Recursion and Human Language 9783110219258, 9783110219241

Citation preview

Recursion and Human Language

Studies in Generative Grammar 104

Editors

Harry van der Hulst Jan Koster Henk van Riemsdijk

De Gruyter Mouton

Recursion and Human Language Edited by

Harry van der Hulst

De Gruyter Mouton

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

ISBN 978-3-11-021924-1 e-ISBN 978-3-11-021925-8 ISSN 0167-4331 Library of Congress Cataloging-in-Publication Data Recursion and human language / edited by Harry van der Hulst. p. cm. ⫺ (Studies in generative grammar ; 104) Based on presentations at a 2007 conference. Includes bibliographical references and index. ISBN 978-3-11-021924-1 (hardcover : alk. paper) 1. Grammar, Comparative and general ⫺ Syntax ⫺ Congresses. 2. Grammar, Comparative and general ⫺ Phonology ⫺ Congresses. P291.R38 2010 415⫺dc22 2010003502

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de. ” 2010 Walter de Gruyter GmbH & Co. KG, Berlin/New York Typesetting: RoyalStandard, Hong Kong Printing: Hubert & Co. GmbH & Co. KG, Göttingen ⬁ Printed on acid-free paper Printed in Germany www.degruyter.com

Table of Contents List of authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii

Preliminaries Re Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Harry van der Hulst

xv

Part I. Discussing the need for recursion on empirical grounds 1. Piraha˜ – in need of recursive syntax? . . . . . . . . . . . . . . . . . . . . Jeanette Sakel & Eugenie Stapert

3

2. The fluidity of recursion and its implications. . . . . . . . . . . . . . . Marianne Mithun

17

3. Recursion and iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fred Karlsson

43

4. Recursion in conversation: What speakers of Finnish and Japanese know how to do. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ritva Laury & Tsuyoshi Ono

69

5. What do you think is the proper place of recursion? Conceptual and empirical issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arie Verhagen

93

Part II.

Formal Issues

6. Recursion and the infinitude claim . . . . . . . . . . . . . . . . . . . . . . Geo¤rey Pullum & Barbara C. Scholz

113

7. Just how big are natural languages? . . . . . . . . . . . . . . . . . . . . . D. Terence Langendoen

139

8. Recursion, infinity, and modeling . . . . . . . . . . . . . . . . . . . . . . . Hans-Jo¨rg Tiede & Lawrence Ne¤ Stout

147

9. How recursive is language? A Bayesian exploration . . . . . . . . . Amy Perfors, Josh Tenenbaum, Edward Gibson & Terry Regier

159

vi

Table of Contents

Part III. Evolutionary Perspectives 10. Was recursion the key step in the evolution of the human language faculty? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Anna Kinsella

179

11. When clauses refuse to be recursive: An evolutionary perspective. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ljiljana Progovac

193

12. The use of formal language theory in studies of artificial language learning: a proposal for distinguishing the di¤erences between human and nonhuman animal learners . . . . . . . . . . . . James Rogers & Marc Hauser 13. Over the top – recursion as a functional option . . . . . . . . . . . . Peter Harder

213 233

Part IV. Recursion and the Lexicon 14. Lack of recursion in the lexicon: The two-argument restriction. . Eva Juarros-Daussa`

247

15. Kinds of recursion in Adyghe morphology. . . . . . . . . . . . . . . . Yury A. Lander & Alexander B. Letuchiy

263

16. Recursion and the Lexicon . . . . . . . . . . . . . . . . . . . . . . . . . . . Jan Koster

285

Part V. Recursion outside Syntax 17. A note on recursion in phonology . . . . . . . . . . . . . . . . . . . . . . Harry van der Hulst

301

18. Cognitive grouping and recursion in prosody . . . . . . . . . . . . . . La´szlo´ Hunyadi

343

19. Becoming recursive: Toward a computational neuroscience account of recursion in language and thought . . . . . . . . . . . . . Simon D. Levy

371

20. Recursion in severe agrammatism . . . . . . . . . . . . . . . . . . . . . . Vitor Zimmerer & Rosemary A. Varley

393

Subject index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Language index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

407 417

List of authors Edward Gibson Department of Brain & Cognitive Sciences Massachusetts Institute of Technology Cambridge, MA 02139 USA [email protected] Peter Harder Department of English, Germanic and Romance Studies University of Copenhagen Njalsgade 130, DK-2300 Copenhagen S Denmark [email protected] Marc D. Hauser Depts. of Psychology, Organismic & Evolutionary Biology and Biological Anthropology Harvard University Cambridge, MA 02138 USA [email protected] La´szlo´ Hunyadi Department of General and Applied Linguistics University of Debrecen Egyetem te´r 1. Debrecen, H-4010 Hungary [email protected] Eva Juarros-Daussa` Department of Romance Languages and Literatures State University of New York at Bu¤alo 910 Clemens Hall, North Campus, Bu¤alo, NY 14260-4620 USA ejuarros@bu¤alo.edu

viii

List of authors

Fred Karlsson Department of General Linguistics PB 24 FI-00014 University of Helsinki Finland [email protected] Anna R. Kinsella Language Evolution and Computation Research Unit, School of Philosophy, Psychology, and Language Sciences, The University of Edinburgh 3 Charles Street, Edinburgh, EH8 9AD, UK [email protected] Jan Koster Department of Linguistics University of Groningen P.O. Box 716 9700 AS Groningen The Netherlands J. [email protected] Yury A. Lander Institute of Oriental Studies, RAS Rozhdestvenka, 12 Moscow 107031 Russia [email protected] D. Terence Langendoen National Science Foundation 4201 Wilson Boulevard, Room 1125 Arlington, Virginia 22230 USA [email protected]

List of authors

Ritva Laury University of Helsinki Department of Finnish Language and Literature PL 3 (Fabianinkatu 33) 00014 University of Helsinki Finland [email protected] Alexander B. Letuchiy Russian Language Institute, RAS Volkhonka, 18/2 Moscow 119019 Russia [email protected] Simon D. Levy Computer Science Department Washington and Lee University Lexington, Virginia 24450 [email protected] Marianne Mithun Department of Linguistics University of California, Santa Barbara Santa Barbara, CA 93106 USA [email protected] Tsuyoshi Ono East Asian Studies 3-31 Pembina University of Alberta Edmonton, AB T6G 2H8 Canada [email protected] Amy Perfors School of Psychology University of Adelaide Level 4, Hughes Building Adelaide, SA 5005 Australia [email protected]

ix

x

List of authors

Ljiljana Progovac Linguistics Program Wayne State University 5057 Woodward Detroit, MI 48202 USA [email protected] Geo¤rey K. Pullum School of Philosophy, Psychology and Language Sciences University of Edinburgh 3 Charles Street, Edinburgh EH8 9AD Scotland, UK [email protected] Terry Regier Department of Psychology University of Chicago Chicago USA [email protected] James Rogers Earlham College Richmond, IN USA [email protected] Jeanette Sakel University of the West of England Frenchay Campus Coldharbour Lane Bristol BS16 1QY UK [email protected] Barbara C. Scholz School of Philosophy, Psychology and Language Sciences University of Edinburgh 3 Charles Street, Edinburgh EH8 9AD Scotland, UK [email protected]

List of authors

Eugenie Stapert Max Planck Institute for Evolutionary Anthropology, Leipzig Deutscher Platz 6 04103 Leipzig Germany [email protected] Lawrence Ne¤ Stout Department of Mathematics and Computer Science Illinois Wesleyan University P.O. Box 2900 Bloomington, IL 61702-2900 USA [email protected] Josh Tenenbaum Department of Brain & Cognitive Sciences Massachusetts Institute of Technology Cambridge, MA 02139 USA [email protected] Hans-Jo¨rg Tiede Department of Mathematics and Computer Science Illinois Wesleyan University P.O. Box 2900 Bloomington, IL 61702-2900 USA [email protected] Rosemary A. Varley University of She‰eld Department of Human Communication Sciences 31 Claremont Crescent She‰eld S10 2TA UK [email protected]

xi

xii

List of authors

Arie Verhagen Leiden University Centre for Linguistics (LUCL) Leiden University P.O. Box 9515 2300 RA Leiden The Netherlands [email protected] Vitor C. Zimmerer Department of Human Communication Sciences University of She‰eld 31 Claremont Crescent She‰eld S10 2TA UK [email protected]

Preliminaries

Re Recursion1 Harry van der Hulst ‘‘We hypothesize that FLN only includes recursion and is the only uniquely human component of the faculty of language.’’ (Hauser et al. 2002: 1569) ‘‘If you already know what recursion is, just remember the answer. Otherwise, find someone who is standing closer to Douglas Hofstadter than you are; then ask him or her what recursion is.’’2 ‘‘An apparently new speech disorder a linguistics department our correspondent visited was a¤ected by has appeared. Those a¤ected our correspondent a local grad student called could hardly understand apparently still speak fluently. The cause experts the LSA sent investigate remains elusive. Frighteningly, linguists linguists linguists sent examined are highly contagious. Physicians neurologists psychologists other linguists called for help called for help called for help didn’t help either. The disorder experts reporters SpecGram sent consulted investigated apparently is a case of pathological center embedding.’’3

1. Introduction The present volume is an edited collection of original contributions which all deal with the issue of recursion in human language(s). All contributions (but one4) originated as papers that were prepared for presentation at a conference organized by Dan Everett on the topic of recursion in human 1. I wish to thank the following people for comments on an earlier draft of or ideas contained in this chapter: Jonathan Bobaljik, Marcel den Dikken, Laszlo Hunyadi, Fred Karlsson, Simon Levy, Marianne Mithun, Geo¤rey Pullum, Barbara Scholz and Arie Verhagen. Needless to say that some points that I did not remove from this final version met with strong disagreement. 2. Attributed to Andrew Plotkin (http://en.wikipedia.org/wiki/Recursion). 3. http://specgram.com/CLI.2/03.bakery.disorder.html 4. The exception is chapter 17. In an earlier draft of this introductory chapter I included a section on phonology that I had to remove because it was too long. The reviewers suggested that I turn this section into a separate contribution to this volume, which I did.

xvi

Harry van der Hulst

language (Illinois State University, April, 27–29 2007).5 For the purpose of this collection all articles underwent a double-blind peer-review process. The present chapters were written in the course of 2008. The characterization of language as a potentially infinite number of expressions that can be produced with finite means has been noted for a long time, among others by the linguists Panini and Wilhelm von Humboldt. Chomsky’s early work (e.g. Chomsky 1955 [1975]) proposes various ways to build recursive mechanisms into the grammar (cf. below) and since then many linguists have adopted one of these mechanisms, namely a rewrite or phrase structure component which contains recursive rewrite rules or recursive rule sets (cf. below). However, no general agreement seems to exist concerning the empirical status as well as the formal status of this ‘characteristic’ of human languages or the grammars that underlie them. Renewed interest in this subject was sparked by claims made by or attributed to, Hauser, Chomsky and Fitch (2002) which I paraphrase as follows:6 (1) a. b. c. d. e.

Recursion essentially constitutes the innate human language faculty7 Recursion is the sole uniquely human trait of human language Recursion is unique to the language faculty Recursion is universal (i.e. present in all human languages) Recursion is unique to the human mind

As one might expect, all these bold claims are controversial. According to the first claim, language results from several mental faculties whose intersection leads to language, as well as from necessary ‘natural laws’ of some kind that take scope over language (and, presumably many other phenomena). However the recursion faculty (also called the narrow language

5. Dan Everett wishes to thank Bernard Comrie and the Max Planck Institute for Evolutionary Anthropology and the College of Arts and Sciences and the Provost of Illinois State University for financing this conference. 6. There is some uncertainty on what the authors of this article say precisely and di¤erent interpretations can be found in reactions to it. My focus here is on the kinds of claims that linguists have attributed to the article. 7. Together with the so-called interfaces to the sensory-motor system and the conceptual system, it forms the ‘narrow language faculty’. This claim constitutes an interesting and indeed minimalist interpretation of Chomsky’s Innateness Hypothesis which, originally, had it that human are born with a richly articulated universal grammar.

Re Recursion

xvii

faculty) is properly contained in this intersection. The second claim adds that all other language-relevant faculties than recursion (which together with recursion make up the broad language faculty) can also be attested in non-human animals, while recursion cannot. Pinker and Jackendo¤ (2005) contest the first claim by pointing to other aspects of language that specifically serve language (being confined to the intersection, which goes against claim a). Thus, for them, the innate language faculty contains more than recursion. They also imply that at least some of these other aspects may also be unique to human minds (which goes against claim b). They also note that recursion seems to play a role in other human cognitive systems such as the ‘mathematical module’, or ‘social intelligence’ which runs against claim c. Hauser, Chomsky and Fitch have responded to their article (Fitch, Hauser and Chomsky 2005) which has sparked a further rebuttal by Jackendo¤ and Pinker (2005). With respect to claim (d), Everett (2005) finds that Piraha˜, a Muran language from the Brazilian Amazon, does not exhibit any recursive structures at the syntactic level. His claim has been called into question in Nevins, Pesetsky and Rodriguez (to appear) to which Everett (2007, to appear) is a response. This claim, which has attracted a lot of attention in the professional literature and in the popular press, has contributed to further interest into the matter of recursion.8 Finally, claim (e) has been called into question by Genther at al. (2006) who report on experiments which show that European starlings can be trained to make a distinction between strings that result from recursive or from non-recursive grammars. These findings, which also gained media attention, have received alternative interpretations (cf. Language Log9; Marcus 2006), notably the idea that making a distinction between anbn and random combination of a’s and b’s may point to an ability to ‘count’ (in itself still a remarkable capacity). In fact, Hauser, Chomsky and Fitch (2002) themselves note that animal navigation capacities can be analyzed as containing recursion, which implies that recursion may not be entirely limited to the human mind. In this volume, the question of recursion is tackled from a variety of angles. It is perhaps fair to say that the conference call invited participants to take a critical stance regarding the claims in, or attributed to Hauser, Chomsky and Fitch (2002) and this is certainly reflected in the present

8. See the discussion on the Edge website (http://www.edge.org/3rd_culture/ everett07/everett07_index.html) 9. http://158.130.17.5/~myl/languagelog/archives/003076.html.

xviii

Harry van der Hulst

very interesting collection.10 Some articles cover empirical issues by examining the kinds of structures in languages that suggest recursive mechanisms, focusing on the question to what extent recursive constructions can actually be attested in natural language use. Others focus on formal issues, notably what kind of devices can be used to describe the apparent recursive expressions, and whether the relevant devices have the specific function of making recursion possible or, perhaps lead to recursion as an epiphenomenon. Most articles discuss syntactic phenomena, but a few involve morphology, the lexicon and phonology. In addition, we find discussions that involve evolutionary notions and language disorders, and the broader cognitive context of recursion. In this introductory chapter, section 2 o¤ers a brief discussion of the use of the notion recursion in linguistics and, for the sake of discussion, some further remarks about the role and source of recursion as a morphosyntactic device. Then, in section 3–9 I will review the content of the chapters in the order in which they appear in this volume.11 In some sections I include some additional discussion, particularly in areas which are less well represented in this collection, such as derivational morphology and phonology. Section 10 summarizes what I believe to be the major claims or considerations that are contained in this volume. 2. ‘Recursion as hierarchical grouping’ allows ‘specific recursion’ as a possibility A general problem with the HCF article is that it does not define precisely enough what it means by recursion (cf. Tomalin 2007: 1796). Tomalin (2007) and Parker (2006) clearly show that the concept of recursion and of recursive functions (in linguistics, mathematics and computer science) can be defined in several di¤erent ways. When one gets down to specifics, mathematical formalizations are intricate and inaccessible to most people, including most linguists. In addition, there are di¤erent notions of recursion around and to disentangle their formal di¤erences is, again, largely a 10. Another conference on recursion was held May 26–28, 2009 at the University of Amherst. This conference reflected as less critical view of the centrality of recursion. The centrality of the recursion topic, or the related topic of complexity is further evident from two other conferences that focus on this issue, June 19–20, 2009 and February 24–26, 2010, both in Berlin. 11. The description of the articles’ content is based on abstracts that were provided to me by the authors.

Re Recursion

xix

topic for advanced mathematical minds. Tomalin (2007) presents an overview of the historical background of this notion in generative grammar, pointing to connections to work outside linguistics as well as di¤erent ways in which the notion appears within generative grammar, and, more specifically, within the Minimalist Program.12 He shows how Chomsky, faced with the problem of designing a grammar that could generate an infinite number of expressions with finite means, introduces di¤erent ‘recursive devices’ in di¤erent parts of LSLT (Chomsky 1975 [1955]). In LSLT, chapter 7, it is suggested that a finite set of rewrite rules can be applied more than once. Then in chapter 8 rewrite rules are considered that have the symbol on the left of the arrow also appearing on the right side of the arrow. It is this notion of recursion that became more widely adopted in generative grammar, by Chomsky and others. Thirdly, in chapter 10, the recursive part of the grammar is located in the transformational component where ‘‘the product of a T-marker can itself appear inside the P-base of a T-marker.’’ (LSLT, 516–518; Tomalin 2007: 1793). Tomalin also distinguishes at least five di¤erent notions of recursion and concludes that the type of recursion that HCF refer to (that conforms to the Minimalist Program) is perhaps best characterized as the idea of providing an inductive definition (indeed also called recursive definition) for linguistic expressions. In the MP ‘‘the operations of CHL recursively construct syntactic objects’’ (Chomsky 1995: 226) which means that every syntactic object (i.e. linguistic expression) can be defined in terms of a combination of smaller syntactic objects, with lexical items being the ‘base case’ syntactic objects (that thus terminate a derivation). This characterization of recursion is more general than what most linguists usually have in mind when they define recursion as ‘embedding a constituent in a constituent of the same type’. However, it could perhaps be argued that this latter notion of recursion (which I here will call ‘specific recursion’) is entailed by the more general notion. Let us agree that linguistic expressions (words, sentences) can be analyzed as hierarchically structured object (‘trees’) for which, following LSLT, we can formulate rewrite rules that capture parts of these structures, such as: (2) A % B þ C (often paraphrased as ‘‘An A can consist of a B plus a C’’) 12. Parker (2006) also provides a detailed overview of the way linguists, mathematicians and computer scientists use or define this notion.

xx

Harry van der Hulst

B and C can be atomic units (morphemes in morphology, words in syntax), or they can be combinations themselves (called complex words or phrases). If that latter option is allowed (as it seems to be in linguistic expressions which do have complex words inside (necessarily) complex words and phrases inside phrases) this means that we have rewrite rules of the following sort: (3) a. b.

Phrase % Phrase þ X Word % Word þ X

‘X’ in (3a) can be another phrase or a word, while in (3b) it can be a word or an a‰x. I assume here the idea that X is the head of the phrase to the left of the arrow and will return to the notion of headedness below. The recursive step of an inductive definition, when formalized as a rewrite rule, can be ‘recognized’ by the fact that the symbol on the left of the arrow also occurs on the right. In (3a) and (3b) the identical symbol is ‘word’ or ‘phrase’, respectively. In fact, if the di¤erence between complex words (morphology) and phrases (syntax) is ignored or even denied, there is only one rewrite rule, e¤ectively ‘concatenate’ or, more fashionable: ‘merge’: (4) Linguistic expression % Linguistic expression þ X If we agree that the designation ‘linguistic expression’ literally implies units of the same type (namely the type ‘linguistic expression’) rule (4) meets the classical characterization for recursion: it has the same symbol on both sides of the arrow. When Chomsky says that CHL recursively constructs ‘‘syntactic objects’’ he characterizes such objects as follows (Chomsky 1995: 243): (5) a. b.

lexical items K ¼ {g{a b}}, where a, b are objects and g is the label of K

Starting out with a set of lexical items, the rule in (5b) recursively constructs (binary) units until all members in the ‘numeration’ are dominated by some Kg. With (5a) being the base case that allows the construction to terminate, (5b) is the recursive step (corresponding to 4). Given the ‘definition’ in (5), a complex syntactic object K, if well-formed, can be recursively defined into combinations of (smaller) syntactic objects (which may be complex of simplex). Let us call this notion of recursion here ‘general recursion’. Again, mostly, linguist identify recursion as cases in which a specific rule

Re Recursion

xxi

(from a set of rules) is applied (directly, or indirectly) to its own output, but if there is only one rule (i.e., rule 4), every expression that contains another non-basic expression involves recursive application of that rule. However, the more specific understanding of recursion presupposes the idea of linguistic expressions being of a variety of di¤erent categories. Words are nouns, verbs, prepositions etc., while phrases are noun phrases, verb phrases, prepositional phrases, etc. Thus, instead of the rule in (4) we have rules as in (6). (Having linguistic expressions of di¤erent categories allows us to formalize the idea that linguistic expressions have heads that determine the category): (6) a. b.

XPhrase % YPhrase þ X XWord % YWord þ X

This allows cases in which X and Y are identical, which then presents the ‘Russian Doll’ e¤ect: a phrase of type X within a phrase of type X, or a word of type X within a word of type X. The di¤erence between general recursion and specific recursion is exemplified in the following quote from Pinker and Jackendo¤ (2005: 10): ‘‘. . . (As mentioned, HCF use ‘‘recursion’’ in the loose sense of concatenation within hierarchically embedded structures). Recursion consists of embedding a constituent in a constituent of the same type, for example a relative clause inside a relative clause (. . . .). This does not exist in phonological structure: a syllable, for instance, cannot be embedded in another syllable.’’

There is, however, no contradiction between these two characterizations of recursion: the ‘specific recursive case’ is an automatic result of having general recursion, which means there is no need to regard specific recursion as a basic property. What is basic is the possibility of containing non-atomic objects within larger non-atomic objects of the same complexity and thus the idea of unbounded hierarchical structure. In a system of this sort specific recursion occurs unless it would be explicitly blocked. In other words, if phrases are allowed to occur inside larger phrases then finding a noun phrase inside a noun phrase is an expected consequence: (7) NP % NP þ X

(where X ¼ N, the head)

Rule (7) is a specific case of rule (6a). All things being equal, there is no reason to complicate a grammar by blocking rules like (7), especially if expressions occur which seems to be the result of rules like (7). Rule (7) produces direct specific recursion, i.e. a NP which contains a

xxii

Harry van der Hulst

NP as one of its daughters. We also expect indirect specific recursion: (8) a. b.

S % NP þ VP VP % V þ S

In this case S contains another S as a granddaughter. The nodes linked in terms of indirect recursion can be separated by an indefinite number of intermediate nodes. Both general and specific recursion are possible because the complex expressions that can be contained in (necessarily) complex expressions are of the same complexity type (although possibly belonging to di¤erent categories). A di¤erent kind of system would result if the general schema for rewrite rules would be as in (9): (9) Phrase nþ1 % Phrase n þ X In this case, self-feeding is excluded because each application introduces a new type of expression. Rewrite rules of this kind, in the more specific form of (10), have been suggested to underlie the so-called prosodic hierarchy (see van der Hulst, this volume): (10) Phrase nþ1 % Phrase n þ Phrase n An indeed, given this rule format, prosodic structure would not be recursive. This also drives home the point that hierarchical structure as such does not entail recursion. In conclusion, general recursion (‘merge’) makes specific recursion a possibility which in turn makes it possible that grammars, being themselves finite, can generate an infinite number of linguistic expressions. In other words, recursive mechanisms are held responsible for the apparent discrete infinity of natural languages in the sense that when languages are thought of as sets of expressions, these sets are infinite. This is what Pullum and Scholz (this volume) call the infinitude claim. The infinitude claim also involves the idea that there is no limit on the potential length of linguistic expressions (‘there is no longest sentence’). However, whether it is ‘true’ that languages cannot be said to have a finite number of expressions is an empirical question, although, as Pullum and Scholz show, linguists generally assume that the infinitude claim is true. And therefore, these linguists design grammars that have recursive mechanisms. Tomalin (2007: 1797–1798) notes that ‘‘if the sole requirement is to generate an infinite number of structures using finite means, then an itera-

Re Recursion

xxiii

tive, rather than a ‘recursive’, process could accomplish this, and while such a procedure may be less e‰cient than a ‘recursive’ procedure, the basic point is that a requirement for infinite structures using finite means is not itself su‰cient to motivate the use of specifically recursive procedures.’’ In defense of recursion it could be said that iteration does not adequately capture the recursive nature of the semantic structure that syntactic objects are supposed to encode. In other words, one could claim that syntax uses recursion because recursive syntactic structures express complex thoughts which themselves display this kind of combinatorial capacity. Note that, if this is so, then this specific recursive capacity is not unique to (morpho)syntax because it also characterizes our ‘conceptual constellations’ (using this as another term for our ‘thoughts’). The conceptual structure of ‘a bird in a tree on the hill’, which itself displays recursion (indeed specific recursion), if needed to be expressed, ‘inspires’ or ‘drives’ a syntactic system that iconically builds similar structures, i.e. structures that allow complex entities to be part of even larger complex entities. The claim that morphotactics uses recursion to directly express the recursive nature of conceptual structures is controversial for those, like Chomsky, who regard syntax as an autonomous system and as such, the core of human language.13 But autonomous does not have to mean unmotivated (or ungrounded).14 Seeing syntactic recursion in this light allows the possibility that iteration (or indeed other mechanisms such as intonation; cf. below) could be used as conventional ways of expressing recursive conceptual structures. Syntactic recursion (if interpreted realistically, i.e. as forming part of the grammars that people have in their heads) may be the best solution because of the achieved iconic isomorphy between syntactic and conceptual structure, but it need not be the only one. This, indeed, is the essence of Everett’s claim about Piraha˜ (Everett 2005, 2007a,b, 2008). The above reasoning (controversial as it may be) locates the source of recursion in the general recursive structure of the conceptual system. It is sometimes suggested that there is a particular conceptual basis for this kind of conceptual recursion, namely the human theory of mind which suggests a kind of embedding: (11) {I think {that she thinks {that he thinks. . . }}}

13. See Uriagereka (2009) for a defense of the idea that the relationship between syntax and semantics works in the other direction: syntax constructs semantics. 14. The claim that syntactic structure is semantically grounded is central to Anderson’s notional grammar (e.g., Anderson 1997).

xxiv

Harry van der Hulst

However, it could be said that the conceptual structure ‘a bird in a tree on the hill’ which does not presuppose a theory of mind has the same kind of embedding: (12) {a bird {in a tree {on the hill}}} In this case, however, conceptually, the tree is not contained in the bird, nor is the hill contained in the tree. If anything, the containment runs in the other direction: the hill, as the larger structure, contains the tree and that tree contain the bird. In any event, whether or not the (emergence of a) theory of mind, underlies conceptual recursion, we do not have to make a fuss about conceptual recursion either; it simply results if the conceptual grammar allows hierarchical grouping. If recursion is a side e¤ect of allowing hierarchical grouping that allows units of the same complexity type to occur inside each other, this applies as much to the conceptual grammar as it does to the syntactic grammar.15

3. Types of specific recursion In addition to the distinctions made in the previous section, various types of specific recursion are usually distinguished; cf. Parker 2005 and Kinsella [Parker], this volume, Karlsson, this volume and Verhagen, this volume. An important distinction is that between nested recursion (center-embedding, central embedding, self-embedding) and tail-recursion (the latter covering left-recursion and right-recursion). It is also important to contrast recursion with iteration. Karlsson (this volume) distinguishing six types of iteration (structural iteration, apposition, reduplication, repetition, listing and succession). Let us first illustrate and discuss nested recursion. (13) Nested recursion [ The man [the boy [the girl kissed] hit] filed a complaint] This ‘classical type of example’ is, however, quite atypical of recursion in language because not a single genuine one has ever been attested (Karlsson, this volume). But other cases of nested recursion do occur; cf. below. In 15. Note that if the grammar of ‘conceptual objects’ is recursive and if this ‘conceptual system’ is not seen as part of ‘universal grammar’, recursion cannot be unique to CHL if this system only comprises recursive syntax.

Re Recursion

xxv

nested recursion, a constituent occurs ‘in the middle’ of another constituent such that the latter constituent has material on either side of the former constituent, which makes the latter a discontinuous constituent. Centerembedding can be further di¤erentiated in several subtypes (cf. De Roeck at al. 1982; Sampson 2001: 13–14), depending on the degree of identity of the embedded units and the unit that contains it. One degree of centerembedding passes by unnoticed, but two degrees or more are often considered ‘di‰cult’ and Reich (1969), in fact, argued that sentences with center-embedding are ‘ungrammatical’; see also Christiansen (1992). Whether or not this is so, it is true that nested recursion is notoriously di‰cult to process if it exceeds one instance of embedding, although some aren’t all that di‰cult (cf. De Roeck et, al. 1982; Thomas 1995), e.g.: (14) [The fact [that the teenager [who John dates] was pretty] annoyed Suzie] In De Roeck et al. (1982), entitled ‘A myth about center-embedding’, it is reported that spontaneous, and apparently perfectly acceptable cases of multiple center-embedding do exist. Sampson (2001) reports various additional real-life cases such as:16 (15) a. b. c.

[but don’t you find [that sentences [that people [you know] produce] are easier to understand]? [the only thing [that the words [that can lose –d] have in common] is, apparently, that they are all quite common] [The odds [that your theory will be in fact right, and that the general thing [that everybody’s working on] will be wrong,] is low]

The first sentence was constructed by Anne de Roeck and then posed to Sampson as a question after he had just claimed, following the dogma at the time, that multiple center-embedding is rare and di‰cult to process. After having collected several examples in the de Roeck study and in his own later study, Sampson concludes that, although his data are not based on language corpora, multiple central embedding is certainly not ungrammatical and in many cases not that di‰cult to find and understand after 16. Marcel den Dikken (p.c.) notes that all three examples have a copular verb in the root sentences which, perhaps makes them easier than examples with noncopular verbs. He also notes that in (15c) the copular does not agree with its subject, a possible indication that the speaker of this sentence lost track of its structure.

xxvi

Harry van der Hulst

all. This leaves open the question what the di¤erences are between cases that are not that di‰cult, even having three degrees of embedding, and cases that apparently are, such as the apparently short and simple artificial examples in (16): (16) a. b.

[The man [the boy [the girl kissed] insulted] left] [I met the man [who saw the girl [who left this morning] this afternoon] two minutes ago]

I refer to Thomas (1995) for a review of attempts to narrow down the properties of constructions that pose clear problems and to Karlsson, this volume. Below I return to nested embedding. But first, we turn to tail recursion. (17) Tail recursion a. Left edge (creating a left-branching structure) [[[John’s] sister’s] dog’s] bone was found in the yard b. Right edge (creating a right-branching structure) [This is the car [that hit the boy [who crossed the street]]] We can clearly see that a critical di¤erence between nested and tail-recursion is that the former involves long distance relationships (i.e. relations between two units that are separated by other units), while the latter does not. This di¤erence, according to many, explains why nested recursion is much more di‰cult to process. It is usually said that tail recursion does not pose problems, although it seems to me that (17a) is certainly harder than (17b); cf. Karlsson, this volume. Informally, the di¤erence is that while we can interpret each new relative clause locally as applying to the adjacent noun, (17a) requires us to accumulate the result of each combination so that the genitive ‘s can be interpreted as a property of the steadily growing combination. Returning to nested recursion, as displayed in the sentences in (16), it is often said that such cases involve a string of the form anbn, which can be generated by the context-free grammar in (18): (18) a. b.

S % aSb S % ab

(18) generates a string in which ‘S’ occurs in the middle of a constituent S. However, it is not clear at all that grammars of real languages have rules like (18a). The rule that is responsible for the center-embedding in (16) is the same rule that causes tail-recursion in (17b): (19) NP % NP þ S

Re Recursion

xxvii

However, the e¤ect of (19) is di¤erent depending on whether it expands an NP that is in subject position or a NP that is in object position. At least in most current syntactic frameworks, there is no mechanism to generate a constituent that is literally in the middle of the constituent that immediately contains it if constituents are always binary (which would disallow rules like 18a). Thus (16a) has something like the following structure: (20)

(15b) on the other hand has the following structure: (21)

xxviii

Harry van der Hulst

In other words, the bracketing in (16) is incorrect. As shown in (20) the recursive node S does not occur inside any constituent, but rather at the right edge of Noun Phrases. What this means is that there is no special mechanism for center-embedding (such as the rule in 18a). Rather centerembedding results in the following situation: (22) a. b.

A % BþC C % AþC

c.

d.

C is a right hand expansion of A while A is a left hand expansion of C (or vice versa)

Whenever we have two rules that meet the criterion in (22d) we have what is called center-embedding. What this means is that once special recursion is available (which it is once groupings can involve groupings), centerembedding is available too, unless rule pairs that meet criterion (20) would be explicitly blocked. The special feature of center-embedding that makes the relevant constructions hard to process, is that it creates discontinuous (long-distance) relationships between B’s and C’s going from both edges inward. However, it should be noted that this feature is logically independent of centerembedding; see also Verhagen’s contribution to this volume. The structure in (23) has no special recursion, thus no center-embedding, yet it creates three discontinuous relationships. Even though this structure is obviously ungrammatical in English it is so because the head complement relations, all being final, are non-English. One would think that in a language with the appropriate head final structure, an example of this sort could be constructed and that it would pose processing di‰culties, just like selfembedding structures do. Focusing on tail-recursion, several scholars have argued that relevant patterns can be analyzed as instances of iteration (Ejerhed 1982, Pulman

Re Recursion

xxix

(23) a.

b.

1986, Parker 2005), and Reich (1969) supports this idea with intonational evidence. Christiansen (1992) draws the conclusion that there are no linguistic phenomena that crucially require recursion if center-embedding is ungrammatical and tail-recursion can be handled with an iterative device.17 However, the claim that center-embedding is ungrammatical as such is wrong since depth 1 certainly is fine as in ‘‘The car I bought cost 1000 dollars’’. Because it is always said that phonology is not recursive, it is perhaps interesting to apply a similar argument in this domain. Assume for the sake of the argument that a word can consist of any number of feet. If true18, there would be an infinite number of possible word forms. However, that does not necessarily require a recursive device. The relevant

17. As just shown, the distinction between self-embedding and tail-recursion may be irrelevant. If specific recursion is blocked (and replaced by an iterative procedure) both phenomena are ruled out. 18. See section 9 and chapter 17 on this matter.

xxx

Harry van der Hulst

property can be characterized by a rule schema that can be collapsed into one rule using an abbreviatory device: (24) Word % Foot* In actual fact, however, a more interesting characterization of the phonological structure of words might appeal to a recursive definition: (25) Phonological words (recursive definition) a. Word % foot (base case) b. Word % Word þ Foot (inductive step) The two approaches would create di¤erent structures: (26) a.

Iterative definition

b. recursive definition

Both types of representation have in fact been proposed. (24b) has been proposed in Metrical Theory (Liberman and Prince 1977), whereas the flat representation in (26a) has also been proposed (e.g. Halle and Vergnaud 1987) in the guise of so-called bracketed grids. The question is which representation accounts best for the properties of phonological words. (26b), for example, can be said to express di¤erent degree of prominence of feet, about which (26a) has nothing to say. In other words, just like semantic properties may be more adequately expressed in recursive morphotactic structures, phonetic properties may be more adequately expressed in recursive phonotactic structures. These issues are discussed in chapter 17. Even without considering whether the semantic and phonetic substance that underlies morphotactic and phonotactic structure is recursive, we must ask whether the properties of linguistic expressions which suggest that recursive devices are called for must necessarily give rise to these devices. In section 2, I referred to this issue when I said that a recursive structure in morphotactics may be optimal as an expression of the presumed recursive nature of conceptual structures, but this does not mean that morphotactics actually achieves the desired isomorphy in all cases, or, in some languages, at all. It may be that the morphotactics is using an iterative device, which puts a bigger burden on semantic interpretation

Re Recursion

xxxi

because the recursive semantics now has to be reconstructed from an iterative morphotactic structure. The same applies to phonotactics. Even if one could argue that the phonetic substance is, in some sense, recursive (in fact, however, I will argue in chapter 17 that it probably isn’t), it does not follow that the phonotactics is capable of mimicking this. I briefly return to the question of recursion in phonology in section 9 and more extensively in chapter 17 in this volume. Several authors in this volume indeed explicitly ask whether the apparent recursive morphotactic structures necessarily require a formal recursive characterization. A crucial (although perhaps all too obvious) point that must be born in mind here is that the tactic devices that a linguist proposes (whatever their formal properties) are hypotheses or models for the system that humans have or use to produce and understand expressions (Itkonen 1976, Tiede and Stout, this volume), i.e., if a realist stance is adopted, rather than an instrumentalist stance (cf. Carr 1990 for discussion). These hypotheses/models may involve a formal system that uses recursion but this does not entail that the cognitive tactic system used by real people also has recursion, even if we all agree that the conceptual structure that the morphotactic structures encode is recursive. After all, the data that lead the linguist to suspect that the morphotactic system is recursive may also arise from another kind of system such as iteration, supported by intonation devices; cf. Stapert and Sakel, this volume for some explicit ideas.19 What the above reasoning misses, though, is the point we discussed in section 2 namely that specific recursion is formally available once we admit that the syntax builds hierarchical structures of a certain kind (cf. 6) which is a claim that perhaps most linguists will agree on. And if recursion is free, then self-embedding is, in principle, available, since it uses the same mechanism as tail-recursion. This means that if one wants to question the modeling choice that entails recursion, one would have to reject the idea that the context-free grammars are required for natural languages (assuming that finite state automata are su‰cient).20 However, this being said, we must also accept that constructions that employ specific recursion, espe19. By the same reasoning we can also not take for granted that semantic stu¤, or phonetic stu¤ is or isn’t recursive. That too is a modeling choice. See section 3 for some discussion of the question as to whether the conceptual system allows self-embedding. 20. One could also reject constituent-based grammars and turn to pure dependencybased grammars which have no hierarchical constituent structure.

xxxii

Harry van der Hulst

cially when leading to center-embedding e¤ects, are clearly avoided in natural languages, since it creates long-distance relationship which call for special computational machinery involving a stack. Whether the causes of this avoidance are ‘dismissed’ as performance factors, seriously studied as processing constraints, or even integrated into the (competence) grammar (if a distinction between ‘competence’ and ‘perfomance/processing’ is maintained) is another matter. I return to these issues, which are also discussed in various chapters in this volume, in the next sections.

4. Discussing the need for recursion on empirical grounds Even though all these formal issues are obviously pertinent, we start this collection with a series of chapters that take an empirical perspective. At the empirical level, the question needs to be raised whether (all) languages display recursive structures since it has been claimed that some languages do not have recursive structures at all, or employ (some kinds of ) recursion very modestly. Also, we need to know what the precise properties of these recursive structures are. Some of these issues were already raised in the previous section. In 1995, the linguist Richard Hudson posted the following question on the Linguist List: Does anyone know of any *empirical* investigations of center-embedding examples (aka self-embedding) such as the following? (1)

The dog the stick the fire burned beat bit the cat.

I’ve found lots of discussions (especially in introductory books – the above example is from Pinker’s ‘The Language Instinct’), but no experimental data (or any other kind of data). There’s no shortage of explanations of the ‘facts’, but there does seem to be a shortage of well-established facts to be explained. (Richard Hudson on Linguist List (4 December 1995).21

Several chapters in this volume provide a di¤erent answer to Hudson’s question than Sampson’s response which I mentioned in the previous section. For languages that have been investigated for the property of recursion using language corpora, it has been found that recursion of more than 21. http://www.linguistlist.org/issues/6/6-1705.html

Re Recursion

xxxiii

one or two steps is far from common. In spoken language, Karlsson (2007a,b; this volume) finds that self-embedding of degree 2 is virtually non-existent (which apparently contradicts Sampson’s admittedly more impressionistic findings). Tail-recursion is not uncommon, especially rightwards, although here actual language data do not show much more than degree 2 recursion. It has of course long been observed that center-embedding of degree 2 and anything higher, even though claimed to be perfectly grammatical (conforming to the design of the linguistic ‘competence’; Chomsky and Miller 1963), is rare.22 This raises the reasonable question how something so marginal can be taken to be so foundational (even definitional as in Hauser, Chomsky and Fitch 2002) for human language. If recursion is a defining feature of human language, as has been claimed, we would expect to find evidence of it in everyday talk, the primary form of language. Chomsky famously asserted that language was not designed to be produced or parsed (understood), meaning that language is not designed to be used as a communication system. Rather it is a system to facilitate thought. Additionally, it is said that language production and parsing belong to ‘performance’ and it is here, as Chomsky claims, that we find the limitations that limit the use of center-embedding. A factor that must be recognized in evaluating the ‘clash’ between the rarity of recursion and its alleged central role in language is that studies of the actual occurrence of recursion focus on specific recursion while Chomsky’s more recent claims about the centrality of recursion seem to emphasize general recursion, i.e. hierarchical grouping. Bearing this point in mind, let us ask, for the sake of discussion, language why should be recursive in the first place. If syntax emerged and exists, primarily, or perhaps exclusively, to facilitate thought, why was recursion required, if we assume that the conceptual system itself had its own syntax (which, I believe, should be an uncontroversial assumption)? Certainly Chomsky does not equate what he calls ‘syntax’ with the ‘syntax of conceptual structures’.23 So, if the two are di¤erent, why does the mind need ‘syntax’ only to duplicate conceptual syntax? What is wrong with the syntax of our conceptual system? This is the point where (most) other 22. Rather than evaluating the degree of embedding, Davis (1995), in particular considers di¤erent types of center-embedding in terms of whether, for example, a relative clause occurs inside a sentential complement, or the other way around. 23. Some authors such as Burton-Roberts (2000) think that Chomsky should. For Burton-Roberts indeed ‘syntax’ is the syntax of thought(s).

xxxiv

Harry van der Hulst

linguists would say that ‘syntax’ did not emerge to organize our conceptual system, but to express or externalize it (cf. Jackendo¤ 2002). Together with what we call phonology, syntax is a mechanism to relate conceptual structure to utterances that can be produced and perceived. In other words, in this view, syntax (and phonology) did emerge to make communication (or at least, externalization) possible. Putting this issue aside as well, and accepting that syntactic objects are distinct from conceptual objects, we must again ask why it is necessarily the case that syntax is formally set up to allow recursion and thus selfembedding? It is said that syntactic objects with self-embedding are di‰cult to process (in performance), but what warrants their existence in competence (if that distinction is made)? Perhaps, as one might say, the problem is not that humans can’t process self-embedding, but rather that our syntactic system does not have this formal trait to begin with. Conceptually, we can understand self-embedding very easily, for example when it is based on visual information. We can see, and understand a circle that contains another circle which contains another circle and so on. Hence our conceptual system permits self-embedding with no apparent limitation, but, playing the devil’s advocate, one might say that there is no evidence that the syntax can mimic that, and if that is so, and the distinction between competence and performance is accepted, we cannot process it either. We will not try to produce it because the grammar does not deliver the appropriate input to the processing systems, and we cannot parse artificially produced examples because no parse will lead to a result that is syntactically well-formed. Clearly, we cannot here resolve all these issues, but I remind the reader that this line of reasoning does not take into account that specific recursion and thus self-embedding come for free once one admits that grammars generate hierarchical objects with groupings inside groupings of the same complexity. Although, as we have just experienced, empirical and formal issues are di‰cult to separate sometimes (among others because ‘data’ always imply an analysis of some sort), the following five chapters focus on empirical issues. These chapters show that (a) recursive structures are not very common and (b) where apparently occurring they may result from specific templatic constructions involving specific lexical items (often derived from clearly non-recursive constructions) rather than abstract recursive mechanisms. Karlsson reports that multiple nested syntactic recursion of degrees greater than 3 does not exist in written language, neither in sentences nor

Re Recursion

xxxv

in noun phrases or prepositional phrases. In practice, even nesting of degree 2 is extremely rare in writing. In speech, nested recursion at depths greater than 1 is practically non-existing. Left-branching tail-recursion of clauses is strictly constrained to maximally two recursive cycles. Rightbranching clausal tail-recursion rarely transcends three cycles in spoken language and five in written language. On constituent NP or PP level both left- and right-branching is less constrained (especially in written language), but e.g. left-branching genitives rarely display more than two recursive cycles ([[[Pam’s] mum’s] baggage]). Laury and Ono also supply data on the use of recursion in real speech. However, they first discuss certain methodological problems that arise when one tries to establish the use of recursion in conversation. First, there is the tendency for certain types of main clauses to become grammaticized as discourse particles, in which case it is not clear whether a given item should be counted as a particle or a clause. Secondly, the authors discuss the problematic nature of the category ‘sentence’. In spontaneous spoken language, it is not always possible to tell whether a clausal unit is embedded within another, because conjunctions also have uses as independent discourse particles and also because two speakers may be involved in the production of a single sentence. Both of these problems challenge analysts who try to identify recursion in conversation. The data that Laury and Ono present show that clausal recursion is a strictly limited phenomenon in spoken language. Embedding beyond the depth of two is extremely rare, and the upper limit of clausal embedding is four. Embedding of a clause within a clause of the same type, typically used to illustrate recursion in the literature, is a vanishingly rare phenomenon. In fact, as their data show, speakers of Japanese and Finnish do not seem to be constructing complex clause combinations but rather joining clauses together one at a time. In other words, the authors conclude, recursion appears to be irrelevant to what actual speakers do. They then suggest that a more obvious characterization of clausal embedding is to simply say that speakers know, for example, how to quote, how to identify and describe referents, and how to give reasons, all in the form of one clause combination at a time. They conclude that the nature of clause combining and the limited extent of clausal embedding in ordinary conversation casts doubt on the status of recursion as a defining feature of human language. Recursion, they say, at least as far as Finnish and Japanese are concerned, may be only a linguist’s category, and not supported by the primary form of language. This remark relates to the ques-

xxxvi

Harry van der Hulst

tion raised earlier concerning the distinction between the actual workings of language (which, unfortunately cannot be established objectively) and the way that linguists try to model these mechanisms. Whereas the two previous chapters report on the limited use of recursion in the investigated languages, the chapter by Stapert and Sakel, drawing on their own fieldwork data, tests Everett’s (2005) claims about Piraha˜ having no subordinations or other syntactically recursive structures. They consider possible alternatives for languages to express complex thought and compare their findings from Piraha˜ with discussions in the recent literature (such as the two above-mentioned chapters) on spoken versus written language and formulaic language use. They also consider the possibility that recursive structures could have been borrowed from Portuguese, with which Piraha˜ is in contact. They argue, finally, that complex ideas can be expressed by other means than syntax, and therefore that syntactic recursion may be common (relative to what we learn from the above-mentioned studies), but not universal in human language. Mithun examines a range of complement and adverbial clause constructions which could or have been characterized as displaying recursion in three genetically and distinct languages. Examples of older constructions are drawn from Central Alaskan Yup’ik Eskimo. The Yup’ik constructions are pervasive in speech and deeply entrenched in the grammar, signaled morphologically. Their origins can still be detected, however, in nominalized clauses. The adverbial clauses are marked by subordinating su‰xes descended from case markers. Somewhat younger embedded constructions can be seen in Khalkha Mongolian. A pervasive complement structure, descended from a quotative construction, is marked by a particle homophonous with a non-finite form of the verb ‘say’. This complement construction is no longer restricted to use with utterance verbs like ‘say’; it occurs with a range of other verbs as well. Still, it has not yet been extended to commentative verbs (‘be sad’, ‘be significant’), modals (‘be able’, ‘should’), achievements (‘manage’, ‘try’), or phasals (‘start’, ‘finish’). The author also discusses various adverbial clause markers in Khalkha that are descended from case su‰xes: locative (‘at N’, ‘when S’), instrumental (‘with N’, ‘as a result of S’, ‘in order to S’), and ablative (‘from N’, ‘because of S’). Examples of the youngest embedding constructions are drawn from Mohawk. Mithun says that it may seem that complement constructions have not yet arisen in the language: Mohawk speakers use simple sequences of sentences where speakers of other languages would use embedding. She then addresses the important role of prosody, showing that once prosody is taken into account, clear patterns

Re Recursion

xxxvii

of embedding are easy to identify. Other young complement constructions include demonstratives or definite articles. Again, a consideration of the prosody shows how they originated and evolved, along pathways di¤erent from those usually assumed for complements marked with demonstratives in Germanic languages. Temporal adverbial clauses appear at first to be lacking as well. Mohawk speakers seem to simply add a sentence beginning with a particle ‘at the time, now, then’, where speakers of other languages would use dependent adverbial clauses. Here once more, prosody shows that they are integrated into larger sentences. Mithun concludes that the variability in space and time calls into question the status of recursion as the basic design feature of human language. It suggests instead that recursive structures are epiphenomenal, the product of combinations of a variety of cognitive processes such as the routinization of frequentlyused structure combinations, the reification of events, and the generalization of functions of markers and constructions to ever more abstract contexts. Mithun concludes that recursive structures that occur in human languages are not uniform cross-linguistically, nor are they static within individual languages. This variety indicates, she argues, that recursion may not be the fixed, fundamental, hard-wired property envisioned. Verhagen sets out to characterize the notion of recursion from an empirical perspective and, in so doing, he shows that the role of ‘recursion’ is rather overestimated. He distinguishes between two di¤erent notions that are related but not identical, and that have played a role in di¤erent stages of 20th century theoretical linguistics: (27) a.

b.

‘Di¤erent parts of a phrase may be separated by other (indefinitely long) phrases’, e¤ectively the notion of ‘long distance dependency’. ‘The specification of certain phrases requires the application of a rule to its own output’.

He claims that if (and only if ) both cases occur in combination we have what computer scientists call ‘true recursion’ (i.e. center-embedding), which requires a special kind of computational architecture, in order to keep the intermediate results of the calling procedure in memory while the embedded instance is being executed. If situation (a) does not hold, embedded phrases occur at the ‘edges’ of embedding phrases and processing architecture capable of handling iteration su‰ces, because special architecture to store and retrieve intermediate results is not required (although their description may still involve a recursive rule).

xxxviii

Harry van der Hulst

Using the above distinctions, Verhagen critically examines three phenomena that have been taken as instantiating recursion in the sense of sentential embedding, viz. embedding of adverbial clauses in other adverbial clauses, nonfinite complementation in causative constructions, and longdistance Wh-movement (cf. examples (26a–c) respectively): (28) a. b. c.

[ S Because our flight turned out to be cancelled [ S when we arrived in Madrid]], . . . He made [ S me understand the situation] Whoi did Mary say [ S that John kissed ti ]

On the basis of actual usage data, Verhagen shows that none of these classic cases actually requires a truly recursive specification. Empirically, a system that uses relatively specific templates is at least indistinguishable from one using general recursion, and is in some respects even more adequate (which has obvious consequences for the issue whether recursion can have been a target of selection). He does acknowledge that recursion is relevant for grammar for some ‘pockets’ of rather specific phenomena, adding that it may very well have been produced by cultural evolution (involving literacy) rather than genetic evolution.

5. Discussing the need for recursion on formal and functional grounds The last three mentioned chapters suggest that many apparently recursive constructions may be the result of specific templates (which figure specific lexical items), rather than free-wheeling recursive mechanisms. As mentioned, this raises a question concerning the di¤erence between formal properties of grammars and the languages that they model. In the following five chapters, this issue is discussed from formal and functional perspectives. Pullum and Scholz start out observing that certain remarks in the linguistics literature over the past few years suggest that some linguists think of infinitude (i.e. the infinity of language if thought of as a set of expressions generated by a grammar) as a universal. That is, these linguists, they say, believe that it has been empirically established that there are infinitely many grammatical expressions in human languages, and that we need recursion in grammars in order to account for this. Pullum and Scholz examine the arguments given for the infinitude claim, and show that they depend on an unwarranted assumption: that the only way to

Re Recursion

xxxix

represent syntactic properties is through a generative grammar with a recursive rule system. They then explore some of the reasons why linguists have been so willing to accept language infinitude despite its inadequate support and its lack of linguistic consequences. These authors suggest that the infinitude claim seems to be motivated chiefly by a lingering adherence to the outdated notion that languages should be regarded as sets. It is not motivated by considerations of the creative aspect of language use, or opposition to associationist psychology, or the putative universality of iterable linguistic structure such as recursive embedding or unbounded coordination (which are probably not universal anyway). Langendoen examines Pullum & Scholz’s argument that there has never been an adequate demonstration for the claim that natural languages contain infinitely many expressions, and concludes that they are correct in asserting that the question remains open. He then proposes a method whereby it can be determined at least under certain conditions whether a language has infinitely many expressions, and finally assesses the claim in Postal & Langendoen (1984) that natural languages contain transfinitely many expressions. As discussed in section 1, Hauser, Chomsky and Fitch (2002) claim that a core property of the human language faculty is recursion and that this property yields discrete infinity of natural languages. On the other hand, recursion is often motivated by the observation that there are infinitely many sentences that should be generated by a finite number of rules. According to Tiede and Stout it should be obvious that one cannot pursue both arguments simultaneously, on pain of circularity. The main aim of their chapter is to clarify both conceptually and methodologically the relationship between recursion and infinity in language. They argue that discrete infinity is not derived, but a modeling choice. Furthermore, many arguments, both for recursion and infinity in language, crucially depend on particular grammar formalisms. Thus, care should be taken to distinguish, on the one hand, whether to derive infinity from recursion or the other way around, and, on the other hand, the role of recursion in language in general from the role of recursion in specific grammar formalisms. Pursuing a similar line of inquiry, Perfors, Tenenbaum, Gibson and Regier state that recursion involves an inherent tradeo¤ between simplicity and goodness–of–fit: a grammar with recursive rules might be simpler than one without, but will predict the sentences in any finite corpus less exactly. As a result, one cannot conclude that any particular grammar or grammatical rule is recursive, given a corpus, without some way to quantify and

xl

Harry van der Hulst

calculate this tradeo¤ in a principled way. They present a Bayesian framework for performing rational inference that enables us to quantitatively evaluate grammars with and without recursive rules and normatively determine which best describe the sentences in a corpus of child-directed spoken English. Their results suggest three main points. First, they suggest that rational principles would favor a grammar with a specific type of recursive rule, even if there are relatively few instances of particular recursivelygenerated sentences in the input. Second, they suggest that the optimal grammar may occupy a representational middle ground between fully recursive and non-recursive. Finally, their results suggest that the optimal grammar may represent subject NPs distinctly from object NPs. According to Harder functional linguists tend to think that recursion as an issue in linguistics is an artifact of a Chomskyan formalism, i.e. from a functionalist point of view, the mathematical properties of models used to describe languages do not automatically qualify as features of real languages. Like everything else, human languages can be simulated by a formal model, but, as we have seen several times, that does not entail that the properties of the model are also properties of the language. However, at the same time, functional linguists do not dispute that languages like English possess syntactic mechanisms that can be formally modeled by recursion. For a functionalist linguist, this then poses the challenge of considering whether recursion has a functional role in language. Recursion, he says, may not be the right way to account for linguistic creativity, but functionalists need to o¤er their own take on the issue – including the question strikingly raised by Piraha˜ (cf. Everett 2005), of what the precise relations are between recursion in culture, cognition and language.

6. Evolutionary Perspectives The HCF paper also addresses the issue of language evolution, a subject that has moved to the center of attention during the last couple of decades, despite Chomsky’s earlier pessimism that anything significant could be said about this subject. By proposing that the innate human capacity for language is not a ‘richly articulated’ mental system (as used to be the case in generative grammar), but a rather minimal system that merely contains a recursive device (in the sense of general recursion; cf. 5), the question of how language came about is considered more manageable by HCF. Human language could have come about abruptly from a simpler

Re Recursion

xli

word-based system (allowing one-word utterances only) because the only thing that is needed is to add a recursive device (cf. Reuland 2009). Hornstein (2009) suggests an even more specific mechanism that made recursion possible: endocentricity or headedness, or what he calls ‘labeling’. Turning back to the rules in (6), while X and Y may be identical (which produces specific recursion), it is usually claimed that X and Z must be identical, which is to say that the category label of a complex syntactic object is projected from one of the daughters, called its head. If a complex word is a noun, one of its daughters must be a noun as well (in derivational morphology this would be a noun-making a‰x, while in compounds it would be simplex or complex word), and if a phrase is a noun phrase it must contain a word of the category noun. According to Hornstein (2009: 59–60) it is the emergence of this specific labeling convention (i.e. projection from the head) which kick started recursion in the evolution of language: (29) a. b.

XPhrase % YPhrase X (e.g., NP % APN) XWord % YWord X (e.g., N % VNaff )

Indeed, it would seem that this rule meets the criterion of being recursive on the understanding that the label of a head and the label of its mother node are strictly identical, which means that we must ignore or eliminate the di¤erence between phrase labels (XP) and word labels (X) and thus replace (28a) by (28b): (30) a. b.

XPhrase % YPhrase X X % YX (e.g., N % AN)

However, by locating recursion in this labeling convention, as Hornstein does, the recursive symbol on the right side of the arrow is the head of the expansion, whereas the usual approach, while accepting the idea of headedness, locates recursion in the dependent (or complement) to the head: (31) a. b.

VP % VPV V % V1V2

If (31b) is the counterpart of (31a), by virtue of eliminating the di¤erence between XP and X, V2 would be the head, while V1 would make the rule recursive (because of its identity to the V to the left of the arrow). In conclusion, it is not clear to me how Hornstein can locate the source of recursion in the head labeling convention, i.e. in endocentricity, if recursion in

xlii

Harry van der Hulst

the usual sense results from identity between the symbol on the left and the non-head symbol on the right of the arrow. Taking this one step further we might say that Hornstein’s idea that headedness entails recursion is correct after all, in the sense that endocentricity leads to what I will call recursion following from adjunction. Indeed, when linguists speak of adjunction they refer to a situation in which a head is expanded with a dependent with the result being a category that is identical to category of the head. Normally this means that the resulting category has the same category and the same complexity as the head. In fact, the rule in (19) (NP % NP þ S) was a rule of precisely this sort. Hornstein seems to include the case in which the head is simplex (i.e. lexical category), while the result is phrasal. But by ignoring this di¤erence one might say, then, that headedness entails recursion following from adjunction. However, this kind of recursion di¤ers from the ‘prototypical’ case in which a combination has the same category status as the dependent, which I will call recursion following from subjunction. It remains unclear how the former type of recursion automatically leads to the latter type. Whatever the merit of Hornstein’s suggestion, there can be no doubt that the step from one word utterances to multiword utterances with hierarchical structure (perhaps mediated by a two-word ‘proto-language’) was a crucial step in the development of human language, but it remains questionable whether this is the only property that sets human language apart from other communication systems (cf. Jackendo¤ and Pinker 2005). The evolutionary angle is examined in the following four contributions. Kinsella evaluates the HCF claim that recursion is the one property which sets human linguistic abilities apart from any other system of communication or cognition which she considers to be fundamentally flawed. Like Jackendo¤ and Pinker (2005) she argues that, first, properties of language independent of its recursive nature are unique to the system and, second, recursion is exhibited in domains outside human language. Thirdly, she argues that language works equally well without recursion. Progovac focuses her attention on what she calls a small clause grammar which co-exists, in English, with a sentential grammar. The latter is a robust system of functional projections and structural relationships (tense, case checking, complementizers), while the former does without any such mechanisms, and without a possibility for recursion/embedding. She proposes that this small clause grammar is a vestige/‘‘living fossil’’ of a previous stage of morpho-syntax which utilized no functional categories (which she claims, are necessary for embedding). Her conclusion is that a relevant functional projection/category is necessary to facilitate embedding. She

Re Recursion

xliii

argues that if a human grammar with Merge, but without recursion, is possible, then recursion cannot be the defining property of human language, and neither can Merge alone be responsible for all its recursive power (contra the hypothesis in Hauser, Chomsky, and Fitch 2002). This perspective opens up a new way of looking at some puzzling constraints in syntax, including Subjacency. Rogers and Hauser discuss acoustic pattern recognition experiments, often called artificial language learning experiments, which, they claim, hold the promise of providing a method for dissecting the ontogenetic and evolutionary building blocks of the language faculty. In particular, by studying the capacity to acquire specific fragments of linguistic computation in human infants and nonhuman animals, it is possible to determine which psychological processes are available to the developmentally immature human learner and which to humans’ evolutionarily ancestors. They specifically explore the formal mathematical structure of these experiments and develop criteria for their design and interpretation based on the Sub-Regular Hierarchy, a hierarchy of complexity classes which correspond to a hierarchy of cognitive capabilities that are relevant to any faculty that processes acoustic stimuli solely as sequences of events.

7. Recursion and the Lexicon In most articles, specific recursion is treated as a property of syntactic, i.e. sentence-level expressions. The following two articles examine the role of recursion with reference to the lexicon, although the authors take very different perspectives and arrive at seemingly contradictory conclusions. Koster defends a lexicalist approach to recursion in syntax. In a lexicalist framework, syntactic base structures are no longer generated independently of the lexicon but are seen as projections from lexical elements. Chomsky (1981, 31–32) discusses the redundancy problem that arises with lexicon-independent rules: these rules would introduce properties as hierarchical phrase structure organization and recursion, while these elements also exist independently, namely as the projection properties of lexical items. Surprisingly, Koster says, Minimalism reintroduced lexicon-independent structure generation in the form of Merge. This, then, brings back the redundancy problem. As a remedy, Koster proposes to maintain a version of X-bar theory, in which structure-building (with hierarchical organization and recursion) is seen as filling the slots that exist as a matter of lexical projection. Syntactic computation is done on the basis of these lexical-

xliv

Harry van der Hulst

cartographic structures and involves the selective, strictly local sharing of properties: by sisters (horizontal) and by daughters and mothers (vertical). By iteration of vertical property sharing (percolation) and in combination with horizontal property sharing, most forms of syntactic construal (including ‘‘movement’’) can be accounted for. In the framework proposed by Koster, recursive Merge is not a rule of grammar, but a characterization of the abstract background capacity that humans built into language by inventing a complex lexicon, as a matter of free, agentive application. In that respect, he claims, language is part of human culture and di¤ers from biological organology, which, clearly, involves non-agentive functionality. Juarros-Daussa’s starting point is the basic fact of argument structure that verbs cannot take more than three core arguments – more precisely, one subject, or external argument, and two objects, or internal arguments, as in the English verb give (32a). Introducing a conceivable additional participant without the help of a lexical preposition (such as for in (32c) below), which contributes its own argument-taking abilities, results in ungrammaticality (32b): (32) a.

[The LinguistList] gave [nice prizes] [to the winners of the challenge] b.* [The LinguistList] gave [nice prizes] [to the winners] [(to) the students] c. [The LinguistList] gave [nice prizes] [to the winners] [for the students]

In her chapter, the above observation is formulated as the Two-Argument Restriction (TAR): ‘‘A single predicate can have at most two internal arguments and one external.’’ The author claims that the TAR is an unrestricted universal (in the sense of Croft 1990) and shows that valencyincreasing operations such as applicative and causative constructions, which present an apparent challenge, do not violate the TAR. She further argues that, since there is no known processing reason not to lexically associate more than three (two) participants to a predicate, the TAR is syntactic in nature, and it is one of a family of architectural constraints that determine and limit possible attainable languages (in this case possible argument structures). Following this idea, she shows that the framework of lexical syntax put forth by Hale and Keyser (2002) is especially suited to derive the TAR. In her proposal, deriving the TAR crucially involves negating the existence of a recursive function in the domain of argument structure.

Re Recursion

xlv

8. Recursion and Morphology As traditionally understood, complex words display direct or indirect specific recursion. This can be most easily demonstrated in the domain of compounding. Consider the following compound rules in English (taken from van der Hulst 2008): (33) Compounding N%NN N%AN N%VN

arm - chair green - house jump suit

V%VV V%NV V%AV

break dance steam roll white wash

A%AA A%NA A%VA

red hot nation wide -

In each of the three blocks of rules, the first one meets the criterion of being recursive: the dependent is of the same type as the whole construction. In the area of derivation we also attest recursion: (34) a.

Su‰xation N % N N a¤ friend - ship, child - hood, host - ess, hand - ful N % A N a¤ tall - ness, free - dom, loyal - ist, real - ism N % V N a¤ sing - er, employ - ee, grow - th, inform - ant V % V V a¤ V % N V a¤ victim - ize, beauti - fy V % A V a¤ black - en A % A Aa¤ green - ish A % N Aa¤ boy - ish, wood - en, nation - al, pain - ful A % V Aa¤ read - able, help - ful, harm - less, act - ive

b.

Prefixation N % Na¤ N anti - war, ex - president, super - structure N % Na¤ V N % Na¤ A -

xlvi

Harry van der Hulst

V % Va¤ V un - do, re - read, mis - align V % Va¤ N en - slave, be - witch V % Va¤ A be - little A % Aa¤ A un - fair, dis - loyal A % Aa¤ N A % Aa¤ V Recursion results not only from rules that are themselves recursive but also from rule sets that form recursive chains which produce indirect recursion: (35)

If even English, with its modest morphology, displays specific recursion, it is to be expected that in polysynthetic languages recursion will play an even greater role in the morphology; cf. Mithun, this volume. Lander and Letuchiy present a survey of various morphological phenomena in Adyghe, a highly polysynthetic language of the Northwest Caucasian family, such as multiple applicatives, multiple propositional operators (e.g., tense markers), and double causatives that all seemingly instantiate recursion. It is argued, however, that the corresponding derivations di¤er in what concerns their ‘recursability’, i.e., their inclination to recursion. Moreover, they propose that the degree to which a derivation is ‘recursable’ correlates with the extent to which a derivation a¤ects the meaning of a stem and the structure of the verb. Since no apparent syntactic restrictions of this kind are found, this suggests that despite the

Re Recursion

xlvii

existence of morphological recursion, the latter is still somewhat more constrained than recursion in syntax.

9. Recursion and phonology It is quite common to read that phonology is not recursive. To illustrate this point it is often mentioned that, for example, we do not find ‘syllables inside syllables’; cf. the quote from Pinker and Jackendo¤ given earlier. On the other hand, Ladd (1986, 1996) argues that higher-level prosodic structure allows (limited) recursion. I also refer to Wagner (2005, 2007a, 2007b) for extensive discussion of these matters, again with specific reference to higher levels of prosodic organization. Since phonology is somewhat underrepresented in this volume, I included a chapter on this subject in this volume. In this chapter van der Hulst first discusses recursion at lower levels of organization (syllable and foot) showing, contrary to popular belief, that it is possible to think of codas as ‘syllables within syllables’ and, also, once this move is made, to reanalyze feet in terms of recursive syllable structure. Second, he turns to recursion at higher prosodic levels reviewing some of the above-mentioned literature. One central issue that emerges from this discussion is that the kind of data that suggest recursion at higher levels could also be taken as supporting the idea that phonological rules (for example, rules for phrasal accentuation) make direct reference to the recursive syntactic structure, in which case we have no argument for recursion in the prosodic structure as such (cf. Wagner 2007a, 2007b). A second central theme regards the fact that the recursion of prosodic structure itself, if such can de demonstrated to exist, would be caused by (a) adjunction of ‘stray’ units (creating one level of recursion) and (b) by the ‘desire’ to make prosodic structure isomorphic to the syntactic structure which may take the depth of prosodic recursion further. However, in this latter case, as has been argued in Giegerich (1985), recursive structure often tends to be replaced by a flatter structure which is more rhythmic. It would seem, then, that rhythmic forces suppress excessive recursion in phonology or that phonological recursion and rhythm occur at di¤erent levels of representation. Hunyadi introduces the principle of tonal continuity to account for the continuous tonal phrasing of discontinuous structures with nested embedding and suggests that what underlies this cognitive computational process is the bookmark e¤ect. He shows that the computational di¤erence between nested recursion and iteration correlates with their prosodic di¤erence, whereas tail recursion and iteration (which are computationally indistin-

xlviii

Harry van der Hulst

guishable) also have a similar prosodic realization. Through grouping experiments with (a) abstract visual elements, (b) abstract prosodic elements and (c) actual linguistic utterances Hunyadi shows that speakers use temporal and/or tonal variation to indicate various kinds of grouping and he attributed the underlying principles to more general cognitive strategies because, apparently, they play a role outside language. For temporal variation, recursive embedding is denoted by pre-boundary shortening and de-embedding by pre-boundary lengthening so that for each embedded phrase, the pause preceding it is shorter than the pause following it and that with each level of recursive embedding these respective pauses decrease at the point of embedding and increase at the point of de-embedding. For tonal variation, recursive embedding is denoted by the pre-boundary lowering of the tone and de-embedding by the pre-boundary raising of the tone so that with each level of recursive embedding a given phrase is embedded at a pitch level lower than the previous one. Hunyadi’s paper, specifically his discussion of the principle of tonal continuity relates to Mithun’s paper in that this principle demonstrates that intonation can be used as a mechanism to encode embedding and this, one might argue, does not necessarily presuppose a syntactic embedding mechanism. In other words, semantic center-embedding can perhaps be encoded in terms of intonation alone, which suggests, in line with several other chapters that syntactic recursion, or, specifically, syntactic center-embedding is not the only grammatical device that can be used to encode semantic center-embedding. A final remark about intonation. Intonation is not ‘phonology’. The intonational tune is not merely part of the phonological side of linguistic expressions. Intonational tunes are the product of an intonational grammar which produces intonational expressions that have morphotactic, semantic and phonotactic components; cf. Gussenhoven (1984). As such, the question can be raised whether the intonation grammar displays recursion in its semantic, morphotactic or phonotactic component. I will not explore that issue here.

10. Recursion outside Syntax Finally, the question arises as to the language-uniqueness of recursion. It has been acknowledged that other human cognitive systems also display recursion. In this connection, some mention the numerical system. Chomsky (2005: 6), however, suggests that the numerical system may be the same recursive merge mechanism that is responsible for linguistic expres-

Re Recursion

xlix

sions (cf. Tomalin 2007: 1794–5). Others refer to ‘social intelligence’, or more specifically, the other mind capacity as being inherently recursive. Does this mean that recursion is a general cognitive faculty that is used in language, as it is in other cognitive modules? Or is it the case that this more general trait occurs in di¤erent modules as, let us say, homologous traits? The contribution by Levy presents a biologically plausible approach to syntax that treats language as a transduction between propositional meanings and sequences of symbols, conceptualized as a sequential planning task. Recursion emerges as part of a general solution to the problem of transducing the fillers of semantic roles (agent, patient, experiencer, theme) in this manner. His model uses high-dimensional, low-precision vectors of random noise as a plausible stand-in for neural activations. Such vectors support the encoding of semantic role/filler bindings and variable substitutions required for language but (unlike the automata corresponding to traditional grammars) degrade gracefully as more complex representations are encoded. The model thereby o¤ers a plausible account of depth limitations and other psychological phenomena associated with recursion, which are treated as a tangential ‘‘performance’’ issue in traditional grammarbased approaches. The author provides an example of how the model learns a simple grammatical construction from a single instance, avoiding the psychological and biological implausibility of previous neurallyinspired approaches. After describing the relationship of this representational scheme to traditional grammatical categories, he concludes with a discussion of the possible origins of linguistic universals not explained by the model. Zimmerer and Varley show that recursion may be inaccessible to an agrammatic patient in his language, or language-like computations while it is readily available in other cognitive domains. Their report presents data from PR, an agrammatic aphasic patient who shows no signs of processing recursion in his syntactic output and comprehension, but performs well in domains claimed to be recursive, namely mathematics and Theory of Mind. One response to this dissociation might be to claim that PR’s language deficits lie in performance only, linguistic competence being intact. To address this claim, the authors present new data from two artificial grammar learning experiments in auditory and visual modalities. After long exposure to strings of stimuli structured by the grammar AnBn, PR showed an atypical pattern when asked to judge new strings on their correctness. The results suggest that severe aphasia can represent a failure of syntactic competence. It is concluded that in the mature mind, even without syntactic competence, recursion in other cognitive areas is possible.

l

Harry van der Hulst

11. Final remarks Several general conclusions can be drawn from the collection of chapters in this volume, although these conclusions are not necessarily shared by all authors. First, to formally represent certain linguistic phenomena in terms of recursive devices is a modeling choice. Other modeling choices are possible and should be considered. Second, there is dispute on what the linguistic phenomena are when one does not wish to rely on ‘grammatical intuitions’ which, supposedly, reflect the principles of competence without being cluttered by performance factors. Many linguists adopt a more empirical stance, and insist on inspecting what occurs in actual language use. They then find that the types of constructions that can be modeled in terms of recursive devices are not that common, and, when present, are of limited depth. Third, what I have called specific recursion (A within A) does not have to be regarded as a basic linguistic mechanism. Rather this form of recursion (as is self-embedding) is implicit in a combinatorial system that allows combinations to enter into further combinations. Fourth, it might be argued that morphosyntactic structures display recursion because these structures want to be isomorphic to semantic structure which is inherently recursive. Phonological structures, on the other hand, accommodate phonetic substance which is not inherently recursive but rather rhythmic which induces a flat, non-recursive structure. Nonetheless, phonological structure seems to tolerate limited recursion (competing with rhythmic flattening) in order to increase isomorphy to morphosyntatic (and thus indirectly semantic) structure. Fifth, semantic recursion can be encoded in terms of morphotactic specific recursion, but other mechanisms, such as intonation can also be used to encode semantic embedding. These various points are not conclusions that all authors agree on. Still, these are arguments that the reader will find in this book which, I believe, deserve serious consideration.

References Burton-Roberts, Noel 2000 Where and what is phonology?: A representational view. In Burton-Roberts, N., Carr, P., Docherty, G. (eds.), Phonological Knowledge: Its Nature and Status. Oxford University Press, Oxford, pp. 39–66.

Re Recursion

li

Christiansen, Morten H. 1992 The (non)necesssity of recursion in natural language processing. Proceedings of the 10th annual conference of the cognitive science society. Indiana University, Indiana: Cognitive Science Society, July/August, 665–670. Chomsky, Noam 1975 The logical structure of linguistic theory. The MIT Press, Cambridge. Massachusetts. 1981 Lectures on government and binding. Foris, Dordrecht. 1995 The Minimalist Program. The MIT Press, Cambridge. Massachusetts. 2005 On Phases. MIT, ms. Chomsky, Noam and George A. Miller 1963 Introduction to the formal analysis of natural languages. In R.D. Luce, R.R. Bush and E. Galanter (eds.), Handbook of mathematical psychology. Vol. II. John Wiley, New York, 269–321. Croft, William 1990 Typology and Universals. Cambridge University Press, Cambridge. Davis, Hugh C. 1995 To Embed or Not to Embed. . . Communications of the ACM (CACM), 38(8): 208–209. De Roeck, Anne, Roderick Johnson, Margaret King, Michael Rosner, Geo¤rey Sampson, and Nino Varile 1982 A Myth about Centre-Embedding. Lingua 58: 327–340. Ejerhed, Eva 1982 The processing of unbounded dependencies in Swedish. In E. Engdahl and E. Ejerhed (eds.), Readings on Unbounded Dependencies in Scandinavian Languages. Almqvist & Wiksell Intl. Stockholm, pp. 99–149. Everett, Daniel 2005 Cultural constraints on grammar and cognition in Piraha˜: Another look at the design features of human language. Current Anthropology 76(4): 621–646. 2007a Cultural constraints on grammar in Piraha˜: A reply to Nevins, Pesetsky and Rodrigues (2007). LingBuzz article archive (http:// ling.auf.net/lingBuzz/000427) 2007b Challenging Chomskyan Linguistics: The Case of Piraha˜. Human Development 50: 297–299. 2008 Don’t sleep there are snakes: life and language in the Amazonian jungle. New York: Pantheon Books/London: Prolile Books. To appear. Piraha˜ culture and grammar: A response to some criticism. Fitch, W. Tecumseh, Marc D. Hauser, and Noam Chomsky 2005 The evolution of the language faculty: Clarifications and implications. Cognition 97: 179–210. Gentner, T. Q., K. M. Fenn, D. Margoliash, and H. C. Nusbaum 2006 Recursive syntactic pattern learning by songbirds. Nature, 440: 1204–1207.

lii

Harry van der Hulst

Giegerich, Heinz J. 1985 Metrical Phonology and Phonological Structure. Cambridge University Press, Cambridge. Gussenhoven, Carlos 1984 On the grammar and semantics of sentence accents. Dordrecht: Foris. Hale, Ken and Samuel Keyser 2002 Prolegomena to a Theory of Argument Structure. The MIT Press, Cambridge, Mass. Halle, M. and J.-R. Vergnaud 1987 An essay on stress. Cambridge, Mass.: MIT Press. 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. Hornstein, Norbert 2009 A theory of syntax. Minimal operations and universal grammar. Cambridge: Cambridge Univesity Press. Hulst, Harry van der 2008 Linguistic structures. Kendall/Hunt Itkonen, Esa 1976 The Use and Misuse of the Principle of Axiomatics in Linguistics. Lingua 28: 185–220. Jackendo¤, Ray 2002 Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford University Press, Oxford. Jackendo¤, Ray and Steven Pinker 2005 The nature of the language faculty and its implications for evolution of language (Reply to Fitch, Hauser, and Chomsky). Cognition 97: 211–25. Karlsson, Fred 2007a Constraints on multiple initial embedding of clauses. International Journal of Corpus Linguistics 12(1): 107–118. 2007b Constraints on multiple center-embedding of clauses. Journal of Linguistics 43(2): 365–392. Kuryłowicz, Jerzy 1952 Uwagi o polskich grupach spo´łgłoskowych. Biuletyn Polskiego Towarzystwa Je˛zykoznawczego 12: 221–232. Ladd, D. Robert 1986 Intonational Phrasing: The Case for Recursive Prosodic Structure. Phonology Yearbook 3: 311–340. 1996 Intonational Phonology. Cambridge: Cambridge University Press. [2nd edition 2008] Liberman Mark and Alan Prince 1977 On stress and Linguistic Rhythm. Linguistic Inquiry 8: 249–270. Marcus, Gary 2006 Startling starlings. Nature 440: 1117–1118.

Re Recursion

liii

Nevins, Andrew Ira, David Pesetsky and Cilene Rodrigues To appear Piraha˜ exceptionality: A reassessment. Parker, Anna R. 2006 Evolution as a Constraint on Theories of Syntax: The Case against Minimalism. PhD Dissertation, University of Edinburgh Peano, G. 1889 Arithmetices principia nova methodo exposita. Opera Scelte 2: 20–55 (Reptinted in Unione Matematica Italiana, 1957–1959. Pinker, Steven and Ray Jackendo¤ 2005 The faculty of language: What’s special about it? Cognition 95: 201–236. Postal, Paul M. and D. Terence Langendoen 1984 The Vastness of Natural Languages. Blackwell, Oxford. Pulman, S.G. 1986 Grammars, Parsers, and memory limitations. Language and Cognitive Processes 2, 197–225. Reich, Peter 1969 The finiteness of natural language. Language 45, 831–843. Reuland, Eric 2009 Language: symbolization and beyond. In R. Botha and C. Knight (eds.). The Prehistory of Language. Oxford: Oxford University Press, 201–224. Sampson, Geofrrey 2001 Empirical Linguistics. London/new York: Continuum Thomas, James Davis 1995 Center-embedding and self-embedding in human language processing. PhD dissertation, MIT. Tomalin, Marcus 2007 Reconsidering recursion in syntactic theory. Lingua 117, 1784– 1800. Uriagereka, Juan 2009 Syntactic Anchors. Cambridge: Cambridge University Press. Wagner, Michael 2005 Prosody and recursion. Ph.D. diss., Massachusetts Institute of Technology, Cambridge, MA. 2007a Prosodic Evidence for Recursion? Ms. Cornell University. 2007b Prosody and Recursion in Coordinate Structures and Beyond. Ms. Cornell University.

Part I.

Discussing the need for recursion on empirical grounds

1. Piraha˜ – in need of recursive syntax?1 Jeanette Sakel and Eugenie Stapert

1. Introduction Since the publication of the article ‘The faculty of language: what is it, who has it and how did it evolve?’ by Hauser, Chomsky, and Fitch (2002) recursion has assumed a prominent place in the discussion of what aspects of our communication system are unique to humans and human language. The hypothesis put forward in that article is that recursion is the only property unique to human language. The entailments of this hypothesis are that recursion would be found in all languages spoken by humans and that a human language that does not use recursive structures would not exist. This claim has provoked a number of reactions, most noticeably Everett (2005), who argues that Piraha˜, the last surviving member of the Muran language family, does not exhibit recursive structures in its syntax, even though it undoubtedly is a language spoken by humans. Piraha˜ is spoken by approximately 450 people in the Brazilian state of Amazonas, in small settlesments along the river Maici. The Piraha˜ live a largely traditional life as hunter-gatherers and rarely seek contact with the outside world. Since the publication of Everett’s (2005) article, there has been an ongoing, mainly web-based, discussion as to whether Piraha˜ exhibits recursive structures, e.g., on LingBuzz by Nevins, Pesetsky, and Rodrigues (2007) and a reply by Everett (2007a), on Language Log by Liberman (2006), Everett (2007b), Sakel’s (2007a) and Slobin’s (2007) Letters to the editor of the journal Human Development and Everett’s reply (2008). The

1. We would like to thank Dan Everett, Nigel Vincent, David Denison, Ted Gibson, Mike Frank, Manfred Krifka, Uli Sauerland, Harry van der Hulst and three anonymous reviewers for comments on earlier versions of this article. We would also like to thank Alison Wray for an inspiring discussion on formulaic language and esoteric communication. This article is based on a presentation at the conference on ‘Recursion in Human Language’ in Normal, Illinois (April 2007).

4

Jeanette Sakel and Eugenie Stapert

issues brought up are far from being resolved. Some of the ambiguity in the present discussion is based on contradictory analyses of Piraha˜ data in two publications by Everett (1986, 2005). Everett’s earlier work (1986) contains an extensive description of embedded clauses in Piraha˜, whereas he rejects the existence of these structures in his more recent work (2005). Due to this discrepancy, researchers who support the hypothesis that Piraha˜ has recursive syntax usually cite data presented by Everett (1986), discarding the new analysis (2005). While the discussion is very focused, few people so far have looked at Piraha˜ language facts other than those given in the two publications by Everett (1986, 2005), comparing them to other indicators as to whether recursion is necessary in a human language. The present article sets out to add new data to the discussion. It is based on our field research among the Piraha˜ and brings together research data from experiments and elicitation, as well as analysis of spontaneous speech.2 We look at first-hand language data in three areas of Piraha˜ grammar which we would expect to be expressed by recursive structures if these existed, and then address our findings in the light of a number of recent findings in linguistics. The article is divided up in the following way: first we will discuss the su‰x -sai, which Everett (1986) reported to be an indicator of embedding, and which features prominently in the recent discussions. We will then look at the other side of the coin and investigate what alternative strategies can be used to express complex situations. Following this, we will address the question whether recursive structures could enter the language through language contact with Portuguese. Finally we will examine our results in the light of recent publications to gain insights into whether recursion is really a necessary notion in human language. This article is a collection of several arguments related to the question whether recursion is necessary in Piraha˜.

2. Our field research was funded by the CHLaSC project (within the EU F6 programme). The fieldwork was facilitated by Dan Everett, who accompanied us to the field, though the research data presented here are entirely our own and based on our own interpretations. Some of the experiments were carried out with Dan Everett as the interpreter, while others were conducted monolingually without his involvement. These settings did not result in any noticeable di¤erences in our language data. We took great care to carry out all experiments as scientifically and objectively as possible.

Piraha˜ – in need of recursive syntax?

5

2. The su‰x -sai The su‰x -sai figures prominently in recent discussions of recursion. Everett (1986: 277) classified it as a nominaliser and an obligatory indicator of embedding in conditional clauses (1986: 264). In his later approach (2005: 21), however, he argues that -sai does not mark syntactic subordination.3 We have studied this marker’s functions, in particular with respect to whether it is an obligatory marker of embedding. We will here look at two very di¤erent constructions in which -sai is used. The first part of our discussion is based on findings by Stapert (2007) and Stapert et al. (in preparation). They tested the functions of -sai experimentally in a sentence repetition task. In this experiment, two clauses representing semantically connected propositions, such as it is raining and I don’t go to the forest were combined. The su‰x -sai was added to either the verb of the first or that of the second clause, cf. (1a) and (1b) and the informants were asked to merely repeat the sentence. Piiboi-bai-sai ti kaha´pi-hiaba. rain-INTENS-SAI 1 go-NEG ‘If it is raining I won’t go.’ b. Piiboi-bai ti kaha´pi-hiabi-sai. rain-INTENS 1 go-NEG-SAI ‘If it is raining I won’t go.’

(1) a.

A total of nine speakers of Piraha˜ – 7 women and 2 men – participated in this language task. In their response, informants attached -sai to the first clause, the second clause, both clauses, or neither of the clauses (cf. 2) independent of the input and with no reported change in meaning or judgement of (un)grammaticality. (2) Piiboi-bai ti kaha´pi-hiaba. Rain-INTENS 1 go-NEG ‘If it is raining I won’t go.’ Out of a total of 39 relevant responses -sai was attached to both clauses in 9 cases, to none in 6, and to one clause – either the rain part or the forest part – in 24 instances4. 3. Everett, p.c., argues that -sai is an old information marker. 4. For a detailed description and analysis of the results of this task see Stapert (2007).

6

Jeanette Sakel and Eugenie Stapert

The alternative in which -sai does not occur in either clause (2) was not part of the input of the experiment. Hence a simple repetition of an ungrammatical sentence from the input is ruled out. This means that the concept can be expressed without the presence of -sai; thus this marker can not be an obligatory marker of embedding. Still unclear, however, is the exact function of -sai in these constructions, but it does not appear to be a marker of subordination, as originally claimed by Everett (1986).5 A related investigation carried out by Sakel and Stapert for the present paper was the analysis of various other constructions with -sai, both in spontaneous speech and elicitation. The results show that -sai occurs most frequently in constructions expressing quotation of the type hi gaisai (3 say-sai) ‘he said’ or ti gai-sai (1 say-sai) ‘I said’. This construction is always followed by direct speech and occurs with great frequency, and indeed in certain discourse contexts in every utterance. While functioning as a quotative in many of these cases, it sometimes appears in contexts that are not directly reportative, cf. (3). (3) Ai hi gai-sai xigihı´ hi xaisigı´aihı´ xaita´hoı´hı´ xoo´. well 3 say-SAI man 3 same sleep forest ‘Well, the same man went to sleep in the forest.’ This example (3) was uttered in the context of elicitation – the story was played out with dolls – where no conversation was directly referred to. Rather than being a quotative, hi gaisai seems to express impersonal reference in this case, such as ‘the Piraha˜ in general do/say this’, detaching the speaker and his responsibility from what is said. The meaning would be ‘one would say it in the following way’. This could point to a possible development away from the mere quotative use of the construction towards a more abstract meaning. Altogether, we can say that hi gaisai and similar constructions function as discourse markers and elements detached from the main content of the clause. Similar claims have been made for constructions in many other languages, including English, cf. Thompson and Mulac (1991) and Thompson (2002) for arguments that I think and I guess have grammaticalised into evidential markers and Hopper (2000) for English pseudoclefts functioning as discourse markers. 5. In the same way as English that, -sai is optional and can be left out. While one could argue that when that is left out in English it is still present in the form of a null complementiser. In Piraha˜ the argument against -sai as a subordination marker is considerably stronger because of the combination of -sai being optional and appearing in di¤erent positions within the clause.

Piraha˜ – in need of recursive syntax?

7

3. Alternative strategies to express complex cognitive structures: Mental verb constructions Similar to -sai above, Piraha˜ employs various other strategies to express cognitively complex concepts without making use of syntactic complexity. A striking example is mental verb constructions, such as sentences containing I think or I doubt in English. These are relevant since they always reflect two perspectives: either from two di¤erent people, or between reality and personal experience of one person. Regardless of whether these are two separate propositions or not, which in itself is a much debated issue, we have to do with a cognitively complex situation and many languages choose to encode this in syntactically complex sentences. Still, non-complex ways of coding mental verb constructions are likewise well-attested, e.g., marking for evidentiality by su‰xes. When looking at the data in Piraha˜, we find that there are no separate verbs expressing mental states. Rather, su‰xes corresponding functionally to English mental verbs or adverbs indicating mental attitudes are used. Table 1 brings together some of the mental attitude su‰xes and their functions, as well as their equivalent translations as complex structures or adverbs in English (for a detailed argument on why mental verb constructions and evidential su‰xes are comparable cf. Diessel and Tomasello 2001; Stapert 2009). Table 1. attitude su‰xes in Piraha˜ Verbal su‰x

Function, meaning

Equivalent in English Mental verb

Adverb

-a´ti -haı´ -ha´

Uncertainty

I doubt, I’m not sure

maybe, perhaps

relative certainty

I think, I guess

probably

complete certainty

I know, I bet, I’m sure

definitely, certainly

-sog -hı´ai

Desiderative

I wish, I want, I hope

hopefully

Hearsay

I heard

apparently, allegedly

-sibiga

deductive

I understand, I suspect I get the impression

apparently, seemingly

-xa´agaha´

observation, matter of fact

I notice, I see, I’m certain (lit. use)

clearly

-bai / -koı´

emphasis, intensifier

I bet, I mean (clarification)

obviously, certainly, for sure

8

Jeanette Sakel and Eugenie Stapert

Example (4) shows the markers -haı´ ‘relative uncertainty’ and hı´ai ‘hearsay’ added to the verb bog-ai ‘he came’ to express ‘doubt’ and ‘hearsay’ respectively. In examples (5) and (6) the meanings of ‘complete certainty’ and ‘observation’ are added in the same way: (4) Garippı´iru bog-ai-haı´-hı´ai. Brazilian.worker come-ATELIC-DOUBT-HEARSAY ‘(I heard that) the Brazilian worker has probably not come here.’ (5) Hi kaga´ihiai koaba´i-p-a´-ha´ 3 jaguar kill-PERF-REM-COMP_CERT ‘(I’m sure) he shot the jaguar’ (6) Piboi-bai hi kaha´pi-hiab-a´agaha´ Rain-INTEN 3 go-NEG-OBSERV ‘It is raining; (I see) he is not going (to the forest)’ In English the concept of uncertainty can be expressed by the adverb probably, as in the translation of (4), or the entire sentence could alternatively be expressed in a double embedded structure such as ‘someone said that he doubts that the Brazilian worker came here.’ These elements function like evidentials, rather than verbs in expressing probability and source of information without having a separate subject themselves. In this way recursive embeddings in English are very di¤erent from constructions with evidentials in Piraha˜. Compare the recursive sentence in (7a) with the non-recursive equivalent using adverbs with similar functions to the Piraha˜ evidentials in (7b): (7) a. b.

He said that I suspected that the students were hung over. Hearsay perhaps the students are hung over.

Summarising, markers of attitude in Piraha˜ can be analysed as expressing semantically complex structures without syntactic embedding. 4. Language contact The examples we have looked at so far were native Piraha˜ language data. Now, we will turn to elements and morphemes outside of the Piraha˜ system that could be introduced by language contact and that could subsequently introduce recursion into Piraha˜ syntax. The hypothesis is that through intensive language contact with Portuguese, markers and structures of embedding, which are common in Portuguese, could be borrowed into Piraha˜. Our reasoning for this is as follows:

Piraha˜ – in need of recursive syntax?

9

Firstly, elements that mark structures of embedding are frequently borrowed in other contact situations. A typological study of grammatical contact phenomena (Matras and Sakel 2007; Sakel 2007b) concludes that function words such as discourse markers, coordinators and subordinating conjunctions are almost always borrowed in situations where a minority language is in contact with a highly dominant language and with prevailing bilingualism. Indeed, in most of these cases subordinating conjunctions were among the borrowed elements, being taken over wholesale with their form and function. Secondly, the Piraha˜ use many Portuguese lexical elements in their language, even though the community as a whole is predominantly monolingual with only a few older men having rudimentary knowledge of Portuguese. This is surprising as the Piraha˜ have been in contact with outsiders for over 200 years. The loanwords from Portuguese include new concepts such as gahia´o ‘plane’ (from Portuguese avia˜o) and kapı´iga ‘paper’ (from Portuguese papel ), as well as a number of elements that already exist in Piraha˜, but that are frequently used when speaking with outsiders, such as bı´i ‘good’ (from Portuguese bem) or ambora ‘away, let’s go’ (from Portuguese embora). When looking at the syntactic structures, however, there is no evidence that Portuguese has had any influence on the grammar of Piraha˜, as there are no apparent grammatical calques. In a number of cases speakers of Piraha˜ incorporate Portuguese grammatical elements into their language, but this is only the case when making conscious e¤orts to speak Portuguese to foreigners, as in (8). keeche hot kwaado aki his-o friio when here sun-LOC cold ai kaba keema ai DM NEG burn DM ai muito braako. DM very white

(8) Ai ai DM DM

aki his-o here sun-LOC6

‘It is hot here in the sun. When it is cold here in the sun, you do not burn. (You are) very white.’ (Portuguese elements in bold)

6. Hiso is Piraha˜ for ‘in the sun’ or ‘on the day’, but it is phonologically similar to local Portuguese constructions with fazer ‘to be (in relation to weather). It may hence be a blend of both languages, accommodated by a similar construction in Piraha˜.

10

Jeanette Sakel and Eugenie Stapert

In (8) the speaker makes use of the Portuguese adverbial clause marker cuando ‘when’ (integrated into Piraha˜ as kwaado). Instead of functioning as an adverbial clause marker, however, it appears to be used similar to the Piraha˜ distance marker -so, which expresses that an event is not happening in the immediate context of the utterance.7 In this case, the speaker expresses that it is not cold at the moment of speech. Comparing this example to typical Piraha˜ sentences, the structure is very similar: relations between clauses are established by simple juxtaposition, combined with distance marking when appropriate. Example (8) is thus an instance of insertion of Portuguese material into a grammatical frame that is purely Piraha˜. This is suggestive of the fact that Piraha˜ has not borrowed recursive structures from Portuguese. Increased contact with the outside world in recent years, and hence increased bilingualism could change this, however.

5. Recursion in Piraha˜? Toward an alternative analysis Let us sum up our findings and discuss to what degree we can expect recursion in Piraha˜. Firstly, does Piraha˜ have recursion? Most structures we have looked at so far have given no evidence of being outright syntactically recursive structures. In most cases clauses are linked by simple juxtaposition and relations between them become clear in the discourse context. However, conclusive support of this negative finding would require more evidence than we presently possess. Thus, our conclusions are necessarily tentative. There are a number of markers, such as -sai and -so, that seem to appear in structures parallel to ‘recursive’ structures in other languages, but these are not outright markers of subordination or recursion in the syntactic sense: more often, these markers are expressing semantic cohesion between parts of the discourse. These markers also indicate relations between what is said and the reality of the speech situation, such as the distance marker -so, which expresses a distance to the current reality. Concepts that are expressed recursively in many other languages are marked by a‰xes in Piraha˜, as in the case of mental verb construction. Language contact has likewise failed so far to introduce recursive structures from 7. We do not yet fully understand the exact functions of -so or its distribution. In the same was as -sai, however, it appears to be optional and is not an obligatory marker for embedding.

Piraha˜ – in need of recursive syntax?

11

Portuguese. Hence, we can not say with any confidence that there is – or for that matter is not – recursion in Piraha˜. Instead of saying that recursion is a core characteristic of human syntax, we believe that it is an important feature of human language which is most likely to be present in languages and language varieties that are used to express complex concepts. Let us discuss a number of recent publications in the field to clarify what we mean. 5.1. Spoken language Recursive structures appear to be far less frequent in spoken language than in written language. Mark Liberman discusses this for English in his entry on Language Log in May 2006, citing the following example from Elmore Leonard’s La Brava: (9) What’re you having, conch? You ever see it they take it out of the shell? You wouldn’t eat it. This is a typical example of a variant of spoken language, though paradoxically in this case it is written language imitating spoken language. Nonetheless, it shows that complex and recursive constructions such as ‘if you had ever seen it being taken out of the shell you would not eat it’ can be replaced by paratactic, non-recursive structures in spoken language. That spoken language makes less use of recursion has also been shown for Finnish and Japanese: Laury and Ono (This Volume) present evidence that when recursive structures appear in spoken language they are generally less complicated than in written language. There is often only one degree of recursion in spoken language, while written language can show many di¤erent layers of subordination (cf. also Karlsson 2007). Similar evidence comes from the analysis of informal talk, where clause chains are preferred to embedding (Pawley and Syder 1983). Comparing these findings with our Piraha˜ data, we can argue that since Piraha˜ is a spoken language exclusively, recursion may be unnecessary or at least far rarer than in written language. 5.2. that-omission in relative clauses On top of the decrease in recursive structures we find in spoken language we can also argue for a parallel case that extends to written language. Two recent approaches to that-omission in English restrictive relative clauses claim that the resulting construction is non-recursive. Fox and Thompson (2007: 293) argue that pragmatic-prosodic factors, as well as frequency,

12

Jeanette Sakel and Eugenie Stapert

can lead to a ‘‘monoclausal’’ nature of the combination of relative clause and main clause. These monoclausal combinations are highly formulaic and processed as one, rather than two clauses, and in these cases that is omitted. In a di¤erent approach, Jaeger and Wasow (2007) argue that the more accessible the relativised element is in English non-subject extracted relative clauses, the more likely it is for the relativiser to be omitted. Thatomission usually takes place when the relativised element is given or definite. Hence that is often absent when the content of what is said is predictable. Another parallel case has been reported by Progovac (This Volume), who argues that certain small clauses do not allow for recursion. In this way, English has constructions that have a non-recursive expression and that appear both in spoken and written language. What if such constructions were the default or indeed the only option in another language, such as Piraha˜? Since English relative clauses work equally well in cases with or without overt syntactic marking for recursion, it is possible to imagine a human language that does not need to have recursive structures. 5.3. Esoteric language use More evidence comes from studies of how human language developed. Wray and Grace (2007) distinguish between esoteric vs. exoteric communication, based on Thurston’s (1987) terminology. Esoteric communication is inward-facing, which means that it is used within a well-defined group. In this type of communication comprehension is facilitated as hearers are likely to know what the speaker is going to say in a given situation. This still means that the language can express novel ideas, but the expression of predictable thoughts is a default. Exoteric communication, on the other hand, is outward-facing. Hence, exoteric communication in the definition of Wray and Grace (2007) would range from using a lingua franca to employing one’s local dialect to communicate with somebody unknown. Speakers have to be clear, since hearers are unlikely to predict what the speaker will talk about. This is possible in a language with simple, unambiguous elements that can be combined by unambiguous rules. Hence it is not surprising that the type of linguistic features found in varieties used for esoteric and exoteric communication are very di¤erent: Wray and Grace (2007) discuss how in esoteric communication suppletion and complex semantic structures are frequent, while language varieties used for exoteric communication often show logical and transparent rules

Piraha˜ – in need of recursive syntax?

13

that are also learnable by adult speakers and that are semantically transparent. They argue that human language probably started as a means for esoteric communication and that rule-based grammar is a cultural add-on that evolved with increased necessity for complex negotiations. Many types of communication are exoteric in the complex and globalised world of today. This is most likely to one reason for recursion being very frequent in the world’s languages. The Piraha˜, on the other hand, are an inward-facing group, and their language is only rarely used with outsiders.8 Explicit recursive syntax may thus not be necessary.

6. Conclusion In conclusion, the Piraha˜ structures we have looked at in this paper have shown no evidence of being syntactically recursive. Instead, Piraha˜ appears to make use of juxtaposition and morphological complexity to express complex concepts. Our conclusion is hence very similar to Everett’s analysis (2005). We have discussed a number of constructions in which even syntactically complex languages prefer non-recursive structures to recursive ones. It is possible that what other languages have as an option is the default in Piraha˜. Further support comes from the fact that Piraha˜ is an exclusively oral language. Spoken language and predictable content are exactly the instances in which non-recursive structures are preferred in other languages such as English. Hence, there is no apparent functional need for recursion in Piraha˜ syntax.

References Diessel, Holger and Michael Tomasello 2001 The acquisition of finite complement clauses in English: A usage based approach to the development of grammatical constructions. Cognitive Linguistics 12: 97–141. Everett, Daniel 1986 Piraha˜. In: Desmond Derbyshire and Geo¤rey Pullum (eds.), Handbook of Amazonian Languages I, p. 220–326. Berlin: Mouton de Gruyter.

8. These outsiders are a handful of linguists and missionaries who speak the language to a degree.

14

Jeanette Sakel and Eugenie Stapert

Everett, Daniel 2005 Cultural constraints on grammar and cognition in Piraha˜: Another look at the design features of human language. Current Anthropology 76(4): 621–646. Everett, Daniel 2007a Cultural constraints on grammar in Piraha˜: A reply to Nevins, Pesetsky and Rodrigues (2007). LingBuzz article archive (http:// ling.auf.net/lingBuzz/000427) Everett, Daniel 2007b Challenging Chomskyan Linguistics: The Case of Piraha˜. Human Development 50: 297–299. Everett, Daniel 2008 Dan Everett replies to Slobin and Sakel. Human Development, letters to the editor 50 (6). (content.karger.com/produktedb/ katalogteile/issn/_0018_716x/hde-letters-to-editor-01-09-2008. pdf ). Fox, Barbara A. and Sandra A. Thompson 2007 Relative clauses in English conversation – relativizers, frequency, and the notion of construction. Studies in Language 31(2): 293– 326. 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. Hopper, Paul 2000 Grammatical constructions and their discourse origins: Prototype or family resemblance? In: Martin Pu¨tz and Susanne Niemeier (eds.) Applied cognitive linguistics: Theory, acquisition and language pedagogy, 109–129. Berlin: Mouton de Gruyter. Jaeger, T. Florian and Thomas Wasow 2008 Processing as a source of accessibility e¤ects on variation. Proceedings of the 31st Berkeley Linguistics Society 169–180. Karlsson, Fred 2007 Constraints on multiple initial embedding of clauses. International Journal of Corpus Linguistics 12(1): 107–118. Liberman, Mark 2006 Parataxis in Piraha˜. Language Log May 19, 2006. (http://itre.cis. upenn.edu/Pmyl/languagelog/archives/003162.html). Matras, Yaron and Jeanette Sakel (eds.) 2007 Grammatical Borrowing in cross-linguistic perspective Berlin: Mouton de Gruyter. Nevins, Andrew Ira, David Pesetsky and Cilene Rodrigues 2007 Piraha˜ exceptionality: A reassessment. LingBuzz article archive. (http://ling.auf.net/lingBuzz/000411). Pawley, Andrew and Frances H. Syder 1983 Two puzzles for linguistic theory: nativelike selection and native-

Piraha˜ – in need of recursive syntax?

15

like fluency. In: Jack C. Richards, Richard W. Schmidt (eds.) Language and Communication, 191–226. New York: Longman. Sakel, Jeanette 2007a

Sakel, Jeanette 2007b

Slobin, Dan 2007

On Piraha˜ and Everett’s claims about recursion. Human Development, letters to the editor 50 (6). (content.karger.com/produktedb/ katalogteile/issn/_0018_716x/hde-letters-to-editor-12-13-2007. pdf ). Types of loan: Matter and pattern. In Grammatical borrowing, Yaron Matras & Jeanette Sakel (eds.) Grammatical Borrowing in cross-linguistic perspective Berlin: Mouton de Gruyter. Context and comments on Dan Everett’s claims. Human Development, letters to the editor 50 (6). (content.karger.com/produktedb/ katalogteile/issn/_0018_716x/hde-letters-to-editor-01-09-2008. pdf )

Stapert, Eugenie 2009 Universals in language or cognition? Evidence from English language acquisition and from Piraha˜. In: David Gill, Peter Trudgill, Geo¤rey Sampson (eds.) Language Complexity as an Evolving Variable, 230–242 Oxford: Oxford University Press. Stapert, Eugenie 2007 Linguistic theory and fieldwork in interaction: the case of Piraha˜. In: Peter K. Austin, Oliver Bond, David Nathan (eds.) Proceedings of conference on Language Documentation & Linguistic Theory. London: SOAS, University of London. Stapert, Eugenie, Michael Frank, Dan Everett, and Ted Gibson in prep. Embedded structures in Piraha: The expression of relative clauses, possessives and conditionals, manuscript. Thompson, Sandra A. and Anthony Mulac 1991 A quantitative perspective on the grammaticization of epistemic parentheticals in English. In: Elizabeth Traugott and Bernd Heine (eds.) Grammaticalization II, 313–339. Amsterdam: John Benjamins. Thompson, Sandra A. 2002 Object complements and conversation: towards a realistic account. Studies in Language 26(1): 125–164. Thurston, William R. 1987 Processes of change in the languages of north-western New Britain. Pacific Linguistics B99, The Australian National University, Canberra. 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.

16

Jeanette Sakel and Eugenie Stapert

Abbreviations 1 3 ATELIC COMP_CERT DM DOUBT EMPH HEARSAY INTENS LOC NEG OBSERV PERF REM SAI

first person third person atelic action/event complete certainty discourse marker expression of doubt emphasis hearsay evidential intensifier locative negation observation perfect remote -sai marker

2.

The fluidity of recursion and its implications

Marianne Mithun 1. Introduction One goal of linguistics is the identification of properties shared by all human languages and no other communication systems. In their muchcited 2002 paper, Hauser, Chomsky, and Fitch proposed a universal: ‘‘FLN (the Faculty of Language in the Narrow sense) only includes recursion and is the only uniquely human component of the faculty of language’’ (2002: 1569). While they provide no explicit definition of recursion, a survey of definitions current in the linguistics and computer science literatures can be found in Parker (2006: 167–190). One that is consistent with much current work in linguistics is that of Pinker and Jackendo¤ (2005: 4), whereby a recursive structure is characterized as ‘‘a constituent that contains a constituent of the same kind’’. Recursive structures are indeed pervasive cross-linguistically. A closer look at the variety of such structures that occur, however, indicates that recursion may not be the fixed, fundamental, hard-wired property envisioned. Much as languages vary in their distribution of structural complexity across the domains of morphology and syntax, they also vary in their distribution of recursion. The most common recursive structures involve noun phrases embedded inside of other noun phrases (the neighbor’s cat’s habits) and clauses embedded inside of other clauses, the type cited by Hauser, Chomsky, and Fitch (Mary believes that S ). As will be seen, such constructions are not uniform cross-linguistically, nor are they static within individual languages. The variability in space and time calls into question the status of recursion as the basic design feature of human language. It suggests, rather, that recursive structures are epiphenomenal, the product of combinations of a variety of cognitive processes. 2. Central Alaskan Yup’ik Languages of the Eskimo-Aleut family contain numerous syntactic constructions containing clause recursion. Examples cited here are drawn from Central Alaskan Yup’ik, a language of southwestern Alaska.

18

Marianne Mithun

2.1. Yup’ik complement clauses The single structure cited by Hauser, Chomsky, and Fitch as an example of recursion is complementation: Mary believes that S. Yup’ik counterparts to this English construction are formed with dependent clauses in the Subordinative mood. The Subordinative mood su‰x -lu- on the second verb in (1) marks the clause ‘that you are well’ as dependent. (1) Yup’ik complement counterpart: Elizabeth Ali, speaker p.c. Qanerutelaranka qaner-ute-lar-gar-nka talk-benefactive-habitual-transitive.indicative-1sg/3pl ‘I would tell them assirluten-gguq . . . assir-lu-ten¼gguq be.good-subordinative-2sg¼quotative [that you are well . . . ]’ Although this construction is now deeply entrenched in the grammar, its origin can still be discerned (Mithun 2008). The marker -lu- developed from an old nominalizer that still persists in derived nouns in all of the Eskimo-Aleut languages, though it is no longer productive in that function. (2) Yup’ik nominalizer -lu tamu‘to chew’ kumeg- ‘to scratch’ uig‘to sample, to taste’

tam-lu kum-lu u-lu

‘chin’ ‘thumb’ ‘tongue’

Further confirmation of nominalization as the diachronic source of the Subordinative comes from the shapes of the pronominal endings that occur with this mood. All Yup’ik verbs end in a pronominal su‰x specifying their core arguments: two for transitives and one for intransitives. In the transitive verb qanerutelara-nka ‘I would tell them’ in (1), the pronominal su‰x is -nka ‘I/them’. In the intransitive verb assirlu-ten ‘you are well’, the pronominal su‰x is -ten ‘you’. The pronominal su‰xes that occur with the Subordinative generally match the possessive su‰xes on nouns.

The fluidity of recursion and its implications

19

(3) Yup’ik possessive and pronominal su‰xes Possessive su‰x on noun Pro su‰x on Subordinative verb angya-ten assir-lu-ten boat-2sg/pl be.good-subordinative-2sg ‘your boats’ ‘your wellness’ ¼ ‘that you are well’ The formation of complement clauses by nominalization is not unusual cross-linguistically. This is no accident. The example in (1), ‘I will tell them [that you are doing well]’ was part of a telephone conversation between a sister and her brother, who was away visiting their parents. She asked how things were going, and he responded at length, saying that things were fine, and he was enjoying the daily round of activities. In closing, she promised to pass along his greetings to their friends. The clause ‘that you are well’ in (1) was not an assertion. It was a referential expression evoking a fact established in the preceding discourse. The reification of the situation is reflected in its syntactically nominal expression. 2.2. Yup’ik adverbial clauses Adverbial clause constructions are often cited as examples of clause embedding (Heine and Kuteva 2007, section 5.3.3, inter alia). Such constructions are pervasive in Yup’ik, marked by a variety of mood su‰xes on verbs. One is the Past Contemporative -ller- ‘when in the past’. The example in (4) is from a description of the silent Pear Film. (4) Yup’ik adverbial clause: Elizabeth Ali, Ellinqegcaarluki tlli-nqegcaar-lu-ki put-completely-subordinative-r/3pl atsaq atsaq fruit

speaker p.c. atauciq, atauciq one

igtellerani igte-ller-ani fall-past.contemporative-3sg

ataam tegukii ellirrarluku-ll’ ataam tegu-ke-ii elli-rrar-lu-ku¼llu again take-tr.prtcp-3sg/3sg put-after-subordinative-/3sg¼and ‘He placed them carefully, when one fruit fell, picking it back up and replacing it.’

20

Marianne Mithun

The origin of the Past Contemporative construction, like that of the Subordinative, is still discernible (Mithun 2008). The source of this mood su‰x is the past nominalizer -ller-/-lleq, which still persists in the language productively its original function. (5) Yup’ik past nominalizer -ller-/-lleq ayagaya-lleq

‘go, leave’ ‘the one who left’

The identification of this past nominalizer as the source of the Past Contemporative mood marker is confirmed by the fact that the pronominal su‰xes used with it have the same shapes as the possessive pronominals þ locative case endings on nouns. (6) Yup’ik possessed locatives and pro su‰xes on Past Contemporatives Possessed locative noun Past Contemporative verb angy-ani igtellerani boat-3sg/sg.loc fall-past.contemp-3sg ‘at/in his boat’ ‘at its falling’ ¼ ‘when it fell’ The locative case has been extended from simple nouns to nominalized clauses, its function generalized and made more abstract, from marking roles of concrete objects to marking the roles of events. Other dependent mood su‰xes in the language have similar histories. Each development reflects the reification of situations, the treatment of events or states as entities, the generalization of markers from individual words to clauses, and the development of more abstract functions, from denoting physical relationships of objects to indicating abstract relationships among events and ideas.

3. Khalkha Mongolian The Mongolian languages, genetically unrelated to Yup’ik and spoken in another part of the world, show a similar richness of clause combining. Examples cited here are drawn from Khalkha Mongolian. 3.1. Khalkha complement clauses A Khalkha complement construction can be seen in (7).

The fluidity of recursion and its implications

21

(7) Khalkha complement clause: Altanzaya Batchuuluun, speaker p.c. Tegeed so manai xu¨u¨xid u¨r’t n’ nadaas manai xu¨u¨xid u¨r’t n’ nad-aas our son earlier foc me-abl [dandaa qar utas avcˇ o¨go¨o¨cˇ. geǰ] dandaa qar utas av-cˇ o¨g-o¨o¨cˇ geǰ always hand phone take-non.finite give-pol comp quidag baisaÐ cˇin’. qui-dag bai-saÐ cˇin’ beg-hab be-past connector ‘Our son always used to beg me [to buy a cell phone].’ This complement construction is pervasive in Khalkha speech, but its diachronic source is still clear. The complementizer geǰ in (7) is a non-finite form of the verb ge- ‘say’. This construction apparently originated in a quotative construction: ‘He begged me, saying S’. The use of geǰ no longer requires actual speech, but its origins still constrain the contexts in which it occurs. In (8) it appears with a verb meaning ‘think’, and in (9) with ‘plan’. (Example (8) also shows extensive noun phrase recursion, with embedded genitives: the development of the religion of Buddhism of Mongolia.) (8) Khalka complement of ‘think’: Erdenebaatar Erdene-Ochir, sp p.c. E udaa bi this time I MoÐqoliiÐ Burxanii sˇasˇnii MoÐqol-iiÐ burxan-ii sˇasˇin-ii Mongolia-gen Buddha-gen religion-gen of Mongolia of Buddha of religion delgerliin tuxai, delger-l-iiÐ tuxai spread-nzr-gen about of the spread about boqa zereg jarij geǰ bodoǰ bain. boqa zereg jari-ǰ geǰ bodo-ǰ bai-na small level talk-nz comp think-non.finite be-non.past ‘This time I’m thinking [that I’ll talk a little about the development of the religion of Buddhism of Mongolia].’

22

Marianne Mithun

(9) Khalkha complement of Manai eeǰ aav manai eeǰ aav our mother father

‘plan’: Altanzaya Batchuuluun, speaker xojr aÐxnaasaa xojr aÐxn-aas-aa two first-abl-rfl.poss

[gurvaÐ xuxedtei boln geǰ] gurvaÐ xuxed-tei bol-no ge-ǰ three child-com be-fut comp jiro¨o¨so¨o¨ to¨lo¨vlo¨o¨gu¨i jiro¨o¨so¨o¨ to¨lo¨vlo¨o¨-gu¨i never plan-neg ‘At first my mother and father never planned [that they would have three children].’ The complementizer geǰ is not yet used in all contexts in which complementizers appear in other languages. Of the matrix predicate types listed in Noonan’s (2007) classification, geǰ appears with utterance predicates (such as ‘say’, ‘tell’, ‘report’, ‘promise’, ‘ask’, etc.), predicates of propositional attitude (‘believe’, ‘think’, ‘suppose’, ‘assume’, ‘doubt’, ‘deny’), pretence (‘imagine’, ‘pretend’, ‘fool’), knowledge and its acquisition (‘know’, ‘discover’, ‘realize’, ‘forget’, ‘see’, ‘hear’), fear (‘fear’, ‘worry’, ‘be anxious’), desire (‘want’, ‘wish’, ‘desire’), manipulatives (‘command’, ‘order’, ‘ask’, ‘allow’), and immediate perception (‘see’, ‘hear’, ‘watch’, ‘feel’), but not with what Noonan terms commentatives (‘be sad’, ‘be sorry’, ‘be odd’, ‘be significant’), with modals (‘be able’, ‘be obliged’, ‘can’, ‘ought’, ‘should’, ‘may’), achievements (‘manage’, ‘chance’, ‘dare’, ‘remember to’, ‘happen to’, ‘get’, ‘try’, ‘forget to’, ‘fail’, ‘avoid’), or phasals (‘start’, ‘begin’, ‘continue’, ‘keep on’, ‘finish’, ‘stop’, ‘cease’). The current limits are exactly what would be predicted on the basis of the diachronic source of the complementizer in a verb of saying. The development of verbs meaning ‘say’ into complementizers is of course not uncommon cross-linguistically (Heine and Kuteva 2002: 261–165). The Mongolian particle has not been generalized as far as complementizers from similar sources in some other languages, nor has it become as abstract. 3.2. Khalkha adverbial clauses Khalkha also shows pervasive use of adverbial clauses. One type can be seen in (10), from a description of the same silent Pear Film seen by the Yup’ik speaker cited in (4).

The fluidity of recursion and its implications

23

(10) Khalkha adverbial clause: Erdenebaatar Erdene-Ochir, sp p.c. Tegeed [neg liir qazar unx-ad] then one pear ground.loc fall-loc then one pear on ground when it fell ‘Then [when one pear fell to the ground], tern-iig av-aad that-acc take-sequential that took and he took it and (rubbed it with his bandana and put it into his basket).’ The origins of this temporal adverbial construction can also still be discerned. The adverbial su‰x -ad has the same shape as the locative case su‰x that appears on nouns. (11) Khalkha locative case on nouns: E. Erdene-Ochir, sp p.c. sagsan-d-aa liir xuraa-ǰ bai-saÐ basket-loc-rfl.poss pear collect-non.finite be-past in his own basket pear collecting was ‘he was collecting pears in his basket’ Other case su‰xes have developed into markers of other kinds of adverbial clauses. (12) Khalkha generalization of cases to adverbial clause markers Locative ‘at NP, when S’ Instrumental ‘with NP, by means of NP/S, as a result of, in order to S’ Ablative ‘from NP/S, because of NP/S As in Yup’ik, the adverbial clause constructions developed when speakers extended the use of case su‰xes from nouns to clauses, generalizing their functions from marking the roles of concrete objects to marking those of more abstract entities: events and states.

4. Mohawk In contrast with Yup’ik and Khalkha, Mohawk appears to show a surprising absence of embedded clauses. All Mohawk clauses are finite and can

24

Marianne Mithun

stand on their own as fully grammatical sentences. The usual formal features that identify embedded clauses in other languages, such as complementizers, omission of coreferential arguments, and non-finite verb forms, do not normally appear in their Mohawk counterparts. 4.1. Mohawk complement clauses Mohawk counterparts of all types of complement constructions distinguished by Noonan (2007) seem to consist of sequences of independent sentences. Example (13) was later translated by the speaker with an English object complement construction, but both Mohawk clauses are perfectly grammatical sentences on their own: ‘He didn’t know it. It happened thus.’ (13) Mohawk sequence of clauses: Kaia’tita´hkhe’ Jacobs, speaker p.c. Iah ki’ the´: tehoterie`n:tare’ iah ki’ the´: te-ho-ate-rien’tar-e’ not in.fact at all neg-m.sg.pat-middle-know-stative not in fact at all did he know (it) na’a´:wen’ne’. na’-a-w-en’n-e’ partitive-factual-n-fall-prf so it happened ‘In fact he didn’t know [what happened].’ The sentence in (14) was translated with an English subject complement. (14) Mohawk sequence of clauses: Cecelia Peters, speaker p.c. Enta`:’on enieka´ria’ke’ tere´ntsho. w-enta’-on en-ie-karia’k-e’ terentsho n-be.necessary-stative fut-indef.agt-pay-prf trente.sous it is necessary one will pay quarter ‘It is necessary [that one pay a quarter]’ ¼ ‘One has to pay a quarter.’ But most language use does not take place under conditions of silence. As soon as the intonational dimension is added, further structure emerges. Both (13) and (14) were pronounced under a single intonation contour, beginning with a high pitch reset and descending to a final fall only at the end of the second clause. These intonation patterns can be seen in the pitch traces in (13’) and (14’).

The fluidity of recursion and its implications

25

(13)’ Mohawk complement construction

‘In fact he didn’t know what happened.’ (14)’ Mohawk complement construction

‘One has to pay a quarter.’ Contrasts are easy to hear between constructions like these and sequences of separate sentences. The sentence in (15) below, ‘He told him to go home’, was pronounced as a single intonation unit, with a high pitch peak on the accented syllable of the first word/clause wahohro´:ri’ ‘he told him’ and much lower peak on the accented syllable of the second word/ clause: aonsahahte´n:ti’ ‘he should go home’. The syntactic cohesion of the two clauses is echoed in the use of the third person pronominal in the second clause ‘he should go home’, indicating indirect speech, rather than a second person ‘you should go home’. (15) Mohawk single sentence: Charlotte Bush, speaker p.c.

26

Marianne Mithun

Wahroro´:ri’ wa-ho-hrori-’ factual-m.sg/m.sg-tell-prf he told him ‘He told him to go home.’

aonsahahte´n:ti’. a-onsa-ha-ahtenti-’ opt-rep-m.sg.agt-leave-prf he should go home

Example (16), by contrast, contains two independent sentences, with a full fall in pitch at the end of the first and a pitch reset at the beginning of the second. The two are further separated by a pause. The syntactic separation of the sentences is reflected in the use of the first person pronominal in the second clause ‘for me to plant’, indicating direct quotation, rather than a third person ‘for him to plant’. (The second sentence also contains a complement of its own.) (16) Mohawk sequence of two sentences

Wahari’wano´n:ton’, wa-ha-ri’w-anonton-’ factual-m.sg.agt-matter-ask-prf ‘He asked, ‘‘Enwa´:ton’ ’’ wahe`n:ron’ en-w-aton-’ wa-ha-ihron-’ future-n-be.possible-prf factual-m.sg.agt-say-prf it will be possible he said ‘‘Would it be possible’’, he said, ‘‘ke`n: entie´ntho’.’’ ken: en-k-ientho-’ here future-1sg.agt-plant-prf here I will plant ‘‘for me to plant here?’’’

The fluidity of recursion and its implications

27

Intonation structure is not a simple mirror of syntactic structure: the two often operate in parallel, but they can also diverge to express di¤erent distinctions (Mithun 2006). The prosodic integration does, however, reflect a cognitive integration, which presumably underlies syntactic integration as well. The integration seen in the prosody is consistent with the status of the second clause in each sequence as a referring expression. Definiteness is not obligatorily distinguished in Mohawk, but there is a particle ne that can indicate that the following nominal is identifiable to the listener from previous reference in the discourse: ‘the aforementioned’. The Pear Film mentioned earlier opens with a man in a tree picking pears. A Mohawk speaker first introduced the pears without ne, since they were new to her listeners. (17) Mohawk absence of ne: Kaia’tita´hkhe’ Jacobs, speaker p.c. Ka´tshe’ ka´hi rahia´kwahs. ka-tshe’ ka-ahi ra-ahi-akw-ahs n-jug n-fruit m.sg.agt-fruit-pick-hab ‘He was picking pears.’ The pear picker repeatedly climbs down out of the tree and dumps the pears into baskets, then climbs back up. A boy comes along on a bicycle and, unobserved, steals one of the baskets. As he rides away, he hits a stone and falls, scattering the fruit. Three other boys come along and help him up. At this point the fruit, now familiar to the audience, is preceded by ne. (18) Mohawk ne with noun: Kaia’tita´hkhe’ Jacobs, speaker p.c. A’the´:rakon sahonne´ta’ ne ka´hi. a’ther-akon sa-hon-net’a-’ ne ka-ahi basket-place.inside rep.factual-m.pl.agt-insert-prf the n-fruit ‘They put the fruit back into the basket.’ The same particle ne occurs before complement clauses under exactly the same discourse conditions. It precedes complement clauses referring to events presumed to be identifiable to the audience from prior mention. Two Mohawk speakers were discussing a translation project. One expressed hope that it might be finished before Christmas. The other gave the response in (19). The anticipation of the final product was already an integral part of the conversation at this point, so the complement clause ‘to see what it will be like’ was preceded by ne. (This particle cliticizes to a following vowel-initial word.)

28

Marianne Mithun

(19) Mohawk ne with clause: Charlotte Bush, speaker p.c. Ne´: ki’ ni: tewakhsterı´henhs ne: ki’ ne¼ı`:’i te-wak-hsterihen-hs that in fact the I duplicative-1sg.pat-hurry-hab it is in fact myself it hurries me nakatka´htho’ ne¼a:-k-at-kahtho-’ the optative-1sg.agt-middle-see-prf the I would see it ni: tsi oh nenio´hton’ wa´hi’. ni: tsi oh n-en-io-ht-on-’ wahi’ how so how partitive-fut-n.pat-be.so-prf tag ‘It hurries me [the for me to see [how it will be]] you know.’ ¼ ‘I’m in a hurry to see what it will be like, you know.’ Similar behavior can be seen with demonstratives kı´:ken ‘this’ and thı´:ken ‘that’. Both can appear with nouns, as would be expected. Recall that the Mohawk speaker cited in (17) first introduced the pears without any preceding particles. As he worked, the pear picker stowed the pears in an apron pocket. At this point the speaker used the proximal demonstrative kı´:ken to refer to the pears, which had just been mentioned in the immediately preceding discourse. (20) Mohawk kı´:ken ‘this’ with noun: Kaia’tita´hkhe’ Jacobs, sp p.c. Tho ki’ iehre´ta’s tho ki’ ie-hr-et’a-s there in.fact translocative-m.sg.agt-insert-hab kı´:ken ka´tshe’ ka´hi. kı´:ken ka-tshe’ ka-ahi proximal n-jug n-fruit ‘[He was wearing something like a bag] and he was putting these pears in there.’ The demonstratives also appear before clauses, where they serve exactly the same function, situating the events denoted by the clauses as proximal or distal in space, time, or discourse. The pear picker climbed down from the tree periodically to transfer the pears into his three baskets. On one trip down, he noticed that a basket was missing.

The fluidity of recursion and its implications

(21) Mohawk kı´:ken before complement: Kaia’tita´hkhe’ Waha´ttoke’ kı´:ken wa-ha-at-tok-’ kı´:ken factual-m.sg.agt-middle-notive-prf proximal he noticed this

29

Jacobs, sp p.c. te´keni khok te´keni khok two only two only

tesewa’the´:raien te-se-w-a’ther-a-ien-’ duplicative-repetitive-n-basket-lie-stative two were basket lying ne ne the the ‘He

kahenta`:ke.’ ka-hent-a’ke n-field-place ground place noticed [this there were only two baskets on the ground].’

When asked later about her use of kı´:ken ‘this’ here, this speaker replied that the demonstrative referred to the fact of there being only two baskets. The proximal form was chosen because the man was noticing this fact as he looked at the baskets right before him. On another occasion, two other speakers were reminiscing about life during the 1930’s. One mentioned an old man, long since dead. This speaker’s use of the distal thı´:ken ‘that’ was appropriate for the distant time. (22) Mohawk thı´:ken ‘that’ with nominal: Tiorhakwe´n:te’ Dove, sp To´ka’ ni:se’ se`:iahre’, to´ka’ ne ´ı:se’ s-ehiahr-’ maybe the you 2sg.agt-remember-prf thike´n:– that Watia’ksne’ke´nha’. Watia’ksne’¼kenha’ name¼decessive ‘Maybe you remember that guy, the late Watia’ksne’.’ A bit later, the same speaker used the distal demonstrative before a clause, referring to an event at the same time in the past.

30

Marianne Mithun

(23) Mohawk distal thı´: ‘that’ before clause: Tiorhakwe´n:te’ Dove, sp To´ka’ ken enhsehia`:rake’ thi: toka’ ken en-hs-ehiahr-ak-’ thiken maybe Q future-2sg.agt-remember-cont-prf that wahshakonahskwawı´hon wa-hshako-nahskw-awi-hon factual-m.sg/3pl-domestic.animal-give-distr ‘Maybe you remember [that he (the government) used to give away livestock].’ (These demonstratives have not developed into complementizers comparable to English that. These Mohawk demonstrative constructions originated in a di¤erent source structure and di¤er syntactically, semantically, and prosodically from English that complements. Details are in Mithun 2006, in press). None of these particles, ne ‘the aforementioned’, kı´:ken ‘this’, or thı´:ken ‘that’, occurs before an independent sentence in isolation: they occur only before the kinds of clauses that could be identified as complements on semantic or intonational grounds. Their contexts of use have been generalized from simple nouns to clauses, indicating that the referents of these clauses, like those of nouns, are identifiable, proximal in space, time, or discourse, or distal. Their presence confirms that the clause is not an independent predication, but rather a referring expression. Mohawk thus contains complement constructions after all, but at an early stage of development. The construction is distinguished in three ways: 1) by coreference between a core argument of the first clause and the entire second clause, 2) by a single prosodic contour, and 3) by the fixed order of the clauses. The cross-linguistic tendency for heavy arguments to be postposed has become crystallized, with the complement always following the matrix. 4.2. Mohawk adverbial clauses Mohawk also contains constructions translated as adverbial clauses. Some of the most common types are temporal clauses set o¤ by the particle o´:nen. (24) Mohawk temporal adverbial clause: Watshennı´:ne’ Sawyer, sp p.c. Nek tsi o´:nen tiakwaristı´:ia’ke ne ok tsi o´:nen i-a’-t-iakwa-rist-iia’k-’ but when tloc-factual-dv-1.pl.excl.agt-metal-cross-prf

The fluidity of recursion and its implications

kwa´h ki: isto`n:ha kwah ki: o-sto`n¼ha quite this n-little¼dim ‘But when we crossed the

31

saion’we´:sen’ne. sa-io-on’wesen-’n-’ rep.factual-n.pat-be.pleasant-inch-prf border we felt a little better again.’

The particle o´:nen also occurs pervasively in independent sentences meaning ‘at the time’, ‘now’, or ‘then’. (25) Mohawk o´:nen ‘now’: Josephine Horne, Charlotte Bush, sp p.c. JH To: ken nitiako´:ien’ o´:nen? to: ken ni-t-iako-ien’ o´:nen how.many Q partitive-cisloc-f.pat-have now ‘How old is she now?’ ´ :nen ki’ CB O iakoierı´:ton tsa´:ta niwa´hsen. o´:nen ki’ iako-ierit-on tsiata ni-w-ahs-en now in.fact f.pat-complete-st seven prt-n-be.ten-st ‘She’s now seventy.’ ´ :nen ki’ JH En:. O ieio´he’ naiotorı´shen. en: o´:nen ki’ ie-io-he’ na-io-at-orish-en yes now actually tloc-n-be.time prt-n.pat-mid-rest-st ‘Yes. It’s time for her to retire now.’ One might be tempted to conclude that since o´:nen can be used as a temporal adverb without relational syntactic meaning as in (25), examples like (24) are simply sequences of independent sentences with the same adverb: ‘But then we crossed the border. We felt better again.’ There is, however, evidence that these are conventionalized constructions, complex sentences with embedded clauses. Like the complement constructions seen earlier, these constructions are pronounced under a single intonation contour, without an intermediate fall or pitch reset. (24)’ ‘But when we crossed the border we felt better again.’

32

Marianne Mithun

Temporal clauses may precede the matrix clause as in (24), or follow it as in (26), without implying an iconic sequence of events as they would in separate sentences. The sentence in (26) does not mean that it was pleasant and then people would cook. (26) Mohawk temporal adverbial clause: Awenhra´then Deer, speaker Ion’we´:sen’s ki’ thı´:ken io-on’wes-en¼’s ki’ thı´:ken n.pat-be.pleasant-stative-distr in.fact that o´:nen enthatikho´n:ni. o´:nen en-t-hati-khw-onni-’ o´:nen future-cislocative-m.pl.agt-meal-make-prf ‘In fact it used to be nice when they would cook.’ Furthermore, adverbial clauses, like complement clauses, can be preceded by particles like ne ‘the’, provided that the events referred to are identifiable to the listeners. The speaker cited in (27) had just been describing the di‰culty some children were having as they kept trying to saddle a horse. (27) Mohawk temporal adverbial clause with ne ‘the’: A. Deer, sp No´:nen enhonte’nie´n:ten ne o´:nen en-hon-ate-’nienten’ the when future-m.pl.agt-middle-try-prf sok tenka´:ta’ne’. sok t-en-ka-t-a’n-’ then duplicative-future-n.agt-stand-inch-prf ‘When they would try, it would stop.’ Finally, the translations provided by speakers confirm the fact that these constructions are interpreted not as sequences of separate, independent sentences, but rather as single, complex sentences, routinized constructions. The fact that we can identify a lexical source for the subordinating marker o´:nen is not surprising. As is well known, the most frequent sources of grammatical markers are lexical items (a tendency already noted by Gabelentz (1891) and Meillet (1912), and the object of the extensive work on processes of grammaticalization over the past several decades). The survival of uses of a particular form with lexical meaning in some contexts does not preclude its development of a grammatical function in others.

The fluidity of recursion and its implications

33

5. Cross-linguistic variation in recursive structures As seen in the preceding sections, Yup’ik, Khalkha, and Mohawk all contain syntactic structures that involve recursion, but their constructions are not identical to each other in function. The Yup’ik Subordinative is broader and more general than English complement constructions: it is used extensively to indicate syntactic and discourse cohesion, as in (4) above: ‘He placed them [subordinative] carefully, and when one fell, he picked it up, placing it [subordinative] in the basket’. The Khalkha geǰ construction, on the other hand, is used more narrowly than English complementation, limited to situations involving words or thoughts. There are also cross-linguistic di¤erences in the density of recursion in particular areas of syntactic structure. Some languages show productive recursion in noun phrases, with multiple levels of embedding, as in the Khalkha example in (8): ‘This time I’m thinking that I’ll talk a little about [the development [of the religion [of Buddhism [of Mongolia]]]].’ Others show little if any noun phrase recursion, expressing similar ideas in di¤erent structures. A Yup’ik speaker opened her description of the Pear Film with the sentence in (28). It contained three appositive nominals, each with an Ablative case su‰x marking its syntactic role in the matrix. (28) Yup’ik appositive noun phrases: Elizabeth Ali, speaker p.c. Qalarteqatartua, qalarte-qatar-tu-a speak-fut-intransitive.indicative-1sg I will speak watua uumek, watua una-mek just.awhile.ago this-abl just awhile ago about this one tarenramek tangllemnek. tarenraq-mek tangerr-lleq-mnek picture-abl see-past.nzr-1sg/sg.abl about picture about my seen one ‘I am going to speak about this picture I have just seen.’ Adding appositive nominals could in principle open the way to infinitely long sentences, but such constructions are actually examples of iteration rather than recursion. Parker (2006: 181) characterizes the di¤erence as follows.

34

Marianne Mithun Simply put, the di¤erence between iteration and recursion is this: the former involves mere repetition of an action or object, each repetition being a separate act that can exist in its entirety apart from the other repetitions, while the latter involves the embedding of an action or object inside another action or object of the same type, each embedding being dependent in some way on the action/object it is embedded inside.

The Hauser, Chomsky, and Fitch proposal pertains to recursion. In Mohawk, relationships among entities are not usually expressed in noun phrases at all. (29) Mohawk: Skawe´nnati Montour, speaker p.c. ´ :nen ken o´nhte’ O now Q perhaps rontate’ken’oko`n:’a, ron-atate-’ken’¼okon’a 3pl-rfl-be.siblings¼distr they are siblings to each other roie`n:’a. ro-ien’¼’a m.sg/m.sg-have as o¤spring¼dim he has him as o¤spring ‘Maybe it’s his brother’s son.’ Some languages show recursion in morphological structure, while others do not. The Yup’ik verb in (30) shows what is termed ‘indirect recursion’ (Heine and Kuteva 2007: 265–6). The past tense su‰x appears twice in the word, but it was not applied directly to its own output. (30) Yup’ik indirect recursion: George Charles, speaker p.c. Ayallrunillruat. ayag-llru-ni-llru-a-t go-past-say-past-transitive.indicative-3pl/3sg ‘They said he had left.’ The past stem ayag-llru- ‘left’ first served as the input to the derivation of a new stem, ayagllru-ni- ‘say left’, which in turn served as the base for the second application of the past tense su‰x. The Khalkha verb in (31) shows what is termed ‘direct recursion’ (Heine and Kuteva 2007: 265). The causative -uul was applied directly to a stem formed by a previous application of the same su‰x.

The fluidity of recursion and its implications

35

(31) Khalkha direct recursion: Erdenebaatar Erdene-Ochir, sp p.c. jav ‘go’ jav-uul ‘send’ (cause to go) javuul-uul ‘have (someone) send’ (cause to send) bi bi

zaxia javuulsaÐ. zaxia javuuluulsaÐ.

‘I sent a letter.’ ‘I had (somebody) send a letter.’

The recursive structure was not built up on a single occasion, through simple iteration of a su‰x. The complex stem jav-uul first became established as a lexical unit: javuul ‘send’. The causative -uul was then added to this stem as a whole. The idiosyncratic meaning of the first derived form persists in the second: the verb javuuluul means ‘have someone send’, not ‘cause to cause to go’. Recursive structures may be pervasive but they are far from homogenous in distribution or nature cross-linguistically. They occur in both morphology and syntax, but not in every language. They occur in noun phrases and clauses, but again not in all languages. Some recursive structures, like the noun phrase recursion seen in Mongolian, are the result of what Harder (this volume), citing Pinker and Bloom (1990: 724) and Parker (2006: 5.5.3), describes as ‘top down’: successive specification is added to already existing constituents, resulting in an ever finer level of precision of description. Others, like that seen in complementation, are the result of what Harder describes as ‘bottom-up’ construction: a situation may be first asserted in a predication, as in the Yup’ik ‘I am fine’, then later reified and referred to in a new predication ‘I will tell them [that you are doing fine]’. Information conveyed by embedding in some languages is conveyed by iteration in others. When the Mohawk speaker described the Pear Film, she used the complement construction in (32). (32) Mohawk complementation: Kaia’tita´hkhe’ Jacobs, sp p.c. Sok kı´:ken raksa`:’a so this boy wa`:rehre’ he thought Iah ki’ na:’ the´: thaha´ttoke’ ki: iah ki’ na`:’a the´:nen th-a:-ha-attok-’ kı´: not in fact guess at all contr-opt-m.sg.agt-notice-prf this

36

Marianne Mithun

osto`n:ha ia´:khawe’ ki: rao`:hi. oston¼ha i-a:-k-haw-’ kı´:ken rao-ahi little¼dim tloc-opt-1sg.agt-carry-prf this m.sg.al.poss-fruit ‘So this boy thought, ‘‘I guess he wouldn’t notice [that I took a little of this fruit].’ Describing the same film, the Mongolian speaker used a series of adverbial clauses. (33) Mongolian adverbials: Erdenebaatar Erdene-Ochir, speaker p.c. Tegeed ter duguitai xu¨u¨xed xaǰuuqaar n’ neg teg-eed ter dugui-tai xu¨u¨xed xaǰuu-qaar n’ neg do.so-seq that circle-com child side-inst foc one zo¨ro¨x gesnee, uh zo¨r-o¨x ge-sen-ee, uh pass-infinitive say-past-rfl hes ter ter that

xu¨nii xu¨n-ii person-gen

liir xuraaǰ baiqaag xaraad, liir xuraa-ǰ bai-qaa-g xar-aad pear collect-non.finite be-prog-acc see-seq sagstai liiru¨u¨d, ter gazar sags-tai liir-u¨u¨d ter gazar basket-com pear-pl that ground sagstai liiru¨u¨diig sags-tai liir-u¨u¨d-iig basket-com pear-pl-acc tegeed teg-eed thus-seq

ter ter that

o¨o¨riig n’ o¨o¨r-iig n’ self-acc foc ter ter that

deer baigaa deer bai-gaa above be-prog

xaraad xar-aad see-seq

xu¨Ð xu¨Ð person aÐzaaraagu¨i boloxoor aÐzaar-aa-güi bol-ox-oor notice-rfl-neg become-inf-inst

liiriig n’ liir-iig n’ pear-acc foc

xulqailǰ xulqai-l-ǰ theft-verbalizer-non.finite

terter that

xu¨u¨xed xu¨u¨xed child

baiqaa bai-qaa be-prog

The fluidity of recursion and its implications

jum jum thing

37

sˇig baiaÐ. sˇig bai-saÐ like be-past

‘Then, it looked like when the child with the bicycle was about to pass by, having seen the man collecting the pears, having seen the pears being in the baskets, that pears with the baskets were on the ground, since the man didn’t notice him, the child was stealing the pears.’ 6. The dynamism of the recursive constructions The constructions that exhibit recursive structure are not static within languages. The most commonly cited examples of recursion in syntax, the type mentioned by Hauser, Chomsky, and Fitch, are complement constructions. The Yup’ik Subordinative construction is deeply embedded in the grammar, marked by verbal su‰xes, but we can still trace its origin in clause nominalization. The Khalkha geǰ construction developed through the generalization of a quotative construction. It is not as advanced in its development as the Yup’ik Subordinative, nor is it used in the same wide range of contexts: it still shows the full formal structure of direct quotation and is restricted to contexts involving words or thoughts. The Mohawk complement construction shows an even earlier stage of development. There is no complementizer, no omission of coreferential arguments, and no reduction in finiteness. The clausal integration is often marked only intonationally. Still, the status of the complement clause as a referring expression rather than an independent predication is confirmed by its ability to be preceded by a determiner. Adverbial clause constructions show similar variability in their origins and stages of development. Yup’ik Past Contemporative clauses (‘when in the past’) have their source in nominalized clauses inflected for locative case. The Khalkha counterpart shows a similar development through the extension of the locative case su‰x from nouns to clauses. The source of the corresponding Mohawk temporal adverbial clause construction can still be traced as well, but it developed in a di¤erent way and is less deeply integrated into the grammar, marked just by the particle o´:nen ‘at the time’ and intonation. The fact that we can see the origins of these embedded clause structures should not be taken as evidence that the languages lacked recursion at an earlier stage. Each of these languages in fact shows evidence of the earlier existence of complex sentence constructions.

38

Marianne Mithun

As is well known, the most common sources of a‰xes are full lexical items. The Eskimo-Aleut languages contain a wealth of modern morphological formations that are apparently descended from complex sentence constructions, like -ni- ‘say’ in the Yup’ik example in (30) above: ‘He said that he had left.’ The Mongolian languages also show morphological formations suggestive of earlier complement constructions, like the causatives based on the su‰x -uul seen in (31): jav-uul ‘cause to go’ ¼ ‘send’. The Iroquoian languages also show traces of earlier complex sentence constructions in their morphologies. Among the derivational su‰xes on verbs are applicatives which add a core argument, such as an instrument, a beneficiary, or a location. One of the instrumental applicative su‰xes is -hst. (34) Mohawk Instrumental Applicative iera’wistakarha’tho´hstha’ ie-ra’wist-a-karha’tho-hst-ha’ one-pancake-linker-flip-inst.applicative-hab ‘one pancake flips with it’ ¼ ‘spatula’ The source of this su‰x persists in the modern Iroquoian languages as the verb root -hst ‘use’, as in Tuscarora kı´hstha’ and Mohawk ka´tsha’ ‘I use it’. After a complement construction has emerged, it can continue to evolve. When the Mohawk speaker cited earlier first began describing the pear picker in the Pear Film, she included two short, barely audible words: ´ı:kehre’ ‘I think’ and o´nhte ‘perhaps’. Both still function in other contexts as matrix verbs, but here they are reduced to barely audible discourse particles, no longer restricted to a position immediately before a complement clause. (35) Mohawk nascent discourse particles: Kaia’tita´hkhe’ Jacobs, speaker Kı´:ken ro´n:kwe, this man kwah kwah just just

´ı:kehre’ k-ehr-’ 1sg.agt-believe-st I believe

ka´tshe’ katshe’ jug jug

ka´hi kahi fruit fruit

o´nhte, a:-w-aht-e opt-n.agt-be.so-st it could be so

rahia´kwahs. ra-ahi-akw-ahs m.sg.agt-fruit-pick-hab he fruit picks ‘This man, I guess, was perhaps just picking fruit.’

The fluidity of recursion and its implications

39

Such gradience in evolving forms is of course common cross-linguistically. A well-known example is the English you know (Thompson 2002).

7. The notion of recursion as the essence of human language Recursive structures, in which one constituent is embedded inside of another of the same type, are pervasive cross-linguistically, perhaps more pervasive than might first appear. But as we have seen, they are neither homogenous across languages nor static within them. The variation and gradience suggest that recursion is not the basic, hard-wired feature proposed. Recursive structures are in a sense epiphenomenal, the products of a host of cognitive abilities. Among these is an ability to integrate elements of what is understood as a larger idea into a larger, coherent structure. We saw this in the prosodic integration of sequences of clauses under a single intonational contour. Another is the capacity to reify events, to refer to situations as entities. We saw this in the formation of complement constructions from nominalized clauses in Eskimo-Aleut. Another is the capacity for abstraction, seen in the expansion of Mongolian quotative constructions to complement constructions involving thoughts and intentions. Still another is the capacity for generalization, seen in the extension of case markers from nouns to clauses in the formation of adverbial clause constructions. Yet another is the ability to routinize patterns, so that constructions need not be created anew each time they are used. Such routinization is behind the continuing evolution of certain complement constructions into discourse markers, as in Mohawk. It is also behind the chunking of morphologically complex stems into lexical units, which can in turn serve as the basis for subsequent derivational processes, as in the Khalkha causative ‘go-cause’ ! ‘send’; ‘send-cause’ ! ‘have send’. In the end, the logical possibility of infinitely long sentences is neither the goal of speakers nor the essence of language. It is an artifact of a view of language as a mathematical system instead of the continually evolving product of human cognitive abilities.

Abbreviations abl ¼ ablative, acc ¼ accusative, agt ¼ grammatical agent, cisloc ¼ cislocative, comp ¼ complementizer, dim ¼ dimunitive, distr ¼ distributive, dv ¼ duplicative, excl ¼ first person exclusive, f ¼ feminine gender, foc ¼ focus, fut ¼ future, gen ¼ genitive, inst ¼

40

Marianne Mithun

instrumental, loc ¼ locative, m ¼ masculine, mid ¼ middle, n ¼ neuter, neg ¼ negative, nz ¼ nominalizer, pat ¼ grammatical patient, prf ¼ perfective, prog ¼ progressive, r ¼ coreferential 3, rfl ¼ reflexive, seq ¼ sequential, transloc ¼ translocative

References Gabelentz, Georg von der 1891 Die Sprachwissenschaft: ihre Aufgaben, Methoden und bisherigen Ergebnisse. Leipzig: Weigel. Harder, Peter This volume. Over the top-recursion as a functional option. 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 22:298.5598: 1569–1579. Heine, Bernd and Tania Kuteva 2002 World Lexicon of Grammaticalization. Cambridge, UK: Cambridge University Press. Heine, Bernd and Tania Kuteva 2007 The Genesis of Grammar: A Reconstruction. Oxford, UK: Oxford University Press. Meillet, Antoine 1912 L’e´volution des formes grammaticales. Scientia 12/26 (Milan). Reprinted 1951 in Linguistique historique et linguistique ge´ne´rale, 130–148. Paris: C. Klincksieck. Mithun, Marianne 2006 Threads in the tapestry of syntax: Complementation and Mohawk. In Jacqueline Bunting, Sapna Desai, Robert Peachey, Christopher Straughn, Zuzana Tomkova´ (eds.), Papers from the Forty-second Meeting of the Chicago Linguistics Society, 213– 238. Chicago: Chicago Linguistic Society. Mithun, Marianne 2008 The extension of dependency beyond the sentence. Language 83.69–119. Mithun, Marianne 2009 Re(e)volving complexity: Adding intonation. In T. Givo´n and Masayoshi Shibatani, (eds.), Syntactic Complexity: Diachrony, Acquisition, Neuro-cognition, Evolution, 53–80. Amsterdam/Philadelphia: Benjamins. Noonan, Michael 2007 Complementation. In Timothy Shopen (ed.), Language Typology and Syntactic Description. 2nd ed, 52–150. Cambridge, UK: Cambridge University Press.

The fluidity of recursion and its implications Parker, Anna 2006

41

Evolution as a constraint on theories of syntax: The case against Minimalism. Ph.D. diss, Department of Linguistics, University of Edinburgh. Pinker, Stephen and Raymond Jackendo¤ 2005 The faculty of language: What’s special about it? Cognition 95: 201–236. Thompson, Sandra 2002 ‘‘Object complements’’ and conversation: Towards a realistic account. Studies in Language 26.1: 125–164.

3.

Syntactic recursion and iteration

Fred Karlsson

1. Introduction1 The nature and origin of syntactic recursion in natural languages is a topical problem. Important recent contributions include those of Johansson (2005), Parker (2006), Tomalin (2006; 2007), and Heine and Kuteva (2007). Syntactic recursion will here be discussed especially in relation to its cognate concept of iteration. Their basic common feature is plain structural repetition: ‘‘keep on emitting instances of the current structure, or stop’’. Their main di¤erence is that recursion builds structure by increasing embedding depth whereas iteration yields flat output structures which do not increase depth. My focus here is on the types of recursion and iteration, and on what empirically determinable constraints there are on the number of recursive and iterative cycles of application. Recursion comes in two subtypes, nested recursion (¼center-embedding) and tail-recursion, the latter covering left-recursion and right-recursion. There are six functionally di¤erent types of iteration: structural iteration, apposition, reduplication, repetition, listing and succession. It will be empirically shown that multiple nested syntactic recursion of degrees greater than 3 does not exist in written language, neither in sentences nor in noun phrases or prepositional phrases. In practice, even nesting of degree 2 is extremely rare in writing. In speech, nested recursion at depths greater than 1 is practically non-existing, thus partly confirming an early hypothesis of Reich (1969). Left-branching tail-recursion of clauses is strictly constrained to maximally two recursive cycles. Right-branching clausal tail-recursion rarely transcends three cycles in spoken language and five in written language. On constituent level both left- and rightbranching is less constrained (especially in written language), but e.g.

1. I thank three anonymous referees for valuable comments and suggestions for improvement.

44

Fred Karlsson

left-branching genitives rarely recurse more than two cycles ([[[Pam’s] mum’s] baggage]).

2. Types of sequential arrangement There are many types of rules and restrictions on the sequential arrangements of words, constituents and clauses. Most important in any language are the principles of basic linearization of obligatory elements. These are word order restrictions and tendencies governing the arrangements of grammatical elements and simple canonical constituents, clauses and larger structures, e.g. (for English typical pragmatically unmarked declarative sentences): – a grammatical subject precedes its main verb, – a grammatical object follows its main verb, – a subcategorized locative adverbial follows it main verb. This obligatory basic linearization of word classes, constituents, heads and dependents, grammatical functions etc. defines the co-occurrence and word-order properties of the grammatical elements and constructions. This core grammar is the foundation of the optional complexity-introducing types of sequential arrangement such as iteration and recursion. Every natural language has basic linearization principles which also determine its classification in word order typologies. These rules are mostly local and simple precedence statements. A special type of basic linearization is that of lexicalized multiword expressions: phraseology (1) and lexical bundles (2), i.e. expressions frequent in a certain register (Biber et al. 1992: 993): (1) you know (2) I was going to Every language has lexicalized multiword expressions, at least lexical bundles. Most languages have semantically non-compositional idioms and phraseology (however, Piraha˜ is claimed by Dan Everett, p.c., to have no idioms nor lexical bundles). Minimal canonical structures may be optionally elaborated by free modification, e.g. by adding adjectival or genitival premodifiers to nouns, intensifiers as premodifiers to adjectives, free adjuncts to verbs, and noncomplement clauses subordinate clauses to main clauses:

Syntactic recursion and iteration

– – – – – –

45

a determiner precedes its nominal head, a determiner precedes an adjectival modifier of the same head, an adjectival modifier precedes its nominal head, a case-marked genitival NP precedes its nominal head, an if-clause prefers initial embedding position, a when-clause may be embedded initially or finally.

There are also information-packaging principles such as inversion, preposing, postposing, extraposition, and ‘‘fat things last’’, which interact with the principles of basic linearization and free modification, adapting the orders of canonical arrangements to fit textual and pragmatic needs. These three mechanisms together define grammatically well-formed and situationally appropriate sentences such as (3). (3) Nat’s fat cat sat on the mat when it spat out a rat. Recursion and iteration are also types of sequential arrangement, but layered on top of basic linearization, free modification and informationpackaging by invoking repetition: ‘‘keep on emitting instances of the current structure, or stop’’. Their main di¤erence is that recursion builds structure by increasing embedding depth whereas iteration yields flat output structures, repetitive sequences on the same depth level as the first instance. Recursion and iteration are the main topics of this paper and they will here be analyzed in conformance with Wilhelm von Humboldt’s insightful early agenda: (4) ‘‘Ist aber auch die [grammatische] Freiheit an sich unbestimmbar und unerkla¨rlich, so lassen sich dennoch vielleicht ihre Gra¨nzen innerhalb eines gewissen ihr allein gewa¨hrten Spielraum au‰nden; und die Sprachuntersuchung muß die Erscheinung der Freiheit erkennen und ehren, aber auch sorgfa¨ltig ihren Gra¨nzen nachspu¨ren.’’ (von Humboldt 1836: 81; italics added/FK) ‘Even if [grammatical] freedom in itself would be impossible to determine and explain, it might still perhaps be possible to find its limits within the confines of a certain latitude reserved for it, and language study must recognize and respect the phenomenon of freedom, but also carefully try to determine its limits.’ In this spirit, it is relevant to ask both what types of recursion and iteration there are, and what empirically determinable constraints (if any) there are on the number of repetitive cycles in both recursion and iteration. I first turn to iteration.

46

Fred Karlsson

3. Six types of iteration On both formal and on functional (partially semantic, partially psycholinguistic) grounds, six types of iterative sequential arrangement can be distinguished: 1) structural iteration, 2) apposition, 3) reduplication, 4) repetition, 5) listing, and 6) succession. They all concatenate elements without additional depth-increasing structure building. The latter feature distinguishes them from recursion, which creates hierarchic structures governing the linear strings it produces, typically by repeatedly subordinating an element to another instance of the same type. 3.1. Structural iteration The most important type of iteration is structural iteration which is overwhelmingly most frequent. It has no clear semantic task and this too sets it o¤ from the other five types of iterative sequential arrangement which are functionally motivated special cases. The main manifestation of structural iteration is coordination, with or without explicit conjunctions. The gist of coordination and all other types of structural iteration is ‘‘optionally add an instance of the same structural type as the current one, with (syndetically) or without (asyndetically) a coordinating conjunction’’. Prototypical structural iteration works much like the Kleene star, ‘‘*’’, in formal language theory. All major constituent types (NP, AP, PP etc.), grammatical functions (subject, object, adverbial etc.) and clause types (relative clauses, ifclauses, infinitive clauses etc.) may be iteratively coordinated, on the same level of structure. Also many free modifiers may be iteratively coordinated, e.g. adjectival modifiers (but not determiners) and case-inflected genitival NPs as premodifiers of nominal heads. It is thus an inherent property of grammatical elements and construction types whether they are iterative, recursive or neither or both. By far the two most important and frequent iterative structures are coordination of clauses in sentences and of NPs in clauses. There are no bounds on the number of constituent-level conjuncts, or on how many coordinated clauses there may be in a single sentence. For example, in Hermann Broch’s (1961: 138–144) novel Der Tod des Vergil there is a 1077-word sentence extending over six pages, with 81 finite and 38 nonfinite clauses, and several hundred instances of coordination on modifier, constituent and clause level. By contrast, observe that depth-increasing recursion with multitudes

Syntactic recursion and iteration

47

(e.g. 100, or even 10) of successively embedded subordinate clauses never is the central mechanism underlying such unusually long sentences. Thus, the maximal clausal embedding depth of Broch’s gigantic sentence is a modest 3, and this is reached only once. In this sentence there are only two instances of subordinate clauses at depth 2. To pick another example, consider the famous chapter of Ulysses where Joyce describes the stream of consciousness of Molly Bloom in eight huge paragraphs without punctuation. One sentence is 12,931 words long. The first ten pages of this chapter (Joyce 1934: 723–732; 5,500 words) contain one instance of recursive final (right-branching) clausal embedding reaching depth 6 (‘‘F’’ ¼ final, the numbers as in F-2 indicate embedding depth): (5) [ . . . she said yes [F-1 because it grigged her [F-2 because she knew [F-3 what it meant [F-4 because I used [F-5 to tell her a good bit of [F-6 what went on between us . . . ]]]]]]] (Joyce 1934 [1922]: 729) But such depths are very untypical. The abnormal length of Joyce’s 12,931-word sentence is caused by extensive use of coordination. One more indication of the prevalence of iterated coordination is the fact that the conjunction and or its equivalent is one of the most common words in many languages. Clause chaining is another variant of iterative clause linkage, closely related to asyndetic coordination. Likewise, certain types of verb serialization are basically iterative in nature (Sebba 1987: 111). 3.2. Apposition Appositions are typically iterated NPs on the same level of depth. Their defining feature is mutual coreference. This property sets appositions o¤ from coordination as a special, semantically motivated type of iteration. (6) is a genuine fivefold Swedish appositive structure (coreferential nominal heads in boldface). (6) Ej la˚ngt da¨rifra˚n sitter ett fo¨rema˚l fo¨r ma¨ngdens ka¨rleksfulla not far away sits an object for crowd’s loving hyllning, ovation Henrik Gabriel Porthan Henrik Gabriel Porthan, akademiens yppersta la¨rare, Academy’s most magnificent teacher

48

Fred Karlsson

den fosterla¨ndska ha¨fdaforskningen fader, the national historical research’s father en man som gaf sitt namn a˚t ett helt tidehvarf i va˚r a man who gave his name to a whole epoch in our kulturhistoria. cultural history ‘Not far away sits the object of the loving ovation of the crowd, Henrik Gabriel Porthan, the most magnificent teacher of the Academy, the founder of national history, a man who gave his name to a whole epoch in the history of our culture.’ The number of cycles in iterated apposition is unconstrained but numbers in excess of a handful are rare. This property too sets apposition o¤ from structural iteration for which the number of cycles is truly unconstrained and easy to document by concrete examples. 3.3. Reduplication Syntactic reduplication, often called repetition, is a motivated type of iteration, where a word (normally a content word) is repeated in order to express meaning modulations such as intensification, augmentation, repetition, diminution, or iterative or continuative action. Usually the construction is asyndetic, but it may be syndetic as well, especially with adverbs, comparatives, and verbs. Here are some originally Swedish examples from Lindstro¨m (1999), in direct translations all of which qualify equally as idiomatic English (and all of which can be found on the internet): (7) a. b. c. d. e. f. g.

There were cars cars cars. a long long way small small creatures much much better played better and better came closer and closer They walked and walked and walked.

In his corpus-based study of syntactic reduplication in 1.7 million words of spoken and written Swedish, Lindstro¨m (1999) noted the predominance of reduplicated pairs with two identical terms. Triples are rare, quadruples extremely rare. Therefore reduplication is indeed the most apt term for this type of iteration. The incidence of asyndetic reduplication (without conjunctions) in Swedish is one per 3000 words, higher in spoken than in written language, and of syndetic reduplication one per

Syntactic recursion and iteration

49

5000 words. Of 620 Swedish instances of reduplication only five were triples. These are clear preferences. On the other hand, they do not strictly rule out longer chains (which are easy to attest e.g. by googling). 3.4. Repetition, listing, succession The iterative concept of plain repetition (or repeat, in the terminology of Biber et al. 1999: 1055) concerns the special, psycholinguistically motivated case of speech planning. The speaker repeats (often involuntarily) some word or syntagm, especially words belonging to closed grammatical classes, in order to gain time to complete planning the message. The outcomes of such repetitions are formally ungrammatical. (8) a. b.

Yeah, it, it, it is it’s it’s it’s it’s good. (Biber et al. 1999: 1055) And and / if if if you know they had none at all. (British National Corpus, FMD 321)

The corpus data of Biber et al. show that the likelihood of the repetition decreases sharply with the number of words repeated. The vast majority of examples contain a single repeat. There are extremely few instances of three or more repeats. Listing (enumeration) is an iterative mechanism predominantly used in restricted lexical taxonomies, see (9a) and (9b): (9) a. b. c.

Smith, Brown, Jones . . . Monday Tuesday Wednesday . . . one two three . . .

A special kind of listing is succession (9c), which underlies numerosity (Heine and Kuteva 2007: 271). Given any number n, the successor function enables production of n þ 1. Repetition, listing and succession of course have no limits. 4. Syntactic recursion: historical and conceptual background Heine and Kuteva (2007: 264–266) discuss several important conceptual distinctions relating to recursion: direct recursion (modeled by rules like A ! AB) vs. indirect recursion (A ! B, B ! A), counting recursion (yielding AABB) vs. mirror recursion (yielding ABBA), and simple recursion (with only once application cycle) vs. productive recursion (with an unlimited number of cycles). The di¤erence between nested recursion and tailrecursion is the most important one from the viewpoint of the present paper. Nested recursion and tail-recursion have often been presumed to be

50

Fred Karlsson

unrestricted (or productive in Heine and Kuteva’s terminology). For tailrecursion this is by and large true, but in Humboldt’s spirit (1) it will be demonstrated below that there are quantitative constraints especially on nested syntactic recursion of clauses, noun phrases and prepositional phrases. The use of nested and tail-recursive rules in grammatical description is ancient because such rules were formulated already around 400 BC, in Pa#n¸ini’s Sanskrit grammar (Bhate and Kak 1993: 79) and in Tholkaappiyar’s Tamil grammar (Baladunarman, Ishwar, and Kumar 2003). In recent linguistic theory recursion was reintroduced around 1950, probably first by Bar-Hillel (1953), under influence from the theory of recursive functions in mathematical logic (Tomalin 2006: 64). In Chomsky’s PhD dissertation (1975 [1955–1956]) recursion is discussed in several connections, e.g. as the source of sentential infinity and as for its proper locus in the grammar. However, Chomsky’s (1956: 65; 1957: 21–23) original published claims concerning the alleged unconstrainedness of nested recursion in English syntax are sketchy and rely only on examples invented by Chomsky himself. It is surprising to note that prior to the current millennium the deeper nature and the various types of recursion in natural languages have not been discussed at great length in any brand of grammatical theory (however, see Halliday 1963 for an early exception). Parker (2006) surveys the prevailing conceptual and terminological variation. Table 1 provides an overview of some relevant concepts and terminology in a mainly generative and computational framework. Table 1. Basic conceptualization and terminology of recursion. Type of rewrite rule (a, b A 0)

Mode (orientation) of recursion

Relation to iteration

Linguistic term

A!A a

left-recursion

left-branching ¼ initial embedding

A!a A

right-recursion

A!a A b

nested recursion

tail-recursion ¼ convertible to iteration2 tail-recursion ¼ convertible to iteration full-blown recursion, not convertible to iteration

right-branching ¼ final embedding nesting, centerembedding

2. Tail-recursion is always convertible to simple iteration (Aho, Sethi, and Ullman 1986: 53).

Syntactic recursion and iteration

51

A standard view of recursion in the formal sciences is the top-down one that a function is recursive if it is defined in terms of successively simpler versions of itself such that a termination condition (variably called the baseline, base case, or degenerate case) guarantees the process to halt. A recursive function is not well-formed if its termination is not guaranteed (Brookshear 2007: 246). A well-formed recursive procedure is designed to test for the termination condition before requesting further cycles of activation. Material introduced by any lower recursive cycle is always contained (embedded) in the material introduced by the immediately higher cycle. Every applicative cycle increases the recursive depth by 1. Thus, the depth of sentence (5) is 6. The typical generative-linguistic conception of recursion on the contrary emphasizes that syntactic recursion lacks termination conditions: this is the very property that guarantees the infinity of sentence length and structural elaboration. Nested recursion (center-embedding) is recursion par excellence. A nested structure Xn is embedded within a structure of the same type, Xn1, with non-null material (excluding subordinating conjunctions, cf. endnote 3) of Xn1 both to the left and to the right of Xn. Nested recursion consumes memory resources and presupposes storage of pending material and return addresses on a stack in order to secure proper unwinding. This sketchy description is equally true of human language processing and its counterparts in formal language theory and computational linguistics. Nested recursion of precisely the same type of element (e.g. two relative clauses, two if-clauses, two NPs) is called self-embedding. Formally, leftrecursion and right-recursion can always be converted to iteration, i.e. to repetition (one more cycle) of the same procedure, without anything pending on the stack (Aho, Sethi, and Ullman 1986: 53). Structures like ‘‘if if if S1, then S2, then S3, then S4’’, i.e. counting recursion (Christiansen and Chater 1999: 168) were introduced into the public theoretical discussion already by Chomsky (1956: 65) and claimed to be unrestricted as for recursive depth. For reasons mentioned in endnote 3, such sentences are here regarded as initially-embedded, not as centerembedded. Karlsson (2007a) demonstrated that there is a strict empirically determinable quantitative constraint on repeated initial embedding of clauses, viz. (10) to be discussed shortly. A final note is in order concerning the ontological nature of recursion. Halliday (1963: 85–13) is clear on interpreting recursion as an integral part of language structure. A structural element is repeated in depth, a series of such elements thus forming a progression. This ontological commitment also is at the heart of generative grammar, where recursion has

52

Fred Karlsson

always been regarded as the very structural feature that caters for the presumed infinity of language (i.e., a category on the left-hand side of a rewrite rule reoccurs on the right-hand side, cf. Table 1). But there are other views. For Linell (2005: 75), in spoken interaction long utterances with multiple embeddings are emergent results of local decisions of moving from one turn-constructional unit to another. They are not part of the grammar per se. The grammar contains just a core syntax plus methods for moving to the next turn-constructional unit. Similarly, Harder (2006) treats recursion as a semantic and functional option of use, a possibility allowed by the whole system, arising as a more or less epiphenomenal consequence or side e¤ect when a functional upgrading operation such as reusing a that-clause after reporting verbs happens to concern a structural type that is already present in the input. Clausal recursion as such is not built into the fabric of grammar. Heine and Kuteva (2007: 265) opt for an intermediate agnostic position. According to them, recursion is not a property of language but rather the product of a theory designed to describe language structure: ‘‘. . . we are not maintaining that language, or language structure, has recursion but rather that there are certain properties to be observed in language that can appropriately be described in terms of a construct [of formal recursion]’’. These views need not be fully contradictory. Properties of languages can be described by formal mechanisms without necessarily renouncing the ultimate functional motivation of the properties. Allowing multiple (even if not fully unconstrained) right-recursion in core grammar seems to be more or less equivalent to Linell’s basic embedding patterns þ methods for moving to the next turn-constructional unit. On the other hand, every theoretical concept is a hypothesis about a phenomenon in some more or less real world and the question whether the concept properly represents the ‘‘God’s-truth Ding an sich’’, or is a plain ‘‘Hocus Pocus construct’’, in the sense of Householder’s (1952) famous and much-debated distinction, can be left open for philosophical discussion.

5. Constraints on clausal recursion Clausal embedding recursion means repeated embedding of clauses in initial, central or final position of the superordinate clause. Initially-embedded clauses (IEs) have either nothing to their left, or a subordinating

Syntactic recursion and iteration

53

conjunction of the superordinate clause.3 Center-embedded clauses (CEs) have superordinate clause constituents to their left (other than subordinating conjunctions) and right. Final embeddings (FEs) have no material rightwards that immediately belongs to the superordinate clause. IE and FE manifest tail-recursion, CE is nested recursion. Karlsson (2007a; 2007b) has empirically established constraints, quantitative limits and qualitative restrictions, on clausal embedding recursion in many European languages. His data were (i) complex sentences in the British National Corpus (BNC), the Brown corpus, and the LOB corpus, (ii) computerized extraction of complex embedding patterns in Finnish, German, and Swedish, and (iii) consultation of 100þ corpus-based syntactic and stylistic descriptions of European languages, especially Latin and older variants of German. Here (10-14) are some constraints operational in many European languages (at least in Danish, English, Greek, Finnish, Latin and Swedish). ‘‘I’’ stands for initial clausal embedding, ‘‘C’’ for clausal center-embedding, and the raised exponent expresses maximal degree of embedding of a sentence, e.g. I2 is double initial embedding. Expressions like C-2 indicate type and embedding depth of individual clauses. ‘‘-w’’ is short for ‘‘written language’’ and ‘‘-s’’ for ‘‘spoken language’’. (10) I2 max: the maximal degree of initial clausal embedding is 2; 100 instances found, as in (15). (11) Qualitative I2-constraints: Double initial clausal embedding strongly prefers a) written language, b) an if-clause as higher embedding, c) a sentential subject, i.e. a what-clause as lower embedding, and d) finiteness. See (15). (12) C3 max-w: in written language, the maximal (very rare) degree of multiple clausal center-embedding is 3; thirteen instances retrieved, (16). (13) C2 max-s: in spoken language, the maximal (very rare) degree of multiple clausal center-embedding is 2; only two instances retrieved, (17). 3. Note that I follow Quirk et al. (1989: 1037) in interpreting clauses like I-2 in sentence (15), embedded immediately after a subordinating or coordinating conjunction, as initially-embedded rather than as center-embedded. The main argument is that subordinating and coordinating conjunctions are not syntactically as tightly integrated and real constituents in their clauses as ordinary full constituents (and relative pronouns) are.

54

Fred Karlsson

(14) Only-postmodifying-self-embedding: only postmodifying (i.e. typically relative) clauses allow self-embedding (16, 17). (15) [[I-1 If [ I-2 what he saw through security] did not impress him] Tammuz . . . ] (BNC) (16) a. Der Ritter von Malzahn, dem der Junker sich als einen Fremden, der bei seiner Durchreise den seltsamen Mann, den er mit sich fu¨hre, im Augenschein zu nehmen wu¨nschte, vorstellte, no¨tigte ihn . . . b. Der Ritter von Malzahn, Main the rider from Malzahn dem der Junker sich als einen Fremden whom the Junker himself as a stranger der bei seiner Durchreise den seltsamen Mann, who upon his journey the unusual man den er mit sich fu¨hre, whom he with himself would bring in Augenschein zu nehmen wu¨nschte, by appearance to judge wanted vorstellte, introduced no¨tigte ihn . . . forced him . . .

C-1 C-2 C-3 C-2 C-1 Main

‘The rider from Malzahn, to whom the Junker introduced himself as a stranger, who upon his journey wanted to judge by appearance the unusual man whom he would bring with himself, forced him . . .’ (Heinrich von Kleist, Michael Kohlhaas) (17) [A lot of the housing [C-1 that the people [C-2 that worked in New Haven] lived in] was back that way.] (Reich and Dell 1976) No genuine triple initial embeddings nor quadruple center-embeddings are on record, suggesting that I2max and C3max-w are firm restrictions on respectively left-branching clausal tail-recursion and nested clausal recursion. Only some 130 double center-embeddings were retrieved among the tens of millions of words checked. The overall rarity and the numerous constraints on multiple IE and CE prompt the conclusion that multiple clausal initial and center-embeddings are not fully recursive at least in the European languages here considered, if ‘‘fully recursive’’ is taken to mean that ordinary language use would provide instances of multiple initial

Syntactic recursion and iteration

55

clausal embedding or nesting degrees beyond the confines of I2 max and C3 max-w. Of special importance is the fact that multiple nested (¼fully productive) clausal recursion (i.e. multiple center-embedding) is practically absent from ordinary spoken language, as stipulated by constraint (13) C2max-s. Multiple nesting cannot therefore reasonably be considered a central design feature of language, as claimed by Hauser, Chomsky, and Fitch (2002).4 On the contrary, empirical facts prove that Reich (1969) and Reich and Schu¨tze (1991) were on the right track in claiming that the limit on clausal nesting is 1. This is practically true for spoken language, whereas written language allows two cycles and extremely marginally three. C3max-w and Only-postmodifying-self-embedding taken together license triple relative self-embedding, as in (16). Final clausal embedding is less constrained but not totally unregimented. In an extensive empirical study Karlsson (ms.) found evidence in several European languages for a tendency called F3-5max: (18) F3-5 max: Syntactically simple genres (such as most of spoken language and some of written language) avoid final clausal embedding in excess of degree 3, complex genres in excess of degree 5. As sentence (19) from the Brown corpus with eight finally-embedded clauses shows, F3-5 max may be overstepped. Thus, it is a stylistic preference. (19) [ . . . it seems . . . [F-1 that . . . it is unlikely [F-2 to bear much relation . . . to the . . . need of [F-3 ensuring . . . [F-4 that there may be adequate opportunity given to the sta¤ [F-5 to do [F-6 what they can [F-7 to get the man [F-8 to stand again . . .]]]]]]]]] 6. Constraints on NP and PP recursion Surprisingly little empirical work has been done even on much-researched languages like English and German on the recursive potential of basic phrases, especially NPs and PPs, to which I here confine myself. The definition of a noun phrase in this context is that its head is a noun or a pronoun, to the exclusion of clauses, infinitives and similar structures. Furthermore, I do not here treat the case of indirect multiple NP and PP 4. This is further support for Pinker and Jackendo¤ ’s (2005) criticism of Hauser, Chomsky, and Fitch’s (2002) claim.

56

Fred Karlsson

recursion arising through embedding of clauses as postmodifiers of head nouns in NPs. Such clauses may of course contain further NPs and/or PPs. All examples below are genuine and emanate from the extensive corpus studies mentioned in Section 5. As for multiple center-embedding of NPs and PPs (which could also be called self-embedding, as they invoke the same types of constituents), the most complex instances found of both are of degree 3. The German 19thcentury specimen of legal language (20) contains one instance of both NP3 and PP3 (Engel 1922: 327). Successive levels of indentation show increasing depth of embedding (in this and later examples). Please note that indentation is not supposed to reflect constituent structure in detail. Material at the same level of indentation is at the same embedding depth (and normally belongs to the same constituent). The rightmost column indicates embedding depth borders (e.g. the first PP-2 marks the start of the PP at depth 2, the second PP-2 marks its end). (20) a. Das Gericht wolle erkennen, der beklagte sei schuldig, mir fu¨r die von mir fu¨r ihn an die in dem von ihm zur Bearbeitung u¨bernommenen Steinbruche bescha¨ftigt gewesenen Arbeiter vorgeschossenen Arbeitslo¨hne Ersatz zu leisten. b. Das Gericht wolle erkennen, der beklagte sei schuldig, mir . . . The court might rule the defendant is guilty to me . . . fu¨r die PP-0; NP-0 for the von mir PP-1a; NP-1a by me fu¨r ihn PP-1b; NP-1b for him an die PP-1c; NP-1c for the in dem PP-2; NP-2 in the von ihm PP-3a; NP-3a by him zur Bearbeitung u¨bernommenen PP-3b; NP-3b to run had taken over Steinbruche PP-2; NP-2 quarry bescha¨ftigt gewesenen Arbeiter PP-1c; NP-1c employed have been workers vorgeschossenen Arbeitslo¨hne Ersatz zu leisten. PP-0; NP-0 prepaid salaries remuneration to pay

Syntactic recursion and iteration

57

‘May the court rule that the defendant remunerate me for the salaries prepaid by me on his behalf to the workers that were employed in the quarry that he had taken over to run.’ Sentence (20) involves clauses that have been nominalized to premodifying adjectival participles and which carry along their original complements as dependents of the participle in the form of PPs with their requisite NPs. In (20) the PP-0 and NP-0 (fu¨r) die . . . Arbeitslo¨hne ‘(for) the . . . salaries’ at basic level (depth 0) are interrupted by a nominalization (von mir . . . vorgeschossenen ‘by me . . . prepaid’), where the PP-1c and NP1c (at the same depth as PP-1a, NP1-a and PP-1b, NP-1b) (an) die . . . Arbeiter ‘(for) the . . . workers’ (nested PP and NP recursion, degree 1) are interrupted by another nominalization (in dem . . . gewesenen ‘in the . . . been’). This nominalization contains the further nestings PP-2 and NP-2 (in) dem . . . Steinbruche ‘(in) the . . . quarry’ (nested PP and NP recursion, degree 2), interrupted by a third nominalization (von ihm . . . u¨bernommenen ‘by him . . . overtaken’), which contains no less than two more nested PPs and NPs, at depth 3, PP-3a and NP-3a von ihm ‘by him’ and PP-3b and NP-3b zur Bearbeitung ‘to run’. Sentence (21), from Kaseler Zeitung 1.5.1889, also manifests triple nested recursion of prepositional phrases and noun phrases. As for NPs, the nesting sequence is [die . . . [der . . . [das . . . [dem Kloster] Hospiz] Insel] Badereisenden]. (21) a. Fu¨r die nach der durch das von dem Kloster Loccum erbaute Hospitz in weiteren Kreisen bekannt gewordenen Insel Langeoog kommenden Badereisenden hat sich . . . b. Fu¨r die PP-0; NP-0 For the nach der PP-1; NP-1 after the durch das PP-2; NP-2 through the von dem Kloster Loccum PP-3; NP-3 by the monastery Loccum erbaute Hospitz PP-2; NP-2 built hospital in weiteren Kreisen bekannt gewordenen Insel Langeoog PP-1; NP-1 in larger circles known become island Langeoog kommenden Badereisenden hat sich . . . PP-0; NP-0 coming seaside guests has REFL

58

Fred Karlsson

‘For the seaside guests coming to the island Langeoog which has become known to larger circles through the hospital built by the Loccum monastery there has . . .’ Of course triple center-embedding of NPs and PPs as in (20) and (21) is extremely convoluted and rare and confined to restricted, formal, stylistically marked, older (obsolete?) genres of writing. It must be emphasized that no genuine examples of triple nesting of PPs and NPs are at hand from the twentieth (not to speak of the twenty-first) century. Even double center-embedding of NPs and PPs is utterly rare so one more genuine example deserves presentation. Here (22) is an Ancient Greek NP from Plato’s The Sophist.5 (22) a. ta` ga`r teˆs toˆn polloˆn psycheˆs o´mmata b. ta` ga`r NP-0 DET because teˆs NP-1 DET toˆn polloˆn NP-2 DET of many psycheˆs NP-1 soul’s o´mmata NP-0 eyes ‘because the eyes of the soul of many’ Insofar as this small corpus (20, 21, 22) can be trusted, the favourite locus of multiple center-embedding of NPs and PPs seems to be after the determiner in the NP. Two quantitative constraints (where again ‘‘-w’’ is short for ‘‘written language’’) can be postulated at least for several European languages: (23) NP3 max-w: In written language, the maximal (very rare) degree of multiple center-embedding of NPs is 3. (24) PP3 max-w: In written language, the maximal (very rare) degree of multiple center-embedding of PPs is 3. Multiple nesting of PPs presupposes the same amount of nested NPs. PPs and NPs thus form a peculiar tightly bonded recursive pair of layered nestings. No genuine examples are at hand of nested NP3s (with nouns or pronouns as headwords) that would not involve at least one PP. 5. This example was helpfully provided by Martti Leiwo.

Syntactic recursion and iteration

59

What about the present-day limits on multiple NP- and PP-nesting, especially in spoken language? The weirdness of the above genuine obsolete German examples from the nineteenth century, and the total absence of similar genuine contemporary examples, certainly indicates that the current limit is below 3. On the other hand it is clear that at least NP-nesting at depth 1 is a living option for example in Finnish (25). (25) a. ta¨ma¨ edellisen hallituksen ehdotus 6 b. ta¨ma¨ this.DET.NOM.SG edellise+n previous.A+GEN.SG hallitukse+n government.N+GEN.SG ehdotus proposal.N.NOM.SG

NP-0 NP-1 NP-1 NP-0

‘this proposal by the previous government’ Native intuitions tell that NP-1 in (25) could be interrupted by another short NP inflected in the genitive case and still remain grammatical and acceptable: (26) a. ta¨ma¨ edellisen pa¨a¨ministeri Lipposen hallituksen ehdotus b. ta¨ma¨ NP-0 this.DET.NOM.SG edellise+n NP-1 previous.A+GEN.SG pa¨a¨ministeri NP-2 Prime Minister.N.NOM.SG Lippose+n NP-2 Lipponen.N+GEN.SG hallitukse+n NP-1 government.N+GEN.SG ehdotus NP-0 proposal.N.NOM.SG ‘this proposal by the previous government (led by) Prime Minister Lipponen’ 6. The NP (25) can be verified by googling on the internet.

60

Fred Karlsson

As a tentative generalization I therefore submit (27), where ‘‘-s’’ is short for ‘‘spoken language’’: (27) NP2 max-s: in spoken language, the maximal (very rare) degree of multiple center-embedding of NPs is 2. Conclusive resolution of this issue, including the limits of PP-nesting in speech, must await additional detailed corpus study of a su‰cient number of languages. As for left-recursion on constituent level, prepositional phrases by their very nature do not have it. NPs are left-recursive, in the form of repeated genitives, where the previous genitive modifies either the subsequent one or the combination of this and the head noun – this is submodification and stacking, respectively, in the terminology of Huddleston and Pullum (2003: 446–447). This feature distinguishes recursive genitives from iterated phrases, which are structurally parallel. The Brown, LOB and ICEGB corpora (3 million words) show that submodifying genitives are rare, brief, and semantically restricted (28a, 28b, 28c). Only some 20 instances were found of submodifying genitive chains of depth 2. Stacked genitives are even rarer. (28) a. b. c.

Hilary’s lawyer’s secretary Burnside’s horse’s belly Pam’s mum’s baggage

Left-recursive open-endedness on constituent level emerges in formal written genres of Finnish where there is a productive construction based on nominalization of verbs and multiple NPs inflected in the genitive or oblique cases. As in (20) and (21), the nominalized verbs act as premodifying adjectival participles in NPs, with their original complements as properly case-inflected dependents of the participle. The postpositional phrase in (29) comes from a regulation adopted by the Finnish government. There are five left-recursive NPs inflected in the genitive case, several of which have adjectivized verbs with their own left-branching case-inflected NPs as dependents or premodifiers, creating an overall depth at word level of 11. (29) a. valtion virkamiehelta¨ virallisesti vaadittavasta kielitaidosta annetun lain ta¨yta¨nto¨o¨npanosta annetun asetuksen 7 pyka¨la¨n 1 momentin mukaan

Syntactic recursion and iteration b.

61

valtio+n state+GEN.SG virkamiehe+lta¨ o‰cial+ABL.SG vaadi+tta+va+sta require+PSS+PCP+ELA.SG kielitaido+sta language proficiency+ELA.SG anne+tu+n give+PSS.PCP+GEN.SG lai+n law+GEN.SG ta¨yta¨nto¨o¨npano+sta implementation+ELA.SG anne+tu+n give+PSS.PCP+GEN.SG asetukse+n regulation+GEN.SG 7 seventh+GEN.SG pyka¨la¨+n paragraph+GEN.SG 1 first+GEN.SG momenti+n subsection+GEN.SG mukaan according_to ‘according to subsection 1 of paragraph 7 in the regulation which was issued concerning the implementation of the law which was issued concerning the language proficiency which is required of a state o‰cial’

Right-recursion of phrase-level constituents is most productive in the interplay between multiple PPs and NPs contained in one another in postmodifying structures such as (30). No clear upper limits are in evidence but recursive PPs in excess of degree 5 are rare. However, the complex (genuine) Swedish NP (31), again from a regulation issued by the Finnish government, contains no less than ten right-recursive PPs, labelled a, b, c . . . in (31b), and headed by the prepositions om ‘concerning’, av ‘of ’, and i ‘in’.

62

Fred Karlsson

(30) [in the soil [on the slopes [of the rock [by the river [in . . . ]]]]] (31) a. statsra˚dets fo¨rordning om a¨ndring av ikrafttra¨dandebesta¨mmelsen i statsra˚dets fo¨rordning om a¨ndring av ikrafttra¨dandebesta¨mmelsen i statsra˚dets fo¨rordning om a¨ndring av statsra˚dets fo¨rordning om utveckling av landsbygden7 b. statsra˚d+et+s government+DEF+GEN.SG fo¨rordning regulation om a¨ndring (a) concerning change av ikrafttra¨dandebesta¨mmelse+n (b) of implementation decision+DEF.NOM.SG i (c) in statsra˚d+et+s government+DEF+GEN.SG fo¨rordning regulation om a¨ndring (d) concerning change av ikrafttra¨dandebesta¨mmelse+n (e) of implementation decision+DEF.NOM.SG i (f ) in statsra˚d+et+s government+DEF+GEN.SG fo¨rordning regulation om a¨ndring (g) concerning change 7. The source of (31a) is Asetus/Fo¨rordning 778/2004 and the Finnish original is: Valtioneuvoston asetus voimaantulosa¨a¨nno¨ksen muuttamisesta valtioneuvoston asetuksessa maaseudun kehitta¨misesta¨ annetun valtioneuvoston asetuksen muuttamisesta annetun valltioneuvoston asetuksen voimaantulosa¨a¨nno¨ksen muuttamisesta.

Syntactic recursion and iteration

63

av (h) of statsra˚d+et+s government+DEF+GEN.SG fo¨rordning regulation om utveckling (i) concerning development av landsbygd+en ( j) of rural area+def.nom.sg ‘the government’s regulation concerning change of the implementation decision in the government’s regulation concerning change of the implementation decision in the government’s regulation concerning change of the government’s regulation concerning the development of rural areas’8

7. Conclusion No evidence for nested syntactic recursion of degrees greater than 3 is at hand, neither on clause level nor on phrase level. Two interesting questions cannot be addressed here: Why is the maximal degree 3 and not 2, 7 or some other number? Why is the maximal degree the same on phrase and clause level? Motivated answers to these questions would presuppose careful consideration of referent processing and short-term memory capacity restrictions which transcend the scope of this contribution. The extant examples of nesting of degree 3 and even those of degree 2 are so few, convoluted, and almost exclusively confined to written language as to practically falsify the hypothesis of the importance of unlimited syntactic nesting (multiple center-embedding), thus also confirming the experimental results of Christiansen and Chater (1999) achieved by connectionist modeling concerning the marginality of multiple center-embedding. No more than two spoken instances of double clausal center-embedding have been documented (Karlsson 2007b). Double NP and PP 8. One referee aptly remarks that the English translation of this NP ‘‘is completely Martian, and nothing like anything that would ever be produced by even the most constipated English-speaking bureaucrat. It sounds like an insane utterance!’’

64

Fred Karlsson

center-embedding have, to the best of my knowledge, not been recorded in speech at all. Thus, in speech, nested recursion in the domains of clauses and constituents is for all practical purposes confined to just one cycle, degree 1, which is what Heine and Kuteva (2007: 268) call simple (as opposed to productive) recursion. Karlsson (2009) demonstrates in philological detail that multiple nesting arose along with the advent of written language. This is also compliant with known psychological facts: multiple nesting taps short-term memory resources and such resources are more readily available in writing and reading than in real-time speaking and listening. Multiple tail-recursion is real both in its rightwards and leftwards manifestations especially in written language, but not fully unconstrained, as shown above e.g. for multiple left-branching of clauses (I2 max) and for genitives (triple genitive chains are extremely rare). One more interesting question will be left open here as for its proper explanation: Why is clausal left-branching strictly constrained to two cycles, but left-branching of certain premodifiers (such as nominalized participle constructions) is not? In the invitation to the April 2007 recursion conference organized by Dan Everett at the University of Northern Illinois, a complex sentence by the late Cli¤ord Geertz was singled out in the call for papers, with the implication that this sentence (32a) is particularly relevant for the conference topic. (32b) is the precise analysis of this sentence in terms of clause structure (C ¼ finite center-embedded, F ¼ finite finally-embedded, f ¼ non-finite finally-embedded). (32) a. The notion that the essence of what it means to be human is most clearly revealed in those features of human culture that are universal rather than in those that are distinctive to this people or that is a prejudice that we are not obliged to share . . . b. The notion that the essence of what it means to be human is most clearly revealed in those features of human culture that are universal rather than in those that are distinctive to this people or that is a prejudice that we are not obliged to share . . .

Main C-1 C-2 f-3 F-2 F-2 F-1 f-2

The analysis shows that (32b) complies with all constraints on clausal embedding outlined in Section 5, in particular with the constraints

Syntactic recursion and iteration

65

C3 max-w, F3-5 max and Only-postmodifying-self-embedding. There is a double center-embedding in (32b), where the lower center-embedding C-2 contains a further non-finite final embedding f-3 (the infinitive construction to be human). The upper center-embedding C-1 contains two more final embeddings F-2 þ F-2. The predicate complement a prejudice contains two more final embeddings, F-1 þ f-2. No NPs nor PPs in (32b) violate NP3 max-w, PP3 max-w, nor NP2 max-s. Thus, Geertz’s sentence is structurally interesting, but fully compliant with all the constraints on clausal and phrasal recursion outlined in this paper.

References Aho, Alfred V., Ravi Sethi, and Je¤rey D. Ullman 1986 Compilers. Principles, Techniques, and Tools. Reading, Mass.: Addison-Wesley Publishing Company. Baladunarman, I., S. Ishwar, and R. Sanjeeth Kumar 2003 Context-free grammar for natural language constructs. An implementation for Venpa class of Tamil poetry. In: The Sixth International Tamil Internet Conference and Exhibition With the Theme ‘‘IT for Education in Tamil’’. Available at http://www. infitt.org/ti2003/papers/19_raman.pdf. Bar-Hillel, Yehoshua 1953 On recursive definitions in empirical science. 11th International Congress of Philosophy 5: 160–165. Amsterdam: North-Holland. Bhate, Saroja, and Subhash Kak 1993 Pa#n¸ini’s Grammar and Computer Science. Annals of the Bhandarkar Oriental Research Institute 72: 79–94. Also available at http://www.ece.lsu.edu/kak/bhate.pdf. Biber, Douglas, Stig Johansson, Geo¤rey Leech, Susan Conrad, and Edward Finegan 1999 Longman Grammar of Spoken and Written English. London: Longman. Brookshear, J. Glenn 2007 Computer Science. An Overview. Ninth edition. Boston etc.: Pearson Addison Wesley. Broch, Hermann 1961 Der Tod des Vergil. Darmstadt: Moderner Buch-Club. Chomsky, Noam 1956 On the limits of finite-state description. MIT Research Laboratory of Electronics, Quarterly Progress Report 41: 64–65. Chomsky, Noam 1957 Syntactic Structures. The Hague: Mouton.

66

Fred Karlsson

Chomsky, Noam 1975 [1955–1956] The Logical Structure of Linguistic Theory. New York and London: Plenum Press. Christiansen, M. H., and N. Chater 1999 Toward a connectionist model of recursion in human linguistic performance. Cognitive Science 23: 157–205. Engel, Eduard 1922 Deutsche Stilkunst. 30th edition. Vienna & Leipzig: Ho¨lderPichler-Tempsky A-G. Halliday, M.A.K. 1963 Class in relation to the axes of chain and choice in language. Journal of Linguistics 2: 5–15. Harder, Peter 2006 Recursion in a functional-semantic grammar. In: Timo Haukioja (ed.), Papers on Language Theory. Publications in General Linguistics 10: 67–105. Turku: University of Turku. 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. Heine, Bernd, and Tania Kuteva 2007 The Genesis of Grammar. A Reconstruction. Oxford: Oxford University Press. Householder Jr., Fred W. 1952 Review of Harris, Zellig S., 1951, Methods in Structural Linguistics. International Journal of American Linguistics 28: 260–268. Huddleston, Rodney, and Geo¤rey K. Pullum 2003 The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Humboldt, Wilhelm von ¨ ber die Verschiedenheit des menschlichen Sprachbaues und ihre 1836 U Einfluss auf die geistige Entwickelung des Menschengeschlechtes. Berlin: Druckerei der Ko¨niglichen Akademie der Wissenschaften. Johansson, Sverker 2005 Origins of Language. Constraints on Hypotheses. Amsterdam: John Benjamins Publishing Company. Joyce, James 1934 [1922] Ulysses. New York: Random House Inc. Karlsson, Fred 2007a Constraints on multiple initial embedding of clauses. International Journal of Corpus Linguistics 12 (1): 107–118. Karlsson, Fred 2007b Constraints on multiple center-embedding of clauses. Journal of Linguistics 43 (2): 365–392. Karlsson, Fred 2009 Origin and maintenance of clausal embedding complexity. In:

Syntactic recursion and iteration

67

Geo¤rey Sampson, David Gil, and Peter Trudgill (eds.), Language Complexity as an Evolving Variable, 192–202. Oxford: Oxford University Press. Karlsson, Fred (ms.)

Multiple final embedding of clauses. University of Helsinki: Department of General Linguistics.

Lindstro¨m, Jan 1999 Vackert, vackert! Syntaktisk reduplikation i svenskan. Studier i nordisk filologi 77. Helsingfors: Svenska Litteratursa¨llskapet. Linell, Per 2005 The Written Language Bias in Linguistics. Its Nature, Origins and Transformations. London: Routledge. Parker, Anne R. 2006 Evolving the narrow language faculty: was recursion the pivotal step? In: Angelo Cangelosi, Andrew D. M. Smith, and Kenny Smith (eds.), The Evolution of Language: Proceedings of the 6th International Conference (Evolang 6), 239–246. London: World Scientific. Pinker, Steven and Ray Jackendo¤ 2005 The faculty of language: What’s special about it?. Cognition 95: 201–236 Quirk, Randolph, Sidney Greenbaum, Geo¤rey Leech, and Jan Svartvik 1972 A Grammar of Contemporary English. London: Longman. Reich, Peter A. 1969 The finiteness of natural language. Language 45: 831–843. Reprinted in Fred Householder (ed.), Syntactic Theory 1. Structuralist, 238–272. Harmondsworth: Penguin. Reich, Peter A., and Gary S. Dell 1977 Finiteness and embedding. In: R. J. Di Pietro and E. L. Blansett Jr. (eds.), The Third LACUS Forum 1976, 438–447. Columbia, S.C.: Hornbeam Press. Reich, Peter A., and Carson T. Schu¨tze 1991 Syntactic embedding: What can people really do? In: Barbara Brunson and Tom Wilson (eds.), Toronto Working Papers in Linguistics 11 (1)–11 (2): 91–97. Sebba, Mark 1987 The Syntax of Serial Verbs. An Investigation into Serialisation in Sranan and Other Languages. Amsterdam/Philadelphia: John Benjamins Publishing Company. Tomalin, Marcus 2006 Linguistics and the Formal Sciences. The Origins of Generative Grammar. Cambridge: Cambridge University Press. Tomalin, Marcus 2007 Reconsidering recursion in syntactic theory. Lingua 117: 1784– 1800.

4.

Recursion in conversation: What speakers of Finnish and Japanese know how to do

Ritva Laury and Tsuyoshi Ono

1. Introduction In their article in Science (2002), Hauser, Chomsky & Fitch (HCF, hereafter) advance a provocative hypothesis that recursion is the only uniquely human component of the faculty of language. That is, HCF in essence proposes that recursion is an innate capacity possessed by all humans which accounts for the central aspects of what human language is like, and in fact what makes it specifically human. There has since been much discussion over this hypothesis both in print and on the internet. In our view as discourse linguists, this hypothesis is seriously flawed at the very least. That is, although HCF assumes that recursion is the defining component of the human language faculty, our experience of closely examining actual human speech leads us to suggest that recursion explains very little of actual human language behavior. In fact, we strongly feel that to evaluate the recursion hypothesis it is necessary to verify that recursion is real for human speech. We will do just that in this article.1 Specifically, we will examine the recursion hypothesis based on clause combining in Finnish and Japanese conversational data. Our rationale behind adopting this approach was that since everyday conversation is the most fundamental form of human language, if recursion is indeed central to human language, as HCF claims, then we should be able to find some manifestation of it in our data. Alternatively, if we find little evidence of recursion in our data, the status of the recursion hypothesis has to be questioned since it would mean that it is a descriptively inadequate hypothesis. That is, in our view, an extremely rare incidence of examples in everyday language use to which recursion might be relevant seriously weakens the HCF

1. One very common research thread in the field is mathematical modeling of recursion. Again, our position is that the first step should be to establish that recursion is an actual human trait.

70

Ritva Laury and Tsuyoshi Ono

hypothesis that recursion is the defining property of human language. It should also be pointed out that much of the investigation into the recursion hypothesis, a universal hypothesis, has been based solely on English and a few other Indo-European languages. For this reason, we felt it important to apply our methodology to two non-Indo-European languages to see if they exhibit similar patterns (see also Everett, 2005, this volume, on Piraha˜, another non-Indo-European language). Below, as a way of presenting our data and methodology, we will first discuss the analytical di‰culties in studying clausal recursion in conversational data. We will then show that even with a very liberal definition of recursion, there are clear limits to it in everyday conversation. We further conclude that multiple embedding is a result of looking at human language as an atemporal object. That is, we find that in ordinary language use, speakers combine clauses one at a time, as already suggested by Pawley & Syder 1983 and Chafe 1988, and that the internal structure of the foregoing material does not seem to be an issue to them. In fact, we suggest that the evocation of the concept of recursion is not necessary to account for the patterns of clause combining in our data, and that our data give no reason to assume that recursion, specifically infinite recursion, is a defining property of human language. Our data consist of approximately five hours of transcribed audiotapes of conversation in Finnish (from corpora collected at the University of Helsinki and the University of Turku2) and five hours of conversation in Japanese (from corpora collected at the University of California, Santa Barbara and University of Arizona3). For our quantitative analysis, in order to provide a good representation of our target populations, we took a random sample consisting of approximately 1000 clausal units from each language, from which we coded all clause combinations, labeling each clause as either a matrix clause or a type of subordinate clause (adverbial, complement, or relative clause), and we coded the depth and 2. The Finnish corpus consists of audiotaped telephone and face-to-face conversations transcribed for and housed in the Conversation Analysis archive at the Department of Finnish, University of Helsinki. Some of the conversations used were originally audiotaped for the Turku spoken language project (see Suojanen 1985). The transcripts have been slightly simplified: for transcription symbols used, see Appendix. 3. The Japanese corpus consists of audiotaped face-to-face and telephone conversations collected and transcribed by researchers at the Universities of California (Santa Barbara) and Arizona and currently housed in Spoken Discourse Research Studio at University of Alberta.

Recursion in conversation

71

direction of embedding for each clause. Our qualitative analysis is more inclusive; we also examined examples found in our data which are not included in the counts.

2. Problems with analyzing embedding in conversational language In analyzing our data, we had to deal with two major problems: the nondiscreteness of linguistic categories and the problematic status of the concept of ‘sentence’ in spoken language. Before we give the main results of the current study, we would like to review these problems in order to allow 1) the reader to properly interpret the type of data we will be dealing with and, we hope, 2) those interested to further pursue this line of research. 2.1. Non-discreteness of categories In actual language use, linguistic categories are non-discrete (Hopper & Thompson 1980, 1984; Langacker 1987, 1991; Hopper 1998) due to the fact that language is constantly changing as it adapts to functional needs. One example of this is the cross-linguistic tendency for certain clause types, clausal markers, and predicates to grammaticize into particles and/ or discourse markers (e.g., I think in English (Scheibman 2003, Ka¨rkka¨inen 2003), iu ‘say’ in Japanese (Suzuki 1999), kato ‘look’ in Finnish (Hakulinen & Seppa¨nen 1992), kedo ‘although’ in Japanese (Nakayama and Ichihashi-Nakayama 1997). This causes problems when one has to make coding decisions. Consider the following examples. The first example comes from a conversation where coworkers are discussing a colleague who is not present. Example (1) Kahvikutsut (Turku spoken language corpus, C-134) 1 EK: tia¨da¨k sa¨a¨ et ha¨n on pyyta¨ny mult know-2SG-Q 2SG COMP 3SG be ask-P.PPLE 1SG-ABL kerran anteeks once apology ‘(Do) you know that s/he has apologized to me once 2

kuule ja siit on muutama pa¨iva¨ aikaa. hear-2SG.IMP and 3SG-ELA be a.few day time-PRT listen and it was only a few days ago.’

72

Ritva Laury and Tsuyoshi Ono

3 JS: a¨la¨:. NEG-2SG.IMP ‘[You] don’t [say] (You’re kidding!)’ ((SEGMENT OMITTED)) 4 EK: tieda¨k sa¨ ma¨[a¨ olin ih]an ha¨mma¨styny know-2SG-Q 2SG 1SG be-PST-1SG quite surprised ‘You know I was really surprised’ 5 X:

[a¨la¨::] NEG-2SG.IMP ‘[You] don’t [say]. (You’re kidding!)’

6 EK: kuule etta¨(.) ha¨nes on tapahtunu: o¨o¨hh oikee hear-2SG-IMP COMP 3SG-INE be happen-P.PPLE PTC really siis (0.7) [su[u:ri PTC great ‘listen that in her there’s been really a great 7 JS:

[suuri ‘great’

8 IW: [just ‘just’ (.) 9 EK: suuri muu- muutos juu:. great change PTC great change yeah.’ In this example, EK uses the expressions tia¨da¨k sa¨(a¨) ‘do you know’ and kuule ‘listen’ twice each. The verbs tieta¨a¨ ‘know’ and kuulla ‘hear’ are full complement-taking verbs in Finnish, but they also function as particles or ¨ stman adverb-like fixed expressions similar to the English you know (O 4 1981) and I think (Thompson 2002 , Ka¨rkka¨inen 2003, Scheibman 2003; see also Sakel & Stapert, this volume, on similar expressions in Piraha˜). It is not always clear which of the two use types one is dealing with. In line 1, tia¨da¨k sa¨a¨ ‘do you know’ is followed by the complementizer et(ta¨)

4. Important evidence for the particle-like functioning of these expressions is that they can and do freely ‘float’ in a sentence; see, for example, Thompson 2002 for examples and discussion.

Recursion in conversation

73

‘that’ and another clause, and thus line 1 is analyzable as a main clause followed by a complement clause (although see Laury 2006, Seppa¨nen & Laury 2007, Laury & Seppa¨nen 2008). However, kuule ‘listen’ in line 2 is not followed by a complement clause, but rather a clause coordinated with the preceding clause with the conjunction ja ‘and’. For that reason, and also because Finnish complement clauses follow their main clauses (Korhonen 1993), kuule is here best analyzed as a particle. The second use of tieta¨a¨ ‘know’, tieda¨k sa¨ ‘(do) you know’ in line 4, is much less main-clauselike than the one in line 1. The pronoun is less stressed and shorter, and the complementizer is missing, although one is normally used in all contexts of finite clausal complementation in Finnish. In line 6, kuule ‘listen’ is followed by an etta¨-initial complement clause. However, kuule does not seem to function here as a complement-taking predicate either. A better analysis would be to consider it here as a particle. The etta¨-initial clause following kuule in lines 6 and 9 is more plausibly, for semantic as well as syntactic reasons, meant as a complement of the clause following tieda¨k sa¨ ‘do you know’, ma¨a¨ olin ihan ha¨mma¨styny ‘I was really surprised’ in line 4. If tieda¨k sa¨ in line 4 is a main clause, we should code the utterance in lines 4, 6, and 9 as an instance of two levels of final embedding (a main clause followed by two complement clauses, the second of which would be embedded within the first one), but if we consider tieda¨k sa¨ a particle, then we have one level of final embedding. The coding decision is essentially arbitrary. For the purpose of this study, we have adopted a maximal definition of embedding and coded this and similar combinations as two levels of final embedding. Consider a similar example from Japanese. In this excerpt, the speaker is recounting an event of talking with a homeless person. Example (2) Homeless 1 shumi de yatten no ka ttsutara hobby as doing NOM Q say:COND ‘When (I) said ‘are (you) doing it (being homeless) as (a) hobby’, 2 shumi de yatteru tteta yo hobby as doing say:PAST FP (he) said ‘(I) am doing that as (a) hobby’ ’ The two utterances end with ttsuta and tteta, reduced forms of a combination of quotative complementizer te and the past tense form of the verb ‘say’ yutta (‘said that . . .’). In these forms, morpheme boundaries are becoming opaque. Further, ttsuta and tteta are produced along with the

74

Ritva Laury and Tsuyoshi Ono

preceding utterance without a pause. For these two reasons, one might suggest that these forms simply mark the preceding utterance as a quote, rather than functioning as full-blown complementizers followed by main clauses or complement-taking predicates. It is, in fact, well known that the ‘say’ verb grammaticizes into a quotative marker cross-linguistically (as in Japanese; Suzuki 1999). In (2), should ttsuta and tteta be considered complementizers with full verbs or quotative particles? In this project, we coded them as verbs in order to capture as many potential levels of embedding as possible. That is, our coding was quite inclusive because we wanted to be careful about dealing with a hypothesis that has been getting quite a bit of attention in the field. Both of these examples thus demonstrate that categories in actual use are not discrete due to the on-going change of human language, which makes our job as analysts very di‰cult and results in many arbitrary coding decisions. This is also an important factor one has to take into account when interpreting our quantitative results. 2.2. The problematic status of the category ‘sentence’ in spoken language Previous studies have shown that in spontaneously produced spoken language, sentence boundaries, even clausal boundaries, are di‰cult to determine, for example when speakers add material (‘increments’ Ford, Fox & Thompson 2002; Couper-Kuhlen & Ono 2007) to already completed clausal utterances. Such expansions of syntactically complete structures can be fitted to what they precede, or they may reanalyze the preceding syntax (e.g., added ‘quotationalizing’ main clauses in Japanese, Tanaka 2001). In fact, the idea of the sentence may be a ‘linguist’s category’, based on written language (Linell 2005) and constructed examples. Sentences do not really present a clear gestalt in speech, although clauses emerge quite clearly as a unit (Helasvuo 2001, Iwasaki & Ono 2001, Thompson & Couper-Kuhlen 2005, Thompson 2006; see also Crystal 1980, Ikola et al 1989, Miller 1995). Consider the following examples from Finnish and Japanese. In example (3), Missu has just said that she has registered for the university. Example (3) Pa¨a¨nsa¨rky (U. of Helsinki Finnish dept. corpus Sig 113_2a1) 1 Missu:
ja to
ta, arvaa mi(h)ta¨ ma¨ shain and PTC guess.2SG-IMP what-PRT 1SG get-PST-1SG tieta¨a¨.h know-1INF ‘And guess what I found out.’

Recursion in conversation

75

2 Kaaka: >no?,[h< PTC ‘What?’ 3 Missu: [.mmthh (0.2) meia¨n (.) P-ELA 1PL-GEN ‘From Paltamo my 4

tota: (.) m:ummun naapuri?h (.) >jonka kanssa ma¨ PTC grandma-GEN neighbor REL-GEN with 1SG um grandma’s neighbor with whom I

5

oon< lapsena leikkiny ¼Tiina Valkiajoki?h be-1SG child-ESS play-P.PPLE T. V. played with as a child. Tiina Valkiajoki.’

6 Kaaka: [tulee sinne kans [(>opiskeleennii.< PTC ‘Yeah.’ In lines 3–5, Missu produces a complex NP, including a relative clause, which she finishes prosodically in line 5. After completing the NP, she adds an appositive proper name, Tiina Valkiajoki. In line 6, Kaaka produces a predication which is syntactically fitted to either of the NPs produced by Missu in lines 3–5. The status of line 6 with respect to lines 3–5 is not entirely clear. Speakers do frequently produce NPs which have no syntactic a‰liation with any neighboring clausal units (so-called ‘free NPs’; see Tao 1996 for Chinese, Helasvuo 2001 for Finnish). The option of continuing with, for example, a new clause containing a pronoun coreferential with the NPs in lines 3–5 is open to Missu at the end of line 5 (see e.g. Auer 2005). Given this, do lines 3–6 form a ‘sentence’? Does the relative clause in lines 4–5 constitute an instance of ‘center embedding’? We have not coded it as such, but the decision not to code clause combinations involving two speakers is an arbitrary one. Our next example comes from a Japanese conversation about playing baseball at the high school level.

76

Ritva Laury and Tsuyoshi Ono

Example (4) Bukatsu 1 U: regyuraa wa kore mo dame jan. regular TOP this also bad TAG ‘(becoming) a starting player, it’s also bad, isn’t it? (I can’t be a starting player)’ 2 M: . . .[un] ‘mhm’ 3 U: [chuu]gakkoo kara yatteru yatsu ga [iru kara]. junior.high.school from do:STAT guy GA exist so ‘because there are guys who have been playing from Junior High School’ 4 M: [un un] ‘mhm mhm’ The utterance in line 1 is a clear complete single clause sentence with a final pitch pattern followed by a pause. The speaker adds another utterance ending with kara ‘because’ in line 3. Please note that kara is originally a conjunctive particle but it has been suggested to be becoming a final particle (Kokuritsu Kokugo Kenkyuujo 1960; Iguchi 1998). So the question is: Is the added utterance an adverbial reason clause ending with kara, forming a sentence with the utterance on line 1? Or is it an independent sentence ending with the final particle kara? In other words, an ongoing categorical change makes the coding decision for (4) essentially arbitrary. We should note that in coding our data, we were extremely inclusive in coding clause combinations as cases of embedding in order to not exclude any possible cases; we coded the kara-clause on line 3 in (4) as an adverbial reason clause associated with the main clause on line 1.

3. Major findings 3.1. Quantitative findings In the Finnish conversational data, there were 997 clauses5 altogether. Table 1 shows the absolute numbers of clause combinations at di¤erent

5. Only finite clauses were counted. That is, we did not include participial or infinitival constructions as clauses in our counts.

Recursion in conversation

77

depths found in the data, and the percentage of these types of clause combinations calculated as a percentage of the total number of clauses in the data: Table 1. Depth of embedding in the Finnish conversational data 146 depth of one embeddings

14.6%

27 depth of two embeddings

2.7%

6 depth of three embeddings

0.6%

4 depth of four embeddings

0.4%

(of the total number of clauses)

These figures show that embedding at a depth of more than two was extremely rare in the Finnish data. The most common type of embedding was at a depth of one (clause combinations consisting of just one subordinate or embedded clause combined with a main clause). Furthermore, and consistent with Karlsson 2007a and 2007b, and Karlsson & Sinnema¨ki 2007, who studied embedding depth in mostly written data from several European languages, including Finnish, there were no clause combinations with a depth of more than four found in the data. This shows that clausal embedding is not unlimited in either spoken or written data. These findings on embedding depth are also consistent with Ikola et al. (1989), a study based on a large dialect interview corpus consisting of 166,262 sentences. Although Ikola et al. found eighteen instances of embedding at a depth of greater than four, those instances were vanishingly rare, given the size of the corpus. Thus, their findings were consistent with ours. Ikola et al. also note, as we do, that the notion of the sentence is problematic for spoken language, and present several examples illustrating why that is so. They further note that all cases of embedding at a depth of more than four from their data are questionable, as it is not clear what is embedded within what in such complex clause combinations (1989: 23). This is a point we will also illustrate below. In our data, and also consistently with Karlsson (2007a & b) and Karlsson & Sinnema¨ki (2007), there were also no cases of more than two relative clauses or adverbial clauses embedded within one another. Even embeddings at the depth of two were limited in number. There were only six cases of two consecutive adverbial clauses, five cases of two consecutive complement clauses, and one case of two relative clauses in a row. Three of these twelve cases were quite marginal and open to interpretation. In particular, in two cases, it seemed clear that the successive complement clauses and relative clauses were not embedded within each other, but

78

Ritva Laury and Tsuyoshi Ono

rather were modifying the same main clause or head noun, but without an overt coordinator. Only one case involving a sequence of three complement clauses embedded within each other was found. In keeping with the typological nature of Finnish as a predominantly right-branching language, most embedded clauses in our data followed their main clauses. There were nine adverbial clauses in the data which preceded their main clauses, but only one initial sequence of two adverbial clauses was found (and even that one was suspect, as shown below, in Example (5). There was one case of center embedding, at the level of one. In the Japanese conversational data, there were 1029 clauses altogether. In these data, there were: Table 2. Depth of embedding in the Japanese conversational data 145 depth of one embeddings

14.1%

48 depth of two embeddings

4.7%

13 depth of three embeddings 6 depth of four embeddings

1.3% 0.6%

(of the total number of clauses)

Replicating the Finnish findings, embedding in Japanese at a depth greater than two was very rare. In fact, we found no clear examples of clause combinations with a depth of more than four. And there were no cases of more than two of the same type of clauses embedded within one another in this data. Further there were 10 cases of two adverbials, three cases of two complement clauses, and one case of two relative clauses embedded within one another. These figures show that embedding is clearly limited in Japanese. Finally, our Japanese data highlights the typological feature of Japanese as a left-branching language; embedded clauses were most typically found to precede their main clause, as shown in (2). However, examples such as (4), where embedded clauses can be interpreted to follow their main clause, were not uncommon in the data (Ono and Suzuki 1992; Ono 2006). To summarize, one general finding of our study is the scarcity of multiple embedding (at most four levels) in conversational data in both Finnish and Japanese (consistent with Karlsson 2007a, 2007b, Karlsson & Sinnema¨ki 2007 for written Finnish, and Ikola 1989 for spoken Finnish). The figures also show that approximately 80% of the clauses in our Finnish and Japanese data showed no embedding at all. We also found that embedding follows the typological tendency of each language, so that

Recursion in conversation

79

Finnish uses predominantly a ‘main clause first’ strategy, while Japanese, uses a ‘main clause last’ strategy. Embedding of more than two of the same type of clause within each other is also extremely rare (only one case consisting of three successive complement clauses in Finnish). The limitations in the depth of clausal embedding may be a result of a cognitively-based processing di‰culty (Miller 1956, Chafe 1994, Gibson 1998, Cowan 2001, Davis 2001). As suggested by Auer (2005: 27), even on the production side, ‘memory for form is shorter than memory for content’, and therefore, ‘speakers do not usually succeed in keeping ongoing syntactic projects alive for a very long time.’6 It might also be the case that clausal embedding beyond a certain number (e.g., four) is something which speakers of these languages have not experienced in their linguistic life and have not produced since there has not really been a need for it. It is simply not a linguistic practice speakers engage in, which undermines the validity of recursion as a component of the faculty of language. 3.2. Clausal embedding is a local, one-clause-at-a-time phenomenon We suggest, based on our data, that in conversational language, examples which might look like instances of recursion are actually created through local operations, and what seems to be more relevant/central to speakers than syntax is semantic processing. That is, clause linkages are made one at a time in conversation. Consider the following three examples. In the first example, Tiina, who is giving rides to an event to a group of people, is negotiating the departure time with Mallu, suggesting that Mallu come earlier than she has suggested herself 7. Example (5) La¨hto¨aika (U. of Helsinki Finnish dept. corpus Sig 112-1b3) 1 Mallu: o