexical item spells out both "v" and Pred, it is presumably correct to label it verb.

4. Parallels to the nominal domain The problem with the restrictive distribution of event-structure participles is parallel to that of the distribution of event-structure nominals (see Grimshaw 1990). Just as with participles, linguists have claimed that event-structure nominals and result (or simple event) nominals differ in the attachment site of a nominalizing morpheme (see e.g. Abney 1987, Kratzer

24

Chapter Two

1996, Alexiadou 2001 and Lundquist 2011). There are good reasons to assume that event structure nouns of the type exemplified in (32a) lack a "specifier", or a subject (introduced by Voice or Pred). 16 I will simply assume that a nominalizer provides an interpretable Gender feature to some structure built in the syntax, or possibly a root, that lacks a Gender feature in its lexical representation. In line with Lundquist (2008) I assume that (interpretable) Gender features are associated with a referential index, i.e. every word with a realized interpretable Gender feature has a referential index (i.e., is a noun, following Baker's definition). Baker (2003) suggests that whereas (32a) is taken to contain some event denoting category, let's call it v, (32b) does not (examples based on examples in Grimshaw 1990). (32)

a. the frequent assignment of easy problems b. [The assignment] lay on the table.

As noted by Grimshaw (1990), one of the characteristics of event structure nouns is that they cannot carry plural marking, or appear in the complement of an indefinite article (33) (examples from Grimshaw 1990), in contrast to result/simple event nouns (34): (33)

a. *the frequent examinations of the students (cf. the frequent examination of the students) b. *The shootings of rabbits are illegal. (c.f. The shooting of rabbits is illegal) c. *A/one shooting of rabbits is illegal.

(34)

a. An assignment lay on the table. b. Many/the assignments lay on the table.

Just like with participles, it thus seems like the internal structure of the nominal determines the external distribution. Event-structure nouns seem to have the distribution of mass nouns, while result/simple event nouns have the distribution of count nouns (see Harley 2009 for a description of the facts in this way). 17 If we assume, following Borer (2005), that a number projection requires a classifier phrase (Cl) in its complement (which functions to individuate a mass noun), we have to conclude that event-structure nouns are incompatible with Cl, in contrast to result/simple event nouns: (35)

a. [ Cl [-mentn [ Root ]]] (Result/Simple event noun) b. *[ Cl [-mentn [ vP ]]] (Event structure noun)

The Category of Participles

25

A classifier projection is thus incompatible with a noun containing event structure just like a degree (or a scale) projection is incompatible with an adjective with event structure: (36)

a. [Degree/Scale [-eda [ Root ]]] (Stative participle) b. *[Degree/Scale [-eda [ vP ]]] (Event-structure participle)

Just as an event-structure noun is still a noun, an event structure participle is still an adjective. Being available for number modification is clearly not a defining characteristic of nouns, since there are mass nouns (like e.g. furniture), just as being available for degree modification is not a defining characteristic of adjectives, since there are non-gradable adjectives (like e.g. chemical).18 The big question that remains to be answered is why nouns and participles with event structure are incompatible with a classifier/scale projection. Verbs can clearly have some scalar structure, as discussed above (and see e.g. Hay et al. 1999), and events can presumably be individuated, for example with the help of an aspect node. One possibility is that verbs (with event structure) already have their own scale and individuation values set, and that these values are incompatible with the additional classifier and scale projections, and not directly accessible by a nominal number projection or certain degree modifiers (possibly due to the embedding under a n/a head). Cross-linguistically, event-structure nouns can sometimes carry number marking, and the availability of number marking is often correlated with certain types of aspect marking, see Alexiadou et al. 2010 for a cross-linguistic overview. It might turn out that some languages have event structure participles that are as gradable as regular adjectives, at least for certain aspectual values, i.e. we may find event-structure participles with the same distribution as regular gradable adjectives.

5. Concluding remarks and remaining issues The restriction on e.g. plural marking on nominalizations has been used as a test for separating event structure nominalizations from result/simple event nominalizations, but little effort has been put into explaining the restrictions (but see e.g. Harley 2009 and Alexiadou et al. 2010 for exceptions). The restrictions on the distribution on event-structure (or verbal) participles have also been used as a diagnostics for separating event structure participles from stative/adjectival participles. For eventstructure/verbal participles, the restrictions have just been assumed to

26

Chapter Two

follow from their category: verbal participles are simply verbs and are thus not expected to appear in typical adjectival positions, as e.g. the complement of seem and remain, and in attributive positions. I have argued above that even event-structure participles have an adjectival distribution; it is only that their internal structure makes them illicit in contexts where only either stative or gradable elements are licensed. In short, event structure participles are adjectives to the same extent that event structure nominalizations are nouns. The important fact is that the presence of an event variable (provided by "v") inside a participle makes the participle unavailable for direct degree modification, just like the presence of an event variable inside a nominal makes the nominal incompatible with a classifier. We now thus have at least the beginning of an answer to the initial question: why is it that the internal structure of a word, rather than the head of the word, determines its distribution. In the story given above, participle morphemes and nominalizing morphemes have very simple functions: a nominalizing suffix adds a referential index, and a participle ending cuts off a functional sequence before a specifier has been added. If they attach to something containing a v-node, the distribution of the derived adjective/nominal will be restricted, due to further limits of degree and number modification. There are at least two concerns that need to be mentioned in this paper. The first one is why there are no lexical event-structure adjectives, i.e., why are there no non-derived adjectives that for example take an agent and a patient argument, like an event structure participle? Part of the answer presumably lies in the fact that every item with an event variable can be used as a verb. In that sense, every possible event-structure adjective would have a corresponding verb, i.e. there would be both a verbal item and an adjectival item connected to the same (complex) concept. In some sense, this is exactly what we see: structures containing event structure can be realized as verbs (e.g. as an infinitive) or as adjectives (i.e., as a participle). It happens to be the case that most event-structure adjectives are formed with the same ending (-ed), but of course there are (semi)irregular (or strong) or zero-derived participles as well (i.e. written or hit, and see also Embick (2004) for a list of participles that have special target state participles). The situation is similar for nominalizations, i.e. eventstructure nominals are derived from verbs, most often with the fully productive affix -ing, but sometimes with a zero nominalizer or a less productive morpheme. The second remaining big issue is the distribution of participles and nominalizations with even more internal verbal structure, like active past participles and POSS-ing, as exemplified below:

The Category of Participles

(37)

27

a. John has given Mark a ball. b. John's/*the giving Mark a ball

In these cases it is probably safe to conclude that the participial/ nominalizing ending attaches in a higher functional domain (as proposed in Abney 1987 and Baker 2005 for gerunds), and that both the participle and the nominalization contain a specifier, where an external argument is introduced. These participles thus contain verbal structure, i.e., a specifier in the sense of Baker (2003) (and see Baker 2005 for a treatment of "dual category" words). 19 Let us for now assume that the nominalizer (i.e., a referential index) and the participial marker attach in a Tense-node, before a specifier has been merged to Tense, as in (38): (38)

-s/have [-ing/-ed [T [vP [ (Ext.Arg) ] v [Root ]]]]

I will assume that the specifier of Tense is the position where a subject can be case licensed. What is interesting is the fact that an even smaller set of determiners/number markers is available in POSS-ing gerunds compared to the event structure nouns discussed above, i.e. not even a definite article (the) can select for a nominal -ing merged in T, and the active past participles don't really have an adjectival distribution at all. However, if (38) is the right structure for an active participle (and a POSS -ing gerund), the active participle is still an adjective, at least in Baker's definition, i.e. it is a word without an external specifier and without a referential index. It is however a unique type of adjective, just as the POSS-ing gerund is a unique type of nominal, in that it contains an external argument. The only way to license these structures seems to be to merge an element that can case-license the external argument, either a possessive -s in the nominal domain, or a possessive verb in the clausal domain.20 In other words, it seems to be possible to treat -ed in active participles and -ing in POSS-ing gerunds as regular adjectivalizers/nominalizers, with their distribution falling out from a more general condition on case-licensing external arguments.

References Abney, S. 1987. The English Noun Phrase in its Sentential Aspect. Ph.D. thesis, MIT, Cambridge, MA. Alexiadou, A. 2001. Functional Structure in Nominals: Nominalization and Ergativity. Amsterdam: John Benjamins.

28

Chapter Two

Alexiadou, A., G. Iordăchioaia, and E. Soare. 2010. Number/Aspect interactions in the syntax of nominalizations, Journal of Linguistics 46: 537-574. Baker, M. C. 2003. Lexical Categories: Verbs, Nouns and Adjectives. Cambridge Studies in Linguistics 102. Cambridge: Cambridge University Press. —. 2005. Gerunds. Ms., Rutgers University. Borer, H., 2005. Structuring Sense: An Exoskeletal Trilogy. Oxford: Oxford University Press. Bowers, J. 1993. The syntax of predication. Linguistic Inquiry 24 4: 591– 656. Bruening, B. to appear. Word Formation is Syntactic: Adjectival Passives in English. To appear in Natural Language and Linguistic Theory. Crystal, D. 1991. A dictionary of linguistics and phonetics. The Language library. Oxford: Blackwell. Embick, D. 2004. Structure of the resultative participles in English. Linguistic Inquiry 35.3: 355–592. Grimshaw, J. 1990. Argument Structure. Linguistic Inquiry Monographs 18. Cambridge: MIT Press. Harley, H. 2005. How do verbs get their names? Denominal verbs, Manner incorporation, and the ontology of verb roots in English. In The Syntax of Aspect, ed. by N. Erteschik-Shir and T. Rapoport, 42–64. Oxford: Oxford University Press. —. 2009. The morphology of nominalizations and the structure of vP. In Quantification, Definiteness and Nominalizations, ed. by A. Giannakidou and M. Rathert, 321–343. Oxford: Oxford University Press. Hay, J., C. Kennedy, and B. Levin. 1999. Scalar structure underlies telicity in degree achievements. In Proceedings of SALT IX, ed. by T. Matthews and D. Strolovitch, 127–144. Ithaca, NY: CLC Publications. Kennedy, C., and B. Levin. 2002. Telicity corresponds to degree of change. Ms., Northwestern University and Stanford University. Kratzer, A. 1996. Severing the external argument from the verb. In Phrase Structure and the Lexicon, ed. by J. Rooryck and L. Zaring, 109–137. Dordrecht: Kluwer. —. 2000. Building statives. In Proceedings of the Twenty-sixth Annual Meeting of the Berkeley Linguistic Society, ed. by L. Conathan, J. Good, D. Kavitskaya, A. Wulf, and A. Yu, 385–399. Berkeley, CA: Berkeley Linguistics Society. Lieber, R. 1980. On the Organization of the Lexicon. Ph.D. thesis, MIT.

The Category of Participles

29

Lundquist, B. 2008. Nominalizations and Participles in Swedish. Ph.D. thesis, University of Tromsø. Lundquist, B. 2011. Restrictions on reflexive and anti-causative readings in nominalizations and participles. In Nordlyd 37: Relating to Reflexives, ed. by T. Strahan. Tromsø: University of Tromsø. Matushansky, O. 2002. Tipping the scales: the syntax of scalarity in the complement of seem. Syntax 5.3: 219–276. Ramchand, G. 2008. Verb Meaning and the Lexicon: A First Phase Syntax. Cambridge: Cambridge University Press. Rapp, I. 2000. The attributive past participle: Stucture and temporal interpretion. In Audiatur Vox Sapientiae. A Festschrift for Arnim von Stechow, ed. by C. Féry and W. Sternefeld, 392–409. Berlin: Academie-Verlag. Taraldsen, T., and L. Medova. 2006. Do toho! Ms., University of Tromsø. Wasow, T. 1977. Transformations and the lexicon. In Formal Syntax, ed. by P. Culicover, T. Wasow, and J. Bresnan, 327–360. New York: Academic Press.

Notes 1

I wish to thank Gianina Iordăchioaia, Gillian Ramchand and two anonymous reviewers for comments and helpful feedback on a previous draft. 2 I will use the term passive to refer to a whole verbal structure, including a copula/auxiliary, that has the same semantic characteristics as a regular active clause (except for the voice-specification). 3 In Kratzer (2000) and Embick (2004) two types of "adjectival" participles are identified: target state/stative and resultant state/resultative participles (see also Parsons 1990 for a more in-depth discussion of the difference between target state and resultant state participles). In this article I will not evaluate these claims, or even give exact structures for different types of adjectival participles. Both Kratzer (2000) and Embick (2004) show that adjectival passives can be phrasal in nature, which makes them difficult to deal with in lexicalist frameworks. For the purpose of this article, it is irrelevant whether participles are formed in the lexicon or the syntax, but given that the arguments for lexically derived adjectival participles are rather weak, we can probably safely assume that participles are syntactically derived (see Bruening to appear). 4 List of abbreviations: CG = Common gender, Def = definite, F = feminine, M = masculine, Nt = neuter, Pass = passive, Poss = possessive, Pst = Past, Refl = reflexive, Pl =plural, Sg = singular. 5 I only know of one language where verbal participles fail to show typical adjectival inflection in contexts where adjectives and adjectival participles do, and that is Danish, where predicative stative/adjectival participles optionally show

Chapter Two

30

number and gender agreement, just like regular adjectives, and eventive passive participles never show agreement (examples from Sten Vikner, p.c.): (i)

a. Dørene blev lukket i går av John. door.Pl.Def. were closed yesterday by John "The doors were closed yesterday by John" b. Dørene er fortsat lukkede/lukket. door.Pl.Def. are still closed.Pl/closed.N "The doors are still closed."

I have no explanation for the Danish pattern to give here. 6 This point was made by Abney (1987) as well, for whom the participial marker always was of the category A. 7 In English this might follow from two independent restrictions: the ban on headinitial prenominal modifiers, and the ban on a PP preceding the head/phrase it modifies. 8 Agent-oriented adverbs are licensed in pre-nominal participle phrases, as shown in (ia-b): (i)

a. the intentionally poorly placed signs b. the intentionally delayed payment

However, an adverb like intentionally does not require an agent present in the semantic/syntactic representation, as shown in (ii): (ii) 9

the intentionally bad joke

It is sometimes possible to have by-phrases, even when the participle is scalarity coerced, as in e.g. this book seems very hurriedly written by its author, as if racing to meet a deadline (G. Ramchand p.c.). See also Bruening (to appear) for more examples of "adjectival" passives with by-phrases. Whether these participles really have event structure is less obvious. It seems sometimes to be possible to add byphrases even to clear stative predications, as long as it is possible to tell from the state who could be the agent of an event leading to this state. 10 In Lieber (1980), it is proposed that adjectival participles contain a phonologically null adjective morpheme, attached outside the participle morphology. The adjectival layer is supposed to give adjectival participles their adjectival distribution. 11 I will have little to say about the early Chomskian view of lexical categories, where the lexical categories where built up by the binary features V and N. See Baker (2003) for discussion and criticism of various theories of lexical categories. 12 Barker's arguments here are far from convincing. He claims he needs Pred to capture the fact that predicative nouns and predicative adjectives sometimes can be conjoined, while predicative nouns/adjectives and verbs cannot. Most of his examples are better explained by assuming that what makes it possible to conjoin

The Category of Participles

31

predicative nouns and adjectives is a shared Scale/Grade P, which is absent in verbs, rather than a PredP. 13 In the Swedish examples, I have sometimes used the morphological passive, which is not based on a participle. This is because the morphological passive is more natural in some contexts. Despite the fact that these passives are "verbal", there is no evidence that they contain a syntactically realized external argument. 14 I give only examples with the morphological passive here, since the participial passive is a bit marked in this context in Swedish, with or without a secondary predicate. 15 I will not say anything about the introduction of the internal argument here. It is possible that the internal argument is introduced in a specifier position linked to either the root or a lower verbal projection. As discussed in Baker (2005), a noun (and presumably an adjective as well) can contain a projection with a specifier, as long as it is further embedded in other material without specifiers. Baker argues that this is the case for verbal gerunds, and it is presumably also the case for more structurally rich participial phrases, like active past participles, and some present participles, which will be returned to in the concluding section. 16 Arguments against the presence of subjects in complex event nominals are given in Abney (1987) and Lundquist (2011). 17 Just as with scales and adjectives with event structure, the facts are more complex, see e.g. Alexiadou et al. (2010). The point here is just that more complex internal verbal structure seems to deprive the noun or the adjective respectively of some of its external distributional properties. The question is whether we can explain this without denying that nominalizations are categorially nouns and that participles are categorially adjectives. 18 Harley (2009) argues that result nominals contain verbal structure as well, at least result nominals containing an overt verbalizer ("v"), like -ate or -ize. According to Harley, the relation between the event noun assignment and the result noun assignment is the same as the relation between the mass noun coffee and the count noun (a) coffee. The absence of e.g. event modifiers in result nouns simply arises as an effect of coercing a mass noun into a count interpretation. If this is the right analysis, the same analysis would work for adjectival participles as well. We have however strong evidence that count coercion (in the nominal domain) and scalarity coercion (in the adjectival/participial domain) removes the evententailments of participles and nominalizations, as can be seen in the absence of argument structure and event-modifiers in result nominals and adjectival participles, at least in English. 19 If we take internal arguments to be merged in the specifier of a lower verbal projection, say V, then even structurally smaller participles and nominalizations embed verbal structure. 20 In gerunds, it is also possible to let the subject come out as a PRO. Why this is not possible for active past participles is not clear. It should be mentioned that it is possible in present participial phrases, like having bought the house, John decided to go home and even verbal passive participles like given the chance, I would leave this country.

CHAPTER THREE DP-DECOMPOSITION ANALYSIS OF JAPANESE DEMONSTRATIVE SO KAORI TAKAMINE NORWEGIAN UNIVERSITY OF SCIENCE AND TECHNOLOGY

1. Introduction1/2 The syntactic status of demonstratives is one of the topics that has not been well studied in the literature of Japanese linguistics. In the eighties, Fukui (1986, 1988) claims that demonstratives are an NP-adjoined lexical category AP. More recently, some linguists argue that demonstratives are functional elements in an extended functional projection DP (Furuya 2008). Furuya (2008) argues that demonstratives in Japanese are generated in SpecDP based on the fact that the presence of the demonstrative so yields definiteness effects in extraction out of a noun phrase. However, a closer investigation of the non-deictic demonstrative so in this paper shows that the two uses of the demonstrative so, the bound variable use (cf. Hoji 1991) and the specific indefinite use, do not exhibit definiteness effects.3 Consequently, the two-way split pattern of so with respect to the definiteness effects, i.e. the presence of the island effects reported by Furuya (2008) and the absence of such effects that is demonstrated in this paper, indicates that neither of the two previous analyses may give a full account for the observed definiteness effects. The present paper proposes that the demonstrative so is uniformly base-generated in the intermediate functional projection of DP from where it may move to SpecDP for checking of definiteness related features. In the present analysis, the two– way split pattern of so is attributed to whether so raises to SpecDP or not. Section 2 briefly summarizes the two opposing previous approaches to demonstratives in Japanese: Fukui’s (1986) AP analysis based on the free word order between a demonstrative and an adjective and Furuya’s (2008)

34

Chapter Three

DP base-generated analysis which is supported by definiteness effects in extraction. In section 3, I consider the two uses of the demonstrative so that are not definite, the specific indefinite so and the bound variable so. I will demonstrate that, unlike the definite so, these two uses of so are not associated with DP based on the fact that they do not exhibit definiteness effects with respect to numeral extraction and NP-ellipsis. On the one hand, the presence of the definiteness effects with the definite so implies that so is in SpecDP; on the other hand, the variation found with the other uses of so strongly indicates that so is not uniformly associated with the Dprojection. In section 4, I propose that so is generated in the intermediate functional phrase between DP and NP from where it moves to SpecDP in order to check the definiteness related features. As a structure for nominal phrases in Japanese, I adopt Watanabe’s DP decomposition analysis and propose that demonstratives are generated in SpecCaseP, which is the locus of reference-fixing. In the present analysis, so that does not induce the definiteness effects stays in SpecCaseP while the definiteness inducing so raises to SpecDP. Section 5 presents a summary.

2. Previous approaches Fukui (1986, 1988) argues that demonstratives in Japanese are adjectives based on the observation that demonstratives may be preceded by an adjective. Note that Japanese does not have a definite article and definiteness is marked by the demonstrative so in (1). (1)

a. so-no nagai hanashi4 so-LIN long story “that/the long story” b. nagai so-no hanashi long so-LIN story “that/the long story”

In addition, unlike English, in which demonstratives cannot appear with articles or possessives, Japanese allows the co-occurrence of demonstratives with possessives.

DP-decomposition Analysis of Japanese Demonstrative so

(2)

35

a. John’s (*this) book b. Taro-no so-no hon Taro-LIN so-LIN book lit. “Taro’s that/the book”

Fukui attributes the difference between Japanese and English in (2) to the presence of the D-projection and the status of demonstratives in the two languages. According to Fukui (1986), demonstratives in English are a D element and hence compete for the D position with the possessive element that is also a D element. On the other hand, in Japanese, both demonstratives and possessives are APs, and hence both can be (freely) adjoined to the NP (cf. Boškoviü 2005 for the non-DP analysis of nondeterminer languages). (3)

a. Taro-no so-no Taro-LIN so-LIN lit. “Taro’s that/the b. so-no Taro-no so-LIN Taro-LIN

akai hon red book red book” akai hon red book

In this approach, Japanese nominals do not project a DP node and demonstratives are adjoined to NP. The presence of a DP projection has been justified in many languages by various empirical evidence since Abney’s (1987) DP hypothesis (Stowell 1989, among others). For instance, it is observed that a definiteness-inducing element such as a demonstrative creates an island and blocks movement out of DP in languages like English (see Stowell 1989). (4)

a. *Whoi did you see that picture of ti? b. Whoi did you see three pictures of ti?

In the standard analysis the SpecDP position in (4a) is filled with the demonstrative that and hence movement out of the DP violates the Minimal Link Condition. By contrast, in (4b) the SpecDP position is available as an escape hatch and movement is allowed. An AP analysis of demonstratives predicts that demonstratives in Japanese do not show definiteness effects that are observed in languages like English in which a noun phrase projects a full DP. Contrary to the prediction, Furuya (2008) reports that island effects are observed with extraction out of the nominal phrase when the demonstrative so is present

36

Chapter Three

(see also Fukuda 1996 for wh-island effects with demonstratives a “this” and ko “that” in Japanese). In (5) the numeral classifier san nin “three persons” may be extracted from a nominal expression to the sentenceinitial position. Note that the existence of the AP modifier does not block extraction of the numeral classifier, indicating that an AP forms no island for movement. The examples in (5) and (6) are adapted from Furuya (2008: 155). (5)

San nini Suzuki sensei-wa [yuushuuna gakusei three CL Suzuki teacher-TOP good student o sikatta ACC scolded “Prof. Suzuki scolded three good students”

ti ]-

On the other hand, extraction of the numeral classifier san nin “three persons” outside the nominal phrase across the demonstrative so-no is disallowed, as illustrated by the example in (6b). (6)

a. Sensei-wa so-no karera san nin-o shikatta. teacher-TOP so-LIN them three CL-ACC scolded lit. “The teacher scolded those them three” b. *San nini sensei-wa so-no karera-o ti shikatta three CL teacher-TOP so-LIN them-ACC scolded (Furuya 2008: 155)

On the basis of definiteness effects for extraction, Furuya (2008) argues for the existence of a DP projection in Japanese. Adopting Campbell’s (1996) small clause analysis for English definite noun phrases to Japanese, Furuya proposes that demonstratives are generated in the specifier position of D in Japanese. (7)

[[DP demonstrative [SC pro [NP ]]]

Furuya (2008) only reports the extraction facts for postnominal numerals. I demonstrate that prenominal numerals, too, are sensitive to islands in the presence of so. The examples in (8) illustrate that a prenominal numeral classifier may be freely moved to the sentence-initial position in the absence of demonstratives.

DP-decomposition Analysis of Japanese Demonstrative so

(8)

37

a. Sensei-wa san nin-no danshigakusee-o hometa teacher-TOP three CL-LIN male.student-ACC praised “The teacher praised three male students” b. (?)San nin-noi sensei-wa ti danshigakusee-o hometa. three CL-LIN teacher-TOP male.student-ACC praised

Although the grammaticality of the sentence in (8b) is slightly more degraded than the corresponding extraction example of the postnominal numeral in (5), what is significant is that the sentence suddenly becomes clearly ungrammatical with the presence of so, as illustrated in the example in (9b). (9)

(Context: John, Bill and Mike are very good students.) a. Sensei-wa so-no san nin-no danshigakuseiteacher-TOP so-LIN three CL-LIN male.studento hometa ACC praise.past “The teacher praised the/these three male students” b. *?San nin-noi sensei-wa so-no ti danshigakuseethree CL-LIN teacher-TOP so-LIN male.studento hometa ACC praise.past

The contrast in grammaticality between (5)/(6b) and (8b)/(9b) in Japanese parallels the grammaticality contrast between (4a) and (4b) in English, which is straightforwardly accounted for if the non-deictic demonstrative so is in SpecDP, blocking cyclic movement of the numerals out of the nominal phrase.

3. The two-way split of so Under a uniform SpecDP analysis of demonstratives, the presence of the demonstrative so is predicted to yield island effects. However, there are at least two uses of so that do not yield island effects. These uses of so exhibit further differences with respect to NP-ellipsis. This section presents the two uses of so, the specific indefinite so and the bound variable so, and argues that these two uses of so are structurally different from the so that yields definiteness effects.

38

Chapter Three

3.1 Non-definite so Section 2 discussed the non-deictic demonstrative so that obtains a definite interpretation. This does not mean, however, that the non-deictic demonstrative so always provides the nominal phrase a definite interpretation. The non-deictic demonstrative so may yield a non-definite interpretation when the co-occurring nominal phrase contains a nondefinite interpretation. Consider an example in (10). The so modified nominal phrase so-no mongoru-no rikishi “the Mongolian sumo-wrestler” is discourse anaphoric (and hence non-deictic) in the sense that it refers to mongoru-no rikishi “Mongolian sumo-wrestler” in the previous context. As the English translation suggests, however, the entire nominal phrase refers to any Mongolian sumo-wrestler and not the uniquely identifiable individual, indicating that so does not contribute a definite interpretation. (10)

Mongoru-no rikishi-wa totemo tsuyoi. TaroMongol-LIN sumo.wrestler-TOP very strong Tarowa so-no mongoru-no rikishi to taisen TOP so-LIN Mongol-LIN sumo.wrestler with match suru do “Mongolian sumo-wrestlers are very tough. Taro will be matched against a Mongolian sumo-wrestler.”

In this paper I call this use of so specific indefinite, because it refers to the nominal expression in the previous context but the referent is not uniquely identified as in the case of indefinite NPs (cf. Fodor and Sag 1982 argue that specific indefinites are ambiguous between referential and quantificational). That a so modified NP may yield a non-definite interpretation is evidenced by the selectional restriction of a certain verb class. An intentional verb like boshuu-suru “seek” requires a non-definite NP as its complement in Japanese. This is demonstrated by the fact that an NP that is typically definite like a proper name and a pronoun is disallowed as a complement of boshuu-suru. (11)

Shi-wa nanninka kyooin/*John/*karera-o community-TOP some.CL.Q teacher/John/they-ACC shiteiru do “The community seeks some teachers/John/them.”

boshuu seek

DP-decomposition Analysis of Japanese Demonstrative so

39

In (11), a noun kyooin “teacher” is forced to have a non-definite interpretation due to the selecting verb boshuu shiteiru “seek”, and it is interpreted as “teachers (in general)”. The specific indefinite so may combine with such a non-definite noun without changing the referential property of the entire nominal phrase. In (12) kyooin in the first sentence refers to teachers in general. So-no kyooin in the second sentence still refers to any teachers and not a specific individual teacher or teachers. The fact that so-no kyooin co-occurs with the verb boshuu shiteiru “seek” also indicates that the so attached NP in (12) yields non-definite interpretation. (12)

Kyooin-wa saikin ninkino takai shokugyo da. teacher-TOP recently popular high occupation COP Shi-wa so-no kyooin-o boshuu shiteiru community-TOP so-LIN teacher-ACC seek do lit. “Teachers are a very popular occupation nowadays. The community seeks a teacher/teachers.”

Another use of so that does not contribute definite interpretation is a bound variable use of so (Hoji 1991, Noguchi 1997, etc.). Hoji (1991) points out that the so modified NP can be bound by a quantificational NP. (13)

Dono zyooshi-mo so-no buka-o every boss-Q so-LIN subordinate-ACC “Every boss criticized his subordinate.”

shikatta criticized

Here so is merely bound by its antecedent and the referent of the so modified nominal expression so-no buka is determined by its antecedent dono zyooshi-mo “every boss”. The so attached NP does not refer to a specific individual and therefore so may be counted as non-definite. The difference between the definite so and the non-definite so can be observed in word order. When a numeral classifier is added, the two instances of so exhibit a different word order. The presence of the verb boshuu shiteiru “seek” signals that each sentence requires a non-definite nominal in the examples in (14). On the other hand, the context in (15) shows that the so modified nominal phrase refers to specific individuals and therefore it is definite.

40

Chapter Three

(14)

(Context: Teachers are a popular occupation nowadays.) a. Shi-wa so-no kyooin san nin-o boshuu community-TOP so-LIN teacher three CL-ACC seek shiteiru do “The community seeks three teachers.” b. #Shi-wa so-no san nin-no kyooin-o community-TOP so-LIN three CL-LIN teacher-ACC boshuu shiteiru seek do

(15)

(Context: The three teachers, Taro, Hanako and John, are very good teachers.) a. Shi-wa so-no kyooin san nin-o community-TOP so-LIN teacher three CL-ACC shoosanshita praised “The community praised the three teachers.” b. Shi-wa so-no san nin-no kyooin-o community-TOP so-LIN three CL-LIN teacher-ACC shoosanshita. praised

The contrast in acceptability between the two word orders in the nominal phrase in (14) and (15) shows that the so-numeral-NP order is unambiguously associated with a definite interpretation, whereas the soNP-numeral order may be ambiguous.

3.2 NP ellipsis It has been observed in the traditional literature of Japanese that null arguments may be given two distinct readings. Consider the examples in (16). The second sentence contains a null subject which may be interpreted either as anaphoric to the subject in the first sentence, i.e. the same student who visited Taro (strict reading), or a different student from the one who visited Taro (sloppy reading). (16)

Gakusee-ga Taro-o tazuneta. e Ziro-mo student-NOM Taro-ACC visited Ziro-also “A student visited Taro. (S)he visited Ziro, too.” “A student visited Taro. A student visited Ziro, too.”

tazuneta visited (strict) (sloppy)

DP-decomposition Analysis of Japanese Demonstrative so

41

According to the literature (Kuroda 1965, Hoji 1985, Saito 1985), the missing argument in a construction like (16) is best analyzed as involving (NP) ellipsis rather than an empty pronoun due to the availability of the sloppy reading. Following the literature, I call the construction that involves a missing argument in (16) NP ellipsis construction. When the antecedent of the missing argument is a pronoun or a proper name that is typically interpreted as definite, only the strict reading, and not the sloppy reading, is available. (17)

John/boku-ga Taro-o tazuneta. John/I-NOM Taro-ACC visisted John/boku -ga Ziro-mo tazuneta John/I-NOM Ziro-also visited “John/I visited Taro. John/I visited Ziro, too.”

(strict/*sloppy)

The contrast between (16) and (17) in terms of the available interpretation indicates that NP ellipsis yields an unambiguously strict reading if the elided NP involves definiteness. So attached nominal phrases behave differently with respect to NP ellipsis, depending on its use. Like a pronoun and a proper name, an appearance of the definite so yields only a strict reading. (18)

(Context: John, Mary and Mike study Japanese.) So-no san nin-no gakusee-ga Taro-o tazuneta. so-LIN three CL-LIN student-NOM Taro-ACC visited So-no san nin-no gakusee-ga Ziro-mo tazuneta so-LIN three cl-LIN student-NOM Ziro-also visited “The three students visited Taro. They visited Ziro, too.” (strict/*sloppy)

In (18) the elided subject so-no san nin-no gakusee-ga “the three students” in the second sentence refers to the same set of students that visited Taro and it does not refer to a different set of students that visited Taro. On the other hand, when the elided nominal phrase contains the specific indefinite so, the sentence may yield either a sloppy reading or a strict reading.

42

(19)

Chapter Three

(Context: A doorstep salesman is not welcome, since (s)he can often be annoying.) So-no seerusuman-ga Taro-o tazuneta. so-LIN salesman-NOM Taro-ACC visited So-no seerusuman-ga Ziro-mo tazuneta so-LIN salesman-NOM Ziro-also visited “The salesman visited Taro. (S)he visited Ziro, too.” (strict) “A salesman visited Taro. A salesman visited Ziro, too.” (sloppy)

In (19) the salesman who came to Taro may either be the same salesman who visited Ziro or a different salesman. Similarly, the bound variable so may yield either a sloppy reading or a strict reading. (20)

Dono hahaoya-mo so-no kodomo-o shikari every mother-Q so-LIN child-ACC scolded.and dono chichioya-mo so-no kodomo-o hometa every father-Q so-LIN child-ACC praised “Every mother scolded her child, and every father praised his child.” (sloppy) “Every mother scolded the child, and every father scolded her/him.” (strict)

Given that NP ellipsis unambiguously yields a strict reading when a definite NP is elided, the fact that the definite so in (18) unambiguously yields a strict reading whereas the specific indefinite so (19) and the bound variable so (20) give rise to ambiguity in terms of the interpretation of the elided NP indicates that the definite so but not the specific indefinite so and the bound variable so involves definiteness.

3.3 . Extraction of numeral classifiers The two-way split of so in terms of definiteness interpretation and NPellipsis seems to be a structural split. More specifically the definite so is a D-element whereas the specific indefinite so and the bound variable so are not associated with the D projection. The main supporting evidence for this claim comes from the extraction of numerals. Unlike the definite so that is reported to yield island effects (Furuya 2008), the specific indefinite so and the bound variable so do not show island effects with respect to numeral classifier extraction. (21) illustrates extraction of a numeral classifier from a nominal phrase with the specific

DP-decomposition Analysis of Japanese Demonstrative so

43

indefinite so. The so attached nominal phrase so-no noruuee-zin-o go nin refers to ‘any five Norwegians’ and in this sense yields a non-definite interpretation. The numeral classifier go nin “five people” is allowed to move outside the nominal phrase boundary without changing the meaning of the sentence, as illustrated by the example in (21b). (21)

specific indefinites (Context: Norwegian people are in general easy-going and easy to entertain.) a. Taro-wa so-no noruueezin-o go nin settaishita. Taro-top so-LIN Norwegian-ACC five CL entertained “Taro entertained five Norwegians.” b. Go nini Taro-wa sono noruueezin-o ti settaishita. five CL Taro-top so-LIN Norwegian-ACC entertained

One may claim that the numeral classifier go nin in (21a) is outside the nominal phrase boundary and hence (21b) is not an example of extraction out of the nominal phrase, given the fact that the numeral classifier follows the accusative case marker o in (21a). However, Watanabe (2006) convincingly argues that the numeral classifier that immediately follows a case marker, as in example (21a), is within the nominal phrase. Note also that extraction of a numeral classifier is possible when the numeral classifier immediately precedes the case marker, which indicates that the numeral classifier is clearly within the nominal phrase. In (22a) the nominal phrase so-no noruueezin go nin-o refers to five non-specific Norwegians, indicating that the prenominal demonstrative so is specific indefinite. Similar to (21b) the numeral classifier go nin in (22b) may be moved out of the nominal phrase without changing the interpretation of the nominal phrase. (22)

specific indefinites (Context: Norwegian people are in general easy-going and easy to entertain.) a. Taro-wa so-no noruueezin go nin-o settaishita. Taro-TOP so-LIN Norwegian five CL-ACC entertained “Taro entertained five Norwegians.” b. Go nini Taro-wa so-no noruueezin ti o settaishita. five CL Taro-TOP so-LIN Norwegian ACC entertained

Furthermore, the bound variable so allows extraction of a numeral classifier out of the nominal phrase.

44

(23)

Chapter Three

a. Dono zyooshi-moi so-noi buka san nin-o every boss-Q so-LIN subordinate three CL-ACC hihanshita. criticized “Every boss criticized three of his subordinates.” b. San nin dono zyooshi-mo so-no buka ti three CL every boss-Q so-LIN subordinate o hihanshita. ACC criticized

The fact that the definite so but not the specific indefinite so and the bound variable so shows island effects in extraction out of the nominal phrase strongly suggests that the definite so, on the one hand, and the specific indefinite so and the bound variable so, on the other hand, are structurally different. Given the notion of a phase and Phase-Impenetrability Condition (PIC) of Chomsky (1999) and the standard assumption that D constitutes a phase head (e.g., Svenonius 2004), the observed island effects in (9) indicates that the definite so is in SpecDP and the subsequent movement of the numeral classifier violates the PIC. On the other hand, the lack of island effects in (21) through (23) implies that SpecDP must be available as an escape hatch in nominal phrases with the specific indefinite so and the bound variable so.

4. Analysis Given that the demonstrative so exhibits definiteness effects depending on its use, it is reasonable to assume that the definite so is structurally different from the specific indefinite so and the bound variable so that do not yield definiteness effects. The next question to ask is the syntactic structure of the demonstrative so that gives rise to two different interpretations in terms of definiteness. The fact regarding island effects in section 3 suggests that SpecDP is available as an escape hatch for extraction in the specific indefinite and the bound variable so nominal but not in specific so nominals. A possible account that would uniformly associate so to the DP-layer may posit that the definite so is in SpecDP whereas the specific indefinite and the bound variable so are in the D head. This analysis, however, can be immediately rejected. Japanese being a strictly head-final language, so cannot occupy the D-head, since this derives an ungrammatical word order with a postnominal demonstrative such as *noruueezin so (compared with so-no noruueezin “the Norwegians”).

DP-decomposition Analysis of Japanese Demonstrative so

45

Another alternative analysis might be to say that the definite so is in SpecDP whereas the specific indefinite and the bound variable so are NPadjoined APs. This alternative faces an immediate empirical problem. Given that the order among adjectives is relaxed in Japanese, the specific indefinite so and the bound variable so are expected to either precede or be preceded by adjectives. Contrary to this expectation, both uses of so have to precede modifiers. The example in (24) illustrates this point with the bound variable so.5 (24)

a. Daremoi-ga so-noi kawaii musume-o kawaigatteiru. everyone-NOM so-LIN cute daughter-ACC love “Everyone loves his cute daughter.” b. *Daremoi-ga kawaii so-noi musume-o kawaigatteiru.

There seems to be no clear syntactic or semantic reason as to why the demonstrative so and the adjective kawaii “cute” are disallowed to cooccur if both of them are adjoined to the NP in (24).6 The order restriction between the demonstrative and the adjective rather seems to suggest that so is not a NP adjunct but it is in a higher position than the adjoined AP. I argue that so is generated in the specifier position of an intermediate functional phrase between DP and NP. From this position the demonstrative moves to SpecDP in order to check the definiteness related features when D contains definiteness related features. As for the structure of nominal phrases, I adopt Watanabe’s (2006) format of the structure of nominal phrases. Watanabe (2006) proposes that nominal phrases in Japanese project at least four extended projections above NP: #P that is the locus of the number specification, CaseP that is responsible for Case features and specificity/presuppositional quantification, QP that is responsible for the mass/count distinction and DP that encodes information about definiteness. (25)

[DP [QP [CaseP [#P [NP ] #] Case] Q] D]

The hallmark of Watanabe’s theory is the assumption that the four different patterns of the sequence of a noun, a numeral classifier and a case marker italicized in the examples in (26) all comprise a single constituent and that they all are derived from the common base structure in (25).

46

(26)

Chapter Three

a. John-wa hon san satsu-o John-TOP book three CL-ACC “John bought three books.” b. John-wa san satsu-no hon-o John-TOP three CL-LIN book-ACC c. John-wa hon-o san satsu John-TOP book-ACC three CL d. John-wa san satsu hon-o John-TOP three CL book-ACC (Watanabe 2006: 4)

katta bought katta bought katta bought katta bought

In his theory, classifiers and numerals are base-generated in different positions the #-head and its specifier. The NP undergoes obligatorily movement to SpecCaseP which is triggered by the agreement relation in the Case features, yielding the word order in (26a). This obligatory movement excludes the ungrammatical word order Num-NP-CL (san hon satsu). If #P further moves to QP for the mass/count specification, stranding the NP in SpecCaseP, the order in (26b) obtains. If the further movement of CaseP to SpecDP takes place, in order to check specificity related features, we get the order in (26c). This movement is motivated by the observation that Case and specificity are intimately related (Enç1991, Giusti 1995, among others). Lastly, (26d) may be derived by a further movement of CaseP outside the DP followed by scrambling of the remnant phrase that contains the numeral classifier. I propose that demonstratives are base-generated in SpecCaseP and may raise to SpecDP to check definiteness related features. In Watanabe’s analysis, both CaseP and DP are responsible for reference fixing: CaseP is responsible for specificity, and DP is responsible for definiteness. The definite so, the specific so and the bound variable so all provide the nominal expression they attach to with reference, via indexing with its antecedent. It is therefore not unreasonable to assume that so is generated in SpecCaseP. Note that the current proposal is in line with the “low source” proposals for demonstratives by Giusti (1993), Brugè (1996, 2002), Bernstein (1997), among others, in which demonstratives are argued to originate in a low position in DP and move to SpecDP to license the D projection.7 Furthermore, the current structure of nominals conforms to Hawkins’ (1983) implicational universal base order of nominals: Demonstrative > Numeral > Adjective > Noun. With the present analysis, the three uses of so will be assigned the structures illustrated in (27) through (29), in which b-examples represent base structures and c-examples represent derived structures.

DP-decomposition Analysis of Japanese Demonstrative so

47

(27)

definite so a. Shi-wa so-no san nin-no kyooin-o community-TOP so-LIN three CL-LIN teacher-ACC shoosanshita praised “The community praised the three teachers” b. [DP [QP [CaseP so [#P san [NP kyooin] [# nin]] [Case o]] Q] DDef] c. [DP soi [QP [#P san tNP nin-no] [CaseP ti [NP kyooin] t#P [Case o]] Q] DDef]

(28)

specific indefinite so a. Shi-wa so-no kyooin san nin-o boshu city-TOP so-LIN teacher three CL-ACC seek shiteiru do “(context: Teachers are a popular occupation.) The city seeks three teachers.” b. [DP [QP [CaseP so [#P san [NP kyooin] [# nin]] [Case o]] Q] D] c. [DP [QP [CaseP so [CaseP [NP kyooin] [#P san tNP nin] [Case o]]] Q] D]

(29)

bound variable a. Dono zyooshi-mo so-no buka san nin-o every boss-Q so-LIN subordinate three CL-ACC shikatta criticized “Every boss criticized his three subordinates.” b. [DP [QP [CaseP so [#P san [NP buka] [# nin]] [Case o]] Q] D] c. [DP [QP [CaseP so [CaseP [NP buka] [#P san tNP nin] [Case o]]] Q] D]

In (27) through (29) so is merged in SpecCaseP, as depicted in the bexamples. This is followed by the obligatory movement of the NP into the lower SpecCaseP (cf. for movement into multiple specifiers, see Richards 1999, 2001). If the demonstrative so further raises to SpecDP (followed/preceded by movement of the numeral to QP), (27) obtains. If the demonstrative so stays in CaseP, (28) and (29) result. Note that the order so-NP-CL-Case in (29a) may also be associated with a definite interpretation, in which case the demonstrative so is raised to SpecDP. Whether the demonstrative raises to SpecDP or stays in-situ in SpecCaseP is determined by the nature of the definiteness in DP. In this paper I adopt Enç’s (1991) definition of definiteness. Enç distinguishes definiteness and specificity with respect to their discourse linking

48

Chapter Three

properties. In her theory the discourse linking relevant for definite NPs is the identity relation which involves the strongest possible linking relation between the referent and its antecedent. The linking relevant for specific NPs, on the other hand, is the inclusion relation which involves a weaker relation, that of being a subset of or standing in some recoverable relation to a familiar object. According to Enç’s distinction, the definite so is definite to the extent that it requires its discourse referent to be linked to previously introduced discourse referents via the identity relation. On the other hand, the bound variable so does not involve any identity linking relation. Likewise the specific indefinite so involves no identity linking relation but it requires that the antecedent of its referent be included in a set that is familiar in the discourse. I suggest that this semantic distinction between definiteness and specificity is represented in the syntactic structure as features (cf. Ihsane and Puskas 2001 who argue on different grounds that specificity and definiteness project separate heads). More concretely, I suggest that a specificity related feature [specific] that yields the inclusion interpretation resides in the Case head and a definiteness related feature [definite] that concerns the strongest linking relation resides in the D head when the DP is definite. I assume that demonstratives always contain the specificity related feature but the definiteness related feature is optional.8 The demonstrative and the D head agree in the [definite] feature when the DP is definite. This agreement relations that hold between the demonstrative and the D head trigger the proposed movement of a demonstrative to SpecDP. Consequently when D contains no [definite] feature, the demonstrative stays in CaseP. In this approach, the asymmetry between the definite so and the specific indefinite so and the bound variable so in terms of extraction possibility is straightforwardly explained. The examples of extraction asymmetry are repeated in (30) through (32). (30)

definite so (Context: John, Bill and Mike are very good students.) a. Sensei-wa so-no san nin-no danshigakusei-o teacher-TOP so-LIN three CL-LIN male.student-ACC hometa praise.past “The teacher praised the three male students.” b. *?San nin-no sensei-wa so-no t danshigakuseethree CL-LIN teacher-TOP so-LIN male.studento hometa ACC praise.past

DP-decomposition Analysis of Japanese Demonstrative so

49

(31)

specific indefinite so (Context: Norwegian people are in general easy-going and easy to entertain.) a. Taro-wa so-no noruueezin-o go nin Taro-TOP so-LIN Norwegian-ACC five CL settaishita entertained “Taro entertaind five Norwegians.” b. Go nini Taro-wa so-no noruueezin-o ti five CL Taro-TOP so-LIN Norwegian-ACC settaishita entertained

(32)

bound variable so a. Dono zyooshi-moi so-noi buka every boss-Q so-LIN subordinate hihanshita criticized “Every boss criticized three of b. San nin dono zyooshi-moi so-noi three CL every boss-Q so-LIN o hihanshita ACC criticized

san nin-o three CL-ACC his subordinate.” buka t subordinate

In (30a), the nominal expression so-no san nin-no danshigakusee “the three male students” is linked to and identified with three particular students, hence it is a definite expression. The definite so moves from SpecCaseP to SpecDP in order to check the relevant definiteness features. The SpecDP position is therefore occupied by so and it is not available as an escape hatch for movement out of DP. Consequently, movement of the numeral classifier san nin in (30b) violates the MLC. In (31a), the nominal expression so-no noruueezin gonin-o may refer to any five Norwegians that are linked to the discourse anaphor by the weak linking relation. In (31b), D does not contain [definite] and therefore the specific indefinite so stays in SpecCaseP. Thus SpecDP is available as an escape hatch for the movement of the numeral classifier out of DP. Similarly in (32b) the so attached NP is not linked to its antecedent by the strong linking relation. It does not therefore raise to Spec DP, allowing the extraction of the numeral expression via SpecDP. Concerning the NP-ellipsis facts, it is standardly argued that NP-ellipsis involves LF copy of the antecedent NP in the elided position, which yields ambiguity (Oku 1998). In the present analysis, it seems that the filled

50

Chapter Three

SpecDP blocks the LF-copying of an elided nominal phrase in the definite so constructions. Interestingly, in his recent work Sato (2012) argues that definiteness blocks LF-copying in NP-ellipsis constructions based on the observation that topic positions exhibit no sloppy readings. The present analysis conforms to Sato’s hypothesis of NP-ellipsis, although the details of the blocking mechanism are yet to be understood. Lastly, the present analysis of so may bring an interesting insight into the treatment of another type of ellipsis phenomenon, known as N’-ellipsis. In N’-ellipsis constructions, a part of a nominal phrase that contains a head noun may be deleted. The example in (33) is taken from Saito et al. (2008: 303). (33)

[Rooma-no hakai]-wa [Kyooto-no hakai]–yorimo Rome-LIN destruction-TOP Kyoto-LIN destruction-than hisan datta miserable was “Rome’s destruction was more miserable than Kyoto’s.”

According to Saito et al. (2008), N’-ellipsis is allowed when the specifier position of the functional head is filled. In their analysis, the nomarked NP Kyooto no in (33), being an argument, moves to SpecDP and licenses N’-deletion. When SpecDP is not filled, N’-deletion is not allowed. The no-phrases, hare no (34a) and the numeral classifier go-satsu no (34b), are adjuncts and cannot raise to SpecDP. The examples in (34), therefore, do not satisfy the licensing condition of N’-deletion. (34)

a. *[Hare -no hi]-wa yoi ga, [ame-no hi]-wa clear-LIN day-TOP good though rain-LIN day-TOP ochikomu feel.depressed “Clear days are OK, but I feel depressed on rainy days.” b. *Taroo-wa ichi-nichi-ni [san-satsu-no hon]-o yomu ga, Taroo-TOP one-day-in three-CL-LIN book-ACC read but Hanako-wa [go-satsu-no hoN]-o yomu Hanako-TOP five-CL-LIN book-ACC read “Taroo reads three books in a day, but Hanako reads five.” (Saito et al. 2008: 306)

Watanabe (2010) examines a wider range of N’-ellipsis phenomena and argues against the N’-ellipsis analysis of Saito et al. (2008). Appealing to Johnson’s (2001) analysis of VP ellipsis, Watanabe proposes that N’ellipsis is licensed by movement of the elided constituent (and not by

DP-decomposition Analysis of Japanese Demonstrative so

51

movement of the arguments of the elided NP) to the specifier position of the functional head. He supports his proposal based on the fact that N’ellipsis with a stranded numeral classifier is possible. (35)

Taroo-wa ichi-nichi-ni [san-satsu-no hon]-o yomu ga, Taro-TOP one-day-in three-CL-no book-ACC read but Hanako-wa [go-satsu hoN]-o yomu Hanako-TOP five-CL book-ACC read “Taroo reads three books in a day, but Hanako reads five.”

In the analysis of Saito et al., the numeral classifier go-satsu in (35) is an adjunct that cannot move to SpecDP. Consequently (35) does not satisfy the N’-ellipsis configuration and the sentence is wrongly predicted to be ungrammatical. Contrary to Saito et al., Watanabe argues that the head NP hon “book” moves to SpecCaseP and licenses N’-ellipsis. Therefore, (35) is correctly predicted to be grammatical. In Watanabe’s analysis the contrast in grammaticality between (33) and (34) is attributed to the interaction of ellipsis with the insertion of the linking element no. He posits the two types of no, a genitive case particle that is retained under ellipsis and a linking element between a modifier and a modified NP that is inserted in the morphological component. No in (33), being a genitive case particle, is retained when N’-ellipsis takes place. No in (34), on the other hand, is a mere linker that requires a following overt NP. The no phrases in (34) are not followed by an overt NP after N’-ellipsis and therefore the examples in (34) are excluded by the ban on the morphological linker that precedes no overtly realized head noun (for the formulation of the rule, see Watanabe 2010: 66). Consider N’-ellipsis with the demonstrative so. The three uses of so display the familiar two-way split behavior in terms of N’-ellipsis. When so is interpreted as a bound variable or a specific indefinite, N’-ellipsis is not available. On the other hand, the definite so allows N’-ellipsis. (36) (Context: Tanaka taught 5 students and Yamada taught 3 students.) Tanaka sensei-ga so-no zibun-no oshieta gakusee Tanaka teacher-NOM so-LIN self-LIN taught student go nin-o yatottaga, Yamada sensei-wa so-no five CL-ACC hired.but Yamada teacher-TOP so-LIN zibun-no oshieta gakusee san nin-o yatowanakatta self-LIN taught student three CL-ACC hired.neg. “Professor Tanaka hired the five students he taught, but Professor Yamada didn’t hire the three students he taught.”

52

Chapter Three

(37)

*Dono zyoseekyooshi-mo so-no seeto sannin-o every female.teacher-Q so-LIN student three.CL-ACC home, dono danseekyoshi-mo so-no seeto praised every male.teacher-Q so-LIN student futari-o shikatta two.CL-ACC scolded “Every female teacher praised three of her students and every male teacher scolded two of his students.”

(38)

(Context: The car industry was reluctant to hire foreign workers before.) *Toyota-ga so-no gaikokuzin gozyuu nin-o yatoi, Toyota-NOM so-LIN foreigner fifty CL-ACC hired, Nissan-ga so-no gaikokuzin yonzyuu nin-o yatotta Nissan-NOM so-LIN foreigner forty CL-ACC hired “Toyota hired 50 foreign workers and Nissan hired 40 foreign workers.”

Is the contrast between (36) and (37)/(38) expected in Saito et al.’s approach? Saito et al. may analyze the demonstrative so as an adjunct and predict that N’-ellipsis is not allowed in (37) and (38). 9 However, the adjunct analysis of the demonstrative would wrongly predict that N’ellipsis is disallowed with the definite so (36). Furthermore, their analysis will face another empirical problem with regard to N’-ellipsis in the DP with a locative phrase. According to Saito et al., locative and temporal phrases may license N’-ellipsis since they are assumed to be basegenerated in SpecDP. Given this, one would predict that insertion of a locative phrase to a DP with the demonstrative so licenses N’-ellipsis. This is not the case, however. In (39) the example that contains the specific indefinite so and the locative phrase Osaka-no does not allow N’-ellipsis.10 (39)

(Context: A NEET is a young person who is not employeed, not in education or training, and not seeking work.) *Tokyo-no so-no Niito-no kazu-wa Osaka-no Tokyo-LIN so-LIN Niito-LIN number-TOP Osaka-LIN so-no Niito-no kazu-o uwamawatteiru. so-LIN NEET-LIN number-ACC exceed “The number of NEETs in Tokyo exceeds the number of Neets in Osaka.”

DP-decomposition Analysis of Japanese Demonstrative so

53

Ungrammaticality of (39) indicates that something other than the filled SpecDP is responsible for licensing the N’-ellipsis configuration. It seems that Watanabe’s account of N’-ellipsis along with the insertion rule of the morphological linker predicts the contrast between (36) and (37)/(38)/(39) in grammaticality. In the present analysis, the elided NPs in (36) through (39) all move to SpecCaseP, satisfying Watanabe’s licensing condition of N’-ellipsis. Recall that Watanabe (2010) distinguishes the two types of no: a morphological linker (34) and a genitive case particle (33), of which the morphological linker requires the following overt NP. The present hypothesis is that the definite so is in SpecDP. Given that genitive phrases are assigned a genitive Case no by being in a spec-head relation with the D-head, it may be assumed that no that follows the definite so in (36) is a genitive Case marker. Consequently, so-no in (36) may be stranded without the following overt NP when N’-ellipsis takes place. On the other hand, similar to the no that follows a numeral classifier, the no that follows the bound variable so and the specific indefinite so, which are in SpecCaseP, is assumed to be a morphological linker. (37), (38) and (39) are, therefore, excluded due to the ban on the morphological linker that precedes no overt head noun. If the present analysis is on the right track, the N’-ellipsis data set with the demonstrative so in (36) through (39) may present additional supporting evidence to Watanabe’s analysis of N’ellipsis.

5. Summary This paper has argued that the demonstrative so is base-generated in the intermediate functional projection between DP and NP from where it may raise to SpecDP in order to check definiteness related features. The main evidence for the intermediate position of so is that the so with a nondefinite interpretation is insensitive to island effects for extraction and lacks definiteness effects in terms of NP-ellipsis. Adopting Watanabe’s (2006) phrase structure format, in which the Case-head as well as D is endowed with reference-fixing features, the paper argues that the demonstrative so is generated in SpecCaseP. Furthermore, the paper adopts Enç’s division of definiteness into strong (or definite) and weak linking (or subset relation) and suggests that the D-head encodes the strongest linking relation between the referent and its antecedent whereas the Case-head specifies the weak linking relation. If the D-head contains the strongest linking related features the demonstrative so moves from SpecCaseP to SpecDP, obtaining the definite interpretation and the

54

Chapter Three

definiteness effects. Consequently, in the present analysis the two-way split pattern of the demonstrative so with respect to the definite interpretation and the definiteness effects is straightforwardly accounted for by the presence or the absence of movement to SpecDP that is triggered by the definiteness related features. The present analysis of the demonstrative so is in line with the “low source” analyses of demonstratives (Guisti 1993, Brugè 1996, Bernstein 1997) in the sense that demonstratives are base generated in a specifier position of a functional head below DP and subsequently raise to SpecDP for licensing the D projection. Finally, the paper provides additional evidence to support the claim that noun phrases in Japanese contain the functional category DP as well as a lexical category NP, i.e. Abney’s hypothesis holds for Japanese.

References Abney, S. 1987. The English Noun Phrase in Its Sentential Aspect. Ph.D. dissertation, MIT. Bernstein, J. 1997. Demonstratives and reinforcers in Romance and Germanic languages. Lingua 102: 87–113. Boškoviü, Ž. 2005. On the locality of left branch extraction and the structure of NP. Studia Linguistica 59.1: 1–45. Brugè, L. 2002. The positions of demonstratives in the extended nominal projection. In Functional structure in DP and IP: The cartography of syntactic structures, ed. by Guglielmo Cinque, 15–53. Oxford: Oxford University Press. —. 1996. Demonstrative movement in Spanish: A comparative approach. University of Venice Working Papers in Linguistics 6.1: 1-53. Carlson, G N. 1977. Reference to Kinds in English. Ph.D. dissertation, University of Massachusetts at Amherst. Campbell, R. 1996. Specificity operators in spec DP. Studia Linguistica 30: 161–188. Chomsky, N. 1999. Derivation by Phase. MIT Occasional Papers in Linguistics 18. Cambridge, Mass. Enç, M. 1991. The Semantics of Specificity. Linguistic Inquiry 22: 1-25. Fodor, J., and I. Sag. 1982. Referential and quantificational indefinites. Linguistics and Philosophy 5: 355- 398. Fukuda, M. 1996. Syntax of Demonstrative Adjectives in Japanese: A Preliminary Study. Kansas Working Papers in Linguistics 21: 21-37. Fukui, N. 1988. Deriving the differences between English and Japanese: A case study in parametric syntax. English Linguistics 5: 249-70.

DP-decomposition Analysis of Japanese Demonstrative so

55

—. 1986. A Theory of Category Projection and Its Applications. Doctoral Dissertation, MIT. Furuya, K. 2008. DP hypothesis for Japanese “bare” noun phrases. Proceedings of the 31st Annual Penn Linguistics Colloquium 14: 149162. Giusti, G. 1995. A Unified Structural Representation of (Abstract) Case and Article: Evidence from Germanic. In Studies in Comparative Germanic Syntax, ed. by H. Haider, S. Olsen, and S. Vikner, 77-93. Dordrecht: Kluwer. —. 1993. Enclitic articles and double definiteness: A comparative analysis of nominal structure in Romance and Germanic. Linguistic Review 11: 241-255. Hawkins, J. 1983. Word Order Universals. New York: Academic Press. Hoji, H. 1985. Logical form constraints and configurational structures in Japanese. Ph.D. Dissertation, University of Washington. —. 1991. Kare. In Interdisciplinary approaches to language: Essays in honor of S.-Y. Kuroda, ed. by C. Georgopoulos and R. Ishihara, 287304. Dordrecht: Kluwer. Ihsane, T., and G. Puskas. 2001. Specific is not Definite. Generative Grammar in Geneva 2: 39–54. Johnson, K. 2001. What VP ellipsis can do, and what it can’t, but not why. In The handbook of contemporary syntactic theory, ed. by M. Baltin and C. Collins, 439–479. Oxford: Blackwell. Kim, Y-K. 1997. Agreement phrases in DP. UCL Working Papers in Linguistics 9: 1-24. Kuno, S. 1973. The Structure of the Japanese Language. Cambridge: MIT Press. Kuroda, S. 1979. (Ko) So, A nituite (On (ko), so and a). In Eigo to Nihongo to (English and Japanese), 41-59. Tokyo: Kurosio Publishers. —. 1965. Generative grammatical studies in the Japanese language. Ph.D. Dissertation, MIT. Noguchi, T. 1997. Two types of pronouns and variable binding. Language 43: 770–797. Oku, S. 1998. A theory of selection and reconstruction in the minimalist perspective. Ph.D. Dissertation, University of Connecticut. Richards, N. 2001. Movement in language: Interactions and architectures. Oxford: Oxford University Press. —. 1999. Featural cyclicity and the ordering of multiple specifiers. In Working Minimalism, ed. By S. D. Epstein and N. Hornstein, 127-158. Cambridge: MIT Press. Saito, M. 1985. Some asymmetries in Japanese and their theoretical

56

Chapter Three

implications. Ph.D. Dissertation, MIT. Saito, M., T.-H. Jonah Lin, and K. Murasugi. 2008. N’-ellipsis and the structure of noun phrases in Chinese and Japanese. Journal of East Asian Linguistics 17: 247–271. Sato, Y. 2012. Argument Ellipsis in Javanese and the Definiteness Restriction on Subjects at the Syntax-Information Structure Interface. Lingbuzz http://ling.auf.net/lingbuzz?_s=cAommuLzn1U ZpBMg&_k=eIQ-U28V80fPHSwD. Stowell, T. 1989. Subjects across categories. The Linguistic Review 2: 285–312. Svenonius, P. 2004. On the Edge. In Peripheries: Syntactic Edges and their Effects, ed. By D. Adger, C. de Cat, and G. Tsoulas, 261-287. Dordrecht: Kluwer. Watanabe, A. 2010. Notes on nominal ellipsis and the nature of no and classifiers in Japanese. Journal of East Asian Linguistics 19: 61-74. —. 2006. Functional Projections of Nominals in Japanese: Syntax of Classifiers. Natural Language and Linguistic Theory: 241-306. Whitman, J. 1981. The internal structure of NP in verb final languages. Chicago Linguistic Society 17:411-18. Zamparelli, R. 2000. Layers in the Determiner Phrase. New York: Garland.

Notes 1

I am grateful to two anonymous reviewers and Isabelle Roy for their helpful comments and suggestions. 2 In this paper the following abbreviations are used in the glosses: TOP = topic marker; NOM = nominative; ACC = accusative; DAT = dative; GEN = genitive; COP = copula; CL = classifier; Q = quantifier; LIN=linker. 3 Japanese has three demonstrative prefixes ko, so, and a, each of which has both deictic and non-deictic uses. The deictic uses are standardly characterized in regard to the distance to the speaker and the hearer: ko is the speaker proximal, so is the hearer proximal and a is the speaker and the hearer distal. The non-deictic uses of these demonstratives are often characterized in terms of the speaker/hearer’s knowledge of the object referred to by the demonstratives (Kuno 1973, but see also Kuroda’s 1979 arguments against the relevance of the hearer). In more recent literature some linguists point out presumable structural differences between so and a/ko. Hoji (1991) presents, independently of these characterizations, that a so modified NP, and not an a modified NP, can be bound by a quantificational NP. This paper concentrates on the demonstrative so that seems to have different syntactic properties from the other two demonstratives. 4 No is a linking element that appears between a modifier and the NP that is modified in Japanese. No is often assumed as a genitive Case marker (e.g., Fukui

DP-decomposition Analysis of Japanese Demonstrative so

57

1986) or a contextual Case marker (Saito et al. 2008). However, it does not seem to be appropriate to uniformly analyze no as a Case marker, since it can be attached to PPs (Watanabe 2010). (i) Haruki-to-no intabyuu Haruki-with-no interview “An interview with Haruki”

(Watanabe 2010: 66)

Watanabe (2010) distinguishes the two types of no: a morphological linker and a genitive Case marker (see section 4). In this paper, I gloss no as LIN(KER) (but see the end of section 4 where I adopt Watanabe’s view of no). 5 The same fact can be reproduced with the specific indefinite so. (i) (Context: Norwegian people are in general very friendly.) a. so-no furendoriina noruueezin-o settaishita. so-LIN friendly Norwegian-ACC entertained “(I) entertained friendly Norwegians.” b. *furendoriina so-no noruueezin-o settaishita. 6

Recall Fukui’s (1986) examples in (1) in which the definite demonstrative so may be preceded by an adjective. With Fukui therefore one may still argue that the definite so is an NP adjunct. Kim (1997) argues that non-restrictive modifiers are base-generated in the specifier position of an intermediate functional phrase in DP, AgrP in her terms, from where they may move to SpecDP due to focus related features. She assumes that this movement is obligatory in head-final languages. It has been demonstrated that the adjective that precedes a demonstrative in Japanese is a non-restrictive modifier (Whitman 1981). With Kim, I assume that the adjective that precedes the demonstrative so is in SpecDP, and not an NP adjunct. Assuming that the definite so tucks in the SpecDP, the AP-demonstrative order observed in (1) can be derived without appealing to the AP analysis of demonstratives. 7 Zamparelli (2000) proposes to split a noun phrase into three layers: SDP (Strong Determiner Phrase), PDP (Predicative Determiner Phrase) and KDP (Kind Determiner Phrase). The highest projection SDP is responsible for referentiality and hosts definiteness, deictics, proper names and personal pronouns. The intermediate projection PDP is a layer that denotes a property and the determiners that can appear in a predicative position such as numerals and indefinites are accommodated in this layer. The lowest projection KDP is responsible for the formation of “kind” concepts in the sense of Carlson (1977) and this layer contains restrictive modifiers and a NP. It is suggested by an anonymous reviewer that the specific indefinite so and the bound variable so may be identified as Zamparelli’s KDP. This amounts to saying that the two types of so are restrictive adjectives since in Zamparelli’s analysis KDP hosts restrictive modifiers and not determiners, which is in line with Fukui’s (1986, 1988) analysis of demonstratives in Japanese. However, this faces an immediate empirical problem with respect to ordering

58

Chapter Three

between the demonstratives and adjective which is pointed out in section 4 (examples in 24) and therefore it is not pursued in this paper. Alternatively, one may analyze the indefinite specific so as PDP and the definite so as SDP. It is not, however, immediately clear how the various word order patterns among demonstratives, numeral quantifiers, and NPs observed in Japanese are derived in his analysis, since Zamparelli excludes floating quantifiers from his analysis. 8 Bare nominals may be interpreted as specific without the presence of demonstratives in Japanese. Such nominals may involve a null demonstrative. 9 Note that the non-elided sentences corresponding to (37) and (38) are grammatical. 10 Note that the non-elided sentence corresponding to (39) is grammatical.

CHAPTER FOUR IS RUSSIAN A VERB CLASSIFIER LANGUAGE? LAURA A. JANDA

UNIVERSITY OF TROMSØ1

1. The Verb Classifier Hypothesis This article focuses on the role of Russian verbal prefixes in forming aspectual correlates (termed “Natural Perfectives” as defined below) for imperfective simplex verbs. The aim is to test the following hypothesis: (1)

The Russian “purely aspectual” prefixes are a verb classifier system.

The Verb Classifier Hypothesis is operationalized in a rigorous series of tests in the body of this article. In aggregate, these tests show that the Russian perfectivizing prefixes in this study conform to the criteria for recognizing classifiers (described below). In addition to purely distributional criteria, the most important issue is to discover whether each prefix is unique in terms of its semantic behavior. The “null hypothesis” would be that there is a random, unpatterned distribution of prefixes across the meanings of verbs. The tests in this study give compelling evidence that refutes the null hypothesis. Section 2 offers a brief discussion of classifier systems, drawing parallels with Russian verbal prefixes. In Section 3 I evaluate the Russian “purely aspectual” prefixes according to standard definitions of classifiers. Sections 4 through 8 present five statistical studies. Conclusions appear in Section 9. I distinguish among four basic types of perfectives in Russian (Janda 2007), three of which are relevant here: x Natural Perfectives, where the lexical meaning matches that of the imperfective simplex verb, as in s-varit’ ‘cook’ which is the

Chapter Four

60

perfective partner verb of the imperfective varit’ which also means ‘cook’. Prefixes used to form Natural Perfectives are often characterized as “purely aspectual” or “semantically empty” (cf. Švedova et al. 1980). I acknowledge as Natural Perfectives the verbs in the Exploring Emptiness database found at http://emptyprefixes.uit.no/. x Specialized Perfectives, where the lexical meaning of the verb is different from that of the corresponding simplex verb, as in razvarit’ ‘cook until soft’. x Complex Act Perfectives, where the prefix sets boundaries on the action named in the simplex verb, as in po-varit’ ‘cook for a while’. I compare the use of prefixes in Natural Perfectives with their use in the other types of perfectives where it is uncontroversial that prefixes have meanings. As indicated above, the set of Natural Perfectives is defined as those included in the Exploring Emptiness database, and the distribution of prefixes in Natural Perfectives is presented in Figure 1 below.

2. Numeral Classifier Systems as a Model Among noun classifier systems, gender systems are the most common type worldwide, followed by the numeral classifier type. Numeral classifier systems are most common in languages of Asia and Central America (Aikhenvald 2000: 98-124). Table 1 shows examples of how numeral classifers are used in Yucatec Maya. Examples ‘un-tz’íit kib’ ‘un-tz’íit che’ ‘un-tz’íit nal ‘un-tz’íit há’as

Glosses one long-thin wax one long-thin wood one long-thin corn one long-thin banana

‘un-wáal há’as ‘un-kúul há’as ‘un-kúuch há’as

one flat banana one planted banana one load banana

Translations “one candle” “one stick” “one ear of corn” “one fruit of the banana” “one banana leaf” “one banana tree” “one bunch of bananas”

Table 1: Examples of numeral classifiers in Yucatec Maya (from Lucy 2000: 329)

Is Russian a Verb Classifier Language?

61

The classifier system in Yucatec Maya has two types of reference: a bare noun without a classifier refers to a substance, while objects are specified by adding classifiers. Russian has two types of verbs: imperfective, which is the default value for a simplex stem, and perfective, often signaled by the addition of a prefix. Janda (2004) showed that imperfective verbs in Russian behave like nouns that refer to substances, while perfective verbs behave like nouns that refer to objects. A simplex Russian verb nearly always refers to the verbal equivalent of a substance. If we want to talk about an event in Russian, we need to add a prefix and perfectivize the verb; this is parallel to using a classifier in Yucatec Maya to refer to an object. Classifiers are associated with the use of numerals. Aspect in Russian, and particularly perfective aspect, has been described as a type of quantification (Jakobson 1957/1971: 136; see also Smith 1991 and Dahl 1985). Shape is the most common parameter conveyed by numeral classifiers cross-linguistically (Aikhenvald 2000: 98 and Lyons 1977: 465-466). Russian verbal prefixes are transparently related to prepositions with spatial meanings. The shape an object takes in space is thus paralleled by the spatio-temporal contours of an event. Nouns in Yucatec Maya are sorted into groups by their classifiers. Russian prefixes likewise divide the simplex verbs into groups. Figure 1 visualizes this sorting effect: each bar represents the number of simplex verbs that form Natural Perfectives with the given prefix, ranging from three Natural Perfectives formed with the prefix v- to 417 Natural Prefixes formed with the prefix po-. Both Yucatec Maya numeral classifiers and Russian “purely aspectual” prefixes have been considered by some linguists to be purely formal, semantically empty markers. Numeral classifiers are now recognized to have a semantic sorting function. The argument that Russian has a verb classifier system is de facto also an argument against the traditional “empty prefix” model. Given the clear parallels between numeral classifiers and Russian verbal prefixes, one might ask: Why has Russian not been recognized as a verb classifier language before? Linguists have only recently become aware of verb classifier systems. McGregor (2002: 404) states that verb classification “has not yet been incorporated into mainstream linguistic knowledge as a category that might be expected in a language”, but that it is “a far from exotic phenomenon” and “[d]oubtless it is not confined to the relatively few languages in which it has been hitherto described, though the extent of its distribution across the world’s languages remains to be charted.”

62

Chapter Four

Figure 1: Distribution of Natural Perfectives across prefixes

In this article I take a comprehensive approach to the collection and quantitative analysis of data to support the Verb Classifier Hypothesis.

3. Distributional Evidence McGregor (2002: 16-22) and Gerner (2009: 708) give a four-part definition of classification (presented in italics below), and here I evaluate how Russian measures up if we assume that “classifiers” are prefixes and “classifieds” are verbs. (i) There are a finite number of ways in which classifiers and classifieds can co-occur. For Russian, the prefixes and verbs co-occur in prefixed verbs. (ii) The group of classifiers has more than one element. Russian has sixteen prefixes that form Natural Perfectives. (iii) The group of classifieds has significantly more elements than the group of classifiers. Russian has 1,429 verbs that are classified by the sixteen prefixes. (iv) At least two of the groups of classifieds that are associated with two different classifiers must be significantly different from each other. Sections 4 and 5 prove in detail that each prefix is associated with a different group of verbs.

Is Russian a Verb Classifier Language?

63

Russian prefixes pass all four tests for classifiers. McGregor (2002: 17) makes an additional requirement that the members of a set of classifiers “must show different behaviours”. The five case studies presented in the following sections give evidence that the prefixes behave differently. Individually the above criteria are certainly not adequate to protect against recognizing as classifiers linguistic units that no reasonable linguist would accept as classifiers. For example, many types of linguistic units fulfull criterion (iv) by presenting different groups of items. It is important, however, that we use all of the criteria in concert, and in addition examine the behavior of the proposed classifiers. This article presents a comprehensive analysis that takes all of the criteria, McGregor’s additional requirement, plus other parallels into consideration in proposing that Russian perfectivizing prefixes function as verb classifiers.

4. Behavioral Evidence 1: Radial Category Profiling of v-, pod-, pere-, pri-, ot-, v(ɨ)z-, u-, iz-, raz-, vy-, ɨ(b)-2 We divide the sixteen prefixes into two groups in order to optimize the analysis. The eleven “small” prefixes presented in this section form few enough Natural Perfectives to facilitate a detailed analysis, but do not present enough data to support statistical models. The empirical method applied in this section belongs to a suite of quantitative methods developed at the University of Tromsø for analysis of form-meaning relationships. Radial category profiling is one of these methods, and it analyzes the distribution of data across radial category networks, such as the related submeanings of a given linguistic unit. Polysemy is characteristic of Russian prefixes, cf. the sample of Specialized Perfectives of raz- in Table 2. Simplex verb pilit’ ‘saw’ toptat’ ‘stamp one’s feet’ katat’ ‘roll’ dut’ ‘blow’ gruzit’ ‘load’

Raz-prefixed verb raz-pilit’ ‘saw apart’ raz-toptat’ ‘trample’ raz-katat’ ‘roll (dough)’ raz-dut’ ‘inflate’ raz-gruzit’ ‘unload’

Meaning of razAPART CRUSH

out

SPREAD SWELL UN-

Table 2: Some Specialized Perfectives and the meanings contributed by raz-

64

Chapter Four

The radial category model of meaning was inspired by the work of Rosch (1978), who showed that human beings organize examples of a concept in radial categories, with a central prototype and other less central examples that bear some relationship to the prototype. Cognitive linguistics (cf. Lakoff 1987, Taylor 2003) has adopted the radial category model for semantic analysis as an effective means for modeling complex networks of meanings. Rather than being defined by features and boundaries, radial categories are defined by relationships to a prototype. The prototype is the meaning that serves as the semantic center of gravity of the category and is its most representative meaning; it is also the meaning that is most densely connected to other meanings (Geeraerts 1995: 25; Croft and Cruse 2004: 78, 81; Lewandowska-Tomaszczyk 2007: 155). Given that the most central meaning is often also the most concrete one, for Russian prefixes, this is often the meaning that is associated with motion verbs since there we typically see the spatial path designated by the prefix. The prototypical meaning is also often, but not necessarily, the most common meaning. Of course all the meanings in a given radial category are related to the prototypical meaning, but this does not mean that they are automatically derived from that meaning. Radial categories are preferred over single abstract features in that they reflect the structure of relationships among meanings in a category and thus better reflect the actual complexity of the data. Radial categories are also preferred over mere inventories of meanings because radial categories reflect the structure of relationships among meanings in a category. A discussion of the relative merits of features, lists, and radial categories is presented in Janda 1986. The radial category profiling method involves comparing radial categories. If we discover that two radial categories share some meanings, the internal structure of the categories will show us whether the shared meanings are distributed in a random fashion or whether they constitute a coherent subset. Overlap that involves coherent subsets shows us that the same radial category is present for the items being compared, though the radial category may be more extensive for one item than for the other. Radial category profiling of prefixes involves comparison of behaviors of Natural Perfectives as opposed to Specialized and Complex Act Perfectives. The analysis involves three steps for each prefix. First the radial category of prefix meanings is established on the basis of the verbs in which prefix meaning is clear and uncontroversial: the Specialized and Complex Act Perfectives, such as those represented in Table 2. Next, the meanings of the simplex verbs that use the same prefix to form Natural

Is Russian a Verb Classifier Language?

65

Perfectives are analyzed in terms of a radial category. Finally, the two radial categories are compared to check for overlap. Radial category profiling has been carried out on the basis of approximately two thousand verbs for all eleven small prefixes, and all the results are available at http://emptyprefixes.uit.no/methodology_rus.htm. This website lists all of the meanings with examples, and gives complete inventories of all of the relevant verbs. In this section, I present the analysis of the prefix raz-. The radial category in Figure 2 is based on analysis of 148 Specialized Perfectives. Figure 2 visually collapses all three steps of the analysis. The boxes represent the meanings of raz-, based on the Specialized Perfectives (SP). The numbers in parentheses represent the number of verbs found in each meaning for each type of perfective, and one example is given for each type. Boxes that show overlap between the meanings of raz- and the meanings of the simplex verbs that form Natural Perfectives with raz- are shaded. For example, meaning 1 is APART; there are thirty-six Specialized Perfectives like raz-pilit’ ‘saw apart’ with this meaning; and there are twenty-two Natural Perfectives like raz-gryzt’ ‘gnaw’ with this meaning.

Figure 2: Radial Category for the prefix raz-

66

Chapter Four

The prototypical meaning of raz-, APART, involves breaking up the unity of either a group or an object, such that individuals or parts go their separate ways. The link between APART and CRUSH is motivated by the fact that when something is crushed, its internal structure is destroyed (taken APART) and the edges may move outward. A cluster of meanings (36) focuses on the dispersal that is inherent in APART, yielding SPREAD, SWELL, and SOFTEN/DISSOLVE. Because excitement tends to spread and things that are excited often swell, we also have a meaning of EXCITEMENT in this cluster. The final meaning is UN-, which is related to the prototype in that undoing something is a kind of taking APART. All of the boxes except 7 in Figure 2 are shaded, indicating that we find both Specialized Perfectives and Natural Perfectives in the majority meanings. There are no Natural Perfectives in the UN- meaning. A Natural Perfective is by definition a verb that has the same meaning as the corresponding simplex verb. The UN- meaning creates prefixed perfectives that have the opposite meaning of the simplex verbs. This clash makes it impossible to form Natural Perfectives from raz- in this meaning. All eleven small prefixes show a similar pattern. For two prefixes overlap is complete, covering all meanings in the radial category: u- and v. For seven prefixes overlap includes most meanings in the radial category: raz-, pri-, ot-, v(o)z-, o(b)-, vy- and iz-. Two prefixes show overlap in a minority of meanings: pere- and pod-. In all instances where overlap is incomplete, the prototypical meaning plus a coherent subset of neighboring meanings exhibit overlap. There is also a clear logic to the pattern of meanings excluded from overlap. The meanings where we do not find Natural Perfectives are incompatible with simple perfectivization in that they involve negation, comparison, quantification, or some other special qualification. The effect of the prefix on the verb is minimized in Natural Perfectives due to overlap. Each prefix selects the simplex verbs that conform best to the meanings in its radial category, and together the prefixes sort the perfectivizable simplex verbs into groups as predicted by the Verb Classifier Hypothesis. Our next task is to see whether there is evidence of prefixal meaning in Natural Perfectives among the remaining five prefixes, and that is the topic of Section 5.

5. Behavioral Evidence 2: Semantic Profiling of po-, s-, nɚ, zɚ-, prɨ-3 The Russian National Corpus (www.ruscorpora.ru, henceforth “RNC”) uses a system of semantic tags based on work by the Moscow Semantic

Is Russian a Verb Classifier Language?

67

School. We look at the distribution of semantic tags for Natural Perfectives and submit these results to statistical tests for significance, effect size, and the degree of attraction (or repulsion) for each combination of prefix and semantic tag. The chi-square test requires a minimum of five expected observations in each cell of a matrix and assumes that all observations are independent. Over 63% of all Natural Perfectives in Russian are prefixed with pɨ-, s-, nɚ-, zɚ-, and prɨ-; thus they provide a large enough quantity of data across a small enough number of prefixes to satisfy the minimum requirement for the test. If we take the semantic tags for which we have at least fifty verbs across the five prefixes, we have four categories for semantic tags: ‘impact’, ‘changest’, and ‘behav’, plus a combined category of ‘sound’ and ‘speech’. The assumption of independent observations means that we have to take two measures in order to make our data suitable for a chi-square analysis, because there are both verbs that select more than one prefix and verbs that have more than one semantic tag in the RNC. After restricting the data to include only the five prefixes and five semantic tags, and only verbs that take exactly one prefix and have exactly one semantic tag, we have 382 verbs distributed as shown in Table 3. A full list of these verbs and details concerning how values were calculated are available on our website (http://emptyprefixes.uit.no/semantic_eng.htm).

‘impact’ ‘changest’ ‘behav’ ‘sound’ & ‘speech’ column totals

pɨ-

s-

nɚ-

zɚ-

prɨ-

11 62 11 37

23 11 23 9

31 3 17 8

47 22 1 1

10 4 0 51

121

66

59

71

65

Row totals 122 102 52 106 Grand total: 382

Table 3: Distribution of prefixes across semantic tags Table 3 is visualized in Figure 3. Po- favors ‘changest’, while pro- is dominated by ‘sound’ & ‘speech’. ‘Impact’ makes a strong showing in s-, na-, and za-, but the balance of other semantic tags is different for each one.

68

Chapter Four

The p-value (probability value) of the chi-square test tells us the likelihood that we would get the observed distribution (or one that is even more uneven) if there was no relationship between semantic tags and prefixes. The chi-square value is 248, with 12 degrees of freedom, and a pvalue of 2.2e-16 (an extremely low value). In other words, the result is statistically very significant. We measure the effect size by calculating the Cramer’s V value, which can range from 0 to 1 and is evaluated according to the following scale: 0.1 is considered “small”, 0.3 is considered “moderate”, and 0.5 is considered “large”. The Cramer’s V value for our data is 0.8, which far exceeds the standard measure for a large effect.

Figure 3: Semantic profiles of po-, s-, na-, za-, and pro-

We have thus found that the distribution of prefixes across semantic tags of verbs is both highly significant and has a large effect size. We can with confidence state that the differences we see in Figure 3 are robust and meaningful. This study, like the previous one, shows that each prefix seeks out a specific group of verbs.

Is Russian a Verb Classifier Language?

69

6. Behavioral Evidence 3: Constructional Profiling of na-, za-, po-4 The Russian verb gruzit’ ‘load’ has three Natural Perfectives prefixed in nɚ-, zɚ-, and pɨ-, and all four verbs can appear in both the “themeobject” construction, as in gruzit’ jašþiki na telegu ‘load the boxes onto the wagon’, and the “goal-object” construction, as in gruzit’ telegu jašþikami ‘load the wagon with boxes’. The objective in this section is to show that the three prefixes behave differently in terms of their distribution across the two constructions. This different behavior provides evidence in support of the Verb Classifier Hypothesis. 1,920 examples of the ‘load’ verbs were extracted from the RNC, and of these 895 are active forms while 1,025 are passive participles. The distribution of active forms is shown in Table 4, and this data is visualized in terms of percentages in Figure 4. Verbs gruzit’ na-gruzit’ za-gruzit’ po-gruzit’

Theme-object constructions 208 34 94 253

Goal-object constructions 78 113 114 1

Total 286 147 208 254

Table 4: Numbers of examples of active forms for the two constructions Each of the ‘load’ verbs has a unique constructional profile: unprefixed gruzit’ prefers the theme-object construction, but is found with the goalobject construction. Na-gruzit’ shows the opposite tendency, since the theme-object construction is less frequent than the goal-object construction. For za-gruzit’ the distribution is nearly balanced between the theme-object construction and the goal-object construction. Po-gruzit’ is found almost exclusively in the theme-object construction. Although this distribution is statistically significant with a large effect size,5 this represents less than half of the data and it appears that there are other factors that have to be taken into account, so a simple chi-square test is not enough.

70

Chapter Four

Figure 4: Examples of active forms as percentages for each verb

The use of passive forms clearly has an influence on the choice of the construction. For passive forms, one construction is preferred in over 95% of examples for every verb. The preference of the simplex verb gruzit’ is the reverse of what it is for active forms: for passive forms the goal-object construction is strongly preferred and the theme-object construction is nearly excluded. Both na-gruzit’ and za-gruzit’ have a preference for the goal-object construction among active forms, and for passive forms this preference is enhanced to the point that the goal-object construction is nearly exclusive. Only the constructional profile of po-gruzit’ remains unchanged, with a strong preference for the theme-object construction. Reduced versions of the constructions, where the theme-object construction names only the theme and the goal-object construction names only the goal, are fairly common and found with all of the prefixes. In all, there are 1,353 examples of full constructions in our database and 567 examples of reduced constructions. We want to discover whether the prefixes in the Natural Perfectives influence the choice of the grammatical construction. Two other factors may be relevant: the use of active vs. passive forms and the use of full vs. reduced constructions. We gauge the contributions of the various factors by using a logistic regression model. In addition to looking at the individual contribution of each variable, which is called a “main effect”, the logistic regression model can discover whether two or more

Is Russian a Verb Classifier Language?

71

independent variables have a combined effect, which is called an “interaction”. Our database and the analysis are available on our website (http://emptyprefixes.uit.no/constructional_eng.htm). The logistic regression analysis shows that all of the independent variables, namely the choice of prefix, the use of passive participles, and the use of reduced vs. full constructions, serve as main effects. In addition there is an interaction between choice of prefix and the use of active vs. passive forms. The following highly significant correlations are listed in order of relative strength, starting with the strongest one: x Na-gruzit’ and za-gruzit’ prefer the goal-object construction. x Unprefixed gruzit’ and po-gruzit’ favor the theme-object construction. x The use of passive participles contributes to the choice of the construction, and this varies according to the choice of prefix. x Full constructions are more often found with the theme-object construction, while goal-object constructions are more likely to be reduced. The model has high correlation strength and correctly predicts the construction for 88.5% of examples. To summarize, the analysis supports the Verb Classifier Hypothesis: the prefixes behave differently in terms of choice of construction. This is confirmed even though there are other factors that show significant relationships.

7. Behavioral Evidence 4: Prefix Variation6 The verb gruzit’ ‘load’, analyzed in Section 6, illustrates prefix variation because it forms Natural Perfectives from not one, but three prefixes. It is essential to measure prefix variation in order to show that it does not conflict with the distributional requirements for recognizing Russian as a verb classifier language. Classifiers typically do show some variation of this type, in that some classifieds can combine with more than one classifier. However, it should be the case that the majority of classifieds use one classifier and variation should be systematic and should reflect semantic patterns. In the Exploring Emptiness database we find 1,429 imperfective simplex verbs in Russian that form 1,981 Natural Perfectives via prefixation. 1,043 simplex verbs select one and only one prefix. The remaining 386 (27%) simplex verbs show prefix variation. Prefix variation can involve up to six prefixes that attach to a single simplex verb. There are 283 simplex verbs that form two Natural Perfectives, 75 simplex verbs

72

Chapter Four

that form three Natural Perfectives, 21 that form four Natural Perfectives, and only 4 and 3 simplex verbs that form five and six Natural Perfectives respectively. Prefix variation involves all sixteen prefixes, and the extent to which a prefix is engaged in prefix variation is roughly keyed to the overall frequency of the prefix. Highly frequent prefixes like po-, s-, and za- are also more often involved in prefix variation and can be combined with all or nearly all other prefixes. A prefix combination is a particular selection of prefixes that are used by one or a group of simplex verbs that participate in prefix variation. Thus for example gruzit’ ‘load’ has the prefix combination na-|za-|pobecause it can form Natural Perfectives with these three prefixes. However, binary prefix combinations are by far the most frequent. Theoretically it is possible to make 120 binary combinations of sixteen prefixes (120 = (16x15)/2), but only a few of them are common. There are thirty-one binary combinations that are used by three or more simplex verbs, twenty-three that are rare and sixty-six that are not found at all. Many of the unattested combinations involve prefixes with clearly opposed meanings, such as pri-|u-, where pri- indicates arrival, but uforms verbs expressing departure. A full list can be found on our website (http://emptyprefixes.uit.no/variation_eng.htm). We examine one case study in order to demonstrate the dynamic relationship between compatible and incompatible meanings found in a prefix combination. Table 5 lists all the simplex verbs than can form Natural Perfectives with both za- and u-, including both binary and larger combinations. We see that most of these verbs can be gathered into semantic groups, listed on the left. The simplex verbs that form their Natural Perfectives with zɚ- and uhighlight the common ground these prefixes share. Zɚ- in its CHANGE TO A FIXED STATE meaning and u- in its HARM meaning both refer to events that reduce capacity and mobility, yielding the DAMAGE group. A similar relationship is present in the CHANGE OF STATE group, which also focuses on both fixed states and reduction. The WRAP group capitalizes on the fact that both prefixes are associated with covering and putting things in containers. The semantic groups in Table 5 intersect. All types of DAMAGE and WRAPPING result in a CHANGE OF STATE. All of the WRAPPING verbs entail loss of mobility, as well as an increase in control. Thus the intersection of DAMAGE, WRAP, and CHANGE OF STATE define nearly the whole group of verbs with the zɚ-|u-combination, with the exception of þinit’ ‘fix’ and platit’ ‘pay’.

Is Russian a Verb Classifier Language? Semantic group

Simplex verb

za- meaning

DAMAGE

davit’ ‘crush’

CHANGE TO FIXED STATE

A

HARM

Prefix combination za-|pɨ-|u-

dušit’ ‘strangle’ morit’ ‘exterminate’ kolot’ ‘stab’

CHANGE TO FIXED STATE

A

HARM

za-|pɨ-|u-

CHANGE TO FIXED STATE

A

HARM

vy-|zɚ-|pɨ-|u-

CHANGE TO FIXED STATE

A

HARM

v-|zɚ-|raz-|u-

vjanut’ ‘wilt’

CHANGE TO FIXED STATE

A

HARM

zɚ-|u-

CHANGE TO FIXED STATE

A

MOVE DOWNWARDS

vy-|zɚ-|u-

CHANGE OF STATE

trambovat’ ‘ram’ vjaznut’ ‘get stuck’

CHANGE TO A FIXED STATE/ ATTACHMENT

KEEP/SAVE

zɚ-|u-

CHANGE TO FIXED STATE

A

REDUCE

za-|pɨ-|u-

CHANGE TO FIXED STATE

A

REDUCE

zɚ-|s-|u-

WRAP

gasit’ ‘extinguish’ molknut’ ‘fall silent’ kutat’ ‘wrap’

COVER

COVER COMPLETELY

zɚ-|o(b)-|u-

motat’ ‘wind’

COVER

COVER COMPLETELY

pakovat’ ‘pack’ þinit’ ‘fix’

COVER

PLACE/FIT

zɚ-|nɚ-|pɨ|prɨ-|uzɚ-|u-

CHANGE TO FIXED STATE

A

HARM

platit’ ‘pay’

CHANGE TO FIXED STATE

A

MOVE AWAY

OTHER

u- meaning

73

zɚ-|o(b)-|pɨ|uzɚ-|u-

Table 5: Simplex verbs that select the prefix combination za-|uIn some cases the two Natural Perfectives are clearly semantically distinct, as in for example kolot’ ‘stab’, which has za-kolot’ ‘slaughter’ vs. u-kolot’ ‘prick’. In other cases the meanings are very close. The Natural Perfectives of the CHANGE OF STATE group can be interchangeable in some contexts, as in (2)-(3). However, these two prefixed perfectives are motivated in slightly different ways: za-vjaznut’ ‘get stuck’ involves getting caught on something as signalled by the ATTACH and CHANGE TO A FIXED STATE meanings of zɚ-, whereas u-vjaznut’ ‘get stuck’ focuses on the downward direction of u- and describes sinking down into something.

74

Chapter Four

(2)

Idet zajac mimo bolota, vdrug vidit -- los’ v trjasine za-vjaz. [Kollekcija anekdotov: zveri (1970-2000)] Goes hare past swamp, suddenly sees -- moose in mire stuck. “A hare walks by a swamp and suddenly sees that a moose has gotten stuck in the mire.”

(3)

Predstavljaeš’, ja segodnja na beregu v gline u-vjaz, a ona menja vyvolokla. [Vladislav Krapivin. Boltik (1976)] Imagine, I today on bank in clay stuck, and she me pulled-out. “Just imagine, today I got stuck in the clay on the riverbank, and she pulled me out.”

Prefix variation is clearly governed by the meanings of the prefixes involved. The simplex verbs that can show overlap with more than one prefix tend to cluster in groups according to their meanings. While many examples of prefix variation are motivated by similarity of meanings across prefixes, we can detect in most cases differences between the Natural Perfectives of a given simplex verb, and often the prefixes focus on possible contrasts within the meaning of the simplex verb. There is a delicate balance between similarity and contrast, since even very near synonyms can have slightly different meanings in some contexts. We also see that meaning is a deciding factor in prefix combinations that are rare or do not exist, since prefixes with meanings that are incompatible avoid prefix variation. Our study of prefix variation supports the Verb Classifier Hypothesis because there is evidence that prefix variation is governed by relationships between the meanings of the prefixes and the meanings of simplex verbs. Prefix variation exists because different prefixes can focus the meanings of a simplex verb in different ways. Prefix variation thus enriches the spectrum of shades of meaning that verbs can express in Russian.

8. Behavioral Evidence 5: Aspectual Triplets An aspectual triplet is a set of three verbs with the same lexical meaning consisting of: a simplex verb, a prefixed Natural Perfective, and a secondary imperfective derived via suffixation of the Natural Perfective. An example is množit’sja, u-množit’sja, u-množ-at’sja, all of which mean ‘multiply, increase’. Aspectual triplets are not traditionally acknowledged in Russian linguistics; it is assumed instead that secondary imperfectives are formed only from Specialized Perfectives, not from Natural Perfectives (cf. Vinogradov et al. 1953: 431-432).

Is Russian a Verb Classifier Language?

75

Our aim is to explore the extent of the triplet phenomenon and to show that it is governed by the meanings of the prefixes. The prefixes show different behaviors in terms of the formation of triplets and this is further evidence in support of the Verb Classifier Hypothesis. The purpose of the Russian imperfectivizing suffixes -yvɚ/-iva, -va, -ɚ/-ja is to supply imperfective partner verbs for perfectives as needed. Ɍhis strategy is nearly universal among Specialized Perfectives, where the perfective verb would otherwise lack an imperfective partner. For example, the Specialized Perfective pere-pisat’ ‘rewrite’ is formed from the simplex pisat’ ‘write’. Because the Specialized Perfective has a clearly different meaning, the simplex cannot serve as its aspectual partner verb, and a secondary imperfective pere-pisyvat’ ‘rewrite’ is created with the help of the suffix -yva. A Natural Perfective formed via prefixation of a simplex verb already has an imperfective partner verb because the simplex shares the same lexical meaning. For example, the Natural Perfective u-množit’sja ‘multiply’ already has the imperfective partner verb množit’sja. There should be no need for an additional imperfective partner verb here. Nonetheless, the secondary imperfective u-množ-at’sja is robustly attested, with nearly 500 examples like (4) in the RNC. (4)

Ⱥ ved’ zazerkal’e -- prodolženie doma, s pomošþ’ju zerkal razdvigajustja steny, u-množaetsja koliþestvo svetil’nikov. [Marija Maganova. ýary zazerkal’ja (2004) // «Homes & Gardens», 2004.12.01] And after-all mirror-use -- extension home, with help mirrors move-apart walls, is-multiplied number lamps. “After all the use of mirrors extends one’s home, with the help of mirrors the walls move apart and the number of lamps is multiplied.”

If we assume, however, that the meanings of the prefix and the simplex verb overlap and that the meaning of the simplex verb is narrowed and focused by the prefix in the formation of the Natural Perfective, there can be slight differences in meaning between the simplex verb and its Natural Perfective. Under these circumstances it would make sense to form secondary imperfectives that emphasize the meaning specific to the Natural Perfective and make it possible to express that meaning with an imperfective verb. Previous studies have presented triplets as a restricted phenomenon, usually acknowledging only forty or fewer triplets (cf. Xrakovskij 2005,

76

Chapter Four

Jasai 2001, Apresjan 1995, Zaliznjak & Mikaelian 2010). We have taken a different approach: we formed hypothetical secondary imperfectives for all 1,981 Natural Perfectives listed in the Exploring Emptiness database and conducted searches for them in both the RNC and the Google search engine. 733 (37%) of the 1,981 possible secondary imperfectives are attested in the RNC, and 1,536 (77%) of the possible secondary perfectives were found in Google. A full inventory of our results can be accessed on our website (http://emptyprefixes.uit.no/triplets_eng.htm). Clearly Russians use secondary imperfectives of Natural Perfectives, even though most of these are not acknowledged in dictionaries. Native speakers must also know what these secondary imperfectives mean and have strategies for how to use them in contrast with the corresponding simplex verbs. We turn now to the meaning of the secondary imperfectives and their use. Unlike other previous scholars who have examined aspectual triplets, Veyrenc (1980) suggested that secondary imperfectives are motivated by a unified meaning, giving particular emphasis to the result of the action. We find evidence in support of this suggestion, because verbs that focus on tangible, intentional, or controllable results favor the use of the secondary imperfective. Wherever aspectual triplets exist, there are two forms -- the simplex imperfective and the secondary imperfective -- that compete as the imperfective partner verbs for the Natural Perfective. For some Natural Perfectives the simplex is the most common choice, for others the secondary imperfective predominates, and sometimes both forms are chosen with nearly equal frequency. Table 6 presents a sample of how this choice is made across a variety of verbs. Each row in Table 6 corresponds to an aspectual triplet for which all three verbs are attested in the RNC. The leftmost columns list the Natural Perfective, followed by the meaning of the prefix in the Natural Perfective. The next four columns represent the competition between the simplex imperfective and the secondary imperfective citing the percentage of use for each. The final column lists the total number of imperfective uses for the triplet found in the RNC. The rows representing triplets are arranged according to the balance between the use of simplex imperfectives and secondary imperfectives, with triplets most preferring simplex verbs on the top and those most preferring secondary imperfectives on the bottom.

Is Russian a Verb Classifier Language? Natural Perfective

Prefix meaning

Simplex imperfective rugat’

Use

vy-rugat’ ‘curse’ vzvolnovat’sja ‘get excited’ ras-tajat’ ‘melt’ o-šalet’ ‘go crazy’

NEGATIVE EXHAUSTION AGITATE

volnovat’sja

SOFTEN /DISSOLVE ACQUIRE A NEW FEATURE CHANGE TO A FIXED STATE

u-bajukat’ ‘lul to sleep’

77

Secondary imperfective vy-rug-ivat’

Use 0.01%

Total # in RNC 7,193

99.97%

vz-volnovyvat’sja

0.03%

7,763

tajat’

99.34%

ras-taj-ivat’

0.66%

3,465

šalet’

87.76%

o-šale-vat’

12.24%

98

žarit’

84.13%

za-žar-ivat’

15.87%

1,821

CONTROL

bajukat’

41.85%

u-bajukivat’

58.15%

454

za-žmurit’ ‘squint’

COVER

žmurit’

37.30%

za-žmurivat’

62.70%

984

o-svjatit’ ‘bless, sanctify’

IMPOSE A NEW FEATURE

svjatit’

20.00%

o-svjašþ-at’

80.00%

130

za-molknut’

CHANGE TO A FIXED STATE

molknut’

0.71%

za-molk-at’

99.29%

1,692

za-žarit’ ‘fry’

99.99%

Table 6: Competition between simplex imperfectives and secondary imperfectives in aspectual triplets We can explore the differences in use among triplets where the two imperfectives are both well represented, such as those headed by the Natural Perfectives u-bajukat’ ‘lull to sleep’ and za-žmurit’ ‘squint’. Examples (5) and (7) show the use of the simplex imperfectives bajukat’ and žmurit’; examples (6) and (8) illustrate the use of the corresponding secondary imperfectives u-bajuk-ivat’ and za-žmur-ivat’. (5)

Devoþka kaprizniþala i trebovala, þtoby vmesto babuški ee bajukal Dimka. [Mark Sergeev. Volšebnaja galoša, ili Neobyknovennye prikljuþenija Vadima Smirnova, ego luþšego druga Paši Kaškina i 33 nevidimok iz 117-j školy (1971)]

78

Chapter Four

Girl made-fuss and demanded, that instead-of grandmother her lulled Dimka. “The girl made a fuss and demanded that Dimka lull her instead of her grandmother.” (6)

Dnem Gusja u-bajuk-ivala doþ’ pod odnu i tu že pesenku... [Tanja Marþant. Kvartiranty (2003) // «Lebed’» (Boston), 2003.10.12] Daytime Gusja lulled daughter under one and same EMPH song... “During the day Gusja lulled her daughter to sleep with one and the same song.”

(7)

On [kot] sidel u nožki stola i žmuril zelenye svoi glaza. [Jurij Družkov (Postnikov). Volšebnaja škola (1984)] He [cat] sat by leg table and squinted green own eyes. “It [the cat] sat by the leg of the table and squinted its green eyes.”

(8)

On mne rasskazyval, þto on mal’þiškoj idja domoj s trenirovki, zažmur-ival glaza, proxodja mimo kioska s moroženym. [Tat’jana Tarasova, Vitalij Melik-Karamov. Krasavica i þudovišþe (19842001)] He me told, that he boy going home from training, squinted eyes, through-walking past stand with ice-cream. “He told me that when he was walking home from training as a young boy, he would squint his eyes when he passed by the icecream stand.”

These examples support Veyrenc’s (1980) suggestion that the secondary imperfectives focus on the result of the action. In addition, we find that the secondary imperfectives are more likely to refer to an action that is not only concluded, but repeated. In (5) with the simplex verb the girl is only interested in who will rock her, not in the result. By contrast, in (6) with the secondary imperfective, lulling her daughter to sleep is something that Gusja does successfully and repeatedly during the daytime, always with the same song. In (7) with the simplex verb, squinty eyes merely describe the cat. Example (8) with the secondary imperfective is part of a larger narrative: the boy was always hungry after sports practice, but he knew that his parents were poor, so he squinted his eyes as he went by the ice-cream stand in order to avoid temptation. These examples illustrate a trend in the distribution of the two imperfectives of an aspectual triplet. The simplex imperfective tends to describe an event where the focus is not on the result. The secondary

Is Russian a Verb Classifier Language?

79

imperfective tends to describe something that creates a result, often an intentional one, in a predictable repeated pattern. The meanings of the prefixes are decisive. Prefix meanings listed at the top of Table 6 do not indicate concrete results that can be intentionally repeated, but the prefixal meanings become more compatible with this pattern as we go lower in the table. Our comprehensive corpus-based analysis of aspectual triplets in Russian supports the Verb Classifier Hypothesis. The use of secondary imperfectives is neither uniform nor random: some triplets feature nearly exclusive use of the simplex verb, some favor the secondary imperfective, and others show a more balanced relative frequency. The choice of simplex vs. secondary imperfective appears to be governed by the meanings of the prefixes. Secondary imperfectives are preferred when the meaning of the prefix motivates focus on a result, preferably produced intentionally and/or repeatedly. Prefix meanings that are less compatible with this meaning of secondary imperfectives reduce or exclude the use of secondary imperfectives.

9. Conclusion I argued that the Russian verbal prefixes that form Natural Perfectives constitute an aspectual classifier system. I presented various kinds of evidence to support the Verb Classifier Hypothesis. These include comparisons between numeral classifier systems and Russian verbal prefixes, distributional facts, and behavioral patterns. There are strong parallels between the use of numeral classifiers and Russian perfectivizing prefixes: both numeral classifiers and aspectual prefixes are associated with quantification; and whereas numeral classifiers sort unbounded substances into objects according to shape, verbal prefixes sort unbounded states and activities into events according to their spatiotemporal contours. The role of verbal prefixes in Russian differs markedly from that in other European languages (e.g., Germanic languages) since Russian prefixes are associated with aspect and quantification in a more specific and consistent way. This fact sets Slavic languages apart and motivates the parallel with numeral classifiers, which is not appropriate for non-Slavic languages in the region. The description of Russian verbal prefixes as a classifier system comports well with a metaphorical description of Russian aspect, according to which imperfective verbs have the properties of metaphorical substances and perfective verbs have the properties of metaphorical objects.

80

Chapter Four

Russian verbal prefixes meet the distributional criteria of a classifier system, with a much larger number of classifieds (= verbs) than classifiers (= prefixes), and significantly different groups of classifieds associated with the various classifiers. Five different studies give evidence that the prefixes show different behaviors, as expected for a classifier system. The radial category profiling and semantic profiling studies show that each of the sixteen prefixes is associated with a different set of simplex verbs, and that the motive for this classification is semantic: the meanings of the prefixes overlap with the meanings of the simplex verbs. The constructional profile study shows that different prefixes have different behaviors in terms of the grammatical constructions they attract. Prefix variation is also consistent with what we find in classifier systems and the patterns of variation are motivated by the meanings of the prefixes. Lastly, the study of aspectual triplets shows differences in the behaviors of prefixes, since some prefixes are more frequently associated with the formation of secondary imperfectives than others, and again this behavior is linked to the meanings of the prefixes. By embracing the hypothesis that Russian prefixes are aspectual classifiers, we gain cross-linguistic support for the proposal that the prefixes are not “empty”, but instead overlap with the meanings of the verbs. The comparison with numeral classifier systems gives us an important insight into the Russian aspectual classifier system: its purpose is to convert amorphous states and activities into discrete events and to sort these into meaningful groups. Recognizing Russian prefixes as verb classifiers brings three kinds of advantages: descriptive, typological, and pedagogical. In terms of description, we gain access to significant patterns in the language that would otherwise be overlooked. Description of the prefixes as classifiers facilitates typological comparisons with other languages and discovery of trends. Traditionally, second language learners of Russian are told that they must simply memorize approximately two thousand seemingly random combinations of simplex imperfective verbs with “empty” prefixes. However, armed with this new understanding of the role of prefixes we can envision new pedagogical materials that would present the behavior of prefixes in a coherent, accessible way.

Is Russian a Verb Classifier Language?

81

References Aikhenvald, A. Y. 2000. Classifiers: A Typology of Noun Categorization Devices. Oxford: Oxford University Press. Apresjan, J. D. 1995. “Traktovka izbytoþnyx aspektual’nyx paradigm v tolkovom slovare”. Izbrannye trudy 2 (Integral’noe opisanie jazyka): 102-113. Croft, W. and D. A. Cruse. 2004. Cognitive Linguistics. Cambridge: Cambridge University Press. Dahl, Ö. 1985. Tense and Aspect Systems. Oxford: Basil Blackwell. Endresen, A., L. A. Janda, J. Kuznetsova, O. Lyashevskaya, A. Makarova, T. Nesset, S. Sokolova. 2012. “Russian ‘purely aspectual’ prefixes: Not so ‘empty’ after all?”. Scando-Slavica 58:2, 231-291. Geeraerts, D. 1995. “Representational formats in cognitive semantics”. Folia Linguistica 29: 21-41. Gerner, M. 2009. “Instruments as verb classifiers in Kam (Dong)” Linguistics 43:3, 697-743. Jakobson, R. O. 1957/1971. “Shifters, Verbal Categories, and the Russian Verb”. In Roman Jakobson: Selected Writings II. The Hague: Mouton, 130-147. Janda, L. A. 1986. A Semantic Analysis of the Russian Verbal Prefixes ZA-, PERE-, DO- and OT- (= Slavistische Beiträge, Band 192). Munich: Otto Sagner, 1986. —. 2004. “A metaphor in search of a source domain: the categories of Slavic aspect”. Cognitive Linguistics 15.4: 471-527. —. 2007. “Aspectual clusters of Russian verbs”. Studies in Language 31.3: 607-648. Janda, L. A. and O. Lyashevskaya. 2011. “Prefix variation as a challenge to Russian aspectual pairs: Are ɡɚɜɹɡɧɭɬɶ and ɭɜɹɡɧɭɬɶ ‘get stuck’ the same or different?”. Russian Linguistics 35: 147-167. Janda, L. A. and O. Lyashevskaya. 2013. “Semantic Profiles of Five Russian Prefixes: po-, s-, za-, na-, pro-”. Journal of Slavic Linguistics 21.2: 211-258. Jasai, L. 2001. “O specifike vtoriþnyx imperfektivov i vidovyx korreljacij”. In Issledovanija po jazykoznaniju: K 70-letiju A. V. Bondarko, ed. by I. V. Nedjalkov, 106-118. St. Petersburg: Institut lingvistiþneskix issledovanij. Lakoff, G. 1987. Women, Fire and Dangerous Things. Chicago: University of Chicago Press. Lewandowska-Tomaszczyk, B. 2007. “Polysemy, prototypes and radial categories”. In The Oxford Handbook of Cognitive Linguistics, ed. by

82

Chapter Four

D. Geeraerts and H. Cuyckens, 139-169. Oxford: Oxford University Press. Lucy, J. A. 2000. “Systems of nominal classification: a concluding discussion”. In Systems of nominal classification, ed. by Gunter Senft, 326-341. Cambridge: Cambridge University Press. Lyons, J. 1977. Semantics, v. 2. Cambridge: Cambridge University Press. McGregor, William B. 2002. Verb Classification in Australian Languages. (= Empirical Approaches to Language Typology 25). Berlin/New York: Mouton de Gruyter. Rosch, E. 1978. “Principles of Categorization”. In Cognition and Categorization, ed. by Eleanor Rosch and Barbara L. Lloyd, 27-48. Hillsdale: Lawrence Earlbaum. Smith, C. S. 1991. The Parameter of Aspect. Dordrecht: Kluwer. Sokolova, S., L. A. Janda, and O. Lyashevskaya. 2012. “The locative alternation and the Russian ‘empty’ prefixes: A case study of the verb gruzit’ ‘load’”. In Frequency Effects in Language, vol. 2, ed. by S. Th. Gries and D. Divjak, 51-86. Berlin: Mouton de Gruyter. Švedova, N. J. et al. (eds.) 1980. Russkaja grammatika, V. 1. Moscow: Nauka. Taylor, J. R. 2003. Linguistic Categorization. Oxford: Oxford University Press. Veyrenc, J. 1980. Études sur les verbes russes. Paris: Institut d’Etudes Slaves. Vinogradov, V. V. et al. (eds.). 1953. Grammatika russkogo jazyka V. 1: Fonetika i morfologija. Moscow: Akademija nauk SSSR. Xrakovskij, V. S. 2005. “Aspektual’nye trojki i vidovye pary”. Russkij jazyk v nauþnom osvjašþenii 9.1: 46-59. Zaliznjak, A. A. and I. L. Mikaèljan. 2010. “O meste vidovyx troek v aspektual’noj sisteme russkogo jazyka”. Dialog 2010. Moscow. http://www.dialog-21.ru/digests/dialog2010/materials/html/21.htm

Note  1 In this article I report on research undertaken together with Anna Endresen, Julia

Kuznetsova, Olga Lyashevskaya, Anastasia Makarova, Tore Nesset, and Svetlana Sokolova. This research was supported by the Norwegian Research Council, the Centre for Advanced Study at the Norwegian Academy of Sciences in Oslo, and the University of Tromsø. 2 For a comprehensive exposition of this study, see Endresen et al. 2012. 3 For a comprehensive exposition of this study, see Janda and Lyashevskaya 2013. 4 For a comprehensive exposition of this study, see Sokolova et al. 2012.

Is Russian a Verb Classifier Language?

83

 5

For the active forms: chi-square = 293.3285, degrees of freedom = 3, p-value < 2.2e-16, effect size (Cramer’s V) = 0.6. 6 For a comprehensive exposition of this study, see Janda and Lyashevskaya 2011.

PART II: ISSUES IN CATEGORY CHANGE

CHAPTER FIVE CATEGORY CHANGE: INTRODUCTION GIANINA IORDĂCHIOAIA UNIVERSITÄT STUTTGART

ISABELLE ROY UNIVERSITÉ PARIS 8–CNRS

KAORI TAKAMINE NORWEGIAN UNIVERSITY OF SCIENCE AND TECHNOLOGY

Modern research on category change has overwhelmingly concentrated on nominalizations, as they have issued one of the most prominent linguistic debates starting with Chomsky's (1970) Remarks on Nominalization and still continuing today: the controversy between lexicalist and syntactic approaches to word formation. After Lees' (1960) and Lakoff's (1965) transformational treatment of all nominalizations as derived from sentences, Chomsky proposes that important differences between the less productive derived nominals and the fully productive gerunds could be accounted for by assuming that a lexicon enriched with syntactic features is the locus of the former, while syntax accounts for the latter. One important property that differentiates, for instance, the derived nominal the/*John's growth of tomatoes from the corresponding sentence John grows tomatoes and the gerund John's growing tomatoes is the unavailability of a causative reading in the former. Chomsky's suggestion is that a feature [+cause] that lexically appears on the verb grow, but not on the nominal growth, could account for this difference. Much of the immediately following literature, especially in the eighties, adhered to the lexicalist view on derived nominals and used lexical rules to account for their apparent idiosyncratic properties as wordlevel formations. It culminated with Grimshaw's (1990) notable conclusion that the presence of argument structure brings a nominal closer to the corresponding verb and, implicitly, plays a crucial role in its interpretation.

88

Chapter Five

At the same time, it became obvious that the question of whether the nominal is derived from the sentence must be reformulated in terms of whether it denotes an eventuality like the base verb (Zucchi 1993). Better articulated syntactic approaches emerge after Baker's (1988) influential proposal to treat word formation as the result of morpheme incorporation, which follows the rules of syntax, in particular, head-tohead movement within the word. In addition, the fine-grained functional syntax of nominal phrases developed in Abney (1987) creates more room for a syntactic account of nominalization and a systematic parallelism between the verbal and the nominal domain. This new perspective makes it possible for nominalizations to be viewed as a mixture of verbal and nominal characteristics (i.e., mixed categories), without being forced to faithfully inherit all the clausal properties of the base verb. Marantz (1997) is the first to systematically re-formulate the syntactic analysis of derived nominals. Translating Chomsky's (1970) X-bar Theory approach into Bare Phrase Structure, Marantz proposes that an uncategorized root can be viewed as combining either with a verbal or a nominal functional head deriving a lexical verb or noun, respectively. If particular verbal properties are inherited by a nominal, its structure must include the corresponding verbal functional projections. For example, in Marantz's terms, the growth of tomatoes is a nominalization of the uncategorized root ¥GROW together with its internal argument tomatoes, while John's growing tomatoes is a nominalization of a causative vP, where v has already categorized the root and also made it causative. In this approach, some nominalizations inherit more verbal structure and others less, or none at all. This is the core idea that is pursued and occasionally adjusted by the various syntactic frameworks that are currently used in word formation: see, among others, Harley and Noyer (1998), van Hout & Roeper (1998), Alexiadou (2001), Fu, Roeper and Borer (2001), Borsley and Kornfilt (2000), Borer (2005) and much subsequent work. For more background on the development and the various forms of the lexicon vs. syntax debate in word formation, the reader is further referred to Carstairs-McCarthy (1992), Siloni (1997), Ackema and Neeleman (2004), Alexiadou (2010), Kornfilt and Whitman (2011), and references therein. As is evident from this chronological presentation, after Chomsky (1970), nominalizations become the test case for a theory of category change and word formation, a so-called "probe into the architecture of grammar" (Alexiadou 2010) or "the perfect prism through which to see modern grammar" (Roeper 2005: 125). There is hardly any work on category change that does not address nominalizations. This state of affairs

Category Change: Introduction

89

is faithfully reflected in the three papers on category change in this volume, which all deal at least partially with this topic. Alexiadou et al. and Roy and Soare are concerned with two issues that are at the core of the study of nominalizations: the similarities and the mismatches between derived nominals and the corresponding verbs in terms of argument realization and eventive interpretation. Fábregas addresses deadjectival nominalizations and denominal adjectives in an attempt to explain various nominal and adjectival suffix combinations that are allowed or disallowed in language. While the first two studies investigate the correlation between the form and the meaning of nominals from a syntactic and syntax-semantics perspective, the last one also touches upon issues at the morphology-syntax interface. The ultimate claims are all formulated in different variants of the syntactic approach to category change. Alexiadou et al. A mismatch between nominalizations and the corresponding verbs concerns the realization of the external argument. It has long been observed that even if certain morpho-syntactic patterns of nominals allow external arguments, this is less systematic than in the verbal domain (Grimshaw 1990, Pesetsky 1995, Marantz 1997, Harley & Noyer 2000, Sichel 2010). For example, while the -tion nominal in (1b) can realize an agent external argument, it disallows the non-agentive causer that is fully compatible with the verbal construction in (1a). (1)

a. The authorities/the approaching hurricane justified the abrupt evacuation of the inhabitants b. The justification of the abrupt evacuation (of the inhabitants) by the authorities/#the approaching hurricane

In a syntactic approach, one might explain the contrast in (1) by arguing that -tion nominals fail to project VoiceP, the layer that hosts the external argument according to Kratzer (1996), as in Alexiadou et al. (2009). However, the cross-linguistic picture that Alexiadou et al. (this volume) bring into discussion indicates that this cannot be the full answer. Looking at various patterns of nominalizations in English, German, Greek, Romanian, Spanish and French, they show that even nominalizations that have independently been argued to project Voice as, for instance, the Romanian supine nominal, the Spanish and the German nominalized infinitives, display this restriction. Similarly, in Greek, Romanian and Spanish external arguments in verbal passives are more restricted than in the active voice.

90

Chapter Five

Alexiadou et al. start by showing that Sichel's (2010) lexical-semantic analysis, according to which the nominal in (1b) expresses a simple event, unlike the verbal accomplishment in (1a), which is a complex event, cannot be on the right track. For instance, this account would predict that in some languages the passive of an active complex event becomes a simple event. Alexiadou et al. offer an account of the restriction in (1) and its cross-linguistic variation that is based on two claims. First, nominalization suffixes, like the verbal passive morphology, are exponents of VoiceP. Second, these suffixes exhibit selectional restrictions with respect to a verbal projection vP-EA that introduces the external argument and determines its thematic relation to the eventive head v. Thus, English passive Voice is compatible with all types of vP-EA's, but the -tion nominal Voice selects a vP-EA that specifies a particular thematic relation to the event. Alexiadou et al. support their analysis with evidence from Hebrew, whose verbal templates are restricted in the realization of external arguments and qualify as different morphological instances of the v-EA head. This account also has the advantage to spell out Grimshaw's (1990) longstanding intuition that nominalization resembles passivization. Roy and Soare. Another important question in the study of nominalization relates to the eventive interpretation: how does eventivity come about and how is it related to the eventivity of the base verb? This issue arises with Grimshaw's distinction between complex event nominals realizing argument structure, simple event nominals, and result nominals. Grimshaw (1990) observes that a nominal like examination has both an entity and an eventive reading, but when it realizes argument structure as in (2b) only the eventive reading remains. Verbal phrases like be on the table and take a long time are predicated of entities and events, respectively, thus distinguishing between the two readings in (2). (2)

a. The examination/exam was on the table. b. The {examination/*exam} of the patients {took a long time/*was on the table}.

Given this difference between argument supporting nominals (ASNs) in (2b) and referential/result nominals (RNs) in (2a), as they label them, Roy and Soare (this volume) further investigate to what extent we can speak of eventivity in simple event nominals (SENs) and participant -er nominals in English and French. They distinguish between a weak eventivity of a lexical/conceptual nature and a strong grammatical eventivity, on the basis of the possibility to combine with aspectual adverbials like for three hours. ASNs allow such adverbials, a hallmark for

Category Change: Introduction

91

grammatical eventivity, but SENs do not. This means that SENs, e.g., meeting, attack in (3a), which are also interpreted as eventive, only contribute conceptual eventivity, just like non-derived nouns with an event denotation as, e.g., concert, movie in (3b), which are also compatible with predicates of events. (3)

a. The meeting/attack (*for three hours) took a long time b. The concert/movie (*for three hours) took a long time

Within Borer's (2003, 2005) framework, Roy and Soare analyze ASNs as involving event structure, which is introduced by dedicated aspectual heads, similarly to verbs. SENs, like RNs, are nominalizations of an uncategorized root. The difference between SENs and RNs has a lexical semantic source: SENs are predicates of abstract entities (conceptual events), RNs refer to concrete ones (individuals). They further argue for a distinction in terms of eventivity among participant -er nominals: while eventive -er nominals like the constant defenders of human rights are taken to include aspectual structure similarly to ASNs, instrument -er nominals like grinder, blender have no event denotation and are derived from the root. Their account for the different patterns of nominalization in relation to the question of eventivity offers an insight into how derived nominals combine lexical and syntactic properties, an issue that has always been at the heart of linguistic research on category change. Fábregas. Besides the interaction between the syntax of nominalizations and their interpretation, a compelling challenge for a theory of category change, be it lexicalist or syntactic, is to explain why certain suffix combinations are possible in language (e.g., sens-ual-ize, sens-itive-ize), while others are not (e.g., *sens-uous-ize, *sens-ible-ize). A well-known study in this respect is Fabb (1988), which shows that the level-ordering analyses in Siegel (1974) and Kiparsky (1982) are not powerful enough for English, and proposes instead lexically encoded selectional restrictions for each suffix. A recurrent restriction Fabb mentions is against attachment to an already suffixed base, which affects ous, -ful, -hood and -ism, among others. In a similar fashion, Fábregas (this volume) pursues an account of adjectivizing and nominalizing suffixes in Spanish and their possibility to combine, for which he, however, employs a syntactic approach to word formation. Fábregas distinguishes between suffixes that only attach to roots (functional functors in Borer's (2003) terms) and those that attach to already suffixed roots. The former cannot change the sortal domain of their complement, but the latter do, that is, they trigger sortal transitions

92

Chapter Five

(in the sense of Ramchand & Svenonius 2012) between the nominal domain of kinds and the adjectival domain of qualities. The suffixes that Fabb identified as rejecting already suffixed bases fall in the first category, as exclusively attaching to roots: e.g., _]¥ ousA, _]¥ ismN. For Spanish, Fábregas identifies several classes of suffixes for which he offers a structural account that heavily relies on their semantics and the relation to the semantics of the base category. For instance, neither of the nominalizing suffixes -itud and -ura attach to already suffixed bases, yet only the former attaches directly to uncategorized roots: _]¥ itudN. Given that -ura involves a scalar property, which is typical of adjectives, Fábregas argues that it must attach to an already adjectivized root (_]¥ _]A uraN): cf. la altitud del faro 'the altitude of the lighthouse' vs. la altura del faro 'the height of the lighthouse'. In addition, the suffix -eza also contains degree information, which means that it attaches to a positive Degree projection (_]¥ _]A _]Deg ezaN): e.g., la ancheza de la calle 'the wideness of the street'. Affixes like the adjectivizer -oso, which attach to any nominalizer (e.g., optim-ist-oso 'optimistic-like'), are taken to instantiate prepositional relations between two nouns. This explains the fact that, like prepositions, they can combine with any nominal independently of how morphologically complex it is. With his approach, Fábregas introduces a variety of semantically and morphologically motivated syntactic categories that account for the possible suffix combinations on the path from a root to a deadjectival noun or to a denominal adjective. In conclusion, the three contributions on category change in this volume build on previous results in both lexicalist and syntactic approaches to word formation in order to explain the behavior of derived words. A common interest relates to the amount of semantic and syntactic properties that are retained from the base, which in syntactic terms translates into an inheritance of original functional structure. Idiosyncratic behavior is usually associated with root-level word formation, while further characteristics typical of the newly formed category are taken to be introduced by specific functional structure. As it becomes clear with each individual study, such an endeavor usually brings about a refinement of basic representations as it happens in the case of Voice and the realization of external arguments in Alexiadou et al, for eventivity in the nominal and the verbal domain in Roy and Soare, as well as for degree and scalarity in adjectives and nouns in Fábregas's contribution.

Category Change: Introduction

93

References Abney, S. 1987. The English Noun Phrase in its Sentential Aspect. Doctoral dissertation, MIT. Ackema, P. and Neeleman, A. D. 2004. Beyond Morphology; Interface Conditions on Word Formation. Studies in Theoretical Linguistics. Oxford: Oxford University Press. Alexiadou, A. 2010. Nominalizations: a probe into the architecture of grammar. Part I: the nominalization puzzle; Part II: the aspectual properties of nominalizations, and the lexicon vs. syntax debate. Language and Linguistics Compass 4, 496-523. —. 2001. Functional structure in nominals: nominalization, and ergativity. Amsterdam: John Benjamins. Alexiadou, A., E. Anagnostopoulou, and F. Schäfer. 2009. PP licensing in nominalizations. Proceedings of NELS 38. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: The University of Chicago Press. Borer, H. 2005. Structuring Sense. Vol. 1: In Name Only. Oxford University Press. —. 2003. Exo-skeletal vs. endo-skeletal explanations. In The Nature of Explanation in Linguistic Theory, ed. by J. Moore and M. Polinsky. Chicago: CSLI and University of Chicago Press. Borsley, R. and Kornfilt, J. 2000. Mixed extended projections. In Syntax and Semantics 22: The Nature and Function of Syntactic Categories, ed. by R. Borsley, 101–131. Academic Press, New York. Carstairs-McCarthy, A. 1992. Current Morphology. (Linguistic Theory Guides). London: Routledge. Chomsky, N. 1970. Remarks on nominalizations. In Readings in English Transformational Grammar, ed. by R. Jacobs and P. Rosenbaum. Waltham, MA: Ginn. Fabb, N. 1988. English suffixation as constrained only by selectional restrictions. Natural Language and Linguistic Theory 6.4: 527-539. Fu, J., T. Roeper and H. Borer. 2001. The VP within nominalizations: evidence from adverbs and the VP anaphor do-so. Natural Language and Linguistic Theory 19.3: 549-582. Grimshaw, J. 1990. Argument Structure. Cambridge, MA: MIT Press. Harley, H. and R. Noyer. 2000. Formal versus encyclopedic properties of vocabulary: Evidence from nominalisations. In The Lexicon/ Encyclopedia Interface, ed. by B. Peeters, 349-374. Amsterdam: Elsevier Press.

94

Chapter Five

Harley, H. and R. Noyer. 1998. Mixed nominalizations, object shift and short verb movement in English. In Proceedings of NELS 28, ed. by K. Kusumoto and P. Tamanji, 143-157. University of Massachusetts at Amherst: GLSA. van Hout, A. and T. Roeper. 1998. Events and aspectual structure in derivational morphology. MIT Working Papers in Linguistics 32: 175220. Kiparsky, P. 1982. Lexical morphology and phonology. In Linguistics in the Morning Calm, ed. by I.-S. Yang, The Linguistic Society of Korea, 3-91. Seoul: Hanshin Publishing. Kornfilt, J. and J. Whitman. 2011. Nominalizations in Linguistic Theory. Special issue of Lingua 121.7: 1159-1314. Kratzer, A. 1996. Severing the external argument from its verb. In Phrase Structure and the Lexicon, ed. by J. Rooryck and L. Zaring, 109-137. Dordrecht, Kluwer. Lakoff, G. 1965. On the Nature of Syntactic Irregularity. Doctoral dissertation. Indiana University, Bloomington. Lees, R. B. 1960. The Grammar of English Nominalizations. The Hague: Mouton. Marantz, A. 1997. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In University of Pennsylvania Working Papers in Linguistics Vol. 4.2, ed. by A. Dimitriadis, L. Siegel, C. Surek-Clark and A. Williams, 201-225. Philadelphia. Pesetsky, D. 1995. Zero Syntax: Experiencers and Cascades. Cambridge: MIT Press. Ramchand, G. and P. Svenonius. 2012. Advanced cartography: Charting elevation. Ms., Tromsø: University of Tromsø. Roeper, T. 2005. Chomsky’s Remarks and the transformationalist hypothesis. In Handbook of Word-Formation, ed. by P. Štekauer and R. Lieber, 125-46. Dordrecht: Springer. Sichel, I. 2010. Event structure constraints in nominalization. In The Syntax of Nominalizations across Languages and Frameworks, ed. by A. Alexiadou and M. Rathert, 151-190. Berlin: Mouton. Siegel, D. 1974. Topics in English Morphology. Doctoral dissertation. Cambridge (Mass.): MIT. Siloni, T. 1997. Noun Phrases and Nominalizations: The Syntax of DPs. Dordrecht: Kluwer Academic Publishers. Zucchi, A. 1993. The Language of Propositions and Events: Issues in the Syntax and Semantics of Nominalization. Dordrecht: Kluwer Academic Publishers.

CHAPTER SIX TOWARDS A SYNTACTIC ACCOUNT OF AFFIX COMBINATIONS: FROM NOUNS TO ADJECTIVES AND VICE VERSA ANTONIO FÁBREGAS UNIVERSITY OF TROMSØ

1. Affix combinations: proposals and principles1,2 Since the work of Fabb (1988), it is a well-known fact that affixes in a language cannot combine in any imaginable way. The 45 English suffixes that Fabb considers in his study should produce 1849 suffix combinations. Eliminating sequences that could be out by categorial selection, the number drops to 614. Once we ignore the combinations that would be impossible given Lexical Strata theory (Siegel 1974, Kiparsky 1982), we should have 459, but the real attested number is around 50 (Fabb 1988: 528-530). This is something that has to be accounted for in any grammatical theory. Besides the descriptive necessity of properly characterising grammars, the problem brings up relevant issues for the theory of language. Potentially, the absence of an affix combination could inform us of the principles that languages use to build the internal structure or words—perhaps, more generally, of principles regulating the syntactic relation between grammatical categories—. This could, in turn, give us some information about what place word formation and morphology occupy in the architecture of language. The goal of this paper is to approach these questions through the study of affix restrictions inside one language (modern Spanish), concentrating on only a subset of the processes that could be approached. For reasons of space, we will only address affixes that turn adjectives into nouns and vice

Chapter Six

96

versa. Two reasons justify this choice. First, these combinations are relatively understudied, so this article—even if its conclusions are not confirmed by further research—can independently contribute to the description of a fragment of the grammar of Spanish. Secondly, nouns and adjectives are two closely related categories which share many properties in Spanish—gender, number, occurrence with copulative verbs…—, so it is more doubtful that the impossibility of having some combination might be due to conceptual or extralinguistic reasons. We independently know that nouns and adjectives can combine with each other, so the absence of some sequence of affixes in this domain must have different causes.

1.1. Four takes on affix combination restrictions As far as we see, the explanations about the impossibility of having a certain affix combination can fall into four groups. (a) Starting with Fabb’s initial suggestion to explain the massive absence of expected combinations, IDIOSYNCRATIC MORPHOLOGICAL RESTRICTIONS in the lexical entry of individual affixes. Fabb identified a number of affixes that cannot combine with any already suffixed base, among them (in English) -ous (ibidem, 532). He proposes a morphological constraint preventing these suffixes from appearing in the environment [word-suffix- ]. Note that such restriction would have to be associated to the lexical entry of the affix, it would be idiosyncratic—the previous suffixes would have it, but not -y—and it would be part of the purely morphological information, as it is stated it is independent of phonological, semantic or categorial constraints, which are not part of the context description. In more modern terms, but perhaps not twisting too much the spirit of Fabb’s proposal, we can imagine that their lexical entry (1), states that they must appear to the right of a root—on the assumption that roots cannot be recursively defined, contra Giegerich (1999)—. (1)

a. ]¥ fulA b. ]¥ ismN c. ]¥ ousA

(b) Another potential account of affix ordering which also treats the problem from a morphological perspective is the use of TEMPLATES, that is, pre-established morphological orderings that group affixes into classes, and states the position of these classes. Templates have been used in the study of Swahili (Schadeberg 1984) and other languages (like Nimboran;

Towards a Syntactic Account of Affix Combinations

97

Inkelas 1993). This kind of analysis requires that a non-violable order can be established between them, such that no affix from class N can follow an affix from class N+1 but can be followed by it. This approach has mainly been applied to inflection (see Table 1), but nothing in principle prevents researchers from applying it to derivation. 1 Negative ha-

2 Subject agr.

3 Neg. si-

4 T/ mood

5 Relative agr.

6 ku-

7 Object agr.

stem

8 plural –ni

Table 1. The Swahili verb template (adapted from Schadenberg 1984: 14) (c) A SYNTACTIC SOLUTION that instead of using idiosyncratic properties of the morphological exponents presents structural reasons for affixes to appear in designated positions inside a sequence, defined by the nature of the objects they select and the kind of information they project as a phrase. In this account, the explanation of the (un)attested sequences involves combining two different properties of structures independently needed in any grammatical theory: on the one hand, the cross-linguistic evidence that syntactic projections are ordered in some way by virtue of their selecting different kinds of objects—subcategorisation—; on the other hand, the proposal that these projections are spelled out by morphophonological items that are associated to sets of syntactic features. This account would explain the combinations of affixes by reference to the structures underlying them, without any intervention of morphologically imposed orders. A sequence of affixes could in principle be represented as in (2), if we motivate the order between the projections (here, LI stands for “lexical item”). In (2) we introduce one item per head; as explained in §2, we will assume a system where any syntactic constituent can be lexicalised by a single item. (2)

WP W LI4

XP X LI3

YP Y LI2

ZP Z LI1

¥ root LI

98

Chapter Six

(d) A PARSING / COMPLEXITY-BASED ORDERING along the lines of Hay (2002) and Hay and Plag (2004). In processing, some affixes are parsed as forming a tighter unit with the base, and tend not to be segmented—due to frequency of single words and patterns, semantic compositionality and phonological transparency, inter alia—, while others are parsed as separate from the base. On the assumption that speakers do not want to break combinations that are felt as a unit (perhaps related to the Atom Condition in Williams 1981), affixes that are more difficult to parse and therefore integrate with the base cannot follow those that are parsed separately. For instance, the English suffix -th is parsed as part of the word, given its low productivity and other properties, while -ness is not, given high productivity. The first suffix can be followed by -less, whose high productivity motivates that it is also treated as non-integrating (3a). However, -th cannot follow -less (3b), but -ness can (3c). The ordering base > non integrating > integrating is thus disallowed. (3)

a. dep-th-less, bir-th-less, tru-th-less… b. *aim-less-th c. aim-less-ness, self-less-ness, piti-less-ness…

The study that follows can be seen as a way to test these accounts under the light of the attested data, and to advance the preliminaries of a syntactic account. This analysis, as we will argue, has the advantage, with respect to the morphological account, that it relies on independent properties of syntax and spell out rather than on idiosyncratic properties of individual lexical items. The result will be a sketch of what can be the typology of possible and impossible noun/adjective relations in Spanish, even though space restrictions will not allow us to get into many interesting details about single affixes.

2. Assumptions about spell out and cartography As they will play a role in our account of terminal suffixes, let us briefly make explicit our assumptions about how syntax is materialised by exponents. We assume a late insertion account (Halle and Marantz 1993) where single exponents can lexicalise the information associated to terminal nodes, but also phrasal projections (Phrasal Spell Out, Caha 2009, Fábregas 2009). In (4), depending on what exponents are available given the syntactic feature, the features can materialise analytically—an exponent per head—or synthetically—one exponent for the whole—. In

Towards a Syntactic Account of Affix Combinations

99

the second case, the exponent is introduced in the phrasal projection and lexicalises all features dominated by that node. (4)

XP

{X, {X, {Y, {Y, Z}}}} Å Exponent A’

X YP {Y, {Y, Z}} {X} Exponent C Y Z {Y} {Z} Exponent B Exponent A We also assume the Elsewhere Principle (Kiparsky 1973): given a particular set of morphosyntactic features, the exponent that is the closest match to that set beats all other competitors, with more specific forms beating all other forms. One important point is what counts as the “closest match” to a set of features when there is no perfect correspondence to an exponent. The Subset Principle (Halle and Marantz 1993) states that the closest match is the exponent that has a maximum number of those features, but no extra features. The Superset Principle (Caha 2009) proposes that the closest match is the exponent that has all represented features, and a minimum of extra features. Here we assume the Superset Principle, as we consider that (a) there is a fair deal of evidence to support it and (b) it does not require extra operations where syntactic features are erased from the representation. Given the set of features in (5), and the exponents in (6), then, the Superset Principle (combined with the elsewhere principle) predicts that B will be used in spell out. The subset would predict insertion of C after the feature X has been somehow made invisible on the way from syntax to vocabulary insertion. (5) Syntax = {X, {Y{Y, Z}}} (6)

a. A {V{V, {W{W, {X{X, {Y{Y, Z}}}}}}}} b. B {W{W, {X{X, {Y{Y, Z}}}}}} c. C {Y{Y, Z}}

100

Chapter Six

2.1. Non-templatic cartography During the last 20 years, a great deal of work has been devoted to identifying a fine-grained structure of functional projections inside domains that used to be viewed as atomic (Rizzi 1997 for the C domain, Cinque 1999 for the middlefield between T and V, Cinque 2010 for adjectives, Beghelli and Stowell 1997 for quantifiers, Zamparelli 2000 for D, etc.). This has created a tension between an adequate description of complex patterns of data and the necessary parsimony of a scientific explanation: a fine-grained Functional sequence (Fseq), with a long series of rigidly ordered projections, seems to be necessary to account for different kinds of facts, but such variety of heads causes trouble for acquisition and imposes extra conditions on merge, which would be constrained by some kind of template dictating the order in which pieces in a numeration are combined with each other. Ramchand and Svenonius (2012) make a proposal to resolve this tension. In their approach, the Fseq is divided into sortal domains that are determined by the interface with general cognition. For instance, the fact that the C domain is over the Tense-Aspect domain, and that is, in turn, over the V domain is not due to properties of narrow syntax, but guided by cognition: propositions—corresponding to C—are built over situations— corresponding to Tense-Aspect—built over eventualities—corresponding to V—. Consider the well-known contrast in (7). (7)

a. John has been eating. b. *John is having eaten.

A templatic cartographic approach might want to explain this contrast by saying that the Perfect head is ordered in the Fseq over the Progressive head—and thus, (7b) would be banned by a condition imposed on merge itself—. Ramchand and Svenonius explain it, however, through a sortal distinction. The idea is that the Progressive head selects for an element belonging to the eventuality domain, while the Perfect head selects for elements belonging to the situation domain. Between the perfect and the progressive, therefore, we have had a transition from one domain to the other. In the simplified schema of (8), the transition is labelled as Asp.

Towards a Syntactic Account of Affix Combinations

(8)

101

T situation, domain of sort s T

Asp Asp

transition: ܑe. R(s, e) V event, domain of sort e

Some projections, thus, would have as their role to allow a transition between one sortal domain and another. The contrast in (7) suggests that the progressive is one form of transitioning between eventualities and situations, but we expect that there would be other heads with that power. Specifically, lexical categorisers would also be transitioners. At this point, the distinction established by Borer (2003) between lexical and functional functors becomes relevant. In Borer’s system, lexical functors (N, V, A) have the power to redefine the category label of the structure they embed. We suggest that these lexical functors have the role of changing the sortal domain of the derivation. In contrast, functional functors (Past, Class, Degree, etc.) do not have that power, and as such they must always combine with members of a particular sortal domain. Inside sortal domains, merge is not rigidly ordered—that is, there is no Fseq imposing a particular order of merge—. Different orderings are possible, and get interpreted at LF in different ways, which have to do with the internal complexity of the set with which an element is merged. Therefore, against Cinque (1999), for Ramchand and Svenonius there would not be two positions (Asp-repetitive and Asp-frequentative) for the two readings of twice in (9); the scope difference would be derived because in each case twice merges with different structures. (9)

a. John twice intentionally knocked on the door (two separate acts of knocking). b. John intentionally knocked twice on the door (one act of knocking, with a repetition) Given this background, let’s move to the data and their analysis.

3. Data collection and empirical generalisations The empirical base of our study relies heavily on data collection, and as such the first concern has to be to make our methodology explicit, as the validity of our conclusions depends on that. The first part of the empirical data collection involved the consultation of several inverse dictionaries, among them the Royal Academy Dictionary,

Chapter Six

102

and other electronic resources. This gave result to a first set of data, where the endings of specific forms that could correspond to affix combinations were examined. The results were then processed manually to eliminate sequences that had the phonological identity of a sequence of affixes but did not correspond to those. For instance, introducing the sequence -ilitud in the database gave a few results, that upon examination were determined not to correspond to -il+itud (10). (10)

disimil-itud unlike-itude “Property of being dissimilar”

Once the form -itud is segmented, the base is the adjective disímil “unlike”. This form does not contain the adjectiviser -il. In clear examples derived with this adjectiviser (infante “child” > infant-il /infant-íl/ “childish”), stress is attracted by the affix, but not in this case (/disímil/, not */disimíl/). Another criterion has been that the form without the sequence that could be an affix must be a base in contemporary Spanish. The suffix -ura cannot follow the adjectiviser -oso in contemporary Spanish. If one introduces the sequence -osura in the inverse dictionaries, one obtains one word (11). (11)

hermos-ura beautiful-nom “Beauty”

This form comes from the adjective hermoso “beautiful”, which prima facie contains the suffix -oso. This is indeed the historical origin of the word. However, the alleged base obtained once we substract the sequence, herm-, does not produce nouns, verbs or other adjectives. We thus conclude that the ending is a phonological sequence that does not correspond to any affix inside the word. It is well-known that corpora give a taste of what is possible in a language, but from there we cannot be sure of what combinations are impossible, as these might have not showed up in the corpus by accident. As a result of that, in a second part of the data collection we presented native speakers, mainly of the European Spanish variety, with words containing the relevant sequences of adjectives (12), so that they could rate whether the form sounded like a possible word of Spanish or not given a plausible meaning—not expressed, to the best of our knowledge, by other

Towards a Syntactic Account of Affix Combinations

103

Spanish words—and a context of use. The form in (12a)—with itud+osowas considered acceptable given the context where it describes a person that points out inaccuracies all the time. In contrast, the form in (12b) was rejected; it was supposed to refer to the quality of supporting Real Madrid—the word in -ista means “that supports Real Madrid”—.3 (12)

a. in-exact-itud-oso in-exact-itude-ful “that points out inaccuracies” b. *madrid-ist-ura madrid-ist-ness “quality of supporting Real Madrid”

3.1. Empirical generalisations After classifying the corpus obtained by the previous methodology, we arranged it as shown in Table 2. In the rows we introduced the first affix of the sequence; in the columns, the second. As can be seen, here we ignore data about combinations of affixes of the same kind in a sequence. Such combinations are sometimes possible; however, here we are interested in the relation between nouns and adjectives. Note that rows and columns are divided in two blocks: in the first column, the first block (from -al to iano) represents the adjectivisers; the second block, from -dad to -dumbre) are the nominalisers. The first row reproduces the same divide. From a general consideration of the table, several kinds of affixes can be identified: (i) A fair amount of adjectivisers and nominalisers cannot follow any other adjectiviser or nominaliser. These are those in (13), where (13a) lists the adjectivisers and (13b), the nominalisers. We call these affixes INTERNAL AFFIXES. Fabb (1988) also identified a number of them in his study on English. (13)

a. -iano, -ivo, -ero b. -or, -arioN, -dumbre, -ero, -itud

(ii) The adjectiviser in (14) can follow all nominalisers considered here. We will call it EXTERNAL ADJECTIVISER.

Chapter Six

104

(14)

-oso

(iii) The nominalisers in (15) can follow -oso (and combine with all or almost all other adjectivisers, with the general exception of -ista). Idad is the most general of them, and -ía is the most restrictive. We will call these four affixes EXTERNAL NOMINALISERS. (15)

-idad, -ismo, -ería, -ía

(iv) A few affixes cannot be followed by any affix of the opposed grammatical category, or almost none. The case of -ista is the most clear: it cannot be followed by any nominal suffix.4 Among the nominalisers, dumbre can only be followed by -oso. Following the standard morphological terminology, we will call these TERMINAL AFFIXES, with the caveat that the term is not completely accurate. 2nd

a l

1st

Al Ario Ero Oso Ivo Il Esco Ento Ista Ico Iano Dad Ez Ía Ismo Ería Ura Itud Ero Or Ario dum -bre

a r i o

e r o

o s o

i v o

I l

e s c o

e n t o

i s t a

i c o

Not considered in the study

x x

x x

x x

x

x x x x x x x x x x x

x

x

x

x

x

x

x

i a n o

d a d

e z a

í i a s m o

x x x x x x x x

x x x

x

x x

x x

x X

Table 2. (Un)attested suffix combinations

x x x x

x

x x x x x x

x

x x

x x

e r í a

u r a

i t u d

e r o

o a d r r u i m o b r e

x

x x

x

Not considered in the study

Towards a Syntactic Account of Affix Combinations

105

Additionally, we can see that affixes that are sometimes described as being identical or almost identical have different combinatorial restrictions. Four adjectives are used to form what descriptive grammars (eg. Rainer 1999) call quality nominalisations: -itud, -ura, -eza and -idad. However, -itud cannot follow any adjectiviser; -idad can follow almost all of them; -ura can follow the adjectivisers that we called Internal affixes (13a); -ez(a) can follow these, plus -il, -al, -iento, -esco and -ario. The table 3 represents an attempt to arrange the affixes in a sequence: in A, we include affixes that do not follow any other affix of the different category; B shows those that can follow at least some of those in A; C, those that can follow at least some in B, etc. -oso is represented twice to indicate it can follow or be followed by those in F. Root Adj. ¥

Nom

A -iano, ivo, -ero

B -il

-or, dumbre, ero, itud

ura

C -al, iento, ista 5 , esco, ario

-

D ico

eza

E oso

F

E’ oso

-idad, ismo, ería, -ía

Table 3. The affixes in a sequence

3.2. The data are not compatible with two theories The generalisations argue against the templatic solution, as the data show two unexpected phenomena for this theory. First of all, the ordering between -oso and the nominalisers in F is free: they can follow or precede each other (16). (16)

a. mal-dad-oso evil-ness-ful “evil” b. aparat-os-idad machine-ful-ness “complication”

The data also make the psycholinguistic approach unlikely. Affixes in A can be analysed as parsed with the base and perhaps those in C could

Chapter Six

106

already be considered parsed independently, but what would be the status of, for instance, those in D that can follow unparsed suffixes but not follow a big number of other parsed suffixes? In the absence of explanations that make the complexity-based approach more compatible with a gradient understanding of the parsed/unparsed divide—and provide researches with tools to measure it—the approach, at least how it is stated now, cannot account for the data. If the development of more fine-grained tests for parsing and statistical tools refine it in the future, then it would plausibly be a serious competitor to the remaining theories.

4. Affixes as the spell out of syntactic projections: a preliminary analysis In this section we will study some of the properties that have emerged from the previous table and propose a syntactic explanation; reasons of space will not allow us to concentrate on many of the contrasts. Here we will limit ourselves to sketching the general guidelines of a complete syntactic analysis of these affix combinations.

4.1. Internal affixes are grammatical functors; the others include a transitioner The first divide that seems to be relevant is the one between the affixes that must be adjacent to the root and those that do not. Their behaviour is expected if only some affixes define a transition from the domain of kinds to the domain of qualities or vice versa. Those that only combine with the root—which lacks grammatical category (Marantz 1997)—but do not appear after nominalisers or adjectivisers, are those that cannot change the sortal domain of their complement. (17a) represents the structure of an adjectiviser that can follow nominal suffixes; it produces a transition between the domain of kinds (K) and the domain of qualities (Q); (17b) shows the opposite case. (17)

a.

AP

ܑQ.R(Q,K)

N

...NP ܑK.R(K, ¥) N -eza ¥

A -il

b.

NP

ܑK.R(K, Q) ...AP ܑQ.R(Q, ¥)

A

¥

In contrast, the affixes that must be adjacent to the root are those which cannot redefine the sortal domain. They would be equivalent to Borer’s

Towards a Syntactic Account of Affix Combinations

107

(2003) functional functors. To represent that they are functional, we use the notation n and a. (18)

a.

nP n {-or, -ario}

b. ¥

aP a ¥ {-ian(o), -iv(o)...}

Consider additional evidence for this proposal, taken from apparent verb > noun derivations. The suffix -or, when it builds nouns from seemingly verbal bases (19), requires the theme vowel (-a, -e) to disappear from the base. This is not the case with -ción, -miento or other verbalisers, which can appear after the theme vowel (20). This would again show that or cannot redefine the domain to which its complement belongs, so the verbalisers have to be removed in order for it to define a noun. (19) (20)

am-a “to love” > am-or “love”; tem-e “to fear” > tem-or “fear”; sab-e “to taste” > sab-or “taste”; ol-e “to smell” > ol-or “smell” actu-a “act” > actu-a-ción “acting”; desplaz-a “displace” > desplaza-miento “displacement”

4.2. External nominalisers as introducing qualia structure Consider now the nominalisers that can follow any adjectiviser. One property that all these nominalisers have in common is in their semantics: in addition to turning the adjective into a noun, they contribute specific meaning notions that can be systematised. We will shortly argue that their semantics is related to what Pustejovsky (1995) calls the qualia structure. The suffix -ía can carry an eventive semantics (“action that displays the property A”), as in osado “brave” > osad-ía “action that shows bravery”. It can be argued that this is related to the agent quale, which denotes the origin of an entity. The base is taken as the origin of an implicit action denoted by the whole word: the property of being brave causes an action showing bravery. In other cases, -ía denotes the science that studies the matters codified by the base, as in all words ending in -logía “logy”, or econom-ía “economy”, and many others economical matters give rise to the science called “economy”. In such cases the notion also corresponds to a qualia: the constitutive qualia, which denotes the matter something is about. In contrast, the suffix -ería is specialised for naming actions related to the base or places where the action is performed (reloj “watch” > reloj-ería “watchmaking” or “watchmaker’s shop”). This seems

Chapter Six

108

to be related to the telic quale: the base denotes an object that defines the goal or purpose of a certain action. As for -ismo, it is specialised in expressing ideologies, movements or arts defined by some property of the base: surreal “surreal” > surreal-ismo “surrealism”, but also activities (cicl-ismo “cycling”). This, again, could relate to the telic quale: the base defines a notion that defines some activity involving that entity. Finally, even though the suffix -idad is traditionally listed as a quality nominaliser, upon further examination, it seems that it is better characterised as a constitutive quale suffix—cristiano “Christian” > cristian-dad “Christianity, group of Christians”—. The same account can be extended to apparent quality nominalisations. Words like generos-idad “generosity” (from generoso “generous”) can be analysed as the entity that contains the quality expressed by the base: “that has generosity”. In that case we would be able to unify the semantic contribution of two apparently separate uses of -idad under the same notion. Our proposal to account for these affixes is dependent on a partial syntactisation of the concept of qualia. We argue that the quale is defined in a designated syntactic head, as shown in (21). This head dominates NP and introduces additional properties of the kind in its relation to events, constitutive parts, etc. We assume that the telic, agentive and constitutive quale are not expressed by different heads, but that the one head represented in (21) is interpreted in any of these ways depending on conceptual semantics. This allows that -ía is sometimes interpreted as the constitutive quale and sometimes as the agentive one. (21)

QuaP Qua

...NP N

...

One reviewer asks the question of how do we know that QuaP is higher than NP. Our argument for its needs—as well as for the convenience of codifying qualia as a syntactic head—is syntactic selection. It is well known that some verbs select for CPs—as they imply operations over propositions—, but also for some NPs (22). Not any NP can be selected in such cases, only those that contain some information as part of their constitutive quale. If we represent the (constitutive) quale as a head over NP, this property follows.

Towards a Syntactic Account of Affix Combinations

(22)

109

a. John knows that John is sick. b. John knows {the menu / the answer / *the rock / *the wind}.

Given this, we propose that these external nominalisers are the spell out of a series of heads whose lowest element is NP and whose highest element is QuaP. Given that the material they spell out includes a transitioner, they are expected to follow any adjectiviser. (23)

QuaP Qua

...NP N

...

We have also affixes that are the materialisation of QuaP without the transitioner NP. The crucial case is -(d)umbre. This affix is internal—only combines with roots—and can form collective nouns, so it can codify a (constitutive) quale: tech-umbre “roof-umbre, set of structures that conform the roof”. Another question is why most adjectivisers cannot follow these nominalisers. In order to explain this property, let us take a look to the ordering between the sortal domains in the DP structure. Cinque (2005) argues convincingly that the order is Dem(ostrative)-Num(eral)A(djective)-N(oun). Independently, we know that prepositions are higher than demonstratives, which gives us a complete ordering P-Dem-Num-AN. Assume, now, that the projection Qua is not structurally placed inside the domain of N, but at a higher sortal domain; this level could be the one to which Num belongs. Evidence for that comes from cases where a single noun does not contain quale structure, but the combination of the noun with some adjectival or prepositional modifier does. There are plenty of examples of this. For instance, the word agua does not have a quale, as evidenced by the impossibility of having it as the direct object of the verb empezar “begin” (24a) unlike other nouns—in order to have a grammatical reading, a verb has to be added (24b)—. However, the combination of agua with angélica “angelic” gives a structure that has a quale (telic: it is a liquid used in traditional medicine to take a purge), as evidenced by (24c). (24)

a. Empezó {*el agua / la sopa}. started.he the water / the soup Intended: “He started (to drink / to prepare) {the water / the soup}”

110

Chapter Six

b. Empezó a beber el agua. started.he to drink the water “He started to drink the water” c. Empezó el agua angélica. started.he the water angelic “He started {drinking / preparing} the angelic water” Given this kind of evidence, the proposal that QuaP is higher than N is at least possible. Assume this is right: then, the adjectivisers that cannot follow these nominalisers that contain quale inside their entry would be unable to do so because Qua defines a structural height which cannot be dominated by members of the domain A, as this would imply the ordering Dem-A-Num. Additional evidence that nominalisers that contain Qua occupy a high syntactic position which defines a domain higher than A comes from the fact that they are the only affixes that can embed an adjective which contains degree. From the comparative mejor “better”, we can get mejor-ía “improvement”; the nominaliser -idad sometimes expresses a particular degree of a property, as in crimin-al-idad “crimin-al-ity, the amount of crimes that take place in an area”. These facts suggest that these affixes can be introduced on top of a structure where the adjectival domain has been functionally expanded; in our account, this is allowed by the nature of the head Qua, that they contain. Note that -oso can follow these nominalisers. The inescapable conclusion, following our line of reasoning, is that -oso is not an adjectiviser, but a head belonging to Dem or P. This will be argued for in 4.3.

4.3. The external adjectiviser as a preposition The adjectiviser -oso can follow any nominaliser. This property is correlated by the fact that it has an extremely underspecified semantics which allows for a wide array of readings. Some adjectives derived with oso have a possessive semantics (25a), some others express the property of producing the noun in the base (25b), while others express the property of staying in the state denoted by the base (25c), among many other readings (Martín García 2007). (25) a. arena “sand” > aren-oso “with sand” b. angustia “stress” > angusti-oso “that causes stress” c. miedo “fear” > mied-oso “that typically experiences fear”

Towards a Syntactic Account of Affix Combinations

111

Other adjectival suffixes have better defined meanings; -esco is specialised in expressing a similative relation (“similar to N”, as in caballer-esco “knight-esco, knight-like”), -ario gives an agentive semantics (“that does something with N”, as in partid-ario “party-ario, that defends a particular political party”), etc. Given its wide array of meanings, and if we want to avoid homophony, the best shot is to associate -oso to an underspecified relational function that associates two entities with each other in a way which is specified inside the conceptual context. One of the entities would be the noun on the base (X) and the other one would be the entity that holds those properties. This second entity would be eventually materialised as the subject of the adjective. (26)

RP DP

R R -oso

X

The structure of (26) is that of a prepositional head, as it is the category that expresses a relation between two entities (Hale and Keyser 2002). Prepositions are used to relate two nouns with each other, and they can take as their complements a variety of noun phrases, including bare NPs when they denote properties rather than arguments (coffee with milk). Like adjectives, they can also be gradable (too inside onself), so they share many surface properties with them. This would account for the fact that -oso is the only adjectiviser that can follow any nominaliser, as prepositions can embed NPs of different structural complexity. Being higher than A and Num, they can also follow -idad, -ería, -ía and -ismo. Why can -idad, -ería, -ía and -ismo follow -oso? If this affix is P, we should perhaps expect the same kind of mismatch between the sortal domains that made impossible for some adjectivisers to follow -ismo or idad. The solution to this problem comes from the fact that prepositions are also combinable with qualia structure. The preposition por “by, through” can have an agentive quale meaning in some syntactic contexts— such as passive constructions—, but not always. This suggests that por is in such cases dominated by QuaP. If this is so, these nominalisers could follow -oso in the same way that QuaP can dominate a preposition like por to add to its locative meaning an agentive sense.

Chapter Six

112

(27)

QuaP Qua

PP P por

...

The suffix -oso is not the only affix that could be analysed as a prepositional structure. The same kind of semantic underspecification is exhibited by many of the adjectivisers that produce relational adjectives such as -ico (sulfúr-ico “sulfuric”) and -iano (ital-iano “Italian”), which have been analysed as prepositional by others (Bosque 2006 labels them as case markers). Clearly, some difference must exist between these affixes and -oso that explains why only the later can follow any nominaliser. We admit that the proposal still needs to be worked out in essential ways, but here is a suggestion: if, following Svenonius (2010), we divide prepositions in a lexical and a functional layer, we could treat -oso as the spell out of a structure that includes a lexical P—therefore, a transitioner— , but the rest as functional Ps (pP), in such a way that they do not contain a projection to transition from nouns to the prepositional domain. Lacking a lexical functor, they would appear in an internal position.6

4.4. -ista Let us now consider -ista, which cannot be followed by any nominaliser. Given that this suffix can follow overt nominalisers like -or, it must contain a lexical categoriser. An examination of its semantics shows that it also incorporates qualia semantics. Adjectives derived with ista have a telic quale: they are used to express the property of working or displaying some activity involving the base noun (28). This restricts the conceptual interpretation of its bases to entities that can play a function in an event. Note, for instance, that the proper name Madrid has to be interpreted not as the city, but as the soccer club when derived with -ista. A person who is simpl-ista “simplistic” is someone who acts in a way that causes simplicity, or someone who is in favour of simplicity. (28)

simple “simple” > simpl-ista “simplistic”; Madrid “Madrid” > madrid-ista “Real Madrid supporter”

These data suggest a structure like (29) for the projections to which the affix is associated.

Towards a Syntactic Account of Affix Combinations

(29)

113

QuaP Qua

AP A

...

Note an interesting consequence of this lexical entry combined with the Superset Principle: -ista should be able to appear with adjectival bases, as it adds an additional layer and its base is an adjectival categoriser. This is what happens in simpl-ista, as simple “simple” is only an adjective in Spanish (30a). In such cases, -ista lexicalises the higher layers and simple is shrunk, by the Superset Principle, to the root, as illustrated in the schema (30b). (30)

a.

AP

A

b. ¥

simple

QuaP Qua

-ista

AP A

¥ simpl-

Consider now why -ista cannot be dominated by any nominaliser. Only those that include N as its lower projection can be considered, because their complement is already categorised. Any nominaliser would need to have N, then, dominating Qualia in the structure (31). However, this would produce an ordering in which Qualia, which belongs to the Num domain, is dominated by N, giving the ordering N > Num > A, which is impossible. Consequently, -ista is correctly predicted to disallow further nominalisation. (31)

*NP N

QuaP Qua

A

However, as noted in endnote 3, -ista is not a terminal affix once we consider also its combination with other adjectivisers. The sequence -ístico is attested (eg. futbol-íst-ico “football-ist-ico, football-related”). Why is this possible? Assume that -ico is the spell out of a functional P layer, as we have proposed. An order in which P dominates Num conforms with the

Chapter Six

114

sortal domains hierarchy P > D > Num > A > N, so such combination is actually expected. As for why p can dominate a structure defined as an adjective, see endnote 5 for some comments. (32)

pP p -ico

QuaP Qua

-ista

AP A

¥ futbol-

Remember that we have analyzed -oso as involving at least a lexical P in its structure. This predicts that -ista can be followed by this suffix, because PP can dominate QuaP; data like (33), found in google, confirm this—remember we did not represent them on the table because we did not consider series of several affixes of the same category. (33)

a. pesim-ist-oso worst-ist-oso “pesimistic-like” b. optim-ist-oso best-ist-oso “optimistic-like”

4.5. Fine-grained differences between similar affixes: an example Different nominalisers and adjectivisers occupy different positions, and a syntactic analysis must find a reason for this beyond idiosyncratic and accidental properties. Let us illustrate the kind of analysis that would be needed in a syntactic view through the three quality nominalisers -itud, ura and -ez(a). Our proposal is that they differ on the size of the element they combine with. In the case of the -itud, it cannot follow another affix; we analyze it as an nP that directly selects a root.

Towards a Syntactic Account of Affix Combinations

(34)

115

nP

n -itud

¥

The prediction is that quality nouns with -itud will not contain characteristics of adjectives. Preliminary evidence that this might be on the right track comes from minimal pairs where -ura and -itud alternate: (35)

a. alto “tall, high” > alt-itud “altitude” / alt-ura “height” b. amargo “bitter” > amar-itud / amarg-ura “bitterness”

Apart from the allomorphic pattern shown by the root, and expected if -itud directly attaches to it (amarg- > amar-), there is a crucial semantic difference. The nominalisation in -itud does not express a scalar property, but the one in -ura does. Amaritud is the property of being bitter in absolute terms (as opposed to sweetness or sourness); altitud is the vertical distance between the position occupied by an object and the sea level, but it does not make any statement about the quality of that object. (36a) identifies the position of an object with respect to the sea level; (36b) determines its extension inside an initially unbounded scale. (36)

a. la altitud del faro the altitude of-the lighthouse “The altitude of the lighthouse” b. la altura del faro the height of-the lighthouse “The height of the lighthouse”

This is expected if, in line with Borer (2005), roots do not carry any specific semantics of their own. In order to be able to denote a scale, the root needs to be categorised. As -itud selects a root, its base does not denote a scale. If -ura selects a categorised root involving A + ¥, its base can at least in principle denote a scale. This also implies that -ura includes a lexical categoriser, which explains that it can follow some adjectivisers. (37)

a. nP n -itud

b. NP ¥ alt-

N -ura

A/aP anch-ura “width” / anch-eza “wideness” b. blando “soft” > bland-ura / bland-ez “softness”

The difference between the members of these two pairs is that the one in -ura denotes the dimension associated to the scale. (39a) does not entail that the street is wide; it could be narrow, but we just say that the street must have some value in that dimension. In contrast, the nouns in -ez(a) also entail that the degree of that property in the subject is higher than the standard of comparison and, thus, (39b) entails that the street is wide. (39)

a. la anch-ura de la calle the wide-ura of the street “The width of the street” b. la anch-eza de la calle the wide-eza of the street “The wideness of the street”

What this suggests is that, while -ura takes a projection denoting a scale, -ez(a) takes a more complex object which already has space to set the positive value of the adjective, that is, already involves some degree projection. We will label this projection, following the spirit in Kennedy (1999) and Bobaljik (2012), Degpos (positive degree). It is the projection taken by positive degree adjectives, and it already incorporates the meaning of a standard of comparison that sets the point in the scale from which the degree counts as possessing that property. Thus, -ez(a) takes as its complement this projection. (40)

NP N -eza

DegposP Degpos

AP A

¥

Towards a Syntactic Account of Affix Combinations

117

5. Conclusions: towards a syntactic map of noun and adjective projections Although admittedly preliminary and sketchy, the overview that we have conducted of nominalizing and adjectivizing morphology in Spanish has provided some initial evidence in favour of a syntactic analysis of affix sequences. The hierarchy that has emerged in this overview is represented in (41). (41)

pP p

PP P

QuaP Qua

aP a

AP A

nP n

NP N

¥

This complies with the sequence that has independently been identified in other work, with P dominating D dominating Num dominating A dominating N. Our addition has been to propose a Qua layer in the Num domain and splitting the P, A and N domains into at least a functional and a lexical layer. We have seen cases of affixes that correspond to functional projections, like nP and aP, and those must combine with bare roots. We have seen cases of affixes that contain a lexical functor, which allows them to be external to another head with categorial features; some affixes include, also, the Qualia layer, while there is at least one case of an affix which is only the spell out of Qua (-(d)umbre). Some affixes, finally, can be analysed as projections of P or p. The order between the affixes is restricted by this hierarchy in its interaction with lexicalisation. A lot can be derived by the sequence in (41) alone, but, of course, the question is whether in addition to being possible, this approach is better than the

118

Chapter Six

morphological approach where the combinations are restricted by idiosyncratic selectional properties of morphemes. As a general consideration, note that a morphological approach would have to accept the existence of some version of the hierarchy in (41) in order to account for word ordering facts in the languages of the world— the kind of facts that led Cinque (2005) to propose his hierarchy—. In addition to that, a morphological treatment has to postulate selectional restrictions in the lexicon, restrictions that would be to a great extent redundant with the information codified already in the syntax through the cognitive-triggered ordering of the sortal domains. Giving up these restrictions would imply for the lexicalist approach to admit that the syntactic hierarchy determines the ordering of (derivational) affixes. Secondly, a morphological approach would encounter serious difficulties accounting for a suffix like -ista. The problem with -ista is that it is a terminal suffix in the sense that it cannot be followed by any nominaliser, despite it producing productively adjectives. In contrast, with respect to other adjectival suffixes, it is not a terminal morpheme. Both oso and -ico can follow it. We cannot see how this property can be accounted for in any principled way in a lexicalist framework; certainly, more idiosyncratic statements could be added to the lexical entry of this affix, specifying that it is terminal for N but not for A, but that comes too close to admitting some kind of homophony between the nominal and the adjectival use of -ista. In our account, on the other hand, we capture the generalisation that only adjectival suffixes which do not carry any specification of a quality, but establish relationships between elements, can follow it: this is because those suffixes are projections of P or p, and thus the only items that can dominate Qua. Let us finish this article by admitting that a lot of work has still to be done. Luckily, the kind of framework that we have adopted—grounded on Ramchand and Svenonius’ non-templatic cartography and nanosyntax—is explicit and constrained enough to let us spot immediately the points where further research has to concentrate. They gravitate around the interaction between P (and p) and the rest of projections. The functional p is able to combine with adjectives, but not with nouns. The why of this partial constraint has to be further explored. Secondly, a structure containing a lexical P can follow Qua (as in the order -dad-oso) or be followed by an N that contains Qua (as in the ordering -os-idad). Initially, this does not seem to be a problem: nouns can be selected by prepositions, but noun phrases can also embed PPs. However, something else might be happening that differentiates the two kinds of combinations. The two sequences might not contain identical features and components: note that

Towards a Syntactic Account of Affix Combinations

119

in the first case -idad is spelled out by -dad and in the second it is necessarily -idad. This might be a sign that the features used in each case are slightly different, which would be a welcome result in a system where the sequence of sortal domains is restricted. This research will presumably be informed by the study of combinations of prepositional phrases inside nouns, the study of adjectival phrases in their combination with prepositional structures and, in general, the typological variation with respect to PPs inside DPs. Lots of details and specific cases have to be studied, but we hope to have at least provided the reader with enough initial evidence to sustain the claim that a syntactic analysis of affix combinations is on the right track.

References Beghelli, F. and T. Stowell. 1997. Distributivity and negation: The syntax of each and every. In Ways of scope taking, ed. by A. Szabolcsi, 71107. Dordrecht: Kluwer. Bobaljik, J. 2012. Universals in comparative morphology: Suppletion, superlatives and the structure of words. Cambridge (Mass.): MIT Press. Borer, H. 2003. Exo-skeletal vs. endo-skeletal explanations. In The Nature of Explanations in Linguistic Theory, ed. by J. Moore and M. Polinsky, 31-67. Stanford: CSLI Publications. —. 2005. In Name Only. Oxford: Oxford University Press. Bosque, I. 2006. Coordinated adjectives and the interpretation of number features. In Studies in Spanish Syntax, ed. by L. Brugè, 47-60. Venice: Libreria Editrice Cafoscarina. Caha, P. 2009. The Nanosyntax of Case. Ph. D. dissertation. Tromsø: University of Tromsø. Cinque, G. 1999. Adverbs and functional heads. A cross-linguistic perspective. Oxford: Oxford University Press. —. 2005. Deriving Greenberg’s Universal 20 and its exceptions. Linguistic Inquiry 36.3: 315-332. Cinque, G. 2010. Syntax of adjectives. Cambridge (Mass.): MIT Press. Fabb, N. 1988. English suffixation as constrained only by selectional restrictions. Natural Language and Linguistic Theory 6.4: 527-539. Fábregas, A. 2009. As argument for phrasal spell out. Nordlyd 36.1: 129168. Giegerich, H. 1999. Lexical strata in English: Morphological causes, phonological effects. Cambridge: Cambridge University Press.

120

Chapter Six

Hale, K. and S. J. Keyser. 2002. Prolegomenon to a theory of argument structure. Cambridge (Mass.): MIT Press. Halle, M. and A. Marantz. 1993. Distributed Morphology and the pieces of inflection. In The View from Building 20, ed. by K. Hale and S. J. Keyser, 111-176. Cambridge (Mass.): MIT Press. Hay, J. 2002. From speech perception to morphology: affix ordering revisited. Language 78.3: 527-555. Hay, J. and I. Plag. 2004. What constraints possible suffix combinations? Natural Language and Linguistic Theory 22: 565.596. Inkelas, S. 1993. Nimboran position class morphology. Natural Language and Linguistic Theory 11: 559-624. Kennedy, C. 1999. Projecting the adjective: The syntax and semantics of gradability and comparison. New York: Garland. Kiparsky, P. 1973. Elsewhere in phonology. In A Festschrift for Morris Halle, ed. by S. R. Anderson and P. Kiparsky, 93-106. New York: Holt. —. 1982. Lexical morphology and phonology. In Linguistics in the Morning Calm, ed. by The Linguistic Society of Korea, 3-91. Seoul: Hanshin Publishing. Marantz, A. 1997. No escape from syntax: don’t try morphological analysis in the privacy of your own lexicon. UPenn Working Papers in Linguistics 4: 201-225. Martín García, J (2007). El significado de las palabras derivadas: los sufijos polisémicos. In Ex admiratione et amicitia. Homenaje a Ramón Santiago, ed. by I. Delgado and A. Puigvert, 759-771. Madrid: Ediciones del Orto. Pustejovsky, J. 1995. The Generative Lexicon. Cambridge (Mass.): MIT Press. Rainer, F. 1999. La derivación adjetival. In Gramática Descriptiva de la Lengua Española, ed. by I. Bosque and V. Demonte, 4595-4645. Madrid: Espasa. Ramchand, G, and P. Svenonius. 2012. Advanced cartography: Charting elevation. Ms., Tromsø: University of Tromsø. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of grammar, ed. by L. Haegeman, 281-337. Dordrecht: Kluwer. Schadeberg, T. C. 1984. A sketch of Swahili Morphology. Dordrecht: Foris. Siegel, D. 1974. Topics in English Morphology. Ph.D. dissertation. Cambridge (Mass.): MIT. Svenonius, P. 2010. Spatial P in English. In Mapping Spatial PPs: The Cartography of Syntactic Structures, Volume 6, ed. by G. Cinque and

Towards a Syntactic Account of Affix Combinations

121

L. Rizzi, 127–160. New York: Oxford. Williams, E. 1981. Argument structure and Morphology. The Linguistics Review 1: 81-114. Zamparelli, R. 2000. Layers in the determiner phrase. New York: Garland.

Notes 

1 We are grateful to Björn Lundquist, Mike Putnam, Josefa Martín and Gianina Iordachioaia for insightful comments, observations and help in getting this article done during the whole process. As usual, the errors that might have remained in this version are my sole responsibility. 2 The following abbreviations are used in this paper: A (lexical adjective), a (functional adjectival projection), agr. (agreement), K (kinds), N (lexical noun), n (functional noun projection), neg. (negation), nom. (nominalisation), Q (quality). 3 We decided to accept all words found in the corpora that matched the previous description (the base without the sequence suspect of being a suffix can produce other words), independently of subjective criteria such as whether the word was considered literary, old-fashioned or restricted. In any case their existence implies that they are at least compatible with the grammar of the language. We did not exclude, either, combinations with a low productivity in the corpus. Even though dad+oso gives only two forms, again, if they exist, this sequence must be possible in the grammar of the language. Thus, we ignored frequency in this study— without wanting to imply that frequency is not useful to determine aspects of language; cf. Janda this volume—. We just choose not to consider it here so that we capture, to the extent that is feasible, the contrast between possible and impossible.  4 A complete account of -ista must also address the fact that it seems to be followed by one adjectival suffix, even if in the sample of data we are not considering A-A combinations. The sequence -ístico is attested in some words— armament-ístico “weapon-ístico, related to weapons”—, and it is plausibly segmented as -íst-ico (given, among other things, that the stress pattern is the same that the affix -ico imposes to its base).  5 We should say right away that -ista can also be used to produce nouns. Words like ten-ista “tennis-ista, tennis player”, real-ista “realist”, or pesim-ista “pessimist” can be used either as nouns or as adjectives. The nominal use of -ista does not perfectly fit in our analysis. Given the hierarchy of domains that will be assumed in §4, we should expect that this nominal use is [Qua [N]], while the adjectival one should be—as argued in §4.5—[Qua [A]]. Given the superset principle, it is not possible to unify those two entries in one single [Qua [A [N]]], because gapping of intermediate projections is not in principle allowed by the Superset Principle. In the nominal use, A should be missing from the structure. One possible alternative would be that the lexical entry of -ista is always [Qua [A]], and in the nominal uses, a phonologically null pronoun is introduced in the

122

Chapter Six

 specifier of QuaP, acting as the subject of the adjective. The whole would work as a noun, and the entry of the affix would be identical in its adjectival and “nominal” use.  6 Let us take a minute to spell out some of the complications of this analysis. The main theoretical complication is in what sense the notion of relation is interpreted isomorphically when we adopt an analysis where P is divided into a lexical and a functional layer: do both heads have the power to establish relationships, and identify these relationships with each other when they appear together, or do each one of them introduce one of the arguments in the relation? Empirically, the complication is the existence of a few affixes used to form relational adjectives, which can also follow some nominalisers. -al is one such example, as in cult-ur-al “cult-ur-al”. One possibility is that these affixes would correspond to a structure [pP p [AP A]], allowing them to have a transitioner which is directly selected by a functional P. This proposal requires further exploration into the relation between Ps and As, as it suggests that no recategorisation is necessary to move from A to P. The relations between adjectives and prepositions have received some attention in the last years, with proposals like Hale and Keyser (2002), which go as far as to say that a prepositional structure is necessary to complete the semantic requisites of an adjective. 

CHAPTER SEVEN EVENT RELATED NOMINALS ISABELLE ROY UNIVERSITÉ PARIS 8–CNRS

ELENA SOARE UNIVERSITÉ PARIS 8–CNRS

1. Introduction1,2 As has been widely argued for in the literature, deverbal nominals fall into two categories, depending on whether they retain the event from their verbal base or not. The original observation is due to Chomsky (1970); see also Lees (1960), Marantz (1997), Borer (2003). For English -ation and -ing nominals (e.g., destruction, examination; forming, examining) and for zero-derived nominals (e.g., form, exam), this has led to distinguishing two classes of nominals: complex-event vs. result nominals (CENs/RNs– Grimshaw 1990) or Argument-Supporting nominals (AS-Ns, cf. (1)) vs. Referential nominals (R-Ns, cf. (2)–Borer 1999): (1)

a. the destruction of the city by the enemy b. the examination of the students by the teacher c. their building new quarters

(2)

a. a complete destruction b. a difficult exam c. an impressive building

The nominals in (1) have an event interpretation (hence, Grimshaw’s 1990 label of CENs). The eventive interpretation has been noted to correlate with various syntactic properties, as, for instance, the (possible) presence of temporal/aspectual modifiers.

124

(3)

Chapter Seven

a. They destroyed the city in two hours. b. the destruction *(of the city) in two hours

The pattern in (3b) also has another important property, which is the obligatory realization of the arguments of the verbal base (hence the term Argument Supporting nominals; cf., Borer 1999, 2003). This property is compulsory when the event structure is activated in the presence of eventrelated modification (see Grimshaw 1990, Borer 1999, 2003, Alexiadou 2001, among others). The correlation between the event interpretation and obligatory realization of argument structure has been an important point since Grimshaw (1990). The fact that both go hand in hand has been claimed in the literature to show that both are realized grammatically and that eventivity and arguments are inherited from the verbal and/or aspectual structure present with AS-Ns but are missing with R-Ns (the latter being simply derived from bare roots)–cf. Borer (2003); Alexiadou (2001); van Hout and Roeper (1998). The syntactic approaches to word formation represented by these scholars take eventivity in AS-Ns to be correlated with the projection of syntactic functional layers detectable through argument structure projection and aspectual modifiers. The source of the eventive interpretation is the presence of a verbal base upon which AS-Ns are built. Nominalizations may thus inherit verbal properties, when (and only when) they involve a verbal / aspectual structure. The term ‘event’ (or ‘eventive’) nominals, however, is often the subject of some misunderstanding, or at least suffers from variable definitions depending on whether it is taken in the syntactic tradition where eventivity is correlated with particular structural properties, or from a (lexical-)semantic point of view where a much larger class of nominals would be considered as ‘eventive’.3 For instance, Grimshaw’s (1990) Simple Event nominals (SENs) fall into this class when semantically defined. Such nouns (e.g., meeting, play) are ‘semantically’ associated to an event interpretation but do not exhibit the common event related properties of AS-Ns described above. SENs cannot take typical verbal / temporal modifiers otherwise found with AS-Ns (even in cases where there is a related verb, with the relevant temporal-aspectual modifiers): (4)

a. They met / played for two hours. b. the meeting / the play (*for two hours)

This suggests that SENs are not structurally derived from verbs, and hence if they involve an event, it cannot be inherited from a VP. Nevertheless, they do have an event interpretation which, it has been claimed in the

Event Related Nominals

125

literature, cf., Haas et al. (2008), can be tested in the context of, e.g. the take place predicate. Take place requires an eventive subject, and is compatible with both AS-Ns and SENs subjects, while rejecting R-Ns, as expected: (5)

a. The destruction took place at noon. b. The movie / meeting took place at noon. c. *The table / form took place at noon.

(AS-Ns) (SENs) (R-Ns)

Another problematic group of nominalizations is also often associated with an event interpretation, namely -er nominals (see Rappaport Hovav and Levin 1992, Alexiadou and Schäfer 2010, Roy and Soare (to appear)). -Er nominals differ from AS-Ns in that they denote individuals (e.g., driver, teacher, scuba-diver). However, recent works by Alexiadou and Schäfer 2010 and Roy and Soare (to appear) have shown that some of them at least are interpreted in association with an actual eventuality. Independently of the typology one accepts (whether retaining three groups: episodic / dispositional / instruments, as in Roy and Soare (to appear) or just two groups based on the episodic / dispositional contrast alone, as in Alexiadou and Schäfer (2010)), there is a common agreement that some -er nominals at least relate to particular events, arguably inherited from their verbal base again. For instance, as commonly noted for English, phrasal –er nominals as in (6a), entail that the individual denoted by the N has taken part in the action expressed by the related verb (i.e., saved lives); whereas such entailment does not exist with the compound nominals as in (6b). One could hypothesize that the source of the event-related meaning for (6a) resides in the presence of an underlying event, plausibly derived from a full verbal phrase. (6)

a. a saver of lives (has saved lives) b. a life-saver (hasn’t necessarily saved lives)

However, as for SENs, ‘eventive’ -er nominals do not take event modifiers otherwise possible with AS-Ns (compare (7) with (3) above): (7)

a. He drove the truck (for two hours). b. the driver of the truck (*for two hours)

Evidently, while there is a strong sense in which both AS-Ns and episodic -er nominals refer to events or are related to an event interpretation, they do not do so in the same way, as the diagnostics for eventivity clearly

126

Chapter Seven

indicate. These facts further illustrate why in the present-day literature on nominalizations there is no consensus on the definition of eventive nominals and their variable properties. The aim of this paper is to gain some understanding into the semantic relationship derived nominals entertain with their verbal base, and to provide a principled analysis of AS-Ns, SENs and (eventive) -er nominals that accounts for the different flavors in which the interpretation of a deverbal nominal is said to be ‘event-related’. We offer an analysis that bears crucially on a difference between strong/grammatical eventuality and lexical/conceptual eventuality cast in terms of a semantic type difference between entity-denoting vs. event-denoting nominals. This paper is organized as follows. Section 2 deals with ‘event’ nominals, i.e., AS-Ns and SENs, providing a summary of the literature that will serve as a starting base for our discussion, and argues for a first distinction between strong/grammatical events and lexical/conceptual events. Section 3 turns to a detailed discussion of the properties of eventive -er nominals, which manifest event properties in a different way from event-denoting nominals. We first show that these nominals are event-related in a strong/grammatical sense, and that a unified structural account is warranted, for event-denoting and for -er AS-Ns as well. We proceed in section 4 to a type semantic analysis of derived nominals that captures the differences between individual nominals and event nominals and the interaction of semantic types with the grammatical vs. conceptual events. In section 5, we extend our proposal to include SENs, and then conclude in section 6.

2. Event nominals 2.1. AS-nominals The literature on deverbal nominalizations starting with Lees (1960); Chomsky (1970); Grimshaw (1990) pays special attention to the question of their ambiguity. Taking, as an illustration, -ation nominals (e.g., examination, destruction, manifestation), they may denote either an event or an entity (i.e., object), which may but must not be the result of an event. As stated above, this is commonly expressed in the literature under the form of the CENs / RNs distinction (initially proposed by Grimshaw 1990), and is implemented broadly in terms of a structural ambiguity, as we will see below. An overview of the properties distinguishing CENs from RNs is summarized in Table 1. Given the observed correlation between argument

Event Related Nominals

127

structure and eventive interpretation, in recent works on nominalizations (cf. Borer 1999, 2003; Alexiadou 2001, 2010a,b; Kornfilt and Whitman 2011, among others) the distinction has been restated in terms of Argument Supporting (or AS)-nominals and Referential (or R)-nominals (terminology from Borer 1999, 2003). The properties in the left column, including obligatory realization of the (internal) arguments and modification with aspectual modifiers, are generally seen as a hallmark for eventivity inside nominals. The properties are exemplified in (8) for ASNs and (9) for R-Ns.

(i) (ii) (iii) (iv) (v)

AS-Ns

R-Ns

event reading obligatory arguments compatible with aspectual modifiers like in three hours constant, frequent with the singular by-phrase is an argument

no event reading arguments not obligatory not compatible with aspectual modifiers constant, frequent possible only with the plural by-phrase is not an argument

Table 1: properties of AS-nominals and R-nominals (8)

AS-Ns a. the examination of the students by the teachers b. the examination *(of the students) (by the teachers) (in three hours) c. the (frequent) examination of the students by the teachers

(9)

R-Ns a. the form; the exam b. the exam (*by the teachers) (*in three hours) c. the frequent exam*(s)

As illustrated above in (8), in presence of event-related modifiers like frequent, constant adjectives or in/for-PPs, AS-Ns obligatorily realize their argument structure. Removing the arguments in the presence of the modifiers would give rise to ungrammaticality. This is not the case with RNs, which are noneventive and do not have arguments (9b). Frequent/constant modification is possible with R-Ns but in the plural

128

Chapter Seven

(9c), which gives rise to an iterative reading only (“frequently giving/taking exams”; compare with (8c)). In/for-PPs are impossible. Furthermore, R-Ns can be selected by predicates that require an entity noun and not an eventive noun: (10)

a. *The examination of the patients was on the table. b. The exam was on the table.

For ambiguous nominals, such as assignment, painting, building, manifestation, the compatibility with be on the table excludes the presence of arguments: (11)

a. *The assignment of the tasks to the participants was on the table. b. The assignment was on the table.

The literature offers different views on the AS-N / R-N ambiguity. Lexicalist approaches take the ambiguity as being stored in lexical entries, and consequently assume the existence of assignment-AS-N / assignment-R-N pairs. This direction has been developed since Halle’s (1973) lexical Word Formation Rules, and continued in Booij (1977), Aronoff (1976), Di Sciullo and Williams (1987), among many others. By opposition, structural/syntactic approaches reject the idea of rampant ambiguity in the lexicon, and see the AS-N/R-N contrast as corresponding to a systematic structural difference in the form of the nominalization. The latter position is argued for by Marantz (1997), Borer (1999), Alexiadou (2001), among others, which endorse a syntactic approach to word formation. Building on Grimshaw’s criteria, it has been argued in this tradition that AS-Ns are derived on the basis of a full structure including verbal / aspectual layers; while R-Ns are simple, root-derived nominals. Syntactic approaches to deverbal nominals formation thus assume that the correlation between the eventive interpretation and the obligatory argument structure must be implemented by assigning different structural representations to the two classes of nominals. The correlation with the argument realization is therefore not accidental but derives from the internal syntactic properties of the relevant nominal expressions. The projection of argument structure inside deverbal nominals is also to be taken as a property of the verbal layers. These different layers have received different labels throughout the literature, from “Event Phrase” (van Hout and Roeper 1998), to different flavors of AspP (Borer 1999, 2003, 2005; Alexiadou et al. 2010); and different executions have been proposed. For instance, Borer (1999, 2003,

Event Related Nominals

129

2005), implements the correlation between event structure and argument structure by proposing that arguments are introduced by functional heads, one of which is also responsible for introducing the event variable. In the structure of AS-Ns in (12) below, AspEv (standing for Aspect of Event) thus introduces the external argument and AspQ (standing for Aspect of Quantity) the internal one (which is likewise severed from the root). AspEv is also responsible for introducing the event variable ev. In this paper, we will by and large adopt Borer’s framework; but see Alexiadou (2001); Van Hout and Roeper (1998) among others for alternative implementations. (12)

DP/NP 3 N AspEvP 3 AspEv AspQP -ation 3 ev AspQ XP 4 form-

(AS-Ns)

R-nominals in (13), in turn, are built directly from a root. They lack verbal structure and, therefore, the event variable introduced by AspEv. (13)

DP/NP 3 N XP -ation 4 form-

(R-Ns)

Recent work on AS-Ns denoting processes have, thus, reached the conclusion that eventivity in these nominals is structurally built-in, and that the projection of argument structure is also a consequence of their functional structure. The syntactic approaches to deverbal nominals formation share the idea that eventivity is encoded in the syntax. Therefore, we will refer to these cases as cases of “grammatical eventivity”, in which the presence of the event is structure-related and results from the presence of dedicated verbal functional projections in the structure of the nominal, identifiable by aspectual and manner modification. We will call this structurally built-up eventivity ‘strong/grammatical’ eventivity.

130

Chapter Seven

2.2. Simple Event nominals In her original typology of deverbal nominals, Grimshaw (1990) distinguished not two, but three classes of deverbal nominals. Besides CENs and RNs, (here replaced by AS-Ns and R-Ns, respectively), her taxonomy includes a third class of so-called Simple Event Nominals (SENs). Nominals in the SENs class also denote events as they can combine with predicates like take place, last x time and be interrupted that take an event as a subject (cf. Haas et al. 2008) (14). According to this test, they pattern with AS-Ns (15) rather than R-Ns (16), leading many current researchers to class them with ‘event’ nominals: (14)

a. The concert/ the movie/ the game took place at nine. (SENs) b. The concert / the movie/ the game lasted three hours. c. The concert/ the movie/ the game has been interrupted.

(15)

a. The examination of the papers by the committee took place today at 5pm. (AS-Ns) b. The examination of the papers by the committee lasted three hours. c. The examination of the papers by the committee has been interrupted.

(16)

a. *The paper/ table took place yesterday. b. *The paper/ table lasted three hours.4 c. *The paper/ table has been interrupted.

(R-Ns)

SENs also pattern with AS-Ns in being satisfactory in the during the N PP construction which calls for a temporal extension: (17)

a. during the movie/concert/game (SENs) b. during the examination of the paper by the committee (AS-Ns) c. *during the paper/table (R-Ns)

However, while the class of SENs represented by concert, movie, game in (14) above, shares with AS-Ns the property of being eventive (in a way to be defined later), they clearly do not necessarily project argument structure. Incidentally, it turns out that, as opposed to AS-Ns (and R-Ns for that matter) SENs may, but need not to, be derived from a verbal base (with or without derivational morphology) (e.g. a movie, a concert, a game vs. a meeting, an attack). SENs are thus eventive by the semantic tests above5, but not in the way AS-Ns are; i.e., presumably not in the

Event Related Nominals

131

strong/grammatical sense defined in the previous section. Recall that standard syntactic views on AS-Ns assume that what we call a ‘grammatical event’ must be contributed by an underlying verbal base (or VP), and is identified by the standard tests summarized in Table 1. SENs, which do not require a verbal base, by assumption, cannot involve a grammatical event. Crucially, and as expected, SENs are not compatible with aspectual PP modifiers (in/for PP), and in that respect pattern with R-Ns (20) rather than AS-Ns (19), independently of the existence of a related verb (18a) or not (18b).6 (18)

a. the meeting/attack (*for three hours) (SENs) b. the concert/movie/boycott (*for three hours/months)

(19)

a. The president met with the Prime Minister (for three hours). b. the meeting of the president with the Prime Minister (for three hours) (AS-Ns)

(20)

the paper/table (*for three hours)

(R-Ns)

Further properties distinguish SENs from AS-Ns. Recall from Table 1 that R-Ns are compatible with frequency modifiers when in the plural only, while AS-Ns accept frequency modifiers in the singular. As (21) shows, in that respect again, SENs pattern with R-Ns in (9c), and not with AS-Ns in (8c). (21)

the frequent concert*(s); the frequent movie*(s)

(SENs)

In general there is no particular restriction on plurality and quantification with SENs (as for R-Ns and other nominals: three exams; many dogs); while certain restrictions on quantifiers, numerals and determiners, as illustrated in (22b), are commonly reported on AS-Ns (see Grimshaw 1990, Snyder 1998).7 (22)

a. many concerts; three movies (SENs) b. *several/two/these elections of John by the department (AS-Ns)

The count properties of SENs can be seen as a result of their being event sortals (cf. Bennett 1988, Snyder 1998). The fact that SENs can be counted is correlated to the fact that they can be individuated, unlike ASNs. The contrast is visible in existential constructions, which require an

132

Chapter Seven

individuated N, and accept SENs but reject AS-Ns (cf. also Mourelatos 1978): (23)

a. There is a movie/ are three movies starting at 5. b. *There is a destruction/ are three destructions of the city by the enemy starting at 5.

In sum, SENs have heterogeneous properties, sometimes similar to those of AS-Ns and sometimes not (and hence making them similar to RNs and other non-derived nominals). In view of these apparently contradictory properties, SENs have always been rather problematic and left aside in many recent accounts. If we follow a structural approach, as discussed earlier, and if we assume that both the projection of argument structure and the event interpretation depends on the presence of a verbal(/aspectual) structure, SENs do not involve a ‘grammatical event’, structurally built up in the nominal, in the sense defined in the previous section. We must conclude that there is another possible source of eventivity inside nominals, which is not structure-related. Nominals can refer to events in the absence of verbal bases and therefore of any verbal layers. We will name this kind of eventivity, which is not inherited from a base predicate, a weak/conceptual(or lexical) eventivity.

2.3. Interim conclusion In order to describe event-related nominals, a first distinction is needed between strong/grammatical eventivity and weak/conceptual eventivity. Among the tests commonly used to identify an underlying event, or eventuality more generally, inside derived nominals, some pertain to the grammatical event, others to the conceptual/lexical event. The tests are split as indicated in Table 2. This understanding of the tests turns crucial to apprehend the properties of SENs specifically, and how they differ from AS-Ns. (More on this in Section 5).

Event Related Nominals

133

Tests for underlying eventuality

Strong/ Grammatical eventuality

Weak/ Conceptual eventuality

(i) subject of be on the table (ii) subject of take place, be interrupted (iii) during the N (iv) obligatory arguments (v) constant, frequent possible with the singular (vi) compatible with aspectual modifiers like in/for three hours

-

-

+ + +

+ + -

+ +

-

Table 2: Tests for strong/grammatical eventuality vs. weak/conceptual eventuality

3. Further issue: individual nominals 3.1. Eventive -er nominals -Er derived nominals (often called ‘Agent’-nominals) present a further difficulty to an already complex notion of eventuality inside nominals. On the one hand, they denote individuals (rather than events); but on the other hand they have a strong/grammatical eventuality that can be linked to an AS-Ns structure (Alexiadou & Schäfer 2010; Roy and Soare (to appear)). Nominals denoting participants in an eventuality have been characterized in the literature as being sensitive to a distinction between eventive and noneventive, on the basis of pairs like saver of lives / lifesaver and mower of the lawn / lawn-mower (Rappaport Hovav and Levin 1992, Van Hout and Roeper 1998, among others). Only in the first case is the participant denoted by the nominal entailed to be involved in an actual event; no such entailment arises with the compound forms. Animacy plays a role here, as instrument -er nominals (grinder, blender) never involve a participation in an event (cf., Rappaport Hovav and Levin 1992; Roy and Soare (to appear)). As argued in Roy and Soare (to appear), when interpreted as eventive, animate -er nominals have properties of strong/grammatical events rather than weak/conceptual events. This can be shown by the tests (iv) and (v) in Table 2, which discriminate between the two types of eventualities. With

134

Chapter Seven

respect to these two tests (and we will come back to test (vi) later), eventive -er Ns pattern with AS-Ns (and not SENs). As argued by Rappaport Hovav and Levin, the eventive interpretation, which is highlighted by the presence of frequency modifiers, correlates with the obligatory realization of arguments. -Er nominals can take frequency modifiers, but only when they realize their arguments.8 In addition, frequency adjectives are possible with the singular (24b). Compare with AS-Ns in (8c) above. (24)

a. the constant defenders *(of human rights) b. this frequent consumer *(of tobacco)

On the basis of the similarities between (24) and (8c), a unified account for eventive -er Ns and AS-Ns, in terms of strong/grammatical eventuality, seems plausible. Recent works have argued for such an account, but assume further distinctions within the class of eventive -er Ns - cf. Alexiadou and Schäfer (2010), Roy and Soare (to appear), in particular, in terms of the episodic/dispositional distinction. According to Alexiadou and Schäfer (2010), dispositional -er nominals have event-related properties and share the same syntactic structure as episodic -er nominals. Both are eventive and involve verb-like internal structure, i.e., Aspect and Voice heads (in a standard Distributed Morphology approach), but they differ in terms of aspectual specifications. Episodic and dispositional meanings are, in this approach, two flavors of an aspectual head necessarily present in eventive nominals. (25)

a. Dispositional -er Ns i. fire-fighter, live-saver, baker, teacher (educated but not necessarily experienced) ii. [nP -er [AspP-DISPO [VoiceP x [vP ev [RootP ¥ ]]]]] b. Episodic -er Ns i. saver of lives, fighter of the fire (necessarily experienced in action) ii. [nP -er [AspP-EPISO [VoiceP x [vP ev [RootP ¥ ]]]]]

For Roy and Soare (to appear), dispositional and episodic -er Ns involve a full verbal structure akin to the one found in AS-Ns, and differ by the type of quantification on the event variable, namely existential vs. generic: (26)

a. Dispositional -er Ns GEN [NP N [AspEvP -er [AspEv’ AspEv ev [AspQP [RootP ¥ ]]]]]

Event Related Nominals

135

b. Episodic -er Ns  [NP N [AspEvP -er [AspEv’ AspEv ev [AspQP [RootP ¥ ]]]]]] They show, on the basis of French data, that both episodic and dispositional Ns allow event-related adjectival modification, which can be of two types; frequency adjectives, allowed by episodic -eur Ns only, and big/happy adjectives with an event-related meaning (cf., Larson 1998), allowed by both dispositional and episodic -eur Ns. The interpretive contrast between episodic and dispositional Ns comes from their internal argument, either specific or nonspecific, leading to a particular vs. generic underlying eventuality. For further details regarding adjectival modification, and their event-related meanings in particular, as well as the correlation between event type and nominal meanings, we refer the reader to Roy and Soare (to appear). (27)

Dispositional -er Ns a. *Nous avons interviewé un vendeur fréquent de voitures/ les we have interviewed a seller frequent of cars the consommateurs fréquents de drogue. consumers frequent of drug "We have interviewed a frequent car-dealer/ the frequent drug users." b. Nous avons interviewé un petit vendeur de voitures/ les we have interviewed a small dealer of cars the gros consommateurs de drogue. big consumers of drugs "We have interviewed a small car-dealer/ the big drug users."

(28)

Episodic -er Ns a. Un consommateur fréquent de plusieurs drogues douces/ de a consumer frequent of several drugs soft of LSD a témoigné au procès. LSD has testified at.the trial "A frequent user of several soft drugs/ of LSD testified in court." b. Un heureux/ gros consommateur de plusieurs drogues douces/ a happy big user of several drugs soft de LSD a témoigné au procès. of LSD has testified at.the trial "A happy/big user of several soft drugs/of LSD testified in court."

136

Chapter Seven

Crucially, no event-related properties can be observed in the case of instrument -eur nominals. Event-related meanings of adjectives are never allowed. Instruments differ in a clear way from dispositional Ns in (27). (29)

Instruments9 a. *Un broyeur fréquent nous serait utile. a grinder frequent us would.be useful "A frequent grinder would be useful to us." b. *Un gros broyeur nous serait utile. a big grinder us would.be useful "A big grinder would be useful to us."

Moreover, instruments never project true arguments. In particular, definite-specific objects are always ruled out, which we take to indicate that instruments do not take arguments (30) and that when they appear with a de-phrase, the latter is a mere modifier (31). The nonargumental status of de-phrases with instruments is further supported by the possibility of substituting them with a purpose à-phrase adjunct (never found with true arguments). (30)

a. L’aspirateur (*de la poussière) n’a pas bien fonctionné. the.aspirator of the dust neg.has not well functioned "The vacuum-cleaner (of the dust) didn’t work well" b. Le photocopieur (*de l’article) a été très efficace. the photocopier of the.paper has been very efficient "The copy-machine (of the paper) has been very efficient"

(31)

a. broyeur de végétaux blender of vegetables "vegetable blender" b. broyeur à végétaux blender at/to vegetables "vegetable blender"

vs. vs.

dresseur de lions tamer of lions "lion tamer" *dresseur à lions tamer at/to lions

Accordingly, instrument -er Ns must be treated on a par with R-Ns, as simple, root-derived nominals. A structurally built-in grammatical event must be assumed in the two classes of eventive (animate) -eur Ns, which must share the structure of AS-Ns in (12). Instrument Ns pattern with rootderived nominals with which they share the structure (13). Accordingly, they have the following structure, respectively:

Event Related Nominals

137

(32)

NP (Eventive -er Ns) 3 N AspEvP -eur 3 -eur AspEv’ x 3 AspEv AspQP ev 3 DP AspQ’ 3 AspQ RootP 4 ¥

(33)

NP 3 N RootP -eur 4 ¥

(Instrument Ns)

However, if episodic/dispositional -er nominals are a form of AS-Ns, involving a case of strong/grammatical event, some differences between process-denoting AS-Ns and individual-denoting eventive -er Ns need to be addressed. We turn to this issue below.

3.2. Some unexplained differences The presence of a grammatical event inside -er Ns has often been questioned and is the subject of some controversies. Proponents of the noneventive view, invoke two types of evidence that they take as arguing against a unified treatment of process As-Ns and individual -er Ns. Baker and Vinokurova (2009) argue on the basis of the grammaticality of adverbials in pairs like (34) for separating process As-Ns and -er Ns, assigning a ‘purely nominal’ status to the latter, which, according to them, are deprived of any internal verbal structure (even in languages in which they are able to assign Accusative case to their object, like Sakha). Absence of adverbial modification is not a reliable test for absence of grammatical events, however, as adverbs are also precluded with many process AS-Ns (even in cases where they are semantically compatible with the base verb) (35).

138

Chapter Seven

(34)

a. finding the wallet quickly b. *the finder of the wallet quickly

(35)

a. the enemy found/destroyed the city quickly b. the finding/destruction of the city by the enemy (*quickly)

Another piece of evidence taken to argue against a unified treatment of process As-Ns and -er Ns concerns the test (vi) in Table 2: eventive -er Ns never allow for aspectual PP modifications typically found with processdenoting AS-Ns. This contrast is taken by Borer (2012), for instance, to suggest the purely nominal (i.e. not verbal based) character of -er Ns altogether. (36)

a. le domptage des chiens (pendant des années) the taming of.the dogs for many years b. la vente du chien (en cinq minutes) the sale of.the dog in five minutes

(37)

a. le dompteur des chiens (*pendant des années) the tamer of.the dogs for many years b. le vendeur du chien (*en cinq minutes) the seller of.the dog in five minutes

This contrast was originally noted for Greek by Alexiadou et al. (2000) (and reported in Alexiadou 2001) and taken as an indication of a diminished verbal character for -er nominals, even in their eventive meaning. (38)

a. *i damastes ton fotonion mesa se/gia enan eona the tamers the-gen photons within for a century b. *o katharistis tu ktiriu epi ena mina telika apolithike the cleaner the-gen building for a month finally got.fired

Alexiadou et al. (2000) suggested that an explanation for the ungrammaticality of aspectual PPs would rely on the fact that -er Ns lack an Asp(ectual) projection, which would rule out adverbial modification across the board–and manner modification can only be spelled out as an adjective. This in turn, as also suggested by Alexiadou (2001), relates to the semantics of -er nominals, which denote individuals, while process nominals denote events. The difference is, thus, expected to the extent that aspect is relevant for processes and not for individuals. However, a precise

Event Related Nominals

139

implementation of the difference has never been proposed. In our view, and since we accept a split between eventive and noneventive (i.e., instrument) -er Ns, the issue remains to understand the ban on aspectual PPs with individual Ns, as they involve a strong/grammatical event. The properties of eventive -er Ns are the combined result, as we shall argue below, of their involving both a grammatical/structurally built-in event and an individual variable. We will turn to this account in the next section.

4. Event nominals and semantic types 4.1. Event vs. individual argument One fundamental difference between eventive -er Ns and process ASNs that, we will argue, has a major bearing on their contrastive properties, concerns their denotations. Evidently, on the one hand -er Ns denote individuals (the driver = the person who drives), whereas on the other hand process AS-Ns denote events properly speaking (the driving of the car = the event of driving the car). The difference is associated to the semantic properties of the nominal suffixes themselves (-er vs. -ation, -ing, etc.), and has structural consequences. As argued in Roy and Soare (to appear), the nominalizing suffix -er, which picks out an individual, is assumed to realize the external argument (i.e., occupies the specifier of AspEvP; cf., (32) above).10 Following Borer (1999, 2003), the nominalizing suffix -ation, for instance, which picks out an event, is the realization of the aspectual head AspEv responsible for introducing an eventuality. Hence, even though eventive -er Ns and process AS-Ns share the same internal structural frame (that of AS-Ns rather than root derived R-Ns), the semantic difference in their denotation is expressed structurally as well. Cf. the representations in (32) vs. (12). We propose that the two types of nominals correspond to a semantic difference between nouns of individuals and nouns of events. Assuming that nominalized constructs are NPs, and that NPs are predicative by nature (while referentiality would require a DP layer), we claim that the former take an individual argument, hence are of type (39); whereas the latter take an event argument, hence are of type (40).11

140

Chapter Seven

(39)

NP 3 N AspEvP –er 3 -er AspEv’ 3 AspEv AspQP 3 DP RootP 4 ¥ (40)

NP 3 N AspEvP –ation 3 DP AspEv’ 3 AspEv AspQP -ation 3 DP RootP 4 ¥

For process AS-Ns, type , the outcome of the nominalization is a (nominal) predicate of events (40). The event semantics is introduced by the AspEv phrase complement of N°. As with AS-Ns, AspEv also introduces the event component in eventive -er Ns (39); however, both differ in the semantics of their nominalizing suffix and hence the interpretation of their resulting nominalization. The individual reading in (39) leads to the participant/agent interpretation as often characterized. Importantly, the source of the eventive interpretation in both (eventive) -er Ns and process AS-Ns is the presence of the event variable associated with AspEv. Both -ation and (eventive) -er suffixes take constituents of type as argument. Other -er nominals, i.e. instruments (see Roy and Soare (to appear)), are noneventive and do not involve an event variable at all. In that case, we will assume a homophonous -er suffix that is noneventive and takes a bare root as complement; the resulting nominalization is consequently of type only:

Event Related Nominals

(41)

141

NP 3 N RootP -er 4 ¥

The important point here is that both process AS-Ns and eventive -er nominals involve an event variable, and hence share a strong/grammatical eventuality interpretation. They only differ in the type of the outcome nominal. We claim in the rest of this section that this difference alone plays a crucial role in explaining the compatibility / incompatibility of the two sorts of eventive nominals ( vs. ) with event-related modifiers, namely aspectual PPs. The relevant data have been presented above and will now be discussed.

4.2. Locality of Predicate Modification The puzzling difference between AS-Ns and eventive -er nominals concerning temporal/aspectual adjunct modifiers, has sometimes been taken as evidence that -er Ns are never eventive. We argue instead that they mark the type difference between and nominals, rather than the absence of underlying eventuality itself. Temporal and aspectual event modifiers are structurally adjuncts. We assume that they combine with the nominal they modify via the rule of Predicate Modification (PM) stated in Heim and Kratzer (1998: 65). (42)

Predicate Modification If Į is a branching node, {ȕ, Ȗ} is the set of Į’s daughters, and [[ȕ]] and [[Ȗ]] are both in D, then [[Į]] = Ȝx ‫ א‬De . [[ȕ]] (x) = [[Ȗ]] (x) = 1.

PM is a conjunction operation. PM amounts to ‘intersective modification’ (i.e., Conjunctive composition) and captures the intersective reading of predicate modifiers: (43)

[[city in Texas]] a. Ȝx ‫ א‬De. [[city]] (x) = [[in Texas]] (x) = 1 b. Ȝx ‫ א‬De. x is a city and x is in Texas

PM predicts that the intersective reading of an adjunct modifier and the nominal is only possible at the level where they merge. Importantly,

142

Chapter Seven

however, PM is only applicable when the two constituents ȕ and Ȗ are of the same semantic type. In their original proposal, both the ȕ and Ȗ elements are of type (for predicates of individuals). Here, however, we extend the system to include a type difference between predicates of events, type and predicates of individuals, type . True to the basic principle, PM is only possible between two expressions of the same or type. If modifiers are adjoined at the AspEvP level (i.e., associated to the underlying grammatical eventuality; cf. (39)-(40)), PM should be applicable in both eventive -er Ns and AS-Ns alike, and both classes of nominals should be equally compatible with the relevant aspectual PPs. The event modifying PPs are possible with AS-Ns but not with -er nominals, however. This suggests that the nominal type plays a role in allowing PM, and consequently that PM takes place at a level where the distinction is expressed. Concretely, it means, we propose, that adjunction takes place at the NP level, i.e. “after” nominalization rather than before nominalization. The ban on aspectual PPs can be explained solely on the basis of the type difference between eventive nominals that take individuals or events. The basic intuition is that driver (type ) is not compatible with for two hours modifiers (type ) simply because it is a predicate of individuals and not of events. With process AS-Ns, the type of the PP modifier matches the type of the nominal and the rule of compositionality can be applied as in (44b). (44)

a. the destruction of the city by the enemy in three days b. NP qp NP PP 3 4 N AspEvP in three days –ation 3 DP AspEv’ 3 AspEv AspQP -ation 3 DP RootP 4 ¥

Event Related Nominals

143

While for eventive -er Ns, the type of the event modifying PP mismatches that of the nominal, which prevents PM from deriving the compositional meaning of the nominal complex: (45)

a. *the painter of the room in three days b. *NP qp NP PP 3 4 N AspEvP in three days –er 3 -er AspEv’ 3 AspEv AspQP 3 DP RootP 4 ¥

By definition, PM is a local compositional rule. Aspectual PPs can only be interpreted intersectively with respect to the predicate they immediately modify. Since aspectual PPs can only be interpreted intersectively with respect to the eventuality (and only the eventuality) they are directly combined with, structurally, it must correspond to the eventuality introduced by the functional projection they are directly adjoined to. Hence, modifiers adjoined at the NP level are interpreted intersectively with NP. This is only possible with AS-Ns: aspectual PPs are intersective with the noun: (46)

[[the destruction of the city by the enemy in 3 days]] a. Ȝe ‫ א‬Dv. [[destruction of the city by the enemy]] (e) = [[in 3 days]] (e) =1 b. Ȝe ‫ א‬Dv. e is a destruction of the city by the enemy and e is (completed) in 3 days.12

As already discussed, this is excluded with eventive -er Ns because of type mismatch. If we wanted to adjoin an aspectual PP in the frame in (44), adjunction (and PM) could only take place at the AspEvP level (due to the type constraints). This situation has, however, one important consequence for event nominals and for the understanding of how eventive -er Ns and process AS-Ns differ in their respective interpretation, even

144

Chapter Seven

though both involve an underlying grammatical eventuality. Very clear predictions are made in terms of interpretation of the adjuncts: the aspectual PPs can only be interpreted intersectively with respect to AspEvP, and not to the NP. In other terms, PPs will be intersective with the embedded event and not with the (complex) NP. Cases of event modifiers inside the NP are possible with -er Ns and have been noted in the literature on French nominalizations (cf., in particular Kerleroux 2007). They behave in a systematic way: they are intersective with respect to the inner event, but non-intersective (and that may include a variety of interpretations; e.g., subsective) with respect to the N itself. Consider the following example: (47)

les pêcheurs sous la glace the fishers under the ice "under-ice fishers"

(=subset of fishers) (due to Kerleroux 2007)

It has been noted that the nominal in (47) can under no circumstances describe individuals that are fishing and located under the ice. However, the important point is that the locative PP is interpreted intersectively with respect to the fishing (and not the fisher(s)): the fishing must take place under the ice, while no location is specified for the fishers. Modifiers at the VP /AspP level will always lead to a nonintersective reading for NP (i.e. a subset). Accordingly, going back to the original examples, (48) is possible but only when understood as subclasses of a (prototypical) class of nominals, and not in an event interpretation. Concretely, (48a) for instance, cannot be interpreted as the x such that x is a runner and x is in 9 seconds (because type mismatch prevents PM from applying at the NP level); but can be interpreted (with more or less pragmatic felicity; cf. (48) vs. (49)) as a subclass of runners (i.e. those than run in less than 9 seconds, with respect to a contextually determined run or running). (48)

les sprinters en moins de 9 secondes the runners in less than 9 seconds "the runners in less than 9 seconds"

(49)

a. le donneur de sang (*?en cinq minutes) the giver of blood in five minutes "the blood donor (*?in five minutes)" b. le dompteur des lions (*?pendant des années) the tamer of.the lions for many years "the lion tamer (*?for years)"

Event Related Nominals

145

To conclude, adjunct PPs (including aspectual in/for-PP) are never allowed with -er Ns, as adjunction is precluded at the NP level. If adjunction takes place it can only be at the AspEvP level, leading systematically to a subsective interpretation with respect to NP (but intersective with respect to the lower AspEvP). The subsective reading, however, is constrained by pragmatic reasons. Nominalization forms a local domain where the meaning is computed first and then further modifiers are possible, provided that they are of the right semantic type for PM to apply. Adjunction at the AspEvP level is possible but is interpreted as intersective with respect to the event only (via PM); adjunction is, in this case, not visible/accessible for the individual.13 Our results confirm what has been claimed in the literature since Grimshaw (1990); Van Hout and Roeper (1998); Borer (1999), namely that aspectual modifiers signal the presence of an underlying aspectual/verbal structure. However, the underlying event is only visible to adjuncts if the nominalization expresses a predicate of events and not of individuals. For a nominal to be built on an underlying event (hence a verbal/aspectual structure) does not equate to express a predicate of events. In that sense, there is no incompatibility between the fact that -er Ns denote individuals and the fact they, nevertheless, involve a grammatical event.

5. Extension to Simple Event nominals (SENs) This view of events inside nominals, forced by semantic compositionality, allows us in turn to gain some understanding of the Simple Event nominals (SENs) traditionally left aside as problematic cases. Recall that SENs are interpreted as related to an event (e.g., movie, meeting, concert), yet they do not pass the tests for a strong/grammatical eventuality, but only the tests for a weak/conceptual eventuality: while they do accept predicates like be interrupted, last x time, and enter in the during the N construction, they do not (i) project arguments, (ii) allow frequency adjectives in the singular, (iii) allow PP aspectual modifiers like in/for x time (cf., Table 2). Since SENs lack a grammatical event, they are not syntactically derived from a verbal/aspectual structure and we assume that they are simply formed from a root (whether they are morphologically complex as in the case of meeting or not, as in concert, movie, attack). Accordingly, they pattern structurally with R-Ns (including instrument -er Ns) rather than AS-Ns. Cf. (41).

Chapter Seven

146

(50)

NP 3 N RootP -ation/Ø 4 ¥

(SENs)

If this is correct, then the difference between SENs and AS-Ns boils down to the same semantic type difference between nominals on the one hand and nominals on the other. What we have described as grammatical vs. conceptual eventuality interacts with the type difference: conceptual eventuality is found with nominals that are of the type, whereas grammatical eventuality is found with event denoting nominals that are of type . In other words, SENs differ from AS-Ns in that they take an individual as argument (rather than an event); it just happens that for SENs that individual variable is an abstract entity, conceptually an event (rather than a concrete object as with table, book, and so on). Accordingly, SENs share properties with other R-Ns that are associated with reference to individual entities (including abstract ones): they are count and allow discrete quantification (three movies, many concerts; cf. (22a)), they do not take frequency adjectives in the singular (*the frequent concert; cf. (21)) and do not take aspectual in/for PPs (52b). They differ, however, from other R-Ns in appearing with predicates like took place, last x time which seek for an event-denoting subject. However, this is only the result of their denoting abstract conceptual events (rather than concrete entities). In terms of the nominalization, for As-Ns, (51a) and (51b) are structurally and semantically related: the nominal form in (b) is derived from the verbal form in (a) and hence they share common semantic features. By contrast, for SENs, (52a) and (52b) are only conceptually related in the sense that their roots share the same conceptual/lexical content, but do not have internal grammatical structure. (51)

a. On a construit la cathédrale (en 100 ans). one has built the cathedral in 100 years b. La construction de la cathédrale (en 100 ans) the building of the cathedral in 100 years

(52)

a. Les membres du projet se sont réunis (pendant 3 heures). the members of.the project refl have met for 3 hours b. La/une réunion des membres du projet (*pendant 3 heures) the/a meeting of.the members of.the project for 3 hours

Event Related Nominals

147

As earlier, the incompatibility of SENs with in/for PPs derives from a type mismatch between nominals and type adjuncts. Nothing special needs to be said of SENs in that respect; nor regarding frequency modification. As expected, SENs accept frequency adjectives when in the plural only, and thus pattern with R-Ns. The frequent modifier is adjoined at the NP level and forces pluralization of the individual with type Ns. By contrast, recall that frequency adjectives do not force the plural on the nominal when they quantify over the inner event introduced by AspEvP in eventive -er Ns and AS-Ns. In sum, SENs can be straightforwardly integrated in our type semantic account as Ns and their apparently mixed properties can be derived from their semantic type and the special kind of abstract object they denote.

6. Conclusion To conclude, event-related nominals (and nominalizations) form a rather heterogeneous group that is traditionally split into three coherent classes: AS-Ns, SENs, eventive -er Ns. They share event related properties that distinguish them from noneventive nominals altogether (R-Ns). The distinct properties of the three groups of event-related nominals can be accounted for on the basis of two interacting notions of event inside nominals. On the one hand, we have argued that a fundamental difference exists between grammatical eventuality, associated to the structural projection of VP/AspP and hence a verbal syntactic base, and conceptual eventuality that is expressed lexically on roots. On the other hand, the semantic type of the outcome of nominalization turns out crucial in distinguishing nominals that denote individuals (type ) and those that denote events (type ). The two notions interact in a very straightforward way, leading to the typology of nominals in Table 3. Nominals of type come in three flavours. They may not involve any event (and thus be interpreted as concrete entities); or involve an underlying eventuality, which can be a conceptual one only (SENs) or a grammatical one (eventive -er Ns). For nominals of type the logically possible combinations are much more restricted, however. As nominals are eventive in the strong/grammatical sense (i.e., they involve an event variable that is introduced, by assumption, structurally), they require a grammatical event: . They cannot involve a conceptual event or no event at all. The attested classes of nominals are, thus, precisely the ones that we should expect in our system.

Chapter Seven

148

NP type

Underlying eventuality

Nominal form

conceptual eventuality

SENs: movie, play, concert

grammatical eventuality: AspEvP

eventive -er Ns: driver, consumer (e.g., the driver of the truck to Paris)

no eventuality

R-Ns (concrete individual entities): table; exam; boy

conceptual eventuality

**impossible

grammatical eventuality: AspEvP

process AS-Ns: destruction, examination, forming (e.g., the destruction of the city by the enemy)

no eventuality

**impossible

Table 3: Typology of eventive and noneventive nominals

References Alexiadou, A. 2001. Functional Structure in Nominals: nominalization and ergativity. Amsterdam: John Benjamins. —. 2010a. Nominalizations: a probe into the architecture of grammar. Part i: the nominalization puzzle. Language and Linguistics Compass 4: 496–511. —. 2010b. Nominalizations: a probe into the architecture of grammar. Part ii: the aspectual properties of nominalizations, and the lexicon vs. syntax debate. Language and Linguistics Compass 4: 512–523. Alexiadou, A., E. Anagnostopoulou, and M. Stavrou. 2000. Deriving words and categories. Paper presented at the 23rd GLOW colloquium, Vitoria-Gasteiz.

Event Related Nominals

149

Alexiadou, A., G. Iordăchioaia, and E. Soare. 2010. Number/aspect interactions in the syntax of nominalizations: a distributed morphology approach. Journal of Linguistics 46: 537–574. Alexiadou, A., and F. Schäfer. 2010. On the syntax of episodic vs. dispositional -er nominals. In The syntax of nominalizations across languages and frameworks, ed. by Artemis Alexiadou and Monika Rathert, 9-38. Berlin: Mouton de Gruyter. Aronoff, M. 1976. Word Formation in Generative Grammar. Cambridge, MA: MIT Press. Bach, E. 1986. The Algebra of Events. Linguistics and Philosophy 9: 5-16. Baker, M. C., and N. Vinokurova. 2009. On agent nominalizations and why they are not like event nominalizations. Language 85: 517–556. Bauke, L., and T. Roeper. to appear. How Phase-based Interpretations Dictate the Typology of Nominalizations. Ms., University of Massachusetts at Amherst. Bennett, J. 1988. Events and their Names. Indianapolis: Hackett Press. Booij, G. E. 1977. Dutch Morphology : A Study of Word Formation in Generative Grammar. Dordrecht: Foris. Borer, H. 1999. The form, the forming and the formation of nominals. Paper presented at the 2nd Mediterranean Morphology Meeting, September 1999. —. 2003. Exo-skeletal vs. endo-skeletal explanations. In The Nature of Explanation in Linguistic Theory, ed. by J. Moore, and M. Polinsky, 31-67. Chicago: CSLI and University of Chicago Press. —. 2005. Structuring Sense, vol. II: The Normal Course of Events. Oxford: Oxford University Press. —. 2012. In the Event of a Nominal. In The Theta System. Argument Structure at the Interface, ed. by M. Everaert, M. Marelj, and T. Siloni, 103-149. Oxford: Oxford University Press. Chomsky, N. 1970. Remarks on nominalization. In Readings in English Transformational Grammar, ed. by Roderick A. Jacobs and Peter S. Rosenbaum, 184-221. Waltham, MA: Ginn. —. 2008. On Phases. In Foundational Issues in Linguistic Theory, ed. by R. Freidin, C. Otero and M. L. Zubizarreta, 133-166. Cambridge, MA: MIT Press. Di Sciullo, A-M., and E. Williams. 1987. On the Definition of Word. Cambridge, MA: MIT Press. Gehrke, B., and L. McNally. 2012. Frequency adjectives as distributional modifiers. Submitted Ms, University of Pompeu Fàbra. Giorgi A., and F. Pianesi. 2001. Ways of Terminating. In Semantics Interfaces: Reference, anaphora and aspect, ed. by C. Cecchetto, G.

150

Chapter Seven

Chierchia, and M. T. Guasti, 211-277. Stanford, CA: CSLI Publications. Grimshaw, J. 1990. Argument Structure. Cambridge, MA: MIT Press. Haas, P., R. Huyghe, and R. Marìn. 2008. Du verbe au nom: calques et décalages aspectuels. In Actes du Congrès Mondial de Linguistique Française. Paris: 2039-2053. Halle, M. 1973. Prolegomena to a Theory of Word Formation. Linguistic Inquiry 4: 3-16. Heim, I., and A. Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Kerleroux, F. 2007. Pêcheurs sous la glace. In Hommages en l’honneur de Marc Plénat, ed. by L. Labrune and E. Delais. Kornfilt, J., and J. Whitman. 2011. Afterword: Nominalizations in syntactic theory. Lingua 121(7)-17: 1297-1313. Larson, R. K. 1998. Events and modification in nominals. In Proceedings of Semantics and Linguistic Theory (SALT) VIII, ed. by D. Strolovitch and A. Lawson, 145–168. Ithaca, NY: CLC Publications, Department of Linguistics, Cornell University. Lees, R. 1960. The Grammar of English Nominalizations. The Hague: Mouton. Marantz, A. 1997. No escape from syntax: don’t try morphological analysis in the privacy of your own lexicon. In Proceedings of the 21st Annual Penn Linguistics Colloquium, ed. by A. Dimitriadis, L. Siegel, C. Surek-Clark, and A. Williams., vol. 4 of Penn Working Papers in Linguistics, 201–225. University of Pennsylvania. Mourelatos, A. P. D. 1978. Events, Processes and States. Linguistics and Philosophy 2:3: 415-434. Rappaport Hovav M., and B. Levin. 1992. –er nominals: Implications for a theory of argument structure. In Syntax and Semantics: Syntax and the Lexicon, ed. by T. Stowell and E. Wehrli, vol. 26: 127-153. New York, NY: Academic Press. Roy, I., and E. Soare. To appear. Naming participants in the eventuality. Submitted Ms., University Paris 8. Snyder, W. 1998. On the aspectual properties of English derived nominals. In The Interpretive Tract: Working Papers in Syntax and Semantics (MITWPL), ed. by U. Sauerland and O. Percus, vol. 25:125–139. Cambridge, MA: MIT Working Papers in Linguistics. van Hout, A., and Thomas R. 1998. Events and aspectual structure in derivational morphology. In Papers from the Upenn/MIT Roundtable on Argument Structure and Aspect, ed. by H. Harley, vol. 32 of MIT Papers in Linguistics: 175–200. Cambridge, MA.

Event Related Nominals

151

—. 2011. Preposing in Nominalizations, Aspect, and Phases. Paper presented at the 4th Workshop on Nominalizations (JeNom4), University of Stuttgart, Germany, June 16-17, 2011. Vendler, Z. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press.

Notes 1

We thank the audience at the CASTL Workshop Categorization and Category Change in Morphology (University of Tromsø, dec. 2011), two anonymous reviewers, Bridget Copley, Florian Schäfer, and Hagit Borer for their helpful comments and discussion. We gratefully acknowledge support from the Program Structure Argumentale et Structure Aspectuelle–CNRS, Fédération TUL. 2 We use the following abbreviations: ev–event; ¥–Root; gen–Genitive, refl– reflexive pronoun. 3 We use the terms ‘event’ and ‘eventivity’ in a loose sense that does not discriminate between events properly speaking and states, and interchangeably with the term ‘eventuality’ introduced by Bach (1986). 4 There is an interpretation for (16b) in which lasted 3 hours applies to the lifetime of the subject (here a concrete entity) and not to an event duration per se. Nevertheless, the conclusion that R-Ns are different from both SENs and AS-Ns is supported by their respective behavior with the other predicates, namely take place, be interrupted, among others. 5 Similarly, the literature invokes the subject position of occur, start or finish (see e.g. Vendler 1967, Snyder 1998 among many others). We take all these tests as broadly indicating semantic eventivity in nominals, and do not broach on further semantic distinctions like the one between events and propositions, pointed at in the literature. 6 As pointed out to us by a reviewer, SENs are compatible with another type of apparently temporal modifiers, namely of x time. Structurally, this modifier is not a VP modifier, however, and involves clear nominal modification; e.g. a meeting/concert/movie of three hours in length; we leave it aside at it is not relevant to the question of an internal grammatical event. 7 Recent works on pluralization have shown that AS-Ns may sometimes accept plurals and discrete quantification when the nominals denote a bounded event; Borer (2005); Alexiadou et al. (2010). What is important here is that no such requirement is made for SENs. 8 As discussed in the literature, frequency adjectives are often ambiguous between (at least) three readings: internal, adverbial and generic interpretations (cf., in particular, Gehrke and McNally 2012). Typical cases of frequency adjective + N, that do not involve argument structure (e.g., an occasional sailor, an occasional beer) are generic or adverbial. We are not concerned with these cases here, and refer the reader to Roy and Soare (to appear) for further discussion.

152

Chapter Seven

9 As specified in Roy and Soare (to appear), the (b) example is ungrammatical under the right, i.e., event related, interpretation and in argumental position. Predicative uses generally seem to render the eventive reading accessible; compare I bought a big grinder (# "a tiny machine that grinds much") vs. This is a big grinder (ok: "a tiny machine that grinds much"). Predication may, however, contribute its own eventuality; for that reason we focus our discussion on argumental nouns exclusively. 10 For the syntactic derivation of -er nominals we assume the results of Roy and Soare (to appear). 11 In type semantics, e stands for entity; v stands for events. We use the following types: for predicates of individuals, for predicates of events, and for a predicate with an open argument position. 12 Preposition in gives the duration of the agent action that brings about the telos, for a telic predicate (Giorgi and Pianesi 2000) 13 A phase-based approach to nominalization seems relevant to account for the domains of adjunction and interpretation (see, Chomsky 2008; van Hout and Roeper 2011, Bauke and Roeper (to appear)). The details of such an analysis would need to be worked out and we leave a proper implementation open for future research.

CHAPTER EIGHT ‘DIRECT PARTICIPATION’ AND ‘AGENT EXCLUSIVITY’ EFFECTS IN DERIVED NOMINALS AND BEYOND ARTEMIS ALEXIADOU UNIVERSITÄT STUTTGART

GIANINA IORDĂCHIOAIA UNIVERSITÄT STUTTGART

FABIENNE MARTIN UNIVERSITÄT STUTTGART

FLORIAN SCHÄFER UNIVERSITÄT STUTTGART

MARIÁNGELES CANO UNIVERSIDAD AUTÓNOMA DE MADRID

1. Introduction1,2 It has been noted that nominalizations show certain restrictions on the realization of their external argument compared to their verbal sources (see Grimshaw 1990, Pesetsky 1995, Marantz 1997, Harley and Noyer 2000, Sichel 2010/2011). In the English examples below, Causers or natural forces are disallowed (1b) although the corresponding verbal forms allow them as external arguments (1a). Human agents, on the other hand, are good in both verbal and nominal contexts (1a, c). (1)

a. The authorities/the hurricane justified the evacuation of the inhabitants

154

Chapter Eight

b. *The justification of the evacuation of the inhabitants by the hurricane c. The justification of the evacuation of the inhabitants by the authorities While initially this was indeed described as an 'agent exclusivity effect', Sichel (2010, 2011) shows that the restriction in English nominals is of a different sort: while verbal forms allow both direct and indirect participants as external arguments, derived nominals require direct participation, which means that the external argument has to be co-temporal and co-spatial with the unfolding event. Following Sichel, we call this the direct participation effect. While human agents can be construed as direct participants quite easily, causers often cannot. Crucially, however, causers sometimes do qualify as direct participants and then they are good as external arguments of nominals (cf. Folli and Harley’s (2008) notion of teleological capability). Note, however, that for certain nominalizations the effect is indeed one of agent exclusivity, e.g. psych nominalizations in English (Pesetsky 1995, Grimshaw 1990, Alexiadou and Iordăchioaia to appear). We consider here the agent exclusivity effect a more restricted version of the direct participation effect; see also Sichel (2011) who notes that agents are some sort of default direct participants. In this paper, we first discuss a lexical-semantic hypothesis proposed by Sichel (2011), who argues that the restriction is related to the size of events that fit into derived nominals. These are necessarily simple events, she argues, and simple events can only combine with direct participants. Based on cross-linguistic evidence from English, German, Greek, Romanian, Spanish and French, but also non-Indo-European languages such as Hebrew and Jacaltec, we argue that Sichel's hypothesis cannot account for the direct participation effect properly. To present here one argument, not only don't we find the effect in all nominalizations across languages, we even find it in the verbal/sentential domain.3 The paper is organized in the following way: In Section 2 we discuss in more detail the direct participation effect in English. Section 3 presents the lexical-semantic account based on event complexity proposed by Sichel (2010, 2011). In Section 4 we discuss some aspect of English nominalizations that we think make Sichel's account problematic. In Section 5, we provide a cross-linguistic landscape of the distribution of the direct participation effect. We show that one has to differentiate between a direct participation effect and an agent exclusivity effect. Next, we show that neither of the two effects is strongly correlated to nominalizations; on the one hand, we find languages where the (smallest) nominalization does

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

155

not show any such restriction on the external argument and, on the other hand, we find languages where even verbal/sentential constructions such as passive and even active clauses show such a restriction on external arguments. In section 6, we outline our analysis.

2. The direct participation effect in English derived nominals As is well known, English nominalizations are morpho-syntactically deficient by comparison to their verbal sources. For instance, they do not allow ECM, double objects, and particle shift (Kayne 1984, Abney 1987), that is, constructions that have been argued to require a VP shell structure (Larson 1988, Harley and Noyer 1998). As the examples (2-5) show, the effects hold for both derived nominals and ing-of gerunds, but not for verbal gerunds. (2) (3)

a. *John’s belief / believing of [Bill to be Caesar] b. John’s believing [Bill to be Caesar] a. *John’s gift /rental /giving (of) Mary of a Fiat b. John’s giving/renting Mary a fiat

ECM Double objects

(4)

a. *John’s persuasion / persuading of Mary [PRO to stay] b. John’s persuading Mary [PRO to stay] Object Control

(5)

a. *John’s explanation (away) of the problem (away) b. John’s explaining (away) of the problem *(away) c. John’s explaining (away) the problem (away) Particle-Shift

The restriction on external arguments that we are interested in here, makes, however, a different cut within nominal structures and, therefore, cannot be related to the morpho-syntactic deficiencies above. While these deficiencies hold for both derived nominals as well as ing-of gerunds, only the former restrict their external arguments to a subset of the external arguments found with the corresponding verb, while ing-of (as well as verbal) gerunds allow the same set of external arguments as their verbal base form (6a-c). (6)

a. The boss/The inflation shrinks his salary b. ?The shrinking of his salary by the inflation/by the boss c. The shrinkage of his salary by the boss /*by the inflation

Chapter Eight

156

While (6a-c) might suggest that nominalizations do not allow causers as external arguments, Sichel (2010, 2011) shows that the relevant constraint does not amount to a strict agent exclusivity effect. This is illustrated by the data in (7-9) (see Sichel (2010, 2011) for more examples) where the accomplishment verbs underlying the nominalizations allow both agents and (direct)4 causers as external arguments. While clear-cut agents are indeed always possible in derived nominalizations (e.g. 9d-e), natural forces are sometimes good (7b-c) and sometimes bad (8b-c, 9b-c). As the examples show, pre-nominal genitives and by-phrases show the same thematic restriction. However, we will focus below on the distribution of by-phrases, as the other languages we will discuss don’t permit transitive nominals with pre-nominal genitives for independent reasons. (7)

a. The hurricane destroyed all our crops b. The hurricane’s destruction of our crops c. The destruction of our crops by the hurricane

(8)

a. The approaching hurricane justified the abrupt evacuation of the inhabitants b. *The approaching hurricane’s justification of the abrupt evacuation of the inhabitants c. *The justification of the abrupt evacuation of the inhabitants by the hurricane d. The authorities justified the rapid evacuation of the inhabitants e. The authorities’ justification of the rapid evacuation of the inhabitants f. The justification of the rapid evacuation of the inhabitants by the authorities

(9)

a. b. c. d. e.

The results/the expert verified the initial diagnosis *The results’ verification of the initial diagnosis *The verification of the initial diagnosis by the results The expert’s verification of the initial diagnosis The verification of the initial diagnosis by the expert

Sichel observes that non-human causers are compatible with derived nominalizations only if they can be construed as direct participants, i.e. forces, which directly bring about the event. This is the case in (7) where the hurricane is the force that directly brings about the destruction of the city. But it is not the case in (8) and (9). While in (8) the hurricane might be understood as the direct cause for the justification event and as such can

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

157

occur as the subject of the verb 'justify' it is not a direct participant in the relevant sense because the act of justification always involves a human agent as most direct participant. Similarly, in (9), the results can be understood as direct causes for the process as verification, but such a process implies an agent as most direct participant. Direct participation is then a stronger restriction than direct causation. Direct participation depends on both the specific event denoted by the nominal and some property inherent to the entity denoted by the external argument (teleological capability, Folli and Harley 2008). A hurricane's inherent force allows it to be understood as a direct participant in a destruction event, but a hurricane has no inherent property, which can make it a direct participant in a justification event. Human agents, on the other hand, typically qualify as direct participants due to their inherent properties such as intentionality or volition.

3. An account based on event complexity In order to account for the difference between nominalizations and ingof gerunds concerning the direct participation effect, Sichel (2010, 2011) proposes the following: Derived nominals of accomplishment verbs do not inherit the event structure of their source verb. While accomplishments are complex events (Grimshaw 1990), their derived nominals denote simple events and, as such, require direct participation of the external argument, which must be co-temporal and in some sense co-spatial with the unfolding event (see below). Ing-of gerunds denote the same complex events as their underlying accomplishment verbs, and as such do not require co-temporal direct participants. The restriction to direct participation for the external argument is derived from a general restriction to simple events. In order to participate in the simple event denoted by the nominalization, the event associated with the external argument (the causing event) has to identify with the simple event. The following restrictions hold on event identification: (10)

Conditions on event identification (Levin and Rappaport Hovav 1999, 2002, Rappaport Hovav and Levin 2001) I. The sub-events must have the same location and are necessarily temporally dependent.

158

Chapter Eight

II. One sub-event must have a property that serves to measure out that sub-event in time; this property is predicated of an entity that is necessarily a participant in both sub-events. This ensures temporal dependence. These restrictions have originally been developed to characterize the difference between bare XP resultatives and lexical causatives (as well as NP-XP resultatives). Bare XP resultatives as in (11) denote simple events in that the activity described by the verb is co-temporal with progress towards the achievement of the result and the sole argument is shared by both sub-events. These simple events derive from the identification of the verbal process with the coming about of the result expressed by the XP following the verb. English lexical causatives as in (12), on the other hand, denote complex events, since the two sub-eventualities (process and result) can be temporally distinct. (11)

a. Carey ran/waltzed out of the room b. The clothes steamed dry c. The kettle boiled dry

(12)

a. Casey's piano playing woke the baby b. Terry shocked Sandy by deciding to run for office c. The widow murdered the old man by putting poison in his soup

The difference between nominalizations and ing-of gerunds can be understood in a similar way, just that this time the causing event associated with the external argument and the (allegedly) simple event denoted by the nominalization have to identify. The two events have to be co-temporal and the argument shared by these sub-events is the external argument. The good and bad nominalizations in (7-9) above can be understood then in the same way as the contrasts in (13a-c): The agent subject 'the teacher' in (13a) can be understood as a direct participant which takes part in the causing sub-event and in the event of separation; the two events are co-temporal and can identify. The event expressed by the causer subject the war in (13b), on the other hand, is understood as a non-co-temporal cause of the separation process and event identification fails. In (13c), on the other hand, the very same causer subject can be understood as co-temporal with the destabilization process.

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

(13)

159

a. The separation of Jim and Tammy Fay by the teacher b *The separation of Jim and Tammy Fay by the war c. The de-stabilization of the economy by the war

Sichel concludes that the relation between the event associated with the external argument and the event denoted by the derived nominal is subject to the condition in (14a) and its corollary in (14b): (14)

a. If a simple event includes an external argument, the participation of the argument is co-temporal with the initiation of the event. b. Corollary: when the participation of the external argument is not co-temporal the event is a complex event.

4. Some problems for an account in terms of event complexity Sichel does not offer an explicit syntax-event structure mapping, and phrases her account in the framework of Rappaport Hovav and Levin (2001). From this perspective, unlike their verbal source, derived nominals of accomplishment verbs have a simple event structure, e.g. [x ACT ]. By contrast, ing-of gerunds maintain the complex event structure of the verb, e.g. [[x ACT] CAUSE [BECOME [y ]]]. In syntactic accounts of event structure such as e.g. Ramchand (2008) and Harley (2011), the former have the structure of activities, i.e. lack a Result State component and simply contain a Process/v head, while the latter contain such a component [v + ResultP], as in (15a vs. 15b): (15)

a. [ProcessP/vP ] b. [ProcessP/vP [ResultP]]

However, the morpho-syntactic composition of English nominals does not seem to match the semantic interpretation attributed to them in most of the cases. Following Harley (2011), e.g. justify in (1) contains the verbalizing affix -ify. This combines with the verbal root that realizes a semicompositional result head (16). (16)

vP v ify

ResultP just

160

Chapter Eight

Similar observations hold for many of the other verbs discussed in Sichel’s work, e.g. unify, separate, verify, destabilize, devastate, which thus always host complex events. Furthermore, derived nominals permit re-prefixation as shown in (17). Crucially, however, it has been argued in the literature that re- only has a restitutive interpretation which means that it attaches to accomplishments, and requires a result state; Wechsler (1990), Marantz (2009/2011). (17)

a. the re-verification of the diagnosis b. a re-justification of former notations

Finally, while Sichel clearly shows that the causing event associated with the external argument has to identify with the process event denoted by the derived nominal, she does not show how the two sub-events denoted by the underlying accomplishment verb (process and result) identify in derived nominals. That is, Sichel's proposal suggests that, independently from the choice of the external argument, the event denoted by an accomplishment verb should differ substantially from the event denoted by the nominal derived from that accomplishment. Accomplishments are made up by two sub-events, a process-event and an unfolding result. Crucially, these two sub-events are not (at least not necessarily) co-temporal and, therefore, cannot be identified. Levin and Rappaport (1999: 32) show this with the following example, where "by putting arsenic in his coffee" modifies the process sub-event contributed by murder. As they note, in (18) "the putting of arsenic in the coffee certainly does not extend to the point of death, and the dying does not start when arsenic is put in the coffee", so the two sub-events do not identify. (18)

The widow murdered the old man by putting arsenic in his coffee

The proposal that derived nominals can only contain simple events, i.e. events where the process event and the coming about of the result are cotemporal to make event identification available, suggests that examples like (18) could not be transformed into derived nominals. While 'murder' does not form an -ation derived nominal (Grimshaw (1990) in fact argues that zero derived nominals in English lack argument structure), assassinate does. It seems to us that in the nominal the process and the result do not have to be co-temporal, exactly as in its verbal source. (19)

a. John assassinated the president by putting arsenic in his coffee

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

161

b. The assassination of the president by putting arsenic in his coffee To conclude then, while we think that the direct participation effect on derived nominals is the correct generalization and can be captured by assuming that the event or the action associated with the external argument (causer or agent) has to identify with the verbal process, we do not think that the inner aspectual event decomposition shifts from that of an accomplishment to that of a process in derived nominals. In the next section we will present further challenges for Sichel's account, based on a preliminary study of the cross-linguistic distribution of the direct participation effect.

5. Nominalizations, active and passive Voice across languages 5.1 The direct participation effect in nominalizations While the morpho-syntactic restrictions on nominalizations in (2-5) and the direct participation effect found with -ation nominals cannot have the very same explanation, on Sichel's account both effects seem to be connected to the intuition that nominalizations are less verbal than the verbs they are derived from. One way to look at this is to say that the nominalizations contain fewer verbal projections than their base verbs. Some of the missing projections would then be responsible for the morpho-syntactic restrictions in (2-5), see Harley and Noyer (1998), and other missing projections might be responsible for the direct participation effect. However, it turns out that this general idea is not supported if we look at a larger set of languages. To this end, we systematically investigated direct participation effects in five further languages, German, Greek, Romanian, Spanish and French. Nominalizations in Greek (20), German (21) and French (22) do not show the direct participation effect. Recall that the corresponding examples are out with English 'verification'. (20)

I epivevosi tis arhikis diagnosis apo/me ta the verification of.the initial diagnosis by/with the apotelesmata tis eksetasis results of.the test “The verification of the initial diagnosis by the results of the test”

162

Chapter Eight

(21)

Die Bestätigung der ursprünglichen Diagnose durch die the confirmation of.the initial diagnosis by the Ergebnisse des Tests results of.the test “The confirmation of the initial diagnosis by the results of the test”

(22)

La vérification du diagnostic initial par les résultats du test the verification of.the diagnosis initial by the results of.the test “The verification of the initial diagnosis by the results of the test”

It could be argued that not all nominalizations are restricted to host simple events. It could be that the nominalizations in (20)-(22), just as English ing-of gerunds, are big enough to host complex events. However, this does not seem to be the case at least for the German and French data in (21) and (22) which involve the smallest nominalizations available in these languages. That is, both languages have a further, more verbal nominalization more akin to the English ing-of gerunds (see Alexiadou, Iordăchioaia and Schäfer (AAS 2011)). In fact, only Greek differs from the other languages in this paradigm as it seems to only have one nominalization type, and in principle this could be more like ing-of gerunds (Alexiadou 2001, 2009). To conclude, the smallest nominalization in a given language does not always show the direct participation effect. The smallest available nominalizations in Romanian (23) and Spanish (24), on the other hand, show the direct participation effect. (23)

justificarea măsurilor de austeritate de către Emil Boc/ justify.Inf.the measures.Gen of austerity by to Emil Boc/ *ratingul scăzut al SUA rating.the down-graded of USA “The justification of the austerity measures by Emil Boc/by the down-graded rating of the USA”

(24)

La verificación de los diagnósticos iniciales por el experto/*por los The verification of the initial diagnosis by the expert /by the resultados results “The verification of the initial diagnosis by the expert/ by the results”

In Romanian, the preposition de către ('by', lit. 'by to') can only introduce agents/direct participants as external arguments in nominalizations, i.e.,

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

163

the restriction is similar to that of the English by-phrase: the wind, the hurricane and insecticides are direct participants in (25b, d), but the bad weather and alcohol are indirect participants in (25a, c) and as thus they are excluded from the nominalizations.5 (25)

a. *schimbarea planurilor de weekend de către vremea rea change.Inf.the plans.Gen of weekend by to weather bad “The bad weather's changing our plans for the weekend” b. schimbarea pozi‫܊‬iei pietrelor de către vânt change.Inf.the position stones.Gen by to wind “The wind's changing the position of the stones” c. *distrugerea lui Amy Winehouse de către alcool6 destroy.Inf.the of Amy Winehouse by to alcohol “Alcohol's destruction of Amy Winehouse” d. distrugerea recoltei de către uraganul puternic/ destroy.Inf.the crops.Gen by to hurricane.the strong/ insecticide insecticides “The hurricane's/the insecticides' destruction of the crops”

In Spanish derived nominals, por introduces agents/direct participants, Picallo (1991, 1999), Varela (2012). Therefore, the external argument is more restricted in nominalizations than in the verbal clauses. In (26c-d), por el huracán is a direct participant; this is not the case in (26b). (26)

a. El huracán justificó la evacuación de los habitantes the hurricane justified the evacuation of the inhabitants “The hurricane justified the evacuation of the inhabitants” b. *La justificación de la evacuación de los habitantes por el the justification of the evacuation of the inhabitants by the huracán hurricane “The justification of the evacuation of the inhabitants by the hurricane” c. El huracán destruyó nuestros cultivos the hurricane destroyed our crops “The hurricane destroyed our crops” d. La destrucción de nuestros cultivos por el huracán the destruction of our crops by the hurricane “The destruction of our crops by the hurricane”

164

Chapter Eight

While derived nominals in Romanian and Spanish behave like their English counterparts, both languages differ from English in important respects. On the one hand, the Romanian supine, i.e. the nominalization that is more similar to English ing-of gerunds (AAS 2011, Cornilescu 2001, Iordăchioaia 2008) also shows the direct participation effect. (27)

a. *schimbatul frecvent al planurilor noastre de weekend de change.Sup.the frequent of plans.Gen ours of weekend by către vremea rea to weather bad “The bad weather's frequently changing our plans for the weekend” b. ?schimbatul frecvent al pozi‫܊‬iei pietrelor de către change.Sup.the frequent of position.Gen stones.Gen by to vântul puternic wind.the strong “The strong wind's frequently changing the position of the stones” c. *distrusul multor vedete de către alcool destroy.Sup.the many.Gen stars by to alcohol “Alcohol's destruction of many stars” d. distrusul frecvent al recoltei de către urganele destroy.Sup.the frequent of crops.Gen by to hurricanes.the din această zonă from this area “The frequent destruction of the crops by the hurricanes in this area”

The situation in Spanish also differs from the one in English. While, as we have seen, Spanish nominalizations restrict their external arguments to direct participants, the verbal infinitive (the nominalization with the most verbal properties in this language, AAS 2011) restricts its external argument to human agents, i.e. it shows a human exclusivity effect. (28)

a. El justificar las autoridades la evacuación the justify.Inf the authorities.Nom the.Acc evacuation “The justification of the evacuation by the authorities” b. El destruir los soldados la ciudad the destroy.Inf the soldiers.Nom the.Acc city “The destruction of the city by the soldiers”

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

165

c. *El justificar el huracán la evacuación the justify.Inf the hurricane.Nom the.Acc evacuation “The justification of the evacuation by the hurricane” d. *El destruir el huracán el puente the destroy.Inf the hurricane.Nom the.Acc bridge “The destruction of the bridge by the hurricane” So while the small nominalization in Spanish shows a direct participant effect, the nominalization with many verbal properties shows an agent exclusivity effect. An agent exclusivity effect can also be found in German. So while the German nominalization in -ung does not restrict its external argument, the nominalized infinitive shows an agent exclusivity effect (Alexiadou et al. 2009). (29)

a. um ein Zerstören der Stadt durch die Soldaten in-order-to a destroy.Inf of-the city through the soldiers zu verhindern to prohibit “In order to prohibit a destruction of the city by the soldiers” b. ?* um ein Zerstören der Stadt durch den Sturm in-order-to a destroy.Inf of-the city through the storm zu verhindern to prohibit “In order to prohibit a destruction of the city by the storm”

Finally, an agent exclusivity effect has actually also been observed in a specific subclass of English nominalizations, namely nominalizations derived from object experiencer verbs. While these verbs allow both agents and causers as external argument, only agents are acceptable in the nominalizations (see Pesetsky 1995, Doron 2003, Alexiadou and Iordăchioaia to appear): (30)

Mary/the event annoyed/amused/embarrassed John

(31)

a. *The event’s annoyance of John b. *The event's amusement of the children c. *The event's embarrassment of the children

(32)

a. ?Mary’s deliberate annoyance of John b. ?Mary’s deliberate amusement of the children c. ?Mary’s deliberate embarrassment of the censors

Chapter Eight

166

To summarize the discussion so far, nominals sometimes but not always restrict their external argument compared to the verbal base. This restriction comes in two variants, either as a direct participant effect or as an agent exclusivity effect. Furthermore, the two restrictions are in no obvious way correlated with the size (or the number of verbal properties) of the nominalization.

5.2 Beyond nominalizations: the direct participation effect in active and passive Voice The idea that a construction is in some sense too small to host complex events and, therefore, the type of external argument is restricted is challenged by the observation that we find restrictions on the type of external argument even in the verbal/sentential domain. In Romanian, the verbal passive shows a direct participant effect if the external argument is introduced by the preposition de către. Since the simple preposition de can also introduce indirect participants, it seems as if this specific preposition (which is the only device used to introduce external arguments in the nominal domain) is responsible for this restriction. Consider the following active/passive pairs:7 (33)

a. Vremea rea ne-a schimbat planurile de weekend weather bad us-has changed plans.the of weekend “The bad weather changed our plans for the weekend” b. Planurile de weekend ne-au fost schimbate de/*de către plans of weekend us-have been changed by/by to vremea rea weather bad “Our plans for the weekend were changed by the bad weather”

(34)

a. Vântul a schimbat pozi‫܊‬ia pietrelor wind.the has changed position.the stones.Gen “The wind changed the position of the stones.” b. Pozi‫܊‬ia pietrelor a fost schimbată de/?de către vânt position.the stones.Gen has been changed by/by to wind “The position of the stones was changed by the wind”

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

(35)

a. Alcoolul a distrus-o pe Amy Winehouse alcohol.the has destroyed-her Acc Amy Winehouse “Alcohol destroyed Amy Winehouse” b. Amy Winehouse a fost distrusă de/*de către alcool Amy Winehouse has been destroyed by/by to alcohol “Amy Winehouse was destroyed by alcohol”

(36)

a. Uraganul a distrus recolta hurricane has destroyed crops.the “The hurricane destroyed the crops” b. Recolta a fost distrusă de/de către uragan crops.the has been destroyed by/by to hurricane “The crops were destroyed by the hurricane”

167

The Spanish ser (eventive) passive always shows a direct participant effect, as shown by the following active-passive pairs. (37)

a. El huracán justificó la evacuación de los habitantes the hurricane justified the evacuation of the inhabitants “The hurricane justified the evacuation of the inhabitants” b. *la evacuación de los habitantes fue justificada por el huracán the evacuation of the inhabitants was justified by the hurricane “The evacuation of the inhabitants was justified by the hurricane”

(38)

a. El huracán destruyó nuestros cultivos the hurricane destroyed our crops “The hurricane destroyed our crops” b. Nuestros cultivos fueron destruidos por el huracán our crops were destroyed by the hurricane “Our crops were destroyed by the hurricane”

Since the Spanish passive uses only one preposition, the question arises whether we should relate this effect to the passive itself or whether we should derive it from a restriction on the preposition as we proposed for Romanian de către. It seems that the former solution is correct as the very same preposition can introduce also indirect participants in other constructions. This is the case with anticausative and inchoative verbs in (39) where the por-phrase clearly introduces indirect participants (see AAS 2006 for further discussion).

168

(39)

Chapter Eight

a. La comida se conservó por la sal the food Rf preserved by the salt “The food preserved through/from the salt” b. La madera se pudrió por la lluvia the wood Rf rotted by the rain “The wood rotted through/from the rain”

Note that in Romanian the preposition de către is out in anticausatives and instead either simple de or the complex preposition de la is used to introduce causers. This is compatible with the idea that de către is lexically restricted to direct participants. (40)

a. U‫܈‬a s-a deschis de (la)/*de către vânt door.the Rf-has opened by (at)/ by to wind “The door opened from/*by the wind” b. Lemnul a putrezit de ??(la)/*de către ploaie wood.the has rotted by (at)/ by to rain “The wood rotted from/*by the rain”

Finally, the Greek passive has been reported to show the stronger restriction on external arguments, i.e. the agent exclusivity effect (AAS 2006). We exemplify this restriction on the basis of de-adjectival verbs (see AAS 2009). This class of verbs gives us a clear contrast between the passive and the anticausative pattern: the former bears non-active morphology, while the latter has active morphology. Only causer PPs are licit in the anticausative, while only agent PPs are licit in the passive (Hebrew is similar; see Doron 2003, Alexiadou and Doron 2012). (41)

a. To pukamiso katharise me to plisimo / apo mono tu / the shirt cleaned-Act with the washing/ by itself / *apo to Jani by the John “The shirt cleaned with the washing / by itself / by John” b. to pukamiso katharistike apo to Jani / *me to plisimo the shirt cleaned-Nact by the John / with the washing “The shirt cleaned by John/ with the washing”

Recall in this connection that the Greek nominalization is not restricted at all. We exemplify this again with the same verb in a passive and a corresponding nominalization. As AAS (2009) observed, we find passive nominals of verbs that lack a verbal passive: the example in (42) only has

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

169

an anticausative but not a passive interpretation, as the licensing of PPs shows. The corresponding nominalization, surprisingly, allows both agent and causer PPs, introduced by ‘apo’ (43): (42)

To dasos kaike apo ti zesti / *apo to Jani the forest burnt-Nact from the heat / *by the John “The forest burnt from the heat / by John”

(43)

to kapsimo tu dasus apo ton Jani / apo ti zesti the burning the forest-Gen by the John / from the heat “The burning of the forest by John/ from the heat”

Finally, even active clauses in some languages can show the restrictions discussed here. Craig (1976) reports what looks like an agent exclusivity effect for Jacaltec, a Mayan VSO language spoken in Guatemala. While subjects of intransitive verbs may be animate as well as inanimate, subjects of transitive verbs are restricted to animate Agents (44a vs. b). Inanimate Causers must be introduced via a preposition combining with the anticausative variant of the verb as in (44c). (44)

a. speba naj te’ pulta close he cl. door “He closed the door” b. *speba cake te’ pulta close wind cl. door “The wind closed the door” c. xpehi te’ pulta yu cake closed cl. door by wind “The wind closed the door” (lit.: The door closed by the wind)

A direct participant effect has been reported (though under a different name/characterization) for Hebrew (Doron 2003).8 While the subject of an active clause in the simple template can be either direct or indirect participant, the subject of an active clause in the intense template needs to be a direct participant.9 Very often this looks like an agent exclusivity effect as in the examples below. (45)

ha-menahel/ macavo ha-bri’uti patar et dani the director/ the state of his health excused-Simpl Acc Dani me-ha-’avoda from the job “The director / the state of this health excused Dani from the job”

170

(46)

Chapter Eight

ha-menahel/ *macavo ha-bri’uti piter et dani the director/ * the state of his health excused-Intns Acc Dani me-ha-’avoda from the job “The director/ the state of his health excused Dani from the job”

However, the decisive notion is not agentivity or animacy. In both sentences in (47), the subject is inanimate. The simple verb produce in (47a) has a reading where the protein is the trigger for antibodies being produced (an indirect participant in our terms). The intensive-template verb in (47b) can only be interpreted such that the protein actually participates in the production process itself (a direct participant). (Note that Doron does not characterize these data in terms of direct/indirect participation but in terms of the contrast agent of an action vs. causer where the agent of an action is not necessarily animate.) (47)

a. ha-xelbon yacar ba-guf nogdanim the protein produced-Simpl in the body antibodies “The protein produced antibodies in the body.” b. ha-xelbon yicer ba-guf nogdanim the protein produced-Intns in the body antibodies “The protein manufactured antibodies in the body.”

To conclude this section, both the direct participant effect and the agent exclusivity effect are not restricted to the nominal domain but they can also be found in the verbal/sentential domain, both in active as well as passive clauses.

6. Towards an account To summarize our discussion so far, the above facts challenge Sichel's account in terms of event complexity, as, in order to account for this distribution of the direct participant restriction, one would have to argue that passivization in e.g. Romanian and Spanish involves a shift in the event structure of the predicate, i.e. an accomplishment should become a simple event under passivization, a point that would be rather hard to provide evidence for. The same would have to be assumed for the Jacaltec active or the Hebrew intensive template (see also fn. 3). We thus conclude that an account in terms of event complexity cannot be on the right track. In addition, the cross-linguistic investigation of nominalizations showed us that an account in terms of small vs. big structures cannot work either, as

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

171

the effect is found in both small and more clausal like nominalizations. Finally, the Romanian data show that occasionally the effect is related to the preposition used: if a language possesses a lexically restricted preposition, the effect will arise independently of the type of construction (e.g. the effect is observed both in the verbal passive as well as in all types of derived nominals that use this preposition). The only correlation that can thus be established is that specific constructions across languages show or do not show the effect, and this is subject to parametrization in the sense that it will not be the same construction across languages that will show the effect. What is then the locus of variation? Following most of the recent literature, we adopt the Borer-Chomsky Conjecture in (48), label due to Baker (2008): (48)

All parameters of variation are attributable to differences in the features of particular items (e.g., the functional heads) in the lexicon.

In the next section we first discuss our assumptions about the decomposition of verbs. Afterwards we turn to the question how specific Voices such as active Voice, passive Voice but also nominalizations which we treat as specific Voices, can be parametrized to bring about agent exclusivity and direct participation effects.

6.1. Verbal decomposition We assume several functional heads involved in the encoding of verbal meaning. The eventive verbal predicate itself is decomposed into two layers, a v head, which is a verbalizer bringing about event implication, and a root, as in (49) (see Marantz 2005, Alexiadou, Anagnostopoulou and Schäfer 2006). (49) is the structural representation of an unaccusative/ anticausative predicate where the root expresses a result state, e.g. the door opened.10 (49)

[vP v [ Ҁopen DP ]]

The external argument is severed from the verbal predicate and is introduced by a functional verbal head on top of vP (called Voice in Kratzer 1996, Marantz 1997, Alexiadou, Anagnostopoulou and Schäfer 2006, Bruening to appear, little v in Hale and Keyser 2003, Folli and Harley 2005, Init in Ramchand 2008, agency in Doron 2003 or originator in Borer 2005). Here we call this head v-E(xternal)A(rgument) to make the relation to the external argument explicit. We assume that v-EA does not

Chapter Eight

172

introduce a further event. It introduces a DP in its specifier (on passives see below) and determines the thematic relation between the external argument and the event expressed by the vP in its complement. The decomposition of a transitive verb is given in (50). (50)

[v-EAP DPSubject v-EA [vP v [Ҁopen DPObject ]]]

We propose that v-EA comes in the three variants in (51) (cf. Doron 2003), which differently specify the relation between the external argument and the verbal event i.e., three different external argument theta roles. More specifically, (51c) imposes stronger requirements on the external argument (human and co-temporal) than (51b) (co-temporal) and (51a) (direct causer) (cf. section 3). Therefore, the three roles are in a subset relation so that what qualifies as an agent will also qualify as a direct participant, and agents as well as direct participants will qualify as indirect participants but not the other way around. (51)

a. v-EA-indirect participant b. v-EA-direct participant c. v-EA-agent

The vP semantically restricts the kind of a v-EA head it can co-occur with (AAS 2006, and Marantz 2009, Schäfer 2012). Since the v-EA heads are in a subset relation, the vP thereby determines the maximal amount of external argument types it can combine with. A detailed theory of such restrictions is beyond the scope of this paper; we just give an exemplary characterization. For example, if the root implies intentionality (murder), only v-EA-agent is possible (52a). Furthermore, indirect participant causers have been argued to occur only in resultative structures (Travis 2005, Folli and Harley 2005, Schäfer 2012). In (52b, c), the indirect participants the wind and the rain are only grammatical if we add a result state to the basically mono-eventive verb (note that these are NP-XP resultatives in the terminology of Levin and Rappaport Hovav 1999). Agents and direct participants, on the other hand, can occur also in monoeventive structures (52d, e). (52)

a. b. c. d. e.

*The storm / John murdered the president The wind rolled the ball ??(across the goal-line) The rain washed the stairs ??(clean) John rolled the ball/washed the stairs John/The train whistled

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

173

6.2. Patterns of variation In section 5 we have seen that Jacaltec restricts the external argument of active accomplishment verbs to agents.11 This suggests that this language lacks the v-EA heads introducing indirect and direct participants (though we think that the difference between indirect and direct participants in this language would need further investigation). Most languages have, however, all three v-EA heads and then the restrictions on the external argument show up only in passives and/or nominalizations. Following Doron (2003), Sigurðsson (2011), Alexiadou and Doron (2012), or Bruening (to appear), we assume that a further layer can appear on top of (50) as in (53). A family of heads, all executing related tasks, can realize this layer. We call these heads Voice-heads (adopting the more traditional understanding of the term as diathesis). (53)

[VoiceP Voice-N/V [vEA v-EA [vP [ Root]]]]

Voice heads do not introduce the external argument (v-EA does) and they also do not introduce a further event. Instead they take an active v-EAP (a default active VoiceP) as their complement and give a Voice (or diathesis) alternation. Morphologically, Voices introduce characterizing morphology, as described in (54) for English. In addition, while verbal Voices leave the category of their complement unchanged, nominal Voices also change the category.12 The three Voice heads that account for the English data discussed in this paper are listed in (54): (54) a. Voice-Passive produces a passive b. Voice-ation produces an ation-nominalization c. Voice-ing-of produces an ing-of gerund Syntactically, Voices also often absorb the external argument. We do not provide the technicalities for this process here but refer to Bruening (to appear) for a proposal how the passive/nominalization can technically absorb/existentially bind the external argument in Spec, vP-EA and optionally reintroduce it in a by-phrase. In this kind of analysis we get an intuition of how nominalization is similar to passivization, as Grimshaw (1990) proposes:13 just like the passive, the nominalization suffix is a Voice head. If the nominalization allows a by-phrase, we have evidence that this nominalization contains the layer that introduces the external argument v-EA, as the default assumption is that the same layer that introduces the external argument in the passive should also be introducing

174

Chapter Eight

the external argument in the nominalization. This is the case in all our 'small' nominalizations, which we take to have the basic structure in (53). With this background, we turn to our proposal how to implement the direct participation/agent exclusivity effect. The central claim is that specific Voices can select for specific vP-EAs. And since the different v-EA heads are thematically in a subset relation, such a choice can either filter out indirect participants, or both, direct or indirect participants as external arguments in this specific Voice construction. To give some examples, the English passive Voice is underspecified for the v-EA it combines with so that all types of external arguments are available. The same holds for the Voice producing ing-of gerunds. The Voice producing -ation nominalizations, on the other hand, selects for v-EA-direct participant and this choice makes indirect causers unavailable in this construction. Finally, in constructions showing an agent exclusivity effect such as the Greek passive, Voice selects for v-EA-agent. Note that we are aware that our proposal is a pure implementation of direct participant/ agent exclusivity effects; from what we know so far about these effects, we cannot predict what Voice selects what v-EA, or why a specific Voice selects a specific v-EA. To our knowledge, these effects have simply not been correlated with any further syntactic or semantic property, which could provide deeper answers to these questions. Hebrew provides morpho-syntactic evidence for this type of decomposition in the verbal domain (see Doron 2003 for details). First, Hebrew provides morphological evidence for the existence of different v-EA heads. The sentences in (55a, b) are both active but differ in the template morphology used and this choice corresponds with the availability of external argument types. Thus in (55a), we see once again that the simple template is underspecified for the type of external argument involved, while the intensive template in (55b) is not. (55)

a. ha-yeladim/ ha-tiltulim ba-argaz šavru the children/the shaking within the box broke-Simpl et-ha-kosot Acc-the glasses “The children/ the shaking within the box broke the glasses” b. ha-yeladim/*ha-tiltulim ba-argaz šibru the children/*the shaking within the box broke-Intns et-ha-kosot Acc-the glasses “The children/ the shaking within the box broke the glasses”

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

175

Furthermore, only the intensive template can combine with Voice-passive. As Doron (2003) states, 'the simple template lacks the passive voice'. This is illustrated in Table 1 from Alexiadou and Doron (2012) for the root Ҁyšv ‘sit, inhabit’. This is expected under an approach to Voice that allows for selection of vP-EAs, as the one put forth here.

VOICE Active Passive

Simple yašav “sit, inhabit” -----------

Agency Intensive yišev “settle, populate” yušav “be settled, populated”

Table 1. Hebrew Template-Voice interaction

7. Conclusions In this paper we showed that there are in fact two restrictions with respect to external arguments, agent exclusivity and direct participation, which are exhibited by various nominal but also verbal constructions across languages. We argued that an account based on event complexity that proposes that nominals can only host simple events makes wrong predictions and that an account in terms of the structural deficiency of nominalizations, which contain less verbal structure than their corresponding verbal bases, fails to explain the fact that sometimes verbal forms are more restricted than nominalizations (e.g. Greek passives vs. nominalizations).14 We proposed an implementation of this variation by means of the interaction between three possible types of heads that introduce external arguments (v-EA) and a Voice head that comes higher in the structure and exhibits selectional restrictions with respect to the v-EA it can combine with.

References Abney, S. 1987. The English Noun Phrase in its Sentential Aspect. Ph.D. dissertation, MIT. Alexiadou, A. 2009. On the role of syntactic locality in morphological processes: the case of (Greek) derived nominals. In Quantification, Definiteness and Nominalization, ed. by A. Giannakidou, and M. Rathert, 253-280. Oxford: Oxford University Press. —. 2001. Fuctional structure in nominals. Amsterdam: John Benjamins.

176

Chapter Eight

Alexiadou, A., E. Anagnostopoulou, and F. Schäfer. 2009. PP licensing in nominalizations. Proceedings of NELS 38. Alexiadou, A., E. Anagnostopoulou, and F. Schäfer. 2006. The properties of anticausatives cross-linguistically. In Phases of Interpretation, ed. by M. Frascarelli, 187-212. Berlin: Mouton de Gruyter. Alexiadou, A., and E. Doron. 2012. The syntactic construction of two nonactive Voices: passive and middle. Journal of Linguistics 48: 1-34. Alexiadou, A., and G. Iordăchioaia. to appear. Psych nominalizations with causers. Proceedings of NELS 42. Alexiadou, A., G. Iordăchioaia, and F. Schäfer. 2011. Scaling the variation in Romance and Germanic nominalizations. In The noun phrase in Romance and Germanic, ed. by P. Sleeman and H. Peridon, 25-40. Amsterdam: John Benjamins. Alexiadou, A., and F. Schäfer. 2006. Instrument subjects are agents or causers. In Proceedings of WCCFL 25, ed. by D. Baumer, D. Montero and M. Scanlon, 40-48. Somerville, MA: Cascadilla Proceedings Project. Baker, M. C. 2008. The Macroparameter in a Microparametric World. In The Limits of Syntactic Variation, ed. by T. Biberauer, 351-374, Amsterdam: John Benjamins. Borer, H. 2013. Structuring sense: taking form. Oxford: Oxford University Press. —. 2005. Structuring sense: the normal course of events. Oxford: Oxford University Press. —. 1993. Parallel Morphology. Ms., University of Massachusetts at Amherst. Bruening, B. to appear. By-Phrases in Passives and Nominals. Syntax. Cornilescu, A. 2001. Romanian nominalizations: Case and aspectual structure. Journal of Linguistics 37:3, 467–501. Craig, C. 1976. Properties of Basic and Derived Subjects in Jacaltec. In: Subject and Topic, ed. by C. N. Li, 99-123. New York: Academic Press. Doron, E. 2003. Agency and voice: the semantics of the Semitic templates. NLS 11: 1-67. Embick, D. 2010. Localism vs. globalism in morphology and phonology. Cambridge: MIT Press. Folli, R., and H. Harley. 2008. Teleology and animacy in external arguments. Lingua 118, 190-202. Folli, R., and H. Harley. 2005. Flavours of v: Consuming Results in Italian and English. In Aspectual Enquiries, ed. by P. Kempchinsky, and R. Slabakova, 95-120. Dordrecht: Springer.

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

177

Grimshaw, J. 1990. Argument Structure. Cambridge: MIT Press. Hale, K., and S. J. Keyser. 2002. Prolegomenon to a Theory of Argument Structure. Cambridge: MIT Press. Harley, H. 2011. An alternative to deficiency approaches to the manneralternation parameter. Paper presented at the Workshop on verbal elasticity, Barcelona, October 2011. Harley, H., and R. Noyer. 2000. Licensing in the non-lexicalist lexicon. In The Lexicon/Encyclopedia Interface, ed. by B. Peeters, 349-374. Amsterdam: Elsevier Press. Harley, H., and R. Noyer. 1998. Mixed nominalizations, object shift and short verb movement in English. In Proceedings of NELS 28, ed. by K. Kusumoto, and P. Tamanji, 143-157. University of Massachusetts at Amherst: GLSA. Iordăchioaia, G. 2008. External argument PPs in Romanian nominalizations. In Working Papers of the SFB 732 (1), ed. by F. Schafer, 71-84. Kayne, R. S. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Kratzer, A. 2003. The event argument and the semantics of Voice. Ms. —. 1996. Severing the External Argument from its Verb. In Phrase Structure and the Lexicon, ed. by J. Rooryck and L. Zaring, 109-137. Dordrecht, Kluwer. Larson, R. K. 1988. On the Double Object Construction. Linguistic Inquiry 19: 335-391. Levin, B., and M. Rappaport Hovav. 2002. The semantic determinants of argument expression: A view from the English resultative construction. In The Syntax of Time, ed. by J. Guéron, and J. Lecarme, 477-494. Cambridge: MIT Press. Levin, B., and M. Rappaport Hovav. 1999. Two Structures for Compositionally Derived Events. Proceedings of SALT 9, 199-223. Ithaca: Cornell Linguistics Circle Publications. Marantz, A. 2011. Syntactic approaches to argument structure without incorporation. Doing the (anti-lexicalist) dance without doing the dance. Paper presented at the Structuring the argument Workshop, Paris, September 2011. —. 2009. Roots, re- and affected agents: can roots pull the agent under little v? Paper presented at Roots II, Stuttgart, June 2009. —. 2005. Objects out of the lexicon: objects as event. Handout. Talk presented at the University of Vienna. —. 1997. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In University of Pennsylvania Working

178

Chapter Eight

Papers in Linguistics Vol. 4.2, ed. by A. Dimitriadis, L. Siegel, C. Surek-Clark, and A. Williams, 201-225. Philadelphia. Picallo, M. C. 1999. La estructura del sintagma nominal: Las nominalizaciones y otros sustantivos con complementos argumentales. In Gramática descriptiva de la lengua española, ed. by I. Bosque, and V. Demonte, vol. I, 363-393. Madrid: Espasa. Picallo, M. C. 1991. Nominals and Nominalizations in Catalan. Probus 3: 279-316. Pesetsky, D. 1995. Zero Syntax: Experiencers and Cascades. Cambridge: MIT Press. Ramchand, G. 2008. Verb meaning and the lexicon. Cambridge: Cambridge University Press. Rappaport Hovav, M., and B. Levin. 2012. Lexicon Uniformity and the Causative Alternation. In, The Theta System: Argument Structure at the Interface, eds. by M. Everaert, M. Marelj, and T. Siloni, 150-176. Oxford: Oxford University Press. Rapport Hovav, M., and B. Levin. 2001. An event structure account of English resultatives. Language 77: 766-797. Schäfer, F. 2012. Two types of external argument licensing: the case of causers. Studia Linguistica 66: 128-180. Sichel, I. 2011. Nominalization, causativization, and category-free syntax. Paper presented at Roots III, Jerusalem, June 2011. —. 2010. Event Structure Constraints in Nominalization. In The Syntax of Nominalizations across Languages and Frameworks, ed. by A. Alexiadou, and M. Rathert, 151-190. Berlin: Mouton. Sigurðsson, H. A. 2011. On the new passive. Syntax 14: 148-178. Tatevosov, S. 2012. Decomposing event structure: evidence from denominal verbs in Tartar. Paper presented at WAFL 8, University of Stuttgart. Travis, L. 2005. Agents and Causes in Malagasy and Tagalog. In The Syntax of Aspect, ed. by N. Erteschik-Shir, and T. Rapoport, 174-189. Oxford: Oxford University Press. Varela, S. 2012. La interacción de las nominalizaciones con la voz, el aspecto y la dimensión temporal. In Tiempo y espacio en la formación de palabras en español, ed. by E. Bernal, M. Emsel and C. Sinner, Munich: Peniope. Wechsler, S. 1990. Accomplishments and the Prefix re-. In Proceedings of the North Eastern Linguistic Society XIX, ed. by J. Carter, and R.-M. Dechaine, 419-434. Amherst: Graduate Linguistic Student Association, Dept. of Linguistics, University of Massachusetts.

‘Direct Participation’ and ‘Agent Exclusivity’ Effects

179

Notes 1

We would like to thank the audience at the IS/CASTL Workshop 'Categorization and Category Change in Morphology', two anonymous reviewers and especially Ivy Sichel for comments and discussion. We also thank Nils Hirsch for help with the final formatting. The research of Cano has been supported by the project FFI2011-23829 (Ministerio de Economía y Competitividad) and a grant FPU (AP2008-00996). The research of the other authors has been supported by a DFG grant to the SFB 732, Incremental specification in context, at the University of Stuttgart. 2 We use the following abbreviations: Acc - accusative case, Act - active morphology, cl - clitic, EA - external argument, Gen - genitive case, Inf infinitive, Intns - intensive template, Instr - instrumental case, Nact - non-active morphology, Nom - nominative case, Pst - Past Tense, Rf - reflexive pronoun, Simpl - simple template. 3 Note here that Sichel (2011) actually observes the same effect in Romance causative formation but concentrates on nominalizations. 4 We follow the standard assumption that the subject of lexical causatives (accomplishments) must be construed as a direct causer (see e.g. Rappaport Hovav and Levin 2012 for recent discussion). 5 It should be noted that for some speakers the preposition de către is more restricted in use: it is excluded with direct participants and only compatible with agents. This is a matter of speaker variation which does not affect our analysis in section 6. The important fact is that de către displays a restriction with respect to the external argument. 6 For alcohol in (25c), we can also build a context where it is a direct participant and thus compatible with de către: (i)

distrugerea ficatului de către alcool destruction liver.Gen by to alcohol “The destruction of the liver by alcohol”

7

The speaker variation mentioned in endnote 5 also applies to verbal passives in Romanian. 8 Tatevosov (2012) provides data from Tatar and Russian active clauses, which look like a direct participant effect. In these languages some accomplishments, which are characterized by specific morphological marking, allow only direct but not indirect participants as their subjects. Consider the following Russian verb pair. (i)

a. Zhizn' bez otca vospita-l-a Volodju siln-ym life without father.Gen bring.up-Pst-F V.Acc strong-Instr “The life without the father made (lit. brought up) Volodja strong” b. ??Smert' otca vospita-l-a Volodju siln-ym death father.Gen bring.up-Pst-F V.Acc strong-Instr “The death of his father made (lit. brought up) Volodja strong”

180

Chapter Eight

Note that Tatevosov shows that the two classes of predicates do not differ in terms of aspectual structure, i.e. they are both accomplishments irrespectively of the type of external argument involved. 9 Hebrew has a third template, the causative template. As this is not only used to form transitive variants of anticausative predicates, but it is also used to derive causatives of transitive predicates, we do not discuss it here, see Doron (2003) for details. 10 In mono-eventive structures the root is a manner modifier of v. 11 On the licensing of causers in Jacaltec anticausatives, see AAS 2006, Alexiadou and Schäfer 2006. These authors argue that causers in anticausatives are not introduced by the layer which we call here v-EA. 12 Here we call both the nominal and the verbal functional head that selects for a particular v-EA 'Voice' to indicate the close selectional relation between the two heads, but in the case of nominalizations one would probably prefer to call it a N(ominalizing head) that selects for a vP-EA of a particular type (indirect, direct or agent participant) (see Bruening to appear). 13 Obviously, this analysis does not capture all types of nominalizations. Some, such as the Spanish verbal infinitive, are active, some do not contain verbal layers at all. See Alexiadou (2001), Alexiadou, Iordăchioaia and Schäfer (2011), and Bruening (to appear), for discussion. 14 Borer (2013) also argues that ing-of gerunds and -ation nominals do not differ in terms of internal verbal structure.

CONTRIBUTORS

Artemis Alexiadou (University of Stuttgart) Mariángeles Cano (University of Madrid) Antonio Fábregas (IS/CASTL, University of Tromsø) Gianina Iordăchioaia (University of Stuttgart) Laura A. Janda (University of Tromsø) Björn Lundquist (CASTL, University of Tromsø) Fabienne Martin (University of Stuttgart) Isabelle Roy (UMR 7023, University of Paris VIII / CNRS) Florian Schäfer (University of Stuttgart) Elena Soare (UMR 7023, University of Paris VIII / CNRS) Kaori Takamine (Norwegian University of Science and Technology)



INDEX

active Voice 89, 171 adjective 4-8, 11-16, 18-20, 23, 2527,29-30 n.5., 31-32 n.12., 3334, 45-46, 57 n.6., 92, 95-96, 98, 100, 102-103, 105, 107, 110118, 122 n.6., 127, 134-136, 138, 145-147 denominal adjective 89 depictive adjective 22 (non)-gradable adjective 16, 18 adposition vii, 4-5 see also prepositions affix 26, 92, 95-98, 102-106, 108114, 117-118, 159 see also prefix affix combinations 95-96, 102, 106, 119 agent exclusivity 153, 154, 156, 165-166, 168-171, 174-175 aspect 7, 23, 25, 61, 79, 100, 129, 134, 138 Boolean category features 4 category change 11, 87-89, 91-92 cartography 98, 100, 118 definiteness 8, 33-37, 41-42, 44-50, 53-54 demonstrative 8, 33-38, 43-48, 5154, 56 n.3., 109 derivation 3, 6, 20-21, 97, 101, 107 derived categories 12 derived nominal 87-89, 91, 123, 126, 128, 132-133, 136, 153, 154-155, 157, 159-161, 163164, 171



direct participation 153, 154-155, 157, 161-162, 164, 166, 170171, 174-175 Distributed Morphology 5-6, 134 DP 8, 33-36, 44-49, 52-54, 109, 111, 119, 129, 137, 139-140, 142-143, 171-172 layered DP 8 English 6, 7-8, 16, 21, 34-38, 89-90, 91, 95, 98, 103, 123, 125, 153155, 158-165, 173-174 event(ivity) complex event (nominal) 90 conceptual event 91, 126, 132133, 145-147 grammatical event(ivity) 90, 91, 126, 129, 131-132, 133, 136-137, 139, 145, 147 simple event (nominal) 23, 25, 90, 124, 130, 145 event structure 12-14, 16-19, 21, 23-27, 91, 124, 129, 157, 159, 170 Exo-skeletal (model) 6 external argument 20-23, 27, 89-90, 92, 129, 139, 153-166, 168, 171175 feature-based approach 4, 6 French 3, 89-90, 135, 144, 154, 161-162 functional head 6, 8, 50-51, 54, 88, 129, 171 functional structure 92, 129 German 16, 89, 154, 161-162, 165 gradability 17-18

184 Greek 89, 138, 154, 161, 162, 168, 174, 175 imperfective 59-61, 71, 74-80 infinitive 4, 7, 21, 26, 89, 164, 165

Index perfective 9, 59-67, 69-80 prefix 8-9, 59-77, 79-80 preposition 4, 6-7, 61, 92, 109, 110112, 118-119, 162, 166-169, 171 quale 107-111, 112

Japanese 4, 6, 8, 33-38, 40, 44-45, 54 lexical category 4-8, 12, 19-20, 33, 54, 101 lexical features vs. functional features 5 lexicalist approach 128 mixed category 4, 6-7, 88 morphology 11, 23, 89, 90, 95, 117, 130, 168, 173, 174 nanosyntax 11, 118 N’-ellipsis 50-53, 56 nominalization 12, 14, 23, 25-27, 87-91, 105, 108, 113, 115,124128, 140, 142, 144-147, 153158, 161-166, 168-171, 173-175 deverbal nominalization 126 -er nominals 90-91, 125-126, 133-134, 137-138, 140-142 noun 3-8, 12-14, 18-20, 24-25, 2627, 33, 35, 39, 45, 46, 50-51, 53, 54, 61-61, 88, 91, 92, 95-96, 98, 102-103, 107, 109, 110, 111112, 115, 116, 117-119, 124, 128, 139, 143 NP-ellipsis 8, 34, 37, 49-50, 53 numeral classifier 8, 9, 36, 39, 4244, 45-46, 49-51, 53, 60-61, 7980 participle 4, 6-7, 7-8, 11-21, 23-27, 69, 71 passive Voice 23, 90, 161, 166, 171, 174-175



referential index 5, 19, 23, 24, 26, 27 Romanian 89, 154, 161, 162, 164, 166-168, 170-171 root 6, 11, 21, 23, 24, 25, 27, 91-92, 96, 97, 105, 106, 109, 113, 114115, 117, 124, 129, 140, 145146, 147, 159, 171-172, 173, 175 Russian 3, 8-9, 59, 61-64, 66-67, 69, 71, 74-76, 79-80 semantics 11, 89, 92, 107, 108, 110-111, 112, 115, 138, 140 Spanish 4, 89, 91, 92, 95, 96, 98, 102-103, 113, 117, 154, 161, 162-165, 167, 170 specificity 45-48 specifier 5, 19-20, 23, 24, 26, 27, 36, 45, 46, 47, 50-51, 54, 139, 172 syntactic derivation 152 syntax 5-6, 11, 24, 87-88, 89, 91, 98-99, 100, 118, 129, 159 Swedish 7, 21 templates 90, 96, "v" (little v) 6, 18, 22-23, 24, 26, 27, 31, 88, 90, 159, 171-172, 173-175 verb 3-9, 11-13, 14-15, 17-20, 23, 25-27, 38-39, 59-76, 78-80, 8791, 96, 97, 102, 107, 108-109, 124-125, 131, 134, 137, 155, 156, 157-158, 159-160, 161, 165, 167, 168-171, 172, 173

Categorization and Category Change verb classifiers 8-9, 59, 61-62, 63, 66, 69, 71, 74, 75, 79-80 word classes 7





185

word formation 6, 87-88, 91-92, 95, 124, 128 root-level word formation 92 syntactic approach to word formation 91, 128