Contemporary Linguistic Parameters: Contemporary Studies in Linguistics 9781472533937, 9781474219549, 9781472525901

Citation preview

Contemporary Linguistic Parameters

ALSO AVAIL ABLE FROM BLOOMSBURY Contemporary Applied Linguistics Volume 1, edited by Vivian Cook and Li Wei Contemporary Applied Linguistics Volume 2, edited by Li Wei and Vivian Cook Contemporary Computer-Assisted Language Learning, edited by Michael Thomas, Hayo Reinders and Mark Warschauer Contemporary Corpus Linguistics, edited by Paul Baker Contemporary Critical Discourse Studies, edited by Christopher Hart and Piotr Cap Contemporary Stylistics, edited by Marina Lambrou and Peter Stockwell Contemporary Task-Based Language Teaching in Asia, edited by Michael Thomas and Hayo Reinders

Contemporary Linguistic Parameters Edited by Antonio Fábregas, Jaume Mateu and Michael T. Putnam Contemporary Studies in Linguistics

Bloomsbury Academic An imprint of Bloomsbury Publishing Plc

Bloomsbury Academic An imprint of Bloomsbury Publishing Plc 50 Bedford Square London WC1B 3DP UK

1385 Broadway New York NY 10018 USA

www.bloomsbury.com BLOOMSBURY and the Diana logo are trademarks of Bloomsbury Publishing Plc First published 2015 © Antonio Fábregas, Jaume Mateu, Michael T. Putnam and Contributors, 2015 Antonio Fábregas, Jaume Mateu and Michael T. Putnam have asserted their right under the Copyright, Designs and Patents Act, 1988, to be identified as the Editors of this work. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage or retrieval system, without prior permission in writing from the publishers. No responsibility for loss caused to any individual or organization acting on or refraining from action as a result of the material in this publication can be accepted by Bloomsbury or the author. British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. ISBN:

HB: 978-1-4725-3393-7 ePDF: 978-1-4725-2590-1 ePub: 978-1-4725-3271-8

Library of Congress Cataloging-in-Publication Data Contemporary linguistic parameters / edited by Antonio Fábregas, Jaume Mateu and Michael T. Putnam. pages cm. -- (Contemporary Studies in Linguistics) Includes bibliographical references and index. ISBN 978-1-4725-3393-7 (hb : alk. paper) -- ISBN 978-1-4725-3271-8 (epub) -ISBN 978-1-4725-2590-1 (epdf). Principles and parameters (Linguistics) 2. Grammar, Comparative and general--Syntax. 3. Language and languages--Variation. I. Fábregas, Antonio, editor. II. Mateu i Giral, Jaume, editor. III. Putnam, Michael T., editor. P295.5.C66 2015 410.1’8--dc23 2015010279 Series: Contemporary Studies in Linguistics Typeset by Fakenham Prepress Solutions, Fakenham, Norfolk NR21 8NN

CONTENTS

Abbreviations  vii

Part I. Prolegomena 1

“Parameters” in linguistic theory: What, where, and how 3 Antonio Fábregas, Jaume Mateu, and Michael T. Putnam

Part II. Main morphosyntactic parameters 2

Morphological parameters Antonio Fábregas 27

3

Case. Ergative languages 47 Michelle Sheehan

4

Head directionality 73 Hubert Haider

5

Parameters and argument structure I: Motion predicates and resultatives 99 Víctor Acedo-Matellán and Jaume Mateu

6

Parameters and argument structure II: Causatives and applicatives 123 María Cristina Cuervo

vi Contents

7

The functional structure of the clause: Main issues 147 Martina Wiltschko

8

Extended projections of V: Inner Aspect 177 Jonathan E. MacDonald

9

Null subject 201 Roberta D’Alessandro

10 Head movement in the clausal domain 227 Ángel J. Gallego and Juan Uriagereka 11 Wh-movement 251 Željko Bošković 12 Topic and focus 281 Jordi Fortuny 13 The functional structure of N 303 Asya Pereltsvaig

Part III. Parameters beyond morphosyntax 14 Parameters in phonological analysis: Stress 335 Marc van Oostendorp 15 Parameters in language acquisition and language contact 353 Nina Hyams, Victoria Mateu, Robyn Ortfitelli, Michael Putnam, Jason Rothman, and Liliana Sánchez References  375 Index  429

ABBREVIATIONS

1

First person

2

Second person

A Agent ABL Ablative ABS Absolutive ACC Accusative AGR Agreement AI

Animate intransitive

ALL Allative ANIM Animate AOR Aorist APPL Applicative ART Article ASP Aspect AUX Auxiliary AV

Active Voice

BEN Benefactive CAUS Causative CL Classifier CLT Clitic COM Comitative CONJ Conjugation DAT Dative DECL Declension DEF Definite

viii Abbreviations

DEM Demonstrative DET Determiner DIM Diminutive DIR Directional E

Event head

EMPH Emphatic ERG Ergative EXPL Expletive F Feminine FACT Factual FAM

Familiarity treatment

FOC Focus Ft Foot FUT Future FV

Final Vowel

GEN Genitive H Heavy HAB Habitual Hd Head HON Honorific HUM Human IMP Imperative IMPF Imperfective IN Inessive INAN Inanimate IND Indicative INF Infinitive L Light LOC Locative M Masculine MAL Malefactive

Abbreviations

MOD Modal NEG Negation NFUT Non-Future NMLZ Nominalizer NOM Nominative NPST

Non-Past

NT Neuter O Object OBJ

Object Agreement

OV

Objective Voice

P Preposition PART Participle PASS Passive PERF Perfect PL Plural POSTES Postessive PRES Present PRFV Perfective PRN Pronoun PROG Progressive PROSP

Prospective

PROX Proximal PRT Particle PST Past PUNC Punctual Q Interrogative QU Quantifier REC Reciprocal REL Relative S Subject SBJ Subjunctive

ix

x Abbreviations

SCLT

Subject Clitic

SG Singular SM

Subject Marker

SUBJ

Subject agreement

ThV

Theme Vowel

TNS Tense TOP Topic VM

Verb Movement

Wd

Prosodic word

μ Mora s Syllable

PART ONE

Prolegomena

1 “Parameters” in linguistic theory: What, where, and how Antonio Fábregas, Jaume Mateu, and Michael T. Putnam University of Tromsø, Autonomous University of Barcelona, and Penn State University

The goal of this chapter is to offer the reader a short introduction to the concepts, distinctions, and notions that will be revisited in this volume, which is dedicated to the use of parameters in contemporary synchronic studies. Our presentation will be short and general, because the readers will find the relevant details in the appropriate chapters that follow. §1.1 introduces the notion of parameter as it was first proposed in generative grammar, and considers some of its properties and the critiques that questioned it. §1.2 introduces some relevant distinctions among parameters, some of them emerging as an answer to the critiques that the classical parameter approach received. §1.3, finally, considers some of the most recent approaches to variation that present alternatives to parameters in the current theoretical universe.

1.1. The classical parameter One of the central issues in the study of formal grammar has been the tension between the proposal that the language faculty is universal—thus, at least to some extent, uniform—and the observation that on the surface specific languages vary, sometimes dramatically, in a wide number of properties. How does variation emerge? What makes it possible? What are its limits? How does the child acquiring a specific language determine the specific range of variation that needs to be considered and find the right choices in a particular language? These questions pose

4

CONTEMPORARY LINGUISTIC PARAMETERS

non-trivial challenges to the notion of universal and uniform aspects of human grammar. In this context, parameters appear progressively during the late 1970s as a proposal with two main goals: (i) to resolve the tension between universal and language particular properties; and (ii) to provide the background to a complete theory of language acquisition. At the earliest formative stages of generative grammar—the (Extended) Standard Theory (EST)—the language system was viewed as a complex system of rules. Variation was simply the effect of different languages possessing independent and unique systems of rules (different transformations, perhaps a different ordering, etc.). The fact that children would acquire a complete system of rules through exposition to a fairly limited set of data was explained by the hypothesis that part of the endowment of Universal Grammar (UG) was a LAD (“Language-acquisition device”). In the words of Chomsky (1965: 31–2): We must require of [an explanatory adequate] linguistic theory that it provide for: (i) an enumeration of the class s1, s2, … of possible sentences (ii) an enumeration of the class SD1, SD2, … of possible structural descriptions (iii) an enumeration of the class G1, G2, … of possible generative grammars (iv) specification of a function f such that SDf(i,j) is the structural description assigned to sentence si by grammar Gj, for arbitrary i,j. (v) specification of a function m such that m(i) is an integer associated to the grammar Gi as its value (with, let us say, lower value indicated by higher number) […] A language-acquisition device that meets conditions (i)–(iv) is capable of utilizing such primary linguistic data as the empirical basis for language learning. This device must search through the set of possible hypotheses G1, G2, … which are available to it by virtue of condition (iii) and must select grammars that are compatible with the primary linguistic data, represented in terms of (i) and (ii). It is possible to test compatibility by virtue of the fact that the device meets condition (iv). The device would then select one of these potential grammars by the evaluation measure guaranteed by (v). This procedure was costly in the sense that it would require a very rich and specific endowment of a predetermined, innate UG; additionally, a theory about the evaluationary measure was never fully developed (even though it was one of the initial goals of Chomsky 1957: Chapter 6). Moreover, generative grammar underwent a radical change during the 1970s, with its goals and analytic procedures moving more away from the description of particular rules and towards the identification of abstract and simple principles that, by interaction, would generate only a subset of the logically imaginable structures and transformations. This move from rules to principles meant that LAD became obsolete, and a new theory for language variation and acquisition was, therefore, needed.



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

5

Universal Grammar was then viewed as a set of highly modularized principles that affected different subcomponents of human language: the Case filter, binding, the theta-criterion, the projection principle, etc. These principles would interact with each other in restricted ways—e.g. the Case filter would only apply to arguments that were also subject to the theta-criterion. Structure-building operations would be universally restricted by virtue of their adhering to these inescapable, inviolable principles. However, there is more than one conceivable way of meeting the demands of a particular principle placed upon neophytes acquiring a grammar. Here is where the notion of parameter comes into play: parameters are particular choices about how a universal principle can be satisfied. Language variation and its tension with Universal Grammar was thus explained in this fashion: variation is a necessary consequence of the fact that UG imposes principles of well-formation but does not specify the specific way in which they have to be satisfied. Variation is, then, restricted by the logical space stated by a universal principle. Accordingly, language acquisition would then behave in a simpler way than the LAD guaranteed: endowed with the universal principles that language must meet, the child must arrive at a hypothesis regarding the particular way in which her language satisfies those principles, and once determined (that is, once the parameter value was set), the right structures of the language—and only the right ones—would be generated without the need to posit specific rules. In a nutshell, this system is commonly known as Principles and Parameters (P&P; cf. Chomsky 1981).1 Let us now concentrate on a particular example in order to see the implications that a classic parametric view has in practice, and also in order to see more closely what the other properties of a classical parameter have.

1.1.1. A classic example: Null objects As an introduction, let us consider Rizzi’s (1986) influential work on null objects. His research starts from a set of data that apparently violate a universal principle of grammar. Consider (1) and (2). (1) (2)

This leads to the following conclusion. Questo conduce alla seguente conclusione this  leads  to-the following conclusion ‘This leads to the following conclusion’

Example (1) is from English, while example (2) is from Italian. The surprising fact about these sentences is that, at least on the surface, the verb lead / conducire is lacking one of its obligatory arguments, the one corresponding to the object. (3) This leads me to the following conclusion. The problem is that this seems to violate a UG principle that imposes conditions on semantic well-formedness; namely, the Theta-Criterion: (4) Each argument bears one and only one theta-role, and each theta-role is assigned to one and only one argument. (Chomsky 1981: 36)

6

CONTEMPORARY LINGUISTIC PARAMETERS

On the surface, examples (1) and (2) seem to violate (4), on the assumption that the verb lead / conducire is specified in the lexicon as assigning at least two theta-roles (as we see in 3). However, by assumption, (1) and (2) must satisfy the theta-criterion in some way, because (a) the theta-criterion is a universal principle, and (b) (1) and (2) are grammatical in English and Italian, respectively. Rizzi (1986) goes on to show that, despite their surface similarities, the underlying structures proposed for examples (1) and (2) are actually quite distinct from one another. First, example (2), but not (1), allows object-control: (5) a. *This leads [PRO to conclude what follows]. b. Questo conduce a [PRO concludere quanto segue]. this  leads  to  conclude what  follows Second, in contexts where there is no apparent object, Italian (but not English) licenses an anaphor that seems to take an object as an antecedent. (6) La buona musica riconcilia ____ con se stessi. the good music reconciles    with oneself Third, in Italian it is possible to have a non-selected small clause licensing the apparently absent object as its subject. (7) Di solito, Gianni fotografa ____ seduti. in general, Gianni photographs  seated ‘In general, Gianni photographs people who are seated’ Fourth, Italian also licenses argument small clauses in this context: (8) Questa musica rende _____ allegri. this  music renders  happy ‘This music renders people happy’ Rizzi’s conclusions are that in Italian there is a syntactically active object, which is able to control and behave as the antecedent of anaphors and of acting like the subject of a (selected) small clause, while in English there is no syntactically represented object. This leads to the proposals that in Italian (2) the object is occupied by a pro, a syntactically active but phonologically silent pronoun, while in English there is no syntactically expressed object. What is the relevant parameter here that simplifies the acquisition task? Rizzi proposes that the parameter has to do with how the theta-criterion is met in English, versus Italian. Rizzi assumes a lexicalist theory, one where the lexicon is not just a repository of lexical entries, but also contains operations and rules with generative power. His proposal is that the theta-criterion can be satisfied in English at a lexical level, but not in Italian. English allows for the operation described in (9), which is active in the lexicon (Rizzi 1986: 509): (9) Assign arb[itrary] to the direct theta-role



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

7

This saturates the theta-role in the lexicon. When the verb projects into the syntactic representation, then, projection of a structural position for the object is not forced, assuming that positions are only made available for unsaturated theta-roles. In other words: by virtue of the availability in English of (9), thanks to a parametric choice, as far as syntax is concerned, the verb can lose one of its arguments. Italian, on the other hand, does not license the parametric choice in (9), and has to create a position available for the object, due to the Projection Principle. Consequently, Italian sentences where an object is seemingly absent are in actuality sentences where the object theta-role is assigned to a non-phonologically overt pro that occupies a standard object position in the clause. Pro was, incidentally, another one of the main proposals that came to light as a result of the Principles and Parameters approach (see Chapter 9). The principle underlying the proposal that some arguments could be expressed through phonologically empty items—empty categories—was known as the Empty Category Principle (ECP) (Chomsky 1981: 250). (10) [a e] must be governed (in some sense) The part of the principle that allowed for different parametric values is, obviously, in what sense an empty category had to be governed. The theory of empty categories in the P&P framework allowed for four subclasses of empty categories, which enabled theorists working in this research program the opportunity to take advantage of the distinction between pronoun and anaphora made available by UG through the principles of Binding Theory. The class of empty categories labeled pro were designated with the featural value: [+pronominal, –anaphoric]. The question that remained focused on the particular conditions that pro must be (properly) governed, that is, the subcase of (10) that applies to pro. Rizzi’s answer is a principle like (11): (11) pro is governed by X0y Parametric variation would be a result of particular languages determining what precise kind of X governs pro, or, in more concrete terms, which categories (T, N, V…) can govern pro. The grammar of Italian specifies that the condition noted in (11) must at least include Inflection and Verb as possible values for y. The result is that Italian would allow subject pro and object pro. English, on the other hand, would not assign any value to y, and therefore the set of categories that license pro would be empty: the result is that, even if pro is in principle made available by UG, the parametric value adopted by English would not license pro in any context, so pro would be absent from that language. Learnability would be ensured as follows, assuming Berwick’s (1982) Subset Principle: the default assumption made by the child would be that no category licenses pro (that is, that the set {Xy} is empty). Exposure to data where lexically required and syntactically active arguments are unpronounced would lead to an increase in the number of values that y can take, extending it to categories like Inf, V, etc., first allowing pro in some context and then extending its distribution to other cases.

8

CONTEMPORARY LINGUISTIC PARAMETERS

Let us now use this example to explicate the properties of classical parameters: i. Parameters are choices on the space of possibilities allowed by Principles. In the case discussed above, two principles are involved. UG dictates that there is a one-to-one correspondence between arguments and theta-roles (the thetacriterion) and that there are specific licensing conditions for empty categories (ECP). Parameters specify the way in which these principles are met: (a) whether the thetacriterion is met lexically or syntactically; (b) the range of categories that license the empty category called pro. ii. Parameters determine minimal modular differences. The theta-criterion is a principle that (depending on the theory) operates over D-structure or LF-structure (Logical Form); the ECP is a principle that acts over the level of representation known as S-structure. Parameters are restricted to choices on the space defined by a principle; as principles are modular, no parameter can refer to two or more different modules simultaneously. iii. A parametric choice can influence the way in which other principles, belonging to different modules of grammar, behave. Even if parameters are logically restricted to one single module, a parametric choice can have—and typically does have—implications for the behavior of other modules. The reason is that grammar modules interact with each other. In our example, the parametric choice that English makes with respect to the theta-criterion (“Assign arb to the object theta-role [in the lexicon]”) has direct implications for the way in which the structure is built after that operation: it prevents the projection of a structural position for the object. This means that the way in which the theta-criterion is satisfied has consequences for another principle, specifically the Projection Principle (Chomsky 1981, 1986a). The Projection Principle—crucial in the so-called endoskeletal theories, see Borer (2003, 2005a, 2005b, 2013) for the characterization of these theories—ensures that the lexical specification of an item must be reflected in its syntax: (12) Representations at each level of representation are projections of the features of lexical items, and given a lexical feature F, F must be projected at all levels. Or, in prose, if a verb is codified in the lexicon as requiring an object which receives a theta-role X, the object has to be represented somehow at D-structure, S-structure, and Logical Form. With respect to our discussion concerning the licensing of pro in English vs Italian above, we have seen that English does not project the object syntactically because arb saturated the theta-role in the lexicon, and this made the property “inactive.” iv. Parametric choices produce clustering effects.



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

9

As a consequence of the proposal that parameters act modularly, but rather modules interact with each other, a parametric choice can have non-local consequences on the grammar of a language. This is perhaps the most obvious difference between a rule-system as the one assumed in Chomsky (1965) and a P&P model (Chomsky 1981): a single parametric choice triggers a chain reaction that makes many structures (un)available by its local effects in its module and its non-local effects through interaction with other modules. In our example, a single parametric choice (whether arb can saturate an object theta-role in the lexicon) had wide-ranging consequences that include but are not restricted to: (a) the availability of control structures; (b) the availability of anaphoric pronouns in generic contexts; (c) the availability of minimal clauses. The second parameter discussed in Rizzi (1986) (the set of categories that govern pro) was associated during the 1980s to an even wider set of consequences (Chomsky 1981: 240, Rizzi 1982). It was proposed that if a language allowed Infl (T) to govern pro, it would also exhibit the following set of properties: (a) missing subjects; (b) free inversion of subject-verb; (c) long wh-movement of subjects; (d) empty resumptive pronouns in embedded clauses; (e) absence of “that-trace” effects. v. Parameter setting requires exposure to natural data. If principles are part of the UG endowment, choices defined on their logical space are not set by UG, hence explaining variation. In our brief discussion above, starting on the null assumption that no category licenses pro, the child will revise her assumptions whenever exposed to cases of null pronouns, thus extending the set of categories that license pro in the set of data she is exposed to.

1.1.2. How many potential values? One important question that has emerged in the course of the discussion on parameters was how many possible values a parameter would (and could) have. In our previous discussion it should have been clear that parameters were not always reduced to binary choices: in the case of the licensing of pro, there are many possible values that the parameter could take (e.g. if we limit ourselves to three potential categories, there are five possible values, and it grows exponentially when considering a wider set of categories). However, parsimonious considerations—especially Berwick’s (1982) proposal that in any parameter there should be one and only one unmarked option assumed by default by the child—gave preference to parametric formulations that would consider only two options. This led once again to the problem that individual languages did not appear to vary in the sharp and clean way that a perfectly binary divide would predict. If we consider pro, again, there are languages which never license empty pronouns, like English, but even Italian, that licenses pro with V and Infl, is unable to license it by P. Indeed (13) is ungrammatical. (13) *Sono sicuro di pro. I.am sure of pro Intended: ‘I am sure of something / everything / what you just mentioned’

10

CONTEMPORARY LINGUISTIC PARAMETERS

That language variation is more fine-grained, and at least in some apparent cases, gradual, is one of the reasons that have led scholars in recent instantiations of generative theory to reject the classic parameter view (see §2), but there have been other proposals that try to capture a more fine-grained nature of parametric variation while reducing parametric values to two. One influential proposal has been to allow for a hierarchical parameter structure where adopting a value of one higher-level parameter might be followed by the default selection the value of a second-level parameter subordinated to the first. Consider the following schema, taken from Roberts and Holmberg (2010: 49), where they assume that the presence of a particular class of features determines the licensing of pro (see D’Alessandro, this volume for more details): (14) Is class X of features obligatory in all probes? No Yes Are those features fully specified in those probes?

Yes

No Are those features fully specified on some probes? Yes

No

One can imagine a continuation of this algorithm where, if a given grammar has the most embedded parameter, we move to yet more specific parameters that divide grammatical categories into natural classes, potentially arriving through a sequence of binary choices to the same result that Rizzi’s formulation would have allowed. Note, however, that this solution has an important consequence that will become relevant later in our introductory comments; namely, the observation that parametric options are stated as choices about the nature of features contained in particular heads.

1.1.3. How many parameters and how well-behaved do they act together? Obviously, the next question would be how many classic parameters are necessary to capture the range of variation in language? General parsimony would favor an answer with a relatively low inventory of such axioms, but even if we leave aside the general preference for elegance and parsimony, the answer would have to be a highly restrictive and limited set of items. The rationale for this position is that classical parameters take a choice among the possible different ways of satisfying



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

11

universal principles imposed by UG. If the number of parameters is high, the number of principles will also have to be equally expansive, and then the endowment of UG would have to be significantly enriched. An outcome advocating an expansive set of parameters is clearly undesirable from the theoretical position of trying to arrive at a finite, restricted set of parameters for the primary purpose of simplifying the language acquisition process. Thus, a further problem for these classic approaches to linguistic parameters would be those proposals where a very high number of parameters is required to correctly capture the cross-linguistic variation. In a well-known paper, Longobardi and Guardiano (2009: 1697) study the properties of DPs in 28 languages, and come to the conclusion that the number of parameters required to correctly capture the variation instantiated just in this very small set is 63 binary parameters, among them: (a) whether the language expresses grammatical person (with separate parameters for number and gender, and also separate parameters to determine whether D has variable person or N expresses number); (b) whether definiteness is expressed grammatically (and, again, with a number of parameters determining whether there is definiteness checking in N, definiteness spread, definiteness on relatives, D-controlled inflection in N, free null partitives, etc.); (c) whether there are cardinal nouns (or cardinal adjectives, or grammatically marked mass-to-count shift, or plural spread from cardinals); (d) whether there is a prepositional genitive (or an inflected one, or whether genitives are postpositional, or whether there is possessive checking with N); (e) whether there is agreement with predicative adjectives (or demonstratives over relatives, or adjectival genitives, etc.); (f) whether there is N-raising with pied-piping (or N movement over A, or over cardinals, or whether D is strong); (g) whether there is strong deixis (or anaphoricity, or D checks demonstratives); (h) whether there are enclitic possessives (and whether there is feature spread to possessives, or whether possessives are checked by D); (i) whether there is a null N-licensing article, etc. Two issues should be noted at this point, though: first, these parameters are not formulated in the classic way. The way in which most of them are presented, as we saw in the case of Roberts and Holmberg (2010), is in the form of what we will call “lexical parameters,” that is, most of them make reference to the feature endowment of single heads (e.g. whether D contains person, whether N contains features that agree with other constituents), and those that are not directly formulated in this way (such as whether there is movement of N to a higher projection) allow for implementations that might use features. The second observation is that, in principle, it is not inconceivable that these 63 parameters could have been accounted for as epiphenomena of a much smaller number of classical parameters, if a hierarchical set of parametric choices (as 14) had been proposed. We could imagine a proposal that takes care of Longobardi and Guardiano’s 63 choices à la Roberts and Holmberg (2010). For example, there would be a first parametric choice, “Does N agree?” that, if answered positively, would be followed by a second parametric option, “Does agreement force N to move?” etc. This would predict, necessarily, implicational relations between the parametric choices: that if parameter P has value “+”, necessarily parameter P’ (hierarchically higher than P) must have a particular value (e.g. “+”), etc. The question is whether the raw empirical data allow us, as linguists, to postulate such implicational relations,

12

CONTEMPORARY LINGUISTIC PARAMETERS

and the answer that comes from Longobardi and Guardiano’s research (2009; see specially table A) seems to be “no.” In the limited survey they conducted, there is virtually no implicational relation between the value of one parameter and the value of other parameters that is systematic across languages (beyond those that are logically and independently tied together, such as whether category X expresses a particular property and whether there is grammatical expression of category X: obviously “+” values of the first will be matched by “+” values of the second, as if there is no category X in the language there will not be any kind of marking in X). Absence of implicational relations is not the only problem that more detailed typological work has presented to classical parameters; the clustering effects that are expected from classic parameters have also been empirically questioned. For instance, in the case of pro-drop, research on a wider set of languages than those initially considered in Chomsky (1981) and Rizzi (1982) has identified cases where (at least apparently) the properties that were supposed to come together “as a package” are at least partially independent of each other (see D’Alessandro, this volume). One of the leitmotivs of the chapters that the reader will find in this volume is, indeed, the empirical evidence that many (perhaps all) of the clustering effects associated to classic parameters are riddled with exceptions. To end this section, let us summarize the set of conceptual and empirical challenges that have presented problems to the classic formulation of parameters: a) The number of parameters that seem to be necessary to account for the attested variation is much higher than the P&P framework would lead one to expect. b) Frequently, there are no clear implicational relations between the values of these parameters (leaving aside logically necessary implications). c) Clustering effects are riddled with exceptions. d) Given the doubtful status of implicational relations, it is unclear whether parameters can always be reduced to binary choices, as language might distribute across a quasi-continuum of properties. This set of problems, combined with changes in the model of grammar assumed during the 1990s—from P&P to the Minimalist Program (MP)—led to a quite radical revision of the concept of “parameter” and its nature.2 We will discuss these revisions in sections §1.2 and §1.3, as they are largely related to the question of what kinds of parameters are proposed and source of the locus of variation in more recent mainstream generativist theorizing.

1.2. Relevant parameter distinctions In this section we will present the different classes of parameters that have been proposed in the current literature. As we will see, some of these distinctions appear as a reaction or a revision of the classical view of parameters as different choices in the way that principles are instantiated.



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

13

1.2.1. Grammatical vs lexical parameters Classical parameters are generally considered to be “grammatical,” in the sense that they operate over the subcomponent of the Language Faculty that is not the lexicon. The assumption is that the language faculty contains at least two components, a lexicon and a computational system; the lexicon is the component where (a) the lexical repertoire of a language is stored; (b) at least some of the formal properties of the lexical items are defined. There is a big debate with respect to whether lexical entries contain all formal features or whether at least some of them are configurationally defined once they enter the computational system, and whether the lexicon is simply an (expandable) list of items or it also contains rules and operations, but these debates are largely orthogonal to our purposes here. A classical parameter (such as the null-object parameter, which we reviewed in the previous section) is defined as a choice among the possibilities that the computational system offers, given the principles that restrict it. An alternative to grammatical parameters is lexical parameters, that is, a proposal where parametric choices are instantiated as choices over the feature composition of individual lexical items, the building blocks of structure, and not over the rules and conditions that determine the well-formedness of those structures. The origin of such an approach is often claimed to be Borer (1984; see specially 27–9). In this seminal work, Borer advances the proposal that a great deal of variation can be reduced to whether specific inflectional rules are available or not in a given language. Inflection—in the context where Borer writes her work—is understood as a set of conditions and rules that act as an interface between the lexicon and syntax and which determine the formal conditions of licensing of grammatical categories, their capacity to license the formal properties of other constituents through case and / or agreement, and their place in the articulation of functional structure. Given this first step, it was only natural that the development of the notion of parameter deriving from it would be one where the parametric differences observed between individual languages would best be captured through the feature endowment of (functional) heads rather than through choices in the grammatical conditions of licensing. If (many) parameters could be understood as the result of the (un)availability of inflectional rules, a natural extension would be to interpret those parameters as differences in the way in which different categories would be structurally licensed in different contexts. One straightforward way of finding a deeper explanation of why the conditions to license, let’s say, an object in two languages were different would be by positing different feature endowments for the relevant head in the two languages. If objects are licensed by V, then the difference between English and Italian could be that V has different features—so Italian is able, but English is unable, to license a null object—or even to say that Italian has an additional functional head in its lexical repertoire that English lacks. This lexical approach has been largely adopted in mainstream generative grammar (see, for instance, Chomsky 1995), and has come to be known as the Borer-Chomsky conjecture: (15) All parameters of variation are attributable to differences in the features of particular items (e.g. the functional heads)3 in the lexicon. (Baker 2008a: 156)

14

CONTEMPORARY LINGUISTIC PARAMETERS

This approach has been pursued in many works (among many others, see Thráinsson 1996; Giorgi and Pianesi 1997; Bobaljik and Thráinsson 1998; Koeneman 2000; Longobardi 2001b). Even those that still favor an approach where there are grammatical parameters (Baker 1996a, 2008a, 2010a) accept that these grammatical parameters have to be complemented with lexically defined parameters in order to provide a proper characterization of the data. The question we touch upon here probes in a little more detail why this approach has been so successful. Three reasons seem to have brought us to this point. The first one is empirical: a lexical parameter is expected to have more restricted effects in the grammar. A grammatical parameter, understood as a choice over the implementation of a principle, is like an umbrella under which constructions of many different classes must fall; a lexical parameter, at the very least, delivers the expectation that the effects of the parameter will be restricted to particular grammatical categories, functions or constructions—e.g. different licensing conditions for null objects and subjects because each one of them is licensed by a different head and the feature endowment of each one can be different in the same language. The critiques levied against the classical notion of parameters having to do with the unexpected absence of full clustering effects has frequently led to a revised version of this program, relying on feature endowment. The second reason has to do with the difficulty of adopting the approach that all grammatical parameters are binary, even when this is done by proposing a parameter hierarchy: as we saw in our discussion of Longobardi and Guardiano (2009), it is difficult to find solid implicational relations between the value of one parameter and the value of another one. A lexical approach to parametric variation does not encounter this problem, because the feature endowment of a head F is in principle independent of the feature endowment of head F’. A potential weakness of a lexical parametric approach is precisely this independence, as it would in principle predict that any combination of lexical properties could be possible (e.g. languages where prepositions agree in person, but verbs do not). However, in a lexical parametric approach it is imaginable that—perhaps as a further constraint imposed by learnability, or perhaps as a result of the way in which heads are related to each other in the lexicon—presence of a feature [f] in a head F implies presence of the same feature in a head F’, explaining at least partial interdependence. A grammatical parametric approach, on the other hand, would at some point have to refer to lexical differences to explain why in many languages V inflects but P does not. The third reason for the success of the perspective of lexical parameters has to do with changes in the framework and parsimony considerations. During the late 1980s and early 1990s, the P&P framework evolved into its most current instantiation; namely, the Minimalist Program. In contrast to the former, Minimalism aims to a reduction of the theoretical machinery used to constrain language structures. This reduction means questioning the existence of some levels (e.g. D-structure is questioned in Chomsky 1995; LF is questioned in Hinzen 2006), the questioning of the existence of some principles (which are explained not as part of UG, but as effects of general cognitive restrictions or, as it is said, third factor considerations), and the move from a representational system to a derivational one. The classic notion of parameter fits well with a representational system where, as in a quasiroad map, grammar evaluates the distance between items and the local context



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

15

in which each element is placed. However, when the system is envisaged to be largely derivational, the legitimate, well-formed structures are not determined by principles that consider the configuration as a whole, but the legitimate configurations themselves emerge as a result of the properties of each item and their licensing conditions. The attention, therefore, moves to the properties of individual items, and these properties are explained through features. Different configurations in language X and Y are, then, effects of the different properties of lexical item LI in each language, which is expressed through a different feature endowment for LI in X and Y. Thus, an orthodox Minimalist framework in principle favors a lexical view of parameters. Once a lexical view of parameters is adopted, we arrive at a different predicted set of properties from those that defined the classical parameter: a) The effects of the parameter setting might be local, in the sense that they only affect structures containing the relevant head and that their effect will be limited to those (local) configurations where the feature remains syntactically active: clustering effects, then, are not expected. b) No implicational relations are expected, in principle and unless imposed by external factors, between the parameter setting (=feature endowment) of head F and that of head F’. c) Two main specific sources of variation are expected: absence vs presence of a head F (e.g. D, Inflection, etc.) and absence vs presence of a feature or a set of features in a head F. Variation cannot be expressed, however, through different conditions of licensing in language X and language Y: licensing will be carried out in the same way, and what can vary is which head licenses, given its feature endowment. But again, the size of the predicted effects of a lexical parameter can vary. Two factors are at play here. The first one is that the term “lexicon” can be interpreted in different ways, and with it, the meaning of “lexical” (see Fábregas, this volume). “Lexicon” can refer to a set of abstract matrixes of features, as we just explained, but it might also refer to the repertoire of exponents (see below), in which case its effects can be quite superficial. In the other extreme, “lexical” can refer to the way in which the building blocks of syntactic structures are organized in a language, and the rules that can operate on them. In this second case, the effects of a parameter could be quite wide ranging: it might affect to the acceptable operations that supress or add arguments to a lexical entry, or to the way in which concepts are mapped into lexical categories. Take, for example, Reinhart and Siloni (2005), who propose that some languages can define lexical operations that reduce the valency of a predicate, with subsequent general effects in the syntax. This takes us to the second factor that might lead us to expect big consequences from a “lexical” parameter. If “lexical” is interpreted as “the general classification of the building blocks of syntax,” a lexical choice can have syntactic consequences in other, apparently independent domains, simply because there is a semantic or conceptual factor that determines the behavior in the two domains at the same time. Consider Stassen’s (1997) work on adjectives. In his typological study, he concludes that there are no languages where predicative adjectives are grammaticalized with an independent strategy (see also Wetzer 1996; Baker 2003). Property-denoting

16

CONTEMPORARY LINGUISTIC PARAMETERS

concepts are mapped reusing patterns that are independently used for verbs, nouns, or adpositions. This choice is not free, however: it is matched by whether the language in question is tensed (Stassen 1997: 350–1). If the language has a grammatical category of tense, which is morphologically bound on verbs and minimally differentiates between past and non-past reference, property concepts will not be mapped into verbs. There is, then, a correlation between treating properties as verbs and lacking strong grammatical tense information, which seems to be a quite wide-ranging effect for a “lexical” parameter. The problem is solved, however, because there is no causative link between tenseness and the verbiness of adjectives, despite appearances: they are both effects of the same independent grammatical parameter, namely the strength of tense as a deictic category. On the assumption that property-denoting concepts are, in principle, incompatible with variable tense specifications, adjectives will be treated as verbs only if tense is not a category that combines with verbs, or is not specified deictically in a way that does not match our default assumptions about the temporal projection of properties; otherwise, adjectives are treated as nouns. The moral is that, when assessing the effects of a parameter, correlation is not enough: a causative link has to be argued for independently. See also Croft (1991), Dixon and Aikhenvald (2004), Himmelmann and Schultze-Berndt (2005), and Verkerk and Lestrade (2008) for different aspects related to the variation found in the category of adjectives, both in attributive and predicative construals.

1.2.2. Macro-parameters vs micro-parameters An overlapping, although in principle independent, distinction has to do with macro-parameters vs micro-parameters. Stricto sensu, macro- and micro-parameters are differentiated not by their effects in a single language (clustering effects vs local effects), but by the kind of typological distinction they want to study and by their methodology. Baker (2008a, 2008b) is currently the main proponent of the macro-parametric approach, while research by Kayne (2005a) is generally identified with the microparametric approach. A micro-parametric approach studies the differences between closely related varieties of what (politically) is considered the same language: in a traditional terminology, micro-parametric approaches study dialectal variation. In contrast, macro-parametric approaches are interested in capturing and explaining the contrasts that divide big families of languages, such as Indo-European vs Semitic. In this context, classical parameters are medium-size parameters—not really macro- and not really micro-, because they typically aimed to study variation in different branches within the same family (Germanic vs Romance, for instance). A macro-parametric methodology involves, then, the comparison of the general distinctive properties of typologically unrelated languages (e.g. whether arguments of a predicate can form one single morphological word with that predicate), while the micro-parametric methodology generally involves the construction-specific comparison of a number of dialectal varieties of the same language (e.g. whether the combination of the verb to be and a particular preposition gives a deontic meaning or not). This distinction is, however, a bit slippery: within the micro-parametric



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

17

approach it is sometimes the case that comparisons are made between different languages—e.g. Kayne’s 2013 selective comparison between (a variety of) English and (a variety of) French. Similarly, the macro-parametric approach does not, by any means, ignore the internal variation inside a language. What is different, however, is the preferred explanation of the distinctions. Micro-parametric approaches tend to propose that the same minimal featural differences that can explain the distinction between two related “dialects” in one “language” should be adopted to explain the distinctions between two typologically not very different languages like French and English. Macro-parametric approaches, in contrast, tend to propose that within a language (family), the parametric values are set in the grammar. Small differences in their internal varieties are due to the “noise” that the idiosyncrasies of specific lexical entries produce in each one of those varieties. There is a further asymmetry between the macro-parametric and the microparametric view. A macro-parametric view commits to a grammatical view of parameters: if the wide-ranging distinctions between language families are to be explained systematically, the lexicon is not the place to look for the explanation of such differences, because, in principle, the lexical repertoire of a language is asystematic, full of idiosyncrasies and is known to vary across the lifespan of speakers—new forms can be listed, new exceptions can be learnt—and across varieties. A micro-parametric view, on the other hand, might favor a lexical view of parameters—because it is easier to ensure mutual intelligibility across varieties through minimal and local distinctions in lexical entries than through wideranging grammatical choices—but could be compatible with a particular view of grammatical parameters provided that the parametric choices can be different across lexical classes (as, for instance, Hyams 2008 proposed for the inflectional properties of verbs). Thus, the macro- vs micro-parameter distinction overlaps, but is not identical to, the grammatical vs lexical parameter distinction.

1.3. Other accounts of variation A related question that again overlaps with the kind of parameter one proposes concerns the level or component of the language faculty which is responsible for variation. We have seen already two kinds of answer that are, in a sense, forced by the 1980s and 1990s view of grammatical and lexical parameters: a) Variation resides in the computational system “grammar,” in the form of choices restricted by universal principles. b) Variation resides in the lexicon, in the form of differences in the feature endowment of heads or even the availability of heads. However, there are other options. As we will see, these options have emerged as a result of changes in the theoretical framework adopted, and to a lesser extent as a result of the discovery of new empirical patterns or data. In some cases, as we will see, these options are proposed in addition to some of the more traditional options, basically implying that language variation is the combined effect of variation in

18

CONTEMPORARY LINGUISTIC PARAMETERS

two or more levels (as in Distributed Morphology (DM); Halle and Marantz 1993; Embick and Noyer 2007); in other cases, these options are meant to substitute completely the traditional view of variation.

1.3.1. Externalization The evolution of the Minimalist Program (Chomsky 1995) has involved the pursue of two ideas that have been gaining progressively more support in this framework (Hauser, Chomsky, and Fitch 2002): (a) the pursuit of the hypothesis that UG is as impoverished as possible, with what previously had been identified as principles actually coming as a result of third factor constraints, that is, general efficiency principles not specific of the language faculty; (b) as a result of the first hypothesis, the view that operations taking place in grammar aim to satisfy some constraint(s) imposed at the interface with the external systems, the sensori-motor system that turns the linguistic structure into a physical signal that can be perceived (e.g. sounds, hand shapes, paralinguistic jesters, etc. ) and the conceptual-intentional system that converts / computes the linguistic structure into a semantic representation. Being open to the possibility for variation in the interface between the narrow language faculty and semantics has quite radical consequences—postulating semantic parameters might be close to saying that different languages impose different logical and conceptual systems, or, at least, that categories might be mapped to different concepts in different languages. Nonetheless, some semantic parameters have been proposed, normally framed inside a rich view of LF where the mapping between categories / configurations and semantic types and operations is non-trivial. Let us highlight some of these proposals. Snyder’s (1995, 2012) compounding parameter is currently understood as a choice with respect to whether syntactic sisterhood can be mapped into general subclass interpretations at the semantic component (“Generalized Modification”). Chierchia (1998a, 1998b) influentially proposed that not all languages map syntactic units (such as noun prhases, NPs) into the same semantic type. Against these approaches, however, one finds the increasing volume of literature arguing against an articulate and complex LF-level (cf. Pietroski 2005; Lohndal 2014). An intermediate position is the one that argues that the narrow language faculty imposes a formal architecture without substantive semantic content, and it is subject to variation which (set of) concepts will be embodied by functional heads (person, time…); cf. Wiltschko (2014 and this volume). In contrast, the other external system—the one that leads to a physical signal—is more open to variation. This is the option pursued in Berwick and Chomsky (2008): the language faculty essentially combines simple concepts into complex structures in order to ensure its properties are designed specifically to meet the requisites of the conceptual-intentional system. However, it is not designed to meet the principles of the sensori-motor system, which has (as stated explicitly in Chomsky 1999) all the non-minimalist properties that one can think of (in the sense that it is not optimally designed, because it has to makes explicit use of organs that were selected for other purposes, such as the throat, the mouth, or the hands). Externalizing the structure into a physical signal is, thus, problematic, and no single perfect solution is available—that is, no perfect solution is defined in the system. Thus, there are



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

19

different imperfect or less than perfect solutions that are conceivable, and variation could emerge from this: languages vary because of the choice of the way in which they solve the problem of how to adapt a linguistic structure into something that can be used by those organs which were not designed with a linguistic purpose. The hypothesis is, then, that one crucial locus of variation is to be found in the so-called PF-branch of the grammar, understood as the interface between syntax and phonology that leads towards a physical signal. Note that, crucially, this kind of variation would be completely free of semantic effects, because it takes place at a point where there is no direct interaction with semantics: accordingly it can be guaranteed that LF remains variation-free. Let us briefly illustrate this change of perspective through Richards’ work (2010: 143–203) in this domain. Richards tries to account for variation in wh-constructions, and specifically whether the wh-word has to move to C or not. This was something that classical parameters (Huang 1982) explained through a parameter that determined the representational level (S-structure or LF) at which movement took place, and that later on was recast as a distinction between two classes of features, strong and weak (Chomsky 1995). Richards, however, pursues a PF-account. All languages have to meet the following PF condition on interrogative constructions: C and the wh-word must be inside the same prosodic phrase. (16) [φ C   wh] Imagine, then, that the initial prosodic phrasing is something like (17), with one or more different prosodic boundaries between C and the wh-word: (17) C [φ ] [φ] [φ wh] To start with, there are two conceivable solutions, and none of them is optimal— and none of them is thus selected by UG or third factor considerations: either the prosodic boundaries in (17) are eliminated, restructuring the whole sequence into one single prosodic unit (as in 16), or the wh-word has to be pronounced in a different position, so that no prosodic boundaries intervene, i.e., “movement.” (18) [wh C [φ ] [φ] [φ wh]] Essentially, wh-movement or in-situ questions are two options that solve the same externalization problem, each one in a different way. The typology is further refined by whether the language defines the left or the right-boundary of a prosodic unit; we refer to Richards’ original work for details. Richards’ view is intended to substitute—not to complement—more traditional parameters inside a new theoretical universe. Not all theories that propose variation in the PF-branch adhere to this view. In Distributed Morphology, for instance, part of the variation is expressed through featural differences in functional heads, but there is a second part that is attributed to the availability of PF-operations. Distributed Morphology proposes that before the insertion of morphophonological exponents in terminal nodes, the syntactic structure is subject to a number of operations, which they call “morphology” in the proper sense. The mapping between

20

CONTEMPORARY LINGUISTIC PARAMETERS

syntax and morphology might be trivial, simply transforming syntactic heads X0 into positions of morphological exponence M0, but it can be more complex. Depending on the particular school of Distributed Morphology that is adopted, not all of these operations are available, but the following have all been proposed at some point or the other: a) Morphological merger, that inverts the ordering between heads. b) Fusion, that maps two syntactic heads into a single morpheme. c) Fission, that maps one syntactic head into two morphemes. d) Impoverishment, that ignores one or several features of a head for lexical insertion. e) Obliteration, that erases a whole head (or maps a head to no position of exponence). f) Insertion of dissociated morphemes, that create new positions of exponence not corresponding to any syntactic head. A second source of variation, thus, would be whether and where these operations are available. Languages could vary, for instance, in whether they create positions of exponence for agreement in different functional heads (v, T, etc.), producing overt agreement or not that might, eventually, be relevant for whether a null argument can be identified.

1.3.2. Size of exponents A proposal that, in a sense, lies between the lexical parametric view (in a particular interpretation of “lexical”) and the externalization account is the one pursued in Nanosyntax (Svenonius, Ramchand, Starke, and Taraldsen 2009). Nanosyntax is a neoconstructionist system which assumes a cartographic approach to grammar (Rizzi 1997, Cinque 1999). In a nanosyntax-approach to parameters, there is a richly defined functional sequence that is universal; moreover, each head in that sequence corresponds to one and only one formal feature. Where more standard theories would claim that there is a D head containing the features [2 person, singular], nanosyntax would argue that there is a hierarchy of heads (for instance, Addressee, Participant, Person, Number—assuming that “singular” is the default interpretation of “Number”). In this approach, then, variation cannot be due to the feature endowment of heads, because the available features are defined by a universal functional hierarchy. Variation is due in this theory to the size of lexical exponents; let us see how. Nanosyntax follows McCawley (1968) with respect to the Phrasal Spell Out proposal: single exponents can lexicalize complex phrases, not just terminal nodes. In the lexicon, exponents are associated to a tree structure; when the syntactic component of the grammar provides this tree, the lexicon introduces the exponent in order to lexicalize the relevant tree. A lexical entry would look like the treelet in (19):



“PARAMETERS” IN LINGUISTIC THEORY: WHAT, WHERE, AND HOW

(19) /exponent/





21

XP X



YP

Y

Z

If the grammar builds the structure in (20), then, the exponent in (19) is introduced: (20)

XP

X

[[±Tense], (Agr)] What this does is to restrict the available feature combinations inside the head Infl, which is assumed to carry both tense and agreement features, as the article is pre-split-Infl (Pollock 1989). In languages such as Spanish, whenever agreement is present, Tense has to be positively specifiied, so all inflected clauses are finite; in languages like Portuguese, where the parameter in (27) is set positively, agreement is compatible with finite and non-finite verbs. Consequently, Portuguese allows inflected infinitives able to license preverbal subjects without the help of other elements (28). (28) Eles aprovar-em a proposta será difícil. they to.approve-agr the proposal will.be difficult ‘For them to approve the proposal will be difficult’ We will not discuss further the details of Raposo’s (1987) proposal, but we want to highlight that the way in which the parameter has to be stated makes it very difficult that it can be considered a macro-parameter. Approaches where there is a parameter that determines whether all heads in one language can or must have uninterpretable features are almost non-existent, but one exception is Huang (2010). This author proposes that there is a macroparameter that explains a number of wide-ranging differences between an isolant language (like Modern Chinese) and others (such as English). Languages that have the following properties (among others) are called languages with high analyticity, and are therefore of the isolating kind.

40

CONTEMPORARY LINGUISTIC PARAMETERS

a) Prevalence of light verb constructions (e.g. da yu, lit. do fish, ‘fish’) instead of single semantically complex morphemes. b) Phrasal compounds: zhuo yu, lit. catch fish, ‘fish’. c) Aspectually complex verbs, like accomplishments, are expressed by more than one morpheme: da po, lit. cause break, ‘break’. d) Count nouns are expressed by overt classifiers (cf. Pereltsvaig, this volume). e) Cannonical Subject-Adjunct-Verb-Complement word order. f) Wh-in-situ. g) No determiner. Huang is not explicit in determining what the ultimate parameter responsible for this cluster of properties is, although he is clear that in his view it must be a macroparameter that later interacts with the specific feature setting of individual heads. His suggestion is that Chinese is a language that lacks formal features that trigger overt movement. This has two immediate consequences: word order would be “classic,” following that imposed by the height at which different participants are introduced, and it will be impossible that a morphologically empty head becomes filled by movement of a lower head. English would allow a situation like (29a), where v is empty, and has a formal feature that triggers movement of the noun; Chinese would not (29b). Movement is impossible in that language for lack of uninterpretable features in v, and therefore no complex word *yu-da ‘fish-do’ can be built. (29) a.

vP

b.

vP

v+N NP v NP [uF] da N … N … fish yu Note that Huang, in the proposal of a macro-parameter, must assume that word formation is a syntactic process. After having reviewed a number of different parametric proposals related to different aspects of morphological variation, in the next section we will evaluate what these phenomena and proposals tell us about the kind of parameters that languages require and their place in the architecture of grammar.

2.3. The locus of variation in morphology The short presentation above shows that different kinds of parameters have been proposed to restrict word-formation rules on a language, determine the features that can be bundled together under the same head or to impose constraints on what counts as a minimal word. Although these proposals show a variety of interests and perspectives, two properties emerge as general of them. First, whenever the proposal involves formal features—specially, but not exclusively, agreement and



MORPHOLOGICAL PARAMETERS

41

other inflectional properties—the parameter tries to relate that property with a syntactic process, typically movement and argument pro-drop. That is, we have seen no proposals where availability of a purely morphological process—e.g. compounding—has an impact for another word-formation process—e.g. categorychanging derivation, or the shape of inflectional paradigms. Secondly, without any exception, all the parameters that deal with morphological processes need at some point or the other to be divided into smaller parameters. Sometimes, the parameter is formulated from the start as a claim relativized to only some heads (functional, as Raposo 1988, or lexical, as Hyams 2008); sometimes, the parameter is formulated as a general principle but the behavior of specific languages suggests that it should be divided into smaller parts (e.g. Baker’s [1996a] own observation that the morphological visibility condition parameter should probably be divided in at least two parameters, depending on the availability of noun incorporation; Huang [2010] also notes that a macro-parametric approach also needs to be complemented with micro-parametric choices to account for apparent exceptions and fine-grained contrasts). The goal of this section is to start from these general properties as a way to assess whether morphological parameters are a good tool to study morphological variation and, if so, what kind of parameters would be necessary. The discussion about whether morphological variation relates to macro- or micro-parameters is necessarily intertwined with the so-called Lexicalist Controversy (e.g. Embick and Noyer 2007 vs Williams 2007). This debate discusses whether morphology is a separate generative component of grammar or not. For Lexicalism (Halle 1973; Aronoff 1976, 1994; Bauer 1978; Lieber 1980; Scalise 1983; Anderson 1992; Stump 1993, 2001, among many others) there is a special component of grammar, the lexicon, which contains word-formation rules by which words are put together—at least, compounds and derived words, with an internal controversy with respect to whether inflection is dealt with in syntax or is also part of the morphological component; such theories are called endo-skeletal (Borer 2003) because they assume that the morphological component determines all grammatically relevant properties of words, which are unanalyzable atoms for syntax: their argument structure, their grammatical category, their semantic information, etc. Neo-constructionist theories (Lieber 1992; Halle and Marantz 1993, 1994; Harley 1995; Noyer 1997; Marantz 1997, 2001; Borer 2005a, 2005b, 2013; Harbour 2007; Starke 2009; Arregi and Nevins 2012, among many others), in contrast, deny that “words” are special kinds of units for syntax: all combination of units, morphemes or not, are performed in the syntactic component of the grammar, which determines properties such as argument structure and grammatical category in the course of the derivation—that is, the approach is exo-skeletal. In principle, a macro-parametric approach is more compatible with a neo-constructionist view or a lexicalist view where inflection is part of the syntactic component. The reason for this assertion is twofold. First, no clear systematic clustering of properties with respect to word formation has been noted across languages: as mentioned before, no researcher has found solid principles that state that if a language has, say, compounding, it also has productive noun to adjective derivation, or inflectional properties of some specific kind. Second, the basic assumption of a lexicalist theory is that one has to recognize a morphological component precisely because of the existence of very restricted patterns of productivity and idiosyncratic

42

CONTEMPORARY LINGUISTIC PARAMETERS

exceptions to otherwise regular processes (Chomsky 1970), so the very fact that one identifies a solid correlation between processes suggests to a lexicalist researcher that the explanation should be in the field of regular, highly productive, syntactic rules, not morphological operations. When one can identify systematic restrictions to word formation, as in high analyticity (§2.2.4) the restriction is so general—“no complex words, not more than one head per exponent”—that the explanation seems syntactic in nature, for instance related to the availability of overt movement in the language. Macro-parametric approaches related to properties that Lexicalism could associate to the morphological component normally tie together an inflectional property, like whether Infl is finite, with one or several syntactic processes, or do not make mention to any syntactic process at all (as Hyams 2008). There is one exception, though, that shows that a lexicalist theory is not logically incompatible with a macro-parametric approach: on the assumption that grammar has two distinct generative components, a parameter can relativize whether a grammatically relevant constraint has to be fulfilled in syntax or in morphology. This is what the Morphological Visibility Condition of Baker (1996a) does: it determines whether the theta-criterion has to be fulfilled morphologically—that is, at a word level, or syntactically—at the level of the phrase. With that exception—the choice between morphology and syntax, which is also implicit in other lexicalist approaches, like Ackema and Neeleman (2004)—it is perhaps fair to say that lexicalism is naturally inclined to a micro-parametric approach. A micro-parametric approach provides the researcher with the possibility of defining at a very fine level of granularity distinct properties for different classes of heads—e.g. whether nouns are inflected for case, whether there are features that allow compounding of categories X and Y. This would make it possible to explain restricted patterns of productivity inside otherwise non-productive systems. For instance, Italian has very restricted compounding possibilities, with the significant and puzzling exception of V-N compounds of the form in (30). A single macroparametric approach to compounding would run into trouble when addressing such cases, but a micro-parametric approach is in principle imaginable, where Italian has an available head with features that allow this kind of combination. (30) lava-piatti wash-dishes ‘dish washer’ The crucial question is whether micro-parameters are also needed in the case of neo-constructionist theories. It is clear that, at least, a micro-parametric approach is logically possible also in a neo-constructionist theory: if micro-parameters are understood as the fine-tuning of the feature endowment of single heads, to the extent that features are responsible for triggering syntactic processes (see Chomsky 1995, 2000, 2004, among many others), micro-parameters are possible in neo-constructionist theories for the same reason that they have become possible in syntactic analysis. The other question is whether they are necessary. Here, most researchers concede that they are, and the question is whether they are necessary in addition to macro-parameters or not. All possibilities have been proposed: Baker (1996a) proposes to divide the



MORPHOLOGICAL PARAMETERS

43

Morphological Visibility Condition in two macro-parameters—determining whether subject–object agreement morphemes are available and whether noun incorporation is available, respectively—and argues that beyond that it is not necessary to relativize the parameter to the feature endowment of specific types of head. The consequence is that Baker (1996a: 11–17) has to treat as stored lexical exceptions the cases of apparent noun-incorporation available in non-polysynthetic languages (e.g. 40b, from Spanish). Whether this move is right or not would depend on the analysis of cases like (40), and whether other data suggest a lexical listing solution or generation by rule (with possible intermediate solutions like analytic listing, see Bermúdez-Otero 2012). (31) a. adicto a las drogas addicted to the drugs ‘addicted to drugs’ b. droga-adicto drug-addict However, other approaches argue that macro- and micro-parameters are both required: a macro-parameter presents a general rule, but this rule can be superseded through lexical exception by specific micro-parameters. This is one possible way to interpret the core vs periphery distinction done in part of the literature—with core-grammar being the field of macro-parameters, and the periphery requiring a micro-parametric explanation. Huang (2010) is also someone who argues that both kinds of parameters are equally necessary: even though high analyticity makes overt movement unavailable in general in Modern Chinese, in particular varieties one can find occasionally cases where such movement seems to be possible. For instance, in Cantonese Chinese (41a), no overt numeral for “one” is necessary, in contrast to other varieties (like Taiwanese, 41b). His proposal is that a micro-parameter allows the exception that the head corresponding to the numeral one ([e] in 42) carries a feature that triggers movement of the classifier to its projection. (32) a. ø ben CL ‘one book’ b. yi ben one CL ‘one book’

shu book shu book

(32) [NumP [e][+F] [ClassP ben [NP shu]]] To conclude, it seems fair to say that given the current state of knowledge microparameters seem to be a better solution to account for morphological variation. However, this is based on the fact that up to know no clear clusters of systematic and exception-free morphological properties have been identified in word-formation. If such clusters of properties are identified in the future, the need for macro-parameters in morphology could be also motivated.4

44

CONTEMPORARY LINGUISTIC PARAMETERS

Next to classical macro-parameters and micro-parametric choices, another current option is to reduce morphological variation to very restricted microparameters operating at the PF-level. Distributed Morphology (Halle and Marantz 1993) admits the existence of a morphological component, but this component is part of the PF-branch of the grammar (Embick and Noyer 2001, 2007). This makes it possible to put the lion’s share of morphological variation as operational choices that determine how a particular syntactic structure is spelled out, on the assumption that vocabulary items are late-inserted into structural configurations. (33) presents a simplified version of Distributed Morphology’s structure of grammar. (33)

Narrow lexicon (abstract bundles of formal features)

  Computational system (syntax)

PF

LF

Morphological operations

Vocabulary insertion This system enables late insertion of lexical items: first, abstract structures are built and modified; then, lexical items are inserted on the available terminals. Morphological variation can, actually, have two sources in this system. It might be due to the features that are bundled together in the narrow lexicon, following the choices over the feature geometries made available by universal grammar: that variation will have an effect over the syntactic structures available. But, crucially and more importantly, variation can be determined by the choice of operations that take place at PF, in which case they have no import in syntax. These operations are viewed as rules that adapt the syntactic structure built by the computational system so that they can be compatible with the lexical repertoire of a language, which is accesses at Vocabulary Insertion. That is: morphological operations regulate the mapping from abstract syntactic structures to morphological structures where the lexical repertoire can be introduced. Given this intermediate position, Distributed Morphology can relate morphological variation to variation found in the lexical component, and reduce both to a PF-branch effect.5

2.4. Some words of conclusion Morphological variation illustrates in a particularly clear way the tension between the need for general explanations and the need for exceptions and reduced patterns of productivity. In a morphological system, there are always some general or quasigeneral principles—which allow to define, for instance, highly analytic languages vs other classes—next to semi-productive patterns, processes with arbitrary gaps, when not pure and simple lexical exceptions. Macro-parametric approaches seem



MORPHOLOGICAL PARAMETERS

45

to be required to capture some of the general properties of the system, but raise the question of whether they should be split into smaller, but also general, parameters to account for different aspects of the system. Micro-parametric approaches are good at defining restricted patterns through the choice of features in a previously defined space, but fail to capture systematicities across categories or across members of the same category. PF operations, still to be fully developed and explored, are an interesting development, because of their intermediate character between syntax and the lexical repertoire, which makes them in principle able to refer to general properties of structures or to highly idiosyncratic pieces of information of lexical items; however, given the architecture of grammar standardly assumed, they cannot have a impact in syntax. Perhaps part of the inherent problem when studying morphological variation is its interaction with the lexical repertoire. It seems, then, that advances in the understanding of variation at this level will be closely related to how much we explore and understand about the interaction between structures and lexical items.

Notes 1

The term “lexicon” is used in most of the literature—specially that which falls inside lexicalism (Halle 1973)—to refer to a generative engine of grammar used to account for the internal structure of words. In this sense, the “lexicon” contains a list of items with idiosyncratic information—what we are calling here the “lexical repertoire”— and a set of word formation rules, which are part of what we are calling here “morphology.” The variation phenomena found in each one of these two components are very different.

2

Given its importance for argument structure and its semantic nature, we do not review here Snyder’s (1995, 2012) Compounding parameter. See Acedo and Mateu (this volume) for the relevant discussion.

3

Julien (2002) rejects the lexicalist hypothesis, and therefore, the proposal that words are grammatical units. Therefore, she considers in her survey both affixes and morphologically free markers; in the table, “+” marks a morphologically dependent form.

4

Feature geometric approaches (cf. Harley and Ritter 2002; Cowper 2005) are articulate attempts to restrict the set of features that the language capacity allows in lexical clusters. For reasons of space, we do not elaborate on these approaches here.

5

See, among many others, Bonet 1991; Noyer 1992, 1997; Oltra-Massuet 1999; OltraMassuet and Arregi 2005, for how morphological operations account for partial systematicities in morphological variation.

3 Case. Ergative languages* Michelle Sheehan University of Cambridge

3.1. Introduction1 The expression of (nominal) case and (verbal) agreement is one of the most obvious points of variation across natural languages. First, it is well known that some (analytic) languages lack morphological case and agreement altogether: (1) (ta) kanjian (ta) le 3SG see 3SG ASP ‘S/he saw him/her.’

[Mandarin]

It is a controversial topic, however, whether such languages nonetheless require a syntactic notion of (abstract) Case (see Li 1985 and Huang, Li, and Li 2009 for arguments that Mandarin does and Sheehan and van der Wal 2015 for a crosslinguistic study of superficially similar languages).2 Other languages have only case or dependent marking and lack agreement or head marking (Japanese) or vice versa (Swahili): (2) Makiko-ga Yoko-o mita Makiko-NOM Yoko-ACC see ‘Makiko saw Yoko’

[Japanese]

(3) (yeye) a-li-mw-ona (yeye) 3sg SM1-PST-OM1-see 3sg ‘S/he saw him/her’

[Swahili]

A final group of languages have both case and agreement, with the two sometimes fitting together in predictable ways (though not always, as we shall see below). Thus in many languages, it is the nominative argument which triggers verbal agreement

48

CONTEMPORARY LINGUISTIC PARAMETERS

in finite clauses. In Icelandic this holds regardless of which argument is nominative (see Bobaljik 2008 for discussion): (4) Jóni líkuðu þessir sokkar Jon.DAT like.PL these socks.NOM ‘Jon likes these socks’

[Icelandic] (Jónsson 1996: 143)

Even among those languages which have case and/or agreement, though, there is considerable variation with respect to what we shall call “alignment.” In many familiar Indo-European languages the subjects of intransitive predicates generally behave like the subjects of transitives in terms of case / agreement, bearing nominative case and triggering agreement on the verb: (5) a. Sie sahen ihn [German] 3PL.NOM see.3PL 3SG.ACC ‘They saw him’ b. Sie kamen 3PL.NOM   came.3PL ‘They arrived’ Such a pattern is usually called “nominative-accusative” or simply “accusative,” and we use the shorter label here.3 According to Chomsky (1995: Chapter 4), case/ agreement morphology are surface realizations of a single deep Agree operation. If this is correct, then in syntactic terms, Japanese, Swahili (and English for that matter), like German, have a basic accusative alignment, whereby both transitive and intransitive subjects agree with T/INFL.4 An apparently less frequent but nonetheless widely attested pattern arises where the subjects of (some or all) intransitives pattern like the objects, rather than the subjects, of transitives in terms of case and / or agreement, bearing absolutive case, for example, in Yup’ik, or triggering object-like agreement on the verb in Chol (see Dixon 1994 for an overview): (6) a. Angute-m qusngiq ner-aa [Yup’ik, Eskimo-Aleut] man-ERG reindeer.ABS eat-TR.3SG/3SG ‘The man is eating (the) reindeer’ b. Qusngiq ner’-uq reindeer.ABS eat-INTR.3SG ‘The reindeer is eating’ (Bobaljik 1993: 3) (7) a. Tyi i-kuchu-yety PRFV 3ERG-carry-2ABS ‘He carried you’ b. Tyi juli-yety PRFV arrive-2ABS ‘You arrived’

[Chol, Mayan]

(Coon and Preminger 2012: 4)

This basic distinction in the behavior of intransitive subjects can be graphically



CASE. ERGATIVE L ANGUAGES

49

described as follows, following Plank (1979) and Dixon (1979, 1994) where A = transitive subject, S = intransitive subject and O = transitive object:

  S A

S P

A

P

Ergative-absolutive Nominative-accusative (ergative) (accusative) FIGURE 3.1  Representation of ergative and accusative alignments

The question of how best to account for ergativity in parametric terms has long occupied generative syntacticians, with various competing analyses having been proposed (see section 3.4, below).5 From the very earliest studies it has been recognized that there are different kinds of ergativity: so-called deep ergativity vs shallow (morphological) ergativity being a common distinction (see Levin 1983, Marantz 1984). As such, it appears that any parametric account of “ergativity” will involve several distinct parameters rather than a single macro-parameter. In other words, while languages in which (some or all) intransitive subjects pattern like transitive objects are often grouped together as “ergative,” this is a simplification of the truth as this group of non-accusative languages is heterogeneous in character. In fact, it has often been claimed that while a language can be fully accusative in alignment, no language is fully ergative (Moravcsik 1978; Dixon 1994). The aim of this chapter is to consider the implications of variation in alignment for parametric theory. To this end, section 3.2 examines the phenomenon of variable alignment, whereby a language appears to display ergativity only in a given domain (sensitive to tense / aspect, clause type or the properties of its DP arguments). Section 3.3 introduces the various types of (split) ergative systems which have been described in the literature, and mentions some of the theoretical challenges they pose. Section 3.4 describes the main generative approaches to ergativity and discusses their implications for parametric theory. Finally, section 3.5 concludes and proposes some areas for future research.

3.2. On variable alignment Traditionally, the term “split alignment” is used to refer to two very different phenomena. On the one hand, a language is called split ergative if it displays only partially ergative behavior, for example if its case pattern is ergative in alignment but its agreement is not, or if only a subset of its intransitive subjects pattern like transitive objects. On the other hand, languages which appear to be ergative only in certain syntactic contexts, for example in the presence of perfective aspect, are also termed “split ergative.” To distinguish these very different phenomena and avoid ambiguity, we shall reserve the term “split ergativity” for the partially ergative

50

CONTEMPORARY LINGUISTIC PARAMETERS

systems described in section 3.3. For the other phenomenon, which we consider in this section, we will use the term “variable alignment.” In many unrelated languages, ergativity (the patterning of some or all intransitive subjects with transitive objects) appears to arise only in certain restricted syntactic contexts, typically sensitive (in cross-linguistically stable ways) to tense / aspect, the matrix / embedded distinction or the features of DP arguments (Dixon 1994). Thus, in many unrelated languages, pronominal transitive subjects fail to be inflected as ergative unlike full DPs, or ergative case surfaces in perfective / non-progressive but not imperfective / progressive aspects. Finally, it has been claimed that some languages display an ergative alignment only in embedded but not matrix clauses. Variable alignment poses potential challenges for a parametric approach to alignment. On the surface, the implication seems to be that the very same language can have multiple parameter settings for alignment relative to distinct syntactic domains. While it is at least possible that different tense / aspect heads might assign different cases in the same language (see Ura 2006 for a proposal along these lines), having a different parameter setting in say embedded vs matrix clauses seems more problematic.6 Interestingly, though, it has recently been argued that variable alignment may actually be an illusion (see Laka 2006a; Coon 2012; Coon and Preminger 2012 for discussion). We consider the three kinds of variability in turn here.

3.2.1. Sensitivity to T/Aspect Consider first the sensitivity to tense / aspect, attested in Basque (isolate), Tsez (Nakh-Daghestanian), Georgian (Kartvelian), Q’anjob’al, Chol (Mayan) and Hindi (Indo-Aryan) among other languages (see Tasaku 1981; DeLancey 1981; Coon 2012; Coon and Preminger 2012). It is very generally the case in such languages that ergative alignment surfaces only in perfective contexts, something that has led to various functionalist explanations. Consider, for example, DeLancey’s (1981) approach which attributes the occurrence of ergative case morphology in the perfective to the need to mark the subject as viewpoint, because, he claims, perfective aspect, like the passive, leads to default viewpoint on the patient. An alternative account of this apparently universal tendency attributes it to a well-known process of diachronic change whereby a passive form is reanalyzed as active and the oblique marker surfacing with the agent is reanalyzed as ergative (see Anderson 1988). As Anderson notes, the fact that ergative alignment is often associated with perfective aspect is thus merely due to the fact that perfectives are often used to form passives, and where a passive is reanalyzed as active it leads to ergative alignment, as passives are typically (though not always) limited to transitive predicates.7 Note that if this is the explanation for language-internal variation, though, we would still need a model with variable alignment parameter settings to model the synchronic status of languages like Basque. Laka (2006a) argues convincingly, however, that (8–9) does not represent an instance of variable alignment but rather an alternation between transitive and intransitive clauses.



CASE. ERGATIVE L ANGUAGES

(8) emakume-a-k ogi-a jaten du woman-DET-ERG bread-DET eating has ‘The woman eats (the) bread’

51

[Basque]

(9) emakume-a [ogi-a jat-e-n] ari da woman-DET bread-DET eat-NMLZ-LOC engaged is Lit. ‘The woman is engaged in eating the bread’ (Laka 2006a: 173) In (8), emakume-a “the woman” is the subject of the transitive predicate jaten “eat” and so it receives ergative case. In (9), however, emakume-a “the woman” is actually the subject of the predicate ari “be engaged in,” which takes a PP object and so functions as an (unaccusative) intransitive predicate. Laka points out that in both cases the object of eat ogi-a receives absolutive case, which would be unexpected if this were really an alternation between ergative and accusative alignment, but which follows if only a contrast in transitivity is at stake. Coon (2012) and Coon and Preminger (2012) argue that similar transitivity-based accounts can be offered for other unrelated languages. The general tendency for ergativity to be coupled with non-progressive/perfective rather than progressive/imperfective aspect in this way, Coon (2012) claims, stems from the fact that progressive aspect is often encoded via locative constructions: (10) a. Nós estamos [a chegar] [European Portuguese] we.NOM are.1PL  P arrive.INF ‘We are arriving’ b. Nós estamos [a comer o bolo] we.NOM are.1PL P eat.INF the cake ‘We’re eating the cake’ In an ergative system, an alternation in transitivity will be reflected by a difference in case/agreement morphology. In an accusative language, however, transitive and intransitive subjects pattern alike in terms of case/agreement and so no such alternation is observed (as in the Portuguese example in 10). Other instances of T/ Asp-sensitive variability have been argued to stem from the special behavior of the internal argument, due either to (i) demotion of the object to PP status (Coon 2012) or (ii) raising of the object to a higher position from which it can trigger agreement (see Patel-Grosz and Grosz 2014).

3.2.2. Sensitivity to the matrix / embedded distinction Next, consider those languages (apparently) displaying variability determined by clause type, whereby ergativity appears to be limited to embedded clauses, as reported for Päri (Nilotic), the Jê languages of Brazil and Tsimshian (Salish) (see Dixon 1994, Coon 2012, Coon and Preminger 2012). In Mẽbengokre, Jê, for example, described by Salanova (2007, 2009), matrix clauses display both ergative and accusative alignments, whereas embedded clauses are always ergative:

52

CONTEMPORARY LINGUISTIC PARAMETERS

(11) ba hadʒu katɛ [Mẽbengokre, Jê] 1.NOM radio break.V ‘I broke the radio’ (Salanova 2009: 8) (12) ba [kutɛ tɛp krẽn] pumũ [Mẽbengokre, Jê] 1.NOM 3.ERG fish eat.N see.V ‘I saw him eating fish’ (Salanova 2009: 6)8 Salanova (2009) argues that the embedded clause in (12) is a nominalization, and that ergative case surfaces as a result. Indeed, Gildea (2008) proposes embedded ergative nominalizations have been reanalyzed as root clauses in a number of unrelated languages of South America, leading to matrix ergative alignment (see also section 3.4.2 below). Again, then, this apparent variability may not actually involve variable parameter settings within a single language, but rather differences in the alignment of nominalizations vs clauses. In an accusative language, an embedded ergative nominalization will give rise to apparently variable alignment, whereas in an ergative language it will not. The result is a pattern whereby an otherwise accusative language displays ergative properties only in an embedded context. Again, the question arises why the opposite trend is not observed. Why do we not seem to find languages which display ergativity only in matrix clauses? This seems to be due to the fact that, for poorly understood reasons, nominalizations tend to be ergative more often than clauses do. Thus even in familiar accusative languages like Greek, French, German, and English, complex event nominalizations display an ergative pattern (see Williams 1987; Alexiadou 2001).

3.2.3. Sensitivity to the features of DP arguments Finally, consider the sensitivity to person, attested in Dyirbal (Pama-Nyungan), Halkomelem (Salish), Mocho’ (Mayan), Kham (Tibeto-Burman), Balochi (NW Iranian), Punjabi (Indo-Aryan) and Cashinawa (Panoan) (see Silverstein 1976, Dixon 1979, 1994, Garrett 1990, Coon and Preminger 2012): (13) Ŋada bayi yaRa balgan I.NOM there.ABS man hit.NFUT ‘I am hitting the man’

[Dyirbal]

(14) Ŋayguna baŋgul yaRa-ŋgu balgan I.ACC there.ERG man-ERG hit.NFUT ‘The man is hitting me’ (Dixon 1972: 60) In such languages, ergative is limited to arguments lower on the Silverstein (1976) scale, a simplified version of which is as follows: (15) First / second person pronouns > third person pronouns > human common nouns > non-human animate common nouns > inanimate nouns



CASE. ERGATIVE L ANGUAGES

53

Once again, there are potential diachronic explanations for this phenomenon. Garrett (1990) offers one such account. He proposes that in null subject languages, where the agent can be omitted, instrumental case can be reanalyzed as ergative, due to the “functional overlap” between instruments and agents. This leads to ergative alignment because instrumentals can function as subjects quite freely in transitive clauses but not in intransitives: (16) John opened the door with the key / The key opened the door (17) John walks with a cane / *A cane walks    (apud Garrett 1990: 265) As instruments are more likely to be towards the right end of the scale in (15), this path of reanalysis will lead to variable alignment whereby only such subjects surface with ergative case. Garrett gives evidence from Hittite and other languages in favor of such a path of grammaticalization. Note that if this kind of explanation holds of (13–14), then Dyirbal would be a language with variable alignment, however that is analyzed synchronically. Once again, though, it is not clear whether examples such as (13–14) are best analyzed as genuine instances of variable alignment. In many languages, it seems, case morphology and / or agreement is used to track semantic or informationstructure-related information, rather than grammatical function. This has been widely studied in relation to objects as differential object marking (DOM) (see Aissen 2002; de Hoop and Malchukov 2007; Dalrymple and Nikolaeva 2011), and somewhat less so in relation to subjects as differential subject marking (DSM) (see de Hoop and de Swart 2008; McGregor and Verstraete 2010). As differential marking occurs in both accusative and non-accusative languages, it is not a specifically “ergative” property. Essentially Dyirbal appears to be a language with both DSM and DOM. Ergative case occurs on 3rd person transitive subjects and accusative is limited to 1st/2nd person objects, as also illustrated by (13–14). Legate (2008) shows, moreover, that DSM in Dyirbal is fairly superficial in nature, not affecting the syntactic properties of transitive subjects, all of which, for example, resist A-bar extraction (a property of syntactically ergative languages, see 3.3.4). As such, at least for some languages, DSM appears to be purely morphological in nature, not affecting the ergative syntax of the language in question. It is not clear whether this is more generally the case for differential marking, however (see Danon 2006 for claim that DOM results from the fact that indefinite NPs do not receive abstract Case). DOM has been explained in functional terms as the requirement for less prototypical transitive subjects to be overtly marked as such (see Haspelmath 2008). This is formalized in Optimality Theoretic terms by Aissen (2003), and refined by Keine and Mueller (2008) to cover alternations where neither exponent is null.9 Aissen (2003) claims explicitly that her account extends to DSM, though Coon and Preminger (2012) take issue with this claim on an empirical basis. Whereas DOM can be sensitive to any of the features on the Silverstein hierarchy (specificity, definiteness, animacy, person), the same is not true, they claim, for DSM, where the dividing line is very generally between 1st and 2nd person pronouns and all other subjects. For this reason, they propose an alternative structural account whereby

54

CONTEMPORARY LINGUISTIC PARAMETERS

1st/2nd person subjects occupy a distinct position in the clause and thus behave like intransitive subjects, assimilating this kind of variability to aspect-sensitive variability. Baker (in progress) provides a different kind of account of differential marking in the context of a dependent case approach to alignment, which we return to in section 3.4.5.

3.3. Ergative splits: Non-accusative alignment types Section 3.2 considered variable alignment in some detail and proposed that it might be illusory. In this section, we turn to split ergativity, whereby a language displays only partially ergative behavior. As will become clear, the various kinds of ergative splits represent genuine instances of parametric variation and very clearly indicate that there is more than one ergativity parameter, something which we will return to in section 3.4, when we consider generative analyses of the phenomenon.

3.3.1. Morphological ergativity One type of “split” involves a morphologically ergative system which is syntactically accusative, as diagnosed by patterns of raising, control, co-ordination and (in some cases) anaphor binding (cf. Anderson 1976). Thus in Tongan, for example, both transitive and intransitive “subjects” can undergo raising, regardless of whether they are marked with absolutive or ergative case: (18) a. ‘oku lava ‘a mele ‘o [hū ki hono fale] PRES possible ABS Mary TNS enter to his house ‘Mary can enter his house’ b. ‘oku lava ‘e siale ‘o [taa’i ‘a e fefine] PRES possible ERG Charlie TNS hit ABS DEF woman ‘Charlie can hit the woman’ [Tongan] (Anderson 1976: 13) As Anderson notes, if the syntactic notion of subject, diagnosed in this way, picks out the same arguments as in accusative systems, then ergative alignment must be a fairly superficial property of the language. In terms of Agree, we might infer that both transitive and intransitive subjects agree with T/INFL in such a language even though surface morphology does not reflect that fact. In fact, it is a matter of some debate whether all ergative systems fit into this category or, in other words, whether syntactic ergativity exists at all, a point to which we return in sections 3.3.4–3.3.5.10



CASE. ERGATIVE L ANGUAGES

55

3.3.2. Split intransitivity A second widely discussed class of non-accusative languages can be described as “split-S” or “active-stative” because intransitive subjects pattern either with transitive objects or transitive subjects, depending on volition / agency / control (Sapir 1917; Harris 1982; Mithun 1991) or other semantic / pragmatic factors (Deal and O’Connor 2010). Consider, for example, the following Chol examples from Coon (2010), which show that the intransitive subject of the unergative verb “sang” is marked via an ergative verbal marker (set A), whereas the intransitive subject of the unaccusative verb “arrive” is marked as absolutive (set B): (19) a. Tyi jul-i-yety PRFV arrive.here-ITV-B2 ‘You arrived here’ b. Tyi a-cha`l-e k’ay PRFV A2-do-DTV song ‘You sang’

[Chol, Mayan]

(Coon 2010: 18)

This kind of split is often called split-S because it concerns only the variable behavior of the intransitive subject (S in Dixon’s 1994 terminology). Other names are active, active-neutral, active-inactive, active-static, or stative-active; agentive or agent-patient; or split intransitive. In some languages, this split appears to track the unergative / unaccusative divide familiar from accusative languages. (Western) Basque, for example, has been argued to pattern along these lines (Laka 2006b): (20) a. Txalupa hondora-tu da boat.DEF.ABS sink-PERF is ‘The boat sank’ b. Klara-k ondo eskia-tzen du Klara-ERG well ski.IMPF has ‘Klara skis well’

[Basque] (Laka 2006b: 376) (Laka 2006b: 379)

Other languages that can arguably be classified as split-S in these terms include: Georgian, Kartvelian (Harris 1982); Hindi-Urdu, Indo-Aryan (Butt and King 2002); Chol, Mayan (Coon 2010); Lakhota, Siouan; and Caddo, Caddoan (Mithun 1991). Mithun (1991) notes that there is cross-linguistic variation regarding the behavior of predicates like “cough,” “hiccup,” “vomit,” and “yawn,” which are performed by an agent but are not controlled. In some languages their subjects pattern with agents (Lakhota) and in others they do not (Central Pomo). It is not clear whether this difference indicates that the languages in question have different kinds of split-S systems or whether they simply make the unaccusative / unergative divide in a different way. Dixon (1994) uses the label “fluid-S” to refer to languages which display the same kind of alternation with some degree of freedom regarding which predicates pattern in which way. Consider, for example, verbs like “fall” in Batsbi (Tsova‐Tush), Nakh,

56

CONTEMPORARY LINGUISTIC PARAMETERS

which can surface with either ergative or absolutive subjects, with a concomitant change in meaning: (21) a. S.a: as vuizhn‐as [Batsbi] 1SG.ERG M.fall‐AOR.1SG.ERG ‘I fell (on purpose)’ b. S.o: so vozhen‐sŏ 1SG.ABS M.fall‐AOR.1SG.ABS ‘I fell (accidentally)’ (Nichols 2008a) A similar effect is observed with the verb pʰitik “belch” in Northern Pomo which can be used with either a nominative-marked or accusative-marked subject, with the expected difference in meaning:11 (22) a. ʔa: pʰitik-čade [Northern Pomo] 1SG.NOM belch-PROSP ‘I’m going to belch’ [I’m going to belch, so there!] b. ṱo: pʰitik-čade 1SG.ACC belch-PROSP ‘I’m going to belch’ [Uh oh, I feel a belch coming on]   (Deal and O’Connor 2010: 176) The existence of so-called “fluid-S” systems is potentially equivalent to the fact that some verbs show variable unaccusative / unergative behavior in languages like Italian (Sorace 2000). Given that many split-S systems seem to permit some degree of fluidity it is not clear to what extent fluid-S should be considered a separate alignment class. Other kinds of split-S system are also attested. The split-S system in Guaraní is famously sensitive to lexical aspect (Aktionsart) rather than agentivity or control (Mithun 1991). The subjects of intransitive eventive verbs (e.g. go, get up) have the same morphological realization as transitive subjects (pronominal prefixes), whereas the subjects of stative predicates (e.g. be sick, be sleepy) share a morphological form with transitive objects (Mithun 1991): (23) a. a-xá ‘I go’ b. še-ropehɨí ‘I am sleepy’ c. še-rerahá ‘It will carry me off’ d. ha upépe a-gařá šupé ‘…and there I caught him’

[Guaraní]

(Mithun 1991: 511)

This kind of split seems more difficult to assimilate to the unaccusative / unergative distinction, and probably requires a further parametric option to be available.



CASE. ERGATIVE L ANGUAGES

57

One question which arises in relation to these data is why split-S systems are grouped with ergative rather than accusative systems. They can be defined as systems in which one subset of S patterns with A and the other with O. The answer is that they usually involve overt marking of transitive subjects (and some intransitive subjects), something which is not typical of accusative systems.12 As we shall see in section 3.4.7, split-S systems have been seen as potential support for the contention that ergative is an inherent (thematic) case.

3.3.3. Case / agreement splits A third kind of split describes a mismatch between case and agreement morphology, whereby a language can have ergative case and accusative agreement but not vice versa (Anderson 1977; Moravcsik 1978; Legate 2002, 2008; Corbett 2006; Woolford 2006b; Bobaljik 2008). An example of the licit case / agreement mismatch comes from Nepali, as described by Bickel and Yādava (2000): (24) a. ma yas pasal-mā patrikā kin-ch-u.  [Nepali] 1SG.NOM DEM.OBL store-LOC newspaper.NOM buy-NPST-1SG ‘I buy the newspaper in this store’ b. maile yas pasal-mā patrikā kin-ē /*kin-yo 1SG.ERG DEM.OBL store-loc newspaper.nom buy-pst.1sg/buy.pst.3msg ‘I bought the newspaper in this store’ (Bickel and Yā dava 2000: 348) Warlpiri displays essentially the same mismatch (Hale 1970), as does Georgian in the aorist (Marantz 1991). Such patterns are potentially problematic for Chomsky’s (1995: Chapter 4) claim that case and agreement are surface realizations of the same abstract Agree relation because, in these languages, agreement does not track case. Interestingly, though, the inverse pattern, whereby a nominative case system co-occurs with ergative agreement alignment appears to be unattested (though see Patel 2006 for one potential example). It remains unclear how best to account for this kind of split ergativity. Marantz (1991, 2000) takes data such as (24) as evidence that case is a purely morphological phenomenon (see 3.4.5), whereas Woolford (2006b) offers an account consistent with the Chomsky (1995) approach to Agree. She claims that there is no such thing as ergative agreement, with all apparent instances of such being pronominal clitics. Agreement on T/INFL must track nominative case, to license structural Case and only if there is no structural nominative to license can T/INFL agree with a DP marked with inherent (thematic) case. See Bobaljik (2008) for a different (morphological) account.

3.3.4. Syntactic ergativity In the recent generative literature, there has been much discussion of the possibility that all linguistic variation might reduce to differences in externalization (see Berwick

58

CONTEMPORARY LINGUISTIC PARAMETERS

and Chomsky 2008). For this reason, the question of syntactic ergativity is of central importance to the focus of this chapter. Dixon’s (1979, 1994) description of Dyirbal gave rise to the term syntactic ergativity as a cover term for the following phenomena: a) Ergative-absolutive topic chaining b) Ergative-absolutive Control pattern c) Ban on A-bar extraction of transitive subjects While the data in evidence of (a-b) have become controversial (see Legate 2012 and section 3.3.5), (c) has proven more robust cross-linguistically. Consider the following example from West Greenlandic, which illustrates that it is impossible to form a relative clause via A-bar extraction of an ergative-marked transitive subject: (25) *angutii [Opi  ti aallaat man.ABS gun.ABS ‘the man who took the gun’

tigu-sima-sa-a] [West Greenlandic] take-PERF-REL.TR-3S (Manning 1996a: 84)

This property has been reported in many Inuit, Austronesian (Manning 1996b) and Mayan languages (Coon, Mateo Pedro, and Preminger 2014), as well as in Dyirbal (Dixon 1994), Halkomelem Salish (Gerdts 1988), Chamorro (Campana 1992), Chukchi (Comrie 1979), Tongan (Otsuka 2006) and Katukina, an Amazonian language (Queixalós 2012). Languages with this restriction use either (a) an Antipassive or Agent Focus construction to demote the agent and render it extractable (see Coon et al. 2014), or (b) as is the case in Tongan, a resumptive subject pronoun to avoid the need for extraction (Otsuka 2006): (26) a. e fefine [na’e tangi] [Tongan] DEF woman PST cry ‘the woman (who) cried’ b. e fefine [na’e fili ‘e Sione] DEF woman PST choose ERG Sione ‘the woman (who) Sione chose’ c. *e fefine [na’e fili ‘a Sione] DEF woman PST choose ABS Sione Intended ‘the woman (who) chose Sione’ d. e fefine [na’e ne fili ‘a Sione] DEF woman PST 3SG choose ABS Sione ‘the woman (who) chose Sione’ (Otsuka 2006: 81) While this restriction has obvious parallels with the that-trace effect observed in some accusative languages (French, English, Nupe), a crucial difference is that it affects only the transitive subject. That-trace effects arise with both intransitive and transitive subjects alike:13 (27) Qui est-ce que tu crois who is-this that you believe ‘Who do you think will leave?’

qui/*que partira [French] who/that leave.FUT.3S



CASE. ERGATIVE L ANGUAGES

59

If, (i) following Hinzen and Sheehan (2013: Chapter 5) we take the definition of narrow syntactic variation to be variation which has a systematic effect on narrow syntactic operations and (ii) we accept that A-bar extraction is narrow syntactic, then it follows that the parameter giving rise to syntactic ergativity must itself be syntactic in nature so that not all variation can be reduced to differences in externalization. It remains unclear how best to account for this property of some ergative languages, but we review some options in section 3.4.

3.3.5. High / mixed absolutives A less-discussed split concerns the source of absolutive Case in ergative systems. Whereas in languages such as Dyirbal, and some Mayan languages, it seems that T/ INFL always assigns absolutive, in other languages the absolutive associated with O seems to come from v (Aldridge 2004, 2007; Legate 2008, 2012; Coon et al. 2014). The evidence for this difference comes from contexts where the structural Case of T is plausibly suppressed (e.g. non-finite clauses). In some ergative languages both objects and intransitive subjects fail to get absolutive case in non-finite contexts. This is potentially the account of the ergative-absolutive (S/O) control pattern observed in Dyirbal (see also Levin 1983):14 (28) [yabu ŋuma-ŋgu giga-n [PRO banaga-ygu]] [Dyirbal] mother.ABS father-ERG tell-NONFUT return-PURP ‘Father told mother to return’ Lit. ‘Father told motheri PROi to return’ (29) [yabu ŋuma-ŋgu giga-n [PRO gubi-ŋgu mawa-li]] mother.ABS father-ERG tell-NONFUT doctor-ERG examine-PURP ‘Father told mother to be examined by the doctor’ Lit. ‘Father told motheri for the doctor to examine PROi’ (30) [ŋuma banaga-Nu [PRO yabu-ŋgu bura-li] father.ABS return-NONFUT mother-ERG see-PURP ‘Father returned in order for mother to see him or father returned and as a result mother saw him’ Lit. ‘Fatheri returned for mother to see PROi’ (Dixon 1994: 168–9) In such embedded contexts, ergative case is licensed but absolutive is not. If control is derived via movement in Dyirbal in the manner proposed by Hornstein (1999) then it follows that transitive objects like intransitive subjects will be required to raise to a higher clause to get Case where T fails to assign absolutive. Aldridge (2008b) reports that in some other ergative languages (e.g. Seediq, citing Aldridge 2004 and Jacaltec Mayan, citing Craig 1977), it is impossible to control into transitive clauses. This follows if these are languages in which control is not derived via movement where PRO, which is necessarily the highest argument in the clause has Case (see Sheehan 2014a for evidence that we need both kinds of control). PRO can receive absolutive in intransitive clauses in such languages and ergative in transitive clauses, but there will be no Case available for the object DP in the latter.

60

CONTEMPORARY LINGUISTIC PARAMETERS

Coon et al. (2014) building on work by Campana (1992), show that “high absolutive” Mayan languages lose the ability to assign absolutive in aspectless embedded clauses, which may be a variant of the same effect (though these languages seem to require nominalization of embedded clauses). In mixed absolutive languages, however, absolutive on S is lost in a non-finite clause, whereas absolutive on O is retained. This gives rise to an accusative control pattern, as illustrated by the following data from West Greenlandic: (31) Miiqqat [PRO Juuna ikiu-ssa-llu-gu] niriursui-pp-u-t children.abs [(erg) Juuna.abs help-fut-inf-3s] promise-ind-intr-3p ‘The children promised to help Juuna’  [W. Greenlandic] (Manning 1996a: 124) While it seems to be the case that all high absolutive languages are also syntactically ergative in that they ban A-bar extraction of the ergative DP, the reverse is not true.15 West Greenlandic is a case in point as it is a low absolutive syntactically ergative language (see (25) and (31)). In section 3.4.8 we see that there are apparently more one-way implications of this kind in the domain of alignment. How best to model them in parametric terms remains an open question but see 3.4.8 for one suggestion.

3.3.6. Summary As we have seen in the preceding sections, there is a great deal of variation in the domain of clausal alignment, even without taking into consideration the lexical cases associated with internal arguments and the behavior of ditransitives and analytic causatives. Note that a given language can display multiple different alignment splits as well as variability. Take for example (Western) Basque, which displays: variability sensitive to aspect (Laka 2006a); split-S alignment (Laka 2006b); plus a mismatch between morphology and syntax, as its syntactic relations seem to be organized along accusative lines, despite the presence of ergative case (Anderson 1976). These splits raise obvious challenges for a parametric account of alignment. Most of all they make it very clear that we are dealing with various micro-parameters rather than a single ergativity parameter, even if these micro-parametric settings are not variable within a given language as discussed in section 3.2.

3.4. Generative analyses of ergativity Many different accounts have been given of (the various aspects of) ergativity in the generative literature, some of which can be considered parametric in nature (see Johns, Massam, and Ndayiragije 2006, Aldridge 2008b). While early accounts posited a thematic parameter, more recent approaches focus on a parametrization of agreement and/or movement operations. In this section we consider the following different kinds of approaches and outline some of the challenges they face:16 (i) (ii)

Thematic ergativity (Levin 1983; Marantz 1984; Dowty 1991) Ergative clauses as nominals (Johns 1992)



CASE. ERGATIVE L ANGUAGES

61

(iii) Parameterized Agree (Bobaljik 1993; Müller 2009; Assmann et al. 2012) (iv) ‘Smuggling’ approaches (Koopman 2012; Roberts 2010c) (v) Ergative as a dependent case (Marantz 1991; Baker in progress, to appear; Stiebels 2000) (vi)  Ergative as an inherent case (Woolford 1997, 2006a; Aldridge 2004, 2008a; Legate 2008, 2012; Sheehan forthcoming) (vii) Phase-based accounts of syntactic ergativity (Aldridge 2004, 2007, 2008a) (viii) A parameter hierarchy approach (Sheehan 2014b) In discussing these approaches we shall consider their generality as well as their ability to deal with variation of the kind described in section 3.3. It may ultimately turn out to be the case, though, that (surface) ergativity is determined by wholly distinct parameters in different languages and that this is the explanation for the various kinds of ergative systems described in section 3.3.

3.4.1. Thematic ergativity Early accounts attributed ergativity to a thematic parameter (at least for ‘deep’ or syntactically ergative languages) (Levin 1983; Marantz 1984; Dowty 1991). Rather than a principle, the Universal Theta-Assignment Hypothesis in (32) was treated as a parameter, so that the deep structures of accusative/ergative languages differ as in (33a–b) (with the notation updated for ease of exposition). (32) Universal Theta-Assignment Hypothesis (UTAH) (Baker 1988: 46) Identical thematic relationships between items are represented by identical structural relationships between those items at the level of D-structure. (33) (a) Syntactic accusativity Agent roles—assigned by predicates Theme/patient roles—assigned by verbs

(b) Syntactic ergativity Agent roles—assigned by verbs Theme/patient roles—assigned by predicates

vP vP Agent

v’

v

Theme VP

V

Theme

v’ v V

VP Agent

This thematic approach to ergativity has now largely been abandoned because there is evidence that even the “deep” ergative languages pattern like accusative languages with regard to binding, incorporation and idiom formation, suggesting a universal thematic structure common to all languages (Baker 1988, 1997; Manning 1996a, 1996b). Thus Dixon (1994: 138) claims that “[i]n every ergative language, as in every accusative language, the ‘antecedent’, i.e. the controller of reflexivity is A (or S,

62

CONTEMPORARY LINGUISTIC PARAMETERS

where it is extended to intransitives).” This is true of morphologically ergative languages like Basque: (34) Gudari-ek elkar hiltzen zuten soldiers-ERG REC.ABS kill AUX ‘The soldiers killed each other’

[Basque]

(35) *Gudari-ak elkarr-ek hiltzen zituen/zituzten soldiers-ABS REC-ERG kill AUX (36) Lagun-ak elkarr-ekin joan dira friend-ABS REC-with go AUX ‘The friends have gone with each other’

(Manning 1995: 2)

It is also true of syntactically ergative languages like Toba Batak (Austronesian), which display the ban on A-bar extraction of ergative DPs. Like many Austronesian languages, Toba Batak has a complex voice system. Following Aldridge’s (2006) account of Tagalog, I take the active voice to be the basic (ergative) alignment. In active voice, O must occur adjacent to the verb, receiving absolutive case. In objective voice, however, which I take to be an antipassive, A must occur adjacent to the verb and receive absolutive case: (37) [Mang-ida si Ria] si Torus AV-see PM Ria PM Torus ‘Torus sees/saw Ria’ (38) [Di-ida si Torus] OV-see PM Torus ‘Torus sees/saw Ria’

si Ria PM Ria

In both cases, A can bind an anaphor in O, and not vice versa, regardless of voice, as Manning (1996b) points out: (39) [Mang-ida diri-na] si John AV-saw self-his PM John ‘John saw himself’

[Toba Batak]

(40) *[Mang-ida si John] diri-na AV-saw PM John self-his ‘Himself saw John’ (41) *[Di-ida diri-na] si John OV-saw self-his PM John ‘Himself saw John’ (42) [Di-ida si John] diri-na OV-saw PM John self-his ‘John saw himself’

(Manning 1996b: 5)



CASE. ERGATIVE L ANGUAGES

63

For this reason, then, it is now generally accepted that, even in syntactically ergative languages, thematic structure is not parameterized (see Baker 1997 for further discussion).

3.4.2. Ergative clauses as nominals Another influential approach to ergativity claims that clauses are actually formally nominalizations in (some) ergative languages (see Johns 1992). Johns’ core proposal is that verbs cannot project to VP in Inuktitut so that “there can be no objects at D-structure” (p. 60). For this reason, all Inuktitut transitive verbs surface with passive morphology which is nominal and so clauses are actually complex nominalizations with two “subjects.” The main evidence for this proposal comes from the morphological make-up of the Inuktitut clause: (43) arna-up angut kuni-ga-a [Inuktitut] woman-REL man(ABS) kiss-PASS.PART-3S/3S ‘The woman kissed the man’ (Johns 1992: 59) Note that in addition to the passive morphology on V, the “ergative” subject actually bears relative (genitive) case, the same case which appears on possessors in NPs: (44) Jaani-up nasa-a John-REL hat-3S ‘John’s hat’ Note also that the agreement morphology in (44) is identical to that in (43) and this is more generally the case. This means that (43) literally has the structure: ‘The man is the woman’s kissed one’. Interestingly, Johns (1992) notes explicitly that this is not really a parametric account but rather an approach whereby ergativity arises from “the interaction between language-particular lexical features and universal principles” (p. 58). Like the analysis in 3.4.1, it implies, nonetheless, that the thematic structure of ergative languages differs dramatically from that of their accusative counterparts, and as such it faces many of the same challenges. It is not clear to what extent this analysis is intended to generalize to other languages, but see Salanova 2009 and Carrió and Salanova 2009 for a similar account of Mẽbengokre (Jê, Brazil) and Mocoví (Guaykurú, Argentina) as well as the discussion in section 3.2.2.

3.4.3. Smuggling approaches Another traditional account of ergativity assimilates it more directly to the passive (see Hale 1970; Estival and Myhill 1988 for discussion), and, unlike the approaches in 3.4.1 and 3.4.2, takes ergativity to be a more superficial “derived” property pertaining to case / agreement only, rather than base-generated structure. Smuggling approaches to ergativity can be considered a modern continuation of this trend. Here we outline the approach in Roberts (2010c), but see also Koopman (2013)

64

CONTEMPORARY LINGUISTIC PARAMETERS

who develops an account of Samoan involving two instances of smuggling or “double passivization.” Roberts (2010c) attempts to address the problem posed by defective intervention of the transitive subject A where T/INFL agrees with the transitive object O.17 (45) T[uϕ: ]

TP vP

DPERG v’ [ϕ] v VP

V

DP [ϕ, uCase]

The problem with (45) is that in many contexts, a DP with inherent case, which presumably lacks an uninterpretable [uCase] feature can nevertheless block an agreement relation between a probe and an active goal. This is true, for example, with dative experiencer DPs of raising verbs in many languages: (46) Gianni sembra (??a Piero) fare il suo dovere [Italian] Gianni seems.3SG to Piero do.INF the his duty Lit. ‘Gianni doesn’t seem to himself to do his duty’ (adapted from McGinnis 1998: 92) (47) Jean semble (??à Marie) avoir du talent [French] Jean seems.3SG to Marie have.INF some talent ‘Jean seems to Mary to have talent’ (adapted from McGinnis 1998: 90) Such examples suggest the existence of something like (48) in natural language: (48) The Defective Intervention Constraint (based on Chomsky 2000: 123) a>b >g (*AGREE (a,g), where a is a probe and b is a matching goal, and b is inactive) Following Collins’ (2005) proposal for the passive, Roberts claims that O is smuggled past A in ergative languages via VP movement to spec VoiceP. This serves to render O available for probing by T, avoiding defective intervention by A: (49) T [VoiceP [VP O V ] Voice [vP A v [VP] ]] The parameter separating ergative from accusative languages, in these terms, is whether derivations are “direct,” involving straightforward argument movement, or “indirect,” involving smuggling of an argument. The approach has the apparent advantage of deriving the lack of SVO ergative languages (Mahajan’s 1994 generalization) from the impossibility of raising a head-initial constituent in this way (ruled out by the independently justified Final-over-Final Constraint).



CASE. ERGATIVE L ANGUAGES

65

There are some potential technical and empirical issues facing such an approach, however. First, after smuggling, O and A are actually equidistant from T, so some other mechanism is needed in order to make sure T agrees with O and A gets ergative case. Indeed, it might be expected that T could agree with both A and O in such a configuration (see Baker’s 1988 proposed analysis of symmetrical double object constructions). On the empirical side, the analysis seems to account only for high absolutive languages. As discussed in section 3.3.5 it seems that in many ergative languages, in transitive clauses absolutive is actually assigned by v.

3.4.4. Parameterized Agree Other parametric approaches to alignment eschew the parallel with the passive and posit a more direct parameterization of Agree relations, without the need for demotion of the subject or smuggling of the object.18 Bobaljik (1993) proposes the Obligatory Case Parameter (OCP) which constitutes the choice between subject- and object-type agreement in intransitive clauses. In a model where Agree relations are encoded as functional heads, following Chomsky (1995: Chapter 3), in accusative languages AgrS is obligatory, whereas in ergative systems AgrO is obligatory: (50) accusative transitive: [AgrSP Agent AgrS [AgrOP Theme AgrO [vP …]]] accusative intransitive: [AgrSP Theme/Agent AgrS [vP …]] (51) ergative transitive: ergative intransitive:

[AgrSP Agent

AgrS [AgrOP Theme AgrO [vP …]]] [AgrOP Theme/Agent AgrO [vP …]]

The result is that in accusative languages the subject of an intransitive behaves like the subject of a transitive predicate, receiving Case from AgrS.19 In ergative languages, on the other hand, the subjects of intransitive predicates behave like the objects of transitives as both receive Case from AgrO. The implication is that ergative is assigned by AgrS, and so is equivalent to nominative, whereas absolutive is assigned by AgrO and is equivalent to accusative. The apparent prediction is that only ergatives will be subject to raising and that absolutive will always be retained in non-finite contexts, where AgrS loses its ability to assign a structural Case. Manning (1996a) raises substantial empirical challenges for these predictions (see also section 3.3.5). The account proposed by Müller (2009) and developed by Assmann et al. 2012 involves a parameterization of the order of Merge and Agree. The crucial idea in this approach is that, in an ergative system, Merge applies prior to Agree. This leads to the ergative / accusative parameter once certain other assumptions are made. First, ergative and accusative must both be Cases assigned by little v via Agree. Secondly, Agree must take place under m-command. Finally, and (most controversially), specifiers must be closer to heads than material in the complement domain. If all these things hold then where Merge precedes Agree, v will Agree with the external argument, resulting in ergative Case being assigned. Where Agree precedes Merge, however, v will Agree with the internal argument, resulting in accusative Case. What is parameterized in these terms, then, is the relative ranking of operations:

66

CONTEMPORARY LINGUISTIC PARAMETERS

(52) ergative alignment:

Merge > Agree

(53) accusative alignment: Agree > Merge An added attraction of Assmann et al.’s approach is that it also provides an account of syntactic ergativity (see section 3.3.4), based on the following three (again fairly controversial) assumptions: (i) there is no intervention so that all DPs in a given complement domain are potential targets for a c-commanding probe; (ii) TP is a phase; and (iii) an ergative-marked DP can nonetheless agree with T. If these assumptions hold then the ergative DP will need to raise through spec TP in order to be A-bar extracted from the TP phase. Upon moving to T, however, it will necessarily “maraud” (i.e. steal) T’s case feature in an ergative system as Merge precedes Agree and specifiers are closer than complements. This leaves the internal argument without Case, leading to a crash in the derivation. The same parameterization which leads to ergative alignment thus also explains syntactic ergativity for free. Despite the attraction of this approach, it faces a number of potential challenges. As Assmann et al. note, without further stipulations, it predicts all ergative languages to be syntactically ergative (banning A-bar extraction of the transitive subject), contrary to fact (see section 3.3). Altering the model to allow for languages such as Basque and Chol requires considerable complication of the elegant model (see their section 5.2). The approach also raises questions concerning the nature of parameters. While the proposal is pleasingly minimal in a certain respect, it involves the parameterization of the ranking of narrow syntactic operations, something which represents a drastic move away from the Borer–Chomsky conjecture that parameters stem from variation in the features of functional heads. To what extent variation can or should be modeled in this way is a matter of ongoing research.

3.4.5. Ergative as a dependent case A very different approach to ergativity was first put forth by Marantz (1991, 2000) and taken up by Stiebels (2000) and Baker (2014, 2015) (see also the proposals in Anderson 1976 and Bittner and Hale 1996). Marantz starts by noting that Burzio’s generalization concerning the distribution of accusative case finds its mirror image in ergative systems: (54) Burzio’s generalization: no accusative case on an object in a sentence with a non-thematic subject position Ergative generalization: no ergative case on a non-thematic subject (i.e. on an argument moved into a non-thematic subject position)  (Marantz 2000: 14) The explanation for this, he claims, is that both accusative and ergative are dependent cases which are assigned in the morphological component in the presence of another DP without lexical case. The parameter determining ergativity is the following:



CASE. ERGATIVE L ANGUAGES

67

(55) Dependent case is assigned by V+I to a position governed by V+I when a distinct position governed by V+I is: a. not “marked” (not part of a chain governed by a lexical case determiner) b. distinct from the chain being assigned dependent case Dependent case assigned up to subject: ergative Dependent case assigned down to subject: accusative20  (Marantz 1991: 25) Abstracting away from the role of government, the effect of (55) is that, where two arguments without inherent cases surface in a c-command relation within the same clause, one of them will be assigned dependent case. By placing case assignment (and agreement) in the post-syntactic morphological component in this way, Marantz denies any syntactic difference between ergative and accusative systems. All that is at stake is a rather superficial fact of morphology. The attraction of such an approach lies in its simplicity and the parallels between accusative and ergative which Marantz describes. As he further notes, a system which places case / agreement in the morphological component is better placed to handle case / agreement mismatches of the kind described in section 3.3.3. Baker (2014, 2015) shows that a variant of this approach can account for a range of facts, including the interaction with differential marking of the kind described in section 3.2.3. A potential challenge for the approach comes from the existence of syntactically ergative and high absolutive languages, which suggests that the syntax of (some) ergative systems differs from that of their accusative counterparts.

3.4.6. Ergative as an inherent Case An approach which has also gained popularity in the recent generative literature is the idea that ergative is an inherent case (see Mahajan 1989; Massam 1994; Woolford 1997, 2006a; Aldridge 2004, 2008a; Legate 2008, 2012). As Woolford (2006a) notes, ergative is often classified as a structural case based on mistaken diagnostics, notably because it is a pervasive feature of transitive clauses. As she notes, independently of ergativity, UG must allow for theta-related cases which are assigned to specific structural positions, because of the existence of dative case which is assigned to goals / beneficiaries / experiencers in many languages. Ergative, she claims, is just the thetarelated inherent case associated with external arguments. Evidence for this position comes from the fact that (i) there is no raising to ergative (Marantz 1991), but (ii) ergative subjects can undergo raising in some languages (Woolford 1997). If ergative is a thematic case associated with little v then the naïve prediction is that all specifiers of little v will be ergative. While this apparently works well for split-S languages such as Basque (see Laka 2006b and section 3.3.2), it is not so obviously true in ergative languages like Dyirbal, as Bok-Bennema (1991) notes. In order to account for such languages a distinction must be made between transitive and intransitive little v, where only transitive v assigns ergative case. While the inherent case approach has been claimed to work for several different ergative languages, Rezac, Albizu, and Etxepare (2014) raise some problems in relation to Basque (see also Preminger 2012). Further challenges arise from the fact that non-Agents receive ergative in a number of different ergative languages

68

CONTEMPORARY LINGUISTIC PARAMETERS

languages. Thus causers in Lezgian (Haspelmath 1993) are ergative as are some experiencers in Hindi (Davison 2014). Such cases are problematic if causers and experiencers are never introduced by little v, something which can be established only via independent (and probably language-specific) diagnostics. Only the careful study of the distribution of ergative case in conjunction with a coherent and motivated theory of thematic structure can really test the validity of the inherent case approach (see Legate 2012 for one such study of Warlpiri see also Sheehan to appear).

3.4.7. Phase-based account of syntactic ergativity As it stands, the inherent case approach to ergativity provides no explanation for syntactic ergativity (see section 3.3.4) or the difference between high and low absolutives (see section 3.3.5). More generally, under any approach to ergativity, it is clear that more than one single parameter is at stake given the kinds of splits described in section 3.3. Aldridge (2004, 2007, 2008a) proposes that a distinct parameter determines syntactic ergativity. An additional EPP feature is present on transitive v in syntactically ergative languages forcing O to (covertly) scramble past A. This movement, she claims, has the effect of (i) triggering a presuppositional reading of O, as it escapes existential closure (see Diesing 1992a) and (ii) trapping the transitive subject inside the vP phase, based on a strong sense of phase impenetrability. Coon, Mateo and Preminger (2014) propose a similar account of syntactic ergativity based on the consideration of a number of Mayan languages. A crucial difference between their approach and that of Aldridge is that they claim syntactic ergativity is a property only of high absolutive languages. In their terms, A gets trapped inside the vP phase precisely because the object must raise to the phase edge to receive absolutive case. Coon et al. state clearly that their analysis is intended to apply only to Mayan. A potential problem with making this proposal more general is that, outside of the Mayan family, there appear to be low absolutive languages which nonetheless display syntactic ergativity (see Aldridge 2007, 2008a).

3.4.8. A parameter hierarchy for alignment Thus far we have considered a number of different approaches to ergativity which attribute it to parameters determining variously: (a) thematic / base-generated structure; (b) the application of passive transformations; (c) Agree relations; (d) morphological case assignment; or (e) the distribution of inherent case. While all of these approaches have their various merits, a major challenge they face comes from the various kinds of split ergativity described in 3.3. At the very least, further microparameters (of the kind in 3.4.7) are needed in order to model the possibility of: (i) split intransitivity; (ii) morphological vs syntactic ergativity; (iii) high / low absolutives; and (iv) case / agreement mismatches. In fact, matters are made more complex by the fact that a number of one-way implications have emerged from generative and typological work in this domain (see also Deal 2013 for an extensive summary): i.

Ergative with unergatives > ergative with transitives



CASE. ERGATIVE L ANGUAGES

69

ii. Syntactically ergative > morphologically ergative (Dixon 1994: 172) iii.  Ergativity in Control > Ergativity in A’-movement > Ergativity in case/ agreement (Deal 2013: 12) iv. Split-S alignment > no restriction on extraction of DPERG (Deal 2013: 12) v. Ergative agreement > ergative case or no case (Anderson 1977; Moravcsik 1978; Corbett 2006; Woolford 2006b) vi. Ergative case > overtly marked ergative case (Deal 2013: 12) vii. Ergative > not SVO (Trask 1979; Mahajan 1994) The status of such implications in generative grammar remains unclear (see Haspelmath 2008). One possibility, though, is that they could be modeled via dependent parameters of the kind discussed by Roberts and Holmberg (2010). Sheehan (2014b) proposes a parameter hierarchy to account for some of (i-vii), based on the approach to ergativity described in sections 3.4.6–3.4.7. The proposed parameter hierarchy looks as follows: a. Basic alignment parameter: Does transitive ‘v’ assign theta-related case (ERG) to its specifier in L?



N    Y

Accusative b. Split-S parameter: Do all ‘v’s in L assign this theta-related (Russian…) case (ERG)?



Y    N

Morphologically split-S (Chol, Basque, Hindi)



c. Active-stative parameter: Do all active ‘v’s in L assign this theta-related case (ERG)?

Y    N

Morphologically active-stative d. Syntactic ergativity parameter: (Guaraní) Does vERG bear an EPP feature in L?

N    Y

Morphologically e. High/low ABS parameter: ergative (Warlpiri) Does vERG lack structural Case (Φ features) in L?



N    Y

Syntactically ergative Syntactically ergative low ABS high ABS (West Greenlandic, Tagalog) (Dyirbal, Q’anjob’al) FIGURE 3.2  A parameter hierarchy for clausal alignment

70

CONTEMPORARY LINGUISTIC PARAMETERS

Figure 3.2 takes the micro-parameters defining (split) ergativity and places them in transitive dependencies deriving the implications in (i-iv). The hierarchy itself, rather than being prespecified by Universal Grammar, is claimed to be emergent (see Roberts 2012). This approach to parametric variation carves a path between micro- and macroparametric models and potentially renders the micro-parametric approach more explanatory. The macro-parameter determining alignment (parameter a) sits at the top of the hierarchy and separates accusative from non-accusative systems. Further dependent parameters give rise to progressively more ergative systems which are more complex and are thus predicted to be rarer, all else being equal. Parameters (b–c) determine the distribution of ergative in a given language across some coherent set of little vs. Strictly speaking these parameters stand in negative dependencies and so are not actually hierarchically ordered: reordering (b–c) would not alter the outputs of the system in any way (see Sheehan, to appear). The same is not true where positive dependencies are involved, as in (a), (d) and (e). These parameters are hierarchically ordered: altering their relative ranking in Fig 3.2 would give rise to different outputs which would in turn fail to capture the implicational universals in (i-iv). The parameters in Fig 3.2 are fairly minimal, involving only the feature make-up of a single class of functional heads (little v). The crucial question is where the dependencies themselves come from. Sheehan (2014b, to appear) proposes that they stem from (i) acquisition strategies (drawing on Roberts 2012) and (ii) the need for convergence. A remaining challenge for this approach is how, if at all, it can account for the remaining implications (v–vii) as well as challenges to the idea that ergative is an inherent case (see Rezac et al. 2014).

3.5. Conclusions and future directions This chapter has considered the nature of various kinds of ergativity in the context of parametric theory. While it is not clear at present whether we need to posit variable parameter settings in a given language, it is clear that far from there being a simple parameter which separates ergative from accusative languages, what we actually have is a series of micro-parameters determining a number of non-accusative alignments. In section 3.4 we have outlined a number of generative approaches to ergativity, most of which can be formulated as parameters, and discussed some of their merits and shortcomings. In order to account for apparent universal implications in this domain, we have also considered the possibility that these alignment parameters stand in dependency relations. What remains unclear at present, given the extent of attested variation, is whether it can be maintained that ergative case has the same (inherent / dependent / preposional / genitive / structural) status in all languages or whether this too varies, complicating the picture even further.



CASE. ERGATIVE L ANGUAGES

71

Notes *

Many thanks to the European Research Council (ERC) for funding research which led to the writing of this chapter under the auspices of the project Rethinking Comparative Syntax (ReCoS) at the University of Cambridge. Thanks also to Anders Holmberg, Jenneke van der Wal, and an anonymous reviewer for comments on an earlier version of this chapter. All errors and omissions are entirely my own responsibility.

1

The special abbreviations used in this chapter are as follows: a = set A (ergative, genitive), av= active voice, b = set B (absolutive), dtv = derived transitive suffix, nmlz = nominalizer, om = object marker, ov = objective voice, sm = subject marker, tv = transitive verb suffix.

2

It is the convention in generative grammar to use capitalization to indicate that Case refers to the abstract syntactic notion as opposed to morphological case.

3

Note, however, that even languages with basic accusative clausal alignment may nonetheless have a series of predicates which assign some other case(s) to certain arguments, for example dative, as in the Icelandic example in (4) (see Woolford 2006a for discussion). While it is a matter of considerable interest how to analyze such examples, we will sidestep such a discussion here for reasons of space, but see section 3.4.6 for a related discussion.

4

There are well-known challenges for this view of case / agreement from (i) languages like Icelandic in which the argument with subject-like properties is not always that which bears nominative case (Zaenen, Maling, and Thráinsson 1985), and (ii) languages in which case/agreement do not match up, a point to which we return in section 3.3.3.

5

Another kind of alignment, termed eccentric agreement by Hale (2001) and agreement displacement by Béjar and Rezac (2009), is determined by the features of both of the arguments in question, subject to Silverstein’s (1976) person hierarchy. We return to the related phenomenon of differential marking in 3.2.3. Needless to say, such agreement patterns pose additional challenges for any parametric approach to case / agreement.

6

Note, however, that if T inherits its phi-features from C, as Chomsky (2008) has proposed, then it is not altogether beyond the realm of possibilities, as matrix and embedded Cs appear to differ considerably in their other syntactic properties.

7

Note moreover that even where passives of intransitives are possible, this is (often) true only of unergative and not unaccusative verbs, providing a potential account for the existence of split-S languages of the kind discussed in section 3.3.2.

8

With the glosses and translation rendered from Portuguese.

9

Further complications arise when one also considers eccentric agreement or agreement displacement whereby agreement patterns are determined by the features of both of the arguments in a transitive clause as mentioned in footnote 5.

10 Note, however, that Tongan does display one feature of syntactic ergativity, the ban on the A-bar extraction of ergative subjects (see Otsuka 2006). We return to this point in 3.3.4. 11 In Northern Pomo a fluid-S system of this kind exists alongside a different, more pragmatically determined split (limited to a small class of verbs) (Deal and O’Connor 2010). 12 Northern Pomo is somewhat exceptional in this regard as it is a split-S accusative system (see example 22).

72

CONTEMPORARY LINGUISTIC PARAMETERS

13 Note that there are other apparent parallels between the two phenomena, with extraction in some syntactically ergative and accusative languages being aided by the presence of a high adverb (see Erlewine 2013). 14 Though see Legate (2012) for a different analysis of Dyirbal. 15 An anonymous reviewer notes that Indo-Aryan languages may be high absolutive without displaying syntactic ergativity. At least in Hindi, however, there is no reason to suppose ABS always comes from T as in non-finite contexts ABS on the object is retained whereas ABS on the subject is lost (see Butt 1995; Davison 2014):

(i) mã=ne

baccõ=se [PRO kıtab-ẽ

paṛʰ-ne=ko] kah-a

mother=erg child.m.pl=abl book-f.pl.abs read-inf=dat say-pf.m.s

In Persian / Farsi, embedded clauses are generally finite and so it is more difficult to test what happens when T loses the ability to assign a structural Case.

16 This list is, of course, non-exhaustive. One obvious omission is the proposal developed by Bittner and Hale (1996). Despite the many merits of this approach, notably its ability to derive two kinds of ergative languages, it relies so heavily on the machinery of GB theory that it is difficult to reformulate in Minimalist parametric terms, though it may be considered a version of the proposal in section 3.4.5. Murasugi (1992) develops a proposal whereby the object raises to spec IP in ergative languages, an account which has much in common with the analyses in section 3.4.3. Campana (1992) develops a variant of the kinds of approaches discussed in 3.4.4 whereby transitive subjects raise to spec AgrOP at LF, whereas objects and intransitive subjects raise to SpecAgrS. Unfortunately, space restrictions prevent a full discussion of these proposals. 17 Note that, as such, it is an account only of high absolutive systems, where T agrees with S/O. 18 For Bobaljik (1993), defective intervention is not an issue as absolutive case is assigned locally by AgrO. It is a potential issue for Assmann et al. (2012) but they define locality in such a way that the problem does not arise. In fact, it is crucial for them that two DPs in a given complement domain are equidistant from a c-commanding probe, as we shall see shortly. 19 As Bobaljik notes, the proposal is not tied to the existence of AgrS/AgrO. See Rezac (2008) for a reformulation without dedicated agreement heads. 20 As Marantz notes, (a) is essentially a stipulation to make sure that accusative is not assigned where a subject has quirky case.

4 Head directionality* Hubert Haider University of Salzburg

4.1. Introduction This chapter is organized as follows. Section 4.2 briefly introduces two aspects of directionality, as a property of heads and as a property of structures. Section 4.3 presents a set of syntactic contrasts that distinguish head-initial and head-final phrases. The explanation for these contrasts is derived and formulated in terms of a directionality theory in section 4.4. Section 4.5 deals with syntactic properties of Slavic languages as examples of unspecified directionality. Section 4.6 briefly compares the directionality properties of phrasal structure with word structure. The appendix summarizes the major differences between two competing accounts. Reference to the position of the head is a convenient criterion for grouping languages. There are strictly head-final languages like Japanese. This means that for any phrase, the head follows its dependents. On the other hand, in strictly headinitial languages (e.g. Austronesian languages), any head precedes its dependents. SVO is the most frequent subtype of head-initial languages. A language of this type is head initial apart from the fact that the position of the subject as an argument of the verb is apparently exceptional. It precedes the verb while the other arguments follow. Cross-categorial uniformity is not obligatory, however. There are categorially differentiating languages. The head position may co-vary with the category of the head. In German and Dutch, verbal and adjectival heads follow; nouns and prepositions precede their complements. Finally, there are languages with seemingly variable head positions. In Slavic languages, for instance, a verb may not only precede or follow its arguments; it may also be “sandwiched” by its objects. Here, scrambling is not the cause but rather the effect in a syntactic setting with characteristics of its own. Arguably, in these languages the directionality of the head relative to its complements is un(der)specified. At least since Greenberg (1963), the search for cross-linguistic correlations has been guided by particular attention to the order of heads relative to other

74

CONTEMPORARY LINGUISTIC PARAMETERS

grammatical properties. Typically, inductive generalizations have been formulated by correlating cross-category patterns. Take for instance Greenberg’s Universal 5: “If a language has dominant SOV order and the genitive follows the governing noun, then the adjective follows the noun.” Had German and Dutch (Greenberg 1963: 67, appendix II: 10) been correctly identified as OV plus V2, Universal 5 could not have been formulated this way. In German, attributive adjectives obligatorily precede; the genitive follows the noun and the nominal arguments of the verb must precede the verb. Greenberg was misled by the V-2 property that typologists had not recognized before Mallinson and Blake (1981).1 (1) dass man [den ehemaligen Präsidenten des Nachbarlandes] nicht einreisen ließ that one [the former president (of) the neighboring-country.GEN.] not enter let A correlation like Universal 5 is an inductive generalization. It describes a clustering of properties of a sample of languages. If correct, inductive generalizations call for an explanation. This means that they have to be paired with general principles of a grammar theory that cover the inductive findings. In grammar theory, linearization properties have been captured in various ways. The position of an item in linear order may be characterized as a reflex of its original phrase-structural position, or as the result of a syntactic process that places an item in a derived position, or even as the outcome of autonomous (post-syntactic) rules of linearization. What would this mean for OV versus VO, for concreteness? In the Principles and Parameters paradigm, based on Chomsky (1981), the position of the verb in the VP has been characterized by a parametric directionality requirement.2 The head determines whether a complement has to precede or follow it. This property is seen as a parametric property of the classes of lexical heads. For VPs with two objects, a parametric choice was assumed between the structural option (2a) and the option (2b). The latter turned out to be empirically inadequate.3 (2) a. [XP [YP V°]V´]VP OV-type b. [V° XP]V´ YP]]VP VO-type (1st guess)

(cf. Chomsky 1981:171)

In (2a, 2b), the linear position of the verb relative to its objects is determined by its directionality requirement. However, most of the syntactic predictions entailed by the left-branching structure (2b) turned out to be empirically inadequate, as Barss and Lasnik (1986) pointed out. Consequently, Larson (1988) suggested replacing (2b) by a derivation ending up at a right-branching shell-structure as an empirically more adequate structure; see §4.4. The current model of Generative Grammar evades directionality parameters especially in those variants that pursue Kayne’s (1984) views. The relation between OV and VO is captured derivationally instead. The base position of heads is assumed to be identical in head-initial and in head-final phrases. As for the VP, the allegedly universal structure is that of the English VP. What appears to vary across languages is believed to be merely the position of the dependents at spell-out. In OV, as for



HEAD DIRECTIONALITY

75

instance in Dutch (Zwart 1994), the originally postverbal dependents are assumed to have moved in order to end up in higher, pre-verbal positions. Once this conjecture is put to test seriously, with sufficient evidence from OV languages,4 straightforward central predictions turn out to be in conflict with empirical evidence; see for instance Haider (2010: Ch. 1) and Haider (2013: Ch. 9). The head-final and the head-initial architecture of phrases produce distinct sets of syntactic properties. Representative examples can be found easily in languages that contain both types of phrases. In German, the VP is head-final while the NP is headinitial. In addition, German word formation provides a fully productive category conversion paradigm. The infinitive form of a verb may be used as a noun. Category conversion therefore yields the closest possible minimal pair context for V° and N° as phrase-final or phrase-initial heads, respectively: (3) a. [ein VerbAcc in ein Nomen konvertierenV°]VP a verb into a noun convert.INF b. das [KonvertierenN° eines VerbsGen. in ein Nomen]NP the [convert.INF a verb into a noun] ‘the converting of a verb into a noun’ In the NP, the preverbal accusative object of the verb corresponds to the post-nominal genitive object of the noun. These are the regular forms of the structural case of the object in a VP and the structural case of an object in an NP. In the VP, the object precedes but in the NP it follows the head. These are the structural aspects of the conversion. And here come some of the concomitant syntactic differences. The German NP as a head-initial phrase behaves syntactically just like an English head-initial phrase, that is, an NP or a VP with respect to scrambling (4), compactness (5), and adjunctions (6), to name just a few syntactic properties that sharply differentiate between head-initial and head-final phrases. (4) a. [ratlose Syntaktiker an ihren wilden Hypothesen erkennen]VP perplexed syntacticians at their wild hypotheses recognize ‘recognize perplexed syntacticians by their wild hypotheses’ b. [an ihren wilden Hypothesen ratlose Syntaktiker erkennen]VP c. das [Erkennen der/von ratlosen Syntaktiker/n an ihren wilden Hypothesen]NP the [recognizeInf. theGen./of perplexed syntacticians by their wild hypotheses d. *das [Erkennen an ihren wilden Hypothesen der/von ratlosen Syntaktiker/n]NP e. *recognize by their wild hypotheses perplexed syntacticians Note that the contrasts in (4) clearly show that scrambling is phrase-specific rather than language-specific. It is not a holistic property of a language. The constraint is nothing but a corollary of the compactness of head-initial phrases (5). For the verb and its object, a scrambled phrase is an intervener and therefore on a par with an intervening adverbial like in (5). (5) a. [ein Abstract für eine Tagung nach der Deadline einreichen]VP an abstract to a conference after the deadline submit5

76

CONTEMPORARY LINGUISTIC PARAMETERS

b. das [Einreichen eines/von Abstracts für eine Tagung nach der Deadline]NP the submit.INF a.GEN/of abstract(s) to a conference after the deadline ‘the submitting of an abstract / of abstracts to a conference after the deadline’ c. *das [Einreichen für eine Tagung eines/von Abstracts nach der Deadline]NP d. *das [Einreichen nach der Deadline eines/von Abstracts für eine Tagung]NP e. *submit for a conference an abstract after the deadline The following set of data illustrates a restriction on adjuncts that occurs only with head-initial phrases as adjunctions sites. The head of the adjoined phrase must be adjacent to the adjunction target. This excludes the adjoining of phrases with post-head material to head-initial phrases. In Haider (2000, 2004, 2013: Ch. 6) this property has been named Edge effect. (6) a. that he had [much more carefully (*than others (thought)) [analyzed the problems]]VP b. dass er viel sorgfältiger (als andere (dachten)) die Probleme analysiert hatte that he much more carefully (than others (thought) the problems analyzed had c. eine [viel sorgfältigere (*als andere (dachten)) [Analyse der Probleme]]NP a much more careful (*than others (thought)) analysis (of) the problems English (6a) contrasts with German (6b). As in German the VP is head-final, the edge effect is absent in (6b), but it shows up with the head-initial NP (6c). In English, both the VP and the NP are head-initial and therefore the edge effect applies to preverbal adverbial phrases as well as to prenominal phrasal attributes. As a consequence, complex attributes with post-head material are invariably post-nominal6 if a language allows attributes on either side of the NP, like in Romance languages.What these facts suggest is a robust syntactic impact of the position of the head within the phrase on the particular syntactic properties of the phrase. Directionality is at issue whenever a structure becomes complex in phrase structure as well as in word structure. In morphology, “perhaps the largest theoretical question is why suffixes are more frequent than prefixes” (Dryer 2013). What this question addresses is a directional asymmetry in word structure. Haspelmath (1992) among others suggests regarding suffixes as heads of complex words, on a par with the right-hand constituents of compounds. The head of a phrase determines the category of a phrase, and the suffix of a complex word determines the grammatical properties of the word, as for instance the category and grammatical gender. Nominal compounding provides a good illustration of the point. In Romance, nominal compounds are head-initial7 but this word formation pattern is strongly constrained (7b). It cannot be applied recursively (Haider 2001). This contrasts with the productive recursive compounding in Germanic languages (7a). (7) a. [N Nh], [N [N Nh]], [N [N [N Nh]]], … (head-final recursive compounding) b. [Nh N], *[[Nh N] N], *[[[Nh N] N] N], … (head-initial recursive compounding)



HEAD DIRECTIONALITY

77

Recursive head-initial compounding would produce left-branching word structures (7b). However, unlike the productive head-final type (7a), the head-initial type (7b) is not used recursively.8 Stacking to the left is productive;9 stacking to the right is not. Stacking to the right would result in left-branching structures. If—as will be discussed in detail below—the structure of complex linguistic structures is right branching, recursive head-initial compounds are restricted to the trivial format of [XN+Y]N.

4.2. Directionality of lexical heads and directionality of structures For lexical heads, directionality information is part of the lexically stored information. The directionality of structures, on the other hand, is a property regulated by the system that determines the properties of phrase structure. Directionality as a lexical property is easy to demonstrate. In German, for instance, relational particles are prepositions plus a handful of postpositions. Some of them alternate between preposition and postposition (8a), but some are grammatical only as postpositions (8b). (8) a. [der Vorschrift nachP°]PP – [nachP° der Vorschrift]PP the regulation.DAT according-to according-to the regulation b. [Gerüchten zufolgeP°] – *[zufolgeP° Gerüchten] rumors.DAT according-to It is clearly not a property of the noun Gerüchte that it has to precede the postposition zufolge. It is a property of zufolge as a lexical head that it is a postposition. Phrased in directional terms, zufolge requires a complement that precedes rather than one that follows.10 The correlation between lexical (classes of) items11 and their directionality property is not limited to relational particles (pre- or post-positions), of course. It is a property of all major lexical categories as well as of functional lexical heads. For instance, there are languages with two sets of complementizers, namely headinitial and head-final ones. An example is Bengali (Singh 1980; Bayer 2001, 1998: 245). This language is representative of a group of Eastern and South-Eastern Aryan languages (Oriya, Assamese, Marathi, Dakkhini-Hindi). Bayer (1998) emphasizes that the final complementizers are base-generated in the clause-final position and argues in detail against hypotheses that surmise that these complementizers might happen to end up in the final position in the course of a derivation that moves the complement of a clause-initial C° to the left, thereby surfacing the complementizer in an apparently clause final position. Koopman (1984: §3.4.2) discusses a similar property in the complementizer system of Vata. Clauses with initial C follow the selecting verb; clauses with clause-final C precede the verb. Bayer (1998) reports the very same distribution for Bengali. As for directionality in structures, it is a graph-theoretic truism that for each binary branching structure, there is a mirror image structure that differs only in directionality. It is no truism but empirically well-founded that the structures of

78

CONTEMPORARY LINGUISTIC PARAMETERS

linguistic expressions do not exploit the mirror image potential of phrase structuring. This is easy to see when we examine the periphery of a clause. There are well-known syntactic phenomena that target the clause initial position, that is, the left periphery. In numerous languages, question formation or relative clause formation places a specifically marked item into the leftmost position. On the other hand, there is no counterpart as a set of languages that target the right periphery for the same class of items, that is, the clause final position. A particular illustrative case is the V2-property of the Germanic languages. A declarative clause starts with a single “arbitrary” constituent followed by the finite verb. Not a single language is known with a mirror image configuration, that is, one that places the finite verb in the penultimate position, followed by a single, arbitrary constituent. There is no evidence that V-penultimate languages could exist. In terms of phrase structure properties, this invites the conclusion that (9b), the mirror image of (9a), is no option for natural language grammars. (9) a. [FP …. [F° … […..]]] b.*[[[…..] … F°] ….]FP In (9), F° is a variable for a functional head position. In general, such a position can be lexicalized either directly by a lexical functional item (e.g. a complementizer) or in a secondary manner, by displacing an independent lexical head with appropriate features into this position, as for instance the finite verb. In the latter case, directionality of the functional head cannot be related to a lexical property of a specific class of lexemes. In German, for instance, the verb as the head of a VP is head-final, but the head position in the V2-structure (9a) is head-initial, that is, it precedes its complement as functional head. The example of V2 is representative of all instances of functional head positions for which there is no primary lexicalization. An analogous case in Romance languages is the functional head position between the subject and the VP that accommodates the finite verb. In English, this position is open only for finite auxiliaries or quasi-auxiliaries. Since the directionality of F° in (9) is not correlated with any lexical property, it must follow from a general system-internal property of phrase structures of human languages. There is no parametric variation of the position of secondarily lexicalized functional heads relative to their complement position, as indicated in (10a, 10b), contrary to empirically insufficiently backed handbook wisdom. Available evidence strongly supports the conclusion that (10b) does not exist. There are no head-final functional phrases that are lexicalized secondarily. (10) a. … [F´ Vi [VP … ei …]] b.*… [[VP … ei …] Vi ]F´ It is telling that (10b) has been tacitly assumed to exist in OV, in spite of the complete absence of any positive evidence of the kind that motivated (10a). The assumption of V-to-T (cf. Gallego and Uriagereka, this volume) in English or French was corroborated by the evidence from intervening items, as indicated in (11a). Corresponding patterns are predicted for OV but they are absent. In head-final



HEAD DIRECTIONALITY

79

clauses, the sequence of clause-final verbs is compact. The predicted order (11b) is ungrammatical, without exception. (11) a. … [F´ Vi [(XP) [VP ei …]]] b. … [[[VP … ei] (*XP)] Vi ]F´ In (11a), a typical instance of XP is an adjunct to the VP as for example a manner adverbial. A case for the XP in (11b) is material that is extraposed to the right edge of a VP. Although extraposition targets the right edge of a VP (12a), this pattern is clearly ungrammatical in the context of (11b). The predicted order would be the unacceptable order (12b). The only acceptable order is (12c). (12) a. [Warten können auf ein Wunder]VP wird er nicht wait can.INF for a miracle shall he not b. *dass er nicht [warten können ei auf ein Wunder] wirdi that he not wait canInf. for a miracle will c. dass er nicht warten können wird auf ein Wunder Extraposition targets the right edge of the containing VP. This is easy to check when a VP is topicalized (12a). If the finite verb were to move out of the VP, up to a clause-final functional head position, the resulting order would be (12b). But this is an unacceptable order. For additional independent evidence12 see Haider (2010: 59–68). The source of the ungrammaticality of (9b) and (10b) is identical. The functional head position that is in need of secondary lexicalization must not follow the phrase whose head is supposed to be displaced into the functional head position. It must precede. This asymmetry is covered by a universal restriction against left-branching (extended) projections (see §4.4, example 23i).13

4.3. Some correlates of the linear position of the head in relation to its dependents This section lists and describes a set of syntactic properties that correlate with the head-initial vs head-final positioning in Germanic languages. By the same token, the list is a challenge for grammar theories and their success in accounting for these phenomena in an insightful way. Directionality-based explanations for the listed properties will be derived in section 4.4. i) Compactness: As illustrated already in the introduction section with examples (4d, 4e), head-initial phrases are compact. Interveners between the head and the nominal complements are ungrammatical. This restriction rules out intervening adverbials just as well as scrambled phrases since they would qualify as interveners as well. Head-final phrases are not compact. ii) Edge effect: The head of a phrase-initial adjunct of a head-initial phrase must be

80

CONTEMPORARY LINGUISTIC PARAMETERS

adjacent to the host phrase. Any material that may follow the head is ungrammatical when the phrase is in an adjunction position. This property can be studied with adnominal attributes or with adverbial phrases, as in example (6) in section 1. There is no edge effect for adjunct adjoined to a head-final phrase,14 and there is not edge effect for phrases in spec-positions either. iii) Particles and result predicates: Without any exception, these elements precede the head in OV and follow the head in VO (13a,b). If the given language allows particle stranding, a particle of a particle verb may be stranded in between two nominal arguments in VO (13c). In OV, the only stranding position is the clausefinal position, that is, the base position of the verb in the VP. The parallel behavior of particles and result predicates is expected, since V+particle combinations typically denote a result relation (13). (13) a. You must cut it loose/off b. Du must es los/ab-schneiden you must it loose/off-cut c. Let’s send her up a drink iv) Higher-type adverbials and negation particles: In VO-languages, sentence adverbials precede the VP (14a). VP-internal placement (14b) is deviant. In OV languages, sentence adverbials arguably are VP-internal. They may be placed in between VP-internal elements and the VP-final verbal head. (14c) merely illustrates the claim. Sufficiently detailed argumentation tends to be lengthy (see Haider 2010: 170f.).15 These patterns will be shown to follow from the directionality system to be explicated in section 4.4. (14) a. He has obviously/unfortunately returned it to the wrong place b. *He has returned it obviously/unfortunately to the wrong place c. Manchmal können Leute damit ja leider/offensichtlich nicht umgehen sometimes can people with-it prt unfortunately/obviously not deal ‘Sometimes, people unfortunately/obviously cannot handle it/deal with it’ The placement of negation particles follows analogous patterns. In VO, the negation precedes the VP. In OV it is VP-internal. In each case, the negation particle closely precedes the head of the VP.16 In both cases, negation minimally c-commands its semantic type domain (Haider 2013: 138). This, but not the VO-pattern, is the cross-linguistically valid constellation. v) In-situ wh-items of a higher type: It is uncontroversial that in English why or how are deviant if they are not fronted (15). Where and when, however, are perfect. Szabolcsi and Zwarts (1993) realized that this reflects a semantic difference. Why and how do not range over individual-type variables, that is, type elements, but over higher types, like properties or propositions? Just like argumental wh-items, when and where operate over individual type variables, namely points of time and place, respectively.



HEAD DIRECTIONALITY

81

(15) a. (Tell me) who called when/*why? b. (Tell me) what happened where/in what manner/*how? In OV languages, this contrast is absent. Any type of adverbial operator may be left in situ (cf. Haider 2010: 118–122). The causal difference is the relative order of adverbs and the verb. In OV, the in-situ wh-elements precede and c-command the (base position of the) finite verb, in VO they do not c-command the (base position of) the finite verb when they follow but they would have to, in order to receive their type-adequate semantic interpretation. vi) Obligatory functional subject position: In Generative Grammar, this property is covered by the EPP. According to Lasnik (2001a: 356), “The ‘Extended Projection Principle’ (EPP) has been […] a pervasive mystery since it was first formulated by Chomsky (1981).” It demands that every clause have a subject in a functional spec-position above the VP. The empirical basis for Chomsky’s generalization was English, and confirming evidence came from European SVO languages. A substantive part of the mystery is that OV languages do not recognize the EPP. The clearest symptom is the necessary presence or absence of an expletive subject. In non-pro-drop SVO languages, a missing subject argument is compensated by means of an expletive. In SOV, subject expletives are ungrammatical in truly subjectless clauses.17 The passive of an intransitive verb produces instances of subjectless clauses. In Scandinavian languages,18 an expletive “saves” the EPP (16a). In German, the expletive in the allegedly universal spec-position for the subject is ungrammatical (16b), but German is not semi-pro-drop, as the obligatory expletive subject in the intransitive middle construction shows (16c). Es would be the appropriate expletive since it is also the obligatory expletive for the clause-initial spec-position in a V2-clause (16d). (16) a. Ofte vart *(det) telefonert often was expl telephoned b. Oft wurde (*es) telephoniert often was expl telephoned c. Hier telephoniert *(es) sich leicht here telephones expl itself easily d. Es wurde oft telephoniert expl are often telephoned

[Norwegian] [German]

What has been overlooked in the various endeavors of shielding EPP against problems raised by German or Dutch is the fact that any SOV language is a problem for EPP and that the EPP itself is merely the name of a problem but not its solution. An EPP feature, strong or weak, does not explain; it merely replaces an empirical generalization by a technical notation in terms of an unexplained feature. Why should OV systematically differ from VO with respect to an EPP feature? There must be a more profound explanation than an ad-hoc feature, especially in the absence of a theory of syntactic features.19 Additional, independent evidence for the structurally different status of subjects in SVO and SOV comes from wh-subjects in situ. In English they are deviant

82

CONTEMPORARY LINGUISTIC PARAMETERS

(Hankamer 1974) and in other SVO languages too; in SOV languages, they are straightforward. Chomsky (1981: 236) acknowledges that clauses like (17a) are ungrammatical because they contain an in-situ wh-subject. The German version is fully acceptable.20 (17) a. *It is not important who thinks (that) who won b. Es ist nicht wichtig, wer glaubt, dass wer gewonnen hat21 it is not important who thinks that who won has The restriction that accounts for the ungrammaticality of (17a) will simultaneously rule out (18a). In each case, there is a wh-subject in situ, but for Chomsky, (18a) is a standard case of the Superiority constraint. Strangely enough, the pattern (17a) has never again been re-examined in the discussions of Superiority violations. As the contrast (18a, 18b) illustrates, Superiority is challenged by data from SOV. (18) a. *It is entirely unclear who(m) what has irritated b. Es ist völlig unklar, wen was irritiert hat it is entirely unclear whom what irritated has

[German]

In sum, the presence or absence of subject expletives as well as the acceptability of in-situ wh-subjects in combination with so-called Superiority violations that involve an in-situ wh-subject are areas of syntactic contrasts between SVO and SOV that call for a more principled explanation than the presently favored ones. In the MP perspective, the structures would be principally alike, with a wh-subject sitting in a pre-verbal spec-position. vii) Order variation among auxiliaries and quasi-auxiliaries in SOV: In VO languages, the auxiliaries precede the main verb in an invariant relative order; in OV they follow.22 It is remarkable that typologists would not immediately recognize the connection with the directionality of the main verb, as Dryer (2009: 204) frankly admits: “The primary conclusion is that there is no obvious explanation for why auxiliary verbs tend to precede the main verb in VO languages but follow in OV languages.” Here is an explanation. Bech (1955) coined the term ‘status government’ and saw the parallel to case government. The status of the dependent verb (e.g. participial or infinitival or aspectual form) is determined by the governing (auxiliary) verb. Just like case government involves directionality, status government does so, too. Hence, O+V languages are V+Aux languages, while V+O languages are Aux+V(P) languages. The main verb is a case governor; Aux is a status governor for the main verb. The directionality of government (viz. for the dependency relation between a verbal head and a depending element) is identical. This was the easy part of verb ordering, and here is the tough issue: In VO-languages the relative order of auxiliaries and the main verb is invariant. In VO languages, the relative order is variable if the given language allows V-movement. Germanic OV-languages meet the latter requirement. The order variation in Dutch and German is illustrated in (19) and (20), respectively.



HEAD DIRECTIONALITY

83

(19) a. *dat hij niets gezien hebben kan that he nothing seen have can b. dat hij niets gezien kan hebben c. dat hij niets kan gezien hebben d. dat hij niets kan hebben gezien

[Dutch] (ans 1984: 1069)

(20) a. dass er nichts gesehen haben wird that he nothing seen have would b. dass er nichts gesehen wird haben c. dass er nichts wird gesehen haben d. *dass er nichts wird haben gesehen

[German]

As SOV languages with immobile verbs (e.g. Korean, Japanese) show, (19a) and (20a) are the base orders, with the dependent verb preceding. A precondition for the variations is the mobility of the verbs. In SOV languages without V-movement (e.g. Asian OV languages like Korean or Japanese), there is a single order, namely the order corresponding to (19a) and (20a). In standard Dutch, (19a) is not available any more. In standard German, on the other hand, the fully inverted Dutch order (20d) is not admitted. There are regional German variants, however, that permit this order (see Schallert 2014: §5.3.1). viii) That-t effect: In English and other SVO languages, fronting the subject across a complementizer is ungrammatical (21a). Swedish (Engdahl 1986: 166) and Danish (Vikner 1995: 12) pattern like English. In Dutch (Den Dikken et al. 2007: 346) or German (21b), the counterpart of English (21a), is grammatical and acceptable. (21) a. Whati do you think (*that) ti has happened in that night? b. Was glaubst du denn dass ti passiert ist in jener Nacht?23 what think you prt that happened has in that night The that-t effect singles out subjects in SVO languages. In these languages, the subject position is preverbal whereas the objects are postverbal. This is a difference in the directionality relation between the verb and its arguments. In SOV, subject and objects are preverbal. There is no difference in directionality, hence no functional subject position and therefore no that-t effect. This is an indication that the appropriate understanding of the that-t effect as an SVO-specific restriction involves directionality (cf. also D’Alessandro, this volume).

4.4. Universal and parametric directionality24 Structural directionality is universal; head-directionality is parameterized. Its values may vary across lexical (sub)classes. Structural directionality is the directional relation between a projecting node on the projection line and its sister node. In principle, phrase structures could be right-branching (22a) or left-branching (22b). Evidently, the structures (22a, 22b) are mirror-images. All available empirical evidence points to the conclusion that a left-branching projection structure (22b)

84

CONTEMPORARY LINGUISTIC PARAMETERS

is not employed by human language grammars. A structure like (22b) is excluded on principled grounds. This is the universal directionality property of structures (cf. 23i below).25 (22) a. [HP X [H´ Y [H´ Z h°]]] right-branching b. [[[h° Z]H´ Y]H´ X]HP left-branching The parameterized property is a property of lexical heads. Complements either precede or follow the head of the phrase. As this formulation tells, this kind of directionality is a property of a relation between the lexical head of a phrase and its complements. Let us call this relation the licensing relation. The directional licensing relation in combination with the universal directionality of structuring is the grammatical source of the syntactic differences between the head-initial and head-final organization presented in the preceding section. The theoretical core assumptions are the following (Haider 1992, 2010: 26, 2013: 3f.): (23) i. Projection lines are universally right-branching26 and endocentric. ii. A dependent phrase is licensed in the canonical direction. iii. The position of a dependent phrase P is licensed =Def. the (projection of the) phrase head h and P minimally and mutually c-command each other. It is the minimal & mutual c-command condition (23iii) that is directly causal for many of the OV/VO contrasts described in the preceding section, namely compactness, scrambling, the edge effect and the need of a functional subject position in VO, with the concomitant syntactic properties of the functional subject position. Let me emphasize that it is the very same principle (viz. merger is universally to the left) applied under parametric directionality that produces the different outcomes for OV and VO. Let us begin with compactness. In keeping with (23i.), the structural grid of a phrase is that of (24a) but not (24b). (24a) offers two alternative foot positions for a head, namely x or y. The actual choice depends on the directionality value of the head. (24) a. … [ … [ … [x y]]] b.*[[[x y] … ] … ] … The value for the canonical licensing direction is parametrical, that is, it is either progressive (“→) or regressive (←). The two implementations in (25) illustrate the directionality difference in the sub-tree that contains the head. (25) a. … [ … [ … [ZP ←V°]]] OV b. … [ … [ … [V°→ ZP]]] VO The crucial differences between OV (25a) and VO (25b) become visible when the phrase becomes more complex. In OV, the canonical direction of licensing is congruent with the universal direction of merger (23i.). In VO, however, the canonical licensing direction is opposite to the universal direction of merger. This is the source of VP-shell formation which in turn is the source of compactness.



HEAD DIRECTIONALITY

85

When a second object is merged in VO (26a), its position is not in the directionality domain of the head. Hence the head needs to be re-instantiated. This amounts to the formation of what is called a VP shell in Generative diction, namely (26b). (26) a. … [ YP [V´ V°→ ZP]] b. … [Vi → [ YP [V´ ei → ZP]]] c. … [Vi → [(*π) [ YP [(*π) [V´ ei

→

ZP]]]]]

A shell is necessarily compact because of the minimality requirement of the licensing condition. Any intervening phrase would destroy the relation of minimal c-command between the verb and YP, or between YP and the trace of the verb. In (26c), π would either disrupt minimal c-command between V and YP or between YP and the lower empty verb position. In each case mutual, minimal c-command is destroyed. Note that YP must minimally c-command the lower V-position. Mutual c-command is fulfilled if V minimally c-commands YP, and YP minimally c-commands a link of the chain of V. In (26c), the lower π would disrupt the minimal c-command of the lower, empty V position by YP. No mutual c-command between V and YP implies no licensing of YP by V. In OV, the situation is different because the canonical directionality of licensing is congruent with the directionality of merger (27). Hence not only the head but any other node on the projection line can serve as a licensing node. As a consequence, there are no shells in OV. In sum, since minimal & mutual c-command is the core of the licensing relation; interveners are excluded in complex head-initial phrases but not in head-final ones. In (27a), V´ as a projection of the head is a licit licenser for YP, but not in (26b) for the simple reason that the canonical directionality domain of V´ includes YP in (27a), but not in (27b). Consequently, the OV structure tolerates interveners (28a), but the VO structure does not (28b, 28c). (27) a. … [ … [YP ← [V´ ZP ←V°]]] b. … [ … [YP ← [V´ π ← [V´ ZP

←

[V´ π V°]]]]]

An adverb, for instance, as an adjunct to V’ would not interfere with minimal & mutual c-command, since there is always a sister node for the next higher argument that is a projection of the head with the required directionality. This is the essential difference between head-initial (VO) and head-final structures (OV) with respect to compactness. In OV, the projection nodes are licensing nodes, in VO they are not, because of the directionality mismatch. In VO, the only element that is able to provide the directionality requirement is the verbal head, and therefore it must be re-instantiated, whence the shell structure (26b) of complex head-initial phrases. As a consequence, head-initial structures are compact and do not leave room for scrambling. (28) a. He would show (*unhesitatingly) someone (*voluntarily) his collection b. Er würde jemandem ohne Zögern seine Sammlung freiwillig zeigen he would someone without hesitation his collection voluntarily show In (28b), the adverb “unhesitatingly” would destroy the minimal c-command relation between V and the object “everyone.” The adverb “voluntarily” disrupts

86

CONTEMPORARY LINGUISTIC PARAMETERS

minimal c-command between the object “everyone” and the lower, empty position of the verb. In each case, minimal & mutual c-command is violated. In (28b), on the other hand, there is always a sister node of the projection line with the required licensing directionality. Minimal & mutual c-command is a trivial property of sister constituents. The very same intervener status blocks scrambling. If an argument is scrambled, this means it is adjoined higher up. This turns the scrambled item into an intervener element. Hence, compactness and the ban against scrambling in VO are just two sides of the same medal. Let us turn to the edge-effect. It singles out adjuncts that precede a head-initial phrase. What makes them special is the fact that these adjunct positions are not in the canonical directionality domain of the head-initial phrase. In head-final phrases, any preceding adjunct is in the directionality domain. Provided that the edge effect is an empirically correct generalization, it is a handy diagnostics for head-initial versus head-final structures, even if its grammatical source has not been fully understood yet27 (see Haider 2013: 144–5 for a proposal). Let us proceed now to the hallmark of SVO languages, namely the functional subject position. The trigger of the EPP property of SVO structures is a directionality mismatch. In SVO, the highest argument in the VP is not in the directionality domain of the verbal head (29a). The canonical directionality is to the right; the directionality of merger in phrases is to the left. Neither the verb nor a projection node can provide directional licensing for the VP-internal subject in (29a). Therefore, a functional head has to provide directional licensing (29b). The functional projection establishes the spec-position that is typical for SVO languages (29b). (29) a. ………………. [VP XPSubj. [Vi → [YP [V´ ei → ZP]]]] b. [FP XPj [F´ F°→ [VP ej [Vi → [ YP [V´ ei → ZP]]]]]] In OV, any argument of a verb is within the directionality domain of the verbal head, whence the absence of this functional projection in the OV-based clause structures. The functional spec-position in (29b) is at the same time the position of expletive subjects. The absence of this functional layer in OV is the grammatical reason for the absence of expletives for this kind of subject position in OV. The EPP is a property of SVO languages. The functional spec-position of the functional head that licenses the VP-internal subject is obligatorily lexicalized in SVO (except for pro-drop). The that-t effect and the restriction against wh-subjects in situ are immediately related to this functional subject position. Therefore it is not surprising that SVO languages differ from SOV languages precisely in these respects. The explanation for the ban against a wh-subject in situ in SVO proposed in Haider (2010: 117) is based on this structural difference. In VO, an in-situ wh-subject is in a functional spec-position. In OV, the in-situ wh-subject is in its VP-internal argument position. A wh-pronoun in a functional spec-position gains operator status. It cannot be interpreted as a dependent wh-element anymore. This rules out patterns like (17a), discussed by Chomsky (1981: 236). However, the acceptability of utterances with an in-situ wh-subject improves once the wh-subject is endowed with operator properties, as Chomsky (1981: 238) admits and Hornstein (1995: 144) observes.



HEAD DIRECTIONALITY

87

(30) a. He wonders what who bought where? b. What did whoi/*j reveal about hisi/j mother? In Chomsky’s example (30a), who is the binding operator for where, just like in Hornstein’s example (30b), who is the binding quantifier for his. Once the wh-subject finds an element that may serve as provider for a variable to be bound, acceptability improves. Let us return now to the rest of the OV-VO contrasts listed above. The distribution of particles and result predicates is a straightforward consequence of the licensing directionality of the verbal head. As predicates that are dependent on the verbal head, they are bound to occur in the canonical directionality domain. Furthermore, these elements are inert, that is, they do not scramble. If they end up distant from their verb, they must have been stranded. In all Germanic V2-languages, fronting of the finite verb strands the particle, but only in the stranding Germanic SVO languages, a stranded particle may appear in between two nominal objects. In the Germanic SOV languages, the particle immediately precedes the phrase final V position. (31) a. He sent them up a drink b. The secretary sent the stockholders out a notice The position of the particle in (31) is the lower, empty verb position in the VP shell. Thus, particle distribution is an immediate source of direct evidence for the shell structure of a head-initial VP, and the absence of a shell structure within head-final VPs. Higher adverbials and negation particles semantically relate to the propositional content of the clause. Tense and mood situate the proposition. Tense and mood are coded functionally, represented formally on the finite verb. Hence, higher adverbials and sentence negation must c-command (a chain link of) the finite verb. In VO, this requirement can be met only in pre-VP positions. In OV, however, any position on the projection line preceding the VP-final head position c-commands the verb. Hence, VP-internal positions are suitable for sentence adverbials or sentence negation in OV but not in VO. Elevating the pattern of VO to the rank of a universal model would be inappropriate. It is not a universal property of negation or sentence adverbials that they have to precede the VP. This is merely a consequence for VO-languages. Verb clustering is related to the right-branching constraint (23i.), too, but only indirectly. Stacking VPs in OV is bound to produce left-branching structures (32a). Cluster-formation avoids the recursive stacking of VPs in favor of recursively clustering the verbs within a single VP (31b). In VO, stacked VPs are right-branching (32c). Grammars are parser-friendly (as a result of cognitive evolution; see Haider 2013: Chapter 2 and Haider in press) and provide clustering (32b) as means of avoiding central embedding structures (32a). (32) a. *… [VP [VP [VP XP V1] V2] V3] b. … [VP XP [[V1 V2]V° V3]V°-cluster] c. … [VP V1 [VP V2 [VP V3 XP]]]

88

CONTEMPORARY LINGUISTIC PARAMETERS

A verb cluster is a complex structure that consists of head-to-head adjoined verbs. Since adjunction preserves the category, a head-to-head adjunction cluster is of the category V°.28 The cluster is a compact unit since head-to-head adjunction leaves no room for phrasal interveners.

4.5. Unspecified directionality? Slavic as an exemplary case Head-initial vs head-final does not constitute an exhaustive classification. There are reasons for assuming a third setting, namely one with variable head positioning. Previously, this option has been dismissed on a priori grounds, presumably in remembrance of the peripherality dogma of x-bar theory which postulates that lexical heads must be phrase-peripheral (see Kornai and Pullum 1990: 35).29 A representative example is the Slavic family. Slavic languages are customarily filed as SVO, mainly because subject-verb-object is a frequent word order in these languages.30 However, an SVO ascription is questionable because typical syntactic characteristics of SVO languages are missing or contradicted. In fact, Slavic languages share more syntactic properties with OV than with VO languages as for instance scrambling, absence of the edge effect, order variation of auxiliaries, no Superiority effect for in-situ wh-subjects, no freezing of preverbal phrases (see Haider and Szucsich in press and the literature cited there). Let me just exemplify the last property since it has not been introduced in the preceding section. In the German example (33a), the object of the scrambled, preverbal infinitival clause is fronted to the initial position of the main clause. In the Russian examples an attribute of a preverbal NP is questioned or topicalized and fronted.31 In an SVO language like English, extractions from these positions are ungrammatical. (33) a. Weni/Ihni hat [damit ei zu konfrontieren]j keiner ej gewagt? whom/him has [with-it to confront] nobody dared ‘Whom has nobody dared to confront with it?’ b. Kakujui Ivan [-i mašinu] kupil svoej žene? whichi Ivan [-i car] bought his wifeDat ‘Which car did Ivan buy for his wife?’ c. Japonskujui Ivan [-i mašinu] kupil svoej žene Japanesei Ivan [-i car] bought his wifeDat ‘A Japanese car, Ivan bought for his wife’

[German] [Russian]

In German, a scrambled phrase remains within the directionality domain of the verb. In English, a preverbal phrase is outside the directionality domain. And what about the preverbal NP in (33b, 33c) in Russian? It behaves as if it was within the directionality domain, and it arguably is indeed. Could a Russian verb be an “ambidextrously” licensing head? In fact, this is what underspecified directionality amounts to. (23ii) refers to canonical directionality, which is parameterized. The values for



HEAD DIRECTIONALITY

89

the parameter are regressive or progressive, that is, to the left or to the right. This is not the whole picture, however. There is an implicit third option, namely the option that the value is not specified. In this case the “canonical” directionality is free. Let us call this setting the third type (T3). Languages of the third type may appear to be of a mixed type, with OV and VO properties. This impression is not wrong, but these languages are not mixed. Their grammars merely generate the properties that are consistent with the unspecified directionality of licensing. Here is an example. The patterns in (34) are genuine T3 patterns. At each point of merger, licensing may be implemented with either of the two directionality values. (34) a. [X ←[Y ←[Z ← V°]]] b. [X ←[Y ←[V°→ Z]]] c. [X ←[Vi°→[Y [ei → Z]]]]

OV-like pattern T3-only pattern VO-like pattern

Pattern (34c) is a frequent one, arguably for information-structure reasons, and provides the basis for misclassifying T3 languages as exceptional VO languages. They appear exceptional because of the patterns (34a, 34b) that are absent in unequivocal SVO languages. These patterns are customarily explained away as scrambling effects. Y and Z (in 34a) or Y in (34b) are said to have been scrambled across the head-verb. Independent evidence for this claim is wanting. The T3 hypothesis correctly predicts the deviations from the typical SVO properties. There is no edge effect, because pre-VP adjuncts are in the directionality domain in each case in (33). The subject is in the directionality domain. So there are no effects triggered by the functional subject position of the SVO languages, viz. no effect for in-situ wh-subjects (e.g. Toman 1981; Bošković 1997; Meyer 2004), no obligatory subject expletives in EPP contexts, and no freezing effects for extractions out of preverbal constituents (Stepanov 2007). The order variation of auxiliaries (Embick and Izvorski 1997) directly reflects the free choice of the directionality value.

4.6. Directionality in morphology Directionality is an issue of word structure, too. Simple inspection of the crosslinguistic order patterns of morphemes tells that suffixing is more frequent than prefixing, both in inflection and derivation, and that right-headed compounding is more frequent than compounding with a left-headed base. This is a first indication that word structures show a strong preponderance of right-branching structures, just like syntactic structures. Williams (1981), who notes an asymmetry of headplacement in morphology, proposed a Right-Hand Head Rule. The word structure of a sequence of morphemes that constitute a word consists of the base (head) and the (structured) elements that are binarily merged with the base. The proper identification of the word structure crucially depends on the correct identification of the base. In (35a), the base is the verb, and it gets prefixed. In (35b) the base is the nominalizing suffix. The category change from verb to noun is a function of the category of the suffix. In (35c) it seems as if the prefix

90

CONTEMPORARY LINGUISTIC PARAMETERS

is category-changing which is a property of a base element. Does this mean that the prefix en- is the base and that the word structure of these English words is left-headed? (35) a. decompose, reiterate … b. composer, iteration … c. enrich, encrust, enlighten… d. ergrünen, verkrusten, belichten… [German] en+green+INF, en+crust+ INF, en+light+ INF If the structure of a word like encrust consisted merely of “en-” plus noun, the prefix would have to be the base that combines with a noun and yields a verb. But there is another, equally plausible analysis. Category conversion (viz. word class change) already applies to the noun (cf. a brush, to brush; a maple-leaf, to maple-leave down) and yields a verb, which gets prefixed. Since English is the only Germanic language that lacks an infinitive suffix, conversion is less easy to identify than in other Germanic languages. The German examples (35d) clearly show an infinitival ending that apparently combines with an adjective or a noun. However, the prefix is a preverbal prefix that is combined with a verb (cf. er-arbeitenV; be-arbeitenV). The verb is generated by category conversion. In this case, the infinitival suffix is category changing inflection (see Haspelmath 1996 for a crosslinguistic investigation of word class changing inflection). In English, conversion cannot be signaled morphologically due to lack of infinitival morphology. At this point of the discussion, let us compare word-structure (word syntax) and phrase structure (phrasal syntax) with respect to branching structure and directionality. In phrasal syntax, left-branching is ruled out for the projection line of a phrase. This amounts to a uniform branching structure, viz. a unidirectional, rightbranching buildup of phrases, with the base position of the head at the end of the phrase which is simultaneously the bottom. Head-final or head-initial is the reflex of the parameterized licensing relation. Licensing is directional and the directionality value is a lexically determined property of heads. Head-initial phrases turn out to be necessarily more complex than head-final ones, since the initial head is in a fronted position and in relation to its trace in the base position. In other words, complex head-initial phrases employ a shell-structure. Word syntax is organized in a parallel fashion, with the exception of shells. Left branching is ruled out and there is parameterized directionality of licensing. However, the counterpart of a shell-structure is necessarily unavailable for complex word structures. Word structure elements are not subject to word-internal displacement rules (viz. antecedent-trace configurations). As a consequence, complex left-headed word structures cannot be generated since this would presuppose a shell-structure. Left-headed word structures are restricted to the trivial format of a head and an element in the sister position (Haider 2001): (36) a. [Aπ X]π b.*[[ Aπ X]π Y]π



HEAD DIRECTIONALITY

91

If (36a) is the structure of a two member compound, branching direction does not come into play, but for (36b) it would. The result would be a left-branching structure, which is ruled out. Immediate confirmation comes from compounding in Romance languages. As mentioned in the introductory section, nominal compounds are left-headed in these languages and recursive compounds are restricted to the format of (36a). Here are examples from Spanish and Catalan: (37) a. perro policía – *policía perro [Spanish] dog police (police dog) b. home aranya – *aranya home [Catalan] man spider (spider man) c. *perro policía mascota [Spanish] dog police pet ‘pet police dog’ Piera (1995) uses examples like (37c) as evidence for the ungrammaticality of recursive nominal compounding in Spanish. The same is true for the other Romance languages. In fact, no language with recursive nominal compounding is known whose compounds are head-initial.32 This contrasts sharply with the productivity of recursive head-final nominal compounds in other languages. In the case of affixation, the ban against left-branching would arise only if an affix had more than one dependent or modifying element, but this is not the case. Neither a structure like (38a) nor (38b) is a recursive left-branching structure. In each case there are two separate affixes, each one combining with a single, possibly complex element. (38) a. [Affβ [Affα X]α ]β b. [[X Affα]α Affβ]β How would genuinely recursive structures with an affix as the base element look like? (39a) is a right-branching structure with a suffix as base. (39b) is the left-branching counterpart with a prefix as the base element. (39) a.

α b. α α

A

B

α A affα

affα

B

The word structure (39a) has a right-branching projection line, while (39b) has a left-branching one. Hence it is ruled out by the universal restriction against left-branching projection lines (see 23i). It is not clear, however, whether one of these structures is implemented at all in natural languages, with an affix as base. If B is a word that gets affixed by joining with the affix, then A is merged with a word, consisting of B plus affix. So A is not in the same relation to the affix as B is. It seems that nominal compounding with a noun as base element (in the place of the affix

92

CONTEMPORARY LINGUISTIC PARAMETERS

in (39) is the only case in which a recursive word structure is needed, and in these cases (39b) is correctly excluded. Note that neither iterative prefixing nor iterative suffixing generates the type of right-branching word structures that are excluded, as (40) shows, except maybe for iterating the very same suffix (40c). The excluded word structure is (36b). A single base element is merged consecutively with material on its right. (40) a. [Affα [Affβ X]β]α prefixes as base elements b. [[X Affβ ]β Affα]α suffixes as base elements c.*[[X Aff α ]α Affα]α iterated suffix as base element In the case of (40c), the iteration of the suffix would produce kind of a recursive left-branching structure. However, it is safe to conjecture that the English prefixiteration in (41) does not have a counterpart in any language in which the analogous affixes are suffixes. (41) a. anti-Semite, anti-anti-Semite, … b. pro-British, pro-pro-British, … A well-known case of suffix-iteration is the formation of diminutives. However, as expected, the iterated diminutive suffixes are different, as the examples from Spanish, Russian and German illustrate (42). Each suffix forms a separate diminutive word. (42) a. burr-o (donkey) > borr-ic-o > borr-iqu-it-o b. Katya > Katyusha > Katyushka > Katyushichka > Katyushenichka c. Esel (donkey) > Eselein > Eseleinchen (*Esel-lein-lein; *Esel-chen-chen) In sum, there is directionality in word structure, too. A category-changing affix that serves as the base of a complex word may be a suffix or a prefix. This corresponds to the parametric directional licensing by lexical heads in phrasal syntax. Recursive left-branching structures are excluded in word structuring just like in phrase structuring. There is no counterpart of complex head-initial projections in word structure, however, since word syntax does not provide the means of forming head-initial shell structures.

4.7. Conclusion Directionality-based conditions are operative in phrase structuring in two different respects. On the one hand, there is the cross-linguistically uniform directionality of merger. Merger is to the left; hence the resulting phrases are right-branching. Left-branching projections lines are not admitted. On the other hand there is the parameterized directionality of licensing. The parameter values are the alternative directions of licensing associated with lexical head classes, which yield head-initial and head-final structures, respectively. The more complex organization of head-initial structures (“shell-structure”) is the direct consequence of the mismatch between the universal branching direction



HEAD DIRECTIONALITY

93

and the licensing direction of the head in head-initial structures in an otherwise uniform mode of licensing (mutual, minimal c-command in the canonical direction). If the lexical directionality value is not specified, phrase structuring allows for alternative positions of the head of a phrase. This is the case in a large sample of the languages of the world. The Slavic languages are representative of this setting. Word structure follows the universal uniformity of branching. They are rightbranching. Recursive expansions are always head-final. The head-initial counterparts of phrase structures do not come into being because word-structure does not provide the means for shell-structures.

4.8. Appendix: The LCA (Kayne 1994) vs the BBC (Haider 1992) In the early nineties, Haider (1991, 1992) and Kayne (1992, 1994) proposed complementary solutions for the very same hypothesis. The hypothesis was this: phrase structures are universally asymmetric. They are right-branching. In Haider (1991, 1992), the asymmetry is covered by the BBC (Basic Branching Constraint), as an axiom of syntactic structuring. Kayne (1992, 1994) derives the asymmetry from an axiomatic trans-derivational constraint, viz. the LCA (Linear Correspondence Axiom). This constraint requires that grammars be organized in such a way that the linearization of the terminals becomes a function, that is, a unique mapping, of structural asymmetries on linearization asymmetries. Interestingly, the LCA can be met by alternative functions from structure to linearization, namely by (i) precede or by (ii) follow. If α asymmetrically c-commands β, then the terminals dominated by α will precede the terminals dominated by β in the implementation (i). However, in the implementation (ii), the terminals of α will follow the terminals of β. The two variants yield complementary grammar families but one family does not materialize in the set of human languages, as it seems. Grammars of natural languages do not employ (ii). So, in the LCA theory, the mapping directionality must be stipulated. In fact, the LCA predicts three types of structuring. Type (i) would uniquely map c-command on “precede.” Type (ii) would uniquely map c-command on follow. Type (iii) would be the type with category-specific mapping relations (precede for one category, follow for another category) that produce differentiated linearization patterns, analogous to German and Dutch head-positioning in NP vs VP. Followers of the LCA theory tacitly assume that languages are only of type (i) and overlook the other options. This tacit assumption is equivalent to the tacit assumption of a directionality requirement. The LCA and the BBC theory handle head-initial versus head-final phrase structures in complementary ways, namely in terms of phrasal chains (LCA) versus head-chains (BBC). The LCA theory derives head-final structures from head initial ones by means of phrasal movement. Any postverbal item is fronted. In the evacuated phrase, the head ends up as the right-most terminal of the phrase. In the BBC account, head-initial versus head-final is a property of the positioning of the head of a phrase. Complex head-initial phrases involve head-chains.

94

CONTEMPORARY LINGUISTIC PARAMETERS

In the LCA system, head-final phrases are more complex than head-initial ones, since they are derivational continuations of head-initial phrases. In the BBC model, head-initial phrases are more complex than head-final ones since complex head-initial phrases consist of shell structures. This perfect complementarity opens numerous ways of empirically testing the relative explanatory success of the two models. These predictions are evaluated in detail in Haider (2013: Chapter 9). It turns out that the LCA-based predictions are sans empirical support. The following table lists the major issues; for the details see Haider (ibid.). LCA predictions (empirically unsupported) BBC predictions (empirically supported) 1. In OV, preverbal phrases are in spec-positions and therefore opaque for extractions like a preverbal VO subject. Hence in OV, all preverbal arguments are opaque.

In OV, preverbal phrases are transparent for extraction since they are in the domain of their selecting head, and therefore transparent like postverbal objects in VO.

2. Scrambling places phrases into preverbal spec-positions and thereby makes them opaque for extractions.

2. Scrambling is adjunction within the directionality domain of the head. Scrambled phrases remain transparent for extraction.

3. OV and VO languages are expected to display the “EPP-effect.”

The “EPP-effect” is an SVO phenomenon and absent in SOV.

4. Diachronically, an SOV language develops out of an SVO language. SOV is the universal structure that gets masked by massive phrasal fronting.

Diachronically, SOV and SVO are inverse settings. Typically, the OV or VO grammar develops out of a T3 grammar, by fixing the directionality of heads.

5. Edge effect is not admitted: Adverbials and modifier phrases (e.g. adnominal attributes) are phrases in spec-positions and behave syntactically like phrases in spec-positions.

Edge effect is admitted: Adverbial and attributive phrases are adjoined to the phrases they modify. The edge effect immediately correlates with directionality.

6. “Immobile” items remain in a postverbal position.

“Immobile” items remain at the directionally canonical side of the head they are dependent on, i.e. preverbal in OV, postverbal in VO.

7. No adjunction. Spec-head-complement is the universal building scheme. Adjunction structures are admitted, neither for phraseto-phrase, nor for head-to-head adjoining.

Adjunction is a standard structural configuration. Adverbial or attributive phrases are instances of phrase to phrase adjunction. Verbal clusters are instances of head-to-head adjunction.

8. No lexically specifiable directionality. Cross-categorial linearization differences are differences in terms of fronting items. Example: Clause-final complementizers are clause-initial, but their complement has been fronted.

Directionality values are lexically specifiable. Cross-categorial linearization differences may be differences in terms of the directionality value. Example: Complementizers with the directionality value “regressive” are clause-final.

9. No nominalizations of the series of verbal Nominalization of verbal clusters in OV are heads in either VO or OV, since they are nominalizations of the same series of verbal heads of stacked VPs in each case. heads that project stacked VPs in VO.



HEAD DIRECTIONALITY

95

Notes *

This paper has benefitted greatly from advice and criticism by anonymous reviewers and the editors. My gratitude does not imply any responsibility on their side for whatever shortcomings may have remained.

1

“The order used for a stylistically unmarked version of ‘John saw Mary’ in German would be SVO, too, but to simply call German an SVO language would disguise the verb-second nature of its word order” (Mallinson and Blake 1981: 129).

2

Parametric directionality has been assumed for case or theta-role assignment (see for instance Koopman 1984: 120).

3

Not only in typological schools is head-initial serialization confounded with the structural notion “left- / right-branching,” as in Dryer (2009: 185): “I argued that the word order correlations reflect a tendency for languages to be consistently left-branching or consistently right-branching.” He obviously means that in these languages phrases tend to be head-initial (“left”-headed) or head-final (“right”headed). This is a statement on word order, not on structure. Positive evidence for the alleged left-branching structures is not delivered, of course.

4

Note that the vast majority of syntacticians working in the Generative Grammar paradigm are native speakers of English or some other Indo-European SVO language and that “every p-theory [present day grammar theoryH.H.] was developed by a native speaker of English” (Dixon 2011: 182).

5

All six permutation variants of the three phrases are grammatical.

6

In Romance, APs are head-initial, hence APs consisting of head plus complement are invariably postnominal. In English, the prenominal position (i) is ungrammatical for a complex AP, just like in Romance languages. So an appositive post-nominal variant is the last resort option (ii). In German (iii.), the complement precedes A°.



i. * a [familiar with Kayardil]AP linguist



ii. a linguist, [familiar with Kayardil]AP



iii. ein [mit Kayardil vertrauter]AP Linguist

7



For instance Fr. café filtre (filter coffee); It. conferenza stampa (press conference); Sp. hombre rana (frog man)

8

Here is an example of recursive N+N compound formation: stone – [corner stone] – [granite [corner stone]] – [ersatz [granite [corner stone]]], …

9

E.g. “Space-speak is an engineering technology concept expression manuscript sentence grammar device.” David McNeill. URL: http://recremisi.blogspot.co.at/2010/08/ linguistics-speaking-of-space.html (18.3.2014).

10 Zufolge, halber, zuliebe are genuine postpositions. Whatever complement they select precedes. This must not be confused with the fact that the defective pronoun “es” is avoided in the Germanic OV languages as a complement of prepositions. Instead, there is a pre-clitic: [mit es] → damit (Ge.); [met het] → ermee (Du.). 11 Directionality may even be item-specific as in the case of “enough” and any of its cognates (Du. genoeg, Sw. nog, Da. nok) It is the “odd-ball” item. Unlike all other measure modifiers, which precede, it obligatorily follows the phrase it modifies: strange enough – sufficiently strange, seltsam genug – genügend seltsam, etc. 12 For German and Dutch, a post-VP functional head as a target of V-movement has been

96

CONTEMPORARY LINGUISTIC PARAMETERS

assumed by mere analogy to the pre-VP functional heads in English. Clear positive evidence has been missing from the beginning. 13 Note that this restriction covers the otherwise unexplained so-called Final-over-Final (FoF) constraint (Haider 1992: §5.2; 2013: 132–5). FoF says: “For all heads {a, b, &} on a single projection line, if a is a head-initial phrase and b is a phrase immediately dominating a, then b must be head-initial. If a is a head-final phrase, and b is a phrase immediately dominatinga then b can be head-initial or head-final.” Biberauer, Holmberg and Roberts (2007) readdress an issue that has originally been opened by Kiparsky (1996: his example 20c): There are no VO languages with a post-VP “I”-position (8b). The BBC eliminates the unattested case: there are no left-sisters of functional heads targeted by head-movement (i.e. functionally extended VPs), neither in OV nor in VO. 14 Note that the edge effect is unexplained if an adverbial phrase is assumed to be accommodated by the spec-position of a covert adverbial functional head. Phrases in Spec do not display an edge effect. 15 A check by means of a simple corpus search count is instructive. A web search for “sent it unfortunately to” produced two hits while “unfortunately sent it to” produced 4,600. There was no hit for “returned it obviously to” but 1,560 for “obviously returned it to” (Google search, March 21, 2014). 16 More precisely, negation particles c-command the item that is their semantic target. For constituent negation it is the negated phrase; for sentence negation it is the item that situates the situation variable, that is, the (base position of the finite) verb. 17 Quasi-arguments, like weather-verb subjects, are not expletive. They are semantically empty arguments and this information is part of the lexical entry of the respective verbs. 18 In English, the obvious candidate for a subject expletive, namely there, became unavailable once it had been tied to a postverbal subject argument in the “thereconstruction.” Grammar change with respect to available expletives has turned English into the only Germanic language without a passive of intransitive verbs. 19 The auxiliary assumption that every OV language is a semi-pro-drop language obviously misses a generalization. Being semi-pro-drop is a contingent property, but the absence of an expletive subject in OV is a universal property. 20 In fact, the counterparts of each of the ungrammatical English sentences of Chomsky (1981: 236) are grammatical in German. Chomsky (1981: 237) attributes the deviance to the that-t effect even though “the presence of the complementizer is immaterial.” See example (17a) and remember that the that-t effect is inoperative when “that” is absent: Whati do you think (*that) ti explains this contradiction? 21 It is easy to locate this pattern in corpora, even if it likely to be infrequent: Wer sagt, dass wer unaufmerksam ist?

http://www.scilogs.de/hochbegabung/hochbegabt-und-aufmerksamkeitsgest-rt/

Wer meint, dass wer meint, er sei Hellseher?

http://de.board.bigpoint.com/seafight/archive/index.php/t-173206-p-20.html

22 Greenberg restricts his statement (Universal 16) to VSO vs SOV because of misclassifications in his sample. German and Dutch are misclassified as SVO and their auxiliary verbs pattern in accordance with OV, obviously.

23 Den Dikken et al. (2007: 345) report that acceptance is highest for Dutch informants when there is material in between the trace and the verbs following it. For me and



HEAD DIRECTIONALITY

97

informants I consulted there is no difference between (21b) and the version with the extraposed adverbial PP preceding the verb “passiert.” Both orders seem to be equally acceptable for speakers of standard German in Southern Germany and Austria. Northerners are known to shy away from extraction of any kind of arguments across a complementizer. 24 What follows is a condensed explication of the theory of parameterized directionality developed in Haider (1992, 2010: Ch. 1, 2013: Chs 3 and 5). 25 This claim has been published for the first time in Haider (1992): BBC = Basic branching conjecture. 26 In other words, the direction of merger within a phrase is universally to the left. 27 Note that under the currently widespread assumption that adverbials are accommodated by specs of empty functional heads, there is no way of capturing the edge-effect. Phrases in spec-positions would not be affected. 28 Note that clusters are not compounds. Head-to-head adjunction structures are syntactic structures that are open for syntactic processes like movement within the cluster (see German or Dutch IPP constructions). A compound would be the result of a word formation process. Its internal structure is not accessible for syntactic processes like movement. This is contrary to Bierwisch’s (1990) attempt to dervive clustering morphologically. 29 Variable head-positioning has been considered by Hoeksema (1992: 121), and by Santorini (1993) for Yiddish. 30 “It is generally acknowledged that Russian is an SVO language in neutral contexts” (Bailyn 2002a: 180). 31 Note that this presupposes that the N-head is able to license to the left, which is predicted if the licensing directionality is “free” in a T3 phrase. 32 Hoeksema (1992: 126) discusses Vietnamese as a potential counterexample. He cites multi-member head-initial compounds but they are not recursive (see Haider 2013: 196). They can be analyzed as two-member compounds whose members are (lexicalized) two-member compounds themselves.



5 Parameters and argument structure I: Motion predicates and resultatives* Víctor Acedo-Matellán and Jaume Mateu University of Minho and Autonomous University of Barcelona

This chapter and the following one study the parametric variation in the expression of argument structure. In this chapter we deal with the parametric variation associated to the expression of directed motion and resultativity, and we explore to what extent both are related. The structure of the chapter is as follows. In Section 5.1 we provide an introduction to the parametric variation in the expression of argument structure. In Section 5.2 we focus on the expression of directed motion. In Section 5.3 we deal with the cross-linguistic variation in the licensing of adjectival resultative constructions. Section 5.4 is an overview of the research done on the diachronic variation in the expression of directed motion and resultativity. Finally, Section 5.5 provides overall conclusions.

5.1. Parameters and argument structure It is self-evident that cross-linguistic variation in argument structure is explained differently according to the particular theory of cross-linguistic variation and the particular theory of argument structure entertained. Moreover, whether crosslinguistic variation arises as a proper theoretical issue depends on how argument structure itself is conceived. Cross-linguistic variation in argument structure was not an issue in the very first stages of the generative enterprise, since argument structure was encoded as an idiosyncratic property of each predicative item of the lexicon of a given language. Thus, in

100

CONTEMPORARY LINGUISTIC PARAMETERS

Chomsky (1965), each verb had a subcategorization frame determining its syntactic environment, with no general constraints formulated on the argument realization possibilities of any lexical item. This made it impossible to consider cross-linguistic variability in argument realization, since there was in fact no object that could vary across languages—no verb classes or any other argument structure universal. It is only when argument structure gains independence from particular lexical items that cross-linguistic variation emerges as a research question. A first step in this direction was the identification of thematic roles as primitives participating in the argument structure of all verbs (Gruber 1965; Fillmore 1968). For instance, when the notion of Agent and Patient became independent of particular verbs, generalizations on the universal expression of these theta-roles could be formulated—through universal thematic hierarchies (Larson 1988; Baker 1997; Grimshaw 1990, and others)—and, likewise, descriptions on how that expression may vary across languages. Since the 1980s cross-linguistic variation in Generative Grammar has been modeled through parameters, that is, principles of Universal Grammar allowing for different values depending on the language (Chomsky 1981). An example of a parameter conditioning argument realization—although not formulated in terms of argument structure primitives—is Snyder’s (1995, 2001, 2012) Compounding Parameter. This parameter encodes the (un)availability of productive root compounding, which in turn determines (dis)allowance of such diverse constructions as resultative constructions and double object constructions (see Section 5.3 and next chapter, respectively). The Compounding Parameter is formulated as an operation available in some languages and not in others. In lexicalist theories, in which argument structure is a property of lexical items, differences in argument structure patterns across languages have often been parameterized in this same way. Rappaport Hovav and Levin (1998) claim that complex resultative constructions in English are possible due to the existence of a lexical rule called Template Augmentation rule. This rule allows the lexical template of an activity verb like bark (as in The dog barked) to be converted into a template integrating the activity event into a change of state event. The new template accounts for the use of the activity verb in a resultative construction (as in The dog barked the chickens awake). Rappaport Hovav and Levin (1998) are not concerned with the fact that complex resultative constructions are not available in all languages. However, Fong and Poulin (1998) observe that these constructions are not possible in French, and argue that this is due to the fact that French does not feature the Template Augmentation rule. Other cases of lexical rules that have been proposed to exist in some languages and not in others, therefore giving rise to variation in argument structure patterns, include Levin and Rapoport’s (1988) Lexical Subordination Rule, Jackendoff’s (1990) GO-Adjunct Rule, Tenny’s (1994) rules of addition of eventive roles to a verb’s aspectual grid (Path + Terminus and Measure), Higginbotham’s (2000, 2009) Telic-pair Formation rule, Zubizarreta and Oh’s (2007) adaptation of Snyder’s (1995, 2001) Compounding Parameter, and Gehrke’s (2008) adaption of Snyder and Lillo-Martin’s (2005) Rule C to the combination of an activity verb and her Predicate Phrase. Since Borer (1984), a standard way of conceiving parametric variation is in terms of variation in the lexical properties of functional categories—the so-called BorerChomsky conjecture (see Baker 2008b). This conception of parameter deviates from that involved in the Pro-drop Parameter (Rizzi 1982) or the Compounding Parameter, for instance. Following Baker (1996a, 2008b), we refer to this latter



PARAMETERS AND ARGUMENT STRUCTURE I

101

notion as macro-parameter, and to the former as micro-parameter (see Section 5.4 for a more fine-grained classification of parameters). Up to the mid-90s a sharp distinction between lexical and functional categories has been assumed and argument structure has been considered a property of the former type of categories. In theories holding such vision of argument structure, the above-mentioned conjecture cannot be implemented, since it involves properties of functional categories. However, in more recent theories argument structure is articulated through functional elements and, therefore, it is possible to work this conjecture out and generate predictions as to the variability in argument realization patterns. In the remainder of this section we present the main theories advocating such articulation of argument structure: Distributed Morphology (Halle and Marantz 1993), the Exo-Skeletal framework (Borer 2005a, 2005b, 2013), and Nanosyntax (Ramchand 2008; Starke 2009). Before considering these approaches we take a look at Hale and Keyser’s (1993, 1998, 2002) theory of l(exical)-syntax, which has undoubtedly inspired the idea that the phenomena pertaining to the domain of argument structure must receive an explanation in the same terms as those phenomena involved in sentential syntax. For Hale and Keyser, argument structure is the syntactic configuration projected by a lexical item. As such, it is constrained by the same abstract principles that underlie the well-formedness of phrases and sentences. Specifically, l-syntactic configurations determining the argument structure of a given lexical item show the head-complement and specifier-head relations familiar from sentential syntax. There are four such configurations, headed by different elements named after the lexical categories of familiar languages like English: (1) Argument structure configurations (Hale and Keyser 1998: 82)

a. V

b. P



h

h spc   h

h   cmp

c. A

d. N

h* spc   h*

h   cmp

h*   h

h

The configurations in (a) through (c) can be combined. For instance, a locative verb like shelve ‘put on a shelf’ emerges as the combination of a verbal l-syntactic configuration (a) taking as complement a prepositional configuration (b): (2) The l-syntactic configuration of the location verb shelve

V

V   P

spc   P P   SHELF

102

CONTEMPORARY LINGUISTIC PARAMETERS

Since the resulting configuration is highly abstract an operation called conflation takes the phonological exponent of the element embedded—in this case, the cmp of the prepositional configuration, the nominal SHELF—and brings it up to the upper head through the P head to yield the actual shelve. The Borer-Chomsky conjecture can be applied to l-syntactic configurations by locating parametric variation in the heads that articulate them. For instance, Mateu (2002) and Mateu and Rigau (2002) recast Talmy’s well-known distinction between satellite-framed languages and verb-framed languages in l-syntactic terms. In particular, this distinction is posited to stem from a difference in the preposition involved in predicates of change of location, the Path. Zubizarreta and Oh (2007) also adopt Hale and Keyser’s l-syntactic theory, but, as pointed out above, these authors explain the satellite- / verb-framed distinction through a version of Snyder’s (1995, 2001) Compounding Parameter, rather than through the lexical properties of some particular head. Other authors have posited cross-linguistic differences in the implementation of conflation. Lin (2001) proposes a Lexicalization Parameter by which languages differ in the degree of conflation –understood by him as a lexical operation– displayed by the verbs: full conflation (English), partial conflation (Japanese) or non-existent conflation (Chinese). This parameter explains the wider availability of unselected arguments in Chinese vs English.1 Huang (2006) adopts this perspective in order to explain differences in resultative constructions in Chinese vs English: while the former allows unergative resultatives as John ran tired, which defy Simpson’s (1983) and Levin and Rappaport Hovav’s (1995) Direct Object Restriction, English does not. Other researchers (Marantz 1997, 2013; Borer 1994, 2005b; Harley 1995, 2005; Pylkkänen 2008; Ramchand 2008, and others) have advocated the view that argument structure is sustained by functional structure, going beyond the Halekeyserian distinction between lexical and sentential syntax. This line of research has been labelled constructivist (Ramchand 2013), since it involves the assumption that the syntactic and semantic properties attributed to argument structure are properties of the construction itself, i.e. do not reside in a lexical item. A distinction is commonly made, though, between neo-constructionism, represented by these theories, and constructionism as represented by Construction Grammar (Fillmore and Kay 1999; Goldberg 1995, 2006): while the latter takes constructions as lexical primitives, the former conceives constructions as being built up through functional elements. With respect to variation in the expression of motion events and resultative constructions, constructionist approaches propose that this variation boils down to the existence or non-existence of a given construction in the lexicon of a given language (cf. Pedersen 2008 for the satellite- / verb-framed distinction) or the capability of inserting specific lexical items in a given construction (cf. Narasimhan 2003 for satellite-framed English vs verb-framed Hindi). By contrast, in neo-constructionist approaches some version of the Borer-Chomsky conjecture is implemented to explain parametric variation. Three different neo-constructionist traditions stand out: Distributed Morphology (Halle and Marantz 1993; Marantz 1997; Harley and Noyer 1999), the Exo-Skeletal framework (Borer 1994, 2005a, 2005b, 2013), and Nanosyntax (Starke 2009; Caha 2009). However, work on cross-linguistic variation has been more prominent in Distributed Morphology and in Nanosyntax. In Distributed Morphology, the



PARAMETERS AND ARGUMENT STRUCTURE I

103

computational component produces hierarchical structures of morphosyntactic feature bundles that are interpreted by the interfaces. These morphosyntactic bundles do not contain phonological information, which is inserted late, at the PF branch of the derivation. Words, as phonological units, are assembled both by syntax and by a series of morphophonological operations that apply also at PF. It is precisely at PF that cross-linguistic variation has been claimed to emerge. Thus, for instance, Harley (2008b) proposes that in the bipartite VP articulating transition events, some languages, and not others, allow the independent lexicalization of the highest verbalizing head. Acedo-Matellán (2010), instead, makes the specific proposal that particular functional heads may be specified as triggering a certain PF operation in some languages and not in others, yielding systematic cross-linguistic differences. The nanosyntactic framework also advocates a neat separation between the syntactic representation and how it is lexicalized. In particular, the assumption is made that lexical items may interpret whole chunks of the syntactic tree, and that it is the (un)availability of a given lexical item or its size what yields cross-linguistic variation (Ramchand 2003, 2008; Folli and Ramchand 2005; Son 2007; Son and Svenonius 2008; Pantcheva 2011; Fábregas 2007; Romeu 2014; see also Folli 2002 for an early matching system along similar lines). Since variation does not depend on the properties of functional features, the concept of parameter, as understood in the light of the Borer-Chomsky conjecture or otherwise, can be claimed to be non-existent in Nanosyntax.

5.2. Cross-linguistic variation in directed motion constructions Probably, the first observation of systematic cross-linguistic differences in the expression of motion events is due to Talmy (1975). Talmy (1975, 1985, 1991, 2000) notes that languages differ in how they lexicalize the components of a motion event. In satellite-framed languages, the path of the motion event is lexicalized (conflated, in Talmy’s 1985 terms) as a non-verbal element associated with the verb that Talmy calls satellite, instantiations of which are Germanic directional particles or Slavic and Latin directional prefixes. In satellite-framed languages the verb can lexicalize a co-event component expressing the manner in which the motion takes place, alongside motion itself. In verb-framed languages, on the other hand, the path, together with the motion component itself, is lexicalized in the verb and any manner co-event has to be expressed as an adjunct. This difference is illustrated in the following English and Spanish examples from Talmy (2000: 223):2 (3) a. The bottle floatedMotion+Manner Co-event outPath b. La botella salióMotion+Path (flotandoManner Co-event) the bottle went.out floating ‘The bottle floated out.’

104

CONTEMPORARY LINGUISTIC PARAMETERS

One of the first refinements of Talmy’s typology was due to Aske (1989), who pointed out that verb-framed languages allowed the expression of the path of motion as a non-verbal element if the path was not a telic secondary predicate, as in (4a) (Aske 1989: 3): (4) a. La botella flotó [hacia la cueva.]Atelic Path the bottle floated towards the cave b. *La botella flotó [a la cueva.]Telic Path the bottle floated to the cave A bunch of works have approached Talmy’s typology of motion events—mostly taking into account Aske’s (1989) qualification—via the adoption of some version of the Borer-Chomsky conjecture on parameters. Klipple (1997) proposes that a VP-internal direction / aspect node may be conflated in prepositional elements in English or Chinese but in the verb in French, explaining the Talmian pattern. Since the hierarchy of categories is universal, according to Klipple, the locus of Talmian variation is morphological: it has to do with the patterns of conflation involving these categories. Mateu (2002) and Mateu and Rigau (2002) follow this approach, although they propose a different structuring of the VP. Adopting an l-syntactic framework (Hale and Keyser 1993, 1998), they propose that the Path head is lexically specified to conflate into a null verbal head only in verb-framed languages (see Spanish cruzó “went across,” in 5). As a result, in satellite-framed languages and not in verb-framed languages, an unergative verb expressing the Manner Co-event, like dance, can be conflated into the null unaccusative verb GO (see 6): (5) Sue cruzó la habitación bailando. Sue went.across the room dancing ‘Sue danced across the room.’ (6) Sue danced across the room. V

V P

V   V   N    P

dance GO Sue P   N

across the room

The parameter is here again located in the properties of some relational head (Path).3 Within the Distributed Morphology framework, Acedo-Matellán (2010) proposes that all cross-linguistic variation must be located in the mapping from the syntactic representation to the morphophonological representation. The Path head in verb-framed languages is specified as forming one and the same morph with the eventive head. This fact is at odds with the eventive head bearing an



PARAMETERS AND ARGUMENT STRUCTURE I

105

adjunct root encoding the Manner co-event. In satellite-framed languages there is no such morphological requirement on the Path. However, in some satellite-framed languages like Latin, Slavic and Ancient Greek, the Path is affixal and must then form a word with the verb. This excludes, Acedo-Matellán (2010, 2012) argues, directed motion constructions with unprefixed verbs and adjectival resultative constructions (see next section). As a result, the class of satellite-framed languages is split in two: weak satellite-framed languages such as Latin and strong satelliteframed languages such as English. In Real-Puigdollers’ (2010, 2013) treatment of the satellite- / verb-framed distinction the parameter is syntactic, although it has also effects at PF. In particular, it consists in the defective / non-defective character of the elements forming the prepositional projection. In satellite-framed languages Path is non-defective and, as such, may lexicalize independently from the verb. In verb-framed languages Path is defective (although Place is not), and can be licensed only by raising to the verb and lexicalizing together with it, which excludes the association of the verb with a co-event root. A further important refinement of Talmy’s typology in the expression of motion events is the acknowledgement that some manner-of-motion verbs such as Italian correre “run” can head directed motion predicates with locative PPs in verb-framed languages, as pointed out by Fong and Poulin (1998), Folli (2002), Horrocks (2004), Folli and Ramchand (2005), Zubizarreta and Oh (2007), Real-Puigdollers (2010, 2013), and others. The following is an Italian example from Folli and Harley (2006: 135):4 (7) Gianni è corso nel bosco in un minuto. Gianni is run in.the woods in one minute ‘Gianni has run into the woods in one minute.’ The existence of predicates such as the one above in verb-framed languages like Romance has been leveled as a counterargument to Talmy’s typology and thus, to (parametric) approaches to this typology, by Folli (2002), Folli and Ramchand (2005), Masini (2005, 2006), Iacobini and Masini (2007), Folli (2008), and Iacobini (2009), among others. However, as pointed out by Mateu and Rigau (2010), and Mateu (2012), motion predicates like the one in (7) or those involving particles in Romance (as Italian saltare dentro “jump in”) are possible only with verbs that already involve directionality, the particle or directional PP further specifying the verbally encoded path. Similarly, Real-Puigdollers (2010) proposes that a verbal root like the one involved in Italian correre “run” is able to fit in the light verbal head into which the defective path head has incorporated, as expected in a verb-framed language. Other roots encoding pure manner, and not “forward motion” in the sense of Nichols (2008b) (cited by den Dikken 2010: 47), such as that in Italian ballare “dance,” are impossible for all speakers in directed motion predicates with locative PPs. Moreover, the sentences like (7) lose force as counterexamples to the Talmian typology when the fact is taken into account that verbs like run seem to function in a similar way in satellite-framed languages, as observed, for Germanic, by Koopman (2000), Thomas (2001, 2003), den Dikken (2010), Tungseth (2006, 2008), Nikitina (2008), Gehrke (2008), Real-Puigdollers (2010), and others.5 Thus, for instance, in the next Norwegian example from Tungseth (2008: 29), involving a non-directional

106

CONTEMPORARY LINGUISTIC PARAMETERS

PP, motion can be interpreted as directional, the field being the final location of the bottle (the locative interpretation of the PP is also possible): (8) Flasken har rullet i åkeren. the.bottle has rolled in the.field ‘The bottle has rolled into the field.’ Acedo-Matellán (2010) argues, in a similar vein, that in Latin verbs like curro “run” can head accomplishment predicates in the absence of a prepositional prefix, otherwise a requirement for this aktionsart to emerge in manner-of-motion predicates in this language (Suet. Otho 8, 2; apud Acedo-Matellán 2010: 98): (9) Ac repente omnes in Palatium cucurrerunt. and suddenly all.NOM.PL in Palace.ACC run.PRFV.3PL ‘Then on a sudden everybody hastened into the Palace.’ Up to now we have revised theories, inspired by Talmy’s (1991, 2000) works, assuming that the cross-linguistic variation in the expression of motion constructions involves two opposing conflation patterns: path-conflation and co-event-conflation. However, there is a different strand of research stemming from Snyder’s (1995 and subsequent work) work on the cross-linguistic allowance of the formation of complex predicates. Snyder (1995, 2001) observes a correlation, both across languages and also in terms of the process of language acquisition, between the availability of productive root compounding (such as frog man “a man related to frogs”) and the availability of productive formation of complex predicates such as particle verbs (e.g. Pick up the book) or adjectival resultative constructions (e.g. Hammer the metal flat). To capture the correlation he formulates the Compounding Parameter, which states that languages differ in whether they productively allow or not endocentric compounds during the syntactic derivation (Snyder 2001: 335). Importantly, the Compounding Parameter and the parameterizable semantic rules proposed in Snyder and Lillo-Martin (2005), Snyder (2012), and Beck and Snyder (2001) are an example of a macro-parameter, since they do not involve properties of specific lexical items. In Snyder’s (1995) seminal dissertation, little attention is paid to events of motion, although it is already suggested that the Compounding Parameter could be extended to cover the variability found in this domain. In particular, Snyder (1995: 57 and ff) proposes that in languages with a positive setting for the Compounding Parameter it is possible to form a compound, in some level of linguistic representation feeding the semantic interface, with a manner-of-motion verb like walk and a bounded path preposition like to. The result of this compounding is a complex predicate interpreted as an accomplishment (e.g. John walked to the summit). Snyder (2001: 336) points out that the availability of productive root compounding is a necessary but not sufficient condition for the availability of complex predicates interpreted as accomplishments. Thus, a language like Basque allows nominal compounds but disallows such complex predicates. It is in Beck and Snyder (2001), Snyder and Lillo-Martin (2005), and Snyder (2012), all of them works in which motion predicates are tackled specifically, that this qualification of Snyder (1995) is formally



PARAMETERS AND ARGUMENT STRUCTURE I

107

expressed. Thus, in Beck and Snyder (2001) a language allows the accomplishment manner-of-motion predicates if it is positively set both for productive compounding and for some Principle R, which allows the semantic integration of the elements in a complex predicate. Basque, therefore, is argued to meet the former requirement but not the latter.6 In Snyder and Lillo-Martin (2005), in which Talmy’s typology is referred to, this idea is reformulated as the “the revised Compounding Parameter,” stating that Rule C (roughly equivalent to the mentioned Principle R) {is / is not} available at the syntax / semantics interface. Finally, in Snyder (2012) the relevant rule of semantic composition is called Generalized Modification. In this work, however, Snyder acknowledges that certain factors can obscure the effects of the positive setting for the availability of the semantic rule. For instance, Japanese is presented as featuring the rule but resembling more Spanish than English as far as the expression of directed motion is concerned. The reason is that Japanese lacks incremental P in the sense of Gehrke (2008) and disallows small-clause complements to v. It is evident, therefore, that a certain amount of micro-parametric variation is allowed, moving away from the macro-parametric desiderata stated in Snyder (1995, 2001).7 An application of Snyder’s (1995 and subsequent work) parameter is carried out by Gehrke (2008), who, although using the satellite- / verb-framed terminology, does follow Snyder in considering that the relevant factor in the typology of the expression of motion events is whether a language allows the composition of a manner-of-motion verb and a PP expressing a bounded goal. In particular, Gehrke (2008: 211) adopts Snyder and Lillo-Martin’s (2005) Rule C, and claims that satellite-framed languages allow the compounding of a manner of motion verb with a functional projection called Pred, which “glues together” the verbal predicate and the PP predicate. Interestingly, however, most works inspired by Snyder’s explanation of the cross-linguistic difference in the expression of motion events as based on his Compounding Parameter (and subsequent semantic rules of composition) do not advocate a compounding of the verb with the resultative predicate or the directional preposition. Rather, what is put forth in these works is the compounding of a null light verb and an adjoined lexical verb or root providing the co-event component. McIntyre (2004: 551), for instance, devises the mechanism of m(orphological)conflation in order to deal with the composition of manner and motion in examples such as Ethel danced into the theater. M-conflation involves compounding a root with a causative or a transition verbal head. The root “names an event which is identical to the initiation or change expressed by those heads.” In the same vein, Zubizarreta and Oh (2007) propose that in Germanic languages like English, a null verb interpreted as “go” or “come” can be compounded with a lexical verb expressing a Manner co-event. The lexical verb acts as a modifier of the null directional verb, in the same way that the noun frog modifies the noun man in the compound frog man. The allowance of motion constructions involving a manner verb and a VP-internal directional PP is thus explained (Zubizarreta and Oh 2007: 46): (10) [John [VP [V dance V] [to [P the kitchen]]]] On the other hand, languages of the Romance type do not allow the productive formation of V-V compounds and, therefore, complex directed motion constructions

108

CONTEMPORARY LINGUISTIC PARAMETERS

such as the one shown above are out in these languages.8 Likewise, Mateu (2012: 273) proposes that Snyder’s (2001) parameter is valid if understood as involving the possibility of combining a root encoding the co-event with a null light verb of GO semantics. To work his proposal out, this author adopts Haugen’s (2009) formal distinction between incorporation and conflation (Haugen 2009: 260):9 (11) Incorporation is conceived of as head-movement (as in Baker 1988, Hale and Keyser 1993), and is instantiated through the syntactic operation of Copy, whereas conflation is instantiated directly through Merge. In satellite-framed languages, conflation of the co-event root into the null light verb is possible. In verb-framed languages, by contrast, this conflation mechanism is not found. Instead, motion predicates are headed by verbs formed through incorporation of an abstract path preposition into the verbal head, as in French entrer “go in.” The conflation / incorporation distinction is further exploited to explain why a sub-class of manner-of-motion verbs, like Italian correre “run,” allow directed motion constructions with locative PPs or particles in verb-framed languages, as pointed out above (see also Mateu and Rigau 2010). Despite appearances, these constructions with correre-verbs involve incorporation of the path into the verbal head. The particle (via in the example) or locative PP further specifies this path. Verbs like ballare “dance” do not admit this construction because they do not express forward motion. As we know, both kinds of verbs are possible in directed motion constructions in satellite-framed languages, since they are formed by conflation: (12) a. Gianni è corso via. Gianni is run away ‘Gianni ran away.’ b. *Gianni è ballato via. Gianni is danced away ‘Gianni danced away.’

(cf. Engl. okJohn danced away)

In the next section we show how the correre / ballare contrast illustrated in these examples is related, via Haugen’s (2009) conflation / incorporation distinction, to Washio’s (1997) semantic distinction between weak vs strong adjectival resultatives. More recently, different works have appeared, particularly within the Ramchandian-nanosyntactic framework, that call into question the Talmyan and Snyderian approaches to the variability in the expression of directed motion, both with respect to their empirical predictions and their theoretical foundations. In general, these new approaches propose that the variation does not signal differences in the underlying syntactic structure of directed motion constructions for different languages, which is assumed to be universal, but in the features of the lexical items that realize that structure.10 Folli and Ramchand (2005), for instance, propose the following universal hierarchy of projections for the VP (see also Folli 2002; Butt and Ramchand 2002; Ramchand 2003, 2008):



PARAMETERS AND ARGUMENT STRUCTURE I

109

(13) vP Initiator  v’

v(= Cause) VP Undergoer V’

V (= Process) RP Resultee  R’

R(esult)

XP

(Folli and Ramchand 2005: 83–6)

Although the order of these projections is universal, the presence or absence of certain projections depends on the type of event. For instance, stative events do not feature the V and R projections. In dynamic events with a goal interpretation, however, these two projections must be present. Once the structure is assembled, it must be identified by lexical items endowed with features matching the heads in the configuration. Italian happens not to count with a purely telic goal preposition as English to, so the construction of telic directed motion must rely on the use of verbs that supply the directional goal component, that is, verbs lexically endowed with the feature +R(esult). These verbs are able to identify the R position in the configurations, such as andare “go” or correre “run.” If such a verb is combined with a locative PP, a telic directional interpretation is readily available for the resulting construction (see (7) above). By contrast, a verb such as galleggiare “float,” not endowed with the +R feature, is not insertable in a structure with the RP projection. As a result, R cannot be identified, so this verb cannot license a directional interpretation with locative PPs (Folli and Ramchand 2005: 97): (14) *La barca è galleggiata sotto il ponte. the boat is floated under the bridge ‘The boat floated under the bridge.’ (Directional interpretation intended) In English, motion verbs are not endowed with R, but this head can be identified through prepositions in whose lexical representations there is an optional +R feature. In Folli (2008) the addition is made that Italian complex prepositions like sotto a “under at” can identify a path position and, hence, license even verbs like galleggiare in directed motion constructions (Folli 2008: 214): (15) La barca è galleggiata sotto al ponte in un secondo. the boat is floated under to.the bridge in a second ‘The boat floated under the bridge in a second.’

110

CONTEMPORARY LINGUISTIC PARAMETERS

In a similar vein, Son (2007) criticizes what she calls macro-parametric approaches to the cross-linguistic availability of directed motion and adjectival resultative constructions, including works within both the Talmyan and the Snyderian tradition. Specifically, she argues against the cross-linguistic correlation of these two constructions on the grounds of a double dissociation of both phenomena: languages like Korean and Japanese allow adjectival resultative constructions and disallow directed motion constructions with manner-of-motion verbs while languages like Hebrew and Indonesian show the reverse pattern. To account for these mismatches and, more generally, for the cross-linguistic variation involved, Son (2007), Ramchand (2008), and Son and Svenonius (2008) propose an approach based on the (un)availability of particular lexical pieces to lexicalize certain positions or stretches of the universal structure, as in Folli and Ramchand (2005).11 A particularly fine-grained analysis is presented by Son and Svenonius (2008), who distinguish a Proc(ess) head, a Dir(ection), head, a Path head and a Place head in the universal hierarchy. As shown in the table below, these distinctions allow them to account for the differences in behavior between the pure directional verbs like go, manner verbs encoding forward motion like run or walk and pure manner-ofmotion verbs like dance, in different languages: (16) Proc KOREAN MALAYALAM ENGLISH



Dir

Path

Place

DP Ground

-ey “at”

cip ‘house’

-ekkə “to”

-il “at”

office

to

behind

the curtain

ka “go” naṭann- “walk” dance

∅

(Son and Svenonius 2008: 395)

Languages like Korean, which do not count with independent lexical items for Dir and Path, only allow directed motion constructions with the first type of verb, which lexicalizes the stretch Proc-Dir-Path. Languages like Malayalam admit directed motion constructions with verbs like walk, since they can lexicalize Path independently, but do not admit the same constructions with verbs like dance, since they cannot lexicalize the Dir head independently. Finally, in English all motion verbs may head directed motion constructions, since both Dir and Path are lexicalized independently (the former through a null lexical item).12

5.3. Types of resultative predication In this section we set off from Washio’s (1997) seminal work on the well-known distinction between strong and weak resultative constructions and we relate his difference to Talmy’s typology. According to Washio (1997: 7), strong resultatives are those ones “in which the meaning of the verb and the meaning of the adjective are fully independent of each other”: e.g. the English examples in (17) can be included in this class.



PARAMETERS AND ARGUMENT STRUCTURE I

111

(17) a. John danced his feet sore. b. John hammered the metal flat. In resultatives of this type, it cannot be predicted from the mere semantics of the verb what kind of state the patient comes to be in as the result of the action named by the verb. Washio (1997: 7) gives a negative definition of weak resultatives: “let us call resultatives that are not strong in the above sense weak [his emphasis] resultatives.” Washio’s (1997: 8) main claim is that “natural languages are divided into two broad types, i.e. those (like English) that permit strong resultatives and those (like Japanese) that do not, though weak resultatives are potentially possible in both types of language.” Some examples of weak resultatives taken from Washio (1997: 5) are given in (18), where the meaning of the verb and the meaning of the resultative phrase are clearly related since the latter specifies the result state already encoded by the former.13 (18) a. boku-wa aisu kuriimu-o katikati-ni koorase-ta. I-TOP ice.cream-ACC solid-ni freeze-PST ‘I froze the ice cream hard.’ b. kare-wa teeburu-o kirei-ni hui-ta. he-TOP table-ACC clean-ni wipe-PST ‘He wiped the table clean.’ Mateu (2012) argues that Washio’s (1997) semantic distinction between strong and weak resultatives can be accounted for by using Haugen’s (2009) formal distinction between conflation and incorporation (see (11) above, and subsequent discussion). Applying these two operations to Talmy’s descriptive typology of satellite-framed languages (e.g. Germanic) and verb-framed languages (e.g. Japanese), a distinction emerges between manner / co-event conflation and path / result incorporation.14 Strong resultatives can be provided with the manner conflation analysis exemplified in (19a), whereas weak resultatives can be provided with the result incorporation analysis exemplified in (19b). Washio’s observation that in strong resultatives the meaning of the verb and the meaning of the adjective are fully independent of each other is captured by the fact that there is no structural relation between the root √dance, which is conflated/adjoined to vdo,15 and the S(mall) C(lause) predicate position occupied by the resultative adjective sore in (19a) (e.g. see Hoekstra 1988, 2004, for so-called Small Clause Results). In contrast, weak resultatives like the Japanese one depicted in (19b), where the secondary predicate katikati-ni “hard” can be claimed to specify the result state encoded in the verb, do not involve manner conflation, but rather result incorporation: the result root comes from the inner SC predicate position and is copied into the null light verb.16 In (19b) the resultative adjective katikati “solid” is understood as adjoined to the original inner position occupied by the result predicate koorase “freeze / frozen,” as also argued by Baker (2003: 222). For example, he claims that the two elements [koorase “frozen” and katikati “solid”] in the more complex structure in (19b’) “work together to describe more precisely the resulting state of the event.” (19) a. [vP [DP John] [v’ [v √dance-vdo][SC his feet sore]]] b. [vP [DP boku-wa] [v’ [v koorasei][SC aisu kuriimu-o [√koorasei katikati-ni]]]

112

CONTEMPORARY LINGUISTIC PARAMETERS

b’. vP

DP    v´

boku-wa v V/PredP cause DP V/Pred’

aisu kuriimu-o V/Pr    AP



be

   A    (A) koorase katikati-ni

Washio (1997) concludes his very influential paper on resultatives by pointing out that Japanese and French (and, more generally, Romance) behave alike with respect to those phenomena which fall under Levin and Rapoport’s (1988) “lexical subordination”: for instance, examples like Beth smiled her thanks are impossible in both languages. Washio adds that “it would not be particularly surprising, therefore, if further research tells us that French [and, more generally, Romance: VAM&JM] does in fact share significantly more such abstract properties with Japanese than it does with English” (1988: 43). Following Washio’s (1997) trend, Mateu (2012) puts forward some structural and semantic parallelisms, on the one hand, between Japanese weak resultative constructions and some Italian phrasal verbs, and, on the other, between English and Chinese strong resultative constructions. Washio’s (1997) strong / weak distinction can then be claimed to be valid not only for adjectival resultatives (see 17 and 18) but also for prepositional-like resultatives, verb-particle constructions included: e.g. cf. (20) and (21). (20) Weak P-verb constructions a. Luca ha lavato via la macchia. Luca has washed away the stain ‘Luca washed the stain away.’ b. Gianni ha raschiato via la vernice. Gianni has scraped away the paint ‘Gianni scraped the paint away.’

[Italian]

(21) Strong P-verb constructions a. John worked his debts off. b. John danced the night away. The strong P-verb pattern in (21), which does not involve any relation between the meaning of the activity verb (work; dance) and the meaning of the resultative particle (off / away), should be distinguished from the weak P-pattern in (20), where the meaning of the verb and the meaning of the particle are related: for example, Masini (2005: 167) claims that the existence of Italian phrasal verbs like lavare via



PARAMETERS AND ARGUMENT STRUCTURE I

113

“wash away” or raschiare via “scrape away” depends on the removal sense of the verb, which Mateu and Rigau (2010) argue is related to the incorporating status of path / result.17 The P-verb construction in (21a) is analyzed in (22a), whose strong character is captured by the fact that there is no structural relation between the root √work, which is conflated / adjoined to vdo, and the SC predicate position occupied by the resultative particle off. In contrast, the P-verb construction in (20a) is analyzed in (22b), whose weak character is captured by the fact that the resultative particle via “away” can be argued to specify the path / result meaning of the verb. Unlike the manner root in (22a), the result root in (22b) comes from the inner SC predicate position and is copied (i.e. incorporated) into the null light verb. (22) a. [vP [DP John] [v’ [v √work-vdo][SC his debts off]]] b. [vP [DP Luca] [v’ [v lavarei][SC la macchia [√lavai via]]] Moreover, it is often noted in the literature that Italian also seems to be somewhat exceptional within Romance languages, in the sense that apparent resultative constructions like (23) are well-formed (see Napoli 1989 and Folli and Ramchand 2005, among others). However, the AP in (23) does not seem to occupy the inner argumental SC predicate position; rather, it is an adjunct. The fact that the apparent resultative AP must be reduplicated gives an adverbial/quantificational flavor that is absent from Germanic resultatives like Maria hammered the metal flat, whereby it seems that we are dealing with two different constructions. Given the contrast in (24), it seems natural to conclude that (23) is a kind of pseudo-resultative construction (see Washio 1997, Mateu 2002 and Levinson 2010, among others, for discussion regarding other cases of “spurious” resultatives like John cut the meat {thin / thick}): (23) Maria ha martellato il metallo piatto *(piatto). Maria has hammered the metal flat flat

[Italian]

(24) a. How did Maria hammer the metal? *Flat. b. Come ha martellato il metallo Maria? Piatto piatto. how has hammered the metal Maria flat flat Mateu (2012) argues that the relevant generalization to be drawn from the Japanese and Italian facts revisited above is that these two languages lack the strong resultative pattern that is found in English and, more generally, in Germanic. Such a generalization is indeed important and nicely squares with Talmy’s (1991, 2000) descriptive typological proposal that verb-framed languages like Italian (and, more generally, Romance) and Japanese lack the co-event conflation pattern that can be found in satellite-framed languages like English or Chinese. On the other hand, an interesting fact that is also nicely predicted by Talmy’s (1991, 2000) typology is that Japanese precisely lacks the resultative V-V compounds that can be found in Chinese: e.g. see the paradigmatic contrast in (25a,b), taken from Nishiyama (1998: 209) (cf. also Sybesma 1999 and Huang 2006, among others):

114

CONTEMPORARY LINGUISTIC PARAMETERS

(25) a. Lisi ba shoujuan ku-shi-le. [Chinese] Lisi ba handkerchief cry-wet-le ‘Lisi cried the handkerchief wet.’ b. *John-ga hankati-o naki-nure-ta. [Japanese] John-nom handkerchief-acc cry-wet-pst Mateu (2012) argues that Chinese V-V compounds like the one exemplified in (25a) involves the same manner conflation process that has also been argued for the English strong resultative construction in (17a) John danced his feet sore, i.e. the one that exemplifies the so-called unselected object pattern: see (26), where word order details have been omitted for the sake of clarity. (26) [vP [DP Lisi] [v’ [v √ku-vdo] [SC shoujuan shi]]] In (26) the result / path can be claimed to be encoded in the subordinate/complement V (cf. the SC), while the root encoding manner √ku can be claimed to be conflated/ compounded with the main null light verb. See also Sybesma (1999), Huang (2006) and Fan (2014), among others, for further discussion. In contrast, Japanese resultative V-V compounds like (27a), taken from Nishiyama (1998: 194), are analyzed differently: in a verb-framed and head final language like Japanese, result / path is incorporated into the main null verb,18 while the subordinate verb (nomi “drink” in 27a) turns out to be left-adjoined to that main verb. Crucially, the subordinate / adjoined verb (nomi) is not compounded with a null verb but with a full one (tubusi “use up”), whereby conflation (i.e. compounding of a root with a null light verb; see Haugen 2009) is not involved. In contrast, the English unselected object construction in (27b) does involve conflation: cf. (28a, 28b). (27) a. John-wa zaisan-o nomi-tubusi-ta. John-TOP fortune-ACC drink-use.up-PST b. John drank his fortune away.

[Japanese]

(28) a. [vP [DP John-wa] [v’ [v nomi [v tubusii]][SC zaisan-o [√tubusii]]]] b. [vP [DP John] [v’ [v √drink-vdo] [SC his fortune away]]] Mateu (2012) argues that Japanese weak resultatives, Italian phrasal verbs, and Japanese resultative V-V compounds fall under Talmy’s (1991, 2000) pattern that involves incorporation of path / result into the verb, whereas English strong resultatives and Chinese resultative V-V compounds fall under his co-event / manner conflation pattern: i.e. the one that involves conflation of a root with a null light verb of {motion / causation}. However, as shown in the previous section, Son (2007) and Son and Svenonius (2008) have criticized what they call macro-parametric approaches like the ones that derive from the Talmyan typological perspective. In particular, when dealing with resultative constructions, they capture the observed variation by positing two functional heads between the process component (always lexicalized by a verb) and the state component (lexicalized by the adjective): they distinguish a Proc(ess) head, a Res(ult) head, a Pred(icative) head and an Adj head in the universal hierarchy. For



PARAMETERS AND ARGUMENT STRUCTURE I

115

example, as shown in the table in (29), these distinctions allow them to account for the differences in resultative formation between light verbs like Spanish hacer “make,” result verbs like Japanese nobasu “spread” and process verbs like Korean twutul “pound” or English pound: (29) Proc SPANISH JAPANESE



Res

Pred

Adj peligrosa “dangerous”

hacer “make” nobasu- “spread”

-ku

usu “thin”

KOREAN

twutul “pound”

-key

yalp “thin”

ENGLISH

pound

∅

flat

(Son and Svenonius 2008: 394)

Accordingly, the difference between a restrictive language like Spanish and a less restrictive one like Japanese is that in Japanese there are morphemes available to license Pred (cf. the affixes -ni and -ku on the adjective). Therefore their prediction is that any verb that licenses Res can be used in Japanese to create a resultative construction—Son and Svenonius (2008) assume that verbs that specify a result state carry a feature reflecting this fact, i.e. Res, cross-linguistically (cf. Ramchand 2008, for further discussion). In Korean, there is a morpheme that can lexicalize both Res and Pred (-key). This allows Korean to form resultative constructions involving verbs that do not independently lexicalize Res. As for English, they assume that it has a null counterpart to Korean -key.19

5.4. Directed motion and resultativity: Diachronic variation Parametric theories of variation have been fruitfully applied to explain diachronic variation (Lightfoot 1991, 2006, Roberts and Roussou 2003, Roberts 2007, Biberauer and Roberts 2012, and others). In a nutshell, what is proposed in these theories is that syntactic diachronic change involves a change in the setting of some parameter in a new generation of language acquirers, due to the possibility that the Primary Linguistic Data (PLD) be analyzed in more than one way compatible with the rest of the evidence. It is a fact that languages change with respect to how they express complex resultativity, be it that involved in motion events or in change of state events. Latin, at least in its early and classical stages, was a satellite-framed language, unlike its daughter languages, and, as such, allowed the result part of accomplishment events to be expressed as a non-verbal element, a prefix (Talmy 1991, 2000).20 The reverse path can be illustrated through Chinese. Thus, Huang (2010) shows

116

CONTEMPORARY LINGUISTIC PARAMETERS

that Classical Chinese had a range of path-conflated verbs such as ru “go in” and chu “go out,” while the modern language expresses the same eventualities with a light verb and a spatial satellite: jin-lai “in-come,” chu-qu “out-go.” These two cases, Latin to Romance and Old to Modern Chinese, have received attention in the literature. As has been pointed out, Talmy (1991, 2000) considers Latin a satellite-framed language, since in this language the path component of motion events is preferably expressed as a verbal prefix (Talmy 2000: 104): (30) Avis in-volavit. bird.nom in-flew ‘A bird flew in.’ Acedo-Matellán (2010) provides an in-depth account of the satellite-framed nature of Latin, as reflected in a variety of constructions that include unselected object constructions such as those described for Germanic languages by Hoekstra (1988) and Mateu (2001). In Latin these constructions feature, indeed, verbal prefixes, which Acedo-Matellán (2010) considers result-encoders, in the spirit of what Arsenijević (2006) and Gehrke (2008) claim for Slavic prefixes. The verbal root is shown to encode the manner co-event, as expected. Acedo-Matellán (2010) and Acedo-Matellán and Mateu (2013) propose that the functional head encoding directionality, Path, was prefixal in Latin. In Romance, by contrast, it is specified as forming one and the same node with v, which precludes the association of v with any co-event encoding root adjoined to it. A nice illustration of this morphophonological process is provided by the historical transformation of the Latin verb exire “go out” in Catalan: (31) ex-i-re > eix-ir out-go-inf > go.out-inf In Latin the elements encoding movement (ire “go”) and path (ex- “out”) are separate morphs. In Catalan they have fused into the same morph (eix- /eʃ/ “go out”). Bartra and Mateu (2005) and Acedo-Matellán and Mateu (2013) show evidence from Old Catalan that verbal prefixes did, to some extent, survive in the earliest stages of the Romance system as path-encoders, in the spirit of work done by Dufresne, Dupuis, and Tremblay (2000), Dufresne, Dupuis, and Longtin (2001), and Dufresne, Dupuis, and Tremblay (2003) on the use of verbal prefixes and verbal particles in Old French.21 Chinese seems to have followed the reversed path, becoming a satellite-framed language in its modern instantiation. Peyraube (2006) traces the history of Chinese from its verb-framed status in the Archaic and Classical periods to its present-day satellite-framed status, allegedly attained around the tenth century.22 Huang (2010) is, in fact, the only work that we have found in which that change is presented as involving a parameter. This author argues that the distinction can be handled through his analytic / synthetic parameter: Old Chinese is a synthetic language, while modern Chinese is an analytic language. This means that in the latter the relation between the morphemes and their overt expression comes closer to a



PARAMETERS AND ARGUMENT STRUCTURE I

117

one-to-one relation. With respect to the specific domain of the expression of resultativity, Huang observes that the result is encoded independently from the verb in the modern language, and not in the old language. Interestingly, considerations on the diachronic change in the kind of predicates dealt with in this chapter may shed light on the nature of the parameter underlying its variation. Biberauer and Roberts (2012) classify parameters in macro-, meso- and micro-parameters, according to the elements affected by the parameter. A macroparameter, such as the one determining whether heads are initial or final, affects all relevant categories. A mesoparameter affects individual syntactic categories: the Null Subject parameter is a case in point, since it involves properties of T. Finally, a micro-parameter affects groups of lexical items. An example thereof is the parameter involved in T-to-C inversion in English conditionals, available for only a group of modal auxiliaries. Crucially, the cline in the size of the class of elements affected by the parameter seems to correlate with a cline in the average lifetime of their positive or negative setting. Mesoparameters are located in the middle of the cline with respect to stability, which may be affected by contact, and we have historical attestation for their resetting. Having this into account, the parameter affecting the expression of motion and resultativity could be a mesoparameter in Biberauer and Roberts’ (2012) terms. Thus, satellite-framed-ness is a hallmark of Indo-European, although it has been lost in Romance.23 Similarly, we have evidence for the reverse change of this alleged parameter in the historical record of Chinese. In both cases the change in the setting of the parameter is comparable to that of the Null Subject parameter. That the satellite- / verb-framed distinction or its equivalents is a mesoparameter would in turn fit with what has been proposed in some of the works dealing with this kind of variation, namely, that it depends on properties of a particular functional head: Path (Mateu and Rigau 2002; Acedo-Matellán 2010; Acedo-Matellán and Mateu 2013; Real-Puigdollers 2013). It remains to be seen whether the attested resetting of this parameter in Latin-Romance and Chinese was brought up through contact, as follows from Biberauer and Roberts’ (2012) definition of a mesoparameter.

5.5. Concluding remarks In this chapter we have provided an overview of the cross-linguistic variation involved in the encoding of directed motion and resultativity. In the influential strands of research originating in seminal work by Talmy (1985, 1991, 2000) and Snyder (1995, 2001), these two notions have been related through the satellite- / verb-framed typology in the former and the Compounding Parameter in the latter. By contrast, in recent years such a correlation has been called into question by other researchers (cf. Kratzer 2005, Folli and Ramchand 2005, Son 2007, Son and Svenonius 2008, inter alia). The debate has emerged in part because no full consensus has been reached regarding the relevant kind of direction motion construction and resultative construction as standard of comparison. For example, Son (2007) and Son and Svenonius (2008) have claimed that Japanese allows resultative constructions but not goal of motion constructions with manner

118

CONTEMPORARY LINGUISTIC PARAMETERS

verbs. However, this claim turns out not to be problematic for Talmy’s typology once the qualification is made that Japanese only allows a certain type of resultative construction, namely, Washio’s (1997) weak resultative, disallowing the strong type. It is the following generalizations that we deem relevant for generative approaches to Talmy’s typology: strong adjectival resultative constructions and strong PP resultatives cannot be found in verb-framed languages; these languages allow resultative constructions to the extent that the verb encodes the result, the AP/PP/particle merely specifying this result. Additionally, there has been a non-consistent use of the terms micro- and macroparameter when evaluating the proposals made to account for the cross-linguistic variation that we are dealing with. Specifically, generative approaches to Talmy’s typology have been labeled macro-parametric on the grounds of the fact that they describe two big groups of languages: satellite- and verb-framed languages. From the point of view of the explanation itself, however, many of these approaches can be called micro-parametric, since they rely on the variability in the properties of a specific functional element (Path), and not on the mere (un)availability of a certain operation (Snyder’s 2012 Generalized Modification). Smaller factors in the debate have to do with the proper characterization of certain elements found in these constructions. For example, so-called until-markers like Japanese made or Spanish hasta have been claimed to license directed motion constructions with manner of motion verbs in verb-framed languages (Inagaki 2002). Some authors, however, have pointed out that PPs headed by until-markers do not qualify as proper goal PPs but are rather delimiters of the event (Beavers 2008a; Real-Puigdollers 2013).

Notes *

We want to thank Antonio Fábregas and an anonymous reviewer for constructive comments and helpful suggestions. All errors are ours. The first author acknowledges the funding of grant SFRH/BPD/90892/2012 (Fundação para a Ciência e a Tecnologia, Ministério da Educação e Ciência, Portugal). The second author acknowledges the funding of grants FFI2011-23356 (Spanish Ministerio de Economía y Competitividad) and 2014 SGR-1013 (Generalitat de Catalunya).

1

The rigidity of argument structure in English compared to its flexibility in Chinese can be exemplified with the verb drive (examples drawn from Huang 2006: 22):



(i) a. Ta

he

kai-le

yi-liang tankeche.

drive-PRFV one-CL tank

‘He drove a tanker.’

b. Ta kai

zuobian, wo kai

he drive left-side, I

drive right-side

‘He drives [on] the left side, I drive [on] the right side.’

c. Ta kai

baitian, wo kai

he drive daylight, I

youbian.

wanshang.

drive night

‘He drives [in] daylight, I drive [at] night.’



PARAMETERS AND ARGUMENT STRUCTURE I



d. Ta kai

jiazhao, wo kai

he drive license, I

shenfenzheng.

drive ID-card

‘He drives [with] a driver’s license, I drive [with] an ID card.’

e. Wo kai I

119

hao wan.

drive good play

‘I drive [for] fun.’

2

Talmy (1985, 2000) also points out the existence of languages, like Atsugewi, lexicalizing the Figure of a motion event (i.e. the element presented as being in motion or stationary) within the verb. Research on this alleged third typological type has not been very abundant, maybe because Talmy (2000: 221ff.) himself privileges the satellite- / verb-framed distinction as the basic cross-linguistic typology. Talmy (1991, 2000) also proposes that static verbs can lexicalize a manner co-event component, as shown by the difference between lie and sit. This pattern falls outside the scope of the present chapter. We are also leaving out the refinement of Talmy’s typology proposed by Slobin (2006 and subsequent work), namely, that besides satellite- and verb-framed languages there exist equipollently framed languages, like Chinese, Nez Perce or Jaminjung, in which “both path and manner receive equal weight” (Slobin 2006: 59). See also Beavers, Levin, and Tham 2010.

3

Mateu (2002) applies this approach to deal with the uneven productivity of the locative alternation across languages. Indeed, it is the case that in satellite-framed languages like English this alternation is licensed by a much wider set of verbs than in verb-framed languages like Romance, as shown by the next contrast between English spray and Catalan ruixar “spray”:



(i) a. Sue sprayed the sheets with perfume.



b. Sue sprayed perfume onto the sheets.

(ii) a. La Sue ruixà



(Change of location alternant) (Change of state alternant)

the Sue sprayed the sheets of perfume

b. *La Sue ruixà



els llençols de perfum.

(Change of state alternant)

perfum

the Sue sprayed perfume

sobre els llençols. (Change of location alternant) on the sheets



Mateu shows that the change of location alternant, which is out for verbs like spray in Catalan (see iib), actually involves a satellite-framed pattern: the direction of movement is encoded by the preposition (onto in (ib)), while the verb encodes the co-event (sprayed). This pattern is not possible in verb-framed Catalan.

4

This example involves an unaccusative structure, where Gianni is not an external argument. Although verbs like correre “run” or volare “fly” select avere “have” in the unergative structure (e.g. Gianni ha/*è corso ‘Gianni has/*is run’), they select essere “be” in the unaccusative one, e.g. in the one containing the obligatory PP nel bosco “in the woods.”

5

As noted in Higginbotham (2009: 117), Talmy (1985; revised and expanded in 2000 and 2007) already observed that while (i) is ambiguous between a locational and a directional interpretation, the Italian equivalent of (ii) only licenses the former interpretation:



(i) The boat floated under the bridge.



(ii) La barca galleggia sotto il ponte.



However, Talmy (1985) had not observed the directional interpretation licensed by verbs of the correre type (see above). To deal with contrasts such as this one, Sugisaki (2003: 55) proposes a syntactic approach in parametric terms:

120

CONTEMPORARY LINGUISTIC PARAMETERS



(iii) “PP is (not) dominated by pP”



Languages like English are positively set for this parameter. As a result, their locative PPs possess an additional layer, headed by the functional preposition p, licensing the goal-of-motion interpretation. French is negatively set for the parameter, so its locative PPs do not license the goal-of-motion interpretation.

6

Later, in Snyder (2012) Basque is not considered to allow root compounding. Rather, his most recent proposal is that Basque compounds involve some functional structure.

7

Similar in spirit to Beck and Snyder’s (2001), Snyder and Lillo-Martin’s (2005), and Snyder’s (2012) semantic rules is the telic-pair formation rule proposed by Higginbotham (2000, 2009). Languages endowed with this semantic rule, such as English or Chinese, can form an accomplishment predicate out of a manner-of-motion verb and a PP.

8

Zubizarreta and Oh (2007: 40) discuss a third type of languages as far as the composition of manner and motion is concerned, namely, those languages like Korean using serial verb constructions:



(i) John-i

hoswu

John-nom lake

hanccok-phyen-ey heyemchi-e ka-ss-ta. one_side-side-loc swim-l go-pst-decl

‘John swam to one side of the lake.’



In this example there appear two verbs, heyemchie and kassta: the former encodes the co-event and the latter, which is the main verb bearing inflection, encodes directionality. However, as far as directionality is expressed in the verbal root ka- (see Son and Svenonius 2008), Korean seems to be more akin to Romance than to Germanic, as originally proposed by Talmy (2000: 60).

9

Cf. Acedo-Matellán (2013) for a technical implementation of this distinction.

10 The idea that the variation in the expression of directed motion is due to the (un)availability of particular lexical items to license the underlying structure can be traced back to Inagaki (2002). Although working within the l-syntactic framework of Hale and Keyser’s (1993), this author proposes that the differences between satellite-framed English and verb-framed Japanese lie in the availability of a variety of directional prepositions in the former and their unavailability in Japanese—except for made “up to” (see Beavers 2008a for relevant discussion). 11 See also Fábregas (2007), in which it is argued at length that, despite appearances, the Spanish preposition a is not directional, but locative, which accounts for the fact that it cannot license the directed motion reading with manner-of-motion verbs. See Inagaki (2002) for an early work incorporating this claim on the nature of Spanish a, which is equalled to Japanese -ni and -de. 12 Outside the accounts relying on Talmy’s typology and Snyder’s Compounding Parameter, Horrocks (2004) and Horrocks and Stavrou (2003, 2007) propose an original explanation for the cross-linguistic variation of both directed motion and resultative constructions. They make the empirical claim that languages with a morphological encoding of the perfective/imperfective distinction in verbs, like Modern Greek, do not allow such complex predicates. See these works for details, and AcedoMatellán (2010) for a critique. 13 See also Kaufmann and Wunderlich (1998) and Takamine (2007), among others., for further discussion and eventual refinement of Washio’s (1997) typology of resultatives. 14 It is perhaps not surprising that German freely allows the conflation pattern but shows some restrictions concerning weak AP resultatives like The butter melted soft;



PARAMETERS AND ARGUMENT STRUCTURE I

121

in contrast, PP resultative constructions like The butter melted into a pool can be expected to be well-formed in German since the Path into remains as satellite (i.e. it is not incorporated into the verb). See Kaufmann and Wunderlich (1998: 20–2) and McIntyre (2004: 554), for further discussion on some important differences between German and English resultatives. 15 Marantz (2013) assumes that the causal relation (cause) always arises structurally, i.e. it emerges from merging a dynamic v (do) with a Small Clause Result. See also Hoekstra (1988) and Zubizarreta and Oh (2007), among others, for similar remarks. 16 Haugen’s (2009) distinction between conflation and incorporation can be recast in minimalist terms: the former operation can be reduced to External Merge and the latter to Internal Merge. For example, in (19a) the root √dance is externally merged with the light verb, whereas in (19b) the root √koorase- (“freeze”) is internally merged. 17 Despite claims to the contrary, verb-particle constructions are not a quirk of Italian but can also be found in other Romance languages (e.g. see Iacobini 2009). For instance, Mateu and Rigau (2010) show that many verb-particle constructions from Dante’s dialect (see Masini 2006: 87–99) can also be found in Catalan and Spanish. This said, it is true that Italian and other languages such as Venetan and Friulan can indeed be considered exceptional among other Romance languages since they have developed a pattern where the verb is not a motion verb (e.g. see the examples in (20), which are not found in Dante’s dialect; see Masini 2006). This notwithstanding, Mateu and Rigau (2010) argue that this innovative pattern is allowed in Italian (and other languages such as Venetan and Friulan) as long as the verbal basis involves an abstract path/result component (cf. also Folli and Ramchand 2005). 18 See Nishiyama (1998: 184), for some arguments that make it clear that the main verb in Japanese V-V compounds is the second one. 19 In reaction to Talmy’s (2000) classification of Korean within verb-framed languages, Son (2007) argues that this language does have complex resultative constructions. However, more research seems to be needed here. The following remarks by Shim and Den Dikken (2007: 21) go against Son’s (2007) and Son and Svenonius’ (2008) claims regarding Korean resultatives and are in fact compatible with Talmy’s typological classification: “Like the Romance languages, Korean does not make English-type bare resultatives by projecting a result state in the complement [our emphasis: VAM&JM] of a lexical verb with a manner component in its semantics. But Korean manages to mimic English quite closely on the surface (though not when one looks beneath the surface) thanks to the fact that it can project a local T outside its result-state-denoting small clause and attach the TP to a projection of the matrix verb as an adjunct [our emphasis: VAM&JM].” 20 Greek seems to have also changed from a satellite- to a verb-framed language. See Horrocks (2004), Horrocks and Stavrou (2007) and Acedo-Matellán (2010) for relevant discussion. 21 Other evidence of the status of Old French as an intermediate stage between the Latin satellite-framed system and the contemporary French verb-framed system is presented by Troberg (2011), on directed motion constructions with manner verbs and locative PPs, and Troberg and Burnett (2014), on weak adjectival resultative constructions. 22 There has actually been a debate in non-generative literature whether (modern) Chinese should be classified as satellite-framed, as claimed by Talmy (2000), Peyraube (2006) and Xu (2006). Tai (2003) argues that it should be considered primarily a

122

CONTEMPORARY LINGUISTIC PARAMETERS

verb-framed language and Slobin (2004, 2006) and Chen and Guo (2009) propose that it is an equipollently framed language, in the sense of Slobin (2006). 23 See Verkerk (2014) for the position that Proto-Indo-European was not satellite-framed, but “had a mixed motion event encoding system” (Verkerk 2014: 40). For reasons of space, we cannot review her work here.

6 Parameters and argument structure II: Causatives and applicatives María Cristina Cuervo University of Toronto

6.1. Introduction This chapter deals with variation in applicative and causative constructions. These two constructions are viewed from diverse theoretical perspectives as the canonical cases of operations that augment the argument-licensing capabilities of a predicate, causatives typically adding a subject, applicatives adding an object. These constructions have been studied in their syntactic properties, their lexical, compositional and discursive semantics and, to a lesser extent, in their morphology. There has also been rich typological and descriptive studies (Dixon 2000; Haspelmath 1993; Peterson 2007, among others). The last decades have shown that cross-linguistic variation in applicatives and causatives is found at all relevant levels, from their availability (e.g. some languages are claimed not to have applicative constructions) to the syntactic coding of arguments (e.g. case, relative hierarchy, linear order, optionality of expression), from their possible interpretation to the morphophonological shape of heads. Additionally, the study of variation in these areas has involved the analysis of intralinguistic argument structure alternations, in particular the dative alternation and the causative alternation, considered by many as operations of argument addition or suppression. Some constructions and argument structure alternations have been part of parameters stated within the Principles and Parameters framework (e.g. double object / applicative constructions figured within the Compounding Parameter of Snyder 1995). There have also been accounts of variation in terms of the denotation and / or formal properties of the causative and applicative heads, including selection and bundling of features (Cuervo 2003; Pylkkänen 2008; Travis 2000; Tubino

124

CONTEMPORARY LINGUISTIC PARAMETERS

Blanco 2010, and others). Other analyses place the burden of the explanation on structural properties or operations (Marantz 1993; Baker 1996a; McGinnis 2001, 2008). It is not clear, however, whether these analyses of variation can be expressed as parameters, or to what extent they are able to account for the diverse levels of inter and intralinguistic variation. In fact, there is not even full consensus on what a causative construction is, or how to decide whether a given verb with a transitive and an intransitive variant is an instance of the causative alternation or some other kind of transitivity alternation. In the domain of applicatives, typologies based on explicit theoretical frameworks have been developed in the last two decades. No definition has been accepted across diverse approaches, however, that could serve as a base to bring together results from studies carried out within diverse approaches to linguistic work. Given this state of affairs, this chapter deals first with a description of variation in each area of applicatives and causatives, including descriptive generalizations, and systematicity and correlations between properties in the various domains are highlighted. The viability of various parametric approaches to cross-linguistic variation is discussed next in light of previous accounts of variation in the two areas.

6.2. Applicatives Applicative constructions involve the licensing of an extra or “non-core” argument into a clause with the morphosyntactic coding of a core argument. Applicative constructions are typically signalled by the addition of a morpheme on the verb. The examples in (1) illustrate Chichewa instrument and benefactive applicatives, respectively. (1) a. Mavuto a-na-um-ir-a mpeni mtsuko [Chichewa] Mavuto SM-PST-mold-APPL-FV knife waterpot ‘Mavuto molded the waterpot with a knife’   [apud Baker 1988] b. Chitsiru chi-na-gul-ir-a atsikana mphatso fool SP-PST-buy-APPL-FV girls gift ‘The fool bought a gift for the girls’  (apud Alsina and Mchombo 1993) This apparently innocent description hides a lot of important controversies and differences of analyses that have illuminated (and sometimes obscure) work on argument structure and on the syntax of numerous languages. As a general approach to applicative constructions, I follow Marantz (1993) in considering that all double-object constructions are applicative constructions.1 With Pylkkänen (2002, 2008), I consider, however, that not all applicative constructions are doubleobject constructions. Pylkkänen’s distinction between high applicatives (merged above the verb) and low applicatives (merged below the verb) based on structural and semantic properties crucially extended the empirical coverage of the notion of applicative, incorporating possessor raising constructions and applicatives in (Japanese) adversity passives. Cuervo (2003, 2010) expanded the typology by



PARAMETERS AND ARGUMENT STRUCTURE II

125

including ethical datives (see also Rivero 2003), dative experiencers and accidental causers with various types of unaccusative predicates, as well as affected applicatives (middle applicatives, merged between two verbal layers) in causative and inchoative constructions. The resulting typology, clearly defined in structural and general semantic terms, but flexible in terms of morphosyntactic coding (word order, agreement and case marking, morphological shape of the applicative head, etc.)2 opened the door to fruitful studies of other Romance languages as well as Slavic, Balkan and Germanic languages, Greek, Arabic, and Basque within this applicative approach (Anagnostopoulou 2005; Arregi and Nevins 2008; Diaconescu and Rivero 2007; Fournier 2010; Kallulli 2006; Marvin 2012; Rivero 2003, 2009; Slavkov 2008; Schäfer 2008, and others). The typology has been recently further extended to cover Peripheral Applicatives in Korean causatives, defined as applicatives merged above Voice (Kim 2012). In order to inquiry on the suitability of parameters as an account of crosslinguistic variation in applicative constructions, a first step consists in laying out the areas in which variation is found. The first contrast is simply that some languages seem not to have applicative constructions while others exhibit a wide array of applied arguments. Focusing from now on on languages that have been claimed to allow applicatives, the next subsections highlight the variation found in the semantics, the morphology and the syntax of applicatives.

6.2.1. Possible meanings of the applied argument Perhaps the most famous applied arguments are Bantu benefactive / recipients and instrumentals, as illustrated in (1). The range of possible meanings is, however, much wider. Applicatives have been identified as expressing locatives, as well as malefactives (i.e. adversely affected arguments) and comitative / associative arguments (2). An applicative argument can also be interpreted as expressing a circumstance, reason or purpose. In addition to these, Peterson’s (2007) analysis of Hakha Lai identifies much less common types, such as prioritive, and relinquitive applicatives. (2) a. n-a-i-lyi-i-a m-ri-nyi k-elya [Kichaga] FOC-1SG-PRES-eat-APPL-FV 3-homestead-LOC 7-food ‘He is eating food at the homestead’    (Bresnan and Moshi 1990: 148) b. rul=niʔ ka-ʔin=ʔaʔ ʔa-ka-luʔ-hnoʔ [Hakha Lai] snake=ERG 1SG.POSS-house=ALL/LOC 3SGS-1SGO-enter2-MAL ‘A snake came into my house on me’ (Peterson 2007: 41) c. ka-law ʔan-ka-thloʔ-pii[Hakha Lai] 1SG.POSS-filed 3PLS-1SGO-weed2-COM ‘They weeded my field (together) with me’ (Peterson 2007: 41) Pylkkänen makes a crucial distinction between benefactives as high applicatives related to an event, and recipients, low applicatives related to an entity / individual. High applicatives add an argument interpreted as the beneficiary / maleficiary of the event, an instrumental or comitative. Low applicatives are interpreted as

126

CONTEMPORARY LINGUISTIC PARAMETERS

recipients of the theme object or as source applicatives, which cover some arguments previously classified as locatives or possessors. Within the same framework, Cuervo (2003) proposes a third type of applicatives, middle or affected applicatives, which appear between two verbal layers in transitive-causative (3a) or intransitive-inchoative change of state (3b) constructions. In these structures an applicative is interpreted as “receiving” the new state of the object. In the latter case, an applicative can alternatively be interpreted as an accidental causer (3b). (3) a. Emilio le rompió la radio a Carolina [Spanish] Emilio CLT.DAT broke the radio Carolina.DAT ‘Emilio broke the radio on Carolina’  (Cuervo 2003: 27) b. A Carolina se le rompió la radio [Spanish] Carolina.DAT se CLT.DAT broke the radio ‘The radio broke on Carolina’ ‘Carolina unintentionally broke the radio’  (Cuervo 2003: 27) Dative experiencers in unaccusative constructions, ethical datives, involuntary state constructions, and causees have also received an applicative analysis (Cuervo 2003; Kim 2012; Roberge and Troberg 2009; Rivero 2003). Have there been attempts to unify applicatives semantically? Marantz (1993) treats them all as affected objects, providing a general definition in terms of entities licensed as the specifier of the applicative head, which are related (affected) either inside or outside “the event that affects the patient/theme” (Marantz 1993: 146). Pylkkänen’s typology unifies applicatives very generally in terms of the nature of the complement the applicative head takes: an entity or an event. This typology is presented as a distinction in denotation, but each semantic type has a direct correlation with the syntactic type of the complement it selects: DP for entities, VP for events. Among low applicatives, which relate an entity to another entity, Pylkkänen distinguishes two subtypes, according to the directionality of the transfer of possession: LowApplTO and LowApplFROM. She does this explicitly, and provides a separate lexical entry for each. The semantics of high applicatives is less specified, although separate lexical entries for beneficiaries, instrumentals, etc. are proposed. Cuervo (2003, 2010) contributes to the typology of applicatives by investigating stative relations in this domain. First, she argues for a third subtype of low applicative, LowApplAT, in which the applied argument is interpreted as the static possessor of the theme DP. This type of low applicative is compatible with stative transitive verbs as well as with dynamic verbs which do not express a transfer of possession, as admire, and wash and lookt at, respectively. (4) a. Pablo le admira la paciencia Pablo CLT.DAT admires the patience.ACC ‘Pablo admires Valeria’s patience’  b. Pablo le lavó el auto Pablo CL.DAT washed the auto.ACC ‘Pablo washed Valeria’s car’

a Valeria Valeria.DAT (Cuervo a Valeria Valeria.DAT (Cuervo

[Spanish] 2003: 71) [Spanish] 2003: 68)



PARAMETERS AND ARGUMENT STRUCTURE II

c. Gil histakel le-Rina et ha-bayit Gil looked-at DAT.Rina ACC the-house ‘Gil looked at Rina’s house’ 

127

[Hebrew] (Landau 1999: ex. 49a)

Additionally, she notes that high applicatives should be distinguished between those which take a dynamic event VP as complement, from those which take a state. Specifically, dative experiencers with psychological (and some non-psychological) predicates are analyzed as high applicative taking a stative vP as complement (e.g. A Daniela le gustan los Rolling Stones ‘Daniela likes the Rolling Stones’). Middle applicatives (Cuervo 2010) also take a stative vP as complement, but are further defined by their being embedded under a dynamic event in causatives and inchoatives/anticausatives, as in (3a-b). Finally, applicatives in anticausative/ inchoative constructions interpreted as accidental causers (3b) correspond to a high applicative which takes the whole complex inchoative as its complement, as opposed to the middle / affected applicative which takes just the result state.3 So far, then, we have seen that applicatives involve some general sense of affectedness as well as possession (of an entity, of a state or situation, of a mental state). The notion of location is also invoked, and possessors and experiencers can be viewed as special cases of locations, usually in combination with animacy and/ or affectedness (e.g. Landau 2010).4 There is also work on applicative constructions which highlights the involuntary, non-volitional nature of high applicatives as non-canonical external arguments distinct from true agents licensed by Voice (see Rivero 2009; Kim 2012, as well as studies of unintentional causers cited above). Can these distinctions be captured as some kind of semantic parameter? Does each type of applicative correspond to a separate lexical entry for the applicative head? Can distinctions be, alternatively, derived as consequences of configurational properties of the structure in which an applicative is inserted? Are there any semantic distinctions that depend on the meaning of lexical elements (e.g. the verbal root) and cannot be reduced to parametric variation in formal properties? These questions are addressed directly in §4, after discussion of variation in causative constructions.

6.2.2. Morphology of the applicative head Applicative heads can be spelled out by a morpheme or they can be phonologically null. This variation is found across languages (English double-objects vs Chaga applicative affix -í-) as well as intralinguistically (Inuktitut Ø- applicative vs the affix -djuti-). Within languages which exhibit a wide range of applicative constructions, some have one unique morpheme (as in Kichaga), while others have more than one applicative morpheme (Kinyarwanda has one applicative form for benefactives and another for instrumentals; Hakha Lai has seven different morphemes, one for each of seven subtypes of applicatives, Peterson 2007). Although most applicative morphemes are affixes invariable in form, it has been proposed that pronominal clitics can be the morphological expression of applicative heads, spelling out the φ-features of the applicative argument. This analysis is found in Cuervo (2003) for Spanish, Arregi and Nevins (2008) for Basque, Roberge and Troberg (2009) for French, among others.

128

CONTEMPORARY LINGUISTIC PARAMETERS

In some languages, the same morphological forms that appear as adpositions can also function as affixes to the verb in constructions analyzed as applicatives. Applicative heads can show syncretism with elements besides adpositions. Passive morphology appears in adversity applicative constructions in Japanese (Pylkkänen 2008). In some languages an applicative morpheme can also mark a causative head. Peterson (2007) presents cases of benefactive / causative isomorphism (e.g. Hualapai, Yuman, Arizona) as well as instrumental or comitative / causative isomorphism (e.g. Shona, Bantu). Jerro (2013) analyzes the instrumental / causative syncretism of Kinyarwanda -ish- on the basis of a common syntax and semantics of adding a causee argument. Some cross-linguistic variation is also found in the position of the applicative morpheme within the verbal complex. Interestingly, variation in morpheme order is also found intralinguistically, as illustrated by Kinyarwanda benefactive -er- which appears as V-Appl-passive-aspect in contrast to the locative -ho applicative which appears as V-aspect-passive-Appl (data in Marantz 1993, from Kimenyi 1980). There is also variation in the relative order of applicative and causative morphemes (see §6.3.1).

6.2.3. Morphosyntactic properties of the applied argument Variation in morphosyntactic properties of applicative constructions has been studied making two kinds of comparisons. One comparison centres on differences in applicative constructions within or across languages. The other line of research has studied morphosyntactic properties of applicative constructions as compared to alternative prepositional or oblique constructions. I discuss below issues of relative hierarchical order and object properties in transitive applicative constructions, compatibility of applicatives with different types of predicates, and variation in availability and obligatoriness of the construction. i) Object properties. The analyses of (transitive) applicative constructions, both of the “high” benefactive type and the “low” double object type, have resulted in the agreement that the applied argument is licensed in a syntactic position higher than the theme argument. Probably one of the most studied points of variation among applicative constructions is the distribution of object properties between the applied argument and the deep/theme object in transitive applicatives. The variation consists on whether (a) both objects exhibit object properties (agreement, passivization, adjacency to verb, structural Case) or (b) only one of the arguments does, dividing applicative constructions into symmetric and asymmetric applicatives, respectively. Both objects exhibit object properties in Chaga, Georgian, Kinyarwanda benefactives, Albanian, and Japanese (Marantz 1993). In contrast, in Chichewa, English, and Swahili only the applied argument exhibits object properties. Several authors provide alternative accounts of this parametric contrast. Baker’s (1988) analysis relies on the interaction between Theta Theory and Case and provides different structures for each case. Marantz (1993) argues that all applicatives have the same initial structure and that the basis of



PARAMETERS AND ARGUMENT STRUCTURE II

129

the contrast is to be found in the way the applicative head and the verb combine in the syntactic derivation, either by adjunction (symmetric) or incorporation (asymmetric). Pylkkänen’s (2002) distinction between high and low applicatives accounts for the fact that those applicatives which combine with unergative verbs are symmetric (high) while those which do not are asymmetric. McGinnis (2001) argues that the high-low distinction coupled with a theory of phases—in which High Appl defines a phase for phonological and semantic interpretation—naturally derives the parametric contrast in object properties, as well as other asymmetries in applicative constructions. These analyses all assume or predict that within asymmetric applicatives only the applied argument can exhibit object properties, the theme object losing the properties it exhibits in monotransitive constructions. This view was challenged by Demonte’s (1995) analysis of Spanish (clitic-doubled) ditransitives as double objects. An applicative analysis of Spanish double objects (Cuervo 2003) explicitly claimed that there exist asymmetric applicatives in which object properties are retained by the theme. Similar asymmetries have been found in other languages, such as Romanian, Bulgarian, and French (Diaconescu and Rivero 2007; Slavkov 2008; Fournier 2010). Within all the variation found in the distribution of object properties in applicatives, there seems to be an object property which always stays with the original theme object: the possibility of incorporation. In ditransitive / applicative constructions of languages with noun incorporation, it is always the direct object, never the indirect object or applicative which incorporates into the verb; an incorporated noun is never interpreted as a goal or benefactive (Baker 1988). This fact is problematic if the applicative in the incorporated construction is identified as low, under the view that in low applicatives the applied argument is always higher than the theme, therefore closer to the root (see Carrier 2014 for discussion).5 ii) Base predicates. Applicative constructions have also been contrasted in terms of their compatibility with different kinds of predicates or base constructions. A contrast was established between applicatives compatible with intransitive predicates and those only compatible with transitive verbs. Baker (1988) proposed that asymmetric applicatives are incompatible with intransitive bases because asymmetric applicatives require structural Case, which (by assumption) intransitive predicates lack, as in Chichewa. This would force an analysis of all applicative constructions with intransitives as symmetrical, but it is not clear what symmetrical would mean for constructions in which the only object is the applied argument. Pylkkanen’s typology addresses and explains this contrast directly: given that low applicatives denote a relation between two entities, low applicatives are incompatible with intransitive structures in which there is no internal argument. A second distinction has been set among intransitives between unergatives and unaccusatives. In particular, some languages can have applicatives with unergatives (with a locative argument), but not with unaccusatives, as reported for Mohawk by Baker (1996a).

130

CONTEMPORARY LINGUISTIC PARAMETERS

(5) a. D-hi-yó’tD-hs-e’[Mohawk] FUT-1SG.S/M.SG.O-work-BEN-PUNC ‘I will work for him’ b. *Sak wahuwayá’tD’-s-e’ ne owirá’a. Sak FACT-F.SG.S/M.SG.O-baby-fall- BEN-PUNC ne baby ‘The baby fell on Sak’ (Baker 1996a) Baker (1996b) argues that applicatives are universally incompatible with unaccusative bases, that a goal and a theme with unaccusatives can only project into a prepositional or oblique construction. Cuervo (2005) explicitly argues against this view, demonstrating that low applicatives with unaccusatives such as arrive, grow, escape and fall are indeed attested in Spanish with the same structural properties and range of meanings they have with transitive predicates. A double object analysis has in fact been proposed by Pesetsky (1995) for get in Susan got the letter. Unaccusative applicatives are possible in Pylkkänen’s typology as long as the notion of the “direct object” required by low applicatives is a deep notion of object as argument of the verbal root—which covers the internal argument of intransitives under the unaccusative hypothesis.6 An even more articulated analysis of verbal constructions has been claimed necessary in order to account for the full range of applicative constructions (Cuervo 2003, 2005, 2010). In particular, transitives must be divided into causative (break, melt) and non-causative (read, send, wash), and unaccusatives into anticausative (intr. break, melt) and simple (arrive, fall). Low applicatives are only compatible with non-causative transitives and simple unaccusatives, because they require an object licensed as complement to the verb. In contrast, low applicatives are incompatible with causatives (e.g. *Peter broke Mary the car) and anticausatives (e.g. *The door opened Mary—intended meaning = The door opened for Mary) because in this case the theme/ patient is licensed, by hypothesis, in a specifier position above the verb, not below it. The structural restriction of low applicatives to arguments in complement position simultaneously accounts for the restrictions of applicatives in causatives and anticausatives together with resultatives, ditransitives, small clauses and external arguments, restrictions noted in the literature but which had not received a unified account. In languages which allow applicatives with causatives, inchoatives, small clauses, etc., the applicatives take a stative vP as complement and are in turn embedded under another vP. These constructions, Middle or Affected Applicatives, are attested in Hebrew, Spanish, German, Greek, etc. (Cuervo 2003; Kallulli 2006; Schäfer 2008, and others). (6) Subtypes of predicates as relevant for a typology of applicatives Transitivity Transitive Causative

Intransitive

Non-causative Unergative

Psych states

Unaccusative Anticausative Simple



PARAMETERS AND ARGUMENT STRUCTURE II

131

iii) Alternative constructions. Another important point of variation centres on the availability and / or obligatoriness of applicative constructions. In one extreme, we have languages claimed to lack applicatives altogether; on the other, languages which exhibit a wide range of applicatives both in terms of their meanings and the types of predicates with which they are compatible. For many authors, dative case, lack of overt verbal morphology, and restrictions in productivity prevent or discourage an applicative analysis of “suspected” applicative constructions (Peterson 2007). This expectation of a particular morphosyntactic coding of applicative heads and arguments (as opposed to definitions in structural or formal semantics terms) is responsible for the initial and still common exclusion of languages like English, German, Russian, Finnish, Japanese, Basque, Guaraní, Spanish, and Kiowa from the lists of languages with applicative constructions, as in Polinsky’s (2013) recent classification. Given this state of affairs, I remain open but agnostic with respect to the possibility (and commonality) of languages lacking applicatives altogether. Applicative constructions vary on whether there is an alternative way of expressing the same or a similar meaning (usually via an oblique or adpositional phrase), or the applicative is obligatory. This variation can be found across languages (obligatory recipient applicatives in Tzotzil vs optional recipient applicatives in Chichewa, Peterson 2007), across constructions (Hakha Lai obligatory benefactive / malefactive applicatives vs optional instrumental and comitative applicatives, Peterson 2007) or across less general or even lexical choices (English optional recipient applicatives vs obligatory but unusual source applicative in That will save you some time / *That will save some time from you). Factors that have been shown to favour applicatives over oblique constructions, when available, include the animacy of the applied argument, inalienable—as opposed to alienable—possession, and pronominalization.7 The dative alternation, that is, the alternation between a double-object / low applicative construction and an adpositional construction, is probably the most studied argument structure alternation in the domain of applicatives. Numerous analyses of the constructions and the factors that affect the alternation in English have been proposed and fruitfully applied to the study of other languages. Under the general view that there is no true synonymy in language, however, the notion of optionality should be taken cautiously. The study of the semantic differences between the alternants of the dative alternation demonstrates that structural alternatives are also semantically distinct (be it at the level of event structure, information structure or discourse-level semantics) and, therefore, not true cases of optionality.

6.2.4. Covariation: Availability, correlations and implicational properties Among the points of variation in applicatives, several correlations and implications have been described. Languages vary greatly in the range of applicative constructions they exhibit. Based on his survey of 100 languages with applicative constructions, Peterson found that benefactives and comitatives are the most common applicatives, and that if a language has no benefactives, it is more likely to have instrumental

132

CONTEMPORARY LINGUISTIC PARAMETERS

applicatives. In terms of a Pylkkänen–Cuervo typology, some languages have low applicatives but not high or middle (e.g. English), some languages have high applicatives but not the others (e.g. Chaga), other languages have the three types low, middle and high (e.g. Spanish, Hebrew, Romanian). A language can have recipient low applicatives but lack source or possessor low applicatives. Beyond the contrast high-low, Pylkkänen (2008) proposes separate lexical entries for each sub-type of applicative head, such as LowApplto, LowApplfrom, HighApplben, LowApplinstr, etc. Among languages with low applicatives, it seems that having source applicatives implies having recipient applicatives, and having possessor applicatives, implies also having source applicative. If the contrast between these sub-types of applicatives merely reflects a language’s selection of functional heads from a universal inventory, these implicational relations are not expected. Observed correlations between types of predicates and type of applicative (e.g. benefactives with unergatives, accidental causers with inchoatives) can be naturally accounted for if the meaning of applied arguments is not fully determined by the licensing head but is also obtained as a configurational meaning. Based on my own work on middle applicatives, I found (in a very small sample of languages) that if a language allows applicatives with causatives, it also allows them with anticausatives, resultatives and small clauses (Cuervo 2005). The hallmark of these affected applicatives is that they take a state as their complement (typically a stative vP). If we consider this proposal in light of Pylkkänen’s reformulation of the denotation of LowApplat to spell out its possibility of combining with both events and states (2008: 44), we might expect a language internal correlation between having middle applicatives and static possessors. I am not aware of work which has explored this possibility, which requires future investigation. At the level of morphology, Peterson found that global morphological properties of a language as a whole—such as whether morphosyntactic relationships are marked on the head, on the dependent or by unbound morphemes—do not seem to correlate with the availability of applicatives (Peterson 2007: 211). In terms of the morphological expression of the applicative head, in combination with the high / middle / low typology, it seems that null applicatives can only be low. This observation holds both across languages (English and Finnish zero-applicatives are low while the applicatives identified as high are always expressed by an affix or clitic), as well as within a language with more than one type of applicative. If a language has recipient and source and / or possessor applicatives, and has overt and null applicative heads, then it is only the recipient applicative which can be null, as in Inuktitut (Carrier 2014). This association may turn out to be related to the contrast noted by Pylkkänen (2008) between low applicatives as closer to adpositions (as in Pesetsky’s 1995 null G) and high applicatives as verbal (as in Marantz’s 1993 double objects).8 Further discussion of these correlations and the viability of accounts in terms of parameters is presented after the section on variation on causative constructions.



PARAMETERS AND ARGUMENT STRUCTURE II

133

6.3. Causatives Causatives are considered by many as a type of valency-increasing operation performed on a verbal basis which can, in principle, be transitive or intransitive, as illustrated with Japanese below (examples from Folli and Harley 2007). (7) a. Calvin-ga Hobbes-o ik-ase-ta Calvin-NOM Hobbes-ACC go-CAUS-PST ‘Calvin made Hobbes go’ b. Calvin-ga Hobbes-ni piza-o tabe-sase-ta Calvin-NOM Hobbes-DAT pizza-ACC eat-CAUS-PST ‘Calvin made Hobbes eat pizza’

Intransitive base Transitive base

For many, the crucial addition in a causative is a participant in the event expressed by the basic clause, a participant systematically interpreted as the initiator. In this view, causatives do not explicitly add a predicate, and causativization is seen as a valency-increasing operation very similar to applicativization. The focus of this type of analysis is therefore on the addition of an argument (the causer) and the formal consequences for the expression of the “original” clause and its arguments. Indeed languages vary in the morphosyntactic properties of the causee (the original subject), such as its alignment with respect to the (complex) verb, its Case, and whether it triggers object or subject agreement. Causatives have also been characterized as the addition of an event, the causing event or initiation of the original event (or state). Within this view, most analyses propose that a causer argument is automatically added together with the causing event (but see Pylkkänen 2008). Within studies of the causative alternation, that is, the case of verbs that appear in a transitive causative structure as well as in an intransitive anticausative-inchoative, as illustrated in (8), the causative variant has been claimed to contain an extra layer of verbal-functional structure above the intransitive structure (Doron 2003; Hale and Keyser 1993; Harley 2008a, among others). (8) a. The boat sank b. The air force sank the boat

Intransitive inchoative Transitive causative

This idea that a causative includes the “corresponding” inchoative structure has been the standard view within various approaches to the alternation within generative grammar, from lexical semantics (e.g. Levin and Rappaport 1995; Koontz-Garboden 2009) to syntax-centred constructionalist approaches (e.g. Hale and Keyser 1993, 2002; Harley 2008a; Alexiadou et al. 2006; cf. Fábregas, this volume) irrespective of whether authors support a derivational or non-derivational analysis of the alternation. Thus, causation arises as the interpretation of a specialized atomic element, either in the argument structure associated with lexical verbs (for lexicalists) or in the syntactic structures in which verbs appear (for constructionalists), but see Cuervo 2003, to appear; Marantz 2005, for a purely configurational meaning of causatives.

134

CONTEMPORARY LINGUISTIC PARAMETERS

Causatives are expressed cross-linguistically via various types of morphology, along a continuum from separate lexical items to internal changes in the verbal root, reduplication, affixation, compounding, and serial or light verbs in periphrastic constructions (Comrie 1981). Causatives also vary inter- and intralinguistically in terms of morphosyntactic coding of arguments and semantic properties of causers, causees and the type of causation itself (Dixon 2000; Svenonious 2005). A central question has been how variation in morphosyntactic properties interact with differences noted in the interpretative properties of causative constructions.

6.3.1. Morphology of causatives As in the case of applicatives, expression of causatives within the verbal complex varies both within and across languages. Descriptively, we can classify the form of causatives into three groups: lexical, morphological, and syntactic (Dixon 2000). i) Lexical causatives. The contrast between a transitive causative and a “corresponding” non-causative intransitive variant of the same event or predicate can be expressed by two different lexical items, as in the following pairs in Dyirbal and English (Dixon 2000: 39). (9) intransitive a. Dyirbal mayi- “come out” gaynyja- “break” jana- “stand” b. English be dead come out lie

transitive bundi- bana- jarra-

“take out” “break” “put standing”

kill take out lay

Another type of lexical causative corresponds to ambitransitive verbs whose “intransitive sense is considered to be primary, e.g. trip, explode, melt, dissolve, walk and march” but take on a causative meaning when used in a transitive clause, as in English, Fijian, and Greek (Dixon 2000: 38). In current morphological theory, the cases illustrated in (9a) would be analyzed as cases of suppletion (if indeed the relation between the verbs or meanings “come out” and “take out” is that of causativization). The case of ambitransitive verbs such as English melt and break can be, and has been, analyzed as zero-derivation (either the transitive deriving from the intransitive or vice versa, or both verbs being the result of combining a root with a verbalizing zero-affix). ii) Morphological causatives. Many languages signal causatives by applying a morphological process to the base verb. Morphological processes can consist of changes internal to the verbal root (e.g. vowel lengthening, tone change, segment addition), reduplication or affixation. Alternatively, a language can form causatives by combining two predicates into a complex expression which acts as a single predicate in a monoclausal structure. Although both predicates can host verbal



PARAMETERS AND ARGUMENT STRUCTURE II

135

morphology, there is only one specification for tense, mood, aspect, polarity and evidentiality (Dixon 2000: 34). This is the case of serial verbs, particle-verb constructions, compounding, and faire causatives. (All data as presented in Dixon 2000). (10) a. Na-ŋa-tar-kwalca-t Yimas serial verb 3SGA-1SGO-CAUS1-rise-PFV ‘She woke me up’ (directly, e.g. by shaking me) b. kə́ə tòk háan múuc Kammu particle-verb 3SG+M CAUS2 die ant ‘He happened to kill an ant’ (e.g. by accidentally treading on it) c. bé-khó-ày-ɔ̀ m Kioawa compounding 2SGA-now-start.off-CAUS+IMP ‘Go ahead and run it (the tape recorder)!’ (Lit. Make it start off) d. Je ferai manger les gâteaux à Jean French faire 1SGA make+FUT+1SG eat+INF the cakes P Jean ‘I shall make Jean eat the cakes’ It is not clear how the reflexive morphology employed in several language families to express the causative alternation (Romance, Slavic, Hebrew) fits within this classification. More generally, this notion of lexical or morphological causatives is biased towards always taking the intransitive as basic, the causative form deriving from it by addition, and it is therefore difficult to account for causative-anticausative pairs in which morphologically the intransitive seems to arise by derivation from the causative (see Haspelmath 1993 for a typological study of morphological marking in the causative alternation, Cuervo 2014; Doron 2003 for analyses). iii) Syntactic causatives. Syntactic causatives refer to causatives formed by periphrastic expressions, each verb belonging to its own clause. Typically, the causative verb is positioned in the higher, main clause while the lexical verb is in a subordinate clause of some kind. Languages with periphrastic causatives vary in the coding properties of the various arguments. In particular, the causee (the original subject of the base, lexical verb) can keep its subject properties (as in Persian), it can take up object properties (as in English) or it can assume both subject and object properties, as in Canela-Kraho (data cited in Dixon 2000). Languages can vary also—intra- or interlinguistically—in the degree of integration of the two verbs and their corresponding clauses, as seems to be the case between the more integrated English make as opposed to causative cause. As noted by Dixon, what starts as two separate verbs can change into a complex predicate in a monoclausal structure (like French faire), and grammaticalize further into a compound causative verb (like Kiowa ɔ̀m). It is important to note that many languages have more than one causative construction. In such cases, two different forms are used, typically each form corresponding to a different “strategy,” such a lexical and periphrastic, affixation and complex predicates, etc., or to two morphemes of different morphophonological weight (e.g. Hindi, Amharic, Tariana; Dixon 2000). Malagasy and Tagalog seem to go against this generalization since, according to Travis (2000), these languages have two causatives with identical morphology but which differ in meaning and site of attachment. As we shall see in section 6.3.4, several proposals have argued for

136

CONTEMPORARY LINGUISTIC PARAMETERS

a systematic association between form of the causative and semantic properties of the construction. Finally, there is some variation in the placement of causative morphemes within a verbal complex. In most Bantu languages, a causative morpheme must appear closer to the root than an applicative: V-cause-appl. In Hiaki, in contrast, both orderings are possible and have a transparent interpretation in which morpheme order reflects syntactic structure and compositionality (Harley 2007). Interestingly, causatives can vary also in their alignment with respect to an event head: in Malagasy, the causative -an- can appear to the right and/or the left of the Event head -f-, the higher causative (to the left of -f-) being interpreted as indirect causation, the lower as direct, as illustrated in (11) from Travis (2000). Variation in morpheme ordering interacts in interesting, systematic ways with the structural and interpretative properties of causative constructions. (11) m-an-f-an-hisatra[Malagasy] T-CAUS-E-CAUS-move.slowly ‘Z make Y move X slowly’

6.3.2. The syntax of causatives Variation in causative constructions is found in the types of predicates that can be causativized in a given language. In relation with the type of verb or clausal structure that a causative can take, variation is found in the morphosyntactic behavior of the arguments, as seen in their case marking, word order and in agreement morphology. In descriptive terms, causatives built on an intransitive base have the causer with subject properties and the original subject, now causee, “demoted” to object of the clause. Cross-linguistically, some causatives are restricted to take only intransitive bases. This is the case for languages which have only lexical or morphological causatives, as opposed to periphrastic causatives, which are very productive. More variation is found in the rearrangement of morphosyntactic coding of arguments (causee and original onject) in a causative built on a transitive or ditransitive base. Dixon (2000:48) presents five alternatives attested in languages of the world. (12) Causative of transitive type causer i. A ii. A iii. A iv. A v. A

original A (causee) special marking retains A-marking has O-marking O non-core

original O O O has O-marking non-core O

The first type indicates that the language has a marking specific for causees, not shared with any other type of argument. This marking can have the form of a case affix or adposition. In type ii, it is possible that two DPs have the same case marking



PARAMETERS AND ARGUMENT STRUCTURE II

137

or that subject properties are distributed between the two (e.g. case marking in the causee DP, verbal agreement with the causer DP). The other three types can be found across languages or can coexist within a language. In fact, it might be difficult to decide whether the two objects in a causative classified as type iii are equally “good” objects. Dixon states that “for most of these languages [of type iii], only one of the arguments has full properties of an O… It is generally the original A which is the full O while the original O has become the second object” (Dixon 2000: 50). This kind of variation in the arguments of a causativized transitive parallels variation found in applicative constructions, both intra- and cross-linguistically. Marantz (1993) explicitly discusses the similarity between applicatives and causatives both at the level of the verbal heads and of the argument phrases. Another dimension in the expression of arguments involves the obligatoriness of argument expression. There is variation in the obligatoriness of both causee and causer. Most analyses of causatives assume or claim that causers are obligatory; for some the addition of the causer is even the crucial element in causatives (Dixon 2000; Doron 2003). In contrast, Pylkkänen argues that a cause head introduces a causing event but does not introduce a causer DP itself; causers are introduced by a “standard” external-argument licensing head, Voice. Japanese adversity causatives and Finnish desiderative causatives are cases of causer-less causatives (Pylkkänen 2008). Causees seem to be obligatory in most causatives, but importantly, not in all. In causatives which distinguish between indicating that “the aim is to get the activity done (by anyone), or to get it done specifically by the stated causee” (Dixon 2000: 73), the causee can be absent in the first type. This is the case of causees in French (and other Romance) faire par periphrastic causatives (and Northern Sámi, Svenonious 2005), as opposed to faire infinitive. In Amharic, a causee can appear in instrumental case, in accusative case, or can be absent altogether (Godfrey 2013: 12). Importantly, variation—particularly intralinguistic variation—in the marking of objects and the obligatoriness of their presence can be associated with contrasts in interpretation and / or underlying syntactic structure, including contrasts in the values of semantic-syntactic features such as specificity, animacy, etc. In Romance periphrastic causatives, faire par has been analyzed as a causative which takes a VP (or a passive clause)—that is, a structure without an external argument—as opposed to faire infinitive constructions, in which the causative takes a full clause vP or VoiceP and the causee must be animate. This structural difference explains the contrast in obligatoriness of the causee, and the fact that if a causee is present in faire par it appears as an adjunct rather than in dative or accusative case. In Japanese, a causative with a causee marked with accusative is interpreted as a direct causative “make” or “force,” while if the causee is marked with dative –ni the causative is indirect, closer in meaning to a permissive “let” (Harley 2007; an unwilling or a willing causee, according to Dixon 2000). The same applies to Spanish causatives with hacer “make” (Strozer 1976). In Korean, the contrast between accusative and dative for the causee of an intransitive base signals whether the action is done fully (accusative) or “to some extent” (Dixon 2000: 73). Variation in the syntax and morphology of causatives has been analyzed by Pylkkänen (2008) as the consequence of parametric cross-linguistic variation in two dimensions, while maintaining a unified account for the denotation of the causative

138

CONTEMPORARY LINGUISTIC PARAMETERS

head cause. One dimension involves the licensing of a causee as a requirement, or not, in the formation of causatives. Pylkkänen argues that cause does not license an argument itself, but that in many languages, including English, the head cause is bundled together with Voice in one syntactic object such that when a causative is formed the addition of a causing event entails the licensing of an external argument (interpreted as causer). As we shall see, bundling can be viewed as a micro-parameter operating at the level of the functional lexicon. A second dimension of variation involves the categorical type of the complement of cause. For Pylkkänen, a causative head can select for one of three possible complements: root phrases, vPs or phases (i.e. a verbal projection with an external argument or high applicative). These three types of complement structures are defined as the main components in the functional architecture of the verbal domain assumed by many within a generative framework. Differences in the types of morphology that can appear closer to the root than the causative morpheme (desideratives, reciprocals, markers of (in)transitivity, stativizers, applicatives), and the types of compatible adverbial modification that various causatives allow can be derived from this selectional parameter. This typology of causatives can also account for differences in the agentivity requirement for the causee. In particular, the restriction to agentive, animate causees—as in faire infinitive and Northern Sámi and Amharic outer causatives—can simply derive from the causative being phase-selecting. The approach, however, does not add much to the explanation of other semantic differences in the type of causation expressed by the different types of causatives, as briefly discussed below.

6.3.3. Types of causation At least two types of causation have been identified in the descriptive and theoretical literature: direct and indirect. In Kratzer’s (2005) terms, direct causation involves an event of causing a state, the event performed by the causer directly brings about the result state. In indirect causatives, the causer brings about the state via some intermediate event. Direct causation is the interpretation not only of English lexical causatives such as break and sink, but also of English resultatives, as in drink the teapot empty. In other approaches, directness is measured not so much as a type of causation, but as whether the causee has or lacks control over the event, or whether it acts willingly or unwillingly. Other researches (or languages) place the emphasis on properties of the causer: whether the causer acts intentionally or not, with or without effort, or if it is directly involved in the event (Dixon 2000). Implementing these distinctions formally is not a straightforward task. Godfrey (2013) argues that an implementation in terms of features such as [+/- direct] [+/intentional] for the causee, causer or causative morpheme is problematic at least for two reasons. First, if several features exist for the specification of causative constructions, we might expect “there could be languages with a wide inventory of causative morphemes” (Godfrey 2013: 5), contrary to fact. Alternatively, there could be only one feature [+/- direct] dependant on cause whose precise semantic implementation may vary from language to language. If this were the case, however, Godfrey argues that causatives in languages with only one type should not exhibit



PARAMETERS AND ARGUMENT STRUCTURE II

139

restrictions in aspects of their interpretations, a situation contradicted by Indonesian single causative, which seems to express direct causation only. He further claims that Kratzer’s extension of types of causation is better avoided if the attested contrasts can be derived from independent principles or from selectional variability in syntactic (as in Pylkkänen 2008; Travis 2000) or semantic terms. In another line of approach, causative meaning arises compositionally in a more literal sense, rather than as the meaning of a particular predicate, the cause head. Different implementations of the idea that causation is the interpretation of a structure in which a dynamic event embeds another event (or state) appear in Cuervo (2003, to appear), Marantz (2005), and Ramchand (2008). Focusing on zero-morphology direct causatives in English and Spanish, Cuervo (to appear) argues that change of state predicates such as break do not mean “cause to become broken” but more directly “make be broken,” an interpretation of a dynamic vdo which takes a state (vPbe) as complement. Direct causation in this view is equated with causing a state, as opposed to indirect causation (expressed by periphrastic causatives in English and Spanish), which embeds a dynamic event (either agentive VoiceP or non-agentive unaccusative vPgo).9 Languages can vary parametrically along these lines.

6.3.4. Further correlations in variation Two main correlations have been noted between dimensions of variation in causative constructions. One correlation has been established between the morphological expression or form of the causative and the directness of the causative meaning, such that the smaller the morphology (from lexical to periphrastic causatives) the more direct the causative meaning (Comrie 1981; Dixon 2000). The second is a robust generalization between the position of the causative and its interpretation, such that the higher the morpheme in the structure, the more indirect the causation. Are these two correlations connected? Are they the expression of the same underlying phenomenon? Can one be subsumed or derived from the other? Svenonious (2005) discusses the differences between “outer” causatives (higher) and “inner” causatives as they emerge from a series of languages which have two distinguishable constructions. Svenonious notes the following association of properties. (13) a. Inner causatives i. Less regular morphology ii.  Many restrictions on the base (e.g. unaccusative, intransitive), less productive iii. Causee is obligatory direct object iv. No restrictions on animacy v. Direct causation vi. No iteration b. Outer causatives i. More regular morphology ii. Few restrictions on the base, more productive

140



CONTEMPORARY LINGUISTIC PARAMETERS

iii. iv. v. vi.

Causee can be optional/oblique Causee is animate Causation can be indirect (e.g. permissive, assistive) Possibly iterative

These associations do indeed group variation in morphological expression with morphosyntactic and semantic properties. The issue is, then, to account for the variation and the observed associations. Would this cluster of properties be best expressed as a macro-parameter, in order to capture the association of properties from various domains of the grammar? Or should we take the properties of one domain to be the crucial point of variation (let’s say height of attachment), properties in other domains deriving as consequences? Pylkkänen’s account in terms of variation in the selectional properties of causatives can naturally account for the association of properties ii-iv, but must be supplemented to incorporate the association of morphological and semantic properties. Similarly, Travis (2000) distinguishes two types of causatives in terms of their merging position, defined as below or above a domain set by an Event head (recall the Malagasy Event head -f- in (11), an approach which does not have clear direct consequences for the direct-indirect semantic distinction nor for the morphological form. Cuervo’s (to appear) proposal that the crucial distinction is between causing a state (lower, inner causatives) and causing an event (higher, outer causatives) would have to be supplemented the same way. Given the similarities in the correlation of height of attachment, morphological expression and interpretation of causatives and applicatives, further discussion of these issues is done together, in the next section.

6.4. Parametric accounts of variation We have identified several dimensions of intra- and interlinguistic variation in the domain of causatives and applicatives: a. Morphological shape of the head b. Order of morphemes within the verbal complex c.  Morphosyntactic coding of arguments (including word order, Case and agreement) d. Compatibility with various types of predicates e. Availability and obligatoriness of the construction f. Semantic interpretation of arguments and event structure. Can this variation be accounted for in terms of parameters? And if so, what kind of parameters would they be? We could be facing macro-parameters if properties of one structure have consequences for or correlate with (properties of) other syntactic structures, forming a cluster of correlations that the macro-parameter should be responsible for. Micro-parameters, in contrast, are better suited to express variation in a more local domain given that they are usually expressed as choices in features



PARAMETERS AND ARGUMENT STRUCTURE II

141

of particular (functional) heads. Although micro-parameters might seem more plausible in a domain in which abstract feature inventory and bundling are relevant, they might still prove inadequate to account for variation expressed as tendencies rather than categorical choices. Current argument structure theory states that verbal meanings are constructed on the basis of a small set of possible syntactic-semantic frames in which particular lexical elements (roots) can be inserted. Applicative and causative heads are also integrated into these structures, which are then interpreted (morpho)phonologically and semantically. Such a constrained theory sets predicted areas of variation as well as the type of variation expected (or not) in each dimension (the lexicon, the morphophonological shape of heads, the morphosyntactic coding of arguments). Variation in these dimensions will reflect “language-particular choices and constraints from a parameter space of universal options” (Marantz 2013: 155). As far as meanings are configurational, they are not expected to vary greatly from language to language (except as derived from available syntactic structures), and semantic parameters in argument structure are not expected. One way to inquiry into the sources of variation and the restrictions found in applicative and causative constructions is to study the dative alternation and the causative alternation as a case of intralinguistic variation. If the alternations are viewed as derivationally related, a controversial but still popular view, then parameters will be behind a certain operation taking place or not (some properties would either trigger or prevent an operation) such as preposition incorporation in Baker 1988, or case checking for possessor raising as in Landau 1999. Within non-derivational approaches to the alternations (basically, constructionalist approaches in which each construction is built without reference to a possible alternant, and in which the semantic principles that may guide clause building are compositional rather than lexically based theta-theory), as in Marantz 1993, Pesetsky 1995, Pylkkänen 2002, the availability and properties of applicative and causative constructions must derive from properties of heads (more specifically, choices in inventory of heads, and in formal features and their grouping). Variation, however, can be more specific, and depend on particular lexical properties, as the restriction on English double objects formed on verb with Latinate roots (give someone a book vs *donate someone a book). This type of idiosyncratic variation must be located at the lexical level (for a theory such as DM, these idiosyncrasies and specific aspects of meaning are associated to roots. Variation which depends on differences in the non-functional lexicon can hardly be considered parametric. Languages make choices with respect to the type of morphology that spells out the various pieces of event / argument structures of sentences, some morphemes spelled out by (possibly null) affixes, clitics, or unbound lexical items. Order of morphemes within a verbal complex is usually the result of syntactic composition (thus reflecting syntactic hierarchies) but can be affected by parametric or more idiosyncratic morphological or phonological constraints. What about the morphological generalizations such that the higher the head, the more regular and/or obligatory overt morphological expression? If this is a true generalization, at what level of parametrization can it be captured? A micro-parameter in this regard would probably make reference to particular heads (e.g. Appl or cause), thus missing the cross-categorical generalization.

142

CONTEMPORARY LINGUISTIC PARAMETERS

Rizzi (2011) revisits the concept of parameters and takes stance against the seemingly arbitrary character of the initial list of parameters proposed within the Principles and Parameters theory. He argues that attention must be paid not only to the locus of parameters (now taken to be the functional lexicon rather than principles of grammar) but crucially to the form. He proposes that a parameter is “an instruction for a certain syntactic action expressed as a feature F on an item of the functional lexicon H and made operative when the item enters syntax as a head.” The format and the possible locus of parameters are as below. (14) a. Format of parameters (Rizzi 2011) H has F {yes, no} b. Locus of parameters Merge parameters Move parameters Spell-out parameters This notion of parameter is very local and, as such, does not seem to be able to express global macro-parameters. Rizzi argues that parameters can interact with each other and with crucial aspects of syntactic structure, from which systemic consequences and articulated theories of variation can be derived. I discuss below accounts of variation in applicatives and causatives in light of Rizzi’s proposal. i) Selection as Merge parameters. An account of variation in terms of (categorical) selectional properties is at the core of Pylkkänen’s analysis of causatives and applicatives. Within Rizzi’s (2011) possible locus of parameters, selection parameters are an instance of Merge parameters, making reference to the type(s) of complements required by the functional item when it enters a syntactic derivation. From this perspective, compatibility with various types of predicates and semantic variation emerge mainly as a result of syntactic features and thus does not require semantic parameters expressing variation in semantic composition. Another type of Merge parameters refers to “cartographic” choices, that is, variation in the relative height of attachment of functional heads, such as Negation, Focus, etc. Proposals on the varying sites of attachment of causatives and applicatives (such as Pylkkänen’s, Cuervo’s, and Travis’) do not seem to be cases of cartographic parameters, however, because this variation—unlike cross-linguistic variation in attachment of Negation—has systematic consequences for interpretation and may be associated with differences in denotation. ii) Bundling of formal features. Feature bundling in one head also depends on the initial choices of formal features individual languages make from a universal inventory (probably constrained by feature geometries), but can the grouping on a certain head be considered a parameter? The closest would be micro-parameters, which deal with feature endowment of heads. Within a Distributed Morphology approach (Halle and Marantz 1993), this takes place in the narrow lexicon, the construction of bundles of formal features into abstract lexical items (not to be confused with vocabulary items with phonological shape). Although feature bundling is not directly represented as a possible type of parametric variation in Rizzi’s approach, it could be considered as a Merge parameter within the functional lexicon with consequences for the availability of



PARAMETERS AND ARGUMENT STRUCTURE II

143

syntactic configurations. For instance, Pylkkänen argues that bundling of Cause with Voice heads in English is responsible for the impossibility of building a causative construction without expressing a causer (see Tubino Blanco 2010 for discussion). Pronominal clitics as the spell-out of applicatives could also be the result of bundling of (uninterpretable) D or φ-features with the applicative head, which allows for the licensing and specific interpretation of null applied arguments in Spanish and Basque. Object properties and movement possibilities in applicative constructions that have been proposed to derive from whether the applicative head sets a phase boundary (McGinnis 2001) can also be captured in terms of feature bundling. iii) Parameter interaction. A promising avenue to explore a syntactic parameter as the source of variation in the morphosyntactic properties of applicatives and causatives is provided by Marantz’s (1993) distinction between two types of combination of a verb and an applicative head. He argues that two verbal heads can combine as Merge-Incorporation M/I or, alternatively, as Raising-Adjunction, R/A. In Chichewa and English asymmetric applicatives, the verb incorporates into the higher applicative and forms a complex predicate, while in Chaga the verb raises and adjoins to the applicative but its trace remains active, maintaining “all the properties of the verb” (Marantz 1993: 131). Although the structures of M/I and R/A before movement are identical, “M/I results in a sort of syntactic clause collapsing that is not characteristic of R/A” (Marantz 1993: 132). We can imagine that languages could make a choice in this respect, and that the type of combination would be a type of Move parameter which could act as a macro-parameter. We then expect that causatives and applicatives should behave the same way within the same language, being both examples of complex predicate formation. Interestingly, a parallel contrast in syntactic combination of verbal heads (roots and functional elements) has been proposed between Incorporation (move and copy) and Conflation (direct merge), which has been exploited in contemporary debates on the manner and result complementarity of interpretation of predicates, and on the formation of denominal verbs (Hale and Keyser 2002; Haugen 2009; Mateu and Acedo-Matellán 2012; Rappaport and Levin 2010). This parallelism highlights the idea that argument structure is built via syntactic operations and, consequently, its interpretation must proceed via transparent semantic composition. A parameter of this type can potentially partly explain the connection between the morphological form of applicative (and causative) heads and their syntactic and semantic behaviour. Let’s recall that, on the syntactic side, asymmetric applicatives (that is, applicatives in which only one object behaves as a direct object) have been analyzed as low applicatives, while high applicatives (which can appear with unergative verbs) are symmetric applicatives. With respect to causatives, outer causatives have been associated with heads which take bigger complements and express causation understood as more indirect than the causation emerging from inner causatives merged closer to the root. On the morphology side, applicative heads which are not phonologically realized, as in English and Finnish, seem to always be low applicatives. In turn, lexical causatives—in the descriptive sense used in section 6.3.1, that is, those causatives formed with a zero-morpheme—are always low, direct causatives. Morphologically more independent causatives, as periphrastic causatives, in contrast, can express indirect causation and embed bigger structures.

144

CONTEMPORARY LINGUISTIC PARAMETERS

This correlation holds not only cross-linguistically, but also within languages which have more than one applicative or causative morpheme, particularly if one of them is phonologically null. As licensing of null pronominals has been linked to particular parametric options (as in the Null Subject Parameter), so the licensing of null heads might require a particular way of combining into a complex predicate. Snyder’s Compounding Parameter is another case of a parameter that deals with complex word or predicate formation (see Fábregas, this volume for discussion). Alternatively, this can be viewed in the reverse: a lower head which moves and incorporates into a root (low applicative), or the first functional head into which a root incorporates (a root-selecting causative) can remain null, as opposed to heads which combine with an already verbalized head. The intra- and cross-linguistic generalizations observed in the association of height of attachment (above or below a root vs above a functional layer) and morphological expression (null vs overt) then, could be expressed in terms of the interaction of a Move parameter and a Spell-out parameter, rather than in terms of features, or properties of specific vocabulary items of particular languages. iv) Other variation. Some of the attested variation seems to escape parametric accounts. Such is the case of choices a language can make from among a list of universally available heads. Thus, whether a language has or does not have instrumental applicatives or source applicatives can be a mere consequence of lexical choices among functional elements, as proposed by Pylkkänen (2008). This approach is supported by the existence in some languages of multiple vocabulary items for various applicatives (even in Finnish null low applicatives, Applto and Applfrom are associated with different Case). Properties of roots can also be at the basis of variation in the availability of alternative constructions, as is the case with the dative alternation in English. Additionally, variation in argument-structure alternations, as well as in the morphosyntactic coding of arguments, can be a consequence of more general syntactic properties of particular languages, such as Case and agreement relations (which, in turn, could be the result of parametric choices).

6.5. Conclusions and pending issues One aspect that must be better understood before we can have better and more comprehensive theories of variation in causatives and applicatives has to do with morphological overtness in general (whether a morpheme has a phonological exponent or not) and particular forms of overtness (whether a head is spelled-out as an independent morpheme, a bound morpheme or is associated with inflectional features of its own). Although the contrast between null and overt heads could be accommodated either in terms of restrictions on zero-derivations or licensing of empty elements, there remains the issue of correlations between morphosyntax and semantics along a continuum of phonological exponence (Comrie 1981; Dixon 2000). This issue is an interesting challenge for Distributed Morphology, given the central claim that phonological properties are not present during the syntactic derivation. Another area that requires, and now can have, detailed investigation concerns the interaction between applicative and causative structures and incorporation, which



PARAMETERS AND ARGUMENT STRUCTURE II

145

can all be considered cases of complex predicate formation. If low applicatives are merged below the verb and in some cases both arguments exhibit object properties, why no language exists (or has been found) in which either object can incorporate into the verb? Incorporation can also interact with causativization. In Southern Tiwa, for instance, the object of a transitive must incorporate before the verb can take a causative suffix (Dixon 2000: 44). More research is needed to account for these interactions. Finally, it is important to recall the role parameters are supposed to have in terms of explaining language acquisition and developmental stages. Comparative studies in the acquisition of causatives and applicatives is a necessary component of a comprehensive, and explanatory theory of causative and applicative constructions.

Notes 1

As in Marantz (1993) this notion of double object construction includes both constructions in which the two DP have the same “direct” case marking, as in English, as those in which one DP is marked accusative (or absolutive) and the other DP is marked with dative or genitive, as in Japanese, Spanish, and Greek.

2

In particular, what allowed for an applicative analysis of constructions not considered applicatives before was (i) the idea that linear word order and use of apparent prepositions were not a true indication of adpositional, oblique constructions as opposed to double object constructions, and (ii) the idea that (dative) clitics can be exponents of applicative heads on a par with non-agreeing verbal affixes and null morphemes.

3

In Cuervo’s analysis, “causer” is the interpretation of external arguments of dynamic vPs (introduced by Voice or HighApplicative) whose v head is not modified by a root.

4

There is an association (or restriction) in some languages between applied arguments and animate arguments, in contrast with a lack of such association in adpositional / oblique arguments (e.g. English and Greek double-objects, Bantu locatives vs recipients, Halkomelem applicatives). This restriction is hardly universal, however, and inanimate applicatives are found in Bantu and Romance languages, and in any language which has instrumental, circumstantial or locative applicatives.

5

This asymmetry is reproduced in nominalizations (there is no recipient nominalizing morphology) and relational adjectives, which typically cannot be understood as recipients (A. Fábregas, p.c.)

6

In fact, Finnish adversity source applicatives show the opposite pattern to (5), unergatives being unacceptable and unaccusative possible with these applicatives (Pylkkänen 2002).

7

For issues of space and scope of this work, I leave aside the role of animacy and pronominalization of the objects. These are factors in availability and properties of the constructions which interact in interesting, non-universal ways, with variation in applicatives.

8

See also discussion of preposition incorporation in the formation of transitive verbs and double objects in Hale and Keyser (2002).

9

For reasons of space, I leave aside discussion of accidental or non-volitional causers. For analysis and discussion, see Cuervo 2003, to appear; Kallulli 2006; Schäffer 2008.

7 The functional structure of the clause: Main issues Martina Wiltschko University of British Columbia

7.1. Exploring language variation In this chapter, I address the question of language variation in the domain of tense, aspect and mood (TAM). Any description of such variation, its analysis, and the theoretical musings based upon it has to be couched within a general understanding of what it means for languages to vary. How do we detect language variation? How do we compare languages to each other? How do we infer language universals? It is a common strategy to determine the common denominator of a category across unrelated languages and conclude that it is a language universal—hence part of universal grammar (UG). The remaining task is to identify the sources and limits of variation associated with this category (see Arregui et al. 2014, for an analysis of imperfectives along these lines). In pursuing this strategy, we have to properly identify the targets of comparison. How do we know that two language-specific categories are two instances of some hypothesized universal category? What are the heuristics to identify the comparison set? It is often assumed that the comparison set is meaning-based. This is the strategy common across various frameworks including the functional typological literature (Haspelmath 2007) the formal semantics literature (von Fintel and Matthewson 2008), as well as some syntactically based typologies, such as for example cartography (Cinque 1999). However, meaning-based comparison inevitably leads to vast variation in formal properties. That is, variation is found in the inventories of categories across languages, but also in the morphological, syntactic, semantic, and pragmatic properties associated with a particular meaning. Consequently, many typologists deny the existence of absolute universals; instead they postulate prototypical

148

CONTEMPORARY LINGUISTIC PARAMETERS

categories with “fuzzy boundaries” (Comrie 1989; Croft 2003) or they deny the existence of universal categories (Evans and Levinson 2009). But denying language universals leaves us without a means for language comparison. That is, languages cannot be compared directly to each other. They must be compared to a third element—a tertium comparationis in the sense of Humboldt—a universal grammar of sorts. Even linguists who deny the existence of UG are using a common basis, at least in the form of glossing conventions. To present data they need to be glossed. The standard guidelines for glossing (the Leipzig glossing rules) contain a set of glosses that are meant to encompass morphemes from all languages: they should be universally applicable. Thus, linguists following these glossing rules assume some universal set of grammatical morphemes. Whether or not they are in fact used in a particular grammar is a different matter. The point is that if they are used they belong to this abstract set of grammatical morphemes which provides us with the tool for crosslinguistic comparison. Within the generative enterprise, language comparison is ultimately a means to an end, namely to find language universals. Here, I adopt this research agenda in the domain of TAM. I argue that exploring language variation (especially in the domain of functional categories) requires us to establish a formal typology, one that takes into consideration how form, meaning, and categorial labels relate to each other. In section 7.2, I give a brief overview of core assumptions regarding the syntax of TAM categories within the generative tradition. In section 7.3, I compare the typological space as delineated across generative approaches towards variation. These frameworks are applied to a case study of TAM categories, comparing Standard German and an Austrian German dialect (section 7.4). I then show how comparison proceeds across the three frameworks (section 7.5). I show that meaning-based approaches have to be supplemented with a typology that is based on the way form relates to meaning. Section 7.6 concludes.

7.2. The history of clausal functional structure 7.2.1. Tense Syntactic analyses of clause-structure frequently assume a functional category Tense (T) which hosts tense morphology. (1)

TP

Subj T



T

[present] [past]

VP

The rationale for postulating T is the fact that the semantic content of the morphology hosted by this head is temporal. The crucial step that initiated this heuristic was



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

149

taken in Pollock’s (1989) seminal paper where Tense was first introduced. The pre-Pollockian clause-structure featured a functional category labelled as INFL, which hosted verbal inflection, including tense and agreement features. What Pollock (1989) found was that word order facts require the postulation of more than one functional category. Thus, the split of INFL was motivated by syntactic and morphological considerations. Since its introduction, Tense has been associated with syntactic and semantic functions that go beyond its original function. One of the syntactic correlates of tense is the licensing of the grammatical subject role. According to X’-theory each head-complement pair introduces another phrasal position (the specifier). Tense marking is commonly assumed to be responsible for nominative case-assignment (Rouveret and Vergnaud 1980). Consequently, Tense is assumed to host in its specifier position the phrase, which bears the grammatical subject role. (2)

TP

Subj T

NOM T [present] [past]

VP

On the semantic side, Tense has also taken on an important function: it serves to establish a relation between the event time, the reference time, and the utterance time. This assumption is rooted in Reichenbach’s (1947) insight that temporal interpretation relies on three distinct times as in (3). (3) Utterance time (UT): the time at which the sentence is uttered Reference time (RT): the time about which a claim is made Event time (ET): the time at which the relevant event takes place While Tense establishes a relation between the reference time and the utterance time, (viewpoint) aspect establishes a relation between the event time and the reference time. Thus, the relation between the event time and the utterance time is mediated by the reference time, which is in turn associated with aspect. There is no necessary connection between the postulation of the functional category tense, which was introduced for syntactic reasons, and the Reichenbachian framework for temporal interpretation.1 However, according to many scholars there is a direct mapping between the syntax and semantics of tense. Tense is commonly viewed as the host for morphemes that locate the reference time relative to the utterance time (Zagona 1990a, 1995; Demirdache and Uribe-Etxebarria 1997; Kratzer 1998; Matthewson 2006, and others). Thus, Tense is viewed as an anchoring category (cf. Enç 1987), which has both formal and interpretive properties: it serves to anchor the event (introduced by the verb) to the utterance. 2, 3 In sum, there are three correlates of the functional category tense. (4) i) hosts tense morphology

150

CONTEMPORARY LINGUISTIC PARAMETERS

ii) licenses grammatical subjects iii) contributes to the temporal interpretation of the clause in a Reichenbachian way tense is commonly assumed to be part of UG (e.g. Matthewson 2006). But it is still an outstanding question what the status of the three correlates in (4) is relative to UG. Minimally at least one of the properties in (i-iii) is universal. Though researchers differ according to which of these correlates they take to be universal— in part depending on the properties of the languages they take as their empirical basis. While all three properties are found in English, not all are universally attested. In some languages, there is no correlation between the presence of tense morphology and the licensing of grammatical subjects. For example, in Romanian grammatical subjects are licensed even in the absence of overt tense morphology, as in (5).4 (5) Ion vrea ca Dan ša resolve problema [Romanian] Ion wants that Dan prt solve the.problem ‘Ion wants Dan to solve the problem’ (Farkas 1985: 2) This suggests that there is no universal correlation between overt tense morphology and the licensing of grammatical subjects. Similarly, many languages do not have overt tense morphology. For example, in Blackfoot there is no overt marking that distinguishes between past and ongoing events, as in (6). (6) Oma píítaawa áípaawaniwa [Blackfoot] om-wa píítaa-wa a-ipaawani-wa dem-prox eagle-prox impf-fly.ai-prox ‘That eagle is/was flying up’ (from Reis, Silva and Matthewson 2007: 8) Such languages are sometimes referred to as tenseless languages or superficially tenseless languages (Matthewson 2006). The list of tenseless languages is pervasive and includes: Mohawk (Baker and Travis 1997), Chinese (Lee 1999; Lin 2002, 2003, 2006, 2010; Smith and Erbaugh 2005), Inuktitut (Shaer 2003; Bittner 2005), Navajo (Smith 1991), Halkomelem (Wiltschko 2003), St’at’imcets (Matthewson 2006). For a recent overview see Lin (2012). In the absence of overt tense morphology, there can by definition not be a correlation between tense morphology and other properties. It is however still possible that there be zero morphemes responsible for temporal interpretation (see section 7.3). Given that the correlates of Tense listed in (4) are not universally attested, it follows that they cannot be universal properties of Tense. Thus, the syntactic category Tense responsible for these correlates must be a construct; but it may also be constructed in other ways. That is, functional categories are constructed from abstract universal categories and language-specific units of language (Wiltschko 2014). This assumption makes three predictions.



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

151

(7) i)  expressions of temporal relations can be constructed in different ways ii) tense can be constructed in different ways, with different languagespecific correlates iii) the same ingredients used to construct tense can be used to construct other categories

7.2.2. Aspect Splitting the head of the clause into two functional categories paved the way for the exploration of the functional structure that makes up the clause. For example, a functional category Aspect has been introduced below Tense (Travis 2010, inter alia). Like Tense it is assumed to fulfill several morphosyntactic and semantic functions. It hosts in its head aspectual morphology (such as perfective and imperfective). Associated with the head of aspect is a phrasal position, which is sometimes assumed to license grammatical objects via the assignment of accusative case (Borer 1994, 2005b; Megerdoomian 2000). This is supported by the fact that in many languages, accusative case-assignment correlates with aspectual morphology. (8)

AspP

Obj Asp

ACC

Asp VP

[perf]

For example, in Finnish perfective predicates assign accusative to the internal argument (9a), whereas imperfective predicates assign dative (9b). Just like with Tense there need not be an overt morphological exponent of Aspect. (9) a. hän luki kirjan he read book.acc ‘He read the book’ b. hän luki kirjaa he read book.dat ‘He was reading the book’ (Comrie 1976: 8)

[Finnish]

In terms of its semantics, Aspect is similar to Tense. It is amenable to a Reichenbachian analysis: it relates the event time to the reference time (Demirdache and Uribe-Etxebarria 1997, among others). In sum, there are three correlates of the functional category aspect. (10) i) hosts aspectual morphology ii) licenses grammatical objects iii) contributes to the aspectual interpretation of the clause in a Reichenbachian way

152

CONTEMPORARY LINGUISTIC PARAMETERS

7.2.3. Mood In many languages, mood is associated with a dedicated verbal (inflectional) paradigm (see section 7.4). For example, a contrast between indicative and subjunctive mood is common in Romance and Balkan languages (Landau 2004; Quer 2006, 2009) while a contrast between realis and irrealis is common in the languages of the Americas (Palmer 2001). Again, following the logic that lead to the postulation of Tense, the existence of verbal inflection expressing contrasts in mood, lead to the postulation of the functional category Mood (for a recent discussion see Amritavalli 2014). Unlike Tense and Aspect, Mood does not appear to be licensing a grammatical role in its specifier position. This is however not too surprising, given that Mood is above the domain where case is licensed. In terms of its semantics, Mezhevich (2008) shows that Mood can be analyzed similar to Tense and Aspect. In particular, she suggests that Mood serves to relate the utterance time to the evaluation time, i.e. the time relative to which the utterance is evaluated. If the evaluation world is the same as the utterance world, the result is a realis interpretation; if the evaluation world is different from the utterance time, the result is an irrealis interpretation. On this view, there are two correlates of the functional category mood. (11) i) hosts mood morphology ii)  contributes to the interpretation of the clause by relating the proposition to the evaluation world The fact that Tense as well as other functional categories in their language-specific instantiations, must be viewed as constructs leads to a methodological problem. If categories have different properties across languages, then we need to have a method of identifying them across languages. This is not a trivial question. If the correlates of a given functional category are language-specific, then it follows that the criteria for determining categorial identities are language-specific. But then how do we determine whether language-specific categories can be analyzed as instantiating the same universal category? This question is approached in different ways across different frameworks.

7.3. Variation across frameworks The postulation of functional categories, such as tense, aspect, and mood (TAM) may have morphological, syntactic, as well as semantic correlates. But languages differ according to which of these correlates manifest themselves. The question regarding what is universal and what varies relative to a given linguistic category must be approached in different ways depending on which of these correlates is taken to be at the core. In what follows, I outline the goals, assumptions, and methods within three different frameworks: cross-linguistic formal semantics (7.3.1), cartography (7.3.2) and the universal spine hypothesis (7.3.3). For each of these frameworks, I briefly outline the core assumptions regarding the universality and variability of functional categories.



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

153

7.3.1. Cross-linguistic formal semantics It is often assumed that there is a straightforward semantic correlate of the functional category tense: it plays a crucial role in anchoring the event time to the utterance time. The TAM categories belong to the set of morphemes that formal semanticists take to be the “semantic glue” which is critical in the composition of meaningful sentences (von Fintel and Matthewson 2008). Thus, cross-linguistic formal semantics shares at least some aspects of its empirical domain with crosslinguistic syntax, namely the exploration of functional categories. From a semantic perspective, this exploration asks how natural languages express meanings. It is widely assumed that the meaning of a complex expression arises compositionally from the meanings of its parts and the way these parts are combined (Frege’s Principle of Compositionality). Hence, it follows that cross-linguistic semantics will have to explore the range and limits of variation as it pertains to the parts (the smallest meaningful elements) as well as the combinatorial principles (the rules of composition). The question as to what may count as a semantic universal is a relatively recent question (von Fintel and Matthewson 2008). The search for such universals is grounded in the principle of effability (Katz 1976) according to which every thought can be expressed in every natural language and that everything that can be expressed in one language can also be expressed in another. Some version of this principle has to be assumed for cross-linguistic comparison of meaning to be possible. Von Fintel and Matthewson (2008: 144) suggest that if a language lacks a simplex form to express a particular meaning, it is (usually) the case that the same meaning is expressed by a complex expression. Thus, in exploring cross-linguistic semantics the comparison set is strictly defined by meaning. It follows, that claims about semantic universals and variation are not to be equated with claims about structure. Applying this approach to the investigation of TAM, one has to proceed as follows. First, a meaning has to be identified to be the target of crosslinguistic comparison. For example, we an take the meaning of a language-specific morpheme (e.g. past or present tense) as the starting point. This is the strategy of Matthewson’s work on superficially tenseless languages. In particular, Matthewson (2006) proposes that the temporal system of St’at’imcets differs from its English counterpart in the form and meaning of the available tense morpheme. While in English there is an overt morphological contrast for present and past, St’at’imcets is analyzed as having a single zero morpheme underspecified for present or past. According to Matthewson (2006: 706) this analysis “fits in with a restrictive view of parameterization and cross-linguistic variation. We can maintain the claim that St’át’imcets parallels English in the basic structures involved in temporal interpretation. Differences between the two tense systems reduce to the lexical entries of specific tense morphemes.” Despite the general tenet of cross-linguistic semantics to compare meanings rather than structures, in practice, it is often assumed that when those compared meanings are similar they map onto the same functional projection. Morphological differences are taken to be purely superficial. This is justified if the identity of a given functional category is defined by its meaning. It points towards the need for a principled way that allows us to determine the mapping from meaning to

154

CONTEMPORARY LINGUISTIC PARAMETERS

the functional architecture, which serves to model the relation between the form, meaning, and distribution of lexical items. A second way in which a comprehensive formal typology needs to be supplemented is by means of a restrictive theory pertaining to lexical entries (cf. Boeckx 2012: 5: i) what counts as a possible lexical entry; ii) what are the sources and limits of variation in this domain; and iii) how do lexical entries relate to the functional architecture? Given the assumption that the functional architecture that comprises the clause is determined based on morphological, syntactic, and semantic properties, heuristics for the identification of variation and universality of functional categories will have to take into consideration all of these properties. Thus, in addition to variation in patterns of meanings, we also need to explore the range and limits of variation in the form-meaning mappings.

7.3.2. Cartography One of the most influential typologies in the investigation of cross-linguistic syntax is the research agenda initiated by Guglielmo Cinque known as cartography (Cinque 1999; see Cinque and Rizzi 2010 for a recent overview). It is grounded in a syntactic view of functional categories: it explores the correlations between meanings, their morphosyntactic expressions, and their structural positions. Cinque explicitly exploits the assumption that functional categories are comprised of a head and a phrasal position (the specifier). The head hosts inflectional morphology, auxiliaries, or particles while the specifier hosts adverbial modifiers. It is one of the core cartographic assumptions that each morphosyntactic feature corresponds to its own head. This heuristic lead to a finer splitting of the functional architecture. Several heads associated with tense, mood, modal, aspectual, and voice distinctions were added. Just for the TAM categories, Cinque (1999: 106) postulates about thirty distinct positions. In its strongest form, cartography postulates the universality of all of these functional categories (their heads and specifiers) and their hierarchical organization. What varies are the types of movement languages admit, and whether the position is overtly realized (Rizzi 1997; Cinque 1999). On this assumption the child during the acquisition process will not have to encounter overt evidence for the presence of a functional category: its presence is assumed to come for free via UG. In this respect, cartography sides with Matthewson’s 2006 view within the formal semantic approach. According to a weaker position, language variation may be a matter of selecting from the set of categories provided by UG (Wiltschko 2003) or bundling more than one feature into a single head (cf. Giorgi and Pianesi 1997; Bobaljik and Thráinsson 1998). Like the pre-cartographic approach, cartography emphasizes the correlation of several properties across different domains (morphological, semantic, and syntactic). For example, in the cartography of Cinque 1999 the functional category tense, splits into three different categories: tensepast, tensefuture, and tenseanterior. This contrasts with earlier proposals according to which the values



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

155

of a morphological contrast are paradigmatically associated with the same functional category. The empirical advantage of the latter view is that it predicts the complementary distribution of such paradigmatic features. Complementarity is the hallmark of identity suggesting that both present and past are instances of the same category, namely tense. Different values of tense are not associated with the same hierarchical position according to cartography. Hence they are not treated as syntactically identical. This raises the question as to what it means for a particular category to bear the same label. Given that different values of tense differ in their syntactic distribution, there is no reason to treat them as the same syntactic category. So on this approach we don’t expect there to be the same correlates of tense as in (4). But if there is no unique category tense, and if instead it is split into three subcategories, then it is not clear which category, if any, is the one that correlates with licensing grammatical subjects. Thus, the typological space predicted by the cartographic enterprise differs from the earlier approach according to which there is a single tense position with morphological, syntactic, and semantic correlates. In particular, the cluster of properties predicted to correlate from the cartographic view is as follows: (12) a. the morphosyntactic expression of F associates with F b. the lexical V (or an auxiliary) may move to F c. an adverbial modifier with a corresponding interpretation may be realized in the specifier of F Thus, if there are morphological, syntactic, or semantic correlates of the meta-category tense (including all of its values) then this would count as evidence against the view that that each of its features is associated with it own functional category. If all functional categories are part of every language then something has to be said about their realization within individual sentences. In particular, not all values of tense can co-occur within a single clause. This means that there must be a distinction between those functional categories that are interpreted but not overtly realized and those that are neither interpreted nor overtly realized. On the cartographic approach it is unclear how to model this distinction. In sum, cartography is an approach towards language variation, which takes into consideration not only the meanings but also the different morphosyntactic architectures within which they can be assembled. It allows for a fine-grained mapping of morphosyntactic structures associated with syntactic and semantic correlates. But it misses the paradigmatic character of certain features. What both approaches thus far discussed have in common is the fact that the comparison set is defined in terms of meaning. This is most explicit in the investigation of cross-linguistic semantics. However, cartography is also mostly meaning based. To see this, consider a simple question: when confronted with a newly studied language, how do we analyze its words and morphemes? How do we determine how this particular language composes its functional architecture? Cartography has it that a past tense morpheme is mapped onto tensepast, a morpheme used to express continuative aspect is mapped onto aspcontinuative, a morpheme used to express irrealis mood is mapped onto modirrealis, and so forth.

156

CONTEMPORARY LINGUISTIC PARAMETERS

This meaning-based heuristic is, however, problematic. Consider the fact that in some languages, past morphemes are not restricted to combine with verbal categories, and thus they are not restricted to the clausal functional architecture. A case in point is Halkomelem, where the past morpheme can attach to auxiliaries and to nouns (Galloway 1993, Burton 1997). (13) a. í-lh qw’eyílex tú-tl’ò AUX-PST dance DET-PRN ‘He was dancing’ b. te-l má:l-elh DET-1SG.POSS father-PST ‘my late father’

[Halkomelem]

Unless we assume a tensepast in the nominal domain (Lecarme 1996), we cannot assume that the mapping of a morpheme to the functional architecture is uniquely determined by its meaning. Furthermore, the past marker of Halkomelem, even if attached to a verbal category, has a different distribution than its English counterpart. In particular, Halkomelem past marking is optional; its absence does not trigger a present interpretation. (14) a. í-lh qw’eyílex tú-tl’ò aux-pst dance det-prn ‘He was dancing’ b. í qw’eyílex tú-tl’. aux dance det-prn ‘He is/was dancing’

(Ritter and Wiltschko, 2014: 2)

In sum, cartography allows us to define a typology that maps out the hierarchical organization of morphosyntactic features, which are in turn defined by their meaning. But it does not capture the distributional differences between morphosyntactic features that have the same meaning. This is the task of a typology that has as its goal the exploration of how form, meaning, and distributional properties relate to each other. To develop a comprehensive formal typology of natural languages, both cross-linguistic semantics as well as cartography have to be supplemented in this way. If we base typological investigations on meaning alone, we may come to the conclusion that categories are not in fact universal since meaning-based classes do not necessarily form a natural class based on their formal and distributional properties.

7.3.3. The Universal Spine Hypothesis The core goal of Wiltschko (2014) is to develop a formal typology for functional categories that does not define the comparison set for cross-linguistic investigation based on meaning. It does so by drawing on generative insights. There are two core departures from previous generative assumptions. First, adopting the work of Ritter and Wiltschko (2009, 2014), Wiltschko (2014) assumes that substantive



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

157

content (like the temporal content that defines tense) is not an intrinsic ingredient of functional categories. Instead language-specific categories are constructed on the basis of a core set of abstract universal categories (k) and language-specific units that add form and meaning as illustrated in (15). (15)

language-specific category

kP k

k … language-specific substantive content While the core universal categories are not defined based on meaning, their function is universally determined. Wiltschko’s functional architecture (16) is comprised of four distinct domains, each associated with a particular function, which is independent of the lexical category it hosts. This derives the well-known fact that there is a systematic parallelism between nominal and verbal functional architectures. The lowest domain is responsible for classification, such as Aktionsart (k:classification). Next comes an area where a point of view can be introduced, such as viewpoint aspect (k:point-of-view). The domain where tense is located is the anchoring domain (k:anchoring) which serves to anchor the event to the utterance. Finally, the topmost layer is dedicated to linking the utterance to the larger discourse (k:linking). This is the category traditionally labelled as C (for complementizer). (16)

k:linking

k:anchoring k:point-of-view UoL

k:classification

These domains echo the three domains that define clausal architecture in terms of the roles nominal arguments play. Within the lowest domain (k:classification) nominal arguments are introduced as event-participants; in the next domain arguments are assigned grammatical roles: k:point-of-view hosts grammatical object while k:anchoring hosts grammatical subjects; and in the highest domain (k:linking), arguments are marked for discourse roles (topic or focus). Thus, the correlation between functional heads and the roles that nominal constituents play within the domain of this head serves as a heuristic in the discovery and comparison of categories.

158

CONTEMPORARY LINGUISTIC PARAMETERS

This function heuristic can be used to locate the absolute position of a given grammatical category while traditional tests for structure (c-command) allow us to diagnose relative position only. If we find a linguistic element that marks event-roles, we can conclude that it is located in k:classification; if it marks grammatical roles, it is located in either k:point-of-view or k:anchoring; and if it marks discourse roles it is located in k:linking. On this assumption, language-specific categories are constructed based on the categories of the universal spine in interaction with language-specific meaningful units. The crucial source of variation is that language-specific units may associate with the spine in different ways and hence may associate with meaning in different ways. In particular, there are three ways in which the association between a unit of language and the spine may differ: In terms of the manner of association (as a head or as a modifier), in terms of the place of association (in which area of the spine), and in terms of the timing association (before or after the syntactic computation).5 These differences define the parametric space within which languages may vary. The traditional notion of a morpheme subsumes three ingredients: sound, meaning, and categorial identity. But the way these three ingredients are associated with each other may differ. For example, some morphemes spell out complex syntactic configurations while others are simplex (i.e. roots). Thus the type of meaning associated with a given morpheme may differ. Recognizing the complexity of the relation between sound, meaning, and categorial identity allows for a typology that encompasses the way these ingredients relate to each other. It will thus serve to supplement meaning-based typologies. This echoes Longobardi’s (2001a: 367) conjecture that semantic parametrization cannot be independent of syntactic parametrization. To explore cross-linguistic variation in the realm of categories, there are two ways to proceed. First, one can identify a particular meaning and explore how it is constructed across languages. For example, Tonhauser (2014) and Bittner (2014) explore how temporal systems are constructed in different languages. Are there limits to variation? Are there universal implicational relations? Alternatively, one may investigate the range of variation that arises based on differences in what type of units associate with a given domain. Thus, another way to develop a typology is to ask whether there are other language-specific categories besides tense that may be constructed based on k:anchoring. Ritter and Wiltschko (2009, in press) argue that the same abstract category INFL (k:anchoring) may be realized as tense, location, and person. This is illustrated in (17). (17)

kP



k

k:anchoring … time place participant

tense location person



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

159

Thus, while the abstract function remains constant, the content substantiating this function varies. Formally this is implemented by postulating an unvalued feature [±coin(cidence)] associated with the head of functional categories. This feature orders abstract situation arguments associated with the specifiers of functional categories: VP hosts the event situation, aspect hosts the point of view situation, and infl hosts the utterance situation.6 The substantive content intrinsic to the units of language that associate with functional heads serves two functions: it values [coin] and it determines the particular facet of the situation argument that is being ordered (time, place, or participant). Thus, present morphology values [coin] as [+coin], asserting that the event time coincides with the utterance time; past morphology values it as [-coin] asserting that the event time does not coincide with the utterance time. Similarly, locative morphology in Halkomelem Salish has the effect of ordering places (event location coincides with the utterance location or not) and person morphology in Blackfoot has the effect of ordering participants (event participant coincides with utterance participant or not). Diagnosing instantiations of k:anchoring involves diagnosing relative position (via c-command, linear ordering, and scope effects) on the one hand and absolute position (via the function heuristic) on the other hand. Diagnosing relative position involves showing that a particular category is higher than categories that instantiate k:point-of-view on the one hand and lower than k:linking on the other hand. Diagnosing absolute position involves showing that a particular category fulfills the anchoring function (i.e. relating the event to the utterance). There is no a priori reason that anchoring proceed via times and temporal relations. Instead, the same function can also be achieved by places and locational relations. Thus, instead of anchoring the event by asserting when relative to the utterance the event took place, in a location-based system, anchoring proceeds by asserting where relative to the utterance the event took place. And participant marking (i.e. person) may equally serve the anchoring function: it is asserted who relative to the utterance participants (the speaker or addressee) took part in the event. Similarly, Amritavalli and Jayaseelan (2005) and Amritavalli (2014) argue that anchoring may arise via mood. An immediate prediction of the assumption that the same abstract category (k:anchoring) can be realized in different ways is that the different instantiations of this function are in complementary distribution; and that the clausal architecture within a single language is comprised of categories that can be related to each other. In particular, given that the content of the morphology which values [coin] determines the aspect of the situation that is being ordered, it follows that two adjacent categories must have the same content; one cannot order times relative to places, for example. Similarly, in a person based system where relations among participants are established both k:anchoring and k:point-of-view have to be person-based (see Bliss et al. 2011; Wiltschko 2014 for evidence that this is indeed so). Furthermore, the assumption of a binary feature [±coin] allows for a straightforward diagnostic tool for zero morphology: it can only be detected if it enters a contrast (Wiltschko 2014).

160

CONTEMPORARY LINGUISTIC PARAMETERS

7.3.4. Conclusion The question how languages vary may be approached in different ways: i) for crosslinguistic semantics, the comparison set is based on meaning; ii) for cartographic approaches, comparison is based on meaning and morphosyntactic and syntactic considerations; iii) according to the universal spine hypothesis, meaning-based comparisons are supplemented with investigations of the form-meaning relation. And in addition to meaning-based comparisons the assumption of a universal spine makes it possible to undertake function-based comparisons. Accordingly, we can define the notion of parametric variation relative to a particular category such as for example tense in two different ways. According to the narrow definition we can explore whether there are different ways in which languages may instantiate the functional category tense. If, however, Ritter and Wiltschko (2009) and Wiltschko (2014) are correct, then tense is always constructed and thus cannot be universal. Thus, this approach is necessarily limited in its empirical scope as not all languages are expected to construct tense; rather languages are expected to differ in the inventories of the categories they make use of. What is universal on this view is the ingredients that are used to construct tense: temporal content and the universal spine. Approaching variation from this end makes for a broader definition of parametric variation and it can be tackled in two different ways: based on meaning or based on function. According to the former approach we would explore the way temporal interpretations are construed across the languages of the world. This approach may or may not be accompanied by explorations of the form-meaning correspondence. In contrast, according to the latter approach—one that defines the comparison set based on spinal functions— comparative investigations explore how languages manifest the domains defined by the spine: how they associate with meaning and sound.

7.4. A case study of TAM in Standard and Austrian German In this section, I compare the TAM categories of Standard German (SG) and Upper Austrian German (UAG). The mere fact that we have here a difference between two dialects does not imply that this is an instance of micro-variation (see section 7.6). The key differences between the two dialects concern the simple past and the subjunctive. While the verbal paradigm in SG includes a simple past (preterite), UAG has no such form. Instead, present perfect morphology is used for past contexts. The second difference concerns subjunctive marking. SG has two types of subjunctive: present and past subjunctive but they are no longer productively used (indicated by ! in Table 7.1), especially in spoken language. This contrasts with UAG, which, does not distinguish between a present and past subjunctive, but has only one form.7 Interestingly, this form is productively used.



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

161

Table 7.1  TAM categories in SG and UAG INDICATIVE PRESENT

SUBJUNCTIVE

PAST

PRESENT

PAST

IMPERF PERF

IMPERF PERF

IMPERF PERF

IMPERF PERF

SG

✓

✓

✓

✓

✓!

✓

✓!

✓

UAG

✓

✓

✗

✗

✓

✓

✗

✗

If we investigate the difference between SG and UAG within a meaning-based system we are lead to conclude that the present perfect has taken over the simple past because it is used in past contexts. This leaves unaccounted for that UAG does not have a present / past contrast in the subjunctive. I show in section 7.5 that an analysis within the USH allows us to understand the distribution of the subjunctive. In particular, Wiltschko 2014 argues that the UAG subjunctive is a different instantiation of the anchoring category. In other words, UAG subjunctive replaces SG past.

7.4.1. Standard German German has a rich system of verbal inflection including several TAM categories. In terms of the morphological contrasts, tense has two values: present and past. Present tense is morphologically unmarked, while past is formed in two ways. For weak verbs, past is marked with the suffix –te which immediately follows the verbal root and precedes person agreement (koch-tepast-st2sg); for strong verbs, past is marked by means of Ablaut (Table 7.2).

Table 7.2  Present and past tense in SG STRONG VERBS komm (“come”)

WEAK VERBS koch (“cook”)

PRESENT

PAST

PRESENT

PAST

1SG

komm-e

kam

koch-e

koch-te

2SG

komm-st

kam-st

koch-st

koch-te-st

3SG

komm-t

kam

koch-t

koch-te

1PL

komm-en

kam-en

koch-en

koch-te-n

2PL

komm-t

kam-t

koch-t

koch-te-t

3PL

komm-en

kam-en

koch-en

koch-te-n

162

CONTEMPORARY LINGUISTIC PARAMETERS

In addition, in SG the perfect may also be used to denote past events. It comes in two forms: present perfect and past perfect. It is formed by means of an auxiliary (haben “have” or sein “be”) and the participial form of the verb. The tense inflection on the auxiliary verb determines whether the complex form is present perfect or past perfect (Table 7.3).

Table 7.3  SG present and present perfect STRONG VERBS komm (“come”)

WEAK VERBS koch (“cook”)

PRESENT PERFECT

PAST PERFECT

PRESENT PERFECT

PAST PERFECT

1SG

bin gekommen

war gekommen

hab-e gekocht

ha-tt-e gekocht

2SG

bi-st gekommen

war-st gekommen

ha-st gekocht

ha-tt-est gekocht

3SG

ist gekommen

war gekommen

ha-t gekocht

ha-tt-e gekocht

1PL

sind gekommen

war-en gekommen

hab-en gekocht

ha-tt-en gekocht

2PL

seid gekommen

war-t gekommen

hab-t gekocht

ha-tt-et gekocht

3PL

sind gekommen

war-en gekommen

hab-en gekocht

ha-tt-en gekocht

As for mood, SG has three values: indicative, present subjunctive and past subjunctive. The latter two are known as Konjunktiv 1 and Konjunktiv 2. All examples thus far are in the morphologically unmarked indicative mood. Present subjunctive is formed on the basis of the simple present, which is in turn suffixed by -e. This suffix has the same distribution as the past suffix: it follows the verbal root and precedes person agreement (koch-e-st). As for past subjunctive, for weak verbs it is formed by suffixing –te to the unmarked verb. Thus, it is identical to the simple past. For strong verbs, it is formed by suffixing –e and the verbal stem is modified by means of Ablaut. For some verbs, subjunctive is associated with a dedicated Ablaut, while for others it is identical to the Ablaut of the past tense (Table 7.4).



STRONG VERBS komm (“come”)

STRONG VERBS geh- (“go”)

WEAK VERBS koch (“cook”)

PRESENT SUBJUNCTIVE

PAST SUBJUNCTIVE

PRESENT SUBJUNCTIVE

PAST SUBJUNCTIVE

PRESENT SUBJUNCTIVE

PAST SUBJUNCTIVE

1SG

komm-e

käm-e

gehe

ging-e

koch-e

koch-te

2SG

komm-e-st

käm-e-st

geh-e-st

ging-e-st

koch-st

koch-te-st

3SG

komm-e-t

käm-e-t

geh-e-t

ging-et

koch-t

koch-te

1PL

? komm-en

käm-en

? geh-en

ging-en

koch-en

koch-te-n

2PL

komm-e-t

käm-e-t

geh-et

ging-et

koch-t

koch-te-t

3PL

? komm-en

käm-en

? gehen

ging-en

koch-en

koch-te-n

THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

Table 7.4  SG mood distinctions

163

164

CONTEMPORARY LINGUISTIC PARAMETERS

For many verbal forms, there is no formal distinction between past and subjunctive. While some forms in the paradigm show that the subjunctive has to be recognized as a dedicated verbal inflection, in many actual forms the distinction is blurred (Fabricius-Hansen and Sæbø 2004). The frequent homophony between past and subjunctive is taken to be the reason for the rare use of this verbal inflection. That is, especially in spoken German, the synthetic subjunctive is replaced with a construction involving an auxiliary. There are two strategies: one is based on the present or past perfect such that the auxiliaries are marked as subjunctive (Table 7.5).

Table 7.5  The analytic subjunctive based on the perfect STRONG VERBS komm (“come”)

WEAK VERBS koch (“cook”)

PRESENT SUBJUNCTIVE

PAST SUBJUNCTIVE

PRESENT SUBJUNCTIVE

PAST SUBJUNCTIVE

1SG

sei gekommen

wär-e gekommen

hab-e gekocht

hät-te gekocht

2SG

sei-st gekommen

wär-e-st gekommen hab-e-st gekocht

hät-te-st gekocht

3SG

sei gekommen

wär-e gekommen

hab-e-t gekocht

hät-te gekocht

1PL

sei-en gekommen

wär-en gekommen

hab-e-en gekocht

hät-te-n gekocht

2PL

sei-d gekommen

wär-e-t gekommen

hab-e-t gekocht

hät-te-t gekocht

3PL

sei-en gekommen

wär-en gekommen

hab-e-en gekocht

hät-te-n gekocht

The second strategy is a complex form involving the past subjunctive form of the auxiliary werden, namely würde (Table 7.6). For the present interpretation of this subjunctive the auxiliary co-occurs with the infinitive; for the past interpretation the verb is in its participial form and is preceded by an infinitival auxiliary.

Table 7.6  The analytic subjunctive based on werden PRESENT SUBJUNCTIVE

PAST SUBJUNCTIVE

1SG

würd-e kommen

würd-e gekommen sein

2SG

würd-e-st kommen

würd-e-st gekommen sein

3SG

würd-e kommen

würd -e gekommen sein

1PL

würd-en kommen

würd -en gekommen sein

2PL

würd-e-t kommen

würd -e-t gekommen sein

3PL

würd-en kommen

würd -en gekommen sein



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

165

TAM indicative present imperf unmarked

subjunctive past

perf Aux+part

present

imperf perf i) -te Aux.past ii) Ablaut +part

imperf -e

perf aux.subj +part

past imperf aux.subj +part

perf aux.past.subj +part

FIGURE 7.1  The SG TAM system

As summarized in Figure 7.1, SG has a complex system of verbal inflection for the TAM categories: in terms of mood it distinguishes between an indicative and a subjunctive. Each of the mood categories is associated with a present and a past form. And within each of the resulting forms there is a distinction between an imperfective form, based on the simple verb, and a perfect form, formed with an auxiliary and a participle.

7.4.2. Upper Austrian German We now turn to the TAM system of Upper Austrian German (UAG). We observe two differences to the SG system. First, there is no morphological difference between present and past, neither in the indicative nor in the subjunctive. There is only one verb that has a present / past contrast, namely the auxiliary sein (“be”) (Table 7.7).

Table 7.7  UAG be sei (“come”) PRESENT

PAST

PRESENT PERFECT

1SG

bin

woa

bin gwe(s)n

2SG

bi-st

woast

bist gwe(s)n

3SG

is

woa

is gwen(s)n

1PL

san

woan

san gwe(s)n

2PL

sats

woats

sats gwe(s)n

3PL

san

woan

san gwe(s)n

For all other verbs, the semantic difference between present and past is expressed by means of the morphosyntactic contrast between imperfective (unmarked) and perfect (auxiliary + participle)—a contrast we have already seen in SG where it

166

CONTEMPORARY LINGUISTIC PARAMETERS

co-exists with the present / past contrast. Interestingly, the absence of the present / past contrast extends to the perfect. While in SG the perfect comes in two forms, a present perfect and a past perfect, UAG has a present perfect only. This is even the case for verbs that form the present perfect with be, which does have a past form (Table 7.8).

Table 7.8  The absence of past perfect in UAG STRONG VERBS kum (“come”) IMPERFECT

PRESENT PERFECT

*PAST PERFECT

1SG

kum

bin kuma

*woa kuma

2SG

kum-st

bin kuma

*woast kuma

3SG

kum-t

is kuma

*woa kuma

1PL

kum-en

san kuma

*woan kuma

2PL

kum-ts

sats kuma

*woats kuma

3PL

kum-en

san kuma

*woan kuma

This establishes that UAG lacks a contrast between present and past tense, both with simple verbs forms and with the perfect. The second difference has to do with the subjunctive, which is productive in UAG unlike in SG. In addition, we also observe a morphological difference. Given that there is no present / past contrast, we expect that in the subjunctive too, the contrast between present and past is absent. This is indeed the case. Instead there is only one subjunctive. For weak verbs, it is formed by suffixing the verb with -at. This suffix is in the same position as the past tense marker of SG: it follows the verbal base and precedes subject agreement (Table 7.9).

Table 7.9  Subjunctive marking in UAG weak verbs WEAK VERB: koch (“cook”) UNMARKED

SUBJUNCTIVE

1SG

koch

koch-at

2SG

koch-st

koch-at-st

3SG

koch-t

koch-at-t

1PL

koch-n

koch-at-n

2PL

koch-ts

koch-at-ts

3PL

koch-n

koch-at-n



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

167

With weak verbs, the subjunctive has three allomorphs: i) Ablaut, ii) suffixation of -at, iii) a combination of Ablaut and -at suffixation (Table 7.10).

Table 7.10  Subjunctive marking in UAG strong verbs STRONG VERB: kum (“come”) UNMARKED

SUBJUNCTIVE: ABLAUT

SUBJUNCTIVE: -AT

SUBJUNCTIVE: ABLAUT -AT

1SG

kum

kam

kum-at

kam-at

2SG

kum-st

kam-st

kum-at-st

kam-at-st

3SG

kum-t

kam

kum-at-t

kam-at

1PL

kum-en

?kam-en

kum-at-n

kam-at-n

2PL

kum-ts

kam-ts

kum-at-ts

kam-at-ts

3PL

kum-en

?kam-en

kum-at-en

kam-at-n

In sum, the UAG TAM system has two morphosyntactic contrasts only: one between indicative and subjunctive and one between imperfective and perfect (Figure 7.2). Thus, relative to the SG system, UAG differs in the number of forms it makes available (Table 7.11). It lacks a morphologically simple past tense. At the same time however, the UAG subjunctive is productive while the simplex subjunctive in SG is no longer productively used (see Table 7.1 above). TAM indicative imperf unmarked

subjunctive

perf Aux+part

imperf i) -at ii) Ablaut

perf Aux.subj +part

FIGURE 7.2  The UAG TAM system

Table 7.11  The difference between SG and UAG TAM systems Perfect/imperfect

Present/past

Indicative/subjunctive

SG

✓

✓

✓

UAG

✓

✗

✓

168

CONTEMPORARY LINGUISTIC PARAMETERS

Given the lack of a simple past, one way of understanding the difference between UAG and SG is to say that that they differ in the availability of the morphosyntactic category tense. In what follows I show how this morphosyntactic difference might be analyzed within the different frameworks introduced above.

7.5. Comparative analyses across frameworks 7.5.1. Meaning-based comparison One way of investigating tense in a cross-linguistic setting is to explore how temporal reference is construed. Under this meaning-based approach the main focus is on the way temporal relations are construed even in languages that lack dedicated morphosyntactic expressions for present, past, or future. From this perspective, the absence of a present / past contrast in UAG does not tell us much about the role tense plays, or how it differs from SG. It is possible that the only difference is that SG has an overt past marker whereas in UAG it is silent. The complementarity of past and subjunctive is insignificant: since they encode different meanings, the two categories are not compared. The crucial empirical question in this context is how temporal relations are construed. In SG we observe that past can be expressed with the simple past (18a) or the present perfect (18b), with no difference in interpretation (Willhelm 2007). Present is expressed with the (unmarked) simple present as in (19). Finally, a future interpretation is expressed either by means of the auxiliary (werden) with the infinitival form of the verb (20a) or by means of the unmarked form, typically classified as the simple present (20b). (18) Past a. Ich koch-te I cook-pst ‘I cooked’ b. Ich habe  gekocht I  have cooked.part ‘I cooked’ (19) Present Ich koch-e I cook-1sg ‘I’m cooking’ (20) Future a. Ich werde koch-en I will cook-inf b. Ich koch-e I cook-1sg ‘I will cook’



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

169

As shown in (21)–(23), the temporal system of UAG is similar to that of SG the only exception being the absence of a simple past tense. (21)

Past I hob kocht I have cook.PART ‘I cooked’

(22)

Present I koch I cook ‘I’m cooking’

(23) Future a. I wead koch-n I will cook.INF ‘I will cook tomorrow’ b. I koch I cook ‘I will cook’ Figure 7.3 summarizes the construal of temporal relations in SG and UAG. Temporal relations are indicated on the timeline. The morphosyntactic categories used to express these temporal relations are represented in square brackets. past     present     future SG

[past] [present perfect]

[present]

[future] [present]

UAG

[present perfect]

[present]

[future] [present]

FIGURE 7.3  Temporal construals in SG and UAG

A formal semantic analysis of this pattern involves the association of lexical entries with the morphological forms that express these temporal relations. Semantic tenses are commonly associated with the denotations in (24) (see von Stechow 2002). (24) SEMANTIC TENSES are symbols of type i which bear time variables as indices. Let c be the context of the utterance with tc the speech time. g is a variable assignment. a. ||NOW|| g,c is the speech time conceived as a point. b. ||PASTj||g,c is defined only if g(j) precedes the speech time tc. If defined, ||PASTj||g,c = g(j). c. ||FUTRj||g,c is defined only if g(j) follows the speech time tc. If defined, ||FUTRj||g,c = g(j). [von Stechow 2002: 402, (19)]

170

CONTEMPORARY LINGUISTIC PARAMETERS

From a comparative perspective, we may ask whether all languages express past, present, and future in the same way. That is, languages may differ in the number of temporal contrasts that may be expressed and in the restrictions that may be placed on any particular semantic tense. For example, we find languages with an explicit past / non-past contrast (German) while other languages have a future / non-future contrast (Comrie 1985: 49). Furthermore, as already noted, some languages have been described as fully tenseless where temporal contrasts are not grammaticized while others are described as mixed tensed language with merely optional tense marking (Smith et al. 2003, 2007; see also Baker and Travis 1997 on Mohawk). In addition, there are also languages, which express temporal relations that go beyond the three-way contrast in (24). Some languages have graded tense systems that distinguish how far back into the past or forward into the future a given event is to be located (see Cable 2013 for recent discussion). Accordingly, one might classify languages in terms of the number of temporal contrasts they encode (Figure 7.4). This type of classification is based on the implicit assumption that only a certain type of temporal expression counts. Typically this includes morphemes belonging to the verbal paradigm, auxiliaries and dedicated particles. For example, SG is typically described as either a past-non / past system (if one determines the number of semantic tenses based on the verbal paradigm alone) or as a three way tense system (if one counts periphrastic expressions including auxiliaries). Note, however that SG has the means to express graded tenses. But whereas in systems classified as graded tense systems, these graded tenses are expressed based on verb-internal modification (Ablaut and / or infixation, Cable 2013), in SG such contrasts are optionally expressed by verb-external modification (i.e. adverbials). Similarly, languages that are classified as tenseless have means to express temporal relations such as aspectual markers, temporal adverbials and the like. The difference between morphemes that are analyzed as instantiating semantic tenses of the type shown in (24) and other forms that establish temporal relations is that the denotation of the former is restricted to being a semantic tense whereas the latter have additional meaning components as well (cf. Tonhauser 2014). Thus, a typology based on semantic criteria does not generate a formal typology that allows us to compare the overt realizations of the functional category tense. past     present     future

No tense

Tenseless

2 way contrast

past

3 way contrast

past

graded tense

remote past

non-past

non-future

future

recent past

FIGURE 7.4  Classifying tense system

present

future

present

immediate future

distant future



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

171

What is missing is an explicit criterion that establishes whether or not a given morpheme is realizing the functional category tense.8 Temporal content is neither a necessary nor a sufficient condition to unambiguously establish this since there are temporal morphemes which do not map onto a functional head tense. Thus, within this approach the fact that UAG lacks an explicit contrast between present and past does not tell us anything about the properties of semantic tense. And consequently it does not tell us much about the source of the difference in the realization of tense in SG vs UAG. There is no explicit way to determine how a given lexical entry relates to the postulated functional categories that make up the clausal architecture. Most telling is von Stechow’s (2002: 402) explicit warning against “fall[ing] into the trap to believe that the Slavic imperfective morphology necessarily expresses IMPERFECTIVE. This morphology may very well be a concomitant with PERFECT or PERFECTIVE in particular contexts.” But how else should we explore parametric variation in the functional architecture of clauses?

7.5.2. The cartography of TAM Cartography aims to draw maps of syntactic configurations. Even though its empirical focus is syntactic, its ingredients—the functional categories—are semantically defined. Cinque’s (1999) hierarchy of TAM categories is given in (25). (25) Cinque’s hierarchy of functional categories: the TAM categories [Moodspeech-act [Moodevaluative [Moodevidential [Modepistemic [Tpast [Tfuture [Modirrealis [Modnecessity [Modpossibility [Asphabitual [Asprepetetive [Aspfrequentative(I) [Modvolitional [Aspcelerative(I) [Tanterior [Aspterminative [Aspcontinuative [Aspperfect(?) [Aspretrospective [Aspproximative [Aspdurative [Aspgeneric/progressive [Aspprospective [Aspsg.completive(I) [Asppl. completive [Voice [Aspcelerative(II) [Asprepetetive(II) [Aspfrequentative(II) [Aspsg.completive(II) ]]]]]]]]]]]]]]]]]]]]]]]]]]]]]] [Cinque 1999: 106] Each functional category is identified by its label, which corresponds to traditional grammatical categories (Aspect, Tense, Modality, or Mood) and a subscript, which identifies the value of that category (past, future, perfect, habitual, etc). The strategy is to map the morphemes that express the values of these categories onto their syntactic position. For our case study, we expect that SG past morphology maps onto Tpast and the future auxiliary werden maps onto Tfuture. The mapping of morphemes that correspond to these functional categories is straightforward and makes clear predictions. In particular, past tense is predicted to be structurally higher than future tense and irrealis mode. But how can we tell? The standard way of mapping within the cartographic approach combines the syntactic makeup of inflectional morphology via head-movement, and the study of the order of arguments and adjuncts with respect to different verbal forms (Cinque and Rizzi 2010: 53). Since German is a verb second language, head-movement is not telling— the verb always moves to C (Den Besten 1983). Thus, the position of the finite verb cannot serve as a heuristic to determine which of the categories in (25) are active in the language.

172

CONTEMPORARY LINGUISTIC PARAMETERS

An immediate question arises in light of the fact that in SG a past interpretation may be achieved by means of either the simple past tense or the present perfect, while in UAG present perfect is the only form available. Under a cartographic approach it is not clear how to analyze this pattern. The meaning-based heuristic would lead us to map the SG past onto Tpast. Since the present perfect has the same meaning the meaning-based heuristic leads us to also map it onto Tpast as in (26a). The morphosyntactic complexity of this periphrastic tense form may be accommodated by assuming that the auxiliary maps onto Tpast while the participle maps onto Aspperfect as in (26b). (26) a. … [Tpast tanzte …] b. … [Tpast hat [Aspperfect getanzt …]] The fact that in many varieties of German the simplex (preterite) form is replaced with the periphrastic form up to the point where the preterite no longer exists would mean that the association in (26a) is no longer available in the language. There are several problems with this view. First, it is unclear how to analyze the difference between present and past perfect in SG. That is, the Cinque hierarchy has no dedicated position for present tense. If the present auxiliary maps onto Tpast to reflect the fact that this form has the same meaning as the simplex past, then how do we analyze the past perfect? A second problem has to do with the fact that in UAG the simple past form has vanished but at the same time the subjunctive is productive. Interestingly, the UAG subjunctive is morphologically similar to the SG preterite. It is thus tempting to view the loss of past as related to the rise of subjunctive (Nübling 1997). The cartographic structure has nothing to say about this relation. Since all mapping heuristics are based on meaning, purely formal similarities between categories remain insignificant. In light of these questions we need to amend the cartographic typology with one that takes into consideration language-specific restrictions on the relation between form and meaning.

7.5.3. A comparison based on the universal spine How do we compare TAM systems within a framework that denies the existence of prefabricated categories, such as the Universal Spine Hypothesis (henceforth USH). One of the predictions of the USH is that meaning is not an intrinsic property of functional categories. Thus, purely meaning based-comparison is not likely to yield a natural class in terms of formal properties. This is because the same substantive content may associate with the spine in different ways, in different positions, and at different times. Within this approach, the difference between SG and UAG may be understood as an instance of different content associating with the same head position, namely the anchoring category: while in SG the morphological contrast between present and past realizes the anchoring category, in UAG it is the morphological contrast between indicative and subjunctive as in (27).



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

(27)

past tense subjunctive mood

kP



173

k

k:anchoring … time realis Thus, in UAG mood anchors the event to the utterance (Wiltschko 2014; see Amritavalli 2014 for Dravidian). On this analysis, subjunctive marking values the abstract coincidence feature as [-coin] and orders the event situation relative to an evaluation situation. Thus, this subjunctive asserts that the event does not coincide with the evaluation world. Accordingly, the subjunctive in UAG is predicted to differ from the subjunctive of the more familiar type. In Balkan languages, the subjunctive associates with the C-domain and consequently can co-occur with tense. The distribution of subjunctive morphology in UAG supports the view that it associates with a position lower than C (Wiltschko 2014). For example, as shown in (28), in UAG agreement associated with the anchoring category is realized on verbs, while agreement associated with the C-system is realized on complementizers. (28) Wonn-st nua du kumm-st… if-2sg only you come-2sg ‘if only you would come…’ Crucially, subjunctive marking in UAG always attaches to verbs (29a), and never to the complementizer (29b), suggesting that it associates with the anchoring category. (29) a. Wonn-st nua du kumm-at-st… if-2sg only you come-sbj-2sg ‘If only you would come…’ b. *Wonn-at-st nua es kumm-(at)-ts… if-sbj-2sg only you.pl come-sbj-2sg Evidence that the UAG subjunctive is formally and functionally similar to other anchoring categories is discussed in Wiltschko (2013, 2014). Among the most striking differences is the fact that the UAG subjunctive is possible in independent clauses and in contexts where the event is ongoing (see Wiltschko 2014 for detailed discussion and analysis). (30) a. Du, ea koch-at You, he cook-subj ‘He is cooking’ b. Du, ea dat koch-n You, he do.sbj cook-inf ‘He’s cooking’

174

CONTEMPORARY LINGUISTIC PARAMETERS

In Wiltschko (2014), I argue that in this context the evaluation world is relativized to he addressee: thus subjunctive marking asserts that the event situation does not coincide with the addressee’s evaluation world. Thus, unlike the Romance and Balkan subjunctives, the UAG is neither dependent nor does it necessarily express an unreal situation. This much establishes that what we call a subjunctive has different semantic and different formal properties. This is not surprising given that grammatical categories are constructed on a language-specific basis. A typology that is based on meaning alone will miss an important facet of language variation in that it will not make any predictions about the nature of the subjunctive in a language that lacks tense. In this way, the formal approach towards variation adds an additional facet to the comparative analysis of categories that is missing from those that are purely based on meaning.

7.6. Coming to terms with language variation within a universalist setting The goal of this chapter was to explore the range of variation available with respect to a core set of functional categories that define the clause: tense, aspect, and mood. Any such exploration must be embedded within a general framework that provides protocols for cross-linguistic comparison. Traditionally, this type of comparative work is the domain of linguistic typology, which concerns itself with the “classification of structural types across languages” (Croft 2003: 1). There are two principal goals: i) exploring linguistic diversity by classifying categories and structures ii) exploring language universals by seeking patterns that occur systematically across languages Within the generative tradition exploration of universal grammar was not always tied to the exploration of diversity. In its early days generative grammar maintained that UG is inferable from a single language. However, with the advent of the principles and parameters framework, the focus of study was broadened to explore the parameters UG. In the early days, these parameters were thought of as yes / no switches which when turned on or off are responsible for a cluster of seemingly unrelated properties. The most famous of these parameters was the pro-drop parameter (Chomsky 1981; Rizzi 1982). It was meant to account for differences among the Romance languages in the availability of an unpronounced subject: yes for Italian and no for French. Within the confinement of the Romance languages a clustering of properties is indeed observed (Jaeggli and Safir 1989). However, the empirical coverage of this parameter does not go beyond the Romance languages. Neither the comparison of dialects within a single language, nor comparison of languages across language families yields the predicted results (Baker 2008b; Gilligan 1987). Consequently, this conceptualization of parameters has been called into question.



THE FUNCTIONAL STRUCTURE OF THE CL AUSE: MAIN ISSUES

175

Instead, many assume, following Borer (1984: 3) that the availability of variation [is restricted] to the possibilities which are offered by one single component: the inflectional component. What does it mean to be part of the “inflectional component,” especially given that inflection is a language-specific morphological type? Isolating languages, for example, would fall outside of this source of language variation. The relevant insight of Borer’s claim is that parametrization is a matter of lexical entries for functional items. Hence her proposal is sometimes referred to as the lexical parametrization hypothesis. The USH falls within this general view of variation: the parametric space is defined by where, how, and when a given unit of language associates with the spine. The logic of this system furthermore allows for different substantive content to associate with the same abstract category. This view paved the way for a more fine-grained approach towards variation known as the micro-parametric view (Kayne 2005b). It is characterized by a difference in the empirical domain as well as a difference in the way cross-linguistic variation is conceptualized. As for the former, it took into consideration the differences across dialects of the same language; as for the latter, the differences are thought to be small, but still affecting the grammatical system of a particular language. If two languages (including dialects) appear to be vastly different, then we are observing the result of a series of micro-parameters. In contrast, the macro-parametric approach, in the sense of Baker (1996a) has it that “there are some parameters within the statements of the general principles that shape natural language syntax” (Baker 2008b: 354). The empirical domain over which a macro-parameter is formulated involves generalizations across different language families. Where is the type of variation discussed in the present chapter situated relative to the division between micro- and macro-parameters? The question regarding variation in the TAM categories is a question about the categorial inventory of a given language. As we saw throughout this chapter, in this area, cross-linguistic comparison is typically substance based. Exploring universals and variation in the domain of TAM typically means to explore the way tense, aspect, and mood are realized in any given language. It is for this reason that typological claims about categorial inventories are pre-dominantly associated with a functionalist perspective. While typologies based on formal properties have been developed within generative traditions, they investigate differences in structural terms such as syntactic relations (Baker 2010b; Baker and McCloskey 2007). But if so, how do we compare categories across languages in a formal setting? This is precisely what the universal spine hypothesis aims at. If the comparison of categories across languages is purely based on meaning we are missing important formal generalizations. Thus, the typical meaning-based comparisons (such as cross-linguistic semantics and cartography) should go hand in hand with a formal exploration of categories. It is not enough to compare meanings across languages; we also want to compare the relation between form and meaning. A formal typology of categorial inventories asks whether there are any systematic correlations between the meanings we find and the way they relate to form. Establishing the significance of such an enterprise was the goal of this chapter. Along the way we saw that at least within the USH, variation across languages can be small or big no matter whether the languages are dialects of the same language

176

CONTEMPORARY LINGUISTIC PARAMETERS

or from different language families. On the one hand, the functional inventory of Standard German is as different from Blackfoot (Algonquian), Halkomelem (Salish) as it is from Upper Austrian German. Standard German associates temporal substantive content with the anchoring category, Blackfoot associates participant marking, Halkomelem associates locative marking, and Upper Austrian associates subjunctive marking. According to Baker’s (2008b) definition this would count as a micro-parametric difference, given that it doesn’t concern the statements of the general principles that shape natural language syntax but instead it concerns the differences in the features of particular items (e.g. the functional heads) in the lexicon. Nevertheless, the surface effects of such variation are rather drastic. In Jakobson’s (1959) terms:“the true difference between languages is not in what may or may not be expressed but in what must or must not be conveyed by the speakers.”

Notes 1

Instead, temporal interpretations may be read off by lexical entries only.

2

In Enç’s (1987) work, anchoring is mediated via the higher functional head (C).

3

For ease of exposition, I abstract away from the role of the reference time associated with aspect (see Demirdache and Uribe-Etxebarria 1997 for discussion).

4

According to Varlokosta (1994), in Greek, nominative case is licensed in the presence of semantic tense; without a requirement for morphosyntactic tense. This may be explained on the assumption that the presence of the syntactic head Tense is conditioned by the presence of semantic tense rather than morphological tense (see Martin 1996).

5

This corresponds to the distinction between early and late insertion within Distributed Morphology (Halle and Marantz 1993).

6

This is an oversimplification: the argument associate with INFL is in fact anaphoric. But in the absence of an appropriate antecedent (as is the case with independent clauses) this anaphoric argument is interpreted deictically (see Ritter and Wiltschko 2014).

7

On the diachronic process of preterite loss, see Fischer (in press).

8

Tonhauser (2007: 858) identifies three criteria for verbal tense markers. Whether these diagnostics serve to identify tense markers that map onto the functional category tense is unclear.



8 Extended projections of V: Inner Aspect Jonathan E. MacDonald University of Illinois at Urbana-Champaign

8.1. Introduction This chapter discusses variation within the domain of inner aspect. The overall approach to this variation is decidedly syntactic, in as much as the chapter focuses on an extended projection of V: Asp(ect). Since inner aspect is a syntax-semantics interface phenomenon, there will, of course, be some discussion of the relevant semantic properties associated with those elements that contribute to the (a)telicity of the verb phrase (VP). As we will see, nevertheless, the range of elements that contribute directly to (a)telicity appear all to be located within a limited syntactic space within the VP, defined by Asp, an aspectual functional projection sandwiched between the external argument introducing head associated with Accusative Case, little v, and the lexical verb, big V.1 The elements that can contribute to the (a)telicity of the VP seem to be limited to Asp and the syntactic space contained by Asp. The main question that this chapter addresses is whether or not the variation found within inner aspect constitutes evidence in favor of classic Principles and Parameter (P&P) style parameters, namely, parameters that have an effect on narrow syntax and are associated with a cluster of properties. I will conclude that the range of variation within the domain of inner aspect does not provide conclusive support for syntactic parameters; to explain this variation, the incarnation of P&P style parameters is not required. We will see that the data can be explained in one of two ways: 1. Whether or not there is a morphological exponent of Asp, and 2. How the relevant aspectual semantic properties are lexicalized. There is no need to appeal to a difference in narrow syntactic structure. In this respect, it is tempting to conclude that variation in this domain is consistent with a strong Minimalist thesis where variation is to morphophonological properties (Sigurdsson 2004; Boeckx 2011) or restricted to the lexicon (what Baker 2008b refers to as the Borer-Chomsky conjecture).

178

CONTEMPORARY LINGUISTIC PARAMETERS

Variation in inner aspect, restricted to the lexicon—and to the morphophonological component—is consistent with the conclusion that micro-parameters are involved, if micro-parameters are related to variable properties of the lexicon (which also tend not to form clusters). Nevertheless, it is not the case that I compare closely related languages, a fruitful methodology espoused by Kayne (2005b), which often has the result of uncovering micro-parameters. Thus, although the locus and size of the points of variation in the domain of inner aspect have the flavor of microparameters, a more detailed look at more closely related languages should be carried out in order to draw this conclusion with greater confidence. On the other hand, it is likewise not clear that the parameters involved in the variation discussed here should be considered macro-parameters either, again, for methodological reasons. This is because, following Baker (2008b), one could easily conclude that, first, a sufficient number of languages is not discussed, and that, second, the languages discussed are too closely related. As Baker (2008b) points out, it may be the case that to uncover macro-parameters, languages that are distantly related should be compared, since languages that are more closely related might, in fact, be operating within the same parameter setting, and consequently, variation related to this setting would be imperceptible. Then again, as Baker (2008b) also points out, there may simply be only a few macro-parameters anyway, which might also explain why so few have been found, or claimed to exist. At the same time, however, even if macro-parameters did appear to emerge within the domain of inner aspect, the possibility still exists that they could be modeled within the paradigm of implicational hierarchies of Roberts and Holmberg (2010). Regardless, in this chapter, no clear evidence is found for macro-parameters or hierarchical relations between parameters within these domains. The variation in inner aspect addressed here is arguably best accounted for by features of the morphonological component and by features of the lexicon. It is important to point out that this chapter focuses on variation that has been previously attested in the literature on inner aspect. I will not bring forth new variation to be considered. I will not attempt to compare or contrast a multitude of unrelated languages, nor will I juxtapose closely related languages for detailed scrutiny. Consequently, this chapter can be seen as a state-of-affairs chapter on variation within inner aspect. My goal is to take a birds eye view with a Minimalist lens of the variation within the domain of inner aspect to determine if there is a need to involve P&P style parameters. Thus, necessarily some relevant details are not included, details which should be thoroughly investigated to arrive at the strongest conclusions. In this regard, this chapter can be considered a point of departure for future research to answer questions regarding the micro- vs macro-parameter question, and is, thus, also partly programmatic. The chapter is structured into two main sections, followed by a brief conclusion section. Section 8.2 deals with the properties of inner aspect. In 8.2.1 I introduce a definition of (a)telicity that will serve throughout the remainder of the chapter. I will define it in terms of (final) endpoints of events, which suffices for the purposes of this chapter. In 8.2.2, I discuss the role of the direct object NP in the aspectual calculation of the VP. It has been noted that, in some languages, the direct object NP influences the (a)telicity of the VP, and, in some languages, it does it not. Moreover, in the languages where it has been claimed not to influence the (a)telicity



EXTENDED PROJECTIONS OF V: INNER ASPECT

179

of the VP, it has likewise been claimed that the (a)telicity of the VP influences the aspectual properties of the NP. The situation seems to illustrate a directionality of aspectual influence, where the aspectual properties of the direct object NP can influence the aspectual properties of the VP, and vice versa, where the aspectual properties of the VP can influence the aspectual properties of the direct object NP. We will see evidence, however, that this directionality of aspectual influence is illusory. The evidence comes from the asymmetrical aspectual interaction between VPs and NPs, an asymmetry that holds for both languages that appear to show an aspectual influence from the NP to the VP and languages that appear to show an aspectual influence from the VP to the NP. I suggest that this apparent directionality might simply be a result of different morphophonological expressions of the same relation between NP and Asp. As I note, however, a more thorough investigation of languages that appear to show an aspectual influence from the VP to the NP (such as Slave, Malagasy, Tagalog, and Russian, as we will see) should be carried out. In section 8.2.3, I discuss the cross-linguistic range of elements that induce telicity. We see that although there are a variety of lexical items crosslinguistically that can induce telicity (English secondary predicates, Chinese verb compounds, Spanish (and Romance) reflexives and Slavic prefixes, for each one of these distinct lexical items, one of two syntactic analyses suffices to explain their telicity inducing properties; thus, the data do not demand an appeal to syntactic variation here either. Finally, in section 8.2.4 I discuss aspectual predicate classes, and simply raise the question of whether there is variation cross-linguistically. I note that one potential locus is the lack of accomplishment predicates in some languages. The answer is left open as a domain for future research, as well, since the presence or lack of variation with respect to accomplishments crucially depends on how accomplishments are defined. Section 8.3 focuses on the interaction of inner aspect with outer aspect, tense and mood. We will see in each case that the relevant properties that interact with these higher syntactic domains appear to be the endpoints of the event denoted by the VP. In 8.3.1 I discuss the interaction of outer and inner aspect; the function of outer aspect is that of focusing in on parts of the event, either including or excluding the event endpoints. In 8.3.2 I discuss the interaction of tense and inner aspect; here we see that the presence vs. absence of event bounds can be tied to the presence vs absence of entailments of completion. In 8.3.3, I discuss the interaction of mood and inner aspect. Again the presence vs. absence of endpoints underlies the patterns observed. In section 4, I briefly conclude the chapter.

8.2. Inner aspect Syntactically, critical to the range of inner aspectual phenomena is an aspectual projection, AspP, assumed by many to be located between vP and VP (Travis 1991, 2010; Baker 1997; MacDonald 2006, 2008a, 2008b.).2 As we will see, several different elements in their contribution to the aspectual interpretation of the VP depend on Asp. This includes the NP that is closest to Asp, as well as prepositional phrases and other secondary predicates that we discuss in more detail below.3

180

CONTEMPORARY LINGUISTIC PARAMETERS

First, however, in order to understand how these elements interact with Asp, a working definition of telic vs. atelic VP must be provided. This is done in section 8.2.1. In section 8.2.2, I discuss the role of the direct object NP that influences the (a)telicity of the VP.4 We will see that the way in which NPs and VPs interact may turn out to be universal, contrary to some previous claim. Any differences may only be a difference in morphophonological expression of the same syntactic relation between the NP and Asp. In section 8.2.3, I discuss a range of telicity inducing elements that have been observed in the literature. In section 8.2.4, I discuss the universality of aspectual predicate classes, noting that at least one class has been claimed not to be universally attested: the accomplishment class.

8.2.1. A definition of (a)telic VP Before beginning the discussion of the range of universal vs. cross-linguistic facts related to inner aspect, it is useful to provide some definitions, namely, telic vs. atelic VP. I take a telic VP to describe an event as containing a final endpoint (or set terminal point following Krifka 1989) and an atelic predicate to describe an event that lacks a final endpoint (or set terminal point).5 The sentences in (1) illustrate. (1) a. Mary drank a glass of wine in five minutes b. Mary carried a glass of wine # in five minutes The in-adverbial can target the final endpoint of an event described by a VP and indicate the amount of time that passes before that endpoint is reached. Thus, in (1a) five minutes passed before Mary finished her glass of wine. Once the glass of wine is completely drunk, the event itself is finished. The in-adverbial identifies a time period of five minutes before the end of that event. Since (1a) is a telic predicate, the in-adverbial is compatible. In contrast, the in-adverbial is incompatible with the VP in (1b) on the interpretation that five minutes passed before the final endpoint of the event. It is incompatible because the VP does not describe an event with a final endpoint that the in-adverbial can target.6 The predicate in (1a) is telic and the VP in (1b) is atelic. Another widely used diagnostic for the (a)telicity of the VP is the (in-)compatibility of a for-adverbial; the for-adverbial is incompatible with a telic VP and compatible with an atelic VP. Thus, a for-adverbial is incompatible with (1a) above, and compatible with (1b) above, illustrated in (2a) and (2b) respectively. (2) a. Mary drank a glass of wine # for five minutes b. Mary carried a glass of wine for five minutes Again, a telic VP is a VP that describes an event that has a linguistically encoded final endpoint. The presence of this endpoint is responsible for the (in)compatibility patterns with the in- and for-adverbials, as discussed above. The presence of this endpoint also gives rise to an entailment of completion. Thus, the entailment of completion in (1a) is that all the wine was drunk, and at the point at which all the wine was drunk the drinking of the glass of wine event finished. This is not the case for (1b), which lacks a linguistically encoded final endpoint. These differences come



EXTENDED PROJECTIONS OF V: INNER ASPECT

181

out when we add an additional phrase that implies that the same event continues to the utterance time, as in (3).7 (3) a. Mary drank a glass of wine (last night) and #she is still drinking it b. Mary carried a glass of wine (all night) and she is still carrying it With this background in mind, we can turn to see what range of elements can influence the (a)telicity of the predicate. We will first look at the role of the direct object NP.

8.2.2. The role of the direct object NP in the (a)telicity of the VP In this section, I discuss the role of the direct object NP that influences the (a)telicity of the VP. Verkuyl (1972) is often cited as the first to observe and to discuss the effect of NPs on aspectual interpretation. This effect is illustrated in (4) below in English. (4) a. John drank a beer #in five minutes/for five minutes b. John drank beer in five minutes/#for five minutes In (4a), there is a quantity, or bounded, NP (a q-NP), and the in-adverbial can target the final endpoint of the event, while the for-adverbial is incompatible.8 The predicate in (4a) is telic. In contrast in (4b) there is a non-quantity, or unbounded, NP (an NP) and there is no final endpoint to the event for the in-adverbial to target; and the for-adverbial is compatible.9 The VP in (4b) is atelic. The sentences in (4a) (4b) differ alone in the nature of the direct object, q-NP vs NP. An NP direct object gives rise to an atelic VP because it denotes an unbounded / non-delimited quantity of material, while a q-NP direct object gives rise to a telic VP, because it denotes a bounded / delimited quantity of material.10 That the NP is influencing the (a)telicity of the VP is the standard way of characterizing what is happening aspectually in sentences like (4). I will discuss the NPs aspectual influence on the VP in more detail below and refine this characterization, but for now, we focus on other aspects related to how NPs influence the (a)telicity of the VP. The direct object has a privileged role in being able to influence the (a)telicity of the VP. Observe in (5) that agentive subject NPs do not give rise to atelic VPs, in contrast to direct object NPs, as illustrated in (4). (5) a. Wildlife drank the bucket of water #for five minutes/in five minutes b. Livestock ate the bail of hay #for five minutes/in five minutes Likewise, NP complements to goal prepositions do not give rise to atelic VPs either.11 (6) a. John carried a goat into water #for five minutes/in five minutes b. John pushed a goat onto mud #for five minutes/in five minutes

182

CONTEMPORARY LINGUISTIC PARAMETERS

MacDonald (2008a, 2008b) takes these facts as evidence that there is a position within the verb phrase, below agents and above the complement to the verb related to the aspectual influence of an NP on the VP. Following previous approaches (Travis 1991) he concludes that there is an aspectual projection between vP and VP, as illustrated in (7). (7)

…vP

v

AspP Asp

VP V

…

Travis (1991, 2010) provides evidence from the verbal morphology of Tagalog for Asp. She shows that a reduplicative completive aspect morpheme appears between an accusative case assigning head with Cause semantics and the lexical verb. The data are given in (8). (8) a. n-m-pag-tumba (nagtumba) Perfective Aspect1-TM-CAUS-V ‘has knocked down’ b. n-m-pag-RED-tumba (nagtutumba) Imperfective Aspect1-TM-CAUS-Aspect2-V ‘ ‘was knocking down’

[Tagalog]

In (8a) there is a morpheme by morpheme break down of nagtumba meaning “has knocked down,” which Travis calls perfective. In (8b), there is a morpheme by morpheme break down of nadtutumba meaning “was knocking down,” which Travis calls imperfective. Morphologically, these two forms differ with respect to the presence of a reduplicative morpheme between pag – which has the semantics of Cause and assigns accusative case – and the lexical verb. Travis claims that the reduplicative morpheme heads a projection related to whether the action is completed or not. This position is Asp in (7). Note here that this is one point of variation in the domain of inner aspect: morphological spell out of the functional projection with which an NP interacts to influence the (a)telicity of the VP. In section 8.2.2 we come back to the relevance of this morphology after we discuss another potential point of variation: whether or not an NP can influence the (a)telicity of the VP as it does in English (among other languages, such as Dutch and Spanish). First, however, we discuss the syntactic nature of the relation between Asp and the NP.12 It appears that the relevant syntactic notion is one of proximity to Asp: the closest NP to Asp can interact. Let us consider the evidence. Borer (2005b: 55–7) observes (see also Hoekstra 1988: 115–16), that there need not be a selectional nor a thematic relation between the verb and the NP that influences the (a)telicity of the VP. The data in (9) illustrate this.



EXTENDED PROJECTIONS OF V: INNER ASPECT

183

(9) a.   John walked the letter to the post office   in 10 minutes/for 10 minutes b.   John walked mail to the post office #in 10 minutes/for 10 minutes c. *John walked the letter/mail In (9a) there is a q-NP the letter and the VP is telic, while in (9b) there is an NP mail and the VP is atelic. Nevertheless, as illustrated in (9c), these NPs influencing the (a)telicity of the VP are not selected by the verb, since they are only licensed by the secondary predicate. Likewise, it is not so clear that these unselected objects are thematically related to the verb. Furthermore, consider the data in (10). It illustrates that for an NP to influence the (a)telicity of the VP, it is not required to be semantically selected by the verb either. These are cases of so-called pseudo-partitives.13 (10) a. John ate a sliver of cheese b. John drank a bottle of wine

in 10 minutes/#for ten minutes in 10 minutes/#for ten minutes

Verbs like eat and drink semantically select for edible and drinkable objects respectively. This explains why, with no established discourse context, the sentences in (11) are infelicitous, as has been previously noted. (See Alexiadou et al. 2007 and references therein for data and discussion.) (11) a. #John ate a sliver b. #John drank a bottle A sliver and a bottle are not semantically selected for by eat and drink, yet, it is their q-NP status that is responsible for the telic VP in (10), since without these expressions, atelicity would result, as illustrated in (12). (12) a. John ate cheese # in 10 minutes/for ten minutes b. John drank wine # in 10 minutes/for ten minutes Thus, there appears to be no need to be thematically/lexically selected by the verb nor to be semantically selected in order for an NP to affect the (a)telicity of the VP. Syntactically, then, the relevant relation between the NP that influences the (a)telicity of the VP and Asp appears to be one of proximity to Asp. In light of this, there are two possible relations: Agree with Asp, or Move to Spec,Asp. Both relations have been proposed in the literature. For the purposes of this paper, either suffices.14 A question arises in this context: Does the NP influencing the (a)telicity of the VP do so universally or is it a point of cross-linguistic variation? Several languages have been observed to show this aspectual influence of an NP; however, it has also been observed that there are some languages where an NP appears not to influence the (a)telicity of the VP. This has been claimed for Russian (Schoorlemmer 1995; MacDonald 2008b), Czech (Filip 2005), Malagasy and Tagalog (Travis 2010), Bulgarian (Slabakova 1997), and Slave (Rice 2000). I will not discuss each of these languages here. I will, however, discuss some properties of Czech, Russian, and Bulgarian that bear on the answer to this question. Ultimately, I suggest that, while at first glance it appears that the direct object NP does not influence the (a)telicity

184

CONTEMPORARY LINGUISTIC PARAMETERS

of the VP as it does in English, there is an aspectual interaction between the direct object NP and the VP that is completely parallel to English. What remains to be seen is whether in Malagasy, Tagalog, and Slave, other languages for which it has been claimed that the direct object NP does not affect the (a)telicity of the VP, the same aspectual interaction between the NP and the VP holds. An investigation of these unrelated languages and their properties would provide an excellent opportunity to address whether a macro-parameter might be involved here. Let us begin by seeing why it has been claimed, in the first place, that the direct object NP does not influence the (a)telicity of the VP in these languages. Then, we will see why it has been claimed that, in fact, these languages seem to show that the direction of aspectual influence is from the VP to the NP direct object, the opposite of languages like English, Dutch, and Spanish. Consider the Russian data in (13). (13) a. Mary čitala {knigu/poèziju} {*za čas/v tečeniji časa} Mary read.IMPF book/poetry *in hour/during hour ‘Mary read a/the book/(the) poetry *in an hour/for an hour’ b. Mary pročitala {knigu/poèziju} {za čas/*v tečeniji časa} Mary read.PRFV book/poetry in hour/*during hour ‘Mary read a/the book/the poetry in an hour/*for an hour’ The verb in (13a) is imperfective and the in-adverbial is ungrammatical, while the for-adverbial is grammatical. In contrast, the verb in (13b) has been perfectivized by the prefix pro, and the in-adverbial is now grammatical, while the for-adverbial is ungrammatical. The VP in (13a) is atelic and the VP in (13b) is telic.15 The direct objects in each example, nevertheless, are identical, at least in surface form. Strictly speaking then, the direct object NP does not influence the (a)telicity of the VP. (See also Gehrke 2008 and references therein for the same conclusions about Czech). Nevertheless, as indicated in the translations of (13), there appears to be an aspectual influence of the VP on the NP, the inverse directionality of aspectual influence from English (Borer 2005b, Filip 1999, Krifka 1992, Schmitt 1996 and references therein). Thus, while the q-NP knigu “book” is interpreted as bounded in each case, this is not so for the NP poèziju “poetry”. Although it has an unbounded interpretation with the atelic VP in (13a), indicated by the parentheses around the, this is not the case in (13b), with the telic VP. This NP is obligatorily interpreted as denoting a bounded amount of poetry, thus the absence of parentheses around the. With telic VPs, then, there appears only to be a q-NP interpretation of the direct object. Krifka (1992: 49) notes an obligatory definite interpretation for Czech, illustrated in (14). (14) Ota vypil víno. Ota drank the wine/*wine

[Czech]

Moreover, as Filip (1999: 228) points out in Bulgarian, a language with a definite article, the article must be present when the verb is perfective (see also Slabakova 1997; MacDonald and Markova 2010; Markova 2007).



EXTENDED PROJECTIONS OF V: INNER ASPECT

185

(15) Toj izpi *kafe / kafeto. [Bulgarian] he.NOM PRFV.drink.PST *coffee.SG.ACC / coffee.DEF.SG.ACC ‘He drank up (all) the coffee’ Thus, there appear not only to be reasons to conclude that direct object NP does not influence the (a)telicity of the VP, but also that the (a)telicity of the VP influences the aspectual interpretation of the NP.16 Comparing what we have seen for a language like English and languages like Czech, Russian and Bulgarian, the situation, in terms of aspect, might be characterized as in (16).17 (16) Directionality of aspectual influence: i. V → DP (Czech, Russian, Bulgarian) ii. V ← DP (English, Dutch, Spanish) This suggests a possible point of variation in the domain of aspect, where directionality of aspectual influence is the key point of variation. We will see, however, by approaching the patterns in a certain way, there appears to be an asymmetrical aspectual relation between VPs and NPs in both sets of languages that suggests that the situation depicted in (16) is not quite correct. In fact, the patterns indicate that (17) would better represent the situation, where there is no directionality of aspectual influence, only an aspectual interaction between (a)telic VPs and (q-)NPs. (17) Lack of directionality of aspectual influence: i. V ↔ NP (Czech, Russian, Bulgarian) ii. V ↔ NP (English, Dutch, Spanish) Consider data that allow us to arrive at this conclusion. First, Borer (2005b) discusses extensively that atelic VPs allow for both a bounded and unbounded interpretation of direct object in Czech. She illustrates this with the data in (18a) and (18b) respectively.18 The data in (18) is from Borer (2005b: 164). (18) a. PilI víno, co mu jeho neunavny hostitel stale dolevalI drank.3sg wine that … ‘He drank (of) the wine that his tireless host kept pouring’ b. PsalI pet dopisu wrote.3SG five letters In (18a), víno “wine” denotes a continuously flowing unbounded quantity of wine, while pet dopisu “five letters” in (18b) denotes a bounded quantity of letters. In both cases, the VPs are atelic. Bulgarian makes this point especially clear.19 Observe in (19) that the container phrase edna čaša “a cup of” satisfies the requirement, noted above in (15), that there be a determiner when the verb is perfective. Data in (19) are from MacDonald and Markova (2010).

186

CONTEMPORARY LINGUISTIC PARAMETERS

(19) Ivan iz-pi kafe-*(to)/edna čaša kafe #edin čas /za edin čas Ivan iz-drank coffee-*(the)/a cup of coffee #one hr./in one hr. ‘Ivan drank *(the) coffee/a cup of coffee #for 1 hr./in 1 hr’ Now, when the VP is atelic, there is no longer a requirement that a determiner be present, although the presence of the container phrase is perfectly grammatical, as illustrated in (20), from MacDonald and Markova (2010). (20) Ivan pi kafe /edna č aša kafe edin č as/#za edin č as Ivan drank coffee/a cup of coffee one hour/#in one hour ‘Ivan drank coffee/a cup of coffee for an hour/#in an hour’ What this sets of facts illustrates is that there appears to be an asymmetry, where telic VPs can only occur with bounded NPs (i.e. q-NPs), while atelic VPs can combine with either bounded (q-NPs) or unbounded NPs (i.e. NPs). What appears to be unattested is an unbounded NP with a telic VP. This is depicted in Table 8.1

Table 8.1  The aspectual relation between VPs and NPs in Czech, Russian and Bulgarian Telic VP

Atelic VP

q-NP

✓

✓

NP

✗

✓

Table 8.1 illustrates an aspectual asymmetry in the interaction between NPs and VPs. Interestingly, there are data from English and Spanish that illustrate the same asymmetry. We start with English. First recall from section 2.1 above that a test for the (a)telicity of a VP is the (in)compatibility of the for-adverbial. It is compatible with atelic VPs and incompatible with telic VPs. Although this is the standard generalization, it has been noted for over thirty years (see Alsina 1999; Jackendoff 1996; Tenny 1987; Verkuyl 1993, although more recently explicitly discussed in Piñón 2008 and Smollet 2005), that the for-adverbial is felicitous with VPs with q-NPs exemplified in (21). Data from Smollet (2005: 50). (21) a. Kathleen ate an apple for a couple of minutes while talking on the phone b. Kathleen ate an apple for a couple of minutes and then she read her novel The relevant felicitous interpretation of the for-adverbial is one in which the objects are not entirely consumed (Piñón 2008) or completed (Smollet 2005), as illustrated by the example in (22), from Smollet (2005: 50), where the additional contextual information brings out the felicitous interpretation to a greater degree. (22) a. The ant ate the apple for a week before it rotted into the ground b. Steven built a Lego tower for three hours



EXTENDED PROJECTIONS OF V: INNER ASPECT

187

In as much as we can understand the felicitous interpretation of for-adverbials with these VPs that have a q-NP direct object as an event that does not necessarily reach its final endpoint, we can understand these VPs as atelic. Interestingly, as Piñón (2008) notes, while there is an atelic interpretation of VPs with q-NPs available, albeit with contextual support, NPs, in contrast, even with contextual support, unambiguously give rise to atelic VPs, as illustrated in (23), from Piñón (2008:183). (23) a. Rebecca ate applesauce #in five minutes b. Bill drank beer #in 10 minutes We find similar facts in Spanish, where a q-NP direct does not unambiguously give rise to a telic VP, as illustrated in (24), from MacDonald and Huidobro (2010).20 (24) a. Juan comió la paella durante una hora/en una hora Juan ate the paella for an hour/in an hour ‘Juan ate the paella for an hour/in an hour’ b. Juan bebió la cerveza durante una hora/en una hora Juan drank the beer for an hour/in an hour ‘Juan drank the beer for an hour/in an hour’ Observe, however, in (25) that when the direct object is an NP, only the for-adverbial is possible. (25) a. Juan comió paella durante una hora/*en una hora Juan ate paella for an hour/in an hour ‘Juan ate paella for an hour/in an hour’ b. Juan bebió cerveza durante una hora/*en una hora Juan drank beer for an hour/in an hour ‘Juan drank beer for an hour/in an hour’ What this set of facts illustrates is that there appears to be an asymmetry, where q-NPs and NPs can occur with atelic VPs, while only q-NPs can occur with telic VPs. What appears to be unattested is an NP with a telic VP. The same table that depicts the Czech, Russian and Bulgarian situation can be used to depict the same aspectual asymmetry. This is illustrated in Table 8.2.

Table 8.2  The aspectual relation between VPs and NPs Telic VP

Atelic VP

bounded NP

✓

✓

unbounded NP

✗

✓

188

CONTEMPORARY LINGUISTIC PARAMETERS

As MacDonald (2010, 2012) points out, the asymmetry in Table 8.2 characterizes both language types.21 What this entails is that in both languages the VP and the NP together play a role in the determination of the (a)telicity of the VP. Viewed from this perspective, there is arguably, no directionality of aspectual influence from the NP to the VP, nor from the VP to the NP. It appears to be illusory. It remains to be seen whether languages such as Tagalog, Malagasy, Slave, and Navajo—languages argued not to show an aspectual effect of the NP direct object—show a similar aspectual asymmetry. Answering this empirical question would shed light on whether this aspectual asymmetry is a cross-linguistically robust phenomenon. Variation in the domain of the role of the NP direct object in determining the (a)telicity of the VP may then simply boil down to morphophonolgy. While technical details remain, one might say that for many Slavic verbs the morphological spell out of the relevant aspectual property that signals telicity is on the verb, perhaps the telicizing prefixes themselves. For languages like English, the morphological spell out of the commensurate aspectual property that signals telicity takes place on the NP. The attested variation, then, could possibly simply be the spelling out different parts of the relevant syntactic relation between an NP and Asp.

8.2.3. Telicity inducing elements In this section, I discuss a range of elements whose presence gives rise to a telic VP. Specifically, I discuss secondary predicates, including goal prepositional and resultative phrases, verb compounding in Chinese, aspectual reflexives in Spanish (and Romance), and Slavic prefixes. Ultimately what we will see is that each of these seemingly different lexical items all share the common semantic property of introducing a final endpoint to the event and consequently inducing telicity. This illustrates one point of variation in inner aspect: elements that lexicalize this property can vary from one language to another. Syntactically, moreover, we will see that each of these elements have all been analyzed in one of two different ways: (i) either being introduced below V, as some class of secondary predicate, or (ii) being merged directly in Asp. Again, we will see that that there is no need in positing variation within the narrow syntactic derivation. An appeal to how final endpoints are lexicalized suffices. I begin with some elements in English. Recall that sentence (1b) from above is atelic; the VP lacks a final endpoint. Observe in (26a), that with the addition of a goal prepositional phrase (PP), the VP is now telic, as evidenced by the compatibility of the in-adverbial. Another example is provided in (26b). (26) a. Mary carried a glass of wine into the bedroom in five minutes b. Mark dragged the rope #(into the garage) in five minutes In each instance, the lack of a goal PP would give rise to an incompatible in-adverbial. Secondary resultative predicates pattern aspectually with goal PPs by being able to induce the telicity of a VP that otherwise would be atelic. Observe that the sentences in (27), which do not have a resultative secondary predicate, are atelic, as indicated by the incompatibility of the in-adverbial and the compatibility of the for-adverbial.



EXTENDED PROJECTIONS OF V: INNER ASPECT

(27) a. Mary wipe the table b. Bill hammered the metal

189

#in five minutes/for five minutes #in five minutes/for five minutes

With the addition of an adjectival secondary resultative predicate, the in-adverbials become compatible and the for-adverbial becomes incompatible, as illustrated in (28). (28) a. Mary wiped the table clean in five minutes/#for five minutes b. Bill hammered the metal flat in five minutes/#for five minutes Other languages that also make use of goal PPs and secondary adjectival resultative predicates are Norwegian, and Chinese, for instance. In Spanish, transitive verbs of ingestion (among others) are compatible with both an in-adverbial (i.e. “en”) as well as a for-adverbial (i.e. “durante”), as illustrated in (29), indicating an aspectual ambiguity. Data from MacDonald and Huidobro (2010: 51).22 (29) a. Juan comió la paella durante una hora/en una hora Juan ate the paella for an hour/in an hour ‘Juan ate the paella for an hour/in an hour’ b. Juan bebió la cerveza durante una hora/en una hora Juan drank the beer for an hour/in an hour ‘Juan drank the beer for an hour/in an hour’ One means that Spanish has to induce an unambiguously telic predicate is via the addition of a clitic pronoun that is homophonous with a reflexive pronoun: se, so-called “aspectual se.” This is illustrated in (30), where in the presence of se, the for-adverbial is no longer compatible. (30) a. Juan se   comió la paella # durante una hora/en una hora Juan seAsp ate   the paella # for an hour/in an hour ‘Juan ate the paella for an hour/in an hour’ b. Juan se bebió la cerveza # durante una hora/en una hora Juan seAsp drank the beers # for an hour/in an hour ‘Juan drank the beer for an hour/in an hour’ Dutch also has this strategy (see Hulk and Cornips 2000), as well as French and Italian (see Campanini and Schäfer 2011). In the Slavic languages it is well known that many prefixes can induce telicity in a VP, which otherwise would be atelic (see Slabakova 1997; Filip 2000; Borik 2006; Gerhke 2008, and references therein). Thus, observe in (31), the non-prefixed imperfective form of the verb is incompatible with the in-adverbial, while compatible with the bare durative phrase (corresponding to the for-adverbial). (31) a. Ja pil vino *za čas/v tečeniji časa [Russian] I drank-IMPF. wine *in hour/during hour ‘I drank wine in an hour/for an hour’

190

CONTEMPORARY LINGUISTIC PARAMETERS

b. Mary čitala knigu/poèziju *za čas/v tečeniji časa Mary read-IMPF. a book/poetry *in hour/during hour ‘Mary read a book/poetry in an hour/for an hour’ Observe in (32) that with the addition of the prefixes vy and pro, we find the reverse incompatibility patterns; the in-adverbial is compatible, while the bare durative phase is not.23 (32) a. Ja vypil vino za čas/*v tečeniji časa I drank-PRFV wine in hour/*during hour ‘I drank wine in an hour/for an hour’ b. Mary pročitala knigu/ poèziju za čas/*v tečeniji časa Mary read-PRFV a book/poetry in hour/*during hour ‘Mary read a book/poetry in an hour/for an hour’ Chinese can induce telicity by recourse to verb-verb compounding structures. Recall that a telic predicate has an inherent final endpoint and due to the presence of this final endpoint, there is an implication that the event described by the predicate is complete. Observe in Chinese in (33a) with the verb kill there appears to be no implication that killing has come about, since the intended victim still survives. In contrast, in the verb-verb compound sha-si, the implication is that the killing was complete and thus there is a murder victim. Data in (33) from Travis (2010: 128).24 (33) a. Zhangsan sha-le Lisi liangci, Lisi dou mei si Zhangsan kill-PRFV Lisi twice Lisi QU NEG die ‘Zhangsan killed Lisi twice but Lisi didn’t die’ b. * Zhangsan sha-si-le Lisi liangci, Lisi dou mei si Zhangsan kill-die-PERF Lisi twice Lisi QU NEG die ‘Zhangsan killed Lisi twice but Lisi didn’t die’ In Tagalog and Malagasy, we find a similar situation of lack of entailment of completion. Consider the example in (34) from Travis (2010: 129). (34) a. nisambotra ny alika ny zaza b. … nefa faingana loatra ilay alika PST.I.captive DET dog DET child but quick   too   that dog ‘The child caught the dog’ ‘… but the dog was too quick’ However, with the addition of the morpheme aha, completion is entailed, as illustrated in (35). (35) a. nahasambotra ny alika ny zaza b. * … nefa faingana loatra PST.a.ha.captive DET dog DET child ilay alika ‘The child caught the dog’ ‘… but the dog was too quick’ In sum, a variety of distinct lexical items can introduce an endpoint which results in an otherwise atelic VP becoming telic.25 These languages have the ability to induce telicity through the addition of an endpoint, yet, they do so via distinct lexical items.



EXTENDED PROJECTIONS OF V: INNER ASPECT

191

This type of variation appears restricted to the lexicon. Moreover, syntactically, for each of these distinct forms, one of two analyses have been proposed. The first is that there is some type of small clause that introduces the additional phrase lower than V. Hoekstra (1988) may be the first to have proposed this class of analysis, but it has also been proposed for resultative predicates, Spanish aspectual reflexives (MacDonald 2004, 2006; Campanini and Schäfer 2011), Chinese verb compounding (Travis 2010), and Slavic prefixes (Svenonius 2004). The second analysis is one where the relevant telicity inducing property is either merged directly in Asp, as claimed for Slavic prefixes by Borer (2005b) and for Tagalog and Malagasy by Travis (2010).

8.2.4. Aspectual predicate classes So far, we have only discussed the role of the direct object and the basic atelic/telic distinction. Nevertheless, the aspectual classification of a VP can be further refined. For instance, Vendler’s (1967) classification is an amply cited and often widely adopted classification. It consists of a first distinction between stative vs. dynamic/ eventive predicates, and then a further division of dynamic/eventive predicates into activities, accomplishments and achievements.26 We start with some diagnostics that distinguish statives vs. eventives. Examples of stative VPs are given in (36) and examples of eventive VPs are given in (37). (36) a. The box contains balls b. John likes cheese (37) a. John ran last night b. John drank a beer c. John dropped the book One way to distinguish stative from eventive VPs relies on the intuition that the action expressed by eventive VPs unfolds, or develops, over time, while the state expressed by stative VPs holds at every moment that the state is expressed to hold. For instance, regarding the stative VP in (38a), it is true at every moment of the fifteen minutes that the box contained the ball.27 This is not the case for the eventive VP in (38b), since if we take a slice of time small enough, which contains, say, a single step (as part of the running) it is not true that during that slice of time, John was running. It is true that he was stepping, however. (38) a. The box contained the ball for 15 minutes b. John ran last night for 30 minutes Two other diagnostics typically used to differentiate statives from eventives are: i) statives cannot appear in the progressive, while eventives can; and ii) statives can be used in present simple tense on a non-habitual, non-generic, non-playby-play interpretation, while eventives cannot. This is illustrated in (39) and (40) respectively.

192

CONTEMPORARY LINGUISTIC PARAMETERS

(39) a. *The box is containing the ball b. John is running (40) a. The box contains the ball b. #John runs While statives can be further subdivided (see for instance Dowty 1979; Maienborn 2005; Husband 2012. and references therein), it is not clear that these subdivisions are relevant to the telic/atelic distinction. That is, stative VPs are atelic. In contrast, eventives can be further subdivided into atelic and telic VPs. The class of atelic eventive VPs is often called activities, some examples of which are provided in (41). (41) a. John ran last night for an hour b. Bill drank wine last night for 15 minutes c. Mike carried a bag for an hour of the trip The class of telic VPs can be further subdivided into accomplishments and achievements. Accomplishments and achievements are typically distinguished in terms of the duration of the event that they express. Accomplishment predicates describes events with some amount of duration, while achievements described events with no duration: they describe punctual / instantaneous events. Examples of accomplishments are provided in (42) and examples of achievements are provided in (43). (42) a. John ran a marathon in four hours b. Bill drank a glass of wine last night in five minutes c. Mike carried a bag into the bedroom in two minutes (43) a. Bill dropped the book #for five minutes b. Mike spotted the plane #for five minutes c. Mary lost her keys #for five minutes28 Observe in the examples of accomplishments in (42) that the same verb is found in the examples of activities in (41). This indicates that accomplishment or activity does not refer to the verb itself, but to the entire VP, since other elements enter into its aspectual calculation. Dowty (1979) offers a host of diagnostics to distinguish each of the VP types. We have already discussed the role of the in-adverbial and the for-adverbial in distinguishing telic vs atelic VPs. Consider one diagnostic that distinguishes accomplishments from achievements: aspectual verb complementation. Accomplishments can be complements to finish, while achievements cannot (see Freed 1979 a detailed discussion). This is illustrated in (44) and (45) respectively. (44) a. Bill finished drinking the glass of wine b. John finished running the marathon c. Mike finished carrying the bag into the bedroom



EXTENDED PROJECTIONS OF V: INNER ASPECT

193

(45) a. #Bill finished dropping the book b. #Mike finished spotting the plane c. #Mary finished losing her keys An interesting question arises as to whether these aspectual VP classes are universal or not. It appears that there is little contention as to the universality of the basic stative-eventive distinction. Moreover, within the eventive VPs, there is little contention that there are telic and atelic VPs. When it comes to variation, there have been proposals that some languages lack accomplishments. An issue that arises, however, is how accomplishments are defined. If they are defined as telic VPs with duration, then, the VPs discussed above by Travis (2010) in section 2.3 that do not entail completion are not accomplishments, since they lack a final endpoint. In contrast, if we follow Travis (2010) who adopts a definition of accomplishment where incrementality is crucial, then, the languages discussed by Travis (2010) are accomplishments, since they have incrementality. MacDonald (2011) suggests that Russian telic predicates are all achievements, not accomplishments since, when atelic imperfective verbs are made telic by the addition of a prefix these verbs can no longer be complements of an aspectual verb (a.k.a. phase verb), which, as we saw in (44) to (45), is a diagnostic for accomplishment VPs. Again, however, the question of definitions still arise, since as Braginsky and Rothstein (2008) show, telic prefixed verbs in Russian are incremental theme verbs. The difference in definition has consequences for the universality of accomplishment VPs, since for verbs of consumption, ingestion, and creation, the object is consumed, ingested or created, in an incremental manner. The discussion is further complicated by claims (see, for instance, Borer 2005b) that even within the same language there is no linguistically relevant difference between accomplishment and achievement. The question arises, then, regarding not only the universality of accomplishments, but their mere existence independently of achievements. This is a fruitful domain for further inquiry that will bear on variation within inner aspect, but which pivots around how to aspectually classify VPs.

8.3. Interaction with inner aspect: Outer aspect, tense, and mood In this section, I discuss some interaction between inner aspect and three other domains: outer aspect, tense, and mood. What we will see is that the relevant elements that enter into an interaction with outer aspect, tense, and mood are the event endpoints which can been taken to essentially define the distinct aspectual VP classes, as discussed above. The most robust interactions come out when contrasting eventive VPs with stative VPs in their interaction with these higher domains. That is, VPs describing events with endpoints show a clear difference in behavior from VPs describing events without endpoints.

194

CONTEMPORARY LINGUISTIC PARAMETERS

8.3.1. Outer aspect Outer aspect is also known as viewpoint (or grammatical) aspect (Smith 1991). Smith (1991) uses the analogy of a camera lens to illustrate the role of viewpoint aspect. The camera lens can focus on the event as a whole, or zoom in to particular portions of the event. In the former, the initial and final endpoints to the event are included within the view; in the latter, the initial and final endpoints do not have to be included in the view. Klein (1995) suggests that the ordering of three times can account for the aspects (and tenses, as we will see in 3.2): an utterance time, an assertion time, and an event time. Assuming this as fundamentally correct, Demirdache and UribeExtebarria (2000, 2007, 2008, 2014) implement Smith’s (1991) analogy in their discussion of outer aspect, where “Aspects are dyadic predicates with inherent lexical meanings (WITHIN, BEFORE, AFTER), establishing ordering relations between two time spans. The relation established between the time span focused by Aspect (the AST-T) and the interval which defines the time of the described event can be one of inclusion, precedence or subsequence” (Demirdache and Uribe-Extebarria 2007: 333–4). Progressive aspect in English, on this approach, can be understood as ordering the assertion time within the event time, where the lens focuses on a period internal to the event to the exclusion of the initial and final event endpoints. Progressive is an imperfective aspect (Pancheva and von Stechow 2004), and contrasts with perfective aspect, where the event time is included within the assertion time, including both endpoints. That is, the event as a whole is within the camera lens. Consider the contrast between the progressive in (46a) and the perfective in (46b). (46) a. John was eating a hamburger. The doorbell rang b. John ate a hamburger. The doorbell rang In (46a) where the progressive is used, it is not entailed that the event of eating a hamburger reached its end. In fact, the follow up sentence in (46a) indicates that the event was interrupted, and thus, there is still hamburger left. This contrasts with the perfective in (46b), where there is an entailment that the event reached its end (see section 8.3.2 below for the relevance of tense here). In fact, the follow up sentence is only interpreted as taking place after the hamburger eating event completed. Thus, a period of time to the exclusion of the event endpoints is asserted in (46a) in the progressive, since outer aspect orders the assertion time within the event time. In contrast, in (46b) in the perfective, since the event is included within the assertion time, the entire event, necessarily including the initial and final event endpoints, is asserted. The event endpoints are crucial in order to establish the relevant ordering relations. Importantly, however, outer aspect does not modify or alter, by eliminating, for instance, the event endpoints themselves. That is, in (46a), for instance, there is still a base accomplishment VP, even when in progressive. This is supported by the compatibility of the in-adverbial in (47). This is not unexpected on Smith’s (1991) analogy, since the focus of a camera lens, likewise, does not modify or alter the object within its focus.29 (47) a. John was eating a hamburger in ten minutes, when… b. John ate a hamburger in ten minutes



EXTENDED PROJECTIONS OF V: INNER ASPECT

195

Finally, as is well known, putting VPs in the progressive is a diagnostic for whether an event has duration. Achievements, which denote events whose endpoints occur (near) simultaneously, and statives, which denote states, which lack event endpoints altogether, both tend to be ungrammatical in the progressive (on a non-habitual interpretation). Examples of stative and achievement VPs are in (48a) and (48b) respectively.30 (48) a. *John is having a headache b. *John is losing his keys

8.3.2. Tense entailments revisited In this section I discuss some limited interaction between tense and inner aspect. Recall Klein’s (1995) approach that assumes the existence of an utterance time, an assertion time, and an event time. The ordering relation on this approach is one where the utterance time and the assertion time are directly ordered relative to each other, and the assertion time and the event time are directly ordered relative to each other. In contrast, the utterance time and the event time are not directly order relative to each other; they are indirectly ordered via their orderings relative to the assentation time. Following Demirdache and Uribe-Extebarria (2000, 2007, 2008, 2014) and references therein, I assume that in the simple tenses, the assertion time and the event time are simultaneous. Thus, a simple past tense orders the event / assertion time prior to the utterance time, and the simple future tense orders the event / assertion time after the utterance time. Assuming that the event time span is determined by the event endpoints (MacDonald 2008b), and event endpoints can account for aspectual VP class, we might expect an interaction between tense and aspectual VP class. More specifically, if past tense orders the event / assertion time prior to the utterance time, in the perfective, where the whole event falls within the view of the camera lens, all endpoints of the event should be ordered prior to the utterance time. This leads to certain expectations regarding the completion of the event. Concretely, if the event has a final event endpoint, then in the perfective, there should be an entailment that the event finishes before the utterance time. If the event does not have a final endpoint, there should be no entailment that the event finishes before the utterance time. Consider first the accomplishment in (49). (49) a. John drank a beer last night b. #And he is still drinking it now The continuation of (49a) in (49b) is infelicitous, because the final event endpoint of drinking a beer is ordered prior to the utterances time. Thus, the event is entailed to be completed before the utterance time. The continuation in (49b), however, expresses that the event does continue to the utterance time, leading to a contradiction. Now recall the observations from Travis (and others) discussed above in section 2.3 where the “completive” interpretation of certain VPs in languages like Malagasy and Tagalog appear not to be entailments. One example from Chinese is repeated below in (50).

196

CONTEMPORARY LINGUISTIC PARAMETERS

(50) a. Zhangsan sha-le Lisi liangci, Lisi dou mei si Zhangsan kill-PRFV Lisi twice Lisi QU NEG die ‘Zhangsan killed Lisi twice but Lisi didn’t die’ b. *Zhangsan sha-si-le Lisi liangci, Lisi dou mei si Zhangsan kill-die-PRFV Lisi twice Lisi QU NEG die ‘Zhangsan killed Lisi twice but Lisi didn’t die’ While in (50a) there is no entailment of that the victim is murdered, in (50b) with the addition of the verb compound sha-si, “there is.” This contrasts with English, as the English translations indicate, and as discussed above, where there is an entailment. In terms of final endpoints, there is a fairly straightforward way to understand the contrast. English verb kill, as part of its lexical specification, has a final endpoint, whereas the Chinese verb sha does not. Once that endpoint is added, the same entailments go through. Chinese verbs are not unique in this respect. In English, there are also examples where the VP denotes an event that lacks a final endpoint. For these VPs, then, like the Chinese examples in (50), we expect that there will be no entailment that the event is completed prior to the utterance time. In fact, MacDonald (2008b) observes that this is exactly what we find with activities and stative VPs. Observe that the activity VP in (51a) does not entail that the beer drinking event completed before the utterance time, as there is no contradiction in the follow up sentence in (51b). For the stative VP in (52a), there is likewise no entailment that the state must no longer hold at the utterance time. (51) a. John drank beer at the bar last night b. In fact, he is still drinking beer at the bar now (52) a. John was sick last night b. In fact, he is still sick now

8.3.3. Mood entailments A parallel set of facts can be found between the interaction of inner aspect and mood, as observed by Zagona (1990b). Zagona (1990b) observes that if the complement of an epistemic modal is stative the time of the state can be simultaneous with the modal evaluation time, or forward shifted relative to the modal evaluation time. In contrast, if the complement of an epistemic modal is eventive only a forward shifted interpretation is available. These contrasts are illustrated in (53) and (54), from Stowell (2004: 624). (53) a. John must/should be in class today (simultaneous or future-shifted) b. Joe must/should leave today (only future-shifted) (54) a. John could/may be at home b. Joe could/may take the train

(simultaneous or future-shifted) (only future-shifted)



EXTENDED PROJECTIONS OF V: INNER ASPECT

197

The same style of analysis that relies on ordering relations can apply here, following Demirdache and Uribe-Extebarria’s (2008) analysis of modal evaluation times. They base their analysis on Condoravdi (2001) who takes a modal evaluation time to be an open ended interval that extends indefinitely into the future. On the assumption that the modal evaluation time of an epistemic is the speech time, it appears that the event time must be ordered relative to the speech time. Thus, the patterns that emerge are essentially the patterns found when there is no modal, as in (55), where, as observed above in section 8.2.5. stative predicates can be used in present simple tense on a non-habitual, non-generic interpretation, non-play-by-play interpretation, while eventives cannot. (55) a. The box contains the ball b. #John runs The relevant properties of the VP that interact with the structurally higher domains of outer aspect, tense and mood, are the event endpoints. As we have seen, the most robust interactions come out when contrasting eventive VPs with stative VPs. That is, VPs describing events with endpoints show a clear difference in behavior from VPs describing events without endpoints.

8.4. Conclusion The central focus of this chapter has been on an extended projection of V: Asp. As I have discussed, there are a variety of different elements that interact with this aspectual functional projection to give rise to the phenomena of inner aspect. One such element is the direct object NP. We have seen that the apparent direction of aspectual influence from the NP to the VP observed in some languages is just that, apparent. Likewise, the apparent opposite direction of aspectual influence observed in some languages is also just apparent. I have concluded that there is no narrow syntactic variation regarding the relation between the direct object NP and Asp. It is likely that this variation can be relegated to the morphophonological component. We also saw that languages vary respect to the how they lexicalize final event endpoints; yet, how the lexical item contributes to the (a)telicity of the VP seems syntactically to reduce to one of two syntactic relations with Asp: i) direct merger with the head Asp, or ii) merger via a secondary predicate, below Asp. Thus, it is not clear that variation here takes place in the narrow syntax either. It appears to be located in the lexicon. Based on these observations, it is not clear that within the domain of inner aspect P&P style parameters need to be invoked. In this respect, a strong Minimalist hypothesis regarding variation is consistent with the data.

198

CONTEMPORARY LINGUISTIC PARAMETERS

Notes 1

Within a framework that adopts acategorial roots whose verbal properties arise via merger with a little v (Marantz 1996, 1997; Arad 2003; Embick and Noyer 2004; Harley 2005 and others), Asp would presumably merge on top of this v, but still below the external argument introducing head.

2

Other proposals exist. See, for instance, Ramchand 1993, 2008; Borer 1994, 1998, 2005b; Ritter and Rosen 1998, 2000; Harley 2005.

3

This chapter focuses decidedly on the syntactic patterns of elements that enter into the aspectual calculation of the predicate. There are several classes of semantic analyses of each of these phenomena. See for instance, Krifka 1989, 1992; Filip 1999; Zucchi and White 2001; Rothstein 2004; Zwart 2005; Borik 2006; Gerhke 2008; Beavers 2008b, 2012a, 2012b; Verkuyl 1993; and references therein. In an architecture where syntax precedes semantics and provides structured representations for interpretation, it might not be the case that the semantics component needs to account directly for variation in inner aspect. That is, if variation results from morphophonlogical features, then there is no effect on syntax or on semantics. If variation results from how semantic features are lexicalized, then the only way we might see variation in the semantics is if these semantic features combine in a different way, or if a language simply does not lexicalize them. In section 8.2.3, I discuss how endpoints, which give rise to telic predicates, are lexicalized in different languages. The existence of a language that has no lexical items that induce telicity is not known to me.

4

I will refer to NPs, and not DPs, since I assume that the relevant property of a nominal that interacts with the aspectual properties of the VP is not at the D level; it is lower, following Borer (2005b). See also MacDonald (2008). As one piece of evidence, note that the presence of a numeral within a DP gives rise to a telic predicate and it appears to the right of the definite determiner: the two beers.

5

Another way to characterize telic vs. atelic is in terms of quantizedness vs. homogeneity, or quantity. See Krifka 1992 and Borer 2005b for formal characterizations. A quantized/quantity predicate gives rise to a final endpoint, while a homogeneous predicate does not. More recently, scales have been instantiated in defining (a)telic VPs, initially to account for degree achievements and their variable telic-atelic interpretation. See Hay, Kennedy, and Levin 1991; Kennedy and McNally 1999; Piñón 2008; Beavers 2008b, 2012a, for the discussion of scales and their relation to (a)telicity.

6

A distinct interpretation has been noted for the in-adverbial in both (1a) and in (1b), one where the initial endpoint of the event is targeted. This becomes more salient when these sentences are used with will. Mary will drink a glass of wine in five minutes. Mary will carry a glass of wine in five minutes. In each case, a period of five minutes passes before the event described by the VPs begin. Importantly, only the final endpoint is relevant for distinguishing a telic from an atelic predicate. Other diagnostics for (a) telicity have been used in the literature as well. See Dowty (1979: Ch.2) for a list.

7

These tense entailments will be discussed in more detail in section 8.2.3 below, where we will see that comparable verbs in different languages do not have this entailment. This is one place where we see language variation. This is also an illustration of the interaction between tense and inner aspect, as discussed below in section 8.3.

8

This is a bit of a simplification. It has been noted repeatedly (see Alsina 1999;



EXTENDED PROJECTIONS OF V: INNER ASPECT

199

Jackendoff 1996; Tenny 1987; Verkuyl 1993; Smollet 2005; Piñon 2008), that the for-adverbial is, in fact, compatible with VPs with q-NPs like those in (4a), where telicity is expected. In these cases an interpretation in which the event proceeded for the time denoted by the for-adverbial arises, but the event does not necessarily come to an end. I discuss the relevance of this fact to variation below. 9

The q-NP vs NP notation is aimed at reflecting Borer’s (2005b) approach where NPs that give rise to telic VPs have a particular aspectual property, q—for quantity—and NPs that do not lack that property. Alternatively, there is a bivalent feature involved (+/-q-NP), as in MacDonald 2006, 2008a, 2008b.

10 There is an extensive literature on the correct semantic characterization of the nouns that give rise to atelic vs. telic VPs. See Verkuyl 1972; Krifka 1989, 1993; Jackendoff 1991; Zucchi and White 2001; Borer 2005b; and references therein, for discussion. 11 There are two potential distinct interpretations of the for-adverbial here. One is an iterative interpretation where in (6a), for instance, the goat was carried repeatedly into the water for a period of five minutes. This iterative interpretation is only available for telic predicates. The second interpretation is one in which the goat was in the water for a period of five minutes, after having been carried there. I set these readings aside throughout this chapter. 12 There have been both semantic accounts of this relationship (for instance, Krifka 1989, 1992; Jackendoff 1996; Piñón 2008; Verkuyl 1972, 1993; Beavers 2008b) and syntactic accounts of this relationship (for instance, Borer 1994, 2005b; van Hout 1996, 2000; MacDonald 2008a, 2008b; Ritter and Rosen 1998, 2000; Thompson 2006; Travis 1991, 2000, 2010). The focus of this chapter is on the syntactic accounts. 13 It is, of course, possible that these expressions can satisfy the selectional properties of a verb, as in John broke a bottle of wine, since what is breakable here is the bottle, not the wine. Importantly this is not the case in (10). 14 The verdict is still out regarding what the relevant relation is. As Husband (2012: 79, n.10) points out, however, if Move to Spec,Asp were the right relation, one would not expect that extraction from said NP would be licit, given the generalization that moved constitutes do not allow extraction (Chomsky 1986b; Takahashi 1994; Nunes and Uriagereka 2000; Merchant 2001; Lasnik 2001b; Stepanov 2007). Observe, however, that extraction is allowed out of the NP that induces a telic VP: What did John eat a bag of in five minutes? 15 It is not the case that there is a one-to-one correspondence between (a)telicity and (im) perfectivity. That a one-to-one correspondence does not exist has been noted more than once. See, for instance, Borik (2006). 16 There have been a variety of proposals regarding these differences between English and Russian (and other Slavic languages). Filip and Rothstein (2005) propose a semantic telicity parameter based on a maximalization operator that applies at the VP in Slavic, but at the V in Germanic. Slabakova (1997), Rice (2000), and Travis (2010) also assume that scope can account for the apparent directionality of aspectual influence. MacDonald (2008b) suggests that the differences between Russian and English with respect to the aspectual influence of the NP on the VP can be explained by the presence of an aspectual function head (Asp) in English and its absence in Russian. Given the conclusions of the present chapter, these proposals do not seem tenable. 17 I include Dutch and Spanish here, although we have not seen the data. Verkuly (1972) discusses Dutch. We discuss some of the properties of Spanish below.

200

CONTEMPORARY LINGUISTIC PARAMETERS

18 The same holds in Russian, as indicated by the translations in (13). 19 First, note that it may be possible to leave the definite determiner out when the direct object is focused. Since focused constituents are definite, it is not clear that this is a counterexample to the generalization. Second, the obligatory presence of the determiner is limited to incremental theme verbs, as illustrated by the following prefixed holistic theme verb: S-mlja brašno #edin č as/za edin č as. “He ground flour #for an hour/in an hour.” 20 Armstrong (2013: 112 n.11) notes that there are speakers who only accept the en-adverbial in the presence of so-called aspectual se. See discussion of examples (29) and (30) below. 21 An independent question is how current theories, both syntactic and semantic can handle this aspectual asymmetry. See MacDonald (2010, 2012) for some discussion. 22 There appears to be some variation with respect to the compatibility of the en-adverbial. See footnote 20. 23 It is generally accepted that Slavic prefixes can induce telicity, but, it is also known that not all prefixes induce telicity. See footnote 8. Additionally, there are unprefixed verbs that are also telic. 24 Japanese also shows the same property. This is one point of variation that occurs cross-linguistically. 25 Dutch, German, and English also have aspectual particles, as in English John ate up the cake. Importantly, however, in English at least, John ate the cake in five minutes illustrates that often, the predicate to which the particle is added is telic independently of the particle. In Dutch, however, the predicate tends to be ambiguous between a telic and an atelic interpretation; in the presence of the particle, only a telic predicate is found. 26 Another class proposed is semelfactive (Smith 1991). I will not discuss semelfactives here, since it is not clear how they are different from (iterated) achievements. See more recently Beavers (2013) for an aspectual classification based on scales of change and affectedness. 27 This is referred to as the subinterval property. See Dowty (1979). 28 There is a felicitous interpretation of the for-adverbial here where the keys were in a state of being lost for five minutes. Some telic VPs give rise to a resultant state that can be idenitifed by for-adverbials, some cannot. For instance: John built the house #for a month. 29 As expected, with an activity in the progressive, the in-adverbial is incompatible: *John was eating hamburger in ten minutes, when … 30 Some stative VPs can take on an eventive interpretation in the progressive, while some cannot: (i.) The baby weighs 7 lbs. (ii.) The doctor is weighing the baby. Likewise, in the progressive some achievements allow for a durative event leading up to the punctual achievement: i.e. John is reaching the summit. For discussion on a variety of topics related to statives, see, for instance, Bach 1981; Dowty 1979; de Swart 1998; Maienborn 2003; Katz 2008; Husband 2012, among others.

9 Null subjects* Roberta D’Alessandro Leiden University Centre for Linguistics

9.1. The Null-subject parameter Null-subject languages (NSLs) are those languages that can leave the subject of a sentence not expressed overtly. One such language is, for instance, Italian, which is a quite well-behaved NSL, since virtually any clause can be uttered leaving the subject unexpressed. In examples (1) and (2) you see the difference between a sentence in Italian and the corresponding sentence in English, which is by contrast a wellbehaved non-NSL. The subject must be expressed in English, but not in Italian (with the exception of imperatives, like in every language, in subject relatives, as we will see below, and some diary forms). (1) (Voi) state leggendo un libro You.PL are reading a book

[Italian]

(2) *(Youpl) are reading a book

[English]

In canonical NSLs, null-subjecthood is independent of the type of clause. Interrogative sentences, as well as exclamatives, negatives, embedded sentences can feature a null subject. This is illustrated through the following examples from Italian: (3) ___ Vai al mare? go.2SG to-the sea ‘Are you going to the beach?’ (4) ___ Sono belle! are.3PL beautiful ‘They are beautiful!’

[Italian]

202

CONTEMPORARY LINGUISTIC PARAMETERS

(5) ___ Non mangiamo carne not eat.1PL meat ‘We don’t eat meat’ (6) Penso che ___ non think.1SG that not ‘I think they can’t speak’

sappiano parlare can.3PL speak

Examples (3) to (6) also show that in Italian (and in full NSLs in general) the person-number specification of the subject is irrelevant, as there is no preference for omitting 2nd person subjects as opposed to 3rd person ones, for instance. Of course, the subject of a clause in a NSL can also be overt. The Italian sentences above with an overt subject in fact are perfectly grammatical. Observe though that pronouncing a subject in a NSL brings about some extra pragmatic effect, which we will not discuss here for reasons of space (see Laia Mayol 2010 and references thereof and Grimshaw and Samek-Lodovici 1998 for more details). By looking at the Italian data and at their English counterpart we can be tempted to conclude that the null-subject (NS) status of a language is a pretty straightforward matter. You must express the subject, like in English, or you do not have to, like in Italian. Empirical evidence shows, though, that pure NS languages are rare, and that there are a number of partial NSLs, i.e. languages allowing NSs under specific structural or lexical requirements. A survey of the known types of NSLs is offered in Section 9.1.1. An interesting research question which has been pursued very thoroughly during the Government and Binding era (Chomsky 1981 and subsequent work) is whether null-subjecthood in a language correlates with other properties of a given language. Parametrically speaking, the question is whether a parameter exists that, once set, determines all other characteristics that correlate with null-subjecthood. This line of research is explored in Section 9.2, where we introduce the classical definition of the null-subject parameter (NSP). Section 9.3 deals with linguistic variation. NSLs and the NS parameter have been the center of linguistic debate, most notably that between typologists and generativists. This section reproduces this debate, examining the empirical evidence which undermines the classical formulation of the NSP and discussing the evidence in favor or against the idea of a NSP. Next, we turn to examine some of the key features of the NSL enterprise. The first one, in Section 9.4, is the status of pro, the evidence for its existence, and the question whether we can do without it. In section 9.5 we consider partial NSLs. The other key issue that has occupied the stage for several years concerns the role of agreement in licensing null subject. Section 9.6 considers the role of rich, impoverished and null agreement in licensing NSs. After this survey of the classical issues, we move on to consider some recent attempts to rewriting the NSP in terms of clusters of micro-parameters, in Section 9.7. Finally, Section 9.8 contains the conclusions. Before turning to the content, one premise is in order. Parameters are often associated to narrow syntax. This implies that the study of the NSP is thought of primarily as a syntactic issue.



NULL SUBJECTS

203

A syntactic NSP is however not the only option. We will see that one alternative can be the parametric option of deleting the subject at PF (given specific syntactic conditions), as proposed by Roberts (2004, 2010b), for instance, or Duguine (2013). In this sense, the parameter could be a syntax-PF parameter. The concept of a syntactic NSP as proposed by Rizzi (1986) also implies the possibility for a language to feature a pro, i.e. an empty pronoun which can fill in the canonical subject position. This is also why NSLs are often referred to as pro-drop languages. In this sense, the parameter could be a morphological/lexical one, rather than a purely syntactic one. Last, the terminology used to refer to NSLs is quite varied. NSLs are also defined as pro-drop, for instance. The choice of the terminology is very much linked to the theoretical assumptions of the researcher: if the author does not believe in pro, for instance, they will never call a language pro-drop. There is also an issue of fashion. Non-inflecting NSLs are now generally referred to as radical pro-drop languages; the terms topic-drop language, or discourse-oriented NSL can also be found. In what follows, we will try to stick to the terminology used by the different authors to refer to the phenomena at issue. When introducing the language types, though, we will mention the different ways in which they are referred to in the literature. When not discussing a specific proposal, we will use the neutral term NSL.

9.1.1. Types of NSL NSLs are very different from each other, for the kind of null subject they admit and for the different structural configurations in which null subjects are allowed. In what follows, we give an overview of the main types of NSL.

9.1.1.1. Canonical NSLs Canonical NSLs, also known as full NSLs, or (full) pro-drop languages, are those languages like Italian, illustrated above, for which a full, referential subject can remain unexpressed, or null. These languages have been the main focus of research on the NSP. Among canonical NSLs we find Greek, all Romance languages excluding French, Turkish, Arabic, Berber, Hausa, and Basque.

9.1.1.2. Radical NSLs Radical NSLs, also known as radical pro-drop languages or discourse pro-drop, are those languages which can leave the subject, as well as the object, unexpressed, and do not have any verbal inflection. Many Asian languages are radical NSLs: Chinese, Japanese, Korean, Thai, and Vietnamese being the best-known. We will discuss radical NSLs and their analysis in Section 9.6. An example of radical pro-drop is offered by Chinese (Huang 1984). In (7), speaker B can reply to the question asked by the speaker A by omitting the subject (7b), the object (7c), both subject and object (7d), (7e) and (7f).

204

CONTEMPORARY LINGUISTIC PARAMETERS

(7) Speaker A: Zhangsan kanjian Lisi le ma? [Chinese] Zhangsan see Lisi LE q ‘Did Zhangsan see Lisi?’ Speaker B: a. ta kanjian ta le he see he LE ‘He saw him’ b. e1 kanjian ta LE ‘[He] saw him’ c. ta kanjian e le ‘He saw [him]’ d. e kanjian e le ‘[He] saw [him]’ e. wo cai [e kanjian e le] I guess see LE ‘I guess [he] saw [him]’ f. Zhangsan shuo [e kanjian e le] Zhangsan say see LE ‘Zhangsan said that [he] saw [him]’ (Huang 1984: 533) Radical NSLs are believed to be of a different sort than canonical NSLs, in that the omitted argument can be retrieved from the discourse, and not from grammar. We will return to this in Section 9.6.

9.1.1.3. Partial NSLs NSs are sometimes restricted to some specific structures or feature composition of subject pronouns. For example, Finnish is a partial NSL in that some referential subjects can be null, but not all: while 1st and 2nd person subjects can be omitted, 3rd person cannot, as shown by the following data: (8) a. Minä puhun englantia [Finnish] I speak.1SG English b. (Sinä) puhut englantia You speak.2SG English c. *(Hän) puhuu englantia He/she speak.3SG English d. (Me) puhumme englantia We speak.2PL English e. (Te) puhutte englantia You speak.3PL English f. *(He) puhuvat englantia They speak.3PL English (Holmberg 2005: 539)



NULL SUBJECTS

205

3rd person pronouns can be null if bound, as the following example shows: (9) Pekkai väittää [että häni/j /∅i/*j puhuu englantia hyvin] [Finnish] Pekka claims that he speaks English well (Holmberg 2005: 539) Partial NSLs have been the object of much recent research, especially by Holmberg and his group. Other restrictions to null-subjecthood can be the generic, indefinite status of the subject (Holmberg 2010a, Holmberg and Sheehan 2010), as illustrated by the following Marathi example: (10) a. Unahlyat lavkar utthavla jato [Marathi] summer-in early wake go.PRES.3SG.M ‘In summer one wakes up early’ b. Ram mhanala ki ghar ghetlə Ram say.PST.3SG.M that house buy.PST.3SG.NT ‘Ram said that he bought a house’ (Holmberg and Sheehan 2010: 125) Partial NSLs are different from expletive NSLs in that in the former subject omission is determined by syntactic conditions, while in the latter it is determined by the nature (referential or expletive) of the subject. Among partial NSLs we have Finnish and Marathi, Russian, Icelandic, Assamese, and Hebrew (according to Biberauer et al.’s 2010 classification).

9.1.1.4. Expletive NSLs Expletive NSLs are those languages in which a full referential subject cannot be null, but an expletive subject can. An example of an expletive NSL is Dutch, in (11): (11) Gisteren werd (er) door het hele dorp gedanst yesterday was there by the whole village danced ‘Yesterday, there was dancing by the whole village’ (adapted from Gilligan 1987: 80)

[Dutch]

The expletive pronoun er can be omitted in the embedded sentence. Similar expletive NSLs are German and Afrikaans. Finnish was included among the partial NSLs. However, this language also features some null expletives, as shown by Holmberg and Nikanne (2002): (12) a. Nyt (se) taas sataa [Finnish] now it again rains ‘Now it’s raining again’ b. (Se) oli hauskaa että tulit käymään it was nice that came.2SG visiting ‘It was nice that you came to visit’

206

CONTEMPORARY LINGUISTIC PARAMETERS

c. (Sitä) on ilmennyt ongelmia there have appeared problems ‘Problems have come up’ d. (Sitä) tuli kiire there came haste ‘We/they were in a hurry’ (Holmberg and Nikanne 2002) While Holmberg and Nikanne attribute this null-subjecthood to other elements, Nicolis (2008) interprets these as instances of expletive null subjects. Nicolis (2005, 2008) also shows that many Creole languages feature expletive NS: Berbice Dutch Creole, Cape Verdean Creole, Jamaican Creole, Haitian Creole, Papiamentu, and Kryol.

9.1.1.5. Non-NSLs Finally, some languages never allow subject omission. Among these languages are English and French, but also Swedish and Sindhi: (13) I den här stolen sitter *(man) bekvämt in this here chair sits one comfortably ‘One can sit comfortably in this chair’

[Swedish]

(14) *(Hik) hini kursi te aaraam-sa vahi sokdo aahen one this chair in comfort-p sit can be-pres ‘One can sit comfortably in this chair’ (Holmberg 2007: 4)

[Sindhi]

As we can see from this first overview, null-subjecthood is not a uniform phenomenon. In general, what all these languages have in common is the possibility of leaving the subject unpronounced in some given contexts without any impact on the semantics of the sentence.

9.2. The (classical) null-subject parameter The classical NSP (null-subject parameter) was first formulated by Rizzi (1982), and subsumed all ideas circulating at the time about NSLs and licensing of null subjects. In what follows we will give an overview of the observations and analyses that fed the NSP hypothesis. These analyses will be presented independently, in order to reconstruct the history of the NSP. Much of these observations are still at the core of the more recent understanding of NSL and licensing of a null subject.

9.2.1. That-t effect The first systematic generative analysis of NSLs and of the conditions for licensing null subjects is put forward by Perlmutter (1971), who notices the correlation



NULL SUBJECTS

207

between null-subjecthood of a language and the possibility of extracting the subject from an embedded clause headed by an overt complementizer; conversely, non-NSL present what is dubbed as that-t effect: the subject of an embedded clause cannot be extracted across an overt subordinating complementizer. Perlmutter brings examples of extraction from several NSLs, among which Spanish, here repeated in (15) and (16): (15) (Nosotros) hemos trabajado todo el día we have.PRES.1SG worked all the day ‘We worked all day’

[Spanish]

(16) a. ¿Quiéni dijiste que ei salió temprano? who say.PRFV.2SG THAT leave.PRFV.3SG early ‘Who did you say that left early?’ b. Las cosasi que dijiste que ei pasaron the things that say.PRFV.2SG THAT happen.PRFV.3PL ‘The things that you said that happened’  [adapted from Perlmutter 1971: 103] (15) shows that Spanish is a NSL, as the full DP subject can be omitted. (16a) and (16b) show that the subject can be extracted from an embedded sentence in the presence of an overt complementizer. This is true both for a wh-subject and for the subject of a relative clause. No difference is attested between thematic (argumental) subjects and non-thematic, expletive subjects.

9.2.2. Inversion In (1980), Kayne lays out a generalization according to which all fully NSLs also have subject inversion. Comparing Italian with English and French, Kayne observes that while Italian, a NSL, can freely invert the subject, i.e. can have a postverbal subject, English and French, non-NSL, must have an expletive in the canonical subject position in the case of free inversion. A language allows free inversion of the subject if the subject can appear after the verb, leaving the canonical subject position empty. (17) a. È arrivato Gianni b. *Est arrivé Jean c. *Has arrived John ‘John has arrived’ 

[Italian] [French] [English]

(18) a. Hanno telefonato molti amici b. Sono arrivati molti amici

[Italian]



[English]

a. *Have telephoned many friends b. *Have arrived many friends

(Roberts 2007: 28)

208

CONTEMPORARY LINGUISTIC PARAMETERS

a. *Ont téléphoné beaucoup d’amis b. *Sont arrivés beaucoup d’amis

[French] (Gilligan 1987: 77)

French and English, non-NSLs, do not allow for subject inversion. Italian, a NSL, does.

9.2.3. Thematic and non-thematic subjects In well-behaved NSLs like Italian, which do not feature expletives, the full subject can be expressed, or it can be left out. Other languages, like English, are fully non-NS, thus the subject must always be expressed, be it a full referential, argumental subject, or an expletive one. There are some exceptions also for English, one being the already mentioned imperative forms. Further exceptions are subject relatives of the kind The boy that e kissed Mary, which do not license an overt embedded subject, and coordinated structures of the kind I ate and e drank. In the early 1980s, when the NSP was taking shape, the difference was made between thematic and non-thematic (null) subjects, i.e. argumental, fully referential DP subjects and expletive subjects. This difference was drawn early on, already by Perlmutter, given the non-canonical behavior of Dutch with respect to the that-t effect. Dutch was considered a non-NSL, like the rest of Germanic. However, subject extraction did not incur in the that-t effect. In fact, Dutch can be NS, but the subject must be an expletive. In this case, it does not comply with the that-t generalization. The discussion on expletive NS languages was brought forward by Maling and Zaenen (1978), who also discussed Icelandic, which does not have null referential subjects but nevertheless has a that-t effect. effect.

9.2.4. Rich agreement The issue of inversion is taken on by Taraldsen (1980) who, starting from a very common assumption which was already made by Latin grammarians (see Roberts and Holmberg 2010: 3), attributes the possibility of extracting the subject in NSLs to their rich agreement inflection. Taraldsen links the that-t effect to the ecp, by stating that the trace of the extracted subject needs to be bound, like an anaphor, and this can only be done by rich agreement.2 According to Taraldsen, in a sentence with extraction like those examined above in (17) and (18), the trace of the moved subject must be bound, like an anaphor (but see Pesetsky 1982 for a different view). Non-NSL don’t allow extraction because the trace of the subject must be governed, but the comp position is already filled by an overt complementizer. Due to the independently assumed doubly filled comp, no other element can be in the compP, hence the trace remains unbound. NSLs have a crucial characteristic that differentiates them from the rest: they have rich agreement, which in Taraldsen’s terms is equivalent to saying that they have an agr head which is “strong” enough to be able to bind the trace of the subject.



NULL SUBJECTS

209

The role of rich agreement will come back over and over again in the course of time. We will discuss it again in Section 9.6, after examining Rizzi’s formulation of the NSP. Before going into that, though, we can notice immediately that a correlation between rich agreement and null-subjecthood, though tempting, does not always hold. In particular, it does not hold for radical NSLs, introduced in 9.1.1.2, which are heavily NS but do not have any agreement.

9.2.5. The NSP The properties of inversion, extraction from an embedded clause, and no that-t effect are separately attributed to NSLs. Rizzi (1982) attributes these three properties to a unique cause: the Null-Subject Parameter. When set to “null,” this parameter determines the co-occurrence of all three properties. According to Rizzi, the role of agr is not crucial in licensing extraction, given that extraction proceeds in two steps, the first of which being inversion. The basic ingredients of the NSP as formulated by Rizzi (1982) and largely adopted for many years to follow are: i) Rich inflection. Like Taraldsen, Rizzi (1982) attributes an important role to rich inflection, not in the sense that it is directly able to license the trace, but in the sense that a rich INFL has a pronominal status. For a null subject to be licensed, it needs to be identified, i.e. it needs to be governed by a head which is [+pronominal] and [+referential]. The NSP is expressed as in (19): (19) a. INFL can be specified [+pronoun] b. INFL which is [+pronoun] can be referential. Only a pronominal INFL can license a null subject.

[Rizzi 1982: 143]

ii) Referential and non-referential NS. Null subjects can be referential or non-referential (expletive). Languages with a pronominal referential INFL can license referential NSs. Languages without a pronominal INFL which is not referential (i.e. which does not have rich inflection) cannot license a referential NS, but they can still license an expletive NS. Dutch and German belong to this second group of NSL, called partial nsl. Recall that referential and non-referential subjects were called respectively thematic and non-thematic by Perlmutter. iii) Free inversion. As noted by Kayne, if a language is NS, it will allow free inversion of the subject. Free inversion was taken at face value by Kayne, but for Rizzi this property is fundamental, because extraction takes place after inversion (see point iv). iv) Extraction. We have seen how Perlmutter observed that subject extraction is only allowed in full NSLs. Taraldsen proposed an analysis according to which the trace of the moved subject must be licensed as if it were an anaphor, and this was done by agr in NSLs. According to Rizzi, this proposal cannot hold. To show this, Rizzi considers the licensing of embedded quantified subjects which also have to be licensed like anaphors. If embedded quantified subjects were licensed by agr, they should be

210

CONTEMPORARY LINGUISTIC PARAMETERS

able to take wide scope (given that this licensing has also an LF-effect), but this is not the case. The sentences that Rizzi examines are the following: (20) a. Non pretendo che tu arresti nessuno [Italian] NEG require that you arrest nobody ‘I require that you arrest nobody’ b. Non pretendo che nessuno ti arresti NEG require that nobody you arrest ‘I don’t require that anybody arrest you’ c. Non pretendo che ti arresti nessuno NEG require that you.2SG.ACC.CLT arrest nobody ‘I don’t require that anybody arrest you’ (Rizzi 1982: 124) In (20a), the quantified object nessuno can have wide scope, because it is licensed by the verb (infl). If the same were true for the embedded subject in (20b), the wide scope reading should also be possible for it, but it is not. This means that infl does not license the embedded subject. This in turn contrasts with the sentence in (20c), where the inverted subject has wide scope, because it is properly governed by infl. This asymmetrical behavior suggests, according to Rizzi, that the position from which the embedded subject must be extracted is the postverbal, inverted one, also for wh-subjects, just like for the negative quantifiers. If extraction takes place through inversion, languages without inversion cannot have extraction. Rizzi’s hypothesis is graphically exemplified by Gilligan (1987: 85) as follows: (21) null thematic < – - – - – - – -> infl subjects [+ref] null subject that-t nonthematic < – - – - -> infl inversion filter subjects [+pnl] violations In his 2007 monograph, Roberts brings some evidence for Rizzi’s analysis of extraction, namely agreement patterns in Florentine. Florentine is a NSL with subject clitics. If the subject occurs in an inverted position (i.e. postverbally), Florentine displays reduced agreement. The subject position is occupied by an uninflected, default, expletive subject, as exemplified in (22): (22) a. Gli ha telefonato delle ragazze [Florentine] S.CLT.M.SG has.3SG telephoned.M.SG some girls.F.PL ‘Some girls phoned’ b. Gli è venuto delle ragazze SCLT is come.M.SG some girls.F.PL ‘Some girls came’



NULL SUBJECTS

211

Observe the contrast with standard Italian, the agreement patterns of which are insensitive to the position of the subject: (23) a. Hanno Have.3PL ‘Some girls b. Sono Are.3PL ‘Some girls

telefonato delle ragazze phoned.M.SG some girls.F.PL phoned’ venute delle ragazze come.F.PL three girls.F.PL came’

[Italian]

Leaving aside the issue of participial agreement (for which the reader is referred to Kayne 1989a, 2000; D’Alessandro and Roberts 2008, and others), we observe that the auxiliary is plural in standard Italian, where it agrees with the inverted subject, while it is singular in Florentine, where the inversion of the subject triggers reduced agreement (Cardinaletti 1997; Roberts 2007). Observe furthermore that Florentine does not allow the Standard Italian agreement patterns. A plural auxiliary and a plural feminine subject clitic are not grammatical: (24) a. *Le hanno S.CLT.3PL.F have.3PL b. *Le son SCLT.3PL.F are.3.PL

telefonato phoned.M.SG venute come.F.PL

delle ragazze. some girls.F.PL delle ragazze. some girls.F.PL

[Florentine]

When the subject is extracted, only the agreement pattern without inflection in (25) is allowed: (25) a. Quante ragazze gli ha parlato con te? How-many girls SCL has talked with you ‘How many girls have talked to you?’ b. Quante ragazze gli è venuto con te? How-many girls SCL is come with you ‘How many girls have come with you?’

[Florentine]

This extends to examples with a wh-embedded subject: (26) a. Quante ragazze tu credi che gli abbia parlato?[Florentine] How-many girls you think that SCL have.3SG talked ‘How many girls do you think talked?’ b. Quante ragazze tu credi che è sia venuto? How-many girls you think that SCL be.3SG come? ‘How many girls do you think have come?’ In subject extraction contexts, thus, agreement follows the subject inversion pattern, not the preverbal subject pattern. This means that the subject must have inverted (causing agreement reduction) before extraction. These data offer empirical evidence that subject extraction is preceded by subject inversion.

212

CONTEMPORARY LINGUISTIC PARAMETERS

The NSP in its classical version elegantly accounted for many features of NSLs, for the fact that they cluster together, by only specifying the [+pronominal] and [+referential] status of the inflectional head INFL. Because of its elegance, this was one of the most cited and accepted parameters of the Government & Binding era. Its all-or-nothing cut, though, gave origin to a number of discussions, based both on its empirical, descriptive inadequacy and on its theoretical formulation. In what follows, we address the debate on the validity of the NSP, which had many participants, trying to focus on the main issues. We do not follow a strictly chronological order, but we isolate four main points and discuss the possible alternatives to the NSP. The points are: i. empirical counterevidence to the NSP (and the rules vs parameters debate) ii. the status of pro iii. non-inflectional NSLs and the role of rich, poor and impoverished inflection iv. partial NSLs After this, we present a more recent formulation of the NSP as a cluster of microparameters, as proposed by Roberts and Holmberg (2010). Observe that a new macro-parameter was recently proposed by Chomsky (2014b), going back to the correlation between the ECP and the NSP.3

9.3. Empirical counterevidence to the NSP and the rules vs parameters debate The NSP was for a long time one of the very few parameters which were agreed upon by most generativists, and therefore it was attacked quite energetically by the anti-Principles & Parameters movement. One such attack came from Gilligan (1987) who disproved the empirical coverage of the NSP, by examining 100 NSLs. Rizzi’s NSP, together with the studies that informed it, was based on a relatively small language sample. The sample selected by Gilligan included languages from all families, and was tested on Rizzi’s (as well as others’) hypotheses. The relevant features of NSLs as identified by Rizzi are, according to Gilligan, the existence of null thematic subjects (which Rizzi calls referential pronominal subjects), the existence of null non-thematic subjects (Rizzi’s non-referential NSs), the possibility of free inversion, and the existence of a that-t effect on subject extraction. According to Rizzi, the only possible languages are those showing the combinations illustrated in the following table: (27) null thematic subjects

null nonthematic subjects

subject inversion

that-t filter violation

+

+

+

+

-

+

+

+

-

-

-

-

[Gilligan 1987: 86]



NULL SUBJECTS

213

Gilligan’s results are expressed in terms of typological implications, and are reproduced in table (28) (where yes-yes means the language has the first property as well as the second; yes-no means that the language has the first property but not the second, and so on): (28) yes-yes

yes-no

no-yes

no-no

ND

pro-EXE

24

0

15

2

61

pro-SI

22

49

11

15

4

pro-THAT

5

3

2

1

89

EXE-SI

14

25

1

1

61

EXE-THAT

7

2

0

1

90

SI-THAT

4

0

3

4

89

(Gilligan 1987: 135) In (28), pro refers to a null referential (thematic) subject, exe to an expletive subject, si dubs subject inversion, that stands for the that-t effect. The table indicates the correlations between these features in a given language. pro-exe refers to a language that has both a referential pro and an expletive pro. pro-si refers to a language that has both a referential pro and free inversion, and so on. Observe that pro was introduced by Rizzi (1986) in a later reformulation of the NSL, which will be discussed in Section 9.5. From this table it appears that the only correlations proposed by Rizzi and others that actually hold in a good number of languages are those between the referential and the expletive pro (recall that a referential pro entails the existence of an expletive pro, according to Rizzi, because the infl licensing the referential pro contains a superset of the features licensing the expletive pro. While the infl licensing expletives is [pronominal] but not [referential], the infl licensing referential pro is both [pronominal] and [referential]); those between pro and inversion, and to a minor extent those between expletive NS and inversion. There is virtually no correlation between inversion and the that-t effect, and between the presence of expletive null subjects and the that-t effect.

9.3.1. That-t effect and free inversion The most striking counterevidence to Rizzi’s NSP is offered, as we observed, by the lack of correlation between the that-t effect and free inversion. One language that Gilligan considers in this respect is Papiamento, where extraction is always possible while inversion never is, as shown by the following examples: (29) a. Maria a disparse. Maria prf disappear ‘Mary disappeared’ b. *A disparse Maria

[Papiamento]

214

CONTEMPORARY LINGUISTIC PARAMETERS

(30) a. Keni bo kere ku ei a bini? who you think that PERF come ‘Who do you think that had come?’ b. e Homberi ku ta evidente ku ei a bini the.man that PROG evident that PERF come ‘The man that it is evident that had come’ (Gilligan 1987: 141) Gilligan concludes that the only possible implication is that if a language has free inversion it must have extraction, but not the other way around.

9.3.2. Thematic and non-thematic null subjects According to Rizzi’s formulation, if a language has thematic null subjects it must also have null expletives. We have seen that Gilligan’s sample shows that this is only partially true. As Camacho (2013) shows, there are cases of languages that contradict this generalization, like for instance Dominican Spanish, which is a NSL but tends to have overt expletives rather than null ones, as the following example illustrates: (31) Ello hay personas que lo aprenden bien (el inglés) [Dominican Spanish] EXPL is people.PL that it learn.PL well the English ‘There are people who learn it well (English)’ (Camacho 2013: 43) A similar phenomenon is attested in contemporary Neapolitan, where the expletive subject has become obligatory (Bichelli 1974, Sornicola 1996, Ledgeway 2000, 2009): (32) E chello mò so’ quase l’unnece and EXPL now are almost the-eleven ‘And now it’s almost eleven’

[Neapolitan] (Ledgeway 2009: 291)

(33) Chello fino a mò è mancata ’a corrente EXPL until to now is missed the electricity ‘There was no electricity until now’ (Ledgeway 2009: 294) Neapolitan and Dominican Spanish are thus also counterexamples to Rizzi’s generalization.

9.3.3. Expletive NS and free inversion Another central generalization attributable to Rizzi (1982) regards the possibility of free inversion in a language that has expletive NSs. In Gilligan’s words, if a language has an exe, it will also have inversion (si), and vice versa. As we have seen, inversion correlates with extraction. It follows, Gilligan points out, that exe implies that-t, which is Taraldsen’s proposal. On checking this implication, Gilligan finds that this also does not hold: there are several languages that have expletive null subjects but



NULL SUBJECTS

215

no extraction (i.e. they have expletive NS and that-t). Finnish and Georgian are, according to Gilligan, languages with expletive NSs and that-t, as exemplified in (34). (34) a. Mari kuuli että Pekka kirjoittaa kirjeen [Finnish] Maru hear that Pekka writes letter ‘Mari heard that Pekka is writing the letter’ b. Mitäi Mari kuuli että Pekka kirjoitaa ei? what Mari hear that Pekka writes ‘What did Mari hear that Pekka is writing?’ c. *Kukai Mari kuuli että ei kirjoittaa kirjeen who Mari hear that writes letter  [Gilligan 1987: 143] While extraction is possible for embedded objects, it is blocked for embedded subjects in the presence of an overt complementizer, as exemplified by (34). Gilligan’s (1987) dissertation constitutes a challenge to the very existence of a NSP, at least in Rizzi’s formulation, for two reasons. The first one is that there are languages that constitute clear-cut counterexamples to Rizzi’s generalizations, as we have seen. The second, more stringent, problem concerns the co-occurrence of all these properties in a given language. If a language is set so that the NSP is “active”, i.e. if a language is NS, all these properties must be present at the same time. There cannot be any exception. This is a general problem for the classical theory of parameters, and has been repeatedly discussed by Chomsky, also recently (Chomsky 2013b), and also in relation to the question whether Greenberg-style typological universals can be the results of principles and parameters. Chomsky’s answer to this question is no, because a principle cannot have exceptions, and once a parameter is set, there cannot be partial counterexamples like those found by Gilligan.

9.3.4. The Newmeyer vs Roberts and Holmberg debate Gilligan’s work is taken as a basis for Newmeyer’s (2004, 2005) attack to the Principle & Parameters enterprise as a whole.4 Newmeyer (2004, 2005) points out that only some of the correlations postulated by Rizzi actually hold, and concludes that the parametric enterprise has failed.

9.3.4.1. Rules vs Parameters Starting from the empirical evidence we have just reviewed, Newmeyer claims that rules are more efficient than parameters to describe linguistic variation. One way to account for the difference between NSLs and non-NSLs, for instance, is through language-specific rules. The difference between a NS Spanish sentence and a non-NS English one just boils down to language-specific rules. The Spanish rule can be “move the wh-phrase of the embedded clause to the front of the sentence,” as in (35); the English rule can say “don’t move the wh- of the embedded clause to the front of the sentence,” as in (36):

216

CONTEMPORARY LINGUISTIC PARAMETERS

(35) Spanish rule (in contemporary syntactic terms): [CP… that [CP wh TP..] > [CPwh- that [CP __ TP..] (36) English rule (in contemporary syntactic terms): [CP… that [CP wh TP..] > *[CPwh- that [CP __ TP..] According to Newmeyer, this is the way to go. Rules can cover the empirical database much better and can account for the microvariation among NSLs (to which we return in the following sections) more accurately. Roberts and Holmberg (2005) reply that rules make no cross-linguistic predictions. The fact that Spanish has a rule allowing extraction does not tell anything about the same rule (or a different one) applying in Italian, or in Chinese. The predictive power of P&P is something that shouldn’t be given up.

9.3.4.2. Empirical inadequacy We have seen that the classical NSP is descriptively inadequate. Newmeyer takes this point as a sign of its failure. Roberts and Holmberg reply, though, that all languages have been considered by Gilligan only with respect to the NSP, putting aside all other aspects that could make the NSP effects opaque. There could be other parametric settings interfering with the NSP and thus shading the effect of the properties correlating to the NSP. While Gilligan did leave aside languages with no inflection and checked the sample carefully, he did not study in detail the whole syntax of the languages he considered, offering thus a necessarily biased survey. Furthermore, the fact that some of the correlations that were predicted on purely theoretical grounds do hold is seen by Roberts and Holmberg as an important result and a sign that parameters are on the right track. The facts remain, according to Newmeyer, that nobody can point out a set of agreed-upon parameters. And that the NSP, as originally formulated, does not account for the range of data we can find.

9.4. The status of pro Rizzi (1986) reformulates the NSP, proposing explicitly the presence of a null pronoun, pro, licensed by INFL of a specific parametric kind under government. pro is a pronoun which is not pronounced. It has all the features of a full DP subject (i.e. it can bear Nominative case, for instance, or it can be singular or plural, etc.) but it is phonetically empty. If INFL is pronominal, it will license a pro. The NSP is reformulated as follows: (37) Licensing pro is Case-marked by X0y, where y is parameterized. Identification pro inherits the f-feature values of X0y (if it has f-features; if not, pro gets a default interpretation, typically arb). (Rizzi 1986: 518–23 in Holmberg 2005: 536)



NULL SUBJECTS

217

The status of this pro has been largely debated by two “teams”: one trying to prove its existence, one denying it. The difference between the two views boils down to considering pro a phonologically silent but syntactically realized head, or to deny its existence altogether. The idea that pro is syntactically present but then not realized (or deleted at PF), is maintained as we have seen by Rizzi (1986) and much of the following GB literature, including Cardinaletti and Starke (1999), Roberts (2004, 2010b), Sheehan (2006), Holmberg (2005), and many others. The opposite view has its origins in Borer (1986), and is implemented in different ways by Alexiadou and Anagnostopoulou (1998), Barbosa (1995, 2009), Fassi Fehri (1993), Manzini and Savoia (2005), Nash and Rouveret (1997), Ordóñez (1997), Platzack (2004), and many others. Cardinaletti and Starke (1999) argue that pro is a weak pronoun based on its semantic and distributional properties. Roberts (2010b) follows Cardinaletti and Starke and brings more evidence that pro is a weak pronoun. According to these scholars, pro is a weak pronoun because it cannot occur in ostension sentences, but it can be used as an expletive. Furthermore, following Rizzi (1982), Roberts (2010b) and Holmberg (2005) show that pro occupies spec, TP. Finnish has in fact both a null referential pronoun and an expletive subject (sitä). In Finnish pro cannot co-occur with an expletive subject, as they compete for the same position, according to Holmberg. Furthermore, pro licenses floating quantifiers, which are only possible with preverbal subjects, and in languages like Florentine and Marchigiano it triggers agreement according to the preverbal subject pattern.5 According to another view, NSs are the result of deletion at PF of full pronominal subjects. This view is maintained mainly by Sheehan (2006), and Roberts (2004, 2010b). According to Roberts and Sheehan, rich agreement corresponds to an uninterpretable D feature on T. This D feature can be deleted through agreement, after the pronoun has been merged in spec, TP, together with the rest of φ-features that enter the Agree relation (Chomsky 2001 and subsequent work). Sheehan proposes that a mechanism of Deletion under feature non-distinctness takes place: (38) Deletion under feature non-distinctness (where non-distinctness refers to the subset relationship): α deletes under non-distinctness of features with β only if β agrees with α After valuation, originally valued features are virtually indistinct from features that are valued after Agree (Frampton and Gutman 2000; Richards 2011). Identical features do not need to be realized twice at PF, hence, according to Sheehan, the features of the specifier, which contains less information than the T head, get deleted (or not realized) at PF. A full pronoun has a D feature, and hence it can be deleted if T has a D feature itself. If T lacks this D feature, i.e. T is defective, not fully inflected, or impoverished, its deletion cannot take place. Yet another point of view on pro worth mentioning is Uriagereka’s (1999), according to which this element is a mere feature, not able to project a specifier.

218

CONTEMPORARY LINGUISTIC PARAMETERS

9.5. Partial NSLs Perlmutter was the first to note that Dutch has non-thematic NSs, but it allows extraction. Some years later, Maling and Zaenen (1978) pointed out that Icelandic is also a partial NSL (in that it only has null nonthematic subjects) and it also allows extraction, but under different structural requirements. Partial NSLs have been largely discussed by the Cambridge group on Null Subjects (Biberauer, Sheehan, Holmberg, Roberts). The results of the project have been collected in Biberauer et al. (2010), to which the reader is referred for a detailed discussion of several NS-related phenomena, and a fine-grained classification of NSLs. Another extensive survey is offered by Camacho (2013). As we have seen in 9.1.1.3., partial NSLs are those languages that do not have full referential silent pronouns, but only have null indefinite pronouns, or null arbitrary pronouns. Finnish, Marathi, Assamese (in the following examples, taken from Holmberg 2010a: 93), as well as Brazilian Portuguese are classified as partial NSLs. Example (39) shows that arbitrary 3rd person pronouns can be dropped in these languages: (39) a. Tässä tuolissa istuu mukavasti this.IN chair.IN sits comfortably ‘One can sit comfortably in this chair’ b. Hya khurchi-war aaramani bushushakto this chair-on comfort-with sit.PRES.3SG ‘One can sit comfortably in this chair’ c. Ei-khon soki-t aramkori boh-ibo par-i this-CL chair-LOC comfortably sit-MOD can-3SG ‘One can sit comfortably in this chair’

[Finnish] [Marathi] [Assamese]

Null arbitrary pronouns are first considered in Rizzi (1986) who proposes a different version of the NSP, as we saw above. In this new version, Case licensing is considered the crucial ingredient for licensing pro, while [+referential] and [+pronominal] features on INFL are replaced by a parameterized “licensing property.” Recall that the formulation of the NSP in Rizzi (1986) is the following: (40) Licensing pro is Case-marked by X0y, where y is parameterized. Identification pro inherits the f-feature values of X0y (if it has f-features; if not, pro gets a default interpretation, typically arb). (Rizzi 1986: 518–23 in Holmberg 2005: 536) The formulation in (40) was such because the main issue of the article were null objects, and Rizzi wished to find a unified account for both subject and object pro. According to Holmberg, this formulation makes one important contribution to the understanding of arbitrary null pronouns. Holmberg (2010a) accurately considers the distribution definite and indefinite



NULL SUBJECTS

219

null subjects, drawing the generalization according to which definite null subjects are common in full NSLs and uncommon, or heavily restrained, in partial NSLs. As for indefinite pronouns, their distribution is the reverse: they are quite common in partial NSLs and not so common (or not existing at all) in full NSLs. An example of this contrast is the following: (41) a. É assim que faz o doce is thus that makes the sweet ‘This is how one makes the sweet’ b É assim que se faz o doce is thus that se makes the sweet ‘This is how one makes the sweet’

[Brazilian Portuguese] [European Portuguese] (Holmberg 2010a: 92)

Holmberg follows Rizzi (1982) in attributing an important role to referentiality, which he however identifies with definiteness. His proposal is that null-subject pronouns are φPs (where φ is a cluster of φ features, see Cardinaletti and Starke 1999; Déchaine and Wiltschko 2002; and others). In NSLs, T is endowed with both unvalued φ features and an unvalued D feature. When T in a NSL probes (Chomsky 2005) for the subject in order to get its φ features valued, these features get copied into the feature matrix of T, giving birth to a sort of incorporation (see Roberts 2010b for more of this conception). Only the highest copy of the subject is pronounced, i.e. that on T, in the form of inflection. As for its reference, the null subject gets it from a null Aboutness topic which Holmberg postulates, following Frascarelli (2007). The story is different for partial NSLs, where T still has unvalued φ features and hence probes a subject, but it doesn’t have an unvalued D feature. This means that when the subject of a non-NSL “incorporates” onto T, it will only have a pronominal nature, in Rizzi’s terms, but it will not be definite, because of the lack of D. Since T does not have a D feature, the reference of the subject cannot be acquired via the chain with the Aboutness topic in partial NSLs. Hence, either their φP subjects, which will not receive their reference, are indefinite, or they are full DPs with their own reference. In this case, they will move to Spec, TP to check the EPP.6

9.6. Rich, impoverished, and no agreement NSLs 9.6.1. The role of rich agreement in licensing NSs One of the core points of the standard formulation of the NSP is that only languages with rich agreement can license null subjects. This was explicitly stated in terms of licensing through agr in Taraldsen’s (1980) work, and indirectly included in both of Rizzi’s formulations of the NSP. In (1982), Rizzi attributed the possibility to license a null subject to the INFL head. This INFL had to be pronominal and referential, where reference was represented by rich agreement. In Rizzi (1986), as we have seen, the NSP was reformulated in terms of a parameterized property on the INFL head, again linked to rich agreement.

220

CONTEMPORARY LINGUISTIC PARAMETERS

Empirical evidence that rich agreement is linked to licensing of thematic, full, referential, Italian-style NSs is offered for instance by Pashto, a split-ergative language spoken in Afghanistan. Pashto shows nominative-accusative alignment in the present tense, and ergative alignment in the past tense. This means for our purposes that in the past tense the transitive verb will show overt agreement with the internal argument / object, as in (42c) and (42d), while in the present tense it will show overt agreement with the external argument / subject, as in (42a) and (42b): (42) a. Jān ra-z-i John DIR-come-3SG.M ‘John comes’ b. Ze maṇa xwr-əm I apple eat-1.M.SG ‘I eat the apple’ c. Jān ra-ǧ -ay John DIR-come-3.M.SG ‘John came’ d. Ma maṇa wə-xwar-a I apple PRFV-eat-3.F.SG ‘I ate the apple’

[Pashto]

[adapted from Huang 1984: 535]

Pashto has null subjects and null objects. If agreement were directly responsible for their licensing, we would expect subject drop in the present and object drop in the past. This is exactly what we find: (43) a. e Ra-z-i DIR-come-3.M.SG ‘[He] comes’ b. e Maṇa xwar-əm apple eat-1.M.SG ‘[I] eat the apple’ c. e Ra-ǧ -ay DIR-come-3.M.SG ‘[He] came’ d. Ma e wə-xwar-a I PRFV-eat-3.F.SG ‘I ate [it (fem.)]’ 

[adapted from Huang 1984: 536]

Observe that dropping the subject in the past tense is not possible (44a), nor is it dropping the internal argument in the present tense (44b): (44) a. *e Maṇa wə-xwar-a[Pashto] apple PRFV-eat-3.F.SG ‘[?] ate the apple’ b. Zə e xwr-əm I eat-1.M.SG ‘I eat [?]’  [Huang 1984: 536]



NULL SUBJECTS

221

Further evidence of the role of rich agreement in licensing null subjects is offered by Ecuadorian Quechua, where null subjects are licensed in root clauses, which show morphological agreement on the verb, but not in embedded clauses, which do not: (45) a. (Ñuka) Marya-ta juya-ni [Ecuadorian Quechua] I-NOM Marya-ACC love-PRES.1.SG ‘I love Marya’ (Cole 1982: 34 in Gilligan 1987: 165) b. Juan-ka [*(ñuka) Marya-ta juya-j-ta] Juan-TOP I-NOM Marya-ACC love-NMLZ-ACC ‘Juan thinks that I love Marya’ (Gilligan 1987: 165) These data are taken as evidence for the relation between referential null subjects and rich agreement.

9.6.2. No agreement and radical NSLs The data from Pashto show uncontroversially that rich agreement plays a key role in licensing null subjects. The problems arise though with languages like Chinese, Japanese, and Korean, and to a certain extent Brazilian Portuguese, which allow null subjects as well as null objects, but have impoverished or no agreement at all. Huang (1984) examines the case of Chinese pro-drop, and proposes the existence an additional parameter separating languages that are discourse-oriented from languages that are sentence-oriented. Following Tsao (1977), Huang includes Chinese among discourse-oriented languages, which are languages in which antecedents of anaphors must be found in the discourse, but not necessarily in the sentence. If a language is discourse-oriented, it will have a zero-topic which will be able to bind empty variables, like null pronouns. This zero-topic is hence a licenser, and performs, according to Huang, the same task as the identifying [reference] feature in Rizzi (1982). Thus, two parameters are involved in null-subjecthood: the discourse-oriented vs. sentence-oriented parameter, and the presence or absence of an antecedent which is “rich enough in content” (a Topic, or rich agr). Discourse is also prominent in Saito’s (2007) analysis of radical pro-drop languages. According to him, radical pro-drop involves a mechanism of PF merging of arguments that are copied directly from discourse elements. In this, pro-drop equals ellipsis. Tomioka (2003: 336) observes the correlation between a language featuring discourse pro-drop and the possibility for it to allow bare NP arguments. He puts forward the following generalization: (46) Discourse pro-drop generalization All languages which allow discourse pro-drop allow (robust) bare NP arguments. Building on this generalization, Barbosa (2011, 2013) proposes an account for NSLs according to which there are two ways of licensing NSs: for pro-drop languages, the functional head bearing agreement is pronominal (see Rizzi 1986). For radical

222

CONTEMPORARY LINGUISTIC PARAMETERS

NSLs, which as we have just seen feature bare NP arguments, NSs are the result of null NP anaphora, which is licensed differently depending on the anaphora-licensing mechanisms available in the language. Along similar lines, Duguine (2013) attempts to explain all pro-drop phenomena as NP/DP ellipsis, by showing that the conditions licensing NSs are the same licensing DP ellipsis in larger constituents. Another attempt to unify radical NSLs with NSLs was put forward by Jaeggli and Safir (1989) with their Uniformity of the Paradigm hypothesis, according to which null subjects are permitted in all and only languages with morphologically uniform inflectional paradigms. Jaeggli and Safir distinguish between derived forms and underived ones. Derived verb forms are constituted by a root and an affix. Underived forms are bare roots. A paradigm that has all derived forms, like Italian or German, can allow null pronouns (referential or expletive). A paradigm with only underived forms, like Chinese, can also license null subjects. Languages like French and English, which have only some derived forms, do not license null subjects. The formulation of the NSP according to Jaeggli and Safir (1989: 29) is the following: (47) Null subjects are allowed in all and only languages with morphologically uniform inflectional paradigms. A NSL must hence have rich agreement or no agreement at all.

9.6.3. Impoverished agreement The setup of the inflectional paradigm is crucial for the NSP also according to Roberts (2004, 2010b). Building on Müller (2005), Roberts (2010b) classifies languages into rich and impoverished agreement ones; rich agreement when each person-number combination for an inflectional morpheme is realized by a different ending, and there has been no pre-syntactic impoverishment creating syncretic forms in the paradigm. If the paradigm is impoverished, T has no D feature, hence it cannot license deletion of the subject pronoun. German, Dutch, and partial NSLs have impoverished paradigms, and that is the reason why they are not fully NS, despite their seemingly rich inflectional paradigms. Italian, Greek, Turkish, and all referential NSLs have a morpheme for every number-person combination: they are uniform regarding the setup of their inflectional forms. Chinese, Korean, and Japanese, on the other hand, are also morphologically uniform, as they do not have any ending at all. Uniformly inflected or uniformly uninflected languages license full pro-drop. Impoverished paradigms can only license partial or expletive NSs. Furthermore, following Saito (2007), Roberts assumes that pro is directly merged at LF in languages without agreement. According to Saito, as mentioned in the previous section, both ellipsis and radical pro-drop involve LF copying of arguments from discourse entities. Discourse licenses pro in radical pro-drop languages. Roberts arrives at the following generalizations:



NULL SUBJECTS

223

(48) a. Radical pro-drop is possible if φ-agreement is not obligatory. b. Consistent null subjects are possible if there is no impoverishment of T’s φ-features.  (Roberts 2010b: 85) (48) restates Saito’s (and Huang’s) discourse licensing of pro in radical NSLs, which lack agreement. (48b) states that among languages with agreement, only those with rich, non-impoverished, agreement can have a null subject. According to Roberts, then, the NSP should be rewritten as follows: (49) a. Are φ-features optional or obligatory on probes? b. If obligatory, are φ-features impoverished on all probes? c. If φ-features can be impoverished, are those of {T, v, …} impoverished? (Roberts 2010b: 86) (49a) separates radical NSLs from the rest; (49b) separates non-NSLs from NSLs, and (49c) separates partial NSLs from full NSLs.

9.7. The micro-parametric approach to NS The classical parametric approach has several flaws, as we have seen. The first and most salient is its rigidity which, as was intensely discussed within the Newmeyer vs Roberts and Holmberg debate, forces an all-or-nothing approach on the data, which are often not so straightforward. With the advent of Minimalism (Chomsky 1995 and subsequent work), the debate on the shape and role of parameters has been put slightly aside: while scholars keep assuming the existence of parameters like the Head Parameter, or the Null-Subject Parameter, not much research has been done on discovering new ones. One notable exception is the reformulation of the concept of parameters dubbed by Baker (2008a) as the “Borer-Chomsky conjecture,” according to which language variation can be ascribed to featural differences on lexical items. This new view on parameters moves variation to the lexicon. Different feature setup on given lexical items determine different languages. For parameters, this means moving from “macro switches” determining many different properties of a given language in one go, to clusters of micro-parameters, each linked to a feature. These clusters have the advantage of being more flexible than classical parameters, and to have wider empirical coverage. The obvious limit of this conception of parameters is it is not clear how and under which conditions this clustering takes place. This limit can be overcome with hypotheses such as the Uniformity of the Input, put forward by Roberts and Holmberg (2010), according to which if a feature has a certain value on a key head (like v), this value will be passed on to all other heads in the clause. If for instance v has a [+EPP] (a feature triggering movement to its specifier), the same value will spread to all other functional heads, resulting in a consistent head-final language.

224

CONTEMPORARY LINGUISTIC PARAMETERS

(50) Generalization of the input If acquirers assign a marked value to H, they will assign the same value to all comparable heads. (Roberts and Holmberg 2010: 41) This parameter resolution mechanism is rather flexible, and accounts both for the fact that languages tend to be harmonic, and for the possible deviation from harmonic order. Languages like German and Dutch, for instance, which are disharmonic, have marked values (i.e. values deviating from uniformity) on some heads. This approach allows more flexibility than the granitic classical parametric one, not only because it permits deviations from the standard, but also because being these micro-parameters independent, and linked to single features, they can easily obscure each other. This in turn makes it possible to have opacization effects, meaning that one parametric setting may be overcome by another, without having to postulate a macro change in the grammar. In this way it is much easier to account for exceptions. Another option is to consider parameters as schemata, available to the language acquiring-child, as proposed by (Gianollo, Guardiano, and Longobardi 2008). For a detailed discussion of the form and quantity of micro-parameters, see Biberauer et al. 2010). For null subjects, Roberts and Holmberg propose the following scheme, which is an updated version of (49): (51) a. Are uφ-features obligatory on all probes? No Radical pro-drop

Yes b. Are uφ-features fully specified on all probes? Yes No

Polysynthesis 

c. Are uφ-features fully specified on some probes? Yes

No

Non-pro-drop  d. Are the uφ-features of {T, v} impoverished? (Roberts and Holmberg 2010: 49)

Radical pro-drop languages are those languages, like Chinese and Japanese discussed above, which allow null subjects and null objects in the absence of verbal inflection. Furthermore, the more impoverished the φ-feature bundle is on T (which is relevant for our case), the more difficult it is to drop the subject. The richer the inflection, the richer the φ-set, the easier it is to drop the subject.



NULL SUBJECTS

225

9.8. Conclusions The idea of a null-subject parameter from which many properties of a given language can descend has always been very appealing. In the GB era, when the attempt to make the parametric components of the UG explicit was strongly pursued, the formulation of an overarching macro parameter which could capture the co-occurrence of several properties of a language was welcomed and accepted. During the years, the aims of generative research have switched slightly, and the idea of a macro NSP has been abandoned altogether, or replaced by a new formulation in terms of micro-parametric clustering. This was attempted because of the urge to account for the increasing amount of empirical counterevidence to the classical, granitic formulation of the NSP, and because of theory-internal considerations. Some observations, though, remain valid through the years: first and foremost, the correlation of null subjects and rich agreement in inflecting languages; the correlation of null subjects and discourse recoverability in radical NSLs; the implication according to which a NSL that has full referential NSs usually also has expletive NSs. Furthermore, partial NSLs have been given a lot of attention in the last years, and the empirical coverage has extended dramatically. In this respect, much has been achieved in the understanding of this phenomenon. Whether the new insights have contributed substantially to what had already been discovered in the early 1980s or not remains an open question.

Notes *

I wish to thank George Walkden, Marco Tamburelli, Jacopo Garzonio and two anonymous reviewers for their valuable comments and suggestions. This research was sponsored by the NWO VIDI program, project 276-70-021 Splitting and clustering grammatical information, which is hereby acknowledged.

1

In the literature, e is the shorthand for “empty category.”

2

Note that this correlation, together with the idea of treating null subjects on a par with empty categories, has been recently reconsidered by Chomsky, in a series of MIT lectures in March–April 2014. See footnote 3.



3

In a recent series of lectures, Chomsky (2014b) reconsiders the NSP within the general issue of labeling. Labeling (Chomsky 2005 ff.) is an operation which assigns a label to a set of two syntactic items, so that the derivation can proceed and selection can take place. In what follows I give a very short overview of the background needed to follow Chomsky’s proposal, which might not be common knowledge yet, and then move on to present Chomsky’s idea on the NSP.   In an “asymmetrical” merge, i.e. when a head is merged with an XP, the label will be given by the head. This is what happens in what was once called a head-complement structure. If V is merged with a DP, the label of the set deriving from this Merge will be V (we will call this set a VP).   One of the problematic cases for the labeling algorithm which Chomsky considers is the XP YP configuration (see also Chomsky 2013a), otherwise known as the specifier problem. In a case of object extraction, the object is internally Merged at the edge of

226

CONTEMPORARY LINGUISTIC PARAMETERS

the vP in order to be extracted to the matrix clause. We have in this case a DPwh-vP structure. In other words we have two XPs merged with each other. This creates a problem for the labeling algorithm.   Chomsky examines several solutions to this problem. For wh-movement, Chomsky follows Moro’s antisymmetry in claiming that when we create an XP YP structure, this needs to be broken via movement of one of the constituents. This, according to Chomsky, generates successive cyclic movement.   Subjects in spec, TP are also a case of XP YP, the subject being a DP merged with a TP. In this case the subject does not move out, though, because it has hit a criterial position (Rizzi 2006, 2007; Chomsky speculates on why the subject does not move forward, but this is not relevant for the present discussion and we leave it aside). The XP YP set created by the subject in spec,TP must be labeled. One way to do it is to give it the label of the most prominent common element, in this case φ. Both the subject and TP share φ features, hence the label of this will be φ.   With this in place, we can now move on to the NSP. Chomsky starts from the observation which is very familiar by now of rich agreement. He says that languages with rich agreement have a “strong T,” i.e. a T which has enough φ-features as to be able to give a label to an XP TP structure in which it is found.   Going once again back to the origin, Chomsky also considers subject extraction as a case of ECP.   Extraction is possible in Italian (ECP is violated in Italian, to put it in Chomsky’s terms) because it is possible to label the intermediate position of extraction (spec,TP of the embedded clause) thanks to rich agreement. This possibility is not given in English, where T is weak, root-like, hence its φ is not sufficient for labeling. T in English can be a label only if it combines with something “labelable,” like a full subject. In that case the label will be the combination of the φ-features of T and the subject.   If T is strong in NSLs, it does not need to be merged with a full subject. This means that Italian and other full NSLs do not have an EPP. No subject is needed in spec,TP in order to continue the derivation. The EPP and the ECP go thus hand in hand: either a language has both, or it has neither.   The NSP regards hence the strength of T, its root-like status vs its full status. From this the possibility descends of having subjectless sentences as well as extracting the subject without incurring in that-t effects. 4

The Newmeyer vs Roberts and Holmberg debate involves several issues, of which we will report only those that are strictly relevant for the NSP.

5

For reasons of space, all the evidence in favor of the existence of pro and its position cannot be presented here. The reader is referred to Roberts (2010b: 70 and ff) for a complete overview of the empirical evidence in favor of pro.

6

So far, we have not addressed the EPP, as it is marginal here. If full NSLs have an EPP, how is this going to be checked in the case of incorporation of a NS? Several alternatives are possible, one being that there is no EPP at all in NSLs (which would mean in terms of our parameters that a T head in a NSL has an unvalued D feature and lacks an EPP – two properties that are not very easy to be derived from the same parameter); the other being checking via verb-movement, following Alexiadou and Anagnostopoulou’s (1998) proposal. We leave the EPP issue aside because not directly relevant for this discussion, and we turn to examine the role of rich inflection and NSLs lacking inflection altogether.

10 Head movement in the clausal domain* Ángel J. Gallego and Juan Uriagereka Universitat Autònoma de Barcelona and University of Maryland

10.1. Introduction Like other hallmark phenomena of the Government-and-Binding framework (GB, henceforth; cf. Chomsky 1981, 1986a, 1986b; Chomsky and Lasnik 1995), the literature on head movement, its locality conditions, and its cross-linguistic variation is vast—almost impossible to summarize in a single paper. This chapter reviews some of the most influential analyses of this phenomenon, restricted to the clausal domain, thus covering an area of empirical work—namely verb movement (VM, henceforth)—that led to various theoretical breakthroughs ever since the 1970s. The interest in VM within the generative literature goes back to Chomsky’s (1957) early analysis of verbal affixes (cf. Lasnik et al. 2000 for ample discussion). Chomsky (1957) was concerned with the fact that affixes, modals, and auxiliaries appear in a specific order in English, which he took to indicate that each auxiliary determines the morphological shape of the following form. Thus, for instance, V1 determines the form of V2 in sequence like ‘V1-V2’, as shown in (1), which in turn entails that there is no possibility of a relation holding between V1 and V3 in a ‘V1-V2-V3’ sequence, as shown in (2) (cf. Biberauer and Roberts 2010): (1)

a. Politicians b. Politicians c. Politicians d. Politicians

were criticiz-ed were criticiz-ing banks have criticiz-ed banks will criticize-∅ banks

228

CONTEMPORARY LINGUISTIC PARAMETERS

(2) a. Politicians {have/*are} been criticized b. Politicians were {being/*been} criticized c. Politicians will be {criticized/*criticize-∅} The study of this paradigm was later on extended by many authors and ultimately gave rise to the postulation of various autonomous functional heads (cf. Chomsky 1981, 1986b; Uriagereka 1988, 1995; Rizzi 1997, 2004; Cinque 1999, 2004), their splitting (cf. Pollock 1989; Chomsky 1991), their locality conditions (cf. Emonds 1978; Koopman 1984; Travis 1984; Rizzi 1982, 1990), and the different parameters associated to them (cf. Emonds 1978; Belletti 1990; Den Besten 1983; Raposo 1987; Lasnik 1999; Roberts 1985, 1993, 1994; Vikner 1995), all of which was further backed up by work on noun incorporation (cf. Baker 1988; Hale and Keyser 1993; Haugen 2009), and clitic movement (cf. Kayne 1975, 1989b, 1991; Sportiche 1996; Uriagereka 1988, 1995; Raposo and Uriagereka 2005; Roberts 2010a).1 These contributions made it possible to improve our understanding of different usual suspects of linguistic theory: locality conditions, parametric variation, left periphery, syntax-phonology interactions, and a long etcetera. In the domain of parametric inquiries, VM played a key role in two main case studies: (i) VM to I (and related functional projections, v, voice, AgrS, Asp, T, and Fin), and (ii) VM to C (and related functional projections, Int, F, Top, Foc, and Force). The first subcase is perhaps best known from Emonds’ (1978) original observation that French has a verb movement rule that is absent in English. Later on, Pollock’s (1989) developed that proposal and put forward the well-known “Split IP Hypothesis,” which took the feature composition of the INFL node to instantiate different projections (AgrSP and TP), arguing for the parameterization of VM in V-to-T languages (French) and non-V-to-T languages (English). The following examples illustrate this kind of parameter (we indicate the verb in bold letters): (3) a. *Jean embrasse souvent Marie *Jean souvent embrasse Marie b. *John kisses often Mary *John often kisses Mary

[French] [English] (apud Roberts 2001: 120)

Although the data in (3) seems conclusive, the clear-cut parameter advocated by Pollock (1989) and its connection to rich morphology has been subject to much controversy, on both theoretical and empirical grounds (cf. Bobaljik 2001, 2008; Biberauer et al. 2010). The second subcase, V-to-C movement is largely concerned with V2 languages (cf. Biberauer 2002; Vikner 1995; Holmberg 2010b, among others), and the empirical intricacies and sub-parameters that appear to play a role in Scandinavian and Germanic languages. Descriptively, V2 refers to the appearance of the verb in the second position, either in main or embedded (finite) clauses (again, the verb is in bold letters):



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

229

(4) a. Jag har ärligt talat aldrig sett huggormar i den här skogen [Swedish] I have honestly speaking never seen adders in this here forest ‘To be honest I’ve never seen adders in this forest’ b. Huggormar har jag ärligt talat aldrig sett i den här skogen adders have I honestly speaking never seen in this here forest c. I den här skogen har jag ärligt talat aldrig sett huggormar in this here forest have I honestly speaking never seen adders d. Ärligt talat har jag aldrig sett huggormar i den här skogen honestly speaking have I never seen adders in this here forest (from Holmberg 2010b: 1) The distribution and restrictions on V2 are fairly intricate, as Holmberg (2010b) shows in detail, and seems to be available in Romance languages in a restricted fashion (cf. Rizzi and Roberts 1989, Rizzi 1996). This can be seen in so-called “obligatory inversion” in present-day Spanish (cf. Suñer 1994; Zagona 1988; Gallego 2010, Depiante and Vicente 2012). Torrego (1984) showed this while discussing V2 in embedded domains, as a side-effect of successive cyclic movement: (5) a. ¿Qué pensaba Juan que le había dicho Pedro que … ? [Spanish] what thought-3.sg Juan that to.him had said Pedro that ‘What did Juan think that Pedro said that … ?’ b. ¿Qué pensaba Juan que Pedro le había dicho que … ? [Spanish] what thought-3.sg Juan that Pedro to.him had said that ‘What did Juan think that Pedro said that … ?’ (from Torrego 1984: 108–9) But V2 was a prominent feature of old varieties of Romance, as the data in (6) indicate: (6) Depues [CP [Cº tornar] lo-as assu logar ] (Old Spanish) afterwards return it.will.2S to+its place ‘Afterwards you will return it to its place’ (Pit 32r21-23) (from Rivero 1993: 231) It is unclear how and why V2 ceased to be operative in Romance in the way it was (and still is in Germanic and Scandinavian languages), which raises the question of how much UG-rooted the V2 parameter is. This chapter is divided as follows. Section 10.2 briefly summarizes the most influential approaches to VM in the literature, discussing its trigger, technical implementation, and relevance for the study of various areas of linguistic research. In section 10.3 we concentrate on the two main case studies of verb movement (to T and to C), and the parameters associated to them. Section 10.4 summarizes the main conclusions.

230

CONTEMPORARY LINGUISTIC PARAMETERS

10.2. Approaches to clausal head movement As already noted, the distribution of verbs along the clausal spine (and its transformational reinterpretation) has been the focus of much research ever since the 50s (cf. Roberts 2001, 2011; Barrie and Mathieu 2014, and references therein). Given that the interest of the parametric nuances of this phenomenon are particularly important in the GB literature, in this chapter we concentrate on the main approaches to this phenomenon within that period and its transition to minimalism. Due to space constraints and for the sake of clarity, we will discuss two popular approaches to VM: one attributing it to morphological richness (which is intimately related to the null-subject parameter; cf. D’Alessandro, this volume), and those taking VM to be triggered by morpho-phonological rules (based on Chomsky’s 2001 claim that head movement is a post-syntactic phonological process). At the end of this section we briefly discuss other influential, but not hegemonic, approaches to VM.

10.2.1. Verb movement in the syntax: The role of morphology Within the IP / TP field, different types of VM can be distinguished, depending on how rich the clause structure functional heads are. Here we will adopt a fairly standard approach whereby the verb can appear in four key positions, as indicated in (7): (7) [CP C(4) [TP T(3) [vP v(2) [VP V(1) … ] ] ] ] According to this restricted scenario, there are three famous case studies of VM that can possibly take place. We list them in (8):2 (8) a. V → v b. v → T c. T → C The option in (8a) may be universal (cf. Chomsky 2004),3 and it is in fact needed for categorization reasons if V is regarded as an acategorial root, as in the Distributed Morphology framework (cf. Halle and Marantz 1993, Embick and Marantz 2008).4 The option in (8b) is more interesting and relevant for our purposes. As pointed out above, v → T movement is typically related to the richness of the morphological paradigm of the verb in different languages in the GB literature (cf. Emonds 1978; Belletti 1990; Chomsky 1991; Pollock 1989; Roberts 1985, 1993, 1994; Rohrbacher 1999; and others). Biberauer and Roberts (2010) note that this leading and fruitful intuition gives rise to at least three classes of languages, following Koeneman and Neeleman (2001):



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

231

(9) Language types according to their inflectional richness a. Strongly rich languages (languages displaying v → T movement and null subjects, such as Spanish, Catalan, Galician-Portuguese, or Greek) b. Mildly rich languages (languages displaying v → T movement too, but not null subjects, such as French or Middle English) c. Impoverished languages (displaying neither v → T movement nor null subjects, such as present-day English or Mainland Scandinavian) (adapted from Biberauer and Roberts 2010: 264) Various pieces of evidence support this seemingly robust correlation between VM and null subjects. For instance, Roberts (2001: 119 and ff) notes that the empirical facts can be reduced to the observation that VP adverbs, clausal negation, and floating quantifiers precede the verb in English, whereas they follow it in French. Let us consider the data again, repeating (3) as (10) for convenience:5 (10) Clausal adverb placement a. *Jean embrasse souvent Marie *Jean souvent embrasse Marie b. *John kisses often Mary *John often kisses Mary (11) Clausal negation a. *Jean (ne) mange pas du chocolat *Jean (ne) pas mange du chocolat b. *John eats not chocolate *John does not eat chocolate (12) Floating quantifiers a. *Les enfants mangent tous le chocolat *Les enfants tous mangent le chocolat b. *The children eat all chocolate *The children all eat chocolate

[French] [English]

[French] [English]

[French] [English] (from Roberts 2001: 120)

Towards the end of the GB framework and the arrival of minimalism, Chomsky (1991, 1993) took VM to be related to two parametric choices, languages being of the verb raising type (French) or of the T lowering type (English), a double option that was replaced by the notion of “feature strength.”6 In brief, the idea was that some features are strong and need to be checked off in the syntax (before the application of the Spell-Out rule), while others are weak (and may therefore wait until LF to be checked off). Let us now consider the case of T → C movement, which covers various constructions involving the “verb inversion” label (e.g. Subject-Aux inversion in English, Aux-to-Comp movement in Italian or Portuguese, subject-clitic inversion in French, and V2 in Germanic languages). Den Besten’s (1983) work is seminal in this domain. In some languages, the V seems to move directly to C, whereas in others some auxiliary element (modal or tense morphemes) does—although the pattern

232

CONTEMPORARY LINGUISTIC PARAMETERS

is different with verbs like have and be in English, an issue we return to in section 10.3.1: (13) a. *¿Rescató a muchos países la Unión Europea? *rescued.3SG A many countries the union european *‘Did the European Nation rescue many countries?’ b. *Rescued the European Union many countries? c. *Did the European Union rescue many countries?

[Spanish] [English]

The asymmetry in (13) is taken to follow from the locality conditions on HM (cf. Travis 1984; Rizzi 1990; Roberts 2010a, 2011). Since T is closer to C than V, only T can move, unless V moves from the latter position, and is pied-piped along with T (as a free rider). Schematically: (14) [CP C … [TP T … [vP v … [VP V … ] ] ] ]

*

Roberts (2001) provides arguments that the verb moves all the way to C in these cases (and does not remain in T or some intermediate projection). To begin with, the verb occupies the position that typically belongs to complementizers (see the German data in 15, where daß is dropped): (15) a. Er sagte, gestern sei er schon angekommen [German] he said yesterday have he already arrived ‘He said he’d already arrived yesterday’ b. Er sagte daß er gestern schon angekommen ist [German] he said that he yesterday already arrived has ‘He said he’d already arrived yesterday’ (adapted from Roberts 2001: 124) A second argument comes from Den Besten’s (1983) work, who noted that the distribution of weak pronouns reinforces a treatment where complementizers (in non-V2 contexts) and verbs (in V2 contexts) occupy the same position. The Dutch examples in (16), which feature the weak pronouns je and ze, are intended to show this: (16) a. … dat {je gisteren / *gisteren je} ziek was [Dutch] … that you yesterday yesterday you sick were ‘… that you were sick yesterday’ b. Was {ze gisteren / *gisteren ze} ziek? [Dutch] were you yesterday yesterday you sick? ‘Were you sick yesterday?’ (adapted from Roberts 2001: 124) Although one may propose different analyses to tackle the facts, V2 was analyzed



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

233

as the result of C having a strong V (or T) feature, which is reminiscent of Pollock’s (1989) idea (cf. Platzack 1987, Holmberg and Platzack 1991, 1995, Roberts and Roussou 1998). Notice that in all these approaches VM is treated as a syntactic operation. It takes place either in the overt syntax (Surface Structure) or else in the covert syntax (Logical Form), but not at Phonetic Form. This view is standard within the GB literature and the first decade of minimalism (until Chomsky 2001). In the next section we discuss the possibility that VM (HM more generally) is a morpho-phonological operation instead.

10.2.2. Verb movement after syntax Departing from previous assumptions (though going back to a view that he held in the 1950s, cf. Lasnik 1995b), Chomsky (2001) casts doubt on the syntactic nature of HM. Chomsky (2001) provides some empirical arguments in order to take HM to be morpho-phonological (cf. Julien 2002; Rizzi 2004; and references therein), such as the fact that Xºs are interpreted in the same way in their base and derived positions, which in the context of V means that Vº is interpreted identically in T and in C. This seems to be different to XP movement, which can feed discourse-oriented readings. Compare (17) and (18) to see this. Whereas the verb has no specific semantic gain after movement to in (17), the DP a man is interpreted in (18) (and arguably favor a specific interpretation): (17) a. You were called by the President b. Were you called by the President? (18) a. The policeman saw a man b. A man, the policeman saw Apart from the problems discussed in Chomsky (2001), the list in (19) summarizes some of the most obvious objections for HM to be a bona fide syntactic operation (cf. Zwart 2001 too). (19)

a. BPS cannot distinguish the segment / category distinction (cf. Harley 2004) b. HM violates the Extension Condition (cf. Chomsky 2001) c. HM violates Uniformity Condition on chains (cf. Chomsky 1995) d. HM is too local (cf. Abels 2003) e. HM violates the A-over-A Condition (cf. Roberts 2010a)

234

CONTEMPORARY LINGUISTIC PARAMETERS

Notice that, technically, the problems that HM faces have to do with its countercyclic nature, which becomes obvious if we compare the outcome of Xº and XP movement: (20) a. Head Movement

XP

WP

b. Phrase Movement

Yº

X’ Xº Xº

XP

YP ZP

X’ Xº

… tYº …

ZP … tYP …

Because of these arguments, Chomsky (2001) argued that HM may, for the most part, fall within the phonological component, a proposal that generated controversy as well as an interesting source for theoretical and empirical discussion (carefully reviewed by Roberts 2010a). As noted by Roberts (2011), the question we face is whether the evidence we have supports the non-syntactic (hence semantically vacuous) status of HM. This author considers the argument provided by Boeckx and Stjepanović (2001), from pseudogapping: (21) Although John doesn’t eat pizza, he does – [VP pasta [VP eats tpasta ] ] (adapted from Roberts 2011: 203) Following work by Howard Lasnik (cf. Lasnik 1999, 2003), this example was taken as evidence in favor of an object-shift operation in English (along the lines of Koizumi 1995; Lasnik and Saito 1999), which yields (21) after VP deletion, bleeding V → v at the same time. If the relevant operations are not ordered in any specific way, the question—Roberts (2011) points out—is why something like (22) is ruled out, assuming that the verb moves to v before ellipsis and object shift: (22) *Although John doesn’t eat pizza, he eats – [VP teats pasta ] Boeckx and Stjepanović (2001) claim that (21) obtains if the object moves to [Spec, VP] in the syntax (more specifically, to the specifier of an AgrO projection), and both ellipsis and VM operate in the phonology. Consequently, ellipsis can apply either before or after VM. If it takes place before, we have pseudogapping; if it takes place after, we have a regular SVO sentence in English. It is important to emphasize, though, that Chomsky’s (2001) observations focus on verbs. However, as has been noted in the literature, the moment other types of heads are taken into account (e.g. pronouns or quantificational elements), a different picture obtains (cf. Lechner 2005; Roberts 2010a; Vicente 2007).7 The data in (23) illustrate this point, since modals, negation, and clitics can interact in ways that seem to be genuinely syntactic: can feeds reconstruction in (23a) (as the preference for the not > can reading indicates), not licenses the NPI any time in (23b), and la



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

235

must be interpreted as denoting a specific DP in (23c) (it cannot substitute a DP such alguna película “some movie”). (23) a. John can not (can) come along today (from Roberts 2010a: 12) b. The witness didn’t see the accused at any time (from Roberts 2010a: 16) c. María no la ha visto, esa película [Spanish] María not her have.3SG seen that movie ‘That movie, María has not seen it’ To recap so far, we have seen that HM must face different technical problems, especially in the context of minimalism. In order to accommodate these problems, one may either regard HM as phonological or else argue that what looks like HM does not display the syntax in (20a). The idea that HM is phonological is consistent with the varying nature of VM from language to language. If parametric variation is, at least partially, determined by the morpho-phonological component (cf. Chomsky 2007, 2010), then the PF nature of VM falls into place. That said, facts like the ones in (23), plus others that Roberts (2010a, 2011) reviews, indicate that things are not so simple. The next section sketches three technical ways to treat HM as syntactic without resorting to (20a).

10.2.3. Other approaches to head movement In the previous sections we discussed the specifics of standard HM within the GB literature, and some of the problems it raised, which led to the possibility that this operation is part of the phonological component. Along with these two poles, there are some technical options in the literature that we should briefly consider. We will discuss the remnant XP movement approach (partially based on Kayne’s 1994 antisymmetry framework), the reprojection approach (developed in different ways by Hornstein and Uriagereka 2002; Donati 2006), and the Agree approach (developed by Roberts 2010a). A first alternative that treats HM as syntactic but without deploying the problems summarized in (19) is based on the claim that what looks like HM in the surface is nothing but XP remnant movement, as depicted I (24) for VM: (24) a. HM (standard approach) b. HM (as XP remnant movement) i. [TP Tº [VP V XP ] ] i. [TP Tº [VP V XP ] ] ii. [TP [Vº+Tº] [VP tV XP ] ] ii. [TP Tº [VP V tXP ] ] iii. [TP [VP Vº tXP ] Tº tVP ] This approach has many followers, including authors working on verb-related phenomena (cf. Koopman and Szabolcsi 2000; Nielsen 2003; Müller 1998, 2004; Kayne and Pollock 2001; Poletto and Pollock 2004; Pollock, Poletto, and Munaro 2003; Carnie, Harley, and Dooley 2005; Collins 2005, among others). The outcome in both (24a) and (24b) is the same: Vº > Tº. The new twist that (24b) gives has the advantage that, since it is an XP that moves, then most of the theoretical shortcomings of HM noted by Chomsky (2001) go away—although new ones emerge, like the

236

CONTEMPORARY LINGUISTIC PARAMETERS

fact that traces / low copies are no longer c-commaned by the moving elements, and freezing effects, in the sense discussed by Wexler and Culicover (1981), should prevent further movement of V to C. Roberts (2011) discusses the case of German “remnant topicalization” as an argument in favor of (24b) (similar arguments are discussed for Romance clitics in Roberts 2011). The facts are as in (25): (25) Gelesen hat er das Buch nicht read has he the book not ‘He hasn’t read the book’

[German] (adapted from Roberts 2011: 207)

Here we can see that the DP das Buch has scrambled out of the VP, which is followed by VP (remnant) fronting to the left edge of the clause, and thus triggering V2, roughly as in (26) (where irrelevant details are omitted): (26) a. [TP hat [ nicht [vP er v [VP das Buch gelesen ] ] ] ] b. [CP [VP tdas Buch gelesen ] hat [TP er [XP das Buch [ nicht tVP ] ] ] ] (adapted from Roberts 2011: 207) Roberts (2011) concludes that the XP remnant (or smuggling) approach opens a new line to approach HM, and in so doing it solves some of the problems pointed out by Chomsky (2001), but there are loose ends too; it is not clear, under this approach, what triggers the movement of the relevant phrase-mates (XP in 24b above, for instance). A second alternative to HM is proposed by authors assuming that Xºs move like XPs (that is, to a specifier position), and then reproject. This option is explicitly ruled out in Chomsky (1995: 256ff.) on Chain Uniformity grounds, but as noted by Hornstein and Uriagereka (2002), most of those grounds are not a problem under a derivation model. The basic intuition that all manifestations of the idea present is shown in (27): (27)

HM (reprojection account) i. [TP Tº [VP Vº XP ] ] ii. [TP Vº [T’ Tº [VP tVº XP ] ] ] (Vº moves to [Spec, TP], no reprojection) iii. [VP Vº [T’ Tº [VP tVº XP ] ] ] (Vº moves to [Spec, TP], reprojection)

Approaches endorsing this type of derivation divide between those that take Vº to project after movement to [Spec, TP] (cf. Donati 2006, Gallego 2010, Hornstein and Uriagereka 2002) and those that take Vº to undergo an affix-lowering-like process with T (cf. Matushansky 2006; Vicente 2007). Donati (2006) provides empirical evidence showing that the label of a CP depends on the size of the category that is attracted to C. If the relevant category is a head, then it projects; if it is an XP, then the previous label is left intact (but see Hornstein and Uriagereka 2002 for arguments that labels may change in this case too, at least in the covert component).8 The asymmetry is shown in (28), where the specifier of CP can be occupied by relative pronouns (which trigger reprojection, according to Donati 2006) and interrogative pronouns (which do not):



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

237

(28) a. I wonder [CP what booki C [TP you read ti ] ] b. I wonder [DP whati C [TP you read ti ] ] If these conclusions are correct, then HM yields hybrid labels (cf. Chomsky 2008, Gallego 2010 and references therein). Notice that, under this account, at least some instances of HM can be regarded as truly syntactic, and not just part of the morphophonological component, as the empirical evidence indicates. A final alternative account is Roberts’ (2010a), which on the basis of cliticization phenomena recast HM as long-distance Agree. As Roberts (2010a) observes, unlike standard Probe-Goal dependencies (see 29), cliticization involves a Probe and a Goal that cannot be teased apart by focusing on the feature composition of v and the clitic (see 30): (29) Agree (v, full DP) a. [v*P John v[person: ][number: ] [VP [V’ met the boy[person: 3][number: plural][Case: ]]]] b. [v*P John v[person: 3][number: plural] [VP [V’ met the boy[person: 3] [number: plural] [Case: Acc]]]] (30) Agree (v, clitic) a. [v*P Luis v[person: ][number: ] [VP [V’ vio los[person: 3][number: plural]]]] b. [v*P Luis los v[person: 3][number: plural] [VP [V’ vio los[person: 3][number: plural]]]] Luis saw them If lexical items are nothing but feature complexes, then narrow syntax cannot differentiate v and the clitic los in (30), and thus will treat them as two instances of the same lexical item (two “occurrences,” as in any complex chain). Once the computation treats v and los as head and tail of a chain, it follows—Roberts (2010a) reasons—that only the higher occurrence will be pronounced, for whatever reason only higher copies of chains are pronounced (cf. Nunes 2004). A crucial assumption that Roberts (2010a) has to make concerns the idea that clitics are φ-defective determiners, lacking Case (cf. Cardinaletti and Starke 1999). Space limitations do not allow a careful evaluation of this proposal (cf. Gallego 2011; Matushansky 2011, and references therein for discussion).

10.3. Parameters associated to head movement In this section we explore the two most salient cases of parametric variation within the realm of VM, namely V → T and T → C movement. We have introduced the basic patterns that underline these case studies, but now we would like to go further and consider the actual accounts that have been put forward in order to show why the relevant parameters are fixed in certain language, but inactive in others.

10.3.1. Variation in V → T movement As noted in section 10.2.1, Emonds (1978) and Pollock (1989), both of which are based on the different position that verbs occupy vis-à-vis elements like adverbs,

238

CONTEMPORARY LINGUISTIC PARAMETERS

floating quantifiers, and negation in languages like English (Germanic) and French (Romance), led to the influential idea that the morphological richness of a verbal paradigm is the key to distinguish languages with V movement from languages with T lowering:9 (31) a. V → T (V movement)

TP

T

b. T → V (T lowering) TP

VP Subj

T V’

V

VP Subj

Obj

V’ V

Obj

A way to formulate the relevant (lexical) parameter would be as in (32), which capitalizes on the notion of “feature strength,” a cornerstone of late GB and early minimalism (cf. Chomsky 1991, 1993, 1995):10 (32) VM Parameter (feature strength version) a. In Romance, φ-features of T are strong (VM applies before Spell-Out) b. In English, φ-features of T are weak (VM applies after Spell-Out) Lasnik (1999, 2003) examines the different analyses of VM within the generative literature. In his account, the relevant parameter is very similar to (32), but he recasts the weak / strong dichotomy by taking verbs to be inflected (in French) or bare (in English) in the lexicon. An important observation, present in Emonds (1978), Pollock (1989), and Lasnik (1999), is that auxiliaries in English behave like regular verbs in French. For instance, they precede adverbs, negation, and floating quantifier (cf. 33a, 33b, 33c), and undergo direct movement to C (cf. 33d): (33) a. John has often kissed Mary b. John has not kissed Mary c. The kids have all eaten the chocolate d. Has John seen Mary? (from Roberts 2001: 120–1) These (a)symmetries are captured by Lasnik (1999, 2003) in (34) and (35), which provide two parameters to capture the V → T / T → V distinction. (33) VM Parameter (feature strength version) a. French verbs are fully inflected in the Lexicon (possibly correlating with the fact that there are no bare forms; even the infinitive has an ending). b. Have and be are fully inflected in the Lexicon (possibly correlating with the fact that they are highly suppletive). c. All other English verbs are bare in the Lexicon. (from Lasnik 1999: 105; Lasnik 2003: 12–13)



(35)

HEAD MOVEMENT IN THE CL AUSAL DOMAIN

239

Affixal T vs Featural T a. T is freely an affix or a set of abstract features. b. Finite featural T is strong in both French and English. c. Affixal T must merge with V, a PF process (distinct from head movement) demanding adjacency. (from Lasnik 1999: 105; Lasnik 2003: 13)

Although the facts discussed above are often taken for granted in the literature, they are not equally robust in all Romance languages. On the one hand, as noted by different authors, time adverbs do not occupy a fixed position in Spanish (cf. Gallego 2010; Depiante and Vicente 2012, and references therein for discussion): (36) a. Mourinho siempre dice lo que quiere Mourinho always say-3SG what wants-3SG b. Mourinho dice siempre lo que quiere Mourinho say-3SG always what wants-3SG c. Mourinho dice lo que quiere siempre Mourinho say-3SG what wants-3SG always ‘Mourinho always says what he wants to’

[Spanish] [Spanish] [Spanish]

On the other hand, notice that, although infinitives do not display morphological richness in any obvious sense, they seem to move to T too—if we build on the standard adverb placement tests (assuming these cannot move; cf. Chomsky 1995; Cinque 1999): (37) Decir siempre lo que se piensa no es bueno say-inf always what SE think-3SG not be-3SG good ‘Always saying what one thinks is not good’

[Spanish]

Following the logic of Chomsky’s (1981) conception of clustering effects (cf. Biberauer et al. 2010), the VM Parameter has been associated to other phenomena. Thus, languages that display VM display pro-drop (cf. Jaeggli 1982, 1984), cliticization (cf. Kayne 1991), restructuring (cf. Rizzi 1982; Cinque 2006; Wurmbrand 2001), and they typically lack VP ellipsis (with the exception of Galician-Portuguese; cf. Lasnik 1999; Goldberg 2005). (38) a. pro van dir que no passaria res pro-drop AUX.3PL say.INF that not would-happen anything ‘They said that nothing would happen’ b. La vam veure cliticization CLT.her AUX.1PL see.INF ‘We saw her’ c. La volem veure restructuring CLT.her want.1PL see.INF ‘We want to see her’ d. A: Vols els llibres? B: *Vull vp ellipsis want.2SG the books want.1SG ‘A: Do you want the books? B: I do’ [Catalan]

240

CONTEMPORARY LINGUISTIC PARAMETERS

The connection between person / number richness of the verbal paradigm and VM has been subject to criticism (cf. Biberauer et al. 2010). In this respect, Bibearuer and Roberts (2010) suggest that the differences between Romance and Germanic are correlated with the richness of the tense distinctions. As they note, these languages differ noticeably in the number of synthetic tense paradigms. The summary in (39) and (40) shows that Romance varieties are considerably richer than Germanic ones: (39) Romance (tense inflection) French: parle (present indicative/subjunctive), parlerai (future), parlerais (conditional), parlais (imperfect), [parlai (preterite), parlasse (past subjunctive)] Italian: parlo (present), parlerò (future), parlerei (conditional), parlavo(imperfect), parli (present subjunctive), parlassi (past subjunctive), parlai (preterit) Spanish: hablo (present), hablaré (future), hablaría (conditional), hablaba (imperfect), hablé (preterite), hable (present subjunctive), hablase (past subjunctive I), hablara (past subjunctive II) (40) Germanic (tense inflection) German: spreche (present indicative/subjunctive), sprach (past), spräche (past subjunctive) English: speak (present), spoke (past) Swedish: snakker (present), snakket (past) (from Biberauer and Roberts 2010: 266) As Biberauer and Roberts (2010) further point out, this new perspective has an effect on the classification we saw in (9), which should be modified as follows: (41) Language types according to their inflectional richness (final version) a. Rich agreement and rich tense inflection: hence v → T and null subjects, e.g. Italian, Greek, Spanish, etc. b. Poor agreement but rich tense: hence v → T, but no null subjects, e.g. French c. Poor tense and poor agreement: hence no v → T and no null subjects, e.g. Modern English, Mainland Scandinavian d. Rich agreement and poor tense: null subjects, but no v → T (no clear example) (adapted from Biberauer and Roberts 2010: 266) Details aside, even if morphological richness (either in person / number or in tense) of T is the ultimate key to these facts, and some variant of agreement operation is needed, it remains to be seen why movement is also needed. In the GB literature, the answer was based on feature strength, but in a framework like Chomsky (2000, 2001) there is no reason to suppose that agreement triggers movement. Perhaps the process is morpho-phonological in nature (as in Chomsky 2001), but some facts suggest that VM is genuinely syntactic, as it has an important role in the licensing of certain word order patterns (cf. Gallego 2013).



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

241

10.3.2. Variation in T-to-C movement In this section we shift our attention to the second key case of VM parameters, namely the one dealing with T → C movement.11 We will dub this parameter “TM Parameter,” merely to indicate the fact that it is T itself or something occupying that position that moves to C—not the verb itself. As noted at the outset, the TM Parameter essentially covers V2 phenomena, regardless of whether it is full (as in Germanic languages of the Dutch type) or residual (as in Romance languages, and English too): (42) Full / Residual V2 phenomena a. Morgen / warm werden sie dieses Buch lesen? [German] tomorrow when will they this book read ‘Tomorrow they will read this book? / When will they read this book?’ b. When will they read this book? c. Quand liront -ils ce livre? [French] when will-read-they this book? ‘When will they read this book?’ (from Roberts 2005: 20) Before discussing bona fide V2 patterns, there is one other case study that should be considered, namely that of V initial languages. These can be VSO or VOS: the former including Semitic and Celtic languages (like Arabic or Welsh), whereas the latter include Austronesian and Mayan languages (like Malagasy or Totzil). For presentation purposes, we will focus on VSO languages, on the assumption that they exhibit the key feature that presumably involves VM to C. It was again Emonds (1980) who first argued for a VM rule in order to account for the VSO word order of languages like Welsh. Standard examples can be seen in (43), where either the verb itself or some auxiliary (do-like) element must occupy a pre-subject position (fe and mi being clause initial particles): (43) a. Fe / mi welais i Megan PRT saw I Megan ‘I saw Megan’ b. Fe / mi wnes i weld Megan PRT did I see Megan ‘I saw Megan’

[Welsh] [Welsh] (from Roberts 2005: 7–8)

Roberts (2005) argues that the right analysis for VSO sentences in Welsh does not involve V2 proper (that is, T → C), a derivation that was argued for by Déprez and Hale (1986) and Doherthy (1996). To defend that view, Roberts reports three pieces of evidence: (i) lack of root-embedded asymmetries in VM, (ii) the type of elements that can be found in the left periphery of Welsh (and Irish), and (iii) the specific properties of the auxiliary verb bod (Eng. “be”), along with the distribution of other tensed verbal forms. Let us concentrate on the first argument, for it has a more close connection with V2. As the literature on V2 has made clear, this process seems to be

242

CONTEMPORARY LINGUISTIC PARAMETERS

restricted to root structures (we come back to this), as shown by indirect questions, which fail to display V2 (compare 44 vis-à-vis 42):12 (44) a. *Ich frage mich, ob morgen wird Maria dieses Buch lesen [German] *I ask me if tomorrow wil Maria this book read b. *I wonder if will she read the book c. *Je me demande si lira-t -elle le livre [French] *I me ask if will-read-she the book (from Roberts 2005: 20) When we consider Welsh (and other Celtic languages, such as Breton or Irish), we readily see that the VSO pattern is not restricted to root clauses (cf. McCloskey 1996): (45) a. Tybed a geith hi ddiwrnod rhydd wythnos nesa? [Welsh] wonder.1SG PRT will.get she day free week next ‘I wonder if she’ll get a free day next week’ b. Goulenn a reas hag-en oac’h eveurus [Breton] asked.3SG PRT did.3SG whether were.2SG happy ‘He asked whether you were happy’ c. Chuir se ceist ort an raibh tii sasta [Irish] asked.3SG he question to-you PRT were you content ‘He asked whether you were content’ (from Roberts 2005: 20–21) In the examples above, the particles a, hag-en, and an are parallel to German ob, English if, and French si in (44). Under the assumption that these elements occupy the C head, the verbs in Celtic languages cannot be there too: they do not compete for the same position, unlike what we see with true V2. One other crucial property of VSO languages that teases them apart from V2 is the fact that the V initial word order is unmarked in declarative sentences—it does not require any other element to undergo movement to the CP layer of the clause. From all this, Roberts (2005) concludes that the verb in the Welsh examples in (43), repeated here as (46) for convenience, should be analyzed as in (47): (46) a. Fe / mi welais i Megan PRT saw I Megan ‘I saw Megan’ b. Fe / mi wnes i weld Megan PRT did I see Megan ‘I saw Megan’

[Welsh] [Welsh] (from Roberts 2005: 7–8)

(47) a. [FinP Mi [AgrSP welais [TP i [VP ti twelais Megan ] ] ] ] b. [FinP Mi [AgrSP wnes [TP I [VP ti weld Megan ] ] ] ] (adapted from Roberts 2005: 40)



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

243

Once VSO languages can be said to lie outside of the TM Parameter, let us focus on V2 phenomena with more detail, which roughly instantiate the abstract pattern in (48): (48) V2 pattern XP Verb [TP … ] As has been noted in the literature (cf. Holmberg 2010b and references therein), V2 is present in Germanic languages (except for English), some Romance languages, Breton (a Celtic language), Estonian (a Finno-Ugric language), Sorbian (a Slavic language), among others. Holmberg (2010b) argues that V2 languages can be divided into two main groups, depending on whether V2 is present in root clauses or extends to all finite clauses (root and embedded). This author calls them “C-V2 languages” and “I-V2 languages” respectively: (49) V2 languages a. C V2 languages (V2 only in root clauses; e.g. Mainland Scandinavian and Continental Germanic languages) b. I V2 languages (V2 in inflected clauses; e.g. Icelandic, Yiddish, or Breton) Following Rizzi (1996) and Rizzi and Roberts (1989), we have previously referred to “residual V2” in order to describe the V2 pattern that surfaces in, for instance, Romance “interrogative sentences” (where the “verb preposing” rule of Torrego 1984 applies). Details aside, this is intended to mean that V2 does not take place in all the relevant contexts. Consider the case of English: (50)

a. Which battery type (would) you (*would) recommend? b. This battery type (*would) I (would) not recommend c. None of them (would) I (*would) recommend d. So good (was) his performance (*was), that he got a standing ovation e. In the sink (*found) John (found) a spider (from Holmberg 2010b: 2–3)

In the examples above, V2 emerges whenever the constituent moved to [Spec, CP] is a wh-phrase or a negative phrase, but not a topicalized element, be it argumental (cf. 50b) or not (cf. 50e). That said, it is also important to point out that the list of elements that move to the CP layer (occupying the first position) and pied-pipe the verb along its way is rather varied (they appear in small capitals below): subjects, objects, adverbials, predicates, negation, and particles. (51) a. André het gister die storie geskryf André has yesterday the story written ‘André wrote the story yesterday’ b. Tidningar läser barnen inte newspapers read the-children not ‘Newspapers, the children don’t read’

[Afrikaans] [Swedish]

244

CONTEMPORARY LINGUISTIC PARAMETERS

c. Heldigvis er den politiske kunst på vej tilbage [Danish] luckily is the political art on way back ‘Luckily, political art is making come-back’ d. Wat hebben jullie besteld? [Dutch] what have you ordered ‘What have you ordered?’ e. Sint er jeg ikke, bare veldig skuffet [Norwegian] angry am I not only very disappointed ‘I’m not angry, just very disappointed’ f. Ekki veit ég hvað ætlar úr þér að verða [Icelandic] not know I what will of you to become ‘I don’t know what will become of you’ g. Es war kein optimales Spiel, trotzdem haben wir gewonnen [German] it was no optimal game yet have we won ‘It wasn’t an optimal game, yet we won’ (from Holmberg 2010b: 7–11) There are, however, a few exceptions to the generalization that V2 languages can only feature one constituent preceding the verb. An interesting case concerns left dislocation, which brings up the possibility (first discussed in Cinque 1990) that it involves base generation (cf. Rubio 2014 for arguments against this, and Ott 2013 for an alternative analysis, involving ellipsis). Germanic V2 languages display two types of left dislocation structures: Hanging Topic Left Dislocation (which is described as being more integrated with the clause, a fact that would explain why it is barred in embedded contexts) and Contrastive Left Dislocation, illustrated in (52a) and (52b) below (cf. Fortuny, this volume): (52) a. Peter, ich werde ihn Morgen sehen Peter I will him tomorrow see ‘Peter, I will see him tomorrow’ b. Die man, die ken ik niet that man him know I not ‘That man, I did not know him’

[German] [Dutch] (from Holmberg 2010b: 14)

In both cases, V2 is not respected, which begs the question of how to “calculate” V2. An account based on purely linear order will not be enough in these cases. Whatever else is needed, there are two other, independent but ultimately related, questions: (i) why the XP element moves to [Spec, CP], and (ii) why the verb moves to C. There seems to be a consensus in the literature (cf. Truckenbrodt 2006 and references therein) that V2 is somehow connected to “illocutionary force” (more precisely, “assertion”), which would explain why V2 is for the most part a root phenomenon—in this respect, embedded clauses with V2 have also been attributed a root nature (cf. Hooper and Thompson 1973; Emonds 1976), a fact that would explain their island status. This can be seen in the example in (53), where the particle att (Eng. “that”) is taken to be irrelevant for V2 calculation.



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

245

(53) Vilken festii sa hon [CP att vi (*behöver) inte (behöver) köpa … [Swedish] which party said she that we need not need buy roliga hattar till ti ] ? funny hats for ‘Which party did she say that we don’t need to buy funny hats for?’ (from Holmberg 2010b: 21) The reason why V2 blocks extraction is far from clear, but it resembles the effects of indicative mood in languages of the Slavic family (cf. Torrego and Uriagereka 1992, Khomitsevich 2007)—an island behavior that is absent in Romance. This is shown in the Russian examples below, where the subjunctive morpheme in C (-by) bars wh-movement: (54) a. ?*Kogo ty dumaeš, [CP čto Maša ljubit ]? whom you think-2.sg that Masha love-3.sg ‘Who do you think that Masha loves?’ b. Kogo ty xočeš’, [CP čtoby Maša poljubila]? whom you think-2.sg that-subj Masha love-3.sg ‘Who do you want Masha to love?’

[Russian] [Russian]

Island effects reappear, interestingly, in covert phenomena, such as time sequences, NEG-raising, NPI licensing, and QR. This is shown in the Spanish data below, taken from by Torrego and Uriagereka (1992): (55) a. Platón quiere [CP que Aristóteles {lea/*leyera} a Sócrates ] Plato want.PRES.3SG that Aristotle read.PRES/PST.SBJ to Socrates ‘Plato wants for Aristotle to read Socrates’ b. Platón dice [CP que Aristóteles {lee/leía} a Sócrates ] Plato say.3SG that Aristotle read.PRES/PST.IND to Socrates ‘Plato says that Aristotle {reads/read} Socrates’ (56) a. Sancho no decía [CP que Don Quijote desvariaba ] Sancho not said.3SG that Don Quixote act.crazily.PST.IND.3SG ‘Sancho did not say that Don Quixote acted crazily’ b. Sancho no quería [CP que Don Quijote desvariara ] Sancho not wanted.3SG that Don Quixote act.crazily.PST.SBJ.3SG ‘Sancho wanted that Don Quixote did not act crazily’ (57) a. *El Cid no dijo [CP que lo vio moro ni cristiano ] *the Cid not said.3SG that him saw.IND.3SG moor nor christian *‘The Cid did not say that a soul saw him’ b. *El Cid no quería [CP que lo viese moro ni cristiano ] *the Cid not wanted.3SG that him saw.SBJ.3SG moor nor christian *‘The Cid wanted that not a soul would see him’

246

CONTEMPORARY LINGUISTIC PARAMETERS

(58) a. Alguien piensa [CP que ha sido solucionado todo problema ] someone think.3SG that have.IND been solved every problem ‘Someone thinks that every problem has been solved’ b. Alguien desea [CP (que) sea solucionado todo problema ] someone wish.3SG that be.SBJ.3SG solved every problem ‘Someone wishes every problem to be solved’ Yet more interestingly, Torrego and Uriagereka (1992) noticed that Romance subjunctives preclude some well-known root features: (i) hanging topics (cf. 59), (ii) emphasis markers (cf. 60), and (iii) focus fronting (cf. 61). (59) a. Creía [CP que, en cuanto a la tragedia, debía haber tres unidades ] thought.3SG that as for to the tragedy there.should.be.3SG three units ‘He thought that, as for the Tragedy, there should be three units’ b. Quería [CP que (,*en cuanto a la tragedia,) hubiera tres unidades ] wanted.3SG that as for to the tragedy there.were.3SG three units ‘He wanted that, as for the Tragedy, there were three units’ (60) a. Creía [CP que sí que la tierra era plana ] believed.3SG that yes that the earth be.IND.3SG flat ‘He believed that indeed the earth was flat’ b. Quería [CP que (*sí que) la tierra fuera plana ] wanted.3SG that yes that the earth be.SBJ.3SG flat ‘He wanted that (indeed) the earth be flat’ (61) a. Dijo [CP que MUCHAS COSAS había visto ] said.3SG that many things had.IND.3SG seen He said that he had seen MANY THINGS b. *Quería [CP que MUCHAS COSAS viera ] wanted.3SG that many things saw.SBJ.3SG ‘He wanted that he saw MANY THINGS’ However the connection between indicatives and V2 is to be established (technically), it seems clear that the same subjacent cause is behind them. Before we conclude, let us go back to the questions we asked above: (i) why the XP element moves to [Spec, CP], and (ii) why the verb moves to C. We didn’t say anything about (i), but it could plausibly be related to the discourse-oriented effects of XP movement (cf. Chomsky 2001). The reason behind (ii) could be the same, under the assumption that C is a Force-encoding head (cf. Rizzi 1997). Be all that as it may, it is unclear why both phenomena go hand in hand. But from a broader perspective, we can easily see that they do not: (62) is an assertion, yet nothing moves to [Spec, CP] (unless we endorse a generalized V2 analysis, as in Den Besten (1983), or else take V2 to apply at Logical Form; cf. Koster 2003), and the verb (auxiliary) presumably remains in T. (62) The journalist will contact his sources A way to make (62) consistent with the idea that assertion lies behind V2 (where of



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

247

course V2 is a misnomer for TM) is to assume that the verb is in C already. This can be done if T and C are the same lexical item. In other words, if (62) is represented as in (63): (63) [CP The journalisti [C’ will [vP ti contact his sources ] ] ] In (63), there are no independent C and T projections, they are one and the same. Or put differently, C is T, an idea that was first sketched by Chomsky (1986b).13 If so, then the verb is in C with no additional commentary. This said, one may actually wonder why the relevant dependency with an assertion projection cannot be done at a distance. It may well be due to the fact that the root structure is actually embedded within a performative shell that has to spell-out its complement (under standard phase-cyclic mechanics). If so, then the head of this shell will require some stuff to occupy its complement domain, for otherwise the process may be vacuous (cf. Kayne 2006 for related ideas).

10.4. Conclusions In the previous pages, we have discussed some well-known (and well-documented) cases of VM in which parametric variation is found. Assuming the standard literature, we have referred to these parameters as VM (V → T) and TM (T → C) parameters, putting aside a great deal of provisos. In general, we can summarize the two relevant situations as in (64) and (65)—recasting what we already said, and taking φ-features to be the source of both steps (although it may well be that T → C is ultimately due to discourse-oriented reasons, of the type Rizzi 1997 and Miyagawa 2010 have argued for on independent grounds): (64) VM Parameter In a language L, the φ-features of T are {weak/strong} (65) TM Parameter In a language L, the φ-features of C are {weak/strong} The fact that we have encountered several counterexamples and qualifications to make may be telling us that terms like “VM,” “VT,” or “V2” (and of course “VX language,” “pro-drop language,” “wh-in-situ language,” “satellite-framed language,” etc.) do not name meaningful, UG rooted, parameters. They may instead serve as methodological labels to capture observable properties. In the case of V2 phenomena, this is expected, since UG should not make reference to linear order (what comes first, second, third, etc.). This is good because it allows us to study languages and come up with generalizations that are somewhat independent of specific frameworks, and bad because we ultimately want to characterize UG, as opposed to specific languages. These tasks need not collide, but they are different, as has often been noted. This chapter has aimed at offering a characterization of the factors that underlie

248

CONTEMPORARY LINGUISTIC PARAMETERS

VM processes. Considering the two broad parameters we have focused on, it is interesting to note that one of them is morphological (v → T), the other (T → C) is not clearly so, which probably indicates that the latter is not a parameter in itself. That would align with the idea that parameters are to be found in the set of externalization strategies (cf. Chomsky 2001, 2010), which are subject to crosslinguistic variation. It would further indicate that many things called “parameters” are something else.

Notes *

We would like to thank Antonio Fábregas for his editorial help. Usual disclaimers apply. This work has been partially supported by grants from the Ministerio de Ciencia e Innovación (FFI2011-29440-C03) and the Generalitat de Catalunya (2014SGR-1013).

1

Various comprehensive and up-to-date works on head movement have appeared in the recent literature (cf. Roberts 2001, 2010a, 2011; Barrie and Mathieu 2014, and references therein).

2

Of course, there are other options, both within the vP area (if we include an AspP, which seems to be necessary to capture properties of Slavic languages; cf. Chernova 2015; Svenonius 2004; Dyakonova 2009) and within the CP area (if we include fine-grained projections, like ModP, FinP, TopP, FocP, and the like; cf. Rizzi 1997, Poletto 2000), but we will stick to this basic picture.

3

There is the issue of whether languages with obligatory auxiliaries (e.g. Basque) involve a separate v head. We put aside this issue. Cf. Ortiz de Urbina (1989).

4

See Lasnik (1999) for some cases in which V can stay in situ under ellipsis conditions. In such cases, the verb is elided as a last resort strategy, to avoid a PF crash that would otherwise occur, given the obligatory status of V → v.

5

This type of evidence assumes the idea that the elements in (10), (11), and (12) occupy a fixed position (cf. Cinque 1999, 2004 for more recent arguments in favor of that perspective). An alternative scenario could be envisaged if they occupied different positions in different languages (cf. Williams 1994) or if they could actually occupy different position even within the same language (cf. Bobaljik 1999; Ernst 2002).

6

This idea has been revamped in Chomsky (2014a), with consequences for null subjects, labeling, and the status of the EPP.

7

Building on work by Matushansky (2006), Roberts (2010a) notes that an independent argument to regard HM as syntactic comes from A-movement, which has weak discourse-oriented effects too.

8

One could think that the representation of (28b) is misleading, for what could be a phrase. However, Donati (2006) provides evidence that what must be a head (i.e. a pronoun), as it cannot take complements:



(i) I read what (*books) you read



An alternative way to understand (i) is to take what to be minimal but nonetheless complex (like proper names in Longobardi’s 1994 analysis).

9

We use the term “paradigm,” which we interpret in a lexical sense. However paradigms are formed, it is not in the syntax (or s-syntax, to use Hale and Keyser’s 1993



HEAD MOVEMENT IN THE CL AUSAL DOMAIN

249

terminology). For a more general discussion about the paradigmatic / syntagmatic cut, we refer the reader to Otero (1976), and Uriagereka (2008). 10 One of the main advantages of feature strength, apart from its easily recognizable status, is that it helped account for the moment when a parameter takes place: before or after Spell-Out. Thus, VM would be akin to object shift or wh-movement, phenomena that can take place covertly or overtly, depending on the language. Feature strenght was dispensed with the moment Chomsky (2000) unified the cycles, but the basic facts of the distinction remained, so Chomsky (2004) brought it back. See Chomsky (2014a) for recent explorations of feature strength in the context of the EPP and labeling. 11 In this chapter we will not consider those languages where T-C takes place in the covert component, as in Romance subjunctives (if Kempchinskys 1987 ideas are correct). Also left aside is an in depth study of the consequences of T-C interactions for extraction, complementizer deletion, and ECM (cf. Pesetsky and Torrego 2001, Gallego 2010). The fact that T-C dependencies are connected to V-T ones suggests a deeper parametric connection between the VM Parameter and the TM Parameter. 12 The availability of V2 in embedded contexts seems possible (in fact, necessary) for some Romance languages, and some English dialects too (cf. Henry 1995). This may suggest, as noted above, that the French / English distinction that Pollock (1989) took as crucial should not be regarded as the baseline. 13 Gallego (2014) develops Chomsky’s (1986b) analysis in the context of Phase Theory in order to derive the process of feature inheritance.

11 Wh-movement* Željko Bošković University of Connecticut

The chapter discusses the typology of wh-questions with respect to multiple wh-fronting, single wh-movement, and wh-in-situ languages. A variety of wh-insitu strategies are discussed since all languages examined in the chapter allow at least some wh-phrases to remain in situ. Superiority effects and single-pair/ pair-list answers are used as tests in establishing the precise nature of wh-fronting. The driving force of wh-movement and successive-cyclic wh-movement are also discussed as well as the related question of wh-islands in light of the fact that some languages allow extraction out of wh-islands in some contexts.

11.1. Introduction The goal of this chapter is to discuss a number of issues pertaining to the typology of wh-movement and wh-questions more generally. The focus will be on multiple wh-questions since they are more conducive to revealing the options languages may employ in question formation. Additionally, given the nature of the volume, the focus will be on establishing language types, i.e. establishing the limits of crosslinguistic variation in this domain, rather than providing detailed analyses of this variation. Still, some analyses will be provided to facilitate the discussion, with references to relevant works for the interested reader.

11.2. Multiple-question typology A superficial examination reveals three language types regarding multiple questions: the English-type, where only one wh-phrase moves, the Chinese-type, where they all stay in-situ, and the Bulgarian-type, where all wh-phrases move (multiple wh-fronting (MWF) languages).1

252

CONTEMPORARY LINGUISTIC PARAMETERS

(1) What did John give to who? (2) John gei-le shei shenme? John give-PRFV who what ‘What did John give to who?’

[Chinese]

(3) Na kogo kakvo dade Ivan? to who what gave Ivan ‘What did Ivan give to who?’

[Bulgarian]

As is often the case, superficial examination does not reveal the true extent of cross-linguistic variation. In particular, it has been argued that French represents a separate wh-type and that there is no separate MWF-type regarding wh-movement itself, languages considered to belong to this type being scattered across the English, Chinese, and French-type.

11.2.1. French French appears to employ both the English and the Chinese option, as in (4). As a result, French was often assumed to be a simple mixture of these two types. (4) a. Qu’ a-t-il what has he b. Il a donné he has given

donné à qui? given to who quoi à qui? what to who

[French]

If this were correct both the English and the Chinese strategy should always be possible in French. However, the in-situ strategy has a very limited distribution in French, which indicates it is of different nature from Chinese wh-in-situ. Wh-in-situ is allowed in French short-distance null C matrix questions (5), but not in embedded and overt C questions (6, 7). (The judgments are given only for the true question reading. Overt C questions are possible only in some dialects.) (5) a. Tu as vu qui? you have seen whom b. Qui as-tu vu? (6) a. Pierre a demandé qui tu as vu. Pierre has asked whom you have seen b. *Pierre a demandé tu as vu qui. (7) a. Qui que tu as vu? whom C you have seen ‘Who did you see?’ b. *Que tu as vu qui?

WH-MOVEMENT

253

Bošković (1998) argues a simple lexical difference between English and French is responsible for this pattern.2 Lexical insertion is standardly assumed to be disallowed in PF and LF. Chomsky (1995) deduces this from Full Interpretation, which requires what is present at the interface levels to be interpretable by the interfaces. A lexical item like John has formal, phonological, and semantic features. If John is inserted in LF the derivation crashes because LF cannot interpret the phonological features of John. If John is inserted in PF, PF cannot interpret the semantic features of John. The only way to derive legitimate PF and LF is for John to be inserted before SS. PF then strips off the phonological features, the semantic features of John proceeding into LF. This allows LF insertion of phonologically null elements (as long as it obeys the cycle, occurring at the top of the tree). Bošković (1998) argues this is what happens with French wh-in-situ, wh-C being inserted in LF. Wh-in-situ is in fact possible in French only where LF insertion is in principle possible, namely, when the C is phonologically null (compare 5a / 7b) and when it is located at the top of the tree (compare 5a / 6b). (5a) is then an IP in overt syntax, CP being inserted covertly. What is then responsible for the contrast between (5a) and English (8)? (8) *You have seen whom? Bošković (1998) suggests matrix interrogative C in English is lexically specified as a PF verbal affix, which is not the case with French interrogative C. Independent evidence for this difference is provided by the fact that, like tense affix -ed, interrogative C in English must be adjacent to a verbal element, which is not the case with French. (9) Qui tu as vu? whom you have seen (10) *Whom you have seen? Since English +wh-C has phonological information in its lexical entry (PF affix specification), it cannot be inserted in LF. Under this view, both French and English have a +wh-C with a strong wh-feature, which under the virus theory of strong features must be checked through wh-movement as soon as it is inserted. The difference between French and English is that French +wh-C has no phonological lexical specification, hence can be inserted in LF, which is not the case with English +wh-C. The simple lexical difference accounts for the contrast in (5a) / (8) (unifying it with (9) / (10). However, since French +wh-C has a strong wh-feature, in contrast to Japanese +wh-C, which has a weak wh-feature (see below), it still triggers overt wh-movement when inserted overtly, which accounts for the contrast between (6b)/ (7b) and Japanese (22). French long-distance questions, however, exhibit rather interesting behavior. (11) a. *Jean et Pierre croient que Marie a vu qui? *Jean and Pierre believe that Marie has seen whom b. Qui Jean et Pierre croient-ils que Marie a vu?

254

CONTEMPORARY LINGUISTIC PARAMETERS

(12) Qui croit que Marie a vu qui? who believes that Marie has seen whom    (Bošković 1998) While (11b), involving overt wh-movement, is acceptable, (11a), involving covert movement of the same element, is unacceptable. On the basis of this and similar facts (including the clause-boundedness of QR), Bošković (1998) argues LF movement is more local than overt movement, which can be captured under Chomsky’s (1995) Move-F proposal. Chomsky observes a natural consequence of the assumption that movement is feature-checking driven is that, all else being equal, Move should apply to features, not to syntactic categories. Overt movement, which feeds PF, still must apply to whole categories, assuming lexical items with scattered features cannot be interpreted at PF. Since PF interpretability is irrelevant to LF, in LF Move applies only to features. Chomsky instantiates feature-movement as adjunction to X0-elements. (11a) then must involve head-movement out of a finite clause, which, as is well known, is disallowed. To account for the contrast in (5) / (11a), Bošković uses the relativized version of the Head-Movement Constraint (Rivero 1991): feature-movement to interrogative C, which involves A’-head movement, crosses an A’-head in (11a) (que), but not (5).3 What about (12)? (12) contains another wh-phrase that is located overtly in SpecCP. This wh-phrase can check the strong +wh-feature of C, so that there is no need for the wh-phrase in-situ to move in LF. It must then be the case that the wh-in-situ in (12) does not move in LF, in contrast to the wh-phrase in (11a), given the clause-boundedness of such movement in French. Bošković (1998) then suggests that qui in (12) is associated with the wh-C without movement through unselective binding. In (11a), the wh-phrase in-situ is the only element that can check the strong +wh-feature of C and is, therefore, forced to undergo LF wh-movement. Unselective binding is not an option, since it would leave the strong +wh-feature of C unchecked. Following Higginbotham (1983, 1985), where N has an index-argument that must be bound, Tsai (1994) and Reinhart (1995) argue wh-NPs have an open position hence can introduce variables in situ.4 As a result, they can be unselectively bound by C. Not being nominal, wh-adverbs do not have an open position, hence cannot introduce variables in situ. From this perspective, German wh-adjuncts are rather interesting. In contrast to English, German allows wh-adverbs in-situ (for an account of the contrast, see Bošković 2000). (13) *Who left why? (14) Wer is warum ekommen?    [German] who is why come Importantly, in contrast to wh-arguments, long-distance wh-adjunct in-situ questions are disallowed. Wh-adjuncts can, however, move long-distance overtly. (15) *Wer hat gesagt daß Fritz warum ein Buch gelesen hat? *who has said that Fritz why a book read has ‘Who has said that Fritz has read a book why?’ (Müller and Sternefeld 1996) (16) Warum hat Hans gesagt daß Fritz t ein Buch gelesen hat?

WH-MOVEMENT

255

Since wh-adverbs cannot be unselectively bound, wh-adverbs in-situ must undergo LF wh-movement even when another wh-phrase is present in +wh-SpecCP. The contrast in (15)–(16) then confirms that covert movement is more local than overt wh-movement, which can be captured under Move-F. The superficial variation regarding locality discussed above, as in the contrast in (11a) / (12) regarding qui, doesn’t require positing any differences in the locality of LF wh-movement: while qui in (11a) undergoes LF wh-movement (i.e. Move-F), qui in (12) doesn’t move. However, there are cases that show there may be variation regarding the locality of LF movement itself. Thus, Brazilian Portuguese (BP) (Zocca 2011) and ASL (Wood 2009) behave like French regarding (5)–(7), which indicates they should also be treated in terms of LF insertion of the strong +wh-C. However, they allow (11a), as illustrated below for BP (as well as other unacceptable cases of wh-in-situ in French (involving quantificational interveners) that Bošković rules out by locality restrictions on Move-F). (17) (*Que) você viu quem?    [Brazilian Portuguese] C you saw who (18) *O Pedro perguntou você viu quem *the Pedro asked you saw who (19) O Pedro acredita que a Maria viu quem? the Pedro believes that the Maria saw who Assuming parallelism between (5)–(7) and (17)–(18) requires adopting the LF C-insertion analysis for BP wh-in-situ (i.e. not treating BP as Chinese), as Zocca (2011) argues, we may conclude that the locality of French and BP LF wh-movement is different. Wood (2009) pursues this account for ASL (see also section 11.2.3. for ASL): in contrast to French LF wh-movement, which involves Move-F, hence its clause-boundedness, ASL LF wh-movement involves Huang-style phrasal movement to SpecCP, which makes intervening A’-heads irrelevant.5 However, Zocca treats French wh-in-situ as involving LF movement of the Q-morpheme (see section 11.2.4) and argues the Q-morpheme can be generated separately from the wh-phrase in BP but not French: Q, generated in CP, then binds quem in (19).6 French and BP do not exhaust variation regarding Romance wh-in-situ. Thus, Spanish differs from French/BP in that its wh-in-situ must be sentence final. (20) *Tú invitaste a quién a tu fiesta? *you invited to who to your party

[Spanish]

(21) Tú invitaste a tu fiesta a quién? Reglero (2007) argues the reason for this is phonological. Assuming lower-copy pronounciation (LCP) is possible if higher-copy pronounciation would induce a PF violation (see section 11.2.4), she argues Spanish wh-in-situ involves overt movement+LCP: the Nuclear Stress Rule and the Focus Prominence Rule require stress assignment to the most deeply embedded copy, which then must be pronounced (note Spanish allows 11a).7

256

CONTEMPORARY LINGUISTIC PARAMETERS

11.2.2. Wh-in-situ languages What about true wh-in-situ languages like Japanese? (22a) should not receive the same analysis as French (5b), since, in contrast to French, embedded in-situ questions are allowed in Japanese. The LF C-insertion analysis would predict (22b) to be ungrammatical. Also, Japanese +wh-C is lexically realized, hence cannot be inserted in LF. (22) a. Anata-ga dare-o mita ka you-NOM who-ACC saw Q b. Peter-wa [anata-ga dare-o mita ka] tazuneta Peter-TOP you-NOM who-ACC saw Q asked Long-distance in-situ questions are also acceptable in Japanese. (23) John to Mary-wa [Peter-ga dare-o mita to] sinziteiru ka John and Mary-TOP [Peter-NOM who-ACC saw that believe Q (22)–(23) indicate Japanese wh-questions cannot involve LF feature-movement/Cinsertion. However, Japanese wh-phrases are not completely insensitive to locality constraints: adjunct wh-phrases are disallowed within islands.8 (24) *John to Mary-wa [Peter-ga

naze kuruma-o naosita kadooka] siritagatteiru ka *John and Mary-TOP [Peter-NOM why car-ACC fixed whether want-to-know Q

Given (24), some kind of movement must occur in Japanese questions. As noted above, this cannot be Move-F. Also, wh-phrases themselves are clearly not moving overtly. There are several possibilities then. One is phrasal LF movement as in Huang (1982). Another option is not to have movement in (22)–(23), only unselective binding. Since wh-adjuncts cannot be unselectively bound, naze in (24) still needs to undergo LF movement, assuming some kind of wh-C / wh-phrase association is necessary. Another possibility is that a null wh-operator undergoes movement, as in Watanabe (1992) (see also Aoun and Li 1993), who argues the difference between English and Japanese regarding what moves is a result of the shape of wh-phrases and their quantificational force (i.e. their internal structure and semantic properties). Under this analysis, interrogative SpecCPs are filled overtly in Japanese, which implies the +wh-feature of C in Japanese is strong. I now turn to MWF, which has been a fertile ground for testing a number of syntactic and semantic issues. 1 will discuss several issues that arise under the typology of wh-fronting established in Bošković (2002) regarding Superiority, single-pair/pair-list answers, and the driving force of MWF.

11.2.3. Multiple wh-fronting Rudin (1988) argues that in spite of superficial similarity, MWF constructions display two different structures. For her, the initial wh-phrase always moves to

WH-MOVEMENT

257

SpecCP. However, she argues MWF languages differ regarding the location of non-initial wh-phrases, i.e. movement that Bošković (2002) calls non-wh-fronting. In Bulgarian, all wh-phrases move to SpecCP, which is not the case in SerboCroatian (SC). As a result, fronted wh-phrases cannot be split by non-wh-material in Bulgarian, but they can in SC.9 (25) ?*Koj, spored tebe, kakvo e kupil? ?*who according to-you what is bought ‘Who, according to you, bought what?’ (26) Ko, po tebi, šta kupuje? who according to-you what is-buying Bošković (2002) makes two modifications of Rudin’s analysis. First, following Stjepanović (1999a), Bošković argues non-wh-fronting involves focus fronting, SC wh-phrases being inherently focused. This follows the line of work originating with Horvath (1986), who makes a correlation between movement of wh-phrases and movement of contrastively focused non-wh-phrases whereby a number of languages that overtly move contrastively focused non-wh-phrases are analyzed as having focus-fronting of wh-phrases. One argument for the focus-movement analysis from Bošković (2002) involves D-linked wh-phrases, which, due to their semantics, do not undergo focusmovement.10 As a result, in contrast to non-D-linked wh-phrases, D-linked wh-phrases can stay in-situ (27). Furthermore, even when they move, they do not move to the same position as wh-phrases that undergo focus-movement, as the contrast in (25) / (28) indicates. (27) a. Koj e kupil koja kniga? who is bought which book b. *Koj e kupil kakvo? *who is bought what

[Bulgarian]

(28) ?Koj, spored tebe, koja kniga e kupil? ?who according you which book is bought ‘Who, according to you, bought which book?’ Even echo wh-phrases must move in the languages in question, which confirms the movement is independent of the movement involved in true questions. (29) ?*Ona će poljubiti KOGA? ?*she will kiss who

[Serbo-Croatian]

(29) is unacceptable on the request-for-repetition echo-question reading. However, Bošković notes it is acceptable on the reading on which it expresses surprise (on different types of echo-questions, see Pope 1976). This follows under the focusmovement analysis since the value of the echo wh-phrase is fully known to both the speaker and the hearer on the surprise reading, but not on the request-for-repetition reading.

258

CONTEMPORARY LINGUISTIC PARAMETERS

Bošković (2002) also argues not all Slavic languages have true wh-movement— some only have focus fronting. He gives several tests for teasing the two apart. Thus, he argues Superiority effects (strict ordering of fronted wh-phrases) are present only with wh-movement (see below for an account). It then follows that Bulgarian (30) involves wh-movement, while SC (31) doesn’t. (30) a. Koj kogo običa? who whom loves ‘Who loves whom?’ b. *Kogo koj običa? (31) a. Ko koga voli? who whom loves b. Koga ko voli? The Superiority test is confirmed by single-pair (SP)/pair-list (PL) answers. Bošković (2002, 2003a) shows languages with overt wh-movement require a PL answer for examples like (32).11 (32) cannot be felicitously asked in the following situation: John is in a store and sees somebody buying an article of clothing, but does not see who it is and does not see exactly what the person is buying. He goes to the sales clerk and asks (32). (32) Who bought what? Whereas German patterns with English, wh-in-situ languages Japanese, Hindi, and Chinese allow SP answers in such questions. Particularly interesting is French, which allows SP answers, but only with in-situ questions like (33a), not (33b). (33) a. Il a donné quoi à qui? he has given what to who b. Qu’a-t-il donné à qui? Bošković thus establishes the generalization that the availability of SP answers depends on the possibility of not moving any wh-phrase to SpecCP overtly. Turning to Slavic, Bošković (2002) observes that Bulgarian, where interrogative SpecCPs are obligatorily filled overtly, patterns with English: (34) requires a PL answer. (34) Koj kakvo e kupil? who what is bought SC patterns with wh-in-situ languages. (35) can have either a PL or an SP answer. (35) Ko je šta kupio? who is what bought Other MWF languages confirm this analysis. Russian, Polish, and Czech also do

WH-MOVEMENT

259

not exhibit Superiority effects and allow SP answers in regular wh-questions, while Romanian and Yiddish show Superiority effects and disallow SP answers (see Bošković 2007b and references therein). So far, we have a division of MWF languages into two groups regarding wh-movement in examples like (3). However, Bošković (2002) argues there are three different types of MWF languages based on the fact that non-wh-movement MWF languages do not behave uniformly in other contexts. Thus, while Russian never shows Superiority effects and Bulgarian always shows them, SC actually shows Superiority effects in some contexts. In contrast to (31), SC shows Superiority effects in long-distance, embedded, and overt-C questions, which are exactly the contexts where French must have wh-movement.12 (36) a. Ko koga kažeš da je istukao? who whom say that is beaten ‘Who do you say beat whom?’ b. *Koga ko kažeš da je istukao? (37) a. Pričali su o tome ko šta mrzi talked are about that who what hates ‘They talked about who hates what’ b. *Pričali su o tome šta ko mrzi (38) a. Ko li koga voli? who C whom loves ‘Who on earth loves whom?’ b. *Koga li ko voli? SC thus has Superiority effects where French must have wh-movement, Bulgarian where English must have wh-movement (all contexts), and Russian where Chinese must have wh-movement (never). This can be captured if SC / Bulgarian / Russian pattern with French / English / Chinese regarding when they have wh-movement; the only difference between them being that the former have additional wh-fronting that involves focus-movement which obligatorily affects all (non-D-linked) wh-phrases. We then have a perfect correlation between Superiority and wh-movement; whenever a MWF language must have wh-movement, it shows Superiority effects. Furthermore, there is no separate MWF-type regarding wh-movement itself. MWF languages are scattered across the English, French, and Chinese-type regarding when they have wh-movement. There is, however, an issue regarding PL/SP readings. Bošković’s (2003a) account of the SP/PL answers generalization leaves room for not filling SpecCP overtly to be necessary but not sufficient for licensing SP answers. There could then be a MWF language without Superiority effects which still disallows SP answers. Grebenyova (2006) reports this is the case for some Russian speakers. Consider Bošković’s (2003a) account of SP/PL answers, which is based on Hagstrom’s (1998) semantics of questions, where a Q-morpheme, an existential quantifier over choice functions which is necessary for an interrogative interpretation, is merged below CP. For Hagstrom, an SP reading results if the Q-morpheme is merged right below CP, with

260

CONTEMPORARY LINGUISTIC PARAMETERS

both wh-phrases in its scope (both wh-phrases are then in the domain of the choice function; cf. (39a), a representation of English questions). The Q-morpheme can also be merged with one of the wh-phrases, where it does not have both wh-phrases in its scope (39b). In Hagstrom’s system this leads to PL answers.13 Bošković (2003a) shows this system captures the damaging effect of overt wh-movement on SP answers. For an SP answer, the Q-morpheme must be introduced above both wh-phrases. In English, the wh-phrase undergoing overt wh-movement then crosses the Q-morpheme, yielding a relativized-minimality effect (an element with a +wh-feature crosses an element with a +wh-feature; see (39a). The problem doesn’t arise in (39b), a PL-answer configuration, where the Q-morpheme is merged with the lower wh-phrase, hence the wh-phrase undergoing wh-movement does not cross it. (39) a. WHi C Q [ti wh] SP answer b. WHi C [ti wh+Q] PL answer Wh-movement thus induces a relativized-minimality violation on the SP reading.14 The problem doesn’t arise in languages without overt wh-movement. Suppose, however, that due to its lexical properties (i.e. selectional requirements), the Q-morpheme requires merger with a wh-phrase. This would rule out the SP reading even in a language without overt wh-movement, the source of variation being lexical. Another more structurally based possibility for Russian is to take advantage of focus-movement. Suppose that for the Russian speakers who disallow SP readings, the landing site of focus-movement is higher than the higher position in which the Q-morpheme is merged on the SP reading. A wh-phrase then crosses the Q-morpheme when undergoing focusmovement, yielding a relativized minimality effect since we are dealing here with A’-movement crossing an A’-element. The SP reading can then be blocked even in a language like Russian, which does not have overt wh-movement. The upshot of this is that an overt wh-movement language will never allow SP answers, while languages without overt wh-movement may, but don’t have to, allow SP answers. However, SP answers are the only option in SC when a lower wh-phrase moves across a higher wh-phrase as in (31b). The reason for this (see Bošković 2003a) is that if the Q-morpheme is merged with the lower wh-phrase, which is what normally happens with PL answers, the lower wh-phrase carries the Q-morpheme under focalization, so that after this movement the Q-morpheme c-commands both wh-phrases, an SP-answer configuration. (There is cross-linguistic variation regarding the strandability of the Q-morpheme, see Hagstrom 1998, Bošković 2003a). Interestingly, Stjepanović (2010) observes that, in contrast to (31b), a PL answer is allowed in (40) (SP answers are also possible). (40) Kakvu je ko dobio ocjenu? what is who gotten grade ‘Who got what grade?’ Only the left-branch moves here, stranding the rest of the object with the Q-morpheme below the subject, thus leaving the Q-morpheme in a PL-answer configuration.

WH-MOVEMENT

261

There are two contexts where SC null-C matrix questions show Superiority effects, sluicing (Stjepanović 1999b) and constructions involving topicalization constituents (TCs, Bošković 2002). (41) Neko nekoga/nekoga neko ljubi. Ko koga/*koga ko? someone.NOM someone.ACC is-kissing who whom (42) a. Tom čoveku, ko je šta poklonio? that man.DAT who is what bestowed ‘To that man, who bestowed what?’ b. ??Tom čoveku, šta je ko poklonio? Recall that in French and SC, +wh-C in questions like (5a) can be inserted overtly or covertly. Overt C-insertion triggers overt wh-movement. Since Superiority indicates wh-movement, it must then be the case that (41)–(42) involve wh-movement to SpecCP. If sluicing involves deletion of the IP-complement of C, this is not surprising regarding (41). As for TCs, Rudin (1993) argues TCs are CP-adjoined (they precede wh-phrases in Bulgarian, which are located in SpecCP). TCs can then be present only when CP is present overtly. Overt C-insertion forces wh-movement, hence the Superiority effect in (42). Russian doesn’t exhibit Superiority effects even with TCs. This is expected: being a Chinese-type language regarding wh-movement, Russian doesn’t have wh-movement regardless of the timing of C-insertion. (43) a. A etomu čeloveku kto kogo predstavil? and that man.DAT who whom introduced ‘And to that man, who introduced whom?’ b. A etomu čeloveku kogo kto predstavil? Furthermore, SC (42a) can only have a PL answer. This is expected: TCs force overt wh-movement, which then forces a PL answer. Significantly, Stepanov (1998) notes Russian (43a) allows an SP answer, as expected given that Russian questions need not involve overt movement to SpecCP. Consider now the reason for the different behavior of focus/wh-movement regarding Superiority. Bošković (1999) shows this can be captured under the economy account where superiority follows from the requirement that features be checked through the shortest movement possible. In the languages under consideration, all wh-phrases undergo focus-movement but only one undergoes wh-movement (movement motivated by checking the +wh-feature of C). To check the +wh-feature through the shortest movement possible, the highest wh-phrase must undergo wh-movement. Assuming movement to SpecCP triggers Spec-head agreement, checking C’s +wh-feature, the highest wh-phrase must move first; otherwise, the +wh-feature would not be checked in the most economical way. Focus-movement involves multiple movement to the same position since all wh-phrases undergo it. Regardless of the order of movements, the same number of nodes is always crossed, hence no order is preferred by Economy. Bošković (1999) explores two ways of stating the focus requirement: as an inadequacy of wh-phrases (where wh-phrases have a strong focus

262

CONTEMPORARY LINGUISTIC PARAMETERS

feature) or an inadequacy of the target, where the target head has the specification Attract-all for focus, which is satisfied by attracting all focalized elements.15 I will adopt here the latter. To illustrate, (44) involves only wh-movement, with the attractor specified as Attract-1-wh, hence the highest wh-phrase moves to check its wh-feature. (44) a. Whati did Mary give ti to who? b. *Whoi did Mary give what to ti? SC (31) involves pure focus-movement. The attractor is specified as Attract-allFocus, which means both wh-phrases must undergo focus-movement. The focus requirement is checked in the same way in terms of nodes crossed regardless of the order of movement of the wh-phrases. In Bulgarian (30), one wh-phrase undergoes wh-movement. Moreover, both wh-phrases are licensed for focus by the interrogative C. For the focus requirement, the order of movements is irrelevant. However, to check the +wh-feature of C in the most economical way, the highest wh-phrase must move first. (Following Rudin 1988, the first wh-phrase in the linear order is the one that moves first). Since, in contrast to wh-movement, focus-movement is not subject to Superiority, if there are three wh-phrases in Bulgarian questions, the order of the second and the third wh-phrase is expected to be free. This is indeed the case. Thus, kogo must move before kakvo when it is the highest wh-phrase before wh-fronting (45), but not when it is not, as in (46), where the highest wh-phrase is koj.16 (45) a. Kogo kakvo e pital Ivan? whom what is asked Ivan ‘Who did Ivan ask what?’ b. ?*Kakvo kogo e pital Ivan? (46) a. Koj kogo kakvo e pital? who whom what is asked ‘Who asked who what?’ b. Koj kakvo kogo e pital? Bošković (2002) took the focus requirement to be the defining property of MWF. While an Attract-all-F head is needed to get MWF (this is the way to force all wh-phrases to move), does the relevant feature have to be the focus feature? Could it be the wh-feature? What would be the difference between an Attract-all-Focus and an Attract-all-wh language? Consider first SC Attract-all-Focus constructions. (47) [FocP [VP non-D-linked-wh… non-D-linked-wh]] Since the wh-phrases are focused, they must move. Since there is no wh-movement, SP readings are available. Moreover, since Attract-all-F heads don’t induce Superiority effects, there are no Superiority effects here. Consider now an Attract-all-wh MWF language, call it Y. Since Y has wh-movement, SP readings are disallowed. However, since the wh-movement-inducing head is an Attract-all head, there should be no Superiority effects. So, Y should look like this: no Superiority effects and no SP readings. There is a caveat though: since SP answers

WH-MOVEMENT

263

can be blocked even in languages without overt wh-movement, we are not ruling out the possibility of Y being an Attract-all-Focus language, like Russian (so far Y actually looks exactly like the variety of Russian that disallows SP answers). There is, however, another difference between Attract-all-Focus languages and Y. Recall D-linked wh-phrases do not undergo focus-movement. However, even D-linked wh-phrases should move in Y since the Attract-all-wh head should not care about whether wh-phrases are D-linked or not. This is then the difference between Attract-all-wh and Attract-all-Focus MWF languages. Consider Hungarian (48) (see Bošković 2007b and Kiss 2002). (48) a. *Ki irt mit? *who wrote what b. Ki mit irt? c. Mit ki irt? d. *Ki irta melyik levelet? *who wrote which letter e. Ki melyik levelet irta? f. Melyik levelet ki irta? Both D-linked and non-D-linked wh-phrases must move here,17 i.e. Hungarian doesn’t show the D-linked/non-D-linked distinction regarding MWF Slavic languages exhibit. This can be captured if Hungarian is an Attract-all-wh language, rather than an Attract-all-Focus language (see Bošković 2007b; Surányi 2005; see also Diesing 2003 for Yiddish). Hungarian MWF questions also disallow SP answers (see e.g. Surányi 2005) and do not show Superiority effects (48b–c), which is exactly the behavior expected of an Attract-all-wh language. However, the standard analysis of Hungarian (e.g. Kiss 2002; Horvath 1998; Puskás 2000; Lipták 2001) holds that the wh-phrase that is closest to the verb in MWF questions undergoes focus-movement, other wh-phrases undergoing movement that non-wh-quantifiers undergo (but see Surányi 2005 for arguments against this analysis). A question then arises whether Hungarian constructions under consideration can be reanalyzed as involving wh-movement.18, 19 Note also that some languages, e.g. Iraqi Arabic and Hindi, don’t allow any wh-phrases to remain in-situ within an embedded finite clause (ignoring dummyscope-marker constructions). Both (11a) and (12) are unacceptable in these languages; no matter how many wh-phrases are located in the same clause as a +wh-C, if one wh-phrase is separated from it by a finite clause boundary the sentence is ungrammatical. This can be interpreted as indicating that in wh-in-situ constructions of such languages C Attracts all wh-phrases in LF, the same LF-strategy as in French being employed, hence the clause-boundedness effect. Finally, I note the rather interesting case of ASL, where in-situ wh-phrases appear to behave differently in single and multiple questions. (49) a. WHO JOHN SEE? ‘Who did John see?’ b. ?JOHN GIVE-UP WHAT HAPPY? ‘What did John happily give up?’ c. *WHO GIVE-UP WHAT HAPPY?

264



CONTEMPORARY LINGUISTIC PARAMETERS

d. WHO GIVE-UP HAPPY WHAT? e. *WHO BUY LAST NIGHT WHICH COMPUTER? f. WHO BUY WHICH COMPUTER LAST NIGHT?    (Wood 2009)

Wood (2009) analyzes ASL as a new type which to some extent combines Bulgarian and SC. At most one wh-phrase undergoes true wh-movement in ASL either overtly (49a) or covertly (49b) (C can be inserted overtly or covertly, see section 11.2.1), other non-D-linked wh-phrases must undergo rightward focus-movement overtly. (49c) is unacceptable since WHAT didn’t undergo focus-movement. The (non-) D-linking contrast between (49d) and (49e) provides evidence that rightward movement involves focus-movement. (49b) is acceptable since, in contrast to (49c), WHAT in (49b) can undergo wh-movement (covertly), hence it is not subject to obligatory focus-movement (see Wood 2009 for discussion of Superiority effects and the focus/wh-movement interaction in ASL).

11.2.4. Back to wh-in-situ MWF languages are also informative regarding the analysis of wh-in-situ. We have seen that various whs-in-situ shouldn’t be analyzed the same way; thus, unmoved wh-phrases in French, BP, Spanish, English, and Japanese exhibit very different behavior (varying even within a single language, compare (11) and (12)). The Minimalist Program (MP) actually has an abundance of proposals regarding how wh-phrases-in-situ can be treated, which may be useful given their varried behavior. MP also has several ways of treating the GB LF wh-movement: a. keeping phrasal covert movement; b. replacing it with Move-F; c. replacing it with Agree; d. replacing it with unselective binding. As discussed above, the freest option, which is however available only for wh-arguments, is unselective binding. Move-F has the locality of head-movement. As for Agree, which involves feature-checking at a distance, there is some controversy regarding its locality. It is subject to relativized minimality/ intervention effects (like phrasal movement and Move-F): X can agree for feature F only with the closest F-bearing element. The issue is, however, whether it is also subject to the PIC and the Activation Condition. While the standard assumption is that it is, Bošković (2007a) argues that it isn’t. Notice, however, that if it is, unmoved wh-phrases cannot be treated in terms of Agree (the PIC would impose such heavy locality restrictions that wh-in-situ could never be found in long-distance questions. Furthermore, we would expect to find languages where subject wh-in-situ (in SpecTP) could be licensed but object wh-in-situ (within VP) could not be. Even phrasal covert movement has several options in MP. In GB, there was a timing difference: covert movement occurs after SS hence follows overt movement. While such treatment can still be maintained, two additional treatments of covert movement arose as a result of elimination of the traditional LF component in MP, as in e.g. the multiple spell-out model. Covert and overt movement then cannot be distinguished by sequential ordering in the derivation: since there is only one cycle, they have to occur on the same cycle. One way to distinguish covert and overt movement is in terms of which copy is pronounced: the highest copy is pronounced in the latter and the lowest in the former (Bobaljik 1995; Groat and O’Neil 1996).

WH-MOVEMENT

265

Another way is to assume that they differ regarding the timing of the transfer to the PF interface (Nissenbaum 2000): they take place on the same cycle, but the phrase hosting the movement can be transferred to the PF interface either after (overt movement) or before (covert movement) the relevant movement occurs. MWF languages provide a clear case of lower-copy pronunciation (LCP), which can be used to test other wh-phrases in situ regarding this option. As illustrated below for Romanian, the second wh-phrase does not move in MWF languages if it is homophonous with the first fronted wh-phrase. (50) a. Cine ce precede? who what precedes b. *Cine precede ce? (51) a. Ce precede ce? what precedes what b. *Ce ce precede?

(Bošković 2002)

Bošković (2002) rules out (51b) by a PF constraint against consecutive homophonous wh-phrases (similar constraints on homophonous elements exist in many languages, see e.g. Golston 1995; Bošković 2001). (51a) involves an intricate phonology/syntax interaction where the need to satisfy a PF requirement apparently overrides the need to satisfy a syntactic requirement, which cannot be implemented in a derivational framework like MP. Bošković (2002), however, shows LCP straightforwardly resolves this phonology-syntax conflict, given Franks’ (1998) proposal that lower-copy pronunciation is possible if higher-copy pronunciation would induce a PF violation (see Nunes 2004 for a deduction of this). Since Romanian has a syntactic requirement that forces all wh-phrases to move overtly, the second wh-phrase must move in the syntax (irrelevant copies are ignored). If, as usual, the higher cei is pronounced, a PF violation obtains (due to a sequence of homophonous wh-elements). This is precisely the situation where a lower copy can be pronounced under Franks’ approach. (52) [ce cei precede cei] LCP thus enables us to derive (51a) without violating the syntactic requirement that forces all wh-phrases to move overtly in Romanian, without look-ahead from the syntax to the phonology, without having phonology override syntax, and without any PF movement. There is evidence that the second ce in (51a) indeed moves in overt syntax. In-situ wh-phrases in languages like English, “true” wh-in-situ languages like Malay, or mixed languages like French differ from their moved counterparts in being unable to license parasitic gaps.20 (53) a. What did John read without filing? b. *Who read what without filing? (54) a. *Kamu aturkan buku yang mana tanpa baca? *you filed book that which without reading b. Buku yang mana kamu aturkan tanpa baca? ‘Which book did you file without reading?’

[Malay]

266

CONTEMPORARY LINGUISTIC PARAMETERS

(55) a. *Il a lu quoi sans classer? *he has read what without to-file b. cf. Qu’a-t-il lu sans classer?

[French] (Bošković 2002)

However, the in-situ wh-phrase in (51a) licenses parasitic gaps (56), as expected under the LCP analysis, where the wh-in-situ in (56) undergoes phrasal movement in overt syntax that doesn’t differ syntactically in any respect from wh-movement of what in (53a). It’s then not surprising that (56) patterns with (53a), not (53b). (56) Ce precede ce fără să influenţeze? what precedes what without subj.particle influence.3p.sg ‘What precedes what without influencing?’ This Romanian wh-in-situ differs from in-situ wh-phrases in non-MWF languages like English and wh-in-situ languages. The latter two then should not be analyzed in terms of LCP given the contrast between (51a) / (56) and (53b) / (54a) / (55a). (More generally, whatever analysis of (51a) is adopted it shouldn’t be applied to the latter, though see Nissenbaum 2000.) Another line of work which was originally developed for wh-in-situ languages should be noted. Hagstrom (1998) argues Japanese questions involve overt movement. However, he argues that the traditional interrogative complementizer -ka is a Q-marker that is generated sentence-internally, and then moves overtly to C. He argues Sinhala has a similar Q-marker, which stays in its base position overtly and undergoes the same movement -ka does covertly. Bošković (2003a) extends this analysis to English, arguing English has a null counterpart of the Q-marker that is generated in the same position as in Japanese / Sinhala and then undergoes movement. As discussed above, the analysis accounts for the distribution of PL/SP answers. Under this view, the Q-syntax / semantics is invariant cross-linguistically, the only difference being whether the Q-morpheme is overt and whether it undergoes overt or covert movement. Wh-movement in languages like English is a result of a separate formal requirement which doesn’t feed semantics, the lower copy of the wh-phrase being interpreted in English. Cable (2010) modifies and extends this analysis more broadly (see also Kotek 2014 for a modification of Cable 2010), arguing that wh-dependencies are quite generally not established between Cwh and the wh-phrase but between Cwh and the Q-particle based on Tlinglit examples like (57), where he argues Q takes the wh-phrase as its complement. QP then undergoes Agree with C and moves to SpecCP. English has a null counterpart of Tlinglit’s Q-morpheme. As for Japanese, where the Q-morpheme is detached overtly from the wh-phrase, he argues that the Q-morpheme is generated adjoined to the wh-phrase, hence can move to C on its own, stranding the wh-phrase. (57) Daa sá i éesh al’óon? what Q your father he.hunts.it

WH-MOVEMENT

267

11.3. Wh-islands Rizzi (1982) observes variation regarding wh-islands: while English exhibits wh-island effects, Italian doesn’t. Interestingly, even languages allowing extraction from wh-islands don’t behave uniformly regarding the phenomenon. Rizzi notes wh-island effects are not fully voided in Italian: Italian disallows extraction from more than one wh-island and extraction from a finite clause embedded under a wh-island. Rudin (1988) shows that Bulgarian is much more permissive than Italian; thus, it allows extraction from more than one wh-island. Rudin also observes SC differs from Bulgarian in that it is wh-island sensitive. (58) Vidjah edna kniga, kojatoi se čudja koj znae koj prodava ti saw-1s one book which-the refl wonder-1s who knows who sells ‘I saw a book which I wonder who knows who sells’  [Bulgarian] (59) *Vidio  sam knjigu kojui se pitam ko prodaje ti ‘I saw a book which I wonder who sells’ 

[Serbo-Croatian]

Rudin interprets this as providing evidence that, in contrast to SC, Bulgarian allows more than one wh-phrase in SpecCP in overt syntax. Kojato in (58) then escapes the Wh-Island Constraint by moving through the embedded SpecCPs, occupied by koj. Since SC disallows more than one wh-phrase in SpecCP overtly, this is not possible in SC. Rudin thus ties the insensitivity of Bulgarian to wh-islands to the availability of a particular type of MWF, where all wh-phrases are located in SpecCP. However, although SC disallows multiply filled SpecCPs in matrix questions, it allows them in embedded questions. Recall that, in contrast to matrix questions, SC shows Superiority effects with embedded questions (37), which indicates that in that context SC has true wh-movement. Furthermore, there is evidence that when SC has wh-movement, it switches completely to the Bulgarian paradigm, with all fronted wh-phrases located in SpecCP.21 While, in contrast to Bulgarian, SC allows non-whmaterial to intervene between fronted wh-phrases in matrix questions, it disallows it in embedded questions, which indicates SC places all fronted wh-phrases in SpecCP in embedded questions. (60) *Pričali su o tome ko, po tebi, šta mrzi. *talked are about that who according to-you what hates However, SC still always exhibits wh-islands effects. Furthermore, Slovenian, a MWF language which disallows multiply filled SpecCP MWF, doesn’t exhibit wh-island effects (see Golden 1997). This also indicates that the grammaticality of (58) should not be tied to the availability of a particular type of MWF. Bulgarian is actually not truly wh-island insensitive. In contrast to relativization, Bulgarian exhibits wh-island effects in questions, as Rudin’s (61) shows. Rudin also observes that (62), containing a D-linked wh-phrase, contrasts with (61). Furthermore, Bošković (2008a) shows extraction of adjuncts from wh-islands is always unacceptable (63).

268

CONTEMPORARY LINGUISTIC PARAMETERS

(61) *Kakvoi se čudiš koj znae koj prodava ti? ‘What do you wonder who knows who sells?’ (62) ?Koja ot tezi knigii se čudiš koj znae koj prodava ti? ‘Which of these books do you wonder who knows who sells?’ (63) **pričinata, poradi kojatoi znae [dali Boris e zaminal ti] ‘the reason for which he knows whether Boris left’ This indicates wh-islands are islands in Bulgarian, hence the wh-island effect should not be completely voided, as in Rudin’s analysis. Furthermore, Bošković (2003b) notes Swedish, which disallows MWF, behaves like Bulgarian: Argument extraction from (in fact multiple) wh-islands is allowed with relativization and D-linked questions, but not with non-D-linked questions. It is also never possible with adjuncts. (64) a. *Vad får gade Jan vem som skrev? [Swedish] ‘What did John ask who wrote?’ b. Det är melodin, som Jan frågade vem som skrev. ‘This is the song that John asked who wrote’ (Maling 1978) c. Vilken film var det du gärna ville veta vem som hade regisserat? ‘Which film did you want to know who had directed?’ (Engdahl 1986) d. **orsaken varför han undrar [vem som lagade bilen t]? ‘the reason why he wonders who fixed the car’ (Bošković 2003b) That Bulgarian, a MWF language, and Swedish, a non-MWF language, exhibit the same behavior regarding wh-islands (referred to below as selective wh-island insensitivity) indicates that the possibility of extraction from wh-islands in certain contexts in Bulgarian shouldn’t be tied to the availability of MWF.22 There is a broader pattern here. Bošković (2008a) shows selective wh-island insensitivity is displayed by Romanian, a Bulgarian-type MWF language (Rudin 1988), as well as Icelandic, Norwegian, Hebrew, and Albanian, all of which are non-MWF languages. What Bulgarian, Swedish, Romanian, Norwegian, Icelandic, Hebrew, and Albanian have in common that differentiates them from languages like SC and English, which do not display selective wh-island insensitivity, is affixal articles. (65) Selective wh-island insensitivity is a property of languages with affixal articles. (Bošković 2008a) (65) confirms that selective wh-island insensitivity should not be tied to MWF. Bošković (2008a) gives a deduction of (65), the gist of which is the following. Phasal heads quite generally allow multiple Specifiers. This means a wh-phrase in SpecCP does not by itself prevent another wh-phrase from passing through SpecCP. However, Bošković (2008a) shows operator-variable creating movements like wh-movement, topicalization, focalization, NPI-movement, and QR cannot feed each other. To account for this, he argues all these movements are driven by an uninterpretable

WH-MOVEMENT

269

operator-feature of the moving element (in Chomsky 2001, this feature activates the moving element for these movements); once X undergoes one such movement, the feature is checked off, hence X cannot undergo another operator-variable creating movement. Focusing on wh-phrases, Bošković argues that due to a feature-sharing property of D in languages with affixal articles (see Lasnik 1995a on feature-sharing between affixes and their hosts), the operator-feature of the wh-phrase is shared by D (this means the wh-phrase is not located in D with relative and D-linking wh-phrases, which is clear in Albanian, where overt article co-occurs with relative and D-linked wh-phrases but crucially not with non-D-linked wh-phrases). As a result, even when the Op-feature of the wh-phrase is deactivated, the Op-feature of D still makes the wh-DP active for further wh-movement (see Bošković 2008a for details, including the reason why the Op-feature of D enables movement from multiple wh-islands [recall that in Italian, a non-affixal D languages, movement from multiple wh-islands is disallowed] and why this strategy is not available with adjuncts). Under the most natural interpretation of this analysis, multiple CPSpecs are expected to be in principle available in all languages, not just Bulgarian-style MWF languages. This enables us to account for wh-island permissiveness of Swedish, Norwegian, Hebrew, Icelandic, and Albanian. Moreover, it is actually hard to prevent the multiple-Spec option in the bare-phrase-structure system. We would then expect it to be available everywhere as long as independent factors do not interfere with it. It is standardly assumed that vP allows multiple Specs. Since subjects are generated in SpecvP, object shift (which also lands in SpecvP), or any movement out of vP (including A’-movement, which must pass through SpecvP given the PIC), requires multiple vPSpecs. The multiple SpecvP option should then be universally allowed.23 Regarding CP, Bošković (2007a) shows that multiple CPSpecs are available even in English by examining constructions where more than one phrase moves out of the same declarative CP, each of which must pass through SpecCP, given the PIC. Wh-islands confirm the multiple-Spec option should always be in principle available. In principle, Swedish, Bulgarian, English, and SC all allow wh-movement through filled SpecCP, (i.e. creation of multiple CPSpecs). Bulgarian and Swedish can use this to void the wh-island effect in certain contexts. Where wh-island effects emerge, movement from a +wh-SpecCP is blocked because the feature that would make the movement possible is checked off. This is always the case with English and SC. The analysis divorces the wh-island effect from the availability of MWF with multiply filled SpecCPs, which is desirable since such association would incorrectly void wh-islands in Bulgarian and SC, but not English and Swedish. However, we cannot allow multiple wh-phrases to surface in the same CP in English and Swedish. Since this option should in principle be available (we shouldn’t simply assume the C disallows multiple Specs), it must then be blocked for independent reasons. For relevant discussion, the reader is referred to Pesetsky (2000) and Bošković (2007a); both works suggest that there is nothing wrong with true MWF in English syntactically; however, the presence of more than one wh-phrase in SpecCP in a final representation in English leads to a PF violation in Pesetsky (2000), and a semantic violation in Bošković (2007a) (see footnote 29).24

270

CONTEMPORARY LINGUISTIC PARAMETERS

11.4. Successive-cyclic movement A question that is related to the issue of wh-islands, which also has important theoretical consequences, is whether intermediate steps of wh-movement, like the one in (66), involve feature-checking / Agree / Spec-Head agreement (SHA) with C. There is evidence even from English for this intermediate step of movement. Thus, McCloskey (2000) shows quantifiers can be floated in this position in West Ulster English. (66) Whati do you think [CP t’i [C’ that Mary bought ti]]? (67) Whati did he say [CP [all ti]j that he wanted tj]? Chomsky (2000) assumes t’ and that undergo agreement in (66a). He develops a probe-goal feature-checking system where the goal must have an uninterpretable feature to be visible for movement (the Activation Condition). Consider (68) (the exact feature-labels are not important): C and what undergo wh-feature checking; the uninterpretable Q-feature, which made the wh-phrase visible for the primary feature-checking relation and movement to SpecCP, is checked off as a reflex of the primary feature-checking relation/movement to C. The wh-phrase is then no longer active for further movement. (68) I wonder whati C iwh uwh uQ EPP

Mary bought ti.

To accommodate successive-cyclic movement, Chomsky adopts the concept of defective heads, which are unable to check off the feature of the goal that has made the goal visible for agreement and movement to the head in question. That in (66) is a defective head; it cannot check off the Q-feature of what, so that what is still active for movement and agreement. (69) You think [CP whati that Mary bought ti] iwh uwh uQ EPP Bošković (2007a) argues that intermediate steps of successive-cyclic movement are not driven by feature checking. Rather, they take place so that the wh-phrase, which has an uninterpretable feature (uK), avoids being sent to spell-out (assuming that CP is a phase and that the IP-complement of C is sent to spell-out after the CP is built). Under this non-feature checking approach to successive-cyclic movement, we need not assume that some heads are defective in that they are unable to check off the uK of their goal since such heads do not undergo feature checking in the first place. All probes can be assumed to delete the uK of the goal that has made the goal active for entering into a relation with the probe, the concept of defective heads being eliminable. Putting aside theoretical issues, Bošković (2007a) argues that and t in (66) do not undergo SHA based on the Lobeck (1990)/Saito and Murasugi (1990)

WH-MOVEMENT

271

generalization that functional heads can license ellipsis of their complement only when they undergo Spec-Head agreement (SHA), i.e. feature-checking. Thus, (70) shows that +wh-C, which undergoes SHA, licenses ellipsis, whereas the non-agreeing complementizer that does not. (70) a. John met someone but I don’t know [CP whoi [C’ C John met ti]]. b. *John believes C/that Peter met someone but I don’t think [CP [C’ C/that Peter met someone]]. Intermediate C cannot license ellipsis, which can be captured if C/that and t do not undergo SHA. (71) *John met someone but I don’t know whoi Peter said [CP ti [C’ C/that John met ti]]. (71) then provides evidence against the feature-checking view of successive-cyclic movement. What about languages that are assumed to have overt reflexes of agreement with intermediate heads under wh-movement? As Boeckx (2003) notes, it is not clear that there are languages with true intermediate wh-agreement, i.e. overt reflexes of agreement between intermediate heads and wh-phrases. In many languages that are traditionally considered to have such agreement, wh-agreement is only indirect (see Boeckx 2003, Chung and Georgopoulos 1988, Georgopoulos 1991, Chung 1998). Wh-movement triggers a morphological change on intermediate verbs and/or intermediate complementizers. However, the change does not reflect any direct relation between a wh-phrase and the verbs / complementizers but a distinct agreement relation holding between the verbs and the intermediate complementizers. Instead of a wh-phrase directly agreeing with an intermediate head, wh-movement affects agreement between intermediate verbs and intermediate complementizers. Consider (72)–(74) from Selayarese, a VOS language. (72) La-alle-i doeʔ-iñjo i Basoʔ 3-take-3 money-the HUM Baso ‘Baso took the money’ (73) Ku-isseʔ-*(i) kuko la-ʔalle-i doeʔ-iñjo i Basoʔ 1s-know-3 C 3-take-3 money-the HUM Baso ‘I know that Baso took the money’ (74) Apa mu-isseʔ la-ʔalle_ i Basoʔ what 2FAM-know 3-take HUM Baso ‘What do you know that Baso took?’

(Finer 1997)

In (72), the verb displays subject (prefix) and object (suffix) agreement. (73) illustrates object agreement between the verb and a clausal object. (74) shows object agreement (and overt C) must be absent with wh-movement. The agreement with the wh-phrase is thus only indirect: wh-movement disrupts agreement between the

272

CONTEMPORARY LINGUISTIC PARAMETERS

verb and the C.25 Building on Finer (1997), Bošković (2008b) analyzes (72)–(74) as follows: Overt object agreement in Selayarese is a reflex of object shift (movement to SpecAgroP). This means the agreeing direct-object NP in (72) and the agreeing direct-object clause in (73) move overtly to SpecAgroP. Takahashi (1994) shows wh-movement out of moved elements is impossible. The wh-phrase then cannot move out of an object-shifted clause. This means wh-movement out of an agreeing clausal object is impossible, which explains why a clause from which a wh-phrase has been extracted cannot agree with the verb. Given the natural assumption that the lack of overt object agreement means the lack of overt movement to SpecAgroP, the reason why wh-movement is possible only when the clause fails to agree is straightforward: only in that case, the clause does not move to SpecAgroP, allowing wh-movement.26 The most plausible candidate for true intermediate wh-agreement seems to be Kinande, where the featural specification of the C covaries with the featural specification of the wh-phrase (numbers represent morphological classes). (75) IyondI y0/ABahI Bo Kambale alangIra who1 that1who2 that2 Kambale saw ‘Who did Kambale see?’ (Rizzi 1990) This agreement occurs with displaced wh/focus-phrases and can be found in every clause on the path of wh-/focus-movement (all the Kinande data are from SchneiderZioga 2005).27 (76) [ekihi kyo Kambale a.si [nga.kyo Yosefu a.kalengekanaya what wh.AGR Kambale AGR.know C.wh-AGR Joseph AGR.thinks [nga.kyo Mary’ a.kahuka __ ]]] C.wh-AGR Mary AGR.cooks ‘What did Kambale know that Joseph thinks that Mary is cooking (for dinner)? Significantly, reconstruction effects indicate Kinande does not have true long-distance wh/focus-movement. (77) a. ekitabu kiwej/k ky’ obul mukoloj a.kasoma _ kangikangi book his wh-AGR each student AGR.reads regularly ‘(It is) Hisj book that [every studentj/k] reads regularly’ b. ekitabu kiwek/*j kyo ngalengekanaya [CP nga.kyo [obuli mukolo]j book his wh-AGR I.think C.wh-AGR every student akasoma _ kangikangi read regularly ‘(It is) Hisk/*j book that I think [every student]j reads regularly.’ c. ekitabu kiwek/*j kyo [obuli mukolo]j alengekanaya [CP nga.kyo book his wh-AGR every student AGR.think C.wh-AGR nganasoma _  kangikangi] I.read regularly ‘(It is) Hisk/*j book that [every student]j thinks I read regularly’

WH-MOVEMENT

273

(77a) shows local A’-extraction allows reconstructed interpretation. However, reconstruction is impossible with long-distance dependencies. Under the standard view which ties reconstruction to movement, this means the focused element undergoes movement from its θ-position to SpecCP in (77a), but not (77b–c). Consider also (78)–(79). (78) is unacceptable due to extraction from an adjunct. Schneider-Zioga (2005) observes that if the extraction site is embedded within a clause headed by an agreeing complementizer, (78) improves (see 79). (78) *omukali ndi yo wasiga [island embere __ wabuga] *woman who wh-AGR you.left before spoke ‘Which woman did you leave before (she) spoke?’ (79) omukali ndi yo

wasiga [island embere Kambale anasi [CP ko.yo _ wabuga]] woman who wh-AGR you.left before Kambale knew C.wh-AGR spoke ‘Which woman did you leave before Kambale knew that (she) spoke?’

The obvious conclusion is that, in contrast to (78), the wh-phrase in (79) does not undergo wh-movement. These data show that there is no wh/focus-movement from clauses headed by agreeing Cs in Kinande: the standard wh-agreement analysis, where a wh-phrase moves to and agrees with an intermediate C, and then moves to and agrees with another C, cannot be correct. While Kinande has local wh/focusmovement, as indicated by (77a), it does not have long-distance wh-movement out of agreeing CPs. The most straightforward analysis here (see Bošković 2008b) is that the focused NP is base-generated in SpecCP, with null Operators, which are co-indexed with it, inserted in intermediate SpecCPs, agreeing with the intermediate Cs. Local A’-movement is possible only from the true θ-position. (77) then has a structure like (80), where only the lowest Op undergoes movement (see also McCloskey 2002 for Op-insertion in “intermediate” SpecCPs). (80) [CP Opi [CP Opi [CP Opi ti ]]] Kinande agreeing long-distance A’-movement examples thus do not involve a wh/ focus-phrase moving clause-to-clause, with a single phrase undergoing agreement with more than one C. Kinande shows it is not the case that intermediate Cs cannot undergo agreement; they can agree. What is disallowed is that after agreeing with an intermediate C, a wh-phrase moves and agrees with another C. Those intermediate Cs in Kinande are actually final Cs, since the phrase in their Spec doesn’t move to another SpecCP. Apparently, once a wh-phrase moves to SpecCP undergoing agreement with the C (even if the C is -wh) it is frozen in this position. The conclusion here is that there are no defective heads regarding feature checking (in the sense that they undergo feature checking but fail to deactivate the goal). All probes delete the uK of the goal that has made the goal active for entering into a relation with the probe (putting aside the above exception regarding wh-islands. This also means there is no feature checking with intermediate heads during successive-cyclic movement.) If there are no defective heads, we can explain the freezing effect that agreement with a C has on wh-phrases in Kinande. If a wh-phrase moves to SpecCP and

274

CONTEMPORARY LINGUISTIC PARAMETERS

undergoes agreement even with an intermediate C like that, its uK feature that makes it active for wh-movement is erased, freezing it in SpecCP. A wh-phrase can then never move from the Spec of a C like that if it undergoes agreement with it. (In English, it doesn’t agree with it, and in Kinande it does). There are then two options to get legitimate long-distance wh-structures: (a) reanalyzing long-distance wh-dependencies as a series of local wh-dependencies, where the declarative C agrees with the element in its Spec, freezing it in place; (b) a wh-phrase moves to the Spec of that without undergoing agreement with that, the movement being driven by the factors discussed above. Kinande takes option (a) in the above examples and English takes option (b). A combination of these options may account for Russian, which disallows long-distance wh-dependencies. Apparently, declarative C in the relevant Kinande examples agrees with the element in its Spec. Suppose this also holds for Russian declarative C. However, suppose that, as in English, the option of reanalyzing long-distance wh-dependencies as a series of local wh-dependencies is unavailable in Russian. The result is that Russian disallows long-distance wh-dependencies.

11.5. The basic wh-movement typology reconsidered Let us see how the basic wh-typology can be implemented formally (the Agree option will be considered for wh-phrases-in-situ in this section). In Chomsky’s (2000, 2001) system, where movement is driven by a property of the target, we can assume that the target head is specified with an Attract-all feature in MWF languages. (I put aside the issue of what the target and the feature are.) English C has an Atract-1-F property. Wh-phrases optionally have the uninterpretable feature that makes them visible for wh-movement (uK). The wh-phrase has uK in Who did she say he gave the book to (or it could not undergo wh-movement), but not in What did she say he gave to who (uK could not be checked due to the PIC, which for Chomsky constrains Agree). As for wh-in-situ languages, the situation is more complicated. Since for Chomsky Agree is constrained by the PIC, wh-dependencies cannot be licensed through Agree in such languages (otherwise, a wh-phrase in-situ could not be dominated by v’ or –wh-C’ that doesn’t dominate +wh-C). They then have to be handled in one of the ways discussed above. Consider then how the multiple-question typology would be stated in Bošković’s (2007a) system, which eliminates the EPP/strength property of the heads which are targeted by movement and where all movement is driven by an uninterpretable feature of the moving element, under the assumption that uninterpretable features must serve as probes, i.e. they can only be checked if they c-command the checker. (X with a uK then moves to probe its checker, since this is the only way to check a uK.) Another property of that system is that the PIC and the Activation Condition are used only to implement successive-cyclic movement, which means they do not hold for Agree. In this system, in a MWF language like Bulgarian, wh-phrases are obligatorily specified with a uF feature (its precise indentity is not important for our purposes). Hence, they all undergo A’-movement. The corresponding F-feature

WH-MOVEMENT

275

of the target (which may not be the same in all languages, I will simply assume it is C below) can be assumed to be uninterpretable in all languages.28 In wh-in-situ languages like Korean, the F-feature of wh-phrases is interpretable, which means wh-phrases remain in-situ, undergoing Agree with the relevant head. (This is possible since in this system the Activation Condition and the PIC do not hold for Agree). Finally, English wh-phrases optionally have uF, with the further assumption that English +wh-C doesn’t allow more than one Specifier.29 Exactly one wh-phrase then always moves to SpecCP in English. Consider (81)–(82). (81) I wonder what John bought (82) *I wonder John bought what If what has uF, it must move to SpecCP to check off the feature by probing the interrogative C. This derivation yields (81). If what doesn’t have uF it remains in-situ, yielding (82). However, this derivation is ruled out because the uF of the interrogative C cannot be checked. The feature is checked against what in (81). However, this is not an option in (82) because what doesn’t have uF. The analysis bears on the controvery regarding whether who in (83) moves to SpecCP. (83) Who left? If who has uF, it moves to SpecCP to probe the C. If who doesn’t have uF, it will not move to SpecCP. However, since the uF of C remains unchecked, this derivation crashes. This means who in (83) must move to SpecCP, as argued by An (2007), Boeckx (2003), and Pesetsky and Torrego (2001).30 The pattern of cross-linguistic variation is summarized below. It is worth emphasizing that the locus of the cross-linguistic variation in question resides in the lexical properties of wh-phrases (though see footnote 29 for one exception. Note that if wh-phrases optionally bear iF, such languages would still be wh-in-situ languages.)31 (84) target

Bulgarian wh-phrases Korean wh-phrases English wh-phrases uninterpretable F uninterpretable F interpretable F (uninterpretable F)

Another proposal is made in Richards (2010), who tries to deduce the wh-fronting typology from the requirement that a wh-phrase and the +wh-C be separated by as few prosodic boundaries (in particular, Minor Phrase boundaries) as possible. In wh-in-situ languages, this can be achieved by manipulating the prosodic structure hence there is no need for wh-movement to satisfy the requirement in question. In wh-movement languages, due to the head-initial nature of C and the placement of prosodic boundaries to the left of wh-phrases, the requirement cannot be satisfied through prosodic manipulation; the wh-phrase then must move close to C to satisfy it. Richards thus treats wh-typology in terms of syntax-phonology interaction, the relevant cross-linguistic variation being reduced to variation in prosody (on which side of the wh-phrase Minor Phrase boundaries are inserted) and whether CP is

276

CONTEMPORARY LINGUISTIC PARAMETERS

head-initial or head-final (the analysis may predict that C-final languages will not involve wh-movement; note that Richards leaves open the case of MWF languages).

11.6. Final remarks The above discussion only scratches the surface of cross-linguistic variation in the domain of wh-questions / movement. It should, however, be clear from the above discussion that much of the rather complex patterns of cross-linguistic variation in this domain can be rather straightforwardly attributed to the lexical properties of elements involved in wh-questions / movement (especially when it comes to its syntactic aspects). This in itself may warrant pursuing the research strategy where all relevant variation is lexically based. Finally, I note that there are areas of variation pertaining to wh-questions / movement that cannot be discussed here in any detail due to space limitations. E.g. languages differ regarding whether they allow partial wh-movement; furthermore languages that allow it do not behave uniformly (see Lutz and Müller 1996), i.e. partial wh-movement exhibits cross-linguistic variation regarding several properties (e.g. what kind of predicates allow it, the form of the higher phrase (typically what, but Russian uses how, see Stepanov 2000) and scope-markers in intermediate clauses).32 Another interesting area of variation concerns what appear to be coordinated wh-phrases in examples like (85). There is considerable cross-linguistic variation in this respect, Slavic languages being particularly promiscuous in that they even allow coordinations like SC (86) (see Gračanin-Yuksek 2007; Citko and Gračanin 2013; Tomaszewicz 2011; Zanon 2014). (85) When and where did you see them? (86) Dali i gdje si ih vidio? whether and where are them seen

Notes *

For helpful comments and suggestions I thank the editors and anonymous reviewers.

1

There are languages like Italian which for independent reasons disallow multiple questions, see e.g. Calabrese 1992 and Stoyanova 2008 (see also Grohmann 2006 for a potential extension to German).

2

For French, see also Boeckx 2000; Cheng and Rooryck 2000; Munaro, Poletto, and Pollock 2001; Zubizarreta 2003; Mathieu 2004, among others.

3

See Bošković (2003c) for a reanalysis of the contrast in (5)/ (11a) where qui undergoes Agree within Bošković’s (2007a) system where the PIC doesn’t hold for Agree and successive-cyclic movement is driven by the need for the moving element to avoid being sent to spell-out (see section 11.5). Assuming only feature labels, not the exact values (+/-), matter for intervention effects, Bošković argues (11a) is ruled out by Agree Closest, because que, which has a -wh-feature, intervenes in the Agree relation (for the

WH-MOVEMENT

277

wh-feature) between +wh-C and qui. The movement of qui to the embedded SpecCP in (11b) takes place for the reason noted above, voiding the intervention effect. 4

For Reinhart, wh-NPs are interpreted in situ via choice functions.

5

Pesetsky (2000) also argues, on different grounds, that languages may differ regarding whether LF wh-movement involves feature or phrasal movement (Pesetsky suggests a Move-F analysis for Japanese, and a phrasal-movement analysis for Chinese).

6

Curiosly, BP allows adjunct wh-in-situ even within islands, though, as in other languages, wh-adjuncts cannot overtly move out of islands (Zocca 2011).

7

Uribe-Etxebarria (2002) gives an alternative account involving overt wh-movement followed by remnant IP-movement (see Reglero 2007 for arguments against this analysis).   There are also differences among French, Spanish, and BP regarding the pragmatic/semantic conditions on the appropriateness of wh-in-situ questions (under particular conditions even English allows them), see Zocca (2011) and Pires and Taylor (2007). 8

While argument wh-phrases are standardly assumed to be island-insensitive, Watanabe (1992) claims some degradation is found with wh-islands (but not in Chinese, see Aoun and Li 1993. See also Bruening and Tran 2006 for a Vietnamese wh-in-situ strategy that is island-sensitive even with wh-arguments.)

9

For an alternative account of Bulgarian where fronted wh-phrases form a constituent prior to wh-movement see Grewendorf (2001). It appears that this account cannot capture the full range of Superiority effects discussed below (e.g. 45–46), since it predicts that the order of fronted wh-phrases will always correspond to their height in the base-generated positions.

10 With D-linked wh-phrases the range of felicitous answers is limited by a set of objects familiar to the speaker and the hearer as a result of it being referred to in the discourse or salient in the context. Their range of reference is thus discourse-given. Due to their “discourse giveness,” they are not inherently focused hence not subject to focus-movement. 11 I ignore D-linking questions and assume non-D-linking readings for non-inherently D-linked wh-phrases. 12 Note that Bošković (2012) establishes a correlation between superiory effects under MWF and articles, which can be generalized in such a way that languages with articles must have true wh-movement. 13 The gist of the system is that when the choice function has both wh-phrases in its domain, we get a set of propositions; when this is not the case we get a set of sets of propositions. 14 If a single wh-fronting language allows SP answers the language is predicted not to have true wh-movement. Rather, it would be fronting wh-phrases to a lower projection (e.g. FocP or TopP). 15 Heads can differ in how many times they attract a given feature: there are Attract-1F and Attract-all-F heads, the latter attracting all elements with the F-feature (which are not located in F-checking positions). 16 Note the parallelism between the wh-phrases in SC (31) and non-initial fronted wh-phrases in Bulgarian regarding Superiority, which follows from the focusmovement analysis, where the movement of the first wh-phrase in Bulgarian differs from the movement of the second and the third wh-phrase, which is in turn the same

278

CONTEMPORARY LINGUISTIC PARAMETERS

as the movement of both wh-phrases in SC (31) (see Richards 2001 for an alternative analysis.) 17 Hungarian often allows both single wh-fronting and MWF. I focus here on a context where the latter is forced (see Kiss 2002). 18 From the perspective of Bošković (2002), the fact that the verb-adjacent wh-phrase in (48e) is D-linked is an argument that the wh-phrase doesn’t undergo focusmovement. Note also that Hungarian passes Rudin’s (1988) main test for locating all fronted wh-phrases in SpecCP: as in Bulgarian, nothing can intervene between fronted wh-phrases in Hungarian (Puskás 2000). 19 Both D-linked and non-D-linked wh-phrases also must move in Basque. Reglero (2003), however, gives a slightly different account of Basque which is also stated in terms of Attract-all, but as a property of a discourse-related projection below CP which attracts both topicalized and focused elements (the underlying assumption being that D-linked wh-phrases are topics (cf. Den Dikken and Giannakidou 2002; Grohmann 1998, 2006). 20 I avoid null object wh-in-situ languages like Japanese due to the difficulty in teasing apart parasitic gaps and null objects. 21 Bošković (2008a) argues that unless the Bulgarian option is taken in the SC constructions in question, focus-movement would feed wh-movement, which is quite generally disallowed. 22 There are aspects of wh-islands where Bulgarian and Swedish do not pattern together. Richards (2001) shows Bulgarian often prefers crossing to nesting paths with extraction from wh-islands, while Swedish appears to prefer nesting paths (see, however, Engdahl 1986: 128–9). At any rate, whether a language prefers crossing or nesting paths does not seem to pattern with (in)sensitivity to wh-islands. (English and Swedish prefer nested paths, and Bulgarian and SC crossing paths.) 23 In Chomsky (2001), languages that disallow object shift do not disallow it because of the unavailability of the landing site for it (additional SpecvP), but because object shift leads to other problems. 24 Another option (from Bošković 2008a), where multiple Specs are not always in principle available, is to assume that heads are lexically specified via the EPP feature for merger with a Spec. However, EPP-features can always be freely given to a phase head to allow successive cyclic movement (Chomsky 2000). Under this approach, intermediate but not final heads in principle freely allow the multiple Spec option. 25 In many languages that are traditionally assumed to have overt reflexes of agreement with intermediate Cs, the paradigm cannot be replicated due to the lack of overt object agreement; however, it is possible such languages behave like Selayarese, the only difference being that the verb does not show overt object agreement, see Bošković (2008b) for such an account of Irish; see also Noonan (1999) for an account of traditional wh-C agreement in Irish which does not involve such agreement at all. Noonan argues that what is traditionally considered to be a special wh-agreeing C in Irish is not a C. For another view, see McCloskey (2002: 201). 26 For details of the analysis, including the question of why a clause can exceptionally remain in-situ to make wh-movement possible and why C cannot be overt in (74), see Bošković (2008b). 27 Nga occurs because monosyllabic Cs are second-position clitics. 28 I assume two uKs can be checked against each other if one of them is valued (Bošković

WH-MOVEMENT

279

2011; Pesetsky and Torrego 2007), which also allows us to dispense with reflex feature checking. 29 This is in fact the only point of variation that implicates the target of wh-movement, not wh-phrases themselves. The Bulgarian / English difference in this respect can also be captured by assuming the F-feature of C disappears (erases and deletes in Chomsky’s 1995 terms) after first checking in English, but not in Bulgarian. Assuming Higginbotham and May’s (1981) absorption for multiple questions, another possibility (which allows multiply filled CPSpecs in all languages) is that multiple wh-phrases located in SpecCPs (not in-situ) in English cannot undergo absorption, whereas this is possible in Bulgarian. Under this analysis, MWF questions in English are ruled out for semantic reasons. There is independent evidence that Bulgarian wh-phrases in interrogative SpecCPs are unusually free regarding absorption, which suggests the proper way of stating the cross-linguistic difference regarding absorption is as follows: a language either doesn’t allow additional wh-phrases in SpecCP to undergo absorption at all, or such wh-phrases are free to undergo absorption in any SpecCP. Dayal (1996) observes kade in (i), which is located in the embedded SpecCP, can take either matrix or embedded scope (i.e. it can be absorbed either with the embedded or the matrix wh-phrase).

(i) Koj znae kakvo kade e kupila Mariya?



who knows what where is bought Maria



‘Who knows where Maria bought what?’



For another perspective see Pesetsky (2000), where English allows multiple SpecCPs syntactically but it doesn’t allow pronunciation of more than one SpecCP in PF.

30 Who left what has the following derivation: who (with uF) and C check their F-feature against each other, while what doesn’t have uF.   Consider also the following Korean / English contrast.

(i) John-i

way ttena-ss-n?

John-NOM why left

(ii) *Who left why?



In Korean, C undergoes Agree with why for the F-feature, which is impossible in English for reasons discussed above. If wh-C must have some kind of a relation with each wh-phrase, we can then use only unselective binding in English. (ii) is then ruled out because adverbs cannot be unselectively bound.

31 There is an alternative where wh-phrases in all languages always have the F-feature, where F is optionally uninterpretable or interpretable in English (English then has both the Bulgarian and the Korean option). Since what in (82) may then have iF, hence can check the uF of C while in-situ, to force wh-movement we can appeal to Cheng’s (1997) clausal typing, on which questions where wh-movement does not occur overtly in English are filtered out due to the failure to type the clause as interrogative. (For Cheng, wh-movement is not necessary to type a clause in wh-in-situ languages.) 32 It’s possible that in partial wh-movement languages, declarative C can check the uF of wh-phrases. Under some analysis, the clause hosting partial wh-movement is treated like a question (Dayal 1996), which means the partially moved wh-phrase moves to a +wh-CP.

12 Topic and focus* Jordi Fortuny University of Barcelona

12.1. Introduction We shall be concerned with two traditional articulations of clauses in natural languages: the topic-comment articulation and the focus-background articulation.1 Let us succinctly present them in this introductory section using some familiar English patterns. i) The topic-comment articulation. The topic of a sentence denotes an entity which is presented as already existing in the discourse and belongs to the common ground shared by communicative agents. The comment denotes the proposition that “is about” the topic. Topichood can thus be conceived as an “aboutness” relationship holding between an argument (the sentence-topic) and a proposition (the comment) relative to a communicative context (Reinhart 1981). It is often claimed that in English the topic is usually realized as the grammatical subject; accordingly, the unmarked realization of topics in English would be by means of the grammatical subject. For example, once question (1-a) has been uttered, the denotation of the subject Peter belongs to the discourse; as a consequence in the response (1-b) the noun Peter is a topic and the predicate is in London the comment on this topic. (1) a. Where is Peter? b. Peter is in London When a topic does not correspond to the grammatical subject, then the basic SVO order may be modified by means of a syntactic operation called topicalization. In current transformational generative grammar, topicalization is commonly conceived in terms of a movement operation that targets a syntactic constituent and displaces it to the front of the clause. Examples (2-a) and (2-b) illustrate, respectively, an object topicalization and a PP topicalization.

282

CONTEMPORARY LINGUISTIC PARAMETERS

(2) a. [This man]i I have never seen ti b. [With Peter]i Mary went to school ti It is thus obvious that the topic-comment articulation is not equivalent to the subject-predicate articulation. Indeed, the constituent with which the verb agrees may be different from the topic, as illustrated, and it is also possible that the comment includes the subject; for instance, in the example (2-a) the comment is the open sentence I have never seen ti, which includes the subject I and the predicate have never seen ti. ii) The focus-background articulation. The focus of a clause denotes an entity that is opposed to a set of alternatives. The background conveys the information that is assumed by the speaker to be shared by him and the hearer(s). In the examples in (3) the background is conveyed by the sentence I invited and your brother is the focus constituent, whose denotation is opposed to other individuals taken as alternatives (say, Mary, John, and Theresa). As a consequence, constructions in (3) would be felicitous corrections to the claim you invited Theresa. We shall follow the convention of using small capitals for focus constituents. (3) a. I invited your brother b. your brother I invited c. It is your brother that I invited In English focus elements are marked by means of a nuclear pitch accent (3-a). As illustrated, it is also possible to displace the focus constituent to the front of the clause (3-b), an instance of movement dubbed focalization, and also to resort to cleft constructions (3-c). In all cases, the focus element preserves focal stress. The topic-comment and the focus-background articulations are relative to a level of interpretation responsible for the packaging of utterances by linking their information to the prior communicative context. Following Halliday’s (1967) and Chafe’s (1976) seminal investigations, as well as subsequent works, we shall name this level of interpretation Information Structure.2 Consequently, focus and topic (and the associated notions of presupposition and comment) are likely candidates for universal categories of Information Structure. On rather uncontroversial grounds, we can tentatively formulate, for the sake of concreteness, the following general principle: Information Structure Principle An I-language provides the expressive means to connect the information of linguistic expressions to the common ground through the topic-comment and the focus-background articulations. However, as we shall illustrate, there is no one-to-one mapping between the pragmatic notions of topic and focus and the way they are grammatically realized across languages. This implies that the above-expressed principle must be somehow parameterized. In the following sections we shall illustrate, with no aim of being exhaustive, the major points of cross-linguistic variation relative to this complex domain of grammar in order to extract relevant conclusions for parametric theory.



TOPIC AND FOCUS

283

In section 12.2 we shall begin by presenting the dichotomy between topicprominent language and subject-prominent languages (Li and Thompson 1976; Kiss 1995) and we shall consider the possibility of coding this distinction in terms of a Topic-Prominent Parameter. As we shall see, this hypothetical parameter would not satisfy certain learnability conditions, whereby its legitimacy is dubious. We shall connect the Topic-Prominent Parameter to the Radical Pro-Drop Parameter, and we shall suggest the possibility that the syntactic properties related to the positive value of these two alleged parameters may be inferred by the learner from a particular morphological cue in the Primary Linguistic Data: the existence of agglutinating morphology on pronouns. In section 12.3 we shall present Büring’s (2009) tripartite linguistic typology of focus marking, which can be easily formulated in terms of three parameters relative to the prosodic marking, the syntactic marking and the morphological marking of focus. We shall suggest that focus parameters may be three-valued, given that focus marking is often optional, and we shall indicate a couple of complications relative to the intricacies of focus interpretation and marking that can be incorporated in the minimal parametric system we describe.

12.2. The Topic-Prominent Parameter and learnability conditions It is commonly claimed that there is an important typological distinction between subject-prominent languages, such as Italian or English, and topic-prominent languages, which include, for instance, Chinese and Japanese (Li and Thompson 1976; Kiss 1995). Subjects in the former type of languages and topics in the latter play a prominent role in the syntactic organization of the clause. This typological distinction leads to the possibility that UG is endowed with the following parameter relative to the Information Packaging Principle introduced above: The Topic-Prominent Parameter Is L a topic-prominent language or a subject prominent-language? This parameter should be assigned one out of two possible values. Accordingly, a target language is either assigned a positive or a negative value, and thus it is analyzed by the learner either as topic-prominent or subject-prominent. Once the learner has fixed the value for this parameter during the acquisition process, then several syntactic properties should be automatically derived. Adapting Li and Thompson’s (1976) and Chafe’s (1976) seminal work to a parametric perspective, the following seven syntactic properties would be derived from setting this parameter to the positive (topic-prominent) value:3 (4) If L is a topic-prominent language, then: a. it will present a surface coding for the topic (a fixed position or a particle), but not necessarily for subjects b. it will lack passivization (or it will show it only as a marginal construction)

284



CONTEMPORARY LINGUISTIC PARAMETERS

c. it will lack subject expletives (as expected, given that the subject grammatical function does not play a prominent role) d. it will allow topic constituent not to have any selectional relationship with respect to the verb e. it will allow only topics to control deleted constituents f. it will be a V-final language

Let us briefly illustrate some of these properties. In the following Japanese (5a) and Mandarin (5b, 5c) sentences we can observe that the topic appears in an initial position; in this sense these two languages present a surface coding for topics. In Japanese, additionally, a topic particle -wa and a subject particle -ga are attached, respectively, to the topic and the subject; Chinese lacks both topic and subject particles, although it shows a sentence-initial position specialized for topics. The topic constituents presented in these three sentences are hanging topics, i.e. topics that are not selected by the verbal predicate. In order to translate these constructions with hanging topics into more or less natural constructions in a subject-prominent language like English we would in general need to introduce an “as for” phrase (“As for school I was very busy”).4 Finally, as illustrated in (5c), only the topic can control an object gap in the second clause. (5) a. Gakkoo-wa boku-ga isogasi-kat-ta [Japanese] School-TOP I-subject busy-PST ‘School, I was busy’ b. Nèike shù yèzi dà [Mandarin] that tree leaves big ‘That tree, the leaves are big’ c. [Nèike shù]i [yèzi dà]j, suŏyi wŏ bu xĭhuān ei/*j that tree leaves big so I not like ‘That tree, the leaves are big, so I don’t like it’ Although the distinction between subject-prominent languages and topic-prominent languages may seem quite sharp, at least three clarifications are in order. First, it is sometimes claimed that subject-prominent languages such as Spanish, Greek or English, also allow hanging topics, as in the following examples borrowed from Alexiadou (2006). However, as is well known, hanging topics in these languages must be resumed by an epithet, like the DP ese idiota, or the pronouns ton and him in the following examples (see Alexiadou [2006] and references cited therein). This is in contrast with hanging topics of topic-prominent languages (see the Mandarin example in 5b). (6) a. Bernardo estoy segura que nadie confia *(en ese idiota) [Spanish] Bernardo am sure that nobody trust that idiot ‘As for Bernardo, I am sure that nobody has confidence in that idiot’ b. Ton Petro *(ton) nostalgo poli [Greek] The Peter.ACC him miss.1SG much ‘Peter I miss him very much’ c. John I like *(him)



TOPIC AND FOCUS

285

Secondly, we must of course reject the idea that a negative value for the TopicProminent Parameter entails that subjects, instead of topics, have a fixed position in subject-prominent languages in general. For instance, Catalan, a null-subject language, favors different positions for subjects depending on thematic properties of the verb. More precisely, as observed by Rosselló (2002), when the whole sentence is focal, the subject of an unaccusative verb (venir, “to come”) must appear postverbally for the sentence to be fully acceptable, whereas the subject of an unergative verb (riure, ‘to laugh’) or a transitive verb (perdre, ‘to loose’) must appear preverbally. (7) Què ha passat? (‘What has happened?’)    [Catalan] a. Ha vingut el president b. #El president ha vingut ‘The president has come’ (8) Què passa? (‘What is happening?’) a. El president riu b. #Riu el president ‘The president has laughed’ (9) Què ha passat? (‘What has happened?’) a. El president ha perdut les claus b. #Ha perdut les claus el president ‘The president has lost the keys’ Therefore, there is no fixed position for subjects in Catalan: the basic position is preverbal for agentive subjects and postverbal for thematic subjects. Thirdly, we must emphasize as well that the absence of subject expletives is also a property of null-subject languages, a subset of subject-prominent languages. Accordingly, the absence of expletives is related to the positive setting of at least two parameters: the Topic-Prominent Parameter and the Null-Subject Parameter. Below we shall argue, though, that only the presence of operations or categories should be derived from parameter setting devices, and not the absence of operations or categories. These two observations (i.e. the position of subjects and the absence of subject expletives) raise the question of how the Null-Subject Parameter and the Topic-Prominent Parameter relate to each other (Ordóñez and Treviño 1999; Rosselló 2000; Biberauer et al. 2010). It is important to our concerns to recall at this point that parameters, at least as they were originally conceived (Chomsky 1981), are not simply tools to describe the attested patterns of syntactic variation. Parameters are assumed to be valued by infants during the language acquisition process, in such a way that, on the basis of the Primary Linguistic Data, an infant decides whether a given parameter has a positive or a negative value in the target language. Different values of the relevant parameters lead infants to different I-languages. Accordingly, it seems fair to conclude, following Fasanella and Fortuny (in press), that the primary goal of parametric models is to determine how parameters can be set during the process of language acquisition on the basis of the Primary Linguistic Data. This would entail

286

CONTEMPORARY LINGUISTIC PARAMETERS

that the descriptive adequacy of the tools we use to describe syntactic variation—i.e. parameters—depends on its explanatory adequacy. Assuming this line of reasoning, it is legitimate to postulate a parameter only if this parameter can be set on the basis of the Primary Linguistic Data. However, as argued by Fasanella and Fortuny (in press), most parametric proposals neglect this important requirement, since they display the following intuitive learnability problems. (10) Locality Problem In order to fix the value of a macro-parameter or a messo-parameter the learner should analyze the data he receives in a global and transverse way, since these types of parameters are defined on highly general properties that are spread across the target language. (11) Linking Problem Micro-parameters, as well as macro-parameters and messo-parameters, are defined over abstract linguistic entities, so the infant still faces the problem of linking these abstract mental representations to actual physical entities in the speech signal (Gervain and Mehler 2010: 194; cf. Pinker 1984). Fasanella and Fortuny (in press) suggest that genuine parameters should satisfy the following two learnability conditions, which ensure that parameters can be set by means of principles of analysis that operate on the Primary Linguistic Data. Conditions (12) and (13) are introduced to avoid respectively problems (10) and (11). (12) Atomicity Condition Parameters must be atomic, they cannot be clusters of properties. (13) Accessibility Condition Parameters must be set by directly inspecting phonological and morphological properties of utterances. It is rather unclear how the Topic-Prominent Parameter could satisfy these two conditions. Note that the learner should be able to detect whether the target language is topic-prominent or subject-prominent; however, in order to decide whether the target language belongs to the former or the latter syntactic type it is necessary to decide first multiple less general properties. For this reason, the parameter under discussion is not atomic. It does not seem to be a genuine parameter, but rather the cumulative effect of previously defining multiple and more local parameters. The Topic-Prominent Parameter does not satisfy the Accessibility Condition either, since the very general property of being a subject-prominent or a topicprominent language cannot always be directly detected from phonological and morphological properties of utterances. It may be the case, for instance, that a topicprominent language displays topic particles, like Japanese, but this is not always the case; as illustrated above, Mandarin lacks them. Accordingly, we cannot conclude that the presence of topic particles is a morphological cue that lets the learner to infer that the target language is topic-prominent.



TOPIC AND FOCUS

287

We must recall that it is a conceptual necessity that the learner segments speech into morphological units. Our claim is that this morphological analysis may be used to infer higher order syntactic properties during the language acquisition process. This working hypothesis has some plausibility if we take into consideration that syntactic and morphological properties tend to correlate. However, it must be remarked that this approach opens several important research questions relative to speech segmentation and language acquisition. Crucially it should be clarified how morphological units are identified in continuous speech during the process of language acquisition. In different words, we should be able to reduce, ultimately, the above given Accessibility Condition to the following formulation: “Parameters must be set by directly inspecting physically accessible properties of utterances.” What we suggest here is that the problem of modeling syntactic variation using parameters must be related to the problem of grounding self-contained linguistic units to physically accessible properties. Our research proposal leads to the idea that in order to meet intuitive learnability conditions such as (12) and (13) and at the same time account for patterns of linguistic variation the Language Acquisition Device must be defined as a discovery procedure that provides a practical and mechanical method for actually constructing a grammar on the basis of the Primary Linguistic Data, as argued recently in Fasanella (2014). For such a discovery procedure to be successful, it is necessary to show how relatively abstract properties may be determined on the basis of relatively less abstract properties in a progressive way. More particularly, the Language Acquisition Device should provide the path to ground syntactic properties on morphological properties, morphological properties on phonological properties, and phonological properties ultimately on acoustic properties. These considerations amount to think that the typological dichotomy between subject-prominent languages and topic-prominent languages cannot be appropriately grasped in terms of a genuine parameter that is set during the language acquisition process. Let us investigate a possible solution to this problematic situation. It seems reasonable to think that the properties described in (4) may be intimately connected to a different parameter, namely, the Radical Pro-Drop Parameter (cf. D’Alessandro, this volume), since languages that can in general drop topics are topic-prominent languages, and vice versa. The Radical Pro-Drop Parameter Verbal arguments and possessors can be dropped when they are topical If this is the case, then we can capitalize on Fasanella and Fortuny’s (in press) elaboration of Neeleman and Szendrői’s (2007) proposals in order to propose a parametric model that accounts both for radical pro-drop and topic-prominence and satisfies intuitive learnability conditions. Let us see how. According to Neeleman and Szendrői’s strong prediction, all languages with agglutinating morphology (cf. Fábregas, this volume) on pronouns have radical pro-drop. In Korean, for instance, pronouns carry the same case particles that nouns do as suffixes. Below we provide the Korean pronominal stems (Table 12.1) and the case endings (14) that can be attached to them.

288

CONTEMPORARY LINGUISTIC PARAMETERS

Table 12.1  Korean pronominal stem 1p

2p

3p

Sg.

/na/

/ne/

/ku/

Pl.

/wuli/

/nehuy/

/kutul/

(14)

Korean case endings a. Nom: /ka/ b. Acc: /(l)ul/ c. Gen: /uy/ d. Dat: /ey/; /eykey/; etc.

We can combine for instance the third person plural pronoun kutul with the nominative case ending ka in order to obtain kutul-ka. This contrasts with the completely fusional pronominal paradigm of Swedish, which presents no case and number markers that can be separated from the stem.

Table 12.2  Swedish pronominal paradigm Nominative

Accusative

Possessive

1 sg.

jag

mig

min

2 sg.

du

dig

din

3 sg. M.

han

honom

hans

3 sg. F.

hon

henne

hennes

1 pl.

vi

oss

vår

2 pl.

ni

er

er

3 pl.

de

dem

deras

Given that the learner may need to segment the continuous speech that surrounds him and obtain a morphological analysis, he must arrive at the conclusion that the target language presents a morph instantiating the category of case or number which is phonologically bound to the pronoun. Fasanella and Fortuny (in press) postulate that, once this morphological analysis is obtained, there may be a morphosyntactic bootstrapping mechanism that takes as input this morphological analysis and infers those syntactic properties of the target language associated to the so-called Radical Pro-Drop Parameter: verbal arguments and possessors can be grammatically omitted when they are topical. We can correctly extend this reasoning to bootstrap as well the syntactic properties associated to the positive value of the Topic-Prominent Parameter.



TOPIC AND FOCUS

289

(15) If the learner has determined that there is a [+bound] head conveying case or number on pronouns in the Primary Linguistic Data, then he can infer that the target language L: a. allows verbal arguments and possessors to be omitted b. will present a surface coding for the topic (a fixed position or a particle), but not necessarily for subjects c. will allow topic constituent not to have any selectional relationship with respect to the verb d. will allow only topics to control deleted constituents e. will be a V-final language Note that the following two properties associated above to the Topic-Prominent Parameter are not included in the bootstrapped properties: (16) a. it will lack passivization b. it will lack subject expletives The reason for not including these properties is that they are relative not to the presence but the absence of grammatical operations and categories: the absence of passivization and of subject expletives. The type of inferential learning mechanisms we postulate is simplified if we consider only properties relative to the presence of operations or elements. Claiming, for instance, that the absence of expletives is inferred is an unnecessary complicated way to say that expletives are not learned, because the relevant piece of evidence is not present in the Primary Linguistic Data. In different words, properties in (18) may be important from a strictly typological view point, but it seems dubious to us that they should be relevant for determining parametric learning systems. We would like to end this section on the so-called Topic-Prominent Parameter and Radical Pro-Drop Parameter emphasizing that in the framework we have sketched they would be generalizations of how certain properties tend to cluster together, but properly speaking they would not be parameters set during the language acquisition process. Our proposal is that, once the learner has discovered that the target language contains a morph expressing case or number attached to pronouns, then the syntactic properties compressed in these two parameters are bootstrapped. As a matter of fact, this approach capitalizes on the traditional intuition that there is a correlation between morphological and syntactical properties and suggests that syntactic properties are inferred once morphological properties have been determined.

12.3. Towards a cross-cutting three-valued parametric system for focus marking It is well known that languages can use morphological, syntactical and prosodic means to mark a constituent as a focus. We shall concentrate on this cross-cutting nature of focus marking in order to investigate the descriptive and theoretical adequacy of parameters.

290

CONTEMPORARY LINGUISTIC PARAMETERS

In this regard, we would like to emphasize that parameters are mainly used by syntacticians with the objective of modeling patterns of variation relative to syntactic phenomena, such as word order, argument omission, and grammatical relations. If understanding language learnability and variability in terms of a parameter fixation problem is a meaningful insight, then it is mandatory to investigate the adequacy of parameters with respect to the variation attested in grammatical levels of representation distinct from syntax. It is also important to take into consideration whether parameters can be used to model those points of grammatical variation that clearly involve mechanisms relative to different levels of representations. In this sense, it seems to us that the case study of the grammatical realization of focus deserves special attention. We shall take as a starting point Büring’s (2009) typology, according to which languages can be classified in terms of boundary languages, edge languages and particle languages with regard to focus marking.5 Let us consider the following definitions: A language L: 1.  is a boundary language if focus is marked by insertion of a prosodic phrase boundary to the left or right of the focus constituent. 2.  an edge language if focus is marked by non-standard constituent order, with the focus in left- or right-peripheral position. 3. a particle language if focus is marked by a special morph. We shall briefly illustrate these definitions.

12.3.1. Boundary languages Chicheŵa and Bengali are classified as boundary languages: the former introduces a boundary to the right of the phonological phrase (pP) that is interpreted as a focus, whereas the latter introduces such a boundary to the left of the focus pP.6 Pp-boundaries can be detected because they block various segmental processes in Chicheŵa; in Bengali they are detectable through various prosodic processes and in addition each pP has one L* pitch accent on its head, and one final boundary tone. Chicheŵa and Bengali are, respectively, SVO and SOV languages. Their respective default phrasings at the pP level are schematized in (19): (17) a. (S) (V O1 O2 Obl) b. (S) (O) (V) In Chicheŵa, a right boundary to mark a specific constituent as a focus is inserted if the default phrasing does not provide it. For instance, in (18) the whole VP is interpreted as focus, whereby the pP contains the whole VP (i.e. three prosodic words). No boundary needs to be introduced. In (19) only the verb is focused; therefore, a boundary is introduced at the right of the verb anaményá ‘hit’ in order to signal that the focus is not the VP but only the verb.



(18)

TOPIC AND FOCUS

291

What did he do? [Chichewa] ([Anaményá nyumbá ndímwáála]F )pP hit house with.rock ‘He hit the house with a rock’

(19) What did he do to the house with the rock? (AnaményáF ) (nyumbá) (ndímwáála) The case of Bengali seems much more unclear. According to Selkirk (2007) (see also Büring 2009), a pP-boundary is inserted to the left of the focus ĉohbir, “pictures,” in (20b), a boundary that is not present in the neutral sentence (20a). (20) a. ami raǰar ĉohbir ǰonno ţaka anlam I king’s picture for money gave ‘I gave money for the king’s picture’ b. ami raǰar ĉohbirF ǰonno ţaka anlam

[Bengali]

With regard to the prosodic marking of focus, we must also take into consideration the case of languages such as English, which apparently lack any consistent segmental or tonal boundary insertion but nonetheless resort to pitch accents to indicate the focus constituent, as we noted above. English is considered a SVO language with (S)(VO) prosodic structure. According to Büring (2009), in English “heads of pPs are realized by pith accents, the intonational phrase head being the nuclear pith accent. […] The primary indicator of the IP-head is that no pitch accents can follow the one marking the iP-head, which yields the familiar effect that a non-sentence-final focus is marked by an early nuclear pitch accent, that is, by the absence of later pitch accents which would otherwise be present” (2009: 188–9). In sum, languages may use pitch accent as the main indicator of focus, such as English; other languages may mark focus by placing a pP-boundary to the right or the left of the focus constituent, such as Bengali or Chicheŵa, respectively.

12.3.2. Edge languages Languages such as Hausa or Hungarian mark a syntactic constituent as a focus by placing it in a position semantically devoted to focus. For instance, in Hungarian a focus constituent appears in the immediately pre-verbal position (Szabolcsi 1981). The focus can be preceded by one topic or more.7 (21) a. Péter [a padlón]F aludt Peter on floor slept ‘Peter slept on the floor’ b. A padlón [Péter]F aludt On floor Peter slept ‘Peter slept on the floor’

[Hungarian]

292

CONTEMPORARY LINGUISTIC PARAMETERS

In languages such as Italian, Spanish, or Catalan, the focus constituent needs to appear in the final position of the clause if it is not contrastive, as illustrated in the following Catalan sentences. In (22b) the whole sentence is focal and its constituents display the canonical S V ADV PP order. The same order is illustrated in (23b): the focus constituent is the commitative PP, which already appears in the final position when the whole sentence is focal (22b). However, when the subject is focal, the sentence is only felicitous if the subject appears in the final position, which means that the basic constituent order is altered (24b, 24c). (22) a. Què va passar ahir? what PAST.3sg happen yesterday ‘What happened yesterday?’ b. En Pere va sortir ahir amb la Maria ART Pere PAST.3SG go.OUT yesterday with art Maria ‘Pere went out yesterday with Maria’ (23) a. Amb qui va sortir ahir with whom PAST.3SG go.OUT yesterday ‘With whom did Pere go out yesterday?’ b. En Pere va sortir ahir ART Pere PAST.3SG go.out yesterday ‘Peter went out yesterday with Maria’

en Pere? art Pere amb la Maria with ART Maria

(24) a. Qui va sortir ahir amb la Maria? Who PAST.3SG go.out yesterday with ART Maria ‘Who went out yesterday with Maria?’ b. Va sortir ahir amb la Maria en Pere PAST.3SG go.out yesterday with ART Maria ART Pere ‘Pere went out yesterday with Maria’ c. #En Pere va sortir ahir amb la Maria ART Pere PAST.3SG go.out yesterday with ART Maria ‘Pere went out yesterday with Maria’ When the focus is contrastive, then a leftward focalization needs to be applied: (25) Amb la Maria va sortir ahir en Pere, i no amb la Carme with ART Maria PAST.3SG go.out yesterday ART Pere, and not with ART C. ‘Pere went out yesterday with Maria, and not with Carme’ Therefore, it seems that focus constituents in edge-languages must appear either in the left periphery (for instance in Hungarian) on in the right periphery (Catalan, Spanish, Italian). When the focus constituent bears a contrastive interpretation in the second type of languages, it can be left-dislocated.



TOPIC AND FOCUS

293

12.3.3. Particle languages It is well known that certain languages mark focus by means of particles. These languages include Wolof, Hausa, and Gúrúntúm. As illustrated below, Gúrúntúm has a focal particle á that is placed right before the focus. The particle is absent in (26), which lacks a focus constituent, and it is placed right before the subject focus and the object focus in (27b) and (28b)—in this latter case, á attaches to the preceding word. (26) Tí bá wúm kẃalíngálá [Gúrúntúm] 3SG PROG chew colanut ‘He is chewing colanut’ (27) Subject focus a. Á kwá bá wúm kẃalíngálá-í? FOC who PROG chew colanut-the ‘Who is chewing the colanut?’ b. Á fúrmáyò bá wúm kẃalíngálá FOC fulani PROG chew colanut ‘[The fulani]F is chewing colanut’ (28) Object focus a. Á kãéã mái tí bá wúmì? FOC what REL 3SG PROG chew ‘What is he chewing?’ b. Tí bá wúm-á kẃalíngálá 3SG PROG chew-FOC colanut ‘He is chewing [colanut]F’ In sum, languages may resort to phonological and syntactic strategies to mark a constituent as a focus, as well as to specialized particles, as we have just illustrated.

12.3.4. A parametric approach Let us tentatively express the following four parameters relative to focus marking, which reformulate Büring (2009)’s tripartite typology sketched in the previous subsection. (29) Let X be a constituent interpreted as a focus. Parameter 1. Does the target language L mark X as a focus by means of a specialized particle Y? Parameter 2. Does the target language L mark X as a focus by means of a pitch accent? Parameter 3. Does the target language L mark X as a focus by means of a prosodic indicator?

294

CONTEMPORARY LINGUISTIC PARAMETERS



Sub-parameter 3.1. If L marks X as a focus by means of a prosodic indicator, is it placed at the right or at the left of the focus? Parameter 4. Does the target language L mark focus by means of a reordering operation? Sub-parameter 4.1. If L marks focus by means of a reordering operation, is the focus constituent placed at the left periphery or at the right periphery?

Admittedly, there are many points of cross-linguistic variation relative to focus marking that are not grasped by this alleged parameters and subparameters. Nonetheless, they can be used to reach some conclusions about how a parametric approach to this grammatical phenomenon would look like and to raise several programmatic questions that future research could address. Note that this parametric system satisfies the cross-cutting requirement we mentioned above in Section 3, namely that cross-linguistic variation relative to focus marking involve mechanisms belonging to different levels of grammatical representation. Observe that in (29) we have begun stating “Let X be a constituent interpreted as a focus”. This statement is necessary to avoid certain learnability problems, and to satisfy certain learnability conditions, discussed by Fasanella and Fortuny (in press) and mentioned in section 2. In order to illustrate this point, let us assume that we do not make that statement and instead we formulate Parameter 1 in the following way: “Does the target language L mark focus by means of a specialized particle?” In this formulation, there is no mention to a category X identified by the learner in the Primary Linguistic Data. In the case that the target language lacked a particle marking focus, we would need to add some constraint that avoided that the learner went on forever searching for some evidence in order to set the parameter to the negative value. In brief, the formulation in (29) intends to satisfy the following learnability condition:8 (30) Positive Evidence Condition Both values of a parameter must be set on the basis of positive evidence. Of course, a different and non-trivial question is by means of what mechanisms the infant determines that a syntactic constituent or a morph is interpreted as a focus, a topic, a question phrase, etc. Learning the meaning of morphs expressing abstract notions, such as grammatical morphs, on the basis of the linguistic context where they are used “requires advanced abilities, such as categorization, understanding referentiality and solving the induction problem for meaning,” as noted by Gervain and Mehler 2010 (cf. also Quine 1960; Nazzi and Bertoncini 2003; Waxman and Gelman 2009). We would like to pay attention on optionality, which poses an interesting complication for parametric approaches in general, and in particular for the system defined in (29). As far as focus marking is concerned, it is well known, for instance, that Hausa may mark a category X as focus by moving it to a left-peripheral position; it can also resort in certain tenses to a specific form of the auxiliary called the relative form, as well as to raising of lexical high tones within X. However, focus in this language can also be left in situ without introducing any acoustically measurable



TOPIC AND FOCUS

295

mark (cf. Büring 2009 and references cited therein). This observation seems to suggest that parameters cannot only be assigned to a positive or to a negative value, but also to a third value expressing focus-marking optionality. Therefore, Parameters 1, 2 and 3 in (29) seem to be three-valued: they can be assigned a value “□” that indicates the necessity of marking focus by means of a given particle (Parameter 1), a prosodic feature (Parameter 2) or a movement operation (Parameter 3), a value “□¬” that indicates the impossibility of marking focus and a value “◊” that indicates the optionality of marking focus. This would be the three-valued (toy) parametric system for topic marking; in the case of Hausa Parameters 1, 2 and 3 would be assigned the ◊ value. Subparameters 3.1 and 4.1 are boolean or two-valued.9 (31) Let X be a constituent interpreted as a focus.

Parameter 1. Does the target language L mark X as a focus by means of a specialized particle Y? □ (Yes necessarily) ◊ (Yes optionally) □¬ (Necessarily not)



Parameter 2. Does the target language L mark X as a focus by means of a pitch accent? □ (Yes necessarily) ◊ (Yes optionally) □¬ (Necessarily not)



Parameter 3. Does the target language L mark X as a focus by means of a prosodic indicator? □ (Yes necessarily) ◊ (Yes optionally) □¬ (Necessarily not)



Sub-parameter 3.1. If Parameter 3 is set to ÿ or to à, is the prosodic indicator placed at the right or at the left of the focus? 0 (‘placed at the right’) 1 (‘placed at the left’)



Parameter 4. Does the target language L mark focus by means of a reordering operation? □ (Yes necessarily) ◊ (Yes optionally) □¬ (Necessarily not)



Sub-parameter 4.1. If Parameter 4 is set to □ or to ◊, is the focus constituent placed at the left periphery or at the right periphery? 0 (‘placed at the right periphery’) 1 (‘placed at the left periphery’)

296

CONTEMPORARY LINGUISTIC PARAMETERS

I would like to mention a further difficulty for the attempt to systematize crosslinguistic patterns of focus marking in terms of parameters. As shown by Hartmann and Zimmermann (2007a, 2007b), several Chadic languages and Kwa languages distinguish between the grammatical marking of subject and non-subject foci. For instance in Fon, a language whose basic constituent order is SVO, when the sentence-initial subject is interpreted as a focus, it must be followed by the particle wɛ̀. However, when the focus is a non-subject constituent (an object or an adjunct), then the sentence either displays the canonical constituent order or the focus constituent is fronted, in which case the particle wɛ̀ must follow the focus. (32) a. Who ate (the) beans? b. nyɔ̀nú ó *(wɛ̀) d ayìkún woman DEF FM eat bean ‘The woman ate the beans’

[Fon]

(33) a. What did the woman eat? b. é d ayìkún 3SG eat bean ‘She ate beans’ c. ayìkún *(wɛ̀) nyɔ̀nú ó d bean FM woman DEF eat ‘It is beans that the woman ate’ These patterns suggest that the valuation process of Parameters 1 and 4 relative, respectively, to the morphological and the syntactic marking of focus can be sensitive to the subject / non-subject distinction. In the case of Fon, Parameter 1 is set to □ for subjects and Parameter 4 is set to ◊ for non-subjects; in this latter case, the learner must infer that if the non-subject focus is left-dislocated, then Parameter 1 is set to ◊, and accordingly the relevant particle may follow the dislocated focus or not. Before concluding our review on how cross-linguistic variation concerning focus marking can be analyzed using the parametric system in (31), we should verify that the parameters contained in (31) satisfy the Atomicity Condition and the Accessibility Condition presented respectively in (12) and (13). It seems quite clear that they do satisfy the Atomicity Condition, since every parametric question is referred to a single property: the presence of a specialized particle (Parameter 1), the presence of a pitch accent (Parameter 2) or a prosodic indicator (Parameter 3), which may appear at the left or the right of the focus (Subparameter 3.1), and the availability of reordering operations (Parameter 4 and Subparameter 4.1). As for the Accessibility Condition, which requires that parameter must be set by directly inspecting phonological and morphological properties of utterances, we must be very cautious. Parameters 1, 2, and 3 seem unproblematic, because they refer to morphological and phonological properties: the presence of a specialized particle in the case of Parameter 1 and the presence of a pitch accent or a prosodic indicator in the cases of Parameters 2 and 3. But Parameter 4 and Subparameters 3.1 and 4.1 are apparently more complex. Consider, for instance, Subparameter 4.1, which asks whether the focus is placed at the left or right periphery. If this



TOPIC AND FOCUS

297

distributional property can be determined on the basis of strictly prosodic hints, then Subparameter 4.1 (and thus Parameter 4) respects the Accessibility Condition in the very restrictive terms we have formulated it. If this were not possible, then we should redefine the Accessibility Condition in order to introduce the required distributional criteria to value Parameter 4.1. This would not be a drastic change for our general proposal, as long as the revised condition continues avoiding the existence of parameters that postulate elements that cannot be linked to speech in any way, such as traces, uninterpretable features, or very general properties about the target language, like for instance whether or not it belongs to the polysynthetic type or not. Recall that the Accessibility Condition simply reminds us that we can postulate only those parameters that can be valued on the basis of the Primary Linguistic Data. The condition should thus be redefined as we learn about what features are accessible in speech analysis and segmentation and how. We must remark that, as argued above, Parameters 1 and 4 may be sensitive to the subject / non-subject distinction. This means that these parameters may not have a single value within a given language (i.e. Fon), but they may be assigned one value when the focus is subject and a different value when it is a non-subject. Consequently, the learner may assign a value to the parameter for subjects, retain this value, and learn that it must assign a different value to the same parameter for non-subjects. Optionality does not seem to present a serious problem for our approach. Simply, the learner determines whether a particular marking for foci is mandatory, illegitimate or optional depending on frequency, i.e. depending on whether such a marking appears in all cases (or practically all cases, if we take into account the inconsistencies of the Primary Linguistic Data), in no case at all, or in some cases.

12.4. On the locus of variation within the architecture of language We would like to discuss the consequences that our previous considerations have in relation to the locus of topic and focus marking within the architecture of language. Recall that we have based our inquiries on the following presumable principle of Universal Grammar: Information Structure Principle An I-language provides the expressive means to connect the information of linguistic expressions to the common ground through the topic-comment and the focus-background articulations. This principle merely ensures that any I-language attained by an infant will have the expressive capacity to connect expressions to the common ground using particularly the two clausal articulations we have studied. This principle belongs to Information Structure, an interpretation level of the Conceptual-Intentional system that interfaces in an intricate way with several grammatical levels, such as narrow syntax, logical form and even phonetic form (see footnote 2). The principle is flexible enough

298

CONTEMPORARY LINGUISTIC PARAMETERS

to permit variability related to how focus and topic are grammatically marked, and indeed, it does not provide any hint about how the infant may determine the grammatical devices used in his target language with the purpose of grammatically marking these notions. It is undeniable that there is syntactic variation relative to topicality. Whether or not verbal arguments and possessors can be omitted and whether or not topics are allowed to control deleted constituents are properties relative to the interaction between syntax and argument structure in the former case and syntax and logical form in the latter. These points of variation cannot be attributed to the externalization systems of language in any sensible way. Similar considerations would apply to focus marking; in this regard, recall that the focalization of quantificational adverbs affect their scopal domain calculated at logical form. Therefore, our previous review strongly suggests that the points of variation relative to topic and focus marking are spread through different linguistic domains including syntax. This is particularly clear in the parametric system we have developed to model focus marking, which takes into consideration not only the syntactic aspects of the phenomenon, but also its phonological and morphological. We can of course assume that syntax consists simply in a combinatorial procedure that generates hierarchical structures that are interpreted in the external systems; however, the syntactic component of different I-languages may generate representations that diverge in relatively important ways, depending on whether or not focus and topics move and form chains, depending on whether any topic can grammatically control an deleted constituent, etc. A different question is whether there exist parameters coded in UG whose nature is inherently syntactic. In relation to topic marking we have argued that postulating a syntactic parameter such as the Topic-Prominent Parameter is problematic because it suffers from certain learnability problems. Instead we have argued for the possibility that the multiple syntactic properties compressed by the Topic-Prominent Parameter and the Radical Pro-Drop Parameter are the result of a bootstrapping mechanism. Therefore, the principle above-mentioned is not associated to a syntactic parameter (the Topic-Prominent Parameter) but to the bootstrapping mechanism formulated in (15) and repeated below in (34) for clarity: (34) If the learner has determined that there is a [+bound] head conveying case or number on pronouns in the Primary Linguistic Data, then he can infer that the target language L: a. allows verbal arguments and possessors to be omitted b. will present a surface coding for the topic (a fixed position or a particle), but not necessarily for subjects c. will allow topic constituent not to have any selectional relationship with respect to the verb d. will allow only topics to control deleted constituents e. will be a V-final language Focus marking, like topic marking, involves cross-linguistic variation regarding several levels of grammatical representation, namely morphology, prosody and syntax. Note, though, that whereas we have modeled cross-linguistic variation



TOPIC AND FOCUS

299

relative to topic marking in terms of a bootstrapping mechanism, in the case of focus marking we have opted for developing a parametric system that affects several domains of grammar. Parameter 1 of this system is relative to morphological properties: “is focus marked by means of a specialized particle?” Parameters 2 and 3 are related to prosodic properties: respectively, “is focus marked by means of a pitch accent or by means of a prosodic indicator?” If the latter is the case, the Subparameter 3.1 asks whether the prosodic indicator appears at the right or at the left of the focus. Finally, Parameter 4 and Subparameter 4.1 are relative to syntactic properties: is the focus constituent placed at the left or at the right periphery? We have shown the path to verify how these parameters may be linked to the analysis of the Primary Linguistic Data assuming a rather restrictive view of the Accessibility Condition that does not provide a full-fledged syntactic analysis of strings of symbols, but solely a morphological and a phonological / prosodic analysis. Finally we would like to remark that placing variation not only in the externalization systems of language but also in the syntax-semantics interface is compatible with Chomsky’s Uniformity Hypothesis, if we can show the path to detect this variation by means of principles of data analysis. Uniformity Hypothesis In the absence of compelling evidence to the contrary, assume languages to be uniform, with variety restricted to easily detectable properties of utterances (Chomsky 2001: 2) In this sense, assuming the Uniformity Hypothesis does not reject the existence of variation in the core aspects of grammar (say, logical form and narrow syntax); it compels us to investigate how this type of variation can be grounded ultimately on the analysis of accessible properties of speech.

12.5. Conclusions and further lines of research In this chapter we have explored the possibility that major points of linguistic variation concerning the grammatical marking of topic and focus constituents may be modeled using parameters. We have argued that the Topic-Prominent Parameter does not satisfy two learnability conditions: the Atomicity Condition and the Accessibility Condition. As a consequence, it is rather mysterious how an infant would be able to value this parameter on the basis of the Primary Linguistic Data. We have suggested the possibility that the properties resulting from the positive value of the TopicProminent Parameter and the Radical Pro-Drop Parameter may be bootstrapped from a morphological cue: the presence of agglutinating morphology on pronouns. If this is correct, the two parameters just mentioned are not genuine parameters; the properties they range over would derive from the operations of the morphological analyzer and the existence of bootstrapping mechanisms that infer syntactic properties of the target language on the basis of morphological cues.

300

CONTEMPORARY LINGUISTIC PARAMETERS

The rationale of rethinking the cross-linguistic observations derived from parametric studies under the light of learnability considerations opens up several important questions relative to language acquisition, speech segmentation and comparative linguistics. Can we link in general syntactic patterns of variation to morphological properties of the target language? If this is the case, how does the infant attain the morphological analysis required to bootstrap syntactic properties if speech is mostly continuous? Can we show that infants really bootstrap the syntactic properties that seem to correlate with the relevant morphological properties? Of course, the observation that certain syntactic properties correlate with morphological properties only suggests that it might be plausible that morphological properties are cues for the bootstrapping of syntactic patterns of variation. This possibility should be experimentally investigated. We have sketched Büring (2009)’s tripartite typology of focus marking and we have shown that it can be easily reformulated in terms of a parametric system. We have insisted on the necessity of defining parametric systems that take into consideration not only syntactic observations, but also prosodic and morphological features, in order to provide a minimally comprehensive parametric model for focus marking. We have defended the introduction of three-valued parameters in order to account for the apparent optionality of focus marking and we have brought into the discussion the observation that, in a given language, the parameters relative to focus marking may be assigned different values for subjects and non-subjects. This provides a minimal parametric system that can be productively enriched. We have finally argued for the view that the observed patterns of crosslinguistic variation relative to topic and focus marking affect not only relatively external domains of grammar, such as prosody and morphology, but also core aspects of narrow syntax and logical form that should be detectable by principles of analysis.

Notes *

This research has been funded by the projects FFI2011-23356, FFI2013-46987C3-1-P and FFI2014-56258-P, awarded by the Spanish Ministerio de Economía y Competitividad and the project 2009SGR-1073, awarded by the Generalitat de Catalunya.

1

We would like to refer the interested reader to the following general studies on these two articulations: Vallduví 1993; Lambrecht 1996; Erteschik-Shir 2007; Reinhart 1981; and Krifka 2008. Needless to say, the current literature on this grammatical domain is extremely rich and diverse. Cinque’s (1990) and Rizzi’s (1997) influential studies of movement possibilities of focus and topic constituents deserve also a very special mention.

2

An important question is how Information Structure is related to other levels of grammatical representation, particularly to Logical Form, where scope relations are calculated on the basis of hierarchical representations. Several authors (Partee 1991; Diesing 1992b; Von Fintel 1994, 2004; Herburger 2000) have shown that the domain of quantification of several adverbs is affected by focalization operations; as a consequence, some of the properties of the representations attained at the level



TOPIC AND FOCUS

301

of Information Structure feed Logical Form, and are thus relevant for the truthconditional interpretation of propositions (Krifka 2008). Cf. Espinal and Villalba (in 2015) for a recent argumentation for the thesis that Information Structure precedes both Logical Form and Phonetic Form and Irurtzun (2009) for a defence of the idea that the inverted Y-model of the architecture of grammar is better suited than “parallel architectures” proposed by Jackendoff (2007). 3

Our approach to the Topic-Prominent Parameter differs from Baker’s definition expressed in the following terms (Baker 2001: 182):



“A sentence may be made up of an initial noun phrase (the topic) and a complete clause that is understood as a comment on that topic (Japanese—i.e. a topicprominent language). or No topic phrase distinct from the clause is allowed (English—i.e. a subject-prominent language).”

Baker seems to assume that this is equivalent to Li and Thompson’s original definition of topic-prominent languages and subject-prominent languages. But this would not be correct. Indeed, as illustrated in (2), English can have topic phrases distinct from the clause, and nonetheless, it is not a topic-prominent language according to Li and Thompson’s typology, because it display none of the properties listed in (4). We do not adopt Baker’s definition of the Topic-Prominent Parameter and instead we remain faithful to Li and Thompson’s generalizations by relating the positive value of the Topic-Prominent Parameter to the properties in (4).

4

This point is exemplified in the following Catalan examples borrowed from Villalba (2000):

(i) a. *Peix, ahir vaig menjar llenguado [Catalan] ‘Fish, yesterday I ate sole’ b. *Formatge, m’agrada molt el Cheddar ‘Cheese, I like Cheddar very much’ (ii) a. Parlant de peix, ahir vaig menjar llenguado ‘Speaking about fish, yesterday I ate sole’ b. Pel que fa al formatge, m’agrada molt el Cheddar ‘As for cheese, I like Cheddar very much’ 5

Cf. also Zimmermann and Féry (2010) and Zimmermann and Onea (2011) as well as the articles contained in the thematic issue 121 of Lingua edited by Onea and Zimmermann for a more detailed exposition on focus marking across languages.

6

In the hierarchy of prosodic units, a phonological phrase is formed by one or more prosodic words. A prosodic word is formed by one or more syllables. One or more phonological phrases form an intonational phrase.

7

It must be remarked that some focus constituents can also appear in situ—i.e. in a postverbal position—in Hungarian (Szabolcsi 1981, 1994; Kiss 1998, 2010; Szendroi 2001). In situ and ex situ foci in Hungarian are argued to have different interpretations: an ex situ focus—but not an in situ focus—is interpreted exhaustively, which means that ex situ foci are equivalent to clefted constituents in English.

8

Alternatively, it could be assumed that the value of a parameter that cannot be set on the basis of positive evidence is the default value. The parameter is initially set to this default value and would only be switched to the non-default value on the basis of positive evidence (Rizzi 1978; Roberts and Holmberg 2010; Fasanella and Fortuny in press, among many others).

302

9

CONTEMPORARY LINGUISTIC PARAMETERS

Note that the optionality of a parameter cannot be accomplished by leaving the value of a parameter unspecified; if a parameter were left unspecified, because the questions expressed by parameters were not answered, then the learner would not be able to decide how the target language works in a particular aspect of grammar. Clearly, optionality and non-specification are different issues.

13 The functional structure of N* Asya Pereltsvaig Stanford University

This chapter is concerned with the systematic variation in the structure of the noun phrase in different languages. There are five main components to this variation (some of which are related). The first one concerns the presence, obligatory or optional, or absence of certain functional elements, such as articles, classifiers, number marking, and so on. The second one concerns the order of these functional elements with respect to each other and with respect to the noun. The third component concerns the expression of certain functional features, such as definiteness or case, by one or more exponents within a noun phrase (such as articles and other elements mentioned in connection with the first aspect of variation above). The fourth component concerns the expression of the argument structure of nouns. Finally, languages also differ as to the position and form of adjectives and other adjuncts. A wealth of analyses have been proposed in the syntactic literature to account for both specific morphosyntactic patterns in individual languages and the variation across languages; a number of functional projections, including DP (Determiner Phrase), NumP (Number/Numeral Phrase), ClP (Classifier Phrase), and others, have been proposed to accommodate various functional elements, as well as attributive modifiers (primarily, adjectives) and in some cases to account for the syntax of arguments inside noun phrases (particularly, of genitive phrases). For reasons of space, this chapter focuses chiefly on the first component of variation in the nominal domain: the presence or absence of various functional elements, with special attention given to articles and classifiers. Differences in word order of nounphrase-internal elements, such as demonstratives, numerals, adjectives, and the like, have been typically accounted for by deriving various orders from one “basic”, or underlying, order by applying either head movement or (remnant) phrasal movement, or both (see Shlonsky 2004; Cinque 2005; Pereltsvaig 2006a, 2006b; Alexiadou et al. 2007, among others). The analysis of double definiteness languages, enclitic definiteness marking, and other types of nominal concord depends on the functional architecture and movement types that one adopts (see Giusti 1994; Julien 2005). For a good summary of variation in the expression of argument structure, the reader is referred to Longobardi and Silvestri (2013).

304

CONTEMPORARY LINGUISTIC PARAMETERS

The rest of the chapter is organized as follows: in section 1, I describe the main aspects of morphosyntactic variation in the nominal domain that the rest of the chapter focuses on and outline two major approaches to such variation. Section 2 focuses on the approach that postulates only as much structure as is justified by the presence of overt functional elements (articles, classifiers) in a given language; section 3 considers the alternative approach that views syntactic structure as universal and places the variation in the lexical component: according to this view, languages differ not as to what structure they have but what morphosyntactic elements they have to realize the structure. Finally, section 4 reconsiders the two approaches in light of the distinction between macro- and micro-parameters.

13.1. Morphosyntactic variation in the nominal domain and the question of parametrization It is well known that languages differ as to what functional elements they employ in the nominal domain. One contrast involves languages, such as English or Italian, which have (overt) articles—whose role is to express (in)definiteness, determinacy, specificity, genericity, identifiability, referentiality, and so on (depending on one’s analysis)—and languages that do not have articles, such as Russian: (1) Il detective ha trovato un indizio. the detective has found a clue ‘The detective found a clue.’

[Italian]

(2) Detektiv našël uliku. detective.NOM found clue.ACC ‘The detective found a clue.’

[Russian]

Another contrast is between languages that have (numeral) classifiers and those that do not.1 English belongs to the latter type, while Mandarin and Japanese to the former type:2 (3) sān bĕn shū [Mandarin] three CL book ‘three books’ (4) ni-satsu-no hon two-CL-GEN book ‘two books’

[Japanese]

Additionally, although it appears that all languages have numerals, not all languages make obligatory use of number marking.3 For example, Japanese does not: the object in the following sentence can be interpreted as either singular or plural (as well as either definite or indefinite), depending on the context:



THE FUNCTIONAL STRUCTURE OF N

305

(5) Hiro-ga hon-o yon-da [Japanese] Hiro-NOM book-ACC read-PST ‘Hiro read {a book / some books / the book / the books}’ In this respect, Japanese contrasts with both English, where plurality must be marked overtly (though the singular form is more often than not morphologically unmarked; cf. book / books), and with languages such as Tatar, where plurality must be overtly marked on accusative-marked direct objects but not on objects that are not marked for case: (6) a. Marat alma-nı aša-dı Marat apple-ACC eat -PST ‘Marat ate the apple’ b. Marat alma-lar-nı aša-dı Marat apple-PL-ACC eat -PST ‘Marat ate the apples’ c. Marat alma aša-dı Marat apple eat-PST ‘Marat ate {an apple / some apples}’

[Tatar]

As can be seen from the Tatar example above, there are some non-trivial interactions of definiteness (or perhaps, specificity) with plurality, as well as with case. (The exact nature of the semantic distinction that is encoded via the presence or absence of the accusative marker is a matter of debate and will be left aside here; see Lyutikova and Pereltsvaig 2013, forthcoming.) Similarly, it has been shown by Cheng and Sybesma (1999) that classifiers interact with definiteness: in Mandarin a bare noun phrase can be interpreted as either indefinite or definite and a classified counterpart can only be interpreted as indefinite; in contrast, in Cantonese a bare noun phrase can only be interpreted as indefinite, while a classified counterpart can be interpreted as either definite or indefinite.4 Moreover, it has long been noted that classifiers interact with plurality: the existence of classifiers in a language correlates bidirectionally with the absence of plural inflection (see Borer 2005a: 94–5 and references cited therein). For instance, both Mandarin and Japanese, the prime examples of classifier languages, do not mark plurality. Borer also notes that in Armenian, where both plural marking and classifiers occur optionally, they never co-occur even though the two elements are not in morphophonological competition with each other, the classifier being prenominal and an independent morpheme and the plural being a suffix: (7) a. Yergu hovanoc uni-m two umbrella have-1SG b. Yergu had hovanoc uni-m. two CL umbrella have-1SG c. Yergu hovanoc-ner unim two umbrella-PL have-1SG d. *Yergu had hovanoc-ner unim two CL umbrella-pl have-1SG All: ‘I have two umbrellas’

[Armenian]

306

CONTEMPORARY LINGUISTIC PARAMETERS

Despite these intriguing interactions, the three types of functional elements— articles, number marking, and classifiers—are typically analyzed as generated in their dedicated functional projections: DP, NumP, and ClP, which constitute the skeleton of the extended noun phrase (cf. Grimshaw 1991), much as the TP and CP are taken to constitute the extended projection of the verb (see Wiltschko, this volume).5 The DP was first proposed by Abney (1987), chiefly to account for the structure of Saxon genitives (John’s car) in English and the possessor agreement in languages such as Turkish and Hungarian (see also Kornfilt 1984). It was followed by the introduction of NumP by Ritter (1991) to accommodate the plural marking in Hebrew, as well as to account for word order facts concerning Hebrew construct state and free genitive nominals. ClP was introduced around the same time in the literature on Chinese (see Tang 1990). Thus, the following structure for noun phrases has emerged (specifier positions are omitted):6 (8)

DP

D

NumP

Num ClP

Cl

NP N

The question that immediately arises in connection with this structure is whether all the functional projections are available in all languages, particularly whether certain projections are available in languages that lack the functional elements that are typically merged there. Do languages that lack articles have the DP projection? Do languages without (obligatory) number marking have the NumP? And do languages without classifiers have the ClP?7 In essence, these are questions about the nature of the relevant parameters: the claim that languages lacking certain functional elements lack the corresponding projections is tantamount to saying that the relevant parameters concern (narrow) syntax (or semantics, see discussion below). Because this type of analysis eschews functional projections that are filled with non-overt material, left unfilled, or remain inactive, I will refer to this approach as the WYSIWYG (What You See Is What You Get) approach. The alternative view takes functional architecture to be universal, which requires locating the relevant parameters elsewhere in the grammar, for instance, in the morphophonological component, PF, and the like; I will refer to this approach as the Universal Structure approach. Both approaches have been explored in connection with the DP and ClP since the late 1990s (see Progovac 1998; Chierchia 1998a, 1998b, inter alia); the rest of this chapter presents evidence that supports or contradicts each of the two approaches.



THE FUNCTIONAL STRUCTURE OF N

307

13.2. The WYSIWYG approach According to the WYSIWYG approach, languages that lack certain functional elements, such as articles, classifiers, or number marking, also lack the projections that typically host such elements. Main advocates of this approach are Chierchia (1998a, 1998b) and Bošković (2005 and later work). A similar view, particularly on the issue of DP in languages without articles, has also been adopted in Willim (1998, 2000), Baker (2003: 113), Mahajan (2003), Trenkic (2004), and elsewhere. In this section, however, I will focus on the proposals of Chierchia and Bošković. Crucially, adopting the WYSIWYG approach leads to one of two possible outcomes: either the isomorphic relationship between syntactic and semantic representations (aka Frege’s “compositionality principle”) must be abandoned, or if it is to be maintained, the semantic load of the relevant functional projection (e.g. referentiality in the case of DP, divisibility [cf. Borer 2005a: Chapter 4] in the case of ClP, etc.) must be placed elsewhere, typically on the noun itself. The proponents of the WYSIWYG approach by and large take the second route: Chierchia does so explicitly, while Bošković does so implicitly. Yet in either case, the relevant parameter is taken to be essentially a semantic one even if it is formulated as being in (narrow) syntax. In the following subsection, I shall begin with Chierchia’s explicitly semantic parameter.

13.2.1. Chierchia’s Nominal Mapping Parameter Chierchia (1998a, 1998b) explicitly argues against the widely accepted view, argued for in Longobardi (1994, 1996) and elsewhere, that DPs are cross-linguistically arguments, whereas bare NPs are always predicates.8 Under this “consensus view,” as Chierchia calls it, “each time one sees a bare NP argument, the category D must have been projected” (Chierchia 1998a: 343). Instead, Chierchia takes the “what you see is what you get” approach: if a language has no D-level elements, it has no DP projection, its load being placed on the noun itself. Specifically, languages differ as to whether their nouns (and consequently their NPs) can be mapped into arguments and/or into predicates.9 Accordingly, he distinguishes three main types of languages.10 The first type of language, [−arg, +pred], is the one in which Longobardi’s view that DPs are arguments and NPs are predicates applies; Romance languages fall into this category. It is possible that a language of this type would have a null D, but like other instances of phonologically null structure it would be subject to licensing conditions. Chierchia uses this possibility to account for the grammaticality of bare mass / plural noun arguments in the postverbal position in Italian and Spanish, in contrast to the ungrammaticality of the same phrases in the preverbal position: in the postverbal position, the null D in such phrases is governed by a lexical head—the verb. (9) a. *[DP Ø [NP Acqua]] viene giù dalle colline water comes down from-the hills b. Viene giù [DP Ø [NP acqua ]] dalle colline comes down water from-the hills Both: ‘Water comes down from the hills’11

[Italian]

308

CONTEMPORARY LINGUISTIC PARAMETERS

The second type of language in Chierchia’s classification is [+arg, −pred]. In such a language, nouns themselves (and thus NPs) are mapped into arguments and not predicates, which means that all nouns are, in a sense, mass (the denotation of such nouns is discussed in detail in Chierchia 1998b). According to Chierchia, there are several immediate consequences to such mapping. First, “plural marking, of the kind familiar from many western languages, will be absent, since with every noun being mass the function PL will have no argument for which it is defined” (Chierchia 1998a: 353). Second, “numerals will not be able to combine directly with nouns: a classifier will be necessary to individuate an appropriate counting level” (ibid., 353–4). Thus, the correlation between a classifier system and the lack of plural marking, noted in section 1 above, follows nicely from this analysis. Chinese and Japanese are languages in this category. Finally, the third major type of language is [+arg, +pred], where nouns can freely be either predicative or argumental. A language of this type has the mass/count distinction: mass nouns map into arguments and count nouns map into predicates. English and most other Germanic languages fall into this category. The crucial difference between a [+arg, +pred] language, such as English, and a [+arg, −pred] language, such as Italian, is that only in the former type of language can mass and plural nouns occur as bare arguments, whereas (singular) count nouns require a determiner in the argument position. Witness the following contrast between English and Italian: (10) a. *Dinosaur is extinct b. Dinosaurs are extinct

[English]

(11) a. *Dinosauro Dinosaur b. *Dinosauri Dinosaur

[Italian]

è estinto are extinct sono estinti are extinct

Finally, a subtype of [+arg, +pred] would be a language with the same type assignment as English but one which happens to lack (overt) articles. According to Chierchia, this is exactly what Russian is: it has the mass / count distinction, which manifests itself in the same form as in English, but bare singular count nouns can occur in argument position, with either definite or indefinite interpretation, and even possibly a kind reference. Thus, the counterpart of the English (10a) is grammatical in Russian, except that the sentence-initial position correlates with definiteness; to receive an indefinite interpretation, the noun phrase is typically placed postverbally: (12) a. Sobaka ljubit igrat’ [Russian] dog loves to-play ‘The dog loves to play’ b. Vo dvore igraet sobaka in yard plays dog ‘A dog is playing outside / There is a dog playing outside’ c. Sobaka — obyčnoe životnoe dog typical animal ‘The dog is a common animal’



THE FUNCTIONAL STRUCTURE OF N

309

Chierchia’s Nominal Mapping Parameter is appealing in that it accounts for several types of languages, that is, two types of languages with articles and two types of languages without articles, exemplified here with English and Italian, and Russian and Chinese, respectively. It also allows for an easy account of language acquisition: a child postulates [−pred] if her language has classifiers (and [+pred] if there are no classifiers), and [+arg] if bare mass/plural nouns can appear preverbally (and [−arg] otherwise). Nonetheless, there are a number of problems with this analysis, particularly as regards the analysis of classifier, or [+arg, −pred], languages.12 The first problem concerns Chierchia’s assumption that all nouns are (lexically marked as) mass in Chinese, which appears to be invalid. Cheng and Sybesma (1999: 516) note that “just like some nouns in languages like English, some nouns in Chinese are inherently semantically partitioned into discrete units […] and others are not” and further claim that “this cognitive reality is represented in the semantics of the noun in all these languages.” Consequently, they distinguish two types of “classifiers”: true (or count-) classifiers, which appear with count nouns, as in (13a), and “mass-classifiers” (or “massifiers”), which appear with mass nouns, as in (13b).13 (13) a. sān bĕn shū [Mandarin] three CL book ‘three books’ b. sān bàng ròu three CL-pounds meat ‘three pounds of meat’ Count-classifiers such as bĕn differ from massifiers such as bang “pound” in two important ways. First, massifiers allow the optional occurrence of attributive modifier marker de (see Rubin 1994), which is otherwise typically attested at phrasal boundaries internal to the DP; count-classifiers do not co-occur with de: (14) a. sān bàng (de) ròu three CL-pound DE meat ‘three pounds of meat’ b. shí zhāng (*de) zhuōzi ten CL DE table ‘ten tables’

[Mandarin]

Second, certain adjectives can modify massifiers but not count-classifiers: (15) a. yì dà zhāng zhĭ [Mandarin] one big CL-sheet paper ‘one large sheet of paper’ b. *yì dà zhī gŏu one big CL dog Intended: ‘one big dog’

310

CONTEMPORARY LINGUISTIC PARAMETERS

It is, therefore, plausible that massifiers in Chinese are lexical heads of their own nominal projections rather than merely functional heads in the extended projection of a noun. A similar proposal has been put forward for massifiers in English in Borer (2005a), who suggested that “the head of expressions such as three drops of blood is arguably drops, rather than blood, and that of blood is arguably a type of complement.” Cheng and Sybesma’s conclusion that the mass / count distinction is represented in the semantics of the noun in both English and Chinese (though in Chinese this distinction is also reflected at the functional level of the classifier) challenges the core of Chierchia’s analysis, which relies on the idea that in contrast to English, where the mass/count distinction is reflected in the semantics of the noun, in Chinese all nouns are mass. The second problem for Chierchia’s analysis lies with the fact that the distribution of bare nouns in Chinese is not as free as he makes it seem. As mentioned above, in languages with articles, bare nominals can occur in argument positions only if they are mass / plural (already in English-type languages; cf. 10), and (in Italian-type languages; cf. 9) if they additionally occur in lexically governed positions. In Chierchia’s system, classifier languages are expected to not have any such restrictions on bare nominals. However, this expectation is not met. As Cheng and Sybesma (1999) show, Mandarin disallows an indefinite reading for bare NPs in preverbal position, and Cantonese allows preverbal bare noun phrases only under generic interpretation. (16) a. Gŏu yào guò mǎl‫‏‬ù [Mandarin] dog want cross road ‘The dog wants to cross the road’, #‘A dog wants to cross the road’ b. * Gau2 soeng2 gwo3 maa5lou6 [Cantonese] dog want cross road ‘{A/The} dog wants to cross the road’ Bare noun phrases in postverbal position are not subject to such restrictions: in both Mandarin and Cantonese, such noun phrases can be interpreted as indefinite (or generic):14 (17) a. húfēi măi shū qù le [Mandarin] Hufei buy book go SFP ‘Hufei went to buy a book/books’ b. Wu4fei1 heoi3 maai5 syu1 [Cantonese] Wufei go buy book ‘Wufei went to buy a book/books’ As Borer (2005a: 90) points out, “under the plausible assumption that weak indefinites involve a null, unbound D, but definites and generics do not, these facts are directly explained, but only provided that D does project in Chinese.” Such effects cast doubt on Chierchia’s claim that in Mandarin, Cantonese, and other classifier languages there are indeed argumental bare NPs. Moreover, unlike classifier languages, which are clearly definable in semantic terms in Chierchia’s system, the distinction between languages with articles and



THE FUNCTIONAL STRUCTURE OF N

311

without articles is not as clear-cut: some languages with articles belong to the [−arg, +pred] type (Romance languages), while others belong to the [+arg, +pred] type (Germanic languages). Similarly, languages without articles may belong to either the [+arg, −pred] type (Chinese, Japanese) or the [+arg, +pred] type (Russian). Thus, the presence or absence of articles does not correlate with either value of [±pred] or [±arg] feature. Furthermore, the difference between English (with articles) and Russian (without articles) in this system is described not in terms of the semantic parameter but in terms of morphophonology: English has overt articles, while Russian does not. This is a departure from a pure WYSIWYG approach in the direction of the Universal Structure approach. But before we turn our attention to that alternative approach, I shall consider another analysis couched within the WYSIWYG view.

13.2.2. Bošković’s Parameterized DP Hypothesis The analysis proposed by Bošković (2005, 2009, 2010; see also Bošković and Gajewski 2011; Bošković and Şener 2014) is similar in spirit, though not in scope or in details, to proposal by Chierchia (1998a), discussed in the previous section. According to Bošković, nominals in article-less languages are bare NPs, with no DP ever projected.15 Implicit in this analysis is the claim that nouns (and NPs) in languages without articles are able to serve as arguments, with no help from the D°. Under this view, a noun in Russian denotes a “determiner + noun” meaning. Simply put, the meaning of sobaka “dog” in Russian is equivalent to the meaning of the dog (or a dog) in English. It follows from this—although Bošković does not discuss this explicitly—that the meaning of the noun sobaka “dog” in Russian is different from the meaning of dog in English. Under this view, demonstratives are analyzed syntactically as attributive modifiers, adjoined to NP but semantically as of type (unlike typical attributive adjectival modifiers, which are of type ). Ordering restrictions between demonstratives and attributive adjectives (i.e. that demonstratives must precede attributive adjectives, as illustrated below) are explained in this system by assuming that semantic composition requires adjectives to be composed before demonstratives. (18) a. ètot interesnyj fakt this interesting fact b. * interesnyj ètot fakt interesting this fact Both:  ‘this interesting fact’

[Russian]

The principal argument for Bošković’s (2008, 2010) analysis comes from the range of purported correlations between the lack of articles, on the one hand, and various other syntactic (or, in some cases, semantic) phenomena, on the other hand. Specifically, Bošković claims that the lack of articles correlates with: (a) the availability of the so-called Left Branch Extraction (LBE, or in more descriptive terms “discontinuous noun phrases”), (b) the lack of superiority effects in multiple wh-interrogatives, (c) the lack of the exhaustivity presupposition in

312

CONTEMPORARY LINGUISTIC PARAMETERS

possessive constructions, (d) the unavailability of two adnominal genitives, (e) the possibility of adjunct extraction from NP, (f) the possibility of scrambling, (g) the availability of second position clitics, (h) the impossibility of negative raising, (i) the unavailability of clitic doubling, (j) the impossibility of the majority superlative reading, (k) the lack of inverse quantifier scope, and (l) the lack of sequence of tenses. Under Bošković’s view, the alleged correlations are tied together into a macro-parameter that accounts for several surface typological properties of a language: property X (i.e. the lack of articles) in a given language implies that certain other syntactic (or semantic) properties Y, Z, etc.; the presence of articles implies the reverse typological picture (i.e. the absence of properties Y, Z, etc.). This makes the proposed DP parameter much like, for example, the pro-drop parameter, which also accounts for the correlation between several seemingly unrelated surface phenomena. As impressive as the list of purported correlations is, it is riddled with problems. First, some languages, such as Lithuanian, do not fit neatly within the DP/NP dichotomy. Although Lithuanian has no free-morpheme articles of the English or Italian type, it does have a definiteness marker which takes the form of a bound morpheme that attaches (optionally) to the highest element in a noun phrase, be it an adjective, a participle, an ordinal numeral, or the noun itself in the absence of premodifiers: (19) a. balt-as šun-uk-as [Lithuanian] white-M.SG.NOM dog-DIM-M.SG.NOM ‘a/the white doggie’ b. balt-as-is šun-uk-as white-M.SG.NOM-DEF.M.SG.NOM dog-DIM-M.SG.NOM ‘the white doggie’ While traditional grammars analyze this morpheme as part of the adjectival inflection (or “long form morphology”, comparable to that of Serbo-Croatian, Slovenian, and Russian), Van Hofwegen (2013) argues that it should be analyzed akin to the Bulgarian definiteness marker -ta, which Bošković labels as an article. Consequently, Lithuanian is expected to pattern with Bulgarian (as well as English and other languages with articles) rather than with article-less languages. The problem is that it does not. What is worse, Lithuanian patterns in some ways with languages with articles, and in other ways with languages without articles, as shown in Table 13.1:



THE FUNCTIONAL STRUCTURE OF N

313

Table 13.1  Some of Bošković’s NP/DP generalizations applied to Lithuanian Generalization

languages without articles

Lithuanian languages with articles

Left branch extraction

yes

yes

no

Adjunct extraction from NP

yes

yes

no

Scrambling

yes

yes

no

Superiority in multiple wh-fronting

no

no

yes

Exhaustivity of possessors

no

no

yes

Adnominal genitive

no

yes

yes

Majority superlative reading

no

yes

yes

(adapted from Van Hofwegen 2013: 16)

Let’s consider five of these putative correlations—LBE, superiority, exhaustivity, adnominal genitive, and majority superlative reading—more closely; as we shall see in the remainder of this section, these five correlations either do not hold, or can be accounted in some alternative way, without a recourse to the distinction between languages that project a DP (i.e. languages with articles) and those that do not (i.e. languages without articles). These facts cast serious doubt on Bošković’s version of the WYSIWYG approach to nominal structure.16 The correlation between LBE and the lack of articles has been noted at least as early as Corver (1992): both Serbo-Croatian and Russian lack articles and allow discontinuous noun phrases, whereas English, which has articles, does not allow such splitting (for ease of reference, here and below, discontinuous elements of a noun phrase are boldfaced):17 (20) a. Dorogie on kupil tufli expensive.ACC he bought shoes.ACC ‘It is expensive shoes that he bought’ b. *Expensive he bought shoes

[Russian] [English]

Bošković’s explanation is based on the idea that the DP in languages with articles constitutes a barrier for extraction out of the noun phrase; in article-less languages, which according to him lack the DP, this barrier does not exist, making LBE possible. But a closer consideration of this phenomenon reveals two problems. First, the correlation does not hold with respect to Lithuanian, which allows discontinuous noun phrases despite having a definite article (cf. Van Hofwegen 2013 for details): (21) Brangi-ą jis mat-ė mašin-ą [Lithuanian] expensive-F.SG.ACC he see-PST car-F.SG.ACC ‘He saw an EXPENSIVE car’

314

CONTEMPORARY LINGUISTIC PARAMETERS

Second, the correlation proves spurious in that the availability of LBE has nothing to do with the lack of the DP projection. Note that the DP-as-barrier explanation works only if discontinuous noun phrases indeed arise as a result of extraction of prenominal elements out of a noun phrase. However, there are at least two good reasons to believe that this is not the case: (a) discontinuous elements of a noun phrase need not be constituents, and (b) noun phrases that constitute well-established islands can be discontinuous. The first problem is illustrated by (attested) examples where complex nonconstituent strings are separated from the noun. In (22a) the non-clitic preposition protiv “against” and the adjective do not form a constituent. Nor is the sequence of an emphatic demonstrative vot èti “these here,” an adjective and what is normally a postnominal PP modifier form a constituent in (22b). (22) a. Protiv sovetskoj on vystupal vlasti [Russian] against Soviet he demonstrated regime ‘It is against the Soviet regime that he demonstrated’ b. Ja prosto probovala vot èti češskie s supinatorami pokupat’ tufli I simply tried here those Czech with arch-supports to-buy shoes ‘I simply tried to buy those here Czech shoes with arch-supports’ One could propose a more complex extraction analysis, whereby the noun (e.g. vlasti “regime,” tufli “shoes”) first vacates the noun phrase, followed by the movement of the remnant; however, it is not the case that the second (right-side) element of a discontinuous noun phrase needs to be a constituent either, as can be seen from the following examples: (23) a. Nezlobivyj u nego soveršenno xarakter [Russian] kindhearted to him absolutely disposition ‘He is an absolutely kindhearted disposition’ b. Odna očen’ est’ èlegantnaja rubaška u Peti one very there-is elegant shirt to Petja ‘Petya has one VERY elegant shirt’ The second problem for the extraction analysis of discontinuous noun phrases is that the presumed extraction can cross island boundaries. For instance, complements of comparatives in Russian do not allow complement extraction, but can be split by LBE: (24) a. *[Po sintaksisu]i ona byla starše [professora ti] [Russian] of syntax she was older professor.GEN Intended: ‘She was older than the professor of SYNTAX’ b. Kak podumaju, ved’ ona let na pjat’ nas starše vot nynešnix How I-will-think EMPH she years on five us older behold present-day byla, kogda my s nej poznakomilis’ was when we with her met ‘How odd: she was about 5 years older than us today, when I met her’



THE FUNCTIONAL STRUCTURE OF N

315

Pereltsvaig (2007b, 2008, 2013) argued for an alternative analysis of LBE, not in terms of extraction but as an instance of copy-and-delete (see also Fanselow and Cavar 2002): first, a legitimate movement of the whole nominal, such as scrambling, creates a second copy, and then parts of two copies are deleted: (25) [Dorogie tufli] on kupil [dorogie tufli] expensive.ACC shoes.ACC he bought expensive.ACC shoes.ACC ‘It is expensive shoes that he bought’ Under this analysis, it is not clear what role a DP could have in preventing a partial deletion of multiple copies. Pereltsvaig (2007b, 2008) shows that the possibility of LBE depends crucially on the availability of N-ellipsis and overt case marking. In fact, Bulgarian and Macedonian—the two Slavic languages that do not allow discontinuous noun phrases—also retain merely remnants of the old declension system: (26) *Novata ja prodade Petko kola [Macedonian] new has sold Petko car Intended: ‘It is a new car that Petko sold’ English and French also lack overt case marking and hence lack discontinuous noun phrases, of the Russian type (on the differences between the French combien-splits and the Slavic-style LBE, see Pereltsvaig 2008). Furthermore, even in languages that allow discontinuous noun phrases, case marking plays a decisive role as to when such splitting is or is not possible. For example, in Russian the first part of a complex masculine foreign proper name need not be overtly case-marked and can appear in the default nominative instead (27a), but in order for a discontinuous proper name to be grammatical, both parts of the name must be overtly case-marked (27b).18 (27) a. {Džejms / Džejmsa} Bonda ona obožaet James.nom / James.acc Bond.acc she adores ‘As for James Bond, she adores him’ b. {*Džejms / OKDžejmsa} ona obožaet Bonda James.nom / James.acc she adores Bond.acc ‘It is James Bond that she adores’

[Russian]

The second correlation that proves problematic on closer inspection involves superiority effects in multiple wh-questions. According to Bošković, languages with articles, such as English or Bulgarian, exhibit superiority effects, resulting in grammaticality contrast between Who brought what to the party? and *What did who bring to the party? (the latter is acceptable only in D-linking contexts; see Pesetsky 1987). Languages without articles, such as Serbo-Croatian, are claimed to be free of superiority effects (cf. Boškovic´ 1997):19 (28) a. Koj kogo e vidjal? who whom is seen b. *Kogo koj e vidjal? whom who is seen Both: ‘Who has seen whom?’

[Bulgarian]

316

(29) a. Ko who b. Koga whom Both:

CONTEMPORARY LINGUISTIC PARAMETERS

je koga video? is whom seen je ko video? is who seen ‘Who has seen whom?’

[Serbo-Croatian]

Other multiple wh-fronting Slavic languages—Slovenian, Czech, Polish, Ukrainian, and Russian—are also said to not show superiority effects. However, the facts, particularly as regards Russian, are not incontrovertible here: as noted in Grebenyova (2004: 179), while superiority does not arise with who/who combinations, it does with who/what combinations (setting aside the D-linking contexts):20 (30) a. Kto kogo priglasil na užin? who whom invited to dinner b. Kogo kto priglasil na užin? whom who invited to dinner Both: ‘Who invited whom to the dinner?’ c. Kto čto kupil? who what.ACC bought d. *Čto kto kupil? what.ACC who bought Both: ‘Who bought what?’

[Russian]

Another problematic correlation concerns the exhaustivity presupposition in possessives. Bošković (2010) asserts that “possessors may induce an exhaustivity presupposition only in DP languages”; for example, in English John’s three sweaters presupposes that John has exactly three sweaters, while corresponding examples in languages without articles are said to not trigger the same presupposition. The original source for this generalization is Partee’s (2006) discussion of the contrast between English possessives (which do trigger the exhaustivity presupposition) and their counterparts in Mandarin Chinese (which do not). Partee also notes in passing that Russian patterns with Mandarin Chinese rather than English in this respect (pp. 271–2). Bošković (2010) extends this generalization to include such languages as Serbo-Croatian, Turkish, Japanese, and Korean, which lack both articles and the exhaustivity presupposition, and Spanish, Brazilian Portuguese, Italian, Hebrew, and Dutch, which have both articles and the exhaustivity presupposition. However, additional cross-linguistic facts defy this generalization: Lithuanian unexpectedly patterns with Serbo-Croatian in not triggering the exhaustivity presupposition (for examples, see Van Hofwegen 2013), while Russian possessives actually do trigger the exhaustivity presupposition, but only if they occur high enough in the structure, above cardinality expressions, which is expected if the exhaustivity presupposition is associated with the DP projection.21 (31) a. Diminy pjat’ knig [Russian] Dima’s.NOM five books ‘Dima’s five books’ (i.e. Dima only has five books) b. pjat’ Diminyx knig five Dima’s.GEN books ‘five of Dima’s books’ (i.e. Dima may have more than five books)



THE FUNCTIONAL STRUCTURE OF N

317

Next, consider the putative correlation between the lack of articles and the impossibility of two adnominal genitives (first noticed by Willim 2000; see also Bošković 2008): English, German, Dutch, Arabic, and Catalan—all of which have articles—allow two genitive arguments, internal and external, whereas article-less languages, such as Polish, Czech, Russian, and Latin, disallow such configurations. Bošković (2008) also adds Icelandic, Swedish, Portuguese, Spanish, French, Basque, Greek, Hebrew, Maltese, Welsh, Maori, Samoan, as well as Bulgarian and Macedonian—the two Slavic languages with articles—to the list of languages that have both articles and two genitives; and Serbo-Croatian, Slovenian, Ukrainian, Chinese, Quechua, and Turkish to the list of article-less languages that disallow two genitives. (32) a. Hannibals Eroberung Roms Hannibal.GEN conquest Rome.GEN b. *zaxvat Gannibala Rima conquest Hannibal.GEN Rome.GEN Both: ‘Hannibal’s conquest of Rome’

[German] [Russian]

However, Gianollo (2005), Giusti and Oniga (2007), and Giusti and Iovino (2011) observed that Latin did allow two adnominal genitives: a subject genitive in prenominal position and an object genitive in postnominal position, making the grammatical Latin example below completely parallel to the German example in (32a) above: (33) veteribus Helvetiorum iniuriis [Latin] old.ABL.PL Helvetii.GEN.PL injuries.ABL.PL populi Romani people.GEN.SG Roman.GEN.SG ‘through the old offences by the Helvetii to the Roman people’ (Caes. Gall. 1,30,2; cited in Giusti and Iovino 2011, their (10b)) Moreover, Willim’s (2000) and Bošković’s (2008) explanation for the contrast is flawed as well. They proposed that in languages with articles two licensors of genitive are available: the noun itself and the D°. In contrast, in languages without articles, the DP is said to be absent and therefore only one genitive can be licensed. Russian facts, once again, throw a wrench into the works: the unavailability of the second adnominal genitive applies only to process (or event) nominals, as in (32b) above and (34b) below. Russian non-process nominals, as in (34a), allow two genitives (examples adapted from Engelhardt and Trugman 1998): (34) a. analiz [poèmy Puškina ] [literaturoveda Pupkina] analysis poem.GEN Pushkin.GEN literary-critic.GEN Pupkin.GEN ‘a literary critic Pupkin’s analysis of Pushkin’s poem’ b. *analizirovanie [poèmy Puškina ] [literaturoveda Pupkina] analyzing poem.GEN Pushkin.GEN literary-critic.GEN Pupkin. GEN Intended: ‘analyzing Pushkin’s poem by a literary critic Pupkin’

318

CONTEMPORARY LINGUISTIC PARAMETERS

If the DP is indeed absent in Russian, the availability of two genitives in (34a) remains mysterious.22 Engelhardt and Trugman’s (1998) analysis relies crucially on the availability of DP in Russian. They claim that Russian has the DP projection, which licenses one of the two genitives in both process and non-process nominals—just as Willim (2000) and Bošković (2008) do for languages with articles. The difference between the two types of nominals, according to Engelhardt and Trugman, lies not in the presence or absence of the DP, but in the availability of another, lexical case licensor. Specifically, with non-process nominals like (34a) the genitive on the external argument literaturoveda Pupkina “literary-critic Pupkin’s” is licensed by D°, and the genitive on the internal argument poèmy Puškina “Pushkin’s poem” is licensed by the N°. In contrast, with process nominals like (34b) the noun cannot license case, due to the fact that the noun is derived via incorporation of the verb (cf. Borer 1993), which has been passivized. Crucially for the present argument, Engelhardt and Trugman’s (1998) analysis (see also Engelhardt and Trugman 2000) requires that nominals even in article-less languages sometimes be projected as DPs, though other nominals may be less than DPs.23 Finally, let’s briefly consider the alleged correlation between the lack of articles and the lack of majority superlative reading. This generalization is also at odds with the Russian facts: despite being a language without articles, Russian does allow the majority superlative reading, which is in fact the most salient reading of the following naturally occurring example: (35) Počemu do six por bol’šinstvo ljudej pjut koka-kolu, [Russian] why to this day most people drink Coca-Cola nesmotrja na to, čto vse davno znajut, despite that all long-ago know čto ona očen’ vrednaja? that it very harmful ‘Why does the majority of people drink Coca-Cola, although everybody has known for a long time that it is very harmful?’ (i.e. why does more than half the people drink Coca-Cola, not why more people drink Coca-Cola than any other drink)24 To recap, Bošković’s approach to languages with and without articles, which places the burden of explanation in syntax (and consequently, in semantics, as long as an isomorphic relationship between syntactic and semantic representations is assumed), is very problematic as many of the alleged correlations do not hold or must be explained in an alternative way. Furthermore, a closer examination of some of these generalizations, for example the adnominal genitive correlation, suggest that postulating the DP projection in languages without articles—contrary to the WYSIWYG approach—may actually be advantageous. In the next section, we turn to examining in more detail the alternative approach, which takes the functional structure to be universal.



THE FUNCTIONAL STRUCTURE OF N

319

13.3. Universal Structure approach Contrary to the WYSIWYG approach, which allows no DP in languages without articles and no ClP in languages without classifiers, the Universal Structure view takes the functional structure to be the same across languages. The main advocate for this view is Borer (2005a), who considered the three functional projections in the structure in (8) above: DP, NumP (Borer calls it #P), and ClP (see also the work in the cartographic framework: Cinque 2002, 2010). There is also a significant body of literature either explicitly arguing for the presence of the DP projection in languages without articles, especially Slavic and Turkic ones, or assuming it; see Progovac 1998; Rappaport 1998, 2001a, 2001b, 2004; Pereltsvaig 2001, 2006c, 2007a, 2013; Rutkowski 2002, 2003, 2007, in press; Franks and Pereltsvaig 2004; Rutkowski and Maliszewska 2007; among others, for Slavic languages; Kornfilt 1984; Lyutikova and Pereltsvaig 2013, forthcoming, 2014; among others, for Turkic languages. According to the strong version of the Universal Structure approach, languages that lack certain overt functional elements (articles or classifiers) either have non-overt ones or to use the relevant structure for other elements (e.g. DP for demonstratives, ClP for number marking, as in Borer 2005a). A weak version of this approach contends that an article-less language may not always project a DP, making some noun phrases in such a language DPs and others incompletely projected, or Small Nominals (see Pereltsvaig 2006c). The same may also be true, however, of languages with articles as well. Consequently, arguments for the Universal Structure approach come from two main empirical areas: (a) parallels between languages with and without articles/classifiers and (b) language-internal contrasts between noun phrases in different structural contexts that can be best accounted for by assuming different amount of functional structure. Let’s consider each type of argument in turn. Languages with and without classifiers may seem radically different syntactically (and semantically, as argued by Chierchia 1998a), yet Borer (2005a) claimed that languages lacking classifiers, such as English, must be assumed to have the ClP projection. While she agrees with Chierchia that all nouns in Chinese are mass, she argues that this is not a fact peculiar to classifier languages, but rather one that generalizes across all languages. For her, all nouns in English are mass, just as they are in Chinese. The divisional function is thus never performed by the nouns themselves; instead it is associated with the ClP. Since English lacks (count) classifiers, this projection is filled in English by plural morphology, Borer contends. She goes as far as to claim that “no parameter is in evidence here” (Borer 2005a: 93), in reference to Chierchia’s Nominal Mapping Parameter. More precisely, for her cross-linguistic variation involves not the semantics of the nouns (or more generally lexical items), nor the structure which nouns and functional elements are plugged into, but the morphosyntactic expression of the universal structure, which is projected regardless of the functional elements that a given language makes use of. This is the quintessence of the Universal Structure approach. One of Borer’s key arguments for treating morphologically plural forms as divisional rather than semantically plural (see Borer 2005a: Chapter 4, for a detailed

320

CONTEMPORARY LINGUISTIC PARAMETERS

technical discussion) comes from the fact that decimal numbers, including those less than 1 and 1.0, must be accompanied by plural inflection in English: 1.5 apples, 0.2 apples, and 1.0 apples rather than *1.5 apple, *0.2 apple, and *1.0 apple. Moreover, the incompatibility of plural inflection with an overt classifier in languages that have both, such as Armenian, illustrated in (7d) repeated below, is taken as a further argument for placing plural morphology in ClP: (36) *Yergu had hovanoc-ner unim two CL umbrella-PL have-1SG Intended: ‘I have two umbrellas’

[=(7d)]

[Armenian]

However, the same treatment of plural morphology can hardly be extended to all languages that lack classifiers: for example, the Russian counterparts of the English examples above contain a singular form of the noun: (37) a. odna celaja pjat’ desjatyx jablok-a one whole five tenths apple-SG.GEN ‘1.5 apples’ b. nol’ celyx dve desjatyx jablok-a zero whole two tenths apple-SG.GEN ‘0.2 apples’ c. odna celaja nol’ desjatyx jablok-a one whole zero tenths apple-SG.GEN ‘1.0 apples’

[Russian]

If, as the examples above suggest, the plural marking does not fulfill the divisional function in Russian, as Borer proposed for English, this leaves a possibility of some other elements occupying the ClP in Russian. As it turns out, although Russian does not require classifiers in the presence of numerals, like Chinese or Japanese do, it does have three optionally-used numeral classifiers: štuk “items” for objects, golov “heads” for animals (especially livestock), and čelovek “persons” for humans (cf. also the archaic classifier for humans, duš “souls”): (38) a. sto (štuk) karandašej [Russian] hundred item.GEN.PL pencils.GEN.PL ‘100 pencils’ b. sto (golov) ovec hundred head.GEN.PL sheep.GEN.PL ‘100 sheep’ c. sto (čelovek) prezidentov kompanij hundred person.GEN.PL president.GEN.PL company.GEN.PL ‘100 presidents of companies’ The earliest analysis of these items as numeral classifiers is Sussex (1976); his arguments are more convincing than those of Aikhenvald (2000: 115–16), who denies them the classifier status.25 First, as expected of classifiers if the structure in (8) above is assumed for Russian, these words normally occupy the position between



THE FUNCTIONAL STRUCTURE OF N

321

the numeral and the noun. Second, although the complement of štuk “items” and its ilk appears in the genitive, just as complements of nouns normally do (cf. 38c), this is not a sufficient evidence for the lexical nature of štuk “items” and the like. Pereltsvaig (2007a, 2013) argues that numerals, which similarly take genitive complements in the absence of štuk “items,” are not nouns either. Third, Aikhenvald’s (2000: 116) argument that the words like štuk “items” in Russian are too few to be considered a category of its own likewise fails: in this respect, Russian is exactly like Akatek, a bona fide classifier language, in that it has only three numeral classifiers: one each for objects, animals, and humans (see fn. 1). The fourth argument for analyzing these words as classifiers concerns the form of the classifier for humans, čelovek: this special count form is used only when it cooccurs with a numeral; in the absence of a numeral, even if a quantifier like mnogo “many” is present, the plural form of čelovek “person, man” is the suppletive form ljudi “people”: (39) a. Ja videl pjat’ čelovek [Russian] I saw five person.PL.COUNT ‘I saw five people’ b. Ja videl (mnogo) ljudej I saw many person.PL ‘I saw (many) people’ c. *Ja videl (mnogo) čelovek I saw many person.PL.COUNT Intended: ‘I saw (many) people’ Finally, štuk “items,” golov “head,” and čelovek “persons” used in this classifier function cannot take any modifiers (cf. Yadroff 1999: 151–2), which buttresses the argument that they are functional rather than lexical items (cf. the discussion surrounding 15 above): (40) a. *desjat’ ten b. *pjat’ five

vesëlyx happy otdel’nyx separate

čelovek persons štuk items

našix oficerov our officers nexorošix slov obscene words

[Russian]

Going back to Borer’s (2005a: 116) analysis of plural morphemes as classifiers, she admits that it cannot be extended to all languages and proposes that in addition to classifiers and plural morphology, the divisional function may be fulfilled by numerals and quantifiers which are merged in ClP and then move to #P, where they fulfill their counting function. This is the case in Hungarian, she claims, where numerals co-occur with singular (unmarked) form of the noun rather than the plural: (41) a. a két kalap (*-ok) the two hat (*PL) ‘the two hats’ b. a kalap -ok the hat -PL ‘the hats’

[Hungarian]

322

CONTEMPORARY LINGUISTIC PARAMETERS

As Borer (2005a: 116) noted, similar facts hold for Turkish (see also Ortman 2000); Tatar facts are essentially the same as well: (42) a. ike kitap (*-lar) two book (*PL) ‘(the) two books’ b. kitap -lar book -PL ‘(the) books’

[Tatar]

Note that Hungarian and Tatar data above illustrate a further parallelism between languages with and without articles: regardless of whether a given languages has articles, its numerals may combine with either plural or singular form of the noun, as summarized in Table 13.2:

Table 13.2  Non-correlation of the noun form used with numerals and the availability of article Form of the noun with numerals

languages with articles

languages without articles

singular

Hungarian

Tatar

plural

English

Russian

Other instances of parallelism between languages with and without articles have been explored in the literature that argued for the presence of DP in the latter type of language. For example, Progovac (1998) showed that noun ⁄ pronoun asymmetries, where the pronouns precede—and nouns (including proper nouns) follow—certain intensifying adjectives (e.g. samu “alone”), obtain in SerboCroatian, a language lacking articles. Progovac accounted for this noun / pronoun asymmetry by raising pronouns but not nouns to D. Similarly, indefinite pronouns precede modifying adjectives, both in English (with articles) and Russian (without articles); the following Russian examples are completely parallel to their English translations: (43) a. novye knigi new books ‘new books’ b. čto-to novoe something new ‘something new’

[Russian]

Advocates of the Universal Structure approach typically adopt Longobardi’s (1994) view that that a nominal can be an argument only if it is projected as a DP. An expectation that arises from this hypothesis is that non-argument nominals are not required to be DPs, regardless of whether the language has articles. While



THE FUNCTIONAL STRUCTURE OF N

323

Longobardi has shown this to be true for languages with articles, such as English or Italian, Pereltsvaig (2001, 2006c, 2007c) showed that predicative nominals in Russian are bare NPs rather than DPs (see also Lutikova 2010 on Russian vocatives being smaller than a DP). Another parallelism between English and languages without articles, such as Polish, concerns the impossibility of extracting prenominal adjuncts, discussed in detail in Rappaport (2004). Contrary to the WYSIWYG approach, which predicts no blocking effect in a language lacking articles (and thus, by hypothesis, lacking the DP projection), extraction of an adjunct—whether a PP or a preposition-less noun phrase—is ungrammatical. Again, the Polish examples below are completely parallel to their English translations: (44) a. *Z czym postrzelili przed domem [chlopca ]? [Polish] with what shot in.front.of house boy ‘*With whati did they shoot [a boy ti] in front of the house?’ b. *Kiedy on opisal [demonstracje ] dla gazety. when he described demonstration for newspaper ‘*Wheni did he describe [a demonstration ti] for the newspaper?’ A different kind of argument for the Universal Structure approach comes from instances of language-internal contrasts that can be (arguably) best accounted for by assuming a different amount of functional structure to be projected in each case. Due to space limitations, only two examples of such contrasts will be considered here, both from languages without articles that correlate with the presence or absence of DP-level material. The first example involves extraction out of certain nominals in Russian, more precisely out of infinitival complements (bracketed in the examples below) inside deverbal nominals (in the following examples, headed by the noun rešenie “decision”). As Lyutikova (2010) observed, such extraction is possible when the nominal occurs as a complement of a light verb (e.g. prinjat’ “adopt, make”) but not as a complement of a lexical verb (e.g. ob”jasnit’ “explain”): (45) a. My znaem, kem Vanya prinjal rešenie [Russian] we know who Vanya adopted decision [stat’ kem kogda vyrastet]]? [to-become who when will.grow.up]] ‘We know who Vanya made a decision to become when he grows up’ b. *My znaem, kem Vanya ob”jasnil rešenie we know who Vanya explained decision [stat’ kem kogda vyrastet]]? [to-become who when will.grow.up]] Intended: ‘We know who Vanya explained (his) decision to become when he grows up’ Note further that the grammaticality of extraction inversely correlates with the possibility of inserting a DP-level element such as a possessive pronoun or a demonstrative: such an element is grammatical only if it occurs in an event nominal

324

CONTEMPORARY LINGUISTIC PARAMETERS

that serves as a complements of a light verb, and not as a complement of a lexical verb: (46) a. Vanya prinjal (*svoë/*èto) rešenie [Russian] Vanya adopted (*own /*this) decision [stat’ lëtčikom]] [to-become   pilot]] ‘Vanya made {a/*his own/*this} decision to become a pilot’ b. Vanya ob”jasnil (svoë/èto) rešenie Vanya explained (own /this) decision [stat’ lëtčikom]] [to-become pilot]] ‘Vanya explained {a/his own/this} decision to become a pilot’ Lyutikova adopted the Universal Structure approach, hypothesizing that nominals that are complements of lexical verbs (45b, 46b) are DPs, whereas nominals that are complements of light verbs in collocations (45a, 46a) lack the DP projection. In contrast, under the WYSIWYG approach, such as that of Corver’s (1992) and Bošković’s (2005), who explained the availability of LBE in Russian but not in English by the blocking effect of the DP in the latter language, the contrast in (46) is unexpected. Thus, these facts buttress the view that some (though not all) nominals in languages without articles are DPs. Another argument in support of the Universal Structure view, similar in spirit, comes from the Differential Object Marking (DOM) in Tatar, discussed in Lyutikova and Pereltsvaig (2013, forthcoming). As mentioned above in connection with (6), in this article-less language direct objects may be either marked accusative or left unmarked for case: (47) a. Marat bala-nı čakır-dı [Tatar] Marat child-ACC invite-PST ‘Marat invited {a (specific)/the} child’ b. Marat bala čakır-dı Marat child invite-PST ‘Marat invited {a child/children}’ The contrast in case marking correlates once again with the availability of DP-level elements, such as strong quantifiers and demonstratives: when such elements are present, the object must be marked accusative. (48) a. Marat här bala-*(nı) čakır-dı [Tatar] Marat every child-ACC invite-PST ‘Marat invited every child’ b. Marat bu bala-*(nı) čakır-dı Marat this child-ACC invite-PST ‘Marat invited this child’



THE FUNCTIONAL STRUCTURE OF N

325

Moreover, Lyutikova and Pereltsvaig (2013, 2014) show that unmarked direct objects fit the profile of a Small Nominal (cf. Pereltsvaig 2006c) and argue that the DOM contrast cannot be explained in terms of object shift for accusative objects nor in terms of pseudo-noun-incorporation of unmarked objects.

13.4. Macro-parameters vs micro-parameters In the preceding sections, I described two approaches to the variation in the functional elements of a noun phrase: what I called the WYSIWYG approach, which says that only the structure that hosts the functional elements actually present in a given language is projected, and the Universal Structure approach, which takes the functional structure to be universal and places the locus of variation in the morphosyntactic or even morphophonological component. Although under both approaches languages from different language families are compared, the two approaches differ as to whether they view such variation through the macro- or micro-parametric lens. Under the WYSIWYG approach adopted by Chierchia (1998a, 1998b) and Boškovic´ (2005 and later work), the variation between languages with and without articles or languages with and without classifiers is attributed to the meaning of the whole class of lexical items, the nouns. For example, in languages without articles nouns are said to be themselves referential (or “argumental”), whereas in languages with articles nouns are predicative (or can be either predicative or referential). Similarly, in languages with classifiers all nouns are said to be mass, whereas in languages without classifiers nouns can be either mass or count. Consequently, different functional structures are available in different types of languages: if the nouns themselves fulfill the functions (e.g. referential, divisional) otherwise associated with functional elements, the corresponding functional structure is not projected. This approach is thus clearly macro-parameteric in nature. As such, it contradicts what Baker (2008) calls “the Borer-Chomsky conjecture” and postulates existence of what Baker termed a “grammatical parameter,” that is a parameter that “concerns principles of grammar that cannot be localized in the lexicon per se.”26 Baker argued that such grammatical parameters indeed exist: in Baker (2008a) he proposed two such parameters that concern agreement configurations; his polysynthesis parameter (Baker 1996a; see also Fábregas, this volume) also falls under this category. It is therefore unsurprising that Baker (2003: 113) takes the WYSIWYG approach to nominal structure. In contrast, the Universal Structure approach, championed by Borer (2005a) and largely shared by syntacticians working within the cartographic framework (see Cinque 2002, inter alia), is micro-parametric in that it attributes the variation in noun phrase structure not to the availability of functional projections in a language but to the morphological or even morphophonological properties of inflectional material, in line with the Borer-Chomsky conjecture. Under the Universal Structure approach, the same functional structure is projected in all types of languages; which functional elements instantiate a given projection depends on the properties of functional heads in the lexicon. For example, the ClP is projected in all languages, but languages vary as to what functional heads can fill the ClP: an obligatory

326

CONTEMPORARY LINGUISTIC PARAMETERS

classifier (e.g. in Mandarin), optional classifier (e.g. in Russian), plural marking (e.g. in English), or numeral which is then raised to a higher functional projection (e.g. in Hungarian). Similarly, the DP is projected both in languages with and without articles, but it can be filled by an overt or null article, a demonstrative, a raised noun, a possessor agreement marker, and so on. Moreover, under the Universal Structure view, the typology of nominal variation is thought to be the result of the accretion of many micro-parameters that concern the individual functional heads. An even more fine-grained micro-parametric approach is taken by Giuseppe Longobardi and his colleagues at the LanGeLin (Language and Gene Lineages) project (Longobardi 2003, 2005; Guardiano and Longobardi 2005; Longobardi and Guardiano 2009; Longobardi et al. 2013). The goals of this project extend beyond describing variation in nominal structure; the parameters (or rather their values) are used as comparanda in determining phylogenetic relationships among languages. For example, Longobardi and Guardiano (2009) examine 63 binary parameters pertaining to the structure of noun phrases in 28 (23 extant and 5 extinct) languages. This set of languages includes some very closely related languages (or dialects), such as Italian and Salentino (the Italo-Romance variety spoken in the provinces of Bridisi and Lecce), some more distantly related Indo-European languages, such as English, Rumanian, and Russian, as well as some unrelated (non-Indo-European) languages, such as Hebrew, Wolof, Hungarian, and Basque. In Longobardi et al. (2013), 26 Indo-European languages are compared with respect to 56 parameters; there is a partial overlap between the two parameter lists, with 24 parameters appearing on both. Given a more detailed description of each parameter discussed in Longobardi et al. (2013), I will focus on that list. Of particular relevance for the types of languages discussed in this chapter are the parameters p1-p20, which “define the syntactic properties of four features typically associated with the functional category that heads nominal structure (D), i.e. Person, Number, Gender, and definiteness” (Supplementary Materials: 4). Parameters p1-p3 discriminate languages with and without grammaticalized person, grammaticalized number, and grammaticalized gender, which Longobardi et al. (2013) believe to be associated with D°. Notably, all 26 Indo-European languages considered, except Farsi with respect to grammaticalized gender, are valued [+] for all three parameters. This means, in Longobardi et al.’s (2013) framework that all languages in the sample have DP, including those that do not have articles (e.g. article-less Slavic languages). Thus, they implicitly adopt the Universal Structure approach, at least as far as some classes of article-less languages are concerned (no languages of the Chinese/Japanese type are considered in either paper, but it appears that they would be analyzed as -p1). Other parameters that distinguish the different types of languages discussed here include p7 and p8, which “ask whether a language must formally mark a definiteness value on its nominals arguments, i.e. express an operator of maximality in the relevant domain of discourse” (SM: 6). Languages with articles (Romance, Greek, Germanic, Celtic, and Bulgarian) are specified as +p7 and +p8, whereas other Slavic languages and the Indo-Iranian languages—all lacking articles—are specified as −p7 and consequently as 0 with respect to p8. Parameter p9 “defines whether attraction to the D area of referential nominal material … is overt … or not” (SM: 7). Romance languages have overt movement of proper names to D° (cf. Roma antica lit. ‘Rome ancient’ vs l’antica Roma lit. ‘the ancient Rome’; see Longobardi 1994 for details), while



THE FUNCTIONAL STRUCTURE OF N

327

Germanic languages do not (cf. ancient Rome). Unfortunately, space limitations prevent me from considering the other parameters in detail. It is worth mentioning, however, that the micro-parametric view of the variation in the nominal domain allows Longobardi and Guardiano to identify a sufficient number of parameters in one limited subdomain. Contrary to the WYSIWYG approach, championed by Bošković, Longobardi and Guardiano (2009: 1687) explicitly assume that the parameters in the nominal domain are “relatively isolated from interaction with parameters external to the set.” This “modularized global parameterization” strategy, first proposed in Longobardi (2003), allows to draw inferences about phylogenetic relationships among languages, Longobardi and Guardiano claim. Application of computational phylogenetic algorithms to such a set of parameter values produces results that “meet most expectations” (Longobardi and Guardiano 2009: 1693): Basque, usually treated as an isolate, is the first outlier; Wolof, a West Atlantic language which has never been connected to any European or Mediterranean language, comes second. Both Basque and Wolof are clearly recognized as external to a node coinciding with the so-called Nostratic grouping.27 The next outmost bifurcation singles out the (West) Semitic subgroup. The Uralic (Finno-Ugric) family emerges correctly as well. The branching within the remaining Indo-European family is overwhelmingly the expected one, although a couple of surprises are to be found. These unexpected groupings typically reflect instances of horizontal transmission (i.e. borrowing): for example, the grouping of English with Norwegian reflects the Scandinavian influence on English. Despite these occasional misclassifications, the Parametric Comparison Method (PCM) has been judged to work “strikingly well” (Longobardi et al. 2013: 124; see also Bortolussi et al. 2011) as the horizontal transmission does not seem to limit the effectiveness of this method seriously enough to undermine the correct representation of the vertical (i.e. phylogenetic) relations. Another important advantage of the PCM is that it is able “to reconstruct chronologically deep phylogenies using exclusively modern language data, often the only available data outside Eurasia” (Longobardi et al. 2013: 123). The possibility of extending the micro-parametric view of noun phrase variation, couched in terms of the Universal Structure approach, to historical linguistics provides another advantage for this approach over the macro-parametric WYSIWYG approach.

13.5. Further lines of research As is clear from the above sections, the exact nature of the DP parameter is far from clear. Much careful consideration of facts both within and across relevant languages is still needed before the debate between the WYSIWYG approach and the Universal Structure approach can be resolved. One obvious observation is that languages that lack overt functional elements of a certain type (such as classifiers or determiners) may use other means to express the semantics associated with the given functional projection. For example, article-less languages may use word order, the form of adjectives, and demonstrative determiners to express the semantics of articles (i.e. definiteness, determinacy, specificity, and the like). A closer

328

CONTEMPORARY LINGUISTIC PARAMETERS

examination of these alternative means is bound to shed new light on the structure and parameterization of the noun phrase. Nevertheless, the research on the DP parameter need not be limited to noun phrase internal elements and their expression in various languages. One conclusion that has emerged from the debate so far is that the issue of the structure of nominals in different types of languages may be connected to—and have important consequences for—other areas of syntax and semantics, both on the nominal and clausal levels. Thus, gaining a better understanding of such phenomena—including adjunct extraction from NP, scrambling, second position clitics, negation, question formation, clitics and clitic doubling, quantifier scope possibilities, sequence of tenses, and more—is vital for the DP parameter debate. Likewise, further insight into the functional structure of nominals comes from considering evidence from other nominal-internal phenomena such as nominal-internal word order permutations, possible adjective positions, determiner choice, etc. Another potentially fruitful area of investigation involves examining the DP parameter in light of evidence from psycholinguistics, including studies of acquisition (L1 and L2) and of loss (aphasic and attrition) of nominal functional categories. Some work in this area has already begun: for example, Trenkic (2004) argued based on evidence from a study on the second language acquisition of English articles by speakers of Serbo-Croatian that DP is not projected on top of (full) nominal phrases in Serbo-Croatian, and by implication in other article-less languages. Another way to look at the issue is to model the prediction of the two major approaches to the DP parameter and then examine whether acquisition/attrition evidence validates the prediction of either major approach. As mentioned above, according to the WYSIWYG approach, whether a language has articles is a syntactic parameter, which must interact with other syntactic parameters, for example, with the hierarchy of parameters as proposed by Baker (2001). How the DP parameter might fit into Baker’s model is suggested by the claim in Bošković (2009) that polysynthetic languages are necessarily article-less. According to the logic of Baker’s model, this means that the DP parameter occupies a very high position in the hierarchy of parameters, above the (currently) top-most polysynthesis parameter. As such, it must be acquired in the earliest stages of first language acquisition, before the polysynthesis parameter. Unfortunately, as Baker himself admits, little is known about the acquisition of polysynthesis, but this issue is certainly worthy of further exploration. For a preliminary discussion of L1 acquisition of the DP parameter, the reader is referred to Koulidobrova (in press).

Notes *

Special thanks to Giuseppe Longobardi, Ora Matushansky, Kate Lynn Lindsey, Antonio Fábregas, and an anonymous reviewer for helpful discussions, comments, and suggestions. All errors are mine.

1

Besides numeral classifiers, considered in this chapter, some languages such as Akatek (described by Zavala 2000) also have separate sortal and noun classifiers. While



THE FUNCTIONAL STRUCTURE OF N

329

numeral classifiers make nominal referents countable or quantifiable, sortal classifiers sort nominal referents by characteristics such as shape and noun classifiers typically sort nouns by material qualities or essences. In Akatek, numeral classifiers distinguish human, animal, and inanimate nouns; sortal classifiers distinguish a dozen or more shapes (“smooth,” “long three-dimensional,” “erect,” “half-circle,” “round,” “wide flat,” “small spherical,” “separate,” etc.); whereas noun classifiers include “man,” “woman,” “animal,” “tree,” “corn,” “water,” “salt,” etc. (i) a. ʔoš-k’on kupan no’ wakaš three-anim half.circle animal cow ‘three cows’ (lying down) b. ʔoš-eb’ kupan ʔišim paat three-inan half.circle corn tortilla ‘three (folded) tortillas’ c. ʔoš-eb’ šoyan ʔišim paat three-inan round corn tortilla ‘three tortillas’

According to Svenonius (2008), the three types of classifiers are merged in their own dedicated functional projections: numeral classifiers in UnitP (comparable to Borer’s 2005a #P, or to NumP in the structure adopted in this chapter), sortal classifiers in SortP (comparable to Borer’s ClP), and noun classifiers in nP.

2

Words in languages like English that are sometimes called “classifiers” (e.g. grain in three grains of rice and the like) are not true numeral classifiers. Similar items in Chinese are also argued not to be true classifiers in the main text below.

3

Pirahã is a particularly controversial exception (see Everett 2005, but also Nevins, Pesetsky and Rodrigues 2009).

4

Here, I set aside the generic interpretation of nominals.

5

Numerals are assumed to be merged somewhere other than “NumP.”

6

Numerous other projections have been proposed to augment the nominal structure and to accommodate various additional functional elements. Due to size limitations, they will not be considered in this chapter.

7

The rest of this chapter focuses on articles and classifiers though number marking is a crucial element in the argumentation for certain analyses, as we shall see below.

8

Longobardi himself argued that arguments are DPs in Germanic no less than in Romance, but the hypothesis that arguments are DPs has since been extended to other languages/language families.

9

Since Chierchia assumes the use of type-shifting operators (iota, etc.) similar to the ones proposed by Partee (1987), one could assume—though Chierchia himself does not propose this explicitly—that such operators are generated in D°, which would make this approach not entirely WYSIWYG. I am grateful to the anonymous reviewer for pointing this out to me.



10 The [−arg, −pred] option is impossible, as it would prevent NPs from having any interpretation at all. 11 According to Giuseppe Longobardi (p.c.), some variation exists among Italian speakers as to the availability of bare nominals in such examples. The nature of this variation is not entirely clear. 12 In what follows, I focus on challenges for Chierchia’s analysis from classifier languages of the Chinese type, which Chierchia himself discusses. As pointed out to

330

CONTEMPORARY LINGUISTIC PARAMETERS

me by the anonymous reviewer, creole languages are also problematic for Chierchia’s classification, since they share features with more than one language type. The reader is referred to Deprez (2005, 2007); Baptista and Guéron (2007); Alexandre and Hagemeijer (2007); Baptista (2007); Grant (2007); Kesler and Schmitt (2007); Kihm (2007); and Zribi-Hertz and Glaude (2007). 13 Borer (2005a) also notes that massifiers such as xiāng “box” may occasionally occur with count nouns: (i) sān xiāng shū three CL-box book ‘three boxes of books’ 14 SFP = sentence-final particle (cf. Cheng and Sybesma 1999). 15 A weaker alternative would be to claim that noun phrases in languages without articles are (or can be) NumPs, treating numerals and/or number marking as occupying Num°. For arguments that numerals in Russian (and similar languages) should not be reduced to the category Noun but should be treated as functional elements, see Pereltsvaig 2007a, 2013). 16 For more detailed discussions of the other relevant phenomena, particularly in Slavic languages, the interested reader is referred to the following works: Stjepanović (1998), Rappaport (2001a), and Bošković (2005) on adjunct extraction from NP; Stjepanović (1999c), Bailyn (2002b, 2003a, 2003b), and Bošković (2005) on scrambling; Franks and King (2000) and Migdalski (2010) on second position clitics; Bošković (2008) and Bošković and Gajewski (2011) on negative raising; Franks and King (2000), Arnaudova (2005), Bošković (2008), Slavkov (2008), and Runić (2011) on clitic doubling; Živanovič (2007) on the interpretation of superlatives; Bošković (2010) on quantifier scope; and Altshuler (2004, 2008) and Grønn and von Stechow (2010) on Sequence of Tense. 17 In Italian, LBE is not entirely impossible (as pointed out to me by Giuseppe Longobardi, p.c.). Particularly, LBE is possible if the when the nominal object is a resumptive clitic of a right dislocated definite nominal: (i) COSTOSE le ha comprate, le scarpe EXPENSIVE them has bought the shoes ‘He has bought EXPENSIVE shoes’ 18 Female foreign surnames do not decline in Russian, and neither do many female foreign given names. With such names, splitting appears to be likewise impossible: (i) *Čelsi ona obožaet Klinton Chelsea she adores Clinton ‘It is Chelsea Clinton that she adores’ 19 Once again, Lithuanian is expected to pattern with Bulgarian, based on its definiteness marking, but it patterns with Serbo-Croatian instead; see Van Hofwegen 2013 for details. 20 Furthermore, several Russian native speakers I have consulted found (30b) to be (nearly) as degraded as (30d), meaning that there is some degree of inter-speaker variation on this matter. 21 Crucially, examples such as (31b) do not involve a pseudo-partitive structure. 22 One alternative route to take is to hypothesize that the DP projection is available in non-process nominals but not in process nominals. It is not clear, though, what the type of nominal (process vs non-process) has to do with the availability of DP, so this approach would be highly ad hoc.



THE FUNCTIONAL STRUCTURE OF N

331

23 For details of the analysis and supporting evidence, see Engelhardt and Trugman (1998). Derived nominals in Slavic are discussed further in Rappaport (1998, 2001b, 2005); Zlatić (1997); Dvořáková-Procházková (2007); Pazelskaya (2007), among others. 24 http://russiafaq.ru/questions/Pochemu-do-sih-por-bol-shinstvo-lyudey-p-yut-koka-kolu-ne-smotrya-na-to--chto-vse-davno-znayut--chto-ona-ochen--vrednaya-.html 25 A similar proposal for corresponding items in Bulgarian is put forward by Cinque and Krapova (2007). 26 Roberts (2012) proposes a way to reconcile Baker’s notion of macro-parameters with the Borer-Chomsky Conjecture “by construing macro-parameters as aggregates of micro-parameters” (p. 320). 27 In a more recent study, using more refined parametric data, Wolof came out as the outlier, with Basque lightly more internal, a configuration that reflects the geographical distinction between African and non-African languages (Giuseppe Longobardi, p.c.).

PART THREE

Parameters beyond morphosyntax

14 Parameters in phonological analysis: Stress Marc van Oostendorp Meertens Institute / Royal Netherlands Academy of Sciences

14.1. Introduction Phonological stress might be one of the most successful testing beds for any theory of linguistic variation and language acquisition.1 One reason is that stress systems— deciding on which syllables in a word languages put primary and secondary stress—seem relatively easy to isolate from other properties of language. We can approximate the stress pattern of a word by a string of numbers (e.g. 2001 is a representation of a four-syllable word with primary stress on the last syllable and secondary stress on the first), and thus view an individual language as a set of such strings, and the grammar as an intensional description of such sets. It seems fairly easy to write grammars of such sets of strings; and then to generalize over such grammars. A second, related, reason is that we have a relatively good overview of stress systems of the world’s languages, for instance, in the StressTyp2 database2 created by Rob Goedemans, Jeff Heinz and Harry van der Hulst. Also the World Atlas of Linguistic Structures3 features several chapters on stress. Thirdly, the parametric space seems relatively simple; it has turned out to be possible to determine a relatively small number of parameters which in interaction can describe it virtually completely. It was probably for this reason that some of the first relatively complete, and relatively fruitful, applications of Chomsky (1981)’s idea of parameters were found in the phonological domain: Halle and Vergnaud (1987) and Dresher and Kaye (1990). The first offered a way of showing how a vast number of languages could be described by using only a small set of parameters. The second furthermore showed

336

CONTEMPORARY LINGUISTIC PARAMETERS

that it was possible to implement these parameters in a computer program that could then simulate the acquisition of a language based on these parameters. Around the same time, however, the parametric model suffered two important blows in metrical theory, one empirical and one theoretical. The empirical blow was that a serious gap in the predicted typology was shown to exist: some of the predicted languages are very rare, or maybe even absent from the database (Hayes 1987, McCarthy and Prince 1986). This led to an approach in which a universal foot inventory took over the central role from parameters; languages would pick their feet from this non-binary inventory. The theoretical blow came from Optimality Theory (OT; Prince and Smolensky 1993), which replaced the view on binary parameters switched “on” and “off” with the idea of universal, but conflicting constraints that would get a different ranking in different languages. Most of the typological as well as most of the acquisition modeling work within formal phonology since then has been set within OT. Parameters formally survived in some non-mainstream models such as Government Phonology; but between 1995 and 2005 constraint ranking seemed pervasive in the stress theory—also because GP practitioners for some reason mostly did not show a lot of interest in stress (see Scheer and Szigetvari 2005 for an exception). Meanwhile, OT evolved, as did some of the dominant ideas on what the proper theoretical account of language variation is. One important direction this evolution took was in the direction of more gradient distinctions between languages and small, stochastic influences on the location of stress within individual languages. This seems currently also to be applied to stress. This chapter aims to describe the developments towards and away from parametric models in some detail. What motivated the various moves? What have we learned about the structure of language variation from this discussion? And where are we now? I follow the course of history that I just sketched. It is not my aim to provide a complete picture of the course of history; I merely want to sketch the main arguments which played a role in the discussion, focusing in particular on the theoretical aspects of the discussion, and what this can teach us about language variation, not just for phonology but from a more general linguistic perspective. My main goal is to explain the relevance of the phonological discussion also for syntacticians and morphologists interested in the formal study of linguistic variation. This does not mean that I will take into account the interesting issues of the interface; in order to not unnecessarily complicate the discussion, I concentrate on stress assignment in monomorphemic words. The interest for practitioners of other branches of formal grammar will be mostly in seeing how formal theories of variation were argued for and against within this particular domain.

14.2. Parameterizing stress The idea that language variation may be reduced to a set of binary parameters, first proposed in Chomsky’s so-called 1979 Pisa Lectures (and published as Chomsky 1981) caught on almost immediately in phonology, and in particular in metrical phonology. Conceptually, this is not very strange, since many of the choices that



PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

337

languages make in this typological space are almost trivially binary. There are two types of (binary) feet: iambs and trochees; languages can be quantity sensitive or not; primary stress can be assigned at the left edge of the word, or at its right edge, etc.4 Very important in this respect was Hayes’ (1980) dissertation, published as Hayes (1985), which proposed that the stress system of languages can be described using parameters such as the following, on binary foot types: (1) a. Dominance [Right/Left] (The head of the foot is on the left / on the right) b. Quantity-sensitive [Yes/No] (Does the language distinguish be- tween ‘light’ and ‘heavy’ syllables?) c. Obligatory Branching [Yes/No] (Should the head be heavy?) d. Labeling Based On Branching [Yes/No] (does headedness switch from a light syllable?) Taken together, these parameters cover quite some typological ground. The first distinguishes between languages with left-headed feet (sw) and those with rightheaded feet (ws). The other three parameters are related to each other in an interesting way. The Quantity-Sensitivity Parameter distinguishes those languages in which the stress of an individual syllable is only dependent on its numerical position in the string (quantity-insensitive languages) and those in which the internal structure—the difference between “heavy” and “light” syllables—does play a role (quantity-sensitive languages in which heavy syllables are disallowed from weak positions). The other two parameters can be seen as dependent on this one: they only apply to quantity sensitive-languages. Languages in which Obligatory Branching is switched on, demand that strong positions are filled by heavy syllables (the logical inverse of Quantity-Sensitivity). Languages of this type are sometimes called Quantity-Dependent. We thus end up with a typology of three possible left-headed feet (and three right-headed feet mirroring them): (2) Quantity-Insensitive Quantity-Sensitive

Quantity-Dependent

Ft    Ft

Ft    Ft

Ft    Ft

σ  or  σs  σw

σ  or  σs   σw

σ   σs   σw

-Quantity-Sensitive

L

H  or H   L

+Quantity-Sensitive

+Quantity-Sensitive

-OB

+OB

The last parameter, LBOB, makes a further division in the set of quantity- sensitive languages. It decides that within a “left-dominant” foot the left syllable is strong if and only if it is heavy; otherwise the right syllable is strong (also if it is light). The resulting foot inventory is a bit complex, and has been disregarded in later work since Hammond (1986).

338

CONTEMPORARY LINGUISTIC PARAMETERS

As far as I am able to see, the parameters idea of stress typology met with almost general acceptance in the years that followed. One difference between a lot of phonological (metrical) literature and its syntactic counterpart within GovernmentBinding Theory was that the latter had at the time abandoned the notion of rules almost completely. Grammatical processes were supposed to be universal and exceptionless, and basically reducible to Move α. The parameters mostly were statements about what kinds of representations were permissible. In phonology, on the other hand, it was often supposed that rules could be parameterized. This could happen in two ways, both of them exemplified in Prince’s (1983) influential paper, in which basically all parameters are defined on the formulation of rules and not on representation. A parameter would specify whether the assignment of metrical structure starts from the left or from the right, whether it starts with a “peak” or with a “trough.” He also discusses an alternative option, viz. of making the presence or absence of a rule (the “End Rule”) parameterizable. The discussion that followed concentrated on a different aspect, viz. the right way to represent stress, e.g. in terms of trees, of grids, which was Prince’s (1983) proposal, or of “bracketed grids.” The latter was the point of departure for Halle and Vergnaud (1987). Like Prince, these authors assumed a primarily parameterization of phonological rules rather than of representations. This is true for instance for the headedness parameters. In both syntax and phonology, we find variation in which element is the head in linearly ordered constituents [XY]. In syntax, there are roughly two ways to represent this. One is to assume that left-headed structures are well-formed in some languages, whereas in others only right-headed structures are acceptable. In principle, all languages can construct both—but the unwell-formed ones will be filtered out. Alternatively, one can assume that the structure-building rule itself is parameterized: languages choose either (3) or (4) (Travis, 1983, 1988, 1989): (3) X’ → X (YP) (4) X’ → (YP) X Since Stowell (1981), the standard conception in syntax on these matters seems to be, however, that the parameters are restrictions on structures, and not properties of rules. That idea did not seem to be accepted very generally, however, in metrical analysis. Hayes (1995), arguably the cumulative standard work on parametric metrical stress theory, seven basic parameters are proposed (along with a rather large number of accessory parameters regulating e.g. extrametricality). Hayes states explicitly that (1995: 54) “[b]y setting all the relevant parameters, one derives a stress rule.” An exception to this derivational view on rules was Dresher and Kaye (1990), which presented the following list of parameters:5 (5) a. The word tree is strong on the [Left/Right] b. Feet are [Binary/Unbounded] c. Feet are built from the [Left/Right] d. Feet are strong on the [Left/Right]



PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

339

e. Feet are quantity sensitive (QS) [Yes/No] f. Feet are QS to the [Rime/Nucleus] g. A strong branch of a foot must itself branch [No/Yes] h. 1. There is an extrametrical syllable [NoYes] 2. It is extrametrical on the [Left/Right] i. A weak foot is defooted in clash [No/Yes] j. Feet are noniterative [No/Yes] Of these parameters, only two, or possibly three, are stated in derivational terms (“built from the Left/Right,” “is defooted,” and maybe “are noniterative”). Dresher and Kaye (1990) provide a rather vast typological space of languages in this way: 10 binary parameters gives 210 = 1024 possible languages, although the parameters are not all independent. For instance if feet are Unbounded, it is no longer relevant whether they are built from the left or the right or whether or not they are assigned iteratively; and if languages are not quantity-sensitive, we do not need to know whether or not they look at the level of the nucleus or that of the rime. Dresher and Kaye (1990) is an important paper for parameter theory also because it provides an explicit learning algorithm and shows how a rather complex set of parameters, that is close to complete in its empirical domain, can be set by observing the stress patterns of a language.6 Following Lightfoot (1989), Dresher and Kaye propose that every parameter has a default status and is accompanied with a cue. The learner pays attention to the cue—some piece of positive evidence—in order to be able to set the parameter rightly (see also Dresher 2012). Consider the quantity-sensitivity parameter. The default setting here will be quantity-insensitive (QI): there is no difference between heavy and light syllables. The learner may set this to quantity-sensitive (QS) if she finds that there are words which have the same number of syllables but a different stress pattern (e.g. páta, patán. Notice that this cue is not an individual word: the learner thus needs to have stored a few patterns, and possibly apply statistics to them to avoid being misled by one misinterpretation, in order to be able to decide. Another property of the parametric system is that the acquisition of parameters can be temporally ordered. For instance, the learner can only decide about the parameter on the word tree once several decisions about feet have been made (since the word stress by necessity is built on top of a foot). Because settings are not completely independent from each other, Dresher and Kaye (1990) calculate that their system generates 216 possible languages. Some parameters are left out of consideration; an important one is lexical stress, which actually causes some problems, because lexical stress systems (in which the location of stress is specified for each lexical item separately) are by definition unpredictable. This means that given a limited amount of input, the learner might be misled to believe that she is confronted with a certain language with predictable stress, just because the words she has heard so far fit within a certain pattern. For instance, it is possible that she has by some accident heard only those words that happen to have stress on the final syllable, even though in some other words of the language, stress can fall on any other syllable in the word. On the other hand, setting the “lexical stress” parameter too quickly will never lead to a resetting of that parameter, since there is no possible word pattern

340

CONTEMPORARY LINGUISTIC PARAMETERS

that violates the resulting grammar. In other words, we can only confidently set the lexical stress parameter once we have explored the full parametric space of predictable stress languages and concluded that our language does not fit into that space. That should make the lexical stress languages the most difficult to learn. At the same time, there is never a guarantee for any learner that she is not actually learning a lexical stress language, especially given the fact that many languages seem to have “exceptions,” e.g. on loanwords. Gillis et al. (1995) give an exhaustive review of Dresher and Kaye (1990). They judge that the learning procedure is more complicated than it would seem at first sight: we found that in formulating the cues, much more was needed than positive evidence; some cues rely heavily on indirect negative evidence, whereas others involve extensive cross-word comparisons and multiple ways of recoding the input material. Furthermore, a rather elaborate ordering of the parameters was set up so that parameter settings could be made dependent upon the value of previously set parameters. Finally, exhaustive search was used in restricted parts of the solution space. […] Both the use of indirect negative evidence and extensive cross-word comparisons in the form of consistency checks require all relevant data to be present when learning is initiated. In the context of a deterministic learner, oversights in the data collection phase may lead to wrong initial decisions. Gillis et al. (1995) also did an empirical test in which they made up (artificial) examples of each of the 216 languages predicted by the parameter set, and submitted them to the learner. About 75–80% of them was assigned a correct analysis, which means that Dresher and Kaye (1990) learning algorithm would be less than perfect.

14.3. Inventories The arguments given in Gillis et al. (1995) did not have a strong impact on phonological theory, at least at first. Around the same time, however, it seems that many ‘mainstream’ phonologists were abandoning the notion of parameters for the explanation of stress patterns. The first step in this development was taken by Hayes (1987, 1995) and McCarthy and Prince (1996). In parallel, but independently from each other, these authors made the same discovery, viz. that there is a gap in the typological space predicted by binary parameter theory: while trochaic languages can be either quantity-sensitive or quantity-insensitive, iambic languages always seem to be of the former type. Obviously, a parametric system containing the parameters QS / QI and trochee / iamb predicts four types of languages. The solution suggested by these authors was that Universal Grammar provides an inventory of three different foot types: (6) Syllabic (QI) trochee: (σσ) Moraic (QS) trochee: (μμ) Iamb: (Lσ) or (H)



PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

341

Trochees are “even” (there is no asymmetry between head and dependent in terms of syllable structure), but come in two types: QI trochees consist of two syllables, regardless of their weight. QS trochees consist of two moras, which means that they either consist of one heavy syllable or of two light syllables. Iambs are always weight-sensitive, and this in a slightly different way than trochees: they consist of a light syllable, followed by another one, or only of a heavy one, if no light syllable is available in the word. This implies that iambic languages always need to have some way to express weight on syllables, and typically they will have a form of iambic lengthening.7 It is clear that this system cannot be expressed by binary parameters, although it could obviously be expressed in a system with a parameter with three possible values. Such a system would not have a lot of explanatory value, however: where do exactly these three values come from? Why do we not find the mirror images of the foot types in (6)? Hayes (1987) suggested that the answer to this question was to be found in what he called the Iambic-Trochaic Law (ITL; see Hyde 2011, for an overview): (7) The Iambic-Trochaic Law a. Elements contrasting in intensity naturally form groupings with initial prominence. b. Elements contrasting in duration naturally form groupings with final prominence. The ITL is assumed to have a wider psychological applicability. It is based on psychological experiments (Bolton 1894; Woodrow 1909) in which informants listened to sequences of (non-linguistic) sounds, and were asked to group these. Some sequences alternated in intensity and others in duration. The results seemed to conform the ITL. Later studies, such as Iversen et al. (2008) question some aspects of this universality; for instance, Japanese listeners seem to react differently than English participants. This development was in several ways representative of certain changes which were in the air towards the end of the 1980s, and in the early 1990s within phonology. The most prominent of these probably was a move away from purely formal accounts towards accounts that were grounded in extralinguistic factors, cognitive or otherwise. In this case, this concerned a psychological law which obviously itself is somewhat mysterious: there is no immediate explanation why the human mind would have a preference for grouping elements contrasting in duration in a different way from other kinds of auditory signal. To this extent, then, the ITL does not really “explain” the typological gaps we find, but it rather suggests that there is a correlation between linguistic typology and the results of a psychological experiment. Another way in which this theoretical turn was typical for phonological research was the interest in gaps in the typology. One could argue that the fact that a certain class of languages—those with QI iambs—has not been found, does not mean that they do not exist, and even if they do not exist, this does not necessarily mean that they could not be generated by the language faculty. Both of these facts could be accidents of history, and have a non-linguistic explanation, as was correctly

342

CONTEMPORARY LINGUISTIC PARAMETERS

observed by Hale and Reiss (2000, 2008). The typological gap argument has played a role in the literature also after Hale and Reiss made their observations. See Buckley (2009) for an interesting discussion. The issue is an interesting one: what are the responsibilities of typological theory? On the one hand, the theory should definitely account for all systems that are attested. At the same time, the theory should also be relatively tight, in order to make predictions and to be falsifiable. Now there seems to be consensus that some systems are definitely to be excluded from linguistic theory. An example that is sometimes cited would be a language in which stress is put on every prime-numbered syllable of the word (so syllables number 2, 3, 5, 7, 11, etc.). Such languages do not exist and it may be assumed that they would be outside of ordinary human linguistic capacities. In a thoughtful paper, Blaho and Rice (2013) discuss the issue of falsification at great length. Their basic argument is as follows. Suppose we dispose of two theories Θ1, which predicts that there are four types of languages based on two binary parameters (languages A, B, C and D), and Θ2, which predicts that there are only three types of languages. Inherently, Θ2 seems preferable; however, it is not clear that this is the case if Θ2 has all of the machinery of Θ1 plus the extra statement ‘language D cannot exist’. One might argue that in the case at hand, setting up an inventory of three different feet is not necessarily really more restrictive than a theory which is conceptually simpler, but makes the extra prediction. As a matter of fact, it could be argued that Θ2 is not really more restrictive in any deep sense: if it would turn out that language D would exist, Θ2 would be very easy to repair, viz. by lifting the restriction, or adding the extra foot to the typology. The prediction thus is not very deep in any way. Θ1, in turn could relegate the explanation of why language D is so rare to other factors, such as learnability or (other) diachronic considerations: D is formally possible, but just unlikely for such reasons. Another objection against the introduction of inventories, is that the same typological gap has also been explained without reference to the ITL. An influential contribution was Kager (1993), arguing that the inventory of feet was indeed symmetric, consisting of syllabic and moraic trochees and iambs, and that the asymmetries which are observed in the typology can be derived from independent requirements, such as the avoidance of “lapses” and “clashes” (sequences of consecutive unstressed or stressed syllables, respectively) on the grid. Take for instance the fact that there are no “uneven trochees,” consisting of a heavy syllable, followed by a light syllable, even though there are “uneven iambs.” If we admit that the first mora in a heavy syllable is the head, an uneven trochee would consist of a head, followed by two non-head, causing a lapse within the foot. An uneven iamb, on the other hand, consists of a weak-strong-weak sequence, so that there is no lapse. See Rice (1992) for another proposal for a symmetric foot inventory. A different approach was taken by Van de Vijver (1998), who argued that there is no specific parameter setting (or constraint) in favor of iambs: languages prefer trochees by default, and iambs emerge only when this can avoid a conflict. This approach is embedded within Optimality Theory, which brings us to the next blow which was brought to the parametric vision of metrical structure.



PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

343

14.4. Constraint ranking As is well known, Optimality Theory has dominated phonological theory for the past two decades. As a matter of fact, OT is, if anything, an alternative to Principles and Parameters Theory, that is to say a theory of the way in which languages can vary: it presents a radically different view of language variation, which is seen in this case not as a matter of binary choices within an otherwise unchangeable frame, but of different rankings of surface constraints. Indeed, within classical OT the claim is that those rankings are the only possible systematic phonological differences between languages: there are no differences in underlying representations, and languages also do not differ in the absence or presence of certain phonological processes. Everything is universal, except for the interaction of constraints on surface representations. One difference between constraint ranking systems and parametric systems is in the numbers. Assuming that parameters are binary and independent from each other, a set of n parameters gives 2n possible languages; a system with n constraints, on the other hand, making the same assumptions, gives n! possibilities, a number that is growing much more rapidly. This means that constraint-based systems need to be set up more sparsely: all else being equal, the number of constraints should be smaller than the number of parameters.8 Another important empirical difference between a parameter theory and OT is that the latter predicts so-called the emergence of the unmarked (TETU) effects. Prince and Smolensky (1993) give the following example. One difference between languages is that some allow closed syllables, whereas others do not. Parameter theory could analyze this as a parameter: “Allow closed syllables yes/no.” OT provides for a constraint NoCoda which is undominated in some languages and dominated by faithfulness constraints against deletion of consonants or insertions of vowels in others. Presumably, monomorphemic VCV sequences are syllabified in all languages as V.CV. This is true even in languages which otherwise allow for closed syllables—in other words in which VC is a well-formed syllable. Within parameter theory, this fact needs a separate explanation: some languages apparently have the parameter ALLOW CODAS switched to YES, so why would they still avoid them? Within OT, this can be explained by the constraint NoCoda. Even in languages in which it is low ranked, it is still present. If the higher-order constraint are irrelevant—like here, where we do not have to be unfaithful in order to satisfy the constraint—we will still see the effects of the constraint. In cases such as these, OT can thus boast an advantage of theoretical elegance: two phenomena which have to be analyzed as separate under a different set of assumptions get a uniform explanation here. I am not aware of explicit TETU arguments against parameter theory from the domain of stress.9 At first sight, they seem also difficult to provide, since TETU arguments, like the one above, usually involve some interaction with faithfulness to underlying representations: a markedness constraint is usually marked by a faithfulness constraint, but when such a faithfulness constraint does not play a role, the markedness constraint shows its working again. In the syllable structure example, NoCoda is ranked below

344

CONTEMPORARY LINGUISTIC PARAMETERS

faithfulness constraints preventing the insertion of vowels or the deletion of coda consonants (two ways to satisfy the constraint), but in VCV words we can satisfy the constraint without deleting or inserting anything. However, in the analysis of stress, underlying representations do not usually play a role: it is assumed that feet are not underlyingly present, so that faithfulness is never applicable to them. At the same time, underlying at least some of the theoretical work on stress typologies within OT, the TETU idea does play a role, as I will show below. At the same time, OT does not fit very well intuitively with the kind of binary options that are typical of parametric systems and, possibly, of the typology of stress. This is a potential disadvantage, as I will now illustrate. Consider, for example, the issue of iambs vs trochees. Some languages have iambs, while others have trochees. How are we going to represent this typological fact? One option would be to have two constraints Iamb and Trochee, of which the requirements are exactly the opposite. Trivial constraint-ranking will then give the required result: (8) a Trochee: Feet are left-headed. b. Iamb: Feet are right-headed c. Iamb >> Trochee: Iambic feet d. Trochee >> Iamb: Trochaic feet

.

Although it is obviously technically possible to do it in this way, it does not seem to be a very insightful way to pursue the analysis. Furthermore, the system seems to predict that there can be languages which mixed foot types: by default there would be iambs, but in case some other higher-order constraint disallows those, trochees show up. Such mixed language types do not seem to exist (but see Houghton 2013 for arguments to the contrary). The same problem arises at other levels of prosodic analysis. For instance, some languages have main stress on the right-most foot, and others have it on the left-most foot; some may have it on the penultimate foot. There are no languages which have stress exactly in the middle of the word, so to some extent this is again a binary choice: stress is drawn to the edges of the word, the only issue is which edge. The problem is traditionally treated in OT by reference to so-called constraint schemata, of which the first instance was the so-called Align family of constraints McCarthy and Prince (1993), which is defined as follows: (9) Align (X, L/R, Y, L/R): Align the left/right edge of every X to the left/right edge of a Y. Constraint schemata are thus constraints with built-in parameters; by setting the parameters, we get constraints: (10) a. Align (Ft, R, Hdσ, R): Feet are right-headed (align the right edge of every foot to the right edge of a stressed syllable. b. Align (Ft, L, Hdσ, L): Feet are left-headed. c. Align (Wd, R, HdFt, R): Main stress is at the right. d. Align (Wd, L, HdFt, L): Main stress is at the left.



PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

345

34Two interpretations are possible of the relation between a constraint schema and its instantiations. One can either believe that all individual constraints are universally present, and they are ranked in individual languages such that the requirement for iambs obscures that for trochees, or the other way around. This is the interpretation which McCarthy and Prince (1993) give, but still does not solve the problems just mentioned. Alternatively, one might assume that the parameters are really set on a languagespecific basis. In my view, that would solve the problems just mentioned: a language would choose once and for all whether it likes right- headed or left-headed feet, and stick to that choice. In Van Oostendorp (1995) I argue for such a position also on other grounds. I observe that there is a typological it split between languages such as French in which many phonological processes (like stress and schwa epenthesis) seem sensitive to the phonological phrase, whereas in other languages such as Dutch those same processes seem sensitive primarily to the phonological word. These processes are independent from each other and probably to be accounted for by independent constraints. Yet those constraints seem to refer to a different parameter setting in French than they do in Dutch. If this were to be accepted—there are no indications that the idea has gained a very wide following, but I would like to bring it to the reader’s attention again—it would mean that languages differ in two kinds of way: in the parameter settings which are chosen in individual constraint schemata, and in the rankings of the resulting constraints. This goes against a central tenet of classical OT, viz. that constraints are universal and that the only difference between languages can be their respective rankings. It has to be noted, however, that the concept of one universal set of constraints seems to have been largely abandoned in a lot of OT literature, where it is often proposed that constraints are constructed by the language learning child in the course of acquisition (Moreton 2008; Hayes and Boersma 2001). In such theories, there obviously has to be some model on which new constraints are based, and these can thus be seen as parameterized constraints. It thus seems that the parameter is playing a role also in OT theories. In any case, the discussion on typologies within OT of the last twenty years seems to have concentrated mostly on issues of representation, although it has been argued that certain representational options work better under the assumption of violable constraints. In particular, it has been argued that the technology of surfacebased constraints seems to lead in particular to a certain type of representational assumption. Gordon (2002), for instance, argues that the typology of quantityinsensitive languages can be generated with only bracketed grids, and constraints against lapses and clashes: (11) a. Clash: A string of more than one consecutive stressless syllable may not occur. b. Lapse: A string of more than one consecutive stressed syllable may not occur. The point is, of course, that languages sometimes do allow for lapses or clashes. For instance, in a word consisting of an uneven number of syllables and binary

346

CONTEMPORARY LINGUISTIC PARAMETERS

feet, there is no way to avoid them. In such cases, one of the two constraints can be violated, but this will happen only when it is necessary. This is thus a TETU argument: even though a language sometimes shows clashes, it does not mean that some parameter Clash is switched “on”: they will still be avoided where possible. Similarly Brett Hyde has argued in a number of publications (Hyde 2002, 2007a, 2007b, for instance) that OT allows us to have overlapping feet, so syllables which are in the weak branch of one foot and at the same time in the strong branch of another foot. The reason why we need violable constraints for this, is that we need to assume that such overlapping feet are usually avoided; they only arise under certain very specific circumstances. This can of course be realized by assuming that the constraint against overlapping feet is universal, but violable. In a parametric approach, one would be forced to assume that overlapping feet are allowed in some languages, but this would not explain why also in those languages they are marked. Both Gordon and Hyde thus rely on an effect that can be understood as a form of TETU after all: some structures are disallowed in some languages and demonstrably dispreferred in languages which have them. This correlation is (part of) the classical definition of markedness, but parameter theory cannot express it. An interesting result in this respect is in the analysis of ternary rhythms by Elenbaas and Kager (1999). In some languages, stress does not fall on every second, but on every third syllable. A well-known example is Cayuvava (Key 1961): (12)

a. ˈda.pa b. ˈto.mo.ho c. a.ˈro.po.ro d. a.ri.ˈpi.ri.to e. ˌa.ri.hi.ˈhi.be.e f. ma.ˌra.ha.ha.ˈe.i.ki g. i.ki.ˌta.pa.ra.ˈre.pe.ha

‘canoe’ ‘small water container’ ‘he already turned around’ ‘already plants’ ‘I have already put the top on’ ‘their blankets’ ‘the water is clean’

Parametric theories will typically need to introduce some special device in order to describe such languages, such as ternary feet or (parametric) “weak local parsing” (Hammond 1984; Hayes 1980, 1995). This implies that the parametric system somehow needs to be extended to incorporate languages of this type. Elenbaas and Kager (1999) however propose that nothing special has to be posited in order to derive such languages. They show how the Cayvava pattern can be derived from the interaction between the constraints Lapse (above), AllFt-R and Parse which all have an independent justification: (13) a. AllFt-R: Every foot should be at the right edge of the word (give a violation for every syllable between the right edge of a foot and the right edge of a word) b. Parse: Every syllable should be inside a foot.



PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

347

(14) σσσσσσ

AllFt-R

Parse

a. ☞ σ(σˈσ)σ(σˈσ)

***

**

b. (σˈσ)(σˈσ)(σˈσ)

****!*

c. (σˈσ)σ(σˈσ)σ

****!

**

**

**

d. σσ(σˈσ)(σˈσ)

*Lapse

*!

Since AllFt-R gives a violation for every foot that is not at the right edge, it will be violated by any word that has more than one foot. In this sense, then, it can also do the work of the unboundedness parameter, giving us languages with only one foot per word. The constraint Parse, on the other hand, prefers all syllables to be parsed in a foot structure, something we need for languages that have exhaustive parsing anyway. The independent use of Lapse we have already seen. When they are ranked in the order Lapse >> AllFt-R >> Parse, these constraints turn out to “predict” weak local parsing.10 We thus do need an extra parameter to account for ternary stress, but not an extra constraint, because we can understand the extra language as a TETU effect. I have pointed out above that adding one constraint to an existing hierarchy of constraints gives much more power than adding a binary parameter (at least if the existing hierarchy has at least two constraints). Together, these two factors imply that it is not so clear which of the approaches is better, because the counting of theoretical parsimony becomes rather difficult. Parameter theory would need an extra device, but can add this in a relatively cheap way; constraint ranking theory does not need an extra device, but already contains a rather vaster typological space. The structure of this typological space is a topic that has been studied quite widely in the literature on stress. It seems to me that, as in the days of parameter theory, stress still is one of the best explored topics in phonological variation under OT, and maybe in language variation generally. It is a typical requirement—although not a standard practice—for any OT paper to explore the factorial typology which is predicted by its constraints. There is also software to do so: two packages which can calculate the typological predictions of a given constraint set are OTSoft11 and OTWorkplace.12 It is clear that the authors of at least the latter have stress typology in mind: their example constraint set contains a number of constraints describing stress. Also learning theory has developed quite extensively within OT, and profited from computer implementations; OTSoft for instance, provides a model for learning OT grammars, as does Praat.13 Different from the parametric studies we observed, learning models are not usually cue-based, but more often error-driven: they assume a grammar and compare it to the actual data of the language, fixing the assumptions when things go wrong, i.e. when the outputs of the grammar are different from the observed data. Optimality Theorists have been rather successful in their typological work, as well as in the development of acquisition models; more work has been performed in

348

CONTEMPORARY LINGUISTIC PARAMETERS

this model than in any other of these topics than in any alternative theory, it seems. An interesting recent development is in unpublished work by Birgit Alber and Alan Prince, who study the structure of OT typologies based on a few very simple stressrelated constraints, and show how we can learn things from those typologies quite easily.

14.5. Stochastic ranking The last decade of phonological research in the phonological mainstream has been characterized by a gradual shift away from categorical models towards more gradience-oriented views of language. This is also having its effect on the study of stress. As far as I am able to tell not a lot of work has been done along these lines, but we may expect the importance of this line of work to grow. An important focus point of this line of work is its emphasis on weak generalizations: languages do not altogether classify completely within a particular typological niche. An example of this can be found in Italian. Cei and Hayes (2012) note that there is controversy whether or not this language has quantity-sensitive stress. For most words with a penultimate heavy syllable, stress is on that syllable. If the penultimate syllable is light, however, stress is (usually) on the antepenultimate syllable. However, there are “half a dozen” words which have a heavy penultimate syllable, yet stress on the antepenult (see also Kraemer 2009): mandorla “almond,” acanto “acanthus,” polizza “little note,” and the place names Otranto, Lepanto, and Taranto. How should we evaluate these “exceptions”? Den Os and Kager (1986) argue that these examples show that Italian is not quantity-sensitive after all. D’Imperio and Rosenthal (1999), on the other hand, argue that these are truly exceptions which fall outside of the system proper, which therefore is quantity-sensitive. Cei and Hayes (2012: 2) say that their opinion is “somewhere in between”: heavy penult is an exceptionally strong factor for semi-predicting Italian stress. They work with a computational model that can deduce surface-based constraints from a set of real Italian words with a stress marking. Those surface-based constraints get some weight assigned to them which is represented as a positive real number. In the evaluation of a new word, the different values for each of the constraints is calculated according to some formula which gives a set of new real numbers, which are each to be interpreted as the likelihood that stress is on some syllable of the word. There is already quite a long-standing tradition in generative phonology studying the way in which lexical exceptions can be fitted into the overall formal theory (see Zonneveld 1978, for an early example). In previous theories, the solution was typically a representational one: the exceptional forms would have some kind of “hidden” structure that would make them behave differently. Formal markedness would correspond to frequency: the more hidden structure, the less likely a form is. Ironically, it is not completely clear how exceptionality works out in the newer approaches: stress would still have to be marked on all forms, but how would that relate to the relevant frequency of forms? However this may be, the constraints can refer to many different aspects of the phonological shape of the word. There is no assumption that this is constrained



PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

349

beforehand, e.g. by foot structure. Instead, anything that could possibly be categorized is taken into consideration, such as the complexity of the rhyme, but also of the onset, as well as the quality of the vowel. These latter factors play an important role in this type of work; see for instance also Ryan (2011) for a more extensive discussion of the role of onset complexity in the role of stress. The important point is that such complexity rarely (if ever) plays a decisive role in the assignment of stress. For traditional parametric theories, this means that we have to assume that they do not play a role in the typology and therefore also not in the grammar of the languages in question. It is therefore the goal of the recent types of work to show that native speakers do indeed show knowledge of these patterns in experimental situations. It may be presently too early to evaluate this evidence, but the development is definitely an interesting one. We could characterize it as a development in the direction of micro- or even nanotypology: stress in a language is not just characterized by a small set of universal parameters of constraints, but by a relatively large number of partly language-specific constraints that can have their origin in universal preferences as well as in the accidents of the history of the individual language. One question that may arise is why some of such constraints (such as those about the structure of the rhyme) make it to the macrotypology, whereas others (such as those about the structure of the onset) do not. The argument would typically be that this is grounded in extragrammatical factors, such as phonetics (in this case the fact that rhyme complexity may be more salient and add more to the length than onset complexity). One could wonder, however, why people would have this kind of stochastic knowledge at all, and why it would be represented in this way. Given that the knowledge is stochastic, people are not completely sure where to put stress when confronted with a new word—if they do not hear that stress to begin with. If they do hear a new word, on the other hand, they will already know where the stress is located. Notice that is partially different under more standard accounts, in which the child learns a stress pattern which then becomes redundant and does not have to be lexically stored. Learning stress is then thus a matter of economy. Under both approaches, the stress system seems most useful in parsing, because it makes it predictable where word boundaries are (e.g. if a language has initial stress on every word, you know that a new word has started once you hear a stress). However, if stress is lexical or only stochastic, it also becomes less useful as a marking device. I think it is safe to say that an important result of decades of work into variation in stress systems therefore is that the existence of stress is something of a mystery.

14.6. Conclusion We can thus conclude that the fact that languages seem to have stress systems at all, and that speakers have knowledge about those systems, is not completely understood. We can add to this an interesting observation made by Hyman (2008): there are languages which do not seem to have word stress at all; but when a language

350

CONTEMPORARY LINGUISTIC PARAMETERS

does have stress, then all lexical words will have it (clitics and other function words of course do not have to have it). Hyman (2008) notes that this is in itself a remarkable fact, given that there are languages with rather complicated stress systems and default rules: stress the rightmost heavy syllable in the word, and if there is no heavy syllable, stress the leftmost light syllable. However, there is no language which would have as a rule “stress the rightmost heavy syllable in the word, and if there is no such syllable, do not assign stress.” This seems to point then in the direction of at least one big macro-parameter: LEXICAL-STRESS (YES/NO). This parameter also makes clear sense from a learning perspective: when switched on, the child knows that she will have to look for stress in every word she encounters. This cannot really be expressed in terms of violable constraints, let alone in terms of stochastics. This widens the gap between the macrolevel and the microlevel of variation even more. It seems fair to see that formal theory has brought a lot of insight to our understanding in particular of the macrotypology of stress systems in human language. Although obviously, many questions and problems remain, existing theories have gained a broad empirical coverage, and this has definitely contributed to the success of parameter theory first and Optimality Theory afterwards. I submit that one important goal for phonological theory in the next decades is to try to find a coherent framework which can express the different kinds of insights of the different approaches that have been tried in a natural way: microand macro-variation, the fact that most variation seems to fit in a system of simple binary choices and the fact that some of these choices seem to be seldom or even never chosen.

Notes 1

Many thanks are due to two anonymous reviewers for their thoughtful comments.

2

http://st2.ullet.net/

3

http://wals.info/feature

4

In some languages main stress seems to be assigned from a different edge than secondary stress; English may be a case in point (Hammond 1999). This topic will not be discussed here.

5

Government Phonology (Kaye et al., 1985, 1990, GP) is a framework which is modeled to a large extent on Government and Binding Theory in syntax, including representational parameters. There is not a lot of work on stress within GP. An exception is (Scheer and Szigetvari 2005), which present purely representational parameters to account for the weight of different kinds of “heavy” syllables.

6

There has been some work also on the feasibility of parameter theory of L2 acquisition. See for instance Archibald and Pater (1995).

7

There is quite some literature arguing for the existence of uneven trochees. See for instance Riad (1992), Mellander (2003) and Morén-Duollja (2013).

8

The problem is obviously complicated by the fact that sometimes a high-ranking constraint can make the ranking among lower-ranked constraints irrelevant. For







PARAMETERS IN PHONOLOGICAL ANALYSIS: STRESS

351

instance, a reviewer points out that by highly ranking a constraint in favor of lexical placement of stress, all constraints about alignment, syllable weight, etc. are becoming unordered with respect to each other. And the same is true for parameter theory: setting the constraint on lexical stress “on,” constraints on alignment, etc., become irrelevant. 9

A reviewer suggests to take a look at “degenerate” feet in languages which allow them, by virtue of a low-ranking FootBin constraint. Such degenerate feet still typically count as marked, something which cannot be independently explained.

10 A slightly different approach is taken by Martinez-Paricio (2013), who argues in favor that ternary feet are actually “recursive,” branching into a weak and a strong branch twice. Also in this case, the recursive feet are presented as “last-resort devices,” “to avoid degenerate feet and ensure exhaustive parsing of syllables.” The idea of recursive feet was also proposed in parametric frameworks, however, as Martinez-Paricio (2013) aknowledges. See for instance Van der Hulst (2010) for an overview; and see Bennett (2013) for another proposal of recursion as a “last-resort” device. 11 http://www.linguistics.ucla.edu/people/hayes/otsoft/ 12 https://sites.google.com/site/otworkplace/ 13 http://www.praat.org/

15 Parameters in language acquisition and language contact Nina Hyamsa, Victoria Mateua, Robyn Ortfitellib, Michael Putnamc, Jason Rothman,d/e and Liliana Sánchezf University of California, Los Angeles,a University of Sheffield,b Penn State University,c University of Reading,d UiT, the Arctic University of Norway,e and Rutgers Universityf

15.1. Introduction: Language acquisition and linguistic theory The Principles and Parameters model (P&P) of Universal Grammar (UG) (Chomsky 1981) seems ideally suited to address the logical problem of language acquisition— how human speakers come to know as much as we do based on limited language experience. As Jackendoff (2011: 268) notes: “the primary goal of modern linguistic theory […] is [to provide] an explanation of the human language capacity and how it enables the child to acquire adult competence in language.” Parameter theory addresses the logical problems by vastly simplifying the language acquisition process: the task of the language learner is to choose among competing (ideally binary) values along an array of antecedently given parameters. The “stages” in acquisition thus represent the instantiation of particular parameter values, correct or incorrect vis-à-vis the target grammar. If a parameter is set incorrectly, it must eventually be reset based on “triggers” in the input, and the resetting gives rise to a new “stage” or grammar. In this chapter, we explore the role of parameters in

354

CONTEMPORARY LINGUISTIC PARAMETERS

current P&P research in first- (L1) and second-language (L2) acquisition contexts as well as contact and heritage grammars. We focus on argument omission parameters (null-subject / null-object parameters), because they provide a good illustration of the general conception of parameter theory (as well as its virtues and shortcomings) and because they are also among the most empirically robust and best studied parameters in terms of acquisition. We begin in the following section by discussing L1 acquisition, comparing grammar-based, parameter(-missetting) hypotheses to performancebased accounts of missing arguments in child language, first null subjects and then null objects. The focus of Section 15.3 shifts to adult L2 acquisition, in particular the “(re)setting” of argument omission parameters. The P&P framework addresses two essential questions concerning L2 acquisition: (i) to what extent is UG accessible to learners after the so-called critical period, and (ii) what is the influence of the native language on L2 development and ultimate attainment? We also discuss performance-based explanations for some L1/L2 differences. Section 15.4 concludes our treatment of argument omission parameters with a discussion of contact and heritage grammars, where we consider the role of UG principles and parameters and also the effects of performance factors in accounting for the outcomes of heritage language learners whose grammars are often qualitatively different from those of L1 children, despite the fact that they acquire language naturalistically in early childhood.

15.2. Parameters in L1 acquisition Parameter theory satisfies several theoretical desiderata with respect to L1 acquisition: First, children’s grammatical “rules” and “errors” are not random nor do they arise from otherwise unmotivated principles. Deviations from the adult target grammar are constrained by the parameter space of UG, much in the way grammatical variation across adult languages is constrained. Parameter theory thereby makes precise the claim that child grammars are not fundamentally different from adult grammars (cf. Hyams 1983; Klein 1982; White 1981), a hypothesis now referred to as the “continuity hypothesis” (Pinker 1984). Second, the deductive structure of parameters subsumes what would otherwise be disparate grammatical properties that would have to be individually learned helping to explain the speed and ease of acquisition, and third, parameter (re) setting provides a partial solution to what Felix (1987) called the “stage-transition question,” viz., what accounts for the transition from one grammatical stage to the next?

15.2.1. Missing subjects in child language The missing subject phenomenon in child language is illustrated in (1): 2–3-year-old children acquiring non-null-subject languages such as English, Danish, and French nevertheless optionally omit them:



PARAMETERS IN L ANGUAGE ACQUISITION AND L ANGUAGE CONTACT

355

(1) a. Want more apple [English] (Brown 1973) b. Ikke kore traktor [Danish] (Hamann and Plunkett 1998) not drive tractor ‘(I, you, he) doesn’t drive the tractor’ c. A tout tout tout mangé [French] (Hamann and Plunkett 1998) has all all all eaten ‘(He) has eaten everything’ There have been various parameter accounts of the null subject (NS) stage in child language, beginning with Hyams’ (1983, 1986) proposal that all children start out with the “Italian” setting of the pro-drop parameter (Rizzi 1982). This analysis captured the missing subject phenomenon and a number of seemingly related properties, but ultimately failed on empirical grounds. In particular, a direct comparison of English- and Italian-speaking children at similar grammatical levels show differences both in frequency and distribution of null subjects (Valian 1991). The English-speaking children showed far fewer NSs (30% vs 70% for Italian children) and far more overt pronouns than would be expected if they were speaking a true pro-drop language. It was also observed that in English NSs failed to occur in subordinate clauses or in (finite) post-wh environments, in marked contrast to Italian child language (Guasti 1996). Similar root/first position effects were found for French, Dutch and German-speaking children (Clahsen, Kursawe, and Penke 1995; Crisma 1992; Levow 1995; Hamann 2000; Haegeman 1995). In a further twist, De Haan and Tuijnman (1988) showed that Dutch and German children—in contrast to English and French-speaking children—also dropped objects from first position, in what looked more like topic drop than pro-drop, an option that exists in adult V2 languages as well. The rise and fall of this early proposal provides a good illustration of the testability or falsifiability of parameter models of L1 development: A child language with a hypothesized parameter value p should approximate an adult (or child) language whose true value is p. More recently, other parameter models have been proposed to account for the NS phenomenon (Hyams 1992; Jaeggli and Hyams 1988; Yang 2002; see Hyams 2011 for review), the most far-reaching of which is Rizzi’s (2005) “root null subject parameter” (RNS) account, which as the name suggests, focuses on the NS phenomenon in non-pro-drop languages showing the root restriction. The RNS parameter specifies that a subject may be null in the specifier of the root.1 Rizzi shows that RNS is also a property of some adult languages, for example certain varieties of Brazilian Portuguese. Thus, children born into languages that have no pro-drop or topic drop options will nevertheless drop subjects in root contexts. The various parameter models differ from each other in grammatical details, and in other respects as well. For example, Yang’s “variational” model incorporates a statistical component and Rizzi’s assumes children’s grammars set an initial null-subject setting under pressure from a computational strategy favoring parametric values that reduce the load on the production system (null subjects are computationally less costly than overt subjects, by hypothesis). Despite these differences, all parameter models argue that children’s grammars differ from the respective adult (non-null-subject) targets in licensing null subjects as a grammatical option.

356

CONTEMPORARY LINGUISTIC PARAMETERS

An alternative perspective is provided by various performance accounts proposing that subject omission is solely an effect of children’s limited production abilities and not a property of their grammar (L. Bloom 1970, P. Bloom 1990, Valian 1991). The performance hypothesis is also consistent with the root property of null subjects: if sentence initial position is the locus of sentence planning then this might create a processing bottleneck which is lessened by dropping the subject, Also, Bloom (1990) found that in English-speaking children the VP length of their utterances decreases as a function of subject “heaviness” with the VP longest when the subject is missing: his hypothesis is that subject omission allows for increased resources to be deployed to the VP.2 In a similar vein, Gerken (1991) has proposed that children’s productions are constrained by a metrical template, favoring (the prevalent in English) trochaic (S-w) over iambic (w-S) feet. This leads children to disproportionately drop pronouns in subject position (he-SINGS) over object position (SEES-him).3 Parameter models do not easily account for the VP length effects.4 At the same time performance accounts do not readily provide a basis for explaining certain syntactic contingencies, for example, the fact that in most languages subject omission is much more frequent in root infinitives than in finite clauses during the same period (see Hoekstra and Hyams 1989 for review of relevant findings). A crucial difference between parameter and performance accounts relates to children’s sentence comprehension: if production constraints are responsible for missing subjects and children do not have a NS grammar then they should reject null subjects in comprehension. Conversely, if they have a NS grammar it would underlie both production and comprehension and thus children who drop subjects in production should also accept them in comprehension. In the next section we report the results of a comprehension study on null subjects.

15.2.2. Comprehension of null subjects in L1 In adult English, null-subject sentences can be interpreted only as imperatives. If children in the NS stage have a grammar that also licenses null-subject sentences as declaratives, we expect them to accept such sentences in comprehension in both imperative and declarative contexts. Orfitelli and Hyams (2012) tested English-acquiring children’s comprehension of null-subject sentences using a modified version of Truth-Value Judgment (TVJ) experiment (Crain and McKee 1985; Crain and Fodor 1993). The experimental scenarios consisted of a story about a pair of pictures. The first picture always showed two older children engaged in a particular activity, such as drawing a picture or playing with blocks, while the second picture always showed two younger children in close proximity to the relevant items (e.g. paper and crayons, or blocks) but not interacting with them. Participants were told that while the four children have the same babysitter, only the younger children had to wait for the babysitter to tell them what to do. The older children are old enough to choose their own activities without permission, and because of this, the babysitter should not tell them what to do. This sets up a mood-based dichotomy in which the pictures of the older children are compatible only with declarative sentences, but not imperatives (because they



PARAMETERS IN L ANGUAGE ACQUISITION AND L ANGUAGE CONTACT

357

are autonomous), while the pictures of younger children are compatible only with imperative sentences (because they are waiting to be told what to do).5 Following the presentation of the story, one of the two pictures was removed, and the participant was asked to judge the appropriateness of one of five sentence types as applied to the remaining picture: habitual declarative sentence (2a), progressive declarative sentences (2b), please-imperative (3a), vocative imperatives (3b), and null-subject sentences (4).6 (2) a. They always play with blocks b. Now they are playing with blocks (3) a. Please play with blocks b. Hey kids, play with blocks (4) Play with blocks Thirty participants were included in the final analysis, 10 each in 6-month intervals between 2;6 and 4;0. All 30 performed well on declarative and imperative trials of both sub-types, showing that they understood that the imperative mood could only be applied to the pictures of the younger children and declarative mood only to the pictures of the older children. On the NS condition, however, a logistic regression model found a significant difference between the youngest and middle age groups (p = 0.013), as well as the between the middle and oldest age group (p < 0.0001). Until approximately 3;6, children were allowing NS sentences to have a non-adult declarative interpretation in addition to the adult imperative interpretation, exactly matching the patterns seen in NS production. These data suggest that the NS stage is caused by a principled grammatical difference between child and adult English, and cannot be explained solely on the basis of processing. When the data from the younger two groups of participants was examined in detail, a striking pattern emerged. Although they accepted NS sentences as imperatives and declaratives, the children in the study did not merely answer true to every NS item. Instead, they treated the NS items as if they were declaratives in approximately 50% of the trials, and provided follow up justifications consistent with this interpretation. On the other 50% of the trials, they interpreted the NS sentences as imperatives, again with appropriate follow up justifications. This suggests that they were not using the context of the scenario (declarative versus imperative) to decide which structure to project. Instead, the children appeared to be resolving the syntactic ambiguity in one direction or the other, and only then considering whether this mood would match the scenario that had just been presented. Why are children unable to integrate the contextual (or other) information necessary to assign the correct interpretation? This may be where processing factors may come into play. Unlike adults, Englishacquiring children initially permit two structures for NS sentences, rendering them ambiguous. When interpreting the meaning of an NS sentence, then, children have to decide between a declarative and imperative representation for the sentence, evaluate the representation relative to the context of the situation, and revise their representation when necessary. This revision process is precisely the kind that children have been previously shown to have difficulty with. In studies of both

358

CONTEMPORARY LINGUISTIC PARAMETERS

lexical (Swinney and Prather 1989) and sentence-level (Trueswell et al. 1999; Syrett and Lidz 2005) ambiguities, children appear to consider only a single option, and have difficulty focusing on multiple cues to lexical or syntactic interpretation. In contrast, the adults in these studies not only show evidence of initially considering both possibilities, but are also able to integrate multiple types of evidence (e.g. lexical and referential information, intonation, etc.) to arrive at a meaning. It seems that in contrast to adults, who can revise incorrect parses to reflect additional contextual and other information, children are unable to revise an incorrect interpretation, even in the face of disambiguating contextual evidence. When children exit the NS stage at approximately 3½ years, their grammar changes, and the processor is no longer faced with the problem of ambiguity resolution. Overall, we have seen that an adequate explanation of the NS stage must reference a fundamental difference between child and adult grammars. Performance factors, however, may also play a role in children’s resolution of the mood ambiguity associated with NS utterances. Children’s limited processing resources may not allow them to integrate all sources of information needed to resolve the ambiguity, as seen in studies of ambiguity in other parts of the grammar.

15.2.3. Missing objects in child language Cross-linguistic studies of object omissions in child language are sparser than those for subject omission. However, there is wide agreement that object omission manifests itself differently and to a different extent, in Romance languages than in English. Studies of English child language typically report object omission at very low frequencies (under 10%) and only at the earliest stages of development (Valian 1991, Wang et al. 1992). In Romance languages, on the other hand, it has been widely documented that null objects constitute a non-negligible portion of children’s utterances in contexts where, by adult standards, an object clitic would have been expected. In Spanish, when a direct object denotes a definite or specific referent that is salient in the discourse, it is typically expressed with an object clitic. Clitics are marked for gender and number and their distribution is limited to positions immediately preceding a finite verb (5a) or immediately following a progressive participle, infinitive, or imperative (5b):7 (5) Context: Talking about some dogs a. Anita *(los) lava Annie them washes b. Anita está lavándo-los Annie is washing-them ‘Annie is washing (them)’

[Spanish]

Persistent object clitic optionality was first observed among typically developing French-speaking children (Clark 1985; Grüter 2006; Müller et al. 1996; PérezLeroux et al. 2008), but the phenomenon has also been found in Italian (Guasti 1993/94; Schaeffer 1997; Tedeschi 2009), Romanian (Avram 2000; Babyonyshev



PARAMETERS IN L ANGUAGE ACQUISITION AND L ANGUAGE CONTACT

359

and Marin 2004), Catalan (Gavarró et al. 2010; Wexler et al. 2004) and Spanish (Bedore and Leonard 2001; Castilla and Pérez-Leroux 2010; Fujino and Sano 2002, Mateu to appear). In all these studies suppliance rates for object clitics seem to reach a productive level (viz. 90%) only between the ages of four and five, converging on the finding that object clitics typically appear later than other functional elements, such as subject clitics and definite determiners. As with the NS stage, theoretical explanations for this inconsistent use of object clitics can be divided into those that attribute the difficulty with object clitics to the competence domain and those that attribute it to the performance domain. Representational accounts claim that the child’s early grammar differs from the adult grammar in that it can syntactically represent a sentence with a null referential object. Müller et al. (1996) and Müller and Hulk (2001) propose a parameter account in which early Romance grammars allow referential null objects in the same way topic-drop languages like Chinese do. In a different account, Pérez-Leroux et al. (2008, 2012) propose that children’s overgeneration of referential null objects results from their failure to restrict the null structure to the appropriate context, i.e. non-referential contexts. A different account is that of Schaeffer (2000), who attributes the delayed acquisition of clitics to a deficit in the child’s pragmatic system which in turn leads to the optional marking of referentiality / specificity and the resulting null-object constructions. Other authors have argued that object clitic omission in children is evidence of their computational limitations or their immature performance system, rather than to a divergent grammatical representation or constraint. For example, Jakubowicz and Rigaut (2000), and Prévost (2006) argue that it is the placing of (pre-verbal) clitics in a non-canonical argument position that creates computational problems for children. More recently, Grüter and Crago (2012), and Mateu (to appear) claim that producing (pre- or postverbal) clitic constructions requires more complex operations and more working memory resources than creating transitive constructions with full DPs, leading to children’s inconsistent use of object clitics. Both types of accounts lead to an expectation that children will omit object clitics in production. But only processing accounts specifically claim that children’s verbal working memory will be a good predictor for their rate of clitic omission. Additionally, as with the NS case, if clitic omission results from an option available in the child’s grammar, we predict that children will accept referential null objects in comprehension. On the other hand, performance accounts predict that children who omit clitics in production will reject referential null objects in comprehension, as clitic omission is due to processing or working memory limitations. In the following section, we discuss results from a study testing these predictions.

15.2.4. Comprehension of null objects in L1 Mateu (to appear) addressed these questions in a study investigating object clitic omission in Spanish-speaking children aged 2–4 (n = 32; mean = 3;5). In her elicited production study, she found that 2- and 3-year-olds omitted clitics at high rates, and that at this same age they also overused full DPs, as illustrated in Figure 14.1.

360

CONTEMPORARY LINGUISTIC PARAMETERS

FIGURE 14. Elicitation Task: rate of clitic omission, clitic production and full DPs in Mateu (to appear).

When investigating the relationship between object clitic omission and several other independent linguistic and non-linguistic measures, including mean length of utterance (MLU), subordination index, vocabulary size (i.e. number of different words), verbal working memory (i.e. non-word repetition span), and age, only one variable showed a significant predictive value for clitic omission: verbal working memory (p = 0.001). These results suggest that limited verbal working memory may impair clitic production, and also that children may produce full DPs as a compensatory strategy, at the expense of violating pragmatic rules. To test comprehension, Mateu used a sentence-picture matching task designed such that the child with a null object grammar would allow the target sentences (e.g. Diego vuela “Diego flies”) to match a transitive picture/interpretation (e.g. Diego lo vuela “Diego flies it”). Results showed that regardless of the high clitic omission rates in the elicitation task in the two younger groups (Figure 14.1), no individual child assigned a transitive interpretation to an intransitive scene in more than 1/6 of items, paralleling Grüter’s (2006) results for French. On the other hand, children’s performance on the clitic conditions (6) was at chance in the younger two groups, despite perfect performance in the control conditions (full DPs).  Notably, performance was significantly lower when the verb was longer, i.e. progressive (6a vs 6b) and when there was an additional constituent at the end (6c vs 6d) (cf. Bloom 1990; Valian 1991). (6) a. Diego lo vuela muy alto Diego it flies very high ‘Diego flies it very high’ b. Diego lo está volando muy alto Diego it is flying very high ‘Diego is flying it very high’ c. Diego está volándo-lo Diego is flying- it ‘Diego it is flying’



PARAMETERS IN L ANGUAGE ACQUISITION AND L ANGUAGE CONTACT

361

d. Diego está volándo-lo muy alto Diego is flying- it very high ‘Diego it is flying very high’ Neither of these results is predicted under the hypothesis that children have a null object grammar. Further analyses confirmed that children’s performance in the clitic conditions was again predicted by the verbal working memory score, consistent with the hypothesis that clitic omission is in fact affected by processing limitations, in line with performance accounts. The results from these studies thus suggest that the linguistic principles that govern object clitic constructions in Romance-speaking children are adult-like from the beginning, but the means for integrating clitics may require more refined and developed mechanisms of memory and language processing. In the section that follows, we shift our attention to the role of parameters in L2 acquisition in adults. Most importantly, we address a question that has been at the heart of L2 acquisition research from a P&P-perspective since its inception; namely, to what extent can parameter settings be (re)established in a developing grammar?

15.3. Second language acquisition and linguistic theory As in the L1 studies highlighted in the previous section, the P&P model of UG has significantly informed studies of second language (L2) acquisition, especially adult L2 acquisition (see White 1989, 2003 for detailed review and also Haznedar and Gavruseva 2008, Haznedar 2013 for review of child L2 studies). Adult L2 learners face a logical problem similar to L1 learners—they come to know much more about the L2 than can be deduced solely from the input they are exposed to, and this knowledge crucially cannot be explained on the basis of L1 transfer (e.g. see Rothman 2008, Schwartz 1998, Schwartz and Sprouse 2013). As is its appeal for theories of child L1 acquisition, the parameter model also provides a way to address the poverty-of-the-stimulus problem in adult L2. Under this approach, the task of adult learners is theoretically no different from children’s; they must select for his L2 the appropriate value among competing ones for the same array of antecedently given parameters. While it is understood that the child’s initial state of acquisition is UG, the initial state for adults and their ability to access UG is less clear. Thus, two questions dominated the early L2 field: (a) do adults continue to have access to UG, and (b) what—if any—is the role of the L1 in L2 development? With respect to parameters, the question becomes: is the initial state of the parameters in L2 acquisition like the child’s (unmarked UG settings—if UG is accessible at all) or is it the settings of the L1? The answer to this question is of great consequence for understanding not only the starting point of the L2 acquisition process, but also its development and ultimate attainment. Not surprisingly, central themes within the early years of generative L2 acquisition theory and beyond have been concerned with initial state modeling and hypothesizing about the extent of UG-accessibility in adulthood. Several initial

362

CONTEMPORARY LINGUISTIC PARAMETERS

stages models within the P&P framework were advanced in the 1990s, including No Transfer / Full Access (Epstein et al. 1996), Full Transfer / Full Access (Schwartz and Sprouse 1996), Minimal Trees (Vainikka and Young-Scholten 1996), Valueless Features (Eubank 1993), among others. These models argued for various degrees of transfer at the level of parametric values (features and their specifications) from the L1. At the same time, they advanced various views of UG (in)accessibility in L2 development, ranging from no accessibility (e.g. Bley-Vroman 1989, 2009; Clahsen and Hong 1995), to limited/partial accessibility (e.g. Tsimpli and Roussou 1991, Hawkins and Chan 1997) through complete full accessibility (e.g. Schwartz and Sprouse 1996; White 1989, 2003). As is the case for child language, the application of the parameter model to L2 acquisition satisfies a number of theoretical desiderata. First, it predicts that adult L2 grammatical “rules,” stages, and “errors,” like children’s, will not be random, unmotivated or otherwise inconsistent with natural language. Although there is some disagreement among scholars who interpret available data differently, we maintain a “continuity” view of L2 grammars, viz. that deviations from the target L2 grammar are constrained by the parameter space of UG (White 2003, 2008). This view does not entail that L2 grammars are the same as L1 adult grammars; rather, the claim is that they are not fundamentally different. Second, the deductive structure of parameters is able to explain how L2 adult grammars, like child L1 grammars, project beyond the confines of the input (e.g. so-called L2 poverty-of-the-stimulus knowledge). Third, a parameter (re)setting approach might provide insight as to why some grammatical properties seem harder than others for L2 learners to acquire, for example, when L1 transfer of parameter setting results in a language that is a superset of the “target” L2. Finally, the parameter approach has provided a framework within which various hypotheses can be formulated and from which precise predictions can be made, often shedding light on complicated and seemingly variable data. The tension between performance and competence explanations has also featured prominently in adult L2 acquisition. For example, in the production of obligatory inflectional morphology L2 learners consistently make both omission and commission type errors. When L1 transfer fails to explain these errors, we are faced with the question of whether these errors are performance-based or reflect a different grammatical representation. Several performance accounts have been put forward, including the Missing Surface Inflection Hypothesis (MSIH) (Prévost and White 2000) and the Interface Hypothesis (IH) (Sorace 2011). The MSIH maintains that L2 learners have difficulty mapping syntactic function to morphophonological form in production. This predicts asymmetries between production and comprehension in much the same way performance-based explanations do for L1, as discussed in the previous section. By appealing to possible processing limitations inherent in managing more than one linguistic system (e.g. the tension of inhibitory control and its release, as well as limited attentional resources), the IH explains difficulties with discourse integration that exist even at the highest levels of L2 proficiency. We can see from this discussion that hypothesizing about performance variables at the level of L2 production is also a central theme in P&P approaches to adult non-native acquisition.8 In this section, as in the L1 section, we focus on argument omission parameters (null-subject / null-object parameters) and their possible role, and interplay with



PARAMETERS IN L ANGUAGE ACQUISITION AND L ANGUAGE CONTACT

363

other deterministic variables, in explaining the facts of L2 acquisition. We begin with the literature on null subjects, which is perhaps the most widely studied linguistic domain in generative L2 research and then turn to the less studied domain of null objects.

15.3.1. Missing subjects in adult L2 The null-subject parameter has long been seen to support both the full transfer and full access hypotheses, based on the interplay between null-subject languages such as Spanish and non-null-subject languages such as English. We see a different developmental trajectory depending on which type of language is the L1 and which is the L2. The case of L2 learners of Spanish who have a [– pro-drop] L1 is relatively straightforward. These learners show mastery of the syntactic environments in which null subjects are permitted—a finding consistent with the availability of parameter resetting, hence full access to UG. However, these learners do not necessarily reach fully native-like attainment on null subjects. In particular, they show difficulties with appropriately restricting the pragmatic licensing of null arguments in both productive and receptive tasks until extremely late in development (Al-Kasey and Pérez-Leroux 1998; Liceras 1989; Liceras and Díaz 1999; Liceras et al. 1999; Rothman and Iverson 2007). This pragmatic result has been taken to indicate some type of difficulty for L2 learners at the at the syntax-pragmatics interface (e.g. Sorace 2004; Sorace 2011; but cf. Rothman 2009). The data from L2 English learners with a [+ pro-drop] L1 raise an empirical puzzle, however, because their performance on certain receptive tasks differs from their production data. On grammaticality judgment tasks, low proficiency learners incorrectly accept ungrammatical referential null subjects in English (Davies 1996; White 1985, 1986). This acceptance declines with increasing English proficiency, consistent with an initial L1-setting of the parameter, followed by a resetting during development. Moreover, White (1985, 1986) finds that French-speaking learners of English accept significantly fewer null-subject sentences than Spanish-speaking learners—a contrast which is predicted under a transfer account, as French is not a null-subject language. In contrast to the judgment data, and unlike child L1 learners, however, adult L2 learners with [+ pro-drop] L1s produce very few NS sentences (in both spontaneous and elicited production), and do so only in the earliest stages of L2 development (e.g. Hilles 1986; Phinney 1987; Ruiz de Zarobe 1998). If acceptance of null subjects in judgment tasks is evidence of transfer, why do we not see the transferred L1 grammar reflected in production? There are two possibilities a priori: if the interlanguage grammar permits null subjects till late in L2 development, then learners’ failure to omit subject in production must be for some independent reason; alternatively, if the grammar does not license null subjects, then perhaps learners accept null subjects in the judgment task because of processing difficulties in their L2. These two possibilities lead to different predictions: if L2 learners permit null subjects in judgment tasks because of processing difficulties in their L2, and not for grammatical reasons, then in a comprehension task they should disallow a declarative interpretation of

364

CONTEMPORARY LINGUISTIC PARAMETERS

a NS sentence as do native English speakers. On the other hand, if transfer has occurred, and their interlanguage grammar is [+ pro-drop] then they should be able to interpret a subjectless sentence in English as a declarative. Using a battery of tests, including a grammaticality judgment task, a production task, and the same comprehension task used in Orfitelli and Hyams (2012) to test L1 English children, Orfitelli and Grüter (2013) tested whether null subjects are licensed in the grammar of L2 learners whose first language is [+ pro-drop]. They replicated previous judgment task results, finding that learners incorrectly accept NS sentences in approximately 30–40% of sentences, and this acceptance declines with increasing proficiency (r(15)= -.80, p