185 99 1MB
English Pages 233 [234] Year 2015
Andreas Trotzke and Josef Bayer (Eds.) Syntactic Complexity across Interfaces
Interface Explorations
| Editors Artemis Alexiadou and T. Alan Hall
Volume 30
Syntactic Complexity across Interfaces
| Edited by Andreas Trotzke and Josef Bayer
ISBN 978-1-61451-785-6 e-ISBN (PDF) 978-1-61451-790-0 e-ISBN (EPUB) 978-1-5015-0101-2 ISSN 1861-4167 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. © 2015 Walter de Gruyter Inc., Berlin/Boston/Munich Typesetting: PTP-Berlin, Protago TEX-Production GmbH Printing and binding: CPI books GmbH, Leck ♾Printed on acid-free paper Printed in Germany www.degruyter.com
Contents
1
Andreas Trotzke and Josef Bayer Syntactic complexity across interfaces | 1
2
Uli Sauerland Against complexity parameters | 9
3
Jan-Wouter Zwart Top-down derivation, recursion, and the model of grammar | 25
4
Leah S. Bauke What small clauses can tell us about complex sentence structure | 43
5
Werner Frey and Hubert Truckenbrodt Syntactic and prosodic integration and disintegration in peripheral adverbial clauses and in right dislocation/afterthought | 75
6
Marlies Kluck On representing anchored parentheses in syntax | 107
7
Tonjes Veenstra The development of subordination | 137
8
9
Tom Roeper Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition | 163 Misha Becker Learning structures with displaced arguments | 193
Index | 227
Andreas Trotzke and Josef Bayer
1 Syntactic complexity across interfaces 1.1 Introduction Syntactic complexity has always been a matter of intense investigation in formal linguistics. Since complex syntax is clearly evidenced by sentential embedding and since embedding of one sentence in another is taken to signal recursivity of the grammar, the capacity of computing syntactic complexity is of central interest to the recent hypothesis that syntactic recursion is the defining property of natural language (Hauser et al. 2002). In the light of more recent claims according to which complex syntax is not a universal property of all living languages (Everett 2005), the issue of how to detect and define syntactic complexity has been revived with a combination of classical and new arguments (Nevins et al. 2009). The existing collections on the nature of syntactic complexity either deal with syntactic complexity from a functional-typological perspective (Miestamo et al. 2008; Sampson et al. 2009) or place a premium on the property of syntactic recursion (van der Hulst 2010; Sauerland and Trotzke 2011; Roeper and Speas 2014). In contrast, the current volume makes a new contribution to the ongoing debate by taking into account the recent development in linguistic theory to approach UG ‘from below’ by referring to both grammar-internal and grammar-external interfaces when explaining design features of the human language faculty (Chomsky 2007). According to this shift in perspective, it is reasonable to assume that UG only contains properties such as recursive Merge, binary branching structure, and the valued-unvalued feature distinction. All other properties of grammar might follow from the interaction between UG and other components within the model of grammar (the phonological and the semantic component; i.e. grammar-internal components) and from the interplay between UG and grammar-external components such as the performance and acquisition systems. As for the interaction with grammar-internal components, the new division of labor among the components of the model of grammar raises new issues for defining and detecting syntactic complexity. In particular, the question of the complexity of grammar has to be answered separately for ‘narrow syntax’ and for the grammar as a whole, including the interface components (Trotzke and Zwart 2014). As for the interaction with grammar-external components, Trotzke et al. (2013) show that systematic properties of performance systems (the ‘performance interface’, according to their terminology) can play an important role within the research program outlined by Chomsky (2005, 2007). In particular, investigations of the performance interface can revise current conceptions of UG by relegating widely assumed grammatical constraints to properties of the performance systems, as recently argued, for instance, by Bever (2009) for the Extended Projection Principle or by Hawkins (2013) for the FinalOver-Final Constraint (Biberauer et al. 2014).
2 | Andreas Trotzke and Josef Bayer Given this conceptual background of approaching the issue of syntactic complexity from the perspective of recent linguistic theory, the volume starts with two contributions that deal with the formal complexity of natural languages in terms of the Chomsky hierarchy, the most prominent complexity measure in formal language theory. These two contributions set the scene for the volume by discussing general aspects of grammar architecture and by turning to the question of whether languages can vary as to their formal complexity. The two papers are followed by three contributions that address specific issues of clausal embedding (small clauses, parentheses, peripheral adverbial clauses, and right dislocation/afterthought constructions). The last part of the volume contains three papers that provide accounts of how to address topics revolving around syntactic complexity in terms of grammar-external interfaces in the domain of language acquisition.
1.2 Syntactic complexity and formal language theory In contrast to the recent typological-functional literature, the comparative complexity of languages is not an issue in formal language theory. The question relevant in this context is where the grammar of natural language is to be placed in the ‘Chomsky hierarchy’, a complexity hierarchy of formal languages. In the 1950s, Noam Chomsky developed formal language theory as a mathematically precise model of language. Chomsky established that behaviorist accounts of language were insufficient to account for the computational properties of natural languages, whereas the phrase structure grammars Chomsky introduced stood a chance to be sufficient. In particular, Chomsky (1956, 1959) showed that the property of self-embedding involves the kind of complexity that requires (at least) context-free grammars, rather than less complex types of grammar (specifically, finite-state devices). Following the lead of Chomsky, theoretical linguists developed concrete phrase structure grammars for specific languages. Crucially, and as should be clear from the above, the discussion in formal language theory focuses on general computational properties of ‘narrow syntax’, a core component of the model of grammar that can be equated with the faculty of language in the narrow sense as defined in Hauser et al. (2002). In addition to this component that applies simple rules merging elements, the model of grammar includes interface components dealing with sound and meaning. Accordingly, the question of the complexity of the grammar has to be answered separately for the grammar as a whole and for the individual components (including narrow syntax); with different answers forthcoming in each case. In recent literature, it is an open question which phenomena are to be associated with which component of the grammar, with current proposals relocating seemingly narrow syntactic phenomena such as head movement and inflectional morphology to the interface with phonology (e.g. Chomsky 2001). By discussing notions of formal language theory, the following two contributions investigate which properties of the grammar should be relegated to the
1. Syntactic complexity across interfaces
|
3
interface components and which features of natural language should be considered as belonging to narrow syntax and, therefore, should be evaluated according to the Chomsky hierarchy. In his contribution “Against complexity parameters,” Uli Sauerland addresses the recent proposal that languages can vary concerning their formal complexity in terms of the Chomsky hierarchy. According to Sauerland, such accounts are essentially proposing that this variation is a parameter choice – the ‘complexity parameter’. Sauerland argues that parameterizing languages in this regard is unwarranted and not supported by the evidence. Based on a discussion of languages such as Swiss German, Standard German, and English, Sauerland makes two claims. First, he argues that certain word order differences between these languages should not be addressed in terms of the Chomsky hierarchy. Instead, as Sauerland argues, these variations can be addressed by independently established word-order parameters, belonging to the domain of the phonological interface and not to narrow syntax. After relegating this issue to variation in the domain of linearization, Sauerland turns to a second argument against complexity parameters by referring to the semantics interface. He claims that the semantics of a non-context-free language would need to radically differ from that of a context-free language. Specifically, he argues that the semantics of language is inherently context-free, and, as a consequence, the standard semantics of scope requires at least a memory system that supports context-free grammars. Since Sauerland takes it for granted that the semantics of natural languages should not vary, he concludes that these properties of the semantics interface provide important evidence against complexity parameters. Jan-Wouter Zwart also takes the Chomsky hierarchy as a starting point. In his paper, “Top-down derivation, recursion, and the model of grammar,” he adopts the strong focus on the role of the interfaces from recent minimalist literature and argues that the issue of syntactic complexity of the grammar has to be answered separately for the grammar as a whole and for the individual components (including ‘narrow syntax’). Given this theoretical background, he claims that linguistic recursion should be understood as the interface-related treatment of a complex string as a single item within another complex string. In particular, he demonstrates that this simplex/complex ambiguity is due to separate derivational sequences (‘derivation layers’). He argues that the grammar creating those strings (‘narrow syntax’) may be of the minimal complexity of a finite-state grammar. Zwart claims that competing views suffer from the unmotivated assumption that the rules and principles of grammar are fed by a homogeneous set of symbols. In contrast, he proposes that the symbols in the alphabet/numeration may themselves be the output of separate derivations. Based on this clarification, he concludes that arguments against the finite-state character of generating phrase structure lose their force. As a consequence, the complexity of natural language should not be addressed, in the first place, in terms of the types of grammar rules, but in terms of interaction among derivation layers, crucially involving the interfaces.
4 | Andreas Trotzke and Josef Bayer
1.3 Syntactic complexity and clausal embedding The following three contributions address specific issues of clausal embedding: small clauses, parentheses, peripheral adverbial clauses, and right dislocation/afterthought. The three papers ask to what extent grammar-internal interface conditions and properties can help to detect and define syntactic complexity. Do interface properties concur with the syntactic complexity ascribed to the phenomena in question? Or do interface-related features of the data even exclude an analysis in terms of syntactic complexity? Leah S. Bauke deals with the issue of small clauses, a prominent case for which syntactic complexity is notoriously difficult to define. Working with a minimalist perspective, she focuses on the question of how basic syntactic operations are determined by interface conditions. In her paper “What small clauses can tell us about complex sentence structure,” she argues for a revised analysis of small clauses. In particular, she claims that agreement between the small clause constituents can be established directly upon Merger and need not be mediated by a functional head. However, within minimalist theory, cases of XP-XP Merger are considered problematic because they pose labeling ambiguities. As a consequence, the input to the operation Merge is suggested to be constrained with the effect that at least one element must be or must count as a lexical item. Bauke demonstrates that this constraint poses no problem for her analysis, in which small clauses are generated by direct Merger of the two constituents that make up the small clause. She adopts the approach that complex syntactic objects already merged in the course of the derivation can be shrunk to lexical items, and, based on this account, she proposes an analysis of so far unaccounted for extraction and subextraction patterns in Russian and English small clauses. The contribution by Werner Frey and Hubert Truckenbrodt focuses on the syntax-phonology interface. In their paper “Syntactic and prosodic integration and disintegration in peripheral adverbial clauses and in right dislocation/afterthought,” they analyze different clausal dependencies in German by bringing together their respective work on peripheral adverbial clauses and on right dislocation and afterthought constructions. Frey and Truckenbrodt analyze these phenomena within a single set of analytical assumptions that relate to the notions of ‘integration’ and ‘root sentence’. In the first part of their paper, they demonstrate that peripheral adverbial clauses require high syntactic attachment. Put more technically, peripheral adverbial clauses are either in the specifier of their host clause or are adjoined to their host clause. The authors show that this converges with phonological evidence. Both prosody and information structure of peripheral adverbial clauses reflect their borderline status between integration and disintegration. In the second part, they show that right dislocated or afterthought constituents are ‘added’ to the clause in the sense that they do not occupy a thematic position in their clausal host. However, these constituents show c-command relations like the elements they resume (‘connectedness effects’). Based on evidence from the prosody and information structure
1. Syntactic complexity across interfaces
|
5
of right dislocation and afterthought constructions, the authors show that a syntactic adjunction analysis, if it aims at generalizing across right dislocation and afterthought constructions, cannot represent the properties of disintegration in a principled way. As an alternative, they propose a deletion analysis, which captures both the property of disintegration and the connectedness effects. Marlies Kluck starts her contribution with the observation that syntactic complexity that does not involve subordination, such as coordinate structures and parentheticals, is still poorly understood. In her paper “On representing anchored parentheses in syntax,” she turns to the questions of how and where ‘anchored parentheses’ are represented in grammar. By ‘anchored’ parentheses, Kluck refers to parentheses that are not freely attached somewhere in their host, but are attached at the constituent-level to an ‘anchor’. In this sense, nominal appositions, nominal appositive relative clauses, amalgams, and sluiced parentheticals belong to this category. Contra ‘orphan’ approaches, which put parentheticals outside the domain of syntax, Kluck argues on both conceptual and empirical grounds for anchored parentheticals as represented at the level of syntax. In particular, she provides two reasons for this claim: First, anchored parentheticals must be part of syntax under common assumptions about the model of grammar. Parentheticals are linearized in their hosts and interpreted relative to their hosts. Second, since anchored parentheses are related to a specific constituent in their host, namely the anchor, their integration should not take place at a post-syntactic level.
1.4 Syntactic complexity and the acquisition interface While the contributions sketched above deal with the interaction between syntax and grammar-internal interfaces, the following three papers focus on grammar-external interfaces in the domain of language acquisition. To keep UG as slim and simple as possible, these interfaces have recently been analyzed as external third factor effects (Trotzke et al. 2013), and include, according to Chomsky (2005: 6), “(a) principles of data analysis that might be used in language acquisition and other domains; (b) principles of structural architecture and developmental constraints [. . . ] including principles of efficient computation.” Following Bever (2009: 280), we use the term ‘acquisition interface’ to refer to these grammar-external conditions in the context of language acquisition (i.e. to specific constraints on learnability). Given this interface notion, the following three contributions ask to what extent grammar-external acquisition processes can contribute to the debate on how to detect and define syntactic complexity. The contributions deal in particular with (i) different acquisition processes operative in the development of syntactic subordination, (ii) the identification of different scales of syntactic complexity by means of acquisition devices such as semantic bootstrapping and (iii) the application of minimalist economic principles to the acquisition problem.
6 | Andreas Trotzke and Josef Bayer Tonjes Veenstra argues that the pidgin-creole cycle provides crucial insights into the development of subordination in natural language. In his paper “The development of subordination,” he starts with the observation that interlanguage varieties and pidgins both lack subordinative structures, and that creoles, by contrast, do exhibit such structures. On the basis of a comparison between Saramaccan (an English/Portuguese-based creole spoken in Suriname) and Fòngbe (its major substrate language), Veenstra shows that the creole patterns cannot be accounted for by substrate influence alone. Concentrating on sentence-embedding predicates and their associated syntax, he argues that the mismatches between the creole and its substrate are due to processes of incipient (second) language learning. Given different acquisition processes operative in creole genesis, Veenstra claims that incipient learning, i.e. early second language acquisition, plays a substantial role. Specifically, he argues that incipient learning accounts both for variable selection patterns of clausal embedding and for the absence of morphologically marked verb forms in embedded contexts. On the other hand, non-incipient learning, i.e. more advanced second language acquisition, accounts for the appearance of an unspecified subordinator. Relexification can explain the specific properties that this subordinator exhibits. Tom Roeper also focuses on the acquisition interface. In his paper “Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition,” he discusses the claim that both small clauses and infinitives lack a CP. More specifically, he focuses on the fact that Germanic languages generally do not permit a wh-word in an indirect infinitival question, while English, as an exception, does. In the context of evidence from first language acquisition, Roeper argues that the child acquiring English can posit a zero scope-marker in the matrix CP and need not posit a new infinitival CP. In support of this view, Roeper presents experimental evidence according to which the child more often interprets the medial wh- as covertly moved to the matrix CP with an infinitive than with a tensed clause. In addition, he refers to the notion of ‘acquisition efficiency’ and postulates ‘Avoid Phase’ as a general economic principle that converges with assumptions in minimalist theory. Adopting the notion of periodic ‘Transfer’ of syntax to a semantic interface interpretation, he claims that periodic Transfer is psychologically costly, and, consequently, the child should limit the number of Transfers by limiting the set of phases (thus the term ‘Avoid Phase’). In other words, the child will maintain a more economical representation by positing as few phases as possible. Given this background, Roeper argues that tensed clauses initiate an interface transfer, while the default representation of infinitives, lacking a CP, does not. Like Roeper, Misha Becker deals with data from first language acquisition. In her paper “Learning structures with displaced arguments,” she starts with the assumption that sentences in which an argument has been displaced with respect to the position associated with its thematic/semantic role are more complex than sentences without such displacement. By focusing on this issue, Becker looks at how children acquire two constructions that involve such a more complex alignment of thematic relations: ‘raising-to-subject constructions’ and ‘tough-constructions’. Becker claims that inan-
1. Syntactic complexity across interfaces
|
7
imate subjects provide a clue to the detection or postulation of a complex structure. Essentially, she extends the scope of semantic bootstrapping and argues that not only can an animate NP serve as a cue to canonical subjecthood, but an inanimate subject can serve as a cue that the subject is displaced, and therefore that the structure is more complex. Becker supports her claims with two types of data: (1) naturalistic input data (child-directed speech from the CHILDES corpus) and (2) controlled experimental input data in simulated learning tasks with both children and adults. Acknowledgement: We thank the participants at the workshop on Complex Sentences, Types of Embedding, and Recursivity (05–06 March 2012, University of Konstanz). Andreas Trotzke acknowledges financial support from the DFG Excellence Initiative (University of Konstanz, project no. 65411). We are grateful to all the scholars who volunteered as reviewers of the contributions to this volume: Markus Bader, Nicole Dehé, Marcel den Dikken, Doreen Georgi, Jeff Good, Jeremy Hartman, Katharina Hartmann, Fabian Heck, Riny Huybregts, Göz Kaufmann, Vincenzo Moscati, Andreas Pankau, Martin Salzmann, and Eva Wittenberg. Last but not least, we thank Uwe Braun for editorial help.
References Bever, Thomas G. 2009. Remarks on the individual basis for linguistic structures. In Of minds and language: A dialogue with Noam Chomsky in the Basque country, Massimo Piattelli-Palmarini, Juan Uriagereka and Pello Salaburu (eds.), 278–295. Oxford: Oxford University Press. Biberauer, Theresa, Anders Holmberg and Ian Roberts. 2014. A syntactic universal and its consequences. Linguistic Inquiry 45: 169–225. Chomsky, Noam. 1956. Three models for the description of language. IRE Transactions of Information Theory IT-2 3: 113–124. Chomsky, Noam. 1959. On certain formal properties of grammars. Information and Control 2: 137– 167. Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A life in language, Michael J. Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. Chomsky, Noam. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22. Chomsky, Noam. 2007. Approaching UG from below. In Interfaces + recursion = language? Chomsky’s minimalism and the view from syntax-semantics, Uli Sauerland and Hans-Martin Gärtner (eds.), 1–29. Berlin/New York: Mouton de Gruyter. Everett, Daniel L. 2005. Cultural constraints on grammar and cognition in Pirahã: Another look at the design features of human language. Current Anthropology 46: 621–634. Hauser, Marc D., Noam Chomsky and W. Tecumseh Fitch. 2002. The faculty of Language: What is it, who has it, and how did it evolve? Science 298: 1569–1579. Hawkins, John A. 2013. Disharmonic word orders from a processing-efficiency perspective. In Theoretical approaches to disharmonic word orders, Theresa Biberauer and Michelle Sheehan (eds.), 391–406. Oxford: Oxford University Press. van der Hulst, Harry (ed.). 2010. Recursion and human language. Berlin/New York: Mouton de Gruyter.
8 | Andreas Trotzke and Josef Bayer Miestamo, Matti, Kaius Sinnemäki and Fred Karlsson (eds.). 2008. Language complexity: Typology, contact, change. Amsterdam/Philadelphia: Benjamins. Nevins, Andrew, David Pesetsky and Cilene Rodrigues. 2009. Pirahã exceptionality: A reassessment. Language 85: 355–404. Roeper, Tom and Margaret Speas (eds.). 2014. Recursion: Complexity in cognition. Dordrecht: Springer. Sampson, Geoffrey, David Gil and Peter Trudgill (eds.). 2009. Language complexity as an evolving variable. Oxford: Oxford University Press. Sauerland, Uli and Andreas Trotzke. 2011. Biolinguistic perspectives on recursion: Introduction to the special issue. Biolinguistics 5: 1–9. Trotzke, Andreas, Markus Bader and Lyn Frazier. 2013. Third factors and the performance interface in language design. Biolinguistics 7: 1–34. Trotzke, Andreas and Jan-Wouter Zwart. 2014. The complexity of narrow syntax: Minimalism, representational economy, and simplest Merge. In Measuring grammatical complexity, Frederick J. Newmeyer and Laurel B. Preston (eds.), 128–147. Oxford: Oxford University Press.
Uli Sauerland
2 Against complexity parameters
Recent work in linguistics proposes that languages can vary as to their formal complexity in terms of the Chomsky hierarchy. In this paper, I argue that this proposal for the parametrization of languages is unwarranted and not supported by the evidence. I address three types of languages: 1) languages where there is evidence for a complexity greater than context-free (Swiss German, Dutch), languages where there is no evidence that their complexity exceeds that of context-free grammars (Standard German, English) and languages where there is no evidence that their complexity exceeds that of regular grammars (possibly Pirahã (Everett 2005) and Teiwa (Klamer 2010), but also large fragments of English). However, all differences concern only weak generative capacity. I first argue that the differences between the three types of languages can be easily addressed by established word-order parameters rather than a complexity parameter. Secondly, I argue that, if one still wanted to maintain the claim that there are languages parametrized to be less complex than context-free, the semantics of such a non-context-free language would need to radically differ from that of a context-free language. Specifically, I argue that the semantics of embedding and other types of scope is inherently linked to non-finite notions of memory such as push-down automata. In this sense, the semantics of language is inherently context-free. This observation makes it easy to test for context-freeness.
2.1 Introduction The study of the formal complexity of language has been an important topic within generative linguistics since its inception (see Trotzke 2008; Lobina Bona 2012; Fitch and Friederici 2012 for recent reviews). Initially, formal complexity considerations have been invoked to rule out theories of language because the theories in question predicted a specific level of formal complexity to be impossible for language (Chomsky 1957; Huybregts 1984; Shieber 1985). More recent work in this tradition compares directly humans and other species using artificial grammar learning. This work is also oriented at finding the level of complexity the human language faculty can attain (Fitch and Hauser 2004; Gentner et al. 2006; Abe and Watanabe 2011; Beckers et al. 2012). Research in neurolinguistics that compares neural activity in linguistic and artificial grammar learning tasks underscores the relation between these two types of work (Friederici et al. 2011). A second line of argument using the term complexity has emerged mostly within historical lingustics (Deutscher 2005; Givón 2009; Heine and Kuteva 2007; Sampson
10 | Uli Sauerland et al. 2009; Wray and Grace 2007), though also in some synchronic work (Everett 2005).¹ McWhorter (2001) proposes furthermore that creole languages lack complexity (but see DeGraff 2003; Bane 2008; Veenstra this volume). At least some of this work targets the notion of formal complexity characterized by the Chomsky hierarchy just as the work discussed in the previous paragraph (see also Trotzke and Zwart 2014 on the relation between the two lines of research). However, the work just mentioned claims that individual languages are parameterized for different levels of formal complexity. So, this discussion also assumes that all humans are capable of learning all human languages. But it proposes that specific languages vary as to which level of the complexity hierarchy they occupy and furthermore that this variation is not just coincidental, but a parameter choice – the complexity parameter. The view that formal complexity is a parameter of language is quite different from the use of formal complexity only to rule out some models of insufficiently rich models of language like regular grammars. In the following, I use the term Complexity Parameter View for this proposal. I attribute this view to Deutscher (2005); Everett (2005); Givón (2009); Heine and Kuteva (2007); Wray and Grace (2007) as all seven authors claim that substantial, interesting variation between languages can be explained by appeal to a notion of syntactic complexity according to which self-embedding of sentences is more complex than concatenation of two independent sentences.² In this sense, the work differs from other work in syntax using only intuitive notions of linguistic complexity (Dahl 2004; Hawkins 2004), which are difficult to evaluate. Selfembedding, however, was shown by Chomsky (1959) to be the crucial property distinguishing context-free languages from regular languages.³ Some of the works may be reinterpreted to view formal complexity not as a parameter, but instead claim that variation in formal complexity is an epiphenomenon of other parameter settings, for instance word-order settings – the view I advocate below. But, as far as I can see, this is not the view actually taken: At least one author (Everett 2005) explicitly claims a restriction that can be characterized directly by the notion of recursion. We argue in Sauerland and Trotzke (2011) that the discussion about recursion following Hauser et al. (2002) only makes sense if we understand the term ‘recursion’ as self-embedding. Furthermore, the seven papers mentioned appeal to a progressive trajectory of increasing formal complexity in natural language syntax in the course of human develop-
1 In language acquisition, too, formal complexity has been invoked. Frank (1998) proposes that children’s language acquisition can be insightfully modeled by different levels of the complexity hierarchy. Roeper and Snyder (2005); Roeper (2011) argue that recursion in compounds must be learned. These proposals are not addressed in this paper, since they primarily concern maturation of language, not the parametrization of mature language. 2 See also work by Jackendoff and Wittenberg (2014). 3 Recall that Chomsky (1959) defines self-embedding as a type of center-embedding on string basis, not in phrase structural terms.
2. Against complexity parameters
| 11
ment.⁴ As far as I can see, the idea of such a trajectory requires a commitment to a complexity parameter. The relevant concept of formal complexity for this paper is that of the Chomsky hierarchy (Chomsky 1956, 1959).⁵ In effect, only three classes of the hierarchy are relevant for the following: 1) regular grammars (also called finite state grammars, Chomsky and Miller 1958) represent the level of lowest complexity on the hierarchy. 2) Context-free grammars (also called phrase structure grammars) are at the level of intermediate complexity. Finally, 3) context-sensitive grammars represent the highest level of complexity relevant for the discussion. The level of context-sensitive grammars may be restricted to the mildly context-sensitive grammars that Joshi (1985) introduced (see Joshi 2003; Graf 2013). Concerning the lower limit of human ability on the complexity, Chomsky (1957) using English evidence showed that language could not be analyzed using only regular grammar formalisms, but required at least contextfree grammars. Furthermore, Huybregts (1976, 1984) and Shieber (1985) used Dutch and Swiss German evidence to show that language goes beyond the context-free level of complexity. Finally, Joshi (1985) hypothesized an upper limit; namely, that mildly context-sensitive grammars represent the highest level of complexity that language can attain (see Michaelis and Kracht 1997; Bhatt and Joshi 2004 for discussion). For identifying the three levels of formal complexity, it is useful to keep the three formal grammars illustrated in (1) in mind. The (ab)n grammar is the regular grammar used in much of the work in artificial grammar learning starting with Fitch and Hauser (2004). But for comparison with the grammars in (1b) and (1c), it is more conspicuos to use a grammar that produces sequences of pairs of two identical terminals. This process is also clearly finite state, as it requires only memory for one symbol. In contrast to this grammar, the mirror grammar in (1b) requires an unlimited push-down memory and is therefore at the context-free level of the Chomsky hierarchy. Finally, the copy grammar (1c) is beyond context-free, but mildly context-sensitive.⁶ (1)
a. b. c.
(xx)n : aa, aabb, aabbcc, aabbccdd, . . . mirror grammar: aa, abba, abccba, abcddcba, . . . copy grammar: aa, abab, abcabc, abcdabcd, . . .
4 Evidence of a superficially similar trajectory exists in other domains: Hay and Bauer (2007) report a positive correlation of phonemic inventory of a language and the size of the population of speakers of the language. Since all languages in prehistoric times were spoken only by a small group, these results entail a positive correlation of historic development and phoneme inventory size, which can be regarded as a form of complexity though Hay and Bauer (2007) do not use this term. Furthermore, Maddieson (2005a,b, 2011) reports a positive correlation of consonant inventory and syllable complexity. 5 Zwart (this volume) assumes a slightly different conception of the distinction between phrasal and terminal symbols, but also derives at the finite state vs. context-free grammar distinction. 6 One reviewer points out that the notation I use here for the pair grammar, (xx)n , is not standard. However, there is no standard comprehensive formula for this grammar as far as I know.
12 | Uli Sauerland Consider now the variation in the expression of infinitival complement clauses in West Germanic: (1) and (2) show representative examples from three West Germanic languages: Swiss German, Standard German, and English (cf. Wurmbrand 2004; Schmid 2005). (We follow the frequent practice to present Swiss and Standard German examples in the word order that they occur in an embedded clause, abstracting away from the effect of verb-second in the matrix clause. This is indicated by the initial ellipsis dots in the following.) (2)
a.
Swiss German (Shieber 1985: 334) . . . mer em Hans es huus hälfed aastriiche . . . we the.dat Hans the house helped paint
b.
c. (3)
a.
Standard German . . . wir dem Hans das Haus anzustreichen halfen . . . we the.dat Hans the house paint helped English We helped John to paint the house. Swiss German (Shieber 1985: 335) . . . mer d’chind em Hans es huus lönd hälfe . . . we the children.acc the.dat Hans the house let helped aastiiche paint
b.
German . . . wir . . . we helfen help
c.
die Kinder dem Hans das Haus anstreichen the.acc children the.dat Hans the house paint ließen let
We let the children help John paint the house.
In both Swiss German and Standard German, there is a syntactic relationship of case assignment between the verb and its object as shown graphically in (4) for the examples in (2): specifically, hälfe/helfen (‘help’) assigns dative case to its nominal object, while aastriche/anzustreichen (‘paint’) does not and therefore the object receives accusative case. Case assignment therefore provides a purely syntactic argument that the grammars underlying the Swiss German and Standard German data must be essentially the copy grammar for Swiss German with its nested dependencies and something like the mirror grammar for the Standard German data with its nested dependencies. Though English has no Dative case marking, even if it did, a regular grammar of the (xx)n type would be sufficient to generate the grammatical English strings involving infinitival complementation as the representation in (4c) shows.
2. Against complexity parameters
(4)
a.
we [the Hans]dat [the house]acc helped paint
b.
we [the Hans]dat [the house]acc paint helped
c.
we helped John paint the house
| 13
For adherents of the complexity parameter view, the natural explanation for the variation in (2) and (3) is to link it to the complexity parameter. In fact, it is often assumed that children start language acquisition with a default parameter setting which has to be the most restrictive setting. Then, children in the face of positive evidence for a different parameter setting in the target language change parameters to a less restrictive setting (Crain 1991). If this view is combined with the complexity parameter, the default setting of the complexity parameter would have to be the one restricted to regular grammars. Children acquiring Standard German would then at some point encounter examples such as those in (2b) and (3b) and reset their complexity parameter to allow context-free grammars. Swiss German speakers would also start out with the regular grammar restriction, and then possibly first move on to the context-free setting, but ultimately would realize that their target language shows evidence for construction beyond context-free. Of course, in English grammar some phenomena exist that also require at least context-free complexity as Chomsky (1957) showed (involving for example relative clauses, either . . . or , or if . . . then). However, the relevant types of examples seem rather rare. For the following argument, I assume (possibly counterfactually) that some English children never encounter such sentences. So, in a sense, we are actually considering a language that shares the syntax of infinitival complementation with English, but does not share the English syntax of relative clauses and other constructions that provide evidence for context-freeness in English. In the following, I will call this hypothetical language the Regular Fragment of English – i.e. a subset of English sentences such that it can be weakly generated by regular rules and includes infinitival complements. In the two following sections, I address two potential applications of the complexity parameter view. In the first section, I consider the difference between languages where there is no evidence against a restriction to context-free grammars (e.g. Standard German) and those where there is such evidence (e.g. Swiss German). I first argue that this case has just as much legitimacy as the one of a regular restriction even though proponents of the complexity parameter view have generally not addressed this difference. However, I argue that what would be parametrized is the method of memory access within syntax, which seems highly implausible. In the second section, I consider the claim that there are natural languages parametrized to be restricted to regular grammars. I first consider what kind of view of syntactic memory this type of parametrization seems to be assume: Namely, one where different aspects of syntactic memory are recruited on demand by the acquisition process. Then, I argue that a lan-
14 | Uli Sauerland guage faculty with such limited memory could not predict the right interpretations for the regular fragment of English. More generally, I then argue that the standard semantics of scope requires at least a memory system that supports context-free grammars. Truly regular languages should display a very different semantics. Since no convincing evidence for such a difference has been shown, the complexity parameter view has no empirical support.
2.2 Context-free vs. context-sensitive languages? The complexity parameter view as far as I know has not paid attention to the difference between languages that require a (mildly) context-sensitive analysis and those for which a context-free analysis might be available at the level of weak generative capacity. If such a division was contemplated, the dividing line might fall between Dutch and Swiss German on one side and English and Standard German on the other for all that is presently known about the formal complexity of these languages.⁷ While my goal is not to advocate the complexity parameter view for the context-free/contextsensitive division either, it is instructive to consider why it is generally agreed that in this case a complexity parameter is not the right way to capture the cross-linguistic variation within West Germanic. Since the conclusion in this case seems to be universally held up, we can then compare this case to the controversial case of the contextfree vs. regular language division. Before the counter-arguments to the context-free/context-sensitive parameter, consider one argument why this might be a strong case for the complexity parameter view: In this case, the languages involved are well-studied and therefore the evidence basis is solid. English is in all likelihood the best studied language and German is also well studied, so if nobody has found strong evidence that a context-free analysis of English or German is impossible, we can be quite sure that it is indeed possible. The amount of evidence available for English contrasts strongly with the languages hypothesized to be of regular complexity which are either extinct (e.g. Old Babylonian) or spoken in remote locations (e.g. Pirahã). For these languages, the amount of evidence available is much more limited and there are no native speakers active in the field of linguistics. So in these cases, there is a non-negligible likelihood that evidence against a regular analysis was overlooked by the existing work on these languages or may simply be missing from the limited historical records. In the following sections, I discuss five potential arguments that militate against a direct parametrization of the context-free/context-sensitive distinction, of which I
7 With respect to English, there has been quite some discussion of the question of whether it requires a complexity greater than context-free. Other candidates of languages that require a context-sensitive analysis are Bambara and Mohawk (see Manaster-Ramer 1986 and references therein).
2. Against complexity parameters
| 15
consider the last three to be convincing. First, I discuss the absence of a clear direction of historical progress in the context-free/context-sensitive division. Second, I address language acquisition including artificial grammar learning. Third, I consider plausible cognitive implementations of a context-free/context-sensitive parameter. The fourth and fifth argument both relate to grammatical evidence that underlies existing analysis of the languages involved. On the one hand, actual grammars of both types of languages both involve mechanisms like movement that go beyond a context-free analysis. On the other hand, existing proposals for parametrization contain parameters that account for the domain of variation that would be addressed by a context-free/contextsensitive complexity parameter. Probably the most important reason the context-free/context-sensitive division has not been viewed from the complexity-parameter perspective is that there is no clear historical development. The four example languages, Dutch, English, Standard German, and Swiss German make this point very well. All four languages are closely related, and only diverged from each other quite recently in their history. In fact, Swiss German, Standard German and Dutch have been viewed as points of a dialect continuum that also includes many Low German dialects that are quite close to Dutch (though I don’t know about their verb clusters) and Upper German varieties of Southern Germany and Switzerland. Furthermore, the level of cultural development across all four languages is not perceived to be substantially different at present time. However, the absence of historical development alone is not a strong reason to reject the complexity parameter view in this case. For example, cultural evolution may still require some time for the greater expressivity of context-sensitive languages to take effect. Or it may be that other technological developments compensate for the expressive handicap that context-free languages carry with them. So, while the lack of evidence for a historical development has sociological importance, it alone should not discourage the complexity parameter view. Consider now evidence from language acquisition and artificial grammar learning. To my knowledge, there has been little effort to investigate whether this domain provides evidence for a context-free/context-sensitive parameter. Such evidence might come from a comparison of Standard German speakers’ acquisition of Swiss German or Dutch with the reverse. If parameter resetting was to require greater effort, we would expect Swiss German and Dutch to be the more difficult to acquire for Standard German speakers than vice versa. As far as I know, no such difference in learning difficulty has been reported, so at least we can conclude that the comparative difficulty is not dramatically different. But we cannot rule out that there may be a small difference at this point – we simply lack the relevant evidence. Furthermore, a similar contrast would be expected with Artificial Grammar Learning: for English or Standard German speakers, the copy grammar ought to be harder to learn than for Dutch and Swiss German speakers. As far as I know, no results on such comparisons are out in print. Tecumseh Fitch (p.c.) has mentioned to me unpublished experiments on English, where speakers found the copy grammar not harder to learn than a center-
16 | Uli Sauerland embedding context-free grammar. Beyond these unpublished data that would argue against a complexity parameter, though, there is just little evidence from acquisition at present bearing on the complexity parameter view. For the first real argument against the complexity parameter view, consider how the context-free/context-sensitive parameter would be implemented. From the production perspective, it seems natural to simply propose that only languages with the context-sensitive parameter setting are permitted to have context-sensitive rules. However, if we took this view, we might as well not have the complexity parameter, but have for each potential context-sensitive rule a parameter as to whether that rule is part of the grammar or not. From a parsing perspective, though, there may be a more natural implementation of the complexity parameter. Specifically, the Chomsky hierarchy has a fairly direct interpretation in terms of a memory structure: finite state automata correspond to regular grammars, push-down automata correspond to context-free grammars, and context-sensitive grammars correspond to linear-bounded automata. Therefore, one might assume that the context-free parametrization corresponds to a mental state where the syntactic parser can only access memory in the manner of a push-down stack. The context-sensitive setting correspondingly would be captured by a freer access to memory within syntax. But, such a view strikes me as implausible, since the method of access to memory would need to be part of the cognitive implementation of memory, and therefore it is plausible that also English and German speakers should be able to access memory freely if other humans are capable of a non-stack like access to memory in syntax. A second real argument against the complexity parameter view comes from the consideration of actual proposals for the grammar of the languages involved. As for example Chomsky (1965) discusses in connection with the distinction between weak and strong generative capacity, the evidence for a specific syntactic structure comes mostly from other sources than the set of grammatical strings. Though the syntactic analysis of English is highly debated, there is general agreement that purely context-free analyses have been on the wrong track, even though they may be technically feasible. All current analyses of English include mechanisms that go beyond those of context-free grammars. One such mechanism is the movement transformation or its equivalents in other grammatical formalisms. In sum, all current work on English grammar argues against the complexity parameter view. The third real argument against the complexity parameter view is that existing parameters make a complexity parameter superfluous. Consider for example the difference between Standard German and Swiss German with infinitival verb clusters illustrated above. For example, Schmid (2005) proposes an optimality theoretic analysis of word-order variation in such clusters. According to this analysis, actually all speakers generate word orders on both sides of the complexity divide as candidates for an optimality theoretic competition and then select one such analysis as determined by other parameters. In general, syntacticians working on infinitival clusters and word-order
2. Against complexity parameters
| 17
have argued that the extent of typological variation in this domain is better accounted for by a set of word-order parameters than by a complexity parameter. In sum, a context-free/context-sensitive parameter to account for the differences between English/Standard German on the one hand and Dutch/Swiss German on the other is indeed implausible. In this way, I concur with a belief generally held among typologists. For this reason, I assume that the reasons I presented in this section are tacitly held by many in the field, since, as far as I know, there are no advocates of the context-free/context-sensitive parameter. Now it is interesting to compare this case with the one of context-free vs. regular languages.
2.3 Context-free vs. regular natural languages? In this section, I consider the proposal that languages are parametrized for the contextfree vs. regular grammar distinction – the other major distinction of the Chomsky hierarchy relevant to our concerns. I argue first that such a parametric claim would only make sense at the level of cognitive mechanisms for similar reasons as those discussed in the previous section. But, from that it follows that not only purely syntactic evidence is relevant, but also semantic evidence for non-regular structures. Throughout I will consider the regular fragment of English within the context of this argument. Recall from above that I assume that this fragment contains only sequences of nominal phrases and verbs such as those in (5), but does not contain the well-known types of structures that establish solely on the basis of weak generative capacity that English requires a context-free grammar. (5)
a. b. c.
We let the children help John paint the house. We helped John to paint the house. John painted the house.
To generate the strings, assume a recursive, right-regular grammar⁸ consisting of rules obeying the schema in (6a), where NP can be any of the terminals we, John, the house, and the children, and V can be any of the terminals let, help, and paint.⁹ (6)
a. b.
S → NP V S S → NP V NP
While the facts about the language Pirahã are hard to ascertain presently (Everett 2005; Nevins et al. 2009), one possible description of the available data (and indeed a
8 As one of the reviewers points out, the term right-linear is also used instead of right-regular. 9 For expository reasons, I abstract away from the difference between inflected and uninflected verbs in the following.
18 | Uli Sauerland plausible one) is that complement clauses in Pirahã have essentially the same grammar as the one in (6). Such a language – essentially English without relative clauses and some complex coordinations such as either . . . or . . . – is quite plausible on the basis of what we know about existing parametric variation. Specifically, the language Teiwa, for which Klamer (2010) provides a detailed description, seems to be a language of this type according to her. Now consider how a language like the regular fragment of English would be captured within an approach assuming a complexity parameter. Recall that regular grammars can be characterized in at least two ways: One characterization is as set of strings that can be generated using only right- or only left-recursive phrase structure rules. The other is as strings that can be parsed by finite state automata. If the complexity parameter was to apply at the level of phrase structure rules, the parameter would govern the linear order of the possible recursion within such rules. However, it would need to apply across all rules of the grammar, while typically word-order parameters can vary from one phrase structure rule to another – for example, German and Japanese are both verb-final, but in German the noun precedes its arguments der Brief an Maria (‘the letter to Mary’), while it follows them in Japanese Mary-e-no tegami. If each phrase structure rule has its own word-order parameters, the complexity parameter would be redundant at this level. A more promising interpretation is to assume that the complexity parameter is implemented as part of the parser. According to this interpretation, the initial state of language would only allow parsing using finitely many memory states like a finite state automaton. But after being triggered by experience in the form of center-embedded strings, the language faculty would make available memory structures that at least allow in principle the processing of context-free grammar structures. Note that this interpretation of the complexity parameter is at odds with the assumption that human memory is limited in general. But, possibly for the present purposes it is sufficient to assume that the two parametrizations link to memory resources of a qualitatively different nature: one kind of memory that is self-contained and thereby finite, while the other can in principle recruit other memory resources without limit. The fact that these other resources are actually limited is not part of the system, but rather a performance limit imposed from the outside. This kind of difference is familiar from other domains: For example, it is well known that individuals’ general working memory is limited (cf. Miller and Chomsky 1963; King and Just 1991). But nevertheless models of memory don’t built in a hard limit. Indeed, if there was a hard limit of n-items to working memory, one would expect that subjects were aware of this boundary and that performance on tasks like a digit span would show a sharp drop-off exactly between remembering 7 vs. 8 digits. As things stand, however, the limit of short-term memory does not seem to be wired in like a computer’s memory limit, but instead the structure of memory is in principle unlimited, but the accuracy of recall drops gradually around
2. Against complexity parameters
| 19
the threshold level.¹⁰ The parameter view then seems to be based on the assumption that the language faculty has access only to a finite memory structure on the parameter setting ‘regular’. The parameter setting ‘context-free’, however, reflects that the language faculty can access a second type of memory, one that is in principle unlimited. A speaker of only the regular fragment of English then should be in the same memory state after processing the sequence NP–V–NP–V as after processing only NP– V: the state where now he either expects an NP to complete the sentence or expects another S. Indeed, this would also be the initial state of such a speaker before starting to process a sentence if his memory is limited to the minimum needed to successfully process the (NP–V)n grammar of the regular fragment of English. What would such a memory limit mean for interpretation? I now argue that a finite-state limited speaker is predicted to exhibit substantially different interpretation for the sentences of the regular fragment of English. Essentially the memory state of such a speaker after each NP–V segment would be identical to that at the beginning of a sentence. As a result, even though the regular grammar in (6) seems to generate embedded sentence structures, the interpretation a finite-state language faculty could associate with these structures would need to differ from that normally assigned to embedded clauses. Since the examples above contain both inflected and uninflected verbs, the point is more easily seen by means of the example in (7). (7)
Joe dreamed Mary read Bill left Germany.
If the language faculty was the minimal finite state system to provide the right syntactic parse for structures like (7), it would be in the same state at the beginning of the sentence and at the beginning of each complement clause, i.e. before the word Mary and before the word Bill. Given the close relation between syntax and semantics, that entails that the meaning associated with (7) could not differ from that of (8), where there are three separate clauses. (8)
Joe dreamed. Mary read. Bill left Germany.
Clearly, this does not hold in English: A speaker who utters (7) need not believe that Bill actually left Germany, but anybody uttering (8) is committed to holding this belief. The semantic difference between (7) and (8) may for example be traced to unpronounced assertion operators that add their complement to the common ground (Stalnaker 1978). If such operators are shown in the structure, (7) would look like (9a) while (8) would look like (9b).
10 This also holds, as far as I can tell, of more articulated, recent models of working memory as containing a number of subcomponents (Baddeley 2010; Makuuchi et al. 2009).
20 | Uli Sauerland (9)
a. b.
ASSERT Joe dreamed Mary read Bill left Germany. ASSERT Joe dreamed ASSERT Mary read ASSERT Bill left Germany.
The semantics of embedding and semantic scope more generally is, in a sense, inherently context-free. If the language faculty was limited to parsing regular structures by a parametric choice, this would predict that these languages ought to be severely limited in their expressivity. In other words, while a regular fragment of English would only differ little from actual English with respect to weak generative capacity, the semantics of the regular fragment ought to differ substantially. In English, the difference is clearly apparent, and in other languages it is very easy to test. Non-factive attitude verbs like dream, say, and think should yield very different interpretations in languages with minimal finite memory. The argument is, as far as I can see, not affected by the fact that some apparent unembedded clauses in English can be interpreted in a way similar to embedded clauses. The relevant phenomena where such an embedded interpretation occurs are modal subordination/telescoping (Roberts 1987; Poesio and Zucchi 1992) as in (10), and free indirect discourse (Schlenker 2004; Sharvit 2008) as in (11). (10)
Each degree candidate walked to the stage. He took his diploma from the dean.
(11)
Yes, his pants were on fire. So, Kaplan decided to call the fire brigade.
These examples are interesting because it seems that scope can be an inter-clausal phenomenon outside of the core of syntax/semantics: In (10), the pronouns he and his in the second clause are apparently interpreted as bound by each degree candidate, and in (11) the first clause represents a belief of Kaplan’s, not of the speakers. But, the authors mentioned above argue that in both cases this is only the appearance and actually a silent embedding operator is reconstructed in the relevant sentences, and therefore both of these are actually properly embedded clauses.¹¹
2.4 Conclusion In this paper, I considered the proposal that languages are parametrized for different levels of complexity on the Chomsky hierarchy of formal languages. I considered two complexity levels at which such a parametrization might occur: the contextfree/context-sensitive boundary and the regular/context-free boundary. I compared this proposal to one where the language faculty across all languages is capable of analyses at the same level of complexity.
11 Note furthermore that the availability of the relevant interpretation is limited in both cases. For instance, free indirect discourse can only represent a third person beliefs, but not his doubts or what he does not believe.
2. Against complexity parameters | 21
I argued that word-order parameters can generally explain the phenomena attributed to a complexity parameter well on the latter type of proposal, and that there is independent evidence for word-order parameters. The proposal that language complexity is directly parametrized, on the other hand, would compel us to a view where the cognitive resources available to the language faculty differ substantially across languages, specifically the structure of memory. Since general cognition independent of language presumably provides a memory structure like that of linear-bounded automata, the parametrization could be stated in terms of a limit of how much of the general memory can be recruited by the language faculty: only a finite state memory in putative regular languages, and only a push-down stack memory in putative context-free languages. Assuming this type of implementation, I argued that the complexity parameter view would predict very different semantics for languages with only a narrow finite memory: namely, the semantics of scope requires unlimited memory at least of the type of a push-down automaton, and therefore languages that do not have this type of memory available to language would be expected to ascribe very different meanings to many recursive structures. As far as I know, this is not the case for any of the languages involved. Therefore I conclude that the complexity parameter is an extreme proposal without any evidence in its favor. Acknowledgement: I have presented related work on a number of occasions, and I especially thank the audience at the University of Konstanz for their comments. Furthermore I am grateful to Andreas Trotzke, Nicholas Longenbaugh, and the two anonymous reviewers for their helpful comments on this paper. The research reported on in this work is in part supported by the German Ministry for Education and Research (BMBF), grant number 01UG0711. The first draft of this paper was written while I was employed by Harvard University.
References Abe, Kentaro, and Dai Watanabe. 2011. Songbirds possess the spontaneous ability to discriminate syntactic rules. Nature Neuroscience 14:1067–1074. Baddeley, Alan. 2010. Working memory. Current Biology 20:R136–R140. Bane, Max. 2008. Quantifying and measuring morphological complexity. In Proceedings of WCCFL, Charles B. Chang and Hannah J. Haynie (eds.), 69–76. Somerville, Mass.: Cascadilla Press. Beckers, Gabriël J.L., Johan J. Bolhuis, Kazuo Okanoya, and Robert C. Berwick. 2012. Birdsong neurolinguistics: songbird context-free grammar claim is premature. Neuroreport 23:139. Bhatt, Rajesh, and Aravind Joshi. 2004. Semilinearity is a syntactic invariant: A reply to Michaelis and Kracht 1997. Linguistic Inquiry 35:683–692. Chomsky, Noam. 1956. Three models for the description of language. IRE Transactions on Information Theory 2:113–124. Chomsky, Noam. 1957. Syntactic structures. The Hague: Mouton. Chomsky, Noam. 1959. On certain formal properties of grammars. Information and Control 2:137– 167.
22 | Uli Sauerland Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, Mass.: MIT Press. Chomsky, Noam, and George A. Miller. 1958. Finite state languages. Information and Control 1:91– 112. Crain, Stephen. 1991. Language acquisition in the absence of experience. Behavioral and Brain Sciences 14:597–650. Includes open Peer Commentary and Authors Response. Dahl, Östen. 2004. The growth and maintenance of linguistic complexity. Amsterdam, Netherlands: John Benjamins. Davis, Steven (ed.). 1991. Pragmatics. A reader. Oxford, UK: Oxford University Press. DeGraff, Michel. 2003. Against creole exceptionalism. Language 79:391–410. Deutscher, Guy. 2005. The unfolding of language. Metropolitan Books. Everett, Daniel. 2005. Cultural constraints on grammar and cognition in Piraha: Another look at the Design Features of human language. Current Anthropology 46:621–646. Fitch, Tecumseh, and Marc Hauser. 2004. Computational constraints on syntactic processing in a nonhuman primate. Science 303:377–380. Fitch, Tecumseh, and Angela D. Friederici. 2012. Artificial grammar learning meets formal language theory: an overview. Philosophical Transactions of the Royal Society B: Biological Sciences 367:1933–1955. Frank, Robert. 1998. Structural complexity and the time course of grammatical development. Cognition 66:249–301. Friederici, A.D., J. Bahlmann, R. Friedrich, and M. Makuuchi. 2011. The neural basis of recursion and complex syntactic hierarchy. Biolinguistics 5:87–104. Gentner, Timothy Q., Kimberley M. Fenn, Daniel Margoliash, and Howard C. Nusbaum. 2006. Recursive syntactic pattern learning by songbirds. Nature 440:1204–1207. Givón, Talmy. 2009. The genesis of syntactic complexity. Amsterdam, Netherlands: John Benjamins. Graf, Thomas. 2013. Local and transderivational constraints in syntax and semantics. Doctoral Dissertation, University of California, Los Angeles. Hauser, Marc D., Noam Chomsky, and Tecumseh Fitch. 2002. The faculty of language: what is it, who has it, and how did it evolve? Science 298:1569–1579. Hawkins, John A. 2004. Efficiency and complexity in grammars. Oxford, UK: Oxford University Press. Hay, Jennifer, and Laurie Bauer. 2007. Phoneme inventory size and population size. Language 83:388–400. Heine, Bernd, and Tania Kuteva. 2007. The genesis of grammar: A reconstruction. Oxford, UK: Oxford University Press. Huybregts, Riny. 1976. Overlapping dependencies in Dutch. Utrecht Working Papers in Linguistics 1:24–65. Huybregts, Riny. 1984. The weak inadequacy of context-free phrase structure grammars. In Van periferie naar kern, Ger J. de Haan, Mieke Trommelen, and Wim Zonneveld (eds.), 81–99. Dordrecht, Netherlands: Foris. Jackendoff, Ray, and Eva Wittenberg. 2014. What you can say without syntax: A hierarchy of grammatical complexity. In Measuring grammatical complexity, Frederick J. Newmeyer and Laurel B. Preston (eds.), 65–92. Oxford, UK: Oxford University Press. Joshi, Aravind K. 1985. Tree-adjoining grammars: How much context sensitivity is required to provide reasonable structural descriptions? In Natural language parsing, David R. Dowty, Lauri Karttunen, and Arnold M. Zwicky (eds.), 206–250. Cambridge, UK: Cambridge University Press. Joshi, Aravind K. 2003. Mildly context-sensitive grammars. In Oxford international encyclopedia of linguistics, 2nd edition. Oxford, UK: Oxford University Press. King, Jonathan, and Marcel A. Just. 1991. Individual differences in syntactic processing: The role of working memory. Journal of Memory and Language 30:580–602.
2. Against complexity parameters
| 23
Klamer, Marian. 2010. A grammar of Teiwa. Berlin, Germany: Mouton de Gruyter. Lobina Bona, David J. 2012. Recursion in cognition: a computational investigation into the representation and processing of language. Doctoral Dissertation, Universitat Rovira i Virgili, Tarragona, Spain. Maddieson, Ian. 2005a. Correlating phonological complexity: data and validation. UC Berkeley Phonology Lab Annual Report (2005). Maddieson, Ian. 2005b. Issues of phonological complexity: Statistical analysis of the relationship between syllable structures, segment inventories and tone contrasts. UC Berkeley Phonology Lab Annual Report (2005). Maddieson, Ian. 2011. Phonological complexity in linguistic patterning. In Proceedings of the 17th International Congress of Phonetic Sciences, 28–34. Makuuchi, Michiru, Jörg Bahlmann, Alfred Anwander, and Angela D. Friederici. 2009. Segregating the core computational faculty of human language from working memory. PNAS 106:8362–8367. Manaster-Ramer, Alexis. 1986. Copying in natural languages, context-freeness, and queue grammars. In Proceedings of the 24th annual meeting on Association for Computational Linguistics, 85–89. McWhorter, John. 2001. The world’s simplest grammars are creole grammars. Linguistic Typology 5:125–166. Michaelis, Jens and Marcus Kracht. 1997. Semilinearity as a syntactic invariant. In Logical aspects of computational linguistics, 329–345. Springer. Miller, George A., and Noam Chomsky. 1963. Finitary models of language users. In Handbook of mathematical psychology, volume 2, 419–491. New York: Wiley. Nevins, Andrew, David Pesetsky, and Cilene Rodrigues. 2009. Pirahã exceptionality: A reassessment. Language 85:355–404. Newmeyer, Frederick J. and Laurel Preston (eds.), 2014. Measuring grammatical complexity. Oxford, UK: Oxford University Press. Poesio, Massimo and Alessandro Zucchi. 1992. On telescoping. In SALT 2: Proceedings of the second conference on Semantics and Linguistic Theory, ed. Chris Barker and David Dowty, number 40 in OSUWPL, 347–366. Ohio State University. Roberts, Craige. 1987. Modal subordination, anaphora, and distributivity. Doctoral Dissertation, University of Massachusetts, Amherst. Roeper, Tom. 2011. The acquisition of recursion: How formalism articulates the child’s path. Biolinguistics 5:57–86. Roeper, Tom, and William Snyder. 2005. Language learnability and the forms of recursion. In UG and external systems: Language, brain and computation, ed. Anne-Marie di Sciullo, 155–169. Amsterdam, Netherlands: John Benjamins. Sampson, Geoffrey, David Gil, and Peter Trudgill, ed. 2009. Language complexity as an evolving variable. Oxford, UK: Oxford University Press. Sauerland, Uli, and Andreas Trotzke. 2011. Biolinguistic perspectives on recursion: Introduction to the special issue. Biolinguistics 5:1–9. Schlenker, Philippe. 2004. Context of thought and context of utterance: A note on free indirect discourse and the historical present. Mind & Language 19:279–304. Schmid, Tanja. 2005. Infinitival syntax: Infinitivus pro participio as a repair strategy. Amsterdam, Netherlands: John Benjamins. Sharvit, Yael. 2008. The puzzle of free indirect discourse. Linguistics and Philosophy 31:353–395. Shieber, Stuart M. 1985. Evidence against the context-freeness of natural language. Linguistics and Philosophy 8:333–343.
24 | Uli Sauerland Stalnaker, Robert. 1978. Assertion. In Pragmatics, ed. Peter Cole, 315–332. New York, N.Y.: Academic Press (reprinted in Davis 1991). Trotzke, Andreas. 2008. Rekursive Syntax & FLN? Eine kritische Betrachtung neuerer Diskussionen zur menschlichen Sprachfähigkeit innerhalb des Generativismus. MA thesis, Freiburg. Trotzke, Andreas, and Josef Bayer (eds.). 2015. Syntactic complexity across interfaces. Berlin, Germany: Mouton de Gruyter (this volume). Trotzke, Andreas, and Jan-Wouter Zwart. 2014. The complexity of narrow syntax: Minimalism, representational economy, and simplest Merge. In Measuring grammatical complexity, Frederick J. Newmeyer and Laurel B. Preston (eds.), 128–147. Oxford, UK: Oxford University Press. Veenstra, Tonjes. this volume. The development of subordination. In Trotzke and Bayer (2015). Wray, Alison, and George W. Grace. 2007. The consequences of talking to strangers: Evolutionary corollaries of socio-cultural influences on linguistic form. Lingua 117:543–578. Wurmbrand, Susi. 2004. West Germanic verb clusters: The empirical domain. In Verb clusters: A study of Hungarian, German, and Dutch, 43–85. Amsterdam, Netherlands: John Benjamins. Zwart, Jan-Wouter. this volume. Top-down derivation, recursion, and the model of grammar. In Trotzke and Bayer (2015).
Jan-Wouter Zwart
3 Top-down derivation, recursion, and the model of grammar
3.1 Introduction An aspect of human cognition that may be central to language is the ability to treat a complex structure as a single item, and to move back and forth between the complex and atomic interpretation of such elements depending on the cognitive task at hand. This ability is identified by Hofstadter (2007: 83) as crucial to the development of the mental lexicon, where complex substructures in e.g. family relations may be subsumed under a single concept like ‘uncle’ or ‘family’. Hofstadter identifies the ability to manipulate such semantic loops as marking a crucial step in the evolutionary development of human cognition. I would like to suggest that the very same ability is involved in linguistic recursion, understood as the syntactic treatment of a complex string as a single item within another complex string. The simplex/complex ambiguity of such strings can be formalized if we understand them to be construed in a separate derivational sequence (a derivation layer), feeding into the alphabet (numeration, in minimalist terminology) for the next derivational sequence. The string under discussion, then, is complex in the context of the earlier derivation, and simplex in the context of the later derivation. If human cognition is able to jump back and forth between a complex and a simplex treatment of strings, the grammar creating those strings (‘narrow syntax’) may be of the minimal complexity of a finite-state grammar. If so, the complexity of natural language is not to be expressed in the type of grammar rules, but in terms of interaction among derivation layers, made possible by the atomic interpretation of complex structures. This paper has the following structure. Section 3.2 establishes that the alphabet/ numeration may be nonhomogeneous, in the sense that some of its symbols may be the output of a previous derivation. Section 3.3 recapitulates arguments for thinking that the phrase structure rules generating structure must be of a higher complexity than finite-state grammar rules or context-free grammar rules. Section 3.4 then proposes that the structure building rules, in their most minimalist conception, are of the finite-state type. After a discussion of the diagnostics for determining that certain symbols must be the output of a previous derivation in section 3.5, we use those diagnostics in section 3.6 to revisit the arguments concerning the type of grammar rules, and argue that once the principle of derivation layering is taken into account, the arguments lose their force. Section 3.7 concludes with a discussion of the nature of re-
26 | Jan-Wouter Zwart cursion in the model of grammar contemplated here, arguing that recursion does not reside in the rules of narrow syntax, but in the interaction among derivation layers.
3.2 The nonhomogenous alphabet of natural language A key entity in formal language theory is ‘symbol’, which is generally not formally defined (cf. Hopcroft and Ullman 1979: 1), even if it features crucially in the definition of other key concepts, such as string, alphabet, or (formal) language. A ‘string of symbols’ is sometimes called ‘sentence’ (Kimball 1973: 2), and an ‘alphabet’, defined as a finite set of symbols (Hopcroft and Ullman 1979: 2), is sometimes called ‘vocabulary’ or ‘lexicon’. It corresponds to what is called ‘numeration’ in linguistic minimalism (Chomsky 1995: 225), the input to the rules of grammar. A ‘grammar’ of a language L is a finite specification of the sentences (strings of symbols) of L (Kimball 1973: 2), so that we may state that a grammar specifies sequences of symbols of a language (typically in the form of rewrite rules). All this implies that we have some understanding of what a symbol is, involving, I think, the common implicit assumption in (1). (1)
An alphabet is a homogeneous set of symbols.
With ‘homogeneous’ I mean ‘of a single type’, where ‘type’ ranges over the set of common linguistic concepts in (2). (2)
{ phoneme, morpheme, word, phrase, clause }
More specifically, it appears to be tacitly assumed that symbols are invariably words (3) and the question of what type of grammar we may construe (4) or what kind of language we observe (5) typically starts from that assumption. (3)
The vocabulary is simply a list of words of English (Kimball 1973: 2)
(4)
. . . the task of constructing a finite-state grammar for English can be considerably simplified if we take [the set of symbols generated] as the set of English morphemes or words [as opposed to phonemes] (Chomsky 1956: 114–115)
(5)
. . . when we consider the human languages purely as sets of strings of words (. . . ), do they always fall within the class of context-free languages? (Pullum and Gazdar 1982: 471)
There is, however, no reason why ‘symbol’ (or ‘word’ as an entity of formal grammar rules) should be equated with ‘word’ (in the sense of a natural language object), or in fact with any single type of natural language object. Intuitively, it is clear that the rules of grammar often combine elements of different types, as discussed at length in
3. Top-down derivation, recursion, and the model of grammar
|
27
Ackema and Neeleman (2004). Ignoring phonemes, we observe that elements of all types in (2) may be combined with each other (examples from Dutch):¹ (6)
a.
b.
c.
d.
e.
word + morpheme vader ‘father’ + -je dim > vader-tje ‘dear/little father’ phrase + morpheme vader en moeder ‘father and mother’ + je dim > [vader-en-moeder]-tje ‘playing house’ phrase + word syntaxis en semantiek ‘syntax and semantics’ + groep ‘group’ > [syntaxis en semantiek]-groep ‘syntax and semantics group’ clause + word ik weet niet wat ik moet doen ‘I don’t know what to do’ + gevoel ‘feeling’ > [ik weet niet wat ik moet doen]-gevoel ‘feeling of not knowing what to do’ clause + morpheme ik weet niet wat ik moet doen ‘I don’t know what to do’ + ge freq/iter > ge-[ik weet niet wat ik moet doen] ‘constantly saying that you don’t know what to do’
Since words and phrases can be combined with morphemes, it would have to be the case that the alphabet (numeration) either consists of morphemes or is nonhomogeneous. If the alphabet is nonhomogeneous, the rules of grammar may combine simplex and complex elements. But the complex elements, being complex, are structured and hence must themselves be derived by the rules of grammar. I am assuming the uniformity hypothesis in (7), referring to the structure-building process of linguistic minimalism (‘Merge’). (7)
Uniformity hypothesis Every structured complex is created by Merge.
We return to the merge mechanism in section 3.4; for now it suffices to think of Merge as the operation that combines two elements in a string (constituent). It follows from (7) that if the alphabet is nonhomogeneous, containing some complex element, that complex element must itself have been created by Merge. The derivation of elements like (6), therefore, must be punctuated: there must have
1 Abbreviations used in the glosses: acc = accusative case, dat = dative case, det = determiner, dim = diminutive, freq = frequentative, inf = infinitive, iter = iterative, masc = masculine gender, neg = negation, nom = nominative case, obj = objective case, pl = plural number, sg = singular number, 1 = first person, 3 = third person.
28 | Jan-Wouter Zwart been an operation (or sequence of operations) Merge creating a complex element and feeding that into the alphabet for the operation Merge that yields the elements in (6). On the other hand, if the alphabet is homogeneous, consisting of morphemes only, we need subroutines creating complex elements out of a subset of elements in the alphabet. This can be done in a number of ways, either enriching the alphabet as we go along (as in Bobaljik 1995) or utilizing simultaneous workspaces (‘parallel Merge’, Citko 2005). (In both cases, the added complexity of the process entails that the grammar can no longer be finite-state.) But what needs to be accounted for is that the complex elements thus derived acquire idiosyncratic properties as to form and interpretation, of the kind that in current minimalism are established at the interface components dealing with sensory/motor (sound) and conceptual/intentional (meaning) aspects of language. This assumes a model of grammar as pictured in (8), where the sequence of operations Merge is contained within the box marked ‘narrow syntax’ (cf. Hauser et al. 2002: 1570). (8) numeration
narrow syntax
interfaces
Thus, whereas a complex like handbook may be derived within narrow syntax (i.e. via Merge) from a numeration containing the elements hand and book, its noncompositional meaning ‘(short) book giving all the important information’ is not just a function of Merge, but must be established at the interfaces. Using handbook in a clause, then, requires a loop feeding handbook, with its special properties, from the interface component into another numeration. It follows that the subroutines needed if the numeration/alphabet is homogeneous must involve the interface components, and hence that these subroutines are in fact full-fledged derivations of the type in (8). Therefore, derivations must be ‘layered’, and there is no need to maintain that the alphabet must be homogeneous at all times.
3. Top-down derivation, recursion, and the model of grammar
|
29
3.3 Arguments relating to the typing of grammar rules The possibility of nonhomogeneous alphabets affects the argumentation regarding the nature of the rules of (phrase structure) grammar in a significant way.² As is well known, the earliest discussions of transformational generative grammar (e.g. Chomsky 1956) contain proofs that the rules of grammar of natural languages like English are not of the finite-state type, and later discussions have centered around the question of whether these rules are of the context-free or context-sensitive type (Huybregts 1976; Pullum and Gazdar 1982; Bresnan et al. 1982; and others).³ Discussions of the type of grammar rules for natural language invariably assume the alphabet to be homogeneous. As we will see in section 3.6, allowing for nonhomogeneous alphabets (in combination with the recursive loop discussed in section 3.2) undermines the proof that the rules of grammar cannot be finite-state. Here we just want to present that proof, as well as a proof showing that the rules cannot be contextfree. The commonly (though not universally) accepted proof demonstrating that English is not a finite-state language (from Chomsky 1956: 115; see also Kimball 1973: 22f; Partee et al. 1990: 480) is in (9), quoted from Langendoen (2003: 26–28). (9)
to account for the syntactic dependency between the words if and then in a sentence like if it rains then it pours, English grammar must contain a rule like S → if S then S. If so (. . .), then English also contains the sentences if if it rains then it pours then it pours and if if if it rains then it pours then it pours then it pours, but not *if it rains then it pours then it pours nor *if if it rains then it pours. That is, English contains every sentence of the form { (if )m it rains (then it pours)n | m = n }, but no sentence of the form { (if )m it rains (then it pours)n | m ≠ n }. Hence English is not a finite-state language.
2 A note of clarification is in order here. Discussion of the formal grammar of natural languages typically focuses on the type of language, rather than the type of rules. That is, the question is whether the sentences of a language (the string-sets) fall within a language-type class. In the model of grammar assumed here (cf. [8]), the sentences of a language are derived by a complex of operations, involving a structure-building process (‘Merge’) that can be described as a set of phrase structure rules, and a variety of interface processes, involving morphophonology, linear ordering, and the establishment of particular sound-meaning pairings, but also unpredictable processes such as reanalysis and recategorization (cf. Zwart 2009). On top of that, we must allow for derivation layering, allowing the output of one derivation to feed the next. My concern here is not with language as a whole, but with a small but important component of the model of grammar, narrow syntax, and with the nature of the rules needed to generate the type of structures we take to be its output. This is relevant to the hypothesis of Hauser et al. (2002) that narrow syntax represents the faculty of language in the narrow sense, a component of the human cognitive system that is arguably species-specific. 3 The assumption of transformational rules, already in Chomsky (1956), significantly reduced the urgency of these questions in mainstream generative grammar, a point that I will side-step here.
30 | Jan-Wouter Zwart Chomsky (1956: 115) likens sentences containing such a (recursive) if/then dependency to sentences of the type in (10), i.e. “sentences consisting of n occurrences of a followed by n occurrences of b, and only these.” (10)
a⌢b a⌢a⌢b⌢b a⌢a⌢a⌢b⌢b⌢b etc.
A finite-state grammar contains rules turning an initial (nonterminal) state A into a terminal a or into a string of a terminal a and a nonterminal B (see (11); the ‘|’ symbol indicates disjunction). (11)
finite-state grammar rule A→a|aB
The nonterminal B in (11) may be rewritten by a subsequent application of the finitestate grammar rule, creating perhaps a terminal b, but nothing guarantees the balanced and potentially infinite accretion of a’s and b’s we see in (10) (see Chomsky 1956: 115). The fact that English has sentences that have exactly the property in (10) shows that English is not a finite-state language, where a finite-state language is defined as the set of strings generated by finite-state grammar rules. We return to this argument in section 3.6. It has been argued that phrase structure rules generating natural language sentences have to be context-sensitive, rather than context-free, based on examples like (12), from Swiss German (Huybregts 1984, extending an argument from Huybregts 1976; Bresnan et al. 1982). (12)
wil mer de maa em chind lönd hälffe because 1pl:nom det:acc man det:dat child let:past.3pl help:inf schwüme (Zurich German; Huybregts 1984: 91) swim:inf ‘because we let the man help the child to swim’
In (12), causative lönd ‘let’ selects the accusative noun phrase de maa ‘the man’, and hälffe ‘help’ selects the dative noun phrase dem chind ‘the child’, yielding crossing case dependencies. Adding further embedded clauses adds further crossing dependencies, in such a way that the noun phrases and the verbs constitute potentially infinite series of crossing dependencies. (The facts in Dutch are the same, except that the dependencies are not expressed in case morphology.) The sensitivity to context that these crossing dependencies display cannot be expressed in context-free rewrite
3. Top-down derivation, recursion, and the model of grammar
|
31
rules, indicating that natural languages are not context-free languages (Huybregts 1984: 93).⁴ Both arguments assume that the grammar of a language contains just rewrite rules (i.e. is a phrase structure grammar). As Chomsky (1956) argued, phrase structure grammars are inadequate in that rules of phrase structure grammar fail to express natural language regularities captured by transformation rules. Cutting corners, we might state that English is not a finite-state (or context-free, or context-sensitive) language because to describe sentences of English, we need transformations in addition to phrase structure rules (see also note 3). Transformations range from morphological adjustment via reordering (displacement) to deletion and insertion. In linguistic minimalism, some of these processes are no longer considered to be part of narrow syntax. Morphological adjustment (or morphological expression in general) is taken to be postsyntactic (Chomsky 1995: 229), and deletion may just be failure to spell out at the interface component dealing with sound (cf. Merchant 2001). On the other hand, insertion and displacement are unified in the single operation Merge, joining some element to the structure being derived (extracted from out of that structure, in the case of displacement). In minimalism, then, the neat separation between structure building (rewrite rules) and structure manipulation (transformations) is gone. Structure is created by a single process Merge, which may involve movement.⁵ In the context of the present discussion, this leads to the question what kind of operation Merge is. I argue in what follows that Merge may be equated to a finite-state grammar rule, and that the complexity of natural language illustrated in (9) and (12) is brought in by the recursive loop process that turns the output of one derivation layer into part of the input for the next derivation layer.
3.4 The nature of Merge Let us assume that the rules of grammar of a language L must derive at least the information in (13) about the sentences of L. (In what follows, we ignore movement entirely; for issues arising in this connection, cf. Zwart 2009.) (13)
a. b. c.
constituency hierarchy dependency
4 For discussion of an alternative conclusion, that languages apparently vary as to the complexity of the rewrite rules of their grammars, see Sauerland (this volume). 5 Sentences, on the other hand, are derived by Merge in combination with a range of additional processes, see note 3.
32 | Jan-Wouter Zwart We derive constituency if the rules allow us to single out a string as a unit. This is typically done by conceiving of the rules (‘Merge’) as combining two elements in a set (or parts in a whole): (14)
common conception of Merge (1) (i) There is a numeration N (ii) There is an operation (‘Merge’) such that a. Merge takes two elements x, y from N, and b. Merge combines x and y
Hierarchy is derived if we allow Merge to combine an element from N with the newly constructed x-y combination, which I call the ‘object under construction’ ((15) supplementing (14)): (15)
common conception of Merge (2) (i) There is a numeration N (ii) There is an object under construction A (iii) There is an operation (‘Merge’) such that a. Merge takes an element z from N, and A, and b. Merge combines z and A
Dependency is derived if a dependent element needs to be in a hierarchical relation to its antecedent, i.e. a relation defined in terms of (15), as in Epstein (1999). The fact that Merge, in its common conception, is either (14) or (15) looks like an imperfection. In fact, the only time we need (14) is the first time Merge applies, as there is no object under construction at that point (Jaspers 1998: 109). But if we allow the object under construction to be empty at that point (Fortuny 2008: 18; Zwart 2009), (14) is redundant, and we can simplify the conception of Merge to (15). Hierarchy and dependency still follow. Still, this conception of Merge seems overly complicated in its two-step process of selecting elements and combining them. To use a common image, it is like both the numeration and the object under construction (the derivation) are in separate spatial locations, and Merge takes an element from the numeration and transfers it to the derivation. I find myself in agreement with Bobaljik (1995: 47), who argues that the common conception of Merge as involving transfer is an artifact of the way the operation is notated, while what is going on instead is that Merge articulates a particular relation among the elements of the numeration. One implementation of this idea, discussed in Zwart (2009, 2011a), is that Merge (now a misnomer) starts with an unordered numeration set, and with each step orders
3. Top-down derivation, recursion, and the model of grammar
|
33
one element from that set with respect to the residual members of the set (‘top-down Merge’).⁶ (16)
Top-down Merge a. b.
There is a numeration N There is an operation (‘Merge’) converting N into ⟨(x ∈ N), (N − x)⟩⁷
This operation yields a constituent (N − x), as well as hierarchical structure, if the operation can subsequently apply to (N − x), i.e. N becomes (N − x). The result is a nested sequence of ordered pairs, and dependency can again be defined in terms of the hierarchical structure thus derived (or, more exactly, in terms of the sequence of operations Merge). The top-down conception of Merge essentially splits a nonterminal element into a terminal element and another nonterminal element. The operation ends when the final remaining element is split off from the residue set, yielding just a terminal. The rule of grammar, then, is the finite-state grammar rule (11), yielding either a string of a terminal and a nonterminal, or a terminal. The discussion would be complicated if we wanted the rules of grammar to express movement. For the common conception of Merge, movement implies a further situation, not covered in (14)–(15), where Merge combines the object under construction A with an element from A. For the top-down conception of Merge (16), some representation of the element x split off from N would have to remain part of N, either as a copy or a trace or a feature, to be split off later, in order to account for the observation that x ‘belongs’ in a position where we do not see it. These complications take us too far afield at this point, so I will continue on the assumption that they will not jeopardize the approach contemplated here, an assumption that may well be off the mark. In what follows, then, I take (16) to be the minimalist conception of the structurebuilding process. Narrow syntax (cf. [8]) merely involves a sequence of operations of this type, suggesting that at least this part of grammar is finite-state.⁸ With that in mind, let us return to the argumentation in section 3.3, concerning the proper char-
6 See Phillips (2003) and Chesi (2007) for an earlier top-down generative model, and Zwart (2009: 165) for discussion of the differences with the present proposal. 7 Informally, the rule splits an element off from the numeration and yields an ordered pair consisting of that element and the numeration minus the element split off from it. 8 Importantly, the conclusion that this part of the grammar, Merge, is finite-state does not entail that the entire grammar, or the language generated by that grammar, is finite-state. If we are correct, a sentence is derived by a network of derivations, in which little finite-state units interact. It follows that the question of formal complexity cannot be resolved simply by inspecting sentences. Rather, the network of derivations involved needs to be identified, and the question of complexity can be asked of each subderivation. See Trotzke and Zwart (2014) for discussion.
34 | Jan-Wouter Zwart acterization of the rules of grammar for natural language, now assuming the alphabet/numeration to be nonhomogeneous.
3.5 The composition of the numeration If the alphabet/numeration can be nonhomogeneous, a simple sentence like (17) may receive a number ([i]-[iv]) of string-set analyses. (17)
The man left (i) (ii) (iii) (iv)
(the) (man) (left) (the man) (left) (the) (man left) (the man left)
The corresponding alphabets are: (18)
(i) (ii) (iii) (iv)
{ the, man, left } { the_man, left } { the, man_left } { the_man_left }
If (17) is analyzed as in (17iv), the grammar (the top-down merge machine) turns the initial state (18iv) into (17) in one step. The grammar then involves a single operation (16b), articulated in (19), which is a finite-state grammar rule of the type in (20), turning the initial state S (= [18iv]) into a single terminal a (= [17iv]). (19)
{ the_man_left } → ⟨ [the man left], ⌀ ⟩
(20)
S→a
In the remaining cases, the machine turns the initial state into one or more intermediate stages before reaching the end stage. The rules of the grammar then all have the form in (21) (= [11]), where A is a nonfinal state, a is a terminal, and B an intermediate stage. (21)
A →aB|a
For example, to derive (17i) from (18i), the rules, terminals, and intermediate stages are as in (22). (22) 1. 2. 3.
initial stage { the, man, left }
rule (21) (21) (20)
terminal the man left
intermediate stage { man, left } { left } ⌀
3. Top-down derivation, recursion, and the model of grammar
|
35
For our purposes, there is the additional question of which of the possible analyses is the correct one, i.e. is the one yielding the correct structure. A structure is cognitive reality brought out by experiments known as constituency tests, indicating which substrings are perceived as units by the speaker. If these units must be brought out by the grammar, a nonambiguous string has a single correct derivation. Since constituency tests identify the man in (17) as a unit, the correct derivation of (17) is the one that yields (17ii), i.e. the one that takes the alphabet for this sentence to be (18ii). Importantly, the string the man itself can be derived by a finite-state machine from the alphabet {(the), (man)}. If so, the alphabet for each string consists of symbols that are primitive relative to the derivation of that string, but not in an absolute sense. It is easy to see that in a system with nonhomogeneous (relativized) alphabets, recursion takes the form of ‘derivational interaction’, in the sense that a symbol in the alphabet for derivation D1 may itself be a string of terminals generated in derivation D2 (see Zwart 2011b). With this in mind, let us return to the question of sentences of English referred to in (9), proving that English is not a finite-state language.⁹
3.6 Revisiting the question of grammar rule types Sentences of the structure if X then Y obviously have the structure in (23): (23)
((if A) ( then (B)) = a b, where a = if A b = then B
An additional conditional construction (i.e. a pair consisting of a conditional clause and a consequent clause), as illustrated in (9), would substitute for A in (23). Since A is a term of a, the analysis is not crucially altered by multiplying the conditionals and consequents. Assuming nonhomogeneous alphabets, then, the string if it rains then it pours can be the output of a derivation over the alphabet in (24), and if if it rains then it pours then it pours can be the output of a derivation over the alphabet in (25). (24)
{ if_it_rains, then, it, pours }
(25)
{ if_if_it_rains_then_it_pours, then, it, pours }
9 The idea of describing complex strings as involving separate (finite-state) subroutines is familiar from the literature on natural language parsing (Woods 1970; Bates 1978; Abney 1996; Roche 1997). As Roche (1997: 269) observes, this cuts into the argument that to describe expressions with a certain level of complexity, we need more powerful formalisms, such as context-free grammars.
36 | Jan-Wouter Zwart Of course, the symbols if_it_rains in (24) and if_if_it_rains_then_it_pours in (25) can themselves be analyzed as outputs of derivations over alphabets, such as (26) for (24) and (27) for (25), and the unanalyzed complex symbols can likewise be further analyzed. (26)
{ if, it_rains }
(27)
{ if, if_it_rains_then_it_pours }
The point is that if (if it rains then it pours) then it pours does not have the structure a⌢(a⌢b)⌢b, or (a⌢a)⌢(b⌢b), with infinite parallel accretion of a’s and b’s, but the structure a⌢b, regardless the complexity of a. The parallel accretion of a’s and b’s results from the recursive inclusion of an ab structure (a conditional-consequent pair) inside a, and the relevant class of constructions is not correctly characterized as in (28), but as in (29), where S can be any clause (including another instance of if it rains then it pours). (28)
{ (if )m it rains (then it pours)n | m = n }
(29)
(if S) then it pours
What remains is the local dependency of a consequent clause and a conditional clause, which is repeated within S in (29) if it contains a conditional construction (a conditional-consequent pair). But such a dependency can be handled within a finite-state grammar, if dependency is a function of Merge, as we have assumed (see section 3.4). Thus, if we allow symbols to be the complex output of separate derivation layers, and we define recursion as the interaction between derivation layers, the rules of narrow syntax (‘Merge’) can be of the ultimate minimalist finite-state type. It remains the case that English is not a finite-state language (see note 3), but given the definition of recursion as derivational interaction, the narrow syntax component of English grammar may still be finite-state. For this conclusion to hold, we have to assume that the human cognitive capacity to treat objects as both complex and atomic, identified by Hofstadter (2007), supplements the faculty of language in the narrow sense as defined in Hauser et al. (2002). It remains to discuss the argument based on (12) showing that the rules of grammar are context-sensitive. Constructions of the type in (12), typical of the Continental West-Germanic languages, are notoriously complicated, with no single analysis of the various types and variations being currently uniformly accepted (see Zwart 1996 and Wurmbrand 2005 for a survey of the phenomena and their analyses). For our purposes, the interesting question is whether particular substrings in examples of this type can be characterized as outputs of separate derivation layers.
3. Top-down derivation, recursion, and the model of grammar
|
37
In Dutch, a verb-second language, strings that appear before the finite verb in independent clauses can be identified as constituents (Zwart 2011c: 21). This test allows us to identify verb clusters as constituents: (30)
a.
Ik heb hem de kinderen niet helpen 1sg:nom have:1sg 3sg.masc:obj det child:pl neg help:inf leren zwemmen teach:inf swim:inf ‘I did not help him to teach the children to swim.’
b.
Helpen leren zwemmen heb ik hem help:inf teach:inf swim:inf have:1sg 1sg:nom 3sg.masc:obj de kinderen niet det child:pl neg (same as a.)
The constituency of the verb cluster helpen leren zwemmen [help teach swim] is consistent with the idea that the verb cluster is the output of a separate derivation, hence a single symbol in the alphabet (numeration) for the derivation of the clause containing it (30a). Moreover, subparts of the verb cluster cannot be separately fronted: (31)
a. *Leren zwemmen heb ik hem de kinderen teach:inf swim:inf have:1sg 1sg:nom 3sg.masc:obj det child:pl niet helpen neg help:inf b. *Zwemmen heb ik hem de kinderen niet swim:inf have:1sg 1sg:nom 3sg.masc:obj det child:pl neg helpen leren help:inf teach:inf (both intended the same as [30a])
This argues against an alternative analysis, where the fronted constituent in (30b) is the remnant of a phrase depleted by extraction of some of its subparts (i.e. an accidental constituent; cf. den Besten and Webelhuth 1987). This kind of derivation, not available in a top-down model (cf. [16]), and also otherwise contested (cf. Fanselow 2002), does not account for the observation that the cluster as a whole behaves like an atomic constituent. In terms of the top-down derivational machine (16): subparts
38 | Jan-Wouter Zwart of the verb cluster may not be split off from the numeration independently, which is explained if the cluster is a single symbol in the relevant numeration.¹⁰ If verb clusters are created in separate derivation layers, noun phrases associated semantically with the verbs in the cluster must be linked with those verbs via ccommand (dependency created as a function of merge), i.e. must be ‘base-generated’ in their surface position, again entirely consistent with the top-down approach of section 3.4. The relation between the noun phrases and the verb cluster, then, is no longer one of cross-serial dependency, but a many-to-one relation. In this context, it is relevant to note that the order of the verbs inside the cluster is subject to considerable variation (both within and among dialects), so that the particular cross-serial dependency configuration is just one of many possibilities, hence arguably more coincidental than systematic.
3.7 The place of recursion in the model of grammar The top-down merge machine (16) essentially performs a series of identical steps, each step identifying a different symbol x from the alphabet/numeration N and in doing so reducing the set of unordered symbols (N − x). Both x and (N − x) are constituents, and the result at each step is an ordered pair. As I have argued elsewhere, dependency phenomena of natural language may be the linguistic interpretation of the asymmetry between the members of such an ordered pair (Zwart 2009: 165f). It is common to think of a sequence of operations Merge as recursive (e.g. Nevins et al. 2009: 366). This is because Merge (under any conception of it) is a rule creating an output that may be subjected to an application of the same rule. However, it is not the case that the structured objects created by Merge (hierarchical syntactic structures) are necessarily created by a recursive process (see e.g. Arsenijević and Hinzen 2012). For example, assuming Merge to involve transfer from a numeration to an object under construction (essentially [16]), hierarchical structures are created by transfering elements one by one, an iterative procedure. Similarly, under the top-down conception of Merge (16), elements are split off from the numeration one by one. In each case, we derive hierarchical structures by means of an iterative procedure. On the other hand, the loop that results from taking the complex output of one derivation layer to be the atomic input to a next derivation layer is inherently recur-
10 A complication is that we have to allow for the finite verb to escape from the cluster to obtain the verb second position (e.g. heb in [31] in the text is not or no longer part of the verb cluster, whereas it is string-adjacent to or included in the cluster in embedded clauses, not illustrated here). This, however, may not be a problem if Chomsky (2001) is correct in identifying verb-second as a phenomenon of the interface component dealing with sound (see also Anderson 1993; Zwart 2005), and if we allow interface phenomena to affect parts of strings that are treated as atomic symbols in narrow syntax (as we must to account for morphological or prosodic marking anyway).
3. Top-down derivation, recursion, and the model of grammar
|
39
sive, in the sense that a symbol in the alphabet for a procedure P is itself the output of P. Here recursivity is not an artifact of the notation of the procedure, but is the unavoidable correlate of the simplex/complex ambiguity of linguistic objects alluded to in the introduction. This view of recursion shows some resemblance to the earliest conception of recursion in generative grammar, involving generalized transformations (Chomsky [1955]1975: 383, 518; Chomsky 1961: 134; Chomsky 1966: 52f). Generalized transformations are used to derive complex sentences; they are overarching transformations that operate on fully developed phrase markers. For example, a clause may function as an object in another clause by substituting for a dummy symbol in that other clause (Chomsky 1961: 52–53 note 2). In the cyclic, layered derivations approach contemplated here, there is no need for a dummy symbol, as the embedded clause functions as a single symbol in the alphabet for the derivation of the embedding clause. In what Palmatier (1972: ix) calls ‘second-generation transformational grammar’ (roughly from the mid-1960s), generalized transformations were abandoned, and recursion was written straight into the phrase structure rules, which were allowed to reintroduce the start symbol S (Chomsky 1966: 63). Since generalized transformations were typically substitutions targeting a dummy symbol, taking the dummy symbol to be S (or some other rewritable category) allows one to eliminate generalized transformations altogether. This development was made possible after Katz and Postal (1964: 120ff) had shown that the generalized transformations themselves made no semantic contribution. In minimalism (e.g. Chomsky 1993: 21), the need to introduce elements of arbitrary complexity directly (i.e. via Merge) was interpreted as motivating a device like generalized transformations once more (essentially in the context of abandoning the distinction between D-structure and surface structure; see also Frank 2002: 9). But while this gave rise to exploitation of parallel derivations, the simple consequence of allowing derivations to feed into the numeration for further derivations was not, to my mind, sufficiently explored, nor were the consequences for the definition of recursion.
3.8 Conclusion In this article, I have argued that common views on the nature of phrase structure rules (finite-state or of a higher complexity) suffer from an unmotivated and unnecessary hidden assumption, namely that the rules of grammar are fed by a homogeneous set of symbols, the alphabet/numeration. Once it is understood that the symbols in a numeration may themselves be the output of a separate derivation, arguments against the finite-state character of the phrase structure rules (‘Merge’) lose their force. I have argued that the minimalist conception of Merge involves the iterative conversion of the alphabet/numeration into a sequence of ordered pairs, each consisting of an element split off from the numeration and the residue of the numeration at that
40 | Jan-Wouter Zwart point, i.e. a terminal/nonterminal pair, mimicking a finite-state machine. Viewed this way, narrow syntax is just the conversion of an unordered set into an ordered n-tuple (cf. Fortuny 2008), and the complexity of the elements involved is not created by recursivity in the rules of narrow syntax, but by the circumstance that some of these elements have their own derivational history, being the output of a separate derivational sequence. I submit that the cognitive ability to treat elements simultaneously as simplex and complex underlies much of the complexity of natural language, allowing us to consider the rules of grammar to be maximally simple. It follows that if Hauser et al. (2002) are right in identifying recursion as the single species-specific property of the human faculty of language, and if we are right here, human language derives its unique properties not from the structure building procedure Merge, but from the circumstance that sequences of operations Merge (derivation layers) may interact, yielding atomic symbols of potentially infinite complexity. Acknowledgement: Thanks to Gertjan van Noord, Riny Huybregts, Jordi Fortuny, Jorike van Werven, and the audience at the Konstanz Complex sentences, types of embedding, and recursivity workshop, March 5–6, 2012, as well as to two anonymous reviewers. This article reports on research carried out within the Dependency in Universal Grammar research program, sponsored by the Netherlands Organization for Scientific Research (program number 360-70-200), which is gratefully acknowledged.
References Abney, Steven. 1996. Partial parsing via finite-state cascades. Natural Language Engineering 2, 337–344. Ackema, Peter and Ad Neeleman. 2004. Beyond morphology: interface conditions on word formation. Oxford: Oxford University Press. Anderson, Stephen. 1993. Wackernagel’s revenge: clitics, morphology, and the syntax-semantics interface. Language 69: 68–98. Arsenijević, Boban and Wolfram Hinzen. 2012. On the absence of X-within-X recursion in human grammar. Linguistic Inquiry 43: 423–440. Bates, Madeleine. 1978. The theory and practice of augmented transition network grammars. In Natural language communication with computers, Leonard Bolc (ed.), 191–254. Berlin: Springer Verlag. Bobaljik, Jonathan D. 1995. In terms of merge: copy and head movement. MIT Working Papers in Linguistics 27: 41–64. Bresnan, Joan, Ronald Kaplan, Stanley Peters, and Annie Zaenen. 1982. Cross-serial dependencies in Dutch. Linguistic Inquiry 13: 613–635. Chesi, Christiano. 2007. An introduction to phase-based minimalist grammars: why move is topdown and from left-to-right. CISCL Working Papers on Language and Cognition 1 (Studies in Linguistics): 38–75. Chomsky, Noam. 1956. Three models for the description of language. IRE transactions on information theory, vol. IT-2 (Proceedings of the symposium on information theory): 113–124.
3. Top-down derivation, recursion, and the model of grammar
|
41
Chomsky, Noam. 1961. On the notion ‘rule of grammar’. Proceedings of the Symposium in Applied Mathematics 12: 6–24. Chomsky, Noam. 1966. Topics in the theory of generative grammar. The Hague: Mouton. Chomsky, Noam. 1975. The logical structure of linguistic theory. New York: Plenum Press. Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The view from Building 20: essays in linguistics in honor of Sylvain Bromberger, Kenneth Hale and Samuel J. Keyser (eds.), 1–52. Cambridge: MIT Press. Chomsky, Noam. 1995. The minimalist program. Cambridge: MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: a life in language, Michael Kenstowicz (ed.), 1–52. Cambridge: MIT Press. Citko, Barbara. 2005. On the nature of merge: external merge, internal merge, and parallel merge. Linguistic Inquiry 36: 475–497. Den Besten, Hans and Gert Webelhuth. 1987. Remnant topicalization and the constituent structure of the VP in Germanic SOV languages. GLOW Newsletter 18: 15–16. Epstein, Samuel D. 1999. Un-principled syntax: the derivation of syntactic relations. In Working minimalism, Samuel D. Epstein and Norbert Hornstein (eds.), 317–345. Cambridge: MIT Press. Fanselow, Gisbert. 2002. Against remnant movement. In Dimensions of movement: from features to remnants, Artemis Alexiadou, Elena Anagnostopoulou, Sjef Barbiers, and Hans-Martin Gärtner (eds.), 91–127. Amsterdam: John Benjamins. Fortuny, Jordi. 2008. The emergence of order in syntax. Amsterdam: John Benjamins. Frank, Robert. 2002. Phrase structure composition and syntactic dependencies. Cambridge: MIT Press. Hauser, Marc, Noam Chomsky, and W. Tecumseh Fitch. 2002. The faculty of language: what is it, who has it, and how did it evolve? Science 298: 1569–1579. Hofstadter, Douglas. 2007. I am a strange loop. New York: Basic Books. Hopcroft, John E. and Jeffrey D. Ullman. 1979. Introduction to automata theory, languages, and computation. Reading: Addison-Wesley Publishing Company. Huybregts, Riny. 1976. Overlapping dependencies in Dutch. Utrecht Working Papers in Linguistics 1: 24–65. Huybregts, Riny. 1984. The weak inadequacy of context-free phrase structure grammars. In Van periferie naar kern, Ger J. de Haan, Mieke Trommelen, and Wim Zonneveld (eds.), 81–99. Dordrecht: Foris Publications. Jaspers, Dany. 1998. Categories and recursion. Interface: journal of applied linguistics 12: 81–112. Katz, Jerrold J. and Paul Postal. 1964. An integrated theory of linguistic descriptions. Cambridge: MIT Press. Kimball, John P. 1973. The formal theory of grammar. Englewood Cliffs: Prentice-Hall. Langendoen, D. Terrence. 2003. Finite state languages and grammars. In The Oxford International Encyclopedia of Linguistics (2nd edition), William J. Frawley (ed.), Vol. 2, 26–28. Oxford: Oxford University Press. Merchant, Jason. 2001. The syntax of silence. Oxford: Oxford University Press. Nevins, Andrew, David Pesetsky, and Cilene Rodrigues. 2009. Pirahã exceptionality: a reassessment. Language 85: 355–404. Palmatier, Robert A. 1972. A glossary for English transformational grammar. New York: AppletonCentury-Crofts. Partee, Barbara, Alice ter Meulen, and Robert E. Wall. 1990. Mathematical methods in linguistics. Dordrecht: Kluwer Academic Publishers. Phillips, Colin. 2003. Linear order and constituency. Linguistic Inquiry 34: 37–90. Pullum, Geoffrey K. and Gerald Gazdar. 1982. Natural languages and context-free grammars. Linguistics and Philosophy 4: 471–504.
42 | Jan-Wouter Zwart Roche, Emmanuel. 1997. Parsing with finite-state transducers. In Finite-state language processing, Emmanuel Roche and Yves Schades (eds.), 241–281. Cambridge: MIT Press. Sauerland, Uli. 2015. Against complexity parameters. In Syntactic complexity across interfaces, Andreas Trotzke and Josef Bayer (eds.), this volume. Berlin: Mouton de Gruyter. Trotzke, Andreas and Jan-Wouter Zwart. 2014. The complexity of narrow syntax: Minimalism, representational economy, and simplest Merge. In Measuring grammatical complexity, Frederick J. Newmeyer and Laurel B. Preston (eds.), 128–147. Oxford: Oxford University Press. Woods, William A. 1970. Transition Network Grammars for natural language analysis. Communications of the ACM 13, 591–606. Wurmbrand, Susi. 2005. Verb clusters, Verb Raising, and restructuring. In The Blackwell companion to syntax, Martin Everaert, Henk van Riemsdijk, Rob Goedemans, and Bart Hollebrandse (eds.), Vol. 5, 227–341. Oxford: Blackwell. Zwart, Jan-Wouter. 1996. Verb clusters in Continental West-Germanic dialects. In Microparametric syntax and dialect variation, James R. Black and Virginia Motapanyane (eds.), 229–258. Amsterdam: John Benjamins. Zwart, Jan-Wouter. 2005. Verb second as a function of merge. In The function of function words and functional categories, Marcel den Dikken and Christina M. Tortora (eds.), 11–40. Amsterdam: John Benjamins. Zwart, Jan-Wouter. 2009. Prospects for top-down derivation. Catalan Journal of Linguistics 8: 161-187. Zwart, Jan-Wouter. 2011a. Structure and order: asymmetric merge. In The Oxford handbook of linguistic minimalism, Cedric Boeckx (ed.), 96–118. Oxford: Oxford University Press. Zwart, Jan-Wouter. 2011b. Recursion in language: a layered-derivation approach. Biolinguistics 5: 43–56. Zwart, Jan-Wouter. 2011c. The syntax of Dutch. Cambridge: Cambridge University Press.
Leah S. Bauke
4 What small clauses can tell us about complex sentence structure
Introduction It is the hallmark of Chomsky (2004: 117; 2008) that Merge is characterized as the only cost-free operation within narrow syntax, that is it can be performed without limitations. In the literature Merge is almost universally described as a recursively applicable structure building operation that forms bigger syntactic objects from smaller ones. In this paper it will be suggested that the structures that are formed by Merge are of only limited complexity: the Merge operation itself can be applied recursively and without limitation, but the input to Merge is restricted in so far as at least one of the elements involved in the operation must be a lexical item (LI) (cf. e.g. Narita 2009, 2011). The logical consequence of this is that cases of XP-XP Merger do not exist (cf. also Kayne 2010; Jayaseelan 2008). It is this restriction alone that allows for a natural explanation of a heretofore unaccounted pattern that can be observed in Russian small clauses (SCs) and that presents a challenging puzzle. In the following chapter I will present the data and briefly discuss what is to be understood by SCs in this paper. The next chapter will give a short overview over existing analyses for SCs and it will be shown that these are incoherent and do not allow for an explanation of the Russian data. Chapter three will deal with cases of XP-XP Merger and the theoretical framework for the analysis will be introduced, which is basically built on ideas introduced in Narita (2009). Chapter four will give a general overview of how recursive structure building that does not lead to unlimited complexity works and chapter five focuses on the analysis for the Russian SCs data. Chapter six will provide some additional evidence for the analysis presented here and chapter seven will conclude the paper.
4.1 The data: Russian SCs SCs are notoriously difficult to define (cf. e.g. Bauke 2012, 2014 for an overview). It is not my intention to add to the discussion here; instead I will use the relatively broad definition in (1), which is a standard definition in the linguistic literature (cf. e.g. Aarts 1992; Moro 2000; den Dikken 2006, 2007a,b and many others): (1)
A SC is a subject-predicate structure lacking tense.
44 | Leah S. Bauke It will become clear in the course of this paper that this definition is not exactly right and needs some refinement. This refinement, it will be shown, follows naturally from the analysis presented here. For the moment this definition will do and I will focus on one type of SC, i.e. what is known as a bare copula SC (cf. Moro 2000), and in a second step the analysis will be extended to believe-type SCs. Examples for both types of constructions are given below, where (2a) is an example for a bare copula construction, and example (2b) is an instantiation of a believe-type SC. For the moment I will abstract away from the question of what – if any – structural differences there are between bare copula and believe-type SCs. I will treat both SC constructions alike for the time being and refer the reader to chapter 5 of this paper, for some comments on potential very minor differences between the two constructions with respect to their subextraction possibilities. For a more detailed comparison between the two construction types the reader is referred to Bauke (2014): (2)
a. b.
This man is the president. This man is believed to be the president.
Parallel examples exist in Russian¹: (3)
a.
Etot muschtzina prezident. this man.NOM president.NOM ‘This man is the president.’
b.
Etot muschtzina oni sčitajut prezidenta. this man.NOM they consider president.ACC ‘They consider this man the president.’
The example in (3a) and also the following example in (4a) shows that in Russian bare copula SCs the present tense copula is a covert element. This is at least what is generally assumed in the literature, and I will follow this assumption here. In order to make the copula overt, it must be in the past tense, as (4b) shows: (4)
a.
Prezident SŠA lider svobodnyx nacij. president.NOM USA.GEN leader.NOM free.GEN.PL nation.GEN.PL ‘The president of the US is the leader of the free nations.’
b.
Prezident SŠA byl [SC Prezident SŠA liderom president.NOM USA.GEN was leader.INST svobodnyx nacij]. free.GEN.PL nation.GEN.PL The president of the US was the leader of the free nations.’
1 All Russian data in this paper were elicited by the author from native speakers. Many thanks to my informants for their kind cooperation.
4. What small clauses can tell us about complex sentence structure
| 45
So in what follows I will focus on sentences where the bare copula is in the past tense. Thus, the copula is overt in these sentences and it has the additional side effect that the two SC constituents bear different case markings, i.e. nominative and instrumental in (4b), rather than a double nominative marking as in (3a) and (4a). As is expected from a multiple wh-fronting language like Russian, it is possible to fully extract the two SC constituents at least when the construction is in the past tense, where the copula is overt: (5)
a.
Kto kem byl kto kem? who.NOM what.INST was ‘Who was what?’
b. *Kto kto? who.NOM who.NOM The ungrammaticality of (5b) is highly reminiscent of the Serbo-Croatian data already discussed in Bošković (2002). In this paper Bošković argues that two homophonous heads cannot both be fronted, whereas two wh-words that are non-homophonous can and indeed must be fronted, as is expected from a multiple wh-fronting language, where wh-fronting is obligatory. This is illustrated in (6a)–(6c) below: (6)
a. *Šta šta uslovljava? what what conditions b.
Ko šta kupuje? who what bought ‘Who bought what?’
c. *Ko kupuje šta? who bought what The example in (6a) illustrates that fronting both of the two words leads to ungrammaticality. In (6b), as is expected, fronting is fine though, because here the two whheads are non-homophonous. (6c) finally illustrates that fronting is obligatory in non-homophonous contexts, because otherwise ungrammaticality results. The solution that Bošković (2002) suggests for the ungrammatical construction in (6a) is given in (7): (7)
Šta šta uslovljava šta? what conditions what? ‘What conditions what?’
Here it is assumed that both šta-heads are fronted in Serbo-Croatian, but only at L(ogical)F(orm). At P(phonological) F(orm) one of the šta-forms is pronounced in the position of the lower copy, which eliminates the homophony effect.
46 | Leah S. Bauke Getting back to the Russian data, we can observe the same phenomenon here. In those cases where the bare copula SC constructions feature a non-overt present tense copula, multiple wh-fronting is not possible: (8)
*Kto kto ⌀? what what?
The strategy that Bošković (2002) discusses for Serbo-Croatian does not apply to these Russian cases either, because pronouncing the lower copy does not do away with the homophony effect, due to the fact that the copula is non-overt – and neither does leaving one of the wh-constituents in the lower position altogether: (9)
a. *Kto kto ⌀ kto? what what? b. *Kto ⌀ kto? what what?
That this is on the right track is suggested by making the present tense copula overt. In these cases the Russian SCs follow exactly the pattern that is illustrated above for Serbo-Croatian: (10)
a.
*Kto kto est’? who who is?
b.
Kto est’ kto? who is who? ‘Who is what?’
c.
Kto kto est’ kto? who who is who ‘Who is what?’
Thus, Russian bare copula SCs behave exactly as expected from a language with (optional) multiple wh- fronting. What is also expected is that it is possible to subextract from each of the two SC constituents. This is illustrated in (11a) and (11b). In (11a) we have extraction from the SC subject and in (11b) extraction from the SC predicate: (11)
a.
Kakoj strany byl prezident kakoj strany liderom which.GEN country was president.NOM leader.INST svobonyx nacij? free.GEN.PL nation.GEN.PL ‘Of which country was the president the leader of the free nations?’
4. What small clauses can tell us about complex sentence structure
b.
| 47
Kakix nacij byl prezident SŠA which nations.GEN was president.NOM USA.GEN liderom kakix nacij? leader.INST ‘Of which nations was the president of the US the leader?’
Additionally, it is possible to combine subextraction from one of the SC constituents with full extraction of the other constituent. This is also expected in a language that allows multiple wh-fronting. In (12a) it is the SC subject that is fully extracted and the SC predicate that is subextracted from, and in (12b) it is the SC subject that is subextracted from and the SC predicate is fully extracted: (12)
a.
Kto kakix nacij byl kto liderom kakix nacij? who.NOM which nations.GEN was leader.INST ‘Who was the leader of which nations?’
b.
Kem kakoy strany byl who.INST which country.GEN was kakoy strany kem? prezident president.NOM ‘Of which country was the president what?’
What is not possible and, prima facie, surprising, is that subextraction from both the SC subject and the SC predicate at the same time leads to ungrammaticality. This observation has not, to my knowledge, been discussed in the literature: (13)
*Kakoj strany kakix nacij byl prezident kakoj strany which country.GEN which nations.GEN was president.NOM liderom kakix nacij? leader.INST
I will argue that the ungrammaticality of the construction in (13) is an immediate consequence of the ban on XP-XP Merger. To see this more clearly and to understand in how far the current analysis differs from existing approaches in the literature, let me first briefly discuss two well-known approaches and point out potential difficulties for these analyses, before returning to cases of XP-XP Merger in the next chapter.
4.2 Analyses: A first approximation to the structure of SCs It has been pointed out in the previous chapter already that SCs are frequently described as subject-predicate structures lacking tense (cf. e.g. Aarts 1992; Moro 2000; den Dikken 2006, 2007a,b and many others). Furthermore, two major approaches to
48 | Leah S. Bauke the structural description of SCs are discernible in the literature. Both will be presented, and merits and potential shortcomings will be reflected upon in what follows. Under the first approach SCs are the result of direct Merger of two phrases (cf. e.g. Moro 2000), one of which is the SC subject and the other is the SC predicate: (14)
SC Subj
Pred
Although widely accepted in the literature, there are a number of issues that remain unresolved under this analysis. What is, for instance, not clear under this approach is which of the two XPs provides the label for the syntactic object that is identified as a SC, i.e. whether it is the SC subject or the SC predicate. Nor is it clear how agreement between the SC subject and the SC predicate is established. In principle this could be achieved by direct agreement between the two constituents as is argued by e.g. AlHorais (2007), Carstens (2000) and others and as can be seen from French and Arabic data in (15) and (16) respectively taken from Al-Horais (2007): (15)
(16)
a.
Il considère [sc la fille intelligente]. he considers the girl.FEM.SG intelligent.FEM.SG ‘He considers the girl intelligent.’
b.
Il considère [sc les filles intelligentes]. he considers the girls.FEM.PL intelligent.FEM.PL ‘He considers the girls intelligent.’
c.
La fille intelligente the girl.FEM.SG intelligent. FEM.SG ‘The intelligent girl.’
d.
Les filles intelligentes the girl. FEM.PL intelligent. FEM.PL ‘The intelligent girls.’
a.
‘aadad:tu Mariam-a thakia:t-an. considered Mary.3SG.FEM-ACC intelligent. 3SG.FEM-ACC ‘I considered Mary intelligent.’
b.
Salma iʔtabarat walad-a-ha wazir-an. Salma considered child-ACC-her minister-ACC ‘Salma considered her child to be a minister.’
Here the assumption is that, under a direct Merge approach, agreement can be established simply because the two SC constituents are directly merged. Thus the gender and number agreement in the French examples in (15) and in the Arabic examples in (16) is the result of direct feature valuation between the SC subject and the SC predi-
4. What small clauses can tell us about complex sentence structure
| 49
cate that is established directly upon Merge. In that sense, the agreement relation in SCs is comparable to what has been suggested for adjectival agreement within DPs. However, Moro (2000: 34–35) illustrates that this need not be the case. In arguing against the assumption, originally presented in Kayne (1984), that an agreement head mediates between the SC subject and the SC predicate of a bare copula SC, Moro provides the following piece of data as counterevidence: (17)
Gianni ritiene [sc [dp questi libri] [dp la John considers this.MASC.PL book. MASC.PL the.FEM.SG della rivolta]. causa cause.FEM.SG of the riot ‘John considers these books the cause of the riot.’
Here the SC subject and the SC predicate DP do not agree either in number nor in gender, which Moro (2000) takes as evidence that a) no functional projection is involved, i.e. that SCs are the result of direct Merger of the SC subject and predicate constituents and b) that agreement between the two SC constituents is not obligatory. However, it should be noted that – despite Moro’s claim – this is not a bare copula SC but rather a believe-type SC and additionally, Moro (2000: 35) himself provides further data, also from believe-type SCs, that show that agreement between the subject and predicate DPs of the SCs is indeed obligatory: (18)
a.
Gianni ritiene [sc [dp queste ragazze] [dp le sui migliori John considers this.PL girl.PL the.PL his.PL best.PL amiche]. friend.PL ‘John considers these girls his best friends.’
b.
*Gianni ritiene [sc [dp queste ragazze] [dp la sua John considers this.PL girl.PL the.SG his.SG migliore amica]. best.SG friend.SG ‘John considers these girls his best friend.’
Moro (2000) does not provide any explanation for why agreement is obligatory in the SC in (18) but not in (17). In any case, both are believe-type SCs, and if the argument that an agreement head is present in this type of SC is valid, then the remaining questions are what distinguishes a believe-type SC from a bare copula SC and why the two believe-type SCs in (17) and (18) respectively behave differently.² So, although perfectly
2 Admittedly Moro (2000) makes a distinction between bare copula and believe-type SCs and argues that the latter differ from the bare copula SCs in so far as they feature a functional projection.
50 | Leah S. Bauke viable, a number of questions remain open under this approach, and it could easily be argued alternatively that agreement is indeed obligatory between the SC subject and predicate constituents, and that this agreement is established directly upon Merge in both bare copula and in believe-type SCs. The difference between obligatory agreement in (15)–(16) and (18) and its absence in (17) could then be related to the fact that the former SCs all denote a reason or origin relation that is absent in (17) (cf. also Bauke 2014). Notice that the puzzle in (13) remains mysterious under this approach. There is no principled reason why subextraction from both SC constituents, in particular in bare copula SCs, should be unavailable. For the believe-type SCs it could be argued that subextraction from a subject position is generally problematic and thus also unavailable in those SC constituents where the SC subject is a specifier of a F(unctional) P(rojection). But for bare SCs this line of reasoning does not hold and subject subextraction combined with predicate subextraction would have to be ruled out for independent reasons. Let me now turn to the alternative approach that is also widely discussed in the literature. This goes back to the already mentioned assumption originally presented in Kayne (1984) that SCs host a functional projection. Here the predication relation is established by merging the subject XP in the specifier of the functional projection and the SC predicate YP in the complement position of the functional projection (cf. e.g. den Dikken 2006, 2007a): (19)
FP/SC Subj
F’/SC F/SC
Pred
Here the functional head is the element that provides the label, thus the labeling problem in (14) does not arise for the structure in (19). However, it is not at all clear what the exact properties of F are.³ Den Dikken (2006, 2007a) argues that the predication relation is mediated by F, which he identifies as a relator head R.⁴ Yet, as Matushansky (2007b) correctly points out, a predication is a
3 Den Dikken (2006) for instance argues that it is a meaningless category, which is a clear violation of the principle of Full Interpretation (cf. chapter 3 for details). Although it should be noted that this only holds if elements that contribute the identity function as their meaning must also violate Full Interpretation. In other words, phonologically null heads that have the semantic property of introducing the identity function into the derivation are not in and of themselves a violation of the principle of Full Interpretation (cf. Sternefeld 2007). 4 Den Dikken (2006) also introduces a FP, but this is needed only for predicate inversion structures in bare copula constructions and thus should not be confused with the FP here.
4. What small clauses can tell us about complex sentence structure
| 51
semantic relation and it is not at all clear how this semantic relation is expressed by such a structure. It could be argued that the functional head establishes the agree relation between the SC subject and predicate constituents, as in Kayne (1984), but this leaves open the question how the data from Moro discussed in (17) and (18) fit into the picture. If, as den Dikken (2006, 2007a) argues, all SCs have the structure in (19), then it is not at all clear why the properties of F can differ considerably. On the one hand it is not clear why F establishes an obligatory agreement relation in (18) but not in (17), nor is it clear whether the functional head F is the same in bare copula SCs, where it is the copula that is hosted by F, and in believe-type SC constructions where F is realized either by zero or by a preposition. On the other hand it is not clear whether a functional head should also play a role in those SCs that do not express a predication relation, such as naming constructions (cf. Matushansky 2007a), equative constructions (cf. Heycock and Kroch 1999) and certain other bare copula constructions (cf. den Dikken 2006). Again, it should be noted that the ungrammaticality of (13) also remains unexplained under this approach. As has been mentioned above, the general unavailability of subextraction from subjects may play a role. However, it accounts for the data only insufficiently, since subextraction from the SC subject alone seems to be unproblematic and also subject subextraction paired with full extraction of the predicate is generally available (cf. (11a) and (12b) above). Why it is exactly the combination of subextraction from the SC subject and the SC predicate that leads to ungrammaticality is left unaccounted for in this approach. Thus, it can be concluded that neither of the two existing analyses is without problems and a number of unresolved questions remain (cf. also Bauke 2014 for a more detailed discussion of these issues). In what follows I will argue that the basic structure for bare copula and believe-type SCs is effectively the one in (14). However, this precise structure runs into the immediate problem of a point of symmetry⁵ that results from the Merger of two XPs. This, I suggest, is illicit. Before showing how this problem can be avoided in the next chapter, let me briefly remark here that the same problem also pertains to the structure in (19). Merging the SC subject (XP) in the specifier of FP is also an instance of XP-XP Merger, because in bare phrase structure terms (cf. Chomsky 2001, et seq.) at the point in the derivation where XP is merged, F’ is maximal (or at least not a head, i.e. a LI for that matter). To be precise, in a more traditional feature-driven system of Merge, a projection is defined as maximal when the head of that projection has no more unvalued features. In that sense F’ is not a maximal projection then. But it is also not a minimal projection, i.e. a head, either. Thus at the point in the derivation where the specifier is merged, this is not an instance of Merger
5 A point of symmetry arises when two constituents of the same type are merged, i.e. either two heads or two maximal projections. In these cases the label of the syntactic object that results from Merger cannot be determined.
52 | Leah S. Bauke of a head and #a maximal projection⁶ but, arguably, of merging a complex S(yntactic) O(bject) with a maximal projection.
4.3 Cases of XP-XP Merger Following recent analyses in Jayaseelan (2008), Kayne (2010) and Narita (2011) I adopt the following principle: (20)
Cases of XP-XP Merger do not exist.
It has already been pointed out in the introduction that Merge is the only structure building operation in narrow syntax that can apply freely and that forms binary unordered sets (cf. Chomsky 2008, 2013). As Narita (2011) illustrates in his dissertation, in a bare phrase structure account the structures that Merge yields are subject to the Inclusiveness Condition, the No-Tampering Condition and Full Interpretation. All these are stated in (21)–(23) in an adapted version: (21)
Inclusiveness Condition No new features are introduced in the course of the derivation.
(22)
No-Tampering Condition⁷ No elements introduced by syntax are deleted or modified in the course of the linguistic derivation.
(23)
Full Interpretation Every constituent of SEM and PHON contributes to interpretation.
It is worth noticing that the postulation a functional projection (as in den Dikken 2006), which is identified as a meaningless functional category that cannot be adjoined to, is in clear violation of (23). But also the symmetric object that is derived in (14) can constitute a violation of the strongest minimalist thesis, which states that syntax is an optimal solution to interface requirements (cf. Chomsky 2000: 96) in so far as any form of (extrinsic) labeling and/or projection would violate (22).
6 As we will see below, this structural configuration is illicit in the system that is developed in this paper, and it should just be noted that it is also not entirely unproblematic in more traditional approaches. 7 As one reviewer correctly points out, (22) is a very strong – potentially too strong – constraint that raises the question how agreement is accounted for in this system. Arguably valuation of phi-features or any other feature checking or probing would constitute a violation of the NTC. However, feature valuation is at the core of syntax and Narita (2009) provides an account under which agree operates on EFs called AgreeEF, where the EF itself, being an unvalued feature that never is valued and never deletes, can act repeatedly as either a Probe or a Goal for agree (cf. ibid: 220).
4. What small clauses can tell us about complex sentence structure
| 53
So only those structures that are produced by Merge that are in strict compliance with the principles in (21)–(23) are compatible with the strongest minimalist thesis and are licit structures that can be derived in narrow syntax (cf. Narita 2011 for detailed discussion). This, however, leads to the question how a structure with a specifier can ever be derived. Put differently, the question is whether the principles in (21)–(23) lead to a situation in which the baby is thrown out with the bathwater, because it would per se be impossible to generate specifiers. In order to deal with this question let me once again consider the properties of Merge in a bit more detail. It has already been said that Merge is an operation that creates bigger syntactic objects from smaller ones. Now the logical next step is to ask what a syntactic object actually is. Here I use the following definition from Narita (2011): (24)
A linguistic representation α is a syntactic object iff: a) b)
α is a LI, or α is a set {β, γ} with β and γ = syntactic objects.
LIs in turn are defined in the following way: (25)
a. b.
Only LIs carry edge features (EFs).⁸ Non-LIs do not carry EFs.
This leads to the following central claim in Narita (2011) that is also the basis for the analysis of the SC structures that are under discussion here: (26)
Only LIs are permitted to be merged with some syntactic object
(27)
The H-α schema: Merge (H, α) → {H, α}
(26) and (27) is a formal way of expressing that XP-XP structures cannot be merged, because for Merge to be operative at least one of the two elements that form the input to the operation must be a LI with an EF. Notice that projecting EFs higher up the tree, as originally assumed in Chomsky (2008), is not an option qua (22). Thus, a different way needs to be found. This is what is discussed in the next section.
8 It should be noted here that the definition of EF differs from the original use of the term in Chomsky (2008). “Edge feature” here does not signify the general availability for internal Merge. Instead, it specifies a property of LIs, which basically signals the general availablity of a SO as input for the operation Merge (internal and/or external Merge, for that matter). In that respect the use of the term EF here is also distinct from earlier conceptions of EPP- features in Chomsky (2000, 2001) or OCC(urrence) features in Chomsky (2004).
54 | Leah S. Bauke
4.4 Structure building 4.4.1 The basic structure Let us look at a sample derivation in order to understand how complex structures can be built that are compatible with the strongest minimalist thesis and qua extension with the principles discussed in the previous chapter. The complications of what is known as the “first-Merge” problem in the literature (cf. e.g Boeckx 2008: 79–84) are immaterial in this approach. All items coming from the lexicon (or the numeration for that matter) are equipped with an EF and are thus automatically susceptible to Merge. First-Merge then is simply a scenario in which not just one of the elements bears an EF, but one in which both bear EFs. The result is a complex syntactic object without an EF. Let us term this syntactic object β. Now, for the derivation to continue, the next element merged with the syntactic object β must be a LI. Since β does not bear an EF (only the elements contained in it each bear EFs, and these do not percolate)⁹, the only option that is left is that the element that β merges with is one that bears an EF. This is what is illustrated in (28): (28)
Merge of α and β; α a LI and β a complex syntactic object: SO αEF
β
Under the H-α schema, α now is the head of the construction and for the derivation to continue, it again has to be a LI that is merged with the syntactic object in (28) which in and of itself again does not bear an EF – but then only specifier-less structures emerge, which of course corresponds to the ban on XP-XP Merger but rules out a large number of constructions that are known to be perfectly licit and among them the SC structures discussed here. Now, let us assume that α in (28) is a phase head and that this phase head spells out and transfers its complement domain. What remains in the derivation then is nothing but (29), adapted from Narita (2011: 46): (29)
Spell-out of the complement domain of α in (28): αEF
So, all that remains in the derivation after spell-out and transfer of the complement domain of α is the phase head α itself. Now, according to Chomsky (2008), the EF on the phase head (or any other LI for that matter) does not delete nor percolate. Thus, the phase head may have checked and deleted (some of) its features with elements
9 In fact, in an approach – like the one I am pursuing here – that follows the theory of Narita (2011), there is no feature projection of any kind, as this is regarded to be a violation of the NTC in (22).
4. What small clauses can tell us about complex sentence structure
| 55
in its complement domain but this does not apply to EFs. Thus, all that is left in the derivation is α, at least with an EF and possibly some other (heretofore undeleted) features. Thus α is available for a further Merge operation with some other syntactic object, because its EF is enough to satisfy the requirements from the H-α schema in (27). This is illustrated in (30) (adapted from Narita 2011: 46). (30)
SO γ
αEF
Here the phase head α (which has spelled-out and transferred its complement domain in a previous step of the derivation) merges with the syntactic object γ via the EF on α. The derivational steps from (28)–(30) are summarized in (31): (31) αEF
β
⇒
αEF
⇒
γ
αEF
Under this view, specifiers of phase heads lose their special status because γ is nothing but a second complement to α. The remaining question is if non-phase heads can have specifiers as well or whether this is exclusively reserved for phase heads. This is what is discussed in section 4.4.2.
4.4.2 Merger of a syntactic object in the specifier of a non-phase head In the previous chapter we have looked at a way to merge specifiers while maintaining the ban on XP-XP Merger at the same time. This solution hinges on the idea that a phase head with a spelled-out and transferred complement domain is nothing but a LI in the remainder of the derivation (cf. also Uriagereka 2012 where such structures are described as gigantic lexicalized compounds). What is not clear yet is whether non-phase heads can host specifiers as well. Prima facie this seems impossible in the current system, because non-phase heads cannot transfer their complement domains, thus their EFs are not accessible for any further operation of Merge and specifiers are non-LIs, i.e. syntactic objects without EFs as well. The standard case for such a construction is TP. Arguably, following Abels (2003) and others, T is not a phase head (but cf. Chomsky 2004; Boeckx and Grohmann 2007; Müller 2004 and others for the opposing view¹⁰) but still, TPs do have specifiers. Again
10 In fact, whether T actually is a phase head or not is somewhat orthogonal to the discussion here. All that matters is that there most certainly are some heads that are not phase heads and that can still host specifiers even in a system that adheres to the H-α schema in basically the same way as is illustrated here for T.
56 | Leah S. Bauke I will follow the analysis in Narita (2011) here in arguing that the element merged in SpecTP itself is a phase head with a spelled-out complement domain. To see this more clearly, let me once again quickly review the derivation step by step. The first part is unproblematic. The non-phase head T can be merged with some syntactic object β via the EF on T. The result is a complex syntactic object without an EF: (32)
SO β
TEF
Now, for the derivation to continue, all that is needed is that the element in SpecTP is a LI with an EF. This is possible in the form of the element being a phase head with a spelled-out complement domain. Assuming that DPs are phases (cf. e.g. the discussion in Svenonius 2004; Bošković 2005; Hiraiwa 2005; Matushansky 2005; Heck 2009; Ott 2008; Samuels 2008, Richards 2012), the following sample structure emerges¹¹: (33)
SO DEF
SO TEF
β
With this much in place, I will now return to the (basic) structures for SCs.
4.5 SC structures 4.5.1 Basics If XP-XP Merger is universally unavailable, then in SCs at least one XP must be merged as a phase head with a spelled-out and transferred complement domain. So let us first consider the SC in (34): (34)
This man is the president.
For this SC I assume the basic structure in (14) repeated here as (35). Note that, under the direct Merge approach, the copula enters the derivation at a later stage, as is shown in what follows. Notice also that in an analysis that adheres to the H-α schema, the labeling problem mentioned in connection with (14) does not arise. Due to the fact
11 I will abstract away from the question whether the DP in the specifier of TP is a case of internal or external Merge, as it is orthogonal to the analysis here. Suffice it to say that the analysis is compatible with both options and that Chomsky (2008, 2013) does away with the distinction anyway.
4. What small clauses can tell us about complex sentence structure
| 57
that cases of XP-XP Merger do not exist, an unlabelable point of symmetry does not even arise: (35)
SC subj (this man)
pred (the president)
Now there are two possible scenarios for the derivation of the SC in (35). A first possibility which itself includes two options is that either one of the two SC DPs is a phase head. So each can, in principle, be merged as a D-head and can thus also be moved. So, both options in (36) are available: (36)
a. b.
This man is [sc this man the president]. The president is [sc this man the president].
It is also possible that both SC DPs (instead of just one) are merged as D-heads. Still the same movement options as in (36) are available then. I tentatively suggest that under the latter scenario an equative construction results, while under the former it is a subject-predicate structure. Both scenarios are schematically illustrated in (37) and (38) respectively: (37)
SC DP this man
(38)
DEF the president
SC DEF this man
DEF the president
The second scenario is that there is a SC that contains not two DPs which are both phase heads, but one DP and one other syntactic object that cannot be reduced to a LI via spell-out, i.e. a maximal projection that is not a phase. A DP-AP structure would be a possible configuration here, or a DP-PP structure:¹² (39)
a.
Etot muschtzina byl [sc etot muschtzina prezidentom]. this man.NOM was the president.INST ‘This man was the president.’
12 I will abstract away from the question whether all PPs are non-phases or not. Currently, this is a hotly debated issue and I will assume that – at least for the examples presented here – P is not a phase head.
58 | Leah S. Bauke
(40)
b.
Etot muschtzina byl [sc etot muschtzina intelligentnim]. this man.NOM was intelligent ‘This man was intelligent.’
c.
Etot muschtzina byl [sc etot muschtzina v parke]. this man.NOM was in the park ‘This man was in the park.’
a. Jack is intelligent/in the rose garden. b. Jack is [SC Jack intelligent/in the rose garden]. c. *Intelligent/in the rose garden is [SC Jack intelligent/in the rose garden].
So, if A and P in (39) and (40) are not phase heads and thus cannot be remerged (contra Abels 2003; Řezáč 2010) then the structures for these constructions can be derived in the following way: (41)
Derivation of this man is the president: step 1 construction of the SO1 = this man and construction of the SO2 = the president [SO1 D N] [SO2 D N] step 2 Spell-out of the complement domain of the phase heads in SO1 and SO2 D1E D2EF step 3 Merge of D1EF and D2EF : [SO D1EF D2EF ] step 4 Merge of the copula (via the EF on the copula): [SO isEF [SO D1EF D2EF ]] step 5 Remerge of either D1 or D2 (for EPP reasons): [SO D1EF [SO isEF [SO D1EF D2EF ]]]
(42)
Derivation of this man is intelligent/in the rose garden: step 1 construction of SO1 = this man and construction of SO2 = intelligent/in the rose garden [SO1 D N] [SO2 AP/PP] step 2 Spell-out of the complement domain of the phase head in SO1 D1EF [SO2 AP/PP] step 3 Merge of D1EF with SO2 (via the EF on D1 ): [SO D1EF [SO2 AP/PP]] step 4 Merge of the copula (via the EF on the copula): [SO isEF [SO D1EF [SO2 AP/PP]]] step 5 Remerge of D1 (for EPP reasons): [SO D1EF [SO isEF [SO D1EF [SO2 AP/PP]]]]
4. What small clauses can tell us about complex sentence structure
| 59
Before we continue, two comments are in order here. First, it is to be noted that spellout of the complement domain of a phase head entails that the complement domain becomes inaccessible for further syntactic operations. Assuming that the complement domain is actually removed from syntax altogether is not an option, because the EPPdriven movement in step five of (41) and (42) respectively is (at least PF) movement of the whole DP, i.e. of the D-head and its complement domain; it is not just movement of the D-head alone. This movement is only possible though, because the D-head has an accessible EF, i.e. because it is reduced, i.e. it rendered its complement domain inaccessible for further syntactic operations. A further indication that points in the direction of inaccessibility rather than full deletion of the complement domain of a phase head is that a certain asymmetry between SOs first merged with a phase head, i.e. traditional complements, must be somehow distinguishable from later instances of Merger of a SO with a phase head with a spelled-out complement domain, i.e. traditional specifiers, in order to maintain the asymmetry between complements and specifiers that allows us to account for certain binding facts etc. (unless of course these asymmetries turn out to be irrelevant for binding). The second point that needs to be taken into account is the EPP-driven movement itself. Under the approach outlined here, Merge is unconstrained. The only prerequisite is that an LI with an EF must be available. And as we will see below, even the reduction of complement domains via phase heads is optional to a certain degree. This raises the question how feature-driven movement, e.g. by an EPP-feature, fits into the overall picture. I do not want to go into the details of a discussion on feature-driven movement here, but it should be pointed out that movement that is driven by features – EPP or others – is still potentially available in this system. In fact, in a Mergedriven system, like the one outlined here, feature driven movement is not per se excluded. As long as feature-driven movement satisfies the requirements imposed by the H-α schema, it is still compatible with the current system. Under this analysis the following data receive a natural explanation as well: (43)
a. *Who do you think in the rose garden is? b. */?In which garden do you think the president is? ¹³,¹⁴
13 Grammaticality judgements differ considerably here. While some native speakers find this structure perfectly grammatical, others mark it as marginal (at most) if not ungrammatical. One reviewer remarks that other PPs can clearly be moved. Notable examples are the following: i ii
With who(m) did you leave? About what do you think that Mr Barrimore lied?
This then leads to the valid question whether it would not be more appropriate to assume that every head is a phase head. This comes close to what is suggested in Marantz (2007) and it reverses the situation. Instead of undergenerating by excluding the grammatical examples listed above, this approach would possibly overgenerate and leave unexplained why TP and NP cannot be moved. However, it is not clear why multiple specifiers should prima facie be excluded. In fact, an analysis along the lines of
60 | Leah S. Bauke c. *Which garden do you think the president in is? d. Which garden do you think the president is in? The ungrammaticality of (43a)–(43c) can be explained on the basis of the analysis presented above. In (43a) it is not possible to wh-extract the SC subject and to also extract the SC predicate. The underlying structure for (43a) is given in (44): (44)
Who do you think in the rose garden is [sc who in the rose garden]?
So, although it is perfectly licit to extract the wh-word alone from the SC, it is not possible to extract it in combination with extraction of a second constituent. The sentence in (45) illustrates that this is on the right track: (45)
Who do you think is [sc who in the rose garden]?
In (45) the EF on the LI who allows for extracting and remerging the LI outside the SC just as expected under an analysis that is in line with the H-α schema. The constituent in the rose garden is not a LI though. Rather it is a syntactic object without EF. Also, it cannot be reduced to a LI, because P is not a phase head that can spell out its complement domain and thus the EFs on P are also not available for Remerger. The example in (43b) illustrates basically the same thing and will thus not be discussed any further here; the data in (43c) can also be explained along these lines. To see this more clearly, let us once again assume that the structure for the SC in (43c) is as in (46). (46)
. . . is [sc the president in the rose garden].
Again, the SC subject the president is a DP. This DP can be reduced to a LI via spell-out of the complement domain of the phase head D. Thus, it is also possible to remerge the D(P) outside of the SC (by virtue of the EF on D). It is not possible, though, to remerge the PP. Once again, this is because the P is not a phase head and thus there is
Narita (2011) would predict that multiple specifiers are indeed fine, as long as the constituents merged in the specifier position are merged as phase heads with spelled- out complement domains. So, given the grammaticality of the fronted PPs, in the examples above, it seems more likely that only designated heads are phase heads. These may include some instances of APs and PPs. Actually, this introduces the immensely broader question of what counts as a relevant phase head, which I will not go into here, because it cannot be done justice in this paper. Suffice it to say that phase properties are tied to certain features and one of the relevant features in this context may be case. This would then account for why certain APs and PPs and all DPs come out as phases. 14 The other question that arises in this example is why the PP cannot simply be pied-piped with the wh-phase head. Or put differently, it raises the absolutely valid question how pied-piping is handled in the current system. Again, this seems to be possible only for certain types of heads in certain languages. Thus, when deciding which types of heads count as phase heads, pied-piping phenomena should also be taken into consideration.
4. What small clauses can tell us about complex sentence structure
| 61
no syntactic object that is just a LI with EFs, which, according to the H-α schema, are a prerequisite for (Re)merger. Now, at first sight, the grammaticality of the data in (43d) is somewhat surprising, but it follows naturally from the approach that is pursued here. Again, it is not an issue that the DP the president is remerged outside of the SC. This is done via the EF on the phase head D, which must have spelled out its complement domain at some point before Remerge can take place. But we can even say more than that. We can even give a fairly good approximation at which point this spell-out operation must have taken place. Notice, incidentally, that the sentence in (47) is also fine. Here the DP the president, which is the SC subject, remains inside the SC in its base position: (47)
Which garden (do you think) is [sc the president in which garden].
So this shows a number of things. First, it is immediately obvious that despite the fact that Remerger is an option, it is not a necessity. So there is a fairly larger amount of optionality in the system presented here than in the original bare phrase structure system devised in Chomsky (2008, 2013). Marc Richards (p.c.) pointed out to me that the present system, which is determined by the logic of the H-α schema, is a Mergedriven system of phasal transfer, while Chomsky’s system is Agree-driven. However, I will not dwell on the implications of this distinction any further at this point.¹⁵ Now, it is important to notice though that the optionality is only an issue as far as Remerger outside of the SC is concerned. For the original Merger operation inside the SC, it is an absolute necessity that the SC subject is merged as a D-head with a spelled-out complement domain and accessible EFs, otherwise it would be impossible to merge the two SC constituents in the first place, because the SC predicate again is a PP and as such a syntactic object without available EFs. So, Merger inside the SC proceeds as schematically illustrated in (48):
15 It should be noted though that this optionality does not mean that certain phase heads, i.e. those that do not reduce their complement domain immediately, also optionally loose their phase status. One reviewer asks in this context, whether the optionality of spell-out is beneficial with respect to locality. Phases are assumed to be restrictors for the downward search space of a head, but as soon as there is a certain amount of optionality for spell-out, the predictions about the inaccessibility of domains diminish. This is, of course, correct. However, phases are also assumed to lead to more economical derivations, likewise, by restricting the search space, and it is never so clear that strictly enforcing spell-out is indeed economical. The picture that emerges here is one in which the enforcement of spell-out itself is rendered dynamic as sort of a last resort option. Only if the derivations would stall, because no EF were available for the next step of Merge, would spell-out be an absolute necessity, otherwise it might even be more economical for the derivation to do nothing (other than continue merging via an EF). Notice also that whenever there is an optionality with respect to the operation of spell-out, there is an interpretive effect, i.e. the distinction between an equative and a predicative interpretation for SCs that consist of two DPs. This is what is expected, given the strongest minimalist thesis.
62 | Leah S. Bauke (48)
SC DEF the
PP in which garden
Now, the data in (43d) and (48) suggest that it is possible to subextract from inside the SC predicate. The wh-constituent which garden can apparently be fronted in both instances. This strongly suggests that on the one hand the wh-constituent is remerged as a phase head with an EF – a property that must hold of basically all wh-constituents in a system shaped by the H-α schema (cf. Narita 2011 for detailed discussion). On the other hand, the PP that contains the wh-constituent cannot be a phase, because if it were a phase it would spell-out its complement domain, making further extraction from inside the complement domain of P impossible (cf. also Fox und Pesetsky 2005; Uriagereka 2012 and many others for recent discussion).
4.5.2 Back to Russian This gives us an angle under which the pattern that can be observed in the Russian SCs, illustrated in section 4.1, can be handled. Let us look at the data in a little more detail and start with the example in (4b) repeated here for convenience as (49): (49)
Prezident SŠA byl [sc Prezident SŠA liderom president.nom USA.GEN was leader.INST svobodnyx nacij]. free.GEN.PL nation.GEN.PL ‘The president of the US was the leader of the free nations.’
The derivation for this SC is illustrated in (50): (50)
Derivation for (49): step 1 construction of the SO1 = the president of the US/Prezident SŠA & construction of the SO2 = the leader of the free nations/liderom svobodnyx nacij: step 2 Spell-out of the complement domain of one of the SOs (here SO1 ): SO1 ⇒ DEF DEF the
president of the US Prezident SŠA
4. What small clauses can tell us about complex sentence structure
step 3
| 63
Merger of DEF with SO2 : SO DEF the
SO2
DEF the
leader of the free nations liderom svobodnyx nacij
So the SC in (50) is the product of Merger of one phase head with a spelled-out complement domain, which thus has an EF available that allows it to be merged to a syntactic object that does not (necessarily) have an EF, i.e. the DP that is the SC predicate.¹⁶ In a similar vein the extraction data can be explained. Full extraction of the two SC constituents as in (5), repeated here as (51), is possible when the wh-words are both remerged as LIs via their EFs¹⁷: (51)
Kto kem byl kto kem? who.NOM what.INST was ‘Who was what?’
Here the structure for the SC is as follows: (52)
SO whEF kto
whEF kem
The fact that both wh-words are characterized by EFs is essential here. This is the reason why both of the SC constituents can be remerged in the first place.¹⁸
16 Recall from the discussion above that it is, of course, also possible to merge the two DPs as phase heads with spelled-out complement domains. This then leads to an equative construction rather than a subject-predicate structure (cf. also Bauke 2014). 17 Note incidentally that in (51) both SC constituents are moved. This then means that the sentence should get an equative interpretation. According to my informants, this is what holds for (51), but in those cases where both SC constituents are nominative marked – and where one of them has to stay inside the SC in order to avoid the illicit kto kto configuration in the present tense – a predicative interpretation is possible and indeed preferred. This is what would be expected under the present analysis: when both SC constituents are reduced, we get an equative interpretation, when only one is reduced, we get a predicative interpretation. 18 It is immediately clear that EFs on wh-constituents are a prerequisite for multiple wh-fronting. However, this cannot be the only prerequisite, because this would imply that all languages in which whwords have EFs are multiple wh-fronting languages. This is clearly not the case. It is not the aim of this paper to contribute to an investigation of the properties of multiple wh-fronting. Thus I’ll leave this
64 | Leah S. Bauke In those cases where we can observe subextraction of one of the wh-constituents, the wh-constituent itself must also be one with an EF, i.e. a phase head with a spelledout complement domain. However, because this is a case of subextraction from a larger syntactic object, this also means that the constituent that is extracted from must not be a phase head that has spelled out its complement domain. In other words, the interior of the syntactic object that is extracted from must still be accessible. Again the data from chapter one are repeated below as (53): (53)
a.
Kakoj strany byl prezident kakoj strany liderom which.GEN country was president.NOM leader.INST svobonyx nacij? free.GEN.PL nation.GEN.PL ‘Of which country was the president the leader of the free nations?’
b.
Kakix nacij byl prezident SŠA which nations.GEN was president.NOM USA.GEN liderom kakix nacij? leader.INST ‘Of which nations was the president of the US the leader?’
(53a) is a case of subextraction from the SC subject. This means that inside the syntactic object that constitutes the SC the subject constituent must still be fully accessible. Now, since the SC predicate as such is also a complex syntactic object and not just a LI with an EF, it must collapse its complement domain via spell-out in order to make the EF on the head of the syntactic object available for Merger with the complex syntactic object that is the SC subject. The step-by-step derivation of this construction is illustrated below: (54)
step 1 construction of the SO that is the SC predicate: liderom svobonyx nacij construction of the SO that is the SC subject: prezident kakoj strany
question unresolved at this point. Suffice it to say that further properties and possibly also parametric variation is involved in these matters. The question which elements may carry EFs is also highly relevant in this context. Again, I will not enter into a detailed discussion here; it is immediately clear that only phase heads – without specifiers for that matter – are potential candidates for Remerge. So movement phenomena other than wh-movement, e.g. topicalization, should also be exclusively available for phase heads with reduced complement domains. Alternatively, it could be argued that topicalization is movement that is motivated differently, e.g. pragmatically, and thus must be distinguished from other types of movement, e.g. syntactic movement and therefore cannot be subject to the restrictions that – among others – wh-movement is.
4. What small clauses can tell us about complex sentence structure
| 65
step 2 spell-out of the complement domain of the SO that is the SC predicate: SO ⇒
DEF
DEF liderom svobonyx nacij step 3 Merger of the SO that is the SC predicate, now reduced to a LI with a spelled-out complement domain and accessible EFs with the SC subject: SO
prezident kakoj strany
DEF liderom svobonyx nacij
step 4 Merger of the copula via the EF on the copula SO SO
bylEF
prezident kakoj strany step 5
DEF liderom svobonyx nacij
Remerger of the wh-constituent from inside the SC subject: SO SO
whEF kakoj strany bylEF
SO
prezident kakoj strany
DEF liderom svobonyx nacij
The derivation for the subextraction of a wh-constituent from inside the SC predicate proceeds in analogous fashion. The only difference is that here it is the SC predicate that must have an accessible domain whereas the SC subject must be merged as a LI with a spelled-out complement domain:
66 | Leah S. Bauke (55)
step 1 construction of the SO that is the SC predicate: liderom kakix nacij construction of the SO that is the SC subject: prezident SŠA step 2 spell-out of the complement domain of the SO that is the SC subject: SO ⇒
DEF
DEF prezident SŠA step 3 Merger of the SO that is the SC subject, now reduced to a LI with a spelled-out complement domain and accessible EFs with the SC predicate: SO
DEF prezident SŠA
liderom kakix nacij
step 4 Merger of the copula via the EF on the copula SO bylEF DEF prezident SŠA step 5
SO
liderom kakix nacij
Remerger of the wh-constituent from inside the SC predicate: SO SO
whEF kakix nacij bylEF
DEF prezident SŠA
SO
liderom kakix nacij
Further, the derivation of those SCs, already introduced in section 4.1, in which one SC constituent is fully extracted and the other is subextracted from, follows naturally along the lines of the current analysis. The only further step that is needed is that the constituent that is merged as a phase head with a spelled-out complement domain inside the SC in the derivations discussed above, i.e. the SC predicate in (54) and the SC subject in (55) is remerged outside the SC. This is possible because the EF on these constituents, which are phase heads with spelled-out complement domains, is always
4. What small clauses can tell us about complex sentence structure
| 67
available and thus always a potential candidate for Remerge. The relevant data from (12) in section 4.1 are repeated here in (56): (56)
a.
kakix nacij? Kto kakix nacij byl kto liderom leader.INST who.NOM which nations.GEN was ‘Who was of which nations the leader?’
b.
kakoy strany kem? Kem kakoy strany byl prezident who.INST which country.GEN was president.NOM ‘Of which country the president was what?’
Here, it should be sufficient to look at the derivation for one of the two sentences. (56a) is the example for full extraction of the SC subject and subextraction from inside the SC predicate, and in (56b) the situation is reversed. Here it is the SC subject that is extracted from and the SC predicate is the constituent that is fully remerged. Let us look at the derivation for the sentence in (56a) for expository purposes: (57)
step 1 construction of the SO that is the SC subject: kto construction of the SO that is the SC predicate: liderom kakix nacij step 2 Merger of the SC subject with the SO that is the SC predicate via the EF on the SC subject: SO whEF kto
liderom kakix nacij
step 3 Merger of the copula via the EF on the copula: SO SO
bylEF whEF kto
liderom kakix nacij
68 | Leah S. Bauke step 4 Remerger of the wh-constituent from inside the SC predicate: SO SO
whEF kakix nacij bylEF
SO
whEF kto
liderom kakix nacij
step 5 Remerger of the wh-constituent that is the SC subject: SO whEF kto
SO
whEF
SO
kakix nacij bylEF
whEF kto
SO
liderom kakix nacij
Finally, the ungrammaticality that results from simultaneously subextracting from the SC subject and the SC object falls out as a natural consequence. Again the relevant data are repeated below: (58) *Kakoj strany kakix nacij byl prezident kakoj strany which country.GEN which nations.GEN was president.NOM liderom kakix nacij? leader.INST From this conflict, it results that both SC constituents, the subject and the predicate, must be accessible for subextraction. This means that none of the two constituents can discharge its complement domain via spell-out, because this would make it inaccessible for subextraction at a later point. On the other hand, at least one of the two constituents must be reduced to a LI with an EF, which can only be achieved via spell-
4. What small clauses can tell us about complex sentence structure
| 69
out of the complement domain, because otherwise it would be impossible to merge the two constituents inside the SC.¹⁹ Note incidentally that the following data from English also confirm this pattern: (59)
a. b. c.
Of which nations is the president of the US the leader? Of which country is the president the leader of the free nations? *Of which country is the president intelligent/in the rose garden?
(59a) is a case of subextraction from inside the SC predicate. This presupposes that the SC subject is merged as a phase head with a spelled-out complement domain, i.e. as a LI with an EF. (59b) illustrates the reverse scenario. Here it is the SC subject that is extracted from and the predicate must thus be merged as a LI with an EF. This is perfectly in line with the analysis presented here, and also the ungrammaticality of (59c) falls into place. Again, subextraction from the SC subject has taken place. This means that the SC subject is not a potential candidate for spell-out and reduction of the complement domain. It is and remains a complex syntactic object without acces-
19 One reviewer asks whether there is a loophole in the constraint that would allow for double subextraction when the subextracted DPs/whPs are moved to the edge of the phrases they are subextracted from. This is schematically illustrated below: i.
[SC [DP1 . . . DP2 . . . ] [DP3 . . . DP4 . . . ]]
Now the complement domains of DP2 and DP4 are reduced via spell-out: ii.
[SC [DP1 . . . D2 . . . ] [DP3 . . . D4 . . . ]]
And then D2 and D4 are moved to the edge of DP1 and DP3 respectively: iii.
[SC [DP1 D2 [D’1 . . . t2 . . . ]] [DP3 D4 [D’3 . . . t4 . . . ]]]
Now spell-out of the complement domains of D1 and D3 leads to the following result: iv.
[SC [ D2 D1 ] [ D4 D3 ]]
And in the next step of the derivation the phase heads D2 and D4 spell out their complement domains leading to the following result: v.
[SC [ D2 D4 ]
And it should now be possible for D2 and D4 to be remerged outside the SC, since they are both LIs with a reduced complement domain and accessible EFs: vi.
D2 D4 . . . [SC [ t2 t4 ]
Two comments are in order here. First, in step (i) of the derivation Merger of DP1 and DP3 as the two SC constituents is impossible, because they both are complex SOs without EFs. So, at least one of the two DPs must reduce its complement domain prior to Merger with the other DP, rendering the DP inside inaccessible for further movement. Second, Merger of D2 and D4 in the edge of DP1 and DP2 respectively does not circumvent the problem but creates at least two new problems. In steps (iii)/(iv) new complex SOs are created. So what looks like subextraction of D2 and D4 in step (vi) of the derivation is actually movement of the complete SO. Additionally the labels for the DPs are reversed. What starts out as DP2 embedded in DP1 ends up as DP2 with DP1 embedded in its complement domain.
70 | Leah S. Bauke sible EFs. This would automatically require the SC predicate to be merged with the SC subject as a LI with an EF. However, the predicate constituent is one which cannot reduce its complement domain, because the relevant head is not a phase head. Thus, it is the impossibility of merging the SC subject with the SC predicate in the first place that leads to the ungrammaticality of (59c), rather than e.g. the impossibility to subextract from inside a subject.²⁰ Subextraction from subjects is possible, as long as these are still accessible, i.e. as long as they have not been reduced to a LI with an EF. Thus, the widely discussed inaccessibility of moved subjects also receives a natural explanation under the current account (cf. also Narita 2011).²¹ For a subject to move, it must be susceptible to (Re)merge. This means that it must either Merge with a LI that has an EF or it must itself be reduced to a LI with an EF via spell-out. However, once this reduction has taken place, subextraction is not an option anymore.
4.6 Some more evidence Further evidence that the account pursued here, under which the operation Merge is subject to the H-α schema, is on the right track comes from the following observations. In the case of the combination of full extraction of one SC constituent and subextraction from another that we have seen for Russian above, superiority effects disappear. This is illustrated by the following data: (60)
a.
Kakix nacij kto byl kto liderom kakix nacij? which nations.GEN who.NOM was leader.INST ‘Who was of which nations the leader?’
20 Notice the apparent difference between the AP and the PP in (59c). It could, of course, be argued that the line of reasoning applied here only really holds for the PP and not for the AP, since the latter consists of just a single LI anyway and should thus be mergeable via the EF on this LI with the SC subject. Then it would be expected that subextraction from the SC subject is still possible, because the complement domain of this SO need not be reduced. Thus, at least the AP construction in (59c) would be expected to be grammatical – contrary to fact. However, this really depends on the status of intelligent. In a system of structure building that is in line with the H-α schema, it might well be that the AP is more complex than it looks, for example, if we assume that LIs enter the derivation as roots that are merged with a categorizing head in a first derivational step. Both the root LI and the categorizer, e.g. a little x-head in the sense of Marantz (2007), bear EFs, but the SO that results from this Merge operation, i.e. the AP intelligent, is a complex SO without EFs. Thus, the ungramaticality of (59c) is again expected and in line with the current analysis. 21 In fact, as Narita (2011: 59) illustrates, this system can account for why non-moved subjects are more susceptible to subextraction than moved subjects, as is common e.g. in Japanese and for postverbal subjects in Romance languages (cf. Narita ibid; Uriagereka 2009). The remaining cases in which non-moved subjects cannot be subextracted from must then be related to independent factors.
4. What small clauses can tell us about complex sentence structure
(61)
b.
Kto kakix nacij byl kto liderom kakix nacij? leader.INST who.NOM which nations.GEN was ‘Who was of which nations the leader?’
a.
Kakoj strany kem byl which countries.GEN what.INST was prezident kakoy strany kem? president.NOM ‘Of which countries was the president what?’
b.
Kem kakoy strany byl what.INST which country.GEN was prezident kakoy strany kem? president.NOM ‘Of which country the president was what?’
| 71
The examples in (60b) and (61b) are the familiar examples that have already been discussed in the preceding section. The examples in (60a) and (61a) illustrate the same type of extraction as their respective b)-counterparts. However, the order in which the two wh-constituents are merged is reversed without any difference in grammaticality or interpretation. This can be taken as an indication that the two SC constituents are directly merged and that there is no intermediate functional projection that mediates between them. Incidentally, superiority effects do not even occur in believe-type SCs (with one little caveat, see below). This again strongly suggests that these types of SCs also do not have a functional projection and that the two constituents are directly merged. Provided, of course, at least one of them can be merged as a LI with an EF, which in turn is a very good indication that the relevant constituent must have spelled out its complement domain and is thus inaccessible for subextraction. Consider the following data in this context: (62)
a.
Kogo kakix nacij ty dumaeš oni sčitajut whom.ACC which nations.GEN you think they consider liderom? leader.INST ‘Who do you think they consider of which nations the leader?’
b. *Kakix nacij kogo ty dumaeš oni sčitajut which nations.GEN who.ACC you think they consider liderom? leader.INST
72 | Leah S. Bauke (63)
a.
Kem kakoj strany ty dumaeš oni sčitajut who.INST which country.GEN you think they consider prezidenta? president.ACC ‘Who do you think they consider of which state the president?’
b.
Kakoj strany kem ty dumaeš oni sčitajut which country.GEN who.INST you think they consider prezidenta? president.ACC ‘Who do you think they consider of which state the president?’
Let me start with the examples in (63). In (63a) the SC subject is fully extracted from inside the SC and the SC predicate is subextracted from. (63b) illustrates the same extraction phenomena; the only difference between the two examples is that in (63a) it is the fully extracted constituent that is remerged in the highest position whereas it is the subextracted constituent in (63b) that is remerged in this position. Again there are no differences in grammaticality and interpretation between the two sentences. In (62) we are looking at the reverse scenario. Here it is the SC predicate that is fully extracted and the SC subject is subextracted from. The data show that subextraction from an instrumental DP is not possible. This can be related to the fact that the exceptional case marking verb believe requires an accusative marking instead of an instrumental. However, apart from this small caveat even believe-type SCs behave completely as expected and subextraction from both SC constituents at the same time is again not possible no matter which constituent is remerged in the highest position: (64)
a. *Kakoj strany kakix nacij ty dumaeš oni which country.GEN which nations.GEN you think they sčitajut prezidenta liderom? consider president.ACC leader.INST b.
*Kakix nacij kakoj strany ty dumaeš oni which nations.GEN which country.GEN you think they sčitajut prezidenta liderom? consider president.ACC leader.INST
4.7 Conclusion The analysis presented in this paper builds on the central insight in Narita (2011) that the H-α schema restricts the complexity of syntactic structure by restraining the combinatorial possibilities of structure building. The binary operation Merge is shaped in such a way that of the two elements that are input to the operation at least one has to be
4. What small clauses can tell us about complex sentence structure
| 73
a LI with an EF. This makes complexity a relative notion, because what counts as a LI is not restricted to those elements that enter the derivation from the lexicon or the numeration (cf. Trotzke and Zwart 2014). Instead, complex syntactic objects already constructed in the course of the derivation can be reduced to LIs with accessible EFs via spell-out. Those LIs that are phase heads can spell-out their complement domain. All that remains (active) in the derivation then is a phase head with an accessible EF that can be remerged with a complex syntactic object. So in a sense all recursive embedding must proceed via LIs with EFs and structure building is achieved by recursively applying Merge. These insights have been applied to the special case of SCs in this paper and it has been shown that under the general implications of the H-α schema there is good evidence that SCs are built by direct Merger of the two constituents that make up the SC, i.e. without any intervening functional projections. Heretofore unaccounted extraction and subextraction patterns in Russian and English SCs can be accounted for straightforwardly under the analysis presented here and follow naturally. Acknowledgement: Many thanks to the audiences at the workshop on complex sentences, types of embedding, and recursivity and the DGfS workshop on syntactic variation where the ideas in this paper have been presented. Many thanks also to Andreas Blümel, Erich Groat, and Thilo Tappe for their comments and remarks and to two anonymous reviewers for their comments. All errors, of course, are mine.
References Aarts, Bas. 1992. SCs in English. The non-verbal types. Berlin: Mouton de Gruyter. Abels, Klaus. 2003. Successive-cyclicity, anit-locality and adposition stranding. PhD diss. UConn. Al-Horais, Nasser. 2007. The categorial status of the SC node: a minimalist approach. NDWPL 13: 96–108. Bauke, Leah. 2014. Symmetry breaking in syntax and the lexicon. Amsterdam: John Benjamins. Boeckx, Cedric. 2008. Bare syntax. Oxford: Oxford University Press. Boeckx, Cedric and Grohmann, Kleanthes. 2007. Putting phases into perspective. Syntax 10: 204–222. Bošković, Zeljko. 2002. On multiple wh-fronting. LI 33: 351–383. Bošković, Zeljko. 2005. On the locality of left branch extraction and the structure of NP. Studia Linguistica 59: 589–644. Carstens, Vicki. 2000. Concord in minimalist theory. LI 31: 319–355. Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Step by step: essays on minimalist syntax in honor of Howard Lasnik, Roger Martin, David Michaels and Juan Uriagereka (eds.), 89–155. Cambridge, MA: MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Ken Hale. A life in language, Michael Kenstowicz (ed.), 1–52, Cambridge, MA: MIT Press. Chomsky, Noam. 2004. Beyond explanatory adequacy. In Structures and beyond, Adriana Belletti (ed.), 104–131. Oxford: Oxford University Press. Chomsky, Noam. 2008. On Phases. In Foundational issues in linguistic theory, Robert Freidin, Carlos Otero and Maria Luisa Zubizarreta (eds.), 133–166. Cambridge, MA: MIT Press.
74 | Leah S. Bauke Chomsky, Noam. 2013. Problems of projection. Lingua 130: 33–49. den Dikken, Marcel. 2006. Relators and linkers: the syntax of predication, predicate inversion, and copulas. Cambridge, MA: MIT Press. den Dikken, Marcel. 2007a. Phase extension. Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics 33: 1–41 den Dikken, Marcel. 2007b. Phase extension: a reply. Theoretical Linguistics 33: 133–163. Fox, Danny and Pesetsky, David. 2005. Cyclic linearization of syntactic structure. Theoretical Linguistics 31: 1–46. Heck, Fabian. 2009. On certain properties of pied-piping. LI 40: 75–111. Heycock, Caroline and Kroch, Anthony. 1999. Pseudocleft conectedness: implications for the LF interface level. LI 30: 365–398. Hiraiwa, Ken. 2005. Dimensions of symmetry in syntax: agreement and clausal architecture. Ph.D. diss. MIT. Jayaseelan, K. A. 2008. Bare phrase structure and specifier-less syntax. Biolinguistics 2: 87–106. Kayne, Richard. 1984. Connectedness and binary branching. Dordrecht: Foris. Kayne, Richard. 2010. Why are there no directionality parameters? Ms. NYU. Marantz, Alec. 2007. Phases and words. Ms. NYU. Matushansky, Ora. 2005. Going through a phase. In Perspectives on phases, Martha McGinnis and Norvin Richards (eds.), 157–181. Cambridge, MA: MIT Press. Matushansky, Ora. 2007a. What’s in a name. Handout from talk at PALMYR: Paris. Matushansky, Ora. 2007b. Predication and escape hatches in phase extension theory. Theoretical Linguistics 33: 93–104. Moro, Andrea. 2000. Dynamic antisymmetry. Cambridge, MA: MIT Press. Müller, Gereon. 2004. Verb-second as vP-first. Journal of comparative Germanic linguistics 7: 179–234. Narita, Hiroki. 2009. Full interpretation of optimal labeling. Biolinguistics 3: 213–254. Narita, Hiroki. 2011. Phasing-in full interpretation. Ph.D. diss. Harvard University. Ott, Dennis. 2008. Notes on noun ph(r)ases. Ms. Harvard University. Řezáč, Milan. 2010. Phi-features and the modular architecture of language. Dordrecht: Springer. Richards, Marc. D. 2012. No phase is an island (?): an attempt to constrain internal ‘free merge’ via the edge feature. Handout from talk at syntax seminar. University of Frankfurt. Samuels, Bridget. 2008. The structure of phonological theory. Ph.D. diss. Harvard University. Sternefeld, Wolfgang. 2007. Zur Interpretierbarkeit von morpho-syntaktischen Merkmalen. Ms. Universität Tübingen. Svenonius, Peter. 2004. On the edge. In Peripheries: syntactic edges and their effects, David Adger, Cécile de Cat and George Tsoulas (eds.), 261–287. Dordrecht: Kluwer. Trotzke, Andreas and Zwart, Jan-Wouter. 2014. The complexity of narrow syntax: Minimalism, representational economy, and simplest Merge. In Measuring grammatical complexity, Frederick J. Newmeyer and Laurel B. Preston (eds.), 128–147. Oxford: Oxford University Press. Uriagereka, Juan. 2009. Spell-bound: dynamical frustration and the accordian model. Ms. University of Maryland. Uriagereka, Juan. 2012. Spell-out and the minimalist program. Oxford: Oxford University Press.
Werner Frey and Hubert Truckenbrodt
5 Syntactic and prosodic integration and disintegration in peripheral adverbial clauses and in right dislocation/afterthought 5.1 Introduction and background 5.1.1 Overview In this paper we bring together two lines of research. This is the work on peripheral adverbial clauses (PACs) by Frey (2011, 2012) and the work on right dislocation and afterthought by Truckenbrodt (to appear). Both relate to syntactic integration, to root clauses, to intonation phrases and sentence stress. In this paper we would like to integrate them into a single set of analytical assumptions that relate to the notion of integration by Reis (1997) and Holler (2008) and simultaneously the notion of root sentence (unembedded root clause), which according to B. Downing (1970), henceforth Downing (1970), triggers obligatory pauses (intonation phrase boundaries). The combined account involves minor adjustments of the preceding independent accounts, and it brings out similarities and differences among the phenomena investigated here. In the remainder of this section we introduce notions of syntactic integration, root sentences, and their consequences for information structure and prosody. In section 5.2 we discuss peripheral adverbial clauses. In section 5.3 we address right dislocation and afterthought. Peripheral adverbial clauses and dislocated constituents in the right periphery are compared in a summary of the results in section 5.4.
5.1.2 Integration and root sentences We begin the discussion with the notion of integration from Reis (1997) and Holler (2008). According to their division, non-restrictive relatives (1a) and continuative whrelative clauses (1b) are not integrated into their host clause. (1)
a.
Ich besuche meine Schwester, die in Augsburg wohnt. I visit my sister who in Augsburg lives ‘I am visiting my sister, who lives in Augsburg.’
b.
Um neun Uhr schlief das Kind ein, woraufhin die Eltern at nine o’clock slept the child in whereupon the parents fortgingen. left ‘At nine o’clock the child fell asleep, whereupon the parents left.’
76 | Werner Frey and Hubert Truckenbrodt Typical integrated clauses are argument clauses (2a), restrictive relative clauses (2b), and temporal adjunct clauses (2c). (2)
a.
Peter hofft, dass Maria die Einladung annimmt. Peter hopes that Maria the invitation accepts ‘Peter hopes that Maria will accept the invitation.’
b.
Peter schreibt jedem Besucher, der seine Adresse hinterlassen Peter writes every visitor who his address left hat. has ‘Peter writes to every visitor who left his address.’
c.
Peter kam erst zur Ruhe, als alle gegangen waren. Peter came only to rest when everyone left were ‘Peter only came to rest when everyone had left.’
We connect the notion of integration from Reis and Holler to the notion of root sentences by Downing (1970). According to Downing, non-restrictive relatives, parentheticals¹, vocatives, and left dislocated elements in English are not syntactically part of the root sentence that they occur with. Furthermore, coordinate sentences are separate root sentences for Downing, and they are not integrated with each other for Reis and Holler. On the other hand, each example in (2) would constitute a single root sentence in Downing’s analysis. The notion of root sentence is crucial for Downing, since he defines “obligatory pauses” in these terms. The rule mapping between syntax and prosody is that each root sentence is delimited by obligatory pauses. Nespor and Vogel (1986) adapted this to the prosodic hierarchy: Each root sentence is separated by intonation phrase boundaries. See L. Downing (2011), henceforth Downing (2011), Kan (2009), Selkirk (2005, 2011) for recent analyses that build on these classical suggestions; for our purposes it is sufficient to work with the classical generalizations. After Downing (1970) embedded root clauses were discovered as well (Hooper and Thompson 1973). We stick with the terminological distinctions from the literature and distinguish root clauses from root sentences. Root sentences are unembedded root clauses. (Unembedded and embedded) root clauses are diagnosed by root clause phenomena such as English topicalization, German verb-second, and in many languages, the occurrence of certain adverbials or modal particles (MPs) that relate to the attitude of the speaker or to the speech act performed. These phenomena can occur unembedded in root sentences or embedded under a range of verbs which encode an illocu-
1 From the perspective pursued in this paper, not all these elements are prosodically of the same kind. Parentheticals, in particular, have been argued to often lack the expected intonation phrase boundaries (Dehé 2009). They also lack the systematic sentence stress that we employ as a criterion below.
5. Syntactic and prosodic integration and disintegration
|
77
tion of their logical subjects. See Heycock (2006) for a review of embedded root phenomena. The root clause phenomena and their distribution suggest that they show a relation to speaker attitude and illocutionary force in some way. Root clauses have been identified with syntactic ForcePs in Bayer (2001), Haegeman (2002, 2004), and Coniglio (2011), among others. While this is compatible with our analysis, we employ the terms CP, root clause, and root sentence in our discussion. We employ the classical definitions of Downing (1970) for root sentences. The definition of root sentences seems not to be contingent on lexical selection, since restrictive relatives as in (2b) and certain adverbial clauses like (2c) are not lexically selected, yet are not separate root sentences. On the other hand, the definition of root sentence also cannot simply be ‘an undominated CP in a phrase marker’. This is because coordinated sentences like [CP1 [CP2 it is snowing] and [CP3 it is cold]] have two coordinated root sentences CP2 and CP3 , yet both are dominated by a higher node, here CP1 . Downing therefore employs an auxiliary definition: the predicate sentence. This, for him, is an S node immediately dominating a VP. Let us here call a predicate sentence a CP with a VP ‘of its own’ in the intuitive sense.² Following Downing, a root sentence is then a clause not dominated by a predicate sentence, as in (3). (3)
A root sentence is a CP that is not dominated by a predicate sentence CPP .
By (3) both CP2 and CP3 are root sentences in the coordinated structure [CP1 [CP2 it is snowing] and [CP3 it is cold]]. They are dominated by another CP, namely CP1 , but CP1 is not a predicate sentence: It has no predicate of its own. The appositive and continuative relative clauses in (1) are non-integrated in the sense of Reis (1997) and Holler (2008) and separate root sentences for Downing (1970). Downing (1970: 142ff) derives non-restrictive relatives from coordinated root sentences, from which they inherit their prosodic boundaries. Reis (1997: 138) employs a structure in which she adjoins non-restrictive relatives to the host clause. We reject both these structures and instead adopt the orphan analysis of non-restrictive relatives from Safir (1986), Holler (2008) and Frey (2011, 2012). In this analysis nonrestrictive relatives are simply not part of the same phrase marker as the host clause, as illustrated in (4).
2 A formal definition is that a predicate sentence is a CP that contains a VP and the CP node is not separated from the VP node by another CP node.
78 | Werner Frey and Hubert Truckenbrodt (4)
CP | TP DP
VP
John1
CP who1
will stop by tomorrow is in town
Following Frey (2011, 2012) this analysis will allow us to distinguish the behavior of non-restrictive relatives from that of peripheral adverbial clauses, which are arguably adjoined to their host clause. In relation to the definition (3), both CPs in (4) are root sentences in the orphan analysis, since neither of them is dominated by a clause with a predicate of its own. In particular, the relative clause in (4) is not dominated by the host clause CP because it is an orphan, i.e. not structurally attached to the host clause.
5.1.3 Information structure consequence of (non-)integration and root sentences Downing (1970), as well as Brandt (1990), Reis (1997) and Holler (2008), see prosodyrelated consequences of root sentences/non-integrated clauses. For Brandt, Reis and Holler, this is captured in terms of separate focus-background structures for nonintegrated clauses. (1b) is repeated here as (5) with some structure added for illustration. Each bracketed constituent is a focus-background structure in this perspective. (5)
[Um neun Uhr schlief das Kind ein] [woraufhin die Eltern at nine o’clock slept the child in whereupon the parents fortgingen]. left ‘At nine o’clock the child fell asleep, whereupon the parents left.’
An important aspect of this perspective, pointed out by Brandt (1990), Reis (1997) and Holler (2008), is that it is not possible to construct a single focus-background structure across the entire utterance. We adopt this here as the principle in (6). Following Rooth (1992) we denote a focus-background structure (or scope of a focus) with the squiggleoperator ∼. (6)
Brandt (1990), Reis (1997) and Holler (2008): A focus-background structure ∼[ . . . ] cannot be larger than a root sentence.
5. Syntactic and prosodic integration and disintegration
|
79
The restriction is illustrated in (7) and (8), where (7) is a comparison case that will be considered first. (7a) is a context question for the focused example (7b). In (7b) there is a single root sentence that contains a fronted subject clause. This root sentence can form a single focus-background structure, indicated by the scope of ∼. In the theory of Rooth (1992) this requires alternatives in the context of the utterance that are calculated in (7b) at the level of ∼, replacing the focus with alternative values. These alternatives must therefore be as shown in (7c). The requirement that these alternatives are given in the context is satisfied because the meaning of the context question in (7a) semantically defines such a set of alternatives, i.e. the question asks essentially which of the alternatives in (7c) is true. (7)
a.
Wie ist es für dich, dass Peter absagen musste? ‘What is it like for you that Peter had to cancel?’
b.
∼ [Dass Peter absagen musste [hat mich enttäuscht]F . that Peter cancel had.to has me disappointed ‘That Peter had to cancel disappointed me.’
c.
{[that Peter had to cancel [has made me sad]], [that Peter had to cancel [has given me relief]], . . . }
Now, we might expect the same to be possible in (8), where ∼ and F in (8b) would lead to the alternatives in (8c) which are, one might think, the options among which the context question (8a) asks for the truth. Yet (8b) is not a nicely matching answer to the question (8a). If the focus-background structure indicated in (8b) were acceptable, we could reasonably expect (8b) to be a nicely matching answer to (8a). (8)
a.
Wie ist es für dich, dass Peter absagten musste? ‘What is it like for you that Peter had to cancel?’
b.
#/*∼ [Peter musste absagen, [was mich enttäuscht hat]F ] . Peter had.to cancel which me disappointed has ‘Peter had to cancel, which disappointed me.’
c.
{[Peter had to cancel [which made me sad]], [Peter had to cancel [which gave me relief]], . . . }
The focus-background structure shown in (8b) is ruled out by (6), the claim of Brandt, Reis and Holler, that non-integrated clauses cannot form a focus-background structure together with their host clause.
5.1.4 Prosodic consequences of non-integration and root sentences For Downing (1970) root sentences are relevant for the prosody. Downing argues at the length of a 215-page thesis for a single generalization across a variety of differ-
80 | Werner Frey and Hubert Truckenbrodt ent syntactic constructions in English. The generalization, his Obligatory Boundary Insertion (OBI) convention, is that the edges of root sentences require the insertion of prosodic boundaries, which he calls phonological phrase boundaries and diagnoses by obligatory pauses. The terminology changed later on, and Nespor and Vogel (1986) adapted the formulation of the OBI to the prosodic hierarchy. In the prosodic hierarchy the prosodic constituents relevant to the OBI are intonation phrases. In (5), for example, each bracketed constituent, i.e. each root sentence, is an intonation phrase. (Additional, optional, intonation phrase boundaries are allowed by the accounts of Downing and of Nespor and Vogel.) We adopt this as well: (9)
OBI of Downing (1970), as adapted by Nespor and Vogel (1986): Each root sentence is mapped to one (or more) intonation phrases.
In addition, Nespor and Vogel (1986) construe the intonation phrase as a stress domain. This perspective is also taken for German in Uhmann (1991) and in much later work on German sentence prosody. In such an account sentence stress is assigned as the prosodic head of the intonation phrase (I). We adopt this as well: (10)
Stress-Domain-I Each intonation phrase carries sentence stress.
Thus in (5), where we have separate root sentences, we have separate intonation phrases on account of (9). Each of them carries sentence stress on account of (10). Notice that (9) and (10) taken together have the effect that each root sentence requires at least one sentence stress. This is formulated and given a name in (11). (11)
Stress-RS: Each root sentence contains sentence stress. (Derived from the OBI in (9) and Stress-Domain-I in (10).)
Since sentence stress is sometimes easier to observe than intonation phrase boundaries, we will often discuss the prosodic effect of root sentences in terms of Stress-RS. Notice that Stress-RS cannot be overridden by information structure. The empirical side of this is that there is no way of getting rid of the two instances of sentence stress in (5). We think that this is related to the intuition of two focus-background structures in (5). In the current account, the impossibility of overriding Stress-RS with information structure has two sides. First, focus cannot override Stress-RS, since (6) prevents this for essentially geometrical reasons. A focus-background structure, if it were to override Stress-RS, would need to be larger than the root sentence, like in (8b). In that structure the first root sentence in (8b), Peter musste absagen ‘Peter had to cancel’, would have a reason to be destressed (which would now go against Stress-RS) because a focus-background structure normally involves stressing the focus and – crucially for the point here – destressing the background, which in (8b) is the first root sentence. However, such an overarching focus-background structure is blocked in principle by (6). Therefore Stress-RS cannot be overridden by focus.
5. Syntactic and prosodic integration and disintegration
|
81
A second way in which Stress-RS might conceivably be overridden is in terms of the effect of contextual givenness, which requires destressing (Ladd 1983). This is formalized in Féry and Samek-Lodovici (2006) in terms of a feature G and its stressrejecting effect as in (12): (12)
*Stress-Given: Do not assign sentence stress to a constituent marked as G (i.e. contextually given).
An example from Höhle (1992) shows that sentences that are contextually given in their entirety nevertheless carry sentence stress. The context (13a) makes the sentence (13b,c) entirely given. It may carry verum focus as in (13b) or not as in (13c). Either way it needs to carry sentence stress. (13)
a.
Ich habe Hanna gefragt, was Karl gerade macht, und sie hat die alberne Behauptung aufgestellt, dass er ein Drehbuch schreibt. ‘I asked Hanna what Karl is currently doing, and she made the silly claim that he is writing a script.’
b.
(Das stimmt.) Karl schreibt ein Drehbuch. ‘(that’s right) Karl is writing a script.’
c.
(Das stimmt.) Karl schreibt ein Drehbuch. ‘(that’s right) Karl is writing a script.’
For representing this we must assume that Stress-RS overrides *Stress-Given. This has the desired effect in (13), where Stress-RS forces sentence stress on the all-given root sentence. Consequently, in (5), the two instances of sentence stress in the two root sentences are also forced by Stress-RS and cannot be prevented from being assigned by contextual givenness, even if one of the two root sentences is given in its entirety. In sum, we have assembled an analysis of syntactic (dis-)integration in terms of root sentences. A root sentence is a sentence node not dominated by the sentence node of a predicate sentence. Coordinated sentences form separate root sentences, as do appositive and continuative relatives and their host clauses. Root sentences delimit focus-background structures. Further, they are delimited by intonation phrase boundaries and carry sentence stress (Stress-RS) in a way that cannot be overridden by information structure. We now turn to analyzing peripheral adverbial clauses and right dislocation/afterthought in these terms.
82 | Werner Frey and Hubert Truckenbrodt
5.2 Peripheral adverbial clauses (PACs) 5.2.1 The syntactic position and c-command relation of PACs Haegeman (2004, 2006) distinguishes central from peripheral adverbial clauses for English. The analysis is extended to German in Frey (2011, 2012), on which the following review is based. Typical central adverbial clauses specify time or manner, or form the antecedents of a conditional. Typical peripheral adverbial clauses, PACs, are adversative or concessive adverbial clauses. (Clauses specifying a reason can be central or peripheral, primarily depending on the complementizer.) All the relevant adverbial clauses are V-final in German.³ A number of c-command tests show that central adverbial clauses can be c-commanded by elements in the host clause, while peripheral adverbial clauses cannot. For example, central adverbial clauses can be in the scope of negation, in contrast to peripheral ones. (14b) does not have a reading in which Peter came but not despite his working duties but despite his tiredness. (14)
a.
Peter wird nicht kommen, sobald er kann, sondern sobald Peter will not come as.soon.as he can but as.soon.as es Clara erlaubt. (central) it Clara allows
b. *Peter wird nicht kommen, obwohl er arbeiten muss, sondern Peter will not come although he work must but obwohl er schlafen sollte. (peripheral) although he sleep should Similarly, central adverbial clauses allow binding of a pronoun in them by a quantifier in the host clause, while peripheral ones do not:
3 The following table classifies some often used conjunctions with respect to the clause types central and peripheral. Some can introduce clauses of either. Conjunction als (‘when’) sobald (‘as soon as’) während (temporal ‘while’) wenn (‘if’, event condition) weil (‘because’) da (justifying ‘because’) obwohl (‘although’) trotzdem (‘even though’) während (adversative ‘while’)
central (per default)
peripheral
(possible, if triggered)
5. Syntactic and prosodic integration and disintegration
(15)
a.
|
83
Keineri hat protestiert, als eri unterbrochen wurde. (central) noone has protested when he interrupted was ‘Noone protested when he was interrupted.’
b. *Keineri hat protestiert, obwohl eri unterbrochen noone has protested although he interrupted wurde. was ‘Noonei protested although hei was interrupted.’
(peripheral)
These differences can be accounted for by adopting the syntactic analysis of Haegeman (2004): PACs are adjoined to CP, and thus not c-commanded by other elements in the host clause. Central adverbial clauses, on the other hand, stand inside of CP, and can thus be c-commanded by elements in the host clause. Despite these indications of disintegration, PACs show a number of signs of syntactic integration. Thus, they can contain a pronoun that is bound by a quantifier in a clause which is higher than the host clause of the PAC, as in (16) (Frey 2011, 2012). (16)
Keineri hat gedacht, [[andere werden bevorzugt] [während eri doch noone has thought others are preferred while he MP der Richtige sei]] the right.one be ‘Noonei thought that others are preferred while hei is the right one.’
This separates PACs from appositive relatives, as shown in (17a), and from continuative wh-relatives, as shown in (17b). The examples in (17) demonstrate that these relative clauses cannot be attached to an embedded clause. Hence (17b) does not have a reading with the continuative relative in the scope of the negation of the matrix clause. (17)
a.
[Peter called everyone. However . . . ] *Keineri hat gesagt, dass Peter (*[der ihni angerufen hat]) ihni noone has said that Peter who him called has him um etwas gebeten hat. for something asked has ‘Noonei has said that Peter (*who had called himi ) asked himi for something.’
b. *Keiner hat gesagt [dass Peter gewonnen hat, worüber noone has said that Peter won has, where.about sich Maria gefreut hat] herself Maria delighted has ‘Noone said that Peter had won, about which fact Maria has been delighted.’
84 | Werner Frey and Hubert Truckenbrodt The orphan analysis of non-restrictive relatives in Safir (1986), Holler (2008) and Frey (2011, 2012) provides a plausible approach to this. The appositive and continuative relatives are assumed to be not structurally connected to their host clause as in (4). They therefore cannot form a constituent with their host clause that could be subordinated as in (17b) and cannot be c-commanded from a higher clause as in (17a). PACs, on the other hand, are structurally attached to the host clause as in Haegeman’s analysis. As (16) shows they can therefore be subordinated together with their host clause and can be c-commanded from the outside by elements c-commanding their host clause. They are just attached fairly high to their host clause. A further observation due to Frey (2011) is that PACs can stand in the German Vorfeld (prefield): (18)
[Während Hans sonst bei schönem Wetter einen Ausflug while Hans otherwise in beautiful weather an excursion macht] ist er gestern zu Hause geblieben. makes is he yesterday at home stayed ‘While Hans otherwise makes an excursion when there is nice weather, he stayed at home yesterday.’
Positioning in the German prefield is taken as an unambiguous sign of its full integration into the clause by e.g. König and van der Auwera (1988). The prefield position is standardly analyzed as SpecCP, and thus a syntactic position genuinely inside of the German clause (see e.g. Grewendorf 2002; Haider 2010). SpecCP in German is also not followed by a systematic intonation phrase break (see e.g. Truckenbrodt 2002, 2005). Thus, revising Haegeman’s analysis, Frey (2011, 2012) argues that PACs are structurally base-generated as part of the host clause in very high positions, namely in SpecCP or by adjunction to CP. What requires the high positioning of the PACs? A hypothesis might be that PACs must have high positions, since they semantically combine with a proposition, the proposition expressed in their host clause. However, this possible reason can be excluded. Clauses introduced by weil ‘because’ also take scope over a proposition (Dowty 1979), normally the proposition expressed by their host clause. We also point out that causal adverbs are base-generated higher than the thematic subject position in German, plausibly for this reason (Frey 2003). Yet elements from the host clause can ccommand a weil-clause. In (19), for example, the bracketed weil-clause relates causally to the proposition expressed by (likewise in brackets) ti protestiert, and the quantifier scopes over the combination of the two bracketed constituents, binding the trace and the coindexed pronoun.⁴
4 An additional trace ti is included as a reminder that German allows A-scrambling; otherwise a weak crossover effect would be expected in this example.
5. Syntactic and prosodic integration and disintegration
(19)
|
85
eri unterbrochen wurde] Keineri hat ti [ti protestiert] [weil protested because he interrupted was noone has ‘Noonei protested because hei was interrupted.’ (i.e. people may have protested for other reasons)
Therefore, taking a propositional argument may motivate attachment at a certain height (above the underlying position of the subject); however, it cannot motivate the considerably height to which PACs are limited (in which no element from the host clause can c-command them). There must then be another reason that forces PACs to take such high positions relative to their host clause.
5.2.2 PACs as root clauses We review arguments from Frey (2011, 2012) that PACs are inherently root clauses, and that the reason for their high attachment is plausibly related to this property. (20)
PACs are inherently root clauses (Haegeman 2004, 2006; Frey 2011, 2012).
PACs attach to embedded clauses only under predicates that license root phenomena. For example, they occur under the root-clause-embedding verb meinen ‘think’ as in (21a) but not under the non-root-clause embedding verb zurückweisen ‘reject’ as indicated by the star in (21b). (21)
a.
Max meint, dass Maria Fußball liebt, während Paul für Opern Max thinks that Maria soccer loves while Paul for opera schwärmt. raves ‘Max thinks that Maria loves soccer while Paul raves about opera.’
b. *Max wies zurück, dass Maria Fußball liebt, während Paul für Max rejected that Maria soccer loves while Paul for Opern schwärmt. opera raves ‘Max rejected that Maria loves soccer while Paul raves about opera.’ Furthermore, Coniglio (2011) and Frey (2011, 2012) observe that PACs allow MPs as in (22), which are arguably a root clause phenomenon (Bayer 2001; Coniglio 2011). The examples in (23), which show this as well, are from Thurmair (1989: 78).
86 | Werner Frey and Hubert Truckenbrodt (22)
Max könnte etwas hilfsbereiter sein [da wir ihn doch höflich Max could somewhat more.helpful be since we him MP politely gefragt haben] asked have ‘Max could be somewhat more helpful, because we asked him politely after all.’
(23)
a.
Gestern ist sie den ganzen Tag zu Hause geblieben, während yesterday has she the whole day at home stayed while sie doch sonst bei schönem Wetter meistens einen she MP otherwise in nice weather mostly an macht. Ausflug excursion makes ‘Yesterday she stayed at home all day while she otherwise mostly makes an excursion in nice weather.’
b.
Er hat die Prüfung nicht bestanden, trotzdem er ja recht he has the exam not passed although he MP quite intelligent ist. intelligent is ‘He didn’t pass the exam even though he is intelligent.’
This separates PACs from central adverbial clauses, which do not license MPs (see Coniglio 2011 and Frey 2011, 2012 for details). The two apparently contradictory properties – integration in the German Vorfeld and a certain independence demonstrated by MPs – may be combined in the same example: PACs with modal particles freely occur in the Vorfeld: (24)
[Trotzdem/obwohl er ja recht intelligent ist] hat er die Prüfung although he MP quite intelligent is has he the exam nicht bestanden. not passed ‘Although he is quite intelligent, he didn’t pass the exam.’
We turn to the analysis. How might the root clause status of PACs relate to their high attachment? Hooper and Thompson (1973) suggested that root phenomena are limited to ‘assertive’ environments: They can occur unembedded as part of the speaker’s assertion, or they can occur embedded under a range of ‘assertive’ verbs in a somewhat wider sense. Today it is known that root phenomena occur in a broader class of cases that include questions and embedded questions under illocutionary verbs like ‘ask’ (cf. e.g. Coniglio 2011). Haegeman (2002, 2006) suggests that root clauses must be anchored to a speaker or a potential speaker. For example, German verb-second (V2) is a root phenomenon (e.g. Heycock 2006). Declarative V2-clauses occur unembedded as
5. Syntactic and prosodic integration and disintegration
|
87
speaker assertions as in (25a) or genuinely embedded under assertive predicates like (25b), though not in the non-assertive embedding in (25c). (25d) shows for comparison that a complementizer-initial clause can be embedded under the predicate that does not embed the V2-clause in (25c). (26) in addition shows that the embedded V2-clause is c-commanded by the matrix subject, which can bind a pronoun in the embedded V2-clause. (25)
a.
Maria liebt Fußball. Maria loves soccer
b.
Max meint, Maria liebt Fußball. Max thinks Maria loves soccer
c. *Max wies zurück, Maria liebt Fußball. Max rejected Maria loves soccer d. (26)
Max wies zurück, dass Maria Fußball liebt. Max rejected that Maria soccer loves
Jederi meint, eri ist eingeladen. everyonei thinks hei is invited
In (25b), then, the embedded V2 root clause occurs in the ‘assertive’ environment that is constituted by the description of Max’s opinion. In Haegeman’s terms, Max is the potential speaker that the root clause is linked to. It is plain that this licensing relation obeys a locality restriction. We formulate this in (27). (27)
A root clause must be locally embedded in an illocutionary context.
For example, in (25c), the local embedding of the embedded root clause is the nonassertive zurückweisen ‘reject’. Non-locally, there is also the speaker assertion of the entire clause in (25c). However, this embedding does not license the embedded root phenomenon, because it does not locally embed the embedded root clause. Since (27) is important for the analysis, we also establish it with the MP ja in (28). It demands that the clause it occurs in is a root clause. This root clause is bracketed in the examples in (28). The bracketed root clause is immediately embedded in the speaker’s assertion in (28a). It is immediately embedded under Max’s assertion in (28b). In (28c) it is not immediately embedded under an assertive predicate (but under a non-assertive one). The MP is also not licensed by the speaker’s assertion in (28c), since this does not locally embed the bracketed clause. A similar point is made in (28d). Here the quantifier jeder ‘every’ enforces a restrictive relative clause. The MP ja cannot be added to such a restrictive relative clause. The reason is that the relative clause is not locally embedded under an assertive embedding. It is not licensed by
88 | Werner Frey and Hubert Truckenbrodt being part of the speaker’s assertion, since the speaker’s assertion embeds the entire utterance, but does not locally embed the bracketed root clause. (28)
a.
[Maria liebt (ja) Fußball] Maria loves MP soccer ‘Maria loves soccer.’
b.
Max hat zu Peter gesagt [dass Maria (ja) Fußball liebt] Max has to Peter said that Maria MP soccer loves ‘Max said to Peter that Maria loves soccer.’
c.
Max bestreitet [dass Maria (*ja) Fußball liebt] Max denies that Maria MP soccer loves ‘Max denies that Maria loves soccer.’
d.
Jeder hier [der (*ja) Fußball liebt] soll die Hand heben. Everyone here who MP soccer loves should the hand raise ‘Everyone here who loves soccer should raise his/her hand.’
Following up on the analysis of Frey (2011), which we minimally modify, we are now in a position to analyze the requirement that PACs must be attached at a very high level to their host clause: Since PACs are inherently root clauses, they must be locally embedded in an illocutionary environment. The illocution of which they participate is normally the speaker’s assertion of the host clause. PACs use this assertion in a parasitic way. To access this assertion, PACs need to also be locally embedded under that assertion in the sense of (27). We make this idea more precise using (29) for illustration. It shows the projection of the host clause CP and the two possible positions for the PAC (likewise a CP, though named PAC in (29)): It can stand as the specifier of CP or right-adjoined to CP. In addition, we indicate the embedding of the host clause CP in the speaker’s Assertion, and highlight this by underlining. (This embedding may also be semantic or pragmatic in nature, but we indicate it in (29) as syntactic.) We assume that the Assertion is induced by CP due to its V2-property (see e.g. Gärtner 2002; Truckenbrodt 2006) and that both CP and the PAC are locally embedded under it in the sense of (27). We maintain that the locality requirement in (27) is the reason why the PAC requires such high attachment. (29) Assertion
CP PAC
CP PAC
C’ C
(rest of clause)
5. Syntactic and prosodic integration and disintegration
|
89
For concreteness, we may enforce the specifics of the high attachment of the PAC and allow its parasitic use of the higher speech act by postulating (30), a more specific version of (27). For the purpose of this paper, we apply the locality requirement in (30) to the underlying position of the root clause. (30)
A root clause must be locally embedded in an illocutionary context. This locality requirement tolerates the intervening presence of a CP node. However, no other syntactic constituents may intervene between the root clause and its illocutionary embedding.
This formulation rules out that the embedded root clauses in (25c), (28c,d) are licensed by the speaker’s assertion (above the main clause). In these cases there would be more syntactic nodes than a single CP intervening between the embedded root clause and the assertive embedding. For example, the object clause in (28c) is underlyingly within VP. It is therefore separated from the speaker’s assertion of the entire utterance by additional nodes of the matrix clause. At the same time, the formulation in (30) allows the immediate embedding of the unembedded root clause in (28a) under the speaker’s assertion and the immediate embedding of the embedded root clause in (28b) under Max’ assertion, with no intervening nodes at all: Assertion [CP Maria liebt ja Fußball] in the case of (28a), and [CP dass Maria ja Fußball liebt] gesagt in the case of (28b). For the crucial PACs, then, (30) leaves a loophole that allows the PAC to parasitically access the speaker’s assertion in (29). Here only a CP intervenes between the PAC and the speaker’s assertion, and so the PAC is locally embedded in the speaker’s assertion by the formulation in (30). At the same time, (30) prevents that the PAC is more deeply embedded inside of the host clause. While the analysis has so far concentrated on the unembedded case, it carries over to examples like (21a), in which the host clause is itself embedded. In (21a), Max meint . . . ‘Max thinks . . . ’ is the assertive embedding that is shared by the host clause (which is the dass-clause) and the PAC. The PAC in (21a) is then likewise a root clause that is in accord with (30), since it is separated from Max meint . . . ‘Max thinks . . . ’ only by the root node of the dass-clause. (21b) is ruled out because the PAC, a root clause, does not have an assertive embedding that it can share in this fashion. This is captured by (30) insofar the PAC is not locally embedded in an assertive context. The local embedding that it could share, Max wies zurück . . . ‘Max rejected’ is not assertive, and the assertive speech act of the entire utterance is not local to the PAC. In summary, it seems that PACs require high attachment because they are inherently root clauses and participate in the assertive embedding of their host clause. We may think of them as speech act parasites. The participation in the assertive embedding of their host clause tolerates only a minimal distance from the assertive embedding, allowing PACs to either stand in the specifier of their host clause or adjoin to their host clause.
90 | Werner Frey and Hubert Truckenbrodt 5.2.3 Prosody of PACs We begin by addressing the prosody of PACs that are right-adjoined to the host clause. In a wide focus context, central adverbial clauses like (31) can carry the sentence stress of the entire utterance, while peripheral adverbial clauses like (32) require separate sentence stress on the host clause (Brandt 1990; Frey 2011). (31)
(32)
What did Maria tell you? a.
[Peter wird kommen [sobald er etwas Zeit hat]]
b.
[Peter wird kommen [sobald er etwas Zeit hat]] (central) Peter will come as.soon.as he some time has ‘Peter will come as soon as he has some time.’
(central)
What did Maria tell you? a. # [Peter wird kommen] [obwohl er keine Zeit hat] b.
[Peter wird kommen] [obwohl er keine Zeit hat] Peter will come although he no time has ‘Peter will come although he has no time.’
(peripheral) (peripheral)
This difference can be accounted for in terms of the notion of root sentences in (3) and Stress-RS in (11). The central adverbial clause in (31a) is part of the host clause. It is dominated by the CP node of the host clause, which is a predicate clause. By (3) the central adverbial clause is therefore not a root sentence. The PAC in (32a) is adjoined to the CP root clause. This host clause has a predicate, i.e. it is a predicate clause. Is the PAC then dominated by its CP node? It is inside one CP segment of the host clause but not inside of all CP segments of that host clause (see (29)). Syntactically, the PAC is partly inside, partly outside of that CP. In the following, we will see effects of both, which we subsume under the following more general formulation: (33)
A PAC that is adjoined to its host clause CP will normally count as outside of CP for the syntax-prosody mapping, but can marginally also count as inside of CP.
In (32) we observe the preference for the PAC to count as being outside of the host clause CP. As long as the PAC counts as being outside the host clause CP, it is itself a root sentence by (3) and requires sentence stress by Stress-RS. This is what we see in (32). Let us then look at some cases in which it appears that the PAC can count as a root sentence together with its host clause. First, when either the host clause or the PAC is contextually given and therefore G-marked, the constraint *Stress-given in (12) can in principle remove sentence stress from them, as in (34) and (35).
5. Syntactic and prosodic integration and disintegration
|
91
(34)
Peter wird kommen. [Er wird kommen]G obwohl er keine Zeit hat. he will come time has Peter will come although he no ‘Peter will come. He will come although he has no time.’
(35)
Peter hat keine Zeit. Aber er wird kommen, obwohl [er keine Zeit Peter has no time but he will come although he no time hat]G has ‘Peter has no time. But he will come altough he has no time.’
This is a piece of motivation for the marked option in (33): Where the PAC (marginally) counts as inside of its host clause, host clause and PAC form a root sentence in the sense of (3) together. In this case Stress-RS requires sentence stress only once in the entire utterance, and so the stress may shift away from a given part to another part of the utterance. Another piece of motivation comes from some complementizers introducing PACs that allow being accented. When such a complementizer is present and when both host clause and content of the PAC are contextually given, though not the complementizer, the given elements are deaccented and the complementizer is accented: (36)
Peter meinte, dass es dunkel war und dass der Mond am Himmel stand. ‘Peter thought that it was dark and that the moon was in the sky.’ Ja, es war dunkel trotzdem/obwohl/gleichwohl der Mond am Yes it was dark although the moon in.the Himmel stand. sky stood ‘Yes, it was dark although the moon was in the sky.’
This, too, is analzyed in terms of the marginal option in (33) that allows the PAC to form a single root sentence together with its host clause. The single root sentence requires only a single instance of sentence stress, which is satisfied in (36). The placement of narrow focus points in the same direction. While it is not straightforward to form a focus-background structure across the PAC and its host, we have constructed the following examples in which it seems to be possible. The main difficulty with finding such examples seems to be to construct a question that asks for the PAC, and indeed the questions in (37) are themselves marked. However, if one accepts the questions, the answers seem to be relatively acceptable. (37)
(?) Welches Geschehens ungeachtet hat Max Maria geküsst? which event disregarded has Max Maria kissed ‘Regardless of which event has Max kissed Maria?’
92 | Werner Frey and Hubert Truckenbrodt Max hat Maria geküsst gleichwohl [das Licht anging]F Max has Maria kissed although the light on.went ‘Max has kissed Maria although the light came on.’ (38)
(?) Welchem Faktor zum Trotz ist die Nacht dunkel? which factor to.the spite is the night dark ‘Despite which factor is the night dark?’ Die Nacht ist dunkel trotzdem/gleichwohl [der Mond am Himmel the night is dark although the moon in.the sky steht]F stands ‘The night is dark although the moon is in the sky.’
We see in this another sign of the relative integration of the PACs: Where they constitute, marginally, a single root sentence together with the host clause, a focusbackground structure can be built across host clause and PAC without violating (6). Consider then also the prediction of this account for the prosody of the PAC when the PAC is in the Vorfeld, SpecCP. Since the Vorfeld is part of the host clause, we expect integrated prosodic behavior. Where we test this with givenness, it is borne out: (39) is acceptable. The PAC counts towards the host clause insofar its sentence stress satisfies Stress-RS for the host clause (otherwise we would see additional sentence stress outside of the PAC, cf. (13) and (32)). (39)
Peter wird kommen. [Obwohl er keine Zeit hat]PAC [wird er Peter will come although he no time has will he kommen]G . come ‘Peter will come. Although he has no time he will come.’
Similarly, a focus-background structure across PAC and host clause is possible, confirming that the initial PAC is in the same structure with the remainder of the utterance. (40)
(?) Welchem Faktor zum Trotz ist es dunkel? ‘Despite which factor is it dark?’ [Trotzdem/obwohl [der Mond am Himmel steht]F ]PAC ist es dunkel although the moon in.the sky stands is it dark ‘It is dark although the moon is in the sky.’
More generally, we expect integrated behavior of the PACs in the Vorfeld. This is compatible with the data we are aware of insofar we have not found any prosodic distinctions between PACs in the Vorfeld and other adverbial clauses in the Vorfeld. It is difficult to construct pairs like (31) and (32) for PACs in the Vorfeld (where we expect that PACs behave like integrated adverbial clauses) because German clauses generally tend
5. Syntactic and prosodic integration and disintegration
|
93
to be followed by an intonation phrase boundary (Truckenbrodt 2005 for German, see also Downing 2011 for Bantu languages), though unlike Stress-RS for root sentences, this can be overridden by contextual givenness. In (31) this seems not, or not obligatorily, to interfere, apparently because the central adverbial clause is attached low enough to be part of the main clause CP, so that the intonation phrase boundary at the right edge of the main clause CP follows the embedded clause. However, a CP in the Vorfeld in an all-new sentence seems to be regularly followed by an intonation phrase boundary and to carry sentence stress. This is plausibly related to the structure insofar a CP in the Vorfeld always has its right edge, where the intonation phrase boundary is inserted, at the end of the Vorfeld. In summary, while the facts are complex, the prosody and the information structure of PACs reflect their borderline status between integration and disintegration. PACs can act prosodically as though not part of the host clause. However, in an option that seems to also be available, they act as though they are part of the host clause.
5.2.4 PACs: Summary It seems that PACs are inherently root clauses and that they are parasitic on the speech act of their host clause. They connect to their host clause structurally but must stand in a very high position because of their connection to the speech act of the host clause. Their prosody and their information structure reflect this very high position insofar they can act either as disintegrated or, at least marginally, as integrated.
5.3 Right dislocation and afterthought 5.3.1 Syntax of right dislocation and afterthought Right dislocation and afterthought involve the resumption of a pronoun or other element in the sentence by an element that follows the clause, as in (41). In right dislocation the resumed element is a personal pronoun and the resuming element is stressless as in (41a). Where the element on the right is stressed as in (41b), it is referred to as afterthought (Ziv und Grosz 1994; Averintseva-Klisch 2006, 2009). (41)
a.
Ich habe siei gesehen, die Mariai . I have her seen the Maria ‘I have seen Maria.’
b.
Ich habe siei gesehen, – (ich meine) die Mariai . (afterthought) I have her seen I mean the Maria ‘I have seen her, I mean Maria.’
(right dislocation)
94 | Werner Frey and Hubert Truckenbrodt Averintseva-Klisch (2009) pursues an analysis of right dislocation in terms of rightadjunction as in (42a). Ott and de Vries (2012) and Truckenbrodt (to appear) pursue an analysis of right dislocation and afterthought in which deletion applies to a biclausal structure as shown in (42b,c). Sluicing as in (42b) (fronting followed by deletion of a constituent) is postulated by Ott and de Vries (2012), building on the sluicing analysis of Merchant (2004) and on a sluicing analysis for Japanese right dislocation in Tanaka (2001). A gapping analysis as in (42c), following essentially Kuno (1978), is explored in Truckenbrodt (to appear).⁵ (42)
a.
Ich habe [[sie gesehen], die Maria].
(right-adjunction)
b.
Ich habe sie gesehen, die Maria habe ich gesehen. I have her seen the Maria have I seen
(sluicing)
c.
Ich habe sie gesehen, ich habe die Maria gesehen. I have the Maria seen I have her seen
(gapping)
The right-dislocated element obeys c-command restrictions as though it was in the position of the pronoun. Three such connectedness effects from Ott and de Vries (2012) are reviewed in (43)–(45). In each case it appears that the initial constituent ccommands the right-dislocated element, as though the right-dislocated element was in the position of the pronoun it resumes. (43) shows this with binding of a pronoun by a quantifier, (44) with reflexive binding according to Binding Condition A, and (45a) with a Binding Condition C effect (Chomsky 1981; Büring 2005). (45b) is a comparison case. (43)
[Did you wave to your students?] Jederi von uns hat ihnen gewunken, seineni Schülern. each of us has them waved, his students ‘Each of us has waved to them, to his students.’
(44)
Jani hat jemanden im Spiegel gesehen: sichi selbst. Jan has someone in.the mirror seen himself self ‘Jan saw someone in the mirror: himself.’
(45)
a. *Siei hat ihn mit einer Anderen gesehen, Mariasi Freund. she has him with a.fem other seen Maria’s boyfriend ‘She saw him with another woman, Maria’s boyfriend.’ b.
Siei hat ihn mit einer Anderen gesehen, ihreni Freund. ‘She saw him with another woman, her boyfriend.’
5 We take gapping to be circumscribed and restricted as in Neijt (1979).
5. Syntactic and prosodic integration and disintegration
|
95
These effects can be represented in an adjunction analysis and in the deletion analyses. For the adjunction theory, this is shown in (46). If the height of right-adjunction mirrors the syntactic height of the resumed pronoun, the right-adjoined element will share the c-command relations to other elements with the resumed pronoun. In (46c) for example, the initial sie c-commands both ihn and the resuming element Maria’s Freund. The Condition C effect will correctly follow in such a structure. (46)
a.
Jederi von uns hat [[ihnen gewunken], seineni Schülern]. Each of us [[waved to them] to his students] b. Jani hat [[jemanden im Spiegel gesehen]: sichi selbst]. Jan [[saw someone in the mirror] himself] c. *Siei hat [[ihn mit einer Anderen gesehen], Mariasi Freund]. *She [[saw him with another woman] Maria’s boyfriend]
(47) shows how a deletion theory in terms of gapping can derive these effects. In all cases, the relevant c-command relations obtain in the elliptical clause between an initial (elided) constituent and the deletion remnant. For example, the deleted clauseinitial pronoun siei in (47c) binds and c-commands the name Maria in the second clause, in violation of Binding Condition C. (47)
a.
Jederi von uns hat ihnen gewunken, jederi von uns hat [seineni Schülern] gewunken. Each of us has waved to them, each of us has waved to his students. b. Jani hat jemanden im Spiegel gesehen: Jani hat sichi selbst im Spiegel gesehen. Jan has seen someone in the mirror, Jan has seen himself in the mirror. c. *Siei hat ihn mit einer Anderen gesehen, siei hat Mariasi Freund mit einer anderen gesehen. *She has seen him with another woman, she has seen Maria’s boyfriend with another woman.
In the sluicing analysis of Ott and de Vries (2012), the effects follow in the same way, though via (independently motivated) reconstruction of the fronting movement that precedes deletion in the elliptical clause. An important distinction between the adjunction analysis and the deletion analyses is that the adjunction analysis treats the dislocated element as part of the host clause (adjoined inside of it), while the deletion analysis treats the dislocated element as genuinely outside of the host clause (derived via deletion from a clause that follows the host clause). Broadly speaking, the adjunction analysis therefore predicts integrated behavior of the dislocated element, while the deletion analysis predicts it to act in a disintegrated way. We believe that the main evidence for disintegrated behavior of the dislocated element comes from its prosody and information structure, which we turn to in the following section. The syntactic and semantic arguments by Ott and
96 | Werner Frey and Hubert Truckenbrodt de Vries (2012) for disintegration of the dislocated element are: (i) The dislocated element does not affect the syntactic well-formedness of the host clause, which is always complete also without the dislocated element. (ii) The dislocated element does not affect the truth-value of the clause preceding it. (iii) The dislocated element is a syntactic island for movement. While these observations are all very much compatible with the deletion analysis, they are not strong arguments against an adjunction analysis. For example, the CED of Huang (1982) predicts that adjuncts are islands for movement. Therefore the adjunction analysis also predicts that the dislocated elements are islands for syntactic movement. We also want to point out that the disintegrated view is compatible with Zifonun et al. (1997: 1647), who view the dislocated element as a kind of doubled version of an element in the clause and in that sense as not part of the preceding clause. In summary, a constituent that is right dislocated or an afterthought is added to the clause in the sense that it does not occupy a thematic position in the preceding clause. It shows c-command relations like the element it resumes (connectedness effects). These can be captured in an adjunction analysis as well as in deletion analyses. Ott and de Vries (2012) argue that disintegration of the dislocated constituent (together with the connectedness effects) supports a deletion analysis. This will be strengthened in the following section on the prosody of right dislocation and afterthought.
5.3.2 Prosody of right dislocation and afterthought The prosody of right dislocation and afterthought is here discussed in comparison with that of extraposed elements. The comparison follows Truckenbrodt (to appear) and extends compatible observations in Downing (1970) for English and Altmann (1981) for German. Extraposition, classically conceived of as movement to the right (e.g. Guéron 1980; see Büring and Hartmann 1997 for a defense of movement accounts), is not thought of as deriving a disintegrated structure. The moving element is inside of the clause before movement, and it is inside of the clause after movement. The syntactic integration is reflected in integrated behavior in the prosody. In (48) and (49) extraposed constituents are shown in wide-focus contexts, where the regular rules of sentence-stress assignment place the sentence stress on the extraposed constituent. The remainder of the clause to the left of the extraposed element does not carry sentence stress in these examples. (48)
[What happened?] Maria hat ein Buch ti gelesen [von Chomsky]i . Maria has a book read by Chomsky ‘Maria read a book by Chomsky.’
5. Syntactic and prosodic integration and disintegration
(49)
|
97
[What will the weather be like?] [Peter hat ti gesagt, [[dass es regnen wird]i ]I . Peter has said will that it rain ‘Peter has said that it will rain.’
On the other hand, the prosody of right dislocation and afterthought shows striking effects of disintegration. The examples (41) are again shown in (50), omitting elements irrelevant here. Right dislocation and afterthought share that there must be sentence stress on the clause that precedes the dislocated element. Crucially, it is not possible to omit that sentence stress, as in (51). Further examples that show this ban are given in (52). (50)
(51) (52)
a.
Ich habe siei gesehen, die Mariai .
b.
Ich habe siei gesehen – die Mariai . I have her seen the Maria
*Ich habe siei gesehen, die Mariai . I have her seen the Maria a. *Peter hat jemanden kennengelernt, die Maria. Peter has someone met the Maria ‘Peter has met someone, Maria.’ b. *Jeder hat sie gerne, seine Mutter. everyone has he dear his mother ‘Everyone likes her, his mother.’ c. *Jan hat jemand im Spiegel gesehen, sich selbst. Jan has someone in.the mirror seen himself self ‘Jan has seen someone in the mirror, himself.’
In this regard, right dislocation and afterthought are strikingly different from extraposition. In the perfectly natural examples (48) and (49), the part of the sentence that precedes the extraposed element lacks sentence stress. We capture this distinction in terms of Stress-RS. It is plain that extraposition does not place constituents outside of the root sentence from which they move. Therefore the assignment of sentence stress proceeds unobstructed by Stress-RS in the examples (48) and (49). On the other hand, the observation in (51) and (52) finds a natural account if right-dislocated constituents are preceded by a complete root sentence: This root sentence will then correctly require sentence stress by Stress-RS. The examples (51) and (52) are then ruled out as violations of Stress-RS because the initial root sentence lacks sentence stress. Here as in (13) Stress-RS is stronger than *Stress-Given. Thus in (53) the clause to the left of the dislocated constituent is contextually given. Nevertheless, it must carry sentence stress.
98 | Werner Frey and Hubert Truckenbrodt (53)
Claudias Mutter sagt, dass Peter sie gesehen hat. ‘Claudia’s mother says that Peter has seen her.’ a. *Ja, [Peter hat sie gesehen, die Claudia]. b. Ja, [Peter hat sie gesehen], die Claudia. c.
Ja, [Peter hat sie gesehen] – [die Claudia]. yes Peter has her seen the Claudia ______given_________ ‘Yes, Peter has seen Claudia.’
(right dislocation) (afterthought)
In this regard, the observation at hand is comparable to the one we saw for nonrestrictive relatives in (1) and (5). It seems that right-dislocated constituents and afterthoughts are disintegrated to the same degree as non-restrictive relatives. For the purpose of stress-assignment in the preceding clause, it is as though the dislocated constituents were not there. We turn to consequences for the syntactic analyses below.
5.3.3 Focus assignment with right dislocation and afterthoughts A strong effect of disintegration can also be observed with the assignment of narrow focus. Notice first for comparison that extraposed constituents can be assigned narrow focus, as in (54) and (55). (54)
Von wem hat Maria ein Buch gelesen? ‘By whom did Maria read a book?’ Sie hat ein Buch t gelesen [F von Chomsky]. she has a book read by Chomsky ‘She has read a book by Chomsky.’
(55)
Was hat Peter gesagt? ‘What did Peter say?’ Er hat t gesagt [F dass es regnen wird]. he has said that it rain will ‘He has said that it will rain.’
This is what we expect: Clause-internal constituents can normally be assigned narrow focus, no matter what their position in the clause. On the other hand, right dislocated constituents and afterthoughts cannot be assigned narrow focus: (56)
Wen hat Marias Vater gesehen? ‘Who has Maria’s father seen?’ *Er hat siei gesehen, [F die Maria]i . he has her seen the Maria ‘He has seen her, Maria.’
5. Syntactic and prosodic integration and disintegration
(57)
|
99
Wen hat Peter kennengelernt? ‘Who did Peter meet?’ *Er hat jemanden kennengelernt, [F die Claudia]. he has someone met the Claudia ‘He has met someone, Claudia.’
One may correctly object that (56) and (57) are independently ruled out by Stress-RS, given our assumption that the dislocated constituents are outside of the domain of sentence stress. However, we can try to remedy this situation and assign sentence stress in the default position of the final verb, as in (58) and (59). Now the examples satisfy Stress-RS. However, they are still not felicitous in their focus-triggering contexts. (58)
Wen hat Marias Vater gesehen? ‘Who has Maria’s father seen? *Er hat siei gesehen, [F die Maria]i . he has her seen the Maria ‘He has seen her, Maria.’
(59)
Wen hat Peter kennengelernt? ‘Who did Peter meet?’ *Er hat jemanden kennengelernt, [F die Claudia]. he has someone met the Claudia ‘He has met someone, Claudia.’
Thus focus assignment provides further evidence for the disintegrated status of rightdislocated elements. More specifically, if the right-dislocated element or afterthought is preceded by a root sentence, it correctly follows from (6) that this root sentence cannot be integrated into a focus-background structure with the dislocated element that follows it. In sum, the preceding two subsections have provided striking evidence for the disintegration of right-dislocated elements and afterthoughts: The requirement on preceding sentence stress and the impossibility of an overarching focus-background structure follow if right dislocated elements and afterthoughts follow a complete root sentence that does not include the dislocated element.
5.3.4 Syntactic consequences of the prosodic observations Assume, then, that a right-dislocated element or afterthought were right-adjoined to the preceding clause as in (60).
100 | Werner Frey and Hubert Truckenbrodt (60)
Peter hat [TP [TP siei gesehen], die Mariai ]. Peter has her seen the Maria ‘Peter has seen Maria.’
In such a structure, there would be no reason why the dislocated element could not carry the only sentence stress of the utterance: As part of the unembedded root clause, it would be eligible for sentence stress like other constituents, given appropriate motivation for choosing this site such as givenness of the remainder of the clause as in (53). Furthermore, there would be no obstacle to focusing the right-dislocated constituent, since it would be syntactically fully integrated with the remainder of the clause. The only way, then, in which the adjunction analysis can be maintained, is with construction-specific assumptions: By maintaining that afterthought has a different structure entirely, and that the structure in (60) is inherently tied to a stressless dislocated constituent. This stipulation will correctly keep sentence stress (and thus also focus) from occurring on the dislocated element, and thus force them to remain in the domain that precedes the dislocation. Yet afterthought shows the same connectivity effects as right dislocation: (61)
a.
Jederi von uns hat einem gewunken – einem seineri Schüler. one.of his students each of us has someone waved ‘Eachi of us waved to someone, to one of hisi students.’
b.
Jani hat jemanden im Spiegel gesehen – sichi selbst. Jan has someone in.the mirror seen himself self ‘Jan has seen someone in the mirror, himself.’
c. *Siei hat jemanden gesehen – Mariasi Freund. she has someone seen Maria’s boyfriend ‘She has seen someone, Maria’s boyfriend.’ d.
Siei hat jemanden gesehen – ihreni Freund. boyfriend she has someone seen her ‘She has seen someone, her boyfriend.’
If an advocate of (60) would therefore extend the structure (60) to at least some cases of afterthought, then there can be no ban on stressing the right-adjoined constituent in (60), and our argument against (60) applies with full force: There is no reason why sentence stress is required before the adjoined constituent in (60), nor why focus could not be assigned to the adjoined constituent. The deletion analysis, on the other hand, can derive the effects of disintegration we observed. It starts from a biclausal structure like (62). By (3) both clauses are root sentences, since both are not contained in a higher clause with a predicate of its own. Since the first clause is not affected by deletion, it remains a root sentence. When Stress-RS applies to it, sentence stress is assigned to it. As we saw, Stress-RS can-
5. Syntactic and prosodic integration and disintegration
|
101
not be overridden by information structure. Furthermore, it correctly follows that no focus-background structure can be built that combines the first clause and remnants of the second clause, since the first clause is a root sentence so that (5) prevents a focus-background structure that is larger than it. (62)
Er hat sie gesehen, er hat die Maria gesehen. he has her seen, he has the Maria seen ‘He has seen her, he has seen Maria.’
We contend that the facts from prosody and information structure are not compatible with an analysis of right dislocation and afterthought in terms of syntactic adjunction. On the other hand, the deletion analysis reconciles the disintegrated status of the dislocated element with the connectivity effects it displays, as argued by Ott and de Vries (2012).
5.3.5 On the distinction between right dislocation and afterthought Ziv and Grosz (1994) argue for English and Averintseva-Klisch (2009) argues for German that right dislocation and afterthought have different properties. While we agree that there are non-trivial distinctions, we treat them here as a natural class insofar both involve a preceding root sentence that requires sentence stress and both are derived by syntactic deletion. In an account of this kind, further distinctions between them relate to different properties of the second clause in (62) before or after deletion. As a first approach to these distinctions, Truckenbrodt (to appear) suggests that right dislocation involves deletion of the second root sentence CP node along with the overtly deleted material, as in (63), while afterthought involves the retention of this second root sentence CP node as in (64). With assignment of prosody following deletion, (63) will then allow a stressless right-peripheral constituent, while the second root sentence CP in (64) still requires sentence stress on the ellipsis remnant. (63)
CP
Er hat sie gesehen he has her seen (64)
CP
Er hat jemanden gesehen he has someone seen
CP
er hat die Maria gesehen he has the Maria seen CP
er hat die Maria gesehen he has the Maria seen
Other syntactic distinctions may follow from the different resulting root sentence structure, though much remains to be explored and spelled out in detail here. As far
102 | Werner Frey and Hubert Truckenbrodt as the prosody, (9) and (10) derive the intonation phrases and sentence stress in (65) and (66) from (63) and (64). (65)
[Er hat sie gesehen]I , die Maria. he has her seen the Maria
(66)
[Er hat jemanden gesehen]I , [die Maria]I . he has someone seen the Maria
What is unusual in (65) is that the final element does not belong to any intonation phrase. However such violations of exhaustive prosodic parsing are also known from other prosodic levels, see e.g. Selkirk (1995). What the structure correctly captures is that the host clause to the left of the dislocated element is a domain of sentence stress, i.e. an intonation phrase. This is also true of (66), where, in addition, the dislocated element is an additional intonation phrase carrying sentence stress. The structures furthermore give us an approach to the position of pauses in German. Empirically and impressionistically, there is no pause preceding right dislocation as in (65), though there is a pause preceding afterthought as in (66).⁶ This can be represented by the distinction between (65) and (66) if a left edge of an intonation phrase leads to a preceding pause, and if right edges of intonation phrases do not trigger pauses: (67)
Insert a pause preceding a left edge of an intonation phrase.
(67) will insert a pause preceding die Maria in (66) but not in (65).
5.3.6 Summary of right dislocation and afterthought The prosody and the information structure of right dislocation and afterthought show that these elements are disintegrated. A syntactic adjunction analysis, if it wants to generalize across right dislocation and afterthought, cannot represent this disintegration in a principled way. The deletion analysis, on the other hand, captures both this disintegration and the connectedness effects.
6 Schneider-Wiejowski (2011) recorded a corpus of spontaneous speech in which she measured pause frequency and pause duration (next to F0 change) for three categories: preceding extraposed nonsentential elements (‘enges Nachfeld’), preceding extraposed sentential elements (‘weites Nachfeld’) and a category that includes the boundary preceding right dislocation and afterthought (‘rechtes Außenfeld’), but was mostly represented by expressions with adverbial function in her corpus. The three-way distinction was also investigated in a perception study. The classification and its terminology are from Zifonun et al. (1997). Because of the different research questions, the results do not directly speak to the issues discussed in the text. However, the distinction Schneider-Wiejowski (2011) finds, in which the elements in the third category are generally more strongly separated than the extraposed elements, seems to us to be broadly compatible with our results.
5. Syntactic and prosodic integration and disintegration
|
103
5.4 Summary PACs cannot be c-commanded from elements in the host clause. They are attached high because they are root clauses that are parasitic on the assertive embedding of their host clause. They are nevertheless attached to the host clause in ways in which non-restrictive relatives are not. This is reflected in their syntactic properties, in their prosodic options and in their options with regard to information structure. In the current analysis, this is captured in the distinction between an orphan structure for nonrestrictive relatives as in (68) and a CP-adjunction structure for PACs which follow their hosts as in (69). (68)
CP1
CP2
Die Sekretärin kontaktierte Peter The secretary contacted Peter (69)
der im Urlaub war who was on vacation.
CP1 CP1 Die Sekretärin kontaktierte Peter The secretary contacted Peter
CP2 obwohl er im Urlaub war although he was on vacation.
The CP-adjunction structure places PACs at the border between being a separate root sentences and not being a separate root sentence. Non-restrictive relatives, on the other hand, are always separate root sentences. Dislocated constituents in the right periphery do not have this borderline status. They are always outside of the host clause. The strongest arguments in favor of this conclusion seem to be their prosody and their information structure. All elements at the right periphery that resume some part of the clause are obligatorily preceded by a host clause that carries sentence stress. No elements at the right periphery that resumes some part of the clause can be focused in a focus-background structure that extends across host clause and right-peripheral element. This correctly follows from a deletion analysis of such right-peripheral elements in which the elided ‘conjunct’ is a separate structure that is not part of the host clause or adjoined to it: (70)
CP1
Die Sekretärin kontaktierte Peter The secretary contacted Peter
CP2
die Sekretärin kontaktierte den neuen Mitarbeiter The secretary contacted the new employee.
104 | Werner Frey and Hubert Truckenbrodt These distinctions were worked out in an analysis that connects observations and suggestions about integration of Reis (1997) and Holler (2008) to the notion of root sentence from Downing (1970). Building on Brandt (1990), Reis (1997) and Holler (2008), a restriction was formulated that a focus-background structure cannot extend beyond a root sentence. From Downing, the restriction that root sentences are delimited by intonation phrase boundaries was adopted, with the crucial consequence at hand that root sentences must carry sentence stress. Acknowledgement: This work was supported by the Bundesministerium für Bildung und Forschung (BMBF) (Grant Nr. 01UG1411).
References Altmann, Hans. 1981. Formen der “Herausstellung” im Deutschen – Rechtsversetzung, Linkversetzung, Freies Thema und verwandte Konstruktionen. Tübingen: Niemeyer. Averintseva-Klisch, Maria. 2006. Separate performative account of the German right dislocation. In: Proceedings of the Sinn und Bedeutung 10, Christian Ebert and Cornelia Endriss (eds.), 15–28. ZASPiL 44. Averintseva-Klisch, Maria. 2009. Rechte Satzperipherie im Diskurs. NP-Rechtsversetzung im Deutschen. Tübingen: Stauffenburg. Bayer, Josef. 2001. Asymmetry in emphatic topicalization. In Audiatur vox sapientiae. A Festschrift for Arnim von Stechow, Caroline Féry and Wolfgang Sternefeld (eds.), 15–47. Berlin: Akademie Verlag. Brandt, Margareta. 1990. Weiterführende Nebensätze. Zu ihrer Syntax, Semantik und Pragmatik. Stockholm: Almqvist und Wiksell (Lunder germanistische Forschungen 57). Büring, Daniel. 2005. Binding theory. Cambridge: Cambridge University Press. Büring, Daniel and Katharina Hartmann. 1997. Doing the right thing. The Linguistic Review 14: 1–42. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Coniglio, Marco. 2011. Die Syntax der deutschen Modalpartikeln: Ihre Distribution und Lizenzierung in Haupt- und Nebensätzen. Berlin: Akademie-Verlag. Dehé, Nicole. 2009. The relation between syntactic and prosodic parenthesis. In Parentheticals, Nicole Dehé and Yordanka Kavalova (eds.), 261-284. Amsterdam: Benjamins. Downing, Bruce Theodore. 1970. Syntactic structure and phonological phrasing in English. Ph. D. diss., The University of Texas. Downing, Laura. 2011. The prosody of ‘dislocation’ in selected Bantu langauges. Lingua 121: 772–786. Dowty, David. 1979. Word meaning and Montague Grammar. Dordrecht: Reidel. Féry, Caroline and Vieri Samek-Lodovici. 2006. Focus projection and prosodic prominence in nested foci. Language 82: 131–150. Frey, Werner. 2003. Syntactic conditions on adjunct classes. In Modifying adjuncts, Ewald Lang, Claudia Maienborn, and Cathrine Fabricius-Hansen (eds.), 163–209. Berlin: de Gruyter. Frey, Werner. 2011. Peripheral adverbial clauses, their licensing and the prefield in German. In Satzverknüpfung – zur Interaktion von Form, Bedeutung und Diskursfunktion, Eva Breindl, Gisella Ferraresi, and Anna Volodina (eds.), 41–77. Berlin: de Gruyter.
5. Syntactic and prosodic integration and disintegration
|
105
Frey, Werner. 2012. On two types of adverbial clauses allowing root-phenomena. In Main Clause phenomena: New horizons, Lobke Aelbrecht, Liliane Haegeman, and Rachel Nye (eds.), 405– 429. Amsterdam: Benjamins. Gärtner, Hans-Martin. 2002. On the force of V2 declaratives. Theoretical Linguistics 28: 33–42. Grewendorf, Günther. 2002. Minimalistische Syntax. Tübingen, Basel: A. Francke Verlag. Guéron, Jacqueline. 1980. On the syntax and semantics of PP Extraposition. Linguistic Inquiry 11: 638–678. Haegeman, Liliane. 2002. Anchoring to the speaker, adverbial clauses and the structure of CP. Georgetown University Working Papers in Theoretical Linguistics 2: 117–180. Haegeman, Liliane. 2004. The syntax of adverbial clauses and its consequences for topicalisation. Antwerp Papers in Linguistics. 107. Current Studies in Comparative Romance Linguistics: 61–90. Haegeman, Liliane. 2006. Conditionals, factives and the left periphery. Lingua 116: 1651–1669. Haider, Hubert. 2010. The syntax of German. Cambride: Cambridge University Press. Heycock, Caroline. 2006. Embedded Root Phenomena. In The Blackwell companion to syntax, Vol. II, Martin Everaert and Henk van Riemsdijk (eds.), 174-209. Oxford: Blackwell. Höhle, Tilman N. 1992. Über Verum-Fokus im Deutschen. In Informationsstruktur und Grammatik, Joachim Jacobs (ed.), 112-141. Opladen: Westdeutscher Verlag. Holler, Anke. 2008. German dependent clauses from a constraint-based perspective. In “Subordination” vs. “coordination” in sentence and text, Cathrine Fabricius-Hansen and Wiebke Ramm (eds.), 187-216. Amsterdam: Benjamins. Hooper, Joan B. and Sandra A. Thompson. 1973. On the applicability of root transformations. Linguistic Inquiry 4: 465–497. Huang, C.T. James. 1982. Logial relations in Chinese and the theory of grammar. Ph. D. diss., MIT. Kan, Seda. 2009. Prosodic domains and the syntax-prosody mapping in Turkish. M. A. thesis, Boğaziçi University, Istanbul. König, Ekkehard and Johan van der Auwera. 1988. Clause integration in German and Dutch conditionals, concessive conditionals and concessives. In Clause combining in grammar and discourse, John Haiman and Sandra A. Thompson (eds.), 101–133. Amsterdam: Benjamins. Kuno, Susumo. 1978. Danwa-no bunpoo [Grammar of discourse]. Tokyo: Taishuukan-shoten. Ladd, D. Robert. 1983. Even, focus, and normal stress. Journal of Semantics 2: 257–270. Merchant, Jason. 2004. Fragments and ellipsis. Linguistics and Philosophy 27: 661–738. Neijt, Anneke. 1979. Gapping: a contribution to sentence grammar. Dordrecht: Foris. Nespor, Marina and Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Ott, Dennis and Mark de Vries. 2012. Right-dislocation as deletion. Ms., University of Groningen. Reis, Marga. 1997. Zum syntaktischen Status unselbständiger Verbzweit-Sätze. In Sprache im Fokus. Festschrift für Heinz Vater zum 65. Geburtstag, Christa Dürscheid and Karl-Heinz Ramers (eds.), 121–144. Tübingen: Niemeyer. Rooth, Mats. 1992. A theory of focus interpretation. Natural Language Semantics 1: 75–116. Safir, Kenneth. 1986. Relative clauses in a theory of binding and levels. Linguistic Inquiry 17: 663–83. Schneider-Wiejowski, Karina. 2011. Prosodische Anbindung nach der rechten Satzklammer. In Satverknüpfung mehrdimensional. Zur Interaktion von Form, Bedeutung und Diskursfunktion in Geschichte und Gegenwart, Eva Breindl, Gisella Ferraresi, and Anna Voladina (eds.), 327–350. Berlin: de Gruyter. Selkirk, Elisabeth. 1995. The prosodic structure of function words. In Papers in Optimality Theory. University of Massachusetts Occasional Papers 18, Jill Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), 439–469. Amherst, MA: GLSA.
106 | Werner Frey and Hubert Truckenbrodt Selkirk, Elisabeth. 2005. Comments on intonational phrasing in English. In Prosodies. With special reference to Iberian languages, Sónia Frota, Marina Vigário, and Maria João Freitas (eds.), 11–58. Berlin: de Gruyter. Selkirk, Elisabeth. 2011. The syntax-phonology interface. In The handbook of phonological theory, 2nd edition, John Goldsmith, Jason Riggle, and Alan Yu (eds.), 435–484. Oxford: Blackwell. Tanaka, Hidekazu 2001. Right-dislocation as scrambling. Journal of Linguistics 37: 551–579. Thurmair, Maria. 1989. Modalpartikeln und ihre Kombinationen. Tübingen: Niemeyer. Truckenbrodt, Hubert. 2002. Upstep and embedded register levels. Phonology 19: 77–120. Truckenbrodt, Hubert. 2005. A short report on intonation phrase boundaries in German. Linguistische Berichte 203: 273–296. Truckenbrodt, Hubert. 2006. On the semantic motivation of syntactic verb movement to C in German. Theoretical Linguistics 32: 257–306. Truckenbrodt, Hubert. to appear. Some distinctions in the German Nachfeld. In Inner-sentential propositional pro-forms: syntactic properties and interpretative effects, Werner Frey, André Meinunger, and Kerstin Schwabe (eds.), Amsterdam: Benjamins. Uhmann, Susanne. 1991. Fokusphonologie. Eine Analyse deutscher Intonationskonturen im Rahmen der nicht-linearen Phonologie. Tübingen: Niemeyer. Zifonun, Gisela, Ludger Hoffmann, and Bruno Strecker. 1997. Grammatik der deutschen Sprache, Band 1. Berlin: de Gruyter. Ziv, Yael and Barbara Grosz. 1994. Right dislocation and attentional state. In The Israel Association for Theoretical Linguistics. Proceeding of the 9th Annual Conference and workshop on Discourse, Rhonna Buchalla and Anita Mittwoch (eds.), 184–199. Jerusalem: Akademon.
Marlies Kluck
6 On representing anchored parentheses in syntax 6.1 Introduction Parentheses constitute a quirky class of constructions that are somehow interpolated in a regular sentence (the host or matrix clause). This ‘motley crew’ (Dehé and Kavalova 2007) ranges from unrelated utterances that are freely interjected in a sentence, to appositives, which clearly express information about a particular matrix constituent. We can more generally distinguish between ‘free’ (1) and ‘anchored’ (2) parenthesis: (1)
Free parenthesis a. b.
c.
(2)
Newton’s Principia – take a seat – was finally published in 1687. [interjection] Einstein’s theory of special relativity – I think – was presented in his 1905 paper. [comment clause] The professor made out with – and we all knew that – lots of students at the party. [and-parenthetical]
Anchored parenthesis a. b. c. d.
Bea kissed Bob, who she has known since high school, at the party. [nominal appositive relative clause/NARC] Bea kissed Bob, her high school sweetheart, at the party. [nominal apposition/NA] Bea kissed someone, I think it was Bob, at the party [sluiced parenthetical] Bea kissed [I think it was Bob] at the party. [amalgam]
Intuitively, there is a clear parallel between these interpolations: they all add some non-restrictive, secondary content to what is expressed by their hosts. However, the parentheses in (1) are not related to anything specific in their hosts, whereas the ones in (2) all add information about a particular (understood) constituent in the host. That is, the parentheses in both (2a), (2b) and (2c) provide extra, more specific information about Bob and someone. The construction in (2d) is an example of so-called sentence amalgamation (Lakoff 1974), and closely resembles the parenthetical in (2c). In this case however, the interpolation consists of a reduced it-cleft, of which the cleft pivot is associated with an null matrix argument, namely the understood object of kissed. This paper focuses primarily on anchored parentheses and their representation.
108 | Marlies Kluck A central hallmark of parentheticals is that they are structurally independent of their hosts. This can be illustrated, for example, by the absence of a condition C effect in (3), an example of free parenthesis: (3)
Hei has kissed, Casanova that Bobi is, many women.
Notice that the truth-conditions of the host are not affected by the parenthetical: (3) is true regardless of whether Bob is a Casanova or not. Parentheses are thus structurally and semantically independent. In addition, various scholars have suggested an interpretive asymmetry between a parenthetical and its host: the information expressed by parentheses is necessarily speaker-oriented. The notion ‘speaker-orientation’ was originally contrasted with ‘subject-orientation’ (Jackendoff 1972; Reinhart 1983), and can roughly be defined as revealing the speaker’s attitude (for instance epistemic stance) with respect to the matrix proposition. Speaker-oriented content thus implies a crucially different status of what is expressed in the matrix (an assertion) and what is expressed in a parenthetical. Potts (2005) formulates this as conventional implicatures, and distinguishes between at-issue and non-at-issue (conventional implicature, CI) content. I will represent the former by ‘→’ and the latter by ‘←’: (4)
Bob has kissed, Casanova that he is, many women. → Bob has kissed many women. ← Bob is a Casanova.
In other words, (4) asserts that Bob has kissed many women, and expresses additionally that Bob is a Casanova. For the anchored parentheses (2), we can make a similar distinction between assertive content and parenthetical content, but the latter is only expressed about a particular constituent in the host. This paper sets out to characterize this relation more precisely. Taking structural and interpretive independence to be the most important properties of parentheses, I focus on the question at which level anchored parentheses are represented in grammar, and how. I argue on both conceptual and empirical grounds that anchored parentheticals must be represented at the level of syntax. For this, I elaborate on the representation of parallel construal in a ‘colon phrase’ proposed in Koster (1999, 2000, 2012) and the idea that the inclusion of parenthetical objects is a primitive in grammar in De Vries (2007, 2012). The paper is organised as follows. In order to establish which kinds of parentheses should be considered ‘anchored’, I present a global overview in §2. This is followed by a discussion of the core properties of anchored parentheses, focusing on the relation between anchor and parenthesis, and their ‘specificational’ interpretation. §3 discusses anchored parenthesis in the domain of parallel construal, and shows why the Kosterian colon phrase is not sufficient to account for them. In §4, I elaborate on the structural independence of parentheticals in the context of De Vries’ syntactic approach, suggesting a parenthetical version of the colon phrase to account for an-
6. On representing anchored parentheses in syntax | 109
chored parenthesis. §5 is a brief discussion of two implications (or: complications) of this analysis. §6 concludes the paper.
6.2 Parentheses: Kinds and properties 6.2.1 A rough typology of parentheses It is clear from the range of examples above that more needs to be said about the distinguishing properties of anchored versus ‘free’ parenthesis. Within the latter type, we can make an (at least) tripartite distinction based on (1). That is, whereas (1a) does not relate to the content of the matrix at all (but rather to what happens in the domain of discourse), (1b) and (1c) do. In addition, the difference between the latter seems to be that (1b) modifies the speech act as a whole, whereas (1c) adds information to the propositional content of the matrix. More specifically, the parenthetical in this case contains an anaphoric pronoun (here that, in boldfaced italics), which refers to the propositional content of the entire matrix (italicized here): (5)
The professor – and we all knew that – made out with lots of students at the party.
Interestingly, such a semantic connection between host and parenthetical is not restricted to the level of the host CP, as can be witnessed in (6), where the parenthetical, containing the anaphoric pronoun him, adds information about the professor: (6)
The professori – and I’ve heard more of these stories about himi – made out with lots of students at the party.
This raises the question whether we should regard examples such as these as instances of ‘anchoring’. After all, I have formulated ‘adding information about a particular host constituent’ as a distinguishing property of anchored parentheses in the introduction of this paper. As it turns out, this is not a sufficient description of anchoring. It has been observed in the literature that NAs may not be separated from their anchors by movement, and that they are opaque for structural operations. This is illustrated in (7): (7)
a. Bill, Bea’s husband, I kissed in an alley. b. *Bill I kissed , Bea’s husband, in an alley.
110 | Marlies Kluck Similarly, the head of a NARC cannot move without moving the relative along (8).¹ Given the status of relative clauses as islands for movement (Ross 1967), it is not surprising that elements cannot move out of ARCs either (9): (8)
a. Bill, who used to be Bea’s husband, I kissed yesterday . b. *Bill, I kissed , who used to be Bea’s husband, yesterday.
(9)
*Bea’s husband, I kissed Bill, who used to be .
Potts (2005) captures this under a more general ‘adjacency’ requirement, and illustrates this further by examples of right-dislocation of apposition and NARC: (10)
a. *We spoke with Lance before the race, the famous cyclist, about the weather. b. *We spoke with Lance before the race, who is a famous cyclist, about the weather.
It thus seems that appositives form a constituent: their parts cannot be separated by some movement operation. Turning back to (6), we can immediately see that such a requirement does not hold. After all, the parenthetical in such cases does not need to follow the constituent that it seems to relate to; it can occur freely in the host clause, as (11) illustrates: (11)
a. b.
The professori made out with – and I’ve heard more of these stories about himi – lots of students at the party. The professori made out with lots of students – and I’ve heard more of these stories about himi – at the party.
In other words: there is a distinction between parentheses that are semantically related to a constituent (or the matrix as a whole as in (5)) via some anaphoric relation, and true, structural ‘anchoring’, which entails constituency, and consequently adjacency of anchor and parenthesis.
1 A careful reader might object that NARCs can be extraposed, i.e. I kissed Bill yesterday, who used to be Bea’s husband. There is an ongoing debate considering the status of extraposition as movement. After all, the conditions on extraposition differ significantly from those on leftward movement. For a recent overview of analyses and discussion, I refer to Webelhuth, Sailer and Walker (2013). In line with Koster (2000), De Vries (2009) and Kluck and De Vries (2013), I assume that extraposition does not involve movement. Instead, it involves base-generation and deletion. Although the issue of extraposition in anchored parenthesis is beyond the scope of the present work, it is expected that the proposal pursued here is suitable to capture these as well.
6. On representing anchored parentheses in syntax | 111
6.2.2 Anchoring in parentheses: Appositives, amalgams and sluiced parentheses In what follows, I elaborate further on the three types of anchored parentheses introduced above, focusing in particular on the less well-known ‘amalgams’. This is introduced by a discussion of the relation between anchor and parenthetical material, which can generally be described as ‘specificational’, and underlies the syntactic representation of anchored parentheses laid out in §3 and §4. Let me first establish an important empirical boundary. The discussion below is limited to anchored parentheses with nominal anchors. The literature about appositions (2b) distinguishes nominal appositions (NAs) from appositions of other categories and ‘unbalanced’ appositions (for discussion and overview, see Heringa 2012). For the present purposes, we will only be concerned with balanced NAs, i.e. appositions of which both anchor and apposition are nominal. Similarly, the generalizations about appositive relative clauses do not extend to ARCs with non-nominal anchors, i.e. so-called which-appositives (cf. Sells 1985; Del Gobbo 2003, 2007; Potts 2002; LaCara 2012). Consequently, I leave open the question to what extent the representation for anchored parenthesis pursued here applies to these cases. Starting out with NAs and NARCs, we can observe that the examples sofar involve definite, referential anchors. Appositives are, however, by no means restricted to definite anchors: (12)
a. b.
Bob kissed an older woman, his former neighbor. Bob kissed an older woman, who used to be his neighbor.
For NARCs, it holds that indefinite anchors can only be understood as a specific indefinite: examples (12) can only mean that there is a particular older woman (who the speaker has in mind) to whom what is expressed in the appositive applies. It is generally agreed in the literature that specific indefinite DPs are used when the speaker has a particular person in mind, but that this does not imply that the DP is in fact a definite in disguise (Reinhart 1997; Von Heusinger 2002; Geurts 2011; a.o.). For the present purposes, I assume that anchors of NAs and NARCs denote individuals (type e). This is further corroborated by the observation that neither NAs nor NARCs allow for anchors with generalized quantifiers, as is illustrated in (13):² (13)
a. *Bea kissed no/every professor, her hero. b. *Bea kissed no/every professor, who she admired.
2 It should be noted here that NAs with unspecific indefinite anchors are not strictly excluded, i.e. Bob would like to once own an elephant, such a huge inhabitant of Africa. However, these seem to be quite exceptional (i.e. notably hard to construe). For a complete overview and discussion, see Heringa and De Vries (2008) and Heringa (2012, chapter 2).
112 | Marlies Kluck For more discussion on the ban on quantificational anchors in NARCs and NAs, see Del Gobbo (2003, 2007), Constant (2012) and Nouwen (2014). Let us then turn to a less familiar type of parenthesis, namely ‘sluiced parentheticals’ (SPs). These were mentioned in the introduction (cf. (2c)), and concern examples such as (14): (14)
a. b.
Bea kissed someone, you’ll never guess who, at the party. Bea kissed someone, I think it was Bob, at the party.
The parentheticals here are clearly clausal, but appear to be incomplete: they are subject to ellipsis. More specifically, (14a) involves a common wh-sluice, and (14b) a reduced it-cleft. In both cases, the anchor (someone) is the correlate of the pivot of the sluice (who and Bob respectively). I assume that they involve deletion of syntactically present material (see also Merchant 2001; Van Craenenbroeck 2010). That is, the clause (IP or CP) in which the pivot is base-generated and out of which it undergoes some type of A -movement, is deleted at PF: (15)
a. b.
Bea kissed someone, you’ll never guess whoi Bea kissed ti at the party, at the party. Bea kissed someone, I think it was Bob that Bea kissed ti at the party, at the party.
SPs are clearly related to so-called ‘amalgams’ (Lakoff 1974; Guimarães 2004; Kluck 2011, 2013). Lakoff distinguishes between two variants: (16a) represents an ‘Andrewsamalgam’ (resembling the SP with a wh-sluice), and (16b) a ‘Horn-amalgam’ (involving a reduced it-cleft, see also (2d) in the introduction). (16)
a. b.
Bea kissed [you’ll never guess whoi Bea kissed ti at the party] at the party. [Andrews-amalgam] Bea kissed [I think it was Bobi Bea kissed ti at the party] at the party. [Horn-amalgam]
The parenthetical part of amalgams (here indicated by brackets) is exactly the same as their counterparts in (14), but amalgams lack an overt anchor.³ However, as is argued at length in Kluck (2011, chapter 5), the evidence that amalgams are instances of sluicing and are therefore related to SPs, is abundant. For instance, amalgams display connectivity effects, pattern with regular sluicing constructions with respect to
3 Unlike NAs, NARCs and SPs, there are no prosodic boundaries that mark the parenthetical part in amalgams. That is, the ‘interrupting clause’ seems to be prosodically integrated in the host clause. Why this is so, lies outside the scope of the present work. It should be pointed out that ‘comma-intonation’ should by no means be taken as a general hallmark of parentheses, as is shown in Dehé (2007) and Döring (2007).
6. On representing anchored parentheses in syntax | 113
P-stranding effects, and the A -movement of the pivot is island insensitive (see also Merchant 2001). It thus seems that sluicing in amalgams involves a null correlate, or, that anchoring in parentheses of this sort is to some phonologically empty (but semantically understood) constituent in the host. For convenience, I adopt the notation of Chung et al. (1995) for implicit arguments, where x stands for an individual variable (an empty element of type e).⁴ (17)
Bea kissed ex [you’ll never guess who/I think it was Bob] at the party.
So, anchored parenthesis (2) can be subdivided into two classes: those with overt anchors (2a)–(2c), and those with null anchors (2d). This then explains why the grammaticality of the host in any case does not depend on the presence of the parenthetical, but it does in case of amalgams if the anchor is an argument in the host: (18)
a. b.
Bea kissed someone (her boss/you’ll never guess who/I think it was Bob) at the party. Bea kissed ex *(you’ll never guess who/I think it was Bob) at the party.
This means that amalgams, too, must be seen as anchored parentheses, albeit with empty anchors. In what follows, I will focus on SPs, and assume that the null anchors of amalgams behave similarly. It was observed above that the anchors of NAs and NARCs are subject to some restrictions: both allow for definite anchors, and if the anchor is indefinite, it is a specific indefinite. Neither allows for anchors with generalized quantifiers. Based on (19), we can observe a further restriction on anchors in SPs, as both universally quantified and definite anchors are disallowed: (19)
a. *Bea kissed the professor, you’ll never guess who/I think it was Bob, at the party. b. *Bea kissed every/no professor, you’ll never guess who/I think it was Bob, at the party.
This is straightforwardly related to a general restriction on correlates that has been noticed in sluicing (Chung et al. 1995; Reinhart 1997; Romero 1998, 2000, 2003): correlates of sluicing must be indefinite. Furthermore, this indefinite must be specific (or: take wide scope). This has been related to a requirement of scope parallellism, i.e.
4 This is merely to avoid confusion between denotations of semantic types (i.e. type e, ⟨e, t⟩ et cetera) and empty syntactic elements (‘e’). It should however be stressed that Chung et al. (1995) and Romero (2000) employ the term ‘implicit argument’. Contrary to the null XP employed here for amalgams, this is strictly to account for instances of ‘sprouting’. Although amalgams may involve sprouting, they are not generally instances of this special variety of sluices. For a detailed explanation, see Kluck (2011, chapter 7) and Kluck (2013).
114 | Marlies Kluck the indefinite correlate must have wide scope because the wh-phrase in the sluice has (Chung et al. 1995; Romero 2000). This is illustrated in examples such as (20) (for more examples of the indefinite outscoping of other scope-bearing elements in sluicing, see the literature cited above): (20)
a. b.
Everyone wanted to hit someone at that party. ∃ > ∀, ∀ > ∃ Everyone wanted to hit someone at that party, but it was unclear who. ∃ > ∀, *∀ > ∃
The same holds for SPs: (21)
Everyone wanted to hit someone, I think it was Bob/you’ll never guess who, at that party. ∃ > ∀, *∀ > ∃
Abstracting away from the exact mechanics of scope parallellism, the relevant observation here is that NAs, NARCs and SPs share the restriction that indefinite anchors must be specific. A final remark should be made about the categories of anchors in SPs and amalgams. I explicitly restricted the discussion of appositives to those with nominal anchors. For the present purposes, I will do the same for SPs and amalgams. This is not to say that non-nominal anchors do not occur in these constructions. For instance, SPs and Andrews-amalgams can have AP pivots, which in the present analysis implies that they have (null) anchors that denote properties (‘P’):⁵ (22)
a. b.
Bea got herself an expensive, you’ll never guess how expensive car. Bea got herself a eP [you’ll never guess how expensive] car.
Merchant (2001:§5.1) discusses similar regular sluices in the context of the Left Branch Condition (LBC), and assumes an extended adjectival projection in which AP is the complement of a degree-head Deg (cf. Abney 1987; Corver 1990, 1997). This explains the presence of the predicate in both the parenthetical and the anchor in (22a). After all, movement of how (= Deg∘ ) would be an illicit type of head movement, which gives rise to the ungrammaticality of (23a). Thus, pied-piping of the AP is obligatory, in whquestions (23b) as well as in wh-sluices (22)/(24): (23)
a. *How is this car expensive? b. How expensive is this car?
5 However, similar variants of Horn-amalgams do not exist. This is most likely related to the traditional observation that it-clefts cannot have predicative (i.e. type ⟨e,t⟩) pivots (cf. Emonds 1976; Hedberg 1990). For examples and discussion, see Kluck (2011: 130ff).
6. On representing anchored parentheses in syntax | 115
(24)
Bea got herself an expensive car, but I don’t know how *(expensive).
Although the proposal for anchored parentheses captures these cases as well (see Kluck 2011, §7.3 in particular), the rest of the discussion will be restricted to those with nominal anchors. Before we discuss these observations in relation to the kind of information that the parenthetical adds to its anchor, Table 6.1 provides an overview of the restrictions on anchored parentheses with nominal anchors: Table 6.1. Anchors in parentheses. nominal anchor quantificational unspecific indefinite specific indefinite definite
apposition
NARC
SP/amalgam
– + + +
– – + +
– – + –
6.2.3 Attribution, identification and speaker-orientation in anchored parentheses It is generally assumed in the literature that parentheses express additional, nonrestrictive information to their host sentences. More specifically, the truth conditions of the host are independent of the parenthetical (Potts 2005). The question is then, what kind of information do anchored parentheses add to their anchors? As will be shown in this section, we should distinguish ‘attributive’ from ‘identificational’ anchored parentheses, and not all types correlate with both types of meaning. Let us again start out with NAs. Perhaps the most salient way to connect the meanings of nominals to each other in natural language, is via a copular construction, i.e. DP1 is DP2 . In the vast literature about copular clauses, a basic distinction is made between the following types:⁶ (25)
a. b.
[ DP1 Ingrid Bergman] is [ DP2 the lead actress in that movie]. [predicational] [ DP1 The lead actress in that movie] is [ DP2 Ingrid Bergman]. [specificational]
The difference is that predicational clauses (25a) say something about the referent of the subject of the clause, whereas specificational clauses (25b) tell us who or what the
6 For reasons of space, this discussion does not include ‘equative’ copular clauses such as She is Ingrid Bergman. Instead, these are taken together with specificational clauses. This is in line with Heringa (2012), who takes only this basic distinction to be relevant in his analysis of appositions. For discussion, see Heringa (2012: 87ff) and the references cited therein.
116 | Marlies Kluck referent of the subject is (Mikkelsen 2005: 1). This insight goes back to Higgins (1979) and Akmajian (1979), who argue that in case of the latter, the subject introduces a variable for which the post-copular expression (here DP2 ) provides a value. I assume something similar for SPs and amalgams, which are all identificational. Heringa (2012) argues convincingly that appositions correlate with copular clauses, such that a distinction can be made between ‘attributive’ (predicational) and ‘identificational’ (specificational) appositions:⁷ (26)
a. b.
[ DP Ingrid Bergman], [ NA the lead actress in that movie], was born in 1915. [attributive] [ DP The lead actress in that movie], [ NA Ingrid Bergman], was born in 1915. [identificational]
Notice that often both readings are available, as most (if not all) DPs allow for both predicational (attributive) and referential (identificational) readings: (27)
Bob, Bea’s former husband, loves Stravinsky.
Interestingly, only attributive appositions correlate with appositive relative clauses, as is observed in O’Connor (2008) and Cardoso and De Vries (2010). That is, NARCs do not have the identificational reading:⁸ (28)
a.
[ DP Ingrid Bergman], [ NARC who is the lead actress in that movie], was born in 1915. [attributive] b. *[ DP The lead actress in that movie], who is Ingrid Bergman, was born in 1915. [identificational]
Heringa (2012) assumes a predicational copular structure (a small clause with a subject of type e and a predicate of type ⟨e, t⟩) to underlie attributive NAs and an ‘equative’ copular structure (a small clause with two expressions of type e) for identificational ones. Contrary to regular copular clauses, where DP1 (subject) and DP2 (predicate)
7 I refrain from discussing so-called ‘inclusive’ appositions here, where the apposition describes only a part of the anchor, such as in exemplification: A monarchy, such as Holland, usually spends a lot of tax money on the royal family. For discussion of those in a semantic classification of appositions, I refer to Heringa and De Vries (2008) and Heringa (2012). 8 A reviewer of this paper brings up an example from English that seems to counter this claim: I switch on my iPod to my favorite artist, who happens to be Amon Tobin. For reasons of space, this note is limited to some preliminary observations. First, this example deteriorates considerably when we use a regular copula, i.e. is instead of happens to be. Second, there is no way to construe a similar example in Dutch. Although the present work is limited to English examples, the facts described here extend to Dutch as well. Clearly, the English case needs more careful examination.
6. On representing anchored parentheses in syntax | 117
have the same information-structural status (25), the anchor and NA/NARC express different kinds of information. That is, the parenthetical material in (26)–(28) is speaker-oriented (‘←’), which is an important hallmark of parenthesis (cf. Jackendoff 1972; Reinhart 1983; Potts 2002, 2005). We can thus discern the asserted from the speaker-oriented content in these constructions: (29)
Ingrid Bergman, (who is) the lead actress in that movie, was born in 1915. → Ingrid Bergman was born in 1915. ← Ingrid Bergman is the lead actress in that movie.
The question is now, if what is expressed by SPs and amalgams can be described in a similar fashion. In the remainder of this section, I argue 1. that amalgams and SPs are always identificational (even if their anchor is an attributive AP), and 2. that we need to adopt a somewhat liberal view of what identification means in the case of Andrews-amalgams and their SP variants. First, consider (30): (30)
a. b.
Bob hit someone, I think it was the Dean, in the face. Bob hit ex [I think it was the Dean] in his face.
These cases clearly involve identificational meaning. This seems directly related to the it-cleft, as it-clefts are traditionally associated with specificational meaning (cf. Akmajian 1970, 1979; Higgins 1979; Hedberg 2000; Reeve 2012 and many others). Concretely, the parenthetical expresses who, in the eyes of the speaker, the x such that Bob hit x is, namely the Dean. Let us assume for simplicity’s sake that the relevant variable is here provided by the indefinite anchor. ⁹ The pivot of the it-cleft in the parenthetical thus identifies the indefinite anchor in the matrix in a similar speaker-oriented fashion as we saw above: (31)
Bob hit someone/ex [I think it was the Dean] in the face. → Bob hit someone in the face. ← The person that Bob hit in the face was the Dean.
Andrews-amalgams and SPs of that type present an interesting case in relation to identificational meaning, as it is hard to see how a wh-element can really identify the indefinite anchor:
9 A careful reader may notice that Horn-cases appear to have a double layer of specificational (or identificational) meaning. After all, this relation does not only hold between anchor and parenthesis, but also between the pivot and it. For the present purposes, I will simply assume that it and the cleft-pivot are part of a small clause, and that it is coreferential with the anchor. For specificational/identificational meaning in (it-)clefts, I refer to Heycock and Kroch (1999), Heycock (2012), Den Dikken (2006, 2009), Reeve (2012), inter alia.
118 | Marlies Kluck (32)
a. b.
Bob hit someone, you’ll never guess who, in the face. Bob hit [you’ll never guess who] in the face.
Still, these examples add information about the variable that is introduced by the indefinite anchor. Informally put, it reflects an opinion of the speaker that the identity of the x such that Bob hit x in the face lies outside some contextual standard or expectation. Kluck (2011: §6.3) refers to this meaning aspect of Andrews-amalgams as diverge, and relates it to the complex predicates that embed the sluices (i.e. complexes like will never guess and won’t/can believe). This is informally paraphrased in (33):¹⁰ (33)
Bob married someone/ex [you’ll never guess who] last year. → Bob married someone last year. ← The person x such that Bob married x is not conform to our expectations given what we know about Bob.
So, although the parenthetical in Andrews-amalgams and Andrews-SPs does not pick out a value for the variable x introduced by the anchor like it does in the Horn-cases above (30)–(31), it does reveal some speaker-oriented information about it: the identity of the (null) anchor is such that it diverges from a contextual standard. Although the present paper is restricted to nominal anchored parentheses, the last bit of this section is devoted to the claim that amalgams and SPs cannot be attributive, not even when the anchor is an adjective in attributive position. The relevant examples resemble the following regular sluice: (34)
Bea got herself an expensive car. You’ll never guess how expensive.
Here, the sluiced wh-phrase how expensive questions the degree to which the car that Bea got is expensive. Recall that I have assumed an extended adjectival projection (in line with Kennedy and Merchant 2000; Merchant 2001, who adopt Abney 1987; Corver 1990 and Svenonius 1992): expensive is the complement AP of a degree head Deg. The assumptions about the extended adjectival projection imply that the anchors in (35) and (36) are DegPs, the latter involving an empty AP (here represented as eP because it denotes a property variable): (35)
Bea got herself an [ DegP [ AP expensive]], you’ll never guess how expensive, car.
(36)
Bea got herself a [ DegP [ AP eP ]] you’ll never guess how expensive car.
10 More specifically, Kluck (2011) argues that the meaning of Andrews-amalgams should not be analysed on a par with wh-exclamatives (Lakoff 1974), but as embedded interrogatives instead. The ‘exclamative’ feel that they have arises from the embedding predicates, which has been observed by other scholars in regular contexts as well (cf. d’Avis 2002 and Sæbø 2010).
6. On representing anchored parentheses in syntax | 119
Turning back to the distinction between attributive and identificational anchored parentheses above, we can now make an interesting observation: the parentheses here can only be seen as identificational. The adjectives in (35) and (36) denote properties, but these properties are not attributes of the anchor. This cannot be, because the anchor denotes a property itself. Instead, they specify (the degree of) the property that is expressed explicitly (35) or implicitly (36) by their anchors. That is, the parentheses in (35)–(36) provide speaker-oriented information about the degree to which the car that Bea got is expensive, namely that this degree diverges from a contextually given standard (similar to what we observed above). In §2.2 it was shown that anchored parentheticals specify the meaning of their anchors. Consequently, anchors are restricted to non-quantificational XPs (see Table 6.1). We can now further distinguish between two kinds of specifying information in anchored parentheticals: they express either attributive or identificational information about their anchors, or both (as is the case in appositions). This is summarized in Table 6.2: Table 6.2. Types of relation between anchor and parenthesis. type attributive identificational
apposition
NARC
SP/amalgam
+ +
+ –
– +
The general picture is clear: there is an asymmetrical relation between the anchor and the parenthetical XP. That is, the parenthetical XP adds more specific (i.e. attributive or identifying), speaker-oriented information about the anchor. The following sections discuss a general proposal of how this relation can be established in syntax.
6.3 Anchored parentheticals in the domain of parallel construal 6.3.1 Specification, (asyndetic) coordination and Koster’s colon phrase Koster (1999, 2000, 2012) discusses a range of constructions that can be analysed in terms of ‘parallel construal’, which includes regular coordination. For the latter, it is generally assumed (since Kayne 1994) that coordinators head their own projections, taking the second (or: final) conjunct as their complement and the first (and other non-final) conjunct(s) as their specifier(s), as in (37): (37)
[ CoP [ DP Bob] [ Co [ Co and] [ DP Bea]]]
Koster argues that phrase structures come in two forms: primary structure and parallel construal. In the first, lexical elements are directly licensed by a functional head.
120 | Marlies Kluck In parallel structure, elements are not licensed in this manner, but rather linked to other elements of primary phrase structure. The idea of ‘parallel structure’ for coordination goes back to Williams (1978). Although regular coordination is perhaps the most canonical example, parallel construal is argued to include extraposition,¹¹ (appositive) relative clauses (to be discussed in §6.4.2), and so-called ‘equatives’ (first described in Ross 1969). For expository reasons, I illustrate the general idea of the colon phrase for equatives, because they prima facie resemble NAs. Consider (38): (38)
John saw something beautiful: a golden igloo.
Coordination in this case is asyndetic, and the second conjunct (the ‘extension’) specifies the first (the ‘anchor’). Parallel structures always consist of an anchor-extension pair. So far, this is very similar to what we witnessed above for NAs in particular, and anchored parenthesis in general. Koster (2000: 21) proposes that parallel construal has a structure similar to coordination (37), with the more general Boolean operator (the colon or :) as its head: (39)
Colon Phrase [ :P XP [ XP [ :’ : XP]]
Something beautiful: a golden igloo in (38) is thus an :P (similar to a coordinated structure), which is used as the complement of saw: (40)
:P :
DP Something beautiful
:
DP
a golden igloo Koster (2012) extends this analysis to both restrictive and appositive relative clauses, appositions, right dislocations and more. The question is if the basic parallel configuration in (39) suffices as an analysis for NAs, NARCs, SPs and amalgams. The short answer to this is ‘no’. The reason for this was already alluded to in the introduction to this paper: parentheticals (in general) are structurally and semantically independent from their hosts. That is, their elements cannot interact with elements of their hosts in terms of c-command-based relations,
11 As mentioned in footnote 1 to this paper, discussion of extraposition in terms of specifying coordination/parallel construal lies outside the scope of the present paper. The reader is referred to Koster (2000, 2012), De Vries (2002, 2006, 2009) and Kluck and De Vries (2013) for extensive discussion.
6. On representing anchored parentheses in syntax | 121
and they express non-restrictive information. As is argued and shown below, this is not a general property of parallel construal: we therefore need to distinguish between restrictive and non-restrictive parallel construal.
6.3.2 Restrictive versus non-restrictive parallel construal The notion ‘colon’ and its presumed status as some kind of general coordinative operator thus require some more refinement. More specifically, we want to be able to derive the distinction between NARCs and regular RRCs. After all, without further assumptions, both would be instances of parallel construal for Koster (2000, 2012). Obviously, the crucial difference between NARCs and RRCs is how they modify the meaning of the relative head. Consider (41a) with NARCs (41b): (41)
a. b.
Bea kissed the professor [ RRC that could be her grandfather] at that party. Bea kissed the professor, [ NARC who could be her grandfather], at that party.
In (41a), the relative restricts the possible set of professors in the relevant domain of discourse to the one that could be Bea’s grandfather, while in (41b), the relative only adds a (speaker-oriented) property to the professor that Bea kissed. In addition, NARCs are opaque for structural relations like quantifier binding (42) and condition C (43), unlike RRCs: (42)
a. No studenti wants to work on the puzzle that hei brought up in class. b. *No studenti wants to work on the puzzle, which hei brought up in class.
(43)
a. *Hei wrote a paper that the professori ended up rejecting later on. b. Hei wrote a paper, which the professori ended up rejecting later on.
This is consistent with the well-known generalization that NARCs are opaque for structural relations, but complicates the unified Kosterian approach.¹² Turning to NAs, these appear to pattern with NARCs. For instance, Heringa (2012: 137) notes that referential appositions fare well in contexts where they are preceded by coreferential material:
12 It seems that the extension in a regular equative (38) must be accessible for the matrix that contains its anchor as well, as they are sensitive to condition C effects: *Hei /the professori cited someone interesting: the professori . Similarly, a matrix element may bind a pronominal equative: hei cited someone interesting: himselfi . Interestingly, Ott and De Vries (2012, 2013) discuss a range of reconstruction effects in right dislocations in general. This includes equatives (which they more appropriately term ‘specifying afterthoughts’). They analyse these as instances of parallel construal (not parenthesis), with internal topicalization and sluicing in the extension, which adds an interesting empirical domain to the data discussed here.
122 | Marlies Kluck (44)
Johni first met Mary, now Johni ’s wife, in the linguistic cafe.
In addition, quantifier binding cannot be established between matrix and apposition. The example is from Potts (2005: 82): (45)
*No reporteri believes that Ames, often the subject of hisi columns, is a spy.
However, the anchor of an NA can bind a reflexive in the apposition (46): (46)
Bobi , merely a shadow of himselfi , presented his work after a sleepless night.
As I show briefly below, the clausal analysis of Heringa (2012) assumes a silent E-type pronoun subject in NAs, which can establish this binding relation, while the otherwise opaque behavior of the apposition is preserved. Furthermore, it has been observed by various scholars that both NARCs and NAs are outside the scope of matrix negation. This is illustrated in (47), which entails that Bach is a famous composer despite of matrix negation (for similar and further discussion of appositives and scope, see Nouwen (2014) in particular): (47)
It is not the case that Bach, (who is) a famous composer, died in 1850.
This further corroborates the idea that NAs and NARCs are opaque. It thus seems that parallel construal comes in two different shapes: restrictive parallel construal, which includes RRCs, and non-restrictive parallel construal, including anchored parentheticals. Although Koster (2012: 14) acknowledges that parallel construal includes both non-restrictive and restrictive constructions, it is only speculated how the structural differences would follow, namely by some ‘restrictiveness’ operator in C in case of relative clauses. In the section below, I adopt a more radical solution, namely by assuming a distinction between regular (restrictive) inclusion and parenthetical (non-restrictive) inclusion, based on De Vries (2007, 2012).
6.4 A syntax for anchored parentheses 6.4.1 Parenthetical inclusion as a primitive in syntax: par-Merge The structural independence (or: opacity) witnessed above is widely accepted as one of the central hallmarks of parentheses in general: parentheticals do not interact with elements of their hosts in terms of c-command-based relations. This has been observed for NAs and NARCs in the earlier literature (Emonds 1979; Safir 1986) as well as in more recent scholarship (Potts 2002; Heringa 2012; see also (41)–(45) above), and for less familiar ‘parenthetical’ constructions in Espinal (1991), Haegeman (1991), Ackema and Neeleman (2004), De Vries (2007) and others. (48)–(49) illustrate this for NPI licensing
6. On representing anchored parentheses in syntax | 123
(and lack of PPI-effect) and quantifier binding in free parenthesis (I return to SPs and amalgams in §6.4.3): (48)
Bob never wanted – like someone/*anyone assumed – to kiss the professor.
(49)
*Everyonei would love – hei told me so – to kiss the president.
The facts have led to opposing views in the literature. Basically, there are two types of theory: those that put parentheticals outside the domain of syntax, i.e. ‘orphan’ approaches found most prominently in Safir (1986), Haegeman (1991) and BurtonRoberts (1999), and those that invoke some kind of special way of integrating parentheticals in syntax, such as Espinal (1991), Ackema and Neeleman (2004) and De Vries (2007, 2012). As will be clear below, I adhere to the latter type of approach for anchored parenthesis. The reason for this is twofold: 1. since parentheticals in general are somehow linearized in their hosts, and interpreted relative to their hosts, as we saw in §6.2.1 of this paper, they must be part of syntax under the common assumptions about the model of grammar, and 2. since anchored parentheses in particular are related to a specific constituent in their host (the anchor), their integration cannot possibly take place at some post-syntactic level. Elaborating on the Kosterian colon phrase, I will adopt De Vries’ (2007, 2012) idea that parenthetical inclusion is a primitive in syntax, in addition to the familiar inclusion that gives rise to dominance. To get an idea of the distinction between dominance-inclusion (via regular Merge) and parenthetical-inclusion (via ‘par-Merge’), consider first (50): (50)
a.
Merge(A, B) → C Merge(C, D) → E
b.
E D
C A
B
Here, D c-commands its sister C and everything dominated by C (A and B). The presence of c-command enables dependencies between elements in a single representation, such as binding (depending on the type of pronoun, binder and domain), NPI licensing, et cetera. If parentheticals are generally opaque for these relations, this could mean that parenthetical inclusion should be set apart from regular (or: restrictive) inclusion. This is the rationale behind De Vries’ ‘par-Merge’ (De Vries 2007, 2012).¹³ Basically, if two elements are included in a derivation via par-Merge, the output of that
13 It should be noted that De Vries (2007) refers to this as ‘b-Merge’ or ‘behindance’ instead of parMerge. This is related to the idea that representations of parentheticals are multidimensional (De Vries 2003, 2008). For related discussion, see also Espinal (1991).
124 | Marlies Kluck merger does not dominate the input. So in (51), C does not dominate A and B. Consequently, D c-commands C, but not A and B: (51)
a.
par-Merge (A, B) → C Merge (C, D) → E E
b. D
C A
B
The traditional understanding of c-command can be maintained for this and can be formulated such that it explicitly refers to the type of inclusion, such as in (52): (52)
A c-commands B iff there is an X such that i. ii.
A and X are merge-mates, and B = X, or B is dominance-included in X.
The question then arises when par-Merge can be applied. Suppose we just apply parMerge to a given XP that is used as a free parenthetical in its host derivation. The moment XP is par-merged with the root (the host derivation), what was merged before becomes inaccessible to what is merged after the par-merger of XP. This is undesirable, obviously elements of the host itself must be able to interact with each other. This is why De Vries (2012) ascribes par-Merge to a specialized parenthetical head Par⊛ , which can be thought of as a ‘discourse connector’. The first merger of Par⊛ with XP then invokes the application of par-Merge (this is what the ‘⊛’ stands for in my notation). This has the welcome consequence that the status of a given XP as a parenthetical is due to a syntactic operation rather than an intrinsic property of this XP. Concretely, the possible uses of the AdvP honestly (53) can now simply be explained by the use of regular Merge that adjoins it to the VP in (53a), and the merger with Par⊛ triggering par-Merge in (53b) (the latter is possible in various positions in the host in addition to the VP, I illustrate just two positions here): (53)
a. b.
I am (honestly) interested (*honestly) in what you’re up to. I am (, honestly,) interested (, honestly,) in what you’re up to.
Free parentheses such as these are thus no more than regular adjuncts that are included via par-Merge rather than regular Merge. In the case of free parentheses, Par⊛ can be seen as monovalent, i.e. it takes a complement but not a specifier, and the ParP itself is a (regularly included) adjunct in the matrix sentence:
6. On representing anchored parentheses in syntax | 125
(54)
ZP Z∘
YP ParP
Par⊛
YP XP
parenthetical Notice that the dashed lines do not reflect the application of par-Merge, but represent the inaccessibility of the par-Merged material as a result of par-Merge.
6.4.2 Non-restrictive parallel construal: Bivalent Par⊛ We are now just one step away from an analysis for anchored parentheses. In these cases, the parentheticals are not free adjuncts in the sentence, but added at the constituent level. After all, they modify their anchor in the ways described in §6.2. I propose that we analyse these as a parallel construal in Koster’s sense, with a ‘bivalent Par⊛ ’. This is thus not headed by :, but by a parenthetical head: (55)
ParP XP anchor Par⊛
Par YP parenthetical
Like coordination phrases, ParP is categorially underspecified. Given that the bivalent ParP is no more than a non-restrictive version of the Kosterian colon phrase, I assume that it inherits its category in a way similar to coordination phrases. It has been argued on various empirical grounds that these inherit their categorial features from their specifier (Munn 1993; Johannessen 1998). Concretely, this means that an anchored parenthetical with a DP anchor functions as a DP in its host, that it behaves as an AP if its anchor is an AP, et cetera (this is particularly important for SPs and amalgams with non-nominal anchors, which lie outside the scope of this paper, cf. Kluck 2011). The position of an anchored parenthetical in its host therefore depends on the category of its anchor. The tree in (57) is then a simplified global representation for the NA/NARC in (56), as is proposed in Heringa (2012) and De Vries (2009, 2012).
126 | Marlies Kluck (56)
I kissed Bill, (who used to be) Bea’s husband
(57)
I kissed
. . . VP V kiss
ParPDP Par
DP Bill Par⊛
XP
(who used to be) Bea’s husband It should be pointed out straight away that the respective proposals in Heringa (2012) and De Vries (2009, 2012) involve some specific steps that are not represented here. For instance, Heringa (2012) argues that (attributive) NAs are in fact clausal. This is corroborated by the fact that they can be introduced by complementizers and may contain speaker-oriented adverbs and adverbs of tense (for data, see O’Connor 2008 and Heringa 2012). So, XP in (57) is a CP with a silent copula and a pro subject in Heringa’s analysis, which may bind a reflexive in the apposition and corefer with a matrix constituent (explaining the striking grammaticality of (46) above). In case of NARCs, the parenthetical XP is in fact a DP, which involves a light (null) head and a regular, restrictive relative. This is related to the intuition that the NARC in (56) is in fact interpreted as Bill, someone who used to be Bea’s husband. The facts illustrated above are consistent with this analysis: the opacity of the parenthetical XP is the consequence of the merger of XP with Par⊛ . In addition, the adjacency of anchor and parenthetical shown in (7)-(10) follows automatically from the proposed structure. After all, they are part of the same phrase. This is discussed in detail in Griffiths and De Vries (2013) for NARCs that are used as fragment answers. These involve fronting of the fragment answer and subsequent sluicing of the rest of the clause (similar to SPs and amalgams below). This then derives contrast between (58) and (59) under the assumption that the appositive relative and its anchor form a constituent, as ellipsis targets a non-constituent in the impossible case (59): (58)
A: B:
What did John steal? Mary’s computer, which crashes all the time John stole ti .
(59)
A: Who stole Mary’s computer? B: *John, stole Mary’s computer which crashes all the time.
It should be noted that (7)-(8), where anchor and parenthetical are separated by some A -movement of the parenthetical XP, are in addition excluded because movement is from a position that is not c-commanded by the destination position in the matrix as a result of XP being par-merged.
6. On representing anchored parentheses in syntax | 127
A desirable consequence of this analysis is that NARCs and NAs, which are intuitively two instances of the same species, are approached in a similar fashion. The section below discusses how this analysis works for SPs and amalgams.
6.4.3 Applying the ParP approach to SPs and amalgams The ParP approach to SPs and amalgams involves two additional factors: 1. in both cases the parenthetical XP (a CP) contains an ellipsis site, and 2. in case of amalgams, the anchor is null. (61) is a basic representation for these constructions: (60)
Bea hit someone/ex [you’ll never guess whoi /I think it was the professori Bea hit ti in the face] in the face.
(61)
Bea hit
... VP VP V
PP ParPDP in the face
DP someone/ex Par⊛
Par CPIC
you’ll never guess whoi / I think it was the professori Bea hit ti in the face Similar to what Heringa (2012) suggests for NAs, the complement of Par⊛ is clausal. This CP includes an embedded CP out of which there is A -movement, i.e. wh- or focusmovement, and the remnant (a layer of CP in the Horn-cases, see Kluck 2011) is the target of sluicing. The presence of an ellipsis site explains an intruiging set of facts in amalgams that for expository reasons were not discussed above. This concerns an accessibility paradox that is illustrated by (62): the referential remnant of sluicing in the parenthetical appears to give rise to a condition C effect with a coreferring expression in the matrix (62a), while an R-expression elsewhere in the parenthetical CP does not (62b). Under the present assumptions, the latter is expected, on a par with (3) and (44). However, (62a) can be explained as a common reconstruction effect: the remnant of sluicing is not c-commanded by the coreferring pronoun in the matrix, but in the ellipsis site (for ease of representation, I use a gray copy to denote the trace of the A -moved constituent):
128 | Marlies Kluck (62)
a. *Hei cited, [I think it was the professori hei cited the professori ]. b. Hei cited [the professori thought it was the most famous physician in the worldj that hei cited the most famous physician in the worldj ].
Reconstruction then explains a variety of connectivity effects in amalgams (amongst which quantifier/anaphor binding, case connectivity and condition B effects). More specifically, Horn-amalgams pattern exactly with regular it-clefts in this regard (see Reeve 2012 for extensive discussion): (63)
The story about himselfi /*himi /*Billi that hei denied was suspicious.
These are thus not the result of interaction with the matrix: the parenthetical itself is opaque. This explains the absence of condition C effects illustrated in (62b), but also the impossibility to bind a subject pronoun inside the parenthetical by a matrix quantifier, or NPI subjects by matrix negation (for a complete dataset, see Kluck 2011). There is another property of amalgams that follows straightforwardly from the parenthetical approach, namely that the interrupting clause exhibits root phenomena. (64)–(66) illustrate the most notorious amongst these: independent illocutionary force, topicalization and V2 in Dutch amalgams (Dutch main clauses are strictly V2, embedded clauses are V-final, and embedded V2 is excluded): (64)
a. b.
Bob found – [did you say it was a Stradivarius?] – in his attic. Bob found – [guess what!] – in his attic.
(65)
a. b.
Bea kissed, [Bobi I think it was ti that Bea kissed ti ], at the party. Bob kissed, [how many womeni you can’t even begin to imagine ti Bob kissed ti ], at the party.
(66)
a.
Bob heeft [je raadt nooit hoeveel koekjes] gestolen. Bob has you guess never how.many cookies stolen ‘Bob has stolen you’ll never guess how many cookies.’
b. *Bob heeft [je nooit hoeveel koekjes raadt] gestolen. Bob has you never how.many cookies guess stolen Unsurprisingly, NARCs and NAs can express independent illocutionary force as well, which fits the traditional impression that they constitute independent phrases/ clauses:¹⁴ (67)
Bob, (who is) a doctor, right?, is kind of stupid.
14 However, NARCs pattern syntactically with RRCs, as they do not allow for internal topicalization, and Dutch NARCs are V-final. This seems consistent with the idea that NARCs are in fact appositions consisting of a null head and a regular restrictive relative clause, i.e. as briefly explained above. For more discussion, see Cardoso and De Vries (2010), De Vries (2009, 2012).
6. On representing anchored parentheses in syntax | 129
Root phenomena constitute the major motivation for orphan approaches to parentheses, for which anchored parentheses constitute a particularly difficult class. A great merit of the non-restrictive parallel configuration for anchored parentheses is thus that it allows us to account for their independent (i.e. root) status while analysing them as part of the same syntactic unit as their anchor.
6.5 Two points of consideration This last section discusses two points of consideration which should be taken as directions of future research. First, non-nominal appositions appear to be able to float around in their hosts. On a more theoretical level, the parallel between parenthetical inclusion and coordination raises some new questions. This paper was purposely restricted to nominal appositions. One of the reasons for this is that non-nominal appositions do not seem to be subject to the adjacency requirement illustrated by (7)–(10), as is illustrated in (68) for adjectival appositions: (68)
a. b.
Bob, completely drunk/my neighbor, knocked on my door in the middle of the night. Bob knocked on my door, completely drunk/*my neighbor, in the middle of the night.
It seems that adjectival appositions can ‘float’ around in their hosts, whereas nominal appositions are connected to their anchor.¹⁵ This implies that the present proposal is not suitable for adjectival appositions. It has been noted that except for the comma-intonation, adjectival appositions are highly reminiscent of depictives (69): (69)
Bob knocked on the door completely drunk.
Traditionally, depictives have been analysed as small clauses with a PRO subject (Stowell 1981, 1983; Legendre 1997). However, as has been pointed out by various scholars, there is a clear difference between regular (restrictive) depictives and appositive adjectives (see Schultze-Berndt and Himmelmann 2004; O’Connor 2008; Heringa 2009). (70) illustrates this for the scope of negation:
15 A careful reader might however object that the adjaceny requirement does not hold for NAs either, as they may appear in the right periphery, i.e. My neighbor knocked on my door in the middle of the night, Bob. However, the resemblance between appositions and such cases is deceptive: right-dislocated appositions are in fact ‘specificational afterthoughts’. As was mentioned before, Ott and De Vries (2012) analyse these in terms of specifying coordination (i.e. coordination at the sentence level, employing the Kosterian colon phrase), with sluicing in the second conjunct. This explains a variety of connectivity effects (in many respects similar to what was shown above for amalgams and SPs).
130 | Marlies Kluck (70)
a. b.
He didn’t leave outraged. He didn’t leave, outraged.
The fact that the predicate is outside the scope of negation in (70b), suggests that adjectival appositions belong to the parenthetical domain. Other important differences pointing in this direction concern the types of antecedents that are allowed, (im)possibility of modification of high adverbs, and the level of modification of the predicate. The parallels and differences between depictives and adjective appositions can perhaps be captured if we integrate the ParP approach with the traditional small clause/PRO analysis for depictives, as is speculated in Heringa (2009). This paper draws a parallel between coordination and parenthesis: bivalent Par⊛ is a non-restrictive variant of Koster’s colon phrase, where the colon stands for a coordinator in the broadest sense. Anchor and parenthetical are thus analysed in terms of parallel construal, whereas free parentheses are merely adjuncts. As is emphasized in Koster (2000: 20) ‘it is very important to realize that parallel construal is not the same as coordination. Parallel construal is a more encompassing notion, with coordination only as a subcase. All parallel construals have certain properties in common (. . .). However, there are also certain differences, depending on the nature of the connecting head (. . .)’. Nevertheless, if the colon is a coordinator in the broadest sense, we expect a correspondence between anchored parentheses and the occurrence of coordinators, and perhaps the absence of such correspondence between coordination and free parenthesis. Neither appears to be the case: not all anchored parentheses can be introduced by a coordinator (a lexicalization of Par⊛ ), and coordinators may also occur in what I have classified as free parentheticals. The latter is already clear from (5) and (6), i.e. the presence of and is the unsurprising distinctive property of so-called and-parentheticals (see Blakemore 2005; Kavalova 2007; a.o.). NAs can easily occur with ‘apposition markers’: elements that express the relation between anchor and apposition. These markers may be the ordinary coordinators and and or (examples from Quirk et al. 1985: 1311–1312):¹⁶ (71)
a. b.
The United States of America, or America for short . . . You could cut the atmosphere with a knife, and a blunt knife at that.
By contrast, the parentheticals in NARCs, SPs and amalgams cannot be introduced by coordinators at all: (72)
a.
Bea kissed the professor, (*and/or) who is a genius, at the party.
16 Other well-known markers for English NAs are namely and that is. Since the latter may co-occur with coordinators in NAs, Heringa (2012) argues that they are adverbial rather than lexicalizations of Par⊛ . It should be noted that attributive appositions are usually asyndetic. They do not occur with regular coordinators, but can be introduced by markers such as as you know. For extensive overviews of markers and types of relations in appositions, see Heringa and De Vries (2008) and Heringa (2012).
6. On representing anchored parentheses in syntax | 131
b.
Bea kissed (someone), (*and/or) I think it was Bob/you’ll never guess who, at the party.
This raises the question to what extent a parallel between coordination and parenthesis really holds, and if the (im)possibility of an overt Par⊛ should be taken as indicative of such a parallel. Another question in this context relates to the traditional Law of Coordination of Likes (LCL, going back to Williams 1981). With the exception of the ParP analysis of NARCs, where both anchor and relative are DPs, the other constructions discussed in this paper do not obey the LCL. That is, the parentheticals in NAs, SPs and amalgams are CPs, while their anchors are not. This suggests that parallel construal involves heads with varying properties, an interesting direction for further research.
6.6 Conclusion This paper brings together a range of constructions I have labelled ‘anchored parentheses’: parentheses that are not freely attached somewhere in their host, but are attached at the constituent-level, namely to an anchor. It was argued that nominal appositions (NAs), nominal appositive relative clauses (NARCs), amalgams and sluiced parentheticals belong to this category, based on the observation that they all add nonrestrictive, speaker-oriented and more specific content to their anchors. Taking into account that the position of anchored parentheses is restricted and not arbitrary within the host sentence, I argue for a syntactic approach (i.e. contra so-called orphan approaches). The proposed analysis of anchored parentheses in terms of a bivalent ParP configuration, the non-restrictive variant of Koster’s colon phrase, seems promising: a linear connection between anchor and parenthetical is guaranteed because they are part of the same phrase, the structural independency of the parenthetical is the consequence of par-Merge. Finally, apparent exceptions to the independent status of an anchored parenthetical are explained by the assumptions about their internal syntax, such as a pro subject in NAs and the presence of elided structure in SPs and amalgams as is standard in regular sluicing. Acknowledgement: This research is funded by the European Research Council (ERC), and is part of the starting grant ‘Incomplete Parenthesis’ (INCPAR). In addition to the workshop Complex Sentences, Types of Embedding, and Recursivity at the University of Konstanz, versions of this paper have been presented at (Mis)matches in Clause Linkage, at the ZAS (Berlin) and the weekly meeting of the UiLOTS Syntax group (SyntaxUiLOTSIf ). This paper benefitted greatly from the feedback on these occasions. In addition, I thank the two anonymous reviewers of this paper for their careful comments.
132 | Marlies Kluck
References Abney, Steven. 1987. The English noun phrase in its sentential aspect. Ph. D. diss., Massachusetts Institute of Technology. Ackema, Peter and Ad Neeleman. 2004. Beyond morphology: interface conditions on word formation. New York: Oxford University Press. Akmajian, Adrian. 1970. On deriving cleft sentences from pseudo-cleft sentences. Linguistic Inquiry 1: 149–168. Akmajian, Adrian. 1979. Aspects of the grammar of focus in English. New York: Garland. Blakemore, Diane. 2005. And-parentheticals. Journal of Pragmatics 37: 1165–1181. Burton-Roberts, Noel. 1999. Language, linear precedence and parentheticals. In The clause in English: In honour of Rodney Huddleston, Peter Collins and David Lee (eds.), 33–52. Amsterdam/Philadelphia: John Benjamins. Cardoso, Adriana and Mark de Vries. 2010. Internal and external heads in appositive constructions. Manuscript, University of Lisbon and University of Groningen. Chung, Sandra, William A. Ladusaw and James McCloskey. 1995. Sluicing and Logical Form. Natural Language Semantics 3: 239–282. Constant, Noah. 2012. Witnessable quantifiers license type-e meaning: Evidence from contrastive topic, equatives and supplements. Proceedings of Semantics and Linguistic Theory 22: 286–306. Corver, Norbert. 1990. The syntax of left branch extractions. Ph. D. diss., Katholieke Universiteit Brabant. Corver, Norbert. 1997. The internal syntax of the Dutch extended adjectival projection. Natural Language and Linguistic Theory 15: 289–368. Craenenbroeck, Jeroen van. 2010. The Syntax of ellipsis. Evidence from Dutch dialects. New York: Oxford University Press. d’Avis, Franz-Josef. 2002. On the interpretation of wh-clauses in exclamative environments. Theoretical Linguistics 28: 5–31. Dehé, Nicole. 2007. The relation between syntactic and prosodic parenthesis. In Parentheticals, Nicole Dehé and Yordanka Kavalova (eds.), 261–284. Amsterdam/Philadelphia: John Benjamins. Dehé, Nicole, and Yordanka Kavalova. 2007. Parentheticals: An introduction. In Parentheticals, Nicole Dehé and Yordanka Kavalova (eds.), 1–22. Amsterdam/Philadelphia: John Benjamins. Del Gobbo, Francesca. 2003. Appositives at the interface. Ph. D. diss., University of California, Irvine. Del Gobbo, Francesca. 2007. On the syntax and semantics of appositive relative clauses. In Parentheticals, Nicole Dehé and Yordanka Kavalova (eds.), 173–201. Amsterdam/Philadelpia: John Benjamins. Dikken, Marcel den. 2006. Relators and linkers: The syntax of predication, predicate inversion and copulas. Cambridge, MA: MIT Press. Dikken, Marcel den. 2009. Predication and specification in the syntax of cleft sentences. Manuscript, CUNY. Döring, Sandra. 2007. Quieter, faster, lower and set off by pauses? In Parentheticals, Nicole Dehé and Yordanka Kavalova (eds.), 285–307. Amsterdam/Philadelphia: John Benjamins. Emonds, Joseph. 1976. A transformational approach to English syntax: root, structure-preserving and local transformations. New York: Academic Press. Emonds, Joseph. 1979. Appositive relative clauses have no properties. Linguistic Inquiry 10: 211– 243. Espinal, M. Teresa. 1991. The representation of disjunct constituents. Language 67: 726–762.
6. On representing anchored parentheses in syntax |
133
Geurts, Bart. 2011. Specific indefinites, presupposition, and scope. In Presuppositions and discourse, Rainer Bäuerle, Uwe Reyle and Thomas E. Zimmermann (eds.), 125–158. Bingley: Emerald. Griffiths, James and Mark de Vries. 2013. The syntactic integration of appositives: evidence from fragments and ellipsis. Linguistic Inquiry 42: 332–344. Guimarães, Maximiliano. 2004. Derivation and representation of syntactic amalgams. Ph. D. diss., Department of Linguistics, University of Maryland. Haegeman, Liliane. 1991. Parenthetical adverbials: The radical orphanage approach. In Aspects of Modern English: Papers presented to Masatomo Ukaji on his 60th birthday, Shuki Chiba, Akira Ogawa, Yasuaki Fuiwara, Norio Yamada, Osamu Koma and Takao Yagi (eds.), 232–254, Kaitakusji, Tokyo. Reprint New York: Routledge, 2009. Hedberg, Nancy. 1990. Discourse pragmatics and cleft sentences in English. Ph. D. diss., University of Minnesota. Hedberg, Nancy. 2000. The referential status of clefts. Language 76: 891–920. Heringa, Herman and Mark de Vries. 2008. Een semantische classificatie van apposities. Nederlandse Taalkunde 13: 60–87. Heusinger, Klaus von. 2002. Specificity and definiteness in sentence and discourse structure. Journal of Semantics 19: 245–274. Heringa, Herman. 2009. How depictives and appositive adjectives are different. Paper presented at TABU 2009, University of Groningen. Heringa, Herman. 2012. Appositional constructions. Ph. D. diss., University of Groningen. Heycock, Caroline and Anthony Kroch. 1999. Pseudocleft connectedness. Implications for the LF interface level. Linguistic Inquiry 30: 365–397. Heycock, Caroline. 2012. Specification, equation and agreement in copular sentences. Canadian Journal of Linguistics 57: 209–240. Higgins, Francis. 1979. The pseudo-cleft construction in English. Ph. D. diss., Massachusetts Institute of Technology. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge, MA: MIT Press. Johannessen, Janne Bondi. 1998. Coordination. New York: Oxford University Press. Kavalova, Yordanka. 2007. And-parenthetical clauses. In Parentheticals, Nicole Dehé and Yordanka Kavalova (eds.), 145–172. Amsterdam/Philadelphia: John Benjamins. Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Kennedy, Christopher and Jason Merchant. 2000. Attributive comparative deletion. Natural Language and Linguistic Theory 18: 89–146. Kluck, Marlies. 2011. Sentence amalgamation. Ph. D. diss., University of Groningen. Kluck, Marlies. 2013. A sluicing account of amalgams. Manuscript, University of Groningen. Kluck, Marlies and Mark de Vries. 2013. Cumulative rightward processes. In Rightward movement in a comparative perspective, Gert Webelhuth, Manfred Sailer and Heike Walker (eds.), 281–317. Amsterdam/Philadelphia: John Benjamins. Koster, Jan. 1999. Empty objects in Dutch. Manuscript, University of Groningen. Koster, Jan. 2000. Extraposition as parallel construal. Manuscript, University of Groningen. Koster, Jan. 2012. Relative clauses: Parallelism and partial reconstruction. Manuscript, University of Groningen. LaCara, Nicholas. 2012. Predicate Which-Appositives. In Proceedings of the Poster Session of the West Coast Conference on Formal Linguistics (WCCFL), Jaehoon Choi, E. Alan Hogue, Jeffrey Punske, Deniz Tat, Jessamyn Schertz and Alex Trueman (eds.), 46–55. Tucson: Coyote Papers.
134 | Marlies Kluck Lakoff, George. 1974. Syntactic Amalgams. In Papers from the 10th regional meeting of the Chicago Linguistic Society, Michael Galy, Robert Fox and Anthony Bruck (eds.), 321–344. Chicago: University of Chicago. Legendre, Géraldine. 1997. Secondary predication and functional projections in French. Natural Language & Linguistic Theory 15: 43–87. Merchant, Jason. 2001. The syntax of silence. Sluicing, islands, and the theory of ellipsis. New York: Oxford University Press. Mikkelsen, Line. 2005. Copular clauses: Specification, predication and equation. Amsterdam/Philadelphia: John Benjamins. Mikkelsen, Line. 1993. Topics in the syntax and semantics of coordinate structures. Ph. D. diss., University of Maryland. Nouwen, Rick. 2014. A note on the projection of appositives. In Formal approaches to semantics and pragmatics: Japanese and beyond, Eric McCready, Katsuhiko Yabushita and Kei Yoshimoto (eds.), pp. 205–222. Dordrecht: Springer. O’Connor, Kathleen. 2008. Aspects de la syntaxe et de l’interpretation de l’apposition à antécédent nominal. Ph. D. diss., Charles de Gaule University - Lille 3. Ott, Dennis, and Mark de Vries. 2012. Thinking in the right direction. An ellipsis analysis of right-dislocation. Linguistics in the Netherlands 29: 123–133. Ott, Dennis, and Mark de Vries. 2013. Right-dislocation as deletion. Manuscript, University of Groningen. Potts, Christopher. 2002. The lexical semantics of parenthetical-as and appositive-which. Syntax 5: 55–88. Potts, Christopher. 2005. The logic of conventional implicatures. New York: Oxford University Press. Quirk, Randolf, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik. 1985. A comprehensive grammar of the English language. London: Longman. Reeve, Matthew. 2012. Clefts and their relatives. Amsterdam/Philadelphia: John Benjamins. Reinhart, Tanya. 1983. Point of view in language: The use of parentheticals. In Essays on Deixis, Gisa Rauh (ed.), 169–194. Tübingen: Narr. Reinhart, Tanya. 1997. Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20: 335–397. Romero, Maribel. 1998. Focus and reconstruction effects in wh-phrases. Ph. D. diss., University of Massachusetts at Amherst. Romero, Maribel. 2000. Antecedentless sluiced wh-phrases and islands. In Ellipsis in conjunction, Kerstin Schwabe and Niina Zhang (eds.), 195–220. Tübingen: Niemeyer. Romero, Maribel. 2003. Correlate restriction and definiteness effect in ellipsis. In The interfaces: deriving and interpreting omitted structures, Kerstin Schwabe and Susanne Winkler (eds.), 263–300. Amsterdam/Philadelphia: John Benjamins. Ross, John R. 1967. Constraints on variables in syntax. Ph. D. diss., Massachusetts Institute of Technology. Ross, John R. 1969. Adjectives as Noun Phrases. In Modern studies in English: readings in transformational grammar, David Reidel and Sanford Shane (eds.), 352–360. Englewood Cliffs: Prentice-Hall. Sæbø, Kjell Johan. 2010. On the semantics of “embedded exclamatives". Studia Linguistica 64, 116–140. Safir, Ken. 1986. Relative clauses in a theory of binding and levels. Linguistic Inquiry 17: 663– 689. Schultze-Berndt, Eva and Nikolaus P. Himmelmann. 2004. Depictive secondary predicates in crosslinguistic perspective. Linguistic Typology 8: 59–131.
6. On representing anchored parentheses in syntax |
135
Schwarz, Bernhard. 2000. Topics in ellipsis. Ph. D. diss., University of Massachusetts at Amherst. Sells, Peter. 1985. Restrictive and non-restrictive modification. CSLI Report 85–28. Stanford CA: CSLI Publications. Stowell, Tim. 1981. Origins of Phrase Structure. Ph. D. diss., Massachusetts Institute of Technology. Stowell, Tim. 1983. Subjects across Categories. The Linguistic Review 2: 285–312. Svenonius, Peter. 1992. The structural location of the attributive adjective. In Proceedings of the 11th West Coast Conference on Formal Linguistics, Erin Duncan, Donka Farkas and Philip Spaelti (eds.), 439–454. Stanford: CSLI publications. de Vries, Mark. 2002. The Syntax of Relativization. Ph. D. diss., University of Amsterdam. de Vries, Mark. 2003. Three-dimensional grammar. Linguistics in the Netherlands 20: 201–212. de Vries, Mark. 2006. The syntax of appositive relativization: On specifying coordination, promotion and false free relatives. Linguistic Inquiry 37: 229–270. de Vries, Mark. 2007. Invisible constituents? Parentheses as b-merged adverbial phrases. In Parentheticals, Nicole Dehé and Yordanka Kavalova (eds.), 203–234. Amsterdam/Philadelphia: John Benjamins. de Vries, Mark. 2008. Asymmetric merge and parataxis. Canadian Journal of Linguistics 53: 355– 385. de Vries, Mark. 2009. Specifying coordination: An investigation into the syntax of dislocation, extraposition and parenthesis. In Language and linguistics: emerging trends, Cynthia R. Dreyer (ed.), 37–98. New York: Nova. de Vries, Mark. 2012. Parenthetical main clauses – or not? On appositives and quasi-relatives. In Main clause phenomena: new horizons, Lobke Aelbrecht, Liliane Haegeman and Rachel Nye (eds.), 177–201. Amsterdam/Philadelphia: John Benjamins. de Vries, Mark. 2012. Unconventional mergers. In Ways of structure building, Myriam UribeEtxebarria and Vidal Valalma (eds.), 143–166. New York: Oxford University Press. Webelhuth, Gert, Manfred Sailer and Heike Walker. 2013. Rightward movement in a comparative perspective. Amsterdam/Philadelphia: John Benjamins. Williams, Edwin. 1978. Across-the-board rule application. Linguistic Inquiry 9: 31–43. Williams, Edwin. 1981. Transformationless grammar. Linguistic Inquiry 12: 645–653.
Tonjes Veenstra
7 The development of subordination 7.1 Introduction Due to the socio-historical circumstances under which they arose, it has been argued that some of the grammatical properties of creole languages are in some fashion more directly reflective of the heart of language than older ones (Bickerton 1981; see McWhorter (2002, 2005) for a more recent statement). In order to take advantage of this special window on language creation, the properties under investigation, however, should trace back to the time of their genesis, rather than being later developments. Furthermore, they should be less susceptible to replacement due to processes of historical change, and they should be innovative in creole languages with respect to the relevant lexifier languages (cf. Muysken and Smith 1990). Sentence embedding (or subordination) constitutes in our view such a property for the following reasons. First, whereas sentence embedding is virtually nonexistent in the Hawai’ian Pidgin English data, Hawai’ian Creole English has sentential complements introduced by two complementizers, fo and go respectively (Roberts 1998; Veenstra 2008a). Second, in Veenstra (2008b), based on Van den Berg and Arends (2004), it was shown that sentence embedding in Sranan, the coastal creole of Surinam, was not only an early reconstitution, but also that the selectional properties of subordination are diachronically relatively stable, i.e. there is no substantial change from the earliest attestations in the sources from the 18th century till present day Sranan. Assuming the traditional pidgin-creole cycle, both the Hawai’an and Sranan data indicate two things: (i) there is no subordination in the pidgin phase, but in the creole phase there is; (ii) this feature is an early reconstitution. The first question, then, is how the system of subordination was reconstituted, and which factors operative in creole genesis are responsible for this reconstitution. We focus the discussion on the roles played by different processes of second language acquisition and relexification in this reconstitution process. The second, related, question is what was reconstituted. We assume the following structure to be basic for sentence embedding: (1)
Predicate [embedded clause (complementizer) (SUBJECT) V-(Aff) . . . ]
The three properties to be reconstituted in this configuration then are: (i) predicate class (and argument structure) of the predicate in the matrix clause taking a complement; (ii) presence/absence of lexical markers of subordination, i.e. complementizers; (iii) form (and/or position) of the embedded verb (e.g. subjunctive morphology in the Romance languages, or embedded Verb Second in some of the Germanic languages). We argue that the creation of the creole system is ultimately due to processes of second
138 | Tonjes Veenstra language acquisition, the properties (i) and (iii) are due to incipient second language learning, whereas property (ii) suggests the importance of advanced second language learning and bilingual usage in the creolization process. It is with respect to the latter property only that relexification can be said to have played a substantial role. The focus of the paper will be on a comparison of the subordination systems of Saramaccan and Fòngbe. Saramaccan is an English/Portuguese-based creole spoken along the Suriname River, Suriname. The language was created by slaves, who fled the plantations towards the end of the 17th century (Price 1983). Currently, the language has 50.000 speakers (Aboh et al. 2013) who reside on the banks of the Suriname River, in Paramaribo, in French Guiana, and in The Netherlands. Fòngbe has been identified as one of the major substrate language (Smith 1987). It is part of the greater Gbe cluster, spoken in Benin by fewer than 2 million speakers. The paper is organized as follows. In section 7.2 we give a quick overview of subordination in creoles in general. In section 7.3 we lay out the model of creole genesis that we assume in this study. Section 7.4 is devoted to the subordination systems of Fòngbe and Saramaccan, respectively. In section 7.5 we discuss the developmental pattern of subordination. Section 7.6 concludes the paper.
7.2 Subordination in creoles As noted in the introduction, sentence embedding is virtually nonexistent in Hawai’ian Pidgin English, whereas Hawai’ian Creole English has sentential complements introduced by two complementizers, fo and go respectively. As noted in Bickerton (1999), virtually all utterances in Hawai’ian Pidgin English show the following similar limitations of grammatical structure: (i) almost complete absence of grammatical items (no TMA markers nor complementizers); (ii) a virtually complete absence of embedded structures; (iii) frequent ellipsis of arguments (and even verbs). A typical example is given in (2): (2)
Pehea you KAITAI, you hanahana all same lili more me where you bastard you make/do all same little more me hanamake you! make-dead you ‘Why, you bastard, if you do that again I’ll kill you!’
In Hawai’ian Pidgin English, go is primarily attested as a marker of the imperative (in addition to its use as a main verb), while fo is mainly used as a preposition. In Hawai’ian Creole English, on the other hand, both are used as complementizers, go marking realized events and fo unrealized events. Examples of fo-complementation are given below:
7. The development of subordination
(3)
a. b.
|
139
I think more better fo I write that letter. ‘I think it would be better for me to write that letter.’ I try fo catch em in between da tumb and finga, but hard fo keep um in dat space when no moa webbing. ‘I tried to catch it in between the thumb and finger, but it’s hard to keep it in that space when there’s no webbing.’ (Chock 1998: 28)
Furthermore, Roberts (1998) notes that fo-complementation developed late and primarily within the population of locally-born speakers. Late in this context does not mean that it is not an early reconstitution, but that the reconstitution did not take place in the pidgin phase. Instead, it indicates that it took place in the population characterized by bilingual language use (see more on this in the following section). Thus, the innovation in Hawai’ian Creole English has two parts: The syntactic innovation consists of taking what are basically a preposition and an imperative marker and turning them into complementizers. The semantic innovation is that the two complementizers distinguish between realized and unrealized complements (Veenstra 2008a). From a wider perspective, it can be observed that this pattern seems to be typical of creoles, irrespective of their superstrate languages or their substrates. In (4) some examples are given: (4)
superstrate
creole
realized
non-realized
French
Haitian Mauritian Saramaccan Jamaican Papiamentu Sao Tomense
Ø ki táa se ku ku
pu pu fu fi pa pa
English Spanish Portuguese
In all these creoles, there is a division between marking realized and non-realized complements. As noted by Bickerton (1981), in most creoles, such non-realized complements are introduced by a complementizer with the phonetic form derived from the word meaning ‘for’ in the superstrate language: fo, fi, fu, or u from English for, pu from French pour, or pa from Spanish/Portuguese para. We take this pattern then to be the typical pattern of creoles in general: a subordination system with two complementizers, marking a realized/non-realized opposition. The question is how this system emerged in the transition from the pidgin to the creole phase. An important issue concerns the role of the substrate language(s) in this transition. Is this subordination system modeled on the one of the substrate language(s)? Do we have here a case of substrate transfer, and if so, which language acquisition processes are ultimately responsible for this?
140 | Tonjes Veenstra Siegel (2000) discusses this issue for Hawai’ian Creole English, and makes a strong case for substrate reinforcement rather than substrate transfer. He argues that the pattern of fo-complementation as found in Hawai’ian Creole English mirrors that of Portuguese, one of the substrate languages in Hawai’i. He shows that the word para ‘for’ (which is also used to introduce complements in creoles with Portuguese as a superstrate, e.g. pa in Sao Tomense) is functionally and syntactically very similar to Hawai’ian Creole English fo. First, both have a meaning roughly corresponding to ‘with a view to’, ‘for the purpose of’, ‘in order to’, or ‘intended for’. Second, both can introduce clausal complements with an overt subject in the nominative case (inflected infinitive, cf. Raposo 1987): (5)
a.
I think more better fo I write that letter. ‘I think it would be better for me to write that letter.’ (Hawai’ian Creole English)
a.
Emprestou-me dinheiro para eu comprar um casaco. lend.PST.PERF.3SG-1SG money FOR 1SG buy.INF a coat ‘He lent me money to buy a coat.’ (Portuguese)
Third, both in Hawai’ian Creole English and in Portuguese there is no TMA-marker in the embedded clause¹. Furthermore, Siegel (2000) adduces evidence from other pidgin/creole varieties of the Pacific region, where Portuguese was not among the superstrate or substrate languages, that we are not dealing with a clear case of substrate influence (or superstrate for that matter), since we find the same pattern in these varieties: (6)
a.
b.
c.
We want some leaf fer wrap it up. ‘We need some leaves to wrap it up.’ (Pitcairnese) Hem i go fo kasem samfala mit. ‘He went in order to get some meat.’ (Solomon Islands Pijin) Ai go singaut Lala po em kam. ‘I’ll go tell Lala to come.’ (Torres Strait Creole)
1 TMA stands for Tense/Mood/Aspect. The acronym is used in creole studies, because this is the order in which the markers that convey temporal, mood or aspectual information occurs, i.e. Tense >Mood >Aspect. In the non-creolist literature, however, one finds frequently TAM. The following abbreviations are used in the glosses: 1/2/3SG ‘first/second/third person singular’; Q ‘question particle’; SUB ‘subjunctive’; IRR ‘irrealis’; LOG ‘logophor’; GEN ‘genitive’; D ‘determiner’; P ‘preposition’; T ‘tense marker’; M ‘mood marker’; A ‘aspect marker’; PRTC ‘participle’.
7. The development of subordination
|
141
This pattern, he argues, is prima facie evidence that we are dealing with substrate reinforcement, i.e. extension of a pattern of grammaticalization that had already begun, rather than ‘direct’ substrate transfer. The conclusion we draw from the discussion so far is that the substrate potentially can account for some of the properties, but that it cannot solely account for all of the properties. In a sense, it has a facilitating role, reinforcing some trends that have been set by other forces and/or processes. Siegel (2000) does not couch his findings in terms of different populations having different acquisition mechanisms at their disposal. If we take Roberts’ (1998) findings serious that the development of the subordination system took place within the population of locally-born speakers, we have to look at those mechanisms that this population were entrenched with. This calls for a contact scenario with different mechanisms, including relexification, which may be grouped in terms of the stages in the process of creole genesis that they are most closely associated with. The topic of the next section is to discuss such a scenario. After that, we then proceed with a closer look at the reconstitution of the subordination system in Saramaccan in order to determine the role played by its major substrate Fòngbe.
7.3 Scenarios of creole genesis All current theories of creolization regard second language acquisition (SLA) as one of the determining factors in creole genesis, although the assumptions about its impact and the way it interacts with other contact phenomena, such as relexification, language shift, and language (re-)creation, differ widely (cf. DeGraff 1999; Mufwene 1990, 2010; Muysken 1981, 2008; different contributions in Lefebvre, White and Jourdan 2006; Plag 2008a, 2008b, 2009a, 2009b; Kouwenberg 2009). Nevertheless, several authors have argued that processes of second language acquisition cannot solely account for the emergence of substrate-related grammatical properties of creole languages (Kouwenberg 2006; Siegel 2008). Although there is ample evidence of such grammatical properties of creoles (e.g. Kouwenberg 1994; Lefebvre 1998; Winford 2003, among many others), Siegel (2008) mentions three problems with transfer accounts of these properties: (i) scarcity of evidence of early transfer in interlanguage studies and in restricted pidgins, i.e. detailed descriptions of interlanguages of second language learners have not provided evidence of the process of transfer that is supposed to be the source of some creole features – especially those features that appear to be superstrate forms with substrate grammatical properties (see also Kouwenberg 2006); (ii) existence of transferred structures in expanded pidgins and creoles that seemingly had nowhere to transfer to, in particular transfer of word order in which there was no congruence between the word orders of the L1 and L2, and, as such, problematic for the Transfer-to-Somewhere Principle of Andersen (1983); (iii) creole features apparently from the substrate that were not found in the preceding pidgin
142 | Tonjes Veenstra (where textual evidence exists from both the creole and the preceding pidgin), e.g. the TMA system as well as for-complementation in Hawai’ian Creole English developed late and primarily within the population of locally-born speakers (Roberts 1998). This implies that adult learners cannot have been the sole contributors of grammatical properties (and structure) to creoles and that a scenario of creole genesis ought to involve multiple agents². Becker and Veenstra (2003) reached a similar conclusion and proposed that the process of creolization requires a three-generational scenario of language shift (cf. Bickerton 1977; Corne 1999; Roberts 2000). The main difference between creolization and other language-shift situations lies in a target shift in the case of creolization (see Baker 1990, 1996 for argumentation of this point). Such a scenario was first proposed in Corne (1994) for Tayo. Further evidence for this approach to creole genesis has been adduced by Jennings (1995) for Cayenne Creole and Becker and Veenstra (2003) for Mauritian Creole. The picture of the shift to creole is then as follows: (7)
Multi-generational scenario of creole genesis (after Becker and Veenstra 2003) G1 (foreign-born immigrants) G2 (first generation locally born) G3 (second generation locally born)
L1 ancestral language(s) L2 pidgin L1 ancestral language(s) L1 pidgin/creole L1 creole (L2 ancestral language(s))
In this scenario, the first generation (G1) consists of immigrants/slaves who pidginize the superstrate language. The next generation (G2) is that of the locally born children of G1. According to Corne (1994: 296), this is the crucial generation. As he puts it, writing about the emergence of Tayo: Members of G2 retain their ancestral L1 . . . but, like the children of immigrants everywhere, they have to acquire the language of the community as a whole. Their problem is that there is no community L1 apart from pidginized L2 French, and they must therefore acquire/create L1 competence on the basis of their exposure to (some subset of) the varieties of pidginized L2 French, variable over time and depending on personal circumstances . . . It is they who participate in the creation of the nascent creole.
Members of the third generation (G3) are only indirectly exposed to their ancestral language(s), due to the presence of G1 and G2 speakers in their community; having
2 Independent evidence comes from studies on computational modeling of creole genesis. In particular, Satterfield (2008) notes that her findings suggest that patterns of adult L2 acquisition presumed under standard L2 creole accounts are not sufficient to explain the emergence of these creoles.
7. The development of subordination
|
143
largely lost their ancestral language(s), they can be considered the first monolingual speakers of the incipient creole. Creolization is thus seen as a complex and multi-dimensional phenomenon, not one that can be accounted for by postulating a single mechanism, but rather one that consists of at least two different processes, each with its own agents (Veenstra 1996: 195). The first process is that of untutored L2 acquisition, of which the first-generation adults (i.e., the different cohorts of slaves imported from West Africa) are the agents. This process is responsible for some of the SLA effects in the grammars of creole languages (Plag 2008a, 2008b, 2009a, 2009b). The second process is that of bilingual L1 acquisition, of which the children (i.e., the second generation, consisting of the creole population born in the colony) are the agents. In this process, we find, among other things, persistence of L1-related properties of grammar, and subsequent reinterpretation of superstrate material (Kouwenberg 2006, 2009)³. From this it follows that substrate influence in the case of creole languages is not a unified phenomenon, but rather we can speak of “Distributed Transfer” (Veenstra 2004)⁴: (8)
a. b.
substrate influence due to G1 (processes of SLA) is restricted to inheritance of general surface patterns but not specific grammatical properties; substrate influence due to G2 (processes of bilingual acquisition) targets specific grammatical properties of the contributing source languages.
One of the leading research questions, therefore, is that when we find cross-linguistic influence whether these continuities from the substrate languages are due to processes of SLA or of bilingual acquisition. One of the advantages of the multi-generational model in (7) is that it can incorporate the process of relexification in a natural way. Relexification is defined as a process of vocabulary substitution in which the only information adopted from the target language in the lexical entry is the phonological representation (Muysken 1981: 61). In the classic definition, the semantic representations of source and target language entries must partially overlap for relexification to occur; otherwise, the two entries would never be associated with each other:
3 Satterfield (2005) even goes so far to claim that ‘prototypical creole’ structures emerge when a very small percentage of older bilingual children were present in the population. 4 Similar proposals can be found in the literature as early as Haugen’s (1950) distinction between importation and calque, via Johanson’s (1992) distinction between partial and global copying, Aboh’s (2006a) pattern vs feature transmission, and Matras and Sakel’s (2007) pattern vs matter replication, to Cabrera and Zubizarreta’s (2004) observation that non-advanced learners focus on the L1 constructional properties of causatives, whereas advanced learners focus on L1 specific lexical properties of verb classes. Furthermore, Sanchez (2003, 2006) discusses the process of functional convergence, which typically takes place with bilingual speakers, and is akin to the process of relexification we know from creole studies (Lefebvre 1998) and contact linguistics in general (Muysken 1981).
144 | Tonjes Veenstra Relexification (9)
(Muysken 1981, 1988; Lefebvre 1998)
Original lexical entry [phonology]i [semantic feature]k [syntactic feature]n
Lexifier language [phonetic string]j used in specific semantic and pragmatic contexts
New lexical entry [phonology]j or [Ø] [semantic feature]k [syntactic feature]n As argued extensively by Muysken (1981, 1997), Media Lengua, spoken in Equador, is a prime example of a mixed language which arose through relexification and cannot be considered to represent a stage in learning Spanish as a second language for the following two reasons: (i) many Media Lengua-speakers also speak fluent Spanish; (ii) Media Lengua is very different from Quechua-Spanish interlanguage. This is shown in the following schematic contrast between Media Lengua and Spanish Interlanguage in central Ecuador (adapted from Muysken 2008): Table 7.1. Differences between relexification and L2-learning. Media Lengua (relexification)
Interlanguage (L2 learning)
Structure
complex
highly simplified
Degree of stabilization
rigid
highly variable
Source
Quechua morpho-syntax and phonology with slight Spanish influence
Spanish morpho-syntax and phonology with Quechua influence
Function
in-group language
interethnic communication
In the creole context it is G2, the first locally-born generation, acquiring simultaneously the emerging contact variety and their ancestral language(s), that is responsible for the relexificational effects that have been amply documented for the Atlantic creoles (see especially Lefebvre (1998) and Lumsden (1994, 1995, 1999) on Haitian Creole)⁵.
5 This conception of the notion of relexification diverges from the one argued for in Lefebvre (1998) along the following two parameters: (i) relexification does not occur with incipient adult SLA; (ii) the
7. The development of subordination
|
145
From the perspective just outlined, the question boils down to whether the reconstitution of subordination is due to processes of incipient second language acquisition or to the relexification process. A first indication that early second language acquisition is not the main factor in the reconstitution of this part of grammar is the observation by Klein and Perdue (1997) that in general there are no subordinated structures to be found in the earliest stages of second language acquisition (at the level of attainment they call “the Basic Variety”). As such, the Basic Variety (BV) resembles the pidgin phase, as exemplified by Hawai’ian Pidgin English above. Siegel (2008) lists the following properties that are shared between BVs and (restricted) pidgins⁶: (10)
a. no inflections; b. lexical items used in invariant form (multifunctionality); c.* invariant forms generally infinitive or nominative (but also some inflected forms); d. lexical items noun-like and verb-like words (with some adjectives and adverbs); e.* most lexical items from the L2 (but some from the L1 and other languages); f. only a few quantifiers; g. a single word for negation; h. only a few prepositions; i.* no complementizers; j. use of temporal adverbs, rather than grammatical TMA markers, to indicate temporality; k. “boundary markers” to express the beginning or end of some situation, such as work finish ‘after work is/was/will be over’.
The interlanguage features most relevant to subordination are the starred ones, repeated below for convenience: (11)
a. b. c.
invariant forms generally infinitive (but also some inflected; most lexical items from L2 (but some from L1); forms); no complementizers.
As noted in the introduction, the following three features of the subordination system can be distinguished: (i) predicate class (and argument structure) of the predicate in the matrix clause taking a complement; (ii) presence/absence of lexical markers of
language pair in relexification is not superstrate - substrate(s), but rather contact variety - substrate. This is more in line with the original characterization, given in Muysken (1981, cf. also 2008). 6 A significant part of these “simplified” structural features of pidgins can therefore be a consequence of processes of early second language acquisition, and some of these simplifications have found their way in the emerging creole (as BV-continuities, cf. Veenstra 2003).
146 | Tonjes Veenstra subordination, i.e. complementizers; (iii) form of the embedded verb. If we bring these two sets of features together, the following can be said on the role of incipient learning in the reconstitution process, assuming the model developed here is on the right track. First, since complementizers are absent from the BV stage and creole languages have them, retention cannot have taken place at this stage. Second, verbs in creole languages in general are not inflected, in matrix as well as embedded contexts. This feature can, therefore, be attributed to processes of incipient learning. Third, since clause-embedding predicates are mostly taken from the L2, it is an open issue whether their class membership and/or their argument structure are also taken from this language, or are copied from their L1 onto the predicates in the creole (as it would be the case in a relexification account). Alternatively, particular features of class membership and argument structure are neither taken over from the L2 nor from the L1, in which case we expect a (high) degree of variability in the selection patterns. So, the verdict is still open with respect to this last feature. Nevertheless, a first conclusion that we can draw from this discussion is that processes of incipient learning are only partly relevant for the reconstitution of this domain of grammar, and cannot account for all of the properties. The predictions of a relexification scenario, on the other hand, are very clear. First, we expect identical (or at least very similar) selection patterns, since the argument structure of the clause-embedding predicate is copied wholesale onto the creole predicate. Second, we expect a high degree of stabilization (at least as stable as the substrate language). If relexification is taken as a Full Access/Full Transfer approach to second language acquisition, as proposed by Sprouse (2006), the prediction is that we find an identical embedded C-domain as well as the same complementizer inventory. In order to evaluate the role played by the different acquisition processes operative in creole genesis then, we focus in the next section our attention on the patterns of subordination in Saramaccan and the Gbe languages, especially Fòngbe, concentrating thereby on sentence-embedding predicates and their associated syntax. As noted in the introduction above, one of the leading questions of this paper is whether the subordination system in Saramaccan strictly follows the patterns found in the substrate language Fòngbe. To answer this question, we restrict the discussion to the set of utterance, emotion, and propositional attitude predicates (Noonan 2007).
7.4 Sentence embedding in Fòngbe and Saramaccan The relation between the matrix predicate and the embedded clause is expressed (or mediated) in the highest projection of the clause, namely ForceP (Rizzi 1997). According to Rizzi (1997), only complementizers that select for a finite clause can head ForceP. Such complementizers are then selected by particular classes of verbs that occur in the matrix clause.
7. The development of subordination
|
147
7.4.1 Fòngbe According to Lefebvre and Brousseau (2002), there are two tensed complementizers in Fòngbe: ɖɔ and nú/ní. They argue that the first one is indicative and occurs with verbs from the SAY-class, whereas the second one is subjunctive and occurs with verbs from the WANT-class: (12)
a.
Kɖkú ɖì ɖɔ Bàyí wá. Koku believe ɖɔ Bayi come ‘Koku believed that Bayi came.’
b.
È nyɔ nú à ní yì. 3SG good NU 2SG SUB leave. ‘It is good that you leave.’
If verbs select for an embedded question, ɖɔ is obligatorily present – suggesting ɖɔ does not determine the typing of the clause: (13)
Ùn kán byɔ ɛ ɖɔ fitɛ yé nɔ nɔ àjí. 1SG ask 3SG ɖɔ where 3PL usually live-Q ‘I asked him/her where they live.’
The two complementizers are mutual exclusive: (14)
*Ùn jló ɖɔ ní à ní wá. 1SG want ɖɔ NI 2SG SUB come ‘I want you to come.’
We take this to mean that they are base-generated in the same syntactic position: (15)
PRED [ForceP ɖɔ/ní [. . . ]]
The subjunctive marker is obligatory present in clauses introduced by nú: (16)
*È nyɔ nú à yì. 3SG good NU 2SG leave
The element ni can also appear as an irrealis mood marker in root clauses: (17)
Bàyí ní ɖà wɔ. Bayi SUB prepare dough ‘Bayi should prepare dough.’
According to Kinyalolo (1993), there is a set of verbs in Fòngbe which is obligatorily followed by a ɖɔ-phrase (cf. Lefebvre and Brousseau 2002 for a caveat – see below). This is shown in Table 7.2.
148 | Tonjes Veenstra Table 7.2. Fòngbe verbs that take a ɖɔ-complement. ɖɔ ‘say, tell’ kan by ‘ask’ mɔn ‘deny’ flin ‘remember’ yi gbe ‘answer’
lin ‘think’ wɔn ‘forget’ do akpa ‘promise’ se ‘hear’ ɖo nukun ‘hope’
kɛn ‘bet’ gbɛ ‘refuse’ tùn ‘know’ zɔn ‘command’ xwlé ‘swear’
kú drɔ ‘dream’ jlo ‘want’ mɔ ‘see’ ɖi ‘believe’ gblɔn adan ‘threaten’
The only exception is ɖɔ ‘say’ itself. Only if the verb is accompanied by a PP, it is obligatorily present, otherwise it is optional: (18)
É ɖɔ (ɖɔ) Sìká ná wá. 3SG say ɖɔ Sika IRR come ‘S/he said that Sika will come.’
(19)
É ɖɔ nú mí *(ɖɔ) Sìká ná wá. 3SG say P 1SG ɖɔ Sika IRR come ‘S/he told me that Sika will come.’
Some of these verbs (obligatorily) select for a complement with a future tenseorientation, like do akpa ‘promise’, gblɔn adan ‘threaten’, whereas other do not seem to have similar temporal restrictions, e.g. flin ‘remember’: (20)
a.
Yè gblɔn adan *(ɖɔ) émi *(ná) hú gbɔ lɛ bi. 3PL utter anger ɖɔ LOG IRR kill goat PL all ‘They threatened to kill all the goats.’
b.
Mi flin *(ɖɔ) Kɔkú hwlá Asíbá sín gbɔ. 1PL remember ɖɔ Koku hide Asiba GEN goat ‘We remember that Koku hid Asiba’s goat.’
Lefebvre and Brousseau (2002) classify jlo ‘want’ differently. According to them, it selects either a tensed complement introduced by the complementizer nú (21a), an infinitival complement without an overt subject (21b), or an infinitival complement with an overt subject (21c) – the last is a case of Exceptional Case Marking (as signaled by the low tone on the pronoun): (21)
a.
Éi jló nú éi/j *(ní) yì. 3SG want NU 3SG SUB leave ‘(S)he wants to leave’ OR ‘(S)he wants him/her to leave.’
b.
Kɔkúi jló PROi/*j *(ná) gbà mɔtó ɔ. Koku want DEF.FUT destroy car DEF ‘Koku wants to destroy the car.’
7. The development of subordination
c.
|
149
Éi jló è*i/j yì. 3SG want 3SG leave ‘(S)he wants him/her to leave.’
We, therefore, leave this verb out of the discussion. What is of interest about the set of verbs from the list of Kinyalolo (1993) is that they do not form a homogenous class of clause-embedding predicates from a semantic point of view. If we adhere to the classification proposed by Rochette (1988)⁷, we can observe that these predicates occur in every subclass in this system. In our view, this indicates that there are no particular strong restrictions in terms of the lexical argument structure of clause-embedding predicates operative on the selection of ?? as the complementizer of the clause embedded under such verbs. The question, then, is how this set of verbs behaves syntactically in Saramaccan?⁸ Do we find the same pattern, i.e. do all these verbs also select for one and the same complementizer? This will be one of the main topics for the remainder of this section. In addition, we will have a closer look at the left peripheral architecture of embedded clauses in Saramaccan, and compare it with what is found in the Gbe languages.
7.4.2 Saramaccan The closest equivalent of Fòngbe ɖɔ is táa in Saramaccan. Like ɖɔ, táa can occur as a verb of speaking (derived from táki) and a serial verb/complementizer (in addition to a quotative marker, cf. Lefebvre and Loranger 2008). Unlike ɖɔ, however, táa cannot be followed by the serial verb/complementizer táa: (22)
*A táa táa a ó gó. 3SG talk SAY 3SG M go
7 According to Rochette (1988), verbs are divided semantically as follows. First, verbs divide into two major classes: effective and non-effective (or reflective) verbs. “Effective predicates are verbs that describe a subject’s relationship, whether causal, potential or other, to the performance of an action (examples: ‘to request’, ‘to see to’). Non-effective or reflective predicates are verbs which express a (human) subject’s judgement concerning a proposition or event” (Rochette 1988: 21). Reflective predicates further divide into two classes: propositional and emotive verbs. Propositional verbs describe judgements of truth-value (examples: ‘to assert’, ‘to negate’, ‘to say’) and emotive verbs describe judgements of personal relevance (examples: ‘to feel’, ‘to look’, ‘to be happy’). We will present the verbs under discussion in this paper according to this classification. 8 It is interesting to note that Papiamentu clearly shows that the distribution of the complementizers ku and pa is dependent on the indicative/subjunctive split, and, ultimately, related to the distinction between propositional and emotive/effective predicates (cf. Kouwenberg and Lefebvre 2007 for details). Alleesaib and Veenstra (in prep.) show that the distribution of the complementizers ki and pu in Mauritian Creole is independent of any verb classification.
150 | Tonjes Veenstra Like ɖɔ, táa is not involved in clause-typing, i.e. it does not type the clause as declarative, since it can co-occur with embedded questions: (23)
De bì tá kuútu táa unfa u ó dú dí sóni akí. 3PL T A judge SAY how 1PL M do D thing here ‘They were judging/wondering how they would do this.’
There are two major patterns of sentential embedding in Saramaccan: sentences introduced by táa and/or fu⁹. Damonte (2002) observes that they have the following distribution. As noted above, táa is a finite complementizer. As such, it cannot be used with a verb that requires an irrealis complement. The lexical item fu, on the other hand, is the so-called subjunctive complementizer, which is derived in (almost) every creole from the preposition meaning for in the relevant European lexifier language. It cannot be used with a verb like know, where a realis interpretation is usually assumed: (24)
a.
A sábi táa/*fu dí wómi bì hóndi dí píngo. 3SG know SAY D man T hunt D pig ‘He knows that the man hunted the pig.’ (Byrne 1987: 148)
b.
A ké faa/*táa kísi dí ogifóu a mátu. 3SG want FOR.3SG catch D owl P jungle ‘He wants him to catch the owl in the jungle.’ (Byrne 1987: 138)
If the matrix verb is compatible with both a realised and an unrealised sentential complement, both complementizers are possible: (25)
a.
A táki táa dí mujée bì gó a dí kéíki. 3SG say SAY D woman T go P D church ‘He said that the woman had gone to the church.’
b.
A bì táki f’én kulé. 3SG T talk FOR.3SG run ‘He told him to run.’ (warning)
The choice of the complementizer affects the interpretation of the embedded sentence. According to Byrne (1987), the (main) difference in interpretation between táa-clauses and fu-clauses is that the event described by the clause introduced by táa is presupposed to have occurred or to occur in the near future, whereas the event described by
9 The base form of the complementizer is fu. It can also be realized as u. The alternation does not depend on any grammatical factors, however. In other words, they are fully interchangeable. In addition, it can amalgamate with pronouns, leading to the following forms: faa TAA TAA > Top/Foc
The higher structure of the C-domain of Saramaccan contrasts with the one we encountered in Fòngbe where the two complementizers are in complementary distribution (see above). The conclusion we draw is that the ordering of elements in the higher
11 Unfortunately, we do not have data at hand in which Topics and focused elements co-occur with both complementizers.
7. The development of subordination
|
153
CP of Fòngbe is not mirrored in Saramaccan, and that, therefore, the left peripheral architecture in Saramaccan has not been modelled after the one found in Fòngbe. Another difference between the two languages is to be found in so-called subjunctive clauses. In Saramaccan, modality markers are not possible in subordinate clauses introduced by fu, as shown in the next set of examples: (32)
a.
Mi ké fu a dú dati. 1SG want FOR 3SG do that ‘I want him to do that.’ (Wijnen and Alleyne 1987)
b. *A ké faa fu kísi dí ogifóu a mátu. 3SG want FOR.3SG FOR catch D owl P jungle (Damonte 2002) c. *Mi ké fu a fu dú dati. 1SG want FOR 3SG FOR do that (Kouwenberg and Lefebvre 2007) As shown in (16) above, mood markers are obligatory present in embedded subjunctives in Fòngbe. Thus, the syntax of fu-clauses in Saramaccan is not identical to the syntax of nú-clauses in Fòngbe. There seems to exist variation between speakers as to whether fu is able to occur in root clauses as a mood marker or not. Thus, some of the informants of Byrne (1987) accept the following type of examples: (33)
% Dí wómi bì fu woóko a dí baáka wósu. D man T FOR work P D foreigner house ‘The man should have worked at the foreigner’s house.’ (Byrne 1987: 136)
Wijnen and Alleyne (1987), Veenstra (1996), Damonte (2002), McWhorter (2005), Van de Vate (2008) were not able to replicate the results of Byrne, however. We will assume that for the majority of Saramaccan speakers fu cannot be used as a Mood marker in root contexts. Fòngbe is also different in this respect: ní can function both as a subjunctive complementizer and as a mood marker occurring in independent (i.e. matrix) clauses. Note, in addition, that it is exactly for the group of speakers that do not have fu as a mood marker that the pattern with the co-occurring complementizers has been documented, explicitly in Veenstra (1996), Damonte (2002), and Van de Vate (2008). This rules out an analysis, hinted at in Aboh (2006b), to treat the fu in (27)–(28) as instances of the mood marker. Summarizing, we can state that the syntactic architecture of the embedded clause in Fòngbe and Saramaccan differ along the following two dimensions: (i) presence vs absence of co-occurrence restrictions on complementizer combinations; (ii) absence vs presence of mood marking in embedded subjunctive clauses. The overall conclusion is, therefore, that the left peripheral architecture in Saramaccan has not been
154 | Tonjes Veenstra modelled after the one found in Fòngbe, or the Gbe languages in general (cf. Aboh 2004). This conclusion contradicts the claims made by Aboh (2006b) and Lefebvre and Lorranger (2008). If we compare the Fòngbe set of clause-embedding predicates with the set of verbs in Saramaccan, we get the following result. Whereas in Fòngbe all the verbs, independent of the different verb classes they belong to, uniquely select for a ɖɔ-phrase (with the exception of jlo ‘want’, as noted above), we find a diffuse selectional pattern in Saramaccan. In total, four different patterns can be identified. The embedded clause is either introduced by the complementizer táa, the complementizer fu, a combination of both complementizers, or by no complementizer at all. Some verbs only select for one pattern, e.g. piki ‘say, tell’, pidi ‘ask’, fia ‘deny’, sunja ‘dream’, and soi ‘swear’ only take embedded clauses introduced by táa, whereas niinga ‘refuse’, ke ‘hope’ and maa tongo ‘command, order’ only select clauses headed by fu. Other verbs show variable behavior. The following combinations are found: táa/zero (e.g. feekete ‘forget’, si ‘see’), and táa/táa fu (e.g. taki ‘say, tell’). Absent is the combination fu/zero. The verb paamisi ‘promise’ seems to exhibit the full range of possibilities, as can be seen from (34) (please notice that in (34a-c) fu is realized as u, cf. fn 10): (34)
a.
Mi paamísi táa u m bì ó gó ku hén. 1SG promise SAY FOR 1SG T M go with 3SG ‘I promised that I would go with him.’
TAA FU
b.
Mi paamísi dí míi táa mi ó kó heépi hén. 1SG promise D child SAY 1SG M come help 3SG ‘I promised the child to help him.’
c.
Mi paamísi dí míi u kó heépi hén. 1SG promise D child FOR come help 3SG ‘I promised the child to help him.’
FU, − SUBJECT
d.
Mi paamísi dí míi fu mi kó heépi hén. 1SG promise D child FOR 1SG come help 3SG ‘I promised the child to help him.’
FU, + SUBJECT
e.
Mi paamísi dí míi mi kó heépi hén. 1SG promise D child 1SG come help 3SG ‘I promised the child to help him.’
TAA
ZERO
The four patterns do not seem to correspond to any verb classification, as can be seen in Table 7.3. We do not find, for instance, a similar split as in Papiamentu between propositional and emotive/effective predicates (cf. Kouwenberg and Lefebvre 2007). The complementizer táa appears with verbs from all the different verb classes, the complementizer fu with verbs from all verb classes except a subclass of the propositional predi-
7. The development of subordination
|
155
Table 7.3. Classification in Saramaccan predicates and their complementizers. Reflective Propositional Stating say taa/taa fu fu/zero swear taa see taa hear taa dream taa deny taa answer taa
Effective Emotive Believing
Knowing
think taa/fu believe taa
forget taa/zero know taa remember taa/zero
hope taa/fu want taa/fu/zero
promise taa/taa fu/ fu/zero command fu threaten taa refuse fu
cates containing verbs of knowing. The combination of the two complementizers is also not confined to one particular verb class. This concludes section 7.4, in which it was shown that the syntax of clauseembedding predicates in Saramaccan does not match up with the one of Fòngbe. In particular, the embedded C-domains in Saramaccan and the Gbe languages do not match. Whereas in Fòngbe the two complementizers ɖɔ and ní/nú are in complementary distribution, the Saramaccan complementizers táa and fu can occur together. Furthermore, the variability in selection patterns encountered in this realm of grammar turns out to be highly problematic for a relexification scenario and seems to favor a scenario in terms of processes of second language acquisition instead.
7.5 Development of subordination We started with the following two questions. First, given the absence of a subordination system in pidgins, how is the system of subordination reconstituted in creoles, and which factors operative in creole genesis are responsible for this reconstitution? Second, what was reconstructed? We identified three properties: (i) predicate class (and argument structure) of the sentence-embedding predicates in the matrix clause; (ii) presence/absence of complementizers; (iii) form of the embedded verb. We focused the discussion on the roles played by processes of early and untutored second language acquisition and relexification in the process of reconstitution. We argued that
156 | Tonjes Veenstra the third property (morphological marking of the embedded verb) is clearly due to processes of early second language acquisition, as well as the first property (lexical specifications of the embedding verb), given the selectional variability of such predicates in Saramaccan in comparison with those from Fòngbe (and English, for that matter). Relexification we argued did not play a substantial role in this part of grammar. Furthermore, we showed that the embedded C-domain in the two languages is decidedly different. Still, the very fact that a subordination system was developed cannot be explained by these processes of early second language acquisition, given that at this level of attainment, Basic Varieties (BVs, see p. 9) do not have any subordination. Finally, we have not said anything yet about property (ii), the fact that creoles developed complementizers. If we look at what happens when second language learners go beyond the BV and start to develop subordinated structures, we get an important cue as to what may have taken place in creolization. Although I am not aware of large-scale studies on this topic, the study by Giacalone Ramat (1999) shows that some of the learners of Italian as a second language she investigated make extensive use of a multifunctional element come ‘as, how’ not only to introduce clauses expressing temporal and causal relations, but also as a complementizer-like element. All these uses are at odds with the grammar of Italian: (35)
anche c’e‘ una albergo eh++eh ha also there is an inn eh++eh have-3SG eh come eh tre: bicchiero di vino eh tu eh as eh three glass of wine eh you puoi passano a questo can-2SG pass-3PL to this
scritto eh+ come write-PRTC eh+ as hai mangiato non have-2SG eat-PRTC not
‘There is also an inn and there is written if you have drunk three glasses wine you may not go beyond.’ Giacalone Ramat (1999) specifically notes that Valentini (1998) takes come as a kind of ‘unspecified subordinator’, which is like a generic placeholder that the learner slowly refines over time in order to fit into the target-language system. We propose that something similar happened in the creolization context. Also in this case, an unspecified subordinator was created. This took place at the moment when the pidgin was treated as a separate (and autonomous) system, different from the superstrate language (Muysken 2001; Becker and Veenstra 2003). This occurrence of a target shift resulted in the acceptance of the pidgin (or BV) as a separate norm.¹² Due to this target
12 As Muysken (2001: 160) notes, “before a pidgin can be creolized, it needs to be accepted as a separate norm. The crystallization of the pidgin and the stability of a pidginized variety may be furthered by: (i) a socially homogeneous community speaking the pidginized variety; (ii) reciprocal native language/target language community communication in the pidginized variety; (iii) use of the pidginized
7. The development of subordination
|
157
shift, this unspecified subordinator or generic placeholder was not refined towards the target-language system over time, however, because there was no access to this system anymore. Thus, the emergence of a complementizer-like element is due when learners go beyond the BV in the case of second language acquisition as well as when learners in the creolization context go beyond the stage in which the pidgin is still treated as a form of the superstrate language. The result is an autonomous grammatical system that has an unspecified subordinator that is involved in linking up clauses. Now, if we compare the complementizer táa in Saramaccan with the complementizer as found in Fòngbe and other West African languages, the resemblance is quite telling. As Lefebvre and Loranger (2008) show, they share the following characteristics: (36) – speech verb ‘to say’ – speech verb ‘to talk/tell/chat’ – serial verb – complementizer – quote introducer – conjunction conveying comparison/manner
táa SA + – – + + +
ɖɔ Fòngbe + + + + + +
sé Twi + – – + + +
bé Ewe – – – + + +
(adapted from Lefebvre and Loranger 2008) The explanation for this resemblance then is that the highly underspecified subordinator was filled in with the grammatical properties of the general complementizer, as found in the Gbe languages (which actually is an areal feature from a larger area in West Africa, cf. Güldemann 2008). The close resemblance between the complementizers in Saramaccan and the Gbe languages comes about through the process of relexification.
7.6 Sentence embedding in Fòngbe and Saramaccan Given our leading questions of this paper, as to how the system of subordination was reconstituted in creoles, the potential role of the substrate language(s) (in terms of transfer or reinforcement), and which factors operative in creole genesis are responsible for this reconstitution, we come to the following conclusions.
variety in communication between different groups of non-target language speakers; (iv) the variety is learned by newcomers as a system; (v) pidginized variety mutually intelligible with target language.”
158 | Tonjes Veenstra With respect to the three properties to be reconstituted in the subordination system of creoles in general, the advantages of our scenario are that we can account for the following findings with respect to the role played by substrate(s) in the case of Saramaccan: (37)
a. b. c.
match in the functioning of the general complementizer; mismatch in selectional restrictions of matrix predicates; mismatch in morphological marking in the subordinated clause.
There is substrate continuity in the feature specification of the ‘unspecified subordinator’, as shown in (36). There is no substrate continuity in the selectional properties of clause embedding predicates. Whereas there is a set of verbs obligatorily followed by a ɖɔ-phrase in Fòngbe (no variation), we find variable selection patterns with this set of verbs in Saramaccan. There is no substrate continuity in morphological marking in embedded clauses. Whereas there is obligatory mood marking in Fòngbe in non-root contexts, mood marking is absent in these contexts in Saramaccan. In terms of different acquisition processes operative in creole genesis, we can state that incipient learning (i.e. early second language acquisition) played a substantial role, accounting for both variable selection patterns of clause-embedding as well as the absence of morphologically marked verb forms in embedded contexts. Non-incipient learning (i.e. more advanced second language acquisition), on the other hand, accounts for the appearance of an unspecified subordinator, while relexification can account for the specific properties this subordinator exhibits. Acknowledgement: Many thanks to the audience of the workshop on Complex Sentences, Types of Embedding, and Recursivity, held at the University of Konstanz, for their comments and questions, especially Tom Roeper. The paper also benefited greatly from the detailed comments of two anonymous reviewers. Funding for this project comes from the Federal Ministry of Education and Research (BMBF) of Germany (grant number 01UG0711), which is hereby gratefully acknowledged. The usual disclaimers apply. The bulk of the data stems from fieldwork conducted by the author in the Netherlands and in Surinam.
References Aboh, Enoch. 2004. The morphosyntax of complement-head sequences: Clause structure and word order patterns in Kwa. Oxford: Oxford University Press. Aboh, Enoch. 2006a. The role of the syntax-semantics interface in language transfer. In L2 acquisition and creole genesis, Claire Lefebvre, Lydia White and Christine Jourdan (eds.), 221–252. Amsterdam: John Benjamins. Aboh, Enoch. 2006b. Complementation in Saramaccan and Gungbe: The case of C-type modal particles. Natural Language and Linguistic Theory 24: 1–55.
7. The development of subordination
|
159
Aboh, Enoch, Norval Smith and Tonjes Veenstra. 2013. Saramaccan. In Survey of Pidgin and Creole languages, Volume I: English-based and Dutch-based languages, Susanne Michaelis, Philippe Maurer, Martin Haspelmath and Magnus Huber (eds.), 27–38. Oxford: Oxford University Press. Alleesaib, Muhsina and Tonjes Veenstra. in preparation. Database of complement-taking predicates in Mauritian Creole. Ms., ZAS. Andersen, Roger. 1983. Transfer to somewhere. In Language transfer to somewhere, Susann Gass and Larry Selinker (eds.), 177–201. Rowley, MA: Newbury House. Baker, Philip. 1990. Off Target? Journal of Pidgin and Creole Languages 5: 107–119. Baker, Philip. 1996. Pidginization, creolization, and français approximatif. Review article of Chaudenson (1996). Journal of Pidgin and Creole Languages 11: 95–120. Becker, Angelika and Tonjes Veenstra. 2003. The survival of inflectional morphology in Frenchrelated Creoles: The role of SLA processes. Studies in Second Language Acquisition 25: 283– 306. Bickerton, Derek. 1977. Change and variation in Hawaiian English. Vol. 2: Creole syntax. University of Hawai’i, Social Science and Linguistics Institute. Bickerton, Derek. 1981. Roots of language. Ann Arbor: Karoma. Bickerton, Derek. 1999. How to acquire language without positive evidence: What acquisitionists can learn from Creoles. In Language creation and language change: Creolization, diachrony and development, Michel DeGraff (ed.), 49–74. Cambridge, MA: MIT Press. Byrne, Frank. 1987. Grammatical relations in a radical Creole. [Creole Language Library 3] Amsterdam: John Benjamins. Cabrera, Monica and Maria Luisa Zubizarreta. 2004. The role of the L1 in the overgeneralization of causatives in L2 English and L2 Spanish. In Contemporary approaches to Romance linguistics, Julie Auger, Clancy J. Clements and Barbara Vance (eds.), 45–64. Amsterdam: John Benjamins. Chock, Eric (ed.). 1998. Growing up local: An anthology of poetry and prose from Hawai‘i. Honolulu: Bamboo Ridge Press. Corne, Chris. 1994. Relativization and thematization in Tayo and the implications for creole genesis. Journal of Pidgin and Creole Languages 9: 283–304. Corne, Chris. 1999. From French to Creole. London: University of Westminster Press. Damonte, Federico. 2002. The complementizer layer in Saramaccan. In Current issues in generative grammar. 10th colloquium on generative grammar, M. Leonetti, O. Fernadez Soriano and V. Escandell Vidal (eds.). Universidad de Henares. DeGraff, Michel. 1999. Creolization, language change and language acquisition: an epilogue. In Language change and creation: Creolization, diachrony and development, Michel DeGraff (ed.), 473–454. Cambridge, MA: MIT Press. de Groot, Antonius. 1981. Woordregister Saramakkaans-Nederlands. Paramaribo: VACO. Giacolone Ramat, Anna. 1999. Functional typology and strategies of clause connection in secondlanguage acquisition. Linguistics 37: 519–548. Güldemann, Tom. 2008. Quotative indexes in African languages: a synchronic and diachronic survey. [Empirical Approaches to Language Typology 34]. Berlin: Mouton de Gruyter. Haugen, Einar. 1950. The analysis of linguistic borrowing. Language 26: 210–231. Jennings, William. 1995. The first generations of a creole society: Cayenne 1660–1700. In From contact to Creole and beyond. Philip Baker (ed.), 21–40. London: University of Westminster Press. Johanson, Lars. 1992. Strukturelle Faktoren in türkischen Sprachkontakten. Sitzungsberichte der Wissenschaftlichen Gesellschaft an der J. W. Goethe-Universität Frankfurt am Main 29:5. Stuttgart: Steiner.
160 | Tonjes Veenstra Kinyalolo, Kasangati K. W. 1993. On some syntactic properties of ɖó in Fòn. Lingua 91: 201–233. Klein, Wolfgang and Clive Perdue. 1997. The basic variety. Second Language Research 13: 301–347. Kouwenberg, Silvia. 1994. A grammar of Berbice Dutch. Berlin: Mouton de Gruyter. Kouwenberg, Silvia. 2006. L1 transfer and the cut-off point for L2 acquisition processes in creole formation. In L2 acquisition and creole genesis, Claire Lefebvre, Lydia White and Christine Jourdan (eds.), 205–219. Amsterdam: John Benjamins. Kouwenberg, Silvia. 2009. The invisible hand in creole genesis: Reanalysis in the formation of Berbice Dutch. In Complex processes in new languages, Enoch Aboh and Norval Smith (eds.), 115–158. Amsterdam: John Benjamins. Kouwenberg, Silvia and Claire Lefebvre. 2007. A new analysis of the Papiamentu clause structure. Probus 19: 37–73. Lefebvre, Claire. 1998. Creole genesis and the acquisition of grammar. Cambridge: Cambridge University Press. Lefebvre, Claire and Anne-Marie Brousseau. 2002. A grammar of Fòngbe. Berlin: Mouton de Gruyter. Lefebvre, Claire, Lydia White and Christine Jourdan (eds.). 2006. L2 acquisition and creole genesis. Amsterdam: John Benjamins. Lefebvre, Claire and Virginie Loranger. 2008. A diachronic and synchronic account of the multifunctionality of Saramaccan taá. Linguistics 46: 1167–1228. Lumsden, John. 1994. Possession: substratum semantics in Haitian Creole. Journal of Pidgin and Creole Languages 9: 25–50. Lumsden, John. 1995. Aspect and lexical semantics in Haitian Creole. The Linguistic Review 12: 123–142. Lumsden, John. 1999. The role of relexification in creole genesis. Journal of Pidgin and Creole Languages 14: 225–258. Matras, Yaron and Jeanette Sakel. 2007. Investigating the mechanisms of pattern replication in language convergence. Studies in Language 31: 829–865. McWhorter, John H. 2002. The rest of the story: Restoring pidginization to creole genesis theory. Journal of Pidgin and Creole Languages 17: 1–48. McWhorter, John H. 2005. Defining creole. Oxford: Oxford University Press. McWhorter, John H. and Jeff Good. 2012. A grammar of Saramaccan Creole. Berlin: Mouton de Gruyter. Mufwene, Salikoko. 1990. Transfer and the substrate hypothesis in creolistics. Studies in Second Language Acquisition 12: 1–23. Mufwene, Salikoko. 2010. Second language acquisition and the emergence of creoles. Studies in Second Language Acquisition 32: 1–42. Muysken, Pieter. 1981. Halfway between Quechua and Spanish: The case for relexification. In Historicity and variation in creole studies, Arnold R. Highfield and Albert Valdman (eds.), 52– 87. Ann Arbor: Karoma. Muysken, Pieter. 1988. Lexical restructuring and creole genesis. In Beiträge zum 4. Essener Kolloquium über ‘Sprachkontakt, Sprachwandel, Sprachwechsel, Sprachtod’ vom 9.10.–10.10.1987 an der Universität Essen, Norbert Boretzky, Hans den Besten and Thomas Stolz (eds.), 193–210. [Bochum-Essener Beiträge zur Sprachwandelforschung 4]. Bochum: Brockmeyer. Muysken, Pieter. 1997. Media Lengua In Contact languages: a wider perspective, Sarah Thomason (ed.), 365–426. [Creole Language Library 17]. Amsterdam: John Benjamins. Muysken, Pieter. 2001. The origin of creole languages: The perspective of second language learning. In Creolization and contact, Norval Smith and Tonjes Veenstra (eds.), 157–174. Amsterdam: John Benjamins.
7. The development of subordination
|
161
Muysken, Pieter. 2008. Creole studies and multilingualism. In Handbook of Pidgin and Creole languages, John Singler and Silvia Kouwenberg (eds.). London: Blackwell. Muysken, Pieter and Norval Smith. 1990. Question words in pidgin and creole languages. Linguistics 28: 883–903. Noonan, Michael. 2007. Complementation. In Language typology and syntactic description (2nd ed.), vol. II, Complex constructions, Timothy Shopen (ed.), 52–150. Cambridge: Cambridge University Press. Plag, Ingo. 2008a. Creoles as interlanguages: inflectional morphology. Journal of Pidgin and Creole Languages 23: 109–130. Plag, Ingo. 2008b. Creoles as interlanguages: syntactic structures. Journal of Pidgin and Creole Languages 23: 307–328. Plag, Ingo. 2009a. Creoles as interlanguages: phonology. Journal of Pidgin and Creole Languages 24: 121–140. Plag, Ingo. 2009b. Creoles as interlanguages: word-formation, Journal of Pidgin and Creole Languages 24: 339–362. Price, Richard. 1983. First time. Baltimore: The Johns Hopkins University Press. Raposo, Eduardo. 1987. Case theory and Infl-to-Comp: The inflected infinitive in European Portuguese. Linguistic Inquiry 18: 85–109. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of grammar, Liliane Haegeman (ed.), 281–337. Dordrecht: Kluwer. Roberts, Sarah. 1998. The role of diffusion in the genesis of Hawaiian Creole. Language 74: 1–39. Roberts, Sarah. 2000. Nativization and the genesis of Hawaiian Creole. In Language change and language contact in Pidgins and Creoles, John McWhorter (ed.), 257–300. Amsterdam: John Benjamins. Rochette, Anne. 1988. Semantic and Syntactic aspects of Romance sentential complementation. Unpublished Ph.D. dissertation, MIT, Cambridge. Sanchez, Liliana. 2003. Quechua-Spanish bilingualism. Amsterdam: John Benjamins. Sanchez, Liliana. 2006. Bilingual grammars and Creoles: Similarities between functional convergence and morphological elaboration. In L2 acquisition and creole genesis, Claire Lefebvre, Lydia White and Christine Jourdan (eds.), 277–294. Amsterdam: John Benjamins. Satterfield, Teresa. 2005. The Bilingual program: evidence for child bilingualism in the formation of creoles. In ISB4: Proceedings of the 4th International Symposium on bilingualism, James Cohen, Kara T. McAlister, Kellie Rolstad and Jeff MacSwan (eds.), 1070–1090. Somerville, MA: Cascadilla Press. Satterfield, Teresa. 2008. Back to Nature or Nurture: Using computer models in creole genesis. In Language evolution: Cognitive and social factors, Regine Eckhardt, Gerhard Jaeger and Tonjes Veenstra (eds.), 143–176. Berlin: Mouton de Gruyter. Siegel, Jeff. 2000. Substrate influence in Hawai’i Creole English. Language in Society 29: 197–236. Siegel, Jeff. 2008. Pidgins/Creoles and second language acquisition. In Handbook of Pidgin and Creole languages, John Singler and Silvia Kouwenberg (eds.), 189–218. London: Blackwell. Smith, Norval. 1987. The genesis of the creole languages of Surinam. Ph.D. dissertation, University of Amsterdam. Sprouse, Rex A. 2006. Full transfer and relexification: Second language acquisition and creole genesis. In L2 acquisition and creole genesis: Dialogues (Language Acquisition and Language Disorders, 42), Claire Lefebvre, Christine Jourdan and Lydia White (eds.), 169–181. Amsterdam: John Benjamins.
162 | Tonjes Veenstra Valentini, Ada. 1998. Le frasi causali e l’emergere della subordinazione in italiano L2: il caso di due apprendenti cinesi. Linguistica e filologia 8 (Università degli studi di Bergamo): 113–148. Van de Vate, Marleen. 2008. I músu fu woóko taánga: Restructuring in Saamáaka. In Tromsoe Working Papers on Language and Linguistics: Nordlyd 35, special issue on Complex Predicates, Peter Svenonius (ed.), 1–28. CASTL: Tromsoe. van den Berg, Margot and Jacques Arends. 2004. A complement to complementation in early Sranan and Saramaccan: logophoricity. Paper presented at the meeting of the Society for Pidgin and Creole Linguistics, 8–9 January 2004, Boston. Veenstra, Tonjes. 1996. Serial verbs in Saramaccan: Predication and creole genesis. Dordrecht: ICG Printing. Veenstra, Tonjes. 2003. What verbal morphology can tell us about creole genesis: the case of French-related creoles. In The phonology and morphology of creole languages. Ingo Plag and Sabine Lappe (eds.), 293–314. Tübingen: Niemeyer. Veenstra, Tonjes. 2004. Distributed transfer. Presentation at the Workshop on Linguistic Borrowing. John F. Kennedy Institute, FU Berlin, July 2004. Veenstra, Tonjes. 2008a. Pidgin/Creole genesis: The impact of the language bioprogram hypothesis. In Handbook of Pidgin and Creole languages, John Singler and Silvia Kouwenberg (eds.), 219–214. London: Blackwell. Veenstra, Tonjes. 2008b. Presentation at ZAS internal colloquium, Sauen. Wijnen, Beppy and Mervyn C. Alleyne. 1987. A note on FU in Saramaccan. In Studies in Saramaccan language structure, Mervyn C. Alleyne (ed.), 41–51. Amsterdam/Kingston: University of Amsterdam/University of the West Indies. Winford, Donald. 2003. An introduction to contact linguistics. London: Blackwell.
Tom Roeper
8 Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition 8.1 Introduction: The acquisition challenge The challenge in acquisition theory, as it has always been, is to articulate how a child simultaneously evaluates expressed sentences, contextual information, and UG parametric options. In Chomsky’s terms (1976), what is “the triggering experience” for the child to instantiate UG and make parametric choices.¹ The emergence of a theory of interfaces promises greater precision in modeling how and where these factors coincide to produce new triggers that advance the grammar. We will first outline an efficient syntactic theory, with simple default principles. A minimal hypothesis – not necessarily correct – is that the acquisition device may be most efficient if secondary factors are mapped onto the representation of cyclic movement after, not before, movement chains are established within the syntax module. We are well aware that auxiliary factors may be present: information structure and implicatures could in principle participate in the triggering mechanism. Nonetheless, they must each undergo their own language-specific acquisition. Whether other modules are part of a trigger for some structure or must be triggered by it is an important question about the nature of the acquisition path. If we can provide an explanatory account without engaging information structure and implicatures in the triggering system for syntax, then we count it as a success in seeking acquisition efficiency. Our account leans only on the interface notion that overt Tense entails propositions.
8.1.1 Interfaces and restrictiveness A theory of interfaces threatens to produce the kind of explosion of grammar options that the “autonomy of syntax” was meant to avoid (see Chomsky 1965). It cannot be that the child collects a whole array of factors, gives them all equal voice, and somehow juggles them all at once to make primary syntactic decisions. Holmberg (2010) argues masterfully for the relevance of numerous factors (e.g. null subject, topicalization, rich inflection) in parametric decisions. Yet that stunning array can lead again
1 See Roeper (1982) for an early model that incorporates context and more extensive discussion of the role of pragmatic compatibility as an ingredient in triggering. In particular, the well-known fact that children learn irreversible passives (milk was drunk) before reversible passives (John was hit) follows if the child uses plausible interface pragmatic assumptions as a part of the trigger for movement in passive.
164 | Tom Roeper to the problem of a geometric increase in the complexity of the decision procedure. Their relevance does not reveal which are triggers and which are consequences of the decisions. An efficient mechanism will operate with a small set of input data variations where some factors have “epistemic priority” (Chomsky 1965), which is to say are independently recognizeable by the child before a parameter is set. For instance, agency and intentionality have independent cognitive representations or, more relevant to the argument below, a situationally recognized completed event is relevant to the projection of the grammatical notion of factivity, but it is not the same thing. In the spirit of minimalism applied to the acquisition problem, we will suggest a very simple syntactic procedure, coupled with default representations, with Internal Merge, and with the capacity to maintain more than one grammar (Multiple Grammars) at the heart of the process. As we elaborate below, it follows a general economic principle of ‘Avoid Phase’.² In brief, the Strong Minimalist Thesis³ guarantees periodic ‘Transfer’ of syntax to a semantic interface interpretation. If periodic Transfer is psychologically costly – an interface hypothesis in itself – then the child could limit the number of Transfers by limiting the set of phases, hence would Avoid Phase. If interpretation and Transfer must occur at phase boundaries, which engages further modules, then the child will maintain a more economical representation by initially avoiding them, and by positing as few phases as possible.
8.1.2 The cacophony of experience and interface logic The first factor to grasp is that children are constantly confronted with sentences that exceed their grammatical capacities. The classic view is that they change their grammar to capture each new sentence. The classic view understates a complex process with many steps. If there is no simple step evident – addition of a rule or a node – then a child must somehow readjust what he hears to fit a form to which he can respond. That is, he must use his current syntax, set of UG defaults, and best-guess inferences to arrive at a meaning that meets the conversational demand for a response.
2 Other theories capture the concept of phase differently. See Bošković (to appear) for discussion of dynamic approaches to phases. While the concept may be in flux, our acquisition approach provides another perspective where we see the consequences of assuming that interpretation is periodic in a derivation whatever theory one adopts. 3 “[T]here are Transfer operations: one hands the SO already constructed to the phonological component, which maps it to the SM [sensory-motor] interface (“Spell-Out”); the other hands SO to the semantic component, which maps it to the C-I [conceptual-intentional] interface. Call these SOs phases. Thus SMT entails that computation of expressions must be restricted to a single cyclic/compositional process with phases. In the best case, the phases will be the same for both Transfer operations.” (Chomsky 2008: 142)
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
|
165
In fact, where early communication is dominated by context-bound suppositions, the grammar may be left far behind. Eventually when syntax, semantics, and pragmatics map cleanly onto the same representation, grammar can be revised. If not, the child’s choices may reveal quite clearly the non-target grammar in current use, which in turn points at default UG assumptions that provide direct insights into which principles are fundamental and work as pivots in the growth of grammar. Our view of interfaces entails the notion that interpretation leads to distinct meanings for each kind of syntactic structure, a view implicit in much of semantics as well. This approach provides a critical form of syntax-semantics mapping efficiency for the child in acquiring grammar. In particular, we argue that the Tense-proposition connection makes infinitivals distinct. Elsewhere, where infinitives carry propositional implicatures, the notion of proposition is a kind of entailment and not an assertion.⁴ A large challenge for acquisition research is to characterize that acquisition path with precision. Once again the child confronts a world of objects, eye-gaze, human relations, and events (something is happening), inferences about human intentions, all outside of language which he must coordinate and map onto a phonological stream with intonational contours. It is very unclear how this mapping should occur, but there are two broad approaches which need to be straightened out before our very narrow claims come into focus.
8.1.2.1 3rd factor effects The first is the proposal that some phenomena that had been assimilated to UG should be represented as 3rd factor effects, outside of grammar. The concept of interface, however, we suggest, plays a critical role here, in determining exactly where 3rd factor effects may function. In other words, a critical role for UG is in the organization of 3rd factor effects.⁵ For instance, the notion of Noun and Verb have obviously external links to action and thing although each can translate into the other (clothes ⇒ clothed (N ⇒ V), and a drink = V ⇒ N). However, the fact that verb phrases dominate noun phrases (verb-object) and not the reverse is what UG must do (see Roeper 1982 for
4 In contrast to the view of a reviewer, we do not think that perception verbs entail propositions, as in John heard Bill sing. This sentence does not entail awareness of the proposition: John heard that Bill sang. He could have heard him sing without knowing who he was. See Roeper (2011) for discussion. 5 Thus we expect many principles of UG to contain connections elsewhere and in that sense are not “outside of grammar.” See Roeper (to appear) for a more extensive discussion of these interfaces, with discussion of Operator Stages in acquisition that lead to tense-doubling, negation-doubling, quantifier-doubling, and complementizer-doubling. See Trotzke et al. (2013) for comparable adult connections to interfaces and doubling phenomena.
166 | Tom Roeper discussion).⁶ Such connections I have labelled ‘Strict Interfaces’ in several papers to underline the notion that UG notions of innateness apply.⁷ Chomsky has used the term interface as well to capture the notion that interfaces achieve a “perfect” connection between the conceptual-intentional system and grammar. A second notion of interface – critical to our analysis – is the interface between components of grammar which Chomsky has labelled the Strong Minimalist Thesis and pivots around the notion that UG specifies phases and phase edges which are specified points where semantics and then phonology apply.⁸ These are interfaces within grammar. The critical ones for our discussion – which we argue in detail below – are the links between the clause boundaries (phase edges) Tense and propositionality. The acquisition path for the UG-link ‘Tense entails propositionality’ places this form of propositionality in a different domain from “actuality implicatures” (see below) which derive propositionality for infinitives as in sentences like John managed to climb the mountain, where the verb entails that he actually climbed the mountain. Therefore the propositional meaning that we associate strictly with the Tense+ proposition link does not cover every kind of proposition. This is a good feature – it fits the notion that there is a tight relation between syntactic choice and semantic interpretation.⁹
6 Another example (raised by Holmberg 2010) is Cinque’s claim that adjective order is determined by external factors (from abstract to specific) which is not a property of grammar. By comparison, however, possessive order can be reversed in different languages: i. John’s friend’s father’s dog’s bone ii. the bone of the dog of the father of the friend of John provide opposite orders, both of which are acceptable, but speakers of Romance find the pre-nominal order difficult to process, while the English speakers find the post-nominal form difficult. Why is the order not fixed the same way by a 3rd factor for both adjectives and possessives? One signal role for innate interfaces is to determine exactly where the 3rd factors apply. 7 See Roeper (2011, to appear) for discussion of Strict Interfaces. Thanks to a reviewer for noting that these uses of the term “interface” need to be clarified. 8 They are obviously not the only points where semantics applies. For instance, lexical heads, i.e. words, are semantically interpretable outside of sentences. One could imagine a grammar where even words are uninterpretable unless embedded in functional structure. 9 Schulz (2003) shows that children do not initially discriminate infinitives and tensed clauses properly. They allow these two sentences to mean the same thing: i. John forgot to buy eggs. ii. John forgot that he bought eggs. This means that they not arrived at a Strict Interface which the acquisition system seeks, and therefore have allowed a general inference to play a role.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
|
167
8.1.2.2 The left-periphery challenge Theoretical work on the left periphery in syntax leads straight to an acquisition challenge: how and where does the child identify and subdivide the CP domain? While there are intriguing suggestions about whether very early syntax involves a CP-less stage (Rizzi 1997), there are important questions about whether, where, and how subdivision of a CP can occur, particularly in complement environments. Wurmbrand (2001, 2012 and references¹⁰) has provided extensive argumentation, linked to the history of claims about clause-union and restructuring, that infinitivals across many languages may or may not possess a CP. Economy of representation immediately suggests that a child should start without one. Our specific goal in this essay is to explain first why 4–5 year old children answer the what in (1a, b) instead of the yes/no answer which inversion ostensibly seeks, and second, why they will give wide-scope interpretation to sentences like (a), but narrow scope to (b): (1)
John told Bill to eat a sandwich, but Bill ate a banana. a.
Did John say what to do [wide scope (what say-do) = what did he say to do, not what he actually did = eat a sandwich]
but for: b.
Did John say what he did [narrow scope = (say what did) = what he actually did = eat a banana = (what)1 did he say what1 he did)]
In syntactic terms, we will argue that the child for the first infinitival case does not tolerate the presence of a wh-word in a medial pre-IP position. He forces it to move to the initial position where there must be a CP. The CP position is occupied by something, generally regarded an empty operator, a position where the wh- can now be interpreted. The child avoids making the infinitive a phase, by not projecting a CP for it, and thereby minimizes Transfer. Second – the critical fact – he chooses to give it scope over say in (a) but not (b). Forced movement is common in grammar, not an exclusively acquisition move. Chomsky (2013) has pointed out that a similar forced movement is required just from the medial CP when it occurs with “test” questions. The middle sentence is excluded: (2)
a. In which city does John think __Bill played basketball. b. *John thinks in which city Bill played basketball. c. John thinks Bill played basketball in which city?
10 See the Relational Grammar tradition for discussion.
168 | Tom Roeper The wh-question can be initial or in situ, but when the wh-word moves to the middle CP, then it must move further to the root CP. Why? Chomsky explains this as a problem with the Labelling Algorithm. The Labelling approach in itself, like the split CP for the left periphery (with language variation), complicates the acquisition problem unless we somehow enrich the model with interface information. For children, we will argue that forced movement reflects a simply stated, but even deeper UG bias against phases in infinitives. In both (1a) and (2b), further movement restores grammaticality.
8.1.3 Covert movement and rhetorical questions A relevant question is how far such operations occur in adult grammar and English in particular. We present data from Oiry (2008, 2011) and Mathieu (1999) in behalf of the view that French contains such expressions quite readily. While “rhetorical” questions are a traditional grammatical concept that captures expressions like (3) where 2 pm is natural and #yes is infelicitous. (3)
Do you know what time it is?
The answer is equal to a covert movement of the wh-word to an initial position [what do you know the time is] which we call an empty operator. What we do not know is how widespread in daily life such constructions are, and how much variation across languages exist. It is certainly the case that in an emergency if you ask (a) or (b): (4)
a. b.
Do you know where the fire extinguisher is? Do you know where to find a fire-extinguisher?
a most probable answer would be behind the door and not #yes or even #yes, behind the door. It is plausible that whenever the fact that YOU know something is irrelevant, then answering the yes/no property would be dropped, and this might be particularly true in interactions with children. Note that this possibility cannot predict a distinction between infinitives (1b) and tensed clauses (1a).
8.2 Initial theoretical background We begin with a broad background that points to where basic principles of default UG representations and operations come into play. We will argue that those principles of UG are at work, which after this overview we will introduce in greater depth. (5)
1.
The Strong Minimalist Thesis (SMT): each phase is an interface domain and must receive an interpretation.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
| 169
When does interpretation occur? Words have a meaning before they are integrated into sentences. The SMT says that while some meanings are computed for the whole sentence, in general meaning is computed at each PHASE, where major constituents like CP, DP, and PP occur. 2.
Transfer: Each phase has an interface connection that engages Transfer to phonology and interpretation.
The notion of Transfer carries out this action: moving information to a new semantic representation, which involves lexical and LF meaning, which in turn is united with independent pragmatics and situational inference.¹¹ 3.
Avoid Phase: a general economy principle, which minimizes Transfer and interface computations.¹²
This concept is the inverse of Transfer, stating in effect, that the frequency of Transfer should be limited.¹³ 4.
Default empty operators: empty operators appear in the typology of Simple Partial Movement¹⁴ and serve many purposes. (Fanselow 2006; Oiry and Demirdache 2006; Mathieu 1999)
11 It is important to realize that ‘interpret each Phase’ is both an intuitive notion and a technical one. The intuitive notion is partly misleading because much of the interpretation is in fact delayed. First the extraction of a wh-word (either to the first CP or further to a higher DP) means that the clause is interpreted with a trace, which means that there is a hole in the interpretation to be interpreted when the extracted wh-word is identified (N. Chomsky pc), see discussion below. Other features in the clause, including intonation, the scope of particles (J. Bayer pc) and pragmatic factors are interpreted at the final phase as well. 12 While CP, vP, and VP are phases, proposals exist for the claim that TP, DP, and PP could be phases. Our acquisition approach advances the general hypothesis that economy leads to Avoid Phase, but it is not yet clear if other domains arise where the impact of interpretation and Transfer make a difference. 13 A reviewer correctly points out that delaying Transfer should therefore entail that other memory demands are increased. This observation must eventually be addressed in terms of the mental demands that the representations behind each kind of memory entail, a question that cannot be formulated until the concept of memory is more tightly linked to particular representations. Why we can remember the content of sentences for years, but forget the syntax in milliseconds is not understood, but clearly some forms of memory representation have greater stability than others. Visual memory, which can be long-lasting, seems inherently more informationally complex than verbal memory. Hence comparative metrics will be very obscure. 14 There are many forms of Partial Movement. Fanselow (2006) defines Simple Partial Movement as “The construction exemplified in Bahasa Indonesian [. . . ]: the structure involves the partial displacement of the wh-phrase only. This construction will be called simple partial movement (SPM).” It captures overt movement to the middle position, which is interpreted sentence-initially as in Indonesian: Bill tahu siapa yang Tom cintai? Bill knows who FOC Tom loves ‘Who does Bill know that Tom loves?’
170 | Tom Roeper Empty operators are needed to explain movement to invisible positions – in our case unspoken wh-positions in the initial clause.
8.2.1 CP background Where does a child start? An obvious possibility, which we will pursue, is that they begin without a CP in the infinitive. That is, an infinitive participates in clause union without a second CP2. This is again precisely what has been extensively defended in a long literature: (6)
CP1 [VP [IP VP ]] =/= CP1 [VP [CP2 [IP VP] start to play
Now the question arises: how, when, and where does a child build a CP2. The issue becomes acute with infinitival clauses in the input which require one like: (7)
John knows who to see.
where the infinitive includes a wh-question. Such clauses are not found in many languages, which simply exclude wh-infinitives, like Dutch, German, and Germanic languages generally, except for English. Interestingly, Dutch allows was zu tun and German allows it in constructions like Ich weiß, was zu tun ist (Google example). The fact that such examples may not instantly generalize fits our argument. It means that one wh-type example that calls for a CP must not permanently introduce the CP category in that position in infinitives. There is, however, widespread disapproval of the isolated form was zu tun, which is also consistent with the notion that language variation or change could be involved.
8.2.2 Triggers and diversity of input A straightforward hypothesis, generally assumed by both intuition-based and acquisition theoreticians, is that a single example (or perhaps just a few) can rewrite a grammatical representation. We argue, however, that a rather minute acquisition process is engaged. Because some verbs “restructure” (Wurmbrand 2012) while others do not, we are led to the idea that English retains both +CP and −CP infinitives that are subcategorized by higher verbs. Second, inside the CP-infinitive, there could be a stagewise evolution of possible wh-expressions. Thus the language may have two representations for infinitives. In principle, then, there are two grammar types available, the basis for claiming that all languages have Multiple Grammars (Kroch, Taylor and Ringe 1997; Roeper 1999; Yang 2002; Chomsky 2005; Amaral and Roeper (to appear); and others). But first we need a richer theoretical background.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
| 171
8.2.3 Propositionality and the Strong Minimalist Thesis The clearest case of a phase is a tensed clause that entails a proposition, which follows from this UG assumption: (8)
Tense: Tense entails a proposition with true/false value.
Thus in (9) the two tenses imply two propositions, each headed by a CP which is subject to interpretation (9)
=/= [CP John hopes [CP that Bill is here Phase Phase
Now what is the status of infinitives? They arguably, but not necessarily, lack Tense, but they may have material which appears in a CP position: (10)
a. b.
John wants to swim. John knows who to see.
Therefore, if they are uniform, we must assume that all infinitives have a CP and should therefore be the objects of phase interpretation. If they are not uniform in UG, then they may escape being objects of interpretation.
8.2.4 Acquisition background: Tense, propositions, and infinitives Does a child assume that every sentence belongs in a true/false framework? What happens when a child moves from early small clauses (11a) to tensed clauses (11b): (11)
a. b.
he big he is big
Potts and Roeper (2006) discuss expressives extensively and argue that we can consider early child utterances (11a) as expressives without the property of truthfunctional assertions (Klein 2006). At the moment Tense is introduced, the child generates a notion of truth, hence potential falsity, at the interpretation interface.¹⁵ 8.2.5 First clue: Long-distance movement and infinitives There is a clue in early experimentation which has remained unexplained for twenty years, namely that very young children seem to prefer LD movement in infinitives. In
15 Jill de Villiers (pc) has anecdotal evidence that the emergence of the first Tense markers in is in it is is followed within a day by is it, which neatly supports the view that propositionality entails a true/false contrast which questions then exhibit.
172 | Tom Roeper tensed clauses, we find that there is an even split between long and short distance movement (see de Villiers and Roeper 1994, 2012): (12)
[he said, in the evening that he fell in the afternoon] When1 did he say __1 he fell __1 48% 50%
Surprisingly with virtually identical kinds of ambiguity, we find that younger children prefer LD movement with infinitives: (13)
[He asked his Dad if he could call his grandmother] Who did he ask __to call __ 4yr olds 50% 50% 3yr olds: 30% 70%
This puzzling anomaly may have a deeper explanation if children shift the syntactic representation of infinitives at 4 yrs, since by age 3.5 we find Adam using expressions with a wh-infinitive that are not possible in many languages, notably favoring how: (14)
adam14.cha:*CHI: show # how to play. adam22.cha:*CHI: I can’t how to put it back together. adam22.cha:*CHI: how to make it. adam23.cha:*CHI: how to get here? adam25.cha:*CHI: I teach him how to walk on dat [: that]. adam25.cha:*CHI: you can know how to play yourself. adam27.cha:*CHI: I don’t know how to take it out? CHILDES
The next question then is what governs the interpretation of medial wh-expressions in infinitives, since in these cases there is little distinction between interpreting how before the upper tensed verb and the lower infinitive. Our experiments create the critical distinction.
8.3 Simple Partial Movement A large range of languages allow medial questions to be directly answered for both tensed and infinitival clauses:
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
(15)
| 173
Ancash Quechua (Cole and Hermon 2000) (Qam) kreinki imata Maria munanqanta José rantinanta? (you) think what M. want J. buy ‘What do you think Maria wants José to buy?’ Bahasa Indonesia (Saddy 1991) Bill tahu siapa yang Tom cintai? Bill knows who FOC Tom loves ‘Who does Bill know that Tom loves?’ Kitharaka (Muriungi 2004) U-ri-thugania ati n-uu John 2ndSG-Tense-think that Foc-who J. a-ring-ir-e-t? Simple-past-beat-Tense-Final-Vowel ‘Who do you think that John beat?’ Kikuyu (Sabel 2000) Ó-𝛾w-eciìri-á Ngo𝛾e a-ú𝛾-írε nóo ate o-on-írε Kaanakε ? you-think N. said FP-who that saw Kanake ‘Who do you think Ngo𝛾e said that Kaanake saw?’¹⁶
Other languages, perhaps all, allow this construction in a more reduced way, such as French, where it is present in the adult language (16) and more evident in child language (16c-f), as Oiry (2008) shows:¹⁷ These questions entail the presence of a discourse where the wh-word reflects the question-under-discussion, but they are not simply echo questions: (16)
a.
Tu crois que Jean a acheté quel livre? you believe that Jean has bought which book ‘which book you believe that Jean has bought’ ⇒ wide scope [wh-believe-bought]
b.
Tu penses que Jean va épouser qui finalement? you think that Jean will marry who in-the-end
French child data with empty operators present exactly the same grammatically acceptable phenomenon from roughly 5yr olds where a question is asked but an initial wh-word is not expressed:
16 In addition: Buli (Ferreira and Ko 2000), Slave (see Rice 1989; Basilico 1998), and other Athabascan languages (Babine-Witsuwit’en, Denham 2000), and in Ancash Quechua (Cole 1982b). 17 See references in Oiry (2008) and Starke (2001).
174 | Tom Roeper c.
Tu penses que j’habite où? you think that I-live where ‘Where do you think I live?’
d.
Tu penses qu’il y a quoi dans la malle? you think that-it there is what in the trunk ‘What do you think is in the trunk?’
e.
Tu penses que quelle fille est cachée en dessous de cette you think that which girl is hidden in below of that boîte? box ‘Which girl do you think is hidden in the box?’
f.
Tu penses que le policier fait quoi? you think that the policeman does what ‘What do you think the policeman is doing?’
Fanselow (2006) reports – which is a crucially relevant fact for acquisition – that Iraqi Arabic allows this construction restricted to infinitive complements. (17)
a. b. c. d.
Mona raadat tijbir Su’ad tisa’ad meno? Mona wanted to force Su’ad to help who? Mona raadat tijbir Su’ad meno tisa’ad? Mona raadat meno tijbir Su’ad tisa’ad? Meno Mona raadat tijbir Su’ad tisa’ad?
It seems that something forces the wh-word to escape just from the infinitive and therefore require wide scope. The natural argument under the SMT is that the infinitive is not a phase and the wh-word cannot be interpreted unless attracted to a CP which is only found in the matrix clause. Its appearance in the medial position could then be by a form of adjunction to IP, possibly in a TopicPhrase.
8.3.1 Structure of Simple Partial Movement (SPM) Fanselow (2006) defines SPM as: (18)
“the structure involves the partial displacement of the wh-phrase only.”
And he explains that there is a three-way correlation between languages with wh-insitu, empty scope-markers, and LD movement:
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
(19)
a. b. c.
|
175
wh-in-situ =>[ you see John did what] empty operators and Partial Movement ⇒ [OP you see what John did] Partial Movement and LD movement ⇒ [What did you see what John did t]
In other words, these three possibilities are linked: (20)
OP1 you see what1 John did (what1) [CP1 OP. . . [CP2 (what) . . . (wh-)
It is exactly where there is robust cross-linguistic typological evidence that we would expect UG default representations.
8.4 Historical and typological perspectives As mentioned, Modern German prohibits wh+infinitive. It is radically ungrammatical to say: (21)
Hans weiß, wohin zu gehen. ‘John knows where to go.’
Dutch allows an idiomatic form wat te (what to). The evolution of wh-infinitive in English seems to be slow. Roger Higgins (pc) reports that the construction first appears in early Middle English, in texts dated to about 1200–1350 – there is no sign of it in Old English. He reports that initially it seems to be limited to how and what. Pintzuk (pc) reports that there were 21 instances of wh+infinitive, in contrast to 2161 wh+finite forms in Medieval English, and 13 were what and a few whether, but no signs of others. Early Modern English jumped to 368 in comparison to 4174 finite forms, or roughly 10%.
8.4.1 Acquisition parallel The acquisition of wh+complements in general appears to follow a one-by-one path as well. De Villiers (1991) shows precisely that with finite wh-complements the use of wh-words emerges one by one over several years and the appearance of each one in a subordinate context parallels their appearance with subject-verb inversion in matrix clauses. That is, initially children exhibit no inversion with an adjoined wh-word, at first what or where (22)
a.
where John go
176 | Tom Roeper Then one-by-one, wh-words appear in complements: what, where, how, when, and last why. As in: b.
John knows how Bill goes why where
In each instance, inversion in the matrix clause questions co-occurs with their appearance in complement clauses [what can John sing/ask what John can sing]. That fits the view that the CP is introduced in finite complement clauses when the wh-word appears, which then allows the wh-word to appear in the CP in matrix clauses as well. While these various strands of evidence are quite incomplete, they are very compatible with a plausible acquisition path, and therefore deserve attention: (23)
1. 2.
The UG default form allows no wh+infinitive The emergence of wh+infinitive will be gradual and proceed both in terms of particular verbs and particular auxiliaries.
Therefore the child will proceed with essentially two grammar types: both a default CP-less infinitive: (24)
G1: CP NP V IP
and a slowly evolving form, at first linked to particular verbs and particular wh-words that allow a CP: (25)
G2: CP1 NP V [CP2 how IP [CP2 what IP
Possibly G1 disappears and when a full complement of wh-words is present, then all infinitives have a CP. The acquisition and historical evidence both point to the view that acquisition of embedded wh-words will appear one by one. This is compatible with the view that the embedded CP itself must be triggered.¹⁸ Logically, this piecemeal acquisition must be true in order to account for possible grammars whose infinitives allow only one or two wh-words, that is, for certain forms of modern German and Dutch. That is, when the child first recognizes know how they
18 A stronger hypothesis is possible: i) the acquisition path exactly recapitulates the historical path in the emergence of wh-words in embedded contexts. There is no direct evidence for such a claim, but it could be true if there is a hierarchy within wh-words beyond the argument/adjunct distinction. (Thanks to a reviewer for pointing out that the gradual acquisition could still be different from the path of historical evolution.)
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
| 177
should assume it is a highly restricted form that will not generalize,¹⁹ as if German were their language. They must then add wh-words as evidence warrants, until finally any wh-word can appear.
8.5 Long-distance movement in theory and acquisition A crucial question arises from this perspective: How are long-distance wh-questions formed where the what-object of the first phase is interpreted later? It looks as if LD questions contradict the phase-based view (and may explain why the claim did not emerge earlier in linguistic theory): (26)
What did Mary say she bought ⇒ What did Mary say [CP t [IP she bought t] Phase2 Phase1
The answer involves a technical division between phase complements (IP which we treat as equal to TP) that are transferred and phase head (CP), which, as in the traditional CP “escape hatch,” is interpreted in the next higher phase. Thus to give the object scope over the higher verb, it must be moved to the CP, the phase head, which is outside the Tense domain and not immediately transferred. This creates an alternating phase structure. (27)
[Phase-Head [Phase [Phase-Head [Phase]]]]
Transfer takes place every other time Merge applies to yield the following pattern which we illustrate with putative phase categories for CP, TP, and vP: (28)
a.
b.
that John said that Bill said that HEAD Complement HEAD Complement IP IP (Head1 (ComplementTransfer2, (Head3, (ComplementTransfer4, (Head5)...)))) = [C [Phase-trans [T [Phase-trans [vP [Phase-trans]]]]]]
Boeckx (2008) provides evidence from islands, pied-piping, and elsewhere for this pattern. We argue that acquisition data provide subtle evidence on behalf of the SMT where tensed IPs are phases, and infinitives are not (see Roeper 2011 and Arsenijević and Hinzen 2012).
19 Compounds show the same phenomenon in modern English. We have a form knowhow, as in he has the knowhow, but there is no form allowing *he has the knowwhere or knowwhen.
178 | Tom Roeper 8.5.1 Previous acquisition results A large range of evidence gathered over the last 15 years (see de Villiers et al. 2011) show that children interpret LD questions as if the wh-word is in the first phase in tensed clauses. While an adult would answer what she said she bought, the child answers: what she actually bought, as if the what1 were interpreted in the initial IP and not moved into the phase-head escape hatch. The exact syntactic characterization of this acquisition moment is an important goal. De Villiers et al. linked it to the verbspecific subcategorization of the clause. In technical terms, it is the requirement that the lower CP inherits an indirect question feature which serves as a landing site for a wh-word (say what): (29)
a. b.
say CP +Indirect Q ⇒ +Indirect Q what1 did she say [what2 she bought] what1 = what2
In fact, Thornton (1990) found that children produced an extra what in the lower clause of LD questions, providing direct evidence of movement through that position: (30)
What do you think what Cookie Monster likes?
Significantly these were found for tensed complements, not infinitivals. Moreover, Oiry and Demirdache (2006), Oiry (2008, 2011), and Strik (2008) in French found acquisition evidence that producing the repeated wh- correlates with the lower clause interpretation. Thus what2 must be inside the phase, below the escape hatch topmost CP. Therefore there must be a first phase position that undergoes Transfer available to children. Under the view that there is a Split CP (Rizzi 1997), it is possible that a Force Phrase is projected, or a Topic Phrase.²⁰
8.5.2 Representation of movement and reconstruction Current arguments about the representation of movement in general have arrived at essentially the conclusion that “reconstruction” for wh-movement should occur. The Copy Theory of Movement (Nunes and Corver 2007) argues that all movement involves copy and delete. This allows for two kinds of traces (S. Cable, pc), building on work by Rooth (1992), Sauerland and Elbourne (2003), and Miyagawa (2006) who show that typical reconstruction shows scope effects from the landing site (as in wh-movement)
20 In fact, Hinzen and Sheehan (2011) and Haddican et al. (2011) argue for assertive properties in embedded clauses and wh-slifting linked to a ForceP which entails comparable effects for adults.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
|
179
as well as the origin. A second kind, Total Reconstruction, returns a word to its original position and scope with no meaning change, as if the movement occurred only in pronunciation (at Phonetic Form (PF)), yielding the contrast: (31)
a. b.
PF-deletion which entails Radical Reconstruction (RR) Typical (Mature) Reconstruction: PF + LF effects (e.g. scope)
Radical Reconstruction (sometimes called Total Reconstruction) occurs only in one module, Phonetic Form, and therefore is a plausible first stage of acquisition. Miyagawa (2006) comments that, “the [RR] chain completes the paradigm of chain spellouts given in the literature. It is easy to see that radical reconstruction fills a logical gap in the paradigm.” This perspective leads to an important acquisition prediction: the tensed clause will allow RR as a first option because the medial trace is available. The infinitival clause, if no intermediate CP is present, will allow no RR. Therefore, following the SMT, which calls for interpretation phase by phase, we predict: (32)
Children will exhibit an interpretive difference between tensed and infinitival complements. That is:
(33)
Children will allow what to have scope over say only for an infinitive complement (a): a.
What did he say to buy?
because they have not interpreted the complement at the first phase, in comparison to that possibility in (33b): b.
What did he say [(what) he bought]? Phase2 Phase1
Therefore it should be less likely for children to give a “reality” or “actuality” interpretation for the content of infinitives. In other words, under RR there is an interpretive option for reconstruction into a tensed clause – the tensed proposition – that is unavailable for reconstruction into an infinitive if there is no CP present. How does a child deal with novel input? Let us imagine in closer detail exactly what the child could do with an unfamiliar input: (34)
Did John say what to bring [what [IP to bring]
180 | Tom Roeper He has available from default UG the option of projecting an empty operator at the root which may contain a [+Q] feature to be satisfied. There is no phase present in the lower clause because there is no Tense node that requires a phase and an interpretation.²¹ The child can now either project a new node, a CP, or utilize the default CP and an operation of Internal Merge to move the wh-word to the matrix clause. Then, however, it will take scope over both say and bring. This is not required for the finite clause which has the potential of being interpreted because it is a phase, even when it is linked by scope-marking to the empty operator. We turn now to new evidence in behalf of this perspective. This analysis introduces important questions for theory: how do we represent the grammar of wh+infinitive at this stage. For the child, if it is only a comprehension challenge, the movement of the wh to the matrix re-analyzes the input into a grammar that is possible for the child. For the Arabic speaker who can utter such sentences, the representation must be different. Following Chomsky, we have suggested that the movement is like the case where think does not subcategorize a node with a Q-feature (a label) and therefore movement is forced. The Arabic case could be the same, but with movement forced only at LF, and phonologically the wh-word can occupy an unlabelled node.²²
8.5.3 Experiment²³ 8.5.3.1 Design In this study we used the single verb say that allows both types of complement, and two question types: the usual wh-question that requires a long-distance interpretation in adult grammar, and a yes/no question with a medial wh: (35)
Yes/No Tensed: Did Jimmy say what Daddy got for the party? Infinitival: Did Jimmy say what to get for the party?
21 The operator effectively cancels the yes/no property of the inversion, just as it does in wh-questions generally. It is also true that by answering a wh-word it entails the yes-answer to the question of do you know. 22 A reviewer asks if it is odd to argue that an exceptional form in the adult grammar would be a default form in acquisition. In fact, most early deviations from the target grammar in developing grammars represent “overgeneralizations” of what survives in marginal ways in the adult grammar. They are usually simple default rules, where Tense, quantification, or negation are treated first like operators, but later given local representations. 23 The following discussion is essentially excerpted from de Villiers et al. (2012).
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
|
181
Wh-question: Tensed: What did Jimmy say Dad got for the party? Infinitival: What did Jimmy say to get for the party?
8.5.3.2 Hypotheses First we predict that tensed clauses would produce higher rates of incorrect whanswers than infinitival clauses as follows. We had three sub-hypotheses that children would give without respect to the tense/infinitival difference: (36)
a.
b. c.
d. e.
More initial wh-questions answered with respect to the final clause than adults only (i.e. premature transfer to interpretation at the first phase): what did Jimmy say he got/to get = what did Jimmy say what he got/to get More medial wh-answers to the yes/no questions: (what) did he say what he got/to get More medial answers with respect to only the lower clause verb: what he got/what to get = ‘what he actually got’ and not ‘what he said he got.’ Younger children would be more likely to give these kinds of answers than older children. Age shift with respect to tense/infinitival contrast: Younger children: we expect some interpretation of wh with only lower scope [what to buy = what actually bought], because the whole sentence is underanalyzed, we expect it to stop at 3yrs. Older children: With tensed clauses [what he bought = what actually bought], medial wh-interpretation may continue to 4yrs or later because it fits the grammar of Partial Movement.
8.5.3.3 Procedure We created four books, two about “home” and two about “preschool” each of which contained eight stories and questions, counterbalanced by type across children. Each child saw two books, totaling sixteen stories/questions. In each case, there is a general set-up, then a speech act, then a counter-action. Keeping this order of events made the comparison across clause types more controlled. Figure 8.1 shows one such story for a nonfinite complement. (37)
A.
Dad asks Jimmy, “What kind of games would you and your friends like for the Halloween party?” Jimmy tells Dad, “Oh! Pick some pumpkins to paint, please!” But the store is having a special sale on all kinds of balloons, so Dad buys them instead of pumpkins.
182 | Tom Roeper
Fig. 8.1. Nonfinite complement story.
a. Yes/no question: Did Jimmy say what to get for the party? b. Wh-question: What did Jimmy say to get for the party? Note: a child would only get one type of question per story (37)
B.
Today, it is Jimmy’s turn to feed the class pet rabbit. Another student asks the teacher, “Can I feed the rabbit today?” The teacher tells him, “I’m sorry, but Jimmy already gave the rabbit a carrot for the day.” But before when the teacher was outside, Jimmy could not find the carrots. So instead, Jimmy gave the rabbit a piece of apple from his lunchbox. a. Yes/no question: Did the teacher say what Jimmy gave the rabbit?²⁴ b. Wh-question: What did the teacher say Jimmy gave the rabbit?
Fig. 8.2. Finite complement story.
24 Note that adults can answer two ways: yes or no.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
|
183
8.5.3.4 Participants Participants were 32 preschool children, boys and girls, aged 3–5 years, mean age 4;3, from 2 local preschools. They were tested individually on two occasions. For the purposes of age analysis we divided them into two age groups: 12 three year olds and 20 four year olds.
8.5.3.5 Coding Wh-questions: We coded the answers by whether the answer was correct (in scope of two verbs), or reflected “reality.” (38)
a.
b. c.
Tensed version: What did the teacher say Jimmy gave the rabbit? (correct = carrots, reality = apple) Infinitival version: What did the teacher say to give the rabbit? (correct = carrots, reality = apple) Yes/no questions: Did the teacher say what to give/what Jimmy gave?
We coded first whether the child answered the yes/no question or the medial whquestion. The yes/no answers were YES or NO. The medial wh-answers were coded by their scope: (39)
2-verb (say + fed): Did the teacher say what Jimmy fed the rabbit? = carrot Or “reality” (what was fed): Did the teacher say what Jimmy fed the rabbit? = apple
8.5.3.6 Results We predicted that tensed clauses would produce higher rates of incorrect wh-answers than infinitival clauses. In particular, we expected more initial wh-questions answered with respect to the first phase, i.e. to the final clause only. The results of repeated measures ANOVA confirm this prediction, with a highly significant difference between tensed and infinitival complements in these errors (F (1,30) = 12.7, p < .001). This difference is seen in Figure 8.3. As the graph shows, 4 year olds were significantly better than 3 year olds at all types of questions, but the tense-infinitival difference did not interact with age. The second sub-hypothesis was: B) that partial movement (medial wh-answers) would be more common for tensed clauses than infinitivals.
184 | Tom Roeper
Fig. 8.3. Initial wh-question answers with respect to first phase only.
This prediction was not confirmed, in fact there were more medial wh-answers with infinitives than for tensed clauses just for four year olds, though this falls short of significance (p = .07) (see Figure 8.4).
Fig. 8.4. Partial movement answers.
The third sub-hypothesis predicted: C) more medial answers interpreted only in the lower clause for tensed versus infinitival complements. That prediction is confirmed (F (1,30) = 12.2, p = .002). When 3 year olds answered the medial wh instead of the yes/no question, for tensed complements:²⁵
25 See de Villiers (1991) for further experimental and statistical details.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
| 185
Fig. 8.5. Proportion of partial movement answers to first phase only.
(40)
47% of their answers were lower verb only, and for infinitival complements, only 9% were (see Figure 8.5).
8.5.4 Further recent evidence for partial movement Grohe, Schulz, and Müller (2011) found particularly explicit evidence that in German, partial movement, which is optional in adult German, is preferred by 4-yr-old German children in elicited production over long-distance movement. Children were presented with the following story: Introduction: Toby and Anna spot two children. The children are holding some forks in their hands and they are hungry already. Test sentence: Wo denkst du, dass die Kinder heute essen? ’Where do you think that the children eat today?’ Children were asked to simply imitate one of three questions: (41)
Scope marker: Was denkst du wo die Kinder heute essen? Copying: Wo denkst du wo die Kinder essen? Long-distance (LD): Wo denkst du dass die Kinder essen?
24% of the children converted the LD question into either copying or a scope-marker form. None of the children converted scope-marker or copying sentences into LD sentences. Moreover, only 59% imitated the copy sentences exactly and 48% imitated the
186 | Tom Roeper scope-marking sentences exactly. But only 24% were able to imitate the LD sentences exactly. The evidence shows that children in the 4–6 range (as in some dialects) clearly disprefer LD movement and favor various forms of partial movement.²⁶
8.6 Theoretical interpretation Our analysis thus far suggests that partial-movement structures are a UG default that appears in child language. We have argued that there is ample evidence for an invisible operator in the scope-marking position to which a child could move an element. The system involves two representations: (42)
a.
Default representation: CP [ ø NP V [wh [IP]]
and a second [+CP] representation which the child builds in a stepwise fashion. b.
Two phase: CP ø NP VP [CP2 wh [IP Tense V]]
In (42a), movement of the wh to ø forces a wide-scope reading, while in (42b), movement of wh to ø can be “reconstructed” into its original position as if had not moved (or moved only at PF). Representation (b) is initially linked to: (43)
Lexical class of verbs and particular wh-words: how, what
At some point, the language keeps both forms, or it collapses infinitives into a CP IP structure.
8.6.1 Intermodular complexity We have deliberately left out a range of other factors – born of clearly relevant theories – to which we now turn. They could interact with our simple representations and participate in the triggering mechanism prior to the point where these representations are established. However, we would like to promulgate the idea that they may not be a part of the triggering mechanism, but rather a consequence. We take the sheer simplicity of these representations as an argument that they reflect a core syntactic mechanism which makes the system efficient. In other words, the acquisition system keeps other factors
26 J. Bayer (pc) based on extensive work says that this is true of adults as well, which is consistent with our argument here. It has classically been regarded as a dialect feature and, one might guess, is ‘corrected’ in written German. The sociolinguistic status of the construction, therefore, is a topic of great interest. See work by Brandner (2000) and dialect studies from the Meertens Institute in Amsterdam for background.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
|
187
systematically outside the system until the first representations – like a skeleton inside a body – have been represented. They are then efficiently extended to include further factors. This should eventually be elevated to an acquisition concept.
8.6.2 Which factors are secondary: The problem of epistemological priority It is very plausible that many other factors participate in the child’s linguistic experience from the outset. We will briefly point to these dimensions without a full explanation in order to give the reader a sense of possible intervening elements: (44)
1. 2.
3.
Tense: does the child use invisible Tense-binding to interpret infinitives? Focus and wh: wh-words interact with information structure in the determination of which languages allow default operators in their final grammars. Does Focus help determine phase boundaries? Pragmatics and implicatures: children are aware of the pragmatic momentum of situations from an early age. Do they use that information to determine “actuality” implicatures to define phases?
Each of these dimensions must eventually be mapped into the representations that children develop. We will sketch the approaches behind each one and provide an initial argument for why they are not the primary organizing instruments that the child uses, but rather are engaged once certain structures are in place.
8.6.3 Dynamic phase theory Recent work by Wurmbrand (2012), Alexiadou et al. (2012), Bošković (to appear) argues that phases are contextually determined in terms of whether Tense in the lower clause is bound by the upper clause. A hidden Tense-binder contributes to whether a verb is episodic or “propositional” (simplified discussion): (45)
a. b.
John likes to sing (implies multiple episodes) John believes Mary to be smart (proposition)
This source for propositionality is distinct from the overt Tense-proposition connection. As far as I know, none of the critical verbs are in the early vocabulary of children (know to, believe to, manage, enable). Where does the child start? Note that bare verbs in small clauses show no evidence of a CP: (46)
a. *I heard how Bill sing b. I heard Bill sing
188 | Tom Roeper but it does allow apparent LD movement: (47)
How did John hear [ Bill sing t]
Small clauses emerge quite early in child grammar (earliest under 3yrs): (48)
brown Adam @ 2;8.0 [see me put de boot on?] bloom Peter @ 2;9.15 [wan(t) (t)a see me rock on the horse? bloom allison @ 2;10.0 [you help me push the bus.] bloom peter @ 2;1.0 [help me make a happy face # xxx.] 3-verb cases: brown eve @ 2;0.0 [they go see me eat lunch.] brown eve @ 2;1.0 [Fraser go see me have my lunch.]
Here clearly the past tense in the matrix verb applies to the small clause as well although it contains a bare verb. Therefore the early perception verb structures fit our claim that the child begins without a CP in infinitives as well. Nonetheless if infinitives do or do not carry invisible Tense-binding, the question again is what the child uses to decide that the wh+infinitive could constitute a separate phase. We argue that the explicit wh-word could be "epistemologically prior" to Tensebinding, but the topic obviously deserves research.
8.6.4 Focus and default operators An important question is how the child decides whether the target grammar will ultimately retain empty operators. Oiry (2012) developing work by Sabel (2000) argues that Focus and wh-movement not only are both generated on wh-words, but either can be strong or weak, where feature combinations serve to predict the various options blocking empty operators. In general if the features are strong, then wh-words must be overtly moved to the matrix CP. Thus we have languages where all wh-words typically move left to the root CP (Slavic), where none move (Chinese), and where some move (French and arguably English, with tremendously intricate interactions with, for instance, echo questions). In principle, Focus could be the critical factor that determines the phase status of an infinitive. However, while the wh-words are overt, their feature strength is not. Therefore it seems less likely that they will have epistemic priority. On the other hand, children certainly grasp emphatic focal stress at an early point, and many situations require children to project contrast sets. A parent who says: eat peas, no apples! will most likely be understood by a 2.5 year old. On the other hand, children’s intonation patterns, while they carry a great deal of social implications, may not clearly associate particular constituents with focal stress. Thus we find that children will often overstress a variety of constituents and the focus operator appears to affect an overarching sentential intonation pattern:
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
(49)
|
189
I DONT want to PLAY with SUSIE TOMORROW.
This suggests that a focus operator is not differentiating particular constituents or phases. We conclude that focus, while present early in determining information structure, may fail to have epistemic priority because its mapping is imprecise.
8.6.5 Pragmatics and implicatures It is often argued that children derive syntactic knowledge from pragmatic or cognitive information. It is clear that enormously rich inference capacities are required from both parent and child in the interpretation of one and two-word utterances. Do they then play a critical role in structuring the syntax? This question may not have a single or a simple answer. Yet, for the topic under concern, the existence of what are called “actuality” implicatures are fairly sophisticated demands. Current literature argues that implicatures are in general late-acquired, though the empirical evidence remains in flux. It is not clear that implicatures guide syntactic acquisition. Thus a sentence like: (50)
John managed to climb the mountain
implies the proposition that he succeeded in climbing the mountain.²⁷ Such factive properties are certainly an important part of how we construe situations, but it is less evident that they are guiding the syntactic structure of infinitives. Rather they seem to be consequences about decisions with respect to Tense-binding that appear after wh-infinitives like how-to.
8.7 Conclusions We have argued that simple notions of Avoid Phase invite the use of Internal Merge to link to a default operator just when a wh-word is adjoined to a lower IP in order to avoid projecting a CP Phase boundary. We have argued further that the concept of Avoid Phase can explain why whinfinitives receive a wide-scope interpretation while wh-tensed complements receive an interpretation within the first phase. The concept critically relies upon the Strong Minimalist Thesis that dictates that tensed clauses initiate an interface transfer, while the default representation of infinitives, lacking a CP, does not.
27 Hacquard (pc UMass talk) discusses other relevant implicatures like “endorsement” where Bill thinks John is capable can be read as an endorsement of a proposition as in According to Bill, John is capable.
190 | Tom Roeper We have pushed the spirit of minimalism in a new direction: the effort to state very simple principles and rules as the fulcrum upon which acquisition must work, setting phase boundaries. By hypothesis, it is only once phases are set, other modules, involving tense-binding, focus, and implicatures are mapped onto those representations. The result is that we have built a mechanism that uses very abstract principles, honors the necessity of interfaces in both theory and the acquisition process, and thereby predicts unusually subtle acquisition data, forced into view by the theory itself. Arguments where the principles are abstract and the consequences are subtle cover a real developmental process are what we should expect of a theory that seeks to be biologically robust. Acknowledgement: Thanks to Jill de Villiers, Ellen Harrington, and Elizabeth Gadilauskas, Magda Oiry, Susan Pintzuk, Roger Higgins, Michael Clauss, Kyle Johnson, audiences at BUCLD, Konstanz, and the LARC group at UMass, and two anonymous reviewers, and thanks especially to Andreas Trotzke and Josef Bayer for many helpful remarks.
References Alexiadou, Artemis, Elena Anagnostopoulou, and Susi Wurmbrand. 2012. Movement vs. long distance Agree in raising: disappearing phases and feature valuation. NELS Poster. Amaral, Luiz, and Thomas Roeper. to appear. Multiple grammars and second language representation. Target Article. Second Language Research, University of Massachusetts Amherst. Arsenijević, Boban, and Wolfram Hinzen. 2012. On the absence of X-within-X recursion in human grammar. Linguistic Inquiry 43: 423–440. Basilico, David. 1998. Wh-Movement in Iraqi Arabic and Slave. The Linguistic Review 15: 303–339. Boeckx, Cedric. 2008. Bare Syntax. Oxford: Oxford University Press. Bošković, Zeljko. to appear. Now I’m a phase, now I’m not a phase: On the variability of phases with extraction and ellipsis. Linguistic Inquiry. Brandner, Ellen. 2000. Scope marking and clausal typing In Wh-Scope Marking, Uli Lutz, Gereon Müller, and Arnim von Stechow (eds.), 45–75. Amsterdam and Philadelphia: Benjamins. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, Noam. 1976. Reflections on Language. New York: Praeger Publishers. Chomsky, Noam. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22. Chomsky, Noam. 2008. On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero, and Maria L. Zubizarreta (eds.), 133– 166. Cambridge, MA: MIT Press. Chomsky, Noam. 2013. Problems of projection. Lingua 130: 33–49. Cole, Peter. 1982. Subjacency and successive cyclicity: Evidence from Ancash Quechua. Journal of Linguistic Research 2: 35–58. Cole, Peter, and Gabriella Hermon. 2000. Partial wh-movement: Evidence from Malay. In WhScope Marking, Uli Lutz, Gereon Müller, and Arnim von Stechow (eds.), 101–130. Amsterdam and Philadelphia: Benjamins. Denham, Kristin. 2000. Optional Wh-Movement in Babine-Witsuwit’en. Natural Language and Linguistic Theory 18: 199–251.
8. Avoid Phase: How interfaces provide critical triggers for wh-movement in acquisition
| 191
de Villiers, Jill. 1991. Wh-questions. In The Acquisition of wh, Thomas Maxfield, and Bernadette Plunkett (eds.), University of Massachusetts Occasional Papers in Linguistics, 155–173. de Villiers, Jill, and Tom Roeper. 1994. Lexical links and the wh-chain. In Syntactic Theory and First Language Acquisition: Cross Linguistic Perspectives. Vol. 2: Binding, dependencies, and learnability, Barbara Lust, Gabriella Hermon, and Jaklin Kornfilt (eds.), 357–390. Hillsdale, NJ: Erlbaum. de Villiers, Jill, Peter de Villiers, and Tom Roeper. 2011. Wh-questions: Moving beyond the first phase. Lingua 121: 352–366. de Villiers, Jill, Tom Roeper, Ellen Harrington, and Elizabeth Gadilauskas. 2012. Tense and truth in children’s question answering. In Proceedings of the 36th Boston University Conference on Language Development 36, Alia Biller, Esther Chung, and Amelia Kimball (eds.), 152–163. Somerville, MA: Cascadilla Press. Fanselow, Gisbert. 2006. Partial wh-movement. In The Blackwell Companion to Syntax. Vol. 3. Martin Everaert and Henk van Riemsdijk (eds.), 437–492. Oxford: Wiley-Blackwell. Ferreira, Marcelo, and Heejeong Ko. 2003. Questions in Buli. Working Papers on Endangered and Less Familiar Languages 4: 35–44. Grohe, Lydia, Petra Schulz, and Anja Müller. 2011. How children “copy” long-distance structures: The production of complex wh-questions in German. In Proceedings of the 35th Boston University Conference on Language Development 35, Nick Danis, Kate Mesh, and Hyunsuk Sung (eds.), 233–245. Somerville, MA: Cascadilla Press. Haddican, William, Anders Holmberg, Hidekazu Tanaka, and George Tsoulas. 2011. English whslifting as an embedded root phenomenon. Unpublished ms., University of Newcastle. Hinzen, Wolfram, and Michelle Sheehan. 2011. Radicalizing the interface: the grammar of intensionality. Unpublished ms., Durham University. Holmberg, Anders. 2010. Parameters in minimalist syntax: the case of Scandinavian. Theoretical Linguistics 36: 1–48. Klein, Wolfgang. 2006. On finiteness. In Semantics Meets Acquisition, Veerle van Geenhoven (ed.), 245–272. Dordrecht: Kluwer. Kroch, Anthony, Ann Taylor, and Donald Ringe. 1997. The Middle English verb-second constraint: a case study in language contact and language change. In Textual Parameters in Older Languages, Susan C. Herring, Pieter van Reenen, and Lene Schøsler (eds.), 353–392. Amsterdam and Philadelphia: Benjamins. Mathieu, Eric. 1999. French wh-in situ and the intervention effect. UCLA Working Papers in Linguistics 11: 441–472. Miyagawa, Shigeru. 2006. EPP and semantically vacuous scrambling. In The Free Word Order Phenomenon: Its Syntactic Sources and Diversity, Joachim Sabel, and Mamoru Saito (eds.), 181–220. Berlin and New York: Mouton de Gruyter. Muriungi, Peter Kinyua. 2004. Wh-movement in Kitharaka as focus movement. Proceedings of the Workshop on the Syntax, Semantics and Pragmatics of Questions, ESSLLI 16. Nunes, Jairo, and Norbert Corver. 2007. The Copy Theory of Movement. Amsterdam and Philadelphia: Benjamins. Oiry, Magda. 2008, 2011. L’Acquisition des Questions à Longue-Distance par des enfants français. Stratégie à dépendance directe versus indirecte et questions alternatives. Editions Universitaires Européennes. Oiry, Magda. 2012. Input and universals in developing grammars. Unpublished ms., University of Massachusetts. Oiry, Magda, and Hamida Demirdache. 2006. Evidence from L1 acquisition for the syntax of whscope marking in French. In The Acquisition of the Syntax of Romance Languages, Vincent Torrens and Linda Escobar (eds.), 289–315. Amsterdam and Philadelphia: Benjamins.
192 | Tom Roeper Potts, Chris and Thomas Roeper. 2006. The narrowing acquisition path: From expressive small clauses to declarative. In The Syntax of Nonsententials: Multi-Disciplinary Perspectives, Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds.), 183–201. Amsterdam and Philadelphia: Benjamins. Rice, Keren. 1989. A Grammar of Slave. Berlin and New York: Mouton de Gruyter. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of Grammar: Handbook of Generative Syntax, Liliane Haegeman (ed.), 281–331. Dordrecht: Kluwer. Roeper, Tom. 1982. The role of universals in the acquisition of gerunds. In Language Acquisition: The State of the Art, Lila Gleitman and Eric Wanner (eds.), 267–287. Cambridge: Cambridge University Press. Roeper, Tom. 1999. Universal bilingualism. Bilingualism: Language and Cognition 2: 169–186. Roeper, Tom. 2011. The acquisition of recursion: How formalism articulates the acquisition path. Biolinguistics 5: 57–86. Roeper, Tom. to appear. Strict interfaces and intermodular contact. In Proceedings of the Past and Future of UG, Language Science, Wolfram Hinzen (ed.). Rooth, Mats. 1992. A theory of focus interpretation. Natural Language Semantics 1: 75–116. Sabel, Joachim. 2000. Partial wh-movement and the typology of wh-questions. In Wh-Scope Marking, Uli Lutz, Gereon Müller, and Arnim von Stechow (eds.), 409–446. Amsterdam and Philadelphia: Benjamins. Saddy, Doug. 1991. Wh-scope mechanisms in Bahasa Indonesia. In MIT Working Papers in Linguistics 15, Lisa Cheng, and Hamida Demirdache (eds.), 182–218. Cambridge, MA: MIT Press. Sauerland, Uli, and Paul Elbourne. 2003. Total reconstruction. PF-movement and derivational order. Linguistic Inquiry 33: 283–319. Schulz, Petra. 2003. Factivity: Its Nature and Acquisition. Tübingen: Max Niemeyer Verlag. Starke, Michal. 2001. Move Dissolves into Merge: A Theory of Locality. Dissertation, University of Geneva. Strik, Nelleke. 2008. Syntaxe et acquisition de phrases interrogatives en francais et nederlandes. Une etude contrastive. Unpublished dissertation, Paris 8. Thornton, Rosalind. 1990. Adventures in long distance moving: The acquisition of wh-questions. Unpublished dissertation, University of Connecticut. Trotzke, Andreas, Markus Bader, and Lyn Frazier. 2013. Third factors and the performance interface in language design. Biolinguistics 7: 1–34. Wurmbrand, Susi. 2001. Infinitives: Restructuring and Clause Structure. Berlin and New York: Mouton de Gruyter. Wurmbrand, Susi. 2012. Tense and Aspect in English Infinitives. Unpublished ms., University of Connecticut. Yang, Charles. 2002. Knowledge and Learning in Natural Language. Oxford: Oxford University Press.
Misha Becker
9 Learning structures with displaced arguments 9.1 Introduction In typical, average, canonical sentences (sometimes called “basic” sentences), NPs bearing thematic roles are adjacent to their selecting predicate both in the underlying structure and on the surface. For example, in a basic active transitive sentence, the syntactic subject is also the “semantic subject” (usually an agent or experiencer) and is adjacent to the main predicate. (1)
John threw the ball.
(2)
S NP John
VP V
NP
threw
the ball
This paper looks at how children acquire two constructions which involve a more complex alignment of thematic relations; specifically, constructions in which the syntactic subject is not the agent/experiencer of the main (adjacent) predicate, but rather is semantically linked only to the embedded predicate. The two constructions to be investigated are raising-to-subject constructions, as in (3a), and tough-constructions, as in (3b). (3)
a. b.
The climberi seemed ti to be stuck. The girli is easy to see ti .
In derivational syntactic frameworks, the subject in both (3a) and (3b) is said to be derived from some lower part of the structure. Even in non-derivational frameworks (e.g. Lexical Functional Grammar), the subject in these constructions is not local with respect to the predicate that selects it. I will refer to the subjects in these constructions as “displaced” subjects, and the predicates that participate in these constructions as “displacing” predicates. Compared to basic sentences like (1), these sentences are complex: they are biclausal (they contain two main predicates, seem/be easy and be stuck/see), and they involve non-canonical semantic relations between the NP arguments and the main predicate. In derivational syntactic frameworks, they involve an extra derivational step of NP-movement from the lower clause to the higher one, rendering their syntactic structure more complex.
194 | Misha Becker The specific question I will address is how children figure out these complex structures; more precisely, how children determine that the main clause syntactic subject in just these structures is a displaced subject and is not semantically selected by the (local) main clause predicate. One could recast the question as a question about whether a predicate takes an agent (or experiencer) subject. I take the view that this translates into a deeper question about argument structure: non-agent/experiencer subjects are generally not projected as external arguments, and so the external argument position is potentially empty, underlyingly.¹ The question I am investigating, at bottom, is how children determine this kind of underlying representation. This question is certainly not new. C. Chomsky (1969: 6) posed this very question by asking how children acquire the structure of sentences in which “the true grammatical relations which hold among the words in a sentence are not expressed directly in its surface structure.” And N. Chomsky (1965: 22–27) famously noted the same learnability conundrum. But the little research that has been done on the acquisition of these constructions in children (including C. Chomsky’s) has focused on the question of what children know about these constructions in their target language, rather than how they come to know it. Thus, the fundamental question of how these constructions are acquired remains open. The learnability problem results from a superficial overlap between the displacing predicates just discussed and predicates that do select an external argument (their subject is not displaced). This overlap is seen here: (4)
a. b.
The climberi seemed [ti to be stuck.] The climberi pretended [PROi to be stuck.]
(5)
a. b.
The girli is easy [(Op) PROarb to see ti .] The girli is eager [PROi to see e.]
The problem for the learner is that prior to knowing what the predicates seem, pretend and so forth mean, these sentences cannot be unequivocally parsed. Following work in the Syntactic Bootstrapping literature (Gleitman 1990; Gleitman et al. 2005, i.a.), I assume that although some cues to predicate meanings may be available from nonlinguistic sources (i.e. observing events in the world), there is not sufficient sentenceexternal information to pin down the unusual argument structures of these kinds of abstract predicates. Rather, the child primarily infers the meaning and argument structure properties of the predicate based on the structure it occurs in. But this brings us back to our original question: how does the child figure out the structure? To put the learning puzzle in perspective, consider the sentences in (6), which contain predicates with unknown lexical meanings.
1 Here I abstract away from copular constructions (John is a man), in which the subject, to the extent it can be considered an external argument, is a theme.
9. Learning structures with displaced arguments
(6)
a. b.
|
195
The climber gorped to be stuck. Max is daxy to see.
What is the semantic role of the subjects in these sentences, and what are the sentences’ underlying structures? The learning strategy I propose is based on the semantic feature of animacy. Notice that the displacing predicates easily permit inanimate subjects, while the nondisplacing predicates do not. (7)
a. The rock seemed to be stuck. b. #The rock pretended to be stuck.
(8)
a. The house is easy to see. b. #The house is eager to see.
While displacing predicates permit both animate and inanimate subjects, if a reader/ hearer encounters the novel predicates in (6) with inanimate subjects, the subjects are less likely to be construed as the agent/experiencer of the main predicate. That is, one is more likely to construe the subjects in (9) as being something that can be stuck or be seen (it is semantically related to those lower predicates) but not as a “gorper” or something that is “daxy.” (9)
a. b.
The rock gorped to be stuck. The house is daxy to see.
Raising-to-subject and tough-constructions are not the only constructions that involve argument displacement. Passives likewise involve a derived subject (the syntactic subject is not an agent/experiencer and it undergoes NP-movement, as in The balli was thrown ti (by John)), and unaccusative verbs are verbs whose lone argument is a theme rather than an agent (Perlmutter 1978); in derivational frameworks, it is an underlying object which raises to subject position (The giftsi arrived ti at the wedding; Burzio 1986). While I will briefly discuss unaccusative constructions at the end of the paper, the main focus will be on raising-to-subject and tough-constructions (please see Becker 2014a for discussion of the passive). Animacy is something children know about early in life. Within the first year of life, babies discriminate people from objects based on behavioral properties (Woodward et al. 1993; Woodward 1998), and there is evidence that between about 6 and 18 months of age babies can categorize conceptually certain animals, like birds, as distinct from other apparently similar inanimate things like airplanes (Mandler and McDonough 1993). More importantly, by the second year of life they attribute intentional states to people but not machines (Meltzoff 1995). So by the time children start acquiring basic sentence structure, and certainly by the time they start acquiring these more complex sentences, the concept of animacy is available to them.
196 | Misha Becker Moreover, animacy plays a significant role in grammatical hierarchical structure, and the ability of displacing predicates (but not non-displacing predicates) to take inanimate subjects is found cross-linguistically. Languages that require the subject to be more animate than the object in a transitive sentence (e.g. Japanese (Kuno 1973), Jacaltec (Craig 1976), Navajo (Comrie 1989) and Blackfoot (Ritter and Rosen 2010)) nevertheless allow an inanimate subject in certain constructions if it is derived. This deep preference for subjects to be animate in human language was part of the rationale for Pinker’s Semantic Bootstrapping Hypothesis (Pinker 1984). Since subjects are often agents, and agents are (virtually) always animate, an animate NP is likely to be in the subject position; an inanimate NP is more likely to be in the object position. Indeed, there is good evidence that children preferentially associate agency with subjecthood (Fisher et al. 1994; Naigles 1990) and can use that association to begin mastering the word order of their language (Hirsh-Pasek et al. 1988). Part of the goal of my work is to extend the scope of Semantic Bootstrapping by considering what learners might do when the subject is in fact inanimate. Thus, my claim is that not only can an animate NP serve as a cue to canonical subjecthood, but an inanimate subject serves as a cue that the subject is displaced, and therefore that the structure is complex. I will support my claim with two types of data: (1) naturalistic input data (childdirected speech from CHILDES) and (2) controlled experimental input data in simulated learning tasks with both children and adults. Both types of data show that even if inanimate subjects are highly infrequent, learners (adults and children) can use this information to make assumptions about the selectional properties of predicates, and therefore to learn complex, non-canonical sentence structures. Before proceeding, it is worth noting that although I adopt a derivational syntactic framework in this paper (because it is the framework I am most familiar with), this research question could be pursued via alternative approaches. Notably, Kirby (2012) takes a Construction Grammar approach (Goldberg 2003, 2006) to the representation and acquisition of raising and control verbs. Such an account is worth exploring in the future, particularly in connection with tough-adjectives. However, Kirby’s particular analysis will not extend to the data covered here. Her account relies on the claim that the highly frequent verb want is the “prototype” verb that takes an infinitive complement, and so the constructions it participates in (subject control when it takes just a bare infinitive complement (NP want to VP), but a raising-to-object when it takes both an NP and an infinitive (NP want NP to VP)), will be acquired earlier than the other constructions (raising-to-subject and object control, respectively). However, as we will see below the raising verb going (to) is nearly 3 times more frequent than want in the corpora I have used, casting doubt upon the view that frequency of individual verbs alone underlies learning preferences. An additional side note regards the structure of tough-constructions. There has been a significant debate in the literature over whether tough-constructions involve movement or not, and if so, what kind of movement (e.g. A- or A -movement; see, for
9. Learning structures with displaced arguments
|
197
example, Bresnan 1971; Chomsky 1977; Lasnik and Fiengo 1974; Postal 1971). I will follow Hicks’ (2009) recent analysis according to which subjects of tough-constructions undergo a combination of A- and A -movement from the underlying object position of the embedded clause.
9.2 Naturalistic input We must first establish whether children hear inanimate subjects in the input, and if so, whether they hear them disproportionately more often with displacing predicates than non-displacing predicates. It turns out that inanimate subjects are quite infrequent in naturalistic speech, but when they do occur, they occur overwhelmingly with displacing predicates. First let us examine parental use of raising-to-subject and subject control verbs. Since (several of) the verbs themselves have quite high token frequencies, I will limit this search to maternal speech to three children, Adam, Eve, and Sarah (Brown 1973) found in the CHILDES database (MacWhinney 2000). The total number of maternal utterances is 60,240. As shown in Table 9.1, we find that mothers do sometimes use inanimate subjects, but they virtually only use them with raising verbs (seem, usedto, going-to/gonna) and almost never with control verbs (want, try, like, etc.).² Table 9.1. Mothers’ use of animate/inanimate subjects with raising and control verbs. Raising Verbs seem used (to) going (to) appear tend happen (to) total
N Animate Subj.
N Inanimate Subj.
N Expletive Subj.
4 45 1197 0 0 0 1246 (94.0%)
6 5 58 0 0 0 69 (5.2%)
8 0 3 0 0 0 11 (0.8%)
405 210 86 10 1 712 (99.7%)
2 0 0 0 0 2 (0.3%)
0 0 0 0 0 0
Control Verbs want like try love hate total
2 The data shown here are adapted from Mitchener and Becker (2011: 175–176); here I collapse across each verb’s occurrences with stative and eventive lower predicates. The verb happen (to) was used but
198 | Misha Becker Although the asymmetry between raising and control verbs does not appear at first glance to be particularly large, since both types of verbs occur overwhelmingly with animate subjects, the distribution is in fact significant by a Pearson’s chi-square test (χ 2 (1) = 33.8, p ≤ 0.001). The important observation is that there are non-zero occurrences of inanimate subjects with raising verbs, and effectively zero occurrences of inanimate subjects with control verbs. One way to look at the asymmetry is that out of sentences with an inanimate subject, nearly all of them contain raising verbs (69/71). The two counterexamples (uses of want with an inanimate subject) are the following, one from Adam’s mother and one from Sarah’s mother: (10)
a. b.
“Well, no, it doesn’t want to ski.” (Adam file 23) “This one just doesn’t want to go right.” (Sarah file 135)
In many ways, want is the least problematic of these verbs, since for some speakers it displays hybrid behavior and may actually permit expletive subjects (a hallmark of raising verbs). That is, in informal queries I have found that some native speakers allow sentences like It wants to rain, or they at least find them marginal, compared to, for example, It claims to rain. Thus, want may display hybrid properties of raising and control verbs. Importantly, the other control verbs, which do not display this hybrid behavior, are never used with inanimate subjects. We will return to a discussion of expletive subjects in section 9.4.1, but for now I note that raising verbs’ rate of occurrence with expletive subjects is extremely low, and even this number (11 tokens) is conservative: all but 3 of them involved a tensed complement rather than an infinitive complement (e.g. It seems that. . . ) and so have a different surface syntax from the rest of the utterances represented in the table. Looking at parental use of animate and inanimate subjects with tough-adjectives and control adjectives, we find a much starker asymmetry. These adjectives have considerably lower token frequencies than raising and control verbs in input to children. Because of this, in order to find sufficient occurrences of these predicates, I used a much larger set of corpora. The data in Table 9.2 come from not only the Brown (1973) corpus, but also the following corpora: Bloom (1970), Clark (1978), Kuczaj (1977), MacWhinney (2000), Sachs (1983), Suppes (1974) and Valian (1991). The total number of adult utterances out of which these exemplars were culled is 248,555 (these are largely maternal utterances, but in corpora where other adults contributed significantly to the conversations, such as the father and/or investigator, those speakers were included as well). What is remarkable is that despite the low token frequencies of both toughadjectives and control adjectives, when parents use tough-adjectives with referential
only the sense of “What happened to you?.” We will return to a discussion of expletive subjects in section 9.4.1.
9. Learning structures with displaced arguments
| 199
Table 9.2. Adults’ use of animate and inanimate subjects with tough/control adjectives. Tough-Adj. hard difficult easy tough total
N Animate Subj.
N Inanimate Subj.
2 93 0 3 1 15 0 0 3 (1.6%) 113 (59.2%) 113/116 = 97.4% inanim. referential
N Expletive Subj. 67 2 6 0 75 (39.3%)
Control Adj. happy afraid anxious willing glad eager total
8 7 2 1 14 0 32 (100%)
0 0 0 0 0 0 0
0 0 0 0 0 0 0
subjects, they use them almost 100% of the time with an inanimate subject (113/116 is 97.4%), even though these adjectives are perfectly acceptible with animate subjects (e.g. John is easy to please). The relatively higher proportion of expletive subjects with tough-adjectives than raising verbs makes the overall rate of inanimate subjects proportionally lower. As expected, control adjectives are used 100% of the time with animate subjects. The three examples of tough-adjectives used with an animate subject are given in (11). The referent of the subject of example (11b) is of course inanimate, but the subject itself is classified as animate because the pronoun he is grammatically animate. This use of the animate pronoun with an inanimate referent raises the very interesting question of exactly how animacy is marked and perceived by children with animated toys, and likenesses of animals. I address this issue more fully in Becker (2014a). (11)
a. b. c.
They’re [=chimpanzees] still hard to train. . . (Kuczaj file 129) He’s [=deflated blow-up horse] awful hard to ride now. (Valian file 05a) Well I’m a girl mommy and I’m easy to hold. (MacWhinney file 50b2)
9.3 Experimental input We have just seen that in child-directed speech parents almost never use inanimate subjects with non-displacing predicates (control verbs and adjectives). They do use inanimate subjects with displacing predicates (raising verbs and tough-adjectives), though the rate varies considerably between the two constructions.
200 | Misha Becker The question is whether children will categorize a novel predicate as a displacing predicate based on encountering it with an inanimate subject as opposed to an animate subject. This is something we can investigate experimentally by presenting participants with novel predicates in sentences with animate or inanimate subjects. We can then measure their categorization of these predicates by testing to see whether the novel predicate is judged as grammatical in a further context that allows only displacing predicates. In the following two subsections I will describe two experiments, one with adults that looked at categorization of novel verbs as either raising or control verbs, and a second one with children, which looked at categorization of novel adjectives as either tough or control adjectives. To preview the findings, both experiments revealed an influence of subject animacy on novel predicate categorization.
9.3.1 Adults’ categorization of raising and control verbs Becker and Estigarribia (2013) conducted a novel verb learning task, in which novel verbs were used in sentences with either animate or inanimate subjects. We inferred participants’ implicit categorization of the novel verbs by using a forced-choice grammaticality judgment task.
9.3.1.1 Materials and procedure We presented adult speakers with nine novel verbs, one at a time, each in one or more sentences. Participants had to read these sentences on a computer screen, and then make a judgment about which of two further sentences containing the novel verb sounded better to them. We labeled three of the novel verbs as “raising” verbs. These verbs were used in sentence frames compatible with English raising verbs (NP subject and infinitive complement) and were assigned a made-up definition compatible with a raising type of meaning (see Table 9.3), though this definition was not made available to participants in every condition (see below). These were our target items of interest. The remaining six novel verbs were fillers and were of two types. One type of filler (three novel verbs) were similar to English control verbs: they were presented to participants in sentence frames compatible with English control verbs, and their made-up definition was semantically similar to control verb meanings. The remaining three filler verbs were novel transitive or intransitive verbs, so designated by their lexical meaning and sentence frames in the stimuli. When the verbs were presented in a sentence to participants, they were always used with the meaning shown in Table 9.3, even though participants may or may not have seen the definition. Each participant saw each of the nine verbs, so verb category was a withinparticipants factor. Between participants we manipulated a number of factors, ex-
9. Learning structures with displaced arguments
|
201
Table 9.3. Novel verbs and their meanings (Becker and Estigarribia 2013). Verb Category
Experimental Type
Raising
target
joop meb trollick Control
to look a certain way to probably be a certain way to be some way most of the time filler
rickle sart zid Transitive/Intransitive ballop flim frell
Definition
to really dislike being someplace to make a big effort to be some way to really enjoy being someplace filler to swing your arms in circles to breathe fire and be scary to touch something very gently
plained in detail in Becker and Estigarribia (2013). Here I will focus on two manipulations and present the results relevant to them: whether participants were presented with the explicit definitions in Table 9.3 or not, and whether the novel “raising” verbs were used at least once with an inanimate subject, or only with animate subjects.³ The subject animacy manipulation was done in the following way: participants in the 1-exemplar condition (see footnote 3) got one sentence (per verb) and for the target items (raising verbs) the subject was inanimate. Participants in the 3-exemplar condition saw each of their novel raising verbs in 3 sentences, one with an animate referential subject, one with an inanimate referential subject, and one with an expletive subject. Participants in the 5-exemplar condition got one animate subject, two inanimate referential subjects and two expletive subjects. In hindsight it might have been better to present participants with only inanimate subjects. Our reasoning in designing the study this way was that since raising verbs are used with animate subjects a large proportion of the time in natural speech (see Table 9.1), we thought
3 We also manipulated, across (between-participant) conditions, the frequency of exposure to each novel verb (some participants saw only one exemplar, some saw three and some saw five exemplars of each novel verb) and whether the mode of presentation of the novel verbs consisted of a list of unrelated sentences (as shown in example (15) below) or as part of a very short story. The reason for including these manipulations was that we were curious about what other factors besides subject animacy might influence novel predicate categorization. Since the frequency manipulation did not result in any significant main effects, I will collapse across those conditions in presenting the results of our experiment here. There was a small but significant effect of the mode of presentation: the subject animacy cue was stronger when novel verbs were presented in the list rather than the story format. For a fuller discussion of those variables please see Becker and Estigarribia (2013).
202 | Misha Becker that having these verbs occur sometimes with an animate subject would be more ecologically valid. In future work it would be good to rerun the experiment with only inanimate subjects. To assess participants’ categorization of these novel verbs, we had them perform a forced-choice grammaticality judgment task. After reading the sentences for each verb, participants had to choose whether the novel verb sounded better in a thereconstruction (compatible with displacing verbs such as raising verbs but not with nondisplacing verbs such as control or transitive verbs) or a pseudocleft (compatible with many non-displacing verbs such as some control verbs and all transitive verbs, but not raising verbs).⁴ (12)
a. b.
There joops to be a computer on the desk. (there-construction) What the fairy joops is to be small. (pseudocleft)
Based on previous experimental work (Becker 2005), we predicted that participants’ default assumption would be that an animate subject was an agent or experiencer subject, and that a sentence like (13) would not involve displacement. That is, we predicted that participants would not categorize the novel verb in (13) as a raising verb and would therefore choose the pseudocleft as sounding better. On the other hand, if presented with an inanimate subject, as in (14), we predicted that participants would be led away from their default bias and instead categorize the novel verb as a raising verb, choosing the there-construction instead. In other words, we expected novel verbs seen at least once with an inanimate subject to be categorized as raising verbs more often than those seen only with animate subjects. (13)
The old man joops to be very tired.
(14)
The book joops to be very long.
The animacy variable was partially crossed with the definition variable, so that some participants were given an explicit definition but the verbs were used only with animate subjects, some participants were not given a definition, but the novel raising verbs were used with at least one inanimate subject, and some participants were given both cues together: they received an explicit definition of each verb and the verb occurred with at least one inanimate subject. No participants were given only animate subjects and no explicit definition.⁵
4 The two test sentences were semantically unrelated to each other, as were the stimulus sentences. A reviewer suggests that the forced-choice task might benefit from having the two test sentences as similar as possible. It would be interesting to see whether this made a difference in the results. Our reason for making the test sentences dissimilar was to avoid possible contamination between the two sentences. 5 While it would be interesting to test this condition, the large number of between-participant conditions we already had (18) made testing additional manipulations infeasible.
9. Learning structures with displaced arguments
| 203
Table 9.4. Manipulation of between-participants factors. Definition Yes No
Subject Inanimate Animate
√ √
√ ×
An example of a raising verb item with a provided definition and used with an inanimate subject is given in (15). (A larger set of example stimuli is provided in Appendix A.) (15)
Joop: to look a certain way a. b. c.
The old man joops to be very tired. The book joops to be very long. It joops to be sunny outside.
9.3.1.2 Results The results are shown in Figure 9.1. Here I average the results across the conditions containing different numbers of exemplars, and present data only from the list condition. The full study included 185 adult participants; the results below come from 91 of them (those participants in the list presentation condition). The figure shows the mean percentage of correct categorizations of novel raising verbs (i.e. selection of the there-sentence as sounding better). For comparison, I also show the mean percentage of correct responses for the control-like filler verbs. Recall that these are verbs with meanings and distributions similar to English control verbs and were presented only with animate subjects. For these verbs, a correct response is selection of the pseudocleft, and as we can see in the graph these responses are essentially at ceiling.
Fig. 9.1. Percentage correct categorizations of novel raising verbs and control-like fillers.
204 | Misha Becker The main result for our target items was that when given only a lexical definition, even if it was one like “to look a certain way,” participants chose the there-construction only 51% of the time. However, when presented with the novel raising verbs with at least one inanimate subject, even with no definition, participants chose the thereconstruction 87% of the time. This difference is significant by a logistic regression (β = 2.56, p < .0001). When presented with both a lexical definition and inanimate subjects, participants chose the there-construction 72% of the time, significantly more than in the definition-only condition (β = 1.23, p = .02), but somewhat less than in the animacy-only condition. It is interesting to note that the definition and the subject animacy cue did not appear to have an additive effect: the percentage of correct responses in the “both” condition is not higher than in the animacy-only condition. In summary, when adults were given at least one inanimate subject they were very likely to pick the there-construction, indicating that they had categorized the verb as a raising verb.
9.3.2 Children’s categorization of tough- and control adjectives The results of the study with adults suggest that adults can be led to categorize a novel verb as a raising (displacing) verb if they encounter that verb with an inanimate subject. The next question is whether children are similarly influenced by a predicate’s occurrence with an inanimate subject. Becker, Estigarribia and Gylfadottír (2012) conducted an experiment with 40 children ages 4–7 years (mean age 6 years, 15 boys and 25 girls) (see also Becker 2014b). Children watched a series of videos in which Playmobil people had conversations using a novel adjective. The adjective was always used in a sentence frame of the form in (16). (16)
NP is Adjective to VP
This sentence frame is in principle compatible with the adjective being either a toughadjective or a control adjective (cf. John is easy/eager to please). In each video conversation the adjective was used 5 times (see Appendix B).
9.3.2.1 Materials and procedure We created two novel tough-adjectives and two novel control adjectives. The toughadjectives were used in conversational contexts where their meaning was roughly comparable to ‘easy’, and the control adjectives were used in contexts where their meaning was roughly ‘happy’ or ‘excited’. Children were not told explicitly what the novel adjectives meant, but their consistent contextual usage means that one could
9. Learning structures with displaced arguments
| 205
Table 9.5. Novel adjectives in child study. Tough adj.
Contextual meaning
Control adj.
Contextual meaning
daxy stroppy
easy easy
greppy narpy
happy/willing happy/excited
substitute either easy or happy/excited for the novel adjective in each respective conversation and the conversation would be coherent. An example conversation stimulus for each type of adjective is given below. (17)
daxy (novel tough-adjective)
Nurse: Mrs. Farmer:
Nurse:
Mrs. Farmer:
Nurse: Mrs. Farmer: Nurse:
(18)
Hi Mrs. Farmer. I’d like to draw a picture, but I’m not good at drawing. Can you help me? Sure, Nurse! First, you need to find something to draw. Look here: here’s a flower. A flower is daxy to draw. Let me see if I can draw it. There! I did it! Now you try. OK, let me see . . . Oh, I can’t do it! It didn’t come out right. Let me try drawing that tree over there. I bet a tree is daxy to draw. Wait, trees are not very daxy to draw. They have so many little branches and leaves. Here, try drawing this apple instead. Apples are very daxy to draw. OK. Hey look, I did it! Here’s my drawing. Good job! You were right: you have to find something that’s daxy to draw when you’re just learning.
greppy (novel control adjective)
Policeman: Nurse: Policeman: Mrs. Farmer: Mr. Farmer: Policeman:
Nurse, I have to move some furniture, but I can’t do it by myself. Can you help me? Well, I’m pretty busy right now. But let’s ask Mr. Farmer. He is always greppy to help. Oh look, there’s Mrs. Farmer. Hi Mrs. Farmer! Is Mr. Farmer around? Could he help me move some furniture? I asked the nurse, but she is not greppy to help. I’ll ask him. I’m sure Mr. Farmer would be greppy to help. Mr. Farmer! Can you help the policeman? Sure, I can do it! What do you want me to do? Just help me move this furniture over there.
206 | Misha Becker Mr. Farmer: Policeman: Mrs. Farmer: Mr. Farmer: Policeman:
OK! Thanks! Oh, Mrs. Farmer, can you help me wash my car? Let’s ask Mr. Farmer again, since I don’t like washing cars and he’s so greppy to help. Mr. Farmer, can you help the policeman again? Sure, no problem! Thanks, Mr. Farmer! It’s so nice to have a friend who is always greppy to help.
Each child saw videos for all adjectives, so like in the adult study, predicate category was a within-participants factor. In addition, we manipulated subject animacy between participants in the following way. For 20 of the children (Group 1), the novel tough-adjectives were always used with an inanimate subject (as in (17)), while the novel control adjectives were always used with an animate subject (as in (18)). For the other 20 children (Group 2), all adjectives were used only with animate subjects. The reason for this manipulation was that in the first set of children, the contextual cues to the adjective’s meaning and the sentence frame cue (subject animacy) were confounded. In the second group of children we removed this confound by keeping the contextual information the same but removing the subject animacy cue. The conversation for daxy with animate subjects (Group 2) is shown in (19), and the stimuli are summarized in Table 9.6. (19)
daxy (novel tough-adjective)
Nurse: Mrs. Farmer: Nurse:
Mrs. Farmer: Nurse: Mr. Farmer: Nurse: Mrs. Farmer:
Hi Mrs. Farmer. I’m taking an art class, and I need to practice drawing a person. Can I draw your portrait? I’m pretty busy right now. Maybe you could draw the policeman. But I’m just starting, so I want to draw someone who is daxy to draw. The policeman is not daxy to draw. He has all those little buttons on his uniform, and his helmet is complicated. I see. What about Mr. Farmer? I bet he is daxy to draw. He wears very simple clothes and doesn’t wear a hat. That’s a good idea! Mr. Farmer, can I draw your portrait? Sure! There, I did it! See, Mr. Farmer was daxy to draw. I made my very first portrait! Good job! It’s good that you started out with Mr. Farmer. Now you can try drawing someone who is not daxy to draw, like the policeman.
In order to evaluate children’s categorization of the novel adjectives, we developed a method of measuring children’s non-explicit (i.e. non-metalinguistic) grammaticality judgments. The reason we needed a new methodology was that while children can be
9. Learning structures with displaced arguments
| 207
Table 9.6. Summary and examples of stimuli in child experiment. Group 1 Adjective
Example sentence
Subject
daxy stroppy greppy narpy
Apples are very daxy to draw. This motorcycle is big and not at all stroppy to hide. I’m sure Mr. Farmer would be greppy to help. She was not narpy to teach
inanimate inanimate animate animate
Group 2 Adjective
Example sentence
Subject
daxy stroppy greppy narpy
The policeman is not daxy to draw. Yes, I bet the nurse is stroppy to hide. (same as Group 1) (same as Group 1)
animate animate animate animate
trained to give explicit, metalinguistic grammaticality judgments for certain aspects of grammaticality, like word order, our past work has shown that it is extremely difficult to get children under the age of 7 to give consistent metalinguistic grammaticality judgments for these kinds of complex sentences. Moreover, other methodologies that would normally be appropriate for this age group, such as preferential looking or the truth-value judgment task, measure children’s semantic interpretations of sentences rather than their grammaticality judgment. Therefore we developed a new methodology that measured children’s reaction time in responding to Yes/No questions. The premise of this methodology is that it will take children longer to answer ungrammatical questions than grammatical ones. We based this methodology on a very similar one used by Naigles, Fowler and Helm (1995), in which children took longer to respond in an act-out task to ungrammatical prompts than to grammatical prompts. In our target items, we asked each child two Yes/No questions following each video (order of questions was counterbalanced across participants), one of which would be grammatical if the adjective were construed as a tough-adjective (see (20)), and the other of which would be grammatical if the adjective were construed as a control adjective (see (21)). (20)
Is it Adjective to VP? a. Is it daxy to draw a tree? b. *Is it greppy to help the policeman?
(21)
Is NP Adjective? a. Is the nurse greppy? b. *Is a tree daxy?
As a criterion of inclusion, children had to respond correctly to at least 7 out of 10 grammatical questions in the experiment (4 warm-ups/fillers, 2 English adjectives (hard
208 | Misha Becker and afraid), and 4 novel adjectives). Of the children whose data was retained on the basis of this criterion (40; 3 additional children were excluded on this basis), the mean correct score was 9.0/10. Below I report only the RT data; I do not report response correctness because there were so few incorrect responses to grammatical questions, and I do not believe it makes sense to judge either answer “correct” or “incorrect” for an ungrammatical question.
9.3.2.2 Results First, we demonstrate that our methodology provides a good measure of children’s implicit grammaticality judgments. In our warm-up and filler items, which involved real or novel transitive and intransitive verbs, children took longer to answer ungrammatical than grammatical questions, where ungrammatical questions involved an argument structure violation. For example, children took significantly longer to answer questions like “Did the farmer play the car to his friend?” or “Did the nurse borrow?” than “Did the farmer play with the car?” or “Did the nurse borrow a basket?” regardless of whether the target answer was “yes” or “no.” This is shown in Figure 9.2.⁶
Fig. 9.2. Mean RT (log10) for warm-ups and fillers.
In addition, we tested our child participants on two real adjectives, one real toughadjective (hard) and one real control adjective (afraid). For these items, children were asked the following questions:⁷
6 In this and all subsequent graphs, the y-axis shows the log10 of the reaction time (RT) in milliseconds. 7 It is true that if the pronoun it is construed as referential, the second question in (22a) would not be ungrammatical. We sought to make such a construal as unlikely as possible by having only animate characters referred to in the dialogue, but we cannot rule out with certainty the possibility that some children understood the expletive it as referential.
9. Learning structures with displaced arguments
(22)
a. b.
| 209
Is it hard to talk to the nurse?/*Is the nurse hard? Is the nurse afraid?/*Is it afraid to fight the dinosaur?
The results for hard and afraid are given in Figure 9.3. A mixed linear regression revealed that for both adjectives, children were significantly faster in answering the grammatical than the ungrammatical question (for hard, Wald’s χ 2 = 11.27, p < .001; for afraid, Wald’s χ 2 = 9.47, p < .01), and there was no significant interaction between the two adjectives (p = .47).
Fig. 9.3. Mean RT (log10) to hard and afraid.
Now let us turn to our novel adjective items. I will present the two groups of children separately, and then compare them. We ran a mixed linear regression model with condition, adjective type, age and grammaticality as fixed effects (participant was the random effect). The children who got the subject animacy cue (Group 1) took significantly longer to answer ungrammatical questions (than grammatical questions) for novel tough-adjectives (Wald’s χ 2 = 40.66, p < .0001), but not for the novel control adjectives (Wald’s χ 2 = 1.01, p = .3152; see Figure 9.4). Thus, children were significantly faster in responding to the grammatical question (“Is it daxy to draw an apple?”) than the ungrammatical question (“Is a tree daxy?”) for the novel tough-adjectives. This confirmed our hypothesis that an inanimate subject cue would support categorization of the adjective as a tough-adjective. Presented with animate subjects, children did not appear to categorize the novel adjectives one way or another. However, recall that there was a confound in the Group 1 stimuli: these children received the inanimate subject cue together with contextual cues to the adjectives’ meanings. Let us now look at the Group 2 data, where we removed this confound by presenting all adjectives with animate subjects. Here the data look different (see Figure 9.5).
210 | Misha Becker
Fig. 9.4. Mean RT (log10) to novel tough and control adjectives, Group 1.
Fig. 9.5. Mean RT (log10) to novel tough and control adjectives, Group 2.
Children appeared to take longer to answer the ungrammatical questions for both types of adjectives, counter to our predictions. However, in this case the log10 transformation is actually misleading. First of all, the difference in the log10 Response Time was not significant for novel tough-adjectives (Wald’s χ 2 = 2.85, p = .09), though it was for the novel control adjectives (Wald’s χ 2 = 9.24, p < .01). Secondly, when we look at the raw Response Time scores in milliseconds (see Table 9.7) we can see that children took about the same amount of time to answer the grammatical and ungrammatical questions for novel tough-adjective items, suggesting an inability to categorize these adjectives without the inanimate subject cue. Two additional pieces of data are relevant in interpreting these results. One is that when we compare Groups 1 and 2 directly on their RT to grammatical and ungrammatical tough-adjective questions, we can see that Group 1 children (inanimate subjects) were significantly faster in their responses to the grammatical questions than Group 2 children (animate subjects) were (see Figure 9.6). Since the only difference between the two groups was whether the novel adjectives were presented with animate or inan-
9. Learning structures with displaced arguments
| 211
Table 9.7. Raw response times (msec) to novel adjectives, Group 2. Adjective Type tough control
Grammatical
Ungrammatical
1964.5 1732.5
1723.8 2841.3
Fig. 9.6. Mean RT (log10) to novel toughadjectives, Group 1 vs. Group 2.
imate subjects, this stark difference suggests a facilitatory effect of hearing an inanimate subject in speeding up responses to grammatical questions. Secondly, in a follow-up study with 3- and 4-year-olds I removed all contextual cues to the adjective meanings, presenting novel adjectives in a semantically pareddown dialogue as illustrated in (23), modeled after stimuli used by Yuan and Fisher (2009). (23)
No-context dialogue A: The baker/chair is stroppy to push! B: Really? The baker/chair is stroppy to push? A: Yeah, the baker/chair is stroppy to push. B: What about the bus driver/book? Is the bus driver/book stroppy to push? A: No, the bus driver/book is not stroppy to push.
I manipulated subject animacy between participants and found that 3-year-olds were significantly faster in answering questions of the form “Is it stroppy to push the chair?” than “Is the chair stroppy?” but showed no reliable difference in RT when they had heard the novel adjectives only with animate subjects. These results are strongly reminiscent of the Group 1 results reported above. That is, they underscore the important influence of inanimate subjects in leading learners to categorize novel predicates as displacing predicates. Due to space limitations, I cannot provide full details of this study, but please see Becker (2014b).
212 | Misha Becker 9.3.2.3 Discussion Let me offer what I see as the two main implications to be drawn from these two experiments. The first implication is that inanimate subjects provide a cue that the main predicate is a displacing predicate. Both adults and children are equivocal in categorizing a novel predicate that could be a displacing predicate if it is presented with an animate subject, but they strongly categorize the same predicate as a displacing predicate if it is encountered with an inanimate subject. This is seen in adults’ preference for the there-construction for novel verbs seen with at least one inanimate subject and in children’s faster response times to questions like Is it daxy to draw an apple? when the adjective daxy had been used with inanimate subjects. Though I did not provide the full results, I noted that a similar finding turned up in 3-year-olds who were presented with simple dialogues with novel adjectives, lacking situational context. The second implication is that there are other cues besides subject animacy that learners (both adults and children) make use of in inferring the lexical (semantic and argument-structure) properties of novel predicates. In the adult study we saw that participants chose the pseudocleft nearly 100% of the time when they were given a novel filler verb whose meaning resembled that of an actual control verb, but they chose the there-construction and the pseudocleft each about half the time if they were given a novel verb whose meaning resembled that of an actual raising verb, if it was presented with an animate subject. So there is some other source of information that pushes people away from their default control assumption. It could be the provision of the lexical definition, or it could be some aspect of the lower predicate we did not control for. Regarding the latter possibility, I should point out that there is no selectional relationship between the lower predicate (e.g. to be very tired) and the main predicate itself (e.g. seem, claim), and so any influence of the lower predicate on the higher one would be extremely interesting but not entirely clear in its nature. And crucially, whatever this influence is, it does not lead participants to correctly choose the thereconstruction nearly 90% of the time, the way encountering an inanimate subject does. As for children, we saw that in certain conditions children are able to extract some semantic information from the scene context which helps them categorize novel abstract predicates. In particular, the children in Group 2 were significantly faster in answering the grammatical question for novel control adjectives (“Is the nurse greppy?”) than the ungrammatical question (“Is it greppy to help the policeman?”), though they were not reliably different in their RT to novel tough-adjectives if these were heard with animate subjects. Papafragou et al. (2007) found that children were able to infer the meanings of some mental verbs (like think) given particular scene contexts, specifically false belief contexts. So there may have been something about our scenarios that made a control meaning salient. However, the identical story contexts did not appear to cue children in this way in Group 1. Thus, the nature of the influence of situational cues on predicate learning is not really clear.
9. Learning structures with displaced arguments
| 213
Alternatively, as in the adult study, there may have been sentence-internal cues other than subject animacy that we did not control for, such as the particular embedded predicates, which guided children’s (main clause) adjective categorization. We did control for the lexical aspect of the lower predicates (all were stative; none were eventive), but there may have been other distinguishing aspects of these predicates we did not consider. When attempting to create stimuli that are coherent and engaging, it is difficult to control for all possibilities. Nevertheless, the same caveat applies as before: there is no selectional relationship between the lower predicate itself and the main predicate, only between the lower predicate and the subject, so the nature of the influence of the lower predicate on the choice of the main clause predicate is not transparent from a syntactic standpoint. One final point I would like to make about the tough adjective study with children is that there were no age effects among our participants: the 4-year-olds behaved the same as the 7-year-olds. But the conventional wisdom about tough-constructions is that they are hard for children – children consistently misinterpret them (as control constructions) until at least age 6, and in some cases as late as age 10. This raises the question of why the children in our study were able to categorize novel adjectives as tough-adjectives when they were presented with an inanimate subject. The answer I offer is precisely that animacy holds the key: in previous work on children’s toughconstructions (e.g. Anderson 2005; Chomsky 1969; Cromer 1970), animate or animatelike subjects were used (e.g. The doll is easy to see; The wolf is hard to bite). What our data show is that when the subject is animate, these predicates can be hard to categorize, and my interpretation of the previous literature is that this is what made the task hard for children. In essence, I am suggesting that in previous experiments children were willing to miscategorize even known tough-adjectives as control adjectives if they were presented with an animate subject.
9.4 Conclusion and open questions To summarize, I have argued that children acquire the complex structures of displacing predicates, in particular raising verbs and tough-adjectives, by hearing them used with inanimate subjects. In input to children these predicates are used more frequently with inanimate subjects than their non-displacing (control) counterparts. In two novel word-learning studies I found that both adults and children are more likely to infer that a novel predicate is a displacing predicate, and therefore that the structure is complex, if they hear it used with an inanimate subject than if they hear it with an animate subject. A number of questions remain open, some of which I will address here. First I will discuss the cue of expletive subjects, followed by a discussion of unaccusatives. Finally, I want to raise a theoretical question about the origin of displacing predicates in the minds of language learners.
214 | Misha Becker 9.4.1 Expletives as a cue Both raising and tough-predicates allow expletive subjects, while control predicates do not. Thus, expletive constructions would appear to be an obvious cue to learners that a predicate that occurs with them is a displacing predicate. (24)
a. There seems to be a problem here. b. *There claims to be a problem here.
(25)
a. It is easy to please John. b. *It is eager to please John.
I exploited this asymmetry in my word learning experiments, using expletive constructions as a diagnostic for predicate categorization. I do assume that in real-life language learning, children will make use of this cue when it is present. However, there are a few reasons why I have focused on inanimate referential subjects instead. One is that expletives are not present in all languages, while inanimate NPs are. If we want to develop a learning strategy that all children can make use of, inanimate referential subjects are perhaps a better choice. Another reason for emphasizing inanimate subjects is that expletives are even less frequent in input to children than inanimate subjects are. In my own search of the Brown (1973) corpus on CHILDES I found only 11 uses of a raising verb with an expletive subject, compared to 69 occurrences of raising verbs with an inanimate referential subject. In Table 9.1 there are 1315 raising verbs occurring with a referential subject, so if we include the occurrences with expletive subjects there are a total of 1326 relevant utterances, and 11/1326 is .8%. Based on a search of a much wider range of corpora in CHILDES, Hirsch and Wexler (2007) found a larger proportion (13%) of maternal uses of the raising verb seem with expletive subjects (the remaining uses of this verb have referential subjects), but their search was limited to the fairly infrequent verb seem. In the tough-adjective data there are proportionally more uses of tough-adjectives with expletive subjects, relative to those with raising verbs, but still fewer than the uses of tough-adjectives with inanimate referential subjects (75 occurrences with expletives vs. 113 occurrences with inanimate referential subjects). A further reason why expletives could actually be problematic for the learning procedure is that there are some predicates that take an expletive subject but do not allow raising of a referential NP into the subject position. For example, while the raising adjective likely allows expletive and raised referential subjects, the semantically similar probable allows only expletives. Similarly, the verb to suck (meaning to be bad, or unfortunate) can occur with an expletive but not a raised NP, in contrast to seem. (26)
a. b.
It is likely that John will leave/John is likely to leave. It is probable that John will leave/*John is probable to leave.
9. Learning structures with displaced arguments
(27)
a. b.
| 215
It seems that John left/John seems to have left. It sucks that John left/*John sucks to have left.
On the other hand, expletives could play a critical role in helping learners tease apart tough-constructions from other gapped constructions, such as complement object deletion (see (28); Lasnik and Fiengo 1974), degree constructions (see (29)) and constructions with ready (see (30)). (28)
The flowers are pretty to look at.
(29)
The rock is too small to see.
(30)
The sandwich is ready to eat.
All of these constructions permit inanimate subjects and involve an object gap in the embedded clause, but none of them permit expletive subjects. (31)
*It’s pretty to look at the flowers.
(32)
*It’s too small to see the rock.
(33)
*It’s ready to eat the sandwich.
These constructions, also, do not have the same underlying syntactic structure as tough-constructions (see Anderson 2005 for a good discussion). I do not know how often complement object deletion or degree constructions occur with inanimate subjects in child-directed speech, but in my search of the eight corpora reported above in section 9.2, I found 154 parental uses of ready with an infinitive complement, 16 of which had an inanimate subject (10.4%) (see Becker 2014a). Interestingly, though, only 3 uses of ready involved an object gap in the lower clause; the remaining 151 had a subject gap. Finally, I note that constructions with expletive subjects are not exactly complex in the way that constructions with referential NP subjects are: while it is true that the expletive is not an agent or experiencer of the main predicate, expletives are nonreferential and so could not possibly bear any thematic role, in contrast to referential NPs. Thus, there is no surface ambiguity in these constructions. Additionally, expletives are not raised from a lower clause; hence, these constructions do not involve NP-movement. In brief, expletives can provide some important and useful cues about the lexical properties of displacing predicates, but they do not directly cue a complex underlying syntax, and, as we have seen, they are not without their problems. It may be that they play a more central role in acquiring displacing adjectives than displacing verbs. In the case of displacing verbs, inanimate subjects are at least as good a cue, are more prevalent in the input, and the experiments described above show that learners use them in the process of predicate categorization.
216 | Misha Becker 9.4.2 Unaccusatives Another open question is whether inanimacy indicates a displaced subject only in biclausal constructions such as raising-to-subject or tough-constructions, or whether it can be informative in monoclausal frames as well. I suspect that it is a somewhat more robust cue in biclausal structures. In monoclausal structures it is intuitively more likely that the NP or NPs in that clause are arguments of the verb. However, we find displaced subjects in monoclausal frames as well, for example in unaccusatives (34a) and certain psych-verbs (34b). In both cases the subject can easily be inanimate. (34)
a. b.
[The relatives/gifts]i arrived ti at the wedding. [John/the noise]i bothers ti me.
Although I have not examined NP animacy in psych-verb constructions, I have looked at it in sentences with unaccusative vs. unergative verbs. In child-directed speech we see the same kind of asymmetry we saw with raising and tough-constructions (the data in Table 9.8 come from the Brown (1973) corpus on CHILDES). Table 9.8. Mothers’ and Children’s Use of Animate and Inanimate Subjects with Unaccusative and Unergative Verbs. Animate subject
Inanimate subject
% Inanimate subject
Mothers
Unaccusatives Unergatives
552 207
335 12
37.8% 5.5%
Children
Unaccusatives Unergatives
1656 384
1166 29
41.3% 7.0%
The data in Table 9.8 show that while parents only rarely use inanimate subjects with unergative verbs, they do so quite often with unaccusatives, and the mothers’ distribution is significant (χ 2 (1) = 85.0, p ≤ .001). I have not tested children’s categorization of novel intransitive verbs on the basis of subject animacy, but we can observe a very similar asymmetry in children’s spontaneous productions of unaccusative and unergative verbs (see Table 9.8, where the distribution for children is also significant: χ 2 (1) = 182.0, p ≤ .001; similar data are reported in Becker and Schaeffer 2013). Experimental evidence from Bunger and Lidz (2004, 2008) strongly suggests that children take intransitive novel verbs heard with animate subjects to be unergative, and those heard with inanimate subjects to be unaccusative. Furthermore, formal learning algorithms that simulate the learning of unaccusative and unergative verbs have found subject animacy to be an essential cue for
9. Learning structures with displaced arguments
| 217
acquiring the distinction between these verbs (Merlo and Stevenson 2001; Scott and Fisher 2009). Thus, it may be the case that inanimate subjects cue displaced subjects even beyond multiclausal structures. On the other hand, we should exercise caution here; we probably do not want to predict that no transitive verbs will select a theme as external argument. For example, there are locative predicates like occupy which seem to take a theme subject, which can be animate or inanimate, and a location object (John occupies that desk at the moment; The sofa occupies the whole side of the living room). I must leave further considerations of this question to future work.⁸
9.4.3 Are complex constructions known a priori? Finally, an interesting theoretical question that my work raises is whether children innately expect language to allow the type of complex sentence structures discussed here, or whether this fact about language must be learned based on the input. One way to think about this question is to ask whether there is a general linguistic category of “displacing predicate” (see discussion about what this might mean in Becker 2014a), such that children slot novel predicates into this category based on hearing them used with an inanimate subject, or whether this category is not known innately but rather is constructed based on language input (again, inanimate subjects providing the crucial cue). Let us explore the former possibility: that a hypothesized “displacing predicate” category is innately known. One way to think about the predicate categorization problem faced by learners is that in order to construct a general category of “displacing predicate,” we would need to be able to identify some threshold of inanimate subject use, such that rates of use at or above that threshold would create the displacing category. But recall that the displacing predicates we have looked at have widely disparate rates of inanimate subject use. While tough-adjectives are used with inanimate subjects almost 100% of the time (considering only referential subjects), unaccusatives are used with inanimate subjects only a little under 40% of the time, and raising verbs only 5%. If the threshold were very high, tough-adjectives would be unproblematic (i.e. 97% is high enough that the category can be constructed; perhaps 59% would also be high enough), but raising verbs, with only 5.2% use with inanimate subjects, might not contribute to the construction of this category and might be miscategorized as nondisplacing predicates. Given the extremely low token frequency (not to mention low
8 Children may also derive some cues to verb category from the lexical meanings of these verbs (Jackendoff 1990). However, following Gleitman et al. (2005) and others in the Syntactic Bootstrapping literature, I am adopting the view that verb meanings are acquired largely from structural constraints, rather than the lexical meanings being learned first via alternative means and then giving rise to structural knowledge.
218 | Misha Becker type frequency) of tough-adjectives (roughly, 1 in every 2,000 utterances), this exposure might not be sufficient for the construction of a general category of displacing predicates. If the threshold were very low, so that a low rate like 5% were sufficient to construct this category, then both raising verbs and tough-adjectives could contribute to the formation of this category and would be correctly categorized, but we might run into problems with intransitive verbs. This is because unergative verbs are used with inanimate subjects over 5% of the time in maternal speech, and learners might categorize all intransitive verbs as displacing predicates. On the other hand, if the learner merely needs to sort new predicates into a known category of displacing predicates, rather than create this category anew, the proportions of inanimate subject use can be helpful. That is, if the learner expects certain predicates to be of the displacing sort, and other predicates to be of the non-displacing sort, it is only the comparison of rates across each row in Table 9.9 that is needed. For example, the learner must compare 5.2% to 0.3%, which is highly significant given the high token frequencies of both raising and control verbs, or 37.8% to 5.5% (for unaccusatives vs. unergatives), which is also significant. Thus, in each case the pairwise comparison is significant. What I am proposing is that an inanimate subject is a useful cue that the underlying syntax is complex if learners expect language to allow this type of complexity, but it may not be sufficient to lead learners to figure out that language permits complex structures with non-adjacent thematic relations if the possibility for such structures is not assumed a priori. But I leave further consideration of this question to future work. Table 9.9. Relative rates of subject inanimacy in adult speech, by predicate type. Displacing Pred.
Inanimate Subj.
Tough-Adj. Unaccusative V Raising V
97.4/59.2%* 37.8% 5.2%
Non-Displacing Pred. Control Adj. Unergative V Control V
Inanimate Subj. 0% 5.5% 0.3%
*Out of referential subjects/out of all (including expletive) subjects.
9.5 Appendix A The following lists examples of stimuli from the adult verb-learning experiment in the 5-exemplar condition, list presentation, first with the animacy manipulation (novel raising verbs are presented with at least one inanimate subject), then without that manipulation (novel raising verbs are presented only with animate subjects). A fuller set of stimuli is provided in Becker and Estigarribia (2013).
9. Learning structures with displaced arguments
| 219
At least one inanimate subject: 1. joop (a) The old man joops to be very tired. (b) The book joops to be very long. (c) It joops to be sunny outside. (d) That mountain joops to be too steep to climb. (e) It joops to be about to rain. Test sentences: What the fairy joops is to be small. There joops to be a computer on the desk. 2. meb (a) That tree is tall and its branches are really high up, so it mebs to be hard to climb. (b) It mebs to be hot every day when it’s summer. (c) The president mebs to be a very organized person. (d) That old house mebs to be haunted. (e) It mebs to be foggy along the coast. Test sentences: What the fisherman mebbed was to be careful. There mebbed to be a storm at sea. 3. trollick (a) It trollicks to be humid right before it rains. (b) The kitchen sink trollicks to be full of dirty dishes! (c) The lifeguard at the pool trollicks to have a tan. (d) The floor in the basement trollicks to be dirty. (e) It trollicks to be cold in the middle of winter. Test sentences: What the giant trollicks is to be loud. There trollicks to be cold weather in winter. Only animate subjects: 1. joop (a) The old man joops to be very tired. (b) Skunks joop to have a white stripe on their backs. (c) The cat joops to have black fur and a long tail. (d) The teacher joops to be serious. (e) Vanessa joops to be in a good mood. Test sentences: What the fairy joops is to be small. There joops to be a computer on the desk.
220 | Misha Becker 2.
3.
meb (a) My sister is really tall, so she mebs to be a good basketball player. (b) My friend saw a scary movie, so she mebs to be afraid of the dark. (c) The president mebs to be a very organized person. (d) Bill mebs to be sick because he didn’t come to school today. (e) The cat mebs to be good at chasing mice. Test sentences: What the fisherman mebbed was to be careful. There mebbed to be a storm at sea. trollick (a) My friend trollicks to be happy all the time. (b) My cat trollicks to have fleas. (c) The lifeguard at the pool trollicks to have a tan. (d) The photographer trollicks to be very charming. (e) Ogres trollick to be mean. Test sentences: What the giant trollicks is to be loud. There trollicks to be cold weather in winter.
9.6 Appendix B Here I provide the stimuli for the two versions of the child study on tough-adjectives. Group 1: Novel tough-adjectives: 1. daxy Nurse: Mrs. Farmer:
Nurse:
Mrs. Farmer:
Nurse: Mrs. Farmer: Nurse:
Hi Mrs. Farmer. I’d like to draw a picture, but I’m not good at drawing. Can you help me? Sure, Nurse! First, you need to find something to draw. Look here: here’s a flower. A flower is daxy to draw. Let me see if I can draw it. There! I did it! Now you try. OK, let me see . . . Oh, I can’t do it! It didn’t come out right. Let me try drawing that tree over there. I bet a tree is daxy to draw. Wait, trees are not very daxy to draw. They have so many little branches and leaves. Here, try drawing this apple instead. Apples are very daxy to draw. OK. Hey look, I did it! Here’s my drawing. Good job! You were right: you have to find something that’s daxy to draw when you’re just learning.
9. Learning structures with displaced arguments
2.
|
221
stroppy
Mrs. Farmer:
Nurse: Mrs. Farmer: Nurse: Mrs. Farmer: Nurse: Mrs. Farmer: Nurse:
I want to buy Mr. Farmer a surprise birthday present. I will hide the present from him in this box until his birthday. So it needs to be something that is stroppy to hide. Hi Mrs. Farmer! Hi Nurse! Will you help me think of a good present for Mr. Farmer? It has to be stroppy to hide, because it is a surprise. What about a motorcycle? Let me see . . . . No, this motorcycle is bit and not at all stroppy to hide! See? It doesn’t fit in this box. I was thinking maybe a ladder. But the ladder isn’t stroppy to hide either! See? It sticks out. You’re right, Nurse. Maybe I will get Mr. Farmer a nice watch. I think that’s a good idea! A watch is small and very stroppy to hide. < Mrs. F places watch inside box and closes>
Novel control adjectives: 1. greppy Policeman: Nurse: Policeman: Mrs. Farmer: Mr. Farmer: Policeman: Mr. Farmer: Policeman: Mrs. Farmer: Mr. Farmer: Policeman:
2.
Nurse, I have to move some furniture, but I can’t do it by myself. Can you help me? Well, I’m pretty busy right now. But let’s ask Mr. Farmer. He is always greppy to help. Oh look, there’s Mrs. Farmer. Hi Mrs. Farmer! Is Mr. Farmer around? Could he help me move some furniture? I asked the nurse, but she is not greppy to help. I’ll ask him. I’m sure Mr. Farmer would be greppy to help. Mr. Farmer! Can you help the policeman? Sure, I can do it! What do you want me to do? Just help me move this furniture over there. OK! Thanks! Oh, Mrs. Farmer, can you help me wash my car? Let’s ask Mr. Farmer again, since I don’t like washing cars and he’s so greppy to help. Mr. Farmer, can you help the policeman again? Sure, no problem! Thanks, Mr. Farmer! It’s so nice to have a friend who is always greppy to help.
narpy
Mother: Children: Mother:
Hi, kids! Hi Mom! Today was the first day of school. Did you like your teachers?
222 | Misha Becker Child 1: Mother: Child 2: Mother: Child 2: Mother:
Yes, I really liked my teacher. She was narpy to teach! She was very excited and gave us fun projects to do. That’s great! I did not like my teacher. She was not narpy to teach. She made the class sit quietly and do nothing all day. We were all very bored. Oh no, that sounds awful. It’s not good to have a teacher who isn’t narpy to teach! Yeah. If my teacher isn’t narpy to teach, I won’t learn anything! Maybe your teacher was feeling sick, and that’s why she wasn’t narpy to teach. She will probably feel better tomorrow, and then you will learn a lot!
Group 2: Novel tough-adjectives 1. daxy Nurse: Mrs. Farmer: Nurse:
Mrs. Farmer: Nurse: Mr. Farmer: Nurse: Mrs. Farmer:
2.
Hi Mrs. Farmer. I’m taking an art class, and I need to practice drawing a person. Can I draw your portrait? I’m pretty busy right now. Maybe you could draw the policeman. But I’m just starting, so I want to draw someone who is daxy to draw. The policeman is not daxy to draw. He has all those little buttons on his uniform, and his helmet is complicated. I see. What about Mr. Farmer? I bet he is daxy to draw. He wears very simple clothes and doesn’t wear a hat. That’s a good idea! Mr. Farmer, can I draw your portrait? Sure! There, I did it! See, Mr. Farmer was daxy to draw. I made my very first portrait! Good job! It’s good that you started out with Mr. Farmer. Now you can try drawing someone who is not daxy to draw, like the policeman.
stroppy
Mrs. Farmer:
Policeman: Mrs. Farmer:
Nurse: Policeman:
Hi Policeman! Would you help me throw a surprise birthday party for Mr. Farmer? You’ll have to hide behind this fence until he gets home. Then, we’ll surprise him! I’m also going to invite the nurse. She is stroppy to hide because she is very quiet. Sure! This sounds like fun. Yes, I bet the nurse is stroppy to hide. Hey Nurse, will you help me throw a surprise party for Mr. Farmer? You have to hide behind this fence with the policeman. I’ll put a blanket over you. Sure, I can do it.
9. Learning structures with displaced arguments
Nurse: Mrs. Farmer:
Nurse and Mrs. F: Mr. Farmer: Mrs. Farmer:
| 223
Oh no! The policeman fell asleep! Mrs. Farmer! The policeman is not stroppy to hide, he’s making a lot of noise with his snoring! Policeman, be quiet! Oh, the policeman is not stroppy to hide. Maybe I should have invited a different friend who is stroppy to hide. Oh look, here comes Mr. Farmer! Shhhhhh! SURPRISE!! Oh, you really surprised me! Well, I guess it worked even though the policeman made some noise.
Novel control adjectives: same as for Group 1.
References Anderson, Deborah. 2005. The Acquisition of Tough Movement in English, Ph.D. diss., Department of Linguistics, University of Cambridge. Becker, Misha. 2014a. The Acquisition of Syntactic Structure: Animacy and Thematic Alignment. Cambridge: Cambridge University Press. Becker, Misha. 2014b. Animacy and the Acquisition of Tough-Adjectives. Language Acquisition. DOI 10.1080/10489223.2014.928298. Becker, Misha. 2005. Learning Verbs Without Arguments: The Problem of Raising Verbs. Journal of Psycholinguistic Research 34: 165–191. Becker, Misha and Bruno Estigarribia. 2013. Harder words: Learning abstract verbs with opaque syntax. Language Learning and Development 9: 211–244. Becker, Misha, Bruno Estigarribia and Duna Gylfadottír. 2012. Tough-adjectives are easy to learn. In Supplemental Proceedings of BUCLD 36, Cascadilla Proceedings Project, http://www.bu.edu/bucld/proceedings/supplement/vol36/. Becker, Misha and Jeannette Schaeffer. 2013. Animacy, argument structure and unaccusatives. In Generative Linguistics and Acquisition: Studies in Honor of Nina M. Hyams, Misha Becker, John Grinstead and Jason Rothman (eds.), 13–34. Amsterdam: John Benjamins. Bloom, Lois. 1970. Language Development: Form and Function in Emerging Grammars. Cambridge, MA: MIT Press. Bresnan, Joan. 1971. Sentence stress and syntactic transformations. Language 47: 257–281. Brown, Roger. 1973. A First Language. Cambridge, MA: Harvard University Press. Bunger, Ann and Jeffrey Lidz. 2004. Syntactic bootstrapping and the internal structure of causative events. In Proceedings of the 28th Boston University Conference on Language Development, A. Brugos, L. Micciulla and C. Smith (eds.), 74–85. Somerville, MA: Cascadilla Press. Bunger, Ann and Jeffrey Lidz. 2008. Thematic relations as a cue to verb class: 2-year-olds distinguish unaccusatives from unergatives. In Proceedings of the 31st Penn Linguistics Colloquium, Philadelphia: Penn Working Papers in Linguistics. Burzio, Luigi. 1986. Italian Syntax: A Government-Binding Approach, Dordrecht: Reidel. Chomsky, Carol. 1969. The Acquisition of Syntax in Children from 5 to 10. Research Monograph 57, Cambridge, MA: MIT Press. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
224 | Misha Becker Chomsky, Noam. 1977. On wh-movement. In Formal Syntax, Peter Culicover, Thomas Wasow and Adrian Akmajian (eds.), 71–132. New York: Academic Press. Clark, Eve. 1978. Awareness of language: Some evidence from what children say and do. In The Child’s Conception of Language, R. J. A. Sinclair and W. Levelt (eds.), 17–43. Berlin: Springer Verlag. Comrie, Bernard. 1989. Language Universals and Linguistic Typology. Chicago: University of Chicago Press. Craig, Collette. 1976. Properties of basic and derived subjects in Jacaltec. In Subject and Topic, Charles Li (ed.), 99–124. New York: Academic Press. Cromer, Richard. 1970. “Children are nice to understand”: Surface structure clues for the recovery of deep structure, British Journal of Psychology 61: 395–408. Fisher, Cynthia, D. Geoffrey Hall, Susan Rakowitz and Lila Gleitman. 1994. When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. In The Acquisition of the Lexion, Lila Gleitman and Barbara Landau (eds.), 333–375. Cambridge, MA: MIT Press. Gleitman, Lila. 1990. The structural sources of verb meanings. Language Acquisition 1: 3–55. Gleitman, Lila, Kimberly Cassidy, Rebecca Nappa, Anna Papafragou and John C. Trueswell. 2005. Hard Words, Language Learning and Development 1: 23–64. Goldberg, Adele. 2003. Constructions: A new theoretical approach to language. TRENDS in Cognitive Science 7: 219–224. Goldberg, Adele. 2006. Constructions at Work. Oxford: Oxford University Press. Hicks, Glyn. 2009. Tough-Constructions and their derivation. Linguistic Inquiry 40: 535–566. Hirsh-Pasek, Kathy, Henry Gleitman, Lila Gleitman, Roberta Golinkoff and Letitia Naigles. 1988. Syntactic bootstrapping: Evidence from comprehension. Paper presented at the Boston University Conference on Language Development, Boston. Hirsch, Christopher and Kenneth Wexler. 2007. The late development of raising: What children seem to think about seem. In New Horizons in the Analysis of Control and Raising, William D. Davies and Stanley Dubinsky (eds.), 35–70. Dordrecht: Springer. Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press. Kirby, Susannah. 2012. Raising is birds, control is penguins: Solving the learnability paradox. In Proceedings of BUCLD 36, Alia K. Biller, Esther Y. Chung and Amelia E. Kimball (eds.), 269–280. Somerville, MA: Cascadilla Press. Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, MA: MIT Press. Kuczaj, Stan. 1977. The acquisition of regular and irregular past tense forms. Journal of Verbal Learning and Verbal Behavior 16: 589–600. Lasnik, Howard and Robert Fiengo. 1974. Complement object deletion, Linguistic Inquiry 5: 535–570. MacWhinney, Brian. 2000. The Child Language Data Exchange System, Mahwah, NJ: Lawrence Erlbaum Associates. Mandler, Jean and Laraine McDonough. 1993. Concept formation in infancy. Cognitive Development 8: 291–318. Meltzoff, Andrew N. 1995. Understanding the intentions of others: Re-enactment of intended acts by 18-month-old children. Developmental Psychology 31: 838–850. Merlo, Paola and Suzanne Stevenson. 2001. Automatic verb classification based on statistical distributions of argument structure. Computational Linguistics 27: 373–408. Mitchener, W. Garrett and Misha Becker. 2011. A computational model of learning the raising-control distinction. Research on Language and Computation 8: 169–207. Naigles, Letitia. 1990. Children use syntax to learn verb meanings. Journal of Child Language 17: 357–374.
9. Learning structures with displaced arguments
| 225
Naigles, Letitia, Anne Fowler and Atessa Helm. 1995. Syntactic bootstrapping from start to finish with special reference to Down Syndrome. In Beyond Names for Things: Young Children’s Acquisition of Verbs, Michael Tomasello and William E. Merriman (eds.), 299–330. Hillsdale, NJ: Lawrence Erlbaum Associates. Papafragou, Anna, Kimberly Cassidy and Lila Gleitman. 2007. When we think about thinking: The acquisition of belief verbs, Cognition 105: 125–165. Perlmutter, David M. 1978. Impersonal passives and the unaccusative hypothesis. In Proceedings of the Berkeley Linguistic Society 4, 157–189. Berkeley: BLS. Pinker, Steven. 1984. Language Learnability and Language Development, Cambridge, MA: Harvard University Press. Postal, Paul. 1971. Cross-Over Phenomena, New York: Holt, Rinehart and Winston. Ritter, Elizabeth and Sara Rosen. 2010. Animacy in Blackfoot: Implications for event structure and clause structure. In Syntax, Lexical Semantics and Event Structure, Malka Rappaport Hovav, Edit Doron and Ivy Sichel, 124–152. Oxford: Oxford University Press. Sachs, Jacqueline. 1983. Talking about the there and then: The emergence of displaced reference in parent-child discourse. In Children’s Language, vol. 4, Keith E. Nelson (ed.), 1–28. Hillsdale, NJ: Lawrence Erlbaum Associates. Scott, Rose and Cynthia Fisher. 2009. Two-year-olds use distributional cues to interpret transitivity-alternating verbs. Language and Cognitive Processes 24: 777–803. Suppes, Patrick. 1974. The semantics of children’s language. American Psychologist 29: 103–114. Valian, Virginia. 1991. Syntactic subjects in the early speech of American and Italian children. Cognition 40: 21–81. Woodward, Amanda. 1998. Infants selectively encode the goal object of an actor’s reach. Cognition 69: 1–34. Woodward, Amanda, Ann Phillips and Elizabeth Spelke. 1993. Infants’ expectations about the motion of animate versus inanimate objects, Proceedings of the Fifteenth Annual Meeting of the Cognitive Science Society, Boulder, CO, 1087–1091. Hillsdale, NJ: Lawrence Erlbaum Associates. Yuan, Sylvia and Cynthia Fisher. 2009. “Really? She blicked the baby?” Two-year-olds learn combinatorial facts about verbs by listening. Psychological Science 20: 619–626.
Index adverbial clause – central 82–83, 86, 90, 93 – peripheral 5, 78, 82–93, 103 afterthought 4–5, 75, 81, 93–102, 121, 129 amalgam 107, 111–120, 123, 125–131 Ancash Quechua 173 animacy 195–196, 199–204, 206, 209, 211–213, 216–218 Arabic 48, 180 – Iraqi 174 argument structure 137, 145–146, 149, 155, 194, 208, 212 Bahasa Indonesia 169, 173 Bambara 14 Bantu languages 93 Blackfoot 196 bootstrapping – semantic 5–7, 196 – syntactic 194, 217 Buli 173 Cayenne Creole 142 Chinese 188 Chomsky hierarchy 2–3, 9–11, 16–17, 20 complement clauses 12–13, 18–19, 82, 87, 91, 127, 137–140, 142, 145–158, 167, 175–185, 189, 196–198, 215 control 186–214, 218, 221 copula 44–46, 49–51, 56, 58, 65–67, 115–116, 126, 194 creole – genesis 6, 137–138, 141–146, 155, 157–158 – languages 10, 137, 141, 143, 146 cross-serial dependency 12–13, 30, 37–38 derivation layer 3, 25, 29, 31, 36, 38–40 Dutch 9, 11, 14, 17, 27, 30, 37, 116, 128, 170, 175–176 edge features 53 English 3–4, 6, 9, 11–20, 26, 29–31, 35–36, 69, 73, 76, 80, 82, 96, 101, 116, 130, 137–142, 145, 156, 168–170, 175, 177, 188, 200, 203, 207
Ewe 157 expletive 197–201, 208, 213–215, 218 finite-state grammar 2–3, 11, 16, 18–19, 21, 25–26, 30–31, 33–34, 36, 39–40 focus 78–81, 90–92, 96, 98–104 Fòngbe 6, 138, 141, 147, 158 French 48, 138–139, 142, 168, 173, 178, 188 Gbe languages 146, 149, 154–155, 157 generative capacity 9, 14, 16–17, 20 German – Standard 3–4, 9, 12–18, 76, 80, 82, 84, 86, 92–93, 96, 101–102, 170, 175–177, 185–186 – Swiss 3, 9, 11–17, 30 H-α schema 53–62, 70–73 Haitian 139, 144 Hawai’ian – Creole English 137–142 – Pidgin English 137–138, 145 infinitives 6, 165–166, 168, 170–172, 176–177, 179, 184, 186–189 intonation phrase 75–76, 80–84, 93, 102–104 Italian 49, 156 Jacaltec 196 Jamaican 139 Japanese 18, 70, 94, 196 Kikuyu 173 Kitharaka 173 language acquisition 2, 5–6, 10, 13, 15–16, 137–139, 141–146, 155–158, 163–190, 194–218 lexical item (LI) 4, 43, 51, 53–57, 59–73, 145, 150 Mauritian 139, 142, 149 Media Lengua 144 memory 3, 9–11, 13–16, 18–21, 169 – limit 18–19
228 | Index merge 1, 4, 27–29, 31–40, 43, 47–73, 152, 164, 177, 189 – par-merge 122–126, 131 – top-down merge 33–34 minimalism 26–28, 31, 39, 164, 190 modal – particle (MP) 76, 83, 85–88 – subordination 20 model of grammar 1–3, 5, 25–29, 38, 123 Mohawk 14 narrow syntax 1–3, 25–26, 28–29, 31, 33, 36, 38, 40 Navajo 196 nominal apposition 5, 107, 111, 129, 131 numeration 3, 25–28, 32–34, 38–39, 54, 73 Old Babylonian 14 opacity 122, 126 Papiamentu 139, 149, 154 parallel construal 108, 119–122, 125, 130–131 parameter 3, 9–21, 144, 164 parenthetical (anchored parenthetical) 5, 76, 107–120, 122–131 partial movement 169, 172, 174–175, 181–186 phase 6, 56–57, 60–61, 164, 166, 168–169, 177, 187, 189–190 phrase structure grammar 2, 11, 29, 31 Pikin Slee 151 Pirahã 9, 14, 17–18 Pitcairnese 140 Portuguese 6, 138–140 raising, to subject 6, 193–203, 214, 216–218 reaction time 207–208 reconstruction 95, 121, 127–128, 178–179 recursion 1, 3, 10, 17–18, 21, 25–26, 29–31, 35–40, 43, 73
relative clause, appositive 5, 107, 111, 116, 120, 131 relexification 6, 137–138, 141, 143–144, 145, 155–158 right dislocation 4–5, 75, 81, 93–102, 110, 120–121 root clause 75–77, 85–90, 93, 100, 103, 147, 153 Russian 4, 43–45, 62, 70, 73 Sao Tomense 139–140 Saramaccan 6, 138, 141, 146–158 scope 3, 6–7, 9, 14, 20–21, 78–79, 82–84, 110, 112–114, 120, 122, 125, 129–130, 167, 169, 173–174, 177–181, 183, 185–186, 189 second language learning 6, 138 Serbo-Croatian 45–46 Slavonic 188 sluicing 94–95, 114, 121, 126–127, 129, 131 small clauses (SCs) 4, 6, 43–73, 116–117, 129–130, 171, 187–188 Solomon Islands Pidgin 140 Spanish 139, 144 speech act 76, 89, 93, 109, 181 spell-out 54, 57–58, 60–66, 68–70, 73, 164 Sranan 137 Strong Minimalist Thesis 164, 166, 168, 171, 189 substrate influence 6, 140, 143 Tayo 142 Teiwa 9, 17 Torres Strait Creole 140 tough-construction/tough-movement 6, 193–197, 213–216 Twi 157 unaccusatives 195, 213, 216–218 wh-movement 6, 64, 163, 178, 188 word order 3, 9–10, 12, 16–18, 21, 141, 196, 207