Representing Structure in Phonology and Syntax 9781501502224, 9781501510663

Citation preview

Marc van Oostendorp and Henk van Riemsdijk (Eds.) Representing Structure in Phonology and Syntax

Studies in Generative Grammar

Editors Norbert Corver Harry van der Hulst Roumyana Pancheva Founding editors Jan Koster Henk van Riemsdijk

Volume 124

Representing Structure in Phonology and Syntax

Edited by Marc van Oostendorp Henk van Riemsdijk

ISBN 978-1-5015-1066-3 e-ISBN (PDF) 978-1-5015-0222-4 e-ISBN (EPUB) 978-1-5015-0225-5 ISSN 0167-4331 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. 6 2015 Walter de Gruyter, Inc., Berlin/Boston Typesetting: RoyalStandard, Hong Kong Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Marc van Oostendorp and Henk van Riemsdijk 1 Introduction Josef Bayer Doubly-Filled Comp, wh head-movement, and derivational economy Norbert Corver Interjections as structured root expressions

7

41

Günther Grewendorf The internal structure of wh-elements and the diversity of wh-movement Jan Koster Relative clauses: Parallelism and partial reconstruction Alan Langus and Marina Nespor On the nature of word order regularities

115

141

M. Rita Manzini On the substantive primitives of morphosyntax and their parametrization: 167 Northern Italian subject clitics Jacques Mehler Language acquisition and the neuroscience of development Kuniya Nasukawa Recursion in the lexical structure of morphemes Marc van Oostendorp Final devoicing in French Markus A. Pöchtrager Binding in phonology

239

255

Ludmila Veselovská and Joseph Emonds The cross-linguistic homes of mood and tense Edwin Williams Multidomination and the coherence of anaphors Language index Subject index

337 339

211

277

315

195

85

Marc van Oostendorp and Henk van Riemsdijk

Introduction 1 The relevance of representations Formal grammars need two parts: a theory of computation (or derivation), and a theory of representation. While a lot of attention in recent decades has been devoted in mainstream syntactic and phonological theory to the former (both Minimalism and Optimality Theory are mostly devoted to issues of computation, ignoring representational issues almost completely), the papers in this volume aim to show that the importance of representational details is not diminished by the insights of such theories. No successful theory of computation can of course in the end neglect the issue what kinds of formal objects the derivation is operating on; and furthermore, the study of representations as cognitive objects is obviously a worthy pursuit in its own right. The reason for this is that our insight into computation ultimately has to rely on representations: we do not know what the outcome of the derivational procedure is if we do not know what the structure is of the objects which the procedure is working on. A brief example may illustrate this: the simple addition operation ‘+’ is not properly defined unless we know whether it operates on real numbers, on imaginary numbers or on strings (‘ta’ + ‘dum’ = ‘tadum’). We would posit that something similar holds for e.g. the operation Merge in syntax (Chomsky 2001, 2014), which supposedly is equally simple as addition, but which is also not properly defined when we do not know what is the nature of the structures which it is merging: are they features? Sets of features? Multidimensional bundles? Another straightforward example comes from phonological theory, in particular developments in Harmonic Serialism. (McCarthy 2010). This theory is a variation of Optimality Theory with one crucial difference: the Generator function does not generate an infinite set of candidate analyses for any given input, but can make only one change at a time to the input. The OT system then picks out the ‘optimal’ of all candidates with one such change in the familiar manner. This winner is then again input to the grammar, until no more ‘optimizing’ changes can be made. Obviously, in order for such a system to work, we need to have a definition what counts as one change: adding a feature? Adding an association line? Adding a feature and an association line at the same time? Etc. The relative popularity of the computational theory of Harmonic Serialism has thus necessarily drawn phonologists’ attention to representational issues.

2

Marc van Oostendorp and Henk van Riemsdijk

2 The basics of representational theory Having accepted, then, the necessity of understanding representations, a formal theory of those in turn has two aspects. First, there is a list of primitives – in linguistics, this might for instance be a list of permissible features such as [±N] or [±Labial]–, and secondly there is a set of structures in which these can be combined, – e.g. feature geometric structures. There is, in other words, a lexicon of primitives and a syntax of how these can be combined.

2.1 Representational primitives The form and structure of the primitives is far from settled in either phonology or syntax. Minimalism has led to a rethinking of the notion of syntactic label. The notion of ‘Merge’, already referred to above and which is now central in structure building, suggests considerable flexibility in the way syntactic nodes are labeled. Quite some work in recent years has gone into uncovering the properties of this labeling operation. But what are the primitives that labels consist of? Here there is little that goes beyond the categorial features ([±N, ±V] and perhaps some level features such as [±Projection, ±Minimal] or some such. What does it mean to say, for instance, that some features are phi features? Does it mean that features in turn have certain properties (they belong to the category ‘phi’, which would suggest that they have internal structure (features can recursively also have features, so that in this case the feature Person can have a feature phi)? Or does the notion phi feature refer to a certain node in a so-called feature geometry? And are there any other kinds of primitives that are not features; and if yes, what are they? And how do they relate to features? Are for instance the so-called formal features features in the same way as features which carry semantic content? As far as we know, there have been very few concrete proposals of a classification of the possible primitives on which Merge or the labeling function can operate. In phonology, on the other hand, there is a much stronger tradition in studying these primitives. At the same time, the points of disagreement are many. An important topic of discussion concerns the ‘substance’ of these features: do they crucially refer to articulatory or acoustic properties of speech sounds, or are they rather purely abstract labels which only become intepreted phonetically after all phonological computation? All three points of view still find their staunch defenders: articulatory interpretation of phonological primitives, for instance, in Articulatory Phonology (Browman and Goldstein 1986, Pouplier 2011); acoustic

Introduction

3

interpretations have been the hallmark of Element Theory (Backley 2011), and the abstract interpretation has been argued for extensively e.g. in Mielke (2011). Note also that there is a rather strong trend which denies the existence of any abstract primitives, arguing instead that cognitive representations of sound are very fine-grained ‘exemplars’ of phonetic representations (Beckman and Pierrehumbert 2003). There are also many other questions, independent of these interpretational issues. For instance, are all features monovalued and privative, or do we need to assume that some or all of them have more than one value? And if the answer to the latter is yes, is the number of values maximally 2, or 3, or unbounded? Are all features, or elements, equal from a formal point of view, or do we find asymmetries or even hierachies between them? The latter question leads us to another issue, viz. the syntax of phonological representations.

2.2 The syntax of linguistic representations At first sight, there seems to be slightly more consensus on the organizational principles that restrict the way in which primitives are bound together. An example is the tree, which for a while was seen as the default structure in both syntactic and (prosodic) phonology. The tree is still sometimes seen as simply the format in which syntactic structures are usually (and in an optically straightforwardly apparent way) represented. The same is true for prosodic structures such as the basic syllable tree: [ σ O [R N C ] ]. But as phonology shows, different ways of defining the basic elements of the syllable (Nucleus, Onset, Rhyme, or moras, or projections of the nucleus) and the ways the relations they entertain with each other may well lead to different conceptions of syllable structure, such as one in which the syllable rather consists of two smaller ‘moraic’ nodes; indeed the canonical syllable tree might turn out to be redundant given the presence of onset and rhyme nodes, as was argued for in classical Government Phonology (Kaye, Lowenstamm and Vergnaud 1987). In syntax, many tree-related notions have been subject to reinterpretation over the course of time. In recent years, for example, the impressionist notion of ‘canonical tree’ has undergone considerable changes with the advent of multi-dominance, parallel merge, parallel domains, overlapping projections, sideward movement, graft, etc. Another example is that the notion of what is a maximal projection is in need to be reevaluated against the background of the multiplicity of functional projections, leading to the question of what exactly an extended projection is.

4

Marc van Oostendorp and Henk van Riemsdijk

In phonology, on the other hand, the notion of the tree has been under a different kind of pressure, with scholars arguing for a variety of reasons that phonology does not need notions such as dominance or sisterhood at all, and that the relevant generalizations can all be framed by exclusively linear notions such as (immediate) precedence (see Van Oostendorp 2013 for critical discussion). One interesting argument put forward by those critics is that there should be a strict division of labour between phonology and syntax, where the former is stated only in terms of the precedence relation and its logical derivatives (like ‘adjacency’), the latter deals only with dominance, constituency and the like, and not with linear precedence relations. The other logical consequence of this point of view is that syntax never refers to linear precedence – which has indeed been proposed, e.g. by Chomsky 2001, 1013. The alternative view traditionally, of course, is that phonology and syntax have at least some of their formal apparatus in common.

3 Positioning the papers in this volume The relative importance of the two representational layers – the inventory and the structure – may be different in the different modules of grammar. Where phonology is rich in its inventory of elements, but poor in hierarchical structure, the situation in syntax appears to be the opposite. Does this mean, as some would have it, that phonology is fundamentally different from syntax, or is it conceivable that fundamental concepts about structure and labeling may carry over from one domain of grammar to the other? Does a modular view of grammar imply that each model has a completely different apparatus, or can different modules consist of very similar primitives? The issues are complex and this may be another reason that they are rarely discussed, next to the fact already mentioned, that the focus tends to be on computational issues anyway. The papers in this volume, however, present several attempts to clarify issues connected to this. Some papers show how interest in representation can clarify relatively new empirical domains. The chapter by Norbert Corver shows, for instance, that there are arguments for taking interjections more seriously as syntactic objects than is sometimes done. He argues in particular that their internal structure is that of bare roots, and that complex interjections can be formed that have the structure of coordinated structures. Other papers look more deeply into the internal structure of syntactic nodes that have been more well-studied and show how the microstructure of nodes can explain larger syntactic behavior such as movement. Grewendorf’s paper

Introduction

5

establishes for instance how different types of overt wh-movement correlates with specific properties of the internal structure of wh-elements. Bayer shows how assuming wh-elements can (sometimes) occur in head-positions makes their behaviour in Continental West-Germanic more understandable. Manzini shows in her paper that the differences between subject clitics in Northern Italian dialects can also be understood in terms of the feature make-up. Similarly, Veselovská and Emonds prove that differences between Czech and English verbal syntax can be understood by reference to different lexical entries for what can and cannot fill V and I nodes in the two languages. Some of the papers also look at the macrostructure of syntax more directly. Williams proposes solutions to two problems for applying multidominance analyses in syntax: interpretative problems for movement analyses, and overgeneration for coordinate sharing. Koster, a long-time proponent of representational solutions to problems that were solved derivationally in the mainstream, gives an analysis of raising phenomena and shows how current analyses of the phenomena are based on obsolete theoretical assumptions. Taking an even wider perspective, Langus and Nespor discuss word order variation in languages of the world, and argue that this is not primarily encoded in a set of parameters but rather the result of different kinds of interactions between autonomous linguistic (and cognitive) modules. Also the three phonological papers in this volume are primarily concerned with macrostructures in one way or another. All three give arguments for considering phonological structures that are very similar to what we know as trees in syntax. While Pöchtrager presents new arguments for trees within the segment in order to be able to address relations between certain vocalic elements that he compares to syntactic binding, Nasukawa argues for a radical revision of our view of recursion above the level of the phonological segment. Van Oostendorp, finally, gives arguments that a difference between Parisian French and Wallonian that at first sight seems purely segmental, should rather be analysed in terms of syllable structure differences between these Romance dialects. The paper by Mehler, finally, gives an interesting personal view on the history of the field, and the connections between experimental and theoretical work.)

4 Acknowledgements The present volume grew out of a Workshop entitled “The Representation of Structure in Grammar” which was held at the Annesso Cartesiano of Villa Salmi in Arezzo, Italy, on July 1–3 2011. The papers by Adriana Belletti & Luigi Rizzi,

6

Marc van Oostendorp and Henk van Riemsdijk

Martin Everaert, Riny Huijbregts, Hilda Koopman, Clemens Mayr, Masayuki Oishi, Dominique Sportiche and Timothy Stowell could not be included in the present volume for a variety of reasons. The workshop was financially supported by a subsidy from the T&GvR Foundation.

References Backley, Eugene. 2011. Introduction to Element Theory. Edinburg: Edinburgh University Press. Beckman, Mary and Janet Pierrehumbert. 2003. Interpreting ‘phonetic interpretation’ over the lexicon. In: John Local, Richard Ogden and Rosalind Temple (eds.) Phonetic interpretation. Papers in Laboratory Phonology VI. Cambridge: Cambridge University Press, 13–38. Browman, Catherine P. and Goldstein, Louis. 1986. Towards an articulatory phonology. In C. Ewen and J. Anderson (eds.) Phonology Yearbook 3. Cambridge: Cambridge University Press, pp. 219–252. Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.) Ken Hale: A Life in Language. (Current Studies in Linguistics 36). Cambridge: MIT Press, 1–52. Chomsky, Noam. 2013. What is language? Journal of Philosophy CX: 645–700. McCarthy, John. 2010. An introduction to Harmonic Serialism, Manuscript, University of Massachusetts. Mielke, Jeff. 2011. Distinctive Features. In Marc van Oostendorp, Colin Ewen, Elizabeth Hume, and Keren Rice, eds., Companion to Phonology. Wiley-Blackwell. Oostendorp, Marc van. 2013. σ strikes back. A defense of headedness and constituency in phonology. The Linguistic Review 30.2:347–371. Pouplier, Marianne. 2011. The atoms of phonological representations. In Marc van Oostendorp, Keren Rice, Beth Hume, Colin Ewen (eds). The Blackwell Companion to Phonology. WileyBlackwell.

Josef Bayer

Doubly-Filled Comp, wh head-movement, and derivational economy 1 Introduction Bavarian and Alemannic are South-German dialects in which the Doubly-Filled Comp Filter (DFCF)1 seems to be suspended. As some traditional as well as some recent empirical work has shown, however, this is not true in general. While uncontroversially phrasal wh-operators tend to require an overt complementizer, pure wh-words tend to reject it. The data are distributed roughly as follows: (1)

[CP wh-phrase [C’ C [ . . . ]]]

(2) *[CP wh-word [C’ C [ . . . ]]] (3)

[CP wh-word [C’ Ø [ . . . ]]]

In this contribution, it will be argued (i) that this distribution is not the result of an accidental spell-out convention, and (ii) that (3) is more successfully analyzed as in (4). (4)

[CP [C wh-word] [ . . . ]]

In other words, it will be argued that it makes a difference whether a wh-word or a wh-phrase undergoes movement, and that the former case may be analyzed as head-movement. In this case, the wh-element is simultaneously a complementizer. As a consequence, merger of a separate complementizer is superfluous and therefore forbidden. The article is organized as follows: After presentation of the core data in 2., it will be shown in 3. under which circumstances the wh-operator may move “head-style”. Head-movement will be shown in section 4 to be a preferred option because phrase structure can be extended on the sole basis of internal merger. The account will be supported in section 5 1 The term is kept although it dates back to a time in which S’’ was thought to dominate two complementizer positions. In the present article, we start out from the later assumption of a functional category C which projects a CP such that C can host a complementizer and SpecCP, a featurally matching specifier.

8

Josef Bayer

with novel data from cliticization to the C-position and complementizer inflection. Section 6 contains a short note on diachrony and cross-linguistic variation. Section 7 gives a sketch of how data from sluicing can be captured. Section 8 presents a note on chain uniformity. A conclusion follows in section 9.

2 DFC and DFCF in South German dialects with special emphasis on Bavarian The DFCF dates back to Chomsky & Lasnik (1977). It says essentially what (5) expresses. (5) *[α β], if α is in SpecCP and β in the C-Position of the same CP, and α and β are overt. This filter holds for many standard languages but may be the result of normative rules which often had no effect in older stages of a language, and which have no effect in many colloquial styles and dialects. Bavarian and Alemannic allow so-called “doubly-filled Comp” (DFC). The following examples show that in modern German, where it is stigmatized, DFC can be found even in the written language. (6) Ich habe mein Handy immer an die Boxen gehalten und 10 I have my cell phone always at the boxes held and 10 Sekunden später wusste ich, von wem dass der Song seconds later knew I from whom that the song wirklich ist. really is. ‘I held my cell phone always to the speakers and after 10 seconds, I knew from whom the song really is.’ http://hitparade.ch/interview.asp?id=55 [written German from Switzerland] (7)

dann then

musst must

du you

dich REF

ernsthaft seriously

fragen, ask

von from

wem whom

dass that

du you

etwas willst. something want. ‘Then you have to seriously ask yourself who you want something from’ http://forum.gofeminin.de/forum/teenfilles/__f3347_teenfilles-Mein-Freindverdammt-hubsch-und-der-Typ-fur-den-mein-Herzschlagt-hasslich.html

Doubly-Filled Comp, wh head-movement, and derivational economy

9

In previous generative work, DFC was taken to be an unconditional option in these dialects (Bayer 1984). However, traditional dialect grammars report that DFC virtually never occurs with was (‘what’) and wer (‘who’), (Schiepek 1899; Merkle 1984; Noth 1993; Steininger 1994). Within their generative studies of Bernese Swiss German, Penner and Bader (1995) and within his generative study of Bavarian, Weiß (1998; 2004) point respectively to a correlation between the size of the wh-Operator and the presence or absence of a complementizer. Schönenberger (2006) finds in a corpus study of Swiss German from Lucerne and St. Gallen that dass occurs almost never in the context of monosyllabic wh-words while it appears with 50% certainty in the context of bi-syllabic whwords in Lucerne German and becomes close to obligatory in St.Gallen German. With genuine wh-phrases like an was für Leute (‘to what kind of people’) dass tends to be obligatory. Since the following discussion will focus mainly on Bavarian, it is important to have empirically reliable information about the occurrence of DFC in this dialect. Judgment studies were carried out with Bavarian speakers from different regions and age/education groups. The task was to value spoken sentences according to the six-point scale of the German school grades according to which 1 = best and 6 = worst.2 Table 1: 10 speakers of Middle Bavarian, age: 40–78, lower educational background; 12 sentences with wh-words; 12 sentences with wh-phrases.

was (‘what’) wie (‘how’, ‘as’) etc. wem (‘who-DAT’) warum (‘why’) P+wh-word, e.g. with what P+wh-phrase, e.g. which NP

WITHOUT COMPLEMENTIZER

WITH COMPLEMENTIZER

1 1,1 1,1 1,25 1,1 1,3

3,5 3,6 2,5 1,25 1,25 1,5

Some speakers had a tendency to shy away from giving bad grades; nevertheless, was and wie are the worst in combination with the complementizer dass; wem and warum are intermediary. A similar task was given to younger dialect speakers with university education and to younger speakers with mixed educational backgrounds. 2 Speakers were instructed that 1 means “I could perfectly use this sentence in my own dialect”, whereas 6 means “I could never use this sentence in my own dialect”.

10

Josef Bayer

Table 2: 3 speakers of Middle to Eastern Bavarian, age: 25–35, university education.

was (‘what’) wer (‘who-NOM’) wo (‘where’) wem (‘who-DAT’) P+wh-word

WITHOUT COMPLEMENTIZER

WITH COMPLEMENTIZER

1,3 1 1 1,2 1

5,7 4,9 5 3,9 1,9

These speakers were more ready to give bad grades. The results are sharper. The wh-words was, wer and wo are inacceptable with dass. Again, wem has a somewhat intermediate status. Table 3: 13 young speakers from Regen (Middle to Eastern Bavarian), mixed educational backgrounds.

was (‘what’) wer (‘who-NOM’) wo (‘where’) wem (‘who-DAT’) P+wh-word

WITHOUT COMPLEMENTIZER

WITH COMPLEMENTIZER

1.5 1.3 1.6 2.2 1.5

4.6 4.7 4.5 4.4 2

The dative pronoun wem cannot be distinguished in this group. Nevertheless, the main effect of wh-word versus wh-phrase remains as stable as in the other investigations. The intermediate status of the wh-words warum and wem can be explained if it is realized that they involve more structure than simplex wh-pronouns. For warum this is obvious because it is bi-morphemic and involves the preposition um. The wh-part is a so-called “R-pronoun” as familiar from discussions of Dutch syntax. According to the structure in (8), warum is underlyingly a PP, although it is a phonological word in terms of phonology. (8)

[PP wa(s)+ [P’ um was]] what for

According to Bayer, Bader & Meng (2001), the dative pronoun wem is likewise more complex than a nominative or accusative pronoun. The latter two can be morphologically primitive as shown by the syncretic form was; the dative, however, must be overtly Case-marked. Consider the contrast in (9).

Doubly-Filled Comp, wh head-movement, and derivational economy

(9)

a.

Welch-em Vorschlag hast du which-DAT proposal have you ‘Which proposal did you object to?’

11

widersprochen? objected

b. *Was hast du widersprochen? what have you objected ‘What did you object to?’ Bayer, Bader and Meng (2001) propose the underlying structure in (10) by which dative Case is syntactically represented as the head of a Kase phrase (KP).3 (10)

[KP K° [NP wem]]

If this is so, there is a reason why speakers fluctuate between a PF-based wordsize and a syntax-based phrase structural parse of these wh-items. Given this, the proper generalization is as in (11). (11)

Descriptive generalization The “size” of the wh-operator determines whether wh can combine with a complementizer or not. Full-fledged wh-phrases can combine with a complementizer; word-size wh must not combine with a complementizer.4

As pointed out by a reviewer, (11) could be misunderstood as a plea for direct influence of phonological weight on syntactic structure and therefore as a determination that would disallow any variation. I will address this issue at the end of section 4.

3 Wh-movement as head-movement What can explain the distribution of the data as described in (11)? Bayer and Brandner (2008a, 2008b) propose that in Alemannic and in Bavarian DFC3 For the motivation of KP in German see Bayer, Bader and Meng (2001); for KP in general see Bittner and Hale (1996). As shown in detail by Seiler (2003), the dative is frequently found in southern German dialects to be “strengthened” by a preposition. One can see this preposition as the spell-out of K. 4 Interestingly, close parallels have been found in V2 and suspended V2 in Northern Norwegian dialects; see Vangsnes (2005); Westergaard and Vansgnes (2005), and the discussion in Bayer and Brandner (2008a).

12

Josef Bayer

dialects the complementizer must be absent if the wh-word itself is the complementizer. If the wh-lexeme itself is a complementizer, merger of a separate complementizer would be superfluous and is therefore banned by economy. Bayer and Brandner suggest a latent categorial feature C on wh-words which is activated under appropriate contextual conditions. (12)

Latent C-feature Wh-items may possess a latent C-feature αC. If α can be set to +, the wh-item is simultaneously C and will project a CP. If α is set to −, the C-feature will delete.

The assumed feature structure of wh-lexemes is as in (13). (13)

Feature structure for simplex wh-lexemes X [wh, . . . , αC ].

αC can only turn into +C if X is merged with TP. If X is trapped in some branching structure, it cannot become a sister of TP. In that case, αC will turn to −C and will ultimately delete. Let us first consider the beginning of a derivation. In (14), wh is merged with V. (14)

Although wh is merged with V° and is therefore in this context a wh XP, its projective status as such remains formally ambiguous between head and phrase. If wh is re-merged with TP, its latent feature αC will be set to +C and will project a CP as shown in (15). (15)

Doubly-Filled Comp, wh head-movement, and derivational economy

13

What being a potential head, the CP in (15) does not result from external merger of C but importantly from internal merger of the wh-word. We assume that internal merger results from the possibility of C to be merged with TP and is as such not feature-driven. C does not come alone but in combination with a wh-feature. This amounts to saying that the projection of a wh-CP results from “self-attachment” rather than from merger of a C which contains an unvalued feature uWh that attracts a wh-phrase.5 As Fanselow (2004: 26) puts it, “the head in question possesses the checking feature and the feature to be checked at the same time.” Merger of what with an NP as in (16) does not do any damage because what fails to become a sister of TP and will therefore never activate the C-feature. (16)

A wh-DP as in (16) will move to the specifier of an independently merged C. In a DFC-language like Bavarian, this C is normally overt, and we observe the DFCphenomenon, e.g. in an embedded sentence such as (17). (17)

I I

mechat want

wissn, [CP [wäichas know which

Physikbuach] [C’ dass [TP d’Sophie physics-book that the-Sophie

g’lesn hot]]] read has ‘I’d like to know which physics book Sophie has read.’ In generative grammar, the proposal of wh-movement as head-movement clearly invokes various questions. The GB-version of X-bar theory that has adopted functional categories suggests a clear division of heads and specifiers. However, Bare Phrase Structure (BPS) as suggested in Chomsky (1995) and following work, 5 Movement of the complementizer has been suggested for independent reasons in the T-to-C movement account of Pesetsky and Torrego (2001). Self-attachment of the verb to its own projection has been explicitly proposed by Platzack (1996), Koeneman (2000; 2002), Bury (2002), Fanselow (2002a), Surányi (2003), Brandner (2004) and van Craenenbroek (2006). Donati (2006) assumes wh-head-movement for independent reasons. It has been implicitly assumed by many more. For a comprehensive overview and detailed theoretical discussion see Georgi and Müller (2010).

14

Josef Bayer

does not and cannot insist on a pre-established division. In BPS, the decision of what is a head and what is a complement or a specifier is made in the course of the derivation. Under the realistic assumption that a lexical item may embrace more than a single feature, we end up with the possibility of a complex feature structure as indicated in (13). One lexical item may simultaneously embrace the feature of C and the feature of wh, the latter of which may still prove to be decomposable as we will argue below.6 It seems that for the purpose of syntactic activation, the features on a lexical item must somehow be ordered. I will turn to this question in the next section.

4 Economy 7 Head-movement has been a controversial issue in minimalist syntax over the last few years. Chomsky (2001) argued that it might be an artifact that can be eliminated from the theory.8 In the meantime, head-movement is back on stage. Chomsky (2010) suggests that, contrary to earlier assumptions, move (= internal merge) may even be preferred over merge (= external merge) because it partially circumvents the notorious numeration problem and thus narrows the search space from which a new lexical item can be drawn for further computation. For the concrete case of DFC in Bavarian, the decision is between (18a) and (18b). (18)

a.

[CP wh [C’ comp [TP . . . wh . . .]]] X external merger of comp X internal merger of wh

b.

[CP wh [TP . . . wh . . .]] X internal merger of wh (= comp)

6 Although the present chapter does not focus on general issues of lexicon and morphology design, it should not be overlooked that there is a clear affinity to the program of Nano-Syntax (NS) as envisaged by Starke (2009) and publications quoted there. In NS, syntax projects from single features building morphemes and phrase structure alike. Thus, a lexical item – as defined by phonology – may associate with a syntactic phrase. The possibility of a combination of features which otherwise often distribute in phrase structure over comp (C) and wh (SpecCP) is expected from this perspective. 7 Thanks to Joe Emonds for his suggestions about this part. 8 The argument was placed in the larger question about movement as such. Movement was seen as an “imperfection” in the design of language. Head-movement was seen as movement that falls outside core syntax, essentially a PF-operation. In the aftermath of Chomsky’s argumentation, Müller (2004) went as far as suggesting reanalysis of a classical and so far undisputed case of head movement, namely Germanic V2, as phrasal movement in disguise (remnant VPmovement).

Doubly-Filled Comp, wh head-movement, and derivational economy

15

If the derivations underlying both structures achieve the same result at LF, (18b) should be preferred over (18a) under minimalist assumptions. (18b) clearly involves fewer computational steps and ergo less structure than (18a). The background assumption of this is, of course, the standard distinction between features and categories. In (18a) the computational system must access the lexicon twice whereas in (18b) it must access the lexicon only once. If we assume that the internal feature structure of the wh-item is the same, (18a) tolerates a redundancy. Comp is merged although it could be activated via the featural make-up of the wh-lexeme. In that case, the feature αC is superfluous and must be deleted. We will shortly see that a wh-element may involve yet another feature. If this is so, the process of external merger will need to be iterated, thus amplifying the economy gap between the two derivations. In (18b), the wh-element is a potential head and involves the sub feature αC. Therefore, it can be “recycled” in a single-step operation of movement. I will shortly turn to a slight modification of this.9 Being primarily concerned with a formal account of grammaticalization, van Gelderen (2004:10) proposes the general economy principle in (19): (19)

Head Preference or Spec to Head Principle Be a head rather than a phrase!

Given that (18b) conforms to (19), and the lexical entry of the wh-item embraces αC as a sub feature, external merger of comp will be blocked, and the derivation underlying (18a) will be discarded. Sentences like in (20), taken from one of the empirical investigations, which native speakers of Bavarian overwhelmingly reject and would never produce spontaneously, can be derived but are excluded by economy. (20) a. *I woass aa ned, wos dass bei de Nachban wieder lous gwen is I know also not what that at the neighbors again on been has ‘I also don’t know what has been going on at our neighbors’ b. *I mechat wissn, wia dass-a dees iwerlebt hod I want know how that-he this survived has ‘I’d like to know how he survived that’ In each case, there is a less costly derivation.

9 I am aware that the space which I can reserve here for the issue of derivational economy is far too small to cope with the problem. See Sternefeld (1997) for detailed discussion.

16

Josef Bayer

Bavarian stands in an interesting contrast with colloquial (substandard) Dutch (E. Hoekstra 1993; Barbiers et al. 2005), Frisian (de Haan and Weerman 1986; Reuland 1990), and West-Flemish (Haegeman 1992), varieties in which the left edge of CP appears to be more articulate than in Standard Dutch. The following Dutch data from E. Hoekstra (1993) show in (21a) complementizer doubling and in (21b) wh-movement on top of complementizer doubling. (21) a. Ik vraag me af [of [dat [Ajax de volgende ronde haalt]]] I ask me PRT if that Ajax the next round reaches ‘I wonder whether Ajax [= the Amsterdam football team] will make it to the next round’ b. Ze weet [wie [of [dat [hij had willen opbellen]]]] she knows who if that he had wanted call ‘She knows who he wanted to call up’ While in Standard German as well as in Standard Dutch, the respective interrogative complementizers ob and of serve simultaneously as polar interrogative markers (typing particles) and as subordinators, the variety of Dutch seen in (21a) spells out the features of polarity (of ) and subordination (dat) with two syntactic heads. In (21b), there is even a tripartite structure in which one could argue that the wh-operator has been moved to the specifier of the interrogative head of.10 The syntactic structure is as in (22). (22)

10 As I point out in Bayer (2006), this division cannot be accidental. It maps rather directly onto the semantic structure of embedded wh-questions for which the partition approach to questions (cf. Groenendijk and Stokhof 1982; Higginbotham 1993, 1997; Lahiri 2002) has argued independently. Wh-questions are like polar (or disjunctive) questions with the difference that they have a gap. John knows whether Bill smiled is true iff John knows that Bill smiled or that Bill did not smile. John knows who smiled is true iff John knows for each individual x (that may be a contextually relevant potential smiler) that x smiled or x did not smile.

Doubly-Filled Comp, wh head-movement, and derivational economy

17

Expanding the proposal in Bayer and Brandner (2008a,b), assume that in Bavarian a word-size wh-lexeme embraces a C-feature which – by virtue of being interrogative – covers also polarity, it is easy to see how virtually the same representation as in (22) can be built solely on the basis of internal merge. Let us suggest here that the feature structure of a wh-item is slightly more complex than in (13), namely as in (23) below. As Georgi and Müller (2010) point out, features on a head must be ordered. In order for the wh-word to be a successful complementizer, the C-feature must be visible. According to Georgi and Müller, it is visible if it is the topmost feature on a stack and will be removed as soon as the head has been re-merged with TP. Adding the feature Pol immediately below C enable the head to undergo re-merged with the C-headed CP. Adding the feature wh immediately below Pol will enable the head to be re-merged with PolP. Let us then revise (13) as in (23) in which features are represented as an ordered feature set; a ≺ b should be understood as “a precedes b”. (23)

Feature structure for simplex wh-lexemes (revised) X {αC ≺ βPol ≺ γwh ≺ . . .} 11

(24)

11 Simplifying somewhat, I assume here that Pol is a subfeature of wh and can be set to + or −. Notice that in German as in various other languages there is lexical ambiguity between an interrogative and an indefinite reading. If there is no interrogative force in the left clausal periphery,

18

Josef Bayer

In (24), the word-size wh-element re-attaches to TP activating C, then re-attaches to CP activating PolP and finally re-attaches to PolP to activate the WhP. In this way, the feature structure of the lexical item that must be assumed for independent reasons unfolds automatically in the process of merger. The process underlying this derivation is a process of “recycling” which is repeated until all the features in the feature configuration are either activated or deleted from the derivation. Deletion can apply only if the feature is set to minus or is deactivated due to semantic interpretation.12 As an intermediary conclusion, it should be clear at this point that wh headmovement is a viable option within the Minimalist Program and especially within the assumptions of BPS. A derivation in terms of re-attachment of a potential syntactic head leads to a more economical derivation than the (coexisting) mechanism of external merger of a new functional head and subsequent attraction for the purpose of feature valuation. One attractive aspect of this approach is that it can capture syntactic variation in an insightful and nonstipulative way. The source of variation rests in lexical differences (cf. Borer 1984). If syntactic differences between related varieties of Germanic such as Bavarian and Dutch, Flemish, Frisian etc. can be traced back to differences in lexical feature structure, such a result seems to be desirable as it would fall into largely understood territory. Lexical differences can, however, also be found within one dialect and even within one idiolect. Thus, it should not be surprising to observe intra-dialectal or even intra-idiolectal variation in the domain of DFC. As one reviewer points out, one can easily find examples of wer dass and even was dass, i.e. of the least favored combinations. Weiß (2004) argues that the DFC-variation can hardly be rooted in core grammar in the sense that a “heavier” constituent would have “more features”. He attributes the distribution wh-pronouns in situ receive an indefinite interpretation. In this case, βPol would be turned into −Pol and get deleted from the structure. (i) Ich habe was gesehen I have wh-thing seen ‘I saw something’ Notice that the system of Georgi and Müller is more complex because it adds to their structure-building (subcategorization/merge) feature [ e F e ] also a probe/agree feature [ F ] that may operate asynchronically in the derivation. For reasons of space I will not elaborate here on this aspect of head-reprojection. 12 Of course, it is not a trivial issue to determine at which point of a derivation a feature is deactivated. Nevertheless it should be clear for the core cases of scope taking C becomes irrelevant after it has been merged with TP; Pol becomes irrelevant after it has been merged with CP; Wh becomes irrelevant after it has been merged with PolP and has been subject to “scope freezing” (cf. Baker 1970 and following work).

Doubly-Filled Comp, wh head-movement, and derivational economy

19

of the data to requirements of the processing system, in particular to the need of quick identification of the clause type by virtue of the complementizer. While I agree that the covariation of phonological weight and number of features is likely to be illusory, I also feel that his explanation should be met with reservation. First it is unclear in which sense the processing system could benefit from early identification of the clause type.13 Secondly it is unclear why a heavy whphrase following a question-embedding verb would qualify as an identifier less successfully than a wh-word. I find it far more plausible to acknowledge that was is the most underspecified wh-element of the German lexicon and as such the top candidate in adopting additional features without running into conflicts, and that there are other wh-lexemes which can do so to a higher or lower degree, and that this variation may be a matter of the individual mental lexicon. By looking at some familiar morphosyntactic processes, the next section will provide independent motivation for the correctness of wh head raising and the approach as it has been developed so far.

5 Cliticization and comp-inflection Unlike Standard German, which may have only strong and weak pronouns (see Cardinaletti, 1999), the South German dialects undoubtedly have clitics. In Bavarian, the only way to express the German examples in (25) would be as in (26). (25)

a.

Wo hat er es dir where has heNOM itACC youDAT ‘Where did he leave it for you’

b.

Ich meine, dass er es I think that heNOM itACC ‘I think that he has left it for you’

denn PRT

hingelegt down-put

dir youDAT

hingelegt down-put

hat has

13 In German, sentences with a V2-complement like (i) are known to hardly create parsing difficulties although there is a firm local ambiguity as shown in (ii). (i) Ich glaube an meine Theorie kann sich niemand mehr erinnern I believe in my theory can REF nobody PART remember ‘I believe nobody can remember my theory’ (ii) Ich glaube an meine Theorie . . . ‘I believe in my (own) theory’ For lucid theoretical discussion see Gorrell (1994; 1995).

20 (26)

Josef Bayer

a.

Wou where

b.

I I

hod-a -s -da -n had-heNOM-itACC-youDAT-PRT

moan think

dass-a -s -da that-heNOM-itACC-youDAT

hῖ:g’legt? down-put hῖ:g’legt down-put

hod has

The clitics attach in (26a) to the finite verb which is in the position of C, the so-called “Wackernagel” position. Their distribution is the same when they cliticize to a complementizer as in (26b).14 As an extension of cliticization, Bavarian also shows in more limited cases what has become known as comp-inflection.15 In this case, a clitic has been reanalyzed as an inflectional suffix, which appears obligatorily on the complementizer. The full pronoun can still follow as seen in (27). (27)

Wenn-st (du) ned foig -st nou if -2SG (you) not obey -2SG then ‘If you don’t listen to me, I’ll lock you up’

schbi:r-a-de lock -I -you

eῖ up

There may be the impression that Bavarian has clitic-doubling. However, doubling is confined to 2nd person singular and plural (and in some dialects also 1st person plural) although there is a pervasive clitic paradigm. There is nothing like clitic doubling for 1st person singular (*wenn-e i: ned foig, if-I I not obey) or 3rd person (*wenn-a er ned foigt, if-he he not obeys). Thus, the 2nd person clitics must at some stage have been reanalyzed as inflectional suffixes. In the mind of a current speaker, they are inflectional. I would now like to demonstrate that external sandhi, consonantal epenthesis and comp-inflection prove consistently that word size wh-elements behave like complementizers, i.e. are like heads resembling the functional head C, and not like syntactic phrases.

5.1 External sandhi: Underlying /r/ In Bavarian, like in many varieties of spoken German, /r/ is consonantal in the onset of a syllable but vocalized in the rhyme (Wiese 1996). Consider the wh-pronoun wer, [vɛɐ] (‘who’). Cliticization crosses a weak prosodic boundary and induces onset maximization. In this case, /r/ will be pronounced as seen in (28b). 14 The clitic particle -n is absent in (26b) as it applies only to questions. 15 Pfalz (1918); Altmann (1984); Bayer (1984); Weiß (1998, 2005); Fuß (2005) among others.

Doubly-Filled Comp, wh head-movement, and derivational economy

(28)

a.

De woass, wea -s is she knows who-it/she is ‘She knows who it/she is’

b.

De woass, we -r -e she knows who-R-I ‘She knows who I am’

21

bin am

The relevant foot structure in (28b) is [φ [σ vɛɐ] [σ re]]. Take now for comparison a bona fide wh-phrase that ends in a vocalized /r/ as in Uhr, [uɐ] (‘clock’). Cliticization to such a phrase leads to a bad result as (29a) shows. One way out could be the avoidance of cliticization as in (29b). The preferred way would be to merge the complementizer dass and thus have a proper host for cliticization ready as in (29c). (29)

a. *De woass [um wiavui Uh -r]-e she knows at how-much clock-R -I ‘She knows at what time I leave’

geh go

b.

De she

woass knows

[um at

wiavui how-much

Ua] clock

i: I

geh go

c.

De she

woass knows

[um at

wiavui how-much

Ua] clock

dass-e that -I

geh go

These data suggest that dass and wer behave alike. Both serve as hosts for cliticization. Genuine wh-phrases such as um wieviel Uhr are flanked with strong prosodic boundaries and show very distinct behavior. These facts are straightforwardly reconciled with the syntactic derivation proposed above according to which a word-size wh-element is internally merged with TP. They can hardly be reconciled with a conventional X-bar theoretic derivation in which an empty complementizer is merged to TP, and the wh-element moves to its specifier.16 The same is true for derivations in which the wh-phrase lands in a distinct higher CP-shell as suggested by Baltin (2008) or by Koopman’s (2000) “Generalized Doubly-Filled Comp Filter”, which bans lexical material in both the head and spec of a given projection. These accounts are by definition blind to the size or

16 The case is reminiscent of wanna-contraction. It has been argued that to cannot cliticize onto want across an intervening trace. This blocks *Who do you wanna die? If this reasoning holds water, cliticization across an empty complementizer should likewise be impossible, contrary to what (28b) shows to be actually the case.

22

Josef Bayer

the shape of the wh-operator that move to the left periphery. Some complex machinery with a number of special assumptions would be required to accommodate the Bavarian cliticization data in such a theory.

5.2 Consonantal epenthesis Consider next lexical elements which terminate in a diphthong that targets the vowel [ɐ]. In Bavarian, these diphthongs embrace [iɐ] and [uɐ]. Attachment of a vocalic clitic to such elements triggers consonantal epenthesis for the avoidance of hiatus.17 In Bavarian, the epenthetic element is [r] as seen in (30b). (30)

a.

wia-s hinte schaut sicht-s as -she back looks sees-she ‘As she looks back, she sees Sepp’

b.

wia-r-e hinte schau sich-e as -R-I back looks see -I ‘As I look back, I see Sepp’

an the

an the

Sepp Sepp

Sepp Sepp

The underlying form of the clitic’s host is wie, phonetically [viɐ]; r is not part of it. Thus, it must be an intrusive element. Interestingly, epenthesis is unattested (and unacceptable) if the host is part of a genuine syntactic phrase. Consider the host Schuh, [ ʃuɐ] (‘shoe’ or ‘shoes’). (31)

a. *Sog-ma [wos fia Schua]-r -e õ:ziang tell-me what for shoes -R -I on-put ‘Tell me which shoes I should put on’ b.

Sog-ma tell -me

[wos what

fia for

Schua] shoes

dass -e that -I

soi should

õ:ziang on-put

soi should

The restriction is the same as in (29). Cliticization applies to a syntactic head but cannot apply to a wh‑phrase. In the presence of a wh-phrase such as wos fia Schua, a functional head, namely dass, is inserted, and consonantal epenthesis does not emerge.18 Thus, data from consonantal epenthesis show again that this process is limited to the environment of a functional head, and that word-size wh-operators behave like such heads whereas genuine wh-phrases do not. 17 Cf. Gutch (1992) for detailed discussion. 18 Alemannic uses n for epenthesis. As Ortmann (1998) shows on the basis of data from the Black Forest area, n is epenthesized for hiatus avoidance in cliticizations to the functional

Doubly-Filled Comp, wh head-movement, and derivational economy

23

5.3 Comp inflection Consider finally comp-inflection for 2nd person sg. and pl. In Bayer (1984), I suggested that inflectional suffixes can target not only heads but also phrases. This suggestion is likely to be wrong.19 Importantly, comp-inflection is completely well-formed on wh-words. (32)

a.

I woass scho, wia-st (du) ausschau-st I know already how-2SG you out-look -2SG ‘I already know what you(sg) look like’

b.

I woass scho, wann-ts (e:s) in-s I know already when-2PL youPL in-the ‘I already know when you(pl) go to bed’

Bett bed

geh-ts go -2PL

Comp-inflection on wh-phrases is dispreferred or downright ungrammatical. In the following examples, there is a strong grammaticality difference between (a) and (b). (33) a. *I woass scho, wos fia Schua-st (du) õ:zong ho -st I know already what for shoes-2SG you on-put have-2SG ‘I already know what kind of shoes you have put on’ b. I woass scho, wos fia Schua dass-st (du) õ:zong ho -st I know already what for shoes that-2SG you on-put have-2SG

heads C and P. Consider the word-size wh-element wo, [vo:] (‘where’) versus a comparable open class noun that is part of a genuine wh-phrase: (i) I weiss it, wo -n-er ani isch I know not where-N-he towards is ‘I don’t know where he went’ (ii) a. *I weiß it [uf wellem Klo] -n -i ga hocke I know not on which toilet-N-I go sit ‘I don’t know on which toilet I will sit’ b. I weiß it [uf wellem Klo] dass-i ga hocke I know not on which toilet that -I go sit 19 Relevant criticism had already been formulated by Marina Nespor (p.c.) at the time of publication of Bayer (1984). At that early stage of GB-theory, the dilemma could not be resolved.

24

Josef Bayer

(34) a. *I woass scho, wia oft -ts (e:s) g’fäit hab -ts I know already how often -2PL youPL be-absent were-2PL ‘I already know how often you(pl) have been absent’ b. I woass scho, wia oft dass-ts (e:s) g’fäit hab -ts I know already how often that-2PL youPL be-absent were-2PL In the context of a wh-operator which cannot be analyzed as a potential head, comp inflection fails, and the complementizer dass has to be externally merged.20 Unfortunately, grammaticality judgments are often less than crystal clear and may be contaminated by Standard German. One cannot exclude the possibility that speakers accept structures as in (33a) and (34a) by virtue of an analogical generalization by which 2nd person inflection is associated not with the host as such but rather with the linear position. To control for that, a production experiment was carried out in which native speakers of Bavarian were 20 A caveat must be added about PPs. It seems that for many speakers there is the possibility to inflect a simplex wh-item such as was even though it is part of a PP. Dialect speakers write on the internet in their dialect, e.g (i) nix verbotenes, und a ned des [an wo] -st du schon nothing forbidden and also not this at what -2SG you already wieder denk -st again think - 2SG ‘Nothing forbidden and not what you already have thoughts about’ http://www.flf-book.de/Benutzer/Partybus.240.htm I tend to say that PP is a potential extension of the category in its complement. If the complement is X°, P+X° is also an X°. Evidence for this comes from the copying strategy in wh-scope extension that is possible in various German dialects. (ii) Wo glaubst du, wo er wohnt? where believe you where he lives ‘Where do you believe he lives?’ Significantly, no copying of genuine XPs is ever possible, – with the exception of PPs of type P+X°. (iii) a. [Mit wem] glaubst du, [mit wem] wir uns treffen könnten? with who believe you with who we REFL meet could ‘Who do you believe we could meet with?’ b. *[Mit welchen Linguisten] glaubst du, [mit welchen Linguisten] wir uns treffen könnten? ‘Which linguist do you believe we could meet with?’ See Bayer and Bader (2007), Barbiers et al. (2010), Pankau (2010) for discussion. The important point in the present context is that PPs with a potential X° complement may be analyzable as syntactic heads: {P, N} % {P, {P, N}}.

Doubly-Filled Comp, wh head-movement, and derivational economy

25

given sentences in Standard German, which they had to turn into their local dialect. The experiment was tape-recorded and transcribed.21 Table 4: Production experiment. 9 speakers from Regen (Middle to Eastern Bavarian); 10 sentences. STANDARD GERMAN INPUT

BAVARIAN OUTPUT

PERCENTAGE %

model (33)

Standard German syntax

14

e.g. [was für DP] du VP what for DP you VP

irrelevant

9

was für split: was-st (du) für DP . . .

64

dass-insertion: was für DP dass-st (du). . .

15

XP+inflection: was für DP-st (du) . . .

–

model (34)

Standard German syntax

30

e.g. [wie(viel) XP] du VP how(much) XP you VP

dass-insertion: wie(viel). . .dass-st (du). . .

70

XP+inflection: wie(viel)-st . . . (du)

–

The results of this experiment could not be clearer: There was not a single case of 2nd person inflection on an XP of type [was für DP] or [wie(viel) XP]. In the first case, the leading strategy was to extract was and strand the für-XP. In this case, was could be inflected and in fact was inflected throughout. In the latter case, this strategy fails: wie cannot be extracted out of wie oft. Here we observe with 70% a clear majority of DFC with dass being inflected throughout. In both parts of the investigation, it is revealing that speakers prefer omission of the inflection, i.e. essentially a reply in Standard German, to inflection of the wh-XP. On the basis of these results, one can be sure that speakers distinguish between wh-words and wh-phrases. Wh-words pattern with externally merged complementizers in their behavior as hosts for inflection. Wh-phrases do not. These facts support the theory according to which wh-words move as C-heads and project CP whereas wh-phrases require external merger of a complementizer. Thus, the classical X-bar picture of DFC may be retained but must be revised for those cases in which a more economical derivation in terms of wh headmovement is viable. Again, it should be clear that a split-CP approach in which wh does not communicate with C at all has nothing to say about the possibility of wh-elements acting as inflected complementizers.

21 The experiment was carried out by Michael Merz in the context of his master thesis, see Merz (2011). More empirical work can be found in Bayer (2014).

26

Josef Bayer

Reis (1985) noticed already that wh words attract weak pronouns and concluded that this is a problem for a theory which places the wh word in a higher position than the C-position. In an HPSG account, Kathol (2000) tries to revive the traditional linear theory of German clause structure that emerged in the late 19th century and became known as the topological fields model (“Theorie der topologischen Felder”, see Höhle (1986)). In this model, wh-phrases and C are always in the same position.22 Kathol seems to be right as far as word-size wh-operators are concerned. On the other hand, the Bavarian data on compinflection suggest that bona fide wh-phrases rely on a separately merged complementizer and therefore move beyond it.

5.4 Consequences We have been able to demonstrate that word-size wh-operators show exactly the same syntactic distribution and the same morphophonological properties as externally merged complementizers and verbs in V2-position. This allows the following generalization. (35)

i.

“Wackernagel-type” morphophono-logical processes – cliticization, consonantal epenthesis, comp inflection – apply uniformly to the C-position.

ii.

If T-to-C movement does not apply, merge a lexical item with the categorial feature C to TP, no matter whether C is a “plain” C or a wh-element with a latent C-feature!

Wh head-movement relies on the C-feature which is needed to project a CP.23 The C-feature cannot be activated at a later stage in which merger with TP is no longer available. Thus, the features [γwh, . . . βPol, αC] in (23) must be ordered in a feature tree that maps onto the order seen in (22) and (24). Once T-to-C movement has applied, i.e. a “V1-structure” has been created, wh cannot be merged with TP, and the prediction is that wh ends up as a specifier rather than as a head. If it is not a head, we do not expect head-typical processes such as epenthesis. Written examples can be found in which the phonological environment could give rise to r-epentheses but in fact does not. 22 “[there] is strong reason to believe that complementizers and wh/d-phrases in subordinate clauses belong to the same natural class in terms of their positional properties.” (Kathol 2000: 111). 23 This cannot be universal, though. Notice that in Hungarian the complementizer hogy precedes wh. It must be merged after wh-movement has applied.

Doubly-Filled Comp, wh head-movement, and derivational economy

(36)

27

Wia is -n des bei engk herendd, Raimund? how is-PART this at you over-here Raimund ‘Hey Raimund, how is it with you over there?’ http://www.google.de/search?q=Bairisch+%22wia+isn%22&btnG=Suche& hl=de&lr=&client=firefox-a&rls=org.mozilla%3Ade%3Aofficial&as_qdr= all&sa=2

Although the wh-item is identical with the one that appears in (30) as a Celement, and although the auxiliary is would be able to undergo encliticization, r-epenthesis is rejected by native speakers that I have consulted with examples like (37).24 (37) *Wia –r -is-n des bassiert? how-R-is -PART this happened ‘How did this happen?’ BPS does not in principle preclude head-movement to a CP (or FinP) that is headed by the finite verb. However, empirical considerations suggest that the wh-element that has been moved in (36) and (37) counts as an XP: Elements in this position (called “Vorfeld”) can be inserted for the satisfaction of the V2constraint, and they can under certain circumstances be dropped if they qualify as discourse topics. Both properties are arguably not attested with heads. We can conclude that even word-size wh-items count as XPs once they are placed before the finite verb in the sense of regular specifiers. As such, they are flanked by a strong prosodic boundary that prevents cliticization and epenthesis.25

24 The same is true for Alemannic n-epenthesis. (i) *Wa -n-isch denn passiert? what -N-is PART happened ‘What happened?’ (ii) *Wo -n-isch de vater ani? where -N-is the father towards ‘Where did father go?’ 25 There is some evidence that a focused wh-word has more structure than an unfocused one. Not too surprisingly then, a focused wh-word may cooccur with a complementizer. Bayer and Brandner (2008a: 93) and Noth (1993) for more details.

28

Josef Bayer

6 A note on diachrony and variation It cannot be overlooked that in many languages, the unmarked complementizer corresponding to English that or German dass is an unmarked wh-pronoun of the language, corresponding to ‘what’. (38)

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

que che что ( ʃtɔ) ço τί (ti) che (ki)

French, Portuguese, Spanish, Catalan Italian Russian Polish Greek, the complementizer being ότι (oti) Persian (‘what’), changes to the complementizer ke Hindi, Guajarati, Marathi, Punjabi and various other Indo-Aryan languages.

If this is not an accident, one must assume that the unmarked wh-operator has been historically reanalyzed toward a neutral, i.e. non-interrogative, complementizer. From research on grammaticalization, the featural impoverishment (alias “bleaching”) of affected elements is a familiar process.26 It is interesting to see in this context that even in Germanic varieties examples can be found in which a wh-word serves either as a polar complementizer or as a non-interrogative (“declarative”) complementizer altogether. (39) is from a Low German dialect reported in Zimmermann (2011). (40) is from Yiddish, reported by Kühnert & Wagner (2004), and (41) is from Bernese Swiss German, reported by Hodler (1969) and Penner (1993).

26 An alternative analysis interprets the homophony of wh-operator and complementizer as identity and argues that complements are actually relative clauses. Kayne (to appear) says “the that that introduces sentential complements is really a relative pronoun, and sentential complements are really relative clauses, in a way that partially recalls Rosenbaum (1967)”. For more discussion of this proposal, which I cannot evaluate here, see Arsenijevic (2009); Manzini (2012); Manzini and Savoia (2003). The proposal leaves many questions open, for instance why German has overwhelmingly a d-word and not a w-word as complementizers although free relatives are as in (i) and not as in (ii). (i) Was du behauptest ist falsch what you claim is wrong (ii) *Das du behauptest ist falsch Another question is why Scandinavian uses a preposition (att) as a complementizer, an element that seems to be an odd candidate for a relativizer.

Doubly-Filled Comp, wh head-movement, and derivational economy

(39)

LOW GERMAN Ik weet nich wat de Bodder al smolten is I know not WAT the butter already melted is ‘I don’t know if the butter has melted already’ http://www.plattpartu.de/kuenst/lueske1_biller.htm; 12.02.07

(40)

YIDDISH veyśtu den nit voz unz Ari know-you PRT not what us Ari ‘Don’t you know that Ari belongs to us?’

(41)

BERNESE SWISS GERMAN I gloub nid {wo / wa I believe not where / what ‘I don’t believe that he will come’

/ /

zu to

was} what

29

gihert belongs

er he

chunt comes

It looks as if there is a grammaticalization path which runs as in (42). (42) Grammaticalization path for wh with increasing featural impoverishment27 +wh phrase > +wh head > polar interrogative head > −wh head 2 3 2 3 2 3 2 3 þwh þwh wh wh 4 þpol 5 4 þpol 5 4 þpol 5 4 pol 5 þres res res res This path maps onto the three layers of the split CP that have been attested in Dutch, see (22) above, and which we could find again in terms of feature structure in Bavarian, see (23) and (24) above. The categorial feature C that was previously introduced as a primitive may turn out not to be primitive but rather the consequence of featural impoverishment that leads to a wh-lexeme which actually lacks the semantic part of the wh-feature and is as a consequence recruited as a complementizer.28 The present account also finds a straight explanation why semantically restricted wh-words do not turn into pure subordinators.29 A form like *I believe when you are depressed can presumably not stand in any language for the meaning I believe that you are depressed. This is so because the semantic restriction,

27 “res” stands for a semantic restriction that appears automatically in a phrase. 28 This squares with the widely known fact that complementizer (C) is a highly inhomogeneous category which embraces at least d-pronouns, wh-pronouns, prepositions and verbs. 29 Thanks to Richie Kayne (p.c.) for drawing my attention to this fact.

30

Josef Bayer

in this example TIME, would in all likelihood resist neutralization. This does not mean though that a wh-word like German wann (‘when’) could not synchronically act as a complementizer and as an operator. The important fact is that it has a sub-structure which it can discharge upon an internal merge operation, and which the syntax can identify as the activation of a subordinative function. This is what we mean by a “C-feature” or a “latent C-feature”. From the viewpoint of diachrony and grammaticalization, the question is different. Here the question is how it can be that a wh-operator changes as a whole into a complementizer. Given what is known about the extreme underspecification of correspondents of “what”, there is no doubt that this element stands a high chance of being recruited as a pure subordinator.30 To draw a conclusion from this brief digression into diachrony and variation, the proposed account of wh-movement as economy-driven head-style movement seems to be highly susceptible to a more general theory which integrates syntactic development across time and syntactic variation. In the last two sections, I will very briefly turn to two potential problems for the analysis proposed above.

7 Sluicing Baltin (2010) argues that DFC does not exist at all, the reason being that it is supposed to make wrong predictions about sluicing. Sluicing (Ross 1967) has been studied in detail in Merchant (2000, 2006) and is standardly described as a deletion operation that spares the wh-phrase as seen in (43b). (43)

a.

Fred gave someone my keys but I can’t remember when

b.

Fred gave someone my keys but I can’t remember when he gave this person my keys

At first sight, sluicing seems to be straightforwardly accounted for by TPdeletion. Assuming now that in DFC-languages the wh-operator moves to the specifier of an overt complementizer, the remnant after TP-deletion would be wh plus comp. It is, however, known at least since Lobeck (1995) that sluicing does not spare comp.

30 A rich source of evidence can be found in Jäger (2000).

31

Doubly-Filled Comp, wh head-movement, and derivational economy

(44)

Hans Hans

hat has

von from

jemandem someone

Geld money

gekriegt, gotten

aber but

ich I

habe have

vergessen von wem (*dass) forgotten from who (that) ‘Hans got money from someone but I’ve forgotten from who’ The proposal to have sluicing target C’ instead of TP is ad hoc and X-bar theoretically unattractive as it affects an intermediate projection. Following Rizzi’s (1997) split CP analysis, Baltin argues that wh and comp are in fact never in the same CP. In this case, one can assume that sluicing targets the lower CP-shell which contains comp but spares the higher CP-shell that contains the wh-operator.31 (45)

a.

[CP2 wh [C2’ t [CP1 t [C1’ dass [TP . . . wh . . .]]]]] CP1-deletion 7

b.

[CP2 wh [C2’ t [CP1 t [C1’ dass [TP . . . wh . . .]]]]]

As we have pointed out already in 5.1, this cartographic account can hardly propose more about the interaction of wh and comp than some spell-out convention. The Bavarian data which had been presented so far, however, have shown that wh and comp interact heavily: Word-size wh-operators compete with comp while wh-phrases do not. Word-size wh and comp but not phrasal wh is the target for cliticization and epenthesis. Word-size wh and comp bear 2nd person inflection whereas phrasal wh does not. On the other hand, Bavarian word-size wh-operators behave fully regularly under sluicing. Consider (46a,b). (46) a. D’Resl woidd kumma owa i woass ned wann the-Therese wanted come but I know not when ‘Therese wanted to come but I don’t know when’ b. Da Hias mechat sei Moped frisian owa er woass ned wia the Matthias wants his moped improve but he knows not how ‘Matthias wants to increase the power of his moped but he does not know how’ Although wh-items such as wann or wia behave like C (due to their latent C-feature), sluicing does not target them. If it did, unrecoverable information would be lost, and the prediction would be that sluicing does not exist in this dialect to begin with. It is easy to see that the present account does not yield 31 Baltin proposes the deletion of FinP, which in Rizzi (1997) would correspond the lower projection which I call here CP1 for reasons of simplicity.

32

Josef Bayer

such a conclusion. Conditioned by the feature bundling seen in (23), a word-size wh-item moves as shown in (24). It is first remerged with TP activating the latent C-feature. This leads to the projection of what we have called CP1 in (45). Thanks to the purely interrogative feature Pol, it is then remerged a second time giving rise to PolP, i.e. quasi another “CP-shell”. Due to its wh-feature, the wh-item will undergo internal merger a third time. This last step leads to the activation of the wh-property and to the projection of a WhP, which corresponds to yet another CP-shell. Sluicing can target the CP-shell immediately dominated by WhP. As such, there is no reason to believe that wh head-movement would face more problems than any other account with respect to sluicing. What about wh-phrases then? Baltin’s worry is that TP-deletion wrongly predicts that the complementizer survives. CP-deletion, on the other hand, would also erase the wh-phrase. Notice, however, that under the assumptions of BPS, there is no predetermined difference between C’ and CP. Therefore, it is not possible to express a restriction against the deletion of an intermediate projection. As soon as dass or an empty complementizer is merged with TP, a CP results. This fact is independent of further attachments which may take the role of a specifier. The important fact is that dass heads a CP. If sluicing wants to get rid of the (informationally irrelevant) complementizer dass, it can do so.32 To conclude, Baltin’s argument against previous accounts of DFC is essentially valid but the cure that he offers in terms of an X-bar based cartographic split CP fails to accommodate central Bavarian data, namely those that point to a competition between complementizers and comp-like wh-items. In the present account, instead, comp-like wh-items have a latent C-feature by which they are remerged as heads. If we assume in an extension of Bayer and Brandner (2008a, b) that the simplex wh-item activates different CP-shells of which the wh-feature is activated in the highest one, sluicing as CP-shell deletion as suggested by Baltin does not present a problem at all. 32 What I am saying here should by no means be misunderstood as my account of sluicing; I don’t have any. If I have understood Merchant (2006) correctly, sluicing remains a partially unresolved construction. Why does it allow multiple wh-phrases in languages which otherwise, unlike Slavonic, allow only a single wh-phrase in SpecCP? Why should island violations as they arise under sluicing be “repaired” by deletion? Why does Hungarian retain the complementizer hogy under sluicing as in Merchant’s (2006) example (i), but only optionally so as Krisztá Szendrői (p.c.) points out to me? (i) A gyerekek találkoztak valakivel de nem emlékszem, (hogy) kivel. the children met someone-with but not I-remember that who-with ‘The kids met with someone, but I don’t remember with who’ As long as these problems – and this is only a selection – can only be addressed with special assumptions, arguments from sluicing against the very existence of the DFC-phenomenon such as those forwarded by Baltin should, in my view, be met with reservation.

Doubly-Filled Comp, wh head-movement, and derivational economy

33

8 A note on chain uniformity As far as BPS is understood at this moment, a wh-element that is merged with a head is declared an XP, and so is a wh-element that is adjoined to some phrase. The Condition on Chain Uniformity (CCU) as formulated in (47) requires the chain links of a wh-chain to be of type XP. (47)

Condition on Chain Uniformity A chain must be uniform, where a given chain C= (α1, . . . , αn) is uniform with respect to P if each αi has property P. (Chomsky and Lasnik 1993; taken from Chomsky 1995: 91)

Wh head-movement as proposed in this study violates the CCU. The problem may be new with respect to wh-movement. However, it is not new with respect to other types of movement. Relevant discussion of the CCU was devoted in the past to clitic movement. A clitic with argument status, say an object clitic, is merged with V and should therefore be relationally defined as an XP. At its landing site, the clitic should, however, have X° qualities. It adjoins to a functional head. The most elementary question is then how an X° can bind an XP. Another question is how an X° can bind XP if – as in the present case of wh-Cprojection – X° has to project? According to standard assumptions, the target does not project. The situation is summarized in (48). (48)

Referring to clitics, Chomsky (1995: 403) says that they “appear to share XP and X° properties, as we would expect on minimalist assumptions”. In other words, the projective status of the moving element is ambiguous. What Chomsky

34

Josef Bayer

says pertains to wh head-movement more than to clitic movement.33 All of the wh-words we have been talking about can wind up in the C-position as well as in the SpecCP position. In the latter case, there is no sign of a head property. As a consequence, word-size wh-operators must be X-bar theoretically ambiguous between XP and X° status. Adopting BPS as well as head-raising and reprojection (Münchhausen style head movement), Bayer and Brandner (2008a) suggest the following relaxation of the CCU. (49)

Condition on Chain Uniformity (as in (47) above) (revised) The chains CH = (X° . . . XP) and CH = (XP . . . X°) are uniform iff X° is surface-equivalent with XP, and XP is surface-equivalent with X°.34

The local requirements of the chain links are met. Upon merger with V, a whpronoun will be definable as a DP; upon remerger with TP it will be definable as a potential C-head. Beyond that, X° can be an antecedent of XP in the sense that X° is formally indistinguishable from XP, and XP can be an antecedent of X° in the sense that XP is formally indistinguishable from X°.35 Given that head movement as such has been a tremendous and also tremendously controversial topic in recent work in minimalist syntax, these brief remarks cannot be more than a suggestion in which way to defend the claim that has been made here, namely that word-size wh-items may be internally merged like C-heads and may then autonomously project their own CP. Whatever the best technical implementation of head movement is, in terms of the Minimalist Program this account seems to me preferable over the minimalist standard account by which a zero phantom complementizer with an unvalued wh-feature has to be merged to TP before a wh-phrase with an interpretable wh-feature can move to its specifier. The observed restrictions on DFC and the morphophonological restrictions in the syntax of Bavarian strongly support economical derivations in which the decisive criterion is the maximization of internal merger.

33 Clitics are phonologically highly special elements whereas the shape of the wh-operators we are talking about remains lexically pretty much the same across environments. 34 By “surface-equivalence” I mean shape identity as it occurs time and again in language, for instance in the German Case system where we find der Frau as a “portmanteau” for genitive singular and dative singular. One can show that a single representative of such a constituent can associate with more than a single grammatical function. 35 See the discussion of chain uniformity in Roberts (2010: 31–33), where it is pointed out that rigid chain uniformity dissolves in the sense of our (49) because the notion of “chain” loses its status as a primitive in a strictly derivational theory.

Doubly-Filled Comp, wh head-movement, and derivational economy

35

9 Conclusion Following the lead of the Minimalist Program, especially the lead of BPS and projective economy, a more fine-grained analysis of the Doubly-Filled Comp (DFC) phenomenon was achieved. The empirical results, based on a number of judgment and production tests, show that DFC in Bavarian is not as pervasive as previously assumed. It is notably absent with the pronoun was and heavily reduced with other simplex wh-words. The results show furthermore that cliticization, consonantal epenthesis and comp-inflection apply uniformly to both complementizers and short wh-elements, suggesting that wh and comp are members of a natural class in terms of their role in the syntax of embedded clauses. Previous X-bar theoretic accounts but also more recent ones which assume an invariable mapping of wh and comp to fixed phrase structural positions can hardly capture these restrictions. This was obviously the reason why traditional ideas about the linear organization of grammar were taken up again in HPSG. The derivational sub-theory of clause structure which was presented here is based on a natural and in fact independently motivated assumption, namely that the feature matrix of wh-words involves a latent comp-feature. This feature gets activated in the course of the derivation as soon as the wh-word is internally merged with TP. Since wh and comp fall together in a single lexical item, external merger of a separate comp, let alone a zero comp, is superfluous and is therefore ruled out by derivational economy. DFC, which is otherwise pervasive in Bavarian, is suspended in these cases. The permission of head-style wh-movement in syntax predicts a close fit with the PF-side. Post-syntactic morphophonological processes which typically apply to functional heads can do so directly, i.e. without having to cope with categorial mismatches, zero elements or non-contiguity.

Acknowledgments Central parts of this article rest on work with Ellen Brandner, see Bayer & Brandner (2008a,b) I wish to thank Michael Merz for the collection of data on compinflection and Annika Nitschke for editorial work. For discussion special thanks to Ellen Brandner, Joe Emonds, Günther Grewendorf, Alex Grosu, Hilda Koopman, Antje Lahne, Rita Manzini, Luigi Rizzi, Manuela Schönenberger, Øystein Vangsnes, Helmut Weiß and Marit Westergaard. Special thanks to an anonymous reviewer whose comments have improved this paper. The responsibility for any errors is exclusively on my side.

36

Josef Bayer

References Altmann, Hans. 1984. Das System der enklitischen Personalpronomina in einer mittbairischen Mundart. Zeitschrift für Dialektologie und Linguistik 51 (2): 191–211. Arsenijevic, Boban. 2009. Clausal complementation as relativization. Lingua 119: 39–50. Baltin, Mark. 2010. The nonreality of doubly filled comps. Linguistic Inquiry 41: 331–335. Baker, Carl L. 1970. Notes on the description of English questions: the role of an abstract question morpheme. Foundations of Language 6: 197–219. Barbiers, Sjef, Hans Bennis, Gunther De Vogelaer, Magda Devos, and Margreet van der Ham (eds.). 2005. Syntactic Atlas of the Dutch Dialects. Vol. 1. Amsterdam: Amsterdam University Press. Barbiers, Sjef, Olaf Koeneman, and Marika Lekakou. 2010. Syntactic doubling and the structure of wh-chains. Journal of Linguistics 46 (1): 1–46. Bayer, Josef. 2014. Syntactic and phonological properties of wh-operators and wh-movement in Bavarian. In Bavarian Syntax, Günther Grewendorf, and Helmut Weiß (eds.), 23–50. Amsterdam: Benjamins. Bayer, Josef. 2004. Decomposing the left periphery – dialectal and cross-linguistic evidence. In The syntax and Semantics of the Left Periphery, Horst Lohnstein, and Susanne Trissler (eds.), 59–95. Berlin: de Gruyter. Bayer, Josef. 1984. COMP in Bavarian syntax. The Linguistic Review 3: 209–274. Bayer, Josef, and Markus Bader. 2007. On the syntax of prepositional phrases. In Interface and Interface Condtions, Andreas Späth (ed.), 157–179. Berlin: de Gruyter. Bayer, Josef, Markus Bader, and Michael Meng. 2001. Morphological underspecification meets oblique case: Syntactic and processing effects in German. Lingua 111: 465–514. Bayer, Josef, and Eleonore Brandner. 2008a. On wh-head-movement and the doubly-filledcomp filter. In Proceedings of the 26th West Coast Conference on Formal Linguistics, Charles B. Chang, and Hannah J. Haynie (eds.), 87–95. Somerville: Cascadilla Proceedings Project. Bayer, Josef, and Eleonore Brandner. 2008b. Wie oberflächlich ist die syntaktische Variation zwischen Dialekten? - Doubly-filled COMP revisited. In Dialektale Morphologie, dialektale Syntax, Franz Patocka, and Guido Seiler (eds.), 9–26. Vienna: Praesens. Bittner, Maria, and Ken Hale. 1996. The structural determination of case and agreement. Linguistic Inquiry 27 (1): 1–68. Borer, Hagit. 1984. Parametric Syntax: Case Studies in Semitic and Romance Languages. Dordrecht: Foris. Brandner, Eleonore. 2004. Head-movement in minimalism and V2 as force-marking. In The syntax and semantics of the Left Periphery, Horst Lohnstein, and Susanne Trissler (eds.), 97–138. Berlin: de Gruyter. Cardinaletti, Anna. 1999. Pronouns in Germanic and Romance languages: an overview. In Clitics in the Languages of Europe, Henk C. van Riemsdijk (ed.), 33–81. Berlin: de Gruyter. Chomsky, Noam. 2010. Restricting stipulations: consequences and challenges. Lecture. University of Stuttgart. 24 March, 2010. de/upload_data_www/534/3637/aufzeichnung.flv? 124,234 Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (ed.), 1–52. Cambridge/MA: MIT Press.

Doubly-Filled Comp, wh head-movement, and derivational economy

37

Chomsky, Noam. 1995a. Bare phrase structure. In Government and Binding Theory and the Minimalist Program, Gert Webelhuth (ed.), 383–439. Oxford: Blackwell. Chomsky, Noam. 1995b. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam, and Howard Lasnik. 1977. Filters and control. Linguistic Inquiry 8: 425–504. van Craenenbroeck, Jeroen. 2006. Transitivity failures in the left periphery and foot-driven movement operations. Linguistics in the Netherlands 23: 52–64. Donati, Caterina. 2006. On wh-head movement. In Wh-movement: Moving on, Lisa Cheng, and Nobert Corver (eds.), 21–46. Cambridge/MA: MIT Press. Fanselow, Gisbert. 2004. Münchhausen-style head movement and the analysis of verb second. In Linguistics in Potsdam 22, [Ralf Vogel (ed.), Three Papers on German Verb Movement]: 9–49. Fuß, Eric. 2005. The Rise of Agreement. A Formal Approach on the Syntax and Grammaticalization of Verbal Inflection. Amsterdam: Benjamins. Gelderen, Elly van. 2004. Grammaticalization as Economy. Amsterdam: Benjamins. Georgi, Doreen, and Gereon Müller. 2010. Noun-phrase Structure by reprojection. Syntax 13:1– 36. Gorrell, Paul. Structural relations in the grammar and the parser. Special issue of Folia Linguistica 28 (1–2). Josef Bayer (ed.), 1995. Syntax and parsing. Cambridge: Cambridge University Press. Groenendijk, Jeroen, and Martin Stokhof. 1982. Semantic analysis of wh-complements. Linguistics and Philosophy 5: 175–233. Gutch, Donald. 1992. Linking and intrusive r in English and Bavarian. In Language and Civilization: A Concerted Profusion of Essays and Studies in Honor of Otto Hietsch. Vol. 1. Claudia Blank (ed.), 555–611. Frankfurt a. Main: Lang. Haan, Germen de, and Fred Weerman. 1986. Finiteness and verb fronting in Frisian. In Verb Second Phenomena in Germanic Languages, Hubert Haider, and Martin. Prinzhorn (eds.), 77–110. Dordrecht: Foris. Haegeman, Liliane. 1992. Theory and Description in Generative Syntax: A Case Study in West Flemish. Cambridge: Cambridge University Press. Higginbotham, James. 1997. The semantics of questions. In The Handbook of Contemporary Semantic Theory, Shalom Lappin (ed.), 361–383. Oxford: Blackwell. Higginbotham, James. 1993. Interrogatives. In The View from Building 20, Kenneth Hale, and Samuel J. Keyser (eds.), 195–228. Cambridge MA: MIT-Press. Hodler, Werner. 1969. Berndeutsche Syntax. Bern: Francke. Höhle, Tilman. 1986. Der Begriff “Mittelfeld”. Anmerkungen über die Theorie der topologischen Felder. In Akten des Siebten Internationalen Germanistenkongresses 1985, Göttingen, Germany, Albrecht Schöne (ed.), 329–340. Tübingen: Niemeyer. Hoekstra, Eric. 1993. Dialectal variation inside CP as parametric variation. In Dialektsyntax, Werner Abraham, and Josef Bayer (eds.), 161–179. (Special issue #5 of Linguistische Berichte) Opladen: Westdeutscher Verlag. Jäger, Agnes. 2000. Unterspezifikation am Beispiel des Pronomens was. Zur Grammatik eines w-Elements. MA thesis, Friedrich Schiller Universität Jena. Kathol, Andreas. 2000. Linear Syntax. Oxford: Oxford University Press. Kayne, Richard S. 2010. Why isn’t this a complementizer?. In Comparison and Contrasts, Richard S. Kayne (ed.), 190–227. Oxford: Oxford University Press. Koeneman, Olaf. 2002. The distribution of declarative verb second in Germanic. In Studies in Comparative Germanic Syntax, Jan-Wouter Zwart, and Werner Abraham (eds.), 175–201. Amsterdam: Benjamins.

38

Josef Bayer

Koeneman, Olaf. 2000. The Flexible Nature of Verb Movement. Utrecht: LOT Publications. Koopman, Hilda. 2000. The Syntax of Specifiers and Heads. London: Routledge. Kühnert, Henrike, and Esther-Miriam Wagner. 2004. Konnektive in der diachronen Entwicklung des Jiddischen. In Indogermanistik – Germanistik – Linguistik, Maria Kozianka, Rosemarie Lühr, and Susanne Zeilfelder (eds.), 261–299. Hamburg: Dr. Kovac. Lahiri, Utpal. 2002. Questions and Answers in Embedded Contexts. Oxford: Oxford University Press. Lobeck, Anne. 1995. Ellipsis. Oxford: Oxford University Press. Manzini, Maria Rita. 2012. The status of complementizers in the left periphery. In Main Clause Phenomena. New Horizons, Lobke Aelbrecht, Liliane Haegeman, and Rachel Nye (eds.), 297–318. Amsterdam: Benjamins. Manzini, Maria Rita, and Leonardo Savoia. 2003. The nature of complementizers. Rivista di grammatica generative 28: 87–110. Merchant, Jason. 2001. The Syntax of Silence. Oxford: Oxford University Press. Merchant, Jason. 2006. Sluicing. In The Syntax Companion, Martin Everaert, and Henk van Riemsdijk (eds.), 269–289. London: Blackwell. Merkle, Ludwig. 1984. Reprint. Bairische Grammatik. 2nd ed. München: Hugendubel. Original edition, München: Heimeran Verlag, 1975. Merz, Michael. 2011. Die [r]-Epenthese im Bairischen – zwischen Phonologie und Syntax. MA thesis, University of Konstanz. Müller, Gereon. 2004. Verb-second as vP-first. Journal of Comparative Germanic Linguistics 7: 179-234. Noth, Harald. 1993. Alemannisches Dialekthandbuch vom Kaiserstuhl und seiner Umgebung. Freiburg im Breisgau: Schillinger. Ortmann, Albert. 1998. Consonant epenthesis: Its distribution and phonological specification. In Phonology and Morphology of the Germanic Languages, Wolfgang Kehrein, and Richard Wiese (eds.), 51–76. Tübingen: Niemeyer. Pankau, Andreas. 2010. Wh-Copying in German and its Theoretical Implications, University of Frankfurt & University of Utrecht. Handout. Penner, Zvi. 1993. W-morphology in the COMP system of Bernese Swiss German and the licensing of empty operators in the prefield position. In Dialektsyntax, Werner Abraham, and Josef Bayer (eds.), 201–212. (Special issue # 5 of Linguistische Berichte) Opladen: Westdeutscher Verlag. Penner, Zvi and Thomas Bader. 1995. Issues in the syntax of subordination: A comparative study of the complementizer system in Germanic, Romance, and Semitic languages with special reference to Bernese Swiss German. In Topics in Swiss German Syntax, Zvi Penner (ed.) 73–289. Bern. Pesetsky, David, and Esther Torrego. 2001. T-to-C movement: Causes and consequences. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 355–426. Cambridge MA: MIT Press. Pfalz, Anton. 1918. Suffigierung der Personalpronomina im Donaubairischen’ Kaiserliche Akademie der Wissenschaften in Wien, philosophisch-historische Klasse, Sitzungsberichte 190. Platzack, Christer. 1996. Germanic verb second – attract versus repel: On optionality, A-bar movement and the symmetrical/asymmetrical verb second hypothesis. In Deutsch – typologisch, Ewald Lang, and Gisela Zifonoun (eds.), 92–120. Berlin: de Gruyter. Reis, Marga. 1985. Satzeinleitende Strukturen im Deutschen. Über COMP, Haupt – und Nebensätze, w-Bewegung und die Doppelkopfanalyse. In Erklärende Syntax des Deutschen, Werner Abraham (ed.), 271–311. Tübingen: Narr.

Doubly-Filled Comp, wh head-movement, and derivational economy

39

Reuland, Eric. 1990. Head movement and the relation between morphology and syntax. In Yearbook of Morphology 3, Geert Booij, and Jaap van Marle (eds.), 129–161. Dordrecht: Foris. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of Grammar, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. Roberts, Ian. 2010. Agreement and Head Movement. Cambridge MA: MIT Press. Rosenbaum, Peter. 1967. The Grammar of English Predicate Complement Constructions. Cambridge MA: MIT Press. Ross, John R. 1969. Guess who?. In Papers from the Fifth Regional Meeting of the Chicago Linguistic Society (CLS), Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan (eds.), 252–286. Montreal: Montreal Working Papers in Linguistics. Schiepek, Josef. 1899. Der Satzbau der Egerländer Mundart. Prag: Verlag des Vereines für Geschichte der Deutschen in Böhmen. Schönenberger, Manuela. 2006. A glimpse of doubly-filled COMPs in Swiss German. In Organizing Grammar: Studies in Honor of Henk van Riemsdijk, Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz, and Jan Koster (eds.), 572–581. Berlin: de Gruyter. Seiler, Guido. 2003. Präpositionale Dativmarkierung im Oberdeutschen. Stuttgart: Steiner. Starke, Michal. 2009. Nanosyntax. A short primer to a new approach to language. Nord-lyd 36 (1): 1–6. Special issue on Nanosyntax, ed. Peter Svenonius, Gillian Ramchand, Michal Starke, and Knut Tarald Taraldsen, pp. 1–6. CASTL, Tromsø. http://www.ub.uit.no/baser/ nordlyd/ Steininger, Reinhold. 1994. Beiträge zu einer Grammatik des Bairischen. Stuttgart: Steiner. Sternefeld, Wolfgang. 1997. Comparing Reference Sets. In The Role of Economy Principles in Linguistic Theory, eds. Chris Wilder, Hans-Martin Gärtner and Manfred Bierwisch, 81–114. Berlin: Akademie Verlag. Surány, Balázs. 2003. Head movement and reprojection. Annales Universitatis Scientiarum Budapestinensis de Rolando Eötvös Nominatae. Sectio Linguistica. Tomus XXVI: 313–342. Budapest: ELTE. Vangsnes, Øystein A. 2005. Microparameters for Norwegian wh-grammars. Linguistic Variation Yearbook 5: 187–226. Weiß, Helmut. 2005. Inflected complementizers in continental West Germanic dialects. Zeitschrift für Dialektologie und Linguistik 72: 148–166. Weiß, Helmut. 2004. Vom Nutzen der Dialektsyntax. In Morphologie und Syntax deutscher Dialekte und historische Dialektologie des Deutschen. Beiträge zum 1. Kongress der Internationalen Gesellschaft für Dialektologie des Deutschen, Marburg/Lahn, 5.–8. März 2003, eds. Franz Patocka and Peter Wiesinger, 21–41. Vienna: Praesens. Weiß, Helmut. 1998. Die Syntax des Bairischen. Studien zur Grammatik einer natürlichen Sprache. Tübingen: Niemeyer. Westergaard, Marit R., and Øystein Vangsnes. 2005. Wh-questions, V2, and the left periphery of three Norwegian dialect types. Journal of Comparative Germanic Linguistics 8 (1–2): 119–160. Wiese, Richard. 1996. The Phonology of German. Oxford: Oxford University Press. Zimmermann, Malte. 2011. Quantificational structures in low German: On the functional structure of DP and the feature content of pronominal quantifiers. Journal of Comparative Germanic Syntax 14 (3): 203–240.

Norbert Corver

Interjections as structured root expressions 1 Knowledge of interjections1 Knowledge of interjections and their grammatical behavior is part of our knowledge of language. Yet, interjections belong to the least studied parts of speech, and according to some rightly so given that “Interjections are among the least important of speech elements;” Sapir (1921: 5). At first sight, there seem to be good reasons for this lack of interest in interjections. From a syntactic perspective, their behavior does not seem to be particularly interesting. They appear to be syntactic isolates that do not connect to other categories within the clause. Also semantically, interjections differ from lexical categories such as nouns, verbs and adjectives. They are often, though not always as we will see, used to express an isolated emotion or sentiment on the part of the speaker. The nature of this expressive meaning seems harder to capture than the more descriptive meaning of nouns, verbs and adjectives. For some reason, the meaning of wow! seems harder to define than that of amazing! or beautiful!. And also the meaning of interjective shit! is harder to grasp than the noun shit in the shit on your shoe. Also at the sound level, interjections display “different” behavior, in the sense that sound properties are found in interjections that are not attested in other parts of speech. For example, the Dutch interjections njah (/nja/; meaning: ‘I (the speaker) affirms X, but with hesitation’), psst (/pšt/; meaning: ‘I want to get your attention’), sht (/št/; meaning: ‘I want you to be quiet’), pf (/pf/; meaning: irony) display sequences of phonemes (here represented in boldface) that are not attested in other parts of speech in Dutch, such as nouns, verbs and adjectives. In other words, interjections have their own phonotactic properties. In spite of this linguistically isolate behavior, it seems that interjections are nevertheless linguistically significant and worthwhile to investigate for a number of reasons.2 First of all, the language learning child is able to identify interjections as language material in the noise that she hears in her environment. In other words, just like the sound sequences cow, couch and loops are 1 Parts of this paper have been presented at the Annesso Cartesiano Inaugural Workshop, Villa Salmi, Stoppe d’Arca/Arezzo. I would like to thank the participants of the workshop and an anonymous reviewer for valuable comments and suggestions. All errors are mine, obviously. 2 For a summary of the study of interjections in the history of linguistics, see De Boer (2008).

42

Norbert Corver

identified as linguistically meaningful units, so are the interjections wow!, ouch!, and oops!. In short, interjections are acquired during language development and become part of our knowledge of language. If it is the linguist’s task to describe and explain our knowlegde of language (see Chomsky 1986), then interjections should also be part of that research agenda. Secondly, interjections are a part of speech that can be found cross-linguistically (De Groot 1963). In other words, it seems to be a universal part of speech. Thirdly, interjections may give us a mirror of a part of the mind that we are less familiar with, viz., our emotion system. Certain interjections highlight the affective or emotional state of a speaker. They constitute what Scherer (1994: 170) calls affect bursts, that is “very brief, discrete, nonverbal expressions of affect in both face and voice as triggered by clearly identifiable events.”3 These linguistic affect bursts differ from expressive signals in animal communication systems in an important way: they are innovative and free from control by external stimuli. One and the same emotion-eliciting event may trigger the use of different but yet appropriate interjections. For example, if you hit with a hammer on your thumb, you may express your negative feelings by using different interjections: e.g., Darn!/Ouch!/Gee!/Argh!. In other words, an emotion-triggering stimulus does not trigger a specific interjection. In this respect, the use of interjective expressions by human beings is radically different from the use of expressive signals in animal communication systems. Besides these more general reasons for including interjections in the study of language, there are also linguistic-analytic reasons. As we will see in this article, it turns out that interjective expressions can have a complex structure. This raises the question as to what this structure looks like; i.e. what are the atoms from which the structured interjective expression is built and what are the mechanisms that underlie this structure building? In order to show the structured nature of interjective expressions, I will base myself on interjective expressions from Dutch, more specifically those interjections that have a soundsymbolic meaning. The article is organized as follows: section 2 shows that interjections can be complex and that these complex interjections are structured expressions (i.e, they display constituency). In section 3 I argue that interjections are root categories, and that complex interjections are coordinations of roots. In section 4 evidence is presented in support of a syntactic, phrase structural analysis of complex interjections. That is, they are not complex words (more specifically, compounds). In section 5 I propose that their phrase structural organization is

3 I think ‘nonverbal’ should be interpreted here as ‘non-sentential/non-propositional’.

Interjections as structured root expressions

43

similar to that of coordinate structures. More specifically, they have a hierarchically organized (i.e., X-bar-theoretic) structure: i.e., ConjP. Section 6 presents arguments against a flat-structure analysis of complex interjections and in favor of a ConjP-analysis. In section 7 it will be shown that the externalization of complex interjections sometimes involves different types of phonological operations that augment the sound representation of the interjective expressions. Section 8 concludes the article.

2 On the constituency of complex interjections In this section I will examine the internal syntax of interjections by focusing on sound symbolic ones, i.e., those interjections whose meaning is a sounddesignating one. I will argue that these interjections can have a complex syntactic structure and, consequently, are structured expressions. Let us start our investigation with the following examples (cf. Pauwels 1952): (1)

a.

b.

c.

d.

De the

auto car

reed drove

Jan Jan

zakte fell

krak CRACK

De the

kogel bullet

vloog flew

De the

vaas vase

viel fell

boem BANG

tegen against

door through zjoef ZJOEF

baf BAM

het the

over over

op onto

de the

de the

muur wall

ijs ice

heen. PRT

zijn his

aan. PRT

hoofd. head

grond. floor

The interjection can be substituted for by synonomous interjective words: (2)

a.

De the

auto car

reed drove

boem BAM

tegen against

de the

b.

De auto reed knal tegen de muur aan.

c.

De auto reed boink tegen de muur aan.

muur wall

aan. PRT

As shown by the following examples, it is also possible to have a sequence of interjections:

44

Norbert Corver

(3) a. Jan reed [boem] tegen de muur aan. against the wall PRT Jan drove BAM ‘Jan drove bang against the wall.’ b. Jan reed [pats boem] tegen de muur aan. against the wall PRT Jan drove WHAM BAM c. Jan reed [pats boem knal] tegen de muur aan. BANG against the wall PRT Jan drove WHAM BAM d. Jan reed [pats boem knal boink] tegen de muur aan. BANG BOINK against the wall PRT Jan drove WHAM BAM The order of these elements is not fixed, which suggests that these sequences of interjections are not unanalyzable, fixed patterns.4 (4)

Jan reed

.....

[pats boem knal boink] . . . . tegen de muur aan. [boem knal pats boink] [boink pats boem knal]

Importantly, the sequence of interjections in (3) and (4) constitutes a syntactic unit (i.e., a constituent). This constituency of the sequence of interjections cannot be shown, however, on the basis of standard constituency tests such as displacement (i.e., does the sequence of interjections behave as a unit with respect to reordering operations?) and substitution (i.e., can the sequence of interjections be replaced by a substituting pro-form?). Let us consider displacement first. At first sight, the examples in (5), which are more or less acceptable, suggest that interjections cán be displaced and that, consequently, displacement can be used as a test for constituency. As one can observe, the interjection immediately precedes the finite verb of the main clause, which is often taken to occupy the head position of CP (i.e., the well-known Verb-second property; see, among others, Den Besten (1983) and Koster (1975)). Thus, one might hypothesize that the interjection in (5) occupies the specifier position of the functional head C, which is the landing site for the displaced finite verb; see (6): (5)

a.

(?)Boem

b.

(?)Krak

zakte Jan door het ijs heen.

c.

(?)Zjoef

vloog de kogel over zijn hoofd.

d.

(?)Baf

reed de auto tegen de muur aan.

viel de vaas op de grond.

4 In the case of frequently used two-membered interjections the order is often conventionalized. For example, the sequence ??boem pats is odd compared to pats boem. See also note 21.

Interjections as structured root expressions

(6)

45

[CP Boem [C’ reed [TP de auto boem tegen de muur aan reed]]]

According to this line of reasoning the sequence of interjections preceding the finite verb in (7a) would constitute a syntactic unit occupying Spec,CP, as in the representation in (7b): (7)

a.

(?)Pats,

b.

[CP Pats, boem, knal, boink [C’ reed [TP de auto pats, boem, knal, boink tegen de muur aan reed]]]

boem, knal, boink reed de auto tegen de muur aan.

The existence of the minimally different patterns in (8), however, raises the question as to whether the patterns in (5) and (7a) really involve displacement of an interjective expression to Spec,CP. In (8), the locative d(emonstrative)-word daar ‘there’ immediately precedes the finite verb and the interjection behaves more like a base-generated (i.e., E(xternally)-merged) left-peripheral element, quite analogously to the vocative noun phrases in (9) that occur in a left peripheral position within the clause. See (10) for the relevant representations. (8)

a.

Boem, BAM

b.

Krak, CRACK ,

c.

Zjoef, ZJOEF,

d.

Baf, BAM ,

(9)

a.

b.

daar there

reed drove

de the

auto car

daar there

zakte fell

Jan Jan

door through

daar there daar there

vloog flew viel fell

de the

de the

kogel bullet

vaas vase

tegen against het the

over over

op onto

muur wall

ijs ice

heen. PRT

zijn his

de the

Jan, daar rijdt iemand tegen Jan, there drives someone against ‘Jan, someone just drove into your gate!’

de the

aan. PRT

hoofd. head

grond. ground

je your

Jan, daar vliegt een vogel over je Jan, there flies a bird above your ‘Jan, there is a bird flying above your head!’

hek gate

aan! PRT

hoofd! head

(10) a. [CP Boem [CP daar [C’ reed [TP de auto daar tegen de muur aan reed]]]] (8a) b. [CP Jan [CP daar [C’ rijdt [TP iemand daar tegen je hek aan rijdt]]]]

(9a)

46

Norbert Corver

One might hypothesize now that the patterns in (5) have the same syntactic structure as those in (8), the only difference being that in (5) the d-pronoun daar is phonetically empty; i.e., we have a null d-word: (11)

[CP Boem [CP daar [C’ reed [TP de auto daar tegen de muur aan reed]]]]

(5a)

The possibility of dropping daar is reminiscent of the phenomenon of Topicdrop in Dutch (cf. Weerman 1989, Zwart 1993). A demonstrative word like daar can be omitted when it is topical, as is exemplified in (12), where the erased d-pronoun (daar) represents the base position of the fronted d-word. (12)

a.

b.

A:

Weet know

jij you

de the

weg way

in in

Parijs? Paris

B:

Jazeker! sure!

(Daar) (there)

heb have

ik I

jaren years

daar

A:

Ben are

jij you

wel PRT

eens once

in in

Parijs Paris

geweest? been

B:

Nee, no,

(daar) (there)

ben am

ik I

nog yet

nooit never

daar

gewoond! lived

geweest. been

In (12), the topical status of the locative-demonstrative daar is induced by the presence of the antecedent in Parijs in speaker A’s utterance. Plausibly, the topical status of daar in (8) is induced by the situational or discourse context; daar refers to the situationally given or available locus where the event takes place. Given the parallelism with the topic-drop phenomenon in (12), it does not seem implausible to say that the examples in (5) feature an empty d-word in Spec,CP; see (11). This implies that the left peripheral interjection has not been moved to Spec,CP but is rather base-generated in the left periphery of the main clause. Further support for the idea that interjections cannot undergo displacement to Spec,CP and are simply base-generated in their surface position comes from patterns involving a long distance relationship. If interjections were accessible to A-bar movement to Spec,CP, one would expect them to be able to undergo long distance A-bar movement. As shown in (13a’,b’), however, such a long distance relationship with a ‘base position’ is impossible.5 Under a base-generation 5 The long distance pattern is also impossible when the main clause has the locative d-word daar in Spec,CP (Compare with (8a)): (i) *Boem daar zei Kees [dat de auto tegen een muur was aangereden]. BAM there said Kees that the car against a wall was PRT-driven

Interjections as structured root expressions

47

analysis of interjections, one might try to interpret this locality effect in terms of scope, i.e., the sound interjection must be part of the (minimal) clausal domain that represents the event associated with the sound. (13) a.

Kees zei dat Jan boem tegen een muur was aangereden. wall was PRT-driven Kees said that Jan BAM against a a’. *Boem zei Kees [dat Jan tegen een muur was aangereden]. b.

Jan zei dat de kogels zjoef over zijn hoofd vlogen. Jan said that the bullets ZJOEF over his head flew b.’ *Zjoef zei Jan [dat de kogels over zijn hoofd vlogen]. So far, I have argued that interjections are not subject to displacement. Consequently, displacement cannot be used as a constituency test for a linear sequence of interjections, like in (3) and (4). Having shown this, I will now turn to substitution and argue that also this constituency test cannot be used for identifying a linear sequence of interjections as a constituent. The impossibility of using substitution for determining whether a string of interjections forms a syntactic unit is exemplified in (15): it turns out that even simplex interjections cannot be substituted for by what seems to be the most appropriate pro-form, viz., the adverbial pro-form zo ‘so’. This is illustrated in (14) and (15), where (14) shows the occurrence of the manner pro-form zo in clauses, and (15) the impossibility of substituting a sound-symbolic interjection by means of this pro-form:6 (14)

a.

Jan reed zo tegen de muur aan. Jan drove so against the wall PRT ‘Jan drove right into the wall/Jan drove into the wall this way’

b.

Jan zakte zo door het ijs heen. Jan fell so through the ice PRT ‘Jan went right through the ice/Jan went through the ice this way’

6 As noted by Pauwels (1952), interjections can sometimes be paraphrased by an adjunct-PP headed by met ‘with’ (ib). In those examples, the met-PP can be substituted for by the proform zo (see (ic)). (i) a. Jan reed boem tegen de muur aan. b. Jan reed [PP met een knal] tegen de muur aan. Jan drove with a bang against the wall PRT c. Jan reed [met een knal] tegen de muur aan en Jo reed zo tegen het hek aan. Jan drove with a bang against the wall PRT and Jo drove so against the gate PRT

48 (15)

Norbert Corver

a. *Jan Jan

reed drove

reed drove b. *Jan Jan zo so

zo so

boem

tegen against

BAM

tegen against

een a

krak

door through

zakte fell

CRACK

door through

de the

een a

boom tree het the

houten wooden

vloer floor

muur wall

aan PRT

en and

Kees Kees

aan. PRT ijs ice

heen PRT

en and

Kees Kees

zakte went

heen. through.

Since substitution of an interjection by a pro-form does not work for simplex interjections, it obviously cannot be used as a constituency test for a linear sequence of interjections either. Fortunately, a few other tests can be used for showing the constituency of the complex interjection patterns in (3) and (4). First of all, a test that hints at the constituency of the complex interjection is the one that states that the string of words should be able to stand alone as a fragment of sentence (Carnie 2008: 18): (16)

A: B:

Hoe how

klonk sounded

het the

geluid noise

van of

[Pats,

boem,

knal,

boink].

WHAM

BAM

BANG

BOINK

de the

botsing? collision

A second piece of evidence for their constituency comes from the fact that two sequences of interjections can be coordinated. In other words, a complex interjection can function as a conjunct (i.e., a syntactic unit) in a coordinate structure: (17)

a.

b.

De the

opgestapelde piled-up

blikken cans

[boem,

pats,

klets]]

BAM

BOINK

WHACK ]

[[Rinkel

de

kinkel]

RINKEL

DE

KINKEL

het the

glazen glass

tafelblad table-top

vielen fell

kapot broken

en and op onto

[[knal, [BANG op onto

[bam

boem

BANG

BAM

de the

de the

grond. floor

beng,

doink]

BAM

BOINK

en and

grond. floor

beng]], BOINK ,

daar there

viel fell

Interjections as structured root expressions

49

A third test that suggests that the sequence of interjections in (3) and (4) forms a structural unit is based on “interruption”: if a string of elements W, Y and Z forms a syntactic unit XP, then it is impossible to move an XP-external element to a position internal to XP. More specifically, it is impossible to intersperse the sequence of interjections with some XP-external element α, e.g., by means of displacement. Schematically:7 (18)

. . . . . .[XP W (*αi) Y (*αi) Z] . . . . ti . . . . . where t is the trace (copy) of the displaced element α

This impenetrability of the sequence of interjections is illustrated in (19c), where the so-called R-pronoun daar is removed from within a PP (see Van Riemsdijk 1978) and moved to a position in between two interjections. As shown in (19a), it is possible for the R-pronoun to occupy a position following the sequence of interjective atoms. (19b) illustrates that the R-pronoun can also occupy a (PP-external) position preceding the complex interjection. As shown by (19c), what is not possible is placement of the R-pronoun in a position in between the various interjective atoms.8 (19) a. Jan is [pats boem knal boink] daar tegenaan gereden. Jan has WHAM BAM BANG BOINK there against driven ‘Jan WHAM BAM BANG BOINK drove against it.’ b. Jan is daari (gisteren) [pats boem knal boink] [ti tegenaan] gereden. against driven Jan has there (yesterday) WHAM BAM BANG BOINK ‘Jan WHAM BAM BANG BOINK drove against it (yesterday).’ 7 At a more technical level, one may try to implement this restriction in terms of c-command. A displaced constituent must c-command its trace-position. If a constituent moves into a phrase XP, then this constituent no longer has c-command over its base position. This also explains the ill-formedness of (ib), where the complement of A (a non-case-assigning category) has been moved into the specifier position of the subject noun phrase, where it may receive genitive case. This displacement operation is illegitimate because the displaced element does not c-command its trace. (i) a.

[Teachers] were [AP afraid John]

(“base structure”)

b. *[Johni’s teachers] were [AP afraid ti] (derived structure) intended reading: ‘Teachers were afraid of John.’ 8 As pointed out by a reviewer, it is sometimes possible to have two separate (complex) interjections with the R-pronoun in between. Consider, for example, (ib), which is a variant of (ia). (i) a. Jan is gisteren zoef zoef zoef hupsakee er tegenaan gereden. Jan is yesterday ZOEF ZOEF ZOEF HUPSAKEE there against driven b. Jan is gisteren zoef zoef zoef eri hupsakee ti tegenaan gereden

50

Norbert Corver

c. Jan is gisteren [pats (*daari) boem (*daari) knal (*daari) boink] [ti Jan has yesterday WHAM (there) BAM (there) BANG (there) BOINK tegenaan] gereden. against driven Compare this with the placement of the R-pronoun daar with respect to a sequence of adverbial elements that function as modifiers within the clause: (20) a. Jan is [waarschijnlijk] [toen] [hard] tegen de muur aan gereden. Jan has probably then hard against the wall PRT driven ‘Jan probably then drove hard against the wall.’ b. Jan is waarschijnlijk toen hard daar tegen aan gereden. c. Jan is waarschijnlijk toen daar hard tegen aan gereden. d. Jan is waarschijnlijk daar toen hard tegen aan gereden. e. Jan is daar waarschijnlijk toen hard tegen aan gereden. These examples show that the R-pronoun can be placed in a position in between two adverbial elements. This follows from an analysis in which the adverbials are merged separately, i.e., not as a single complex unit, with different layers of the clausal structure (see, for example, Cinque (1999)). Schematically, where UP, WP, YP and ZP are layers in the clausal structure: (21) . . . .[ZP [waarschijnlijk] [YP [ toen] [WP [hard] [UP [PP daartegenaan] gereden]]]] A final phenomenon suggesting the constituency of a string of interjections as in (3)–(4) comes from their occurrence as “quotative” complements in the construction type zo van X (‘so of X’), where zo is kind of a cataphoric adverb whose contents is provided by the quotative material following van. The represented direct speech approximates (i.e., ‘is like’) the contents of the speech (cf. (22a,b)) or facial expression (cf. (22c))) used by the agent (i.e., the subject of the matrix clause). Clearly, the complement of (zo) van is a syntactic unit, e.g., a clause (22a,b) or a noun phrase (22c). (22) a. Mijn vader zei zo van “Drink niet teveel bier!” my father said so of “Drink not too-much beer!” ‘My father said like: “Don’t drink too much beer!”’

Interjections as structured root expressions

51

b. Hij vroeg zo van “Heb jij je huiswerk al gemaakt?!” he asked so of “Have you your homework already made?” ‘He asked something like the following: “Have you finished your homework?”’ c. Hij keek me aan met een blik zo van: “Wat een eikel!” gaze so of: “What a jerk!” he looked-at me PRT with a ‘He looked at me with a gaze expressing: “What a jerk!”’ As exemplified in (23), (complex) interjections can also occupy the X-slot of the zo van X-construction. This suggests that the sequence of interjections forms a syntactic unit. (23) a. Jan reed [zo van [pats boem knal]] tegen een muur aan. wall PRT Jan drove so of WHAM BAM BANG against a b. De opgestapelde blikken vielen [zo van [knal, beng, doink]] the piled-up cans fell so of BANG BAM BOINK op de grond. onto the ground In sum, although certain constituency tests cannot be used for showing the constituency of a string of interjections like (3), there are a number of other tests that support an analysis according to which a sequence of interjections forms a syntactic unit (a constituent). Besides complex interjections like those in (3), which consist of a range of different interjective words, there are also complex interjections that are entirely built up on the basis of the same lexical item. This reduplicative pattern is exemplified in (24b) and (25b): (24)

a.

b.

(25)

a.

b.

Jan Jan

sloeg hit

[boem]

Jan Jan

sloeg hit

[boem

boem

boem]

BAM

BAM

BAM

De the

kogels bullets

vlogen flew

De the

kogels bullets

vlogen flew

BAM

op on

de the

[zjoef ] ZJOEF

trommel. drum

over over

op on

de the

zijn his

trommel. drum

hoofd. head

[zjoef,

zjoef,

zoef ]

ZJOEF

ZJOEF

ZJOEF

over over

zijn his

hoofd. head

52

Norbert Corver

Interestingly, besides the pattern in (24b,25b), in which the interjections are juxtaposed, we find patterns in which they are separated from each other by means of a linking element de: (26)

a.

b.

Jan Jan

sloeg hit

[boemer

de

boem]

BOEMER

DE

BOEM

op on

de the

trommel. drum

Jan sloeg [boemer de boemer de boemer de boem] op de trommel.

In order to show that this sound symbolic interjective pattern manifests itself in various guises, I add a few more examples:9 (27)

a.

b.

c.

d.

e.

Jan Jan

viel fell

holder

de

bolder

(de

bolder

de

bolder

HOLDER

DE

BOLDER

( DE

BOLDER

DE

BOLDER

de

bolder

DE

BOLDER

Jan Jan

zakte went

naar to

de

krak

DE

KRAK

De the

hakken heels

de

klak

DE

KLAK

De the

kinkel

DE

KINKEL

Ze they

krakker

(de

krakker

de

krakker

( DE

KRAKKER

DE

KRAKKER

door through

ruit window

de

beneden. downstairs

KRAKKER

reden drove

de

bobbel

DE

BOBBEL

de the

tikten tapped

op on

de the

. .) . .)

stoel. chair

klikker

(de

klikker

de

klikker

KLIKKER

( DE

KLIKKER

DE

KLIKKER

. .) . .)

vloer. floor

brak broke in into

. .) . .)

rinkel

(de

kinkel

de

kinkel

RINKEL

( DE

KINKEL

DE

KINKEL

. .) . .)

stukken. pieces

hobbel

(de

bobbel

de

bobbel

HOBBEL

( DE

BOBBEL

DE

BOBBEL

over across

de the

. .) . .)

keienweg. stone-way

9 In these complex interjective expressions containing the linking element de, there is a strong preference for the non-final members to end in -er: boemer de boem versus ?*boem de boem; krakker de krak versus *?krak de krak. The pattern rinkel de kinkel does not have a counterpart in which -el is absent on the first interjective atom: *rink de kinkel. In section 7 I will briefly come back to the possible “role” of -er and -el in these complex interjective expressions.

53

Interjections as structured root expressions

Also with these examples, it is impossible for a displaced element (e.g., the R-pronoun daar) to interrupt the sequence of interjections. This is shown in (28) for the iterative pattern lacking the linking element de, and in (29) for the pattern featuring the linker de. This non-interruptability suggests that the string of interjections constitutes a syntactic unit XP that cannot be penetrated by XPexternal material. (28)

(29)

[PP

daar there

op]. on

a.

Jan sloeg boem boem boem BOEM BOEM BOEM Jan hit ‘Jan BAM BAM BAM hit on it.’

b.

Jan sloeg boem (*daari ) boem (*daari ) boem [PP ti op].

c.

Jan sloeg daari boem boem boem [PP ti op].

a.

Jan sloeg boemer de boemer BOEMER DE BOEMER Jan hit ‘Jan BAM BAM BAM hit on it.’

de

boem

DE

BOEM

[PP

daar there

b.

Jan sloeg boemer de (*daari ) boemer de (*daari ) boem [PP ti op].

c.

Jan sloeg boemer (*daari ) de boemer (*daari ) de boem [PP ti op].

d.

Jan sloeg daari boemer de boemer de boem [PP ti op].

op]. on

There is another important property exemplified by the patterns in (27), and also by those in (24b) and (25b). The complex interjective expressions can be arbitrarily long; i.e., they display the property of discrete infinity. In section 5 I will propose that the complex interjective expressions in (3), (4), (24b), (25b), (26) and (27) have a hierarchically organized, right-branching (tail-)recursive structure and are built by a generative procedure (Merge) that takes interjective atoms as its input. More specifically, I will argue that these interjective expressions have a coordinate structure, as is illustrated in (30a) for the pattern pats boem knal boink in (3d) and in (30b) for the pattern boemer de boemer de boemer de boem in (26b). As shown by these representations, the coordinate structures (ConjP) (i) knal boink, (ii) boem, knal, boink, (iii) boemer de boem and (iv) boemer de boemer de boem are embedded within a larger coordinate structure (ConjP). Notice further that in (30a), the coordinate structure is asyndetic (i.e., there is no overt coordinating conjunction), whereas in (30b) it is not, given the presence of the overt connecting element de.

54 (30)

Norbert Corver

a.

[ConjP pats [Conj’ [Conj ø] [ConjP boem [Conj’ [Conj ø] [ConjP knal [Conj’ [Conj ø] [ConjP boink]]]]]]]

b.

[ConjP boemer [Conj’ [Conj de [ConjP boemer [Conj’ [Conj de] [ConjP boemer [Conj’ [Conj de] [ConjP boem]]]]]]]

Summarizing, on the basis of diagnostic tests such as substitution (15), independent occurrence (16), coordination (17), reduplication (24–25) and (unbounded) expansion with added interjective atoms (e.g., (27)), I have tried to show that complex interjective expressions, which appear to be fixed unanalyzable units, can be structurally decomposed into smaller units (interjective atoms). The combination of these units yields a complex interjective expression that I take to have a hierarchically organized recursive structure (cf. section 5). This structural decomposition of Dutch interjective expressions brings us to the next question: what is the categorial nature of the interjective atoms that function as constituents in complex interjective expressions? This question will be addressed in the next section.

3 Sound-symbolic atoms as roots In the spirit of Borer’s Exo-Skeletal Model (2005a,b) and much recent work on Distributed Morphology (Halle and Marantz 1993, Harley and Noyer 1999), I will assume that interjective atoms such as boemer, rinkel, pats, boem, boink, et cetera are roots, i.e., lexical vocabulary items that are not specified for categorial information (or any other formal-syntactic features).10 According to these theories, the categorial status is determined by the structure on top of the root. In DM, for example, it is a categorial marker (e.g., n, v) that merges with the root and determines the categorial nature of the projected structure. The root (dog becomes nominal by combining with a categorial head no: [n n [(dog]]. In Borer’s Exo-Skeletal Model, it is not an empty categorial head that determines the categorial status of the ‘nominal projection’ but rather a functional category like the definite article the as in the dog: [DP [D the] [(dog]]. Some further illustrations that functional material ‘brings in’ categorial information is given in (31) for the Dutch root (wit ‘white’. In (31a), the nominal nature of the phrasal structure het wit is determined by the definite article het. In (31b), the verbal nature of witte is determined by the past tense morphology (T) 10 See also Myers (1984) for the idea that inflection determines category in cases of conversion. For criticism of this approach to conversion, see Don (1993).

Interjections as structured root expressions

55

–te on (wit. In (31c), it is the attributive adjectival inflection –e that determines the adjectival nature of witte. (31)

a.

Marie is in het wit. Marie is in the white ‘Marie is dressed in white.’

b.

Marie Marie

witte whitewashed

c.

Marie Marie

droeg wore

een a

de the witte white

kamer. room jurk. dress

Along similar lines, an interjective root atom like tsjoek can be ‘assigned’ categorial status on the basis of locally available functional material. For example, the numeral twee and the plural morphology –s in (32a) define the nominal nature (and count interpretation) of the phrasal expression twee tsjoeks. The past tense morphology –te (i.e., T) in (32b) defines the verbal character of tsjoekte.11 Along the same lines, a sound symbolic expression like roemer de boem can get a nominal (33a) or verbal (33b) ‘flavor’.12 11 See also De Belder (2011:10) for the root status of interjections and for the idea that the structural environment determines the categorial status of the root. Her example: (i) Ik hoef al [dat ge-hé] niet. I need all that GE - HÉ not ‘I don’t like that saying hi all the time.’ 12 A reviewer points out that attachment of plural morphology and tense morphology can also be found with elements that seem to have a categorial value already, since they are complete phrases. (i) Twee ik-weet-het-niet-meer-s is er een teveel (CP used as noun) two I-know-it-not-anymore-PL is there one too-many ‘Two I-don’t-knows’ is one too many.’ (ii) Jan ik-weet-het-niet-meer-t de hele dag. (CP used as verb) Jan I-know-it-not-anymore-PRES . 3.SG the entire day On the basis of (i)–(ii), one might conclude that the combinability of an element with plural or tense morphology does not tell us very much about the categorial, or category-less, status of that element. If so, the combinability of an interjection with tense/plural morphology would not necessarily lead to the conclusion that they are roots. Although an in-depth analysis of the patterns in (i)–(ii) falls beyond the scope of this article, it may be useful to point out here that in the literature two approaches towards the definition of roots can be found: (i) an approach that takes roots to be lexically defined (e.g., Borer 2005a,b) and (ii) an approach that takes them to be structurally defined (e.g., Halle and Marantz 1993,

56 (32)

(33)

Norbert Corver

a.

Na [twee tsjoeks] stond de TSJOEK .PL stood the after two ‘After two chuffs the train stopped.’

b.

De trein [tsjoekte] het station the train TSJOEK .PST.3.SG the station ‘The train cuffed out of the railway station’

a.

Na after te to

b.

De the

[twee two

roemer

de

ROEMER

DE

boems] BOEM .PL

trein train

stil. still

uit. out

begon began

de the

drumband marching-band

marcheren. march drummer drummer

[roemer

de

ROEMER

DE

boemt] BOEM .PRES .3.SG

de the

hele entire

dag. day

On the basis of the above-mentioned facts, I conclude that interjective atoms constitute (bare) roots ((). In this respect they differ from other so-called parts of speech, such as nouns, verbs, adjectives and prepositions, which constitute complex syntactic objects of the type f-( (i.e., functional category + root): n-(, v-(, a-(, p-(. This root-status of interjections may also be one of the reasons that it is often quite difficult to paraphrase the exact meaning of interjections (see Kaplan 1999, Kratzer 1999). For example, the meaning of Wow! seems less transparent than the meaning of cow, and also the meaning of the interjection boy in Boy, is he smart! is less transparant than the meaning of boy in The boy is smart. The presence of functional material “narrows down” Harley and Noyer 1999). According to the former approach, a root is defined lexically by the merger of a vocabulary item that does not contain any syntactic features; according to the latter approach, a root is defined structurally if it is a terminal node that does not contain any category-specific features. Under the structural approach, a Root-feature functions as a placeholder that signals to the post-syntactic mechanisms (late insertion) that vocabulary material can be inserted in this position. Since the root position is defined in purely structural terms, it is irrelevant whether the vocabulary material realizing this position carries any syntactic features. As De Belder (2011: 41–45), for example, points out, functional vocabulary items (e.g., pronouns, subordinators) can be used in a root-position, as in Er zijn nog twee waaroms (lit.: there are still two why-s; ‘There are still two why-questions that need to be answered’). Importantly, the syntactic features (e.g., the +wh-feature of waarom) of the late-inserted vocabulary item waarom do not play any role in the syntax of the containing clause. Along the same lines, one might try to analyze the ‘clausal roots’ in (i) and (ii). Essentially, these expressions provide lexical contents to the root-feature. The syntactic features (e.g., categorial features like C) of the “clausal” root do not play any (morpho)syntactic role within the sentence of which it is a part.

Interjections as structured root expressions

57

the meaning of the root; very roughly, you get an object-reading (nouns), eventreading (verbs), property-reading (adjectives) or location-reading (prepositions). Absence of this f-layer in an interjective linguistic expression arguably makes its descriptive meaning less transparant. Another consequence of my interpretation of interjections as being roots is that we get a slightly different perspective on so-called ‘secondary interjections’, i.e., interjections that are said to be derived from lexical categories such as verbs and nouns; see, for example, Ameka (1992), Wilkins (1992), and Poggi (2009) for discussion.13 In Dutch, for example, words such as jonge(n) (lit.: ‘boy’) and kut (lit.: ‘cunt’) can be used as interjections; see (34). The sentences in (35) contain the interjections kijk (lit.: ‘look’) and zeg (lit.: ‘say’). Rather than saying that these are interjections “derived from” (i.e., grammaticalized on the basis of) nouns and verbs, we might simply say that they are simply roots that haven’t turned into nouns and verbs by means of a category-defining functional head (n, v).14 (34)

(35)

a.

Kut, ik ben mijn jas vergeten! cunt (lit.), I have my coat forgotten ‘Damned, I forgot to take my coat with me!’

b.

Jonge(n), wat hebben boy, what have ‘Boy, did we laugh!’

a.

Kijk, dat vond ik nog look (lit.) that found I PRT ‘Well, that was quite nice of her!’

eens PRT

b.

Zeg, nu heb je wel say (lit.) now have you PRT ‘Well, you’ve had enough now!’

genoeg enough

wij we

gelachen! laughed

aardig! kind

gehad! had

So far, I have argued that complex interjective expressions can be structurally decomposed into smaller units (atoms) and that these units are roots. This 13 In the linguistics literature a primary interjection is defined as an interjection that does not have a lexical category as its basis: e.g., Wow!, Oops!, Mmm! 14 Notice that these secondary interjections display the property of iteration, which under my analysis involves an asyndetic coordinated structure (See (30a)): (i) a. [Kut, kut, kut], ik ben mijn jas vergeten!

(cf. (34a))

b. [Kijk, kijk, kijk], dat vond ik nog eens aardig! (cf. (35a))

58

Norbert Corver

state of affairs raises an interesting question: if the computational atoms for building complex interjective expressions are roots, how do we ever get such a complex structure, since roots are generally taken to be invisible to syntactic computational operations (Chomsky 2011a,b)?15 One might hypothesize that roots can combine at the level of words and that complex interjective expressions form compounds rather than syntactic (i.e., phrasal) units. However, I will show in section 4 that complex interjective expressions behave like syntactic units rather than word units (i.e., compounds). In section 5 I will propose that coordination is the formal mechanism that is at the basis of the formation of complex interjections. What makes coordinators optimal functional elements for building complex interjections is the fact that upon merge they leave the root status of elements intact. That is, root (e.g., (boemer) plus root (e.g., (boem) ‘is a‘ (complex) root (i.e., [( (boemer Conj (= de) (boem]; boemer de boem).

4 Complex interjections as syntactic objects A first question that can be raised as regards the internal structure of complex interjections is the following: is this structured expression compound-like or phrase-like (in the sense of having a phrasal syntax). The placement of phonological stress (indicated by capital letters) is helpful here, since the stress pattern in Dutch compounds is different from that of syntactic phrases (Booij 1977:70; De Haas and Trommelen 1993:360). This is exemplified in (36), where the a-examples represent the compounds and the b-examples the syntactic phrases: (36)

a.

een a

LANGoor long-ear, ‘a rabbit’

a.’

een a

lang long

b.

een a

ZWAARgewicht heavyweight, ‘a prizefighter’

b.’

een a

zwaar heavy

OOR ear geWICHT weight

15 One might try to relate the idea that interjections –being roots– lack syntactic combinatorial properties to their distributionally free behavior. If they lack syntactic combinatorial properties, you can more or less stick them in (“interject”) anywhere at the level of clauses and other types of phrasal expressions. In Chomsky (2001), the interjective nature of an expression is associated with the absence of what he calls an ‘edge-feature’. An edge-feature (EF) of a Lexical Item (LI) is a feature that allows the LI to enter into a computation, merging with some syntactic object SO. As Chomsky notes: “If an LI lacks EF, it can only be a full expression in itself; an interjection.” In later work (e.g., Chomsky 2007:8), Chomsky proposes that the possibility of an element Z to enter into further computation depends on the label of Z; see also Chomsky (2011a,b).

Interjections as structured root expressions

59

The compounds in (36a,b) show that stress is on the leftmost element of the composite word. The (nominal) phrases in (36a’,b’) show that phrasal stress falls on the rightmost element, i.e., the nominal head of the complex noun phrase. Interestingly, young children and parents talking to young children, sometimes use sound symbolic patterns like boem boem in order to refer to objects; boemboem, for example, designates a drum. As shown in (37a), if used in this way, the complex expression displays the compound stress pattern, i.e., stress on the first syllable. (37b,c) give some additional illustrations of these reduplicative patterns. (37)

a.

Mijn my

[BOEMboem] bam-bam (= drum)

b.

Waar where

c.

Daar there

is has

je your

rijdt drives

een a

is is

kapot. broken

[WOEFwoef ] bowbow (= dog)

gebleven? stayed

[TSJOEKtsjoek]. choo-choo (= train)

When we compare the stress pattern in (37) with the stress pattern of the complex interjections in (38), we notice a clear difference: in the latter, main stress falls on the final element. (38)

a.

Jan Jan

sloeg hit

[boem

boem

BOEM]

BAM

BAM

BAM

op on

b.

Jan sloeg [boemer de BOEM] op de trommel.

c.

De the

ruit window

brak broke

de the

[rinkel

de

KINKEL]

RINKEL

DE

KINKEL

trommel. drum

in into

stukken. pieces

The phrasal stress pattern also makes it possible to use these complex interjections in (partially) nonsense rhymes. Consider, for example, the following rhymes in which the first line is a complex sound symbolic interjection and the second line a phrasal unit (CP in (39a,b,c), DP in (39d)) that is made up of “normal” lexical items:16

16 (39c) is part of a rhyme that children use at the start of the playground game called tikkertje (‘tag’). (39d) is taken from a so-called children’s Sinterklaas-song.

60 (39)

Norbert Corver

a.

rommer de BOM Dit is mijn TROM. this is my drum

b.

holder de BOLDER. Jan is op ZOLDER. Jan is in attic (‘Jan is in the attic.’)

c.

pief paf POEF. [Sound of shooting] Jij bent de BOEF. you are the villain

d.

rinkel de KINKEL. wat een geRINKEL! what a chinking (‘what chinking!’)

In short, interjective expressions display the same nuclear stress pattern as other phrasal expressions in Dutch. That is, nuclear stress falls on the most deeply embedded element on the recursive side (Cinque 1993). In the coordinate structures in (30), this is the interjective atom that is the most deeply embedded conjunct within the hierarchically organized coordinate structure; see also sections 5 and 6.

5 Complex sound interjective expressions as coordinate structures Having argued that complex interjective expressions have a phrasal syntax, we can now investigate what their syntax looks like. In this section I will develop an analysis, already hinted at at the end of section 2, according to which complex interjective expressions are coordinations of roots. Following Thiersch (1993, 1994), Kayne (1994) and Johannessen (1998), among others, I will assume that a coordinate conjunction (Conj) projects a constituent structure containing a complement position and a specifier position. For a coordination like Pieter en Koen, in which two proper names (DPs) are coordinated, this yields the syntactic configuration in (40a). When we apply this structural analysis to an interjective expression like boemer de boem, we get the configuration in (40b): (40)

a.

[ConjP [DP Pieter] [Conj’ [Conj en] [DP Koen]]]

b.

[ConjP [( boemer] [Conj’ [Conj de] [( boem]]]

Interjections as structured root expressions

61

Recall from section 2 that I analyzed interjective atoms as roots. Now why would a coordinating conjunction (Conj) be an adequate functional element for building complex structures on the basis of roots? An important characteristic of coordinators is that they connect constituents of the same kind to form a larger constituent of that same kind.17 For example, in (41a), two DPs –de hond and de kater– are connected to each other and form a larger constituent that has DP-like properties. That is, the coordinate structure as a whole displays the grammatical behavior that is characteristic of its coordinates. Thus, de hond en de kater displays the same grammatical behavior as its conjuncts de hond and de kater.18 For example, if two coordinated DPs bear accusative case, as exemplified by the German example in (41b), then this case property is also associated with the entire coordinate structure: (41) a. Jan heeft [ConjP [DP [D de] (hond]] [Conj’ en [DP [D de] (kater]] geslagen. Jan has the dog and the tom-cat beaten b. Jan hat [ConjP [DP [D denacc] (Hund]] [Conj’ und [DP [D denacc] (Kater]] geschlagen. This inheritance behavior of the coordinate structure can be acounted for by taking coordinate conjunctions to be underspecified feature bundles that act as identity operators with respect to certain grammatical features, such as categorial features, case features and the bar-level property (cf. Thiersch 1993, 1994; see also Johannessen 1998: Chapter 5). Thus, [X1 and X2] “is-an” X. This means that [DP1 and DP2] “is-a” DP, and that [( and ( ] “is-a” (. In other words, the conjunction does not project its lexical nature onto the projected structure. Or to put it differently, coordinators do not have a labeling function (Chomsky 2011a,b). In this respect, the functional category Conjunction differs from other functional heads such as v, C(omplementizer), n and D, which do have a labeling function. This inability to label structures possibly makes conjunctions excellent candidates for connecting roots. The complex structure that results from coordination of roots “is-a” root. This possibly also explains why it 17 It should be noted here that categorial identity is not always necessary, as in Bill is [[a democrat] and [proud of it]]. Importantly, the coordinated phrases must be of the same semantic type; in casu, the noun phrase and the adjective phrase both function as predicates. See Sturm (1986) for a discussion of coordination of categorially distinct phrases. 18 I abstract away here from the fact that coordination of two singular (sg) DPs yields a coordinate structure that has plural (PL) interpretation; see (i). What is important for us is that the coordinate structure has number-properties, just like its DP-conjuncts. met elkaar. (i) [ConjP/pl [DP/sg De hond] en [DP/sg de kater]] vochten the dog and the tomcat fought-PL with each-other

62

Norbert Corver

is possible to have complex roots such as roemer de boem as input to attachment of plural morphology (33a) or present tense morphology (33b). The structural configuration that corresponds to these forms is given in (42).19 (42)

[[ConjP [( roemer] [Conj’ [Conj de] [( boem]]] -s/-t]20

19 One might object to the idea of root-level coordination on the basis of examples like (i) which show that inflectional morphology –e.g., tense morphology (ia), and adjectival inflection (ib)– cannot be attached to a coordination of roots: (i) a. *De verpleegster [[was en weeg]t] de baby. (OK: wast en weegt) the nurse wash and weigh -s the baby ‘The nurse washes and weighs the baby.’ b. *een [[mooi en snel]e] auto a beautiful and fast -e car ‘a beautiful and fast car’

(OK: mooie en snelle)

It should be noted, though, that patterns of coordination exist in which right peripheral attachment of inflectional morphology to a root-level coordination is more, or even fully, acceptable (see (iia,b); (iia) is drawn from the internet). An important characteristic of these coordinations is that the two members form a fixed combination. In other words, they have an idiomatic flavor. (ii) a. Ex-bondscoach José de Cauwer wik en weegt de kansen van de former-national-coach José de Cauwer think and weigh-s the chances of the Belgische ploeg. Belgian team ‘Former national coach José De Cauwer weighs up the chances of the Belgian team.’ b. Geef een kort en bondige beschrijving van de voortgang. give a short and brief-e description of the progress ‘Give a very brief description of the progress that has been made.’ Another type of expression that hints at the possibility of root-level coordination are expressions such as vader-en-moeder-tje (father-and-mother-DIM) and cowboy-en-indiaan-tje (cowboy-andindian-DIM) in (iii), which refer to children’s games. In these examples, diminutive morphology gets attached to the coordinations vader-en-moeder and cowboy-en-indiaan, respectively. (iii) Piet en Ellen spelen vader-en-moedertje / cowboy-en-indiaantje Piet and Ellen play father-and-mother-DIM / cowboy-and-indian-DIM For the purposes of my article it is sufficient to show that root-level coordination with addition of inflection outside the coordination is possible to a certain extent, namely in the case of idiomatic coordinations and other fixed combinations of coordinated elements. The question, of course, remains as to why the patterns in (i) are il-formed. The fact that the two coordinated elements do not correspond to a single “meaning unit” seems to be relevant. I will leave an indepth analysis of this issue for future research 20 For the sake of simplicity, I have labeled the projections of the coordinator with the X-bar theoretic labels Conj’ and ConjP. Crucially, the labeling information is determined by the conjuncts. If these are roots, then the entire configuration (ConjP) is a root syntactic object. Consequently, tense morphology can attach to this complex root.

Interjections as structured root expressions

63

Now that I have argued that conjunctions are appropriate functional categories for building complex structures on the basis of roots, let us see whether there is any further support for this coordination analysis of complex interjective expressions. In what follows I will discuss a number of phenomena that hint at the presence of an underlying coordinate structure in the case of complex interjective expressions. First of all, the Dutch coordinator en ‘and’ is a linking element that combines two elements that intuitively have an “equal” status semantically; that is, one element is not semantically more prominent than the other, as for example in predicate-argument relationships, where the predicate is semantically more prominent in the sense that it selects its arguments, or modifier-modifiee relationships, where the modifier is less prominent than the modifiee in the sense of being optional. Thus, the nominal elements Pieter and Koen in the coordinate structure Pieter en Koen (as in Pieter en Koen slapen ‘Pieter and Koen sleep’) are “semantically symmetric”, i.e., one of the elements does not have an asymmetric role (argument, modifier) with respect to the other element. Similarly, holder and bolder in an expression like holder de bolder arguably have a semantically equal status; see section 6, though, for some formal phenomena that hint at an asymmetric relationship between the two interjective atoms.21 A second parallel between coordination and the (d)e-pattern concerns the possibility of extending the pattern in an unbounded fashion. In other words, the property of discrete infinity applies to both patterns. Coordination displays the property of recursion in the sense that a well-formed coordinate pattern can

21 It should be noted that holder de bolder, as opposed to the coordinate structure Pieter en Wouter has a fixed order. Thus, Wouter en Pieter is a possible pattern in present-day Dutch, but bolder de holder is not. This more restrictive order does not mean that a coordinate-like analysis of holder de bolder should be excluded. Also in coordinations of names we sometimes find order restrictions. In English, for example, Laurel and Hardy (comic duo) and Simon and Garfunkel (singer-song writers) have a fixed order. In Dutch, we find similar fixed patterns of coordination: Peppi en Kokki (comic duo) and Suske en Wiske (duo in comic magazines). Language comparison also makes clear that the order seems quite arbitrary: Dutch pijl en boog (arrow and bow) vs. ??boog en pijl, but English bow and arrow; also, Dutch peper en zout (pepper and salt) vs. ??zout en peper, but English salt and pepper.Also coordinations in idiomatic expressions display a fixed order. The Dutch idiomatic expression in zak en as zitten (in bag and ash sit, ‘to be very sad and disappointed’), for example, does not permit the reverse order of the two conjuncts: *in as en zak zitten. What is important is that these “fixed” coordinations display properties of “normal” coordinations. For example, the requirement that in a monosyndetic coordination the conjunct appear in between the last two coordinated elements, holds both in “normal” coordinations (e.g. John, Peter, Bill and Harry versus *John, Peter and Bill, Harry) and “fixed” coordinations (e.g. Crosby, Stills, Nash and Young versus *Crosby, Stills and Nash, Young).

64

Norbert Corver

be used as a building block in an even bigger coordinate structure. In other words, language allows for recursively constructing more and more nested coordinate structures. Along the same lines, a sound-symbolic interjective expression can be used as a building block in an even bigger interjective expression. (43)

(44)

a.

[Roemer en Koen]

b.

[Wouter en [Roemer en Koen]]

c.

[Pieter en [Wouter en [Roemer en Koen]]]

a.

[boemer [de boem]]

b.

[boemer de [boemer de boem]]

c.

[boemer de [boemer de [boemer de boem]]]

A third similarity concerns the possibility of creating iterative patterns, i.e., patterns in which one atom is used repeatedly. In (45), this iterative use is illustrated for coordinate patterns. Iteration here encodes the aspectual property of ‘delimitedness’ (unboundedness); see Verkuyl (1993). In (45a,b), there was no end to the event of crying and raining, respectively. It went on for an indefinite period of time. (45)

a.

Jan huilde en huilde en Jan cried, and cried and ‘Jan was crying all the time.’

b.

Het regende en it rained and ‘It kept on raining.’

regende rained

huilde. cried.

en and

regende. rained

In (46), the phenomenon of iteration is exemplified for interjective expressions: (46)

a.

boemer de boemer de boemer de boem

b.

rinkel de kinkel de kinkel de kinkel

This iterative use of a sound symbolic interjective atom has the effect that the length (and possibly intensity) of the sound event (i.e., the roll of drums, the sound of breaking glass) is represented. A fourth property that the complex interjective patterns featuring de and the coordinate structures featuring en share is the possibility of leaving the linking

Interjections as structured root expressions

65

element –de in interjective expressions and en in coordinate expressions – phonetically unrealized between some of the conjuncts. In (47), for example, there is no overt coordinator present in between the first three conjuncts. The coordinator en only appears in between the last two conjuncts. Given the single occurence of en in this complex coordinate structure in (47), this coordination pattern can be characterized as monosyndetic coordination. Along the same lines, we may call complex interjections like (48a,b) monosyndetic interjective expressions. With the exception of the last two interjective atoms, the atoms are not separated from each other by means of the connecting element de. (47)

(48)

a.

Jan huilde, huilde, huilde en huilde.

b.

Het regende, regende, regende en regende.

a.

boemer boemer boemer de boem

b.

rinkel rinkel rinkel de kinkel

As noted in Munn (1993) and Kayne (1994), there is a contrast in wellformedness between the partially asyndetic coordination in (49a) and the one in (49b). (49)

a.

I saw [John, Bill and Sam].

b. *I saw [John and Bill, Sam]. As shown in (50), the same contrast is found in Dutch: (50)

a.

Ik I

zag saw

[Roemer, Roemer,

Pieter Pieter

en and

Koen]. Koen

b. *Ik zag [Roemer en Pieter, Koen]. If complex sound symbolic expressions have a coordinate-like structure, we may expect to find a similar contrast. Interestingly, for all Dutch speakers whom I have asked for their judgments, there is a subtle but clear contrast in acceptability between the a-examples and the b-examples in (51)–(52). For some speakers, this contrast seems to be slightly less strong, though, than the contrast between the a-example and b-example in (50).22 I will interpret the contrasts 22 As a reviewer points out, this weakening of the contrast may be due to the fact that (51a) and (51a) are rather marginal to begin with compared to the examples of partially asyndetic coordinations as in (50a). The reviewer agrees that (51a) and (52a) are better than the examples (51b) and (52b).

66

Norbert Corver

in (51) and (52) as corroborating evidence for the coordinate-like structure of complex interjective expressions. (51) a. Ze reden ?hobbel, bobbel de bobbel over het keienpad. they drove HOBBEL , BOBBEL DE BOBBEL across the stone-road b. Ze reden *hobbel de bobbel, bobbel over het keienpad (52) a. Het glazen tafelblad viel ?rinkel, kinkel de kinkel kapot the glass table-top fell RINKEL KINKEL DE KINKEL broken op de grond. onto the floor b. Het glazen tafelblad viel *rinkel de kinkel, kinkel kapot op de grond. A fifth similarity between iterative coordinate patterns featuring en and sound symbolic expressions featuring de concerns the prosodic phrasing of those structures. More specifically, they are prosodically “flat” in the sense that the atoms connected by the linkers en and de are separated by prosodic boundaries of equal strength, which, following Wagner (2005, 2010), I will indicate with the pipe symbol ‘|’. This “balanced” prosodic pattern causes a rhythmic effect. Furthermore, the last atom of the complex iterative structure is phonologically the most prominent one; that is, it carries the nuclear stress of the complex expression. Starting with this last prosodic property, consider the following examples:23 (53)

(54)

a.

Pieter en | KOEN | ‘Pieter and Koen’

b.

Roemer en | Pieter en | KOEN | ‘Roemer and Pieter and Koen’

c.

Roemer en | Pieter en | Wouter en | KOEN | ‘Roemer and Pieter and Wouter and Koen’

a.

roemer de | BOEM |

b.

roemer de | boemer de | BOEM |

c.

roemer de | boemer de | boemer de | BOEM |

23 See also Wagner (2010: section 5.4) for the statement that prosodically flat structures have a nuclear accent on the last element in the sequence.

Interjections as structured root expressions

67

The examples in (53) show that the last conjunct (KOEN) is always accented (indicated here by capitals), and the b-examples show that the last interjective atom (BOEM) is phonologically the most prominent. Given the parallelism between coordinate structures like Roemer en Koen and complex interjections like roemer de boem, I conclude that the latter have an internal syntax that is similar to that of the former construction type. Thus, complex sound symbolic interjections are coordinate structures. The interjective atoms constitute the conjuncts of the complex structure. Following, among others, Thiersch (1993, 1994), Kayne (1994) and Johannessen (1998), I will assume that a coordinate conjunction (Conj) like en projects a phrasal structure containing a complement position and a specifier position. Thus, (53a) has the structure in (55a), (53b) the structure in (55b), and (53c) the one in (55c). (55)

a.

[ConjP Pieter [Conj’ [Conj en] KOEN]]

b.

[ConjP Roemer [Conj’ [Conj en] [ConjP Pieter [Conj’ [Conj en] [ConjP KOEN]]]]]

c.

[ConjP Roemer [Conj’ [Conj en] [ConjP Pieter [Conj’ [Conj en] [ConjP Wouter [Conj’ [Conj en] [ConjP KOEN]]]]]]]

Patterns (55b,c) are recursive in the sense that the Conj-head en takes a phrasal ConjP as its complement. In other words, we have a self-embedded structure: a ConjP is embedded within a larger ConjP. Under the assumption that asyndetic coordinations have the same structure but are headed by a silent Conj-head, here represented as Ø, we have the following hierarchical structure for an asyndetic coordination like Roemer, Pieter, Koen: (56)

[ConjP Roemer [Conj’ [Conj Ø] [ConjP Pieter [Conj’ [Conj Ø] [ConjP KOEN]]]]]

The conjunct that carries nuclear accent is the most deeply embedded conjunct on the recursive side. The idea that nuclear stress placement is directly related to structural embeddedness was proposed in Cinque (1993). Given the parallelism in placement of nuclear stress in complex sound symbolic interjections (cf. (54)) and coordinate structures (cf. (53)), we may assign the following structures to the examples (54a,b,c): (57)

a.

[ConjP roemer [Conj’ [Conj de] BOEM]]

b.

[ConjP roemer [Conj’ [Conj de] [ConjP boemer [Conj’ [Conj de] [ConjP BOEM]]]]]

c.

[ConjP roemer [Conj’ [Conj de] [ConjP boemer [Conj’ [Conj de] [ConjP boemer [Conj’ [Conj de] BOEM]]]]]]

68

Norbert Corver

The asyndetic complex expressions in (3c) and (24b) arguably have the same structural representation, with the only difference that the Conj-head is silent; i.e., Conj does not externalize as de.24 (58) a. [ConjP pats [Conj’ [Conj Ø] [ConjP boem [Conj’ [Conj Ø] [ConjP KNAL]]]]]

(= 3c)

b. [ConjP boem [Conj’ [Conj Ø] [ConjP boem [Conj’ [Conj Ø] [ConjP BOEM]]]]] (= 24b)

6 Against a flat structural analysis In section 5 I have tried to give evidence for the structured nature of complex interjective expressions by drawing a parallel with syntactic coordinations. I crucially built on the hypothesis that coordinations involve a hierarchically organized phrase structure (ConjP) in which the Conj is the head and the conjuncts are in hierarchically different positions; one is the specifier and one is the complement of ConjP (cf. (57a)). The question arises as to whether there is any independent support for the hierarchically asymmetric placement of the interjective atoms that make up the complex interjections. That is, is there any evidence that in an expression like roemer de boem the first interjective atom is in a hierarhically higher position (viz. Spec,ConjP) than the second interjective atom, which is in the complement position of Conj. Of course, the same question can be raised for the interjective atoms in (57b,c), each of which occupies a different hierarchical position in a structure involving multiple coordination. In view of the iterative nature of certain interjective expressions (e.g., the repetitive pattern boem boem boem), one might, for example, propose that the interjective atoms are not in hierarchically asymmetric relation with respect to each other but simply constitute a linear arrangement of elements at the same level of depth (cf. Uriagereka 2008, Karlsson 2010). Under such a syntax-less analysis, the complex interjective expressions in (57) and (58) would simply have a flat structure, as in (53) and (54): (59)

a.

[roemer de boem]

b.

[roemer de boemer de boem]

c.

[roemer de boemer de boemer de boem]

24 See also Van den Toorn (1960) for the idea that complex interjections such as o jee (‘Oh gush!’) and ach ach (expression of pity) involve coordination of interjections.

Interjections as structured root expressions

(60)

a.

[pats boem knal]

b.

[boem boem boem]

69

So, what argues in favor of the hierarchical ConjP-structures in (57)–(58) and against the flat structures in (59)–(60)? One type of phenomenon that Johannessen (1998) takes to support a hierarchical structure in which the two conjuncts are in hierarchically different syntactic positions (i.e., specifier and complement) is what she calls ‘unbalanced coordination’. Unbalancedness holds when one of the conjuncts displays expected, normal grammatical behavior with respect to some grammatical property (e.g., case, agreement), while the other displays deviant behavior. For example, in Stavanger Norwegian, conjoined subjectpronouns display asymmetric behavior as regards case: the first conjoined pronoun bears the expected nominative case (han), while the second pronoun is marked by accusative case (meg); cf. Johannessen (1998: 18). In other words, only the specifier conjunct receives the grammatical feature (in casu nominative case assigned by T) associated with the whole ConjP (cf. Johannessen 1998: 8). (61)

[Han og meg] var he.NOM and me.ACC were ‘He and I were in it together.’

sammen together

om about

det it

Besides unbalancedness of the receiving type, Johannessen distinguishes unbalancedness of the assigning type (see Johannessen 1989: 8): one conjunct assigns a grammatical property to some ConjP-external element, in spite of it possibly having conflicting features with other conjuncts. This is exemplified for Czech in (62), where the first conjunct determines (i.e., “assigns”) the agreement properties of the finite verb (Johannessen 1998: 9): (62)

Pujdu tam [já a will.go.1.Sg. there I and ‘You and I will go there.’

ty] you

As Johannessen (1998: 143–154) shows, unbalancedness is also attested in coordinated constructions with more than two conjunctions. Under a ConjP-analysis, multiple coordination may look as follows: [ConjP1 Spec [Conj’ Conj ConjP2]]. Johannessen (p. 144) notes that there is usually only one conjunct with normal features (e.g., case, agreement) in each multiple ConjP. One of the examples she gives in support of this comes from English presentation constructions. As shown in (63), the verb agrees only with the first conjunct – i.e., the specifier (a man) of the (highest) ConjP. (63)

There is/*are [a man, a woman, and a cat] waiting outside.

70

Norbert Corver

The question now arises as to whether signs of unbalancedness (of the receiving type or assigning type) can be identified within complex interjections. That is, are there patterns in which the interjective atom in the specifier position displays grammatical behavior that is different from that of the interjective atom in the complement position. Importantly, Johannessen’s unbalanced coordination typically involves a relationship between one ConjP-internal conjunct and a ConjP-external element (e.g., the finite verb). Since interjective expressions typically behave like isolates as far as their external syntax is concerned – i.e., they do not seem to enter into any specific morphosyntactic dependency with some clause-internal element – it is quite difficult to test the phenomenon of unbalancedness on the basis of their external syntax. One might explore, though, the possibility of unbalancedness at the syntax-phonology interface level. As Johannessen (1998: 24) points out, also phonological features may be distributed unevenly in a ConjP. The example she gives comes from the phenomenon of soft mutation in Welsh: some words get a different consonant when preceded by certain categories. As shown in (64), only the first conjunct is subject to soft mutation: fara is a mutated form of bara, while menym has the citation form, and thus is not mutated into fenyn. (64)

Bwytais i [ fara, menym a ate I bread butter and ‘I ate bread, butter and cheese.’

chaws] cheese

Interestingly, certain signs of unbalancedness can be found at the phonological level in Dutch complex interjective expressions. Consider, first, the following simplex interjections: (65)

a.

jonge!

Wat ben jij what are you ‘Boy, you are so stupid!’ JONGE

b.

dom! stupid

(amazement)

ja! Wat moet ik hiervan denken? yes! what must I here-of think ‘Well, what should I think of this?!’

(doubt)

The interjections jonge and ja can be phonologically augmented by adding an alveolar stop /t/ at the beginning of the sound sequence: tjonge! and tja! The stop that is added at the beginning of the sound sequence contributes to the expressive/affective flavor of the interjective expression. When we now turn to complex interjective expressions, we see something interesting happen: besides

Interjections as structured root expressions

71

the ‘bare‘ reduplicative patterns (66a,b), we can have the (partially) augmented reduplicative in (67a,b), but not those in (68) and (69). (66)

(67)

(68)

(69)

a.

jonge, jonge, jonge! Wat ben jij dom!

b.

ja, ja, ja! Wat moet ik hiervan denken?

a.

tjonge, jonge, jonge! Wat ben jij dom!

b.

tja, ja, ja! Wat moet ik hiervan zeggen?!

a.

*jonge, tjonge, jonge! . . . .

b.

*ja, tja, ja! . . . .

a.

*jonge, jonge, tjonge! . . . .

b.

*ja, ja, tja! . . . .

What these examples show is that if one of the interjective atoms gets augmented phonologically by /t/, this sound augmentation can only be applied to the first interjective atom, as in (67). Adopting a ConjP-analysis of complex interjections, I can interpret this phenomenon of phonological augmentation as another illustration of Johannessen’s unbalanced coordination. In the mapping from syntax onto phonology, only the first conjunct (i.e., the highest specifierconjunct) gets augmented phonologically. Under a (non-binary branching) flat structure analysis, one would have to postulate an additional linearity principle to account for this phonological augmentation rule. This special status of the first interjective atom in a complex interjective expression is also suggested by certain other phenomena. Consider, for example, the following minimal pairs: (70)

(71)

a.

rinkel de kinkel

b.

rinkel de kinkel de kinkel

c.

??rinkel

a.

holder de bolder

b.

holder de bolder de bolder

c.

??holder

(sound of breaking glass)

de rinkel de kinkel

de holder de bolder

(sound of something that falls down, e.g. from the stairs)

72 (72)

Norbert Corver

a.

hobbel de bobbel

b.

hobbel de bobbel de bobbel

c.

??hobbel

(sound of something that moves on a bumpy road)

de hobbel de bobbel

The a-examples are complex interjections with two interjective atoms. The bexamples show that reduplication can apply to the second interjective atom, resulting in a pattern in which the first interjective element has a unique form (e.g., rinkel, holder, hobbel). The c-examples show that patterns in which the first interjective atom is reduplicated and the last atom has a unique form are considered less acceptable. This uniqueness of the first interjective atom is again compatible with a phrase structural analysis in which the first interjective atom occupies the specifier position of the highest ConjP. In other words, the phenomenon illustrated in (70)–(72) fits in the patterns of unbalanced coordination that are observed in Johannessen (1998). Another phenomenon that possibly hints at the special status of the first interjective element may be called “expressive concord”. To explain what I mean by this consider first the examples in (73) and (74): (73)

a.

roemer de boem

(sound of a drum)

b. *roem de boemer (74)

a.

hieper de piep (hoera) ‘hip hip hurray!’

(expression of joy)

b. *hiep de pieper (hoera) These examples show that in a bi-atomic interjective expression (roemer de boem, hieper de piep), the sequence er (pronounced as /ər/) typically occurs on the first atom.25 Given this observation, one might propose that in more complex expressions like those in (75a,b), the er that appears on the intermediate interjective atoms is somehow dependent on the er of the first interjective atom (roemer/hieper). One might interpret this sharing of the er-property as a kind of concord phenomenon. From the perspective of the ConjP-hypothesis, one could interpret this concord in terms of the structural relation of c-command: the interjective in the highest specifier position enters into a concord relationship with the interjective atoms in its c-command domain (except for the last interjective atom: boem/piep). This is represented in (76a,b), where the boldface er indicates the “base form” and the underscored er’s the concording elements. 25 In certain interjective expressions, the coordination is balanced in the sense that both interjective atoms carry -er/-el; e.g., holder de bolder and rinkel de kinkel.

Interjections as structured root expressions

(75)

(76)

73

a.

roemer de boemer de boemer de boem

b.

hieper de pieper de pieper de piep (hoera)

a.

[ConjP roemer [Conj’ [Conj de] [ConjP boemer [Conj’ [Conj de] [ConjP boemer [Conj’ [Conj de] boem]]]]]]

b.

[ConjP hieper [Conj’ [Conj de] [ConjP pieper [Conj’ [Conj de] [ConjP pieper [Conj’ [Conj de] piep]]]]]]

So far, I have discussed some features of unbalancedness in the domain of complex interjections. Following Johannessen’s interpretation of unbalanced coordination as an indication of a hierachically organized coordinate structure, I will take the above-mentioned phenomena also to be suggestive of a hierarchical organization of complex interjective expressions (see (57)–(58)). In what follows I will discuss two more arguments that seem to support a hierarchical organization of complex interjective expressions; one argument is syntactic, the other phonological. A syntactic argument that seems to favor a hierarchical analysis of complex interjections over a flat-structural analysis comes from patterns involving echoquestion formation by means of the wh-word WAT ‘what’. As shown in (77), the echo-wh-word used by speaker B “refers” to a string of elements in speaker A’s utterance that B has not heard or understood properly. This string of elements typically corresponds to a syntactic unit (i.e., a constituent). (77)

Speaker A:

[twee two

[ foto’s pictures

[van of

[koffers suitcases

[op on

[wieltjes]]]]]] wheels

Speaker B: a.

twee two

foto’s pictures

van of

koffers suitcases

b.

twee two

foto’s pictures

van of

WAT? WHAT

c.

twee two

WAT? WHAT

op on

WAT? WHAT

(WAT = wieltjes)

(WAT = koffers op wieltjes)

(WAT = foto’s van koffers op wieltjes)

If the echo-wh-word WAT typically replaces a string of words that correspond to a syntactic constituent, then the following interjective expressions of which a part has been questioned by the echo-wh-word WAT hint at a hierarchical organization of the interjective atoms that make up the complex interjective expression.

74

Norbert Corver

(78)

Speaker A: roemer de boemer de boem

a.

Speaker B: roemer de boemer de WAT?

(WAT = boem)

b.

roemer de WAT?

(WAT = boemer de boem)

As a final argument in support of a hierarchical organization of complex interjective expressions I mention (again) the assignment of phonological stress. Under a flat-structure analysis of complex interjective expressions stress asignment must be regulated by a separate phonological rule, namely one that places phonological prominence on the last interjective atom in a concatenation (i.e., a linearly ordered sequence) of interjective atoms (e.g., boem boem BOEM; boemer de boemer de BOEM; pats boem KNAL). Besides the question as to why stringbased stress assignment would have this “(linearly) rightmost” effect, it is clear that this linear rule of stress placement would be redundant with respect to the phrase structure senstitive rule of nuclear stress placement. An approach in which we can do without this “extra” string-based phonological rule seems preferable. Summarizing, in sections 5 and 6 I have argued that complex sound interjections display similarities to coordinate patterns. In a way, complex interjections are lists of interjective atoms. It was proposed that the syntax of complex interjections is the same as that of patterns of coordination: a conjunction (connector) heads a phrasal projection containing a complement and a specifier. In coordinations, we have a lexical conjunction en, whereas in complex interjections, we may find the conjoining element de. When the conjunction remains phonetically empty, we have an asyndetic pattern. According to this analysis, the atoms of complex interjective expressions should not be treated like beads on a string. Complex interjections are structured expressions that have a nested structure; that is, they display recursive embedding. I further argued that interjective atoms are roots and that a coordinator is an optimal functional element for building a complex syntactic object on the basis of roots. The outcome of conjoining two roots is a complex root expression.

7 Augmentative sounds In the previous sections I have argued that complex interjective expressions have a coordinate structure and that the coordinated interjective elements are roots. In certain interjective expressions, we have an asyndetic coordination in

Interjections as structured root expressions

75

the sense that there is no overt conjoining element in between the two interjective atoms (e.g., [ConjP boem [Conj’ [Conj Ø] boem]]). In other expressions, we find an overt linking element between the two interjective atoms, as in boemer de boem. So far I have interpreted this element as a realization of the coordinator: [ConjP boemer [Conj’ [Conj de] boem]]. In this section I will try to give a more precise characterization of this connecting element. What I will propose is that the sounds d and e (i.e., /ə/; schwa) are augmentative sounds that contribute an emphatic/expressive flavor to the externalized interjective expression (cf. Overdiep 1937: 113–114, 157–161, Corver 2004, 2006). The alveolar obstruent /d/ will be analyzed as a paragogic sound that gets attached at the end of certain interjective atoms. The schwa-sound will be analyzed as an externalization of the functional head Conj. I start my discussion with the sound schwa (i.e., /ə/). As shown in (79)–(80), schwa occurs “on its own” as a linking sound in certain sound symbolic expressions involving reduplication of lexical material. Consider, for example, the following patterns: (79) a. De trein reed [tsjoek tsjoek tsjoek] het station uit. the train drove TSJOEK TSJOEK TSJOEK the railway-station out b. De trein reed [tsjoek e tsjoek] het station uit. c. De trein reed [tsjoek e tsjoek e tsjoek e tsjoek] het station uit. (80)

a.

b.

c.

De the

telefoon phone

rinkelde rang

De the

telefoon phone

rinkelde rang

De the

telefoon phone

rinkelde rang

[tring

tring

tring].

TRING

TRING

TRING

[tring

e

ling].

TRING

E

LING

[tring

e

ling

e

ling

e

ling].

TRING

E

LING

E

LING

E

LING

Adopting the coordination analysis discussed in section 5, a sequence like tring e ling and tring e ling e ling can be represented as in (81a) and (81b), respectively: (81)

a.

[ConjP tring [Conj’ [Conj -e] ling]]

b.

[ConjP tring [Conj’ [Conj -e] [ConjP ling [Conj’ [Conj -e] [ling]]]]]

Now what about the alveolar obstruent /d/, as in rinkel de kinkel and roemer de boem? I propose that /d/ is a paragogic alveolar obstruent that can be

76

Norbert Corver

inserted at the end of certain interjective atoms, viz., those ending in -el and -er. The insertion of a paragogic alveolar obstruent at the end of a word results from a general tendency in Dutch dialects to end in a segment that is as consonantal as possible (see Corver and Van Oostendorp 2005). These paragogic alveolar sounds are typically attached to words ending with n, l, and r. Van Haeringen (1938) gives the following examples of Dutch words containing a paragogic dental: iemand ‘someone‘ (iemand < ieman), dubbeld ‘double‘ (dubbeld < dubbel), sedert ‘since‘ (sedert < seder).26 27 In the context of our discussion of interjective expressions, especially the last two examples are relevant since interjective atoms like rinkel (see (27d)) and boemer (see (26)), respectively, also end with the sound sequence el (i.e., /əl/) and er (i.e., /ər/). Also in dialectal varieties of Dutch, the insertion of a paragogic alveolar obstruent at the end of words ending with /əl/ or /ər/ is quite common. Van Oostendorp (2000), for example, notes that in Utrecht Dutch we find words like brommert and gozert, which result from attachment of the voiceless alveolar stop /t/ to the words brommer ‘moped‘ and gozer ‘bloke’, respectively. Ter Laan (1953: 138–143) gives the following examples from Groningen Dutch: (82)

a.

Ik kom der moar enkelt. I come there just only-t ‘I just come there now and then.’

b.

Wiegert (paragogic t after (e)r) Wieger-t ‘Wieger’ (= proper name)

(paragogic t after (e)l)

26 The words iemand and sedert are part of the Standard Dutch lexicon, the word dubbeld is archaic but is still found in various Dutch dialects (see e.g., Ter Laan (1953: 142) for Groningen Dutch). 27 As shown by the examples, the alveolar obstruent is sometimes orthographically represented as d or t. They are all pronounced as /t/, i.e., as a voiceless alveolar stop, in those examples. In the examples I give, I use the orthography that is used in the written sources that I consulted. It should be noted here that (variants of) Dutch has a phonological devoicing rule that states that voiced obstruents become voiceless at the end of a word. This final obstruent devoicing rule is also known under the German name of Auslautverhärtung. For example, it is generally assumed that iemand ‘someone’ has an underlying phonological representation with a /d/ sound at the end, which actually is also used in the ortographic representation. This phoneme /d/ surfaces, for example, when iemand gets the plural morpheme -en attched to it, resulting in a structure in which the dental obstruent is no longer at the word end (see (i)). Thus the /t/ pronounced at the end of the singular form iemand results from final obstruent devoicing. (i) Met twee iemanden mag je nooit spreken, namelijk Jan en Bob. with two someones may you never talk, namely Jan and Bob ‘You are not allowed to talk to two specific persons, namely Jan and Bob.’

Interjections as structured root expressions

77

As noted by Van Haeringen (1938), a paragogic alveolar obstruent can sometimes get an expressive value (with intensifying meaning). An example of this is the insertion of -d, pronounced as a voiceless alveolar stop /t/, after the agentive nominalizing suffix -er with words like knoeier ‘bungler’, vreter ‘greedy-guts’, suffer ‘dullard’. Thus, we get forms such as: een knoeierd (‘a bungler’), een vreterd (‘a greedy-guts’) and een sufferd (‘a dullard’). It does not seem implausible that the paragogic d found in complex interjective expressions also contributes to the expressive and intensifying meaning of the interjective expressions. The insertion of a paragogic alveolar obstruent is also attested with substantively used (possessive, demonstrative and interrogative) pronominals in various dialects of Dutch (see Corver and Van Oostendorp 2005; Ter Laan 1953: 138–139; Overdiep 1937: 280–284). Consider, for example, the following equivalents of English mine in a number of Dutch variants: (83) Mijn broertje vindt zijn fiets de mooiste, maar ik . . . I ... my brother-DIM finds his bike the beautiful-SUPERL , but ‘My brother finds his bike the most beautiful one, but I. . . a. de mijne (Standard Dutch) the my-e ‘mine’ b. mient my-t

(Onstwedde Dutch)

c. miende my-d-e

(Giethoorn Dutch)

d. de miende (Hooghalen Dutch) the my-d-e The Standard Dutch pattern in (83a) features a schwa right after the possessive pronoun mijn, which is the form of the attributively used possessive pronoun (as in mijn fiets ‘my bike’). The Onstwedde Dutch form mient has a paragogic t (i.e., a voiceless alveolar stop /t/) right after the possessive pronoun mien. In the Giethoorn Dutch and Hooghalen Dutch examples, we find a pattern that displays both a paragogic voiced alveolar stop (viz., /d/) and a schwa (/ə/). (83c) and (83d) differ from each other in the absence versus presence of the definite article before the possessive pronominal form. Superficially, the sound sequence de (i.e. /də/) in (83c) and (83d) is similar to the sequence de (i.e., /də/) in boemer de boem and holder de bolder. The question arises as to whether the grammatical make-up of forms such as (de) miende

78

Norbert Corver

may give us insight into the internal make-up of the sequences roemer de and rinkel de. Although a full-fledged analysis of the internal syntax of substantively used possessive pronominal forms is beyond the scope of the present article, I would like to propose that the schwa in (de) miende is the same schwa as the one in Standard Dutch de mijne. For the latter form, Corver and Van Koppen (2011) argue that e is a weak pro-form that spells out the functional head n (i.e., little n). More specifically, they assign the representation in (84a) to the pattern de mijne. I propose that the patten (de) miende, featuring the paragogic dental obstruent right after the possessive pronoun, has the form in (84b): (84)

a.

[DP de [PosP mijn [nP [n e] NPø ]]]

(de mijne)

b.

[DP (de) [PosP miend [nP [n e] NPø ]]]

((de) miende)

According to this analysis of substantively used possessive pronominals in Dutch the sound schwa can function as a weak pro-form that realizes (i.e., externalizes) a functional head (in casu n). In this pattern of NP-ellipsis (i.e., the NP-complement, or better the root, remains phonetically unrealized), the categorial flavor of the elided material is recoverable on the basis of the spelled out functional head n. In the spirit of this approach towards substantively used pronominals, I would like to propose that e (i.e., schwa) in patterns like boemer de boem, rinkel de kinkel and tsjoek e tsjoek spells out the functional head Conj. This means that boemer de boem, for example, starts out as the syntactic (coordinate) structure in (85a) and is mapped onto a sound representation (externalization) involving two phonological processes, viz., insertion of the paragogic sound /d/ after the interjective root boemer and insertion of /ə/ as a PF-realization of the functional head Conj (see (85b)). Prosodically, this sequence of sounds is mapped onto a syllabically organized structure like (85c), in which the two augmentative sounds /d/ and /ə/ form together a syllable: (85)

a.

[ConjP boemer [Conj’ Conj boem]]

b.

[ConjP boemerd [Conj’ [Conj e] boem]]

c.

[σ boe] [σ mer] [σ de] [σ boem]

(e = /ə/)

It does not seem implausible to put this occurrence of schwa on a par with what I descriptively call augmentative schwa, i.e., the schwa that is added to an element for emphatic and expressive purposes. As exemplified in (86) and (87), certain pronouns and degree words can be phonologically augmented by means of schwa in (variants of) Dutch (see Corver 2004, 2006).

Interjections as structured root expressions

(86)

(87)

a.

ik

(I)

a.’

ikke

(I-e)

b.

dat

(that)

b.’

datte

(that-e)

c.

wat?

(what)

c.’

watte?

(what-e)

a.

Ik vond die som zo verrekt(e) I found that sum so damned(e) ‘I found that sum very difficult to solve.’

b.

Ik vind Jan zo verdomd(e) I find Jan so damned(e) ‘I find Jan a really nice man.’

79

moeilijk. difficult

aardig. nice

I propose that the augmentative schwa in these examples spells out (i.e., externalizes) a functional head position. More specifically: (88)

a.

[nP [n -e] [NP ik]]

b.

[DegP zo [QP verrekt [Q’ [Q -e] [AP moeilijk]]]]

I take the augmented pronominal form ikke to be derived by means of N (i.e., root)-to-n movement, where n externalizes as schwa. In (88b), the degree modifier verrekt occupies the specifier position of QP (see Corver 1997a,b). The Q-head (optionally) externalizes as schwa, yielding a doubly filled XP-configuration; that is, both the Spec-position and the head position are pronounced. Analogously to the augmentative schwas in the structural environments in (88), I take the schwa that appears in complex interjective expressions to be a “minimal” sound that realizes a functional head, viz., Conj. Let me finish this section with a brief and more speculative remark about the sequences /ər/ and /əl/, which typically precede the augmentative sound combination d+e (see also note 9). As is clear from the examples in (89), these complex interjective expressions have a frequentative/repetitive meaning to them. This repetitive meaning is, first of all, linguistically encoded by the iterative use of an interjective atom (e.g., bolder in holder de bolder de bolder de bolder and kinkel in rinkel de kinkel de kinkel). But also the elements -er and -el in expressions like holder de bolder and rinkel de kinkel, respectively, seem to contribute to this frequentative/repetitive meaning. As exemplified in (89), the sound sequence er also occurs in verbs that have a frequentative/repetitive meaning (see also Van Langendonck 1979):

80 (89)

Norbert Corver

fladderen ‘to flutter’, flodderen ‘to flounder’, debberen (dialectal) ‘to potter’, bibberen ‘to shiver’, ploeteren ‘to dabble’, dabberen ‘to stamp’, lebberen ‘to sip’, snotteren ‘to snivel’.

That -el contributes a frequentative/repetitive meaning aspect is suggested by minimal pairs such as huppen (‘to hop’) vs. huppelen (‘to hop (repeatedly)’) and duiken (‘to plunge/dive’) vs. duikelen (‘to tumble/somersault’).28 Summarizing, I have argued that complex interjective expressions like boemer de boem and rinkel de kinkel consist of two coordinated roots and that a paragogic alveolar obstruent (/d/) is attached at the end of the first root and that a schwa spells out the functional head Conj. Prosodically, the sound sequence /də/ surfaces as a syllabic unit (i.e., [σ də]). I further proposed, building on Van Haeringen (1938), Overdiep (1937) and Corver (2004), that these instances of /d/ and /ə/ can be characterized as “augmentative sounds” having an expressive/emphatic flavor.

8 Conclusion As noted in section 1, Sapir (1921: 5) argued that “Interjections are among the least important of speech elements.” De Groot (1963) argued exactly the opposite: “The interjection is one of the most important and most curious word classes” (taken from the English summary at the end of his article written in Dutch). Although I don’t consider the question about the ranking of interjections on the scale of importance to be very relevant, I have tried to show, on the basis of a case study on complex interjections in Dutch, that interjective expressions are worthwhile to investigate from a syntactic point of view. In this article, in which I focused on the internal syntax of interjective expressions rather than on their external syntax, I have argued (i) that complex interjective expressions are structured expressions, (ii) that their structure is phrase-structural rather than “word-structural” (i.e., compound-like), (iii) that their phrasal organization is based on coordination (i.e., ConjP), and (iv) that the computational atoms of complex interjective expressions – i.e., the simplex interjective atoms – are roots (i.e., (). Thus, complex interjections of the type discussed in this paper have the following structure: [ConjP ( [Conj’ Conj (]]. It was further shown that Conj sometimes surfaces overtly as the sound schwa, which in certain phonological contexts was accompanied by the phonological process of d-insertion after roots

28 I would like to thank Marcel den Dikken for the pointing out the minimal pair huppen vs. huppelen.

Interjections as structured root expressions

81

ending in -er/-el. In this article, I have tried to lay bare certain properties of complex interjections in Dutch. On the basis of my discussion I hope to have shown that interjections deserve further reflections.

References Ameka, Felix. 1992. Interjections: The universal yet neglected part of speech. Journal of Pragmatics 18: 101–118. Belder, de Marijke. 2011. Roots and Affixes. PhD dissertation, Utrecht University. Besten, Hans den. 1983. On the interaction of root transformations and lexical deletive rules. In W. Abraham (ed.), On the Formal Syntax of the Westgermania. 47–131. Amsterdam: John Benjamins. Boer, Minne G. de. 2008. Talking about interjections. Bulletin of The Henry Sweet Society for the History of Linguistic Ideas 50: 31–44. Booij, Geert. 1977. Dutch Morphology. A Study of Word Formation in Generative Grammar. Lisse: Peter de Ridder Press. Borer, Hagit. 2005a. In Name Only, Vol.1: Structuring Sense. Oxford: Oxford University Press. Borer, Hagit. 2005b. The Normal Course of Events, Vol 2: Structuring Sense. Oxford: Oxford University Press. Carnie, Andrew. 2008. Constituent Structure. Oxford: Oxford University Press. Chomsky, Noam. 1986. Knowledge of Language. New York: Praeger. Chomsky, Noam. 2001. On Phases. In C. Otero et al. (eds.), Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud, 134–166. Cambridge, MA: MIT Press. Chomsky, Noam. 2007. Approaching UG from below. In U. Sauerland and H.-M. Gärtner (Eds.), Interfaces + Recursion = Language?, 1–29. Berlin/New York: Mouton de Gruyter. Chomsky, Noam. 2011a. Problems of Projection. Lecture at LUCL, Leiden University, 14 March 2011. (see http://www.chomsky.nl/chomsky-in-nederland-maart-2011/ for videotaped lecture) Chomsky, Noam. 2011b. Problems of Projection. Ms., MIT (to appear in a special issue of Lingua, Luigi Rizzi (ed.)). Cinque, Guglielmo. 1993. A null theory of phrase and compound stress. Linguistic Inquiry 24: 239–297. Cinque, Guglielmo. 1999. Adverbs and Functional Heads. Oxford: Oxford University Press. Corver, Norbert. 1997a. The internal syntax of the Dutch extended adjectival projection. Natural Language and Linguistic Theory 15: 289–368. Corver, Norbert. 1997b. Much-support as a last resort. Linguistic Inquiry 28: 119–164. Corver, Norbert. 2004. Some notes on emphatic forms and displacement in Dutch. In A. Breitbarth and H. van Riemsdijk (eds.), Triggers, 137–172. Berlin/New York: Mouton de Gruyter. Corver, Norbert. 2006. Proleptic agreement as a good design property. In J. Costa and M. C. Figueiredo Silva (eds.), Studies on Agreement, 47–74. Amsterdam/Philadelphia: John Benjamins. Corver, Norbert and Marc van Oostendorp. 2005. Low Saxon possessive pronominals: Syntax and phonology. In J. Doetjes and J. van de Weijer (eds.), Linguistics in the Netherlands 2005, 73–86. Amsterdam/Philadelphia: John Benjamins.

82

Norbert Corver

Corver, Norbert and Marjo van Koppen. 2011. NP-ellipsis with adjectival remnants: A microcomparative perspective. Natural Language and Linguistic Theory 29: 371–421. Don, Jan. 1993. Morphological Conversion. PhD dissertation. Utrecht University. Groot de, A.W. 1963. De Interjectie. In A.W. de Groot & H. Schultink (eds.), Studies op het gebied van het hedendaags Nederlands, 13–22. Den Haag/The Hague: Mouton & Co. Haas, Wim de and Mieke Trommelen. 1993. Morfologisch handboek van het Nederlands. SDUuitgeverij: ’s-Gravenhage. Haeringen, Coenraad B. Van. 1938. Over z.g. ‘paragogische’ consonanten in het Nederlands. De Nieuwe Taalgids 32: 261–273. Halle, Morris and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In K. Hale and S.-J. Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 111–176. Cambridge, Mass.: The MIT Press. Harley, Heidi, and Rolf Noyer. 1999. “State-of-the-Article: Distributed Morphology”, GLOT International 4.4: 3–9. Johannessen, Janne Bondi. 1998. Coordination. Oxford: Oxford University Press. Kaplan, David. 1999. What is meaning? Explorations in the theory of Meaning as Use. Brief version–draft 1. Ms. UCLA, Los Angeles, CA. Karlsson, Fred. 2010. Recursion and iteration. In H. van der Hulst (ed.), Recursion and Human Language. 43–68, Berlin/New York: Mouton de Gruyter. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: The MIT Press. Koster, Jan. 1975. Dutch as an SOV language. Linguistic Analysis 1:111–136. Kratzer, Angelika. 1999. Beyond ouch and oops: How descriptive and expressive meaning interact. (http://semanticsarchive.net/Archive/WEwNGUy0/), A comment on David Kaplan’s paper, Cornell Conference on Theories of Context Dependency. Laan, Kornelis ter. 1953. Proeve van een Groninger Spraakkunst. Winschoten: Van der Veen. Langendonck, Willy van. 1979. De persoonsnaamgeving in een Zuidbrabants dialekt. Deel II De sociolinguïstische component (Mededelingen van de Vereniging voor Limburgse Dialect- en Naamkunde, Nr. 13). Hasselt. Munn, Allan. 1993. Topics in the Syntax and Semantics of Coordinate Structures. PhD dissertation, University of Maryland. Myers, Scott. 1984. Zero-derivation and inflection. In M. Speas and R. Sproat (eds.), MIT Working Papers in Linguistics 7: 53–69. Oostendorp, Marc van. 2000. Coronalen in het Stad-Utrechts en de structuur van het fonologisch woord. Taal en Tongval 53: 110–128. Overdiep, Gerrit S. 1937. Stilistische Grammatica. Zwolle: N.V. Uitgevers-maatschappij W.E.J. Tjeenk Willink. Pauwels, J.L. 1952. Klanknabootsende en bewegingsschilderende tussenwerpsels en bijwoorden. Taal en Tongval 4: 48–58. Poggi, Isabella. 2009. The language of Interjections. In A. Esposito et al. (eds.), Multimodal Signals, LNAI 5398, 170–186. Berlin Heidelberg: Springer-Verlag. Riemsdijk, Henk van. 1978. A Case Study in Syntactic Markedness. Dordrecht: Foris. Sapir, Edward. 1921. Language. An Introduction to the Study of Speech. New York: Harcourt Brace. Scherer, Klaus R. 1994. Affect bursts. In S. van Goozen, N. van de Poll & J. Sergeant (eds.), Emotions: essays on emotion theory, 161–196. Hillsdale, NJ: Erlbaum. Sturm, Arie. 1986. Primaire syntactische structuren in het Nederlands. Leiden: Martinus Nijhoff.

Interjections as structured root expressions

83

Thiersch, Craig. 1993. Some remarks on asymmetrical coordination. In Drijkoningen, F. and K. Hengeveld (eds.), Linguistics in the Netherlands 1993, 141–152. Amsterdam/Philadelphia: John Benjamins. Thiersch, Craig. 1994. On some formal properties of coordination. In Carlos Martin-Vidé (ed.), Current Issues in Mathematical Linguistics, Amsterdam: North Holland. Toorn, Maarten C. van den. 1960. De interjectie als woordsoort. De Nieuwe Taalgids 53: 260– 264. Uriagereka, Juan. 2008. Syntactic Anchors. On Semantic Restructuring. Cambridge: Cambridge University Press. Verkuyl, Henk. 1993. A Theory of Aspectuality. The Interaction between Temporal and Atemporal Structure. Cambridge: Cambridge University Press. Wagner, Michael. 2005. Prosody and Recursion. PhD dissertation, MIT. Wagner, Michael. 2010. Prosody and recursion in coordinate structures and beyond. Natural Language and Linguistic Theory 28: 183–237. Weerman, Fred. 1989. The V2 Conspiracy. Dordrecht: Foris. Wilkins, David. 1992. Interjections as deictics. Journal of Pragmatics 18: 119–158. Zwart, Jan-Wouter. 1993. Dutch Syntax. PhD dissertation. University of Groningen.

Günther Grewendorf

The internal structure of wh-elements and the diversity of wh-movement* 1 Introduction It is well-known that what has traditionally been analyzed as wh-movement covers a collection of different types of movement to the left periphery, each targeting a different landing site in the left periphery and being subject to specific constraints. In particular, (overt) “wh-movement” in wh-questions occurs as topic movement, focus movement, and force (type)-movement. In this paper it will be shown that the different types of movement that wh-elements undergo correlate with specific properties of the internal structure of wh-elements. Following proposals for a split-DP analysis, I will argue in chapter 2 that wh-elements have an internal left periphery that roughly corresponds to the layering of functional categories in the left clausal periphery. After offering evidence for the presence and the hierarchical order of the categories SpecificP, FocP and WhP as part of what we traditionally call a wh-phrase, it is shown in chapter 3 that assuming such a “split-wh” structure provides us with the possibility of attributing a number of interesting restrictions on wh-movement in languages such as English, Italian, Hungarian, and Serbo-Croatian to the interaction of a “split-wh” structure, the structure of the left clausal periphery, and properties of a checking/ probing approach.

2 The internal structure of wh-elements 2.1 Motivating a “split-wh” analysis The crucial idea of this paper is that the type of movement that a wh-element undergoes is determined by the featural make-up of functional categories that are by assumption representations internal to wh-phrases. Following ideas in Giusti (2006), Poletto (2008) and Grewendorf and Poletto (2011), I assume that like the structure of the clause, the internal structure of DPs contains a left * I am grateful to Željko Bošković, Damir Cavar, Eric Fuss, Andreas Pankau, and Ede Zimmermann for help with empirical questions and for revealing discussions about several theoretical problems addressed in this paper. Thanks also go to two anonymous reviewers for their helpful comments and suggestions.

86

Günther Grewendorf

periphery that is comparable to at least part of the left clausal periphery. In particular, I will present evidence that wh-phrases contain functional categories in their left periphery that represent their topical, focal, and modal properties, the latter understood as relevant to the typing of the clause. The topical properties of wh-elements are not to be understood in the sense of left-dislocated topics. Recalling Stowell and Beghelli’s (1994) analysis of specific quantifiers, we can characterize the topical properties of wh-elements in terms of the notion of specificity as described in Enç (1991) and applied to wh-elements in É. Kiss (1993). According to Enç’ definition, specificity is a property of NPs that expresses the familiarity of a discourse referent to which the referent of the NP bears a subset relation. A wh-element can be characterized as specific if it quantifies over a set that is familiar to the participants of the discourse. Specificity can be either inherently associated with a wh-expression, as in the case of which NPs, or acquired as a result of contextual factors. As pointed out by É. Kiss, the wh-pronoun who is inherently easier to attribute a specific interpretation to than the wh-pronoun what although what type/kind of x can also be regarded as inherently specific.1 Proceeding from Chomsky’s (1995) idea (see also Cinque 1999) that each functional head is assigned exactly one feature, and extending ideas of Rullmann and Beck (1998), Boeckx and Grohmann (2004), and Grewendorf (2012), we can represent the specificity of a wh-element as a head Specific, which takes a whelement as a complement. In many languages, wh-movement displays the properties of focus movement. In Italian wh-questions, fronting of wh-phrases is in complementary distribution with the fronting of contrastive focus (Rizzi 1997). In languages such as Tuki, where wh-questions can be formed with wh-elements ex situ or whelements in situ, fronted wh-elements are morphologically marked with a focus particle (Sabel 1998). I will consider the focal property of a wh-element as a genuine operator property, which provides alternatives and is semantically responsible for the type-shifting typically associated with contrastive focus and which creates sets of propositions (either the set of true answers along the lines of Karttunen or the set of possible answers along the lines of Hamblin). I take this property to be represented as an empty operator in a wh-internal Focus projection. The featural correlate of this operator in the left clausal periphery is located in Foc and provides the basis for the interrogative semantics of the clause. 1 É. Kiss (1993) characterizes the specificity of an operator in a more general way, whereby the set to be quantified over is familiar not only when the individuals in this set are known directly to the speaker and the listener but also when the speaker and the listener share some criterion of classification according to which they can exhaustively partition this set in an identical way.

The internal structure of wh-elements and the diversity of wh-movement

87

In addition to the focus feature, we need a representation of the type of the wh-element (e.g. interrogative, relative, indefinite) as either determined by its lexical properties or through a checking/Agree relation with the typing feature of a relevant left-peripheral head. In other words, we need a sentence-type operator that determines interrogative/relative scope and provides the typing of the clause (similar to Lipták’s 2001 [+wh]-feature). As far as the syntactic reflex of this feature is concerned, two head positions are available in Rizzi’s (1997, 2001) analysis of the left clausal periphery to host the typing feature of an embedded interrogative clause and to act as the target of selection: the Force head and the Int head. I will tentatively assume that the relevant head is Force and that the corresponding feature of a wh-element is represented within a wh-phrase as an operator feature in a functional projection WhP. The head of this projection can be conceived of as a covert instance of Cable’s (2010) “Q-particle”. Cable shows that in Tlingit, a language spoken in the southeastern part of Alaska, an overt particle sá, which heads its own projection, selects a wh-DP. As Cable demonstrates (Cable 2010: 67f), Q-particles are crucial for the typing of the clause and are interpreted as variables over choice functions, which in wh-questions are bound by an interrogative operator (represented by an interrogative C-head). Note that in Tlingit, the Q-particle sá can be followed by a focus particle, with which it then morphologically forms a portmanteau. Against the background of these considerations, the internal structure of a wh-phrase such as which computer might thus be represented as in (1): (1)

[SpecificP [FocP Op . . . [WhP Op . . . [DP which computer]]]]

The features projecting these categories can be assumed to be interpretable. Consequently, they need not undergo a “checking” process, but constitute the interpretable correlate that serves to check corresponding uninterpretable features in the left clausal periphery. It should be clear that the head Specific projects only if a wh-element is in fact characterized by the property of specificity. Before dealing with arguments for the specific hierarchy of the projections internal to wh-elements, I will present independent evidence for the feature [specific] as well as for the dissociation of the sentence-type feature and the focus feature.

2.2 Evidence for wh-internal functional heads Morphological evidence for the existence of a SpecificP within the nominal domain comes from the language Gungbe, investigated by Aboh (2004). Aboh shows that in this language, specific noun phrases must occur to the left of the specificity marker lɔ́ . Since specificity is related to topicality, the sequence

88

Günther Grewendorf

consisting of the noun and the specificity marker can undergo topic movement to the left periphery of the clause and be checked against a Topic head in the left clausal periphery. This can be seen from the example in (2) (Aboh 2004: 2, ex. (3a)): (2)

Gungbe [Lɛsì Gúkɔ́mὲ tɔ̀n lɔ́] yà é nɔ̀ víví Rice Gukome POSS DET [+spec; +def ] TOP 3SG HAB sweet ‘As for the aforementioned rice from Gukome, it is very sweet.’

gbáú very

Aboh concludes from this observation that “there is topic specification both within the nominal left periphery and the clausal left periphery” (Aboh 2004: 2). Independent evidence suggests that there is a high position in the left periphery of the clause that is specifically designed to host wh-elements with topic-like properties. It can be shown (cf. Jaeger 2004, Grewendorf and Poletto 2011) that this position is higher than Rizzi’s TopP and corresponds with the “Referential Phrase” (RefP) that Stowell and Beghelli (1994) assume for quantifier phrases with specific reference and which they localize above the root CP. I will assume that specific wh-phrases target the specifier position of RefP (which may be located above ForceP in verb-second languages and below ForceP in non-verbsecond languages; cf. Grewendorf 2010). É. Kiss (1993) has argued for a specificity filter according to which an operator OPi that has scope over another operator OPj and binds a variable in the scope of OPj must be specific. The specificity filter constrains wh-extraction out of a wh-island, predicting extraction to be possible only for specific whelements (cf. (3)), and implying for multiple wh-questions that the operator with wide scope cannot be represented by a non-specific wh-element (cf. the Hungarian examples in (4)): (3)

a. ??What personi do you wonder which present to give to ti? b.

(4)

Which personi do you wonder which present to give to ti?

Hungarian a. *Miért/??hogyan kit választottak why/how whom elected-they ‘*Who did they elect why/how?’ b.

Kit miért/??hogyan választottak whom why/how elected-they ‘Why/how did they elect who?’ (É. Kiss 1993: 101)

meg? PERF

meg? PERF

The internal structure of wh-elements and the diversity of wh-movement

89

The specificity filter accounts for further semantic and syntactic properties of multiple wh-questions in Hungarian, where only one of the preposed wh-elements functions as an interrogative operator; the wh-element that has wider scope functions as a distributive universal quantifier similar to each: (5)

a.

Kinek mit hozott János? whodat whatacc brought John ‘For each person, what did John bring for him?’

b.

Mit kinek hozott János? whatacc whodat brought John ‘For each thing, who did John bring it for?’ (É. Kiss 1993: 98)

Without anticipating the phenomena analyzed in the next section, I will briefly point out an important empirical consequence of this analysis of specific whelements. If it turns out to be true that a wide scope wh-operator of a multiple wh-question functions as a distributive universal quantifier, we can make a prediction about the presence of Superiority effects. Wh-phrases that can be analyzed syntactically as universal quantifiers are not subject to wh-movement proper. If fronting of a wh-phrase W2 is not subject to the same kind of movement as fronting of another wh-phrase W1 that c-commands W2, and if the target positions of W1 and W2 can be shown to differ, then we expect fronting of W2 not to be associated with a Superiority effect. This prediction is borne out by the well-known fact that an unmoved specific wh-phrase does not violate Superiority in multiple wh-questions of English.2 Evidence for a DP-internal Focus projection can be derived from the fact, pointed out in Alexiadou, Haegeman, and Stavrou (2007), that in a language like Greek, focus-fronting can occur DP-internally. In (6b) the boldfaced constituent has undergone DP-internal focalization:

2 See the contrast in (i): (i) a. *What did who buy? b. What did which linguist buy? Pesetsky’s (2000) analysis of examples such as (i) accounts for the absence of a Superiority effect in terms of the assumption that it is actually the wh-subject that undergoes movement first (in the form of feature movement).

90 (6)

Günther Grewendorf

Greek a. to vivlio tu the book thegen ‘Chomsky’s book’ b.

Chomsky Chomsky

tu Chomsky to vivlio (Alexiadou, Haegeman, and Stavrou 2007: 80)

Similar observations about DP-internal focus-fronting in Bulgarian can be found in Dimotrova-Vulchanova and Giusti (1998). Morphological evidence for the claim that the nominal system may involve not only a Topic projection but also a Focus projection can again be obtained from Gungbe. This language displays nominal question-marking in the sense that complex question words consist of a noun phrase and a question marker tɛ́ , analyzed by Aboh (2004) as a nominal focus marker. As expected, the interrogative expression thus formed undergoes overt fronting to the left clausal Focus projection (CFocP) in order to check the clausal Focus feature (Aboh 2004): (7)

Gungbe [Távò xɔ́xɔ́ tɛ́] wɛ́ Kòfí xɔ́? Q FOC Kofi buy table old ‘Which old table did Kofi buy?’ (Aboh 2004: 7)

Finally, the parallelism between interrogative clauses and interrogative DPs in Greek suggests that there is a wh-projection within DP that may function as the target position for DP-internal wh-fronting. Compare clausal wh-fronting in (8) with DP-internal wh-fronting in (9):3 (8)

Greek a. Ekane ti? did3sg what ‘He did what?’ b.

Ti ekane? what did3sg ‘What did he do?’

3 DP-internal wh-fronting is also familiar from English examples such as (i) (i) How intelligent a piece of advice is this?

91

The internal structure of wh-elements and the diversity of wh-movement

(9)

a.

to vivlio tinos? the book whogen ‘whose book?’

b.

tinos to vivlio? whogen the book ‘whose book?’ (Alexiadou, Haegeman, and Stavrou 2007: 81)

The argument for the dissociation of the sentence-type feature and the focus feature is related to interesting properties of wh-imperatives, as investigated by Reis and Rosengren (1992). In a wh-imperative such as (10), the fronted whphrase does not take interrogative scope over the matrix clause, but determines the interrogative force of the embedded clause.4 The reading of (10) is thus equivalent to that of an imperative clause containing an embedded wh-clause, as in (11): (10)

German Wohin where to

sag tell

mir me

bitte please

doch mal modal part.

gleich right away

dass that

Peter Peter

gegangen ist. gone is ‘Tell me please right away where Peter went.’ (Reis and Rosengren 1992: 80) (11)

Sag tell

mir me

bitte please

doch mal modal part.

gleich right away

wohin where

Peter Peter

gegangen ist. gone is ‘Please tell me right away where Peter went.’ The relevant fact about the dissociation of type feature and focus feature is provided by Reis and Rosengren’s observation (1992: 96, 105) that the fronted 4 According to Reis and Rosengren (1992), this is why only those bridge verbs permit whimperatives which are (also) subcategorized for [+wh]-complements (note that the verbs subcategorized exclusively for [+wh]-complements are not bridge verbs). That is, wh-imperative movement is possible from complements of verbs like sagen (‘tell’), vorstellen (‘imagine’), erklären (‘explain’), mitteilen (‘inform’) but not from complements of verbs like glauben (‘believe’), denken (‘think’), wünschen (‘wish’), versprechen (‘promise’). In other words, whimperatives are possible only with [+wh]-complement clauses.

92

Günther Grewendorf

wh-element in wh-imperatives may take wide quantifer scope over a quantifier in the matrix clause (although this reading is not easy to get), despite the fact that the fronting of the wh-phrase does not determine interrogative force of the matrix clause. This can be seen from the fact that an example like (12a) may have a reading as indicated in (12b): (12)

a.

Wen whom

stell imagine

getroffen met b.

sich

mal mod.part.

jeder everyone

vor part.

dass that

ich I

habe. have

For every person x that I met, everyone should imagine that I met x.

Thus, wh-fronting with wh-imperatives and wh-fronting with wh-questions behave alike with respect to quantifier scope effects, although they differ with respect to interrogative scope. That the force/type feature and the focus feature of wh-elements may operate independently of each other can also be seen from the fact that the fronted wh-element in wh-imperatives, while not determining interrogative scope in its target position, may nevertheless trigger a Weak Crossover effect, which is typically associated with focus movement, cf. the contrast in (13): (13)

a.

Weni who

erklär tell

du you

mal modal part.

deiner your

Chefin boss

dass that

sie she

Chefin boss

dass that

sie she

entlassen soll. fire should ‘Tell your boss who she should fire.’

ti

b. *Weni who ti

erklär tell

entlassen fire

du you

mal modal part.

seineri his

soll. should

2.3 The hierarchy within “split-wh” Let us now turn to evidence for the specific hierarchical structure assumed in (1). Ideally, morphological marking of wh-elements and DPs more generally would tell us which of the possible hierarchical orders is correct. However, even in languages that display morphological marking of focus and specificity

The internal structure of wh-elements and the diversity of wh-movement

93

on DPs, it is often unclear whether the morphological marker is lexically associated with the DP or is located in the corresponding functional head in the left clausal periphery, attaching to the DP through movement. Yet, we have found morphological evidence from Gungbe that DP-internal Topic and Focus projections are located in a high peripheral position in the DP. Obviously, in those languages where there is evidence for a lexical representation of DP-internal information structure, there is a tendency for marking of DP-internal information structure to occur at the edge of the DP, possibly subject to cross-linguistic variation, as assumed by Dimitrova-Vulchanova and Giusti (1998). Ihsane and Puskás (2001) show that the highest projection in the left periphery of nominal categories is associated with the encoding of specificity. Another argument that comes to mind is semantic in nature. Intuitively, the property of being specific may affect a wh-element as well as a focused category. Thus, it seems plausible from a semantic point of view for this property to be represented in a position higher than FocP and WhP. The assumption that FocP is in a higher position than WhP also has some intuitive plausibility given that it is the DP rather than the FocP that is lexically characterized as an interrogative element. Furthermore, the assumption that the FocP dominates the WhP is supported by Lipták’s (2001) generalization that a language with overt focus movement has overt wh-movement as well. Finally, recall Cable’s (2010: 22) observation that the Q-particle, which we have taken to be an overt counterpart of the wh-head that selects a wh-DP, can be followed by a focus particle in Tlingit. Let us now take a closer look at the syntactic consequences of the proposal made in (1), seeing how it enables us to account for several mysterious properties of wh-movement in a variety of languages. The extent to which the proposal in (1) provides us with interesting analyses will further support the specific hierarchy of functional projections assumed in this proposal.

3 Varieties of wh-movement 3.1. Wh-movement as focus movement A common observation cross-linguistically is that wh-movement shows similarities with movement of a contrastive focus. Wh-elements and contrastively focused constituents move to the same root-clause target position in a wide range of languages (Lipták 2001). This is the preverbal position in Kashmiri (Bhatt 1999), Italian (Rizzi 1997), Hungarian (É. Kiss 1987), and Basque (Ortiz de Urbina 1989); the clause-initial position in Greek (Tsimpli 1995) and Russian (Bošković 2002);

94

Günther Grewendorf

the position in front of the left-peripheral focus marker in Gungbe (Aboh 2004); the postverbal position in the Chadic languages (Tuller 1992); and the clausefinal position in Tangale and Ngizim (Tuller 1992). In a number of languages, such as Tuki (Biloa 1995; Sabel 1998) and Gungbe, fronted wh-elements bear a morphological focus marker, just as fronted focused elements do. That whelements front to a focus position in Italian is evident from their complementarity with non-wh foci in the same clause, as shown in (14), as well as from the fact that wh-elements, like contrastive foci, follow left-dislocated topics, as shown in (15) (Rizzi 1997): (14)

a. *A TO

GIANNI GIANNI

che cosa what

hai have-you

detto told

(, non a Piero)? (not to Piero)

b. *Che cosa A GIANNI hai detto (, non a Piero)? (15)

a.

A to

Gianni, Gianni

che cosa what

b.

A to

Gianni Gianni

QUESTO THIS

gli him gli him

hai have-you

detto? told

dovrete you-should

dire. tell

The fact that a fronted wh-element is marginally compatible with a contrastive focus in embedded wh-questions suggests that the wh-element can occupy an independent position lower than the focus position in embedded interrogatives: (16) ?Mi domando A I wonder TO (, non a Piero). (not to Piero)

GIANNI GIANNI

che cosa what

abbiamo we-have

detto said

The incompatibility in Italian of wh-fronting with focus-fronting, even if the fronted wh-element is specific, suggests an analysis along the following lines. If overt wh-fronting is obligatorily focus-fronting, then we have to assume that wh-internal DP-raising cannot target a position higher than wh-internal FocP.5

5 I will tentatively propose that wh-internal overt DP-movement to WhP, FocP, and SpecificP correlates with overt movement of WhP, FocP, and SpecificP to the left clausal periphery; cf. Aboh’s (2004: 10) conjecture that nominal topicalization may “favour or license clausal topicalization”. The underlying idea may be expressed as the assumption that the “strength” of the relevant features or the combination of the feature with the EPP is similar in the clausal and the wh-internal left periphery.

The internal structure of wh-elements and the diversity of wh-movement

95

It is then always the wh-internal FocP that raises overtly to left-peripheral clausal FocP (CFocP), with the remnant SpecificP either being checked in situ or covertly raising to RefP. If there is in fact a parametric option with whinternal DP, then we can account for the data in (16) by saying that in this case, wh-internal DP-raising only reaches the specifier of wh-internal WhP, followed by overt WhP-extraction to a left-peripheral position lower than FocP. This may be the specifier of FinP, where the selectional requirements of the matrix verb are satisfied through a checking relation between Force and Fin. English also displays a complementarity of wh-fronting and focus-fronting if we assume that negative inversion constitutes movement to a left-peripheral focus position. The relevant observation here is that negative inversion and whfronting are incompatible within the same clause in root as well as in embedded sentences (Den Dikken and Giannakidou 2002: 50): (17)

a.

*What under no circumstances should he do?>

b.

*John wondered what under no circumstances should he do.

Even if we assume that in English, similar to Italian, wh-fronting and negative inversion target the clausal Focus projection, we still have to account for an obvious difference between the two kinds of fronting: While both trigger subject-auxiliary inversion in root clauses, only negative inversion gives rise to subject-auxiliary inversion in embedded clauses. Den Dikken and Giannakidou (2002) assume that in embedded wh-questions, the head of Foc moves overtly to the higher interrogative head C “before Foc does anything itself”; this, they claim, satisfies Foc’s strong feature, preventing an auxiliary from raising to Foc. They also take Foc-to-C movement to require a [+foc]-bearing overt specifier in SpecCP. The result is that SpecFocP is not projected, thus accounting for the impossibility of negative inversion in embedded wh-questions. This proposal, however, is clearly countercyclic, which casts doubt on its correctness. I will thus suggest a different explanation for the absence of subject-auxiliary inversion with embedded wh-questions, which is similar to the analysis of the Italian example (16). Let us maintain the assumption that both wh-fronting and negative inversion target (or pass through) the left-peripheral FocP in root contexts, which explains their complementarity. As regards embedded wh-clauses, we can assume a parametric option similar to what we assumed for Italian: wh-internal DP-raising only reaches the specifier of the wh-internal WhP and only the wh-internal WhP undergoes overt raising and targets the specifier of FinP. This again satisfies the selectional requirements of the matrix verb through a checking relation between

96

Günther Grewendorf

Force and Fin. The impossibility of subject-auxiliary inversion can then be explained similarly to the impossibility of verb second in embedded wh-interrogatives in a verb-second language like German. What (17b) shows is that negative inversion in embedded clauses cannot target a position below FinP and (unlike Italian (16)) cannot target the higher FocP either since (again unlike Italian) English negative inversion requires verb movement to Foc. Verb movement to Foc, however, is not possible, since the verb cannot pass through Fin. In Hungarian root and embedded clauses, wh-phrases raise to the position reserved for foci. This can again be seen from their complementarity with nonwh foci as well as from their placement after topics (É. Kiss 1987):6 (18)

(19)

Hungarian a. *Kit whoacc

MARI MARI

hívott invited

meg? preverb

b. *MARI MARI

kit whoacc

hívott invited

meg? preverb

a.

Mari MariTOP

kit whoacc

hívott invited

meg? preverb

b.

Kiváncsi vagyok hogy Mari curious I am that MariTOP ‘I wonder who Mari invited.’

kit whoacc

hívott invited

meg. preverb

Further syntactic similarities between wh-fronting and focus-fronting consist in their both displaying subjacency effects and both licensing parasitic gaps. There is an obvious similarity between the Hungarian facts and those in Italian, which suggests that in Hungarian too it is the wh-internal FocP that moves overtly to the left-peripheral Focus projection CFocP. In other words, it is 6 Cable (2008) argues that the feature that triggers wh-fronting in Hungarian is not identical with the feature that triggers focus-fronting since unlike focus-fronting, wh-fronting is not associated with ‘exhaustive identification’. He assumes instead that obligatory wh-fronting in Hungarian is due to a fronting-operation triggered by some morpho-syntactic feature of the wh-word. Cable’s argument suffers from an unclear notion of ‘exhaustivity’; furthermore, even if his alternative is on the right track (for which no independent evidence is presented), his argument does not crucially affect the analyses proposed in the present paper. On the one hand, Cable does not deny that in many languages, wh-operators and ‘focused’ phrases target the same structural position; on the other hand, even if the relevant morpho-syntactic feature of wh-words were different from a focus feature, this could easily be incorporated into my analysis.

The internal structure of wh-elements and the diversity of wh-movement

97

again the specifier of the wh-internal FocP to which the wh-internal DP raises. The operator that is associated with clausal typing – that is, the operator within WhP – undergoes checking in situ or is moved to ForceP. If the wh-element is specific, the remnant SpecificP moves to the higher RefP (or is checked in situ).

3.2 Intervention effects If this analysis is on the right track, we can make an interesting prediction. After the wh-internal FocP moves to the left-peripheral CFocP, there is still a syntactic relation between the operator in WhP and features of Force. This relation is not present when a non-wh focus has been moved to clausal CFocP, since a fronted focus has no impact on clausal typing. We therefore expect that it is only in the former case that an operator intervening between ForceP and clausal FocP might have a negative effect on the grammaticality of a wh-question. This prediction is in fact borne out: While a left-peripheral focus can be preceded by a quantifer, a fronted wh-element cannot (Lipták 2001: 67):7 (20)

a.

Minding PÉTERT hívtam meg. always Peteracc I-invited preverb ‘At all times, I invited Peter only.’

b. *Minding always

kit whoacc

hívtam I-invited

meg. preverb

If we assume that in non-verb-second languages like Hungarian, the RefP or any other target position for (topical) left-peripheral quantifiers is located below

7 Lipták (2001: 83) points out that the pattern exhibited by the Hungarian examples in (20) can also be observed in Romanian, where quantifiers can precede fronted focus but not fronted wh-elements: (i) Romanian a. *Nimeni ce nu-ti va aduce? nobodynom whatacc not-you will bring ‘What will nobody bring you?’ b. Nimeni nici ATENŢIE nu-i dădea. nobodynom not an attentionacc not-him/her gave ‘Nobody was paying THE LEAST ATTENTION to him.’ (Motopanyane 2000)

98

Günther Grewendorf

ForceP,8 then our analysis attributes the contrast between (20a) and (20b) to the following consideration: Unlike (20a), (20b) requires a checking/probing relation between Force and the WhP (within the fronted FocP), which, however, cannot be established because of the intervening quantifier.9 The structural contrast between (20a) and (20b) is represented in (21), the offending intervener being underlined in (21b) (21)

a.

[ForceP Force [RefP minding [CFocP [FocP PÉTERT] [FinP . . .]]]]

b.

[ForceP Force [RefP minding [CFocP [FocP kit [WhP Op]] [FinP . . .]]]] C

The hypothesis that fronted wh-elements and fronted non-wh focus are in complementary distribution in Hungarian (as well as in Italian main clauses) seems to face an empirical problem. This is that unlike multiple focus-fronting, multiple wh-fronting is possible in Hungarian: (22)

a.

Ki whonom

kit whoacc

b. *PÉTER MARIT Peternom Maritacc (Lipták 2001: 63)

hívott invited hívta invited

meg? preverb meg. preverb

To solve this problem, let us recall É. Kiss’ (1993) assumption that in Hungarian multiple wh-questions only one wh-element occupies left-peripheral CFocP; other fronted wh-elements are interpreted as specific and thus constitute universal quantifiers, which move to a topical position higher than left-peripheral CFocP (arguably our RefP). While multiple focus-fronting is ungrammatical, given that there is only one focus position available in the left periphery of the clause, multiple wh-fronting is not subject to this restriction. This, according to É. Kiss (1993), is because the higher wh-element in (22a) occupies a position different 8 Evidence for this assumption emerges from the observation in É. Kiss (1987: 59) that whelements preceding the wh-operator in CFocP must follow a non-wh topical element (which may occupy ForceP): (i)

Neked [mikor ki [mit mondott]]? to-you when who what said

(ii) *Mikor ki neked [mit mondott]? when who to-you what said 9 For an account in terms of an intervention effect, see also Lipták (2001: 81).

The internal structure of wh-elements and the diversity of wh-movement

99

from CFocP. But if the higher (specific) wh-element is interpreted as a quantifier, why does it not trigger an intervention effect similar to that triggered by the quantifier in (20b)? The answer to this question is that the fronted specific whelement, which occupies a topical position below ForceP, itself contains a WhP that may satisfy the requirements of clausal typing. Although it intervenes between ForceP and the wh-element in the left-peripheral CFocP, this intervention is irrelevant, since its own WhP can undergo movement (checking) for clausal typing. The relevant left-peripheral structure is represented in (23): (23)

[ForceP Force [RefP [ SpecificP ki [FocP [WhP Op]]] [CFocP [FocP kit]] [FinP ]]]] C no intervention

This account provides independent support for our analysis of the internal structure of wh-elements. Our account of the contrast in (20) can also provide an explanation for an interesting generalization about English wh-the-hell phrases given in Den Dikken and Giannakidou (2002: 35) (originally due to Lee 1994). Den Dikken and Giannakidou argue that wh-the-hell phrases (as well as carrying a presupposition of negative attitude) constitute Focus phrases, extending the domain of quantification to include familiar and novel values and thus cover the entire interpretation domain D. Accordingly, Den Dikken and Giannakidou assume that wh-the-hell phrases occupy the specifier of the left-peripheral CFocP.10 This assumption correlates with Bošković’s (2002) observation that focus movement licenses single-pair readings of wh-interrogatives, since unlike regular wh-phrases, wh-the-hell phrases support only single-pair readings (Den Dikken and Giannakidou 2002: 35) (note that wh-the-hell questions do not allow for non-echo interpretations in multiple wh-structures): (24)

a.

Who is in love with who? [single-pair echo or pair-list]

b. (?)Who the hell is in love with who? [single-pair echo only] 10 Actually, Den Dikken and Giannakidou make use of Pesetsky’s (1989) insight that the position of wh-phrases in English is not uniform: in root wh-questions, a topic precedes a fronted wh-element, while in embedded wh-questions a topic must follow a wh-phrase: (i) a. ?A book like this why should I buy? b. ?Bill doesn’t know why a book like this, he should buy. If the position of topics is held constant, then wh-phrases must raise to different positions in root and embedded wh-clauses. Based on the cartographic analysis, according to which FocP is located below TopP and TopP below CP (in the sense of ForceP), Den Dikken and Giannakidou (2002) assume that wh-movement targets SpecCP in embedded clauses but SpecFocP in root contexts.

100

Günther Grewendorf

The interesting generalization, then, is that unlike other fronted wh-elements, wh-the-hell phrases always have wide scope with respect to a universal quantifier: (25)

a.

What did everyone buy for Max?

b.

What the hell did everyone buy for Max?

While (25a) allows two readings, with either narrow or wide scope for the universal quantifier, (25b) has only the reading whereby the universal quantifier takes narrow scope with respect to the wh-the-hell phrase. This contrast can be accounted for in much the same way as the contrast between (20a) and (20b): raising of the universal quantifier to a position above CFocP (where the wh-thehell phrase is located) would intervene between Force and the WhP inside the fronted wh-the-hell phrase, triggering an intervention effect that would block the checking/probing relation required for clausal typing. Notice that this account implies that the fronted wh-element in (25a) cannot occupy CFocP but must have moved to a higher position, probably ForceP (contrary to the claim in Pesetsky 1989; cf. n. 15).11 A related phenomenon has been observed in Italian. Rizzi (2004) points out that preposed adverbials such as improvvisamente (‘suddenly’) and rapidamente (‘quickly’) cannot, unlike genuine topics, naturally precede fronted wh-elements if they are used as quantificational elements rather than topics: (26)

Italian a. Rapidamente, hanno fatto i quickly they-have done the ‘Quickly, they did the homework.’ b. ??Rapidamente, che cosa quickly what (Rizzi 2004: 239)

hanno have-they

compiti. homework

fatto? done

Rizzi concludes from the ungrammaticality of (26b) that the target position of preposed adverbials (in their non-topical use) is lower than the position filled by the wh-elements. There is, however, an alternative. It may also be the case that the target position of the fronted adverbials in (26b) is higher than the position of the wh-elements. The ungrammaticality of (26b) might then be due to an intervention effect similar to what we assumed for the Hungarian data in 11 One may assume (Den Dikken and Giannakidou 2002) that quantifier raising targets a position higher than ForceP when associated with a wide scope reading.

The internal structure of wh-elements and the diversity of wh-movement

101

(20). Evidence for this account comes from the fact that it correctly predicts the co-occurrence of the preposed adverbials and a fronted non-wh focus. Since a fronted non-wh focus does not require an Agree relation between a typing feature of Force and a corresponding feature of the element in CFocP, we predict that preposed adverbials like rapidamente can precede a fronted Focus phrase. This prediction is borne out by sentences such as (27):12 (27)

Rapidamente SOLO UN GENIO lo potrebbe rapidly only a genius it could ‘Rapidly, ONLY A GENIUS could solve it.’ (Rizzi 2004: 245)

risolvere. solve

3.3 Multiple wh-fronting in Serbo-Croatian In Serbo-Croatian, a well-known multiple wh-fronting language, wh-movement likewise seems to display the properties of focus-fronting. Bošković (2002) argues that focus movement in Serbo-Croatian is involved whenever wh-movement is not associated with Superiority effects.13 Superiority effects do not appear in short-distance null-C matrix questions, as in (28), but do appear in overt-C questions like (29), as well as in embedded and in long-distance questions, as in (30):

12 A reviewer raises the question as to why there is no intervention effect in examples like (16), where the clausal FocP intervenes between Force and Fin such that checking of the type feature of the wh-element should be blocked. (16) is repeated here as (i): (i) ?Mi domando A GIANNI che cosa abbiamo detto I wonder TO GIANNI what we-have said (, non a Piero). (not to Piero) My answer to this question is that in (i) Agree takes place between CFoc and the focused nonwh phrase, while the wh-phrase remains in SpecFinP (no FocP being generated within the wh-element). The type feature of Force percolates to Fin, and the type feature of WhP is checked within the FinP. Notice that feature percolation is not subject to intervention effects (as can be seen from multiple complementizers). The situation is different with (26b), where the WhP is located higher than FinP (as part of FocP it is located in CFocP), such that in this case there should be an Agree relation between the type feature in Force and the type feature of the WhP. 13 Bošković (2002) assumes that focus movement is insensitive to Superiority.

102 (28)

(29)

Günther Grewendorf

Serbo-Croatian a. Ko koga voli? who whom loves ‘Who loves whom?’ b.

Koga ko voli?

a.

Ko li koga voli? who C whom loves ‘Who on earth loves whom?’

b. *Koga li ko voli? (Bošković 2002: 353f) (30)

a. ?Ko koga kažeš da je who whom say that is ‘Who do you say beat whom?’

istukao? beaten

b. *?Koga ko kažeš da je istukao? Serbo-Croatian thus shows Superiority effects in those contexts in which French requires overt wh-movement. Bošković (2002) accordingly concludes that only in those contexts is wh-fronting in Serbo-Croatian an instance of wh-movement to interrogative SpecCP (SpecForceP). In the case illustrated in (28), wh-fronting is obligatory but as an instance of focus movement. Bošković (2002) takes as further evidence for this conclusion the interpretive properties of multiple wh-questions. If it is true that overt movement to interrogative SpecCP (SpecForceP) forces pair-list answers, then the availability of single-pair answers in short-distance null-C matrix wh-questions such as (28) shows that the wh-phrases do not move to interrogative SpecCP in this case. Bošković assumes that here clausal typing can be carried out within the focuslicensing projection. I do not disagree with Bošković’s claim that focus-fronting is involved in the examples in (28). However, I do disagree with his argument for this claim. The wh-elements in (28) are said to constitute instances of focus-fronting, since no Superiority effect is involved. The presupposition of this argument is that focus movement does not trigger a Superiority effect. Unfortunately, this cannot be independently verified, since most languages with overt focus-fronting do not permit multiple focus-fronting. Moreover, focus-fronting triggers a Weak Crossover effect, thus revealing itself to be operator movement. But the assumption that two instances of operator movement are not associated with a Superiority effect is far from plausible.

The internal structure of wh-elements and the diversity of wh-movement

103

It would thus make sense to consider an alternative analysis which proceeds from the assumption that focus-fronting is indeed associated with a Superiority effect. We can then conclude that at least one of the fronted wh-elements in (28) has not undergone focus-fronting. Let us tentatively dispense with the dubious “tucking in” strategy (Richards 2001, Chomsky 2008) and assume instead that the rightmost wh-element has been fronted first (as assumed for multiple wh-fronting in Hungarian). This raises the question of why (28a) does not trigger a Superiority effect. To answer this question we must take into account the fact that wh-elements may be specific or non-specific. Let us first consider the case where the wh-subject is specific (constituting a SpecificP). As already pointed out above, there is independent evidence that specific wh-subjects do not trigger a Superiority effect. This can be seen from the fact that in English, D-linked wh-subjects do not trigger this effect (Comorovski 1996): (31)

a. *What did who read? b. *Which book did who read? c.

What did which student read?

It can also be seen from the Hungarian example (22a), repeated here for convenience:14 (32)

Hungarian Ki kit whonom whoacc

hívott invited

meg? preverb

In (32), the wh-object moves to the left-peripheral focus position CFocP, thereby crossing the wh-subject – which, as shown by É. Kiss (1993), must then be specific. Further evidence for the claim that specific wh-subjects do not trigger a Superiority effect comes from the observation given in Den Dikken and Giannakidou (2002: 53), that in Hungarian multiple wh-questions, wh-the-hell phrases cannot occur clause-initially. (33)

a.

Ki who

mi a fenét what the hellacc

b. *Ki a fene who the hell

mit whatacc

vett? bought vett? bought

14 See also È. Kiss (1987: 60) for the general claim that Hungarian does not display Superiority effects.

104

Günther Grewendorf

If the wh-the-hell phrase occupies left-peripheral CFocP, as argued by Den Dikken and Giannakidou, it must first have undergone wh-fronting in (33a); this means that it would have crossed the wh-subject, which by Kiss’ assumption must be specific.15 If the wh-the-hell phrase is the subject, as in (33b), it cannot be crossed by the wh-object, since a wh-the-hell phrase cannot be specific. As for the Serbo-Croatian example in (28a), we can conclude that the whobject kogo has crossed the specific wh-subject and reached its destination in CFocP without a violation of Superiority. The specific wh-subject ko, constituting a SpecificP, may then move to a topical target position higher than CFocP. This accounts for the absence of a Superiority violation in (28a) if the wh-subject is specific. Let us now consider the case where the wh-subject is non-specific, thus constituting only a FocP. Since the wh-subject is closer to CFocP than the wh-object, first movement of the wh-object to CFocP should be banned by Superiority. Note, however, that Slavic multiple wh-fronting languages permit the formation of wh-clusters, as argued for on independent grounds by Grewendorf (2001) (Saito 1994 reaches a similar conclusion for multiple wh-fronting in Japanese). If wh-cluster formation can only affect two wh-elements with the same categorial status, we can assume that the FocP of the wh-object adjoins to the FocP of the non-specific wh-subject to undergo overt fronting with the wh-subject, thereby avoiding a Superiority violation. Assuming left-adjunction (Kayne 1994), we can conclude that the wh-cluster reaches CFocP (possibly passing through FinP), as shown in (34): (34)

Serbo-Croatian [CFocP [(koga (ko))]]

The head of this cluster is then extracted from the cluster, moving to ForceP in order to satisfy the requirements of clausal typing.16 This yields the leftperipheral configuration in (35), again deriving (28a) without violating Superiority: (35)

[ ForceP ko

[CFocP (koga (tko))]]

15 An explanation for the absence of Superiority violations with specific wh-subjects might be seen in the fact that the FocP within the SpecificP of the wh-subject does not c-command the FocP of the wh-object. 16 Actually, I assume that it is only the WhP-part of the wh-element ko that moves to ForceP. Note that unlike Hungarian, Serbo-Croatian does not display obligatory fronting of contrastively focused elements; given our assumptions, this means that there is no overt wh-internal movement to SpecFocP. It follows, then, that the wh-internal raising that the wh-DP undergoes does not reach a position higher than SpecWhP.

The internal structure of wh-elements and the diversity of wh-movement

105

Notice that the adjoined wh-object cannot be extracted from the cluster. This means that (28b) can be derived only if it is the wh-subject (its FocP) that moves first to CFocP; the implication is that the wh-object can undergo long movement to the left clausal periphery only if it is specific and can thus move to the leftperipheral topic position RefP. The requirements of clausal typing are then satisfied by the WhP of the fronted wh-object. Interesting evidence for this analysis can be found in Bošković’s (2002) observation of a contrast between (36a) and (36b): (36)

a.

Tom čoveku, ko je šta poklonio? that mandat who is what bestowed ‘On that man, who bestowed what?’

b. ??Tom čoveku, šta that mandat what (Bošković 2002: 362)

je is

ko who

poklonio? bestowed

Bošković’s account of this contrast is that Serbo-Croatian displays Superiority effects with a fronted topic constituent even in short-distance null-C questions. However, this account is not available to us, given our rejection of the “tucking in” strategy. On our assumptions, (36b) cannot constitute a Superiority violation, since it is the wh-subject ko that has undergone first movement to the left periphery; so we are forced to offer a different analysis for the ungrammaticality of (36b). In fact, our account of (28b) provides us with a way to explain this ungrammaticality without making reference to Superiority. Recall from this account that long movement of the wh-object to the left clausal periphery is impossible unless the wh-object is specific. If we now make the rather harmless assumption that the specific wh-object and the specific topic in (36b) target the same kind of topical position in the left periphery (there being only one RefP), we can attribute the ungrammaticality of (36b) to the complementary distribution of these two topical elements. Our account of the Serbo-Croatian examples in (28) can also be applied to multiple wh-questions in Russian and Polish, which do not exhibit any Superiority effects: (37)

Russian a. Kto kogo ljubit? who whom loves ‘Who loves whom?’ b.

Kogo kto ljubit?

106 (38)

Günther Grewendorf

Polish a. Kto co kupił? who what bought ‘Who bought what?’ b.

Co kto kupił? (Schmidt 2008)

As pointed out by É. Kiss (1993: 108), an account that attributes the absence of a Superiority effect in certain cases to the specificity of wh-elements provides an alternative explanation of the famous contrast in (39), analyzed by Kayne (1983: 176) in terms of connectedness: (39)

a.

*I’d like to know whati who hid ti there.

b.

?I’d like to know whati who hid where.

While (39a) is ruled out as a Superiority violation, the addition of the third whphrase where in (39b) enables the in situ wh-element who to assume a specific interpretation and thus to escape the effect of the Superiority condition. Let us now return to (29), repeated here for convenience: (40)

Serbo-Croatian a. Ko li koga voli? who C whom loves ‘Who on earth loves whom?’ b. *Koga li ko voli? (Bošković 2002: 353f)

Notice again that on our approach, the ungrammaticality of (40b) cannot be analyzed as a Superiority violation, since it is the wh-subject that has undergone fronting first. In order to account for the contrast in (40), a brief consideration of the semantics of fronted wh-elements followed by the particle li is in order. With this kind of question, the speaker expresses doubts that the hearer will be able to answer the question.17 The fact that the addition of the particle li to the wh-element turns the wh-question into a “deliberative” question is reminiscent 17 Thanks to Damir Cavar for pointing this out to me. Obviously, the semantics of wh+li is similar to that of Russian čto za-N expressions, as investigated by Polinsky (2007). According to Polinsky (2007), expressions like who on earth, wh+li, and čto za-N do not evoke an answer and form constructions that are “pragmatically closer to exclamatives”.

The internal structure of wh-elements and the diversity of wh-movement

107

of the function of echo wh-words. Adopting Den Dikken and Giannakidou’s (2002) analysis of these wh-words, I will assume that the particle li is attached to a wh-element. This makes wh-elements hosting li distinct from regular whphrases in introducing the presupposition that the speaker does not expect the hearer to provide an answer. This analysis of the wh-element and the particle li as a cluster receives independent support from a fact pointed out to me by Željko Bošković, namely, that li occurs between the wh-element and its extension in a D-linked wh-phrase (like which li book). Note that the presupposition associated with wh-elements followed by li signals another semantic property of these wh-elements, which Bošković glosses as ‘who on earth’. This implies that wh-elements hosting the particle li can be taken to be non-specific.18 Given the wh-internal structure argued for in this paper, we can express this observation by saying that wh-elements marked with a li particle never form a SpecificP. The fact that the wh+li combination is still interpreted as an interrogative constituent taking wide scope can be captured by arguing that li bears an uninterpretable feature that must enter a checking relation with a typing feature in Force. The assumption that wh-elements form a cluster with the particle li and have to move to SpecForceP also captures the fact that exactly one wh-element is allowed to precede the particle li. With these considerations in mind, we can analyze the contrast between (40a) and (40b) along the following lines. Let us first turn to (40a). On our assumptions, the configuration in (40a) rather than the one in (40b) should produce a Superiority violation, since the wh-object has fronted first and thus crossed the non-specific wh-subject+li, which we analyze as a FocP. Two cases should be distinguished. While ko+li cannot be a SpecificP, the wh-object koga may represent a SpecificP or a FocP. Let us first consider the second case. The fact that both wh-elements constitute a FocP implies that the wh-object cannot cross ko+li due to Superiority. Thus, the only way for the wh-object to undergo overt fronting is by first forming a wh-cluster with the wh-subject, along the lines suggested in Saito (1994), Grewendorf (2001), and others. Still assuming left adjunction (Kayne 1994), we can obtain the wh-cluster formation shown in (41): (41)

18 Polinsky (2007) shows that wh-elements like who on earth are not associated with any special reference to a presupposed set of people but instead have a non-referential property denotation and are thus related to intensional reference only.

108

Günther Grewendorf

The cluster in (41) then moves to left-peripheral CFocP (possibly passing through FinP), which results in the configuration in (42): (42)

[CFocP (koga (ko+li))

[FinP t(koga (ko+li))

[TP t(koga

(ko+li))]]]

The head ko+li in the adjunction structure of (42) (actually its WhP) then undergoes further movement to ForceP, yielding (43), which derives (40a): (43)

[ForceP [WhP ko+li]

[CFocP (koga (tko+li)) [ FinP . . .]]]

In case the wh-object is a SpecificP, it is able to bypass FinP and FocP and move directly to the high topic position. Thus in that case, the wh-subject ko+li can undergo stepwise raising to CFocP, and the wh-object is allowed to bypass FinP and FocP and raise directly to a higher topic position. The final step is then the raising of the WhP ko+li (out of its FocP) to its final destination in ForceP, which again derives (40a): (44)

[ForceP [WhP ko+li] [RefP [ SpecificP koga] [CFocP tko+li [ FinP tko+li . . . tSpecificP]]]]

One question that arises here is why the uninterpretable feature of the Force head cannot be checked by the SpecificP koga, which would result in the ungrammatical order (45): (45)

Koga ko li . . .

The answer to this question is that the derivation of (45) is ruled out on independent grounds. If the typing feature in Force were checked by movement of koga, the uninterpretable feature of the particle li would remain unchecked, which would cause the derivation to crash.19 Let us now turn to the problem of how to account for the ungrammaticality of (40b), which we cannot analyze as a Superiority violation according to the assumptions made above. We thus have to show that the derivation of (40b) is ruled out on independent grounds. The starting point of such a derivation is the configuration in (46): 19 Note that there are other ways to attain the same result. We could claim that the presence of a wh+li phrase means that other wh-elements of the clause can no longer be specific either. Although I have not tested this prediction, it seems to me to be rather counterintuitive that the wh-object in a deliberative question such as (40a) could not have a specific reading. Another possibility would be to assume that the operator elements of a SpecificP that has undergone topic checking in a RefP are no longer subject to operator checking. Even if plausible, this assumption is just not necessary to achieve the desired result of ruling out (45).

The internal structure of wh-elements and the diversity of wh-movement

(46)

109

[TP ko . . . koga+li]

Again, we have to distinguish two cases depending on whether or not ko is specific. Let us assume first that ko is non-specific and thus constitutes a FocP. If this FocP raised to CFocP, the wh-object koga+li would be confined to its base position: since it does not represent a SpecificP, it could not undergo long topic movement. As a consequence, the uninterpretable feature of li could not be checked in ForceP and the derivation would crash. We still might assume that koga+li is adjoined to the wh-subject ko and that the cluster so formed raises to CFocP, yielding (47): (47)

[CFocP [koga+li [ko]] [FinP . . .]

In that case, however, the adjoined element koga+li could not be extracted from the wh-cluster and undergo further movement to ForceP. Only the head ko of the adjunction structure, or the entire cluster, could raise to ForceP. In either case, the uninterpretable feature of li could not be checked: it would either be left behind in CFocP or be buried inside the cluster. We can thus assume that wh+li clusters can never adjoin to another wh-element and thereby undergo whcluster formation. It is a consequence of this assumption that (40b) cannot be derived. The same conclusion follows if the wh-subject in (40b) is specific. If it is the FocP of the wh-subject that undergoes movement to CFocP, the wh-object koga +li must stay in situ (since as a FocP it cannot undergo long topic movement) and li remains unchecked. On the other hand, the wh-object cannot form a cluster with the specific wh-subject, since as a SpecificP the latter has a different categorial status. Thus, there is no way for the wh-object koga+li to have the uninterpretable feature of li checked.20 A problem arises if the wh-object koga+li moves to CFocP and this movement is followed by long topic movement of the specific wh-subject. The resulting configuration parallels the configuration in (44): (48)

[ForceP [RefP [SpecificP ko] C ?

[CFocP [FocP koga+li] [ FinP tSpecificP . . . tkoga+li ]]]]

20 Note that cluster formation in FinP is likewise impossible. Even if the FocP of the whsubject passes through SpecFinP, adjunction of the wh-object to the wh-subject in SpecFinP would be countercyclic.

110

Günther Grewendorf

Recall from our analysis of Hungarian multiple wh-fronting that a wh-subject that constitutes a SpecificP does not trigger a Superiority effect. It should thus be possible for the wh-object koga+li to move to left-peripheral CFocP and then for the specific wh-subject to undergo long topic movement. At this point in the derivation, nothing prevents the wh-object in CFocP from undergoing movement to ForceP, as was the case in (44). The result is the ungrammatical order given in (40b). At present, I see two possibilities for ruling out (48). The first possibility is the requirement that in multiple wh-questions, wh+li expressions can reach the left clausal periphery only through cluster formation. This strategy would rule out (44) (without any negative consequences) as well as (48). The other possibility is the weaker requirement that in multiple wh-questions, a wh-object with li can reach the left clausal periphery only through cluster formation. This proposal would only rule out (48). I will leave this problem open, pending further independent evidence. The final problem with Serbo-Croatian wh-questions concerns the ungrammaticality of (30b) with long distance wh-extraction, which is analyzed by Bošković (2002) as a Superiority effect. (30) is repeated here as (49): (49)

a. ?Ko koga kažeš da je who whom say that is ‘Who do you say beat whom?’

istukao? beaten

b. *?Koga ko kažeš da je istukao? As shown by Saito (1994) and Grewendorf (2001), the only way for multiple wh-elements to leave an embedded clause is through wh-cluster formation. If both wh-elements are FocPs, the cluster is formed within the embedded TP by adjunction of the wh-object to the wh-subject, yielding the configuration in (50): (50)

[FocP [(koga (ko))]]

Following Rizzi (2009), we can assume that the wh-cluster reaches CFocP of the matrix clause through successive cyclic movement via the embedded CFocP position, since Rizzi regards this position as a purely formal, uninterpretable counterpart of the matrix criterial CFocP. From the embedded operator position CFocP, it can only move directly to the matrix operator position CFocP to fulfill the requirements of proper movement. The configuration in CFocP of the matrix clause would therefore be (51). (51)

[CFocP [FocP [(koga (ko))]]]

The internal structure of wh-elements and the diversity of wh-movement

111

Subsequent movement of the WhP ko to ForceP derives the well-formed (49a). The same kind of cluster formation is possible if both wh-elements are SpecificPs. Recall that cluster formation is possible only with two wh-elements of equal categorial status. Thus, if the wh-subject is a FocP and the wh-object a SpecificP, the FocP of the latter can adjoin to the wh-subject. But if the whsubject is a SpecificP and the wh-object a FocP, no cluster formation is possible. As a result, if a wh-cluster reaches CFocP in the matrix clause, it is identical to the cluster in (51). In other words, only (49a) can be derived.

4 Summary In this paper I have tried to show that wh-elements have an internal functional structure that correlates with the layering of the left-peripheral structure of the clause, as assumed by Rizzi (1997, 2001). On the basis of such a “split-wh” hypothesis, I have suggested derivations for the different types of left-peripheral movement which wh-fronting may belong to: topic movement, focus movement, and force movement. It is a crucial claim of the analysis that the different types of wh-fronting each target a different landing site in the left periphery and are associated with specific wh-internal DP-movement to appropriate specifier positions. I have shown that these structural assumptions provide us with a new account of mysterious differences in the behavior of focus-fronting of nonwh-elements and wh-elements in Hungarian and Italian. Finally, intriguing generalizations on Superiority effects in Serbo-Croatian could be attributed to an interaction of the “split-wh” structure, the functional structure of the clausal left periphery and the checking/probing mechanisms operating on their functional features.

References Aboh, Enoch 2004. “Topic and focus within D”, Linguistics in the Netherlands 21, Amsterdam: John Benjamins, 1–12. Alexiadou, Artemis, Haegeman, Liliane and Stavrou, Melita. 2007. Noun phrase in the generative perspective, Berlin, New York: Mouton de Gruyter. Bernstein, J. B. 2001. “Focusing the ‘right’ way in Romance determiner phrase”, Probus 13, 1–29. Bhatt, Rajesh. 1999. Verb Movement and the syntax of Kashmiri, Dordrecht: Kluwer. Biloa, Edmond. 1995. Functional categories and the syntax of focus in Tuki, München: Lincom Europa.

112

Günther Grewendorf

Boeckx, Cedric and Grohmann, Kleanthes. 2004. “SubMove: towards a unified account of scrambling and D-linking”. In: D. Adger et al. (eds.), Peripheries, Dordrecht: Kluwer, 241– 257. Bošković, Željko. 2002. “On multiple wh-fronting”, Linguistic Inquiry 33, 351–383. Cable, Seth. 2008. “Wh-fronting (in Hungarian) is not focus-fronting”, ms. University of British Columbia, Vancouver, BC Cable, Seth. 2010. The grammar of Q, Oxford: Oxford University Press. Chomsky, Noam. 1995. The minimalist program, Cambridge Mass.: The MIT Press. Chomsky, Noam. 2008. “On phases”. In: Robert Freidin, Carlos Otero and Maria Luisa Zubizarreta (eds.), Foundational issues in linguistic theory, Cambridge, Mass: The MIT Press, 133– 166. Cinque, Guglielmo. 1999. Adverbs and functional heads. A cross-linguistic perspective, Oxford: Oxford University Press. Comorovski, Ileana. 1996. Interrogative phrases and the syntax-semantics interface, Dordrecht: Kluwer. Den Dikken, Marcel and Giannakidou, Anastasia. 2002. “From hell to polarity: ‘Aggressively non-D-linked’ wh-phrases as polarity items”, Linguistic Inquiry 33, 31–61. Dimotrova-Vulchanova, Mila and Giusti, Giuliana. 1998. “Fragments of Balkan nominal structure”. In: Artemis Alexiadou and Christopher Wilder (eds.), Possessors, predicates and movement in the determiner phrase, Amsterdam: John Benjamins, 333–361. É. Kiss, Katalin. 1987. Configurationality in Hungarian, Dordrecht: Reidel. É. Kiss, Katalin. 1993. “Wh-movement and specificity”, Natural Language and Linguistic Theory 11, 85–120. Enç, Mürvet. 1991. “The semantics of specificity”, Linguistic Inquiry 22, 1–25. Giusti, Giuliana. 2006. “Parallels in clausal and nominal periphery”. In: Mara Frascarelli (ed.), Phases of interpretation, Berlin: Mouton de Gruyter, 163–184. Grewendorf, Günther. 2001. “Multiple wh-fronting”, Linguistic Inquiry 32, 87–122. Grewendorf, Günther. 2010. “On the typology of verb second.” In: Thomas Hanneforth and Gisbert Fanselow (eds.), Language and Logos, Berlin: Akademie-Verlag, 70–94. Grewendorf, Günther. 2012. “Wh-movement as topic movement”. In: Laura Brugè et al. (eds.), Functional heads, Oxford: Oxford University Press, 55–68. Grewendorf, Günther and Poletto, Cecilia. 2011. “Hidden verb second: the case of Cimbrian”. In: Michael T. Putnam (ed.), Studies on German-language islands, Amsterdam: John Benjamins, 301–346. Ihsane, Tabea and Puskás, Genoveva. 2001. “Specific is not definite”, Generative Grammar in Geneva 2, 39–54. Jaeger, T. Florian. 2004. “Topicality and superiority in Bulgarian wh-questions”. In: Olga Arnaudova et al. (eds.), Annual workshop on formal approaches to Slavic languages, The Ottawa Meeting 2003, Ann Arbor: Michigan Slavic Publications, 207–227. Kayne, Richard S. 1983. Connectedness and binary branching, Dordrecht: Foris. Kayne, Richard S. 1994. The antisymmetry of syntax, Cambridge Mass.: The MIT Press. Lee, Felicia. 1994. “Negative polarity licensing in wh-questions”, MA-thesis, UCLA. Lipták, Aniko. 2001. On the syntax of wh-items in Hungarian, Universiteit Leiden: LOT. Motopanyane, Virginia. 2000. “Parameters for focus in English and Romanian”. In: Virginia Motopanyane (ed.), Comparative studies in Romanian syntax, Amsterdam: Elsevier, 265– 294. Ortiz de Urbina, Jon. 1989. Parameters in the grammar of Basque, Dordrecht: Foris.

The internal structure of wh-elements and the diversity of wh-movement

113

Pesetsky, David. 1989. “Language-particular processes and the Earliness Principle”, ms. MIT. Pesetsky, David. 2000. Phrasal movement and its kin, Cambridge Mass.: The MIT Press. Poletto, Cecilia. 2008. “Doubling as a spare movement strategy”. In: Sjef Barbiers, Olaf Koeneman, Marika Lekakou and Margreet van der Ham (eds.), Microvariation in syntactic doubling”, Bingley: Emerald, 38–68. Polinsky, Masha. 2007. “What on earth. Non-referential interrogatives”. In: Nancy Hedberg and Ron Zacharski (eds.), The grammar-pragmatics interface, Amsterdam: John Benjamins, 245–262. Reis, Marga and Rosengren, Inger. 1992. “What do wh-imperatives tell us about wh-movement?”, Natural Language and Linguistic Theory 10, 79–118. Richards, Norvin W. 2001. Movement in language. Oxford: Oxford University Press. Rizzi, Luigi. 1997. “The fine structure of the left periphery”. In: Liliane Haegeman (ed.), Elements of grammar, Dordrecht: Kluwer, 281–337. Rizzi, Luigi. 2001. “On the position ‘Int(errogative)’ in the left periphery”. In: Guglielmo Cinque and Giampaolo Salvi (eds.), Current studies in Italian syntax, Amsterdam: Elsevier, 287– 296. Rizzi, Luigi. 2004. “Locality and left periphery”. In: Adriana Belletti (ed.), Structures and beyond, Oxford: Oxford University Press, 223–251. Rizzi, Luigi. 2009. “Movement and concepts of locality”. In: Massimo Piattelli-Palmarini, Juan Uriagereka and Pello Salaburu (eds.), Of minds and language. A dialogue with Noam Chomsky in the Basque country, Oxford: Oxford University Press, 155–168. Rullmann, Hotze and Beck, Sigrid. 1998. “Presupposition projection and the interpretation of which-questions”. In: Devon Strolovitch and Aaron Lawson (eds.), Semantics and linguistic theory, Ithaca: Cornell University, 215–232. Sabel, Joachim. 1998. “Principles and parameters of wh-movement”, Habilitationsschrift, University of Frankfurt/Main. Saito, Mamoru. 1994. “Additional-wh effects and the adjunction site theory”, Journal of East Asian Linguistics 3, 195–240. Schmidt, Ewa. 2008. “Mehrfache Wh-Fragen”, MA-thesis, University of Frankfurt/Main. Stowell, Tim and Beghelli, Filippo. 1994. “The direction of quantifier movement”, paper presented at the GLOW conference, Vienna. Tsimpli, Ianthi M. 1995. “Focusing in Modern Greek”. In: Katalin É. Kiss (ed.), Discourse configurational languages, Oxford: Oxford University Press, 176–206. Tuller, Laurie. 1992. “The syntax of postverbal focus constructions in Chadic”, Natural Language and Linguistic Theory 10, 303–334.

Jan Koster

Relative clauses: Parallelism and partial reconstruction 1 The rise and fall of transformationalism It is generally assumed that linguistics underwent a revolutionary change since the 1950s. One of the original core ideas of this revolution was the following: (1)

Transformationalism Sentences have more than one level of representation, connected by transformations

Well-known examples of levels are deep structure and surface structure (later Dstructure, S-structure and Logical Form). The deepest level, according to this kind of view, is a full reconstruction of lexical elements in their interpretively most natural and complete form and order (as in so-called ‘kernel sentences’). I will refer to this view as ‘transformationalism.’ Transformationalism goes back to work by Zellig Harris, who was concerned with ‘normalization’ in discourse analysis, i.e., reconstruction of a level at which ‘co-occurrence’ relations are optimally stated (Harris 1957). Passive sentences, for instance, were ‘normalized’ to active sentences, Wh-questions to declaratives, etc. Reconstruction, in this sense, was thought to be ‘total.’ Modern generative grammar gave a new twist to these ideas but inherited the idea of total reconstruction, most saliently in the various accounts for displacement phenomena (‘movement’). Take, for instance, the following examples (– is the reconstruction site): (2)

a.

Bill was arrested –

b.

Who did Mary see –

Total reconstruction in post-Harrisian grammar meant that there was an underlying level of representation in which displaced forms could appear in their full lexical form: (3)

a.

. . . was arrested Bill

b.

. . . Mary (did) see who

116

Jan Koster

However, total reconstruction was by no means limited to movement transformations. It also characterized substitution transformations (4) and deletions (5): (4)

Reflexivization a. John saw John b.

(5)

John saw himself

Equi-NP Deletion a. John wants [John to go] b.

==>

==>

John wants [Ø to go]

Around 1970, it was discovered that such forms of total reconstruction do not work, particularly not with quantified NPs: (6)

a.

Everyone saw everyone ≠ Everyone saw himself

b.

Everyone wants everyone to go ≠ Everyone wants (PRO) to go

In order to overcome such problems, total reconstruction by transformations was replaced by partial reconstruction in the form of rules of construal. Consider reflexivization: (7)

a.

John saw himself

b.

Everyone saw himself

In this new approach, reflexives like himself were no longer fully reconstructed to the shape of their antecedent at the underlying level. Instead, the anaphoric form himself was already present as such at this level and total reconstruction was replaced by partial reconstruction. In this case, that meant reconstruction not of the lexical shape of the antecedent but of its referential index. A similar approach replaced Equi-NP Deletion (5), where the deletion site was replaced by a base-generated empty subject: PRO, with anaphoric/pronominal properties. In this case, too, full reconstruction was replaced by partial reconstruction. In short, the alternative to transformations with their total reconstruction was more adequate from an empirical point of view and more generally applicable. Curiously, movement transformations were excluded from this new approach. This was truly remarkable because there are good empirical arguments that indicate that reconstruction in the case of ‘movement’ is also partial rather than total (see Koster 1987: 54–56). Moreover, the rare arguments against a generalized construal approach to displacement phenomena are obviously invalid, as we will see.

Relative clauses: Parallelism and partial reconstruction

117

In retrospect, it is the rediscovery of the lexicon (Chomsky 1965, 1970) that should have finished off transformational grammar. In my opinion, it completely undermined the idea of the revolutionary discontinuity of generative grammar with respect to earlier traditional and structuralist approaches to syntax. True, generative grammar also often meant a shift from sterile instrumentalist (‘positivistic’) theorizing to psychological-biological realism. But that is a matter of choice, as pre-generative theories could have been (and sometimes were) interpreted realistically as well. Technically speaking, the main innovation of generative grammar was transformationalism and the connected multi-level representation of sentences. Apart from the perspective of generalized partial reconstruction (‘construal’), there were two other developments in the early 1970s indicating that syntax is monostratal rather than multi-level and transformational. Both developments can be seen as a form of structure-preservingness. The best-known form of that is the insight of Joseph Emonds (1970) that most major movement transformations move elements to positions independently given by X-bar theory. At the time, many linguists saw that as a vindication of transformationalism, in particularly of the idea that transformations are ‘constrained.’ However, a less biased conclusion should have been that Emonds’s structure-preservingness showed that movement transformations were superfluous in the first place (but see Brame (1978), Bresnan (2001), Sag and Pollard (1994), Koster (1978a, 1987)). The second aspect of structure-preservingness, proposed roughly in the same period, was trace theory. Trace theory was the idea that movement transformations like Wh-movement, leave a ‘trace’ in their original position ([NP t ] being the trace): (8)

What did you see [NP t ]

As in the case of Emonds’s structure-preservingness, the very notion of a ‘trace’ points in the direction of a transformational bias. What we could, or rather should, have learned from Aspects (Chomsky 1965) is that verbs have an internal structure as expressed, for example, by their subcategorizaton frame. Thus, [NP t ] in (8) is not a trace of movement. In fact it has nothing to do with movement at all: it just makes explicit the object slot that is a property of the verb see. Here, again, is where the notion of partial reconstuction (‘construal’) comes in. In (8), we do not have to reconstruct the object position for see because this object position is already present, namely as a lexical property of see. The only thing we have to do is to reconstruct the lexical content of [NP t ], which, clearly is a matter of partial reconstruction. This is just an instance of how all construal works, namely an partially incomplete element’s sharing of

118

Jan Koster

properties with an antecedent in the local environment (see Koster 1987: 8ff.). As with anaphors and PRO, it is the lexical incompleteness of the object [NP t ] that drives the completion. The minimalist copying theory of displacement, in contrast, is a residue of transformationalism and its ‘total’ reconstruction paradigm. Copies cannot syntactically interact because they are equally complete and therefore lack the need for completion-by-sharing that characterizes contextual dependencies in general. One could of course say the usual, namely that Merge applies pre-Spell Out and that only one copy is lexicalized (spelled out). But since generalized construal applies in a perfectly straightforward way in the case of displacement, this would add only more ad hoc stipulations to the grammar. Whereas incompleteness triggers completion by a local antecedent, copies can only lead to grammatical inertness. More generally, syntax is mainly a form of local property sharing on the basis of a hierarchical-recursive skeleton. Many lexicalizations only partially identify the categories to which they apply, while further identifying features are ‘borrowed’ from the local environment. Unlike what is falsely claimed for the copying theory, contextual completion of [NP t ] by what comes entirely for free in (8), thanks to the generality of property sharing. Partial reconstruction (as implied by property sharing) was first proposed in detail for anaphora (see, for instance, Dougherty 1969), replacing the ‘total reconstruction’ of the pronominalization and reflexivization transformations of earlier theories. Similarly, the total reconstruction involved in Equir NP Deletion was, on empirical grounds, replaced by the interpretation of empty subject (PRO, see Jackendoff 1969 and Wasow 1972): (9)

John wants [PRO to go]

Perhaps the most decisive death blow to transformationalism was given by Higgins (1973) in his treatment of pseudo-clefts: (10)

[What John saw – ] was himself

According to the ‘total reconstruction’ logic of transformationalism, himself in (10) would be reflexivized ‘John’ and have its deep structure position in the object position of see (indicated by –). This is indeed how pseudo-clefts were treated in Chomsky (1970: 209). However, Higgins, and particularly Blom & Daalder (1977) showed conclusively that total reconstruction plus movement of the focused element is impossible in such cases. In reality, – is interpreted as what (by property sharing) and the pronominal what is interpreted as himself.

Relative clauses: Parallelism and partial reconstruction

119

It is important for what follows that idioms like to make headway can also occur in pseudo-cleft constructions: (11)

[What John never made – ] was any significant headway

Like in (10), it is impossible to ‘totally’ reconstruct the focus element (headway) in the position indicated by – . We only have to reconstruct the position in which headway is interpreted, which is done via the incomplete elements – and what and is a form of partial reconstruction (because the reconstruction does not involve a ‘physical’ deep structure position for headway). The underlying sequential integrity of idioms like make headway was one of the major arguments for the underlying ‘total’ reconstruction level of Dstructure, at least until Chomsky (1981). Most relevant in the present context, it also was the direct motivation for the raising analysis of relative clauses, which has been popular until this very day: (12)

The headway we made –

The reasoning is that headway must have been ‘raised’ to its position in (12) because it has its ‘physical’ origin in the position indicated by –. Only in that position are the two words of the idiom ‘physically’ adjacent: made headway. Given the facts about pseudo-clefts, it should have been clear immediately that the original motivation of the raising analysis, whatever its other merits, is not valid. In what follows, I will make the stronger claim that raising analyses are in fact impossible in certain cases. In other words, like the facts of pseudo-clefts, the facts of relative clauses form strong counter-evidence to transformationalism and its core idea of full-reconstruction levels. Unfortunately, the field took a different course in the 1970s and the multilevel, full-reconstruction approach was still emphatically defended in Chomsky (1981). In fact, the approach was even expanded with an ‘invisible’ dimension in the 1980s, namely by the development of the level of Logical Form (LF). All of this led to stagnation, in my view, rather than lifting the field to the next level. Chomsky explicitly rejected the alternative to transformationalism, most prominently in the last section of Chomsky (1973). The argument (Chomsky 1973: note 70) is based on the following cases (where – is the reconstruction site, representing the deep structure position of the fronted Wh-phrases): (13)

a.

[Which pictures of each other] were the men looking at –

b.

[The pictures of each other] that the men were looking at –

120

Jan Koster

Full reconstruction was said to be necessary here for the men to bind (c-command) each other. This kind of argument is without force and should not have been accepted after Higgins (1973) and similar studies. The ‘trace’, indicated by –, is in fact the object slot associated (as one of its properties) with looking at. Since this object slot is phonologically empty it has to ‘borrow’ its identifying lexical features from its local environment, i.e. from the fronted Wh-phrase that c-commends it. The empty slot – is in the domain of the men. Since it shares properties with the fronted Wh-phrase, the latter (which includes each other) can also be said to be in the domain of the men. Local property sharing suffices. In the remainder of this article, I will target the raising analyses of relative clauses, one of the few areas in which the ‘total’ reconstruction of transformationalism still reigns.

2 Analyzing relative clauses The classical raising analysis for relative clauses is supposed to create structures like the following: (14)

The booki (whichi) we read [NP –]i

At the deepest level of representation, the head of the relative construction, the book, is generated in the position of [NP –]i. From here, it is moved to the position taken by whichi . From here it is further promoted to its final position, leaving behind either and empty place (trace) or the relative pronoun which. This kind of derivation immediately leads to problems (see Smits 1988: 119ff.). First of all, raising of the book to its head position would introduce an entirely new lexical item, namely the relative pronoun which. This is not a copy of the book nor a trace and therefore usually not seen as a permissible operation. Current versions of the raising theory (since Kayne 1994) have sought to remedy this. A second problem that is still with us is that it is not clear if what is raised is a full DP or the NP dominated by it. Consider the original motivation of the raising analysis – full reconstruction. It fails when the full DP is reconstructed: (15)

a.

The headway they made –

b. *They made the headway

Relative clauses: Parallelism and partial reconstruction

121

Vergnaud (1974) gave examples like (16): (16)

a.

The Paris I like –

b. *I like the Paris If, on the other hand, we only raise the NP, reconstruction fails with ordinary DPs (see Borseley 1997 for discussion): (17)

a.

The picture Bill liked

b. *Bill liked picture Bianchi (2000) offers a solution for this problem (in a Kaynean framework), but this solution is ad hoc in the sense that it involves arbitrary stipulations and only applies to the case it is supposed to account for. Even more importantly, as we will see, the Kaynean revisions of the raising analysis (including Bianchi’s) encounter many other problems. It is therefore fair to say that the elementary problems posed by (15–17) are still with us. Before I will go into the Kaynean revision, I will first mention a few other mismatches between the head of the relative clauses and the alleged reconstruction site – mismatches, it seems, that are highly problematic for all raising analyses. It is not too difficult to find relative clauses with heads that never occur in any other position, including the ‘reconstruction’ site. Consider Dutch relatives with datgene as their head: (18)

a.

Datgene wat je – hier That-AF what you here ‘What you saw here is new’

b. *Je You

hebt have

datgene that-AF

hier here

hebt have

gezien seen

is is

nieuw new

gezien seen

Another problem is the mismatch we find when the relative pronoun is an R-pronoun in Dutch. By and large, preposition stranding is only possible with R-pronouns in Dutch (Van Riemsdijk 1978). R-pronouns precede the prepositional head (aan in (19)): (19)

Het boek waar ik [PP – aan] [-R] [+R] [+R] the book where I on ‘The book that I worked on’

gewerkt

heb

worked

have

122

Jan Koster

This relative clause involves preposition stranding, which is only possible for R-pronouns. The problem is that the head of the relative clause, het boek, is not an R-pronoun but an ordinary [-R] DP, which cannot be moved from a PP (with preposition stranding) at all in Dutch. There simply is no PP-internal reconstruction site for ordinary DPs (or NPs). I consider such examples lethal to all raising analyses, including the revisions inspired by Kayne (1994) and the matching analyses based on ellipsis (Salzmann 2006). Let us consider these revisions next. If we abstract away for the moment from the special treatment Kayne gives to NPs without specifier (the analysis modified by Bianchi 2000), a Kaynean analysis of the picture which Bill liked looks as follows: (20)

[DP the [CP [DP [NP picture]k [DP which tk ]]i [IP Bill liked ti ]]]

According to this analysis, the relative pronoun is no longer inserted during the derivation as a kind of lexical trace but is the determiner of the NP that is part of a fronted DP. The traditional head of the relative clause now consists of an external determiner (the in (20)) and the NP (picture) that is moved to the left of its determiner (which). There are four kinds of problems with this type of analysis. First of all, unlike which in (20), most relative pronouns are not plausible determiners of NPs. In many cases, such an analysis is downright impossible. Second, movement of the head across the relative pronoun (seen as a determiner of the head) would lead in most cases to forms of illicit movement. Third, there is massive evidence, particularly from languages like Dutch or German, that the external determiner (the in (20)) forms a constituent with the head noun (picture in (20)). Fourth, there is equally overwhelming evidence that the head thus seen as a constituent, does not even have to c-command the relative clause. The first and fourth of these problems can also be held against the ellipsis-based matching analysis of Salzmann (2006) and others. Let us discuss these problems in turn. In (20), the relative pronoun which is analyzed as the underlying determiner of the head NP (picture). This combination is independently attested in phrases like ‘which picture.’ However, most relative pronouns cannot be plausibly analyzed as underlying determiners of the NP head. Take genitive relative pronouns, like whose: (21)

The woman whose father I have known

Relative clauses: Parallelism and partial reconstruction

123

In this case, the relative pronoun is the determiner of father, so that there is no slot left for the head NP woman. Similar observations can be made about (19), repeated here for convenience: (19)

Het boek waar ik [PP – aan] [-R] [+R] [+R] the book where I on ‘The book that I worked on’

gewerkt

heb

worked

have

In this case, the relative pronoun is an R-word. An R-word is not the determiner of an NP but replaces it in the relevant contexts (see Van Riemsdijk 1978). This R-word also is the ‘filler’ of a gap followed by a postposition (aan in (19)). There are no known cases in which such a gap has a full NP as its ‘filler’. Example (19) is therefore a strong counterexample to a Kaynean-style raising analysis (or to the matching analysis of Salzmann 2006 for that matter). Illicit movement is another serious problem for the kind of raising analysis at issue: (22)

A book [PP about which – ] everybody talked

If book is raised from the gap position (–), the movement would be illicit because extraction from a PP is only possible in English from somewhere at the end of a VP. If a PP is fronted in English, it becomes a very strong island: (23)

a.

You said that [to Bill], nobody talked anymore

b.

**Who did you say that [to – ], nobody talked anymore

In other words, (22) would violate an otherwise exception-less generalization about PP islands in English. The class of illicit movements that a raising analysis would meet can be considerably expanded when we consider SOV languages like Dutch (and German). Unlike what we see in, say, the Romance languages (Cardoso 2010), extraposition of relative clauses is very natural in these languages. What makes things particularly interesting in these languages is that the relative clause is ‘physically’ separated from its head by the verb (DP the head, CP the relative clause): (24)

. . . .DP. . .V. . .CP. . .

A typical example is the following:

124 (25)

Jan Koster

Ik heb [DP de vrouw] GEZIEN [CP die I have the woman SEEN who ‘I saw the woman who knew everything’

alles everything

wist] knew

These structures are interesting because they show beyond reasonable doubt that the external determiner (de in (25)) forms a constituent with the head NP (vrouw in (25)). The DP de vrouw can undergo scrambling, for instance, showing that it is a constituent. The extraposed clause (CP) must be a constituent, too. All of this is incompatible with the Kaynean analysis (20), according to which neither the external determiner and the following NP, nor the remnant of the relative clause (to the right of the NP) form a constituent. There is no obvious way to make facts like (25) compatible with an analysis like (20) or any other raising analysis (the analysis of Salzmann (2006), to its credit, analyzes the external determiner and the following NP as a constituent). Things are even worse than they look at first sight. According to the antisymmetry theory of Kayne (1994), rightward movements, like extraposition, are not possible. I agree with that because it has been observed since the 1970s that extraposition is not constrained by core properties of ‘movement’, like Subjacency (see Koster 1978a: 48ff. for discussion). So, can the head DP in (25) be split from the relative clause DP? Assuming that rightward movement is impossible, we might wonder if leftward movement of the head DP would work (as was originally proposed by Kayne). However, leftward movement of the head would be entirely illicit (see Kaan 1992), as it would involve landing sites in argument positions that, moreover, do not even c-command their underlying position (the ‘gap’ or trace). It is, for instance, entirely unproblematic in Dutch to have the head of the relative clause inside a PP: (26)

Ik heb [PP met [DP de vrouw]] GESPROKEN [CP I have with the woman talked wist] knew ‘I talked with the woman who knew everything’

die who

alles everything

The head DP (de vrouw) can be embeddded even more deeply: (27) Ik heb [PP met [NP de moeder [PP van de vrouw]]] GESPROKEN I have with the mother of the woman talked wist] [CP die alles who everything knew ‘I talked with the mother of the woman who knew everything’

Relative clauses: Parallelism and partial reconstruction

125

Such examples make all leftward movement of the DP head illusory. We must therefore conclude that the separation of the head and the relative clause can be the result neither of rightward movement nor of leftward movement. More generally, we must conclude at this point that none of the existing raising analyses for relative clauses work. The same must be concluded for the matching analyses as proposed by Salzmann and others. In that type of analysis, relative clauses are analyzed as follows: (28)

[DP the book]i [CP [DP which booki]j [ C [ . . .tj. . .]]]

This type of analysis is similar to Kayne’s in that the relative pronoun is a determiner of the head NP. However, instead of moving this NP, it is deleted if it matches an earlier occurrence. As mentioned, this analysis has the virtue of making the DP head and the relative clause (CP) constituents, so that they can be separated in extraposition contexts. This analysis also seems to avoid the problem of illicit movement. However, there is no independent evidence for deletion under the control of a deeply embedded, non-c-commanding DP, as in (27). Most devastating, this kind of analysis inherits all the problems caused by the fact that the relative pronoun is seen as a determiner. Particularly, the R-pronoun example ((19) above) has no possible site of deletion controlled under matching. I therefore reject the matching analysis as well. All in all, as neither existing raising analyses nor deletion analyses work, it is time to consider an alternative.

3 The alternative: Parallel Structure 3.1 Introducing parallel structure The name ‘Parallel Structure’ was suggested by the notation Edwin Williams (1978) proposed for coordination, in which the conjuncts were presented in columns, before linearization: (29)

CP

and

DP

John Mary

reads

DP

a newspaper a book

In this representation, the first row can be seen as the primary structure and the second row as the parallel structure. The primary structure introduces the lexical-functional frame of the sentence, in this case with John filling the subject

126

Jan Koster

slot and a newspaper filling the object slot. The defining characteristic of the parallel structure (second row) is that it does not introduce new functional slots. Instead, it provides extensions of the slots introduced by the first row and is, in that respect, functionally dependent on the primary structure. Columns, in this way, represent material assigned to the same functional slot: (and) Mary is an extension of the subject slot introduced by John and (and) a book is an extension of the object slot, introduced by a newspaper. I will refer to elements of the first row as ‘anchors’ and elements of the second row as ‘extensions.’ The nature of the extension is largely determined by operators, like ‘and’. Columns with only one element (like the column containing reads in (29)) indicate material shared by primary and parallel structure. I will not go further into technicalities here but only introduce the general idea: primary structure defines the functional slots of the sentence (the anchors), parallel structure extends the contents of these slots without introducing new slots (the extensions). Functional slots are properties of lexical items. Parallelism is about how these slots can be realized by more than one element, where the elements are connected by operators. The idea that parallel structure (then not under this name) is also relevant for extraposition phenomena and appositions (see Klein 1977) can be found in a very incipient form in Koster (1978a: 48ff.) and, for PP extraposition even earlier, namely in work since 1976 by Jacqueline Guéron (see Guéron 1980). Such ideas were revitalized by reflection on Kayne’s LCA framework (Kayne 1994), which excluded rightward movements like extraposition (see Kaan 1992, Koster 1995 and 2000). Further exploration of parallel structure, sometimes in a somewhat different form, can be found in Rijkhoek (1998), De Vries (2002), Kluck (2011) and Heringa (2012). Examples like (29) are relatively complex, involving multiple anchor-extension pairings. Most examples I will consider in the remainder of this article involve single, non-sentential pairings, like: (30)

Ik heb en

Jan Piet

gezien

I have

John seen and Peter ‘I have seen John and Peter’ In this case, Jan is the anchor and en Piet the extension. It can be linearized in two ways in Dutch, with the extension directly adjoined to the anchor (31a) or to its containing VP:

Relative clauses: Parallelism and partial reconstruction

(31)

a.

Ik heb [Jan [en Piet]] gezien

b.

Ik heb [[Jan gezien] en Piet]

127

In general, an extension can be either directly adjoined to its anchor or to one of the constituents containing the anchor, as long as we remain in the same clause (a constraint known as the Right Roof Constraint, to be briefly discussed below).

3.2 A preliminary classification of parallel structures Like what used to be called A’-movement structures, parallel structures form a family of constructions that is homogeneous in the sense that they are characterized by a distinctive set of properties. Like with A’-movements, a number of subtypes can be distinguished. Anchors are usually arguments, but not necessarily so (see Rijkhoek 1998 for parallelism with comparatives and result clauses). The further distinctions depend on the nature of the operator and the anchor-extension pair. Without attempting a complete listing, I propose the following preliminary classification: (32)

classification of parallel structures (i) argument extensions a. conjunctions (with ‘and’) b. disjunctions (with ‘or’) ... (ii)

predicative extensions a. restrictives 1. restrictive rel. clauses 2. adjunctive PPs ... b.

non-restrictives 1. equatives 2. appositions 3. appositive rel. clauses 4. right dislocations 5. placeholder specifications 6. result clauses ...

128

Jan Koster

Before discussing the common properties of parallel structures, I will first give a few illustrative examples of the elements listed in (32). Argument extensions expand the number of DPs assigned to the slot of the anchor. I have already given examples with the conjunction and. Constructions with disjunctive or are very similar. Predicative extensions do not add arguments but expand the properties of the anchor. Properties can be restrictive (‘set intersection’) or non-restrictive (‘set union’), also in contexts other than parallel structure: (33)

The industrious Dutch

This can mean that all Dutch people are industrious (‘non-restrictive’) or that only a subset of the Dutch is industrious (‘restrictive’). We find the same division among the predicative extensions. The best-known example is the distinction between restrictive (34a) and non-restrictive relative clauses (34b): (34)

a.

I saw [DP no one ] [CP who knew everything]

b.

I saw [DP John], [CP who knew everything]

I will assume that the operator in relative clauses is the head of CP, the category C, which can be [+restrictive] or [-restrictive]. In both cases, the head of the relative clause —a DP— is the anchor, while the relative clause —a CP— is the extension. In general, I will use bold face for the anchor and italics for the extension (as in (34)). Next to restrictive relative clauses, other common restrictors are the various adjunctive PPs in DPs: (35)

a.

I saw [[DP no one] [PP from India]]

Following the general pattern of anchor-extension combinations, DP and PP can form a constituent, but they can also be separated by other material of the sentence (formerly referred to as ‘PP extraposition’): (35)

b.

I saw [DP no one], yesterday, [PP from India]

With PP restrictors, I assume that the head P of the PP is the operator. If that assumption is correct, relative clauses and restrictive PPs distinguish themselves from most of the other extensions of (32) in that they have a projecting head as operator (C projecting CP and P projecting PP, respectively). As is well-known,

Relative clauses: Parallelism and partial reconstruction

129

conjunctions like ‘and’ do not project in this way. Rather, two conjoined phrases have the distribution of the first conjunct ([DP [and DP]] behaves like a DP, etc.). However, a projecting operator is not unique to the class of restrictive predicative extensions. There are also argument extensions with a projecting head, like constructions with the Dutch preposition behalve (‘except’): (36)

Niemand no one

heeft has

iemand anyone

gezien, [PP seen

behalve except

Jan John

Marie] Mary

Jan is an argument extension of niemand and Marie is an argument extension of iemand. Nevertheless, the phrase headed by behalve has PP distribution rather than DP distribution. The non-restrictives in (32) are mostly specifications, i.e., a sentence with these extensions always entails the simpler sentence without the extensions. For example, (34a) does not entail I saw no one, whereas (34b) does entail I saw John. Specifications can have an overt operator, like the word namely (in: I saw someone, namely Mary) but very often there is a silent operator, like in the equatives of Ross (1969): (37)

a.

He saw something beautiful: a golden igloo

inspired by Ross’s notation, I have proposed a silent operator &:, also known as a colon phrase (Koster 1995, 2000): (37)

b.

He saw [something beautiful [&: a golden igloo]]

I will assume this colon phrase for most specifications with silent operator, but I do not exclude further refinement. Consider, for example, appositional constructions, recently studied in detail by Heringa (2012): (38)

John McClave, my neighbor, is a nice guy

As Heringa shows, there is quite some semantic variation among appositions. In examples like (38), the relation between anchor and extension is more properly characterized as ‘simple predication’ than as identificational specification (as in (37)). For simple predications, we could assume a silent operator ‘BE’ instead of specificational &:. The question to what extent we should make further refinements in the class of operators I will leave for further research.

130

Jan Koster

Another example of specification is known as right dislocation (Ross 1967): (39)

I have not seen him there, that boy

Unlike what we see with equatives (with indefinite anchor), the definite anchor leads here to a defocused extension (indicated by a falling intonation, often referred to as ‘backgrounding’). A very important class of examples is formed by what I would like to call ‘placeholder specification.’ It is doubtful whether CPs, apart from factives, can ever occur in case positions in languages like English or Dutch (cf. Stowell’s Case Resistance Principle in Stowell (1981). Similarly, on the left periphery of the sentence, so-called subject sentences usually specify a pronominal DP placeholder in subject position (see Koster 1978b). I see the relation between the subject placeholder it in (40) and the ‘extraposed’ CP along the same lines: (40)

[DP It] is clear [CP that he will come]

This fits in well with an anchor-extension analysis, as indicated by bold face and italics in (40). Although much more common in Dutch than in English, this kind of placeholder specification also occurs with objects: (41)

We may resent it, perhaps, that he will come

There is some evidence that non-factive CPs are never in DP position, even if there is no overt pronoun (see Koster 2000). I will not go into this matter here but will simply assume that CP complements are extensions specifying pronominal, possibly silent, anchors in parallel structure. A last example is certain comparative extensions and result clauses studied as parallel specifications by Rijkhoek (1998: 138): (42)

Die man heeft meer films gemaakt dan documentaires that man has more movies made than documentaries ‘That man made more movies than documentaries’

(43)

Janna heeft zo snel gelopen dat niemand haar kon bijhouden Janna has so fast walked that no one her could up-keep ‘Janna walked so fast that no one could keep up with her’

As Rijkhoek shows, the combinations in question (indicated here again by bold face and italics) have the properties of parallel structure. Let us now have a closer look at what these properties are.

Relative clauses: Parallelism and partial reconstruction

131

3.3 Properties of Parallel Structure If we limit ourselves to single, linearized pairs, we can say that Parallel Structure has the following general form: (44)

General form of Parallel Structure: . . . [X] . . . [ω Y] . . . where X and Y are projections (construction-specific), ω is an operator (and, or, C, P, &:, BE, etc.) and where [X] is the anchor and [ω Y] the extension

Parallel structures have the following general properties (cf. Koster 1995, 2000): (45)

General properties of Parallel Structure: a. [ωY] can be ‘separated’ from [X] b. [ωY] c-commands [X] c. [X] does not necessarily c-command [ωY] d. Right Roof Constraint e. Preservation of parallelism f. Islandhood of extensions

These properties, very different from the properties of ‘movement’ constructions, apply to all parallel structures. I will mostly limit illustration to –prototypical– coordination with en (‘and’) and the relation of relative clauses with their head. The first property of (45), separation, has traditionally been known under the name ‘extraposition.’ I use examples from Dutch, an SOV language in which, as we saw, the separation is clearly marked by the verb (in capital): (46)

heb have

Jan John

a.

Ik I

en and

Marie Mary

b.

Ik heb Jan GEZIEN I have John seen ‘I saw John and Mary’

en and

GEZIEN seen Marie Mary

We see the same separation with relative clauses: (47)

heb have

de the

vrouw woman

a.

Ik I

die who

alles everything

b.

Ik heb de vrouw GEZIEN die I have the woman SEEN who ‘I saw the woman who knew everything’

wist knew

alles everything

GEZIEN seen wist knew

132

Jan Koster

The second property (45b) highlights an intriguing fact about separation: it cannot be brought about by normal leftward ‘movements,’ such as scrambling (48b), passive (48c) and topicalization (48d) (see also Büring and Hartmann 1997): (48) a. Ik heb de vrouw die alles wist GEZIEN I have the woman who everything knew seen b. *Ik heb de vrouw gisteren – die alles wist I have the woman yesterday who everything knew GEZIEN seen c. *De vrouw werd – die alles wist GEZIEN the woman was who everything knew seen d. *De vrouw heb ik – die alles wist nooit GEZIEN the woman have I who everything knew never seen We see the same pattern with coordination: (49)

Ik I

heb have

Jan John

en and

b. *Ik I

heb have

Jan John

gisteren yesterday

a.

c. *Jan John

werd was

–

en and

d. *Jan John

heb have

ik I

–

Marie Mary

GEZIEN seen –

Marie Mary en and

en and

Marie Mary

GEZIEN seen

GEZIEN seen

Marie Mary

nooit never

GEZIEN seen

Interestingly, all leftward ‘movements’ of the anchor are possible when the extension is in postverbal position: (50)

a.

Ik I

heb have

Jan John

GEZIEN seen

b.

Ik I

heb have

Jan John

gisteren yesterday

c.

Jan John

werd was

GEZIEN seen

d.

Jan John

heb have

ik I

nooit never

en and

en and

Marie Mary

GEZIEN seen

en and

Marie Mary

Marie Mary

GEZIEN seen

en and

Marie Mary

Relative clauses: Parallelism and partial reconstruction

133

The same with postverbal relative clauses: (51) a. Ik heb de vrouw GEZIEN die alles wist I have the woman seen who everything knew b. Ik heb de vrouw gisteren – GEZIEN die alles I have the woman yesterday seen who everything wist knew c. De vrouw werd – GEZIEN die alles wist the woman was seen who everything knew d. De vrouw heb ik – nooit GEZIEN die alles wist the woman have I never seen who everything knew This pattern is explained if we assume that the extension must c-command the anchor. Recall that we assumed earlier on that linearization adjoins the extension either directly to the anchor or to a phrase containing the anchor, like VP, IP or CP (see (31)) above). For coordination, that can lead to the following structures (assuming that the conjunction en (‘and’) projects to enP): (52)

a.

[DP John [enP en Mary]]

b.

[VP [VP John V ] [enP en Mary]]

In both cases, the extension en Mary c-commands the anchor John. Leftward ‘movements’ of the anchor, however, create structures in which the extension no longer c-commands the anchor. Consider scrambling over and adverb: (53)

[VP John [VP Adverb [VP [enP en Mary] V ]]]

Assuming that the extension en Mary must c-command the anchor John, this is no valid parallel structure. With the extension postverbally adjoined to the VP, as in (52b), there is no problem: (54)

[VP [VP John [VP Adv V ]] [enP en Mary]]

In this structure, the extension does c-command the anchor, ruling the configuration in as a valid parallel structure. In all parallel structures listed in (32), there is direct evidence that the extension can be adjoined to the VP (as in (54)), because the extension is moved along under VP fronting:

134 (55)

Jan Koster

[VP[VP

Jan John

gezien] seen

en and

Marie] Mary

heeft has

hij he

maar only

zelden rarely

In all cases I know of, the extension must c-command the anchor. Interestingly, the opposite is not true: the anchor does not necessarily c-command the extension. This was already briefly discussed above in connection with the illicit movements that would be required for the raising analysis of relative clauses to work: alles wist (56) Ik heb [PP met de vrouw] GESPROKEN die I have with the woman talked who everything knew Since the anchor (de vrouw) is in a PP, it does not c-command the extension (the relative clause). As concluded before, such structures make it practically impossible to maintain raising analysis of relative clauses. As expected, we see the same pattern with coordination, where the first conjunct – the anchor– can be inside a PP as well: (57)

Ik I

heb [PP have

met with

Jan] John

GESPROKEN talked

en and

Marie Mary

The facts in (56) and (57) illustrate a more general point, namely that normal bounding conditions (like Subjacency or some variant of it) do not apply to extraposition phenomena (Koster 1978a: 48ff.). With the exception of R-words, PPs are very strong islands for ‘movement’ in Dutch, which excludes an analysis of (56) and (57) in terms of a rightward movement rule like extraposition. There is, however, a locality principle that does apply to Parallel Structure, namely the principle known as the Right Roof Constraint. Anchor and extension cannot be separated by a clause boundary. Thus, a conjunct like en Marie in (58a) can appear at the right periphery of the clause that also contains the anchor, but it cannot occur in the next clause up (58b): (58)

a. [CP Dat that

ik I

met with

Jan GESPROKEN John talked

HEB en have and

ik I

met with

Jan GESPROKEN John talked

HEB] is duidelijk en have is clear and

Marie] Mary

is is

duidelijk clear b. *[CP Dat that Marie Mary

135

Relative clauses: Parallelism and partial reconstruction

As predicted, we find the same pattern with relative clauses and their heads: (59)

a.

b.

[CP

Dat that

ik I

alles everything

wist] knew

*[CP

ik I

Dat that

duidelijk clear

de the

met with

die who

is is

met with

vrouw woman

gesproken talked

heb have

die who

duidelijk clear de the

vrouw woman

alles everything

gesproken talked

heb] have

is is

wist knew

The next property of parallel structures is perhaps the most interesting because it involves cases of ungrammaticality totally unexpected under other theories. These cases, however, form striking confirmation of the theory pursued here. The property in question might be dubbed ‘preservation of parallelism, ‘also known as Kaan’s generalization (Kaan 1992). In the terminology of the current discussion it can be formulated as follows: (60)

Kaan’s generalization: A VP containing an extension cannot be fronted without fronting the anchor along

Here are some examples, with the traditional bracketing indicating that the DP Jan can be scrambled out of the VP (Den Besten and Webelhuth 1990): (61)

Ik I

heb have

Jan John

gisteren [VP yesterday

gezien seen

en and

Marie] Mary

Fronting the VP with the extension only and leaving the anchor behind, leads to totally unexpected, severe ungrammaticality: (62)

**[VP

Gezien seen

en and

Marie] Mary

heb have

ik I

Jan John

This is unexpected because, as shown by Den Besten and Webelhuth, normally DPs can be stranded under VP fronting when the DP is scrambled out of a minimal VP:

136 (63)

Jan Koster

[VP

Gezien] seen

heb have

ik I

Jan John

gisteren yesterday

niet not

Under the parallelism analysis, however, these facts are as expected. The traditional bracketing (as in (61)) is just wrong because the extension (en Marie) does not c-command the anchor (Jan), as would be required for a valid parallel structure. As discussed in connection with (54), a correct bracketing of the VP in (61) would be as follows: (64)

[VP [VP Jan [VP gisteren V ]] [enP en Marie]]

The V and the extension en Marie do not even form a constituent, explaining the severe ungrammaticality when they are fronted together (as in (62)). As expected again, we see the same pattern with relative clauses and their heads: (65)

a.

Ik I

heb have

b.

**[VP

de the

Gezien seen

vrouw [VP woman die who

gezien seen

alles everything

die who wist] knew

alles everything heb have

ik I

wist] knew de the

vrouw woman

The severe ungrammaticality of (65b) was always unexplained under a traditional extraposition theory but follows straightforwardly from the general conditions on parallel structure. The last property of parallel structure that I will discuss has to do with the strong islandhood of relative clauses: (66)

**Alles Everything

heb have

ik [DP I

de the

vrouw [CP woman

die who

–

wist]] knew

gezien seen

Traditionally, this was covered under Ross’s (1967) Complex NP Constraint and later on subsumed under Subjacency (Chomsky 1973). However, there never was a good explanation for the fact that after extraposition, the relative clause remains as opaque as before: (67)

**Alles Everything

heb have

ik [DP I

de the

vrouw] woman

gezien [CP seen

die who

–

wist} knew

This was unexpected because extraposition (conceived of as rightward movement) seems to bleed the configuration required for Subjacency to apply. Under

Relative clauses: Parallelism and partial reconstruction

137

the current analysis, however, no difference between (66) and (67) is expected because both involve valid parallel structures. We can subsume both cases now under whatever explains Ross’s (1967) Coordinate Structure Constraint and generalize it to all parallel structures.

3.4 Summary The conclusion so far is that ‘extraposition’ of relative clauses (across V in Dutch) cannot be accounted for by the traditional rightward movement rule of that name. Instead, the relation between head and extraposed relative clause was shown to have the properties of the kind of parallel construal also found in coordination and many other constructions (see (32)). Traditional extraposition does not work for these constructions because the defining locality principles of ‘movement’, like Subjacency, do not apply. Instead, the relevant locality principle appeared to be the Right Roof Constraint. Furthermore, a traditional extraposition analysis cannot account for the severe ungrammaticality of the sentences falling under Kaan’s generalization. The same can be said about the unexpected islandhood of extraposed sentences. Under the alternative, the analysis in terms of parallel structure, all facts are accounted for in a straightforward way. If the proposed alternative is correct, relative clauses ‘extraposed’ across V (CPs) are in the simplest cases related their head (DP) in the following way, with DP as the anchor, CP as the extension and C as the operator: (68)

. . . DP . . . V . . . [CP C [IP]] . . .

Not only rightward extraposition (of CP) is impossible, leftward movement of the DP (from the CP) is impossible as well, particularly because the DP can be embedded – in principle even fairly deeply – in a PP: (69)

. . . [PP P DP] . . . V . . . [CP C [IP]] . . .

Movement of a DP to a position inside a PP, where it is semantically selected by the P and does not c-command the gap of its position of origin, such a movement is illicit in all theories of possible movements. Although I consider the raising analysis of relative clauses untenable, it is not without defenders, even vis-à-vis the facts of Dutch. De Vries (2002) is a serious attempt to rescue raising for Dutch. According to this analysis, to which I cannot do full justice here, relative clauses involve two extended V projections

138

Jan Koster

(like AgrOP) related by the specification operator &P:. The second projection has Kaynean raising but undergoes deletion up to the point that only the relative clause remains. Apart from being overly complex, this analysis inherits most problems of raising analyses (like illicit movements). Most importantly, the relation between head and restrictive relative clause cannot be seen as specification, as it would make the wrong predictions about entailment. Thus, the Dutch sentence Ik heb niemand gezien die alles wist (‘I saw no one who knew everything’) does not entail Ik heb niemand gezien (‘I saw no one’), therefore violating the defining characteristic of specifications as discussed above.

4 Conclusion Raising analyses of relative clauses have their origin in the original ‘full reconstruction’ paradigm of transformational grammar. Although this program gradually collapsed since about 1970, its residues, like ‘move alpha’ and Internal Merge (with copying) tenaciously live on. Raising analyses of relative clauses have always been problematic but seem untenable when confronted with the facts of Dutch. Milder forms of full reconstruction, like the matching analyses proposed by Salzmann (2006) and others, fare somewhat better, but must ultimately be rejected as well. Such analyses have several problems with Kayne-style raising analyses in common, particularly the problem of impossible sources. For relative clauses, the limitation to partial reconstruction leads to remarkably traditional analyses. These are based on taking literally the idea that relative pronouns are pronouns. The anaphoric force of relative pronouns and their corresponding gaps suffices to provide all that is necessary for partial reconstruction. Consider a relative clause with idiom once more: (70)

[[DP the [NP headway]]i [CP [DP e ]i C [we made [DP e]i]]]

By what is called Wh-movement in the generative tradition, a sentence is turned into a property with the form: (71)

[CP [DP e ]i C [we made [DP e]i]]

By parallel construal, this property is interpreted as an extension of the anchor DP, the headway. The indices are not essential but convenient devices to indicate which elements share properties. The empty anaphors [DP e ]i mediate property sharing, guaranteeing that the object of made can share whatever properties it

Relative clauses: Parallelism and partial reconstruction

139

needs from the DP the headway. As the lexicon has an idiomatic entry ‘to make headway’, only a subpart of the DP, the NP headway is shared. Property sharing is subject to locality conditions and is usually partial, depending on the ‘needs’ of the elements that share properties. Nominative case of a subject, for instance, is neither shared with the reflexive in Hei saw himselfi, nor with the object slot in a passive like Hei was arrested [DP e ]i. More generally, we must conclude that the raising analysis is a residue of the total-reconstruction paradigm of transformationalism, which effectively died 40 years ago.

References Besten, Hans den and Gert Webelhuth. 1990. ‘Stranding.’ In Scrambling and Barriers, Guenther Grewendorf and Wolfgang Sternefeld, (eds.). Amsterdam: John Benjamins. Bianchi, Valentina. 2000. The raising analysis of relative clauses: A reply to Borseley. Linguistic Inquiry 31: 123–140. Blom, Alied and Saskia Daalder. 1977. Syntaktische Theorie en Taalbeschrijving. Muiderberg: Dick Coutinho. Borseley, Robert. 1997. Relative clauses and the theory of phrase structure. Linguistic Inquiry 28: 629–647. Brame, Michael. 1978. Base generated syntax. Seattle: Noit Amrofer. Bresnan, Joan. 2001. Lexical-Functional Syntax. Malden, Mass.: Blackwell. Büring, Daniel and Katharina Hartmann. 1997. Doing the right thing – extraposition as a movement rule. The Linguistic Review 14, 1–42. Cardoso, Adriana. 2010. Variation and change in the syntax of relative clauses: New evidence from Portuguese. PhD diss., University of Lisbon. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1970. Remarks on nominalization. In Readings in English Transformational Grammar, R. Jacobs and P. Rosenbaum (eds.), Waltham, Mass.: Ginn and Company, 184– 221. Chomsky, Noam. 1973. Conditions on transformations. In A Festschrift for Morris Halle, S. Anderson and P. Kiparsky (eds.), New York: Holt Rinehart and Winston, 232–286. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Dougherty, Ray. 1969. An interpretive theory of pronominal reference. Foundations of Language 5: 488–519. Emonds, Joseph. 1970. Root and structure preserving transformations. PhD diss., MIT. Guéron, Jacqueline. 1980. On the syntax and semantics of PP extraposition. Linguistic Inquiry 11, 637–678. Harris, Zellig. 1957. Co-occurrence and transformations in linguistic structure. Language 33: 283–340 Heringa, Herman. 2012. Appositional constructions. PhD diss., University of Groningen.

140

Jan Koster

Higgins, Roger. 1973. The pseudo-cleft construction in English. PhD diss., MIT [Published by Garland, New York, 1979] Jackendoff, Ray. 1969. Some Rules of Semantic Interpretation for English. PhD diss., MIT. Kaan, Edith. 1992. A minimalist approach to extraposition. MA Thesis, University of Groningen. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Klein, Maarten. 1977. Appositionele constructies in het Nederlands. PhD diss., University of Nijmegen. Kluck, Marlies. 2011. Sentence amalgamation. PhD diss., University of Groningen. Koster, Jan. 1978a. Locality Principles in Syntax. Dordrecht: Foris. Koster, Jan. 1978b. Why subject sentences don’t exist. In Recent Transformational Studies in European Languages, Samuel J. Keyser (ed.). Cambridge, Mass.: MIT Press. Koster, Jan. 1987. Domains and Dynasties: The Radical Autonomy of Syntax. Dordrecht: Foris. Koster, Jan. 1995. Extraposition as coordination. Ms., University of Groningen. Koster, Jan. 2000. Extraposition as parallel construal. Ms., Groningen: University of Groningen [see: http://www.let.rug.nl/koster/1999.htm]. Pollard, Carl and Ivan Sag. 1994. Head-driven Phrase Structure Grammar. Chicago: University of Chicago Press. Riemsdijk, Henk C. van. 1978. A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases. Dordrecht: Foris. Rijkhoek, Paulien. 1998. On degree phrases and result clauses. PhD diss., University of Groningen. Ross, John Robert. 1967. Constraints on variables in syntax. PhD diss., MIT. Ross, John Robert. 1969. Adjectives as noun phrases. In Modern Studies in English: Readings in Transformational Grammar, David Reibel and Sanford Schane (eds.). Englewood Cliffs, NJ.: Prentice-Hall. Salzmann, Martin. 2006. Resumptive prolepsis: A study in indirect A’- dependencies. PhD diss., University of Leiden. Smits, Rik. 1988. The relative and cleft constructions of the Germanic and Romance languages. PhD diss., Tilburg University. Stowell, Tim. 1981. Origins of phrase structure. PhD diss., MIT. Vergnaud, Jean-Roger. 1974. French relative clauses. PhD diss., MIT. Vries, Mark de. 2002. The syntax of relativization. PhD diss., University of Amsterdam. Wasow, Thomas. 1972. Anaphoric relations in English. Cambridge, Mass.: PhD dissertation MIT. Williams, Edwin. 1978. Across-the-board rule application. Linguistic Inquiry 9: 31–43.

Alan Langus and Marina Nespor

On the nature of word order regularities1 1 Introduction It has been argued that the human faculty of language is modular and that it is possible to identify different cognitive systems responsible for specific linguistic tasks (Chomsky, 2000a; Jackendoff, 1997; Fodor, 1983). The production and comprehension of language (either spoken or signed) require at least three taskspecific cognitive systems: the conceptual system (semantics) that provides and interprets the meaning of linguistic utterances; the sensory-motor system (phonology and phonetics) that produces and perceives the actual sounds and signs of language; and the computational system of grammar (syntax) that links meaning with sounds (or signs) by generating the structure of sentences (Hauser, Chomsky, & Fitch, 2002; Pinker & Jackendoff, 2005). The computational system has been the main focus of linguistic research because only in human language, with the possible exception of birdsong (cf. Langus, Petri, Nespor, & Scharff, 2013), the interface between the sensory-motor and the conceptual system must necessarily be mediated by the computational system of grammar (Chomsky, 1957; Chomsky & Lasnik, 1977). For mapping meaning to sound, the world’s languages use several structural regularities to signal the grammatical relations within a sentence, e.g. those expressing the relations of ‘who did what to whom’. Minimally these include phrase structure, recursion, word order and morphological marking (Pinker & Jackendoff, 2005). The relative importance of word order compared to morphological marking has been shown in recent findings (Erdocia et al., 2009) that suggest that even a language classified as non-configurational (Hale 1978, Chomsky 1981),2 like Basque (de Rijk 1972), has a basic word order that facilitates language processing. This may suggest that also other classical examples of non-configurational languages, such as Mohawk (Baker 1996) and Walpiri (Hale 1978), may have an underlying word order that is not discernable from their surface structure. Furthermore, evidence from language acquisition suggests 1 The research leading to these results has received funding from the European Research Council under the European Union’s Seventh Framework Programme (FP7/2007-2013) / ERC grant agreement n° 269502 (PASCAL). 2 Non-configurational languages have been claimed to lack syntactic configurations, to various degrees (Hale 1978). These include, but are not limited to, free word order, the use of syntactically discontinuous expressions, as well as null anaphora (Hale, 1983).

142

Alan Langus and Marina Nespor

that while word order may be acquired pre-lexically (Gervain, Nespor, Mazuka, Horie, & Mehler, 2008), grammatical devices such as morphology emerge relatively late (Hochmann, Endress, & Mehler, 2010; Hochmann & Mehler, 2012; Hochmann, 2013) and may require the knowledge of the basic word order of the language of exposure (Langus, Mazuka, & Nespor, submitted). These pieces of evidence strongly suggest a primary role of word order in conveying the grammatical relations within a sentence. There are six logically possible ways of arranging words in sentences according to their basic grammatical functions of Subject, Object and Verb (SVO, SOV, VSO, VOS, OVS, OSV), all of which have actually been attested – though to very different degrees – among the world languages (Dryer, 2005). While the Human Faculty of Language must be capable of generating and interpreting all the attested structures of the world’s languages, in the generative framework of grammar, there is disagreement about how this is achieved. For example, a fully representational approach to grammar, such as the Principles and Parameters theory (Chomsky, 1980), assumed that the structural regularities related to signaling the grammatical relations are fully represented in the computational system of grammar (Chomsky, 1957; Jackendoff, 1997; Pinker & Jackendoff, 2005). This implies that all the logically possible six word orders must be encoded in the computational system and the language-acquiring child would learn the word order of its native language through setting the word order parameters (e.g. Nespor, Shukla, Vijver, Avesani, Schraudolf, & Donati, 2008; Gervain, Nespor, Mazuka, Horie, & Mehler, 2008). If all the grammatical diversity is represented in the computational system, then, presumably, all the grammatical configurations attested among the world languages should be widely attested, and almost equally frequent. However, of the six logically possible word orders, SVO and SOV characterize the basic orders of the great majority of the world’s languages (86%: SOV and SVO) (Dryer, 1996), and orders in which the object precedes the subject are rarely attested. Some researchers have thus argued that OV and VO languages share many structural similarities (Lehmann, 1973, 1978; Vennemann, 1974, 1976) and that there is a general preference for word orders where the Subject precedes the Object (Greenberg, 1963; Greenberg, 1978). Of the remaining four word orders, in fact, VSO is the third more frequent, after SOV and SVO. Thus in order to explain the nature of the computational system, one does not only have to account for the scope of the grammatical diversity observed among the world languages, but also explain preferences for some structural configurations over others, and do so as economically as possible. In the Minimalist program, Chomsky (1995) initially argued that not all of the structural

On the nature of word order regularities

143

diversity must be represented in the computational system. Instead, the preference for certain structural regularities is evidence that there is only one basic structure (Kayne, 1994; Chomsky, 1995; Haider, 2000; Moro, 2000) and that all the surface variation observed among the world’s languages is derived from a handful of operations (e.g. movement, recursion and combinatorial processes) that map conceptual knowledge to the sensory-motor programs. However, considering that the SVO and SOV orders are almost equally frequent among the world languages, the frequency of occurrence of specific structural relations does not clearly point to a single preferred structure. Many accounts of the basic structure of the computational system therefore rely on the fact that some of the basic word orders of S, O and V co-occur more frequently with certain functional dependencies such as the order of adpositions, relative clauses, and adjectives relative to the noun. For example, Kayne (1994) has proposed that all branching is binary, all complements follow their head, specifiers precede it, and movement is to the left (see also Moro, 2000; Chomsky, 1995). In this perspective, the specifier-head-complement configuration corresponds to the SVO order and all other orders are derived from it through movement. Haider (1992, 1995), instead, has argued that all structures in natural languages are right-branching, that SOV is the basic word order and all other orders are derived from it through head-movement (see also Larson, 1988). There is thus no agreement on the preferred word order of the computational system, and on how different structural regularities and their occurrences observed among the world languages can help establish it. In fact, at its most extreme, the Minimalist program disregards all structural regularities, including those of word order Chomsky (2000b) assumes that the computational system of grammar generates structured expressions over an infinite range by using a single computation called Merge (narrow syntax). That is, all structural configurations are derived in the sensory-motor system (Chomsky, 1995). For example, given that the physical realization of language is sequential, the ordering of syntactic elements could only emerge because we linearize our linguistic utterances in the sensory-motor system when we externalize them. Since according to Chomsky (2011) the computational system of language is perfectly designed to interface with the Conceptual System but not so with the Sensory-Motor one, the two most frequent word orders would be explained in terms of modality conflicts, where the SOV order is preferred for signs and SVO for speech. There is no doubt that relegating all structural regularities observed among the world languages to the sensory-motor system does considerably simplify the innate aspects of the computational system. If nothing needs to be represented and everything is computed by Merge, the computational system could have

144

Alan Langus and Marina Nespor

evolved in a single sweep. However, because cross-linguistic structural regularities do exist, they also have to be encoded somewhere. The problem of complexity therefore appears to have been shifted from the computational system to the motor-system and the complexity in the Language Faculty as a whole remains unchanged. For example, if only one evolutionary change is necessary to yield Merge, in the case mentioned above – i.e. preference of word order depending on modality – at least two evolutionary changes would be necessary to explain the preferences for SVO order for speech and SOV for signs. Furthermore, if the computational system emerged from pre-existing computational capacities then it is difficult to see why the computational system should be perfectly suitable for the conceptual system, and not have its own structural preferences that might very well conflict with all the other cognitive systems. It is therefore difficult to see how the fully computational grammars are any more minimal in terms of the human ability for language than are the fully representational grammars such as those of the Principles and Parameters approach. In the following, we will take a new look at word order variation and provide an account for the two most common word orders observed among the world languages: SVO and SOV. We will first demonstrate that genetically predetermined and culturally un-modulated word order regularities do exist in cases in which language is acquired without a consistent linguistic input. We will then show that these word order regularities do not emerge from a single cognitive system but rather from the ways in which different cognitive systems involved in the human faculty of language interact. We will conclude by arguing that the simplest account for the systematic structural diversity observed among the vast majority of the world languages is that it is distributed across the different cognitive systems that comprise the language faculty.

2 Word order universals The majority of linguistic studies examining the basic word order of human language rely on the distribution and co-occurrence of structural regularities observed in the world languages. However, the diversity of the world languages, especially the differences between their grammars, is also a fertile ground for finding exceptions (Evans & Levinson, 2009). By focusing solely on the differences between languages, there are researchers who believe that human language neither was subject to biological adaptation nor is it pre-determined in the individual through language specific biological constraints. By using methods from evolutionary biology to track the historical processes of language evolution as

On the nature of word order regularities

145

languages split and diversify, Dunn et al (2011) argued that word order universals – such as the relative order of subject-verb, object-verb, relative clause-noun, adposition-noun – show correlated changes only within, but not across four large language families. For example, the changes in word order relating to the position of Subject and Object with respect to the Verb are claimed to be strongly correlated only in the Uto-Aztecan family (Dunn et al. 2011). Because these functional dependencies are at the core of the reconstruction of the basic universal structure in Kayne (1994), Chomsky (1995) as well as Haider (2000), the failure to observe strongly correlated changes across language families may appear to suggest that linguistic structure emerges only through lineage-specific cultural co-evolution (Dunn et al, 2011). It is thus worthwhile to spend some time to show that word order universals do actually exist. Contrary to Dunn et al. (2011) who failed to find strongly correlated crosslinguistic changes in word order relating to the position of Subject and Object with respect to the Verb, a myriad of studies in historical and comparative linguistics suggest that at least in the case of word order, change is primarily dictated by biological biases. Analyses of how languages change over time show that when word order changes independently of language contact, it is unidirectional from SOV to SVO (Givon, 1977; Newmeyer, 2000; for a recent overview see Gell-Mann & Ruhlen, 2011). The SOV order is likely to have been, in fact, much more common at earlier stages in history than it is today (Dryer, 2005). By reconstructing the word order changes in the seven linguistic macrofamilies Gell-Mann and Ruhlen (2011) argued that out of the seven primary nodes in the phylogenetic tree of world languages, five were originally SOV (Congo-Saharan, Indo-Pacific, Australian, Dene-Caucasian, Nostratic-Amerind), one (Khoisan) either SOV or SVO, and one (Austric) was SVO. This strongly suggests that the language from which all natural languages originate was SOV (Givón, 1977; Gell-Man & Ruhlen, 2011). Considering that the SOV and SVO orders are by far the most frequent orders among the world languages today, the findings of Gell-Man and Ruhlen (2011) also suggest that, while many languages have remained SOV, the majority of changes in word order must have been unidirectional from SOV to SVO. The only two putative exceptions to the unidirectional language internal change from SOV to SVO concern Mandarin Chinese (Li, 1977) and Niger-Congo (Heine, 1976). Li and Thomson (1974) have suggested that Mandarin has been undergoing a change from SVO to SOV through grammaticalization of serial-verb constructions. However, in contemporary Mandarin the SOV constructions are heavily marked and VO constructions vastly outnumber OV constructions, showing that SVO is the basic word order (Li, 1990). The direction of change is less clear in Niger Congo languages. Because many of the languages in the

146

Alan Langus and Marina Nespor

Niger-Congo family are SVO, but have traces of an OV order, the linguistic evidence available does not allow to clearly determine whether the Proto-NigerCongo was SOV and changed to SVO when modern languages emerged (Hyman, 1975; Givón, 1975) or whether it was SVO and its modern descendants acquired some OV characteristics when they split and diversified (Marchese, 1986; Heine, 1976). Leaving aside these only two putative exceptions, the overwhelming direction of change is from SOV to SVO (Givón, 1977; for a recent overview see van Gelderen, 2011), and the change from SVO to SOV results primarily from contact induced language infusion (Givón, 1977; Tai, 1976; Faarlund, 1990; Thomason & Kaufman, 1988; Heine & Kuteva, 2002; Van Gelderen, 2011). With no clear counter-evidence for unidirectional diachronic word order change from SOV to SVO, Dunn et al.’s (2011) failure to detect strong correlated changes in word order relating to the position of Subject and Object with respect to the Verb, suggests that the Bayesian phylogenetic method, as it is applied in Dunn et al., may be inadequate for explaining the ways in which the structure of languages changes. Several studies have applied the Bayesian phylogenetic method with apparent success for determining the date of divergence of languages by using the percentage of shared ‘cognates’ across languages both in the Indo-European (Gray & Atkinson, 2003) as well as in the Austronesian (Greenhill et al., 2010) language families. Dunn et al. use these estimated dates to hypothesize the correlated changes in linguistic structure. However, they have chosen to reduce the structural diversity observed among the world languages to individual parameters that can only vary in binary choices. For the basic word order regularities, Dunn et al. have chosen to compare the ordering of Subject and Object relative to the Verb by correlating two binary parameters (i.e., S before or after V; O before or after V). However, because the VO and OV languages are almost equally distributed, the Bayesian approach that selectively reduces linguistic structure to binary parameters, and neglects the possibility of unidirectional word order change, cannot detect a clear general bias for either the OV or the VO orders across languages. Furthermore, there is a strong crosslinguistic bias to have the Subject preceding the Object and the Verb (86% SVO and SOV), a correlation that could have been tested by comparing the position of Object and Verb relative to the Subject, but which Dunn et al. have chosen to ignore. Finally, in Dunn et al. (2011), the Bantu language family, that shows no diversity in many linguistic structures, like the ordering of Object and Verb, has been disregarded. Structural similarities, possibly the strongest cues to biological forces, are thus treated as uninformative. It seems clear that, before tackling undocumented linguistic changes in single languages and language families where historical record is scarce or non-existent, Dunn et al. should have checked the validity of the Bayesian phylogenetic method for investigating the change of

On the nature of word order regularities

147

linguistic structures by modeling the well documented language changes. Beginning with the word order change in the well-documented Indo-European language family could have been a way to eliminate what seem to be strong methodological biases for only detecting lineage-specific cultural evolution of linguistic structure. It should thus be clear that relying on the robustness of the Bayesian phylogenetic method for estimating the dates of divergence of languages (cf. Greenhill et al., 2010), is not sufficient to establish the robustness of the method when applied to historical changes in linguistic structure. Evidence from historical and comparative linguistics suggests strong biological biases for word order change and evolution. Evolutionary traces are clearly discernable in five of the seven macro-families regardless of the lineage-specific evolution of modern languages that suggest SOV as the basic order of the original proto-language. The strong trend in all the seven macro-families for a shift from SOV to SVO order, suggests that at least two strong biological biases are shaping linguistic structure in terms of word order over long periods of time: one keeping languages SOV and the other causing the change to the SVO order. The considerably less frequent – if at all existent – changes from SVO to SOV that appear to counteract these biological biases are not caused by lineage-specific (vertical) processes when languages split and diversify, but by horizontal processes of assimilation through language contact and infusion. If anything, within a language family, the cultural factors in word order change should result in uniformity, rather than diversity, as Evans and Levinson (2009) assume. Taken together, word order change suggests two forces that shape linguistic structure, one of which preserves the SOV languages and the other that causes them to change to SVO.

3 Atypical language acquisition The previous discussion suggests that cross-linguistic distributional information about the structural regularities may not be sufficient to definitively answer questions about the innate factors of human language. However, if word order universals exist, they should be discernable in the individual. One of the core implications of the culture-driven approaches is that the language acquiring child must discover the structure of its language of exposure through domain general mechanisms that constrain, rather than dictate, language learning (Tomasello, 2003; Evans & Levinson, 2009; Dunn et al. 2011). Were we thus to find a child who for one reason or another was deprived of linguistic input, we would expect it neither to communicate nor to show any of the structural

148

Alan Langus and Marina Nespor

characteristics typical of natural languages. This is, very clearly, not what happens. There are at least two well-attested cases of atypical language acquisition where children receive either fragmentary linguistic input, or are completely deprived of it. In both cases, they appear to develop a communication system that parallels to different degrees that of natural languages. Importantly, the structural regularities that have been the core of generative linguistics, especially systematic word order regularities, are among the core characteristics of these communication systems. The first case where consistent word order emerges in the absence of cultural transmission is constituted by Creole languages. Creoles are new fullyfledged languages that arise in communities where children are exposed to a pidgin – a rudimentary jargon created by people who must communicate without sharing a native language (Bickerton, 1981). The first generation of children that has a pidgin as their communication input, develop a fully-fledged language in a single sweep. Importantly, Creoles do not only have phrase structure, embedded clauses and recursion, but have all been observed to use a systematic SVO order in culturally and geographically distinct locations (Bickerton, 1984). While some researchers have argued that the structural coherence of Creole languages may be the result of children over-generalizing the pidgin input (Hudson Kam & Newport, 2005) that contains fragments of the grammars that the pidgin originated from (cf. DeGraf, 2005), the example of Berbice Dutch Creole that has a SVO order even though it emerged from two SOV languages: Dutch and Ijo (Kouwenberg, 1993; 1994), shows that Creole genesis must at least to a certain degree depend on a biological structural endowment. While the precise mechanisms that lead to the emergence of Creole languages are unclear, there are two prominent views concerning their genesis, both of which rely on infant’s innate abilities. On the one hand, some researchers believe that the invariant structural characteristics of different Creole languages, such as the SVO order, are evidence, in language emergence de novo, that children simply fall back on the default settings of Universal Grammar (Bickerton, 1984). On the other hand, it has also been argued that infants discover the structure of the original languages by analyzing the fragmented pidgin input (DeGraf, 2003; 2004). The latter view implies that infants use domain-general cognitive mechanisms, such as over-regularization of irregular linguistic input (see Hudson Kam & Newport, 2005), that yield a consistent Creole grammar. However, even if over-regularization enables children to achieve this remarkable feat, it is difficult to explain why the resulting word order would be SVO across different Creoles without this particular order being the basic order of natural languages, as the example of Ijo (SOV) strongly suggests.

On the nature of word order regularities

149

Creole languages are not the only example of consistent word order emerging in the absence of coherent linguistic input during language acquisition. A preference for a different word order – SOV – has been found in Homesign – a communication system developed by deaf children who are born to hearing parents and not exposed to any sign language. Homesigning children have been found to create their own gestural vocabulary to communicate with their parents and peers (Goldin-Meadow & Feldman, 1977). The Homesign systems in United States and China have been found to use the SOV order (Goldin-Meadow, 2005; Goldin-Meadow & Mylander, 1998), even though English and Chinese are both SVO languages. Interestingly, the Homesign systems that these children create are not determined by the gestures of their hearing parents (GoldinMeadow & Mylander, 1983). Instead, the vocabulary and the SOV structure are the children’s own invention (Goldin-Meadow, 2003), thus providing a strong argument for an innate bias for word order in the absence of linguistic exposure. Interestingly also the new sign languages that have emerged from homesign appear to be organized in the SOV order (Senghas, Coppola, Newport, & Supalla, 1997; Sandler, Meir, Padden, & Aronoff, 2005). Such cases have been observed in the school for the deafs in Nicaragua, where generations of Homesigning children failed to learn finger-spelling and through communicating with each other in Homesign developed what is today known as the Nicaraguan Sign Language (Shenghas et al., 1997). A similar situation emerged in a Bedouin village in Israel (Sandler et al., 2005) where a group of congenitally deaf individuals began communicating with Homesgin and through passing this Homesign system on to new generations developed the Al-Sayyid Bedouin Sign Language (Sandler, Meir, Padden & Aronoff, 2005). Both of these new sign languages are SOV despite the fact that none of the surrounding languages had this word order (Kegl, 2008), suggesting that the SOV order of Homesign can be assumed to be the basic word order of two new languages in geographically and culturally diverse locations. The children who give birth to Creoles and Homesign systems are remarkable in several ways. Unlike feral children such as Genie, who suffer severe cognitive deficits due to complete deprivation of linguistic input and social interaction (Curtiss, 1977), the children creating Creoles and Homesign have a normal cognitive development (Goldin-Meadow & Feldman, 1977). Considering that both ways of communicating in the absence of linguistic input appear to be the children’s own creation (Goldin-Meadow & Mylander, 1983; Goldin-Meadow, 2003), human language must at least in part emerge from the intent to communicate with our parents and peers. By relying on this instinct in the absence of linguistic input, children create communication systems that look in many ways like natural languages, with universal fixed word order – SOV for Homesign

150

Alan Langus and Marina Nespor

and SVO for Creoles – as the most notable of these language-like characteristics. Unmodulated by cultural processes and difficult to explain in terms of domain general learning mechanisms, these two cases of atypical language acquisition environments thus warrant caution towards theories that see the origins of linguistic structure solely in terms of cultural processes: children clearly bring to language acquisition a robust set of language specific innate biases that emerge even without linguistic input.

4 Cognitive systems and their structural preferences The evidence reviewed thus far clearly shows that there are strong biases for word order regularities that cannot be accounted for by cultural evolution or transmission. In fact, the evidence from language change combined with the cases of atypical language acquisition point to remarkable parallels in word order regularities that must be innate in nature. On the one hand, word order change suggests two forces that shape linguistic structure, one of which maintains the SOV languages, and the other that causes them to change to SVO. On the other hand, the two specific cases of atypical language acquisition point towards the same two orders: SOV in Homesign and SVO in Creole genesis. In the remainder of this section, we will consider the differences between Creoles and Homesigns, and show that while both of these orders are innate, they emerge from different cognitive systems. The differences between Creoles and Homesign, suggest that while Creoles are syntactically fully-fledged languages (Bickerton, 1984; Muysken, 1988), the nature of the gesture systems of homesigning children points to a pre-linguistic preference that emerges outside the computational system of grammar (Langus & Nespor, 2010). The SOV order that is dominant in Homesign has also been found in normally hearing adults who are instructed to use only gestures to describe simple scenarios. For example, Gershkoff-Stowe and Goldin-Meadow (2002) showed that when English-speaking adults were asked to describe simple vignettes, they tended to gesture the gestures for Actors and Objects before the gestures of Actions. This Actor-Object-Action order parallels the SOV order found in Homesign. In fact, the SOV order has been found in the production of improvised gestures of normally hearing English (SVO), Chinese (SVO), Spanish (SVO) and Turkish (SOV) speaking adults (Gershkoff-Stowe & Goldin-Meadow, 2002; Goldin-Meadow, So, Ozyurek, & Mylander, 2008, Langus & Nespor, 2010). Because the SOV order is ungrammatical in SVO languages like Italian, English, Chinese

On the nature of word order regularities

151

and Spanish, it has been suggested that SOV is a natural order for describing simple events in the absence of language (Gershkoff-Stowe & Goldin-Meadow, 2002; Goldin-Meadow et al., 2008, Langus & Nespor, 2010). Unlike in language, where the mapping between signal and meaning has to necessarily be mediated by syntax, in improvised gestural communication the mapping between the signal (the gestures) and its meaning may be achieved without the intervening syntactic computations (Langus & Nespor 2010). Several studies with adult speakers learning a new language show that they do not abandon their native grammar (Odlin, 1989). For example, immigrant workers learning Dutch – a language with SOV order in subordinate clauses and SVO order in main clauses – tend to use the SVO order if their native language is Moroccan Arabic (SVO), and the SOV order if their native language is Turkish (SOV) (Jansen, Lalleman, & Muysken, 1981). The fact that normally hearing English (SVO), Chinese (SVO) and Spanish (SVO) speaking adults produced gesture strings in the SOV order and failed to transfer their native SVO order to gestures, suggests that they must have bypassed their native linguistic structures. This may suggest that it is possible to communicate simple events in a prelinguistic way, i.e. without relying on the computational system of grammar. To test whether the computational system mediates improvised gesturing and whether the gestural utterances have any internal language-like hierarchical organization of constituents, Langus and Nespor (2010) asked Italian- and Turkishspeaking adults to produce more complex gesture strings. When using their native language, Italian (SVO) and Turkish (SOV) speaking adults described complex vignettes with sentences that contained both a main clause and a subordinate clause: i.e. the Turkish speaking participants produced spoken sentences where the subordinate occurred before the Verb of the main clauses, as in Turkish [Adam çocuga [kızın balık yakaladığını] anlatır] (equivalent in English to [man child-to [girl fish catches] tells]); and the Italian speaking participants produced sentences where the subordinate clause followed the Object of the main clause, as in Italian [l’uomo dice al bambino [che la ragazza prende il pesce]] (equivalent in English to [the man tells to the child [that the girl catches fish]]. However, when asked to gesture, gestural communication failed to trigger SOV language-like constructions when describing more complex vignettes in the same participants. Neither Italian- nor Turkish-speaking adults produced even a single gesture-string that conformed to the structure of complex sentences typical of SOV languages, like Turkish. Both Italian and Turkish speakers gestured the main clauses before the subordinate clauses – a construction typical of SVO but not of SOV languages. Importantly, these results do not support the proposal of Haider (2000) according to which the basic structure of human language is SOV and all

152

Alan Langus and Marina Nespor

branching is to the right (as in SVO languages). Were it so, we would have expected the subordinate clauses to follow the Indirect Object, not the Verb, of the main clause. Instead, the results of Langus and Nespor (2010) suggest that improvised gestures lack the internal hierarchical structure of natural languages: participants thus had to break the main clause and the subordinate clause into two separate and syntactically unrelated gestural utterances. This shows that the SOV order in improvised gesturing does not generalize to more complex SOV language-like constructions and it thus does not instantiate the typical syntactic hierarchical organization of constituents. Instead, improvised gesturing relies on the direct link between the conceptual and the sensory-motor systems without the intervention of the computational system of grammar. That the two most prominent orders among the world languages are not caused by simple modality differences while linearizing linguistic utterances is also evident in the fact that the SVO order does not emerge from the sensorymotor system as Chomsky (2011) claims. Langus and Nespor (2010) compared not only the production, but also the comprehension of simple gestural and spoken utterances in Italian and Turkish speaking adults. Both gestured and spoken utterances occurred in all the six possible logical orders. Participants’ reaction times showed that when perceiving gesture utterances, both Italian and Turkish speaking adults were fastest on the SOV order and that, in general, OV orders elicited shorter reaction times than VO orders. Interestingly, the opposite pattern emerged when participants perceived artificially synthesized and prosodically flat word strings in their native language. Both groups were faster in their native orders, showing that they used the computational system of grammar to interpret the spoken utterances. However, when we consider all six orders together, both Italian and Turkish-speaking adults show on average shorter reaction times with word orders where Subject is in the initial position and with word orders where the Verb precedes the Object (VO). In order to account for the preferences for OV in gestures and VO in spoken language, one could assume that the two orders are just effects caused by using either the manual or the oral modality (Chomsky, 2011). According to this view both orders would emerge in the sensory-motor system when one linearizes utterances while externalizing them. However, the results of Langus and Nespor (2010), that showed the preference for SOV in gestures, and SVO in spoken language emerges also in comprehension, suggest that the ordering regularities do not emerge due to linearization of linguistic utterances because linearization can only enforce its preferred structure in production. Furthermore, GershkoffStowe and Goldin-Meadow (2002) showed that the same Agent-Patient-Action order that prevails in improvised gestures emerges also when participants have to stack together transparencies depicting individual constituents, a task that is

On the nature of word order regularities

153

clearly non-linguistic in nature. The results demonstrate that the SOV order is generated independently of the constraints of the specific modality. Thus, while improvised gesturing relies on the direct interaction between the sensory-motor and the conceptual system, language must additionally make use of the computational system. These two studies thus strongly suggest that the difference between the SOV and SVO orders emerges because they are preferred by different cognitive systems: SVO by the computational system and SOV by the direct interaction of the sensory-motor and the conceptual systems. Since the SOV order cannot be a simple modality effect, there is no reason to believe that it emerges from the sensory-motor system alone. Gentner and Boroditsky (2009) have argued that relational terms – such as verbs – require the presence of the entities they link – such as nouns, suggesting that the SOV order may originate from the requirements imposed by semantic relations in the conceptual system of grammar. Along the same line, Bickerton (1992) has argued that while most nouns have concrete counterparts in the environment, the correspondence of a verb to an action is considerably more vague and therefore more abstract. This is also visible in language acquisition where nouns are also acquired more readily than verbs (c.f. Gillette et al., 1999). The concepts that nouns represent may therefore be more accessible than concepts pertaining to verbs (Bock & Warren, 1985), causing the conceptual system to feed the nouns that correspond to Subject and Object into the computational system in language, and to the sensory-motor system (in improvised gesturing), before the verb, thus causing a preference for a Noun-Noun-Verb order. It must be noted that proposals concerning the preferences for certain linguistic structures over others in the computational system of grammar have been somewhat controversial. For example, it has been argued that recursive structures are easier to understand and process in SVO languages like English and Italian, characterized by rightward embedding, than they are in SOV languages like Japanese and Turkish, characterized by center embedding (Frazier & Rayner, 1988). However, it has also been shown that Japanese-speakers can disambiguate multiple centre-embedded clauses without any apparent difficulty (Mazuka et al., 1989). Thus the preference for one order over the other in the computational system of grammar does not emerge from the inability of the system to process certain syntactic structures. The preferences for the SVO order and in general for right-branching syntactic structures may actually arise from the optimality with which they are processed in the computational system of grammar. For example, Hawkins (1994) has noted that left-branching languages are likely to violate the branching direction with syntactically heavy embedded clauses, which are often postposed to the right. Because this construction – where subordinate clauses follow the main clauses – is typical of SVO languages, it has been claimed that there is a performance advantage for the SVO

154

Alan Langus and Marina Nespor

order. The speech comprehension results of Langus and Nespor (2010) are the first experimental evidence of cross-linguistic preferences for one relative order of verb and object over the other, and show that these are even more fine tuned than previously thought: participants show a preference for Verb–Object orders even with simple artificially synthesized three-word strings in their native language, independently on the language’s word order of the participants.

5 The struggle for structure In terms of the nature of the human language faculty, the discussion above suggests that the structural diversity observed in the world’s languages does not emerge from the computational system of grammar alone, as suggested by fully representational approaches to grammar (Chomsky, 1980). The computational system responsible for generating and interpreting unambiguous structures prefers Verb–Object orders, and is possibly limited to the SVO order (Chomsky, 1995; Kayne, 1994; Moro, 2000). This implies that all the alternative grammatical configurations must originate from elsewhere in the language faculty. For example, the SOV order – that is at least as widespread as the SVO order in the world’s languages – emerges from the direct interaction between the sensorymotor and the conceptual system (Langus & Nespor 2010). It is therefore likely, that also other word orders, may originate outside the computational system of grammar. According to such a view, the structural diversity observed among the world’s languages may be the result of a struggle between different individual cognitive systems and their interactions trying to impose their preferred structure on human language. The idea that the different individual cognitive systems that comprise the Language Faculty are in conflict is found also within the Minimalist tradition. Chomsky (2000; 2011) has argued that all the difficulties in perceiving and parsing filler-gap sentences, island constructions, multiple embedding and garden-path sentences, emerge from the generative processes functioning without considerations for the parser. According to Chomsky, garden-path sentences such as “the old man the boat” are syntactically well-formed, but require multiple attempts at parsing because both syntactic and conceptual processes do not consider the possibility that man could be interpreted by the parser as a noun (the old man) rather than the verb (to man), as we will further discuss in the next section. However, the mapping between syntax and semantics is not straightforward either. In sentences such as “Everybody is to judge for themselves” the semantic clarity can even cause speakers to form syntactically incorrect sentences where

On the nature of word order regularities

155

the grammatical number of everybody and themselves disagree. Jane Austen, for example, has used this mismatch between singular indefinite pronouns and plural “they”, “their”, “them”, “themselves” 75 times in her six novels (for more examples see http://www.pemberley.com/janeinfo/austheir.html). This suggests, contrary to Chomsky (2011), that a conflict between communicative and computational efficiency may not be limited to the interaction between the sensorymotor and the computational system. One cognitive system can of course also impose its preferences on another system without any conscious recollection of conflict. Endress and Hauser (2010) taught English-speaking participants simple repetition based grammars (Marcus et al. 1999) that are readily learned by newborns (Gervain, Macagno, Cogoi, Peña, & Mehler, 2008) and rats (Murphy et al., 2008) and are therefore considered as perceptual primitive in speech perception (Endress, Nespor, & Mehler, 2009). However, when participants listened to three-word sequences that either started or ended with two words from the same syntactic category (e.g., AAB noun–noun–verb and verb–verb–noun or ABB noun-verb-verb and verb-noun-noun), participants only learned the repetition patters consistent with syntactically possible structures (AAB: Noun-Noun-Verb and AdjectiveAdjective-Noun; ABB: Verb-Noun-Noun and Noun-Adjective-Adjective) but not the syntactically impossible ones (AAB: VVN and AAV; ABB: NVV and VAA). Importantly, participants failed to learn these grammatically illegal repetition patterns over syntactic categories, even when explicitly instructed to look for them. When hearing a sequence of nouns and verbs, the computational system enforces an interpretation and, as a result, listeners fail to perceive the simpler pattern of repetitions preferred by the sensory-motor system. We do in fact find many more examples of imperfect interactions. For example, electrophysiological evidence has demonstrated that semantic factors can influence unambiguous syntactic interpretation (for an overview see Kuperberg, 2007). In sentences such as “The novelist that the movie inspired . . .”, the verb “inspired” assigns a thematic role of Agent to the inanimate noun (movie) and the role of Theme to the animate noun (novelist), creating a measurable conflict with the thematic roles more commonly taken by animate (Agent) and inanimate (Theme) nouns (Weckerly & Kutas, 1999; Traxler et al., 2002; Caplan & Chen, 2006; Chen et al., 2006). This suggests that the individual cognitive systems compute their outputs opportunistically and incrementally (Jackendoff, 2007; Kuperberg, 2007), creating an indefinite number of possible conflicts that can not only influence the processing of human language but may also shape its structure. This fits well with the assumption that the individual cognitive systems in the human faculty of language evolved through recycling pre-existing cognitive

156

Alan Langus and Marina Nespor

capabilities for the use of language (Hauser, Chomsky, & Fitch, 2002). Evolution does not produce novelties from scratch. Just like a leg is made from a fin, or a part of an ear from a piece of the jaw, also higher cognitive functions must have emerged from pre-existing brain mechanisms (Jacob, 1977). This kind of recycling can be seen in mental arithmetic that is evolutionarily too recent to have evolved its own cortical areas and therefore uses the pre-existing ones responsible for spatial attention (Knops, Thirion, Hubbard, Michel, and Dehaene, 2009). It is likely that also for human language, the three cognitive systems were recruited from pre-existing cognitive capacities. Contrary to Chomsky (2000b; 2004; 2011), it is therefore likely that none of the interactions in the Human Faculty of Language are perfectly designed. The preferences for SVO in the computational system and SOV in the direct interaction between the sensory-motor and conceptual systems, suggest that each cognitive system has its own structural preferences. The patchwork of the blind watchmaker (Dawkins, 1986) resulted in a human faculty of language where different structural preferences dynamically pull human language to different structural configurations and the actual changes in linguistic structure will be the result of the strength of one, or several coinciding preferences, winning over others.

6 Cues from prosody While the preferences between individual cognitive systems can shape linguistic structure over long periods of time, the acquiring infant must be able to converge on the order of its language of exposure. How can this be achieved? The signal of speech is not simply a sequence of syllables, but it also contains prosody characterized by changes in duration, intensity and pitch (Cutler, Dahan, & van Donselaar, 1997; Lehiste, 1970). Speakers can intentionally manipulate these acoustic cues to convey information about their emotional states (e.g. irony or sarcasm), to define the type of statement they are making (e.g. a question or a statement), to highlight certain elements over others (e.g. by contrasting them), or to contrast the meaning of words (e.g. word stress is phonemic in Italian, e.g. méta ‘aim’, metà ‘half’; phoneme duration is phonemic in Estonian, e.g. ma ‘I’, maa ‘land’; pitch is phonemic in tonal languages like Mandarin, e.g. mà ‘mumbler’, má ‘hemp’). Speakers also selectively insert pauses when uttering long sentences (e.g. a sentence as All the children of the many friends of Guinevere will get together to sing in the choir for the end of the year may be restructured into two Intonational Phrases in at least two different places – e.g. with a break either after Guinevere, or after together).

On the nature of word order regularities

157

However, there is also systematic structure in prosody that is hierarchical, and its phrasal constituents go from the phonological word up to the utterance (e.g. Beckman & Pierrehumbert, 1986; Hayes, 1989; Nespor & Vogel, 1986, 2007; Selkirk, 1984). Phrasal prosodic structure is automatically mapped from syntactic structure (Nespor & Vogel, 1986), and can disambiguate strings with an identical sequence of words, but different syntactic structures (e.g., [bad] [boys and girls] vs. [bad boys] [and girls]) (Lehiste, 1974; Lehiste et al., 1976; Price et al., 1991; Streeter, 1978). Thus syntax imposes constraints on prosodic structure (c.f. Shattuck-Hufnagel & Turk, 1996). In fact, garden-path sentences such as “the old man the boat” constitute a conflict between the communicative and computational efficiency only when they are read, because in spoken language phonological phrase boundaries (PP) signal the appropriate constituency (e.g. [the old]PP[man the boat]PP). Prosodic boundaries are also responsible for the fact that many sentences that are ambiguous when written, are not so when uttered (Nespor & Vogel 1986; Christophe et al., 2003, 2004; Millotte et al., 2007, 2008). This suggests that garden-path sentences, in spontaneous speech, may not occur at all. Importantly, there is evidence from the prosodic structure of natural languages that corroborates the claims of prosody is encoded in the cognitive system of human language. Nespor et al. (2008) compared the phrasal prosody of French (SVO) and Turkish (SOV) and showed that the iambic-trochaic law determines the physical realization of main prominence within phonological phrases that contain more than one word: if it is realized mainly through pitch and intensity, it is in a phonological phrase that is stress-initial and corresponds to a complement-head structure (OV); if it is realized mainly through duration, it is in a phonological phrase that is stress-final and corresponds to a head-complement structure (VO). Importantly, the different realization of the two types of prominence in phonological phrases was also found within a language. In German, where both orders of head and complement are found, the prominence is marked through pitch and intensity in OV constructions and through duration in VO constructions. This suggests that the two most prominent word orders among the world languages – SVO and SOV – are also correlated to specific acoustic realizations of phrasal rhythm, crucially not only across but also within languages. Importantly, it is difficult to envision how these prosodic regularities could emerge if word order were determined culturally and not encoded in our cognitive repertoire. Prosody does not only function as a cue for finding individual constituents in the speech stream (e.g., either words or phrases), but could also be a viable cue for bootstrapping into the hierarchical syntactic structure (i.e., finding constituents at different levels) (see Langus et al. 2011 for an overview). Gerken

158

Alan Langus and Marina Nespor

(1994) presented 9-month-old infants sentences where: (a) Phonological Phrase boundaries either signal major syntactic boundaries (e.g. in My neighbor/never walks his dog, where a Phonological Phrase boundary signals the major syntactic boundary between the Subject Noun Phrase and the Verb Phrase); and (b) sentences where Phonological Phrase boundaries fail to cue major syntactic boundaries (e.g. in He never walks his dog, where there is no Phonological Phrase boundary between the pronominal Subject and the Verb Phrase even though this corresponds to a major syntactic boundary). Importantly, pauses were inserted either after the Subject (a major syntactic boundary) or after the verb (a minor syntactic boundary) in all sentences. Nine-month-old infants showed longer looking times to sentences where pauses were inserted at major syntactic boundaries that were signaled by prosodic boundaries (sentences in (a)), but no differences were found when pauses were inserted at syntactic boundaries that were not signaled by prosodic boundaries (sentences in (b)). This shows that 9month-old infants follow prosodic constituents over syntactic ones. Thus these results suggest that, in early acquisition, prosody provides an important cue for understanding the hierarchical structure of the speech stream – even if the prosodic grouping principles do not perfectly predict the syntactic structure from which they are derived. The different realizations of phrasal stress could thus be cues exploited by infants to set the relative order of heads and complements – an iambic rhythm in phrasal stress signals that complements follow their heads; a trochaic rhythm signals that heads follow their complements (Nespor et al., 2008). Bion, Benavides, and Nespor (2011) showed that 7-month-old Italian infants segment sequences of syllables alternating in pitch into trochaic units with high pitch on the first syllable. The trochaic preference for sequences alternating in high and low pitch has also been found with English infants (Jusczyk, Cutler, & Redanz, 1993; Thiessen & Saffran, 2003). However, sensitivity to duration as a possible cue for iambic grouping is not found in 7-month-olds (Bion et al., 2011). These findings suggest that the trochaic preference could enable pre-lexical infants to discover the ordering of the SOV order from linguistic input during language acquisition. Considering that the SVO order is the preferred order of the computational system (Langus & Nespor, 2010) and may be the basic construction of all human languages (Kayne, 1994; Chomsky, 1995), the failure of young infants to group syllables according to the iambic preference, may suggest that they do not need to set the SVO order from the signal because it is the default setting for human language (Kayne, 1994; Chomsky, 1995). They do, however, need to set the SOV order, and their sensitivity to trochaic grouping might aid them very early on in this task.

On the nature of word order regularities

159

7 Conclusions In conclusion, the picture that emerges from the above considerations suggests that the Human Faculty of Language is structurally much simpler than previously thought: the grammatical diversity observed among the world languages does not have to be genetically encoded into the structure of the computational system of grammar, but rather, it emerges from the interaction between the individual cognitive systems that make up the Language Faculty. In fact, the preceding discussion has assumed that the Human Faculty of Language is modular in the broadest sense of the term, i.e. on the level of individual cognitive systems responsible for the sounds and signs of language (phonology), the meaning of utterances (semantics) and the structure of words and sentences (morphosyntax). However, it is possible that encapsulated and modular processing occurs also within these individual cognitive systems. For example, within phonology, consonants appear to aide word processing whereas vowels serve primarily for syntactic processing (Nespor, Peña & Mehler, 2003; Bonatti, Peña, Nespor & Mehler, 2005, 2007; Hochmann et al., 2011). Within semantics, a line can be drawn between word classes such as nouns and verbs (Bickerton 1981; 1992; Jakcendoff, 1992). In other words, all parts of grammar that can be identified in terms of underlying cognitive processes are possible sources of conflicting preferences and may thus also trigger the grammatical diversity observed among the world’s languages. The idea that grammatical diversity can emerge from the conflicts between the specific preferences of individual cognitive systems, or even individual cognitive processes, provides concrete suggestions for further research that will have to be empirically validated. For example, how do the remaining four (of the six logically possible orders) emerge? Will we be able to explain all the systematic grammatical diversity in terms of the conflicts between modules in the Language Faculty or of different cognitive systems imposing their preference? How can the strength of the preferences of the individual cognitive systems and how they affect language change be established? While many questions still remain unanswered, the previous pages may remove some of the road-blocks on the path of linking linguistic structure to the human cognitive abilities.

References Beckman, M., & Pierrehumbert, J. 1986. Intonational structure in Japanese and English. Phonology Yearbook, 3, 15–70. Bickerton, D. 1981. Roots of Language. Karoma Publishers.

160

Alan Langus and Marina Nespor

Bickerton, D. 1984. The language bioprogram hypothesis. Behavioral and Brain Sciences, 7, 173–221. Bickerton, D. 1992. Language and Species. Chicago University Press. Bion, R. A. H., Benavides, S., & Nespor, M. 2011. Acoustic markers of prominence influence adults’ and infants’ memory of speech sequences. Language & Speech, 54, 123–140. Bock, J., & Warren, R. 1985. Conceptual accessibility and syntactic structure in sentence formulation. Cognition, 21, 47–67. Bonatti, L., Peña, M., Nespor, M., Mehler, J. 2005. Linguistic Constraints on Statistical Computations: The Role of Consonants and Vowels in Continuous Speech Processing. Psychological Science, 16, 451–459. Bonatti, L., Peña, M., Nespor, M., Mehler, J. 2007. On Consonants, Vowels, Chickens, and Eggs. Psychological Science, 18, 924–925. Caplan, D. & Chen, E. 2006. Using fMRI to discover cognitive operations. Cortex, 42, 393–395. Chen, E., West, W.C., Waters, G., & Caplan, D. 2006. Determinants of bold signal correlates of processing object-extracted relative clauses. Cortex, 42, 591–604. Chomsky, N. 1957. Syntactic Structures. Mouton: The Hague. Chomsky, N. 1980. Principles and parameters in syntactic theory. In N. L. Hornstein & D. Lightfoot (Eds.), Explanation in linguistics: The logical problem of language acquisition, London and New York: Longman, 32–75. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht. Foris. Chomsky, N. 1995. Minimalist Program. Cambridge: MIT Press. Chomsky, N. 2000a. Linguistics and brain science. In A. Marantz, Y. Miyashita, & W. O’Neil (Eds.), Image, Language, Brain. Cambridge, MA: MIT Press. Chomsky, N. 2000b. Minimalist inquiries: the framework. In R. Martin, D. Michaels, & J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT Press. Chomsky, N. 2011. Language and Other Cogniive Systems. What is special about language? Language Learning and Development, 7, 263–278. Chomsky, N., & Lasnik, H. 1977. Filters and control. Linguistic Inquiry, 8, 425–504. Christophe, A., Nespor, M., Guasti, M. T., & van Ooyen, B. 2003. Prosodic structure and syntactic acquisition: The case of the head-direction parameter. Developmental Science, 6, 211–220. Christophe, A., Peperkamp, S., Pallier, C., Block, E., & Mehler, J. 2004. Phonological Phrase boundaries constrain lexical access I. Adult data. Journal of Memory and Language, 51, 523–547. Curtiss, S. 1977. Genie: a psycholinguistic study of a modern-day “wild child”. Boston: Academic Press. Cutler, A., Dahan, D., & van Donselaar, W. 1997. Prosody in the comprehension of spoken language: A literature review. Language and Speech, 40, 141–201. Dawkins, R. 1986. The Blind Watchmaker. New York: W. W. Norton & Company. DeGraf, M. 2003. Against Creole Exceptionalism. Language, 79, 2. DeGraf, M. 2004. Against Creole Exceptionalism (redux). Language, 80, 4. DeGraf, M. 2004. Morphology and word order in ‘creolization and beyond. In Cinque, G. & Kayne, R. (eds.) The Oxford Handbook of Comparative Syntax, Oxford, Oxford University Press, 293–372. Dryer, Matthew S. 1996. “Word Order Typology.” In Handbook on Syntax, Vol. 2, edited by J. Jacobs. Walter de Gruyter Publishing, 1050–1065.

On the nature of word order regularities

161

Dryer, M. S. 2005. The order of subject, object and verb. In M. Haspelmath, M. S. Dryer, D. Gil, & B. Comrie (Eds.), The world atlas of language structures, Oxford, Oxford University Press, 330–333. Dunn, M., Greenhill, S.J., Levinson, S.C., & Gray, R.D. 2011. Evolved structure of language shows lineage-specific trends in word-order universals. Nature, 473, 79–82. Endress, A.D. & Hauser, M.D. 2009. Syntax-induced pattern deafness. Proceedings of the National Academy of Sciences of the USA, 106(49), 21001–21006. Endress, A.D., Nespor, M. & Mehler, J. 2009. Perceptual and memory constraints on language acquisition. Trends in Cognitive Sciences, 13(8), 348–353. Erdocia, K., Laka, I., Mestres-Missé, A., & Rodriguez-Fornells, A. 2009. Syntactic complexity and ambiguity resolution in a free word order language: Behavioral and electrophysiological evidences from Basque. Brain and Language, 109, 1–17. Evans, N., & Levinson, S. C. 2009. With diversity in mind: Freeing the language sciences from universal grammar [Author’s response]. Behavioral and Brain Sciences, 32(5), 472–484. Faarlund J.T. 1990. Syntactic and pragmatic principles as arguments in the interpretation of runic inscriptions. In Fisiak, J. (Ed). Historical Linguistics and Philology, Berlin: Mouton de Gruyter, Berlin, 165–186. Fodor, J. A. 1983. The Modularity of Mind. MA: Bradford Books. MIT Press. Frazier, L., & Rayner, K. 1988. Parametrizing the language processing system: Left- vs. rightbranching within and across languages. In J. Hawkins (Ed.), Explaining Language Universals. Oxford: Basil Blackwell. Gell-Mann, M. & Ruhlen, M. 2011. The origin and evolution of word order. Proceedings of the National Academy of Sciences of the USA, 108, 17290–17295. Gentner, D., & Boroditsky, L. 2009. Early acquisition of nouns and verbs: Evidence from Navajo. In V. C. Mueller Gathercole (Ed.), Routes to Language: Studies in honor of Melissa Bowerman, New York: Taylor & Francis, 5–32. Gerken, L. A., Jusczyk, P. W., & Mandel, D. R. 1994. When prosody fails to cue syntactic structure: Nine-month-old’s sensititivy to phonological vs. syntactic phrases. Cognition, 51, 237–265. Gershkoff-Stowe, L., & Goldin-Meadow, S. 2002. Is there a natural order for expressing semantic relations? Cognitive Psychology, 45, 375–412. Gervain, J. Nespor, M., Mazuka, R., Horie, R., Mehler J. 2008. Bootstrapping word order in prelexical infants: a Japanese-Italian cross-linguistic study. Cognitive Psychology, 57, 56–74. Gervain, J., Macagno, F., Cogoi, S., Peña, M., Mehler, J. 2008. The neonate brain detects speech structure. Proceedings of the National Academy of Sciences of the United States of America, 105(37), 14222–14227. Gillette, J., Gleitman, H., Gleitman, L., & Lederer, A. 1999. Human simulations of vocabulary learning. Cognition, 73, 135–176. Givón, T. 1977. The Drift from VSO to SVO in Biblical Hebrew: the Pragmatics of Tense-Aspect. In Li, C.N. (Ed.) Mechanisms of Syntactic Change, Austin: University of Texas Press, 184–254. Goldin-Meadow, S. 2003. The Resilience of Language. Psychology Press, New York. Goldin-Meadow, S. 2005. Watching language grow. Proceedings of the National Academy of Sciences of the United States of America, 102, 2271–2272. Goldin-Meadow, S., & Feldman, H. 1977. The development of language-like communication without a language model. Science, 197, 401–403. Goldin-Meadow, S., & Mylander, C. 1983. Gestural communication in deaf children: Non-effect of parental input on language development. Science, 221, 372–374.

162

Alan Langus and Marina Nespor

Goldin-Meadow, S., & Mylander, C. 1998. Spontaneous sign systems created by deaf children in two cultures. Nature, 391, 279–281. Goldin-Meadow, S., So, W. C., Ozyürek, A., & Mylander, C. 2008. The natural order of events: How speakers of different languages represent events nonverbally. Proceedings of the National Academy of Sciences of the United States of America, 105, 9163–9168. Gray, R.D. & Atkinson, Q.D. 2003. Language-tree divergence times support the Anatolian theory of Indo-European origin. Nature, 426, 435–439. Greenberg, J. H. 1963. Universals of Languages. Cambridge: MIT Press. Greenberg, J. H. 1978. Universals of Human Language. Syntax. Stanford: Stanford University Press. Greenhill, S.J., Drummond, A.J., & Gray, R.D. 2010. How Accurate and Robust Are the Phylogenetic Estimates of Austronesian Language Relationships? PLoS ONE 5(3): e9573. Haider, Hubert. 1992. ‘Branching & discharge,’ Arbeitspapiere des SFB 340 23, University of Stuttgart; revised version in Lexical Structure, ed. by Peter Coopmans and Jane Grimshaw. Haider, Hubert. 1993. Deutsche Syntax, generativ. Vorstudien zur Theorien einer projektiven Grammatik. Narr, Tübingen. Haider, H. 2000. OV is more basic than VO. In P. Svenonius (Ed.), The derivation of VO and OV, Amsterdam: Benjamins, 45–67. Hale, K. 1978. On the position of Walbiri in a typology of the base. ms. MIT. Hale, K. 1983. Walpiri and the grammar of non-configurational languages. Natural Language & Linguistic Theory, 1(1), 5–47. Heine, B. & Kuteva, T. 2002. Language Contact and Grammatical Change. Cambridge, Cambridge University Press. Hauser, M. D., Chomsky, N., & Fitch, T. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science, 298, 1569–1579. Hawkins, J. A. 1994. A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. Hayes, B. 1989. The prosodic hierarchy in meter. In P. Kiparsky & G. Youmans (Eds.), Phonetics and Phonology. Rhythm and Meter, San Diego: Academic Press, Vol. 1, 201–260. Hochmann, J-R. 2013. Word frequency, function words and the second Gavagai problem. Cognition, 128, 13–25. Hochmann, J-R., Benavides-Varela, S., Nespor, M., & Mehler, J. 2011. Vowels and Consonants in Early Language Acquisition. Developmental Science, 14, 1445–1458. Hochmann, J-R., Endress, A. D., & Mehler, J. 2010. Word frequency as a cue for identifying function words in infancy. Cognition, 115, 444–457. Hochmann, J-R., & Mehler, J. (2012). Recent findings about language acquisition. In M. PiattelliPalmarini, & R.C. Berwick (Eds.), Rich Languages from Poor Inputs (pp. 107–114). Oxford, United Kingdom: Oxford University Press. Hudson Kam, C.L., & Newport, E.L. 2005. Regularizing unpredictable variation: The roles of adult and child learners in language formation and change. Language Learning and Development, 1, 151–195. Jackendoff, R. 1997. The Architecture of the Language Faculty. Cambridge: MIT Press. Jackendoff, R. 1992. Languages of the Mind, Bradford: MIT Press, 1992. Jackendoff, R., 2007. A Parallel Architecture perspective on language processing. Brain Research, 1146, 2–22. Jacob, F. 1977. Evolution and tinkering. Science, 196, 1161–1166.

On the nature of word order regularities

163

Jansen, B., Lalleman, J., & Muysken, P. 1981. The alternation hypothesis: Acquisition of Dutch word order by Turkish and Moroccan foreign workers. Language Learning, 31, 315–336. Jusczyk, P. W., Cutler, A., & Redanz, N. 1993. Preference for the predominant stress pattern of English words. Child Development, 64, 675–687. Kayne, R. S. 1994. The Antisymmetry of Syntax. Cambridge: MIT Press. Kegl, J. 2008. The case of signed languages in the context of pidgin and creole studies. In S. Kowenberg & V. Singler (Eds.), The Handbook of Pidgin and Creole Studies. WileyBlackwell. Knops, A., Thirion, B., Hubbard, E. M., Michel, V., & Dehaene, S. 2009. Recruitment of an area involved in eye movements during mental arithmetic. Science, 324, 1583–1585. Kouwenberg, S. 1993. A Grammar of Berbice Dutch Creole, Berlin; New York: Mouton de Gruyter. Kouwenberg, S. 1994. “Berbice Dutch”. In Jacques Arends, Pieter Muysken & Norval Smith (Eds.) Pidgins and Creoles: An Introduction. John Benjamins. 233–243. Kuperberg, G.R. 2007. Neural mechanisms of language comprehension: Challenges to syntax. Brain Research, 1146, 23–49. Langus, A., J. Petri, M. Nespor, and C. Scharff. 2013. FoxP2 and deep homology in the evolution of birdsong and human language. In Botha, R. and M. Everaert (ed.) The Emergence of Language. Oxford University Press. 223–242. Langus, A., Mazuka, R., & Nespor, M. (submitted). Order or Morphology? Syntactic constraints on statistical learning. Langus, A. & Nespor, M. 2010. Cognitive systems struggling for order. Cognitive Psychology, 60:4, 291–318. Langus, A., Marchetto, E., Bion, A.H., & Nespor, M. 2011. Can prosody be used to discover hierarchical structure in continuous speech? Journal of Memory and Language, 66, 285– 306. Larson, R. 1988. On the double object construction.’ Linguistic Inquiry, 19, 335–391. Lehiste, I. 1970. Suprasegementals. Cambridge: MIT Press. Lehiste, I. 1974. Interaction between test word duration and the length of utterance. Ohio State University Working Papers in Linguistics, 17, 160–169. Lehiste, I., Olive, J. P., & Streeter, L. 1976. Role of duration in disambiguating syntactically ambiguous sentences. Journal of the Acoustical Society of America, 60, 1199–1202. Lehmann, W. P. 1973. A structural principle of language and its implications. Language, 49, 42– 66. Lehmann, W.P. 1975. Proto-Indo-European Syntax, Austin: University of Texas Press. Lehmann, W. P. 1978. The great underlying ground-plans. In W. P. Lehmann (Ed.), Syntactic typology, Austin: University of Texas Press, 3–55. Li, C. N. 1977. Mechanisms of Syntactic Change. Austin: University of Texas Press. Li, Y. H. A. 1990. Order and Constituency in Mandarin Chinese. Dordrecht: Kluwer Academic Publishers, 14–15. Li, C. N., & Thomson, A. 1974. An explanation of word order change SVO/SOV. Foundations of Language, 12, 201–214. Marcus, G. F., Vijayan, S., Bandi Rao, S., & Vishton, P. M. 1999. Rule learning in seven-monthold infants. Science, 283, 77–80. Mazuka, R., Itoh, K., Kiritani, S., Niwa, S., Ikejiru, K., & Naito, K. 1989. Processing of Japanese Garden Path, center-embedded, and multiply-left embedded sentences: Reading time data from an eye movement study. Annual Bulletin of the Research Institute of Logopedics and Phoniatrics, 23, 187–212.

164

Alan Langus and Marina Nespor

Millotte, S., Frauenfelder, U. H., & Christophe, A. 2007. Phrasal prosody constraints lexical access. AmLap – 13th Annual conference on architectures and mechanisms for language processing, Turku, Finland. Millotte, S., Rene, A., Wales, R., & Christophe, A. 2008. Phonological Phrase boundaries constrain the online syntactic analysis of spoken sentences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 43, 874–885. Moro, A. 2000. Dynamic antisymmetry. Linguistic inquiry monograph series, Cambridge: MIT Press. Murphy, R.A., Mondragon, E., & Murphy, V.A. 2008. Rule learning by rats. Science, 319, 1849– 1851. Muysken, P. 1988. Are creoles a special type of language? In F. J. Newmeyer (Ed.), Linguistics: The Cambridge survey, Cambridge: Cambridge University Press, 285–302. Nespor, M., M. Peña and J. Mehler. 2003. On the different roles of vowels and consonants in speech processing and language acquisition. Lingue e Linguaggio. 2. 221–247. Nespor, M., Shukla, M., van de Vijver, R., Avesani, C., Schraudolf, H., & Donati, C. 2008. Different phrasal prominence realizations in VO and OV languages. Lingue e Linguaggio, 2, 1–29. Nespor, M., & Vogel, I. 1986. Prosodic Phonology (1st ed.). Berlin: Mouton de Gruyter [Dordrecht, Foris]. Newmeyer, F. J. 2000. On the reconstruction of ‘Proto-world’ word order. In C. Knight, J. R. Hurford, & M. Studdert-Kennedy (Eds.), The evolutionary emergence of language: Social function and the origins of linguistic form. Cambridge: Cambridge University Press. Odlin, T. 1989. Language transfer: Cross-linguistic influence in language learning. Cambridge: University of Cambrdige Press. Pinker, S., & Jackendoff, R. 2005. The faculty of language: What’s special about it? Cognition, 95, 201–236. Price, P. J., Ostendorf, M., Shattuck-Hufnagel, S., & Fong, C. 1991. The use of prosody in syntactic disambiguation. Journal of the Acoustical Society of America, 90, 2956–2970. Sandler, W., Meir, I., Padden, C., & Aronoff, M. 2005. The emergence of grammar: Systematic structure in a new language. Proceedings of the National Academy of Sciences of the United States of America, 102, 2661–2665. Selkirk, E. 1984. Phonology and syntax: The relation between sound and structure. Cambridge: The MIT Press. Senghas, A., Coppola, M., Newport, E. L., & Supalla, T. 1997. Argument structure in Nicaraguan sign language: The emergence of grammatical devices. In E. Hughes & A. Greenhill (Eds.), Proceedings of the Boston university conference on language development, Somerville: Cascadilla Press, 550–561. Shattuck-Hufnagel, S., & Turk, A. E. 1996. A prosody tutorial for investigators of auditory sentence processing. Journal of Psycholinguistic Research, 25, 193–247. Steele, S. 1978. Word order variation: A typological study. In J. H. Greenberg, C. A. Ferguson, & E. A. Moravcsik (Eds.), Universals of human language. Syntax, Stanford: Stanford University Press, 4, 585–623. Streeter, L. A. 1978. Acoustic determinants of phrase boundary perception. Journal of the Acoustical Society of America, 64, 1582–1592. Tai, J.H.Y. 1976. On the change from SVO to SOV in Chinese. In Steever, S.B., Walker, C.A., & Mufwene, S.S. (Eds), Papers from the Parasession on Diachronic Syntax, Chicago: Chicago Linguist Society, 291–304.

On the nature of word order regularities

165

Thiessen, E. D., & Saffran, J. R. 2003. When cues collide: Use of stress and statistical cues to word boundaries by 7- to 9-month-old infants. Developmental Psychology, 39, 706–716. Thomson, S. & Kaufman, T. 1988. Language Contact, Creolization, and Genetic Linguistics. Berkley: University of California Press. Tomasello, M. 2003. Constructing a Language: A Usage-Based Theory of Language Acquisition. Harvard University Press. Traxler, M.J., Morris, R.K., Seely, R.E. 2002. Processing subject and object relative clauses: evidence from eye movements. Journal of Memory and Language, 47, 69–90. Van Gelderen, E. 2011. The Linguistic Cycle. Oxford University Press. Vennemann, T. 1973. Explanation in Syntax. In Kimball, J.P. (Ed.). Syntax and Semantics, New York: Seminar, 2, 1–50. Vennemann, T. 1974. Analogy in generative grammar: The origin of word order. Proceedings of the Eleventh International Congress of Linguists, 79–83. Vennemann, T. 1976. Categorial grammar and the order of meaningful elements. In A. Juilland (Ed.), Linguistic studies offered to Joseph Greenberg on the occasion of his sixtieth birthday, Saratoga: Anma Libri, 615–634. Weckerly, J., Kutas, M. 1999. An electrophysiological analysis of animacy effects in the processing of object relative sentences. Psychophysiology, 36, 559–570.

M. Rita Manzini

On the substantive primitives of morphosyntax and their parametrization: Northern Italian subject clitics 1 The general issue In a recent article, Evans and Levinson (2009) argue that language universals are a myth, in contrast with both typological and generative approaches to language, despite the fact that these are otherwise incompatible in fundamental respects. Generative grammar is mentalist, predicated on the idea that languages correspond to systems of mental representations/ computations. These have formal properties that can be modelled by axiomatic systems, built on a restricted set of substantive primitives and of primitive operations/ relations. A natural link between the notions of primitive of the system and that of universal is provided by what Chomsky (2002: 147) calls the Uniformity thesis, namely that language “is a system that is, as far as we know, essentially uniform. Nobody has found any genetic differences . . . since its emergence there has not been any significant evolution. It has stayed that way”. Given the Uniformity thesis, the notions of linguistic primitive (i.e. primitive of the axiomatic system) and of linguistic universal are potentially identified. Interestingly, conclusions comparable to the Uniformity thesis emerge from non-mentalist typological approaches; for instance Nichols (1992: 227) states that “languages from typologically very different areas have the same latent structural potential. . . this survey has uncovered no evidence that human language in general has changed since the earliest stage recoverable by the method used here”. The issue of universals and variation has been clouded over in the history of generative grammar by two logically distinct problems. One of them, i.e. the specifically linguistic nature of the relevant universals, has been considerably clarified by the introduction of minimalism. As discussed by Chomsky (2002, 2005), Hauser, Chomsky and Fitch (2002), Fitch, Hauser and Chomsky (2005), most of the concepts and principles covered by the umbrella of Universal Grammar may very well be shared with other cognitive faculties. The other relevant issue is that of innateness. The strict connection of innatism to mentalism in Chomsky’s work has meant that critics have generally lumped them together. Again, recent work shows that holding a mentalist non-innatist position is at least logically possible (Perfors, Tenenbaum and Regier 2011).

168

M. Rita Manzini

Given this rich conceptual articulation, it is hard to evaluate statements of Evans and Levinson (2009) such as the following (found in their Abstract): “Although there are significant recurrent patterns in organization, these are better explained as stable engineering solutions satisfying multiple design constraints, reflecting both cultural-historical factors and the constraints of human cognition”. As we just saw, the idea that “constraints of human cognition”, rather than specifically linguistic principles, may largely organize grammar is at the core of minimalism. So we are left with the claim that “cultural-historical factors” play a role in linguistic explanation, a central tenet of typologicalfunctionalist approaches, to which Evans and Levinson fully belong in this respect. We are interested in one particular aspect of this well-rehearsed debate – concerning the substantive primitives of grammar, i.e. its ‘categories’ or ‘features’, and the way they enter into patterns of morphosyntactic variation. Our data base of Romance (specifically Italo-Romance) dialects (Manzini and Savoia 2005, 2007, 2008, 2011a) prompts some preliminary remarks. Italian dialects provide a rich and articulated picture of variation that contrasts with that of other well studied languages such as English. The view we take is that dialects exemplify the kind of variation we expect in natural languages in the absence of external constraints. It is linguistic situations such as those in Britain that represent a somewhat misleading picture of variation, reflecting not only the internal shaping forces of language development, but also external mechanisms of social and political standardization. Similar conclusions are reached by Nichols (1992: 23) from her quite different perspective: “a residual zone or a set of residual zones will contain a good deal of the world’s possible linguistic diversity in microcosm, and both the existence of internal diversity and its actual profile are stable and obviously very natural situations. Diversity of a particular kind may even be regarded as the state to which a group of languages will naturally revert if left undisturbed. . . Spread zones, in contrast, are typically highly divergent from one another, but each is internally quite homogeneous. . . Just which language spreads in a spread zone is a matter of historical accident, and this historical accident can distort the statistical distribution of linguistic types in an area”. In our work, we have addressed in particular inflectional categories/ features. Standard generative treatments of inflectional morphology yield considerable opacity at the syntax- PF interface, leading to the postulation of an autonomous morphological component such as the Distributed Morphology of Halle and Marantz (1993). The latter runs against the grain of more than one minimalist postulate; in particular Late Insertion violates Inclusiveness, while morphological repair violates no backtracking. In our previous work we have

On the substantive primitives of morphosyntax and their parametrization

169

argued that reforming traditional categorizations is necessary and often sufficient to uphold bare minimalist architecture (projection from the lexicon, no backtracking) Our take on variation depends on the Lexical Parameterization Hypothesis (Manzini and Wexler 1987, cf. Borer 1984, Chomsky 1995) to the effect that parameters are associated with lexical entries, in the presence of an invariant computational component (on the latter see Boeckx 2011). We do not share an important assumption of much current theorizing (e.g. Distributed Morphology, but also the ‘silent’ categories approach of Kayne (2010)), namely that there is a fundamental distinction between the substantive lexicon (lexical categories proper) and the functional lexicon (functional categories). For instance, within the Distributed Morphology framework, Embick (2000:187) assumes a “distinction between the functional and lexical vocabularies of a language . . . functional categories merely instantiate sets of abstract syntacticosemantic features”. These abstract nodes yield a sort of universal structural organization with respect to which linguistic variation would appear to be mere flatus vocis. On the contrary, we pursue a unitary model of the lexicon – of the type traditionally associated with the substantive lexicon: there is a conceptual and grammatical space to be lexicalized and variation results from different partitions of that space, i.e. “lexicons are . . . ways of partitioning an abstract categorial space” (Manzini and Savoia 2011a: 7). The so-called functional domain is just like all other conceptual space; the distinction between functional (i.e. grammatical) contents and other concepts, to the extent that it can be defined, is at best an external one. We illustrate our general approach in relation to an apparently trivial variation problem, involving the lexicalization of subject clitic pronouns in Northern Italian varieties. We show the empirical inadequacy of views based on implicational generalizations (Renzi and Vanelli 1983) despite the fact that they are upheld by recent formal work (Cardinaletti and Repetti 2008). Empirical problems are not completely resolved under the Distributed Morphology view of Calabrese (2008b, to appear) either, involving markedness hierarchies and repairs. We argue that a simple Boolean organization of categorial (i.e. conceptual) space is best at modeling variation.

2 Case study: Subject clitic paradigms in Northern Italian varieties Subject clitics in Northern Italian dialects display complex alternations between specialized forms, highly syncretic forms, and zero slots both in proclisis and in

170

M. Rita Manzini

enclisis. We take both proclitics and enclitics to be subject clitics, following Manzini and Savoia (2002a, 2005, 2007), and more recently Cardinaletti and Repetti (2008), who present extensive arguments against the alternative view that enclitics are an inflectional paradigm. As in Manzini and Savoia (1997, 2005, 2007), Cardinaletti and Repetti (2008), we also adopt the view that varieties with only some subject clitic forms represent instances of partial pro-drop1.

2.1 Renzi and Vanelli’s (1983) generalizations; Cardinaletti and Repetti’s (2008) cartographic rendering In pioneering work, Renzi e Vanelli (1983) propose several implicational generalizations holding of Northern Italian subject clitics, including (1) and (2), which we reproduce in the rendering by Cardinaletti and Repetti (2008). (1)

(2)

a.

Generalization 1: If a variety has at least one subject clitic, it is 2sg.

b.

Generalization 2: If a variety has two subject clitics, they are 2sg and 3sg.

c.

Generalization 3: If a variety has three subject clitics, they are 2sg, 3sg, 3pl.

If interrogative sentences are formed via subject-inversion, a. the number of enclitic pronouns found in interrogative sentences is equal to or greater than the number of proclitic pronouns found in declarative sentences, and b.

the subject pronouns found in proclitic position are also found in enclitic position.

Renzi and Vanelli’s empirical base is relatively small (30 dialects), and a larger database brings out a few classes of systematic counterexamples. In particular Manzini and Savoia (2005: §2.3) have several varieties where 3rd person subject clitics are realized, but not the 2nd singular, for instance (3), violating the generalizations in (1)2. 1 By contrast, the large cartographic literature concerned with variation in Northern Italian dialects (e.g. Poletto 2000) follows earlier theories (e.g. Rizzi 1986) in taking subject clitic languages to be consistently pro-drop. 2 Other counterexamples include: Stroppo/Macra, Pradleves, Acceglio, S. Pietro Val Grana, Tuenno, Vermiglio.

On the substantive primitives of morphosyntax and their parametrization

(3) el/la

dɔrmi dɔrmes dɔrm dorˈmiΝ

171

Livo (Trentino)

dorˈmio i/le dɔrm ‘I sleep, etc.’ Things do not improve when it comes to enclisis-proclisis generalizations. A pattern representing a counterexample to (2), involves the lexicalization of 2nd person singular clitic t- in declaratives (proclisis) but not in questions (enclisis) (Manzini and Savoia 2005: §3.6). In a fairly large set of varieties, exemplified in (4), this may be argued to involve a phonological or morphological process of haplology (OCP-avoidance) since the verbal inflection for the 2nd singular is also –t3. But this is not the case in varieties of the type in (5), which present a sigmatic –s ending for the 2nd singular verb inflection4. In fact, the phenomenon is equally found in Tuscan varieties with a non-sigmatic, non-consonantal ending for the 2nd singular, as illustrated in (6)5. (4)

(5)

(6)

a.

ta you

dormet sleep.2sg

b.

dormet sleep.2sg ‘Are you sleeping?’

a.

te you

b.

nɔwa dɔrmεs where sleep.2sg ‘Where are you sleeping?’

a.

tu maɲɲawe you eat.impf.past.2sg ‘You were eating’

b.

ke mmaɲɲawe what eat.impf.past.2sg ‘What were you eating?’

dɔrmes sleep.2sg

Grumello (Lombardy)

Mulegns (Grisons)

Vagli (Tuscany)

3 Other examples include: S.Fedele Intelvi, S.Bartolomeo Cavargna, Premana, Borgo di Terzo, Cataeggio, Adrara, La Strozza Valle Imagna, Odolo, Passirano, Pozzaglio/ Cicognolo, Cevo. 4 Other examples include: Montereale, Cantoira. 5 Other examples include: Sillano, Dalli

172

M. Rita Manzini

What is more, in many varieties, specialized proclitics alternate with syncretic subject enclitics. Hence there are proclitic forms that do not appear in enclisis, violating at least (2b). For instance, in (7a) the specialized 3rd person singular proclitic is lost in enclisis, where a clitic syncretic with the 2nd plural, as in (7b), appears instead; comparison with declarative forms is provided in (7a’–b’)6. Similarly, in varieties of the type in (8) a rather rich proclitic paradigm alternates with an undifferentiated lɔ enclitic7. (7)

a.

u/la drom-ni he/she sleep-ClS ‘Does s/he sleep?’

a’.

u/la dro:m he/she sleep-ClS ‘S/he sleeps’

b.

a druˈmim-ni ClS sleep.1pl-ClS ‘Shall we sleep?’

b’.

a drumima ClS sleep.1pl ‘We sleep’

(8) ty a/i nu u i/la: ‘Am I

Felizzano (Piedmont)

dyrmiu-ˈlɔ Prali (Piedmont) dyrmie-ˈlɔ dyərmə-ˈlɔ dyərmən-ˈlɔ dyrmie-ˈlɔ dyərmən-ˈlɔ sleeping? etc.’

In the spirit of Evans and Levinson (2009) one could use dialectological data like (3)–(8) not only to deny the validity of (1)–(2), but more generally to cast doubts on the existence of implicational universals. Vice versa, much theoretical

6 Other varieties tabulated by Manzini and Savoia (2005: §3.6.4) which present an enclitic n form covering at least one 1/2P and one 3rd person slot include: Viguzzolo, Casei, Castellinaldo, Cortemilia, Garessio, Fontane. 7 Other varieties tabulated by Manzini and Savoia (2005: §3.6.4) which present an enclitic l- form covering one or more 1/2P slots include: Pamparato, Corsaglia, Mombercelli, Montaldo, Antona.

On the substantive primitives of morphosyntax and their parametrization

173

ingenuity has been deployed in the strong universalist tradition of generative grammar in order to show that suitably constrained formal models actually yield implicational universals as a result (e.g. Cinque 2005). This approach characterizes Cardinaletti and Repetti’s (2008) discussion of Renzi and Vanelli’s (1983) generalizations. In particular, Cardinaletti and Repetti (2008) argue that the implicational hierarchy 2nd singular < 3rd singular < 3rd plural that can be deduced from Renzi and Vanelli’s generalizations in (1) depends on a structural hierarchy of positions, as in (9). They propose that in (9) the 2nd singular position is licenced by verb movement to it. In turn, both the 3rd singular and the 2nd singular positions are licenced by verb movement to the 3rd singular, and so on. This means that no position can be licenced unless 2nd singular is; nor is it possible to licence, say, 2nd singular and 3rd plural to the exclusion of 3rd singular. Therefore 3rd singular can be lexicalized only if 2nd singular also is; it is not possible to lexicalize 2nd singular and 3rd plural to the exclusion of 3rd singular; and so on. (9)

[3pl

[3sg

[2sg

[TP]]]]

Examples like (3), showing that it is perfectly possible for 3rd person to be lexicalized when 2nd singular is not, represent a direct counterexample to the structural hierarchy in (9), or at least to Cardinaletti and Repetti’s (2008) construal of it8. If, on the other hand, we allow the hierarchy to vary from language to language, the analysis loses its explanatory edge over merely listing the forms available in a given language. In fact, it seems preferable to maintain a bare statement of the facts, i.e. subject clitic present in x person, absent in y person in language L – than to translate this into a more complex stipulation, i.e. verb movement triggered by x person, not by y person in L. Cardinaletti and Repetti also propose a structural derivation for Renzi and Vanelli’s generalizations in (2). Because by hypothesis, enclisis is created by movement of the verb to a sufficiently high position to leave all clitics to its right, all clitic positions in (9) will be licenced in enclisis, letting clitics surface. Hence enclitics are at least as many as proclitics (2a) and all proclitics are also enclitics (2b). Again the phenomena in (4)–(8) represent a direct counterexample not only to Renzi and Vanelli’s generalizations in (2), specifically to (2b), but also to their structural encoding by Cardinaletti and Repetti. 8 Cardinaletti and Repetti quote Manzini and Savoia (2002a) as sharing the view of Poletto (2000) that there is a hierarchy of subject clitic positions, where the (1/2)P position is lower than Num (i.e. pl) and N (i.e. 3rd sg.). This conclusion is abandoned in Manzini and Savoia (2002b) and subsequent works, where object clitics enter the picture and a single subject clitic position is recognized.

174

M. Rita Manzini

A more general point raised by this discussion is whether it is in fact expected that implicational universals should hold within the framework defined by generative grammar, and be derived by the model. It seems to us that especially in the minimalist, biolinguistic perspective (Chomsky 1995, 2005), we expect to find that natural languages attest all and only the patterns allowed by the Narrow Faculty of Language (Hauser, Chomsky and Fitch 2002) and by the conceptual and motor systems aggregated around it. Given the extremely elementary content of the FLN, a considerable variation space is presumably generated by the categories of the conceptual system and their realizations by the motor (externalization) system. In this sense, a picture of language variation like that provided by our dialectal data seems to fit predictions well. In other words, it is expected that generalizations like (1)–(2) do not hold and that the admissible combinations of referential properties and overt lexicalizations are many more than (1)–(2) allow for. This is not to say that all logically possible combinations are allowed; however we expect that instances where they are excluded are more difficult to find than typological work implies. In general, a mentalist approach is not consonant with the strong functionalist bias of typological tradition. Therefore, rather than finding ways around ‘exceptions’ to typological generalizations, we may welcome these exceptions as the proof that the underlying picture is altogether different.

2.2 Calabrese’s (to appear) morphological approach Another important tradition in approaching substantive universals and variation is reflected by the treatment of Northern Italian subject clitics by Calabrese (2008b, to appear). Calabrese takes as his empirical basis the 187 varieties (his count) reported in Manzini and Savoia (2005: §2.3). Manzini and Savoia point out that there is a close correspondence between instances of partial pro-drop and instances of syncretism. By and large subject clitics are absent for a given set of person and number forms iff a syncretic realization is attested for the same set. For instance, by Calabrese’s count, more than a third of the dialects in the corpus present a pattern with a syncretic subject clitic in the 1st singular, 1st and 2nd plural (65/187); correspondingly, subject clitic drop in the same persons is fairly popular (39/187)9. It is syncretisms that Calabrese sets out to 9 These dialects have specialized subject clitics in the 2nd singular and in the 3rd person, as in Renzi and Vanelli’s generalization (1c). Of the remaining dialects about 30/187 conform to the Renzi and Vanelli’s pattern (1b) and only 3/187 have only the 2nd singular clitic, as in (1a) (Calabrese’s count). So about 50/187 dialects are out of the picture.

On the substantive primitives of morphosyntax and their parametrization

175

account for. The connection that he establishes with partial pro-drop will become evident directly below10. Calabrese’s analysis is again based on a person hierarchy, namely 2sg < 3sg < 3pl < 1sg < 2pl < 1pl, closely mimicking the Renzi-Vanelli and CardinalettiRepetti hierarchies. For Calabrese, this hierarchy corresponds to a set of constraints, each of which blocks the realization of the relevant person forms, as in (10). For instance, the activation of constraint (10f) means that the feature cluster [+speak, +augm], i.e. 1st plural, is excluded. This in turn triggers morphological readjustment, in order to allow for lexicalization, yielding syncretism. For instance *[+speak, +augm] can be repaired by changing the value of the [augm] feature from positive to negative, yielding the wellformed combination [+speak, -augm]. This forms the basis for a syncretism between the 1st plural (i.e. the starred feature combination) and the 1st singular (i.e. its repair). Alternatively, the activation of a constraint can lead to ‘obliteration’, i.e. lack of the relevant lexicalization, hence to partial pro-drop. (10)

In the context [[AgrS ____ ] + V a. *[+part, -speak, –augm] b.

*[-part, –augm]

c.

*[-part, +augm]

d.

*[+speak, -augm]

e.

*[+part, -speak, +augm]

f.

*[+speak, +augm]

Like Cardinaletti and Repetti’s structure in (9), Calabrese’s system allows only for certain syncretisms or partial pro-drop patterns, since the constraints are forced to apply in the order given, from more marked to less marked, i.e. from bottom to top of the hierarchy in (10). Thus there can be no variety where the 3rd singular (10b) and the 1st plural (10f) are syncretic, or which have pro-drop in these persons only, to the exclusion of the forms that appear between them in the hierarchy. From an empirical point of view, by imputing the person hierarchy to the morphological, rather than to the syntactic level, Calabrese reaches a better match to the data – yet some problems remain. First, Calabrese recognizes that in enclisis the hierarchy of persons may be different, as witnessed by the drop of the 2nd singular in (4)–(6). For this reason, the hierarchy in (10) is restricted to proclitics; for enclitics, Calabrese hypothesizes that the hierarchy may need to 10 For Cardinaletti and Repetti (2008), syncretic proclitics are not subject clitics– hence vice versa syncretism reduces to an instance of partial pro-drop.

176

M. Rita Manzini

be partially reversed to 3sg < 2sg. Even allowing for this, extending the approach in (10) from proclitics to enclitics will prove less than straighforward. For instance, languages like (7) instantiate the syncretism between 3rd singular and 1st plural whose exclusion (in proclisis) Calabrese claims as an argument in favor of his approach. Even limiting ourselves to proclisis, varieties like Livo in (3) require some special stipulation, since they violate the hierarchy, instantiating missing forms for 2nd singular, which precedes 3rd person in the hierarchy. Calabrese also notes that his system does not deal with the proclitics of a variety like Prali in (8), since only the 1st singular is missing and all other forms are specialized – counter once more to his hierarchy. From a theoretical point of view, we already commented in section 1 on the fact that the Late Insertion and the morphological repairs of Distributed Morphology violate minimalist postulates such as Inclusiveness and no backtracking. It is possible that these hold in syntax and not in morphology, but the result is in any case an enrichment of grammar. Summing up, syntactic hierarchies of the Cardinaletti and Repetti type are too rigid to model the variation picture attested to by Northern Italian dialects. Morphological hierarchies of the type introduced by Calabrese provide a much tighter picture of the observed data. Yet a few aspects of the variation in subject clitic paradigms still escape him and can be covered only by extra stipulations. In other words, it is not obvious that the price paid in theoretical terms by (Calabrese’s version of) Distributed Morphology pays a worthy dividend when it comes to the modelling of variation. A possible diagnosis of what is amiss comes from the observation that Calabrese (to appear), Cardinaletti and Repetti (2008) and Renzi and Vanelli (1983) all share the assumption that a hierarchical arrangement of features (or positions) is involved in the realization of partial pro-drop and of syncretism in subject clitic paradigms. Hierarchies are at the core of the implicational approach that typological studies take to language universals. It is far from obvious that hierarchies are either the best available tool to describe language variation or that they should somehow be incorporated in (or derived by) formal models of linguistic competence, i.e. generative grammars. Our recent work (Manzini and Savoia 2005, 2007, 2011a) argues in fact that despite the descriptive richness that they encapsulate, hierarchies typically fail in describing language variation – where single varieties easily jump certain particular links of the hierarchy. In this spirit, Manzini and Savoia (2011a, 2011b) argue for a treatment of auxiliary choice according to person (have vs. be) that avoids hierarchies (cf. Loporcaro 2007, Legendre 2010, also Sorace 2000 for the opposite view). Similarly Manzini and Savoia (2010a, 2011a, 2011c) reject Case hierarchies (cf. Calabrese 1998, 2008a, also Caha 2009) in their approach to case syncretism. In both instances we argue that the facts can be better derived by discrete cate-

On the substantive primitives of morphosyntax and their parametrization

177

gorial splits, whose fundamental ranking is provided by the elementary Boolean logic of super- and sub-categories. In the next section, we apply the same general research strategy to the domain of Northern Italian subject clitics.

3 Manzini and Savoia (2005): spelling out the proposal Our general framework of assumptions, as implied by the discussion in sections 1–2, involves a classical approach to lexical entries, conceived of as mappings between an LF content and a PF one. From these we project morphosyntactic structures, in keeping with minimalist postulates. On grounds of simplicity of the theory (both formal and regarding the underlying ontology of language) we further commit ourselves to positively specified properties only, i.e. privative categories. The linguistic literature is ripe with observations as to the fact that 3rd person is in fact a non-person, and generative systems encode this typically with a negative feature value. Thus 3rd person is [-participant] for Calabrese (to appear) in (10). In underspecification systems (e.g. Harley and Ritter 2002), lack of specification for the feature ‘participant’ is effectively interpreted as a negative specification for that feature. Consider however what lexical evidence tells us. The non-syncretic, singular paradigm of object clitics of Italian is summarized in (11a–b). The 1st and 2nd person forms in (11a) are each characterized by a specialized lexical base m-/ t-, denoting ‘speaker’ and ‘hearer’. In turn the 3rd person forms in (11b) have a recognizable lexical base l- followed by nominal class inflections -o/ -a. As is well known, the same lexical base l- turns up as the determiner of nouns, in which case its referential value is definiteness, as in (11c); incidentally the nominal class endings –o/-a are the same seen on nouns (here zi- ‘uncle/ aunt’). Thus at the morphophonological interface, separate lexicalizations for ‘speaker’ m-, ‘hearer’ t- and ‘definiteness’ l- are instantiated11. (11)

a.

mi/ me/

ti you

b.

lo/ him/

la her

c.

lo zio/ the uncle/

la zia the aunt

11 In Harley and Ritter (2002, Appendix), several languages are listed as having suppletive (demonstrative) 3rd person forms. However the Romance languages are not among them.

178

M. Rita Manzini

On the interpretive interface, it is worth quoting Kratzer (2009: 221): “the alleged ‘‘3rd person’’ features are in fact gender features, a variety of descriptive feature . . . If [a descriptive feature] is to grow into a pronoun, it has to combine with a feature [def] that turns it into a definite description. If [def] is the familiar feature that can also be pronounced as a definite determiner in certain configurations, it should head its own functional projection, hence be a D . . . Descriptive features . . . are nominal, hence Ns”. Manzini and Savoia’s (2002b, 2005, 2007) categorization for so-called 3rd person pronouns, largely based on the morphological interface, is essentially identical to Kratzer’s, i.e. a D category for the Definiteness morphology (l- in Romance) embedding an N, i.e. nominal class category, for its inflections. In turn, a super-category participant can be defined for ‘speaker’ and hearer’ denotations, corresponding to the P (Person, i.e. 1st/2nd Person) of Manzini and Savoia (2005). In the proposed system, reference accrues to 3rd person forms not through their lack of P categorization, but through their positive D categorization. In other words, the logical space is partitioned not into P and not-P but into P and D. No empirical advantage is gained at the lexical or semantic interface by a [–participant] or [0participant] characterization. In this respect Kayne’s (2010) work provides an important cross-reference to ours, since it also depends on raising morphological features to syntactic (hence privative) categories (cf. Kayne (2000) on the analysis of pronouns, Manzini and Savoia (2010b, 2011d) for further discussion). Manzini and Savoia (2005: §2.3.1) provide two separate resumptive tables for subject proclitics. One table summarizes possible patterns of syncretism and partial pro-drop in the P paradigm. The other table summarizes the same patterns over the classical 6 persons paradigm, but keeping constant what turns out to be the prevalent P system, namely a specialized lexicalization of the 2nd singular and a syncretic or zero lexicalization of the 1st person and 2nd plural (cf. fn. 9). Due to space limitations, and since we are interested in covering both proclitics and enclitics, here we will concentrate on variation in the P paradigm. To further simplify the discussion we concentrate on alternations between P forms and non lexicalized forms, in other words on partial pro-drop, avoiding the discussion of syncretic patterns (but see section 3.2).

3.1 The core proposal The logical possibilities for combining four person denotations with two choices for lexicalization (1/2P vs. zero/ syncretic form) are sixteen. In the absence of further constraints we expect to find all of them. Manzini and Savoia (2005: §2.3.1) however tabulate 6 possible proclitic patterns with 1st/2nd person, which means that there is a considerable amount of missing combinatorial slots, as

On the substantive primitives of morphosyntax and their parametrization

179

shown in (12). In (12) the dialect naming each existing pattern is one exemplified in the present text or else the first one found in the relevant subgroup in Manzini and Savoia (2005: §2.3). The minus sign refers to the fact that the relevant P reference is not lexicalized or not by a specialized P form. (12) 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16.

Prali Sarre Càsola * * * * * Faeto Sillano * * Livo * * *

1st – – – – P P P P P P P P – – – –

2nd P P P P – – – – P P P P – – – –

4th P – – P P – – P P – – P – P – P

5th P – P – P – P – P – P – – P P –

Before considering the missing languages in (12) we briefly review the enclisis data, for which Manzini and Savoia (2005: §3.1, §3.13.2), report about one hundred paradigms. Enclitic paradigms are only about half the proclitic ones partially because several subject clitic varieties form questions without subject clitic inversion. Again crossing four persons with two choices (P vs. zero/non-P form), we obtain sixteen possible combinations. Sieving through the examples of Manzini and Savoia, we can find evidence for perhaps fourteen existing varieties12, as 12 In line 14 of table (12), we parenthesized La Strozza, because in this variety the 1st plural is lexicalized by an impersonal, hence by a 3rd person (εn dørm-ei lit: ‘one sleeps?’ for ‘Shall we sleep?’; cf. Italian si va? lit: ‘one goes?’ for ‘Are we going?’). The question therefore is whether there could be a comparable variety with a morphologically 1st plural P enclitic. One pattern, namely that in line 10, is not instantiated under partial pro-drop. We associate it with Comeglians where the plural (durˈmin-o ‘Are we sleeping?’, durˈmi:z-o ‘Are you sleeping?’ duarˈmin-o ‘Are they sleeping?’) involves a single syncretic –o enclitic lexicalization. Castellazzo in line 11 gives an idea of the decisions involved in segmentation. Manzini and Savoia (2005) analyze the 1st person singular dɾwɔm-ju ‘Am I sleeping?’ as involving a specialized P clitic; the superficially identical 3rd plural is analyzed as involving mesoclisis of the subject clitic i between the verb base and the inflectional ending –u, i.e. dɾwɔm-j-u ‘Are they sleeping?’.

180

M. Rita Manzini

shown in (13). Manzini and Savoia (2005) do not provide any grouping or tabulation of the data (as we just saw, there are no major gaps). We constructed the table in (13) by simply listing the first example for a given pattern found in their data set. (13) 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16.

Villa di Chiavenna Chioggia Colle S.Lucia (impf.) * Tuenno (impf.) Passirano * Vermiglio Barcis Comeglians Castellazzo Bormida Forni di Sotto Pozzaglio (La Strozza) Odolo Cataeggio

1st – – – – P P P P P P P P – – – –

2nd P P P P – – – – P P P P – – – –

4th P – – P P – – P P – – P – (P) – P

5th P – P – P – P – P – P – – P P –

The question is whether the variation in (12)–(13), including the gaps observed in particular in (12), can be made to follow from a simple set of parameters, namely choices open within UG. What we are looking for is something no less restrictive than implicational scales, yet flexible enough to yield all of the observed cross-linguistic variation, as well as enclisis-proclisis variation within the same language. We assume that parametric choices involve interface properties and their interaction with the universal rules/ principles component. In particular, parametric choices are effected by the rule/ principle of Externalization in (14) pairing a conceptual cluster with a motor realization. As far as we can tell (14) is the notion of externalization intended by Berwick and Chomsky (2011). (14)

Externalization Pair LF x with PF y

(14) is not subject to variation. Parameter settings are the interface choices that (14) effects. For example, a language may lexicalize (externalize) all P forms, as

On the substantive primitives of morphosyntax and their parametrization

181

in line 9 of tables (12)–(13) – or not externalize any of them, as in line 13. At the same time a language may externalize along a finer fault line, that between ‘speaker’ participant and ‘hearer’ participant. This may result in the externalization of just ‘speaker’ reference (cf. in particular line 8 of table (13)) or of ‘hearer’ reference (cf. in particular line 3). The ‘parameters’ interacting with (14) are therefore the categorial splits ‘speaker’ vs. ‘hearer’, P (1st/2nd person) vs. D (3rd person), etc. This simple picture may be sufficient to explain table (13), where more or less all logically possible values are attested, but it is not sufficient to explain Table (12), where languages with 2nd singular not lexicalized are excluded. In the terms of Manzini and Savoia (2011a), the problem is that there are not only ‘reversible’ parametric choices, i.e. choices yielding mirror-image languages (e.g. line 3 and line 8 of Table (13)), but also ‘irreversible’ ones. An example of an ‘irreversible’ parameter is the salience of speaker reference in (15) – or conversely the fact that “the ‘hearer’ is singled out by the lack of any referential property, including pragmatic salience, beyond mere anchoring in the discourse” (Manzini and Savoia 2011a: 212, 214). (15)

(pragmatic) salience of speaker reference

(15) is a parameter in the sense defined here, i.e. a categorial cut, but an intrinsically asymmetric one. It could equally well be expressed as a two-members hierarchy, namely (16) (cf. Noyer 1992). Vice versa, however, a minimal hierarchy of the type in (16) can always be stated as a categorial split. The latter is intrinsically restricted to a binary choice, while a hierarchy can in principle be expanded, and in fact generally is. If one wishes one could restrict hierarchies to minimal pairs – but the point is always the same, namely that there is no evidence for real ranking (pace Optimality Theory), i.e. ultimately for counting. (16)

1



This LF Cancellation feature allows an elegant lexical entry for the (always Realis) finite copulas: (46)

English Past Copula {were, +PL / was }, -M, +T, Φ

Since a copula has no inherent feature other than its category, the ICP (35) permits the position of be insideVP to be silent, and so by Economy it must be; see again note 22.

26 In this perspective, an additional label ±STATIVE used only in semantics is redundant.

302

Ludmila Veselovská and Joseph Emonds

(47)

We draw attention to a contrast between these copular forms followed by V = Ø and the I treated earlier such as would, could and do. In the latter cases, because lexical features of the Vs in VP are interpreted (either as Activities or States), the following V must be overt. Another question concerning AR of a feature F is when F’s Canonical Position must be empty. This appears to be the unmarked option if the two positions are both in the same extended projection. With the exception of do-support, empty Canonical Positions are required for all uses of AR focused on in this paper; hence the auxiliaries was/were in I cannot precede overt stative verbs.27 English and Czech number agreement and Czech case-marking alternatively realize categories inside different extended projections, e.g. [+PL] from DP appears under V in (47), so in these cases AR can then “double” the overt [D, +PL] in its Canonical Position. (ii) The appearance of have under I. A second English grammatical verb that can surface under I with the feature ±T is the stative verb have. Like be, have is an otherwise unmarked stative verb; but like do (and unlike be), have is transitive, that is, its lexical specification includes a context features __ as

27 According to this idea, number agreement between overt Ds and Ns inside a single DP as in (47) is a marked option. If some general condition for when AR allows doubling cannot be found, lexical entries of AR morphemes must stipulate whether they can double or not (Emonds 2000: Ch. 4).

The cross-linguistic homes of Mood and Tense

303

well as the LF cancellation feature Φ introduced in (44).28 If this Φ is alternatively realized under I, then have has no interpretation, and the ICP (35) allows its canonical position in VP to be empty. (48)

English stative have: { have/ had, +T }, V, Φ, ___

Since underlining notation permits but does not require AR, the features V and Φ in (48) can be either canonically realized under VP, giving rise to main verb behavior such as (49a), or alternatively realized under I with an empty V in the VP, yielding rather patterns as in (49b). (49)

a.

Sue doesn’t have much of a chance, does she? Does Ann have as many friends as Bill does?

b.

Sue hasn’t [V Ø] much of a chance, has she? Has Ann [V Ø] as many friends as Bill has?

Since the two alternatives involve exactly the same structures, namely I + VP with a single word have under either I or V, it appears that both are equally economic. This syntactic optionality seems realized in different English styles and idiolects.29 In addition, the English verb have has a further grammatical use with a VP complement, whose head V-en is an active perfect participle. We don’t propose a 28 In two lexical contexts, have replaces the main verb do and so has an Activity sense: (i) before DP objects in Grimshaw’s (1990) “Simple Event Nominals” and (ii) in a causative. (i) have a bite/ drink/ listen/ look/ nap/ pee/peek/ rest/ sniff/ snooze/ taste/ trip/ vacation/ walk Cf. *do a bite/ drink/ listen/ look/ nap/ pee/peek/ rest/ sniff/ snooze/ taste/ trip/ vacation/ walk (ii) Bill will soon have/ *do the kids wash the dishes. In these collocations, have like do is an Activity verb and lacks the cancellation feature Φ; i.e. its category V is interpreted as Activity in its canonical position. Consequently, in these uses have never appears under I, and like any other verb inside VP, requires do-insertion in questions and negation: (i) Does Bill often have a short nap? (iii) *Has Bill often a short nap? (ii) Bill doesn’t have his kids wash the dishes. (iv) *Has Bill his kids wash the dishes? 29 In current English, have under I seems to be losing its status as an alternatively realized V. In place of this, have is being used in idioms as a lexical I, but not allowed under V. For example, had in the context ___better^VP is an [I, +T] but never a V: Sue had/ *will have better see a doctor. Similarly, have as in (49b) is being replaced by idiomatic have in I in the context ___got^XP: His parents (*may /*used to/ *seem to) have got a lot of troubles

304

Ludmila Veselovská and Joseph Emonds

lexical entry for this highly specific combination. Nonetheless, this have must appear in any I (with both values ±T) that contains no other interpretable word such as a Modal. If I were empty and perfective have headed its own VP, such an extra embedded VP would be structurally less economic than the simple combination I+VP. So we have now seen how the marked English feature [+T], a Generalized Non-Present, comes to be alternatively realized under I in six special forms (would, could, dared, did, was/were, had), in addition to its cross-linguistic canonical position under V. In many previous analyses of English, a restricted set of grammatical verbs “raise to I” and various suffixes “lower to V” (and sometimes even go back to I!) under ad hoc conditions. The resulting far from elegant picture has resulted from overreliance on head to head derivational movements, in attempts to force lexical particularities into a model suggested by the regular movements of English I to C and French V to I. Our proposal factors out all the language-particular variations, and minimally and elegantly expresses them in lexical entries that exploit the device of Alternative Realization (33). Maximally simple entries for English auxiliaries such as (36), (41), (45), (46) and (48) mention only the barest minimum of itemparticular properties, succinctly capturing the peculiar and non-trivial aspects of English auxiliaries. They require no recourse to idiosyncratic movements.

7 Czech and English Futures In the quest for a universal grammatical system for Mood and Tense, Sections 4–5 have shown that the feature Tense in both Czech and English is canonically on the highest V in a VP, where it also usually gets phonetically realized, except for a few English auxiliaries. In other words, though some familiar patterns (was/ were, auxiliary did, etc.) spell out Generalized Non-Present [+T] under the higher head I, all regular and productive verb morphology in both languages locates Tense under VP. Generative syntax has previously not properly expressed this pervasive pattern. As seen in Section 6, a worked out theory of the grammatical lexicon can specify the realizations of Generalized Non-Present under I as minor variations on otherwise uniform canonical pairings of I with Mood [±M] (±Reaslis) and V with Tense [±T] (Generalized (Non-)present).

305

The cross-linguistic homes of Mood and Tense

7.1 Two grammatical forms of Czech futures (i) Futures of imperfective verbs. It is curiously interesting that, with respect to the same universal grammatical system, the surface positions of so called Future Tense in the two languages are reversed. In English the Modal will transparently exemplifies Canonical Realization of +M in the I position; cf. lexical entry (36). However, the surface patterns of Czech future forms at first glance do not seem to point in the same direction. Veselovská (2008) argues that according to the criteria discussed in Section 3, the Czech future auxiliary stem bud- ‘≈will’ is a regularly conjugated copular V within VP. In contrast to the properties of AuxA in I/C, illustrated in (7)–(14), the future Aux bud- accepts the negative prefix, can be used for VP ellipsis and can be focused. The (a) examples in (50)–(52) illustrate the contrasting behaviors of (a) the future auxiliary stem bud- ‘≈will’ with (b) the conditional AuxB in (6) – demonstrating the location of the former inside VP and of the latter in I/C.30 (50) Negation as a prefix on V: a. Já ne-budu chválit

Hanu.

b. Já (*ne)-bych

(ne) chválil

Hanu.

I not-will1S praiseINF Hana

I (*not)-AuxB,1S (not) praisedpastPRT Hana

‘I won’t praise Hana.’

‘I would(n’t) praise Hana.’

(51) Ellipsis of VP using only V: a. Zítra Tomorrow

budeš

chválit

i

Petra? – Ano, budu / *Ano, chválit.

will2S

praiseINF

even

Peter – Yes, will1S. / *Yes, praiseINF

‘Will you praise even Peter tomorrow? b. Včera

bys

pochválil

Yesterday AuxB,2S praisepast

i PRT

– Yes, I will.’

Petra? – *Ano, bych.

/ Ano, pochválil.

even Peter – *Yes, AuxB.1S. / Yes, praisedpast

tPRT

‘Would you praise even Peter yesterday?

(52)

Focus/ Stress of V in initial or final position: a. (Chci/ Budu) Já (chci/ budu) pracovat (want/ will1S) workINF (want/ will1S) I ‘I want to/ will work.’

(chci/budu). (want/ will1S)

30 There are more properties of especially the AuxB worth discussing in comparison with both lexical verbs and the other auxiliaries. See e.g. footnote 14 on the possible incorporation/ alternative realization of AuxB under some “inflected” complementizers. The discussion here concentrates on contrasting the distinct positions of the Czech conditional and future auxiliaries.

306

Ludmila Veselovská and Joseph Emonds

b.

(*Bych/* Jsem) Já bych/ jsem I AuxB/A.1S (*AuxB/A.1S) ‘I would work / I worked.’

pracovala workedpast

PRT

(*bych/ *jsem). (*AuxB/A.1S)

The lexical entries for the Czech auxiliaries below in (53) can be compared with earlier ones for English auxiliaries in I in (36), (41) and (46). Notice that (53a/b) include the feature contrast already suggested for the Czech Aux in (6), namely the feature [±M], and that both of them are properly followed by a VP whose head is the same [+T], i.e. the -l of the “past” participle. (53)

a.

Czech preterit AUXA:

js-, -M, __< +T >

b.

Czech conditional AUXB:

by-, +M, __< +T >

c.

Czech future AUX:

bud-, V, Φ, +M, ___

Note also that “Futures” in both English (36) and Czech (53c) are (unmarked) Irrealis forms of [+M], with no special feature needed for this “Tense.” The above entry for the Czech future brings together several devices of the grammatical lexicon used throughout this study. First, as the data above show, the canonical Irrealis [+M] feature of I must here be alternatively realized on V. Second, the stem bud- uses the cancellation feature Φ which, again as earlier, indicates that this V is stative. And third, the context feature ___, where Y stands for the lexical head categories N, A, V, P, shows that bud- has the broad range of complements typical of a copula, which will be demonstrated in Section 7.2. (ii) Futures of perfective verbs. Though Aspect is not a main focus of this study, the feature PERF (≈perfective) plays a central role in the Czech verbal system and interacts with the Tense system. Czech grammatical tradition has consistently argued that ±PERF is an inherent feature of (at least some) verbs; therefore, in our terms, PERF is canonically located on the lexical head V. We can see the contrast between imperfective and perfective forms of Czech Pasts in (54a-b): the perfectivity is marked on this particular stem with the prefix do(different stems require different prefixes and no generalization avoids a number of idiosyncratic exceptions). (54)

a. Já jsem stavě-l. I AuxA.1S builtpast PRT.-PERF ‘I was building./ I built.’

b. Já jsem I AuxA.1S ‘I built up.’

do-stavě-l. up-builtpast

PRT.+PERF

Notice that apart from the prefix do-, the past Tenses of both Perfective and nonPerfective stems are formed identically, combining the identical jsem AuxA with the -l participle.

The cross-linguistic homes of Mood and Tense

307

Returning now to future forms, the grammatical expression of the future in Czech depends on the presence or absence of perfectivity (and/or telicity) of the verb; the future auxiliary bud- is incompatible with a following +PERF infinitive. To form a future of a perfective V, Czech does not use the Aux bud-, but rather agreeing present Tense morphology on the lexical stem. Compare the identical underlined endings in (55a/b) and the contrasting interpretation forced by the presence of a perfective prefix. The following examples (55c/d) show that the auxiliary bud- combines with only the imperfective V infinitives and that combining the future Aux bud-and a Perfective (+PERF) infinitive is ungrammatical.31 (55)

a.

b.

[-T],[-PERF]:

[-T],[+PERF]:

Já stav-ím. ‘I am building/ I build.’

= Generalized Present

Já do-stav-ím. ‘I will build up.’

= “future’”

c.

[-T],[-PERF]:

Já budu stavět. ‘I will/ build.’

= “future”

d.

[-T],[+PERF]:

*Já budu do-stavět.

=*

Given the interpretation of the formally “present” inflection in (55b), the [+PERF] verbs in Czech have no possible “present” interpretation. The trees in (56) are plausible representations of the forms in (55b) and (55c). The highest Vs in both these trees realize the canonical feature [–T], i.e. the “Generalized Present” feature which in (56a) is realized in canonical position in the future Aux bud-. In (56b) the same feature is alternatively realized on a lower V, together with the AR of [-M] of I. (56) a.

b.

31 Infinitival structures in both English and Czech may well contain a kind of defective I, e.g. realized as semantically empty to in English. In general, however, contrastive realizations of different feature values in these infinitival I are ruled out. For the grammar of Czech infinitives, see Veselovská and Karlík (2009) and other works cited there.

308

Ludmila Veselovská and Joseph Emonds

It is tempting to explain the Aspectual restriction by saying that some inherently PERF characteristic of the future Aux blocks the PERF feature on V. We briefly return to the topic at the end of this section, suggesting that the restriction is rather a result of Economy.32

7.2 Lexical entries of Auxiliaries and Copulas Consider now some examples of English and Czech copulas expressing past and future time. (57a) shows that the English past form was serves as a copula in addition to being located in I. On the other hand, the future of the English copula be is analytic (57b), requiring both the I will and the infinitival V be. In Czech the situation is the opposite: The past in (57c) is analytic (I+V), while the future (57d) is synthetic, located in V.33 (57) a. I was [V Ø] at home /a student. c. Já jsem byl I AuxA1S bepast

PRT

b. I will be at home /a student.

doma /student. d. Já budu doma /student. at-home /a student I [I Ø] will-be1S at-home /a student

In (58) we repeat (53c), i.e. the entry for the Czech future auxiliary bud- which is conflated with that of a future Copula. The status as a copula makes the variety and pattern of its possible complements similar (though not identical) to the English verb be in (59b).34 (58)

Czech future Auxiliary/ Copula:

bud-, V, Φ, +M, ___< YP >

For comparison with English we repeat the entry for the Modal will in (59a) and give entries for the English copulas be and was/ were in (59b/c). Recall that Φ is

32 Though we won’t further discuss the following examples, note that the restriction on perfective infinitives extends to several other Czech verbs, e.g. to temporal aspect verbs like start and stop/finish, though not to Czech Modals. i. Já musím /budu /začnu /přestanu stavět. ii. Já musím /*budu /*začnu /*přestanu do-stavět. ‘I must /*will /*start /*stop to-build -up+PERF .’ ‘I must / will / start / stop to-build-PERF.’ 33 The Czech future Aux bud- is thus like a synthetic future form of a copula such as French/ Spanish ser- ‘will be’. 34 The selection feature ___ is not quite as general as suggested by (58). English is plausibly like other languages in which copulas are not transitive V that assign accusative case, so the context feature __ probably does not include DP. For more justification of this point, see Emonds (2000: Ch. 8). In Czech the copulas keep a limited ability to assign case, in particular instrumental case; see Veselovská (2008).

The cross-linguistic homes of Mood and Tense

309

the LF Cancellation Feature (44) that characterizes stative verbs, including copulas. (59)

a.

English “future” Modal:

will, M

b.

English non-finite Copula:

be, V, Φ, ___< YP >

c.

English past Copula:

{were, PL / was }, -M, +T, Φ

These lexical entries express what is different in the two languages, i.e. what native speaking children must learn. The basic feature inventories and their canonical positions in trees are, on the other hand, completely uniform.

7.3 The role and distribution of the feature PERFECT The general nature of the feature PERF now calls for clarification. Among other things, this will allow us to explain why it never occurs with the Czech auxiliary bud. The “semantics-based” tradition for defining features, which we have already departed from in our treatment of [±T] and [±M], claims that PERF more or less directly stipulates “completed action.” Instead, we propose a basic syntactic feature on V whose semantic import is less direct, and defined as follows.35 (60)

Definition and locus of +PERF. The canonical feature +PERF on a lexical V reports that a verbal Event/ State holds at a “point of time” different from the deictic Now.

That is, Events or States that pragmatically can cover a time span are conceptualized differently with the feature PERF; they are rather reported as occurring at a point of time, not as “lasting.” We do not mean that the “point of time” must be an instant however; it can be for example a given year. (61) Chomsky napsal své Syntactické Struktury v roce 1956. ‘Chomsky wrote (and completed) his Syntactic Structures in the year 1956.’ The only restriction is that the point of time cannot be Now. However, it can be either future or past, as we now see.

35 For familiarity we retain the traditional label PERF, which has the (for us misleading) connotation of “Aktionsart,” i.e. a “kind’ of verbal action.

310

Ludmila Veselovská and Joseph Emonds

According to (60), the perfective of even stative verbs must relate them to a “point in time,” even if their basic meanings are such that they typically hold over a continuous period.36 The definition (60) holds the key to why Czech perfective verbs with present Tense agreement morphology ([–M] and [–T]) refer to future time. If a clause’s Tense is Generalized Present, formally [–T], then its verb’s Event Time is a “potential Now” (i.e. neither Past nor Conditional), as specified by (1a); it can be only Present or Future. But because its lexical V is +PERF, this Event Time cannot by (60) be the actual Now. Because the only potential Now that are not limited to the actual Now are future times, the Event Time of a morphological Czech “perfective present” is future. The interpretation of the much analyzed Czech perfective pasts is equally natural in our system. Since these forms are [–M, +T], they necessarily refer to Realis Events in the Past, and so automatically satisfy the condition that they do not hold Now. By virtue of the feature [+PERF], they report events as complete and hence completed, i.e. they don’t continue beyond the verb’s Past Realis Event Time. (This is the gist of traditional studies which link the Czech perfect to “completed actions.”) In general then, our system of general syntactic features is crucially not linked to common sense semantic time spans. This has allowed us to unlock the perennial mystery of why Czech “perfective presents” refer not to the Past but are rather the standard way of making futures for these verbs. The difference between the two forms of Czech future concerns their Mood features: the analytic futures with bud- are Irrealis [+M], while the prefixed perfect forms are Realis [-M]. It thus appears that a speaker can conceptualize an expected future event either way, as Realis or not, with essentially no entailed pragmatic distinction.37 In fact, English has futures in both Moods as well, a standard Irrealis with will [+M] and also a Realis future in the form is/ are going to. However, the nature of the contrasts between the two types is quite different in the two languages. While they are in complementary distribution in Czech, depending on the Perfectivity of the lexical V, in English the Irrealis future with will and the 36 As a result, perfective forms of stative verbs take on related meanings that can be conceived of as punctual, for example, a perfective of know is interpreted as find out. 37 It is natural to ask, why do English present perfects, whose auxiliary have may well be specified as +PERF, not refer to the future? The answer is because this V have must occur with a participle with the separate morphology of –en, a morpheme which when active spells out Past, i.e. [+T]; for arguments to this effect, see Emonds (2012). This Past -en [+T] ensures that the lexical verb’s Event Time as always prior to the “Reference Time” expressed by the Tense of have. In contrast, Czech PERF prefixes are directly linked to the Verb’s Event Time, which can be [+T] or [–T].

The cross-linguistic homes of Mood and Tense

311

Realis future with going to are close to synonymous; it is next to impossible to find contexts where one is well-formed and/or appropriate and the other is clearly not (though nuances may differ). We conclude then that the two LFs of futures in both languages are the same, and that the difference in (otherwise unmarked) Mood values [±M] combined with [-T] is nearly meaningless. All the expressions for the future are [–T], and all specify in LF an Event Time different from Now.38 As a final point, our system now easily accounts for why Czech bud- does not occur with a verb’s Perfective form. If it did, it would be entirely redundant indication of the future by a sort of “double tense marking,” and so the combination is ruled out by Economy.

8 Czech and English: finite variations on a universal theme Within Indo-European, one could hardly ask for a system more superficially different from English than Czech. In contrast to English, Czech exhibits scrambling, i.e. free constituent order in clausal domains, including extractions of nominal pre-modifiers. Czech is a full pro-drop language; it has rich agreement inflections with lexical categories (N, A,V), including both person and gender agreement with the subject in two analytic Tenses. We have moreover examined the Czech verbal prefixes that express perfective vs. imperfective aspect and mentioned some facts related to the Czech (Slavic) system of C2 (“second position clitics”) for verbal objects. On the other hand, Czech completely lacks articles (as overt realizations of the D category), as well as counterparts to hallmarks of English IP structure, such as the non-verbal modal words and the auxiliary do. To add to this sketchy list of properties of English not shared by Czech, one can mention also parasitic gaps, multiple and long distance WH movement, possessive anaphors, raising 38 In our system, the futures with bud- and will correspond, because they both realize the same features: [+M, -T] as well as -PERF. The correspondence of the Realis futures in the two languages is less direct. As just discussed, in Czech the [+PERF] becomes “future” by shifting the verb’s Event Time away from Now, while still keeping it as a Generalized Present [-T]. In English and also French and Spanish, unmarked verbs of Motion away from the spatial deictic center “Here” (go/ aller/ ir) are metaphorically transposed into motion away from the temporal center “Now.” Thus, the feature PERF in Czech and a feature of temporal Motion in English have the same effects in LF: movement of the clausal Event Time away from the actual Now into a potential Now of the future.

312

Ludmila Veselovská and Joseph Emonds

of the negation particle, and raising of noun phrases to object and subject positions. Despite the above and many other differences, this study has demonstrated that general syntactic principles plus a few simple language specific lexical entries (in any case needed independently of a given framework) provide accurate and parsimonious analyses of the superficially very different Czech and English verbal systems, and allow at the same time a systematic comparison. With respect to the architecture of verbal projections in Czech and English, we have argued that the same general principles and categorical system of Universal Grammar furnish a unified framework for parallel extended V-projections, i.e. VP and IP. In particular, the (binary) feature content of the lexical head V and the related functional head I is the same in the two languages. We argued that contrary to generally accepted ideas, the functional I projection canonically houses only the feature of mood [±M] ,which basically expresses the contrast Realis vs. Irrealis. On the other hand, a clause’s temporal specification using [±T] (Generalized Present vs. Generalized Non-Present) and [±PERF] is canonically located (and interpreted) not on I, as standardly assumed, but on verbal heads inside the VP projection.39 The proposed analysis demonstrates that apart from occupying their canonic positions in I and V, their respective principal features [±M] and [±T] can also be alternatively realized (though not interpreted) on adjacent head nodes, both higher and lower. We discussed alternative realizations of e.g. the features of I on both V and C. We conclude that unawareness of the option of AR has confused previous analyses, which have inadvertently taken the surface positions of English Tense in I as revealing some general property of Universal Grammar. In arguing for both functional and lexical projections in Czech clauses (for the existence of separate I and V) we illustrated lexical entries of items located in both positions. Then, comparing the Czech and English auxiliaries and copulas, we demonstrated that the lexical items with similar feature content can be located in distinct categorical positions. For example, the Czech present Tense copula js- ‘is/are’ is in V, while its English finite counterparts (is/are) are in I. In an interesting contrast, the Czech future auxiliary bud- is inside VP (with a regular finite person and number paradigm), while the English future auxiliary will appears only under I, and like other English Modals accepts no inflections. 39 In this study we used only one functional head related to VP and labeled it I. If an adequate framework requires separate heads for each individual feature, there may be several functional heads in both Czech and English – a very high AGR being the most plausible one for Czech within both VP and IP. However, not much in this paper depends on the (non-)existence of separate proxy functional heads above and below VP. Neither would the system here change much if the verbal heads were relabeled – given that the arguments here mainly illustrate and depend on structural relations and don’t concern the actual symbols in a taxonomy.

The cross-linguistic homes of Mood and Tense

313

The future morphemes of both languages seem to lack infinitive forms because they both realize [-M], i.e. Irrealis, and at least in Czech and English, infinitive forms appear to prohibit any realizations of specific features of I. Therefore in spite of several differing language-particular realizations, the interpreted LF representations of (non-perfective) Czech and English futures are identical: [I, +M] + [VP, -T]. In conclusion, this study demonstrates that the two languages share similar clausal structures consisting of lexical and functional domains. Moreover, in spite of their specific characteristics they are subject to the same universal principles, in particular Alternative Realization, governing the behavior of closed class items, including bound morphology. The framework employed succeeds in providing analyses of the verbal systems of the two languages which are both simple and explanatory.

References Chomsky, Noam. 1957. Syntactic Structures, The Hague: Mouton. Chomsky, Noam. 1965. Aspects of the Theory of Syntax, Cambridge: MIT Press. Chomsky, Noam. 1991. “Some Notes on Economy of Derivation and Representation,” Principles and Parameters in Comparative Grammar, R. Freidin (ed.), Cambridge: MIT Press, 417– 454. Collins, Chris. 2001. “Economy Conditions in Syntax,” The Handbook of Contemporary Syntactic Theory, M. Baltin and C. Collins (eds.), Oxford: Blackwell Publishers, 45–61. Embick, David and Ralph Noyer. 2001. “Movement Operations after Syntax,” Linguistic Inquiry 32,555–595. Emonds, Joseph. 1976. A Transformational Approach to English Syntax, New York: Academic Press. Emonds, Joseph. 1978. “The verbal complex V’-V in French,” Linguistic Inquiry 9, 151–175. Emonds, Joseph. 1987. “The Invisible Category Principle,” Linguistic Inquiry 18, 613–632. Emonds, Joseph. 1994. “Two principles of economy,” Paths towards Universal Grammar: Studies in Honor of Richard S. Kayne, G. Cinque et al. (eds.), Washington: Georgetown University Press. Emonds, Joseph. 2000. Grammar and Lexicon: the English Syntacticon, Berlin: Mouton de Gruyter. Emonds, Joseph. 2012. “The Single Morpheme -ed/-en of the English Past/ Passive,” Acta Linguistica Hungarica, 59 (1–2), 1–26. Grimshaw, Jane. 1991. Argument Structure, Cambridge: MIT Press. Huddleston, Rodney and Geoffrey Pullum. 2002. The Cambridge Grammar of the English Language, Cambridge: Cambridge University Press. Halle, Morris and Alec Marantz. 1993. “Distributed Morphology and the Pieces of Inflection,” The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale and S. Keyser (eds.), Cambridge: MIT Press. Lees, Robert. 1960. The Grammar of English Nominalizations, The Hague: Mouton.

314

Ludmila Veselovská and Joseph Emonds

Lieber, Rochelle. 1980. On the Organization of the Lexicon, MIT doctoral dissertation. Palmer, Frank. 1986. 2nd edition, 2001. Mood and Modality, Cambridge: Cambridge University Press. Roberts, Ian. 1993. Verbs and diachronic syntax: a comparative history of English and French, Dordrecht: Kluwer Publications. Toman, Jindřich. 1980. “Weak and Strong: Notes on be in Czech,” Wege zur Universalien Forschung, Sprachwwissenschaftliche Beitrage zum 60 Geburstag von Hansjacob Seiler. G. Brettschneider and C. Lehmann (eds.) Tubingen: Gynter Narr Verlag. Veselovská, Ludmila. 2002. “Struktura subjekt-predikátové shody,” Čeština-univerzália specifika 4, Hladká and Karlík (eds.), Brno: Masaryk University. Veselovská, Ludmila. 2008. “The Extended verbal projection in Czech: Three Variants of the verb be,” Formal Description of Slavic Languages (the Fifth Conference, Potsdam 2003), G. Zybatow, L. Szucsich, U. Junghanns and R. Meyer (eds.), Berlin: Peter Lang. Veselovská, Ludmila and Petr Karlík. 2004. “Analytic Passives in Czech,” Zeitschrift fur Slavistik 49 (2), 163–243. Veselovská, Ludmila and Petr Karlík. 2009. “Infinitive Puzzle,” Czech in Formal Grammar, M. Dočekal and M. Ziková (eds.), Munich: Lincom, 197–215. Zagona, Karen. 1988. Verb Phrase Syntax: A Parametric Study of English and Spanish, Dordrecht: Kluwer Publications.

Edwin Williams

Multidomination and the coherence of anaphors* Multidomination (MD) has been proposed for two sorts of relations in natural language – movement and coordinate sharing constructions like Right Node Raising, gapping, ATB movement, and parasitic gaps. Both applications of MD have big problems. The MD theory of movement has an interpretive problem in all implementations. The MD theory of coordinate sharing constructions leads to enormous over-generation. In this paper I will suggest solutions to each of these problems.

1 Multidomination theories of movement Multidomination theories of Movement share a problem with copy theories of movement – the moved thing is of a different semantic type than the movedfrom position. So, for example, we have the following representation in Fox’s (2002) copy theory of movement: (1)

[Which book] did John read [which book]

Here, underlining marks the copy. The problem is that the copy has which in it. The copy will be interpreted as a variable bound by the moved phrase, but what role could which play in the identification of the copy as a variable? Clearly none. So Fox proposes the following rules to re-adjust the copy: (2)

Trace Conversion a. Variable Insertion: (Det) Pred N -> (Det) [Pred λy(y=x)] b.

Determiner Replacement: (Det) [Pred λy(y=x)] -> the [Pred λy(y=x)] Fox (2002) p. 68

* I want to thank R. Kayne, H. van Riemsdijk, A. Szczegielniak, and the audience of a lecture series I gave on Multidomination in Vienna in June 2011 for helpful discussion of the material herein. They are irresponsible for any errors. I am also grateful to an anonymous reviewer for many sharp observations.

316

Edwin Williams

These two rules will effect the following conversion: (3)

a.

[Which book] did John read [which book]

b.

[which book] λx did John read the book

==>

x1

The rules give the right results – the copy is now something like a bound variable – but the rules are fishy and in fact they violate the premise of “the indistinguishability of copies” that is tacitly assumed to constrain copy theories. The rules in (2) work on one copy without affecting the other copy. A true MD theory of movement would seem to have the “the indistinguishability of copies” built right in, since there are not two copies, but rather a single thing which is dominated multiple times by the structure which contains it. And yet even here, ingenuity finds a way. Below is a representation of Wh-movement implemented as MD: (4)

But of course the same problem arises for Multidomination as arises with Copy theories of movement: we want a bound variable as the object of see, but we want a quantificational expression in the SpecCP. Since there are not two copies, but only one thing, it would seem impossible to manipulate the “trace” separately from the moved constituent. And yet, here are the adjustments to the structure in (3) that Johnson (2012) suggests: 1 In addition to the rules in (2), Fox has a rule which inserts λx just below the moved constituent.

Multidomination and the Coherence of Anaphors

(5)

317

a.

The upper instance of the MD’ed DP is ‘housed’ in a QP: [Q DP]QP

b.

The bottom instance is not housed in a QP, and DP is interpretable as a variable

The problems are manifold: the presence of which in the bottom copy does not square with that copy’s role as a variable; the [Q DP] structure at the top is not interpreted compositionally (by Johnson’s own admission); the upper instance does not C-command the lower instance; and the relation between the dominating QP and its DP is stipulated thusly: “Let’s assume that the index on QP is copied from the index that comes with which”. This is just as fishy as Fox’s copy theory adjustment rules because it is in fact the same fish. A parallel variant of movement as Multidomination appears in Engdahl (1986)2. The structure she proposes for questions are parallel in every respect to the structures in Johnson (2012) as in (4) above, and with the same problems. Engdahl’s solution (p. 49 ff) is parallel to Johnson’s and Fox’s, but resides in how rules of interpretation apply to structures. In a discussion of “Which table _ did Mary put a book [on _ ]?”, Engdahl says, “the only way we can interpret the PP is by storing the variable binding operator and inserting a variable in its place”. This corresponds precisely to Johnson’s and Fox’s adjustments (though preceding them chronologically). In all three of these theories, structures are first built, and then semantic interpretation is the compositional interpretation of those built structures, and in all three, special adjustments must be made to reconcile the thesis that structures are compositionally interpreted with the behavior of Multidomination structures, or for that matter, of movement itself. I think that a coherent theory of movement cannot be built unless the thesis that structures are compositionally interpreted is given up. This does not mean that semantic compositionality must be given up. It is the interpretation of structures that must go. Instead, we will interpret the derivation of those structures, not the structures themselves. In previous work I have described a mode of derivation which attains the MD theory of movement without the problems just described. The part that will make this solution hard to swallow for some is that “which book” in (1) is not an interpreted constituent; it is rather a purely morphosyntactic constituent, with no more claim to mirror meaning that “think’ll” has in “Who do you think’ll go”. As I have detailed this mode of derivation elsewhere (Williams 2011, 2010,

2 Thanks to a reviewer for bringing to my attention the relevance of Engdahl’s proposals.

318

Edwin Williams

2003, 1998), here I will only present the essential components that implement a MD theory of movement without the fishiness. The basic idea is that a DP is not built up to its top and then embedded as, say, the object of a verb; rather, the core of the DP (N or NP) is embedded immediately as the object of V, and then N and V together “grow” throughout the rest of the derivation. I called this mode “Cogeneration” in Williams (2003), and opposed it to “Level Embedding”, the mode for clausal embedding. Cogeneration is implemented with “pointers” – that is, at the initial stage of derivation, N and V are not themselves merged together; rather, a pointer to N is merged with V, and N stays in the workspace. It must stay in the workspace if it is to be subject to “growth” by further merges (or any other operations), as only members of the workspace can be worked on: (6)

{. . . ,N, . . . ,V, . . .} => {. . . ,N1, . . . ,[V 1], . . .}3

Various further operations can affect N1 and [V 1] – for example, Agree can apply between them, and so forth. Now, how does WH movement fit into this? One can immediately see that a MD theory of movement is in the offing, implemented with the very pointer that was merged with V at the initial stage of derivation. At a certain point, we will have the following in the workspace: (7)

{. . . . , N’’’1, . . . . , [V 1]’’’, . . . . .}

The ticks are meant to represent the further applications of rules like Agree that add words and morphemes to the original N and V. At this point, which enters the picture. Two things happen – which triggers a rule of semantic interpretation, and which triggers a morphosyntactic operation. The key idea here is that these happen simultaneously, like in Montague grammar (Montague (1973)). Semantically, which, being a (generalized) quantifier, takes two arguments, a restrictor and a scope; the N’’’ and [V 1]”” will be these, respectively (or rather, their meanings will be). Morphosyntactically, which prefixes the restrictor to the scope, and prefixes itself to the restrictor. The whole thing can be illustrated in the following diagram:4 3 Curly braces enclose workspaces. 4 Since each derivational step computes a new constituent and its meaning, the workspace really consists of form-meaning pairs, with Combine creating new forms and the Semantic function creating their new meanings. Here I have left the meanings out of the representation of the workspace.

Multidomination and the Coherence of Anaphors

(8)

319

restrictor scope / / Semantics: which( [N’’’1], [[V 1]’’’]) / which {. . ..N’’’1 . . ..[V 1]’’’’ . . . . .} ==> { [j [V . . . . . j]]VP , [which N]j } \ Morphosyntax: 1. Combine( which, N) = [which N]j 2. Combine( 1, [V 1]’’’) = [1 [V 1]’’’]

Which maps one workspace to the next. On the top is the semantic side of the derivation; which takes its two arguments, the meaning of N’’’1, and the meaning of [V 1]’’’’. At the same time, and independently, a morphosyntactic function, called Combine, applies to the two arguments (the forms, not their meanings) of the semantic function. Combine(which, restrictor) syntactically prefixes “which” to the restrictor argument. Combine(1, scope) prefixes the index of the restrictor to the scope, giving the representation [1 [V 1]’’’’]. As “1” now points to [which N]1, this will now be “read out” as the correct surface form: (9)

a.

New Workspace: {. . . ,[1 [V 1]’’’], . . . ,[which N]1, . . .}

b.

read out as: [[which N’’’]1 [V 1]’’’’]

Movement here is MD, in that the second Morphosyntactic operation prefixes to the scope argument a second pointer to N’’’, and the convention is that the MD’ed item is pronounced, or “read out”, at the position of the most recent pointer. This implementation of MD does not have the problem of the ones discussed earlier, because “which N” is not an interpreted constituent, it is rather a purely morphosyntactic form created by Combine, a rule whose input is interpreted, but whose output is not. The surface has the constituency in (9), but the semantic constituency is (which (N’’’, [V 1]’’’)), and Combine is responsible for the discrepancy between the two. There are many problems to be solved here – how do islands arise, how does pied-piping work, etc., see Williams (2003, 2010, 2011) for proposals. Also, what kinds of things can the Combine function do? Clearly there are more possible outcomes than the one illustrated – e.g. in some languages Wh does not move, so the second Combine operation does not apply. Again, see the works cited for a specific proposal about how Combine is parameterized for different outcomes. If Combine is the only syntactic/morphosyntactic operation, then one could say that syntax/morphosyntax is the science of how the surface of language fails

320

Edwin Williams

to mirror the meaning, and that the parameters governing Combine are the channels of language variation (see Williams (2011) for further discussion). The narrow point here is that the mode of derivation illustrated in (8) avoids the interpretive problems that arise in MD and copy theories of movement like Fox’s, Johnson’s, and Engdahl’s.

2 Multidomination theories of coordinate sharing structures If MD is a good candidate for movement, then it is surely a good candidate for various sharing structures, like Right Node Raising (RNR), ATB movement, gapping, parasitic gaps, and coordinate reduction structures, as none of these pose the interpretive problem just discussed for MD theories of movement – in all coordinate cases, the MD’ed item has the same semantic type in each of its instances. But there are other problems. In the next several sections I will explore a particular instance of a general problem – that the different instances of a shared phrase must “match” in their local function in a way that is not explained by extant theories of MD. In particular I will demonstrate that an NP (or DP) cannot be shared between predicative and referential positions except in a very limited way, and I will ask how that matching effect can be forced by different theories. The over-generation for sharing constructions is acute, especially given Grosz’s (2009) findings about RNR. It has been known at least since Abbot (1976) that RNR was not limited to constituents (e.g. “John mailed, and Bill couriered, [packages] [to Mary]”). But Grosz unearthed a nightmarish extreme, which I think can justly be described as “micro-sharing” – any number of constituents of arbitrarily small size can be shared in RNR structures. His telling example is the following: (10) [ CNN claims that a man3 is __] and [the BBC claims that a woman7 isn’t __ ] likely to be assassinated t3/7 by the serial killer. Grosz (2009) The problem posed by the example is that if the underlined phrase is the RNR’ed constituent, there is no consistent way to label the trace that it contains. Grosz’s solution: the underlined phrase is not RNR’ed as a whole, but in many little pieces: “by the serial killer” is shared, “assassinated” is shared, “to” is shared,

Multidomination and the Coherence of Anaphors

321

“be” is shared, etc, everything is shared but the trace(s), which have to be different. I in fact think that Grosz’s analysis of this example is not correct, but my present concern is what would stop it? That is, what principle would bar microsharing?

3 Predicative and referential NPs Predicative and referential NPs are similar in their internal structure, but utterly different in their function, and many distinguishing properties have been observed. For example, predicative NPs do not passivize: (11)

*A doctor was become by Mary.

The generalization we will explore is that an NP is barred from being shared between a predicative and a referential position in every situation in which the opportunity would arise: (12)

a.

RNR: *John married and Mary became an alcoholic prostitute

b.

ATB: *What a glamorous movie star John married and Mary is becoming

c.

PG: *What a famous movie star John married before becoming

d.

VP-Del: *John saw a communist in the woods and he is too

e.

Relativization: *John hired a comedian that my father was

The question is, why does (12) hold? One could simply stipulate that a shared phrase must play the same role in all of its contexts. But this is too general; for example, a shared NP can be a theme in one context and a goal in the other: (13)

John married and Bill gave money to an alcoholic prostitute

So the notion of “sameness of function” needed for (12) reduces to the predicative/referential distinction, and explanation is diluted to nil. One would hope that referential and predicative NPs would have different internal structure, and that that would make sharing impossible. Again, a number of differences have been noted; for example, predicative NPs cannot have the full range of quantifiers:

322 (14)

Edwin Williams

*John and Bill are every friend of mine.

A plausible idea about how predicative and referential N/DPs differ is the following: (15)

A referential DP is a full DP, whereas a predicative NP is just an NP, where NP is a subpart of DP (that is, [. . .NP. . .]DP is the structure of a referential NP)5.

But the force of this, or any other, structural difference in the constitution of predicative and referential NPs is vitiated by micro-sharing; no matter what the differences in structure, an example like (12) can be generated by micro-sharing “an”, “alcoholic”, and “prostitute”, and not sharing anything else in the structures: (16)

John married [X an Y alcoholic Z prostitute W]RefNP and Mary became [X’ an Y’ alcoholic Z’ prostitute W’]PredNP

It won’t matter what the structural difference is between Predicative and Referential NPs – it can be much more radical that what I suggested above – microsharing will still allow sharing of the audible parts. Whatever differences there are will be in the silent parts, but those will not be shared. I think that micro-sharing must be given up. This means that some alternative to Grosz’s analysis of (10) must be found. But in fact, even barring microsharing might not be enough to rule out (12) under the assumption in (15). Supposing that “an alcoholic prostitute” is an NP and not a DP, what will stop the following sharing?: (17)

John married [. . .NP. . .]DP and Mary became NP [an alcoholic prostitute]NP

5 I ignore here the fact that predicative NPs are liable to second order quantification in examples like “This house has been every color”; see Williams (1983) for discussion.

Multidomination and the Coherence of Anaphors

323

Even restricting shared material to single constituents, which we know is overly restrictive in any case, will not solve the problem in (12) under the assumption in (15). In the next sections I will propose a way to restrict sharing in a way that is consistent with (15).

4 Coherence First, I want to document that when a predicative NP occurs, there is always higher “clausal” structure dominating it that is not present when an NP is part of a referential DP. We may compare the following: (18)

a.

John is usually a bully.

b.

I was attacked by usually a bully

c.

I was usually attacked by a bully

d.

I was attacked by someone who is usually a bully

The simple point here is that “usually a bully”, if a phrase, is not an NP, and cannot occur as a subpart of DP. Importantly, (b) cannot have the meaning paraphrased in (d), but only the paraphrase in (c). Put in Cinque’s (1999) terms, usually in (b) is used as a focusing adverb, and not as a direct modifier of “a bully”; that is, usually does not form part of the predicate that “a bully” is the (ultimate) head of. Cinque identifies focusing uses of adverbs as lying outside of his program of regimenting adverb orders. That is probably a mistake; he is right though that their surface position does not represent their “F-structure” role.6 Wherever predicative NP occurs, adverbs in non-focusing-roles are admissible: (19) a. Small Clause: I consider John usually a bully b. Absolutive: With John usually a bully, we had to put him on Zoloft c. Nonrestrictive appositive: John, usually a bully, became a mouse on Zoloft d. Restrictive appositive: The person usually the leader is here to see you.

6 See Williams (2003), (2010), (2011) for discussion; In the last two citations, Combine(Adverb, X) does not always prefix Adverb to X, but places it inside of it.

324

Edwin Williams

Here we have a nominal predicate of a small clause, of an absolutive construction, of a nonrestrictive appositive, and of a restrictive appositive, and in all cases an adverb is allowed. Conclusion: no matter how small the clause, there is room for adverbs above NP, and these adverbs cannot occur above NP in DP. There are some apparent exceptions, but they are subject to different analysis: (20)

a.

I was attacked by possibly the biggest bully in the school

b.

I was attacked by someone who was possibly the biggest bully in the school

c.

I was possibly attacked by the biggest bully in the school.

Here (a) is accurately paraphrased by (b), and not by (c), so possibly here is not a displaced focusing adverb. Since possibly is only allowed for superlative NPs, it is likely that possibly modifies the superlative operator directly, and not the NP which contains it. I conclude from the above discussion that there is no predicative use of NP which does not have F-structure dominating it, no “pure” use of NP as a predicate; so “. . .” in the following is always non-empty, even if silent: (21)

with [John . . . [a bully]NP]

And further, whatever the “. . .” is in (21), it does not occur above NP in referential DPs. This does not by itself solve the sharing problem exposed in the previous section, but sets the ground for it. The needed principle is “Coherence”. It can be illustrated with VP ellipsis. (22)

a.

John arrived and Bill did [ ] too

b.

John arrived yesterday and Bill did [ ] too.

c.

John arrived yesterday and Bill did [ ] today

From (a) it does not follow that John and Bill arrived at the same time, or on the same day. From (b) it follows that John and Bill both arrived yesterday; from (c) it follows that John and Bill did not arrive on the same day. How do these differences arise? Let us suppose that there is a bit of F-structure (X below) dominating VP which pertains to event time:

Multidomination and the Coherence of Anaphors

(23)

325

[[arrive]VP X]VP

Now, when X is silent it is interpreted as “at some time”; and so (22a) is rendered (24)

John arrived at some time and Bill arrived at some time too

It is not obvious whether the “at some time” in the second clause originated there as a silent X, or came via the antecedent. But (22b) and (c) are relevant to this question: (c) shows that VP ellipsis can target the smaller VP in (23). Given that, there is no reason why (b) cannot have the following structure: (25)

John arrived yesterday and Bill did [[]VP X]VP too

where silent X is again rendered as “at some time” giving (26)

John arrived yesterday and Bill arrived at some time too

But it does not follow from (25) or (26) that Bill arrived yesterday. There must be something wrong with [[ ]VP X]VP where X is silent with default interpretation and VP is anaphoric. I propose the following principle7 (27)

Coherence: If X is silent and adjacent to a silent anaphor, X must be part of that anaphor.

By (27), when X is silent in [[ ]VP X]VP, the smaller VP cannot be an anaphor; rather, the anaphor must be the larger VP, including X; X will then take on the value of its antecedent, whatever that is, including the time reference – it might be the generic “at some time” if the antecedent of X is silent also. This incidentally resolves the question raised earlier about the interpretation (22a): the “at some time” modifier of the second VP is interpreted via antecedence. Importantly, the identity of the time reference of the elided VP in (22b) as “yesterday” does not arise simply because the time reference is silent; in the following the second time reference is not necessarily identical with the first (i.e. both yesterday): 7 The principle appears in Williams (1997) and again in Williams (2009). In Williams (1997) the principle said that if X was a silent anaphor and adjacent to another silent anaphor Y, then X and Y must be “Co – anaphoric”, to use the terminology of that paper; here I seek to extend that principle to cases where X is not necessarily an anaphor, but is merely silent.

326 (28)

Edwin Williams

John arrived yesterday and Bill arrived too

Rather, together with (22), (28) show that it is very specifically the adjacency of a silent element to a silent anaphor that triggers Coherence. But how can this help with RNR, where we have shared material, not elided material with an antecedent? We can define Coherence for shared material as well: (29)

Coherence: If X is silent and adjacent to a shared element, X must be shared too.

I will leave it for later work to find out how (27) and (29) can be combined. I note here that there is some indeterminacy in what it means to be adjacent to a shared element, which will be resolved shortly. Now, Coherence rules out Grosz’s analysis of his telling case (10), as it requires the traces to be shared, and I will return to that later. But it also correctly rules out the sharing of an NP between a predicative and a referential position, under the assumption about their structure that we have made, namely, that predicative NP always occurs at the bottom of an F-structure possibly hosting adverbs, but NP in DP does not. Let us assume that those adverb positions can be silent and then get default values, “at some time”, “for some reason”, etc. Now let’s look at a case of RNR where Coherence does some work: (30)

John bought

X and Mary ate

cookies today

Here, “cookies” is shared, but “today” is not shared, it is only dominated by the VP of the second clause; in fact X occupies the “time” position in the F-structure of the first clause and is meant to receive its default interpretation. But it does not. Coherence says why this cannot be: X is silent, and X is adjacent to a shared item, so X must be shared as well; given its position in F-structure it can only count as a shared time expression (either silent or not), and so X must be “today”, giving both constituents as shared. Next let us return to our central problem, the sharing of an NP between referential and predicative positions:

Multidomination and the Coherence of Anaphors

327

(31) a. *John married [ . . . [ ]NP]DP and Sally became [ . . . [ ]NP]VP [an alcoholic prostitute]NP b. *Sally became [ . . . [ ]NP]VP and John married [ . . . [ ]NP]DP [an alcoholic prostitute]NP

We can now understand (31a and b) as violations of Coherence. The “. . .” in the VP of the second conjunct of (a) contains one or more silent elements adjacent to the shared NP. Under Coherence, these must be shared as well. But in the first clause, the “. . .” in DP does not contain corresponding positions, and so sharing is impossible. We have argued that no matter how small the clause, the predicative NP does not itself constitute the entire structure of the clause, so there will always be a residue that cannot be shared with the upper structure, silent or not, of a DP. The (b) example is just the reverse of the (a) example, but the logic is the same: Under Coherence, the “. . .” after become must be shared with the “. . .” after married, but again that is impossible. Importantly, it is the adjacency of silent material to one of the sharing positions (“[ ]NP”) that triggers Coherence, not adjacency to the spelled-out shared phrase itself.8 We do expect to have sharing of an NP between two predicative positions, or between two referential positions, as the “. . .” residue can be shared in such cases: (32) a. Sue is [. . .[ ]]VP and Mary will become [. . .[ ]NP]VP [an alcoholic prostitute]NP b. The CIA is [Bill’s. . .[ ]]DP and the FBI is [Sally’s. . .[ ]NP ] [favorite charity]NP c. The CIA is [Bill’s [ ]]DP and the FBI is [Sally’s [ ]NP ] . . . [favorite charity]NP

Note that in both (a) and (b) it is an NP, not a DP, that is shared. The reason that (b) is valid is that the “. . .”s in the two clauses are the same, and so can be shared, and under Coherence, must be shared; so in fact the correct representation of (b) is actually (c), with everything shared including silent material (in fact it is likely that “. . .[favorite charity]” must be shared as a single constituent, but that is beyond the point needed here). The fact that NP can be shared between two DPs shows that the failure to share between predicative and referential positions does not reduce to a failure of NP inside of DP ever to be shared, in case you were thinking that that was the answer.

8 Thanks to a reviewer for pointing out the possibility for misconstrual here.

328

Edwin Williams

5 Coherence and relativization There is a final construction in which Predicative and Referential NPs cannot be mixed, and that is relative clauses9: (33)

a.

I hired the comedian that your father hired

b.

I am the comedian that my father was

c. *I hired a comedian that your father was Of special interest is (c), which would have a grammatical derivation under some general assumptions. I will show how Coherence blocks it, but first the general assumptions. The first assumption is our dual assumptions about the relation between predicative and referential NPs – that predicative NPs occur without a DP shell, and that referential DPs contain an NP which is identical in form and interpretation to predicative NPs. Secondly we need to assume something about the mechanics of relativization, and there are many ways to go. There are 3 relations to take into account in the analysis of relative clauses: the relation of the trace to the CP of the relative, the relation of the CP to the head of NP, and the relation of the Wh-phrase to the head of the NP. For simplicity, I am going to adopt the “head raising” analysis proposed by Vergnaud (1974), in which there is a chain of movement relations from the original trace position to the SpecCP to the head position of the NP. If other assumptions about relativization are chosen, the account about to be given for (c) would need to be adjusted accordingly. Under the head raising analysis, (c) will have the following structure: (34)

a.

I hired [ [ [ ]NP ] [ that your father was [a comedian]NP]DP =>

b.

I hired [ [ [a comedian]NP ] [t that your father was [ ]NP]DP

There is no good reason that this derivation should fail. However, since movement is implemented as Multidomination, we must consider the silent elements in (b) to judge how Coherence will apply: (35)

I hired [. . .N [ [a comedian]NP ] [t that your father was . . . V [ ]NP]DP

9 These examples were inspired by Vergnaud (1974) though he did not make the point about them made here.

Multidomination and the Coherence of Anaphors

329

As we have shown that there is always verbal functional structure above a predicate nominal (the “. . . v” here) and as that material is both silent here and adjacent to a shared element, it must be shared as well. But as the NP in the head of DP is either dominated by no silent functional material or is dominated by entirely different silent functional material (the “. . .N”), the Coherence condition cannot be met. So (c) is akin to all the other cases of bad sharing we have looked at.

6 Coherence and F-structure Some examples supplied by a reviewer point to the possibility of a fruitful exploration of the interaction of F-structure and Coherence: (36)

a.

John arrived yesterday but Bill never will.

b.

John will read this tomorrow because Mary hasn’t.

These were in fact presented by the reviewer as challenges to the Coherence principle itself: in (a), the adverb yesterday adjacent to the antecedent is not understood as part of the elided material; and in (b), tomorrow likewise is not. In both cases, if the adverb were included as part of the elided material, nonsensical readings would arise (*Bill never will arrive yesterday; *Mary hasn’t read this tomorrow). We could conceivably revise Coherence to take into account “sensicality”; that is, Coherence would apply just so long as the readings derived were “sensical.” I think that is the wrong path. I believe in fact that the cases cited by the reviewer have a direct answer in terms of the theory as already stated. At the same time, examples similar to the ones cited by the reviewer need a deeper investigation than I can give here. In this section I will outline the kind of solution that sticks to the present understanding of Coherence, and I will illustrate the kinds of predictions that need further investigation. First let’s look at solutions to the examples in (36). I will take (b) first, as its solution is simpler and will serve as a model for (a). The structure for (b) is: (37)

John will X read this tomorrow because Mary not has Y [read this]

Here X stands for the (largely or possibly totally) silent F-structure that intervenes between “will” and “read”, and Y is similarly interpreted. To understand (36b), we need only assume that “has” is low enough in F-structure that Y does

330

Edwin Williams

not include time adverbs like tomorrow. We know that has is lower than will because we have “will have” (and not *have will), and we may safely assume that the surface order of auxiliaries reflects their true scope. It is plausible then that the position for time adverbs like tomorrow is above have, and so even if Coherence forces Y to be anaphoric with material in X, X will not itself contain tomorrow. In short, if we have the ordering in (a), then the anaphora forced by Coherence will be (b): (38)

a.

will > tomorrow > have > [Y VP]

b.

will [tomorrow

[X VP]] (antecedent clause) have [Y VP ]] (clause with elision) Y anaphoric to some low segment of X

Since X does not include tomorrow, Y cannot either. Example (36a) can be made to work similarly, if some assumptions are made about the relative F-structure points at which yesterday and never come into the derivation. Unfortunately we cannot take the surface order of adverb and auxiliary to determine scope (where scope is equivalent to F-structure position); for example, with never and can, the order of interpretation is the opposite of the surface order: (39)

John could never do that

never > could

In our terms, the order of entry into the derivation, and hence the scope, must be never>could (that is, never has scope over could, or, equivalently, never enters the derivation after could), but the rule for the morphosyntactic placement of at least some adverbs is to place them after the first auxiliary verb of the structure that serves as their scope (see Williams (2010), (2012) for discussion of Fstructure as a “clock” timing derivational events, rather than a structure itself.) So, let us assume the following F-structure ordering: (40)

Tense > will > tomorrow/yesterday ≥ never > VP

The crux of the solution to (36a) then is that never is as least as low as tomorrow/ yesterday, and possibly lower. This is a plausible ordering to the extent that when both tomorrow and never appear in the same sentence, never is understood to be in the scope of tomorrow, and not the reverse: (41)

John will never go there tomorrow (never < tomorrow; *tomorrow < never)

Multidomination and the Coherence of Anaphors

331

If we make the assumptions in (40), then (36a) will have the following structure: (42)

John Past Marry Pres will

yesterday never

`

X Y

arrive arrive

(antecedent clause) (clause with elision)

Since Y is smaller than X, it will be forced to be anaphoric with some lower segment of X, but that will not include yesterday, and so the observed fact is accounted for. For both examples in (36) then, assumptions about the F-structure positions of elements for auxiliaries and adverbs “capped” the size of the material adjacent to the elided VP in such a way that the application of Coherence was appropriately constrained and the nonsensical readings were not derived. The question is, can this kind of solution work generally for examples like (36)? The answer here must be tentative, but is yes. A hint that we are on the right track is shown by what happens if we “reverse” the kinds of examples exemplified by (36). So alongside (36a) we should consider the following: (43)

*John will arrive tomorrow and/but Mary did (too)

Under the assumptions made earlier, this has the structure (44)

John Mary

Present Past

will

tomorrow X

VP VP

(antecedent clause) (clause with elision)

Here, Coherence will force X to be anaphoric to everything between Present and VP in the antecedent clause, yielding a nonsensical interpretation (*Mary Past will tomorrow arrive). Tellingly, (43) can be made grammatical by inserting an adverb like already: (45)

John will arrive tomorrow and/but Mary already did

If we assume that tomorrow ≥ already in F-structure terms, then (45) has the following structure: (46)

John Present Mary

will Past

tomorrow already

X

VP VP

(antecedent clause) (clause with elision)

Here Coherence will force X to be anaphoric to only the material lower than Falready in the antecedent clause, and so will not contain tomorrow, since

332

Edwin Williams

tomorrow ≥ already. So the adverb already serves to “cap” the size of the silent material adjacent to the elided VP that Coherence applies to. Coherence thus explains why the adverb is essential in distinguishing the grammaticality of (43) from (45). It is not enough to say that the time adverb in (45) is obligatory to meet some kind of parallelism or balanced-focus requirement. (36b) itself shows that no adverb is needed in the second clause simply because there is an adverb in the first clause. Rather, the reason that no adverb is needed in (36b) is exactly because the size of the silent material X adjacent to the elided VP is already “capped” by the auxiliary have, but in (43) there is no such independent capping, so an adverb is needed to do that job. Thus the distinctions drawn are quite fine, but again, I believe that these few examples can do no more than point the way to a more thorough investigation. The promise of such investigation is that Coherence might give another avenue of research into the character of F-structure.

7 Some relevant differences between pronouns and ellipsis The pronoun one (as in “a good one”, and the second one in “one good one”) is of special interest in our inquiry because it can occur in predicative NP positions, and in those positions can have either predicative or DP-contained antecedents; and when it occurs in DPs, it also can have either predicative or DP-contained antecedents: (47)

a.

John is a taxidermist, and I saw one downtown yesterday as well

Pred, Ref

b.

I saw a taxidermist downtown, and Joe is one too

Ref, Pred

c.

John saw a taxidermist and I saw one too

Ref, Ref

d.

John is a taxidermist and I am one too

Pred Pred

The promiscuity in (a) and (b) is perhaps surprising given what we know from previous sections: that predicative and referential NPs are not sharable nor can they enter into mixed ellipsis relations. But given the Coherence principle, it is not at all surprising, as Coherence is only triggered by silent elements adjacent to silent anaphors.

Multidomination and the Coherence of Anaphors

333

The following facts are relevant: (48)

a.

John has a good taxidermist, and I have one too.

b.

John knows a local taxidermist, and I think I can find one in Tuscon.

c.

John has a good taxidermist, and I do too.

The most natural interpretation of (a) is that one = “good taxidermist”, not simply “taxidermist”, and this suggests that one is forcing Coherence. But in (b), most naturally one = “taxidermist”; in fact, one = “local taxidermist” does not even make sense. Importantly, “in Tuscon” is not part of the NP that one is head of, rather it is a VP modifier; it could, for example, be replaced with “when I go to Tuscon”. So it seems that nonsilent anaphors do not in general trigger Coherence of the kind illustrated in (25) with ellipsis. And even if (a) exhibits a preference for including the adjective in the antecedent, it should be compared with (c) where the adjective is absolutely included. We do not need to redefine Coherence in light of these findings, it already anticipated them, as did the versions in Williams (1997) and (2010); but they do confirm it.

8 Grosz’s example Coherence disallows Grosz’s micro-sharing analysis of his example ((10) above): (49) [ CNN claims that a man3 is __ ] and [the BBC claims that a woman7 isn’t __ ] likely to be assassinated t3/7 by the serial killer. Grosz (2009) The trace, being silent and adjacent to shared material, must be shared. This means that neither “a man” nor “a woman” can be its antecedent. As an alternative to micro-sharing, I would suggest instead that Grosz’s example should be analyzed as a kind of “sloppy identity” and should be handled by one of the number of accounts of that phenomenon. For example, suppose that the shared expression [likely. . .] has the following structure: (50)

[X likely tX to be assassinated tX]

and that this expression is interpreted as a closed lambda expression: (51)

λX [X likely tX to be assassinated tX]

334

Edwin Williams

See Williams (1977) for such a proposal. The shared constituent is now a property that can be shared as a whole between the two positions in (49), and micro-sharing is not needed. Since micro-sharing is inconsistent with Coherence this is welcome. The challenge to Grosz is to devise cases that are not plausibly open to a “sloppy identity” treatment. I have not been able to find one myself.

9 An open issue: the definition of Adjacency In the following, we find a large contrast between Wh-movement and VP ellipsis: (52)

a.

I don’t know how many rabbits Mary caught t last week and John [ate t] TIME

b.

Mary caught rabbits last week and John did [ ]VP TIME too.

(TIME is the unpronounced position at which adverbs like “last week” would be pronounced). In (a), there is far less pressure to construe John’s rabbit-eating time as “last week” – that is, although TIME is adjacent to t, it does not seem to cohere with it to the same extent as in (b). The obvious way to draw the difference is through the presence of the pronounced verb in (a) versus the unpronounced verb in (b) – “ate” in (a) intervenes between t and TIME and so Coherence is not enforced. Similar remarks apply to RNR’ed structures like: (53)

Mary caught t last week and John ate t, a dozen rabbits

where the tendency for “last week” to be shared is weak at best; as pointed out by a reviewer, (53) contrasts with (54)

Mary caught t t, and John ate t t, a dozen rabbits last week

in which “last week” is obligatorily shared, as discussed earlier. If we adopt the suggestion that (52a) and (52b) differ because the intervention of the pronounced verb blocks Coherence, then we cannot appeal to Coherence for (54), as we did earlier, and so an alternative account of that case must then be sought. One possibility is that the RNR “pivots” (here, “a dozen rabbits” and “last week”) are both outside of the conjunction of clauses, and that the sharing of “last week” is due purely to the “across the board” constraint, rendering Coherence possibly superfluous for this case. This position of the pivots in RNR structures

Multidomination and the Coherence of Anaphors

335

(inside the last conjunct vs. outside the whole conjunction) is an ongoing controversy, at least in a language like English, but work by Sabbagh (2007) based on quantifier scope and “antecedent contained deletion” phenomena strongly supports the second possibility, making viable an ATB account of the obligatory sharing in (54). In that case, in (52), only (b) would fall under Coherence.

Appendix: Is Coherence the same as MaxElide? Several people (irritating p.c.) have suggested that the principle of Coherence is akin to MaxElide, as proposed in Merchant (2008): (1)

MaXElide (Definition): Let XP be an elided constituent containing an A’-trace. Let YP be a possible target for deletion. YP must not properly contain XP (XP not a proper subset of YP).

The principle is meant to explain contrasts like the following (2)

a.

They said Nick heard about a Balkan language but I don’t know [CP which (Balkan language) [IP they [VP said [CP Nick [VP heard about twh ]]]]].

b. *They said Nick heard about a Balkan language but I don’t know [CP which (Balkan language) [IP they did [VP say [CP Nick [VP heard about twh ]]]]]. Such contrasts are real, but do not follow from Coherence, and the facts that follow from Coherence do not follow from MaxElide. See Williams (1977) for a different treatment of the contrast in (2) in terms of free variables in ellipsis sites. To see that MaxElide is an entirely different principle from Coherence, let’s generalize it as much as possible: (3)

MaxElide2.0: Let XP be an elided constituent, and let YP be a possible target for ellipsis. If XP is in YP then YP must be elided.

MaxElide2.0 says that once you start eliding you can’t stop, constrained only by identity. MaxElide2.0 is obviously too general, but so much the better to make my point. Now consider the following:

336 (4)

Edwin Williams

a.

?John arrived today and Bill did today too.

b.

~&John arrived today and Bill did too.

MaxElide2.0 rules out (a), because today could have been elided, but wasn’t; Coherence says nothing about (a), since today is not silent. On the other hand Coherence insists that (b) is not ambiguous (i.e. Bill arrived today), but MaxElide2.0 says nothing about it. So the principles are entirely different. (By the way (a) is also ruled out by the Disanaphora principle of Williams (1997); see also Williams (2009)).

Bibliography Abbott, B. 1976. “Right Node Raising as a Test of Constituenthood” Linguistic Inquiry 7, 639– 642. Cinque, G. 1999. Adverbs and functional heads: a cross-linguistic perspective, Oxford University Press. Engdahl, E. 1986. Constituent Questions D. Reidel Publishing Company, Dordrecht. Montague, R. 1973. “The Proper Treatment of Quantification in Ordinary English” in Jaakko Hintikka, Julius Moravcsik, Patrick Suppes eds., Approaches to Natural Language, Dordrecht, 221–242. Merchant, J. 2008. “Variable island repair under ellipsis,” in Topics in Ellipsis, edited by K. Johnson. Cambridge: Cambridge University Press, 132–153. Grosz, P. 2009. “Movement and Agreement in Right-Node Raising Constructions” ms MIT. Fox, D. 2002. “Antecedent-Contained Deletion and the Copy Theory of Movement” Linguistic Inquiry 33.1, 63–96. Johnson, K. 2012. “Toward deriving differences in how Wh movement and QR are pronounced” Lingua 122:529–553. Sabbagh, J. 2007. “Ordering and linearizing rightward movement”. Natural Language and Linguistic Theory 25, 349–401. Vergnaud J.-R. 1974 French Relative Clauses MIT dissertation. Williams E. 1977. “Discourse and Logical Form” Linguistic Inquiry 8.1, pp. 101–139. Williams E. 1983. “Syntactic vs. Semantic Categories” Linguistics and Philosophy 6.3, 423–446 Williams E. 1997. “Blocking and Anaphora” Linguistic Inquiry 28.4 577–628. Williams E. 1998. “Economy as Shape Conservation”, MIT Press’s “N. Chomsky’s 70th Birthday Celebration” Web Page, J. Keyser and K. Hale, eds. Williams E. 2003. Representation Theory MIT Press. Williams E. 2009. “The Locality of Focus and the Coherence of Anaphors”, Information Structure, Ad Neeleman and Ivona Kucera eds., Cambridge University Press. Williams E. 2010. Regimes of Derivation in Syntax and Morphology, Routledge, Leading Linguists series. Williams E. 2011. “Generative Semantics, Generative Morphosyntax” Syntax 16.1.

Language index Alemannic 7, 8, 11, 22 n., 27 n. Al-Sayyid Bedouin Sign Language 149 Bantu 146, 215, 267 n. Basque 93, 141 Bavarian 7–36 Catalan 28, 203, 241, 243, 245, 253 Chinese 145, 149–151 – Mandarin 156 Cilungu 215, 228–230 Congo-Saharan 145 Czech 5, 69, 277–8, 280–6, 288–90, 292–4, 296 – Moravian 284 Dene-Caucasian 145 Dutch 10, 16, 18, 29, 41–2, 46, 54, 57–8, 60, 63, 65, 70, 76–8, 80–1, 121–131, 137–8, 148, 151, 201, 203, 207, 239, 241, 243, 245 – Giethoorn 77 – Groningen 76 – Hooghalen 77 – Onstwedde 77 – Utrecht 76 English 5, 28, 63 n., 69 n., 77, 85, 89, 90 n., 95–6, 99, 103, 112 n., 123, 130, 149–51, 153, 155, 158, 168, 185, 199, 201–3, 212 n., 228–31, 234, 239, 240 n., 155–6, 258–60, 262, 265–7, 271–4, 277–294, 296–312, 335 – Yorkshire 246 n. – British 255–6 Finnish 273 Flemish 16, 18, 239, 245 – West 16 French 5, 28, 102, 157, 198–203, 207, 239– 253, 273, 278–9, 282–3, 288–91, 296 n., 304, 308 n., 311 n. – Standard 240–53 – Vimeu Picard 249–53 Frisian 16, 18, 31

German 7–39, 61, 76 n., 91, 96, 122–3, 157, 204, 239–41, 245, 247, 266, 267n. – Low 28–9 Greek 28, 89–90, 93 Gungbe 87–88, 90, 93–4 Hindi 28 Homesign 149–50 Hungarian 26 n., 32, 83, 88–9, 93, 96–8, 101, 103, 104 n., 110–1 Ijo 148 Indo-Pacific 145 Italian 5, 28, 85–6, 93–6, 98, 100, 111, 150– 3, 156, 158, 167–70, 174, 176–7, 179 n., 187, 201, 203, 207–8, 225, 262 – Càsola 179 – Castellazzo Bormida 180 – Chioggia 180 – Colle S. Lucia 180 – Comeglians 180 – Faeto 179 – Forni di Sotto 180, 186 – Grumello 171 – La Strozza 179 n., 180 – Livo 171, 176, 179 – Mulegns 171 – Odolo 147, 171 n., 180 – Passirano 171 n., 180 – Pozzaglio 177 n., 180 – Prali 172, 176, 179 – Sarre 179 – Sillano 171 n., 179 – Tuenno 170 n., 180 – Vermiglio 170 n., 180 – Villa di Chiavenna 180 – Zonza 184 Japanese 110, 153, 201–4, 208, 228–31, 264, 270, 272, 283 Kashmiri 93 Khoisan 145

338

Language index

Lithuanian 248 Marathi 28 Mohawk 141 Ngizim 94 Norwegian 69, 273 n. – Stavanger 69 Nostratic-Amarind Persian 28 Polish 28, 105–6, 203, 248 Portuguese 28 Proto-Niger-Congo 145–6 Punjabi 28 Putonghua 260, 262, 264–7, 270–2 Romanian 97 n. Russian 28, 93, 105, 106 n., 200–1

Serbo-Croatian 85, 101–111 Slovenian 248 Spanish 28, 150–1, 201, 203, 278 n., 283, 294 n., 308 n., 311 n. Swiss German 9, 29–9 Tangale 94 Tlingit 87, 93 Tuki 86, 94 Turkish 150–3, 157, 272 n., 273 Uto-Aztecan 145 Walloon 245–249 – Liège 246, 248 – Bastogne 246 Warlpiri 141 Welsh 70 Yiddish 28–9 Yoruba 228

Subject index Acquisition 42, 86, 141–2, 144, 146–50, 153, 158, 195–209, 221, 239, 251, 300 Atypical 147–50 Alternative Realization 286, 294–313 Bilingualism 207 Binding 5, 33, 88, 120, 255–276, 295, 317 Brain-Imaging 205 Case Resistance Principle 130 Chain uniformity 8, 33–4 Clausal typing 16, 19, 28 n., 58 n., 86–7, 91, 97, 99, 100–2, 104–5, 107–8, 115–139, 156, 170, 185–6, 292, 328 Clitics 5, 8, 19–27, 31, 33, 34 n., 166–93, 283 n., 296 n., 311 – subject 166–193 Cogeneration 318 Coherence 149, 315–336 Complementizer (COMP) 7–39, 101 n., 183, 289 n., 305 n. – inflection 8, 19–27, 35, 305 n. Complexity 10, 14, 18, 22, 42–3, 48–81, 90, 126, 136, 138, 144, 151–2, 169, 173, 196–7, 224–6, 232, 255–60, 265, 281, 285, 290 n. Concord 72 – expressive – see Concord Conditional 277, 279–82, 285–92, 297–8, 305–6, 310 Coordination 42, 54, 58, 60–81, 125, 131–4, 137 – asyndetic 53, 57 n., 65, 67–8, 74 – monosyndetic 63 n., 65 – unbalanced 69–74 Copula 282–5, 300–2, 305–6, 308–9, 312 Diphthongs 22, 255–267 Distributed morphology 54, 168–9, 176 Diversity 85–111, 142–7, 149, 154, 159, 168 Doubly Filled COMP Filter (DFCF) 7–39 Element Theory (ET) 3, 220–236, 255–74 Ellipsis 78, 122, 281, 283, 295, 299 n., 305, 324–5, 332–6

Exo-skeletal model 54 Externalization 43, 68, 75, 78–9, 152, 174, 180–2, 186, 189–90 Faculty of Language 139, 195, 174, 196, 211, 219 Features – categorical 1–3, 5, 15–9, 26, 31, 34–5 – inflectional 19, 26, 35, 54, 117, 177–8, 168, 182, 277–312 – phonological 2, 3, 19, 70–120, 211, 213, 215, 217, 218–21, 227, 243–5, 253 Final Devoicing 76 n., 239–254 Focus position 27 n., 85–111, 118–9, 130, 283, 305, 323, 332 Government Phonology 3, 226–230, 234, 244, 255–274, 267 n. Imperatives 91–2, 184–5, 286–8 Implicational universals 172–4, 176, 180 Invisible Category Principle (ICP) 295–7, 301, 303 Language Acquisition Device (LAD) 197, 199 Level embedding 318 Locality 47, 134, 137, 139, 271–2, 299 Melodic Structure 214–36, 255–74 Merge 1–3, 7, 12–18, 21–26, 30, 32–35, 45, 50, 53–4, 56 n., 58, 118, 138, 143–4, 214, 292, 295 n., 318 – external 13–15, 18, 24–26, 35 – internal 13–15, 17, 21, 30–1, 34–5, 138 – re 12, 17, 32, 34 Micro-sharing 320, 322, 333–4 Modality 143–4, 152–3, 186, 278–81 Modals 91–2, 186, 197–8, 277–281, 286–7, 291, 297, 304–5, 308–9, 311–2 Montague grammar 318 Morphological hierarchies 92–3, 175–6, 190–1, 218–9 Movement – copy theory of 49, 117–8, 120–5, 315–7, 320–1, 326, 328, 333, 335

340

Subject index

– head 7–35 – multidomination theory of (MD) 315–320, 328; see also Multi-dominance – phrasal 7–8, 85–111, 315–336 Multi-dominance 3–5, 315–336; see also Multidomination Negation 95–6, 287, 298, 303, 305, 312 – clausal 281–2 – partial (VP) 184–5, 282 Paragogic alveolar obstruent 75–8, 80 Parallel structure 67, 90, 109, 115–139, 312, 317, 332 Periphery 62, 134, 232 – left 17 n., 22, 45–6, 85–8, 93–9, 103–5, 108, 110–1, 130, 183 Projection 2–3, 7 n., 12–3, 14 n., 18 n., 21, 25–6, 31–5, 54, 60–2, 67, 74, 86–90, 93, 95–6, 102, 128–9, 131, 133, 137–8, 169, 177–8, 183, 203, 232, 235, 261, 264, 295 – extended 137, 302, 312 Property sharing 72, 117–8, 120, 138–9, 334, 256 Prosody 3, 20–1, 27, 66, 78, 80, 152, 156–8, 185 n., 200–1, 204–6, 213, 215–8, 222, 226–8, 235–6, 242–3 Reconstruction 115–139, 145 – total 115–6, 118–20, 139 Recoverability 31, 78, 167, 181–2, 187, 189– 90

Recursion 2, 5, 53–4, 60, 63–4, 67, 74, 118, 141, 143, 148, 153, 211–238 Recycling 15, 18, 155–6 Reduplication 54, 72, 75, 217 – reduplicative pattern 51, 71 Relative clauses 28 n., 87, 115–139 – (head) raising analysis of 5, 19, 119–25, 134, 137–9 Right Roof Constraint 127, 131, 134, 137 Roots – bare 4, 41–83 – category 41–83, 88, 93, 95–6, 99 n., 212 n., 216, 220, 227 R-pronoun 10, 49–50, 53, 121–3, 125, 129, 134 Sonority 208, 244 Structure preserving 117, 131, 135, 147, 202, 244 Superiority 89, 101–11 Too Many Repairs Problem 250 Trace conversion 315–6 Uniformity thesis 167 Universals 26 n., 42, 89, 98, 100, 144–50, 167, 228, 255, 286, 291, 304–5, 311–13 Wackernagel position 20, 26 Wh-fronting 90, 92, 94–6, 98, 101–11 – multiple 98, 101–11 Word order 5, 63 n., 141–59, 204, 311