Complexity in Language: Developmental and Evolutionary Perspectives 1107054370, 9781107054370

Citation preview

Complexity in Language Developmental and Evolutionary Perspectives The question of complexity, as in what makes one language more complex than another, is a long-established topic of debate among linguists. Recently, this issue has been complemented with the view that languages are complex adaptive systems, in which emergence and self-organization play major roles. However, few students of the phenomenon have gone beyond the basic assessment of the number of units and rules in a language (what has been characterized as bit complexity) or shown some familiarity with the science of complexity. This book reveals how much can be learned by overcoming these limitations, especially by adopting developmental and evolutionary perspectives. The contributors include specialists of language acquisition, evolution and ecology, grammaticization, phonology, and modelling, all of whom approach languages as dynamical, emergent, and adaptive complex systems. salikoko s. mufwene is Professor of Linguistics and a member of the Committee on Evolutionary Biology and the Commitee on the Conceptual and Historical Studies of Science at the University of Chicago. He is the author of The Ecology of Language Evolution (Cambridge University Press, 2001), Créoles, écologie sociale, évolution linguistique (2005), and Language Evolution: Contact, Competition and Change (2008). He has edited several books, including Iberian Imperialism and Language Evolution in Latin America (2014). He is the founding editor of Cambridge Approaches to Language Contact. His research includes the emergence of creoles, the phylogenetic emergence of language, and globalization and language vitality. christophe coupé is a researcher in cognitive science at the Centre National de la Recherche Scientifique (CNRS) and the University of Lyon, France. With a background in computer science, cognitive science, and psychology, he has been involved in several multidisciplinary programs focusing on language complexity, language origins, and language change. His contributions have mostly consisted in the design and analyses of databases, and in statistical or computational models of linguistic evolution and diversity. In 2003, he received the Prize of the Young Researcher of the city of Lyon for his Ph.D. dissertation on the origins of language. In addition to other papers on phonological complexity, information rate and functional load, he has coedited the collective volume Approaches to Phonological Complexity (2009). françois pellegrino is a senior researcher in linguistics and cognitive science at the Centre National de la Recherche Scientifique (CNRS) and the University of Lyon, France. He has coordinated several projects on language complexity and has been the coordinator of the “Laboratory of Excellence,” Advanced Studies on Language Complexity (ASLAN) since 2011. For more than ten years, his research has focused on the structure and dynamics of phonological systems in the light of the science of complexity and of Shannon’s information theory. He has co-edited the collective volume Approaches to Phonological Complexity (2009) and, over the last fifteen years, he has authored or co-authored about 80 journal articles, book chapters or conference papers.

Cambridge Approaches to Language Contact Founding Editor Salikoko S. Mufwene, University of Chicago Co-Editor Ana Deumert, University of Cape Town Editorial Board Robert Chaudenson, Université d’Aix-en-Provence Raj Mesthrie, University of Cape Town Lesley Milroy, University of Michigan Shana Poplack, University of Ottawa Michael Silverstein, University of Chicago

Cambridge Approaches to Language Contact is an interdisciplinary series bringing together work on language contact from a diverse range of research areas. The series focuses on key topics in the study of contact between languages or dialects, including the development of pidgins and creoles, language evolution and change, World Englishes, code-switching and code-mixing, bilingualism and second language acquisition, borrowing, interference and convergence phenomena. Published Titles Salikoko S. Mufwene, The Ecology of Language Evolution Michael Clyne, Dynamics of Language Contact Bernd Heine and Tania Kuteva, Language Contact and Grammatical Change Edgar W. Schneider, Postcolonial English Virginia Yip and Stephen Matthews, The Bilingual Child Bernd Heine and Derek Nurse (eds.), A Linguistic Geography of Africa J. Clancy Clements, The Linguistic Legacy of Spanish and Portuguese Umberto Ansaldo, Contact Languages Jan Blommaert, The Sociolinguistics of Globalization Carmen Silva-Corvalán, Bilingual Language Acquisition Lotfi Sayahi, Diglossia and Language Contact Emanuel J. Drechsel, Language Contact in the Early Colonial Pacific Enoch Oladé Aboh, The Emergence of Hybrid Grammars Zhiming Bao, The Making of Vernacular Singapore English Ralph Ludwig, Peter Mühlhäusler, and Steve Pagel (eds.), Linguistic Ecology and Language Contact Braj B. Kachru, World Englishes and Culture Wars Bridget Drinka, Language Contact in Europe Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino (eds.), Complexity in Language: Developmental and Evolutionary Perspectives Further Titles Planned for the Series Rakesh Bhatt, Language Contact and Diaspora Gregory D. S. Anderson Language Extinction Kingsley Bolton, Samuli Kaislaniemi, and Anna Winterbottom (eds.), Language Contact and the East India Company Sarah Roberts, The Birth of a Language Ellen Hurst and Rajend Mesthrie (eds.), Youth Language Varieties in Africa Cecile Vigouroux, Migration, Economy, and Language Practice

Complexity in Language Developmental and Evolutionary Perspectives Salikoko S. Mufwene University of Chicago

Christophe Coupé University of Lyon

François Pellegrino University of Lyon

University Printing House, Cambridge CB2 8BS, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 4843/24, 2nd Floor, Ansari Road, Daryaganj, Delhi - 110002, India 79 Anson Road, #06-04/06, Singapore 079906 Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning, and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781107054370 DOI: 10.1017/9781107294264  C Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino 2017

This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2017 A catalog record for this publication is available from the British Library Library of Congress Cataloging-in-Publication Data Names: Mufwene, Salikoko S., editor. | Pellegrino, Franpcois, 1971– editor. | Coupbe, Christophe, 1977– editor. Title: Complexity in language : developmental and evolutionary perspectives / [edited by] Salikoko S. Mufwene, Franpcois Pellegrino, Christophe Coupbe. Description: Cambridge ; New York : Cambridge University Press, [2016] | Series: Cambridge Approaches to Language Contact Identifiers: LCCN 2016041226 | ISBN 9781107054370 Subjects: LCSH: Complexity (Linguistics) | Discourse analysis. | Linguistic analysis (Linguistics) Classification: LCC P128.C664 C65 2016 | DDC 410–dc23 LC record available at https://lccn.loc.gov/2016041226 ISBN 978-1-107-05437-0 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

Figures Tables Contributors Acknowledgments

page vii ix xi xii

1 Complexity in Language: A Multifaceted Phenomenon salikoko s. mufwene, christophe coupé, and françois pellegrino 2 How to Explain the Origins of Complexity in Language: A Case Study for Agreement Systems luc steels and katrien beuls 3 Complexity in Speech: Teasing Apart Culture and Cognition bart de boer 4 A Complex-Adaptive-Systems Approach to the Evolution of Language and the Brain p. thomas schoenemann 5 Evolutionary Complexity of Social Cognition, Semasiographic Systems, and Language william croft 6 To What Extent Are Phonological Inventories Complex Systems? christophe coupé, egidio marsico, and françois pellegrino 7 A Complexity View of Ontogeny as a Window on Phylogeny barbara l. davis

1

30 48

67

101

135

165

v

vi

Contents

8 Language Choice in a Multilingual Society: A View from Complexity Science lucía loureiro-porto and maxi san miguel

187

9 Complexity and Language Contact: A Socio-Cognitive Framework albert bastardas-boada

218

Index

245

Figures

2.1 Results of an agent-based simulation page 39 2.2 Comparison of a population that uses lateral inhibition and one that does not 40 2.3 Experiment comparing the learning efficiency Sg between a strategy that reuses existing words as meaningful markers and one that does not 42 2.4 (a) Overall reduction of form complexity in a population of agents over 50,000 games (b) The -naeamo marker (expressing feature v-2–1) 43 2.5 Reduction in inventory complexity 45 3.1 The different time scales in language and their interactions 54 3.2 French vowels and Hindi plosives illustrating feature economy 56 3.3 Schematic illustration of vertical transmission (top) and horizontal transmission (bottom) 57 3.4 Three processes through which experimental participants create new signals 60 4.1 Increase in ratio of possible combinations of areas for human vs. ape 76 4.2 Comparison of semantic network density for ape vs. human corpora 86 5.1 The logical relationship among types of common ground, its basis, and coordination devices 127 6.1 Distributions of segmental associations according to the value of the binomial test 150 6.2 Graphical representation of the relations between vowels 152 6.3 Distributions of feature associations according to the value of the binomial test 157 7.1 Consonant place assimilation patterns 180 7.2 Consonant manner assimilation patterns 180 7.3 Labial, coronal, and dorsal assimilation pattern changes over time 181 8.1 One-dimensional regular lattice of degree k = 2 199 vii

viii

Figures

8.2 Square regular lattice of degree k = 4 8.3 Simulation tool designed to visualize the effects of bilingualism, prestige and volatility 8.4 Bilingual speakers on the borders of monolingual domains (regular lattice) 8.5 Islands of monolingual speakers linked by a small number of bilinguals 8.6 Abrams-Strogatz Model vs. Bilinguals Model, when a = 0.1 (high volatility) 8.7 Snapshots of the Abrams-Strogatz (top) and the Bilinguals Model (bottom) in a small-world network (bilinguals = white) 8.8 Abrams-Strogatz Model (left) and Bilinguals Model (right) in a network with community structure (t = time) 9.1 Main principles of the complexity perspectives in contrast with the more traditional scientific ones

199 200 201 202 204 207 208 223

Tables

3.1 Languages from UPSID451 (Maddieson, 1984; Maddieson & Precoda, 1990) with inventory sizes smaller than or equal to 16 phonemes. Although there appears to be an important influence of area or language family, nevertheless, languages with small phoneme inventories are spoken in different, linguistically unrelated regions of the world page 49 4.1 Nouns used in sentences Kanzi responded to correctly 81 4.2 Verbs used in sentences Kanzi responded to correctly 82 4.3 Proper names used in sentences Kanzi responded to correctly 82 4.4 Pronouns used in sentences Kanzi responded to correctly 83 4.5 Adjectives, adverbs and prepositions used in sentences Kanzi responded to correctly 83 4.6 Comparative semantic web density for select words in human vs. chimp 87 5.1 Evolutionary complexity based on the evolution of semasiographic systems 125 6.1 Observed frequencies of voiced labial fricatives and voiced coronal fricatives across languages in UPSID 142 6.2 Observed frequencies of /a/ and /ã/ across languages in UPSID 143 6.3 Categorization of the different pair associations of components of phonological inventories according to our statistical approach 146 6.4 Comparison of four different approaches to multiple testing for vocalic features 148 6.5 Comparison of four different approaches to the statistical issue of multiple testing for consonants 149 6.6 Estimation of pair interactions at the segmental level 149 6.7 The 20 vowel-vowel associations with the smallest p-values 150 6.8 The 20 consonant-consonant associations with the smallest p-values 151 6.9 Vowel-consonant significant associations in the inventories 154 6.10 Distribution of the segments involved in the significant associations between a vowel and a consonant 154 ix

x

Tables

6.11 Estimation of pair interactions at the feature level 6.12 The 20 associations of vocalic features with the smallest p-values 6.13 The 20 associations of consonantal features with the smallest p-values 6.14 Pairs of consonantal features characterized by mutual attraction and represented in the 20 most significant associations 7.1 Comparison of consonant vowel association levels within syllables 7.2 Intersyllabic infant-language matching results for intrasyllabic properties 8.1 Typical features of complex systems

155 156 157 158 176 177 189

Contributors

katrien beuls, AI Laboratory, Vrije Universiteit Brussel albert bastardas boada, Universitat de Barcelona bart de boer, Vrije Universiteit Brussel christophe coupé, Centre National de la Recherche Scientifique and University of Lyon william croft, University of New Mexico barbara davis, University of Texas at Austin lucía loureiro-porto, University of the Balearic Islands maxi san miguel, University of the Balearic Islands and IFISC (CSIC-UIB) salikoko s. mufwene, University of Chicago françois pellegrino, Centre National de la Recherche Scientifique and University of Lyon p. thomas schoenemann, Indiana University luc steels, ICREA/Institu de Biologia Evolutiva (UPF-CSIC), Barcelona

xi

Acknowledgments

This book is largely the fruition of the 10-month fellowship that the lead editor Salikoko S. Mufwene enjoyed at the Collegium de Lyon from October 2010 to July 2011. It enabled him to obtain funding from the same institution to cohost, in collaboration with the other two editors and with Jean-Marie Hombert and Egidio Marsico, all at University of Lyon and CNRS, a workshop with the same title as the present book. We express our hearty thanks to the then Collegium’s director Alain Peyraube for his interest in the subject matter, for his encouragements and advice during the preparations, and for ultimately making it possible for this book to materialize. We are also deeply indebted to the Collegium’s then administrative coordinator Marie-Jeanne Barrier for singlehandedly managing the logistics that enabled the Workshop from the time the funding was allocated, through actually coordinating the details of the meeting (for instance, making sure that we had all the equipment and the meals on time), to the bookkeeping details she graciously attended to after the event. We likewise thank the Ecole Normale Supérieure de Lyon for lending us facilities for the Workshop. We are equally grateful to other participants who underscored the significance of the Workshop but did not submit their papers for publication: William S.-Y. Wang (Joint Research Centre for Language and Human Complexity, Chinese University of Hong Kong), Vittorio Loreto (Sapienza University of Rome) and Francesca Tria (Institute for Scientific Interchange, Turin), Jean-Marie Hombert (CNRS and Université de Lyon), Fermin Moscoso del Prado Martín (then a CNRS fellow at Université de Lyon), and Ramon Ferrer-i-Cancho (Universitat Politècnica de Catalunya, Spain). Our exchanges would have certainly been less productive without their engaged contributions. Additionally, Christophe Coupé and François Pellegrino are grateful to the LABEX ASLAN (ANR-10-LABX-0081, French program “Investissements d’Avenir” ANR-11-IDEX-0007) of the University of Lyon and to the Laboratoire Dynamique du Langage (UMR5596) for their support. Lucía Lureiro Porto and Maxi San Miguel thank Vincent Blondel and IOP Publishing for permission to reproduce Figure 8.5. xii

1

Complexity in Language: A Multifaceted Phenomenon Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

1

Complexity in Linguistics

1.1

Linguistics and the Science of Complexity

Complexity has attracted a great deal of attention in linguistics since 2001, at a rate that proportionally far exceeds its invocations in the field since Ferdinand de Saussure, the father of our discipline, in the early twentieth century. The number of books bearing complexity in their title is remarkable, suggesting that there may be an emergent research area whose focus is complexity in Language. The dominant question that the relevant linguists have addressed is the following: To what extent does complexity as observed in different languages or in different modules of the language architecture display both crosssystemic variation and universal principles? This has entailed asking whether there are languages that are more complex than others and explaining the nature of differences. One is struck by the sheer number of book-length publications alone,1 and even more when the numerous journal articles and chapters in edited volumes are added to the total count, regardless of whether or not they include complex(ity) in their title. On the other hand, one is also shocked by the scarcity of works that explain what complexity is, apparently because it is assumed to be known.2 This is quite at variance with publications outside linguistics, which are devoted to explaining various ways in which the notion can be interpreted. 1

2

Book titles containing the term language or linguistic(s) include: Dahl (2004), Hawkins (2005), Risager (2006), Larsen-Freeman (2008), Miestamo et al. (eds., 2008), Sampson et al. (eds., 2009), Givón (2009), Givón & Shibatani, (eds., 2009), Pellegrino et al. (eds., 2009), Aboh & Smith (eds., 2009), Faraclas & Klein (eds. 2009), Cyran (2010), Trudgill (2011), Robinson (2011), McWhorter (2012), Kortmann & Szmrecsanyi (eds., 2012), Housen & Kuiken (2012), Blommaert (2013), Culicover (2013), Massip-Bonet & Bastardas-Boada (eds., 2012), Newmeyer & Preston (eds., 2014), Berlage (2014), Kretzchmar (2015), and Baerman, Brown & Corbett (2015). A noteworthy exception is Ellis and Larsen-Freeman’s (2009) “Language as a complex adaptive system,” which is derived from the lead and seminal chapter by Beckner et al. (also identified as “The Five Graces Group”), “Language is a complex adaptive system: A position paper.”

1

2

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

The subject matter has actually also evolved into what is identified by some as “complexity theory” or the “science of complexity” (see footnote citations). Thus, a convenient starting point for this chapter and this book is to explain what is meant by complexity as it applies both to linguistics and other research areas. Etymologically, the term complexity, as a nominalization from complex, can ultimately be traced to Latin complexus, a past participle of the deponent verb complecti ‘embrace, comprise,’ according to Webster’s Collegiate Dictionary, and also confirmed by the French Petit Robert, which translates it as contenir ‘contain.’ According to the Online Etymology Dictionary, the adjective complex ‘composed of parts’ was borrowed from French complexe ‘complicated, complex, intricate’ (seventeenth century), from Latin complexus ‘surrounding, encompassing,’ past participle of complecti ‘to encircle, embrace.’ In transferred use, the verb meant ‘to hold fast, master, comprehend’, from com- ‘with’ and plectere ‘to weave, braid, twine, entwine.’ The noun complex evolved to mean ‘a whole comprised of parts.’ This etymological definition remains very generic. Beyond it, it appears that no strong consensus has emerged in the science of complexity itself about what complexity means (see, e.g., Strogatz 2003; Gershenson, ed. 2008; Mitchell 2009). There are nonetheless some common themes and properties that recur in the relevant literature. They include the following, which overlap in some ways: (1) Complexity arises from the coexistence of components that interact with each other, not necessarily from the fact that a space or a system is populated with several components or members; it is therefore interactional. (2) Complexity arises from the dynamics of activity coordination or synchronization that integrate individuals as members of a population (e.g., ant colonies, bird flocks, and fish schools); thus, it is dynamical. (3) Complexity emerges from nonlinear evolution, which is driven by multiple factors whose significance may vary at different stages of the evolutionary process; its effects are not constant, but subject to the changing values of the relevant variables. (4) Complexity lies in what brings order out of chaos,3 through what is also known as “self-organization” and was formerly referred to as an “invisible hand” (Smith, 1776).4 (5) There is complexity in any system where the properties of the whole do not amount to the sum of the properties of the components.

3

4

“Chaos” is used here as in “chaos theory,” which studies systems whose behavior is highly sensitive to initial conditions, in the sense that small differences in initial conditions may produce quite divergent evolutions or outcomes. It also seeks to capture emergent patterns from the interplay between order and disorder, from which complexity arises. For a more linguistic take on the “invisible hand,” see Keller (1994).

Complexity in Language: A Multifaceted Phenomenon

3

(6) Finally, complexity is the peculiarity of emergent patterns in a system in constant state of flux between disorder and transient order (or equilibrium). In other words, complexity arises from the dynamics of coexistence and interaction or cooperation of components toward generating the properties of whole. As we are reminded by Loureiro-Porto and San Miguel (Chapter 8), complex should not be confused with complicated (pace the etymology of complex(ity) cited earlier). For instance, airplanes are complicated rather than complex pieces of engineering. Despite the very large number of parts, each part has a clear function that makes it possible – and to some extent easier – to predict its contribution to the whole. On the contrary, it is not evident which role each component of a true complex system (such as an ant colony or a flock of birds) plays in the behavior or function of the whole system, or what it specifically contributes as a unit to the larger, integrated whole. Connecting these interpretations to Language, the idea that a system consists of interacting components is not new. It was indeed at the core of the structuralist program, in which phonemes, words, and other linguistic units were primarily considered as components of structures. An important peculiarity of this tradition is that a language was typically construed as an autonomous system, independent of its speakers and the wider ecology in which it and the speakers evolve. Thus, internal forces and their interactions were paid much more attention than external ones. Although this suits the etymology of the term complexity and some of the earlier points, it understandably omits other interpretations made evident in the science of complexity, especially regarding the dynamical aspects. Complexity arises not just from how the different parts interact with each other but also from how they respond to external pressures of the environment, or the external ecology (Mufwene 2001), outside the system. However, decades later, despite the increasingly interdisciplinary nature of the relevant scholarship in other research areas, most linguists deal with complexity almost as if hypotheses in those other areas couldn’t possibly apply to languages. This observation does not include modelers, often coming from the field of artificial intelligence, who have invoked multi-agent systems or network theory to investigate language emergence and change (e.g., de Boer & Zuidema (2010), Ke et al. (2008), Kirby (2000), Steels (1998, 2011a). Aside from them, others such as Massip-Bonet and Bastardas-Boadas, eds. (2012) or Kretzschmar (2015) have also highlighted the dynamic aspects of language behavior. Overall, linguists still have to ask themselves what interpretation of Language they subscribe to, what complexity may mean under that particular interpretation, whether the architecture of Language and linguistic behavior are exceptional in relation to the common properties of complexity that have been observed in other aspects of nature, and how they contribute to this expanding, more inclusive research endeavor.

4

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

1.2

Components, Structures, and Domains in Linguistic Complexity

Likely because most of them have ignored work in the science of complexity, linguists have remained faithful to the interpretation of a complex system as ‘a whole consisting of several parts.’ Thus, the more parts the whole consists of, the more complex it is assumed to be, regardless of how the parts interact with each other. This explains why an approach commonly found in the literature consists in counting the number of elements of a linguistic system in order to evaluate its complexity. Depending on the specific study, the focus may be the number of phonemes, morphemes, or words, but also relations among variants of such units (allophones, allomorphs, or near-synonyms), or yet the number of categories, rules, or constraints that can be posited in a system. This has come to be called “bit complexity” and has been criticized as uninformative (DeGraff 2001, 2009). Indeed, this approach does not pay attention to possible relationships between the components, and goes counter to the well-known idea that “simply more does not mean more complex.” For example, a set of five bodies moving randomly – in the absence of any interaction force – is not as complex as, let alone more than, a set of five bodies, or even three or four, moving according to gravitational forces they exert on each other (Poincaré 1891). The possibility that a whole with fewer parts engaged is several multilateral interactions can generate more interactive complexity cannot be accounted for in a bit-complexity approach. A good example is when an item generates different interpretations depending on what other item it is combined with. This is illustrated with the particle up in combinations with various verbs such as in pick up, give up, show up, and look up. While the item up is basically the same particle in all these constructions, its contribution to the meaning of each phrase appears to vary. This variation suggests that the particular dynamics of each combination produce the meaning of the whole phrase. The overall meaning of each phrase is not the sum of the meanings of its parts (see, for instance, Victorri 1994 on the dynamics of such constructions). To be sure, some linguists have shifted from counting elements to assessing how they make a system together. These linguists have first attempted to identify the patterns of interactions between the components and then infer linguistic complexity from the interactions. Their approach has involved building mathematical graphs and then quantifying their “structural complexity” with an appropriate measure. The vertices of the graph correspond to the components of the system, while the edges connecting them reflect how they can be related meaningfully. Another strategy is presented in Coupé et al. (Chapter 6), in which patterns of co-occurrences of phonetic segments are evaluated with respect to the individual occurrences of these segments, in order to detect significant interactions.

Complexity in Language: A Multifaceted Phenomenon

5

Rather than focusing on (mostly pairwise) interactions between elements, another systemic view seeks to describe the linguistic system in terms of regularities and irregularities. A classical concept here is Kolmogorov complexity, which is the length of the shortest program that can produce the description of the system (given a programming language). Behind it is the central idea that the more compressible a piece of information is, the lower its complexity is as well. This algorithmic approach to complexity (Dahl 2004:42) is more processual than the previous one, because the algorithm has to be run in order to get the description. While it has been proved that the Kolmogorov complexity cannot be computed, reasonable approximations can be obtained with standard compression algorithms and be applied to compare complexity between objects. The size of an archive containing the compressed version of the initial description of the system and the means to decompress it (i.e., a selfextracting archive) is inversely proportional to the complexity of the system. Rissanen’s (1978) “minimum description length” is another possible approximation to Kolmogorov complexity. Such approaches have been applied especially to measuring the complexity of morphological systems (e.g., Bane 2008, Walther & Sagot 2011). Some linguists have also echoed Gell-Mann’s (2003) concern that Kolmogorov complexity is highest in the case of totally random expressions, while we intuitively do not see totally unordered systems as complex. His counterproposal for measuring complexity effectively, named effective complexity, is the length of the shortest description of the set of regularities of the system. Along these lines, Newmeyer and Preston (2004:182) also state that “the more patterns a linguistic entity contains, the longer its description, and then the greater its complexity.” Although these quantitative approaches offer more refined considerations of linguistic complexity, they rely on the descriptions that linguists can provide of a linguistic system. When different options compete in this regard, the quantification methodologies themselves do not help. This echoes Edmonds’ (1999) statement that complexity lies before all in the eye of the interpreter of the system. Another way of considering this is that a descriptive account of complexity can be at odds with a more functional approach: Does the complexity of the description of an utterance always correspond to the amount of difficulty the hearer experiences in processing it? Does the description capture adequately the nature of the neural and psychological encoding, and its consequences in terms of processing? Is such an approach to complexity informative about the overall complexity of a language? Attempts to assess the complexity of a whole language present an additional difficulty with the same endeavor in other, physical or cultural systems. For example, ferromagnetic materials, a well-known physical system in which selforganization of microscopic magnets can occur in the absence of strong external magnetic field, are only composed of identical elements without hierarchical

6

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

structure. By contrast, in the case of languages, different modules and levels of analysis – phonology, morphology, syntax, semantics, and so on – can be distinguished, and the hierarchical integration of their elements (starting from meaningless sound units) into phrases, sentences, and discourse raises a number of issues. Complexity can be investigated in each module independently but also between these modules, raising research questions such as whether the lesser complexity of a module will be balanced by the greater complexity of another. Such a conception of equilibrium can also be considered within a domain, if, for example, one attempts to check whether the greater complexity of the consonant or vowel system is counterbalanced by the lesser complexity of the syllabic structures (Maddieson 2011). Technically, comparing distinct domains such as phonology and morphosyntax is uneasy beyond simply counting elements, which, as remarked earlier, typically disregards the interactions between them. On a more theoretical level, what is obviously missing from the relevant literature is the interactional complexity that arises from the division of labor and cooperation between different components, including members of the same module. However, it is not evident how many modules must ultimately be posited to account for how the production and interpretation of utterances work in a language. As a matter of fact, Lieberman (2012) goes as far as rejecting the idea of modules, arguing that the neurons of the brain are connected in a way similar to (though more complicated than) the parts of an automobile engine. However, if one subscribes to the modular architecture of language, it becomes important to understand how the modules interface with each other during the production and interpretation of utterances, certainly not in a linear way (McCawley 1998). According to the latter, the modules work concurrently rather than sequentially, as is made evident by, for instance, the correction of false starts while speaking and self-corrections of the interpretations of utterances as the discourse evolves. Let’s assume that the materials of a language fall in one or another module (viz., phonology, morphology, syntax, etc.), each of which makes a clear contribution to the overall system, while its components (such as individual sounds in a phonemic system) do not. We may then have to wonder whether languages do not fall in between complicated and complex systems, consistent with Loureiro-Porto and San Miguel’s distinction (Chapter 8). The question is difficult to answer within the bit-complexity approach. On the other hand, determining whether a language is complex or complicated becomes rather pointless without reference to something that it can be compared with. This explains why linguistic complexity has typically conjured up cross-linguistic research. Any measure of complexity presupposes or

Complexity in Language: A Multifaceted Phenomenon

7

entails some cross-linguistic comparison of phonological inventories, morphological systems, and so on, as is evident, for instance, in discussions about whether or not creoles’ grammars are simpler than those of other languages (e.g., McWhorter 2001, DeGraff 2001). This approach also suggests that a language may exist that has no, or very little, complexity built in it. However, from the point of view of the interaction of modules, McWhorter’s (2001) and Gil’s (2001, 2009) claims about the simplicity of the grammars of creole vernaculars and Riau, Indonesia, respectively, beg the question. However, see Gil’s (2009) reaction discussed below. 1.3

From Static to Dynamic Linguistic Systems

The complexity of a linguistic system can be assessed synchronically, relative to a given time, regardless of what the system was like before. However, languages are constantly changing and being adapted to satisfy various communication pressures, including those that index speakers and the circumstances of their interactions. Beyond structures that may be assumed to be a static response to a fragile assemblage of structural constraints – in the spirit of a saying usually attributed to Ferdinand de Saussure, “la langue est un système où tout se tient” – linguistic systems are in a constant state of flux, with new components appearing and older ones evolving or disappearing. It thus makes sense to ask how complexity evolves under these ecological pressures, and see languages or their subsystems as complex dynamical systems (Bruckner et al. 2009). Self-organization and emergence express how order and regularities arise from an initially chaotic state (as defined in Chaos Theory). They are fundamental processes in the study of physical and biological complex systems, for instance, how ants may build optimal paths between their nest and a food source, how microscopic dipoles can align to create magnetic domains, and how traders’ activity at a market can result in macroscopic events such as economic bubbles or crashes. These concepts can be invoked to account for linguistic phenomena such as the emergence of new language varieties. For example, both self-organization and emergence can be invoked to explain how elements from several languages have been selected, in varying proportions, into a new variety, called creole. Linguistic systems can merge – with some features being selected and possibly modified, and others rejected – into what appears to be a new dynamic equilibrium. Unlike in the science of complexity, linguistics stands out also by the limited attention that has been given to how complexity emerges, that is, from a diachronic perspective, relative to language development and to the phylogenetic emergence of language. Exceptions include Wang et al. (2004), Givón (2009), Lee et al. (2009), Mufwene (2012), and some of the authors contributing

8

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

to this volume, especially Bart de Boer, P. Thomas Schoenemann, and Luc Steels, the first and the last based on modeling the emergence of language. It is to this fold of linguistic complexity that they were invited to contribute. Different time scales relate to different evolutionary processes: from the ontogeny of complexity in child language acquisition, to its modifications in language change and the evolution of languages, to its rise during the phylogenetic emergence of Language. In ontogenetic and phylogenetic evolution, the focus is especially on how a system develops/evolves from architecturally poorer to richer structures (see, e.g., Dahl 2004, Givón 2009, Givón & Shibatani, eds. 2009). Regarding historical language changes, phenomena such as grammaticalization can be studied with a focus on whether they increase or decrease the complexity of the system (e.g., Heine & Kuteva 2007). Other attempts yet derive diachronic models from synchronic constraints, and observe how system coherence and complexity evolve between lower and upper bounds (Coupé et al. 2009). From a phylogenetic point of view, one should not dodge the question of how complexity arose during the transitions from vocalizations to naming and the rise of phonetic systems, to predication and the emergence of simple sentences, all the way to modern linguistic systems (e.g., Mufwene 2013). The linguistics discourse has generally overlooked the dynamics of the linguistic elements in relation to each other, such as what may happen when a new sound is added to the phonetic inventory of a language; or when a preposition is used as the syntactic head of the predicate phrase (like in dis buk fuh you ‘this book [is] for you’ in Gullah), whereas a verb has traditionally been required in this position in English (Mufwene 1996). That is, while speakers/signers modify the extant system with their innovations, the latter may trigger other adjustments in the system. This is the case in the sentence You bin fuh come ‘you had/were expected to come’ in Gullah, where, because it can function as head of a predicate phrase, the preposition fuh has also been coopted as a marker of obligation, the counterpart of a modal verb in English. (As head of the predicate phrase it can also be modified by the anterior tense marker bin, regardless of whether it functions as a preposition or as a modal marker.) Such a change by cooption of extant materials is undoubtedly true of other cultural systems, which are also adaptive but depend primarily on the activities that shape them, those of the practitioners of the culture. Future research should return to this issue, which arises also from some of the contributions to the present book. An important question in such systemic adaptations is: What are the forces or constraints responsible for linguistic change? Answering this question offers complementary and enriching perspectives on linguistic complexity. Indeed, the previous approaches can all fit a framework where linguistic structures are considered in isolation and studied on the basis of their internal (possibly dynamic) patterns of occurrences or interactions. But considering the various

Complexity in Language: A Multifaceted Phenomenon

9

dimensions – social, cognitive, and pragmatic – of what may be called the ecology of language (Mufwene 2001, 2008; Coupé 2016) opens further avenues toward a more complete understanding of linguistic complexity. 1.4

Complexity and Language Ecology

Language ecology has usually been invoked in relation to social factors (Mufwene 2001, Lupyan & Dale 2010).5 Indeed, a language does not exist outside its social environment; it is a communal creation, with structures shaped through speakers’ communicative acts. It displays emergent patterns, which linguists have attempted to capture in the form of rules and constraints, from a synchronic perspective. However, there are ecological factors that arise from within the system itself that also influence the evolution of a language (e.g., frequency, transparency, regularity, and length of particular variants). They determine which variants will prevail and which ones will remain minority alternatives or will be given up.6 Innovations and their replications (or copies) compete among themselves, subject to these and other ecological factors, social and otherwise (Mufwene 2001, 2008; Blythe & Croft 2009). This is especially noticeable in cases of language contact, when a new variety (such as a creole) emerges and retains only a subset of the variants in the prevailing language (called lexifier) and only some of the competing substrate features are selected into the emergent language variety. Equally, if not more, interesting are cases where the competition7 is not resolved. For instance, in (standard) English, the primary stress in the word exquisite may be placed on the first or second syllable; a relative clause may start with a null complementizer, with the complementizer that, or with a 5

6

7

Mufwene actually applies the term ecology to a wider range of factors, both internal and external to particular languages, some direct and others indirect, that influence the evolution of a language, including its vitality. He applies the term to any factor that may be considered as (part of the) environment relative to a language (variety) or a linguistic feature being discussed. Relative to language evolution, some ecological factors are economic and historical. Relative to the phylogenetic emergence of Language, Mufwene (2013) singles out the human anatomy and the brain/mind as critical ecological factors. Linguists such as Weinreich et al. (1968), McMahon (1994), and Labov (2001) have invoked actuation (similar to but not exactly the same as actuator in physics) in reference to the particular combination of factors, which are indeed ecological, that produce particular changes at specific places and at specific points in time. This tradition of course underscores the need to approach language evolution from the point of view of complexity and emergence, as these notions are construed in the science of complexity, in dynamical terms. As explained in Mufwene (2008), “competition” is used here in the same sense as in evolutionary biology, applying to variants, organisms, or species that ecology may not sustain equally, favoring one or some but disadvantaging the other(s). In languages, variants for the same function (including languages spoken in the same community) are often rated differently by their speakers or signers, a state of affairs that explains why some disappear. From an evolutionary perspective, and even that of language ontogenetic development (influenced by who the learner interacts with), different speakers/signers may not rate the variants in identical ways.

10

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

relative pronoun (a wh form); and some speakers say I want you not to come, whereas others prefer I want you to not come. There are indeed a host of similar examples not only in English but in other languages too. What also appears evident in such cases is not only the number of distinctions that are made but also the ways in which the distributions of the competing expressions are articulated. No assessment of complexity in a language should ignore this interactive aspect of the system, which is consistent with Saussure’s notion of opposition between forms or between constructions: the expressions derive their meanings from how they are opposed to or distinguished from each other. How is variation managed in a language or speech community? Can it remain free, in the sense that a speaker/signer may use any variant or another without any communicative or social consequences? Or is it constrained by other factors that are social, such as age, gender, ethnicity, profession, and level of education, or those that stem from the precise context of interaction? Are the constraints rigid or flexible? The interfacing of systems (consisting of structural units and rules) and social constraints emanating from the communities using and shaping the languages appear to foster alternative interpretations of linguistic complexity, which also explains variation in the way that linguists discuss it, as is evident in, for instance, Sampson et al. (eds., 2009) and MassipBonet and Bastardas-Boada (eds., 2012). To the extent that languages can be construed as communal systems, complexity arises at least as much from the dynamics of interaction within the population associated with the language, as from the actual system hypothesized by the linguist (or any analyst). Linguistic complexity therefore conjures up complexity of linguistic structures and external constraints exercised by ecological factors, including specific kinds of social interactions and the particular business or social networks in which one operates. No speaker/signer has complete knowledge of their communal language as an ensemble of idiolects (Mufwene 2001), while they all use it and adapt their respective idiolects relative to other users. The speakers’/signers’ mutual accommodations and their respective responses to novel communicative pressures (which are similar to adaptive responses of elements of better understood complex adaptive systems) drive change or evolution. This peculiarity explains the claim that languages as both practices and systems are in a constant state of flux, hardly staying in equilibrium, and are therefore emergent phenomena. These dynamic aspects of complexity are hardly quantifiable. They also make it obvious that, as stated by Beckner et al. (2009), the agents of the emergence of complexity are the speakers/signers who manipulate the system. They are the ones that modify it, innovate new forms and structures, introduce new dynamics of interaction among the different components of the system on different levels, and therefore modify the patterns of complexity in one way or another. On the other hand, this agency also sets the discourse on linguistic

Complexity in Language: A Multifaceted Phenomenon

11

complexity at odds with epistemological models of complex systems in the science of complexity, such as ant colonies and flocks of birds, in which the agents are parts of the system rather than its users/manipulators. Speakers and signers largely differ from birds flying in flocks or ants living in colonies in that they do not reposition themselves physically but modify their idiolectal characteristics to approximate those of the other speakers or signers they wish to align with socially or professionally. In other words, a bird situates itself as part of a system, whereas a speaker or signer realigns his or her linguistic productions and uses these modifications to situate him or herself socially or professionally but not in the linguistic system itself. Also, while a bird never changes the innate rules that dictate how, while flying, it adapts its speed and direction to the speeds and directions of neighboring birds (see, e.g., Hildenbrandt et al. 2010), humans are much more flexible when it comes to adapting their behaviors – that is, linguistic strategies – which are not what they are as individuals. At the root of this behavioral flexibility obviously lies highly developed cognitive capacities, which allow us to internalize part of the complexity of the linguistic system we are immersed into. Crucially, they enable us as speakers to anticipate the possible effects of our words on the hearer’s mind, and to reconstruct as hearers what was in the speaker’s mind when they produced the message we just received. Beyond social aspects, cognitive aspects also shed light on linguistic complexity. As much as languages are social, communal constructions, they are also processed and stored in individual minds. In a way similar to the aforementioned algorithmic complexity, linguistic complexity can be estimated based on the cognitive efforts required by the mind (conceived of as the state of the brain in activity) to produce or process a message. The question of how difficult specific items or subsystems are to learn especially points to a learnability complexity. Along these lines, Dahl (2009) makes a distinction between, on the one hand, “absolute complexity,” ascribed to the mechanics of the system, and, on the other, “relative complexity,” based on how much difficulty a learner experiences in learning a language. What does this “relative complexity” tell us about the inherent complexity to learn a given language? The answer appears to be negative, because “relative complexity” is predicated on the fact that speakers of particular languages find some other languages harder to learn for various reasons, for instance, the tone contrasts are too difficult to replicate faithfully, or there are too many morphological distinctions (especially inflections) to remember accurately, or there are too many constraints regulating when particular variants can or should not be used. This complexity must be assessed variably, depending on which speaker of which language is learning which other language. However, all languages are

12

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

effectively learned and spoken by their speakers, regardless of their very different strategies when it comes to conveying information. There is no evidence that it takes speakers of, say, tonal and agglutinating languages longer to develop native competence in them than speakers of toneless and English-like languages that are not agglutinating or polysynthetic. What is generally overlooked in this respect is also inter-individual variation in learning skills, even in the acquisition of one’s mother tongue. For instance, while the English article system (including usage of bare nouns in the singular form) and distinction between the preterit and present perfect seem too illusive to many non-native speakers, there are also many others who have no serious problem with them. Going further, we must also consider the fact that, beyond moderate interindividual differences, all living humans are endowed, through their common biological phylogeny, with the capacity to learn modern language(s), identified by Noam Chomsky as the “biological endowment for language.” Assuming that this capacity is unique to the human mind, the cognitive aspects of linguistic complexity are relevant when it comes to the phylogenetic evolution of human societies and communication. For instance, Mufwene (2013) argues that, as the hominine mental capacity increased and hominines developed larger social organizations in which they had to manage their modes of coexistence and norms of interactions, the pressure for more informative communicative systems grew. According to MacNeilage (1998), neurobiological changes and displacements of communicative functions in the brain supported these changes. Mufwene (2013) also assumes that languages as communicative technology developed incrementally without foresight of the emergent structures, with every speaker and signer contributing in different ways to the process (some more successfully than others), exapting the current system not necessarily according to the same principles. Thus, languages appear to have evolved “chaotically” (i.e., without a master plan, according to the science of complexity) toward more complexity both in the architecture and size of the system and in the quantity of interactions between components within and between their different modules.8 The modules themselves may have instantiated complexity, in that they may be assumed to have emerged by self-organization, with the mind dividing the labor to be run concurrently, for faster production and processing of utterances (Mufwene 2012, 2013). 8

This is not a denial of regularities within the different modules of languages. Exaptations are based on analogies between, on the one hand, the new meanings to be conveyed or the new function to be played and, on the other, some meanings or functions currently in use, except that different speakers/signers do not necessarily perceive the same ones when they innovate. Thus, although they introduce variants that compete with each other, they also introduce regularities (Mufwene 2008). Because the resolution of competitions between variants depends on various factors that are not so predictable, the situation is comparable to chaos in complexity.

Complexity in Language: A Multifaceted Phenomenon

13

A last overarching ecological factor bearing on linguistic complexity is the pressure to communicate adequately or accurately. Most of the contributions to Ellis and Larsen-Freeman (2009, best articulated by Beckner et al.) and to Massip-Bonet and Bastardas-Boada (2012) stand out in subscribing to the idea that languages are adapted to the communicative needs of their speakers/signers, thus they reflect changes in the latter’s cognition and adaptations to their social and other ecologies. They keep changing over time, also reflecting the accommodations that the speakers/signers make to each other toward the emergence of some communal norms (however transient these may be), including how they respond selectively to each other’s innovations (see also Mufwene 2001, 2008). Along this line, some usage- and agent-based investigations in artificial intelligence attempt to shed light on the emergence of complexity in linguistic systems. They focus on how individual communicators (often identified by modelers as “agents”) exapt current structures for novel communicative needs and generate new structures (e.g., Lee et al. 2009). However, very little of this (e.g., Wang et al. 2004) is integrated in ways that also address the role of ecological factors identifiable in the anatomical, mental, and social aspects of communication (as discussed above) that contribute to the emergence of various interpretations of linguistic complexity. Different individuals may have (noticeably) different communicative needs. How exaptations are made depends on the context of use of the extant communication system and on individual speakers’/signers’ skills, influenced by their ontogenetic trajectories. But the evolution of languages by exaptation applies to both the phylogenetic and the ontogenetic development of linguistic systems. We assume that our hominine ancestors from one million years ago, for example, needed less expressive communication systems, owing to their less developed cognitive/mental capacities and less complex social organizations. Similarly, young infants have fewer communicative needs than adults, and they make fewer nuances, especially regarding the pragmatics of utterances. Their cognitive capacities and social skills are definitely less developed in the sense that they have not yet developed a full “theory of mind” or mindreading capacity, which forms the basis of the inferential adult communication (Sperber & Wilson 2002). They do not do as well as adults regarding, for instance, metaphors, irony, implicatures, and the significance of connotations (in addition to denotations) in linguistic communication. Their range of skills for mature competence is not fully in place by puberty and will certainly continue to expand until adult life. Hawkins (2009) introduces the concept of effective complexity,9 according to which an utterance with more words may be easier or faster to process 9

Not to be confused with Gell-Mann’s “effective complexity” introduced in the beginning of this chapter.

14

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

than an alternative with fewer words. This appear to be puzzling, as the longer an utterance is, the more taxing it is on the short-term memory for processing, in part because the number of (morpho-)syntactic relations within the utterance increases. Or, to repeat our quotation from Newmeyer and Preston (2004:182), “The more patterns a linguistic entity contains, the longer its description, and then the greater its complexity.” However, according to Hawkins (2009:259), “Complexity in form processing is matched by simplicity with respect to the processing functions performed by rich case marking and definite articles.” That is, what is structurally more complex is not necessarily more complicated. The position suggests that we should not overlook the computational aspect of processing utterances, as one may have to infer more when less is said explicitly, such as in the case of the so-called “restricted code.” However, Gil (2009:24) denies that there is evidence of such “hidden complexity” in Riau, Indonesia, which, he claims, has very little syntax. Koster (2009), Mufwene (2013), and Dor (2015) see “language[s] as technology” developed to meet humans’ communicative needs. However, as there are alternative ways of solving the same problems, different populations have not developed their languages in identical ways, which account for typological variation. As with other technologies, it is indeed legitimate to ask whether some languages are simpler than others. However, as observed by Hawkins (2009), among others, what must one measure in assessing complexity: only the physical, mechanical parts of the language technologies or also their abstract aspects involving various structural and pragmatic rules/principles? The particular way in which one answers this question should help determine whether comparing different languages relative to their complexity, as opposed to trying to understand how complexity arises in language and how to explain it, is an intellectually rewarding exercise. Moving closer to practitioners of the science of complexity (viz., outside linguistics), how variably does complexity arise from the different ways in which different populations shape their languages through their communicative acts? From an evolutionary perspective, languages are perceived not as designed deliberately by particular populations and in different ways, but rather as arising spontaneously from attempts by different populations to communicate using phonetic or manual means (Mufwene 2013). This piecemeal emergence and evolution, in unpredictable, “chaotic” ways, favor treating them as complex adaptive systems. This is also the kind of position that arises from proponents of usage-based or construction grammar, including Croft (2000, 2001, 2009), Steels (2011b, 2012), and Kretzschmar (2015), among others. From this perspective, students of linguistic complexity must explain the consequences of thinking of languages as complex adaptive systems. For instance, is the addition of a phoneme to the phonemic inventory of a language, or the deletion of one from it, as adaptive as the addition a new meaning or word to the

Complexity in Language: A Multifaceted Phenomenon

15

lexicon of a language, or even the loss of one meaning or word? Are the systemic consequences the same in both cases? What about grammatical rules? Overall, the various folds of linguistic systems call for an approach to the global complexity of languages that is distinct from assessing the complexity of particular linguistic structures. That is, the adaptation of linguistic structures to their ecologies (i.e., the social, cognitive, and interactional contexts in which speakers evolve) may justify positing a usage complexity, rooted in the actual use of language rather than in more theoretical structural considerations. A linguistic system that fully responds with its structures to the communicative needs and the context of use may be seen of little complexity, while a complex linguistic system may be characterized by structures that do not reflect these needs. For example, in different languages of the North Pacific Rim, dominant winds play a significant role in the linguistic description of space and spatial directions (Fortescue 2011).10 An adequate coupling exists between these linguistic systems expressing space and their context of use, which can relate to a low complexity. But due to climatic change in these regions, wind patterns have recently been changing (Gearheard et al. 2010) and some linguistic systems may therefore now be at odds with their context of use. In such cases, the coupling is now characterized by a greater complexity, and change is needed to restore a better adequacy of the linguistic system (e.g., move away from winds to describe directions). Such a complexity seems to echo a cognitive complexity, in the sense that non-adapted forms, given their discrepancy with the actual physical and mental world of the individual, may require more cognitive efforts to be processed or learned. Because speakers’ natural and socioeconomic ecologies constantly change, as do their communication needs, a language always has to be adapted to these changes. This coupling, which is typically unplanned and ad hoc, cannot be perfect, partly because of the time needed for the linguistic system to be adapted to new social and pragmatic evolutions (and possibly cognitive evolutions in a phylogenetic perspective); additionally, the coupling is imperfect because speakers/signers have no foresight of the ultimate consequences of their current communicative behaviors for the overall system of their language.11

10 11

Expressing directions according to winds can also be found in other languages, for example in the Oceanic family (Palmer 2007). As made evident by the literature on child language development and on the phylogenetic emergence of language, the changes affecting speakers/signers have to do with the increase of their mental capacity and the richer experiences they develop with their social and other aspects of their ecologies, including the climates of their settings, the fauna, and the vegetation. If coordinating their social lives is the primary function of language (Chapter 5 in this volume), then all these factors exert constant pressure on language to keep adapting to new communication needs, which Mufwene (2013) characterizes as adaptive responses to changing ecological pressures.

16

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

1.5

The Debate Over the Equal Complexity of Languages

A classic issue in linguistics pervades the various approaches to linguistic complexity discussed in the previous sections: it is the question whether one (type of) language is more complex than another (Sampson 2009, Gil 2009, Trudgill 2009). Some of the recent publications have specifically been driven by McWhorter’s (2001) claim that “the world’s simplest grammars are creole grammars.” The debate is most evident in Sampson et al. (eds., 2009), in which the contributors articulate different views. As is made more obvious by especially Bisang (2009) and Deutcher (2009), most, if not all, claims that one (type of) language is more, or less, complex than another depend on what modules of a language are considered as (most) representative of the architecture of language and sufficient to justify one’s conclusion that a (type of) language is more complex than another: morphology and syntax only or also the phonology and semantics? For the proponents of equal complexity, an argument is that compensations may take place between these modules: if morphosyntax is more complex in language A than in language B, semantics in B will be less complex than in A. However, this literature has typically focused on “bit complexity” (i.e., a system having more units than another), which again does not do justice to the interactions between components. If one considers the issue more globally, what does one make of the pragmatic considerations that help us decide whether, say, drop the ball must be interpreted literally or as an idiom? Does a language boil down to the mechanics that help us code and decode information (Hawkins 2009)? Or does it also include the principles that guide the encoding and decoding processes, including the choices that one must make among competing variants and the extent to which one must rely on context during these processes? Both Bisang (2009) and Deutcher (2009) conclude that there is no well-articulated measure of what the global complexity of a language is; therefore claims that some languages are more, or less, complex than others amount to what Deutcher compares to “urban legends.” This is not to say that the relevant literature has not taught us anything about complexity. Some studies shed light on interesting ways in which some languages vary in the mechanics of their forms and constructions, regardless of whether or not these are considered as compensations: for example, languages that are spoken at a faster syllabic rate – Spanish and Japanese compared to Mandarin and English – tend to need more syllables to convey a given semantic content, and vice versa (Pellegrino et al. 2011). Introducing speech rate – and thus language usage – in the equation, this study enabled the authors to shift the debate from a putative (un)equal complexity of languages to whether languages have an equal communicative capacity.

Complexity in Language: A Multifaceted Phenomenon

17

If languages are complex adaptive systems relative to the communication needs of their practitioners (according to Mufwene 2013, in ways similar to technologies with varying designs and developed incrementally to solve similar problems),what is to be gained from comparing, for instance, the overall complexity of a pidgin to that of a non-pidgin? After all, they are not used in the same ecologies of interactions nor have they evolved to meet identical communication needs: one is used strictly as a lingua franca for limited communication needs (typically basic informal trade transactions), whereas the other is used as a vernacular for a broader range of communicative functions. Is the structure of a pidgin less modular? Aren’t pidgins modular and generative like other languages? Don’t utterances in a pidgin involve compositionality and therefore some form of syntax, although this may not be as elaborate as in non-pidgin languages? Haven’t pidgins emerged in a nonlinear fashion? Don’t they respond to novel communicative pressures in the same way as non-pidgin languages? Besides, Gil (2001, 2009), for instance, argues that limited syntax is not an exclusive peculiarity of pidgins. We must ask whether traditional discussions do not simply suggest that there are alternative ways of developing or evolving a language relative to the communication needs it is intended to satisfy. In other words, while it is indeed legitimate to question the assumption that all languages are equally complex, it is not clear that most of the current scholarship can help linguists answer questions such as the following: Why are human languages (including incipient pidgins and child language) more complex than animal means of communication, and in what specific ways? How do human languages as emergent phenomena or complex adaptive systems vary among themselves from the point of view of complexity? Aside from the obvious fact that they differ typologically in various ways, do they display different patterns of complexity? In terms of the complexity of the overall linguistic system, are there informative ways of articulating why a language may be claimed to be more complex than another and how? Do these challenges call for an operational definition of Language, which can be assumed by all who engage in measuring the overall complexity in particular languages or some of their modules? 2

The 2011 Workshop on Language Complexity

The following short history will help the reader put the contributions in this book in the relevant perspective. The book evolved out of a successful workshop also titled Complexity in Language: Developmental and Evolutionary Perspectives that we the editors, along with Jean-Marie Hombert and Egidio Marsico, hosted at the Collegium de Lyon in May 2011, when the lead editor was a fellow there. We had invited a select slate of experts in evolutionary linguistics and child language development, in paleontology, and in artificial intelligence

18

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

to assess the state of the art, focusing on ontogeny and phylogeny, without necessarily overlooking synchrony. The invitation stated the following goals: For the “Complexity” workshop, our aim is to sort out things about the most useful way(s) to conceive of complexity in language. Need there be only one way or can there be several ways specific to particular research objectives? For instance, should the interpretation in relation to the phylogenetic emergence of language be the same as in comparisons of structures of modern languages? Indeed, can one claim that one language is more complex than another? If the answer is affirmative, how does he/she go about demonstrating it? If the answer is negative, what are the arguments in support of the position? Does the scholarship on language complexity measure up to the current scholarship on complexity theory? Can one discuss complexity without discussing emergence as understood in complexity theory? We would like to address some of these questions and/or any others that you may think of.

To be sure, some authors such as Östen Dahl and Talmy Givón discuss some of these aspects of complexity briefly, the former especially in relation to emergentism and the latter in relation to the phylogenetic evolution of language. Nonetheless, we think that modern linguistics may benefit from more exchanges of ideas, especially those also engaging colleagues from other disciplines who are modeling various dynamical (systemic and social) aspects of language. We want to emphasize that our goal is not to downplay the relevance of those approaches that focus on different aspects of structural complexity. Rather, it is to shed more light on the other, interactional/dynamical and emergentist aspects of complexity that deserve just as much attention and provide us a better sense of how linguistic communicative systems differ from their nonlinguistic counterparts both systemically and socially. The contributors to this book address complexity from the perspectives of both the evolution and the ontogenetic development of language. They focus on social dynamics involving decisions that speakers or signers make (not necessarily consciously) during their interactions with others and on the dynamics that produce systems out of the different units or constructions they use frequently in their utterances. This approach helps us address the question of whether, say, pidgins (leaving creoles alone) still exhibit some complexity and remain generative, in the sense that they can generate new structures and thus be adapted to the expanding communication needs of their speakers, as is evidenced by expanded pidgins such as Cameroon and Nigerian Pidgin Englishes, Tok Pisin, and Bislama. 3

The Chapters

The body of the book starts with the chapter by Luc Steels and Katrien Beuls. Focusing on the origins and evolution of grammatical agreement as a case study, they use multi-agent modeling to explore how various aspects of complexity

Complexity in Language: A Multifaceted Phenomenon

19

(especially in the system and in forms, among others) emerge in language. Their working assumption is that complexity arises gradually from innovations produced by interactants largely to meet their new communication needs. It arises also from the competing variants (phonetic, lexical, syntactic, and semantic) they introduce during the process, owing largely to imperfect copying. On the other hand, this communal form of complexity decreases as the speakers’ emergent idiolects converge toward some norm (which maintains less variation), the outcome of their repeated successful interactions. According to Mufwene (2001), the mutual accommodations that speakers/signers make to each other are indeed among the mechanisms that drive selection in language evolution, in particular the emergence of new language varieties such as creoles. Steels and Beuls illustrate another fold of complexity by discussing, with some examples, the way in which ambiguity (in simpler forms or structures) increases complexity in processing. This arises from the fact that the hearer has to eliminate references that may be associated with particular constructions or interdependences between constituents that are not relevant in a multiword utterance. They show what a critical role agreement markers play in disambiguating utterances. This suggests that, although they have typically been interpreted as adding complexity to the structure of a language, agreement markers actually decrease complexity in processing. If complexity is assessed in terms of processing time (Newmeyer & Preston 2014), morphological complexity does not appear to be particularly costly when it enables speakers to express more information compacted in a short form, as with fusional markers such the Latin –arum inflected on a noun to indicate that it is plural, feminine, and in the genitive/possessive. This appears to corroborate Hawkins’ (2009:259) position that “complexity in form processing is matched by simplicity with respect to the processing functions performed by rich case marking and definite articles.” Likewise, the cooption of some current forms for new grammatical functions, such as in grammaticalization, is said to be a case of “damping complexity,” as the strategy reduces guesswork in figuring out the new meaning or function. So is the erosion of forms or constructions that follows for ease of production, supporting their hypothesis that speakers tolerate just the necessary amount of complexity they need to communicate efficiently in their language; otherwise they dampen it. Steels and Beuls’ discussion highlights the fact that the architecture of a language is multi-modular (though it is not evident how many modules one must posit) and that complexity can be assessed differently, depending on the work that the module is assumed to do. Though we need not subscribe to the traditional position that all languages are of equal level of complexity, we may need a multi-dimensional metric for assessing the extent to which a language is more, or less, complex than another.

20

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

Chapter 3 focuses on the following question: When a language can function with just a dozen contrasting sounds (e.g., Hawaian), why does the average phonetic inventory of the world’s languages amount to 29 sounds? This leads Bart de Boer to start with the observation, “Languages are more complex than is strictly necessary for their communicative function.” Assuming that this makes it possible to tell which language is more complex than another phonetically, he focuses on determining “which aspects of linguistic complexity are due to cultural processes, and which aspects are due to cognitive biases.” One may want to entertain the question of whether the emergence of languages can really be attributed to cultural processes. If culture is understood roughly as the particular ways in which members of a particular population behave and do things, is this question well formulated? Is a linguistic system not a cultural system enabled by the particular evolutionary trajectory of its practitioners and shapers? Not only is cultural evolution not mutually exclusive with biological evolution, it also presupposes it. Only animals endowed with uniquely generative and highly adaptive mental/cognitive capacities (viz., the hominine species) have produced human cultures, aspects of which include culture-specific languages (Mufwene, in press). We want to clarify that de Boer does not want to exclude the role of biological evolution in language evolution cum cultural evolution. What he means by “cultural processes” appears to be related to the fact that nobody really builds a language with foresight and based on a master plan. If we can borrow from William Croft (this volume, Chapter 5), a linguistic system emerges in the same way as other “emergent phenomena” (the way systems are seen in the science of complexity), through the addition or disuse of the strategies that the interactants develop in the here and now of their communicative acts, as they integrate gradually into a system. De Boer concludes tentatively that the “complexity of phonological systems is due to cognitive mechanisms that re-use and generalize building blocks.” This appears to be the consequence of transmission through learning by inference, which replicates unfaithfully and introduces (more) variation, as well as of the nonlinear way in which linguistic systems evolve. In Chapter 4, P. Thomas Schoenemann argues, in ways consistent with Bart de Boer, that “the complexity of a language is the result of the evolution of complexity in brain circuits underlying our conceptual awareness.” According to him, modern languages have evolved from the complex interactions of biological evolution, cultural evolution, and successions of ontogenetic development in several generations of individuals in particular populations. Linguistic systems, with their patterns, have been facilitated by humans’ “sociallyinteractive existence,” which is reminiscent of Steels and Beuls’ discussion of how communal norms emerge. From this perspective, Schoenemann argues that language complexity can best be understood when it is grounded in an

Complexity in Language: A Multifaceted Phenomenon

21

evolutionary perspective, focused on interactions of the biological evolution with the changes in the ecologies in which the interacting agents evolve. Highlighting differences and similarities between humans and chimpanzees in particular, notably in the ways they conceptualize about the world, Schoenemann also concludes that the differences can be correlated with differences in the anatomies of their brains. However, some of the similarities also suggest that humans’ ability to conceptualize is pre-linguistic, suggesting that the emergence of Language and the complexification of its architecture are the consequence of the further complexification of the human mind, beyond the chimpanzee’s under the conditions of “social-interaction existence.” He observes that ontogenetically, “the development of expressive grammatical complexity appears to be an exponential function of the size of the lexicon.” Assuming that phylogenetic language evolution proceeded gradually, William Croft argues in Chapter 5 that “at least some elements of the structural complexity of modern human languages are the consequence of the cognitive complexity of the conceptual structures being communicated.” He also argues that “it is only in its social interactional context that the evolution of linguistic complexity can be understood,” thus, that “the evolution of social-cognitive complexity (in terms of joint action) is a prerequisite for the evolution of structural complexity of linguistic signals.” Language as a communal phenomenon is the product of joint action; it is more than the sum of the actions and systems of its practitioners. Thus it satisfies the characterization of a complex system according to practitioners of the science of complexity, especially since it can work only if every member of the community cooperates toward its successful behavior/practice. From an ecological perspective (Mufwene 2001, 2013), Croft also highlights the role played by the material in determining the shape of the emergent semasiographic system (“encod[ing] information in a lasting, visual medium”), for instance, the role of clay in reducing the number of iconic signs. If this hypothesis is correct, one may assume that the shapes of the Chinese logographic characters were largely influenced by the use of papyrus and ink. Overall, Croft’s general observation is that writing systems, which have evolved from simpler nonlinguistic and more iconic semasiographic conventions, emerged gradually, becoming more arbitrary as they were increasingly being used to represent speech. It is, of course, debatable whether the Chinese system has evolved to serve speech, as the same graphic representation can be read equally in any Chinese variety (e.g., Mandarin or Cantonese). Croft also argues that “writing did not express grammatical elements until centuries after its first emergence. In other words, substantial common ground between author and reader was required to interpret just the linguistic form encoded by early writing.” Illustrating how exaptation works in cultural

22

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

evolution, Croft shows through his discussion of musical notations how elements of the current system are recruited for new functions, to help expand it (to the satisfaction of the performers/practitioners). This is indeed reminiscent of the exaptation that takes place during grammaticalization. Based on how semasiographic systems have evolved, gradually and restricted to a wider range of functions, Croft hypothesizes that “language began in highly restricted functional domains, and its extension to become a general-purpose communication system was a long and gradual process in human prehistory.” He further concludes that, like the semasiographic systems, “language initially functioned simply as a coordination device for joint action,” conveying minimal information.12 It is only later that it became more explicit, conveying richer information, and developed more complexity in its architecture, especially as it developed “displacement” (Hockett 1959), the capacity to convey information about entities and states of affairs that are not present. In Chapter 6, Christophe Coupé, Egidio Marsico, and François Pellegrino start with a historical synopsis of the interest of linguists in complexity since Ferdinand de Sausure’s (1916) characterization of a language as a system consisting of interacting parts. Then they consider the particular ways in which scholarship in complexity theory, as practiced in especially physics, mathematics, and cybernetics, has inspired some of the current research in phonology. They also underscore the fact that “a language is an aggregate of individual idiolects” (comparable to Mufwene’s 2001 idea that it is “extrapolation from idiolects”). As spoken of in linguistics, individual languages are reductions of convenience, which overlook interidiolectal variation, which is more closely matched by Michel Breal’s (1897) idea that every idiolect is somewhat a separate language. Both interidiolectal variation and the breakdown of sounds into (articulatory and acoustic) features add complexity in the ways phonological systems can be thought of. Coupé, Marsico, and Pellegrino also highlight the difference between acknowledging that a system is complex and assessing the extent or level of complexity, while citing some studies that have proposed particular metrics. They warn against importing uncritically hypotheses developed by physicists and mathematicians, which are typically based on simplified models of reality, although they are useful research tools. This is consistent with their basic position that any research field (including linguistics) can contribute to the science of complexity. Applying the statistical method to 451 phonetic inventories, the authors address the question of whether phonological systems worldwide present evidence of preferred interactions among segments that may be based on manner 12

Croft uses coordination in a way related to cooperation in theories of human and cultural evolution, in reference to members of a population engaging in joint actions.

Complexity in Language: A Multifaceted Phenomenon

23

or point of articulation, nasality and orality for vowels, or any other phonetic features. That is, are there any particular features that are more significant than others in the emergence of phonological systems? Their conclusions include the observation that, from an evolutionary perspective, “it [is] ( …) difficult to conclude in favor of strongly non-linear interactions between either features or segments.” They also note that it is difficult to assess the overall complexity of a language using tools developed for physics and biology, as they do not transfer faithfully to language. While it is evident that traditional, typologically oriented discussions of linguistic complexity do not capture the full picture, linguists should consider a more informative metric for addressing the question of whether or not different languages display the same level of global complexity. Barbara Davis focuses in Chapter 7 on the ontogenetic development of the phonological component of language to explore the kind of light the analysis may shed on the phylogenetic evolution of language. Like Schoenemann, she grounds her discussion in the interpretation of complexity in the science of complexity. According to her, “Within the tenets of complexity science, phonological knowledge and behavioral patterns can be seen as emerging from connections enabled by general-purpose child capacities such as learning and cognition as opposed to language-dedicated modular mechanisms.” The emergence of a complex phonological system in the child is driven by both their cognitive-neural capacities and the production system capacities that work in cooperation. She refers to the complex interaction between the environment and the child’s brain in the gradual emergence of his or her phonological system, which appears to call for an approach similar to the analysis of emergent phenomena in the science of complexity, which may presuppose only the disposition of a mind sensitive to complex interactions and ready for complex systems rather than specifically for language. Davis’ observations are similar to those of Schoenemann. She terms this approach “biological-functional approach to phonological acquisition,” according to which “outcomes of phonological acquisition result from multiple interactions between heterogeneous aspects of a complex system.” She moves on to explain the significance of change in both the ontogenetic development and the phylogenetic emergence and evolution of language, especially in introducing complexity throughout the adaptations that the emergent system undergoes to satisfy current communicative pressures. The acquisition of phonology is thus characterized as “‘change’ in infant output capacities.” Thus, the “progressive diversification in the inventory of sound types and how they are produced in sequences relative to ambient language patterns is usually considered a critical index of increasing complexity toward mature phonological capacities.” In Chapter 8, Lucía Loureiro-Porto and Maxi San Miguel approach complexity in language practice from the point of view of language choice in a

24

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

multilingual setting, especially those that may result in language loss. That is, they focus on complexity that arises not from interactions between the different components or modules of language as a system but from various factors external to the system that influence speakers’ choices in their discourses, especially when they have to alternate between languages in a bilingual setting. Their study involves modeling as a simplified tool for addressing certain specific questions regarding linguistic behavior in this particular case. From the outset, the authors articulate the distinction they make between a complex system from a complicated system (such as an airplane), which is “composed of many parts, each [of which] has a clear, identifiable function which makes prediction possible.” Complexity has to do largely with the unpredictability of the properties of the whole from those of the parts. From the point of view of language practice, the whole regards the vitality of a language, as it depends on language choices that speakers make when they interact with each other, without foresight of the ultimate consequences of these decisions regarding the languages in competition. It is the whole ecosystem in which the language belongs, in coexistence with other languages, which is of concern. What are the factors that individually or in combinations determine the choice of one or another language (variety) on specific occasions of social interactions? Things are made more unpredictable by the fact that speakers are not necessarily coordinated about their decisions in the typically dyadic or triadic interactions they are most often engaged in. Loureiro-Porto and San Miguel’s modeling reveals the significance of local interactions in bilingual settings, regarding how they reduce the chances of sustaining the vitality of both of the languages in competition. Other interesting questions arise too, as language evolution is not uniform from one bilingual setting to another. One of them regards when multilingualism spells the endangerment of the less prestigious language(s) and when it does not. In the real world, the explanation can be found in differences between the population structures of the multilingual settings: for instance, those fostering assimilation also favor endangerment, whereas those that are socially segregated according to language groups do not. Because it simplifies reality, modeling helps us become more aware of the complexity of factors that influence the linguistic behavior of (members of) a population and thus bear on language vitality. Underscoring the complexity of actuating factors is the fact that even those population structures that are assimilationist do not endanger the disadvantageous languages at the same speed either. Loureiro-Porto and San Miguel’s modeling reveals differences between small-world networks, regular lattices, and networks with community structure. As the authors conclude, “The kind of network in which interactions take place is a strong influencing factor on language dynamics, as it plays a central role in the potential survival or disappearance of one of the languages in competition.”

Complexity in Language: A Multifaceted Phenomenon

25

Closing the book, Albert Bastardas-Boada approaches linguistic complexity both from the parts to the whole and from the whole to the parts. The aspect of complexity he focuses on is that which arises from the interaction of the system with its ecology, including the socio-conceptual matrix of the speakers’ interactions, economic pressures, the distribution of political power, and the effects and language policies. Complexity increases as a consequence of the fact that populations are not uniform and foster variation, which obtains not only interindividually and between groups, but also intergenerationally. According to the author, a language must be conceived of “as a historical and, therefore, temporal phenomenon, with earlier events playing a major role in how the phenomenon evolves.” History shapes and may provide some explanation for the present, including current linguistic behavior. There are indeed other aspects of complexity that this book, like the dominant literature in linguistics, still does not tackle, despite our focus on developmental and evolutionary perspectives. One of these is the extent to which increase in population size affects complexity in the communal language, perhaps more in the pragmatic and social aspects of its usage than in its structures. Another is whether contact with (an)other language(s) reduces or increases structural complexity, and under what specific conditions. Is contact the only explanation for why major world languages such as Modern English and Modern French have lost most of the inflections of Old English and Old French, respectively? On the other hand, hasn’t contact also increased complexity in their systems in other ways, such as in introducing alternative grammatical rules or changing some of the rules while preserving some exceptions? These are all interesting topics for future studies. REFERENCES Aboh, Enoch O. & Norval Smith, eds. 2009. Complex Processes in New Languages. Amsterdam: John Benjamins. Baerman, Matthew, Dunstan Brown, & Greville Corbett, eds. 2015. Understanding Morphosyntactic Complexity. Oxford: Oxford University Press. Bane, Max. 2008. Quantifying and Measuring Morphological Complexity. In Proceedings of the 26th West Coast Conference on Formal Linguistics, ed. by Charles B. Chang & Hannah J. Haynie, 69–76. Somerville, MA: Cascadilla Proceedings Project. Beckner, Clay, Richard Blythe A., et al. 2009. Language is a Complex Adaptive System: Position Paper. In Ellis & Larsen-Freeman (eds.), 1–26. Berlage, Eva. 2014. Noun Phrase Complexity in English. Cambridge: Cambridge University Press. Bisang, Walter. 2009. On the Evolution of Complexity: Sometimes Less is More in East and Mainland Southeast Asia. In Sampson et al. (eds.), 34–49. Blommaert, Jan. 2013. Ethnography, Superdiversity and Linguistic Landscapes: Chronicles of Complexity. Bristol, UK: Multilingual Matters.

26

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

Blythe, Richard A. & William Croft. 2009. A Usage-Based Account of Constituency and Reanalysis. In Ellis & Larsen-Freeman (eds.), 47–63. Bréal, Michel Jules Alfred. 1897. Essai de Sémantique: Science des Significations. Paris: (Original publisher unknown). Coupé, Christophe. 2016. Commentary: Defining and Assessing Constraints on Linguistic Forms. Journal of Language Evolution 1(1): 52–55. Coupé, Christophe, Egidio Marisco, & François Pellegrino. 2009. Structural Complexity of Phonological Systems. In Pellegrino et al. (eds.), 141–169. Croft, William. 2000. Explaining Language Change: An Evolutionary Approach. London: Longman. Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Croft, William. 2009. Towards a Social Cognitive Linguistics. In New Directions in Cognitive Linguistics, ed. by Vyvyan Evans & Stéphanie Pource, 395–420. Amsterdam: John Benjamins. Culicover, Peter. 2013. Grammar and Complexity: Language at the Intersection of Competence and Performance. Oxford: Oxford University Press. Cyran, Eugeniusz. 2010. Complexity Scales and Licensing in Phonology. Berlin: Mouton de Gruyter. Dahl, Östen. 2004. The Growth and Maintenance of Linguistic Complexity. Amsterdam: John Benjamins. Dahl, Östen. 2009. Testing the assumption of complexity invariance: The case of Elfadian and Swedish. In Sampson et al. (eds.), 50–63. DeGraff, Michel. 2001a. On the Origin of Creoles: A Cartesian Critique of neoDarwinian Linguistics. Linguistic Typology 5: 213–310. DeGraff, Michel. 2009. Language Acquisition in Creolization (and Language Change): Some Cartesian-Uniformitarian Guidelines. Language and Linguistics Compass 3&4: 888–971. Deutcher, Guy. 2009. “Overal Complexity”: A Wild Goose Chase? In Sampson et al. (eds.), 243–251. Dor, Daniel. 2015. The Instruction of Imagination: Language as a Social Communication Technology. Oxford: Oxford University Press. Edmonds, Bruce. 1999. What is Complexity? The Philosophy of Complexity per se with Application to some Examples in Evolution. In The Evolution of Complexity, ed. by Francis Heylighen, Johan Bollen, & Alexander Riegler, 1–18. Dordrecht: Kluwer. Ellis, Nick C. & Diane Larsen-Freeman, eds. 2009. Language as a Complex Adaptive System. Malden, MA: Wiley-Blackwell. Faraclas, Nicholas & Thomas B. Klein, eds. 2009. Simplicity and Complexity in Creoles and Pidgins. London: Battlebridge Publications. Fortescue, Michael. 2011. Orientation Systems of the North Pacific Rim. Copenhagen: Museum Tusculanum Press. Gearheard, Shari, Matthew Pocernich, Ronald Stewart, Joelie Sanguya, & Henry P. Huntington. 2010. Linking Inuit Knowledge and Meteorological Station Observations to Understand Changing Wind Patterns at Clyde River, Nunavut. Climatic Change 100(2): 267–294. Gell-Mann, Murray. 2003. Effective Complexity. In Nonextensive Entropy – Interdisciplinary Applications, ed. by Murray Gell-mann & Constantino Tsallis, 387–398. Oxford: Oxford University Press.

Complexity in Language: A Multifaceted Phenomenon

27

Gershenson, Carlos, ed. 2008. Complexity: 5 Questions. Denmark: Automatic Press. Gil, David. 2001. Creoles, Complexity, and Riau Indonesian. Linguistic Typology 5: 325–371. Gil, David. 2009. How Much Grammar Does it Take to Sail a Boat? In Sampson et al. (eds.), 19–33. Givón, T. 2009. The Genesis of Syntactic Complexity. Amsterdam: John Benjamins. Givón, T. & Masayoshi Shibatani, eds. 2009. Syntactic Complexity: Diachrony, Acquisition, Neuro-Cognition, Evolution. Amsterdam: John Benjamins. Hawkins, John A. 2005. Efficiency and Complexity in Grammars. Oxford: Oxford University Press. Hawkins, Hohn A. 2009. An Efficiency Theory of Complexity and Related Phenomena. In Sampson et al. (eds.), 252–268. Heine, Bernd & Kuteva, Tania. 2007. The Genesis of Grammar: A Reconstruction. Oxford: Oxford University Press. Hockett, Charles F. 1959. Animal “Languages” and Human Language. Human Biology 31: 32–39. Housen, Alex, Folkert Kuiken, & Ineke Vedder, eds. 2012. Dimensions of L2 Performance and Proficiency: Complexity, Accuracy and Fluency in SLA. Amsterdam: John Benjamins. Keller, Rudi. 1994. On Language Change: The Invisible Hand in Language. London, New York: Routledge. Kortmann, Bernd & Benedikt Szmrecsanyi, eds. 2012. Linguistic Complexity. Berlin: Mouton de Gruyter. Koster, Jan. 2009. Ceaseless, Unpredictable, Creativity. Biolinguistics 3: 61–92. Kretzchmar, William A, Jr. 2015. Language and Complex Systems. Cambridge: Cambridge University Press. Labov, William. 2001. Principles of Linguistic Change: Social Factors. Malden, MA: Blackwell. Larsen-Freeman, Diane. 2008. Complex Systems and Applied Linguistics. Oxford: Oxford University Press. Lee Namhee, Lisa Mikesell, Anna Dina L. Joaquin, Andrea W. Mates, & John H. Schumann. 2009. The Interactional Instinct. Oxford: Oxford University Press. Liebertman, Philip. 2012. The Unpredictable Species: What Makes Humans Unique. Princeton: Princeton University Press. Lupian, Gary & Rick Dale. 2010. Language Structure is Partly Determined by Social Structure. PLoS ONE 5(1): e8559. MacNeilage, Peter F. 1998. The Frame/Content Theory of Evolution of Speech Production. Behavioral and Brain Sciences 21(4): 499–511. Maddieson, Ian 2011. Phonological Complexity in Linguistic Patterning. In Proc. of the 17th International Congress of Phonetic Sciences, 17–21. Massip-Bonet, Angels & Albert Bastardas-Boada, eds. 2012. Complexity Perspectives on Language, Communication and Society. Heidelberg: Springer. McCawley, James D. The Syntactic Phenomena of English. Chicago: University of Chicago Press. McMahon, April. 1994. Understanding Language Change. Cambridge: Cambridge University Press. McWhorter, John H. 2001. The World’s Simplest Grammars are Creole Grammars. Linguistic Typology 5: 125–166.

28

Salikoko S. Mufwene, Christophe Coupé, and François Pellegrino

McWhorter, John H. 2012. Linguistic Simplicity and Complexity: Why do Languages Undress. Berlin: Mouton de Gruyter. Miestamo, Matti, Kaius Sinnemäki, & Fred Karlsson. 2008. Language Complexity: Typology, Contact, Change. Amsterdam: John Benjamins. Mitchell, Melanie. 2009. Complexity: A Guided Tour. Oxford: Oxford University Press. Mufwene, Salikoko S. 1996. Creolization and Grammaticization: What creolistics Could Contribute to Research on Grammaticization. In Changing Meanings, Changing Functions, ed. by Philip Baker and Anand Syea, 5–28. London: University of Westminster Press. Mufwene, Salikoko S. 2001. The Ecology of Language Evolution. Cambridge: Cambridge University Press. Mufwene, Salikoko S. 2008. Language Evolution: Contact, Competition and Change. London: Continuum Press. Mufwene, Salikoko S. 2009. Postscript: Restructuring, Hybridization, and Complexity in Language Evolution. In Aboh & Smith (eds.), 367–399. Mufwene, Salikoko S. 2012. The Emergence of Complexity in Language: An Evolutionary Perspective. In Massip-Bonet & Bastardas-Boada (eds.), 197–218. Mufwene, Salikoko S. 2013. Language as Technology: Some Questions that Evolutionary Linguistics Should Address. In In Search of Universal Grammar: From Norse to Zoque, ed. by Terje Lohndal, 327–358. Amsterdam: John Benjamins. Mufweme, Salikoko S. in press. The Evolution of Language as Technology: The Cultural Dimension, in Beyond the Meme, ed. by William Wimsatt & Alan Love. Minneapolis, MN: University of Minnesota Press. Newmeyer, Frederick J. & Laurel B. Preston, eds. 2014. Measuring Grammatical Complexity. Oxford: Oxford University Press. Newmeyer, Frederick J. & Laurel B. Preston. 2014. Introduction. In Newmeyer & Preston (eds.), 1–13. Palmer, Bill. 2007. Pointing at the Lagoon: Directional Terms in Oceanic Atoll-Based Languages. In Language Description, History and Development. Linguistic Indulgence in Memory of Terry Crowley, ed. by Jeff Siegel, John Lynch, & Diana Eades, 101–118. Amsterdam/Philadelphia: John Benjamins. Pellegrino, François, Christophe Coupé & Egidio Marsico. 2011. A Cross-Language Perspective on Speech Information Rate. Language 87(3): 539–558. Pellegrino, François, Egidio Marsico, Ioana Chitoran, & Christophe Coupé, eds. 2009. Approaches to Phonological Complexity. Berlin: Mouton de Gruyter. Poincaré, Henri. 1891. Sur le problème des trois corps et les équations de la dynamique. Bulletin Astronomique 8: 12–24. Risager, Karen. 2006. Language and Culture: Global Flows and Local Complexity. Clevdon: Multilingual Matters. Rissanen, Jorma. 1978. Modeling by Shortest Data Description. Automatica 14(5): 465– 471. Robinson, Peter. 2011. Second Language Task Complexity: Researching the Cognition Hypothesis of language learning and performance. Amsterdam: John Benjamins. Sampson, Geoffrey. 2009. A Linguistic Axiom Challenged. In Sampson et al. (eds.), 1–18. Sampson, Geoffrey, David Gil, & Peter Trudgill, eds. 2009. Language Complexity as an Evolving Variable. Oxford: Oxford University Press.

Complexity in Language: A Multifaceted Phenomenon

29

Smith, Adam. 1776. An Inquiry into the Nature and Causes of the Wealth of Nations. London: W. Strahan and T. Cadell. Sperber, Dan. 1994. Understanding Verbal Understanding. In What is Intelligence? ed. by Jean Khalfa, 179–198. Cambridge: Cambridge University Press. Sperber, Dan & Deidre Wilson. 2002. Pragmatics, Modularity and Mind-Reading. Mind and Language 17: 3–23. Steels, Luc. 1998. Synthesising the Origins of Language and Meaning Using Coevolution, Self-Organisation and Level Formation. In Approaches to the Evolution of Language: Social and Cognitive bases, ed. by James R. Hurford, Chris Knight, & Michael Studdert-Kennedy, 384–404. Cambridge: Cambridge University Press. Steels, Luc. 2011a. Modeling the Cultural Evolution of Language. Physics of Life Reviews 8: 339–56. Steels, Luc, ed. 2011b. Design Patterns in Fluid Construction Grammar. Amsterdam: John Benjamins. Steels, Luc, ed. 2012. Experiments in Cultural Language Evolution. Amsterdam: John Benjamins. Strogatz, Steven. 2003. Sync: The Emerging Science of Spontaneous Order. New York: Theia. Trudgill, Peter. 2009. Sociolinguistic Typology and Complexification. In Sampson et al. (eds.), 98–109. Trudgill, Peter. 2011. Sociolinguistic Typology: Social Determinants of Linguistic Complexity. Oxford: Oxford University Press. Victorri, Bernard. 1994. La Construction Dynamique du Sens. In Passion des Formes: Dynamique Qualitative, Sémiophysique et Intelligibilité, ed. by Michèle Porte, 733–747. Fontenay-Saint Cloud: ENS Editions. Walther, Géraldine & Benoît Sagot. 2011. Modélisation et Implémentation de Phénomènes Non-Canoniques. In Traitement Automatique des Langues, 52(2) (Vers la morphologie et au-delà), N. Hathout & F. Namer (eds.), p. 91–122. Wang, William Shi-Yuan, Jinyun Ke, & James W. Minett. 2004. Computational Studies of Language Evolution. In Computational Linguistics and Beyond, ed. by C. R. Huang, & W. Lenders, 65–106. Academia Sinica: Institute of Linguistics. Weinreich, Uriel, William Labov, & Marvin I. Herzog. 1968. Empirical Foundations for a Theory of Language Change. In Directions for Historical Linguistics: A Symposium, ed. by Winfred P. Lehman & Yakov Malkiel, 97–195. Austin: University of Texas Press.

2

How to Explain the Origins of Complexity in Language: A Case Study for Agreement Systems Luc Steels and Katrien Beuls

1

Introduction

It is useful to make a distinction between five different types of complexity in language: 1. Inventory complexity is about the number of conceptual, phonological, lexical and grammatical building blocks in use by an individual or by a particular language community. It pertains to the number of phonemes, the number of concepts (e.g., color categories, action categories, spatial relations, temporal relations, etc.), the number of lexical items (words and morphological elements), the number of grammatical categories (syntactic classes, cases, classifiers, grammatical functions and types of phrase structures), and the number of grammatical constructions. 2. Form complexity is measured in terms of statistics over the length of words and the length of utterances. 3. Processing complexity is defined as the cognitive effort involved in parsing and producing utterances: How much memory is needed? How many processing steps are required? How much combinatorial search is unavoidable? How much ambiguity is left before semantic interpretation? Processing complexity depends on the architecture of the language processing system, the nature and complexity of the grammar, the complexity of forms, and the ecological complexity of the environment in which language users operate. 4. Learning complexity refers to the amount of uncertainty that learners face when acquiring new words or constructions. Is it always possible to uniquely guess unknown meanings and functions, or does a search space get generated? How big is this search space? 5. Population-level complexity is concerned with the properties of the evolving communal language as a whole: How much variation is there in the population with respect to the usage or knowledge of linguistic forms? Or, conversely, what is the coherence, that is, the degree of sharing? What is the resilience of particular constructions in the process of cultural transmission? How intense is language change? 30

How to Explain the Origins of Complexity in Language

31

The central thesis of the present chapter is that human languages evolve in such a way as to minimize complexity at all these levels while providing enough expressive power to handle all the meanings relevant to the community. Minimizing complexity is necessary to keep the language viable; otherwise it would become too complex for regular usage and would no longer be learnable (Hawkins 2005). Minimizing complexity does not happen by intelligent design. Language processing is mostly subconscious and even linguists have no total explicit understanding of the systems underlying a language, let alone that they could rationally expand it. There is no central committee that designs or has total control over the use of English, Chinese or any other language. Speakers are aware of dialectal differences but in general they do not have the power to directly influence others. The most plausible alternative to intelligent design is a Darwinian selectionist process mapped to the cultural level (Mufwene 2001; Steels 2012b). Language users unavoidably generate variation, often because they do not know which alternatives already exist in the population. They also unavoidably generate complexity, for example because they do not know of a more compact way to express their message or they are unknowingly using a non-standard variant. But each language user balances this with strategies to minimize complexity, for example, by simplifying grammatical structure, leaving out some of the phonemes of a word, adopting the most frequently occurring variant, and so on. Variants for concepts, words and constructions compete and those that lead to higher communicative success and less cognitive effort have a higher chance of survival, simply because speakers and hearers want to understand each other and they do not want to spend more effort than is necessary. There is no optimal solution, and so languages keep moving on a linguistic fitness landscape (van Trijp 2014), sometimes optimizing one aspect (e.g., minimizing form complexity) while relaxing another one (e.g., processing complexity). Let us look in more detail at the many ways in which language users minimize the different types of complexity introduced earlier: 1. Damping inventory complexity. Language is an open system. New meanings constantly come up and need to become expressible. Routine expressions tend to lose their force and are then replaced by new ones. Consequently, there is a steady renewal in the inventory of a language: New words are invented or existing words are used for new meanings. New grammatical constructions are introduced as well, sometimes based on new paradigms or on the application of a paradigm to a new domain. Existing constructions may also be coerced into new uses or words may be reanalyzed to fit with existing constructions. At the same time, inventories must be kept within bounds because larger inventories take up more linguistic memory, increase access time, and require longer time for learning. How are the inventories kept in check? Many words

32

Luc Steels and Katrien Beuls

or constructions simply get out of fashion after a while. They disappear from usage and are forgotten, particularly after a few generations. There is also a general tendency to reuse existing forms as much as possible even if they have already existing functions, a process similar to exaptation in biology. If a new invention is based on the exaptation of an existing word or construction in a slightly different context, then there is a higher chance that the hearer might guess this new meaning than if a radically new invention is made. Hence the exapted invention has a higher chance to propagate and survive in the communal language and it contributes to keeping the language inventory in check. 2. Damping form complexity. Novel words tend to be pronounced in full, but once introduced, their production gets eroded over time due to articulatory optimization or errors, reducing both the length of words and the length of utterances in which they appear. Some words end up being morphemes, then clitics and later affixes. A similar form of optimization process happens at the level of constructions. The first time new meanings are expressed it is usually done in an elaborate, circumscriptive way. But with routine usage, the combination of constructions that was used to build the more elaborate phrase gets collapsed into a single construction to achieve faster processing, a process usually called chunking. Words within a chunk may then start to disappear and the phrase may progressively become idiomatic. For instance, the English collocation by name as in I know all of my students by name is a shorter version of the full noun phrase by the/their name. 3. Damping processing complexity. Strategies for reducing inventory size and form complexity may lead to syncretism (the same form gets multiple meanings) and hence syntactic ambiguity. Syntactic ambiguity leads in turn to combinatorial search, which needs to be kept in check because it causes an exponential increase in memory and processing time and hence fluent rapid speaking and listening becomes more and more difficult. Inventory reduction by reuse of existing forms may also lead to semantic ambiguity because the forms in the utterance no longer unequivocally communicate the intended meaning. Processing complexity can be dampened by adding additional syntactic structure. For example, as discussed later in this chapter, grammatical agreement is ways in which language users signal which words belong together in the same sentence and hence it reduces the combinatorial complexity of parsing. More generally, we argue that the reduction of cognitive effort in parsing and interpretation is one of the main motivating forces for the origins of syntactic structure. 4. Damping learning complexity. Language learners need to guess the meaning and function of unknown words or constructions. The more possibilities there are, the more hypotheses the learner has to consider. The learning task

How to Explain the Origins of Complexity in Language

33

gets more manageable if the learner can make strong use of context, which reduces the set of possible meanings, and if the speaker, acting as tutor, helps by scaffolding the complexity of utterances and by correcting wrong uses of words and constructions. The structure of language itself can also help. For example, syntactic structure may not only decrease the complexity of parsing but also help to constrain the possible meanings of an unknown word, leading to syntactic bootstrapping. For example, the adjectives in a nominal phrase are typically ordered based on semantic classes and hence this ordering can help constrain the meaning of an unknown adjective, thus reducing learning complexity. 5. Damping population-level complexity. Individuals in a population will never share the same inventory because language learners have to independently acquire the language based on their own history of interactions with others and their own needs, and each language user has in principle the right to extend the language. Moreover, language is a social tool that is used by speakers to stress their identity as belonging to a certain community or to prevent themselves from adopting ongoing changes (Labov 1966). Hence we see tremendous variation. On the other hand, language variation is detrimental to the efficiency of a language because language users need to be able to parse and interpret expressions that deviate from their own usage and possibly even store these variations. Instead of learning a single language they have to learn a multitude of overlapping languages. So we need a strong force that dampens populationlevel complexity and ensures that idiolects become shared into a common language. We will see later that appropriate learning strategies (such as a lateral inhibition learning rule) can act as such a force. 2

Agent-Based Modeling

The far from exhaustive list of suggestions provided in the previous section certainly sounds plausible, but how can we make them more concrete and test them with scientific methods? There are several ways to study language dynamics: linguistic analysis deconstructs the kinds of rules that must intervene in the parsing and production of utterances, studies in historical linguistics show how a language has been shaped and reshaped over time by its users, research into language typology and sociolinguistics maps the diversity and variation in language, psychological experiments measure delays in response to utterances or difficulties of learning, and neuroimaging experiments measure brain activation while parsing, producing or learning utterances. All of these approaches are valuable. There is also another approach, which is gaining momentum since the mid-1990s, namely agent-based modeling (Steels 1995; Smith et al. 2003; Loreto et al. 2011; Steels 2011).

34

Luc Steels and Katrien Beuls

Agent-based approaches model a population of linguistic agents engaging in situated interactions, either embodied or simulated, called language games, and test the behavior of the model through computer simulations or robotic experiments. An agent-based model defines a set of possible situations and an interaction pattern between agents that contains both verbal and non-verbal aspects (such as pointing). It also contains a set of mechanisms that the agents in a population can use for producing and understanding utterances, such as categorization, pattern matching, associative retrieval, and so on. The agents are also endowed with a Language Acquisition Device in the form of strategies they can use to build up (i.e., learn or extend) their language systems. Then a series of games is being played between randomly chosen agents from the population. Each agent can play both the roles of speaker and listener. As a side effect of a game, the participating agents may expand their conceptual or linguistic inventory, or adapt it, based on the outcome of the game. We can track different measures, both for individual agents and for populations of agents, in order to tell us whether a model works, in the sense of whether the desired linguistic structures indeed arise in model simulations. If they do not, the mechanisms and strategies initially put into the agents are insufficient and they are changed for the next experiment until we know what causal mechanisms generate the phenomena we want to explain. For example, experiments trying to explain how a set of color terms and color concepts could emerge, would involve agents that are initially endowed with concept formation mechanisms and strategies for inventing, acquiring or aligning associations between words and color concepts (Steels & Belpaeme 2005). At the start of the experiment there is neither a shared color language nor any color concept inventory, but if we have been able to come up with the necessary and sufficient mechanisms and strategies we expect to see after a number of games that a shared color language emerges (which indeed happens in these computer simulations). If that is the case, we have evidence that the given mechanisms and strategies (plus the interaction patterns and ecological conditions) explain the emergence of color terms and of the color concepts they express. An agent-based model typically focuses on one aspect of language; for example, color terms, names for actions, tense-aspect systems, quantifiers, expression of information structure, phrase structure, case grammar, or grammatical function. (See Steels 2012a for a representative sample.) Each time a particular language game is chosen that brings out the aspect of language being studied. For instance, if we want to study the emergence and evolution of color terms we obviously need an environment with objects of different colors and the color should matter in the interaction between the agents. Agent-based models have numerous advantages compared to verbal theorizing; it is therefore surprising that this methodology is still so controversial, even

How to Explain the Origins of Complexity in Language

35

though it is widely accepted in sociology, biology, economics and many other scientific disciplines (de Landa 2011): r Agent-based models require a comprehensive mechanistic model of language processing, which is usually lacking in linguistic or psychological research. Most often grammars are not formally represented, and if they are, the formalization is often not in a form that can be used by processing algorithms. Various learning strategies are assumed but they are only described vaguely and not operationalized, so that we cannot know whether they are effective. Agent-based models can only be built with effective computational models of language processing and because these models need to cope with emerging and evolving languages, they challenge several dogmas in formal and computational linguistics (Steels 2012b). In that sense, agent-based models push not only the state of the art in language evolution but also in computational linguistics and linguistic theory in general. r When doing simulations using agent-based models it is possible to measure precisely the complexity of the languages generated with a particular strategy at each of the different levels of complexity described earlier. We can do repeatable experiments and examine the statistical distribution of the results. It is possible to selectively add or take out mechanisms or change a strategy so that we can determine the causal relationships between mechanisms, strategies or parameters of the model, and the structure of an emergent language system. These methods are standard practice in science and there is no reason why the study of language evolution should not use them. Agent-based models are occasionally criticized because the strategies and processing mechanisms are stated at an abstract, mathematical/computational level, but abstraction is a characteristic of every scientific model. They are also criticized because such models are not validated by psychological observations or neuroimaging data. However, we are still far removed from effective models of language processing and even if we would have them, we would not be able to validate them, given the precision of current neuroimaging instruments. Instead, agent-based models should be seen by psychologists and neuroscientists as theoretical hypotheses that they can test using their observational techniques. Indeed, some cognitive psychologists have already been replicating similar conditions as used in agent-based models with results and phenomena comparable to what we see in agent-based simulations (Galantucci 2005; Selten & Warglien 2007; Pickering & Garrod 2013). The observations made by historical linguists remain the ultimate empirical target of agent-based models of cultural language evolution. Historical linguists have identified many of the processes that are active in language emergence and change, such as lexical category shifts, phonological erosion of recruited word forms, damping of synonymy, and so on (Heine & Kuteva 2007). Similar processes have been identified in the formation of creoles (Mufwene 2008). Agent-based models explicate

36

Luc Steels and Katrien Beuls

the mechanisms and strategies behind these proposals and test their explanatory power. Of course we cannot directly replicate precisely the historical evolution of Old English into Middle English because there are so many contingencies, but, just like in evolutionary biology, we should be able to demonstrate what strategies and cognitive mechanisms have been playing a role. A good example is agent-based models about the changes to the German article system from the cleaner old German system to contemporary German (van Trijp 2013). 3

Case Study for Agreement Systems

The best way to illustrate both the theory of language evolution by linguistic selection and the methodology of agent-based modeling is to look at concrete examples. We will do this now for grammatical agreement, building further on agent-based models reported earlier (Beuls & Steels 2013). Grammatical agreement occurs when two linguistic units (typically a word or a phrase) receive a marker that indicates their relatedness (Lehmann 1988). For instance, in the French utterance une belle fille – “a pretty girl” – both the article and the adjective have a feminine form, marked by the final vowel -e, which is governed by gender of the head noun fille, which is specified as feminine. Agreement systems can be very complex and messy and many linguists (and second language learners) have been puzzled as to why they are there. We now look at the different types of complexity discussed earlier and study the origins of agreement systems from that point of view. 3.1

The Language Game

In the current experiment, agents play a game of reference (known as the Naming Game), which has been used in many earlier investigations of language dynamics (e.g., Steels 1995; Dall’Asta et al. 2006). All agents can play the role of speaker or hearer and have an equal chance of playing a game. The agent chosen as speaker goes through the following steps: (i) The speaker selects a subset of the objects in the current situation to act as the topic of the utterance. (ii) The speaker looks for a distinctive combination of properties for each of the objects in the topic, where a distinctive combination is a set of properties that are true for the object but not for any other object in the current situation. (iii) The speaker retrieves the minimal set of words in the vocabulary that cover the chosen properties, which implies that words with the largest coverage are preferred. The speaker then utters these words. Although there is unavoidably a sequential ordering to the words, this does not carry any meaning, that is, agents use a word-order free language. Next, the agent chosen as hearer goes through the following steps: (i) The hearer looks up the words in the vocabulary and thus reconstructs what

How to Explain the Origins of Complexity in Language

37

properties have been communicated by the speaker. (ii) The hearer identifies which objects in the present situation satisfy these properties and points to them. The game is a success if the objects pointed to by the hearer are those initially chosen by the speaker. The game fails if this is not the case or if the utterance remains semantically ambiguous, that is, if there is more than one possible interpretation that fits with the current situation model. 3.2

Processing Complexity

Although this language game looks deceptively simple, there is the potential for a combinatorial explosion and semantic ambiguity. The hearer does not know how many objects the speaker is talking about and the utterance does not communicate which words are about the same object. Hence, all possible combinations must be tried by the hearer to find those that fit with the current situation. This obviously implies a very high degree of processing complexity. Processing complexity can be measured in different ways and it always depends on the architecture of the language processing system that one has implemented. It could be the nodes in the search tree during production or parsing of a particular sentence, the number of constructions needed to build or understand an utterance, or the accessibility of the constructions that are used to parse or produce the utterance, for example, frequently used constructions can be made more accessible (i.e., primed). All of these measures tell us something about the cognitive effort of the speaker or the hearer. In the case of agreement the most relevant complexity measure is the number of hypotheses that remain after parsing a particular utterance. Suppose there is neither word order nor any other kind of grammar. Then the utterance brown blue cat chair big, could have meant: r (cat) (big blue brown chair); r (brown cat) (big blue chair); r (big cat) (blue brown chair); r (blue cat) (big brown chair); r (big brown cat) (blue chair); r (big blue cat) (brown chair); or r (big blue brown cat) (chair) and so on. And this still assumes that there are only two objects. If an utterance can be about more than one object, then the set of possible combinations is still much larger. More generally, the number of possibilities Bn is equal to the number of partitions of the set D of words in an utterance of size n. Bn is known as the Bell number and can be defined using the following equation (Bell 1938): n    n Bk Bn+1 = k k=0

38

Luc Steels and Katrien Beuls

with B0 = 0 and B1 = 1. Bn grows exponentially with the number of words. It means that the sentence you are now reading (which contains 20 words) generates 51,724,158,235,372 partitions and hence possible interpretations. This definition is only true if all words in an utterance are semantically equivalent (i.e., can be grouped together in every possible way), which is not the case in real language use where ontological selection criteria help to determine which words should be grouped together. Yet, selection restrictions reduce the number of solutions but this number remains exponential as the length of the utterance increases (Garcia Casademont & Steels 2014). However, now suppose that the agents adopt a strategy to use agreement markers, that is, morphemes are added to every word that refers to the same object. An example utterance could now be brown-ki blue-ba cat-ki chair-ba big-ba. It is now directly clear which words belong together and the combinatorial explosion disappears. Agreement markers are therefore an efficient means to dampen processing complexity as all the combinations that are evoked by the Bell number are hereby reduced to a single one. We can demonstrate this with an agent-based simulation, in which a population of agents starts without an agreement system but each agent is endowed with a strategy to invent new markers when needed (as speaker) or to adopt them (as hearer). The markers are in this case purely formal (i.e., they don’t express number, gender, case or any other semantic feature). We see in Figure 2.1 that the number of hypotheses gets drastically reduced to a single one as soon as the agents share a sufficient set of markers. Clearly an agreement strategy leads to a language that requires less cognitive effort in parsing and interpreting utterances; hence this explains, from a functional point of view, why we find agreement systems in human languages. 3.3

Population-Level Complexity: Language Variation

When speakers invent new markers, there is unavoidable variation, because individuals are not supposed to have a telepathic general overview of all interactions happening throughout the population. Agents end up with a working system, but there are many more markers than is absolutely necessary and hence agents have to learn many more markers. This in turn slows down the emergence of the agreement system and makes it harder for new agents coming into the population to acquire the existing set of markers. This unnecessary language variation can be damped when agents use lateral inhibition as part of the marker strategy discussed in the previous paragraph (de Vylder & Tuyls 2006; Steels 1998). Agents maintain a score σm between 0.0 and 1.0 for every marker m. The initial score of a new marker is σ = 0.5. The hearer (but not the speaker) increments this score whenever a marker mi

How to Explain the Origins of Complexity in Language

39

Figure 2.1 Results of an agent-based simulation, with a population of 10 agents playing 200 language games in total, which means an average of 40 per agent. The x-axis shows the number of games played; the left y-axis the number of hypotheses left after parsing and the right y-axis the number of markers. When agents use the agreement marker strategy, the number of hypotheses decreases rapidly thus damping drastically the processing complexity. (Error bars indicate standard deviation of 10 runs.)

appears in an utterance and decreases the score of all other non-used markers mj using the following equations with alignment rate γ = 0.2: σm ← σm (1 − γ ) + γ i

i

σm ← σm (1 − γ ) j

j

When choosing which marker to use, the speaker should prefer the marker with the highest score that was not yet used in the same utterance. This establishes a positive feedback loop between usage and marker preference, which leads to a shared minimal marker system. The resulting process is similar to many other self-organizing processes found in natural systems in which large-scale structures arise from local interactions through random behavior influenced by positive feedback loops (Camazine et al. 2001). The effect of using lateral inhibition is illustrated in Figure 2.2. When agents simply invent markers or acquire them from others, the number of markers becomes stable with an inventory of 10 markers. When they use lateral inhibition as part of the agreement strategy, they end up with the optimal number of three markers, because utterances in this experimental run have been limited

40

Luc Steels and Katrien Beuls

Figure 2.2 Comparison of a population that uses lateral inhibition and one that does not. The use of lateral inhibition gives a selectionist advantage because it leads to a smaller marker inventory at the population level, that is, less unneeded variation.

to contain maximally three objects. When there are more than three objects, the population would settle on more markers but always the minimal number needed. This simulation demonstrates again that the adoption of a particular strategy can lead to a more efficient language system in the sense that it dampens complexity along a particular dimension. In this case, lateral inhibition dampens variation in the population, so that each agent needs to keep fewer markers in memory and new agents need to learn fewer markers. 3.4

Learning Complexity

So far the strategies that we investigated use only formal markers, that is, symbols without any meaning. A formal marker such as –ki in the previous example purely functions as a grouping label. It indicates that words carrying this marker refer to the same object in the world. Formal markers are used in some languages, including sign language. One region in the space of the signer (and it can be any region) is temporarily designated to stand for an object and other signs that later refer to the same object are then produced in the same region (Aranoff, Sandler et al. 2004). However, spoken human languages only rarely use formal markers. For example, the Swahili marker ki- used as an agreement marker in the phrase ki-kapu ki-kubwa ki-moja (ki.sg-basket ki-large

How to Explain the Origins of Complexity in Language

41

ki-one) (Welmers 1973) is used when the referent belongs to the class of inanimate objects, which contains artifacts such as baskets. The Latin marker -arum expresses plural, feminine and genitive. Meaningful markers lead unavoidably to a larger inventory of markers, because a marker has to fit with the object being referred to. For example, if gender is used as a meaning distinction, then we need at least two markers (one for feminine and the other for masculine – and possibly a third one for neuter). And if both objects being referred to are masculine we need markers along another dimension, for example for number (with two markers for singular and plural). On the other hand, a marker can introduce additional meanings on top of the meaning supplied by words. This is the case with the Latin -arum, which not only carries out the agreement function but also conveys that the referent is plural and genitive. Hence meaningful markers make it possible to express more meaning with fewer linguistic forms, which is a way to dampen form complexity. Studies in grammaticalization have shown that in the historical record, agreement systems initially arise by reusing existing words as markers (Fuss 2005; Givón 1976). Why would that be? We argue that this dampens learning complexity. If an arbitrary label is used as a meaningful marker, then the meaning of this marker needs to be guessed and agreed upon, and there is a higher chance that different agents introduce different markers. But if an existing word is used, then the hearer can make a much better guess of the possible meaning of the marker and hence the search space for learning the marker is much smaller. The learner only has to become aware that the word is now used not only lexically but also grammatically. We have conducted another experiment in the context of agreement to explore this phenomenon. Learning efficiency is measured as follows. Let the inventory size Ig be the total number of distinct marker constructions invented by the whole population up till game g and Ug be the number of markers being in used for the same window, then the learning efficiency Sg= Ug/Ig captures how well superfluous inventions could be avoided within this interval. This gives an indication of the learning efficiency because the fewer constructions learners add to their inventory (as a possible hypothesis of the shared language), the lower Sg will be. Figure 2.3 compares two strategies: one in which new markers are just random symbols (the no-reuse strategy) and a second strategy where new markers are based on existing words (the reuse strategy). We see that the learning efficiency is considerably different: only 20 percent of the constructions are still circulating with the reuse strategy, whereas 60 percent of the constructions survive in the case of no-reuse (Figure 2.3). This is another example where a particular strategy allows agents to dampen complexity, in this case along the learning dimension.

Luc Steels and Katrien Beuls

learning efficiency

42

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

500

1000

1500

2000

2500

games no reuse

reuse

Figure 2.3 Experiment comparing the learning efficiency Sg between a strategy that reuses existing words as meaningful markers and one that does not. We see that the first strategy is much more efficient in the sense that learners generate fewer hypotheses about the possible meanings of markers that are then later not used.

3.5

Form Complexity

Another phenomenon that is observed quite often in language evolution is that the form of a word gets shortened and thus its form complexity reduced. This has also been observed with agreement markers, which are often shortened so that the original word is no longer recognizable. We have therefore run additional experiments in which agents use this strategy, in addition to the strategies discussed earlier (see also Beuls 2014 for a more elaborate discussion). Speakers optimize articulation by leaving out the last consonant or vowel of a marker with a certain probability ε = 0.1. Hearers are flexible enough in their parsing of markers to recognize that a truncated form is a variant of an existing marker, as long as it deviates for only one consonant or vowel. This maintains an adequate level of communicative success and does not diminish the effectiveness of markers to cut down combinatorial complexity and semantic ambiguity. But how can we explain that a variant might itself become the norm and in turn become the object of further optimizations? Figure 2.4 shows the outcome of a computer simulation of this phonological reduction strategy. An agreement system based on meaningful markers is first emerging using the meaningful marker strategy. But after agents reach a stable level of performance (in the experiment this is typically after 200 games per agent), they occasionally introduce phonological reductions with probability ε = 0.1 and this leads to the erosion of the original markers. Figure 2.4(a) shows that the average marker length decreases from an average of seven to two

How to Explain the Origins of Complexity in Language

43

7

marker length

6 5 4 3 2 1 0

0

5000 10000 15000 20000 25000 30000 35000 40000 45000 50000 games marker length (a)

1

0.8

0.6

0.4

0.2

0

0

10000 -naeamo-(v-1-2) -naeam-(v-1-2)

20000

30000

-naea-(v-1-2) -nae-(v-1-2)

40000

50000

-na-(v-1-2) -n-(v-1-2)

(b)

Figure 2.4 (a) Overall reduction of form complexity in a population of agents over 50,000 games. (b) The -naeamo marker (expressing feature v-2–1) starts off with six letters and erodes down to a single letter “-n.” Longer versions of the original marker (“-naea” and “-nae”) are still occasionally used.

consonants and vowels, without affecting performance. There is greater variation Vg in the population because there are always different variants of the same marker in use, but this generally does not have an impact, because agents are able to recognize them as such, that is, variants within their own norm.

44

Luc Steels and Katrien Beuls

3.6

Inventory Complexity

Agents need a bigger inventory in the case of meaningful markers because a marker can only be used for certain types of referents. The question now is: how can this inventory be further reduced? One possibility is to ‘bleach’ the semantic features of a marker. For example, it is widely attested that the markers for gender (like masculine/feminine) are initially entirely biologically motivated (noun phrases referring to females get feminine and to males masculine) but once this is adopted, genders must be assigned by convention to non-animate beings as well if one wants to avoid additional markers. For example, table in French is assigned to be feminine. The assignment is conventional because another language might use another gender. Indeed, in German, Tish ‘table’ is masculine. We have constructed an agent-based model of this as well, based on the process of coercion: when a marker cannot be used with a given word group because the meaning of the marker is incompatible with the referent of the word group, then the controller1 of the word group can be coerced to become compatible by assigning it the features of the marker. Agents use again a lateral inhibition dynamics to settle on which conventionalization to adopt. Figure 2.5(b) shows the winner-take-all dynamics for a single controller. This strategy is discussed in more detail in Beuls and Steels (2013). Figure 2.5(a) shows the results of our simulations. If agents are allowed to coerce controllers to be compatible with the agreement markers they already have, then the inventory size is reduced by 73 percent, which means that fewer marker constructions need to be consulted in parsing and production and a smaller inventory needs to be learned. 4

Conclusion

The theory of cultural language evolution through linguistic selection (Mufwene 2001, 2008; Steels 2012a) argues that language users are able to come up with different strategies for building their language systems but that they will tend to prefer those strategies that help them construct a better language system. ‘Better’ means first of all that it satisfies their communicative needs, that is, that there is enough expressive power to convey the meanings that need to be expressed. Second, it means that language complexity is damped as much as possible along the different dimensions discussed here: inventory complexity, form complexity, processing complexity, learning complexity and population-level complexity (i.e., language variation). Damping complexity is needed to keep the language viable and culturally transmittable. 1

The controller is the element that determines the agreement, most often the head noun of the noun phrase or the subject noun phrase in subject-verb agreement (Corbett 2006).

How to Explain the Origins of Complexity in Language

45

(a)

(b) Figure 2.5 A reduction in inventory complexity from 15 to 4 markers (a) is possible thanks to the feature coercion in controllers that are marked conventionally for one of the four possible feature values (b).

This chapter did not discuss how new strategies get formed or how agents choose themselves between different strategies based on assessment criteria that they potentially compute themselves. Steps in this direction are shown in other research (e.g., Bleys & Steels 2001; van Trijp 2013). Instead, we focused on giving concrete examples of agent-based models to demonstrate that certain universal features occur in human languages because they help to dampen

46

Luc Steels and Katrien Beuls

complexity along various dimensions. In particular, we developed explanations for the use of agreement markers, the use of meaningful as opposed to formal markers, the recruitment of existing words as markers, the erosion of markerforms and the conventionalization of markers. 5

Acknowledgments

The authors acknowledge the following financial support: Luc Steels received an ICREA fellowship to work at the Institut de Biologia Evolutiva (UPF-CSIC) and a Marie Curie Integration grant to finance collaboration with the VUB AI Lab. Katrien Beuls received a grant from the IWT (Belgium) for her Ph.D. research at the VUB AI Lab in Brussels. REFERENCES Aronoff, Mark; Irit Meir; Carol Padden; and Wendy Sandler. 2004. Morphological universals and the sign language type. Yearbook of Morphology, ed. by Geert Booij and Jaap Van Maerle, 19–39. Dordrecht: Kluwer. Bell, Eric Temple. 1938. The iterated exponential integers. The Annals of Mathematics. 39, 539–557. Beuls, Katrien, and Luc Steels. 2013. Agent-based models of strategies for the emergence and evolution of grammatical agreement. PLoS ONE, 8(3), e58960, 03 2013. doi: 10.1371/journal.pone.0058960. Beuls, Katrien. Spirals in language evolution. Proceedings of the Tenth International Conference on the Evolution of Language, Vienna, 389–390. Bleys, Joris and Luc Steels. 2001. Linguistic selection of language strategies. Advances in AI – LNCS, 5778, 150–157. Corbett, Greville. 2006. Agreement. Cambridge: Cambridge University Press. Camazine, Scott; Jean-Louis Deneubourg; N. Franks; J. Sneyd; Eric Bonabeau; and Guy Theraulaz. 2001. Self-Organization in Biological Systems. Princeton: Princeton University Press. Dall’asta, Luca; Andrea Baronchelli; Alain Barrat; and Vittorio Loreto. 2006. Nonequilibrium dynamics of language games on complex networks. Physical Review E, 74(3), 036105. Delanda, Manuel. 2011. Philosophy and simulation. The emergence of synthetic reasoning. New York: Continuum. De Vylder, Bart; and Karl Tuyls. 2006. How to reach linguistic consensus: A proof of convergence for the naming game. Journal of Theoretical Biology, 242(4), 818– 831. Fuss, Eric. 2005. The Rise of Agreement: A Formal Approach to the Syntax and Grammaticalization of Verbal Inflection. Amsterdam: John Benjamins. Galantucci, Bruno. 2005. An experimental study of the emergence of human communication systems. Cognitive Science, 29(5), 737–767. Garcia Casademont, Emília; and Luc Steels. 2014. Strategies for the emergence of firstorder constituent structure. Proceedings of the Tenth International Conference on the Evolution of Language, Vienna, 50–57. Givón, Talmy. 1976. Topic, pronoun and grammatical agreement. Subject and Topic, ed. by Charles N. Li, 149–188. New York: Academic Press.

How to Explain the Origins of Complexity in Language

47

Hawkins, John A. 2005. Efficiency and Complexity in Grammars. Oxford: Oxford University Press. Heine, Bernd; and Tanja Kuteva. 2007. The Genesis of Grammar. Oxford / New York: Oxford University Press. Labov, William. 1966. 2006, Second Edition. The Social Stratification of English in New York City. Cambridge: Cambridge University Press. Lehmann, Christian. 1988. On the function of agreement. Agreement in Natural Language: approaches, theories, descriptions, ed. by Michael Barlow, and Charles Albert Ferguson, 55–65. Stanford: Center for the Study of Language and Information, Stanford University. Loreto, Vittorio; Andrea Baronchelli; Animesh Mukherjee; Andrea Puglisi; and Francesca Tria. 2011. Statistical physics of language dynamics. Journal of Statistical Mechanics, 4, P04006. Mufwene, Salikoko. 2008. Language Evolution: Contact, Competition and Change. New York: Continuum. Mufwene, Salikoko. 2001. Competition and selection in language evolution. Selection, 3(1), 45–56. Pickering, Martin and Simon Garrod. 2013. An integrated theory of language comprehension and production. Behavioral and Brain Sciences 36, 329–392. Selten, Reinhard and Massimo Warglien. 2007. The emergence of simple languages in an experimental coordination game. Proc. Natl Acad. Sci. USA, 104(18), 7361– 7366. Smith, Kenny; Simon Kirby and Henry Brighton. 2003. Iterated learning: a framework for the emergence of language. Artificial Life, 9(4), 371–386. Steels, Luc. 1995. A self-organizing spatial vocabulary. Artificial Life, 2(3), 319–332. Steels, Luc. 1998. The origins of ontologies and communication conventions in multiagent systems. Journal of Agents and Multi-Agent Systems, 1(2), 169–194. Steels, Luc. 2011. Modeling the cultural evolution of language. Physics of Life Reviews, 8(4), 330–356. Steels, Luc. 2012a. Experiments in Cultural Language Evolution. Amsterdam: John Benjamins. Steels, Luc. 2012b. Self-organization and selection in language evolution. Experiments in Cultural Language Evolution, ed. by Luc Steels, 1–37. Amsterdam: John Benjamins. Steels, Luc and Tony Belpaeme. 2005. Coordinating perceptually grounded categories through language. a case study for colour. Behav. Brain Sci., 24, 469–529. Van Trijp, Remi. 2013. Linguistic assessment criteria for explaining language change: A case study on syncretism in German definite articles. Language Dynamics and Change, 3(1), 105–132. Van Trijp, Remi. 2014. Fitness landscapes in cultural language evolution: A case study on German definite articles. Proceedings of the Tenth International Conference on the Evolution of Language, Vienna, 334–341. Welmers, Wim E. 1973. African Language Structures. Berkeley: University of California Press.

3

Complexity in Speech: Teasing Apart Culture and Cognition Bart de Boer

1

Introduction

Languages are more complex than is strictly necessary for their communicative function. For example, the smallest repertoires of speech sounds found in language have about a dozen different contrasting phonemes (see Table 3.1) whereas in the sample used for the UCLA Phonological Segment Inventory Database (UPSID451 , Maddieson, 1984; Maddieson & Precoda, 1990) the median number of speech sounds is 29 (with a first quartile of 23 and a third quartile of 36). This indicates that in general, languages tend to use more than twice the number of phonemes that appears to be minimally necessary for full language. This makes it possible to define a notion of relative complexity: one language is more complex than another if a language user needs to learn more in order to use the one language than the other.1 . The question then arises of why languages differ in relative complexity and how they become more complex. This chapter focuses on the second of these questions (as it is felt that it is more susceptible to empirical investigation) and more specifically on how one determines which aspects of linguistic complexity are due to cultural processes, and which aspects are due to cognitive biases. Cultural processes that shape language are the way language is used for communication and the way it is transferred from generation to generation or from group to group. Functional pressures, such as constraints on production and perception or information theoretical issues, may play a role here as well. Some linguistic structures may be easier to understand or to produce and will therefore be preserved and transmitted better. Proponents of such a usage-based theory of language (Goldberg, 2003; Tomasello, 2003) explain linguistic structure from these cultural processes, while the underlying cognitive mechanisms are generally considered less important. In other words, even if the cognitive mechanisms would be rather different, still the same universals would be found.

1

It should be noted that in this definition complexity of languages can only be compared for a given aspect of the language, such as the phoneme inventory size in the example.

48

Complexity in Speech: Teasing Apart Culture and Cognition

49

Table 3.1 Languages from UPSID451 (Maddieson, 1984; Maddieson & Precoda, 1990) with inventory sizes smaller than or equal to 16 phonemes. Although there appears to be an important influence of area or language family, nevertheless, languages with small phoneme inventories are spoken in different, linguistically unrelated regions of the world Phonemes

UPSID Name

ISO 639–3

UPSID Family

11 11 13 13 14 14 15 15 16 16 16 16 16 16

Pirahã Rotokas Hawai’ian Nasioi Roro Taoripi Gadsup Ekari Ainu Gugu-Yalanji Bandjalang Yidiny Dyirbal Koiari

Myp Roo Haw Nas Rro Tqo Gaj Ekg Ain Gvn Bdy Yii Dbl Kbk

South American, Paezan East Papuan Polynesian East Papuan Malayo-Polynesian Trans-New Guinea Trans-New Guinea Trans-New Guinea Isolate Pama-Nyungan Pama-Nyungan Pama-Nyungan Pama-Nyungan Trans-New Guinea

Other linguists do explain properties of language from language-specific cognitive constraints. In their view, the way language is used plays a secondary role, and the structure of language is mainly due to properties of the brains of language users. An example of a model that explains structure of language from very language-specific cognitive biases is the principles and parameters model (Chomsky, 1981; Baker, 2002, 2003). In this chapter, we will not argue for or against these positions, but rather address the issue of how one can investigate empirically what aspects of language may be due to cultural processes and what aspects of language may be due to cognitive biases. This in turn may help answer the question of how languages become complex: is it the result of cultural and functional processes, or do we have cognitive biases that tend to make language more complex? The chapter addresses this issue from an evolutionary perspective. It adopts Dobzhansky’s (1973) point of view that “nothing in biology makes sense, except in the light of evolution.” What this means is that if one is interested in the ultimate causes of biological facts, one has to investigate how and why they have evolved. An evolutionary perspective has always been implicit in the study of language from a cognitive point of view: the properties of the human brain that linguists tend to focus on are the ones that are different from other animals,

50

Bart de Boer

and that are considered essential for language. This is an implicit way of investigating which cognitive adaptations have evolved for language. Over the last twenty years or so, the evolutionary perspective on language has become increasingly explicit and influential (Pinker & Bloom, 1990; Hauser et al., 2002; Fisher & Marcus, 2005; Fitch, 2005; Mufwene, 2008; Fitch, 2010). The evolutionary perspective is doubly important for language: in order to understand potential biological adaptations for language, one has to study why and how they evolved biologically. In addition, if one wants to understand the properties of a particular language, one has to understand their history, that is, how they evolved culturally. Cultural evolution is thus responsible for language change and usage-based emergence of structure, while biological evolution is responsible for the cognitive constraints that help humans learn language. The evolutionary perspective sees cultural processes and cognitive biases as two sides of the same coin, rather than as mutually exclusive positions. However, when observing modern language, one only sees the final product of the interaction between culture and cognition. It is therefore difficult to tell which factor is responsible for observed linguistic phenomena. This chapter therefore introduces a recently emerging experimental paradigm – experimental cultural learning (Kirby et al., 2008; Galantucci, 2009; Scott-Phillips & Kirby, 2010). Experimental cultural learning attempts to re-create cultural evolution in a laboratory setting, and in this way to tease apart the effects of cultural processes and the effects of cognitive biases. It will be illustrated how this can be applied to the question whether complexity in phonemic systems is created by cognitive biases (feature economy) or rather by cultural processes (self-organization for distinctiveness). In the next section, the recent debate about the role of biologically evolved cognitive constraints versus culturally evolved properties of human language is described in more detail. This puts the experiments that are described in the section after that in perspective. Finally we discuss how these experiments can be refined and how they may be applied to other aspects of language. 2

The Controversy Between Biology and Culture

Every linguist is aware that there is both a cultural and a biological component to language. Because there are so many different languages, and because children who are not exposed to a language do not spontaneously develop a language on their own,2 it is clear that languages are the products of culture. On the other hand, because only humans (and not, chimpanzees, elephants or dolphins, to name a few intelligent animals) learn language, it is clear that one 2

Interestingly, children not exposed to a language do appear to develop a new language if they grow up in a group, as illustrated by the case of Nicaraguan Sign Language (Polich, 2005).

Complexity in Speech: Teasing Apart Culture and Cognition

51

needs a human brain to learn language. As in many areas of the study of human behavior, there is a debate about the role of nature (biology) versus nurture (culture) in linguistics. However, because the previous observations incontestably show that both culture and biology are important this debate is about the relative importance of both. There are two independent questions in this debate: how specific to language are the cognitive mechanisms that are used in language and how much of the complexity and diversity of language is due to cultural factors versus biological factors. These questions are independent, because whether cognitive mechanisms are specific to language or not does not determine whether cultural processes are most important in shaping language. Thus Chomsky (1957, 2007) proposes both highly language-specific cognition and an important role of these cognitive factors in shaping language. Tomasello (2003) on the other hand proposes highly general cognitive mechanism and an important role for culture. Baker (2003) proposes highly language-specific cognitive adaptations, but an important role for culture nevertheless. Griffiths and Kalish (2007) investigate a model that has a very general learning mechanism, and show that the properties of the language directly reflect the properties of the learning mechanism, thus giving less influence to cultural processes. Even so, the position that there are highly language-specific cognitive adaptations tends to go together with the position that complexity in language is mostly determined by cognitive factors, while the position that language is based on general cognitive mechanisms often goes together with the position that complexity in language is mostly determined by cultural processes. This can perhaps be explained from the fact that general linguistics attempts to derive properties of cognition from observation of language(s) (as first widely advocated by Chomsky, 1957). It is then not extremely fruitful to look for evidence for language-specific cognition in language if one does not believe that languages are determined by those cognitive properties. On the other hand, if one does not believe in language-specific cognition, then one is also less worried about the fact that properties of language may not be determined by cognitive biases. Although these combinations of points of view are therefore methodologically understandable, they are in fact independent, and which combination obtains in reality is an empirical question. Nonetheless, linguistics does seem to oscillate between these two positions. Recently, the position that there are no language-specific cognitive adaptations and that languages are shaped by cultural processes appears to be gaining in popularity. Evans and Levinson (2009) have investigated many proposed universals of human languages and find exceptions for almost all the universals they investigate. They conclude that language universals are a “myth” and therefore that they cannot reflect strict innate restrictions of the human capacity for language. They do find that some patterns occur (much) more often than

52

Bart de Boer

other possible patterns, but they explain this as the result of cultural evolution under functional pressure. In reaction to their article, Tomasello (2009) has said, “Universal grammar is, and has been for some time, a completely empty concept.” To give further arguments against the possibility of evolved languagespecific cognitive adaptations, Christiansen, Chater and Reali (Christiansen & Chater, 2008; Chater et al., 2009) have shown that non-functional biological adaptations for language cannot become fixed through biological evolution, because language changes too quickly through cultural evolution. Because cultural evolution works much faster than biological evolution, language is a moving target. Although in biology it is possible for acquired behavior to influence the biological makeup of an organism through the Baldwin effect (Baldwin, 1896) this requires the environment in which the behavior is acquired to be stable for a very long time. The Baldwin effect has been proposed by linguists as a way in which genetic adaptations to language can emerge (Pinker & Bloom, 1990) but Christiansen and Chater (2008) argue that language changes too quickly for this to work. It must be noted that Christiansen, Chater and Reali (Christiansen & Chater, 2008; Chater et al., 2009) do allow for the possibility that properties of language that tend to recur because they are functionally useful may be stable enough that they influence the biological evolution of the brain. However, in the way their work is cited, for example by Evans and Levinson (2009), it often appears that this nuance is lost, and that it is impossible for language-specific cognitive adaptations to evolve at all. This leads to the impression that biological evolution has not played an important role in the evolution and complexification of language. Interestingly, a very similar point of view is advocated by Chomsky (2007) when writing about evolution, even though (as pointed out earlier) Chomsky has advocated the importance of language-specific innate properties of the brain for understanding language. His point of view is that recursion is a crucial and language-specific property of human cognition (as pointed out in Hauser et al., 2002). However, because there is not such a thing as partial recursion (according to Chomsky, 2007), there is no evolutionary scenario by which recursion could have evolved through incremental improvements. As Chomsky (at least from his minimalist perspective) considers other aspects of language as peripheral, he therefore does not see an important role for biological evolution in the emergence of language, rather he proposes that language may be the result of an evolutionary accident or of exaptation from different functions (e.g., complex cognition). The unexpected convergence of viewpoints of researchers starting from very different perspectives about the lack of importance of biological evolution for understanding language has been remarked upon by Berwick (2009). Of course, many linguists have a more nuanced position, and see language as

Complexity in Speech: Teasing Apart Culture and Cognition

53

the result of complex co-evolution between culture and biology (e.g., Mufwene 2013) and this is the position taken in this chapter as well. In fact, discarding biologically determined and language-specific cognitive mechanisms risks throwing the baby out with the bathwater. If there are no language-specific cognitive mechanisms, then why would one study language from a cognitive point of view at all? It would then be much more efficient to investigate the general cognitive mechanisms separately, and leave language to descriptive linguists. However, Chater et al. (2009) leave open the possibility that functionally relevant cognitive adaptations can evolve. In addition, it may be possible that linguistic cognition is less directly reflected in properties of language than Evans and Levinson (2009) appear to assume, or that there are properties of language that are universal, but that have been considered uninteresting by most linguists (e.g., de Boer, 2014). Therefore, it may be too early to declare language-specific biological adaptations irrelevant. One problem is that the search for language-specific cognitive adaptations appears to have focused on properties of language that do not have a clear function. This is understandable if one searches for language-specific cognitive adaptations: universals of language that cannot be explained from functional factors must be attributed to cognitive factors. However, this does not mean that properties of language that have a clear function, such as a large lexicon, are not associated with corresponding cognitive adaptations. On the contrary, Christiansen, Chater and Reali’s (Christiansen & Chater, 2008; Chater et al., 2009) work appears to indicate that these are the only aspects of language that are stable enough over time to influence biological adaptation. And indeed, coming back to the example of large lexicons, it would seem plausible that humans have adaptations to learning large lexicons rapidly. However, the fact that we need to look for properties of language that are functional, makes it even more difficult to tell the effects of cultural and biological evolution apart. Another problem is that the interaction between cultural evolution and biological evolution makes language one of the most complex phenomena in biology (Steels, 2000; Smith et al., 2003; Beckner et al., 2009). In fact, in order to understand language completely, one must take into account that language is influenced by phenomena on three time scales (Kirby & Hurford, 1997). The shortest time scale is that of individual language learning, which takes place on the time scale of a decade. Then there is the time scale of cultural evolution and language change, which takes place on a scale ranging from decades to millennia. Finally, the slowest time scale is that of biological evolution, which takes place on time scales of the order of thousands to tens of thousands of years. These three processes are not independent, but influence each other, as illustrated in Figure 3.1. This means that it can be hard to tell apart whether properties of language are due to functional constraints, cultural evolution or

54

Bart de Boer

A

Phylogeny

Biological evolution

Stabilizes or destabilizes targets for adaptation

B

Glossogeny

Cultural evolution

Ontogeny

Learning

Sets boundary conditions

Biological evolution

Good communication helps survival

Determines biases and mechanisms Synaptic weights

Language

Cultural evolution

Language in the brain and genes are the only physical correlates of these processes

Keeps language learnable

Determines cultural fitness

Language learning genes

Figure 3.1 The different time scales in language and their interactions. Panel A illustrates the time scales, and the terminology coined by Kirby and Hurford (1997). Panel B illustrates the effects the different processes have on each other.

biological adaptations. This is one of the root causes of the enormous disagreement about these issues among linguists. The newly emerging experimental paradigm for studying cultural evolution in a laboratory setting may help to address these issues empirically. This may help re-interpret linguistic data, which by its very nature is always the result of a long process of cultural evolution under biological constraints. Even in the very few cases where emergence of a new language could be observed when it happened (Senghas et al., 2004; Polich, 2005; Sandler et al., 2005), the situation was insufficiently controlled to be able to draw general conclusions. The next section will introduce this new paradigm and illustrate it with a case study. 3

Culture and Cognition in Speech

In order to illustrate how effects of cultural processes and cognitive biases can be disentangled, an experiment on the emergence of structure in acoustic signals (Verhoef & de Boer, 2011; Verhoef et al., 2011a; Verhoef et al., 2011b) and its theoretical background will be presented in some detail. 3.1

Cultural and Cognitive Explanations of Speech Sounds

Human language is characterized by the use of a limited number of building blocks in order to produce an unlimited number of utterances. Interestingly, the inventories of building blocks that are used are not random. The building blocks appear to be constructed using a set of basic distinctive features (Jakobson & Halle, 1956). For example, the consonants [b] and [p] differ in the feature voicing: [b] is voiced and [p] is voiceless, but they are the same for other features, for example they are both labial consonants. The vowels [i] and [u]

Complexity in Speech: Teasing Apart Culture and Cognition

55

differ in the features backness and rounding where [i] is front and unrounded, while [u] is back and rounded. They are the same however for the feature height, being both high vowels. Some building blocks and some feature distinctions are very common in human languages, others are very rare. Rare features and speech sounds are called “marked” by linguists. Ever since the conception of the idea of distinctive features, there has been a debate about whether this is due to cognitive constraints or functional factors. Jakobson and Halle (1956) and Chomsky and Halle (1968) among others have argued for the position that distinctive features reflect cognitive biases. They use observations from language acquisition and aphasia to argue that there are learning biases that favor certain speech sounds and features over others. Other researchers argue that there are functional reasons why sound systems are the way they are. Liljencrants and Lindblom (1972) show that optimization of distinctiveness leads to realistic vowel systems for small vowel system sizes. Stevens (1972) argues that certain speech sounds are inherently more robust and less subject to small errors in articulation than others, and are therefore preferred in language. More recent work using agent-based computer simulation has shown that optimization for distinctiveness in vowel systems can emerge through self-organization in a population (Berrah et al., 1996; de Boer, 2000). Apparently, innate distinctive features are not necessary to explain small inventories of speech sounds. However, because the models work less well for larger inventories, other factors must also play a role. One factor appears to be feature economy (Clements, 2003, who also gives a historical overview of the concept) or the maximal use of available distinctive features (Ohala, 1980). If one analyzes the sound systems of many languages, one finds that the features that distinguish speech sounds are generally not used to distinguish just two speech sounds in the repertoire, but whole series of sounds. For example when nasalization is used in vowel systems, it is generally not used for only one vowel, but for a series of vowels. Similarly, consonants tend to occur in series: when voiced and voiceless plosives are used, generally these occur for all places of articulation used in the plosive system. Examples of the vowel system of French and the plosive system of Hindi are given in Figure 3.2. Thus distinctive features tend to be used in a maximally economic way to create large systems of speech sounds. It is possible that feature economy is the result of cultural processes, but it is equally possible that it is the result of a cognitive bias. Lending support to the interpretation of feature economy as the result of self-organization are the computer simulation results of Berrah and Laboissière (1999). These show emergent economic use of features in a population of simulated agents that have to communicate as effectively as possible. There is no explicit mechanism for economic use of features, and the authors therefore conclude that it is an emergent effect of functional constraints and cultural evolution. In a very different

56

Bart de Boer

Figure 3.2 French vowels and Hindi plosives illustrating feature economy. French vowels have different levels of height, but at most levels of height three vowels occur. In addition, all low vowels are nasalized. Hindi plosives occur at three places of articulation, but at each place of articulation, a plosive can be voiced or voiceless and aspirated or unaspirated.

line of work however, Maye et al. (2008) show that infants appear to generalize features from familiar to unfamiliar contrasts. This appears to indicate that infants do have cognitive mechanisms for extracting distinctive features from speech data and for generalizing these features across different speech sounds. Because Maye et al. did not investigate feature economy, one should be careful with extending their results to this area. Nevertheless, the mechanisms they uncover are an important prerequisite for an explanation of feature economy based on cognitive factors. There is therefore evidence that both cultural and cognitive processes may play a role in the explanation of feature economy. The challenge is therefore to experimentally tease apart the effect of culture and cognition. 3.2

Experimental Cultural Learning

The experimental paradigm of iterated (cultural) learning that has been developed over the last few years (Galantucci, 2005; Griffiths et al., 2008; Kirby et al., 2008; Smith & Kirby, 2008; Scott-Phillips & Kirby, 2010) may help to unravel the relation between culture and cognitive biases in language. Iterated learning is the repeated learning of culturally transmitted information in a group of agents, and usually focuses on human learning of language. The idea of studying the effect of repeated learning in a group was originally developed by computer modelers of the evolution of language. Two variants were proposed: Steels (1995, 1997, 1998, 2011) initially focused on interactions within a group of agents that in this way negotiate a shared communication system. This kind of cultural learning is alternatively called a language game, social coordination or horizontal transmission. Kirby and Hurford (Hurford, 1989; Kirby & Hurford, 1997; Kirby, 1999) initially focused on transfer of linguistic information across generations, which usually consisted of one agent. In their model a parent agent produces linguistic utterances that are learned by a child agent. After

Complexity in Speech: Teasing Apart Culture and Cognition

57

Vertical Transmission

Horizontal Transmission

Figure 3.3 Schematic illustration of vertical transmission (top) and horizontal transmission (bottom). In vertical transmission, participants are either speaker or hearer sequentially, and only have one role at the same time. In horizontal transmission, each participant plays the role of both speaker and hearer simultaneously.

a while the parent is removed, the child becomes the new parent, and a new child is added to the population. This approach is alternatively called3 “iterated learning,” “vertical transfer” or a “diffusion chain.” Both types are illustrated in Figure 3.3. After initial successes with these computer models, an increasing desire to link the simulations to real-world phenomena led to recreating them in a laboratory setting with human participants (Galantucci, 2005). In such experimental iterated learning tasks, participants have to learn sets of signals produced by other participants in the experiment. The fact that stimuli are produced by other participants in the experiments instead of coming from a carefully set crafted by the experimenter is an important distinction between experimental iterated learning and more classical artificial language learning experiments. In most experimental iterated learning experiments, participants are required not just to learn and (passively) recognize a set of signals, but also to reproduce these signals. Many variants are possible: experiments can either implement horizontal or vertical transmission (cf. Garrod et al., 2010 for a critical comparison). Signals can be used communicatively or not and they can have meaning or not. Experiments can be about strings of discrete symbols (letters) and thus address 3

As experimental iterated learning is a newly emerging paradigm, there is confusion about terminology. Kirby et al. prefer to call all these approaches iterated learning and distinguish between diffusion chains (studying vertical transmission) and social coordination (studying horizontal transmission). Other authors equate iterated learning with vertical transmission and call the general approach cultural learning.

58

Bart de Boer

questions related to syntax or about continuously variable signals and thus address questions about phonology and phonotactics. These and other variants are reviewed in several recent papers (Kirby et al., 2008; Smith et al., 2008; Galantucci, 2009; Scott-Phillips & Kirby, 2010). The results from iterated learning experiments can be analyzed in two ways: by looking at either the system of signals that emerges from the experiments or by looking at the ways in which participants solve the problem of learning and reproducing sets of signals. Analyzing the system of signals that emerges is comparable to analyzing existing real languages: they are the product of cultural evolution, language-specific cognitive processes, and general cognitive processes combined. The ways in which participants learn and generalize utterances, however, make it possible to directly observe the individual behaviors that shape cultural processes. By manipulating the experimental setup, or by complementing iterated learning experiments with specially crafted artificial language learning experiments, it is possible to zero in on whether these behaviors are driven by functional factors or by cognitive factors and whether they are driven by language-specific or domain general cognition. These methods will be illustrated with a small case study discussed next. 3.3

Iterated Learning of Speech

The work presented here is from Verhoef and de Boer (2011) and Verhoef et al. (2011a, 2011b, 2014). These iterated learning experiments investigate the emergence of combinatorial structure in systems of acoustic signals. Combinatorial structure is the ability to construct an unlimited number of utterances from a limited set of speech sounds. In human languages this happens according to a set of learned rules that are generally referred to as the language’s phonotactic structure. As explained earlier, some researchers have claimed that combinatorial structure is the result of cultural processes, while others have proposed that it is due to cognitive biases related to language learning and language use. The hypothesis that structure is (mainly) due to cultural processes makes radically different predictions from the hypothesis that structure is (mainly) due to cognitive processes. A mainly cultural process, as modeled for example in de Boer (2000) and Zuidema and de Boer (2009) results in slow changes that spread in the language community because the systems containing the changed sounds are better at avoiding confusion. A mainly cognitive process, such as feature economy discussed by Clements (2003) and maximal use of available distinctive features discussed by Ohala (1980), results in rapid emergence of (sets of) new building blocks and re-use of these building blocks in the utterances that make up the language. Verhoef et al.’s (Verhoef & de Boer, 2011; Verhoef et al., 2011a; Verhoef et al., 2011b; Verhoef et al., 2014) experiments directly investigate this issue.

Complexity in Speech: Teasing Apart Culture and Cognition

59

These experiments investigate vertical transmission of small systems of initially unstructured signals. Participants are exposed to a set of 12 different signals and are then asked to reproduce them four times. Their last set of reproductions is used as training data for the next participant in the chain. If one of the signals that the participants produce is too close to one they have already made, this is flagged, and they are asked to retry to produce a signal. It should be noted that no meaning is associated with the signals. In order to avoid interference from preexisting linguistic knowledge as much as possible, the signals are not spoken sounds, but are short whistles produced by means of a slide whistle. Because most participants are not familiar with the use of a slide whistle, they can practice for a short period before the actual experiment. Four different chains of ten participants each have been investigated so far. It turns out that systems of signals become significantly more learnable over the generations (as measured by Page’s trend test on the recall error of the participants, Verhoef & de Boer, 2011). Also, the signals in the systems become more similar (as measured by Page’s trend test on the average distance between a signal and its nearest neighbor in the set, Verhoef & de Boer, 2011). In addition, it appears that the entropy of the building blocks as used in the set of signals also decreases over the generations (Verhoef et al., to appear 2012). The first observation would be expected for both the cultural and cognitive hypotheses: more distinctiveness predicted by the cultural hypothesis would make it easier for participants to keep signals apart, whereas more structured sets of signals predicted by the cognitive hypothesis would be more easily learned. The second observation is in contradiction with the cultural hypothesis, which predicts emergence of more distinctive signals, and therefore a larger distance between signals. The cognitive hypothesis does not really make a prediction here, although increased re-use of building blocks could make signals constructed out of these building blocks more similar. The third observation is predicted by the cognitive hypothesis. However, it has been shown that cultural processes can also result in sets of signals that appear to reuse building blocks (de Boer & Zuidema, 2010). Taken together, the observations do appear to indicate a problem for the cultural hypothesis, but it is necessary to look more closely at what happens during learning and reproduction in order to find out which hypothesis (if any) can be rejected. When we look at individual reproductions of learned stimuli, it is much clearer that this involves a process of re-use and generalization of building blocks. Participants have a hard time learning the initial, unstructured set. However, when reproducing signals, they are forced to produce 12 distinct signals. This process results in subconscious generalization and re-use of observed patterns. Three processes by which participants do this are illustrated in Figure 3.4. The first process is the combination of (parts of) one signal into a new signal. Verhoef and de Boer call this “borrowing.” The second process is reversing

60

Bart de Boer 1s 1 octave

generation n

generation n + 1

“borrowing”

n

n+1

“mirroring”

m

m+1 “duplication”

Figure 3.4 Three processes through which experimental participants create new signals in their imperfect reproductions of a learned set of signals. In the left panel, borrowing is illustrated through the recombination of one signal of the learned set with the second half of another signal from the learned set. In mirroring, a signal of the learned set appears both in its original form and in a temporally mirrored form. In duplication, a signal from the learned set is analyzed as consisting of two elements. One of these elements is duplicated in reproduction. Signals were redrawn from (Verhoef & de Boer, 2011).

the pitch pattern of a signal and including this as a new signal in the new set of signals. Verhoef and de Boer call this “mirroring.” The third process creates new signals by repeating elements from existing signals. Verhoef and de Boer call this “duplication.” These processes are clearly indicative of (conscious or subconscious) manipulation of elements of learned signals and must therefore be based on cognitive processes. They are very different from the small and random shifts that are used in models of cultural processes, such as the simulations by Zuidema and de Boer (Zuidema & de Boer, 2009; de Boer & Zuidema, 2010). It must therefore be concluded that in this experiment, cognitive processes play a leading role in the emergence of structure. This does not mean that cultural processes do not play a role whatsoever. In fact, structure emerges only gradually over the experimental generations, indicating that structure is amplified by cultural transmission of the system of signals. In addition, the different chains of transmission from participant to participant appear to converge on differently structured sets of signals. It is currently tested experimentally whether human listeners can identify this culturally transmitted structure (Verhoef, 2013, ch. 5). In these experiments, participants (different from the ones who participated in the original experiment) are exposed to signals that emerged from a particular chain. Then they are tested with a mixture of signals that emerged in the same chain they were trained with (but that they have not heard before) and signals that emerged from different chains. It is measured whether participants can reliably distinguish between signals that emerged from their training chain and those that did not. These experiments appear to indicate that chains of culturally transmitted signals converge to distinguishable sets of signals (thus stressing that cultural effects are important in addition to the cognitive processes described previously).

Complexity in Speech: Teasing Apart Culture and Cognition

4

61

Discussion

This chapter has attempted to illustrate a possible method, with origins in the study of language evolution, of teasing apart the role of cultural and cognitive processes in the emergence of complexity in language. It is a central question in linguistics which aspects of language are due to cultural processes, which aspects are due to language-specific cognitive processes and which aspects are due to general cognitive processes. In the case of general cognitive mechanisms, the evolutionary pressures that shaped these mechanisms have not been specific to language, whereas in the case of language-specific cognitive mechanisms, there must have been evolutionary pressures specific to language. It is difficult to assess the role of these different mechanisms, as existing languages are the product of continuous interaction between the three mechanisms. The paradigm of experimental cultural (or iterated) learning was developed by researchers of the evolution of language in order to study the effects of cultural processes in a laboratory setting. It consists of repeated interactions between participants in which language-like stimuli are learned, reproduced and (possibly) invented. In this way it can be studied directly how languagelike systems change due to interactions between users of the system. Human behavior can then be compared with predictions both of models that are based on cultural processes and of those that are based on cognitive processes. This approach has been illustrated for the case of emergence of complex combinatorial structure in acoustic signals. It was shown that human behavior corresponds more closely to behavior predicted by models based on cognitive processes (comparable to feature economy or maximal use of available distinctive features) than by models based on purely cultural processes (selforganization under pressure of distinctiveness). However, as the different learning chains converge to sets of signals that are structured according to different rules, cultural processes must play an important role as well. If only cognitive processes would play a role, then one would expect the learning chains to always converge to very similar sets of utterances. These experiments both illustrate the interaction of cognitive and cultural processes, as well as the ability to tease them apart. The experiments conform to observations of systems of speech sounds in existing languages and the way they change: people have a tendency to apply the same processes (duplication and borrowing in the experiments; phonological processes in real languages) in different utterances and to re-use existing building blocks. In addition, humans appear to apply similar processes to the building blocks themselves. In the experiments this is observed in the case of mirroring of building blocks, and in real languages it happens when one feature spreads over different phonemes or when utterances are re-analyzed in terms of new sets of building blocks. In this way, whole sets of new speech sounds can

62

Bart de Boer

appear (or disappear) in a language in a single operation. We can therefore tentatively conclude that complexity of phonological systems is due to cognitive mechanisms that re-use and generalize building blocks. Due to the cognitive biases to learn and generalize complex sets of speech sounds, change of (systems of) speech sounds is not a completely random process. This does not alter the fact that language change can be viewed as a process of cultural evolution, just that the operations that introduce variation are not as random as those of biological evolution. This is a well-known potential difference between cultural evolution and biological evolution (e.g., Dennett, 2006) but it is one that is sometimes overlooked by linguists applying evolutionary theory to language change (e.g., Blevins, 2004). Experimental investigation of language creation, change and complexification as illustrated in this chapter, may help to clarify how variation is introduced and how complexity is created in language. A discrepancy between the experiments and real language is that in the experiments rather radical changes happen quickly and frequently, whereas real languages are more stable over generations. This is most likely due to a much stronger conformity bias in the case of real language: language users are exposed to a much larger set of linguistic stimuli, and therefore learn much more accurately than in the experiments. Moreover, language users cannot change their language too radically as otherwise they would be unable to communicate with others in the community. The work described here is just the beginning of the experimental investigation of how culture and cognition interact in the emergence of phonological complexity, and a rather rough beginning at that. The work does form part of a movement to apply similar techniques to different aspects of language, most notably syntax and morphology, but also pragmatics and semantics (Kirby et al., 2008; Galantucci, 2009; Scott-Phillips & Kirby, 2010). There are a large number of issues that still need to be addressed, including: What is the effect of adding meaning to the tasks? What is the effect of interaction between speakers? Which phenomena are due to learning and reproduction and which phenomena are due to invention of utterances? Which processes are due to general cognition and which processes are due to language-specific cognition? Finally, there is the issue of the extent to which it is possible to exclude all influence from the linguistic knowledge that the participants of the experiments already have. All these questions notwithstanding, the approach of experimental cultural learning promises to be a useful new way to investigate old questions. Acknowledgments This work was financed by the NWO vidi grant 276–75–007 “modelling the evolution of speech” and the ERC starting grant project 283435, “ABACUS.”

Complexity in Speech: Teasing Apart Culture and Cognition

63

The author wishes to thank Tessa Verhoef and the editors for comments on the manuscript.

REFERENCES Baker, Mark C. (2002). The atoms of language. Oxford: Oxofrd University Press. Baker, Mark C. (2003). Linguistic differences and language design. Trends in Cognitive Sciences, 7(8), 349–353. Baldwin, James M. (1896). A new factor in evolution. The American Naturalist, 30, 441–451, 536–553. Beckner, Clay, Blythe, Richard, Bybee, Joan, Christiansen, Morten H., Croft, William, Ellis, Nick C., Holland, John, Ke, Jinyun, Larsen-Freeman, Diane & Schoenemann, Tom (2009). Language is a complex adaptive system: Position paper. Language Learning, 59(Suppl. 1), 1–26. Berrah, Ahmed-Redah, Glotin, Hervé, Laboissière, Rafael, Bessière, Pierre, & Boë, Louis-Jean. (1996). From form to formation of phonetic structures: An evolutionary computing perspective. In Terry Fogarty & Gilles Venturini (Eds.), ICML ‘96 workshop on evolutionary computing and machine learning, bari (pp. 23–29). Berrah, Ahmed-Redah, & Laboissière, Rafael (1999). Species: An evolutionary model for the emergence of phonetic structures in an artificial society of speech agents. In Dario Floreano, Jean-Daniel Nicoud & Francesco Mondada (Eds.), Advances in artificial life, lecture notes in artificial intelligence (Vol. 1674, pp. 674–678). Berlin: Springer. Berwick, Robert C. (2009). What genes can’t learn about language. Proceedings of the National Academy of Sciences, 106(6), 1685–1686. Blevins, Juliet (2004). Evolutionary phonology. Cambridge: Cambridge University Press. Chater, Nick, Reali, Florencia, & Christiansen, Morten H. (2009). Restrictions on biological adaptation in language evolution. Proceedings of the National Academy of Sciences, 106(4), 1015–1020. Chomsky, Noam (1957). Syntactic structures (second, 2002 ed.). Berlin: Mouton de Gruyter. Chomsky, Noam (1981). Lectures on government and binding. Dordrecht: Foris Publications. Chomsky, Noam (2007). Of minds and language. Biolinguistics, 1(1), 9–27. Chomsky, Noam, & Halle, Morris (1968). The sound pattern of English. Cambridge, MA: MIT Press. Christiansen, Morten H., & Chater, Nick (2008). Language as shaped by the brain. Behavioral and Brain Sciences, 31, 489–558. Clements, George N. (2003). Feature economy in sound systems. Phonology, 20, 287– 333. de Boer, Bart (2000). Self organization in vowel systems. Journal of Phonetics, 28(4), 441–465. de Boer, Bart (2014). Biological Adaptation to Cultural Traits. In: Cartmill, Erica A., Roberts, Seàn, Lyn, Heidi & Cornish, Hannah (Eds.), The Evolution of Language, Proceedings of the 10th International Conference (EVOLANG10) (pp. 419–420), New Jersey: World Scientific.

64

Bart de Boer

de Boer, Bart, & Zuidema, Willem (2010). An agent model of combinatorial phonology. Adaptive Behavior, 18(2), 141–154. Dennett, Daniel (2006). From typo to thinko: When evolution graduated to semantic norms. In Stephen C. Levinson & Pierre Jaisson (Eds.), Evolution and culture (pp. 133–145). Cambridge, MA: MIT Press. Dobzhansky, Theodoszius (1973). Nothing in biology makes sense, except in the light of evolution. The American Biology Teacher, 35, 125–129. Evans, Nick, & Levinson, Stephen C. (2009). The myth of language universals: Language diversity and its importance for cognitive science. Behavioral and Brain Sciences, 32, 429–492. Fisher, Simon E., & Marcus, Garry F. (2005). The eloquent ape: Genes, brains and the evolution of language. Nature Reviews Genetics, 7, 9–20. Fitch, W. Tecumseh (2005). The evolution of language: A comparative review. Biology and Philosophy, 20, 193–230. Fitch, W. Tecumseh (2010). The evolution of language. Cambridge: Cambridge University Press. Galantucci, Bruno (2005). An experimental study of the emergence of human communication. Cognitive Science, 29, 737–767. Galantucci, Bruno (2009). Experimental semiotics: A new approach for studying communication as a form of joint action. Topics in Cognitive Science, 1, 393–410. Garrod, Simon Fay, Nicolas, Rogers, Shane, Walker, Bradley, & Swoboda, Nik (2010). Can iterated learning explain the emergence of graphical symbols? Interaction Studies, 11(1), 33–50. Goldberg, Adele E. (2003). Constructions: A new theoretical approach to language. Trends in cognitive sciences, 7(5), 219–224. Griffiths, Thomas L., & Kalish, Michael L. (2007). Language evolution by iterated learning with Bayesian agents. Cognitive Science, 31(3), 441–480. Griffiths, Thomas L., Kalish, Michael L., & Lewandowsky, Stephan (2008). Theoretical and empirical evidence for the impact of inductive biases on cultural evolution. Philosophical Transactions of the Royal Society of London B, 363(1509), 3503– 3514. Hauser, Marc D., Chomsky, Noam, & Fitch, W. Tecumseh (2002). The faculty of language: What is it, who has it, and how did it evolve? Science, 298, 1569–1579. Hurford, James R. (1989). Biological evolution of the saussurean sign as a component of the language acquisition device. Lingua, 77(2), 187–222. Jakobson, Roman, & Halle, Morris (1956). Fundamentals of language. The Hague: Mouton & Co. Kirby, Simon (1999). Function, selection and innateness: The emergence of language universals. Oxford: Oxford University Press. Kirby, Simon, Cornish, Hannah, & Smith, Kenny (2008). Cumulative cultural evolution in the laboratory: An experimental approach to the origins of structure in human language. Proceedings of the National Academy of Sciences, 105(31), 10681– 10686. Kirby, Simon, & Hurford, James R. (1997). Learning, culture and evolution in the origin of linguistic constraints. In Phil Husbands & Inman Harvey (Eds.), Fourth European conference on artificial life (pp. 493–502). Cambridge, MA: MIT Press. Liljencrants, J., & Lindblom, Björn (1972). Numerical simulations of vowel quality systems. Language, 48, 839–862.

Complexity in Speech: Teasing Apart Culture and Cognition

65

Maddieson, Ian (1984). Patterns of sounds. Cambridge: Cambridge University Press. Maddieson, Ian, & Precoda, Kristin (1990). Updating UPSID. UCLA Working Papers in Phonetics, 74, 104–111. Maye, Jessica, Weiss, Daniel J., & Aslin, Richard N. (2008). Statistical phonetic learning in infants: Facilitation and feature generalization. Developmental Science, 11(1), 122–134. Mufwene, Salikoko S. (2008). Language evolution: Contact, competition and change. London: Continuum International Publishing Group. Mufwene, Salikoko S. (2013) Language as technology. In: Lohndal, Terje (Ed.) In Search of Universal Grammar: From Old Norse to Zoque. (pp. 327–358). Amsterdam / Philadelphia, PA: John Benjamins Publishing Company. Ohala, John J. (1980). Moderator’s introduction to symposium on phonetic universals in phonological systems and their explanation. In Proceedings of ICPHS IX, vol 3 (pp. 181–185). Copenhagen: Institute of Phonetics. Pinker, Steven, & Bloom, Paul (1990). Natural language and natural selection. Behavioral and Brain Sciences, 13(4), 707–784. Polich, Laura (2005). The emergence of the deaf community in Nicaragua. Washington, DC: Gallaudet. Sandler, Wendy, Meir, Irit, Padden, Carol, & Aronoff, Marc (2005). The emergence of grammar: Systematic structure in a new language. Proceedings of the National Academy of Sciences, 102(7), 2661–2665. Scott-Phillips, Thom C., & Kirby, Simon (2010). Language evolution in the laboratory. Trends in Cognitive Sciences, 14, 411–417. Senghas, Ann, Kita, Sotaro, & Özyürek, Aslı (2004). Children creating core properties of language: Evidence from an emerging sign language in Nicaragua. Science, 305, 1779–1782. Smith, Kenny, Brighton, Henry, & Kirby, Simon (2003). Complex systems in language evolution: The cultural emergence of compositional structure. Advances in Complex Systems, 6(4), 1–22. Smith, Kenny, Kalish, Michael L., Griffiths, Thomas L., & Lewandowsky, Stephan (2008). Introduction. Cultural transmission and the evolution of human behaviour. Philosophical Transactions of the Royal Society of London, 363(1509), 3469– 3476. Smith, Kenny, & Kirby, Simon (2008). Cultural evolution: Implications for understanding the human language faculty and its evolution. Philosophical Transactions of the Royal Society of London B, 363, 3591–3603. Steels, Luc (1995). A self-organizing spatial vocabulary. Artificial Life, 2(3), 319–332. Steels, Luc (1997). The synthetic modelling of language origins. Evolution of Communication, 1(1), 1–34. Steels, Luc (1998). Synthesising the origins of language and meaning using coevolution, self-organisation and level formation. In James R. Hurford, Michael Studdert-Kennedy & Chris Knight (Eds.), Approaches to the Evolution of Language (pp. 384–404). Cambridge: Cambridge University Press. Steels, Luc (2000). Language as a complex adaptive system. In M. Schoenauer, K. Deb, G. Rudolph, X. Yao, E. Lutton, J. Merelo & H.-P. Schwefel (Eds.), Parallel problem solving form nature, PPSN VI (Vol. 1917, pp. 17–26). Berlin: Springer. Steels, Luc (2011). Modeling the cultural evolution of language. Physics of Life Reviews, 8(4), 339–356.

66

Bart de Boer

Stevens, Kenneth N. (1972). The quantal nature of speech: Evidence from articulatoryacoustic data. In E. E. J. David & P. B. Denes (Eds.), Human communication: A unified view (pp. 51–66). New York: McGraw-Hill. Tomasello, Michael (2003). Constructing a language: A usage-based theory of language acquisition. Cambridge, MA: Harvard University Press. Tomasello, Michael (2009). Universal grammar is dead. Behavioral and Brain Sciences, 32(5), 470–471. Verhoef, Tessa (2013). Efficient coding in speech sounds: Cultural evolution and the emergence of structure in artificial languages, Ph.D. thesis, University of Amsterdam. Verhoef, Tessa, & de Boer, Bart (2011). Cultural emergence of feature economy in an artificial whistled language. Paper presented at the International Conference on Phonetic Sciences, Hong Kong. Verhoef, Tessa, de Boer, Bart, Del Giudice, Alex, Padden, Carol, & Kirby, Simon (2011a). Cultural evolution of combinatorial structure in ongoing artificial speech learning experiments. CRL Technical Report UCSD, 23(1), 3–11. Verhoef, Tessa, de Boer, Bart, & Kirby, Simon (2012). Holistic or synthetic protolanguage: Evidence from iterated learning of whistled signals. In The evolution of language: Proceedings of the 8th international conference (EVOLANG8) (368–375). Hackensack, NJ: World Scientific. Verhoef, Tessa, Kirby, Simon, & Padden, Carol (2011b). Cultural emergence of combinatorial structure in an artificial whistled language. In L. Carlson, C. Hölscher & T. Shipley (Eds.), Proceedings of the 33rd annual conference of the cognitive science society (pp. 483–488). Austin, TX: Cognitive Science Society. Verhoef, Tessa, Kirby, Simon & de Boer, Bart (2014). Emergence of combinatorial structure and economy through iterated learning, Journal of Phonetics 43C, 57–68. Zuidema, Willem, & de Boer, Bart (2009). The evolution of combinatorial phonology. Journal of Phonetics, 37(2), 125–144.

4

A Complex-Adaptive-Systems Approach to the Evolution of Language and the Brain P. Thomas Schoenemann

1

Introduction

Language has arguably been as important to our species’ evolutionary success as has any other behavior. Understanding how language evolved is therefore one of the most interesting questions in evolutionary biology. Part of this story involves understanding the evolutionary changes in our biology, particularly our brain. However, these changes cannot themselves be understood independent of the behavioral effects they made possible. The complexity of our inner mental world – what will here be referred to as conceptual complexity – is one critical result of the evolution of our brain, and it will be argued that this has in turn led to the evolution of language structure via cultural mechanisms (many of which remain opaque and hidden from our conscious awareness). From this perspective, the complexity of language is the result of the evolution of complexity in brain circuits underlying our conceptual awareness. However, because individual brains mature in the context of an intensely interactive social existence – one that is typical of primates generally but is taken to an unprecedented level among modern humans – cultural evolution of language has itself contributed to a richer conceptual world. This in turn has produced evolutionary pressures to enhance brain circuits primed to learn such complexity. The dynamics of language evolution, involving brain/behavior co-evolution in this way, make it a prime example of what have been called “complex adaptive systems” (Beckner et al. 2009). Complex adaptive systems are phenomena that result from the interaction of many individual components as they adapt and learn from each other (Holland 2006). They have in common the emergence of interesting higher-level patterns that do not initially appear obvious given the behavior of individual actors in the system. Classic examples of complex adaptive systems are ant colonies, in which the behavior of the colony as a whole is highly intelligent, flexible and adaptive, even though the colony itself is composed exclusively of individuals with very simple, rigidly stereotyped behavior (Holland 1998). Those that take the view that “the mind is what the brain does” are essentially making the same kind of argument: the action of each neuron is very simple, and no single neuron 67

68

P. Thomas Schoenemann

“knows” anything about the mind, yet the sum of their actions is the mind. From a complex adaptive systems perspective, it is a mistake to assume that the patterns and behaviors of the whole system are in each individual agent. Instead, the patterns emerge from the sum of actions of sets of adaptively interacting agents and are therefore more properly understood as existing between agents. The argument here will be that language is similarly not the result of neural circuits innately coded in the mind of each individual, but instead that these patterns are the result of complex interactions at three levels: biological evolution, cultural evolution, and the ontogenetic development of individuals. The complexity of language is the result of a biocultural evolutionary process, or one that is neither exclusively biological nor exclusively cultural (Christiansen & Chater 2008; Evans & Levinson 2009). In particular, it will be argued that the patterns of language use – grammar and, more specifically, syntax – are more properly understood as being emergent characteristics of increasing conceptual complexity of individuals who are embedded in an intensely socially interactive (i.e., communicative) existence (Savage-Rumbaugh & Rumbaugh 1993; Schoenemann 1999). This intense social interactivity is a legacy of our being primates and long predates the origin of our species – let alone language. The complexity of language has been studied from a variety of perspectives in linguistics. One dominant view, identified with Noam Chomsky and followers (e.g., Chomsky 1972; Jackendoff 2002; Pinker & Jackendoff 2005), has been that linguistic complexity is best understood as the result of some formal (logical) model, with the assumption being that a correctly described model will be innately instantiated in the brain in some way. There is no doubt that regularities exist in specific languages. Whether there are truly any language universals across languages, however, has been called into question (Evans & Levinson 2009). The complexity and fuzziness of the phonological systems within and between languages is so great that even the basic assumption that speech can appropriately be divided into discrete packages (called phonemes) has been called into question (Port 2010; Port & Leary 2005). Furthermore, those who believe in Universal Grammar do not agree about exactly what formal model best describes language. Chomsky’s own models have notoriously changed a great deal over the past half-century, with the latest incarnation emphasizing a so-called “minimalist” view of the underlying cognitive mechanisms (Chomsky 1995). However, the minimalist program has been criticized by other formalists (e.g., Pinker & Jackendoff 2005), highlighting the lack of agreement about what such a model of language should look like even among linguists subscribing to Chomsky’s general approach. In any case, the goal in some areas of linguistics has been to try to understand language complexity on its own terms, assuming that this could be validly studied independent of how the brain actually works and that language

A Complex-Adaptive-Systems Approach to the Evolution of Language

69

structures so derived would then necessarily be evident in the brain in some fashion. Language complexity has also usually been studied without a deep understanding of how the evolutionary process has led to various other kinds of behavioral change. This lack of evolutionary grounding is particularly unfortunate, because language did in fact evolve, and, as a consequence, any model of language complexity that is not easily explained from an evolutionary perspective cannot then be considered a legitimate model of language itself (Schoenemann 1999). Understanding the complexity of language necessitates understanding how the brain works in creating meaning, within the context of how the evolutionary process would mold both brain and language. 2

Constraints on Language Adaptation

The evolutionary changes that occurred in our lineage to make language universal are presumably the result of natural selection (Pinker & Bloom 1990). For natural selection to operate, however, there must be some environmental influence favoring these changes. The environment relevant to language ability is the social environment, which is of course made up of interacting individuals. Evolutionary changes relevant to language will only be beneficial to an individual to the extent that they help that individual communicate better with others. This in turn presumes that these other individuals already have similar enough abilities to begin with (before selection operates), such that any new changes introduced by one individual will actually increase this individual’s communication ability with others. This dynamic constrains the possible ways in which evolutionary change can occur, constantly biasing changes toward modifications of pre-existing abilities. This in turn predicts that language will have been built on a cognitive foundation that we share with other species, and furthermore that these foundations should still be evident. One critical component of this foundation is conceptual understanding: how we perceive, experience and make sense of the world. Aspects of this conceptual understanding are, of course, exactly what we are trying to share and communicate with others using language. 3

Types of Complexity

In order to assess complexity in language systems, it is important to define what we mean by “complexity.” One place to look for such a definition is in the field of Complexity Theory, where a number of definitions have been proposed (e.g., Horgan 1995; Mikulecky 2001). A system in a high degree of disorder (i.e., having high entropy) is, in some sense, complex, but it is not necessarily complex in a particularly interesting way (Horgan 1995). One popular notion in the field has been that the most interestingly complex systems exist at “the edge

70

P. Thomas Schoenemann

of chaos” (Langton 1990), somewhere between complete chaos and complete order. Langton argued that systems in such a state are the most potentially useful for computation (Langton 1990), though subsequent work has not particularly supported this idea (Mitchell et al. 1993). One intriguing suggestion is that complexity should be understood as “the property of a real world system that is manifest in the inability of any one formalism being adequate to capture all its properties” (Mikulecky 2001: p. 344). This definition fits nicely with the observation that it has so far been impossible to find a single formalism, even among committed formalists, that describes the patterns of language grammar to everyone’s satisfaction (Pinker & Jackendoff 2005). If instead we view language as the result of a complex adaptive system, in which interacting biological and cultural evolutionary systems – each with their own constraints, influences, and partly interdependent histories – conspire over evolutionary time to produce a system of communication, the problem of language evolution becomes tractable. Mikulecky (2001) suggests that the phenomenon of “emergence” seen in complex adaptive systems may simply be “a result of the limits of a dominant formalism” (p. 344). In other words, perhaps it is our fascination with trying to box complex systems into single formal models that leads to our surprise at the “emergent” behavior of these systems. If we view language evolution as the result of a complex interplay of influences of different kinds (each described, imperfectly, by their own unique formalisms), the emergence of language becomes much less miraculous. For the purposes of this chapter, we will emphasize the following senses of complexity: 1) number of different kinds of individual things (actions, objects, etc.); 2) number of individual interactions between things; 3) number of types of interactions between things; and 4) levels of hierarchical interaction between sets of things. 4

Conceptual Complexity

What exactly is meant by conceptual complexity? It can be understood to be a function of: (1) the number of different dimensions the brain can meaningfully distinguish, and (2) the number of possible interactions between these dimensions (Schoenemann 2010). “Meaningfully distinguish” in this context may be defined as any internally detectable difference in pattern(s) of brain activation (whether caused by external stimuli or by internal neural activity); and “dimensions” may be defined as aspects of reality that the brain is sensitive to (e.g., wavelengths of light, types of molecules, temperature, etc.) or internally creates (e.g., emotions, patterns of thought, etc.). One can imagine that there are organisms that have a much simpler conceptual understanding of the world than

A Complex-Adaptive-Systems Approach to the Evolution of Language

71

we do. Some invertebrates such as jellyfish, for example, do not have eyes that produce images, but rather have very simple light-sensitive eyespots (“pigment spot ocelli”) that simply detect light coming from a particular direction (Hudson 2010). This suggests a very simple conceptual awareness of visual information, vastly simpler than what humans (and most vertebrates) have available. 5

The Neural Basis of Concepts

To understand why a comparative cross-species assessment of brain structure could imply something important about likely degrees of conceptual complexity, it is important to understand how concepts are instantiated in the brain. Neural processing is thought to be the result of networks of neurons in different states or temporal patterns of activation (Baars & Gage 2007). The brain is organized with a degree of regional specialization of function. The cortex itself, which is the seat of conscious awareness (in humans at least), is divided up into different regions that specialize, to one degree or another, in particular types of processing (Baars & Gage 2007). The characteristics of the neural circuitry in different regions (“cytoarchitecture”) differ enough to be reasonably identifiable across individuals, forming the basis for the identification of specific brain regions. These cytoarchitectural characteristics led the early neuroanatomist Brodmann (1909) to suggest a classification scheme for cortical areas (called “Brodmann areas”) that is still used to this day. The details of what exactly each of these cytoarchitectural areas does, and how they interact, is by no means completely understood, but there are wellstudied pathways that are known to specialize in different kinds of information (Baars & Gage 2007). For example, it is possible to distinguish separate areas and neural pathways specialized for visual, auditory, somatosensory (i.e., touch, temperature, pain), olfactory, and taste information. Within these, there are typically further subdivisions of function. For example, cortical processing of visual information starts in the primary visual cortex in the occipital lobe (at the far back of the brain), but then divides approximately into (1) a dorsal “where” pathway (extending superiorly into the parietal lobe) that specializes in movement and spatial aspects of visual information, and (2) a ventral “what” pathway (extending anteriorly into the temporal lobe) that specializes in object identification (Baars & Gage 2007). To be sure, there are additional areas and pathways that integrate this information in various ways, and basic sensory processing in one sensory domain can affect basic sensory processing in another domain, such that these primary areas are not completely independent of each other. One very nice example of this is the McGurk effect (McGurk & MacDonald 1976), where the basic auditory perception of a syllable (e.g., ba) is fundamentally changed by concurrent visual input of someone saying a different syllable (e.g., ga). Nevertheless,

72

P. Thomas Schoenemann

there are areas known as primary sensory areas that are known to specialize in the processing of specific types of sensory information. Very simple concepts, like colors, smells or tastes, are thought to be the result of particular patterns of neural processing within different primary sensory areas that combine different types of signals from external sensors. The retina of the eye, for example, has cells that are tuned to respond to particular wavelengths of light. Different colors are distinguished by different patterns of stimulation of sets of these basic retinal cells. More complex concepts are, at some level, based on patterns of activation of different networks subserving often very different kinds of information. The concept of coffee, for example, binds together a number of sensory components, involving not only taste and smell, but also – for many people – visual components (e.g., shape and color of coffee beans), somatosensory components (e.g., warmth of the fluid, physical features of the cup typically used to drink it), and even more abstract components, such as the sense of well-being that many feel as a result of drinking it (Damasio & Damasio 1992). There remains some question over how concepts are actually represented in the brain at the neural level. One suggestion is that there are specific individual neurons that represent specific concepts. Such hypothetical neurons are usually referred to as “grandmother cells” (after a tongue-in-cheek parable by the neurobiologist Jerry Lettvin that introduced the term; Gross 2002). The idea of individual concept-specific neurons is a logical extension of work on how the visual system identifies objects (Gross 2002). At the level of the retina, individual ganglion cells are sensitive to the activity of specific, very simple patterns of photoreceptor cells: for example, a small spot of light surrounded by darkness. A straight line can therefore be detected as the coincident activation of a unique set of these “spot detector” ganglion cells. This coincident activation can be detected by a single neuron that fires only when the unique set of ganglion cells are also active (for a basic discussion see Goldsmith & Zimmerman 2000). Such a neuron would be the neural representation of a line (a simple concept). However, note that for such a “line detector” neuron to be activated, a network of retinal ganglion neurons must also have been activated. For increasingly complicated, subtle, and interesting concepts, larger and larger networks of neural activity will be involved, connecting regions specializing in processing different kinds of information relevant to those concepts. Whether unique “grandmother cells” (or even specialized networks that we might call “grandmother circuits”) actually exist for more complicated concepts, and if so, whether these should be considered the neural instantiations of concepts, is not critical for the argument here. The important point is that networks of brain activation form the foundation for concepts in the brain. Congruent with this, Barsalou (2010) has argued that the “core representations in cognition” – what we are calling here “concepts” – are not “amodal

A Complex-Adaptive-Systems Approach to the Evolution of Language

73

data structures that exist independently of the brain’s modal [basic sensory] systems,” but are instead fundamentally grounded in “the environment, situations, the body, and simulations in the brain’s modal systems” (p. 717). He refers to this as “grounded cognition,” and notes a number of perspectives that support this idea. Lakoff and Johnson’s (1980) analysis of linguistic metaphors led them to argue that abstract concepts are specifically grounded in bodily experience. Gibson’s (1979) work on visual perception led him to argue the external environment plays a fundamental role in perception. Work by Paivio (1971), Shepard and Cooper (1982), and Kosslyn (1980) suggested that mental imagery was important to perceptual representations. Studies of brain activity while subjects are simply imagining an object (that is not actually present) have shown that the same areas are activated as when the object is actually being viewed (Damasio et al. 1993; Kosslyn et al. 1993; Kosslyn et al. 2003). All of this is consistent with the idea that conceptual understanding is grounded in basic perceptual information. It is also important to note that at least some basic conceptual understanding seems to be instantiated in brains independent of one’s language. For example, Le Clec’H et al. (2000) showed that for bilingual subjects the same specific brain networks for particular concepts (i.e., number processing vs. body-part processing) appeared to be activated regardless of the language used. This is consistent with the idea that languages map onto underlying conceptual networks in the brain. Because of the difficulties of doing brain scanning in awake primates, little work has been done outside of humans with respect to the neural instantiation of concepts. However, there is evidence that grounded cognition holds across species as well (Barsalou 2005). A study of brain activation in Rhesus macaques (Macaca mulatta) showed that species-specific social calls resulted in increased brain activation in auditory processing areas, visual areas, areas of the temporal lobe associated with facial expression and visual motion, and emotionprocessing areas (ventro-medial prefrontal cortex, amygdala, hippocampus), as compared to activation recorded while hearing unfamiliar sounds (Gil-daCosta et al. 2006). Thus, simply hearing socially relevant auditory input led to increased activation in monkeys in a whole range of other areas known to be involved in social information processing. All of this suggests that the more complex a concept is, the greater the number of distinct brain networks will be activated either concurrently or temporally in a causal manner. Thus, we may understand conceptual complexity as a function of the number of differentiable network activation states of a brain (with the understanding that some sets of physically unique network activation states may be functionally equivalent – i.e., will not be differentiable – because they do not make a difference to the organism’s awareness). This notion is congruent with Tononi’s definition of consciousness as being simply the integration of information from different areas of the brain, and therefore a function of

74

P. Thomas Schoenemann

effective connectivity (Tononi 2010), as well as general models of cognition that emphasize network connectivity (e.g., McIntosh 2000). This leads to the likelihood that the degree of network complexity of a species’ brain is proportional to the complexity of their conceptual universe, or their understanding of the world. 6

Species Differences in Network Complexity

Given that concepts are in some fundamental way dependent on brain network activation states, comparisons of the neuroanatomical differences across species, in particular between humans and our closest ape relatives, suggest significant differences in their respective degrees of conceptual complexity. Differences in brain size are one obvious feature of potential relevance to network complexity, but there are a number of correlates of brain size that also suggest important ape/human differences in network complexity, and therefore in conceptual complexity. The human brain is about three times the size of a chimpanzee brain, in absolute terms (Jerison 1973). Although it is true that brain size is correlated with body size across mammals, it is not clear that relative brain size (i.e., brain size corrected for body size) is a better marker of behavioral ability than is simply absolute brain size (Schoenemann 2006). Empirically, in fact, absolute brain size is a better predictor of general cognitive ability than is relative brain size, for primates at least (Deaner et al. 2007). For one thing, larger brains have greater numbers of neurons (Haug 1987, Herculano-Houzel 2012), leading to a greater total number of connections. Interestingly however, the connections do not increase at a rate fast enough to maintain the same degree of interconnectivity between regions (Ringo 1991). In other words, as brain size increases, neural processing in given areas becomes increasingly independent of the processing in other regions (though never completely so, of course). This leads to the increased likelihood of specialization of function of areas, simply as a consequence of increasing brain size. Empirically, the size of a species’ brain predicts the number of anatomically distinct brain regions (which are assumed to be functionally distinct as well, Changizi & Shimojo 2005; Northcutt & Kaas 1995). Estimating from brain size, the number of distinct areas for humans is approximately 150, compared to the estimated number for apes of approximately 100 (Changizi & Shimojo 2005).1 In addition, the number of connections between areas appears to increase as a function of the square of the total number of areas (Changizi & Shimojo 2005). Gibson (2002) has argued such 1

During production of this book, a new article was published arguing that the human brain actually has 180 distinct cortical areas (Glasser et al. 2016). This occurred too late to allow for recalculations of the estimates discussed in this chapter, but would in any case only magnify the expected human/chimp difference in conceptual complexity as proposed here.

A Complex-Adaptive-Systems Approach to the Evolution of Language

75

evolutionary changes in our brain led to a significant increase in the ability to mix “actions, perceptions, and concepts” (p. 10), which she sees as important for language evolution. To give a sense of what a difference in the number of distinct processing areas might mean for differences in degrees of conceptual complexity, assuming it is in some manner a function of degree of the number and types of network activation states, we can look at the increases in possible network activation states made possible by the additional 50 or so cortical areas in humans. First, consider that some concepts, such as individual colors, are likely the function of processing at single areas, rather than interactions among more than one area. The difference between human and ape complexity for concepts like this would be a function of the ratio of distinct cortical areas (using estimates derived from Changizi & Shimojo 2005), or 1.5 times (i.e., 150/100). In other words, everything else being equal, and assuming individual areas in the two species are equally complex in their internal processing (which is not necessarily likely – see later discussion), we would expect humans to have about 1.5 times as many of these basic concepts as do chimpanzees. We can think of this ratio as an estimate of possible increased network complexity, and therefore of possible increased conceptual complexity. Note that a ratio is probably a safer comparison for these purposes than is the absolute increase in numbers of areas, because our knowledge of how processing in individual areas is connected to specific concepts is limited. The use of a ratio for comparison just assumes the process of concept formation is essentially the same across species. Not all concepts are likely the result of processing in single cortical areas, however. Several concepts likely rely on the interaction of at least two distinct areas. For example, our perception of flavor is actually a mix of olfactory (smell) and gustatory (taste) processing, though additional neural systems often contribute to our sensation of flavor as well (Shepherd 2006). To get a crude estimate of the possible complexity of conceptual awareness that could result from combinations of two different areas, we can calculate the possible combinations of sets of two areas given either 150 (human) or 100 (ape) total distinct areas.2 The corresponding ratio of increased complexity in this case would be around 2.3x (11175/4950). This ratio continues to rise as we consider sets of 3, 4, 5, and so on combinations of these areas (Figure 4.1). If concepts for both species can be a function of as many as 11 distinct areas, this would 2

This is calculated in the following way: For a given number of areas assumed per concept, the number of possible combinations (ignoring the order) equals n!/[(k!)(n-k)!], where n = total number of individual areas, and k = number of areas allowed per combination. As one considers larger possible subset sizes, the total number of all possible combinations is the sum of combinations for each value of k up to and including the largest subset size of interest: [total number of possible combinations]=[combinations for k=1]+[combinations for k=2]+ …[combinations for largest subset size].

76

P. Thomas Schoenemann

ratio of distinct combinations for human compared to ape

100 90 80 70 60 50 40 30 20 y = 0.9651e0.4186x

10 0 0

1

2

3

4

5

6

7

8

9

10

11

maximum number of distinct processing areas used to create a given concept

Figure 4.1 Increase in ratio of possible combinations of areas for human vs. ape. Figure 4.1 illustrates how many more combinations of areas there could be, in theory, for humans vs. apes, assuming humans have 150 distinct areas, and apes have only 100. The ratio advantage for human brains increases exponentially as the maximum numbers of areas possible per unique combination of areas is increased (X-axis). These are illustrations of possible differences given a simple model of concept formation, and are not estimates of actual ratios of conceptual complexity in humans vs. apes. See text for details and references.

translate to more than a 100-fold advantage for humans in number of theoretically possible concepts. Caution should be taken in assessing these calculations, of course, as they are based on a number of simplifying assumptions. For example, we don’t know how the maximum number of different areas that could potentially contribute to a concept, so exactly how far out on the X axis of Figure 4.1 we should consider is not clear. It is also entirely possible that larger brains are actually better at integrating increasingly larger numbers of areas, and thus, that some human concepts might be formed from a larger number of areas than occurs in apes. This would mean that we should not be comparing

A Complex-Adaptive-Systems Approach to the Evolution of Language

77

equivalent cortical area set sizes between them, as Figure 4.1 assumes. However, note that this would magnify the probable difference in conceptual complexity between humans and apes. These calculations are not meant to be taken as definitive measures, but simply to highlight the possible effects of anatomical differences between humans and apes on conceptual complexity. In addition to the increased number of cortical areas found in human compared to ape brains, individual areas themselves appear to be larger in humans. Calculating from Changizi and Shimojo’s (2005) regressions of average percentage size of a neocortical area as a function of absolute brain size across mammals, one can calculate that human areas are on average approximately 2.3 times larger, in absolute terms, than what we would expect for corresponding areas in apes. There is substantial variation across particular regions, of course. The human primary visual cortex (V1, or Brodmann area 17) is only 1.6 times as large as that found in chimpanzees (data from Stephan et al. 1981), while Brodmann area 10 of the prefrontal is 6.3 times as large in absolute terms (data from Semendeferi et al. 2001). Does this necessarily translate into increased complexity of processing? One reason to suspect it does comes from Penfield’s (1950) pioneering work mapping the primary motor and somatosensory cortices. He found that there were substantial differences in the amount of cortex devoted to different areas of the body, with areas demonstrating greater sensitivity or degree of motor control having correspondingly larger cortical representations. He strikingly illustrated this with images in which body parts are drawn in proportion to the size of the cortical representation for each area. For example, the lips of these homunculi are very large, consistent with the fact that we have much greater sensitivity to our lips than to many other parts of our body. This indicates that the size of a cortical area does in fact have a functional consequence for primary motor and somatosensory areas within humans. Additionally, some studies have reported that the size of localized areas of the cortex correlate with degrees of ability for behaviors mediated by those areas (e.g., Schoenemann et al. 2000; Thompson et al. 2001). This all suggests that larger cortical areas do in fact correspond to more complex processing of information mediated by those areas in humans. Across species there also appear to be associations between the proportion of cortex that mediates a particular behavior and the degree of elaboration of function of that behavior. For example, Star-nosed moles (Condylura cristata) that live most of their life underground and consequently have very poor eyesight have correspondingly small primary visual cortices, whereas ghost bats (Macroderma gigas) that depend heavily on echolocation devote about half their cortex to auditory processing (Krubitzer 1995). Thus, both across species and within humans, the size of a cortical area appears to be associated with the degree of function of that area (Schoenemann 2010). All of this suggests not only that humans have a greater number of distinct cortical areas with more connections between areas, but also that individual

78

P. Thomas Schoenemann

areas are likely capable of more complex types of processing. This would further magnify the expected difference in complexity, subtlety, and richness of concepts in humans compared to apes – beyond what one might expect, based on differences in the number of distinct areas alone. Finally, during human evolution there appears to have been a biased increase specifically in areas known to be relevant to language processing. For example, the temporal lobe, which is known to mediate connecting concepts to words (Damasio & Damasio 1992), is about 23 percent larger than expected, based on primate scaling trends (Rilling & Seligman 2002). In addition, other areas that participate in language processing, including prefrontal cortex, appear to have increased disproportionately during human evolution (see Schoenemann 2012 for a review). All of this leads to the conclusion that there was a dramatic increase in conceptual complexity during our evolutionary lineage. Given that all of this increase occurred within the context of an intensely socially interactive group existence, it is hard to imagine that this increase in conceptual complexity was not fundamentally important to driving the evolution of language. 7

Conceptual Awareness in Other Species

The behavior of other species is consistent with the notion that they also have concepts, and further, that many of their concepts overlap in important ways with our own. However, an organism’s conceptual understanding of the world will inevitably be influenced by the kinds of sensory information they have evolved to be aware of. Dogs do not have the same range of color information as humans do, but they appear to have better discrimination in low light and better differentiation of shades of grey (Miller & Murphy 1995; Neitz et al. 1989). Echolocating bats can hear frequencies far in excess of humans, and they can use echoes from these sounds to recreate spatial relationships between themselves and their flying insect prey (Jones & Holderied 2007). Elephants can produce and respond to sounds much lower than humans can hear (Garstang 2010). To the extent that species differ in their sensory awareness, their conceptual understanding of the world will likely be different. There is, however, substantial overlap in the types of sensory information that humans have with other species, and the ways in which this information is categorized into conceptual information often appears to match that shown in humans. Experiments with categorization of images by pigeons, for example, suggest they organize visual information into categories corresponding essentially to people and trees that are very close to our own (Herrnstein 1979). Pigeons and monkeys have been shown to be able to correctly categorize still pictures of animals vs. non-animals, which is a fairly abstract concept (Roberts & Mazmanian 1988). Wasserman et al. (1988) found that pigeons learn to

A Complex-Adaptive-Systems Approach to the Evolution of Language

79

categorize images much faster if they are organized into sets that correspond to human language categories (e.g., cats, flowers, cars and chairs), as compared to arbitrary sets. This suggests that human languages and cultures make use of categories that are “real” to a wide variety of animals (Schoenemann 2005). Monkeys appear to have several conceptual categories that correspond at least partly to those found for humans. Monkeys and apes are able to recognize faces of individuals in their groups (Parr 2011). Monkeys also act as if they understand complex social relationships, including hierarchical matrilineal kin groupings, not only with respect to their own positions but also that of others in their social group (Bergman et al. 2003; Seyfarth & Cheney 2000). Several species of monkeys and at least one small ape species (gibbon) have been shown to have different alarm calls for different predators (Clarke et al. 2006; Seyfarth et al. 1980; Zuberbuhler 2000). In order for these primate species to have such calls, they must have separate concepts that relate differentially in some way to each type of predator (Schoenemann 2005). Cheney and Seyfarth (2005) argue these alarm calls are more properly thought of as propositions, as opposed to simple nouns identifying predators. Monkeys and apes (at least) do seem to have concepts that correspond to more than just things. For example, individuals respond differently to others in their social group depending on the context. Apes have been reported to hug other individuals when they are distressed (de Waal 2008). Monkeys respond with reconciliatory behavior after an aggressive encounter with a dominant individual if that individual gives a specific kind of vocalization (Cheney & Seyfarth 1997). These kinds of behaviors show that they are sensitive to the behaviors displayed by others, which suggest that primates have conceptual understanding that differentiates actions from actors (or things). This is important part of the conceptual distinction marked by the verbs vs. nouns in human language grammar. The clearest evidence that apes have concepts comes from ape language studies, which have fairly convincingly shown that apes have the ability to use hand signs or lexicons to represent concepts of various kinds. Double-blind tests show that chimpanzees (Pan troglodytes) can correctly name objects with hand signs (Gardner & Gardner 1984; Gardner et al. 1989). Premack and Premack (1972) showed that chimpanzees could use lexigrams to answer questions about concepts related to objects. For example, when the subject Sarah was asked – via lexigrams – to name the color of “apple” (represented by a blue triangle in her sign system), she responded with the lexigram for “red.” Perhaps the most extensive and impressive work so far with apes has been with Kanzi, a bonobo (Pan paniscus), by Savage-Rumbaugh and colleagues (Savage-Rumbaugh et al. 1993). In one study, Kanzi was presented with more than 660 novel and unusual sentences in controlled blind tests, including things like: “Pour the Coke in the

80

P. Thomas Schoenemann

lemonade,” “Go get the can opener that’s in the bedroom,” and “Kanzi is going to chase Rose,” and he responded correctly to 72 percent of these, which is far above chance given the complexity of the sentences. While there is controversy over the extent to which these show evidence of incipient grammar knowledge (e.g., Wynne 2008; but see Savage-Rumbaugh et al. 2009), Kanzi’s understanding of a few hundred spoken English words – and by extension the concepts underlying them – is not seriously debated. We can get some idea of the range of kinds of concepts that Kanzi appears to understand by looking more closely at the words in the sentences that he responded immediately and correctly to, as listed in Savage-Rumbaugh et al. (1993). Note that this leaves out cases where the researchers nevertheless believe Kanzi understood the sentence but responded imperfectly to it. For example, when Kanzi was told: “Put the carrot in the water,” he picked up a carrot, made a vocalization, took a bite of the carrot, and then put it in the water. The researchers scored this “not immediately correct” because he ate some of the carrot beforehand (Savage-Rumbaugh et al. 1993: p. 162). Restricting our analysis only to sentences Kanzi immediately and correctly responds to also leaves out cases where he was only partially incorrect. For example, when instructed: “Give the big tomato to Liz,” he picked up both the big and the little tomato and gave them to the researcher (Savage-Rumbaugh et al. 1993: p. 163). However, he didn’t pick up any of the numerous other objects that are in front of him, and he didn’t interact with any other researchers present, thereby suggesting that he only misunderstood a single word in the sentence, that is, “big,” and not all the other words in the sentence. Restricting the analysis here just to those sentences to which Kanzi responded immediately and correctly is therefore highly conservative and strongly biased against overinterpreting Kanzi’s understanding of spoken English. In the view of those who work with him, Kanzi almost certainly understands more than is indicated here. In total, Kanzi responded correctly and immediately to 368 of the novel sentences in this study, involving a total of 2,354 word tokens. Parts of speech from these sentences were identified using tools from LingPipe, using the Brown corpus (Alias-i 2008), and then checked by visual inspection. Tables 4.1–4.5 list all the individual words in these sentences, divided into parts of speech categories, with the number of times each word appeared in the sentences also indicated. In total, there were 225 different words used across the sentences in this particular test. This included 119 nouns, 49 verbs, 32 adjectives/adverbs/prepositions, 11 pronouns, and the names of 14 different individuals. (Note that these counts collapse similar words into one instance, e.g., hid, hide, and hiding, are counted as one verb.) The fact that Kanzi responded correctly and immediately to all these sentences is of course not evidence that he necessarily understood every single one, or that he knows the concept adverb, for example.

A Complex-Adaptive-Systems Approach to the Evolution of Language

81

Table 4.1 Nouns Used in Sentences Kanzi Responded to Correctly Number of Occurrences

Nouns (119)

51 32 27 19 18 17 16 14 13 12 12 12 11 10 9 8 7 6

ball outdoors doggie bedroom refrigerator water room rock shot microwave oil potty milk, needles, snake, tomato, toothpaste gorilla, pine, umbrella TV banana, orange, sparklers bowl, collar, colony, juice, lighter, mask, raisins, stick, toy backpack, bunny, can, carrot, melon, monster, mushrooms, opener, play, soap, vacuum apple, cereal, ice, keys, rubber, yard, wipie(s) band, blanket, clay, hat, hose, picture, pillow, potato, teeth, telephone balloon, cabinet, coke, dog, door, hand, head, hotdogs, Jello, lemonade, money, paint, peas, phone, pineapple, shoe book, cleaner, coffee, food, hotdog, knife, orang, paper, tickle bananas, bands, bark, bottle, bubbles, butter, cane, car, clovers, diaper, egg, fire, flashlight, Fourtrax, grapes, hair, hammer, hug, lettuce, mouth, mushroom, onions, outside, oven, Perrier, potatoes, shirt, straw, sugar, surprise, tape, yogurt

5 4 3 2 1

However, Kanzi’s performance on these sentences is strong evidence that, at a cognitive level, he not only understands that arbitrary sound patterns emanating from the researchers’ lips can refer to specific concepts, but also that his underlying conceptual understanding is quite a bit broader than this vocal communication would otherwise suggest. His conceptual world appears not to be limited to individual items or people (nouns), but also includes concepts relating to actions (verbs), and even location (e.g., on, with) and temperature (hot). His performance on tasks like this depends on some non-trivial degree of shared conceptual understanding of the world, along with a deep desire to be a part of a social group in which communication between individuals is important.

82

P. Thomas Schoenemann

Table 4.2 Verbs Used in Sentences Kanzi Responded to Correctly Number of Occurrences

Verbs (49)

97 73 62 58 52 48 15 13 12 11 8 7 6 5 4 3 2 1

put get go/going take can give pour show open, tickle is/are bite hid/hide/hiding could, knife [cut with a knife], make, wash brush, scare chase, grab, see, slap/slapping do, eat, hit, need, tell, throw, want drink, gonna, groom, hug, keep, play, stab bring, carry, close, feed, hammer, let, sit, squeeze, start, think, turn, would

While it is true that Kanzi and other ape language subjects are in highly unique circumstances (for non-human apes) and are not representative of what apes are doing in the wild, they do nevertheless allow us to understand what an ape brain is capable of, given the right sociocultural developmental circumstances. They are our best guess about what the cognitive capacities of our common ape ancestors would have been like. This research shows that animals, and in particular our closest primate relatives, have concepts and can use arbitrary signs to refer to these concepts in interesting and important ways. Table 4.3 Proper Names Used in Sentences Kanzi Responded to Correctly Number of Occurrences

Proper Names (14)

31 21 17 16 6 4 3 2 1

Rose Kelly Liz Kanzi Linda, Sue Matata Panbanisha Karen, Panzee Austin, Krista, Panban, Sherman

A Complex-Adaptive-Systems Approach to the Evolution of Language

83

Table 4.4 Pronouns Used in Sentences Kanzi Responded to Correctly

8

Number of Occurrences

Pronouns (11)

63 45 15 14 5 3 1

you your me it I her, him, them, we my

The Human Elaboration of Conceptual Complexity

While the evidence outlined here suggests that other animals, and in particular apes, have concepts, they nevertheless appear to be limited in the diversity, subtlety and complexity of their conceptual understanding. In his review of ape language studies, Snowdon (1990) remarks: Although the abilities of Kanzi and his companions are remarkable and come very close to some of the capacities shown by young children, there still appear to be limitations. Bonobos [pygmy chimpanzees] and chimps appear to be more limited in the topics that they find interesting to communicate about (p. 222, italics added).

This intuitive assessment is consistent with simple numerical differences in the number of signs that apes appear to be capable of learning compared to the number of words that humans typically understand. The previous analysis of Table 4.5 Adjectives, Adverbs and Prepositions Used in Sentences Kanzi Responded to Correctly Number of Occurrences

Adjectives/Adverbs/Prepositions (32)

451 92 79 54 40 38 22 19 4 3 2 1

the to in on and some that, with a down, out hot, now, of away, big, by, for, this an, at, back, good, hide, if, new, off, over, real, somewhere, sweet, there

84

P. Thomas Schoenemann

the sentences that Kanzi immediately and correctly responded to, for example, suggests he broadly understands the meaning behind at least 225 different words. This is broadly similar to what other ape language studies have reported. For instance, the orangutan Chantek reportedly learned 127 different signs (Miles 1990). By contrast, humans have a working vocabulary that is several orders of magnitude larger. Miller and Gildea (1991) estimate that the average high school student knows the meanings of about 40,000 dictionary entries, and that adding proper names would likely double this number. This suggests there is at least a 100-fold increase in the ability to use arbitrary signs to refer to underlying concepts in humans compared to apes. However, to what extent should we expect these apparent differences in the sheer number of lexical items (or communicative signs) to actually reflect underlying differences in the richness, subtlety, and complexity of conceptual understanding, as opposed to simply reflecting a difference in the ease or ability of attaching arbitrary signs to the underlying conceptual meanings? It is true that Kanzi’s use of signs for communicative purposes reportedly increased at a slower rate than is typical for normal human children (Savage-Rumbaugh & Rumbaugh 1993). However, both Kanzi and his half-sister Panbanisha continued to learn new words into adulthood, in a social context in which words were used by their caregivers (Lyn & Savage-Rumbaugh 2000). Slower learning of communicative signs in apes might reflect a difference in their ability to infer the possible meaning from a complex environment, rather than some specific difficulty associating signs with concepts. However, given the differences in brain anatomy outlined earlier, and their relevance to the possible richness of conceptual understanding, it is likely that a large part of the difference in number of signs for communication highlighted here does reflect a difference in underlying conceptual complexity. To get a visual sense of the possible difference in complexity of the human semantic network compared to that for Kanzi, semantic similarities were calculated on the corpus of Kanzi’s correct sentences (described earlier) and compared to those calculated on an adult human speech corpus: the Charlotte Narrative and Conversation Collection. This is a corpus of 95 narratives, conversations and interviews from residents of Mecklenburg County, North Carolina, and surrounding North Carolina communities, from the American National Corpus Consortium (www.americannationalcorpus.org). The relative semantic similarity between words was calculated using the BEAGLE method (Bound Encoding of the Aggregate Language Environment), which relies on statistical redundancies within a corpus to build a semantic space representation of meaning (Jones & Mewhort 2007). Because the human corpus used here contains around 8,700 words, a single figure depicting the estimated semantic network is too dense to assess; only a subset of the human network can be

A Complex-Adaptive-Systems Approach to the Evolution of Language

85

displayed. As a simple comparison, corresponding human/ape semantic networks were plotted connecting all words estimated by the BEAGLE method to be semantically closer to the word “milk” that is the word “oil.” There were 16 such words in Kanzi’s corpus (Figure 4.2(a), compared to 108 for the human corpus (Figure 4.2(b). As is immediately obvious, the semantic network for these broadly equivalent subsets are dramatically different in the two species, with the human network much richer and covering a much larger portion of possible semantic meaning space. The density of the semantic space implied by these corpora, as estimated by the BEAGLE method, varies quite a bit in different areas, however. Table 4.6 illustrates the relative densities of the same areas of semantic space for Kanzi vs. humans. The numbers represent the total number of words estimated to be closer to the target word than is the comparison word for Kanzi’s corpus and the human corpus. As can be seen from these examples, in some regions the semantic density seems to be much greater in the human corpus than the corresponding area in Kanzi’s corpus (e.g., the space around the target word “apple”), but for the others listed the difference is smaller, and in one case even reversed (e.g., Kanzi’s estimated semantic space has 25 words closer to “eat” than is “get,” but for the human corpus “get” is the closest word to “eat”). This comparison is imperfect for a variety of reasons, and should only be seen as a suggestive first attempt at a numerical comparison of semantic richness between the species. Kanzi’s corpus was taken from sentences specifically selected to assess his understanding of spoken English, and as such were thought to be sentences he had never encountered before. The human corpus represents a sample of spoken English, but obviously does not consist of sentences designed to test knowledge of spoken language. Nevertheless, Kanzi’s corpus is one of the broadest that has been published for apes, and as such this comparison is the best look at species differences in richness of semantic meaning that is currently available. Future work exploring the semantic space between species will hopefully refine the comparisons, but this initial assessment is at least consistent with a dramatic difference in the degree of conceptual complexity, as is predicted by comparisons of brain structure and function between humans and apes. Finally, note that the basic argument here is not inconsistent with Deacon’s (1997; 2012) view that there is a fundamental qualitative difference between humans and all other animals in the types of signs we are able to use. Deacon relies on Peirce’s (1867) taxonomy of types of signs, in which “icons” are signs that directly resemble their referents, “indexes” are signs that correlate reliably with their referents, and “symbols” are signs that refer to their referents only in non-iconic, non-indexical ways, such that they are completely arbitrary conventionalizations. Deacon believes that only humans and language-trained apes like Kanzi truly understand and use signs that are symbols in this sense, and for

Figure 4.2 Comparison of semantic network density for ape vs. human corpora. These figures show a subset of the estimated semantic meaning space for: (a) Kanzi’s corpus of sentences he responded to immediately and correctly, and (b) the Charlotte Narrative and Conversation Collection corpus (see text). These figures plot the meaning space that contains all the words estimated to be as close to, or closer than, “milk” is to “oil.” There were 17 such words for Kanzi’s corpus and 110 in the Charlotte Narrative and Conversation Collection corpus. Semantic meaning space was calculated using the BEAGLE method (see text). Lines between word nodes represent estimates of some level of shared semantic meaning.

A Complex-Adaptive-Systems Approach to the Evolution of Language

87

Table 4.6 Comparative Semantic Web Density for Select Words in Human vs. Chimp Number of Words Closer to Target Word than the Comparison Word for Target Word

Comparison Word

apple milk milk carrot eat

raisin(s) paper oil flashlight get

Human

Kanzi

6747 354 108 267 0

7 40 16 120 25

Numbers in the columns represent the total number of words estimated to be closer semantically to the target word than is the comparison word. Semantic network was estimated using the BEAGLE method. See text for details on the human and Kanzi corpora used for this analysis.

him this is the key difference. But because he agrees that apes can learn to use symbols in particular developmental circumstances, the transition to symbolic behavior is for him fundamentally a behavioral one. Deacon also argues that symbols are built up dynamically from indexical relationships (Deacon 2012). The more complex the underlying conceptual structure, the more potential there would likely be for symbolic behavior in Deacon’s sense.

9

Emergence of Syntax

The evidence laid out here supports the contention that there is a large difference between humans and our closest relatives in the degree of underlying conceptual complexity. Considered in the context of primate interactive sociality, it is hard to believe this would not have had a fundamental influence in driving the evolution of language. As an enhanced communication system, the usefulness of language is partly a function of the usefulness of the underlying conceptual system it is used to convey. While some theorists have suggested that language should not be thought of as primarily a communication mechanism (e.g., Berwick et al. 2013), this is really just a reflection of these theorists’ conflation of underlying conceptual structure (as viewed here) with a definition of language itself. A significant part of the underlying conceptual structure is almost surely pre-linguistic, and even pre-hominin. This is not to say that it is identical in non-human primates, but rather that it formed the basis for language (as a communication mechanism) in the first place. There would not have been any reason to evolve any kind of enhanced communication system

88

P. Thomas Schoenemann

(leading ultimately to language) absent the development of a rich underlying conceptual structure to begin with. What would be the point? It is also sometimes claimed that monkeys have very limited communication systems (e.g., Berwick et al. 2013). Typically, what is actually meant by this is that they only have a few identifiable vocal calls. However, this ignores the tremendous subtlety of primate nonverbal communication (Suomi 1997) and, as a result, is vastly too simplistic a view of primate communication. The reason non-human primate vocal communication systems are so simple is likely that individual survival is not enhanced by extensive vocal communication beyond a few calls specific to their most dangerous predators, as well as the variety of calls that play important roles in social signaling. It is quite clear that primates have very rich understanding of social relationships (Cheney & Seyfarth 2005; de Waal 1982). Cheney and Seyfarth (2005) specifically argue that “upon hearing vocalizations, listeners acquire information about their social companions that is referential, discretely coded, hierarchically structured, rule-governed, and propositional” (p. 135). Given what Kanzi and other ape sign communication projects have demonstrated, it appears to be possible for much (if not most) of this rich underlying conceptual understanding to be coded into some more elaborate vocal communication system, given the right developmental environment, and if it were specifically adaptive to do so. It seems likely therefore that at some point our conceptual system became complex enough that – in the context of an intensely socially interactive existence, and in the relaxation of strong selective pressures against overt signaling (i.e., because risk of predation became sufficiently reduced) – an enhanced, complicated communication system involving grammar and syntax would have been inevitable. This would not, however, have required the evolution of dedicated, innately specified grammar circuits. The complexity of the grammatical system can be seen as an emergent feature of this process (Savage-Rumbaugh & Rumbaugh 1993; Schoenemann 1999). But what evidence is there that human language grammar could actually result simply from increasing conceptual complexity? The details of how this might have happened have not been worked out in detail, but several additional areas of study point in this direction. First, it should be noted that the claim of a clear distinction between grammar and underlying conceptual understanding has been disputed. Some linguists have specifically emphasized the fundamental interconnectedness of grammar and semantics (e.g., Haiman 1985; Langacker 1987; O’Grady 1987). To the extent that these alternative models of language are correct, we would expect increasing conceptual complexity to, in effect, essentially mean the same thing as increasing grammatical complexity. These models fit an evolutionary framework much more elegantly than do those from the formalist tradition. Second, proposed substantive features of Universal Grammar actually look suspiciously like descriptions of how we conceptualize the world, rather than

A Complex-Adaptive-Systems Approach to the Evolution of Language

89

specific rules about how components of language are structured (Schoenemann 1999). For example, Pinker and Bloom (1990) produce a list of “uncontroversial facts about substantive universals, the building blocks of grammars that all theories of universal grammar posit …” (p. 713). One they list is: “Phrase structure rules (e.g., “X-bar theory” or “immediate dominance rules”) force concatenation in the string to correspond to semantic connectedness in the underlying proposition, and thus provides linear clues of underlying structure …” (p. 713). This amounts to simply saying that all languages have rules about how to translate complex, multidimensional conceptual meaning to a linear string of sounds. What is not universal is a particular, specific set of rules that all languages adhere to. Instead, languages differ, sometimes substantially, in what these particular rules look like (which is why Pinker and Bloom are not able to list specific rules shared universally). What this substantive universal amounts to is simply that all languages have some way of coding complex, multidimensional internal conceptual meaning into a serial channel (usually an auditory signal) in a conventionalized way. However, this is exactly what one would expect to occur purely through cultural evolutionary processes alone, if the purpose of language is to communicate meaning between individuals who share a basic conceptual understanding of the world. To take another example, Pinker and Bloom (1990) posit that in all languages “Verb affixes signal the temporal distribution of the event that the verb refers to (aspect) and the time of the event (tense) …” (p. 713). This amounts to saying that all languages have rules that help express when some action occurred (tense), and how it was/is occurring (aspect). If the function of language is to allow communication between individuals sharing the same underlying conceptual understanding, then given that humans share a similar conceptualization of the passage of time, and given that time is highly relevant to humans, one would specifically expect conventionalized rules to emerge through cultural evolutionary processes that would allow speakers to mark this important information for listeners. Assuming one important function of language has been the communication of conceptual understanding, we should expect that the structure of this conceptual understanding would have molded the structure of the grammar. This would be true whether or not the original purpose of language was thinking (e.g., Berwick et al. 2013). Language clearly has been used for communication for a long time and therefore its structure would necessarily be expected to have been significantly molded by shared conceptual understanding among speakers. What is lacking in arguments for the evolution of grammar from the formalist tradition (e.g., Jackendoff 2002; Pinker & Bloom 1990) is the demonstration that cultural evolutionary processes cannot account for the emergence of shared grammatical rules from a common foundation of shared

90

P. Thomas Schoenemann

conceptualizations across individuals played out in the context of individuals in a socially interactive existence trying to share information. Third, empirical studies of human children learning language suggest the development of grammatical knowledge is not actually independent of the development of lexical knowledge. Bates and Goodman (1997) have shown that complexity of grammatical knowledge in children’s speech production is tightly correlated with the development and complexity of their (nongrammatical) lexicon. If grammar were truly an independent system, it should show evidence of being highly decoupled in development, yet it does not. In fact, linguistic development in children has been argued to be essentially item-based: young children use language in a way that suggests they do not understand and use abstract grammatical categories correctly until they have learned the holistic meanings of many isolated phrases first: the “verb-island” hypothesis (Tomasello 2000; Tomasello 2003). For example, in an experimental setting, two-year-old children rarely used novel verbs transitively if they had only been introduced to them in intransitive sentences, and vice versa, even though they understood the holistic meanings of particular transitive and intransitive sentences (Tomasello & Brooks Patricia 1998). This is consistent with the view that children construct their grammatical knowledge by first learning the meanings of whole structures, and then deduce the grammatical rules later based on repeated patterns they experience over being exposed to many independent tokens. This is not consistent with the idea of abstract innate grammatical categories independent of meaning. If this were the case, once a child had knowledge of transitive and intransitive sentences of any kind, they should be able to use all verbs in both ways; yet they do not. There remains debate in the field of language learning, however, about whether or not children could learn all of syntax without at least some of it in some sense being specifically built-in (e.g., Universal Grammar) independent of the lexicon. Gleitman and colleagues (Gleitman 1990; Gleitman & Gleitman 1997; Gleitman et al. 2005), for example, have argued that children make use of syntactic knowledge to help them learn the meanings of words through a process they call “syntactic bootstrapping.” Although they believe that some non-trivial amount of syntactic knowledge is innate in children, this claim is of course logically independent of whether or not syntactic bootstrapping occurs. Such bootstrapping would be expected to occur even if syntactic structure is simply a reflection of underlying shared conceptual understanding. From an evolutionary perspective, it makes more sense to suppose that conceptual understanding has molded language syntax, rather than believing that syntax has an independent evolutionary origin. Lidz, Gleitman and colleagues have published a number of experiments that they believe demonstrate that children have expectations about what syntax looks like that they could not have learned from input. Their studies have

A Complex-Adaptive-Systems Approach to the Evolution of Language

91

focused specifically on learning verb-argument structure (Lidz et al. 2003a; Lidz & Gleitman 2004a; Lidz & Gleitman 2004b) and anaphoric reference (Lidz et al. 2003b). However, an array of non-innatist criticisms have been offered for their findings, including questioning whether the studies actually demonstrate something about innate syntax rather than simply their understanding of word meanings (Tomasello 2004), whether structure is really absent from the linguistic input children are exposed to (Foraker et al. 2009; Regier & Gahl 2004), and whether simple pragmatic constraints would be sufficient (Goldberg 2004). Whether or not some aspects of syntactic knowledge are specifically coded innately, there is broad general agreement among child language learning researchers that syntactic knowledge and lexical knowledge develop together in children. For example, Lidz et al. (2003a) state: “As is now well attested, the verbs of the exposure language are acquired in lockstep with acquisition of those features of the clause-level grammar having to do with the relation between a verb’s semantic argument structure and its syntactic structure” (p. 152; see also Berwick et al. 2013). Thus, both major models of language learning in children agree that lexical and grammatical knowledge are fundamentally linked. Of particular interest to the central thesis of this chapter, Bates and Goodman (1997) review empirical evidence that the development of expressive grammatical complexity appears to be an exponential function of the size of the lexicon, such that grammatical complexity increases very slowly up to a vocabulary of around 200 words, and then begins to accelerate beyond that (Bates & Goodman 1997). In fact, children in the tenth percentile of grammatical complexity for vocabulary sizes of between 200–300 averaged a grammatical complexity score of zero (Bates & Goodman 1997). Note that ape sign communication studies typically report that subjects know about this many signs. It is thus not uncommon for children with vocabulary sizes matching those claimed for apes to similarly also show limited evidence of grammar in their production. A study of Kanzi’s productive communicative sequences (as opposed to his comprehension abilities) when he was five years old showed that they were mostly limited to two words, thus limiting the complexity that could be expected (Greenfield & Savage-Rumbaugh 1991). Even so, some simple patterns were evident. Furthermore, language comprehension also precedes language production in human children, which in turn precedes evidence of grammar usage (Bates & Goodman 1997). Thus, the fact that Kanzi’s comprehension precedes his production, and that his use of grammatical structure in expressions is limited compared to his comprehension of grammatical structure (Greenfield & Savage-Rumbaugh 1991; Savage-Rumbaugh et al. 1993) is not a valid reason to suspect a qualitative difference between apes and humans with respect to language. All of this is consistent with the emergence of grammatical complexity from increasing underlying conceptual understanding.

92

P. Thomas Schoenemann

Last, the possibility of a purely cultural evolutionary transition to grammatical structure as a consequence of expanding lexicon has also been explored in computer simulations. Such simulations necessarily make a number of simplifying assumptions, but they serve as important tests of proof-of-concept, which is particularly important in the case of evolutionary arguments because human intuitions about evolutionary dynamics are often incorrect. Computer simulations of language evolution have generally shown that a surprising degree of emergent structure can occur from cultural evolution alone (for a review, see Gong et al. 2014; Steels 2011). Using an iterated learning agent-based model, Smith et al. (2003) showed that compositionality in communication systems (a precondition of grammar where the meaning of an entire signal sequence is a function of the meanings of subparts of the sequence and their order in the sequence) occurs only when the meaning space that agents are trying to communicate itself exhibits structure. In other words, structure in the communication system emerges from structure in the underlying meaning space (equivalent to conceptual complexity as it is used here). Earlier agent-based simulations (Batali 1998; Goroll 1999) had reported emergent compositionality deriving entirely from cultural evolution, though the effects of the structure of the meaning space were not investigated (e.g., Batali 1998). More recently, Gong (2009; 2011) using an elegant model of interacting agents endowed only with general learning abilities, an ability to create signals of arbitrary type, and a simple interactive social environment, was able to simulate not only the emergence of compositionality but also consistent sequential order of lexical items, again, solely through cultural evolutionary mechanisms. Spranger and Steels (2012) further report the emergence of a communication system displaying incipient hierarchical structure and grammatical marking for spatial information in a simulation of robots playing a cooperative spatial identification guessing game. This occurred even though these features had not been built into the system to begin with. Although no simulation studies have yet reported the emergence of anything approaching the complexity of natural language syntax, the exploration of the effects of increasing conceptual complexity on emergent syntax through cultural evolution has only begun. These initial results are very promising, and suggest a great deal is yet to be learned about the possibility of the evolutionary emergence of grammar from changes in the underlying conceptual system. 10

Conclusions

The model of language evolution presented here suggests that increasing complexity of the meaning space during human cognitive evolution drove the development of syntax and grammar through cultural evolutionary processes. Language complexity is seen as a result of the complexity of the conceptual world

A Complex-Adaptive-Systems Approach to the Evolution of Language

93

it evolved to map and communicate. This model is not only more parsimonious than those requiring separate, independent genetic evolutionary scenarios for both our conceptual system and our linguistic grammatical system, but is also specifically supported by research in a number of areas of study. Our understanding of human brain evolution, when placed in the context of how concepts are instantiated in brains, leads to the conclusion that there has likely been a dramatic increase in conceptual complexity during our evolutionary history. Apparent differences in primate cognition, deriving from both studies of animals in the wild as well as those of captive animals, suggest this is true as well. In particular, it is apparent that other animals (non-human primates in particular) have concepts that meaningfully overlap with some of our own. The difference appears to be one of degree, not of kind. The tight connection between the size of the lexicon and the degree of grammatical complexity in speech production in human children actually fits very well with the data from studies of ape sign communication. Furthermore, several (perhaps all?) universal features of grammar found in languages around the world can be seen as inevitable cultural-evolutionary conventions resulting from a common underlying conceptual system, rather than requiring independent language-grammarspecific genetic constraints. Computer agent-based simulations have begun to investigate possible ways in which this model could work. Future work stands to flesh out these ideas, and demonstrate their full power in explaining the evolution of language. This process highlights language as a complex adaptive system (Beckner et al. 2009). Grammar evolved through cultural evolution, making use of preexisting, non-linguistic general cognitive abilities, and was driven by increasing complexity of underlying conceptual understanding played out in the context of an intensely socially interactive existence. Each of the parts of this equation were (and continue to be) influenced by the others, in a complex interactive feedback system, thus forming a complex adaptive system. Increasing conceptual complexity was itself presumably driven by an increasingly complicated social and technological existence. As social lives got more complicated, leading to new emergent social patterns (such as elaborated kinship systems and social institutions), new forms of conceptual understanding of these emergent patterns followed. Similarly, new technologies created new conceptual understanding. Technological advances, in addition to simply adding conceptually new devices to be named, also changed the social dynamics themselves. Agriculture, for example, dramatically increased population density, which in turn had profound effects on human sociality. More recently, technology applied in the social domain (social media) appears to be in the process of further elaborating and redefining sociality in a variety of interesting ways. The usefulness of language, and grammar in particular, was likely also central to these social and technological developments. Thus, language, conceived

94

P. Thomas Schoenemann

in the broadest sense, itself facilitated further increases in conceptual complexity. Grammar and conceptual understanding influenced each other’s evolutionary trajectories synergistically. They adapted to each other. Although the model outlined here is consistent with what we know about the evolution of brain structure and function, comparative studies of primate cognition, child language learning, and with theories about language itself, the details of how increasing conceptual complexity itself could have led to complexity of language grammar remains to be described in detail. No doubt part of the story involves understanding how grammar itself evolves (culturally), as this suggests ways in which grammar could have emerged in the first place (Bybee 1998). Further work on exactly how children form grammatical concepts from the constructions they hear, and how their conceptual understanding might guide this process, will also be critical. Additionally, if this model is correct, it must be possible to instantiate and probe its dynamics with agent-based computer models of language evolution. The model predicts that elaborating the underlying conceptual understanding of the agents should have profound effects on the grammatical systems that emerge from the simulations. Such simulations would not prove language evolution occurred exactly this way, but simply provide proof-of-concept, and allow for a better understanding of what is possible from this perspective. Unraveling the mystery of language evolution is central to understanding the origin and development of our species. The goal should be to explain as much as possible using cultural evolutionary mechanisms, since this change will occur faster, and will therefore be favored at each evolutionary time-step, compared to biological adaptation (Christiansen & Chater 2008). Because shared conceptual understanding is the foundation that language communication is based on, recognizing the importance of evolutionary changes in this system, and how it would play out over evolutionary time in a socially interactive existence, is likely to be fundamentally important to the explanation. Taking seriously language evolution as a complex adaptive system, involving interactions among many components, is a critical first step.

Acknowledgments The ideas in this chapter have benefited from numerous conversations over the years with William S.-Y. Wang, Vincent Sarich, James Hurford, Morten Christiansen, John Dolan, and John Holland. I thank Salikoko S. Mufwene for initially inviting me to the Workshop on Complexity in Language: Developmental and Evolutionary Perspectives, Lyon, France, May 23–24, 2011, where a sketch of these ideas was presented. Comments by the editors have helped improve the text, though any errors that remain are of course my own.

A Complex-Adaptive-Systems Approach to the Evolution of Language

95

REFERENCES Alias-i. 2008. LingPipe 4.1.0. http://alias-i.com/lingpipe. Baars, Bernard J. & Nicole M. Gage. 2007. Cognition, Brain, and Consciousness: Introduction to Cognitive Neuroscience Amsterdam: Elsevier. Barsalou, L. W. 2005. Continuity of the conceptual system across species. Trends Cogn Sci 9, 309–11. Barsalou, Lawrence W. 2010. Grounded Cognition: Past, Present, and Future. Topics in Cognitive Science 2, 716–24. Batali, J. 1998. Computational Simulations of the Emergence of Grammar. Approaches to the Evolution of Language: Social and Cognitive Bases, ed. by J. R. Hurford, M. Studdert-Kennedy & C. Knight, 405–26. Cambridge: Cambridge University Press. Bates, Elizabeth & Judith C. Goodman. 1997. On the inseparability of grammar and the lexicon: Evidence from acquisition, aphasia, and real-time processing. Language & Cognitive Processes 12, 507–84. Beckner, Clay, Richard Blythe, Joan Bybee, Christiansen Morten H., William Croft, Nick C. Ellis, John Holland, L. C., Jinyun Ke, Diane Larsen-Freeman & Tom Schoenemann. 2009. Language Is a Complex Adaptive System: Position Paper. Language Learning 59, 1–26. Bergman, Thore J., Jacinta C. Beehner, Dorothy L. Cheney & Robert M. Seyfarth. 2003. Hierarchical classification by rank and kinship in baboons. Science 302, 1234–36. Berwick, Robert C., Angela D. Friederici, Noam Chomsky & Johan J. Bolhuis. 2013. Evolution, brain, and the nature of language. Trends in Cognitive Sciences 17, 89– 98. Brodmann, K. 1909. Vergleichende Lokalisationslehre der Grosshirnrinde in ihren Prinzipien dargestellt auf Grund des Zellenbaues Leipzig: Johann Ambrosius Barth Verlag. Bybee, Joan. 1998. A functionalist approach to grammar and its evolution. Evolution of Communication 2, 249–78. Changizi, M. A. & S. Shimojo. 2005. Parcellation and area-area connectivity as a function of neocortex size. Brain, Behavior and Evolution 66, 88–98. Cheney, D. L. & R. M. Seyfarth. 1997. Reconciliatory grunts by dominant female baboons influence victims’ behaviour. Animal Behaviour 54, 409–18. Cheney, Dorothy L. & Robert M. Seyfarth. 2005. Constraints and preadaptations in the earliest stages of language evolution. The Linguistic Review 22, 135–59. Chomsky, Noam. 1972. Language and Mind New York: Harcourt Brace Jovanovich, Inc. 1995. The Minimalist Program Cambridge, M A: The MIT Press. Christiansen, Morten H. & Nick Chater. 2008. Language as shaped by the brain. Behavioral and Brain Sciences 31, 489–509. Clarke, E., U. H. Reichard & K. Zuberbuhler. 2006. The syntax and meaning of wild gibbon songs. PLoS ONE 1.e73. Damasio, Antonio R. & Hanna Damasio. 1992. Brain and language. Scientific American 267, 89–95. Damasio, H., T. J. Grabowski, A. Damasio, D. Tranel, L. Boles-Ponto, G. L. Watkins & R. D. Hichwa. 1993. Visual recall with eyes closed and covered activates early visual cortices. Society for Neuroscience Abstracts 19, 1603. de Waal, Frans. 1982. Chimpanzee Politics Baltimore: Johns Hopkins University Press.

96

P. Thomas Schoenemann

de Waal, Frans B. M. 2008. Putting the Altruism Back into Altruism: The Evolution of Empathy. Annual Review of Psychology 59, 279–300. Deacon, Terrence William. 1997. The symbolic species: the co-evolution of language and the brain. New York: W. W. Norton. Deacon, Terrence W. 2012. Beyond the Symbolic Species. The Symbolic Species Evolved, ed. by T. Schilhab, F. Stjernfelt & T. Deacon, 9–38. Netherlands: Springer. Deaner, R. O., K. Isler, J. Burkart & C. van Schaik. 2007. Overall brain size, and not encephalization quotient, best predicts cognitive ability across non-human primates. Brain, Behavior and Evolution 70, 115–24. Evans, Nicholas & Stephen C. Levinson. 2009. The myth of language universals: Language diversity and its importance for cognitive science. Behavioral and Brain Sciences 32, 429–48. Foraker, Stephani, Terry Regier, Naveen Khetarpal, Amy Perfors & Joshua Tenenbaum. 2009. Indirect Evidence and the Poverty of the Stimulus: The Case of Anaphoric One. Cognitive Science 33, 287–300. Gardner, R. A. & B. T. Gardner. 1984. A vocabulary test for chimpanzee (Pan Troglodytes). Journal of Comparative Psychology 98, 381–404. Gardner, R. A., B. T. Gardner & T. E. van Cantfort. 1989. Teaching Sign Language to a Chimpanzee Albany: State University of New York Press. Garstang, Michael 2010. Elephant infrasounds: long-range communication. Handbook of Mammalian Vocalization; An Integrative Neuroscience Approach, ed. by S. M. Brudzynski, 57–67. London: Academic Press. Gibson, J. J. 1979. The ecological approach to visual perception New York: Houghton Mifflin. Gibson, K. R. 2002. Evolution of human intelligence: the roles of brain size and mental construction. Brain, Behavior and Evolution 59, 10–20. Gil-da-Costa, R., A. Martin, M. A. Lopes, M. Munoz, J. B. Fritz & A. R. Braun. 2006. Species-specific calls activate homologs of Broca’s and Wernicke’s areas in the macaque. Nature Neuroscience 9, 1064–70. Gleitman, L. 1990. The structural sources of verb meanings. Language Acquisition 1, 3–55. Gleitman, Lila & Henry Gleitman. 1997. What is a language made out of? Lingua: International Review of General Linguistics 100, 29–55. Gleitman, Lila R., Kimberly Cassidy, Rebecca Nappa, Anna Papafragou & John C. Trueswell. 2005. Hard words. Language Learning and Development 1, 23–64. Goldberg, Adele E. 2004. But do we need universal grammar? Comment on Lidz et al. (2003). Cognition 94, 77–84. Goldsmith, Timothy H. & William F. Zimmerman. 2000. Biology, Evolution, and Human Nature New York: John Wiley & Sons. Gong, T. 2009. Computational simulation in evolutionary linguistics: a study on language emergence Taipei: Institute of Linguistics, Academia Sinica. Gong, Tao. 2011. Simulating the coevolution of compositionality and word order regularity. Interaction Studies 12, 63–106. Gong, Tao, Lan Shuai & Menghan Zhang. 2014. Modelling language evolution: Examples and predictions. Phys Life Rev 11, 280–302. Goroll, Nils. 1999. Master’s thesis, (The Deep Blue) Nile: Neuronal Influences on Language Evolution. Edinburgh, England: University of Edinburgh.

A Complex-Adaptive-Systems Approach to the Evolution of Language

97

Greenfield, Patricia M. & E. Sue Savage-Rumbaugh. 1991. Imitation, grammatical development, and the invention of protogrammar by an ape. Biological and Behavioral Determinants of Language Development, ed. by N. Krasnegor, D. M. Rumbaugh, R. L. Schiefelbusch & M. Studdert-Kennedy, 235–58. Hillsdale, NJ: Lawrence Erlbaum. Gross, C. G. 2002. Genealogy of the “Grandmother Cell.” The Neuroscientist 8, 512–18. Haiman, John (ed.) 1985. Iconicity in Syntax (Typological Studies in Language). Amsterdam: John Benjamins. Haug, H. 1987. Brain sizes, surfaces, and neuronal sizes of the cortex cerebri: A stereological investigation of man and his variability and a comparison with some mammals (primates, whales, marsupials, insectivores, and one elephant). American Journal of Anatomy 180, 126–42. Herculano-Houzel, S. 2012. The remarkable, yet not extraordinary, human brain as a scaled-up primate brain and its associated cost. Proceedings of the National Academy of Sciences, 109 (Supplement 1), 10661–10668. Herrnstein, R. J. 1979. Acquisition, generalization, and discrimination reversal of a natural concept. Journal of Experimental Psychology: Animal Behavior Processes 5, 116–29. Holland, John H. 1998. Emergence: From chaos to order. Oxford: Oxford University Press. 2006. Studying complex adaptive systems. Journal of Systems Science and Complexity 19, 1–8. Horgan, John. 1995. From complexity to perplexity. Scientific American, 104–09. Hudson, Arthur J. 2010. The evolution of the eye from algae and jellyfish to humans: How vision adapts to environment. Lewiston, NY: Edwin Mellen Press. Jackendoff, Ray. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution New York: Oxford University Press. Jerison, H. J. 1973. Evolution of the Brain and Intelligence New York: Academic Press. Jones, G. & M. W. Holderied. 2007. Bat echolocation calls: adaptation and convergent evolution. Proc Biol Sci 274, 905–12. Jones, M. N. & D. J. Mewhort. 2007. Representing word meaning and order information in a composite holographic lexicon. Psychological Review 114, 1–37. Kosslyn, S. M. 1980. Image and mind. Cambridge, MA: Harvard University Press. Kosslyn, S. M., N. M. Alpert, W. L. Thompson, V. Maljkovic, S. B. Weise, C. F. Chabris, S. E. Hamilton, S. L. Rauch & F. S. Buonanno. 1993. Visual mental imagery activates topographically organized visual cortex: PET investigations. Journal of Cognitive Neuroscience 5, 263–87. Kosslyn, S. M., G. Ganis & W. L. Thompson. 2003. Mental imagery: against the nihilistic hypothesis. Trends Cogn Sci 7, 109–11. Krubitzer, Leah. 1995. The organization of neocortex in mammals: are species differences really so different? Trends in Neurosciences 18, 408–17. Lakoff, George & Mark Johnson. 1980. Metaphors we live by. Chicago: University of Chicago Press. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Stanford: Stanford University Press. Langton, Chris G. 1990. Computation at the edge of chaos: Phase transitions and emergent computation. Physica D 42, 12–37.

98

P. Thomas Schoenemann

Lidz, Jeffrey, Henry Gleitman & Lila Gleitman. 2003a. Understanding how input matters: Verb learning and the footprint of universal grammar. Cognition 87, 151–78. Lidz, Jeffrey & Lila R Gleitman. 2004a. Yes, we still need Universal Grammar. Cognition 94, 85–93. Lidz, Jeffrey & Lila R. Gleitman. 2004b. Argument structure and the child’s contribution to language learning. Trends in Cognitive Sciences 8, 157–61. Lidz, Jeffrey, Sandra Waxman & Jennifer Freedman. 2003b. What infants know about syntax but couldn’t have learned: experimental evidence for syntactic structure at 18 months. Cognition 89, 295–303. Lyn, Heidi & E. Sue Savage-Rumbaugh. 2000. Observational word learning in two bonobos (Pan paniscus): ostensive and non-ostensive contexts. Language & Communication 20, 255–73. McGurk, H. & J. MacDonald. 1976. Hearing lips and seeing voices. Nature 264, 746–8. McIntosh, A. R. 2000. Towards a network theory of cognition. Neural Networks 13, 861–70. Mikulecky, D. C. 2001. The emergence of complexity: science coming of age or science growing old? Computers and Chemistry 25, 341–8. Miles, H. Lyn White. 1990. The cognitive foundations for reference in a signing orangutan. “Language” and Intelligence in Monkeys and Apes: Comparative Developmental Perspectives, ed. by S. T. Parker & K. R. Gibson, 511–39. Cambridge: Cambridge University Press. Miller, George A. & Patricia M. Gildea. 1991. How children learn words. The Emergence of Language: Development and Evolution, ed. by W. S.-Y. Wang, 150–58. New York: W. H. Freeman. Miller, P. E. & C. J. Murphy. 1995. Vision in dogs. Journal of the American Veterinary Medical Association 207, 1623–34. Mitchell, Melanie, Peter T. Hraber & James P. Crutchfield. 1993. Revisiting the Edge of Chaos: Evolving Cellular Automata to Perform Computations. Complex Systems 7, 89–130. Neitz, J., T. Geist & G. H. Jacobs. 1989. Color vision in the dog. Visual Neuroscience 3, 119–25. Northcutt, R. G. & J. H. Kaas. 1995. The emergence and evolution of mammalian neocortex. Trends in Neurosciences 18, 373–9. O’Grady, William. 1987. Principles of Grammar & Learning Chicago: University of Chicago Press. Paivio, A. 1971. Imagery and verbal processes New York: Holt, Rinehart & Winston. Parr, Lisa A. 2011. The evolution of face processing in primates. Philosophical Transactions of the Royal Society B: Biological Sciences 366, 1764–77. Peirce, C. S. 1867. On a new list of categories. Proceedings of the American Academy of Arts and Sciences 7, 287–98. Penfield, Wilder & Theodore Rasmussen. 1950. Cerebral Cortex of Man, a Clinical Study of Localization of Function. New York: Macmillan. Pinker, S. & R. Jackendoff. 2005. The faculty of language: what’s special about it? Cognition 95, 201–36. Pinker, Steven & Paul Bloom. 1990. Natural language and natural selection. Behavioral and Brain Sciences 13, 707–84.

A Complex-Adaptive-Systems Approach to the Evolution of Language

99

Port, Robert F. 2010. Language as a Social Institution: Why Phonemes and Words Do Not Live in the Brain. Ecological Psychology 22, 304–26. Port, Robert F. & Adam P. Leary. 2005. Against Formal Phonology. Language 81, 927– 64. Premack, Ann James & David Premack. 1972. Teaching language to an ape. Scientific American 227, 92–99. Regier, Terry & Susanne Gahl. 2004. Learning the unlearnable: the role of missing evidence. Cognition 93, 147–55. Rilling, J. K. & R. A. Seligman. 2002. A quantitative morphometric comparative analysis of the primate temporal lobe. Journal of Human Evolution 42, 505–33. Ringo, James L. 1991. Neuronal interconnection as a function of brain size. Brain, Behavior and Evolution 38, 1–6. Roberts, William A. & Dwight S. Mazmanian. 1988. Concept learning at different levels of abstraction by pigeons, monkeys, and people. Journal of Experimental Psychology: Animal Behavior Processes 14, 247–60. Savage-Rumbaugh, E. Sue, Jeannine Murphy, Rose A. Sevcik, Karen E. Brakke, Shelly L. Williams & Duane M. Rumbaugh. 1993. Language comprehension in ape and child. Monographs of the Society for Research in Child Development 58, 1–222. Savage-Rumbaugh, E. Sue & Duane M. Rumbaugh. 1993. The emergence of language. Tools, Language and Cognition in Human Evolution, ed. by K. R. Gibson & T. Ingold, 86–108. Cambridge: Cambridge University Press. Savage-Rumbaugh, Sue, Duane Rumbaugh & William M. Fields. 2009. Empirical Kanzi: The Ape Language Controversy Revisited. Skeptic 15, 25–33. Schoenemann, P. T., T. F. Budinger, V. M. Sarich & W. S. Wang. 2000. Brain size does not predict general cognitive ability within families. Proceedings of the National Academy of Sciences of the United States of America 97, 4932–7. Schoenemann, P. Thomas. 1999. Syntax as an emergent characteristic of the evolution of semantic complexity. Minds and Machines 9, 309–46. 2005. Conceptual complexity and the brain: Understanding language origins. Language Acquisition, Change and Emergence: Essays in Evolutionary Linguistics, ed. by W. S.-Y. Wang & J. W. Minett, 47–94. Hong Kong: City University of Hong Kong Press. 2006. Evolution of the Size and Functional Areas of the Human Brain. Annual Review of Anthropology 35, 379–406. 2010. The meaning of brain size: the evolution of conceptual complexity. Human Brain Evolving: Papers in Honor of Ralph Holloway, ed. by D. Broadfield, M. Yuan, K. Schick & N. Toth. Bloomington, IN: Stone Age Institute Press. 2012. Evolution of brain and language. Progress in Brain Research, ed. by M. A. Hofman & D. Falk, 443–59. Amsterdam, The Netherlands: Elsevier. Semendeferi, K., E. Armstrong, A. Schleicher, K. Zilles & G. W. Van Hoesen. 2001. Prefrontal cortex in humans and apes: a comparative study of area 10. American Journal of Physical Anthropology 114, 224–41. Seyfarth, Robert M. & Dorothy L. Cheney. 2000. Social Awareness in Monkeys. American Zoologist 40, 902–09. Seyfarth, Robert M., Dorothy L. Cheney & Peter Marler. 1980. Monkey Responses to Three Different Alarm Calls: Evidence of Predator Classification and Semantic Communication. Science 210, 801–03.

100

P. Thomas Schoenemann

Shepard, R. N. & L. A. Cooper. 1982. Mental images and their transformations New York: Cambridge University Press. Shepherd, Gordon M. 2006. Smell images and the flavour system in the human brain. Nature 444, 316–21. Smith, K., S. Kirby & H. Brighton. 2003. Iterated learning: a framework for the emergence of language. Artificial Life 9, 371–86. Snowdon, Charles T. 1990. Language capacities of nonhuman animals. Yearbook of Physical Anthropology 33, 215–43. Spranger, Michael & Luc Steels. 2012. Emergent functional grammar for space. Experiments in Cultural Language Evolution. Amsterdam: John Benjamins. Steels, L. 2011. Modeling the cultural evolution of language. Phys Life Rev 8, 339–56. Stephan, Heinz, Heiko Frahm & Georg Baron. 1981. New and revised data on volumes of brain structures in Insectivores and Primates. Folia Primatologica 35, 1–29. Suomi, Stephen J. 1997. Nonverbal communication in nonhuman primates: Implications for the emergence of culture. Nonverbal Communication: Where Nature Meets Culture, ed. by U. C. Segerstråle & P. Molnár, 131–46. Mahwah, NJ: Lawrence Erlbaum Associates. Thompson, P. M., T. D. Cannon, K. L. Narr, T. van Erp, V. P. Poutanen, M. Huttunen, J. Lonnqvist, C. G. Standertskjold-Nordenstam, J. Kaprio, M. Khaledy, R. Dail, C. I. Zoumalan & A. W. Toga. 2001. Genetic influences on brain structure. Nature Neuroscience 4, 1253–8. Tomasello, M. 2000. The item-based nature of children’s early syntactic development. Trends Cogn Sci 4, 156–63. Tomasello, Michael. 2003. Constructing a language: a usage-based theory of language acquisition Cambridge, MA: Harvard University Press. 2004. Syntax or semantics? Response to Lidz et al. Cognition 93, 139–40. Tomasello, Michael & J. Brooks Patricia. 1998. Young children’s earliest transitive and intransitive constructions. Cognitive Linguistics 9, 379. Tononi, G. 2010. Information integration: its relevance to brain function and consciousness. Archives Italiennes de Biologie 148, 299–322. Wasserman, E. A., R. E. Kiedinger & R. S. Bhatt. 1988. Conceptual behavior in pigeons: Categories, subcategories, and pseudocategories. Journal of Experimental Psychology: Animal Behavior Processes 14, 235–46. Wynne, Clive. 2008. Aping language: A skeptical analysis of the evidence for nonhuman primate language. Skeptic 13, 10–14. Zuberbuhler, K. 2000. Interspecies semantic communication in two forest primates. Proceedings of the Royal Society of London Series B-Biological Sciences 267, 713–18.

5

Evolutionary Complexity of Social Cognition, Semasiographic Systems, and Language William Croft

1

Introduction

After a long period in which the study of the evolutionary origins of the human language capacity was avoided, some linguists now speculate or use computational modeling to try to infer the evolutionary process that led to modern human language. It is assumed that the evolution of language represents an increase in some sort of linguistic complexity: “pre-language” was less complex than modern human language in some ways, and – probably gradually – acquired the sort of complexity that modern human languages display. Complexity in language is frequently measured in two ways: structural complexity of communicative signals, and the social-cognitive complexity of the interactional situations in which language or language-like communication is used. Some linguists are now attempting to measure structural complexity of contemporary human languages, after a period in which it was assumed that all modern languages were equal in overall structural complexity. There are undoubtedly significant differences among contemporary human languages in terms of the obligatory expression of certain grammatical semantic categories, and the formal morphological complexity of that expression. Nevertheless, all attested human languages display similar degrees of structural complexity in “design features” of language (Hockett 1960). These design features include the combinability and recombinability of meaningful units; some kind of hierarchical structure; recursion (but see Everett 2005, 2009), and duality of patterning. Many of the design features of modern human languages, which presumably evolved from a lesser degree of complexity, are not specific to human language, but rather iconically reflect certain aspects of human cognition, which may or may not be specific to our species (see also Schoenemann 1999). In fact, even the categories of obligatorily expressed grammatical inflections and constructions, although not always obligatorily expressed in every language, are the result of the processes by which we verbalize our experience (Chafe 1977a, 1977b; Croft 2007a); these processes will not be discussed here for reasons of space. 101

102

William Croft

Morphosyntactic complexity largely reflects conceptual complexity. Kirby (2002) argues that in an iterated learning model of language evolution, a grammar made up of recombining, hierarchical, and recursive units can emerge, given only a predicate-calculus-like semantic representation and a preference for an iconic mapping between form and meaning. But it is really the semantic representation and semantics-morphosyntax mapping that brings about the characteristic syntactic structures of modern human languages. The cognitive or conceptual capacities that lead to these morphosyntactic structures are an ability for conceptual analysis and compositionality; conceptual grouping (Langacker 1997); and a preference for diagrammatic iconicity in symbolic (form-meaning) mapping (Bybee 1985; Haiman 1980). Recursive grammatical structures, another type of structural complexity found in most, if not all, modern human languages, commonly occurs in two semantic contexts. Noun-phrase and adpositional-phrase recursion is found in nested figure-ground relations specifying spatial locations (and metaphorical extensions thereof), for example, the bowl in the cabinet above the stove. Sentence recursion is found in certain constructions when events are functioning as modifiers (relative clauses), arguments (verb complements), and adjuncts (adverbial clauses). In particular, mental space builders (Fauconnier 1985) such as believe, want, and hope introduce states of affairs in the “built” mental space – the space of beliefs, desires, hopes, and so on – and these are usually expressed structurally in terms of embedded clauses. So the emergence of recursion presupposes the emergence of these relatively complex cognitive structures. There are in fact other means of packaging these conceptual structures in sentences that do not necessarily involve embedding. But presumably the conceptualization of displaced experience and of other people’s minds is a basic human cognitive ability (see Corballis 2011, esp. chs. 7–9). Also, it should be noted that recursion is not necessarily infinite, as assumed in formal syntactic theories. Spoken language rarely exceeds two levels of embedding (e.g., Croft 1995, 2007b). So the emergence of syntactic embedding may be gradual. On the other hand, duality of patterning does not have a basis in the moreor-less iconic expression of cognitive or conceptual structures. Duality of patterning is defined here as the fact that utterances are made up of combinations of sounds and simultaneously made up of combinations of symbolic units. (See Mufwene 2013 for discussion of the different ways that duality of patterning has been defined.) Duality of patterning is presumably a response to selectional pressures brought about by the expansion of the inventory of signals. Unfortunately it is unclear at what point in language evolution the increase in the number of signals might lead to duality of patterning. I argue in §4.5 that language probably began as a restricted, domain-specific communication system but then evolved to a general-purpose communication system.

Evolutionary Complexity of Social Cognition, Semasiographic Systems 103

A general-purpose communication system has to have the ability to express any situation that arises and any type of entity that comes along. But duality of patterning is a response to the means to express a large number of situations. Duality of patterning could have emerged at a pre-language stage lacking word combination (a one-word stage). Phonological recombination would facilitate production (and remembering) of a larger vocabulary. But duality of patterning could also have occurred at a stage when multiword utterances occur. Multiword utterances reduces the need for a larger vocabulary, but the continued expansion of language to a general-purpose communication system still requires a large vocabulary, and perhaps at this point duality of patterning emerged.1 I have argued here that at least some elements of the structural complexity of modern human languages are the consequence of the cognitive complexity of the conceptual structures being communicated. The leap from conceptual structures to linguistic structures is not an automatic step. The dependence of linguistic structures on conceptual structures only means that the relevant cognitive/conceptual structures must be in place for the formal linguistic structures to emerge. There still remains the mechanism for conceptual structures to be verbalized in language or a pre-language communication system. That is, there has to be a selectional pressure to lead humans to express concepts publicly, with something like language. That selectional pressure is provided by language’s supporting role in the achievement of joint action, which is the foundation of human culture and society. It is only in its social interactional context that the evolution of linguistic complexity can be understood. (Indeed, some might argue that selectional pressures from social interaction led to the evolution of the conceptual structures described earlier, as well as leading to the linguistic structures to communicate them.) That is, the evolution of socialcognitive complexity (in terms of joint action) is a prerequisite for the evolution of structural complexity of linguistic signals. The nature of joint action and language’s role in joint action is described in §2. Joint action presupposes a rich and complex social structure, even if it is “merely” the most basic structure for the emergence of human culture. In §3 I offer some cautionary examples to suggest that the evolution of linguistic complexity may be both gradual and slow. In §4, we look at the nearest sort of communicative system to language whose evolution is documented, namely the evolution of semasiographic systems. Commonalities among the evolutionary paths of different semasiographic systems provide further suggestions about 1

It has been argued that individual speakers do not have that large a vocabulary; e.g., Cheng (2000) has suggested that individuals have knowledge of around 8,000 words. A speech community’s lexical stock is quite a bit larger, but it may imply that there are other selectional pressures leading to duality of patterning than simply the communicative range of a general-purpose communication system.

104

William Croft

the evolutionary path of modern human language. In §5, I offer speculations as to how modern human language may have gradually evolved, based on the observations in §§2–4. 2

Language as Joint Action

Language is fundamentally a joint action. Actually, language is just a part of the process of joint action, but it is language as joint action that determines most of the fundamental design features of language. To describe language as joint action is another way of saying that language can only be properly understood in its social interactional context. The nature of language follows from the part it plays in joint action. The structure of joint actions has been analyzed by social psychologists such as Clark (1996) and philosophers of action such as Bratman (1992); I follow Bratman’s analysis here. Loosely, what makes a joint action joint is that it is more than just the sum of individual actions performed by separate persons; in particular each individual involved must take into consideration the other individual’s beliefs, intentions, and actions in a way that can be described as cooperative. It is this level of cooperation, and the features it entails that are described later, that appears to be a distinctive characteristic of human behavior and has allowed for the emergence of human culture and society (Tomasello 2008). Although great apes engage in some degree of sharing, reciprocity, and collaboration, it appears to be only the first steps beyond essentially competitive social interactions (see Tomasello 2011; Tomasello and Vaish 2013, and references cited therein). Human beings of course engage in all sorts of actions, including actions that involve other human beings. However, an action involving more than one person must satisfy certain conditions in order to count as a joint action – in Bratman’s terms, shared cooperative activity. Each participant, for example each of two people performing a musical piece, intends to perform the joint action, that is, their intention extends beyond just their own individual actions as part of the joint action. The joint action is intended by each person to be performed in accordance with and because of their meshing subplans: that is, the overall intention of performing a musical piece requires each individual to have subplans for contributing to the overall joint action that mesh with the other’s subplans. Neither individual is coercing the other; the cooperation must be voluntary. Each participant is also committed to mutual support: that is, if the other participant falters in some way, the first participant will step in to help the other perform her part and thereby allow the joint action to be successfully achieved. Two other conditions are essential. All of these intentions and commitments on the part of the participants are shared knowledge among them, or as I will say following Clark (1996), it is part of the participants’ common ground. Finally,

Evolutionary Complexity of Social Cognition, Semasiographic Systems 105

in the execution of the joint action, there is mutual responsiveness of the participants. That is, in executing the joint action we will coordinate our individual actions in order to ensure that they mesh with each other in execution and hence the joint action will be successfully carried out (to the best of our abilities).2 Coordination is essential in carrying out joint actions successfully. Yet coordination of our individual actions to achieve a joint action is a problem, because of another fundamental characteristic of human beings: we cannot read each other’s minds directly or literally; we can only interpret the external behaviors of others. This is not always a bad thing, as a moment’s thought will demonstrate. But much of the time, we want or need to carry out joint actions, so we must find a way to overcome this problem. The means to solve coordination problems are coordination devices (Lewis 1969). There are a number of coordination devices that people use to solve coordination problems. One of the most effective coordination devices is communication: we communicate to each other our intentions, thereby making them part of our common ground and allowing us to carry out our joint action. What precisely is communication? According to Grice, communication happens when one of us recognizes the other’s intentions. Communication is itself a joint action, however, as its Latinate etymology (comm¯unic¯are ‘to make common’) implies. One person has to signal her intention and the other person has to recognize the intention. In other words, communication solves one coordination problem, but poses another coordination problem in order for communication itself to succeed. Fortunately, there is an effective coordination device for communication, although it must satisfy specific conditions in order to be usable. This is convention, which has been analyzed by both Lewis (1969) and Clark (1996). The Lewis and Clark definition of convention has five parts: (i) a regularity in behavior, (ii) which is partially arbitrary (iii) and is common ground in a community, (iv) which serves as a coordination device (v) for a recurrent coordination problem. For example, shaking right hands is a regularity in behavior that is partially arbitrary (one could shake left hands, or kiss on the cheek, etc.), and it is common ground in the American community that shaking hands serves as a coordination device for the recurrent coordination problem of greeting someone. Many human behaviors are conventional. In addition to conventions for greeting someone, there is the convention of driving on the right, or left, side of the road and the convention of paying for a meal before, or after, eating it. And language is a vast inventory of conventions. Every word and construction is a regularity in behavior – producing those words and constructions in utterances – that solves a recurrent coordination problem, namely communicating 2

The concept of coordination in joint action is unrelated to the concept of coordination in syntactic structure.

106

William Croft

a particular intention or meaning. So language involves conventions for communication for the achievement of joint actions. Powerful as convention is as a coordination device, it is not enough. Convention can work only under certain circumstances. A convention must become established in a community, that is, become common ground in that community. Hence it cannot be the first use of this coordination device (the particular linguistic form) for this coordination problem (the particular meaning intended). For a convention to be established, the behavior must be used first to solve the coordination problem via nonconventional coordination devices (Lewis 1969). One type of nonconventional coordination device is precedent: a behavior serves as a coordination device because it served before and was successful. For example, a new word was coined by a commentator in The Economist in its Lexington column in the May 10, 2003, issue: “Laugh at Bill Bennett, the erstwhile virtuecrat, but don’t forget his message. …” Assuming that the reader understood what the commentator meant, he uses the new word again later in the column: “Who needs satire when you have the social conservatives? … Now Bill Bennett, the capo di tutti capi of the virtuecrats, has been caught … with his hand glued to the slot machines of Las Vegas and Atlantic City.” Precedent is not yet convention. It operates on the principle of “Hey, it worked once, let’s try it again!” Although it is logically possible that precedent does not even require shared knowledge of the precedent between the individuals, the definitions of precedent found in the literature on coordination devices (Lewis 1969: 36; Clark 1996: 81) are formulated as shared precedent: it worked for us once, so it might work for us again. I believe that the concept of shared precedent is the one that is relevant to human behavior and human evolution (see §5). Coordination via precedent, of course, requires the precedent, and that precedent has to succeed via some other coordination device. And at this point we reach the grounding of all joint action: joint salience, made possible by the existence of joint attention among human beings (Tomasello 1999). For example, while I was driving in the rain a few years ago, my mother, who was a passenger, said, “You need more wiper” – a novel use of the count noun wiper in a mass noun construction, not a convention of English. Her novel coordination device (more wiper) succeeded because our joint attention was focused on the contextually salient fact that the windshield was becoming obscured because the wiper wasn’t going fast enough, and on our shared knowledge that one can adjust the speed of the wiper. Joint salience is required for all utterances, for at least two fundamental properties of utterances (Clark 1996). First, almost every word in every utterance has multiple senses, leading to denotational ambiguity (if this is doubted, just examine a comprehensive dictionary like the Oxford English Dictionary), and that ambiguity has to be resolved “in context,” that is, via joint salience.

Evolutionary Complexity of Social Cognition, Semasiographic Systems 107

Second, almost every utterance is communicating a specific event happening to a specific person or thing: I broke the teapot is about a specific speaker, a specific teapot, and a specific event at a specific time (expressed by the past tense verb form). Yet the words used – I, break, teapot – are linguistic conventions for general types, not specific tokens of the type. Using a word for a general type to refer to one specific individual and not another only succeeds “in context,” that is, via joint salience. Linguistic convention is not yet enough in another way. Words and constructions must be realized in a jointly perceptually salient form: auditory, as in speech, or visual, as in signed language or writing. Only at this point is language fully grounded in joint salience. Thus, the full joint action using language has four levels (Clark 1996) (the paired actions are a reminder that all levels are joint): r proposing and taking up a joint action, via r signaling and recognizing the communicative intention, via r formulating and identifying a conventional proposition, via r producing and attending to a perceivable utterance. Central to joint action and to convention, including linguistic convention, is common ground: shared knowledge, beliefs, and attitudes of members of the community. Common ground requires a shared basis (Clark 1996). Some common ground is shared between persons because of common experiences in their face-to-face interaction: things they have seen or done together. The shared basis for that common ground is joint perceptual salience. Other common ground is shared between persons because of what they have communicated to each other over the course of their interactions, even if it wasn’t experienced together. The shared basis for that common ground is the actional basis – their communicative acts, including linguistic acts. Like the shared basis of joint perceptual salience, the actional basis of shared communication requires direct interaction between the persons. Clark calls common ground founded on perceptual and actional basis personal common ground. A third basis for common ground does not require face-to-face interaction. We share some knowledge simply by being members of the same community: chess players, birders, environmentalists, Mormons, linguists, University of New Mexico employees, and so on. By virtue of membership in these communities, we have shared expertise that serves as a communal basis for common ground. For instance, as chess players, we can assume knowledge of the rules of chess and many strategies and moves in playing the game. But that is stating it a bit too glibly: where does that shared expertise come from? Shared expertise is also not a state, but a process as well: the communities we belong to are communities of shared practices. Wenger (1998) defines communities of practice as possessing mutual engagement (i.e., joint action), a joint enterprise (i.e., the purpose for the joint action), and a shared repertoire

108

William Croft

(i.e., commonly performed joint actions to carry out the joint enterprise). Shared expertise emerges from communities of practice. Conversely, communities persist and expand via the transmission of shared practices and shared expertise to new members. Even personal common ground is ultimately grounded in a process: the shared experiences and communications between the persons. Language is joint action, but it is really part of a joint action that extends far beyond what linguists usually think of as language. Two persons wish to engage in a joint action. This requires a shared intention, meshing subplans, absence of coercion, and mutual support. Above all, it requires coordination of the individual actions of the persons. Coordination of a joint action can be achieved at least in part by joint salience. For example, two musicians look at each other to coordinate their playing. The joint action itself, like its coordination, requires common ground, which also emerges from joint salience but also a history of shared practice in the community to which the participants belong. Much joint action, certainly many sophisticated ones, use communication for coordination. Again, joint salience can facilitate communication. But convention, particularly linguistic convention, is highly effective and widely used – though only after joint salience and precedent have allowed the establishment of the convention in the speech community, and only when supplemented by precedent and joint salience. Finally, linguistic convention must ultimately be realized in a perceivable utterance, whose joint perceptual salience allows the linguistic convention to successfully serve its purpose. Language is only one part – albeit a very useful part – serving a much broader process. 3

The Evolution of Linguistic Complexity

The description of joint action and social cognition in §2 describes joint action in modern human social interaction. It looks very much like a process that is tightly integrated, where each element depends on every other element. It seems that once the basic elements of joint action have evolved – common ground, helpfulness, and so on – all the rest will emerge automatically, and modern human language will therefore emerge with it. It is a little easier to conceive of a gradual evolution of the complexity of the signal described in §1 (and perhaps this is why so much attention has been focused on it). But one cannot separate the evolution of the complexity of the signal from the evolution of social cognitive complexity. Nevertheless, we shouldn’t overestimate how rapidly or suddenly a complex suite of social cognitive skills can evolve. Examples from the evolution of semasiographic systems – which encode information in a lasting, visual medium, of which more is discussed in §4 – and the ontogenetic development of social

Evolutionary Complexity of Social Cognition, Semasiographic Systems 109

cognition show that hindsight is too powerful to form the basis of an evolutionary scenario. For example, it is universally agreed that humans had fully modern language by fifty thousand years ago, if not much earlier. Around that same time (fifty thousand years ago if not earlier; see for example d’Errico et al., 2001), representational art appears in the archaeological record, including artistically quite sophisticated representations. One might think that putting the two together – auditory human language and visual representation – is not an enormous next step. Yet although all human societies have modern human language and representational art, it took tens of thousands of years before the first writing systems emerged; and the evolution of writing systems happened independently twice or at most four times, out of the thousands of human societies that must have existed five thousand years ago. Clearly, putting the two together for the first time was not easy at all. The enormous time lag was probably due to the absence of selection pressure, which probably arrived in the form of large-scale stratified societies where writing emerged. Even so, the actual emergence of writing appears to have been a gradual process where it is documented (see §4.2). Another far more mundane example illustrates the difficulty and gradualness of evolution of semasiographic systems, which probably is characteristic of all human invention. Musical notation, another semasiographic system that will be described in §4.3, had evolved considerably by the seventeenth century. This included representation of notes of different time values (longer and shorter), conventionally described as half-notes, fourth-notes, and so on. By the seventeenth century musical notation represented irregular note lengths, such as a note lasting for three-sixteenths followed by a single sixteenth note (3 + 1), by adding a dot after the longer note: in this case, an eighth note followed by a dot denoted three-sixteenths (“one and a half eighth notes”), and it was followed by a sixteenth note. At the same time, a larger note value could be divided into three shorter notes instead of two notes or four notes. But it wasn’t until after the middle of the nineteenth century that a notation was devised for describing a 2 + 1 combination, even though composers wrote music with 2 + 1 note values. Instead the 3 + 1 notation was used (for example, in music as late as Chopin’s Polonaise-Fantaisie, published in 1846). Even though there was selectional pressure to fix this notational problem – composers produced both the 2 + 1 and 3 + 1 note value combinations frequently in their compositions – it still took some two centuries for it to be solved. The moral is that even when the cognitive parts are there – that is, they are part of human cognitive ability at that point in human history – putting the parts together to produce a new behavior does not happen automatically. In fact, it may take a long time before it even starts to happen at all. There has to be a reason for it to happen, that is, a selectional pressure in evolutionary terms – and even then, it takes a long time, even multiple human generations, for the new

110

William Croft

behavior to emerge and be propagated in the society. The selectional pressure is most likely to be social, not purely cognitive. An example that suggests that social cognitive evolution is also gradual is the ontogenetic development of the human conceptualization of others. Tomasello argues that in ontogenetic development, a “revolution” occurs at around nine months of age, which leads to the ability of the infant to achieve joint attention with other human beings (Tomasello 1999: 61–70). But other types of mental reasoning that are built on the notion of other minds having beliefs other than our own (called the “theory of mind”) take several further years to develop (e.g., Tomasello, Kruger, and Ratner 1993: 499–500). The conclusion we can draw from these various examples is that it is eminently plausible to infer that the evolution of social cognition as described in §2 took some time, possibly a very long time – hundreds of thousands if not some millions of years. Tomasello suggests that although primates and some other animal species appear to engage in joint cooperative activity, the apparently “coordinated” behavior may be due to individual responses to causal inferences based on practical reasoning about conspecifics’ intentions and goals (Tomasello 2008: 44–49, 173–85). Hence, it may have taken up to five million years (from the branching off of the hominin line to no later than fifty thousand years ago) for modern human social cognition to have evolved. Unfortunately, there is no direct empirical evidence regarding intermediate stages of complexity in human cognition between primate group activities and signals and modern human joint actions and language. This is the reason that scholars avoided the question of modern human origins until recently. In the next section, I will look at the evolution of semasiographic systems (the nearest language-like systems for which we have direct empirical evidence) to see what they might tell us about stages in the evolution of language (cf. Coupé 2012, who compares the development of language to the development of photography). After taking a glance at language acquisition I will speculate on the evolution of language from pre-language in §5, using the logical priority of different elements of the model of joint action described in §2 in combination with the insights from the evolution of semasiographic systems. 4

The Evolution of Semasiographic Systems

Semasiographic systems are used for “the communication of relatively specific ideas in a conventional manner by means of permanent, visible marks” (Boone 1994: 15; 2004: 313). Semasiographic systems include images, numerals (counting), maps, calendars, bookkeeping systems, tables, mathematical notation, music and dance notation, graphs, and, as a special case, writing. As noted earlier, semasiographic systems have been around for at least fifty thousand years, with the first production of images and statues. Semasiographic

Evolutionary Complexity of Social Cognition, Semasiographic Systems 111

systems are similar to language in several respects. As noted before, they encode information, as linguistic expressions do. More specifically, they encode information in such a way that it can be shared, just as language involves sharing individuals’ intentions. Semasiographic systems employ, or evolve to employ, relatively complex information encoded by recombinable units, also like modern human language. Most useful for the goals of this chapter, semasiographic tokens survive and leave a record, allowing us to observe stages of their evolution in at least some cases. In this section, I will briefly describe salient properties of the evolution of number systems, writing systems, musical notation, and dance notation, and offer some generalizations over the evolution of these semasiographic systems. 4.1

Numbers

Number appears to be the earliest semasiographic system after images. Artifacts from ten to twenty thousand years ago, such as the Ishango Bone (Marshack 1991: 22; Rudman 2007: 62) appear to involve tallies, the earliest type of number notation.3 In tallies, the number of repeated symbols (such as strokes or notches) denotes the equivalent number of entities. That is, tallies represent an additive number representation system (Pettersson 1996: 796). Additive number representations are essentially iconic: the number of signs reflects the number of objects being counted. Another example of an additive numeral system is represented by the undecorated tokens found in Middle Eastern sites dating back to 8000 BC and apparently used to count goods (Englund 1998: 46). A later development is the employment of distinct signs for larger numbers, such as 10 and 60, which effectively function as bases in the numerical semasiographic system. This development occurs in Mesopotamia at around the same time as protocuneiform around 3200–3100 BC (see later; Nissen et al., 1993: 25). Pettersson (1996: 796) calls these sign-value systems, the earliest of which are also additive, repeating each sign to express multiples of the sign value. Familiar examples are the later Roman numerals, such as XXXIII for 33. However, the signs for each value are largely arbitrary: although signs for larger values are sometimes larger in size, or require more strokes, than signs for smaller values, the actual numerical value denoted by the sign is arbitrary. One striking fact (to modern eyes) about the Mesopotamian sign-value notation is that the numerical signs varied depending on what is being counted or measured, and the same sign may have a different numerical value depending on what is being counted or measured (Nissen et al., 1993: 25–27). For 3

Claims that the marks represent more complex number notation are highly speculative (Rudman 2007: 62–65).

112

William Croft

example, the sign resembling a dot represents 10 of the smaller units for discrete objects, 6 units for dry measures of cereals, and 18 units for surface measures (ibid., 131). This is an instance of the high degree of context-dependence that is commonly found in the earliest stages of semasiographic systems. A still later development of number notation is a positional system (Pettersson 1996: 796), in which the position of the sign determines its value by specifying a base amount that it multiplies. In a sign-value system, one must keep creating new symbols for larger quantities, but in a positional system, the only symbols necessary are the values up to the base (e.g., 10 in the modern decimal system); the position of the symbol determines the magnitude of the number. In Mesopotamia, a positional system evolved around 2000 BC (Pettersson 1996: 798) – several thousand years after the appearance of undecorated tokens, and more than a thousand years after the emergence of a signvalue notation system. The counting board or abacus, which dates back at least to seventh century B.C. Greece (Menninger 1969: 301), and the quipu system of knots, used in the Inca empire (fifteenth century AD; Menninger 1969: 252), also use the positional method to represent numbers. (Interestingly, the contemporary ancient Greek written numerals did not make use of the positional system found in its counting boards.) The evolution of Mayan numeral notation also apparently proceeded through the same stages: additive, then sign-value (with different glyphs for powers of 20), then positional (Rudman 2007: 118; he uses the term “additive” for signvalue systems). The Ancient Greeks used a sign-value system adopting written letters for signs (Pettersson 1996: 803), illustrating another characteristic of semasiographic evolution that we will encounter in other systems, namely, the adoption of signs from one semasiographic system for use in another semasiographic system. Hindu-Arabic numerals appear not to have been adopted from another semasiographic system. The earliest system (ca. 200 BC) was partly additive (1–3 were one to three strokes), and partly sign-valued; a positional system emerged around 600 AD (Pettersson 1996: 804). Again, the modern Hindu-Arabic numeral system is the result of a gradual evolutionary process. 4.2

Writing

Writing essentially involves the adaptation of a semasiographic system in order to represent spoken language. Writing evolved independently at least twice. Undoubtedly independent are its evolution in Mesopotamia and Mesoamerica. Egyptian writing evolved at the same time as Mesopotamian writing. It was earlier thought to be inspired by Mesopotamian writing, but its structure is so different, and the timing so uncertain, that it is now thought more likely to have evolved mostly if not entirely independently (Baines 2004: 175–76). The extant fragments of earliest Chinese writing – oracle bone inscriptions and clan names on bronze vessels from ca. 1200 BC – are a full-fledged writing system in

Evolutionary Complexity of Social Cognition, Semasiographic Systems 113

structure (Bagley 2004: 198). There are no clear precursors in the archaeological record (ibid., 190); so the early evolution of Chinese writing is unknown. (It is possible, though now considered less likely, that Chinese writing developed via diffusion from the Near East.) By far the best documented evolution of writing is from Mesopotamia, because writing was mostly done on clay. The emergence of the writing system probably ultimately began with undecorated stone tokens referred to earlier. By 3300–3200 BC, these undecorated tokens were enclosed in clay envelopes impressed with seals, presumably used for accounting purposes (Englund 1998: 48). Seals were semasiographic representations of specific persons (or institutions such as a temple), that is, they denoted names of individuals. Thus the clay envelopes combined two semasiographic systems, the tokens counting goods and the seal impression designating the person or institution responsible for the goods. This is an example of a multimodal semasiographic system. Some of the clay envelopes were also impressed with signs representing the number of tokens in the envelope and their form. There also occur numerical tablets, tablets with numeral signs (and seal impressions) but not enclosing tokens that they designate. These devices – sealed clay envelopes containing tokens, envelopes containing tokens but impressed with the number and type of tokens, and numerical tablets – are attested at around the same time and may reflect either a sequence of developments or alternative systems for accounting for goods. The envelopes bearing external signs illustrate another characteristic in the emergence of semasiographic systems: the often direct association of the signifier with the signified. The next step occurred around 3200–3100 BC (the Late Uruk period and Uruk IV writing phase), in both Uruk and Susa (the seat of the ancient Elamite kingdom). In this period are found tags – small tablets – with ideographic signs that may represent personal names (Nissen et al., 1993: 20), though some ideographs may denote the goods (Englund 1998: 57). In addition, there are numero-ideographic tablets, containing seal impressions, numeral signs, and ideographic signs that designate different types of goods (Englund 1998: 51– 53). The numero-ideographic tablets mark the beginning of proto-cuneiform. This step involved the combination of a largely preexisting set of numerical signs combined with a largely preexisting set of ideographic conventions (Cooper 2004: 77).4 The number of ideographic signs quickly expanded to 4

Citing Buccellati (1981), Cooper (2004: 77) considers the ideographs to derive from glyptic art. Schmandt-Besserat (1996) argues that the proto-cuneiform ideographs derive from decorated tokens, which contrast with the undecorated tokens referred to earlier. However, the decorated tokens are not found enclosed in the clay envelopes whose markings appear to have evolved into the precursors of proto-cuneiform, namely the numerical tablets (Englund 1998: 47). Englund suggests that the iconic nature of the proto-cuneiform pictographs and the decorated tokens are instances of convergent evolution, rather than derivation of the former from the latter (Englund 1998: 47 [fn. 92], 53).

114

William Croft

some 900 signs (Englund 1998: 68; Cooper 2004: 68), while the contextsensitive system of sign-valued numerals described previously constituted around 60 signs. The proto-cuneiform signs are almost all pictographic, that is, iconic, in nature (Englund 1998: 71) or at least iconic-indexical (i.e., pictographic representation via metonymy and synedoche; Cooper 2004: 84), although some are abstract representations that had been in use for some time. It is doubtful whether the proto-cuneiform tablets represent “writing” in the sense of reflecting in graphic form a spoken language. Both numerals and ideograms do not denote words in a specific language in the way that a writing system that records phonetic content, even only partly so, does. The protocuneiform tablets do not match spoken language syntax but “the ‘grammar’ of the archaic accountants’ syntax” (Englund 1998: 63; 79). The information is structured instead by the shape of the tablet, the ordering of cells in the tablet, and other nonlinguistic properties (Cooper 2004: 80–81). In other words, writing originated as a semasiographic system for purposes other than to record spoken language: …no writing system was invented, or used early on, to mimic spoken language or to perform spoken language’s functions. Livestock or ration accounts, land management records, lexical texts, labels identifying funerary offerings, offering lists, divination records, and commemorative stelae have no oral counterparts. Rather, they represent the extension of language use into areas where spoken language cannot do the job (Cooper 2004: 83).

One did not find written royal inscriptions in Mesopotamia until around 2700 BC, in literature until around 2600 BC, and in letters until around 2400 BC (Cooper 2004: 83). In other words, the semasiographic systems that evolved into writing originated in specific domains of use, such as the accounting functions of early Mesopotamian writing. The signs then came to be construed as denoting word-like concepts (the semasiologographic systems of Trigger 2004: 47), and eventually words in a specific language. Thus, it is not surprising that proto-cuneiform did not have a language-driven syntax, and did not express grammatical morphemes. Indeed, it has been suggested that perhaps proto-cuneiform was not language-specific because ancient Uruk was a polity inhabited by speakers of different languages, although the Sumerians were the dominant group at first (Englund 1998: 73–81). It was not until around 2400 BC that the “spoken language determined the order of the script” (Nissen et al., 1993: 123), and it took until the end of the third millennium BC for grammatical morphemes to be written (ibid., 116): “Full writing was the byproduct of a gradual discovery of new applications” (Bagley 2004: 233). Over time, the signs evolved from a more iconic to a more arbitrary symbolic form (i.e., no longer iconically representing the concept denoted by a word, not its phonetic form). This was in part due to the Mesopotamian writing

Evolutionary Complexity of Social Cognition, Semasiographic Systems 115

technology, namely the use of the stylus on soft clay, which made it difficult to draw curved lines. The origins of writing in Egypt have some similarities to and some differences from the origins of writing in Mesopotamia (Baines 2004; see also Stauder 2010). Its earliest known uses do not seem to be entirely economic. The earliest known ideographic objects are bone tags found in Tomb U-j in the Umm al-Qa’ab cemetery near Abydos (no later than the Naqada III era, ca. 3200 BC; Baines 2004: 153, 154). The tags bear numerals or “representationally based signs,” but normally not both at once, unlike the numero-ideographic tablets of contemporary Mesopotamia (ibid., 157). The tags were probably attached to goods, that is, the signifier is associated with the signified. The tags (apart from the numeral tags) are very likely to denote names, perhaps of prestigious entities (ibid., 164). A small number of pictographic (iconic) signs are attested, mostly different from later hieroglyphics. There are also jars, probably containing valuable oils or fats, with large painted signs similar to those on the tags (but no numerals; ibid., 157, 159). There is nothing language-specific about this semasiography, not unlike proto-cuneiform. The slightly later Hunters’ Palette (Naqada IIIa) and Scorpion Macehead (Dynasty 0) have similar signs (ibid., 168–69). In Dynasty 0 (before the beginning of the third millennium BC) the number of signs greatly expands to more than a thousand, again not unlike proto-cuneiform, before being consolidated to a few hundred by the middle of the third millennium (ibid., 172). In Dynasty 0, one finds larger tags that combine pictorial representation and linguistic (or at least semasiologographic) forms (ibid., 173) – another sort of multimodality. The function of the writing system expanded to other administrative functions in the 1st Dynasty (ca. 2950–2775 BC), but “continuous discourse and full syntax were not notated for another couple of centuries” (ibid., 174), again like Mesopotamian writing. As noted earlier, little is known about the early evolution of the Chinese writing system. The earliest writings are already linguistic in the way that the earliest Mesopotamian, Egyptian, and Mesoamerican systems are not. For example, the signs are fairly arbitrary in form, although more iconic than later versions of the signs; literate Chinese speakers retain an awareness of their original iconicity (Cooper 2004: 89). Mesoamerica represents an undoubtedly independent origin of writing systems, but the earliest history of writing in Mesoamerica is very complex and fragmentary. Part of the complexity is that the best-documented writing system, that of the Maya, is in fact a later adaptation of earlier writing system(s) used for other languages. All of the earliest signs have resisted decipherment, or their decipherment is highly controversial. The earliest occurrence of glyphs that look like later writing signs are found on Olmec carved heads dating from 1150–900 BC (Houston 2004: 288). The glyphs on the heads probably represent

116

William Croft

names of the individuals represented by the heads (ibid., 289). The direct association of signifier and signified was not decoupled until 900–600 BC, when a stela is found with the glyph next to, not directly on, the headdress of the head (ibid., 290). By ca. 500 BC there is a monument, La Venta Monument 13, which contains writing, defined by Houston as representing language, detached from its signified, and occurring in linear sequences (ibid., 292). Nevertheless, the writing occurs on stelae denoting individuals who are probably named by the writing, as in the case of San José Mogote Monument 3 (ibid.), where a captive appears to be named by a calendrical date. This last monument illustrates a common Mesoamerican association of numerical, specifically calendrical, information with a logographic (or semasiologographic) system – not unlike the origins of Mesopotamian writing in a combination of numerical and semasiologographic signs (viz. the numero-ideographic tablets). The Isthmian (or “Epi-Olmec”) texts from 36 BC–162 AD are also associated with calendrical information (hence the precise dating; ibid., 296–97). The Mayan scripts are adapted from the earlier Mesoamerican writing systems for the Mayan language(s). The scripts evolved over time, which makes decipherment of the fragmentary earliest texts difficult even though much is known about the more widely attested Classic Maya writing system. The earliest Mayan texts are lists of deities and name tags (ibid., 304), not unlike the earliest Mesopotamian and Egyptian systems. Again, grammatical morphemes are not encoded until the beginning of writing phase 1B (150 AD; ibid., 305), around 250 years after the earliest Mayan texts. We can make some general observations about the evolutionary emergence of writing from the origins of Mesopotamian, Egyptian, Chinese, and Mesoamerican scripts. The emergence of writing is a gradual process, both in the transition from pre-writing semasiographic systems to writing as a representation of spoken language, and in the further development of writing to express grammatical structure as well as words (syntax and grammatical morphemes). The signifier is often directly associated with the signified, or a representation thereof (e.g., the image of a ruler and his name). Pre-writing is often multimodal, involving some combination of numerals, ideograms, seals, and pictorial representations. Some of this multimodality may have been a trigger for the graphical representation of language, that is, the combination of discrete linguistic units. In pre- or proto-writing phases, the encoding of proper names appears to have been an important step. Proper names are indexical in function (picking out an individual) but do not operate via deixis. The earliest ideographs and logographs were pictographic, that is, iconic, or iconic-indexical (via metonymy and synecdoche). Even the development of phonetic representation originates in iconic signs metonymically via the rebus principle (a sign originally standing for a concept is extended to represent the associated phonetic form of the word denoting that concept). The form

Evolutionary Complexity of Social Cognition, Semasiographic Systems 117

of signs later becomes more arbitrary. Another aspect of pictographic representation for language is that it adapts the signs of another semasiographic system, namely that of images. In this context it is not surprising that later writing systems emerged by adapting existing signs from writing systems that originally evolved for other languages, even using signs for different phonetic values than the originals, such as the Greeks adapting consonant signs to represent vowels. The domains of use for writing were at first very limited, often for centuries, even though spoken language of course is a general-purpose communication system for all sorts of joint actions. This is because writing originates to solve specific information storage and communication problems, not to faithfully reflect spoken language. Even when the domains of use of writing expanded to include narrative, writing left much of the language out. Writing did not express grammatical elements until centuries after its first emergence. In other words, substantial common ground between author and reader was required to interpret just the linguistic form encoded by early writing (this is still true of abjads, a writing system that indicates only the consonants; one must know the language to fill in the vowels). 4.3

Musical Notation

Musical notation appears to have been invented independently or partly independently several times in the Near East and Europe. The earliest known fragment of musical instruction is an Old Babylonian text that gives the incipit of a hymn to Ishtar. The notation names the sixth string of the Babylonian lyre or harp, certain intervals that involve the sixth string, and the mode in which the hymn is to be played (Kilmer and Civil 1986: 96). This fragmentary text appears to be a linguistic description of how to play the hymn rather than any specific musical notation. Later Babylonian texts, from 1250–1200 BC (West 1994) are less fragmentary but remain difficult to interpret. The notation was designed for a ninestringed instrument, probably the bovine lyre (West 1994: 166), and for the seven standard tunings of the instrument (ibid., 164). The texts bearing notation are in Hurrian but apparently in a Babylonian musical tradition. The notation consists of names of intervals followed by a numeral sign (ibid., 171). The interpretation of the numeral is disputed, but may indicate the repetition of notes. Nevertheless, we may make some general observations. The notation is adapted from other semasiographic systems – writing and numbers. As such, it is also multimodal. The music is also meant to be sung, and the lyrics are included. This provides another degree of multimodality, because sung music is also multimodal (words and melody). The notation is very domain-specific: it only can represent music for the lyre, and only in one of the seven modes. Finally, the

118

William Croft

rhythm and other details of performance are not specified (ibid., 176); that is, common ground about the songs is required to perform it properly. Musical notation was reinvented by the Greeks by the third century BC, and lasted no later than the fourth or fifth century AD (West 1992: 254). It is generally agreed that the musical notation developed gradually. Pitches are notated by letters borrowed from the Greek alphabet for use in instrumental music. Some of the pitch letters are Ionian, but other letters appear to derive from the Argive script, a local Greek script of the sixth and fifth centuries BC (ibid., 261). The notation appears to have originated for the aulos (a flute). Vocal notation based on the Ionian alphabet was developed probably in the late fifth or fourth century BC (ibid., 263). Rhythm came to be notated in some detail in the second century AD, employing a separate set of symbols indicating length (ibid., 268). The history of Greek musical notation illustrates the gradual evolution of the notation over time. It is restricted at first to the aulos. Finally, it again presupposes familiarity with the music itself, and its performance style. The notation was probably used only by professional musicians; and was rarely used at all, as most lyric and dramatic texts lack musical notation (West 1994: 270). Musical notation was again reinvented in Europe in the middle ages. (Apparently Boethius [sixth century] was aware of Greek notation using letters for pitches, but his writings did not lead to a similar notation in the middle ages; Taruskin 2005: 17.) The earlier medieval musical notation is in the form of staffless neumes from the nineth and tenth centuries. Staffless neumes are basically dots and lines that indicate the overall pitch contour, but without the staves that allow one to identify musical intervals. Nor do staffless neumes indicate rhythm. Instead, individual neumes appear to have specified particular melodic formulas (by their shapes, which were given Latin names) that were associated with particular chants and modes (Taruskin 2005: 22). Hence staffless neumes presuppose much common ground: they do not indicate details of melody or rhythm (let alone other aspects of performance, including polyphonic accompaniment). The user of the musical notation must already know the music, which is still transmitted essentially orally; the notation at best serves as a reminder or an aid (Taruskin 2005: 17). Musical notation was restricted at first to sacred music and later extended to elite secular music, namely the courtly songs of the troubadours, trouvères, and minnesänger. Neumes were positioned on a staff starting around the early eleventh century (Taruskin 2005: 16) – that is, two centuries after the first surviving documents of staffless neumes. This allowed relative pitch (i.e., intervals) to be represented. (Even so, manuscripts with staffless neumes were produced in the monastery of St. Gall until the fifteenth century.) But rhythm would not be represented until the beginning of the thirteenth century, another two centuries later (Taruskin 2005: 176). Rhythm was first notated for the polyphonic music of Nôtre-Dame

Evolutionary Complexity of Social Cognition, Semasiographic Systems 119

of Paris, by recruiting the neume shapes or combinations (known as ligatures) to express rhythmic patterns instead of melodic formulas. The rhythmic interpretation of ligatures was context-sensitive, depending on the overall pattern of ligatures (ibid., 177). By this point, polyphonic music was being notated, but there did not exist explicitly notated means of aligning the voices such as the later bar lines, apart from the iconic visual alignment of voices in the manuscripts. (Bar lines first appeared in keyboard notation in the early sixteenth century; Apel 1953: 9.) In the middle thirteenth century, so-called Franconian notation was developed for notating the music of the motet style of the time (namely a style in which all voices are equally melismatic, but a text or texts were applied to the melismatic melodies). Again, the existing neume shapes and combinations were recruited, this time for a less context-sensitive way to represent note length and hence rhythm (Taruskin 2005: 212–17). While each specific ligature (combination of note shapes and stems) has a unique rhythmic interpretation in Franconian notation, the individual note shapes did not, except in the context of the ligature. Ars Nova notation (fourteenth century) is less context-dependent for its rhythmic interpretation, and the notation developed by the beginning of the seventeenth century is essentially the same as modern notation with respect to rhythm (apart from odd anomalies such as the one described in §3). Even with the development of more explicit notation of pitch intervals and values, and rhythm, other aspects of musical performance were left to common ground (leading to much debate over the “historically accurate” interpretation of medieval and renaissance music by modern early music performers). From the seventeenth to the twentieth century, notation of articulation, dynamics and tempo were gradually added to musical scores. Much was still left to common ground and an oral tradition, such as how to ornament melodic lines or fill out a figured bass in Baroque music. While a general notation system for European music evolved over the past twelve centuries, specialized notations continued to emerge. These specialized notations were mostly designed for instrumental music, and hence specific to an instrument. The most widespread example was lute tabulature, since lute was the most popular instrument in the Renaissance (Apel 1953: 54–86). Lute tabulature used staff lines, but the staff lines indicate the strings of the lute, not specific pitches. Numbers or letters were adapted to indicate which frets to use, and notes above the staves were adapted from general musical notation to indicate the rhythm. In other words, lute tabulature was multimodal, drawing from general musical notation, letters, and numbers to indicate strings, frets (which together specify pitch), and rhythm. A similar tabulature was used for German organ music through its flowering in the seventeenth and early eighteenth centuries (Apel 1953: 21–47).

120

William Croft

4.4

Dance Notation

Dance notation is not standardized; numerous systems have been invented in the West. Like spoken language and music, dance requires strict temporal sequence (and hence linear order). But dance is far more complex a phenomenon to represent. The notations used in the West vary tremendously; many if not most choreographers invented their own notation. Nevertheless, the types of notation used present a historical sequence. The earliest known dance notation was devised for the Renaissance basse danse (low dance) (Guest 1989: 1). Letters were adapted to indicate the steps to perform (letter notations continued for 200 years; ibid., 4). As with other semasiographic systems, the earliest dance notation was contextually highly restricted (only basse danse steps can be represented by the letters) and required much common ground – they can only be interpreted by those who know the steps to the dance. The next types of dance notation in Europe (in terms of when it was first devised) were iconic in form. In the seventeenth century, a track notation was devised, which iconically represented the choreography as couples moved around the floor (Guest 1989: 12–27). This notation developed in part because the dances of this period involved such movements. The notation also imagistically represented bodily movements, though it also used arbitrary symbols to indicate men vs. women (ibid.). Again, the notation was restricted to representing Baroque dances. In the nineteenth century, another iconic notation was devised, namely, stick-figure notation (Guest 1989: 28–68). Stick-figure notation, unlike the earlier notation, was less restricted in the types of dance movements it could represent. All of these types of dance notation were multimodal in that the music to which one danced was also notated along with the dance steps and movements. In the late nineteenth and early twentieth centuries, a music-based dance notation was invented. It adapted musical notation for dance steps: note values indicate time – a dimension of dance not well notated by earlier systems. The notes were placed on different musical staves that indicated different parts of the body. Direction of movement was often indicated iconically, while modifications of the form of the musical note expressed bodily movements, sometimes iconically and sometimes not. In the 1920s, purely abstract dance notations developed (Guest 1989: 102– 62). Guest also notes that an earlier abstract system was devised by a choreographer in 1831 but was an isolated invention. Unlike the Baroque and nineteenthcentury notations, abstract dance notation was a completely symbolic, and in some instances anti-iconic, representation of dance movements, although some symbols were iconically motivated in part. On the other hand, the abstract systems are also highly general in representing human bodily movement,

Evolutionary Complexity of Social Cognition, Semasiographic Systems 121

timing, and direction of dance – possibly a response to the freer dance styles of twentieth-century dance and the greater variety of movements that they encompassed. The evolution of music and dance notation exhibits all of the properties of the chronologically prior evolution of number and writing. It was gradual (extending over centuries in the case of Western musical notation); it is multimodal, and adapts signs from other semasiological systems; its domain of use was very restricted at first, before expanding; the interpretation of some signs was context-dependent; and the notation also required much common ground to interpret, becoming more explicit as it evolved. The chief difference is that the signifier was not directly associated with the signified, except when using musical scores in performance. This difference is presumably due to the fact that these semasiographic systems represent human performances rather than the conceptual structures represented by mathematics and language. 4.5

Some Commonalities Across the Evolution of Semasiographic Systems

The evolution of different types of semasiographic systems reveals some common features, which may offer clues for the evolutionary origin of the language capacity. To begin with, the evolution of semasiographic systems, where we have documentary evidence, is gradual and incremental. Contemporary semasiographic systems for number, writing, music, and dance are very different from their first notation, and the process leading to contemporary semasiographic systems included many intermediate stages and alternative notations. This observation reinforces the view expressed in §3, viz., that the evolution of social cognition and the language that facilitates joint actions was also gradual and incremental. The earliest uses of semasiographic systems are very domain-specific and context-dependent. They are used for only a restricted functional domain of social (joint) actions. Numbers were used only for counting and for arithmetical functions, before being elaborated to more and more mathematical functions. Initial uses of writing were for specific communicative acts, and only later came to be used in more and more domains. In fact, only with the advent of electronic media is it increasingly used in the most casual conversational interactions. Initial uses of musical notation were restricted to specific types of music, for specific social or religious functions, or for specific instruments, as in the case of the creation of lute notation. Dance notations were only for specific genres of dance; and even now no general dance notation has come into widespread use, although the latest notations are highly general in their representation of bodily movements. Thus we might expect that language began in highly restricted functional domains, and its extension to become a

122

William Croft

general-purpose communication system was a long and gradual process in human prehistory. Another respect in which early semasiographic systems are contextdependent is the minimal expression of what they signify. The earliest writing did not encode grammatical inflections, or all of the sounds of a language. Early European musical notation did not encode exact intervals or rhythms, let alone dynamics. Early dance notation was even more minimal in the information that it encoded. In other words, the interpretation of the semasiographic representation required much common ground; it served to minimally evoke what the relevant information or action actually was. Likewise, language initially functioned simply as a coordination device for joint action. Thus the “meaning” it conveyed was probably minimal – just enough for the (restricted) purpose of coordinating a particular type of joint action. Eventually, language came to be used to share knowledge, that is, to increase common ground in and of itself. At that point, language had to become more semantically specific (although as we saw in §2, language can never be totally “precise”). But this was probably a late stage in the evolution from pre-language to language. In some semasiological systems, the signifier is often directly associated with the signified. This was the case for Mesopotamian tokens and clay envelopes, which were associated with the objects they enumerated or denoted; and for Egyptian and Mesoamerican writing, which were often at first names associated with representations of the persons or deities that they signified. The semantic relationship is thus associative at first. However, the situation is somewhat different with performance notation. Musical notation at first was at most a mnemonic for the melodies that it encoded. In later periods, music was frequently performed from notation, but it is not clear if that was true at the beginning. Dance notation is not directly used in performance. Music and dance notations are more likely to be used for teaching rather than for final performance; they serve as a record of how to perform the composition in question. Still, the cases in which the signifier is directly associated with the signified suggests that the earliest uses of pre-language or language was to denote elements of immediate experience, and only later was language (or pre-language) extended to denote displaced or imagined experience, which must be inferred from the semantics of the sign. (This last suggestion is a common speculation about the evolution of language.) Semasiographic systems are essentially coordination devices: graphic coordination devices. With respect to that function, the coordination device for one semasiographic system is often parasitic on another semasiographic system. For example, writing symbols were appropriated for Babylonian and Greek music notation, for Greek numerals, Renaissance dance, and arithmetic symbols. Writing systems designed for one language were adapted for another language (Olmec for Maya, Phoenician for Greek, Sumerian for Akkadian,

Evolutionary Complexity of Social Cognition, Semasiographic Systems 123

and then Hurrian, Hittite, Persian, and other ancient languages; Chinese for Japanese and Korean; Greek for Slavic languages via Cyrillic, and the Roman alphabet for many languages). And images, the earliest semasiographic system, were adopted for medieval music, for Baroque and nineteenth-century dance, and for the earliest writing and counting systems. Thus we might expect that the earliest linguistic signals might be adopted from some other communicative system. It has been argued that primate vocalizations cannot be the source of modern human linguistic vocalizations, because primate vocalizations and their “meanings” (alarm calls or whatever) are innate. While modern human language definitely is not innate in the specific linguistic structures and vocalizations employed – it is extremely diverse across speech communities – it could still be the case that at some point in the evolution of pre-language, vocalizations or manual and facial gestures were adapted from some other function, including perhaps even innate vocal or other gestures. (This is the process known as exaptation, a phenomenon well attested in contemporary language change; see Lass 1990; Croft 2000, Mufwene 2001.) Semasiographic systems also appear to evolve to a (more) arbitrary symbol system. Again, this reflects a common speculation about the evolution from pre-language to language. Pre-language may originate with indexical or iconic gestures or vocalizations, before being supplemented with more arbitrary symbols. Of course, modern human language retains indexical and iconic expressions, along with more arbitrary symbolic forms. Indexicals form a central body of linguistic expressions in modern human languages: demonstratives are among the first words to be learned, are very ancient (not clearly derived from other words), are central to structuring discourse, and grammaticalize into ubiquitous grammatical functions, including definiteness, reference, and subordination (Diessel 2006). Diagrammatic iconicity is exhibited by the great majority of syntactic structures, and many signed language gestures are iconic. In terms of the complexity of the signals themselves, the signs in the earliest semasiographic systems are frequently multimodal. Early writing combined symbols for numbers and for objects. Medieval and later music combined words and notes (pitches); later music combined notes for pitch and duration with other types of symbols for dynamics and articulation. All dance notations combine music with symbols for bodily movements. Baroque, music-based, and abstract dance notation uses different types of symbols for different types of bodily movements. Hence it might be suggested that the earliest language was multimodal. There is an ongoing speculative debate about the priority of vocalization or manual gesture in the evolution from pre-language to language. It is perhaps most likely that both were used at first, without the separation that we currently conceive for the two (Arbib 2012; McNeill 2012). And of course modern spoken language is accompanied by manual gestures, and some have argued that intonation represents a vocal gesture that is at least partly separate

124

William Croft

from language in the sense of the latter being a segmentally based symbolic system of communication (Bolinger 1985). Finally, semasiographic systems exhibit virtually no duality of patterning. The closest thing to duality of patterning is the use of determinatives or classifiers and phonological complements in logographic writing systems (e.g., Cooper 2004: 89; Nissen et al., 1993: 115). The absence of duality of patterning may be due to the restricted domains of expression of most semasiographic systems, even late in their evolution, in contrast to the general-purpose communicative function of modern human language (see §3). But it might also suggest that the earliest pre-language did not exhibit duality of patterning at first either. One important difference between semasiographic systems and modern human languages in their spoken or signed medium is that language in use must be processed in real time, while semasiographic systems, being lasting, are not. Thus the selectional pressures for semasiographic systems and language might be somewhat different, and the evolutionary characteristics may also differ. However, the properties drawn out from comparison of the evolution of different semasiographic systems – gradualness, domain-specificity and context dependence, minimal expression, association of signifier and signified, function as coordination devices, exaptation of functions, multimodality and lack of duality of patterning – are independent of the medium of communication. In fact, some of these properties (in particular context-dependence, minimal expression, association of signifier and signified) are often treated as more typical of the real-time medium of spoken or signed language than of writing. Before proceeding to speculation on the stages of the evolution from prelanguage to language, it is worth taking a glance at the process of language acquisition. Although ontogeny doesn’t always recapitulate phylogeny, there are some parallels in early language acquisition with the observations made on the evolution of semasiographic systems that are worth noting in supporting speculations on language evolution. Language acquisition is a gradual process. This has been repeatedly demonstrated by recent research on both phonological and grammatical acquisition (Vihman 1996; Lieven, Pine, and Baldwin 1997; Tomasello 2003). The earliest use of language is also restricted in functional domain, to a restricted set of activities that the caregiver and child engage in. It is also minimal, initially consisting of single words that evoke the relevant interaction; productive grammatical inflections do not emerge until later. The signifier and signified are directly associated at first – the first utterances are about immediate experience. Finally, the earliest communicative gestures are multimodal – both vocalizations and bodily gestures. Thus, language acquisition also shares many features with semasiographic systems. Of course, there are some important differences between child language acquisition and the evolution of pre-language from language. The caregivers

Evolutionary Complexity of Social Cognition, Semasiographic Systems 125

Table 5.1 Evolutionary Complexity Based on the Evolution of Semasiographic Systems Less “complex”

More “complex”

Context dependent: specialized function minimal coding immediate experience (associative) extensional (indexical) multimodality

Context independent: general purpose explicit coding displaced experience (inferred) intensional (iconic/symbolic) duality of patterning

that children interact with are already users of modern human language. And children, at least after the “nine month revolution” (see §3), have most of the social cognitive capacity that underpins modern human language. But the parallels to the evolution of semasiographic systems observed here (admittedly, also created by humans with modern social cognitive abilities) lend support to the speculations offered in the next section. 5

Speculations on the Evolution of Language

The observations on semasiographic systems, and the parallels with child language acquisition, suggest that some functions are likely to be earlier and less complex in an evolutionary sense, while other functions are likely to be later and more complex in the same sense (see Table 5.1). I have used the term “complex” in scare quotes because complexity here is based on evolutionary sequence. Also, it is not based purely on the structure of the signifier but on how the signifier expresses the concepts being communicated. Nevertheless, it seems that the phenomena in Table 5.1 do reflect lesser or greater degrees of cognitive complexity. A context-dependent system is going to be less complex because it involves a small number of more specific concepts, whereas a context-independent system will be more complex because it will involve abstracting categories across a wide range of concepts. Explicit coding will also require more signal complexity (grammatical inflections, more complex constructions, etc.). Displaced experience is more complex to communicate than immediate experience because it requires conceptualization of an experience apart from the speech act situation. Intensional symbols are more complex than extensional ones in that they require the formation of a category. Duality of patterning requires two alternative parsings of the same signal structure (phonological and morphosyntactic), whereas even a multimodal signal does not operate on two levels at once; both modalities are being used to convey the intended concept.

126

William Croft

These asymmetries in communicative means can be used to suggest possible stages in the evolution from pre-language to language, as already suggested in §4.5. But at least as important, the logical structure of the suite of cognitive abilities described in §2 also suggests the evolutionary priority of some social cognitive abilities over others. Joint action, coordination, and communication all presuppose the presence of common ground. Hence we will take the evolutionary emergence of common ground – the ability to recognize that others have mental states that can be shared (Tomasello 2008) – as our starting point. In §2, we saw that common ground comes in different types that have different bases. Communal common ground emerges from shared practice among humans in a social group. If so, then personal common ground, based on common perception and action, seems prior: it is based on direct interpersonal interaction. But the actional basis presupposes the communal common ground of shared communicative conventions, in particular a shared language, in the community. Hence the actional basis must follow, or be part of, the shared practice that underpins communal common ground. The perceptual basis for personal common ground can arise from communal common ground resulting from shared practice among members of the social group. That is, the way that we categorize elements of our perceptual experience may be influenced by shared practice. But there is a more fundamental basis for common ground, namely a human being’s individual interaction with the world, including one’s own self (body, bodily functions and needs, etc.). We can call this natural common ground. Natural common ground is made up of an individual’s interaction with the world and one’s body. Its basis is a human’s perceptual recognition of species identity with another human being: this maximally inclusive “community” allows “natural” knowledge to be shared knowledge, that is, common ground. Hence, an ordering of types of common ground based on logical priority would be something like this: r natural common ground: individual ways of interacting with the world and one’s body, shared via recognition of other humans as conspecifics; r perceptual personal common ground, shared via joint attention to and joint salience of the interacting individuals’ environment and actions; r communal common ground, shared via shared practices between the members of the community for various social and cultural purposes; r actional common ground after the emergence of language and other human communicative conventions. With respect to coordination devices, a clear logical priority exists, as described in §2. Convention presupposes precedent: a convention cannot emerge until a coordination device is repeated due to precedent and then becomes common ground in the community. And precedent presupposes joint salience: joint

Evolutionary Complexity of Social Cognition, Semasiographic Systems 127 Basis

Common Ground

Coordination Devices

Being in the World+ Conspecific Recognition

Natural Common Ground

Joint Salience

Shared Environmental Interaction + Joint Attention

Perceptual Personal Common Ground

Shared Precedent

Shared Practice

Communal Common Ground

Convention

Conversation (Language Use)

Actional Personal Common Ground

Language (Linguistic Convention)

Figure 5.1 The logical relationship among types of common ground, its basis, and coordination devices.

salience allows the possibility of first-time (nonrandom) success in coordination, which can then be followed as a precedent. Joint salience and precedent require both current and prior interactions between the individuals (i.e., personal common ground). Convention is shared in a community, that is, conventions are part of a community’s communal common ground. And language is a conventional system for communication, so it also presupposes communal common ground. The logical relationships between types of common ground and types of coordination devices are given in Figure 5.1 (the dotted line indicates the relationship between joint attention and joint salience, and the double role it plays in the achievement of joint actions). These logical relationships suggest the types of coordination devices that one might expect with different types of human interactions with the world, one’s self, and with others, which suggest stages in the evolutionary emergence of coordination devices including, eventually, language. An evolutionary account of the emergence of these social cognitive abilities requires some sort of selectional pressure for hominins to develop each successive stage of these abilities. The obvious place to look is in the evolution of joint actions. In §2, we presented an analysis of joint actions that requires coordination and common ground, but also a degree of helpfulness, or commitment to the other’s success in carrying out their part of the joint action (Bratman 1992: 336–38; Tomasello 2008, §5.2.1). Tomasello and colleagues have more

128

William Croft

recently developed an evolutionary account of the evolution of human morality as a basis for the emergence of joint actions (Tomasello 2011; Tomasello et al., 2012; Tomasello and Vaish 2013). We will use this account to motivate the emergence of common ground, coordination devices, and thence language. Tomasello and colleagues argue for two stages in the evolution of human morality beyond great ape “morality” (their interdependence hypothesis). They argue that while great ape sociality involves some degree of sharing, reciprocity, and revenge, they are not committed to helping the other in collaborative activities in the way that humans are (Tomasello and Vaish 2013: 236). The first stage in the evolution of human morality is brought on by a selectional pressure for genuinely collaborative action (rather than just reciprocal altruism; Tomasello et al., 2012: 673), which they take to be an environmental need to engage in collaborative foraging in order to survive (ibid., 674). This stage, which they call second-personal morality (Tomasello and Vaish 2013: 240), involves helpfulness and commitment – treating the other as an equal, not as a “social tool” (ibid., 236) – but specifically with individuals known personally to the agent. Reputation, critical to avoiding free riding, is based on direct experience of social interactions. The effect is that collaborative joint actions emerge, but only at a small scale. The next stage is what Tomasello and colleagues call group-mindedness (Tomasello et al., 2012: 681) or norm-based morality (Tomasello and Vaish 2013: 245). They argue that this stage emerged as the result of increasing size of groups, and pressure from competing groups also increasing in size (Tomasello et al., 2012: 681). The outcome is the ability to engage in collaborative joint activity with individuals who are not directly known to the agent, and hence do not have a reputation acquired by direct experience of interaction, but instead can be trusted by virtue of group membership. At this stage, there come to exist social norms and institutions that enable large-group collaboration. Second-personal morality is probably the selectional pressure leading to the emergence of perceptual personal common ground and shared precedent as a coordination device (namely, the second level from the top in Figure 5.1). Perceptual personal common ground comes from direct interactions between the individuals, and shared precedent also requires direct experience. On the other hand, norm-based or group morality is probably the selectional pressure leading to communal common ground and convention as a coordination device. Collaboration is based on group membership; the common ground and coordination devices also have to be established groupwide, not by direct experience of interpersonal interaction. Figure 5.1 implies that logically there may have been two other discrete stages in the emergence of coordination devices, including language. The last stage is clearly an elaboration of group-mindedness, where the cultural transmission of knowledge through language (and, eventually, writing) proceeds at

Evolutionary Complexity of Social Cognition, Semasiographic Systems 129

an abstract level and a large scale in a large society. The first stage may be an initial stage of second-personal morality. It presupposes at least some level of collaboration that would first occur only in small groups of individuals known to each other – perhaps a kin group. Hence it may have been kin recognition rather than conspecific recognition that initially allowed for the emergence of (natural) common ground. Finally, we may speculatively map more concrete proposals about prelanguage and language based on the evolution of semasiographic systems and language development in Table 5.1 onto the logical sequence of common ground and coordination devices in Figure 5.1. The first stage involving joint action, as a number of other scholars of the evolution of language have proposed, is most likely to employ only indexicals. Indexicals rely solely on natural common ground and joint salience or attention as the coordination device. Indexicals involve association and hence are restricted to immediate experience; they are context-dependent and typically multimodal – even linguistic demonstratives in their indexical uses are multimodal, accompanied by gestures (Diessel 2006: 469–71). Iconic gestures might also serve as coordination devices at this stage, because they also require only natural common ground, particularly when used for immediate experience. They are also likely to be multimodal. The sort of joint actions that are coordinated by indexicals, icons, and natural common ground would have to be very simple, immediate actions. However, it is likely that individuals would engage in joint actions of this sort only with close associates, perhaps only kin. Second-personal morality might represent a more extensive group that engages in collaborative joint actions: known individuals, but not necessarily kin. Precedent as a coordination device would allow the emergence of more stable signals. One type of signal might be proper names for individuals, because individual identity is important for second-personal morality (proper names are also among the first things to appear in the evolution of writing). These proper names would be used at least as vocatives (i.e., in immediate experience) or possibly in displaced contexts. Proper names might begin with a gesture iconic of an associated property of the individual but may evolve to arbitrary symbols. Naming would facilitate the performance of polyadic joint actions. As such, naming might lead to vocalization as a more prominent or even primary modality: the broadcast nature of vocalization compared to visually perceived bodily gestures may be favored for selection in a polyadic joint action situation (cf. Tomasello 2008: 231). Stable symbols within the group might also lead to the emergence of semantic content in symbols (in fact, iconic gestures may also be used with semantic content). The joint actions are everyday activities that involve a number and variety of objects as well as agents (so are more complex in that sense), requiring explicit categorizing and labeling.

130

William Croft

The advent of larger groups gives rise to communal common ground and conventional signals in that community. It seems likely that the evolutionary success of such groups would require engaging in planned joint actions. These joint actions are complex activities requiring subplans that need to be made explicit (i.e., they are not simply part of communal common ground and cannot be executed by everyone without explicit negotiation or instruction). The content for communication is not just immediate experience but also a restricted set of imagined and displaced experience, namely the intended or desired subplans that must occur in the future in order to achieve the total planned joint action, and the subplans that have just been executed. Successful coordination of planned joint actions of this sort would require combinations of signals (different combinations may involve gesture + gesture, gesture + vocalization, or vocalization + vocalization). Combinations are required to construct imagined experiences, namely the intended subplans. Because these experiences are not in the here and now and may not have existed before, they need to be evoked by combining units that do denote objects and actions in the here and now, but in combinations representing a novel situation. At this stage, combinations are likely to be structured by order, not grammatical inflection; the sort of finegrained conceptual detail that calls for grammatical inflection (see Croft 2007a) is not needed for coordination of everyday joint actions, even these more complex actions. The final stage is the emergence of sharing of knowledge. The joint actions are complex, but now include the joint action of sharing propositional knowledge, beliefs, desires, evaluations, and so on. In other words, they are not coordinating some physical everyday activities but enlarging the stock of knowledge (including cultural knowledge and attitudes) in the community. The content to be communicated is now not only immediate experience and imagined experience geared to executing complex actions, but also recalled experience and other types of imagined experience not leading to immediate action. At this point, actional common ground is produced by the joint action. Of course, the coordination devices include all three of joint attention, precedent, and convention. The signaling system includes the signals from the earlier stages and combinations of such signals. At this stage, the combinations of signals serve to communicate displaced experience and the full range of imagined experience. The generalization of the function of the signaling system to include sharing of knowledge of all kinds will lead to the emergence of grammatical inflection and derivation, and the sort of grammatical constructions familiar from modern human languages. This generalization of function will also lead to duality of patterning in the structure of the signal, though it is possible that duality of patterning emerged earlier (see §1 and Mufwene 2013). Finally, this generalization of function leads to the primacy of vocalization in the signaling system (among hearing persons). In other words, the final stage leads to

Evolutionary Complexity of Social Cognition, Semasiographic Systems 131

modern human language, though I would guess that this stage also undergoes gradual and incremental evolution, and we might only be able to speak of modern human language at the end of the fifth stage described here. This account of stages of language evolution is very speculative (perhaps modeling would give some clues). Nevertheless, by grounding the account of the evolution of language in its function to facilitate joint actions among individuals in a community, we can develop plausible scenarios for the stages it must have gone through on the way to modern human language.

REFERENCES Appel, Willi. 1953. The notation of polyphonic music, 900–1600 (5th ed.). Cambridge, MA: The Medieval Academy of America. Arbib, Michael. 2012. How the brain got language: the mirror system hypothesis. Oxford: Oxford University Press. Bagley, Robert W. 2004. Anyang writing and the origin of the Chinese writing system. Houston 2004a, 190–249. Baines, John. 2004. The earliest Egyptian writing: development, context, purpose. Houston 2004a, 150–89. Bolinger, Dwight. 1985. The inherent iconism of intonation. Iconicity in syntax, ed. John Haiman, 97–108. Amsterdam: John Benjamins. Boone, Elizabeth Hill. 1994. Introduction: writing and recording knowledge. Writing without words: alternative literacies in Mesoamerica and the Andes, eds. Elizabeth Hill Boone and Walter D. Mignolo, 1–26. Durham, NC: Duke University Press. Boone, Elizabeth Hill. 2004. Beyond writing. Houston 2004a, 313–48. Bratman, Michael. 1992. Shared cooperative activity. The Philosophical Review 101:327–41. Buccellati, Giorgio. 1981. The origin of writing and the beginning of history. The shape of the past: studies in honor of Franklin D. Murphy, eds. Giorgio Buccellati and Charles Speroni, 3–13. Los Angeles: Institute of Archaeology and Office of the Chancellor. Bybee, Joan L. 1985. Diagrammatic iconicity in stem-inflection relations. Iconicity in syntax, ed. John Haiman, 11–48. Amsterdam: John Benjamins. Chafe, Wallace. 1977a. Creativity in verbalization and its implications for the nature of stored knowledge. Discourse production and comprehension, ed. Roy Freedle, 41–55. Norwood, NJ: Ablex. Chafe, Wallace. 1977b. The recall and verbalization of past experience. Current issues in linguistic theory, ed. Peter Cole, 215–46. Bloomington: Indiana University Press. Cheng, Chin-Chuan. 2000. Frequently used Chinese characters and language cognition. Studies in the Linguistic Sciences 30:107–17. Clark, Herbert H. 1996. Using language. Cambridge: Cambridge University Press. Cooper, Jerrold S. 2004. Babylonian beginnings: the origin of the cuneiform writing system in comparative perspective. Houston 2004a, 71–99. Corballis, Michael C. 2011. The recursive mind: the origins of human language, thought, and civilization. Princeton: Princeton University Press.

132

William Croft

Coupé, Christophe. 2012. Semiotic investigations into early forms of symbolism and language. The evolution of language: proceedings of the 9th International Conference (Evolang 9), eds. Thomas C. Scott-Phillips, Monica Tamariz, Erica A. Cartmill and James R. Hurford, 80–87. Singapore: World Scientific. Croft, William. 1995. Intonation units and grammatical structure. Linguistics 33:839– 882. Croft, William. 2007a. The origins of grammar in the verbalization of experience. Cognitive Linguistics 18:339–82. Croft, William. 2007b. Intonation units and grammatical structure in Wardaman and English. Australian Journal of Linguistics 27:1–39. Croft, William. 2000. Explaining language change: an evolutionary approach. Harlow, Essex: Longman. d’Errico, Francesco, Christopher Henshilwood and Peter Nilssen. 2001. An engraved bone fragment from c. 70,000-year-old Middle Stone Age levels and Blombos Cave, South Africa: implications for the origin of symbolism and language. Antiquity 75:309–18. Diessel, Holger. 2006. Demonstratives, joint attention, and the emergence of grammar. Cognitive Linguistics 17:463–89. Englund, Robert. 1998. Texts from the late Uruk period. Mesopotamien: SpäturukZeit und Frühdyntastische Zeit (Orbis Biblicus et Orientalis, 160/1), eds. Joseph Bauer, Robert Englund and Manfred Krebernik, 15–233. Freiburg: UniversitätsVerlag/Göttingen: Vandenhoeck & Ruprecht. Everett, Daniel L. 2005. Cultural constraints on grammar and cognition in Pirahã: another look at the design features of human language. Current Anthropology 76: 621–46. Everett, Daniel L. 2009. Pirahã culture and grammar: a response to some criticisms. Language 85:405–42. Fauconnier, Gilles. 1985. Mental Spaces. Cambridge, MA: MIT Press. Guest, Ann Hutchinson. 1989. Choreo-graphics: a comparison of dance notation systems from the fifteenth century to the present. New York: Gordon and Breach. Haiman, John. 1980. The iconicity of grammar: isomorphism and motivation. Language 54:565–589. Haiman, John. 1985. Natural syntax: iconicity and erosion. Cambridge: Cambridge University Press. Hockett, Charles F. 1960. The origin of speech. Scientific American 203:88–96. Houston, Stephen D. (ed.). 2004a. The first writing: script invention as history and process. Cambridge: Cambridge University Press. Houston, Stephen D. (ed.). 2004b. Writing in Early Mesoamerica. In The first writing: script invention as history and process. Cambridge: Cambridge University Press, 274–309. Kilmer, Anne Draffkorn and M. Civil. 1986. Old Babylonian musical instructions relating to hymnody. Journal of Cuneiform Studies 38:94–98. Kirby, Simon. 2002. Learning, bottlenecks and the evolution of recursive syntax. In: Ted Briscoe (ed.), Linguistic evolution through language acquisition, 173–203. Cambridge: Cambridge University Press. Langacker, Ronald W. 1997. Constituency, dependency, and conceptual grouping. Cognitive Linguistics 8:1–32.

Evolutionary Complexity of Social Cognition, Semasiographic Systems 133 Lass, Roger. 1990. How to do things with junk: exaptation in language change. Journal of Linguistics 26:79–102. Lewis, David. 1969. Convention. Cambridge, MA: MIT Press. Lieven, Elena V. M., Julian M. Pine & Gillian Baldwin. 1997. Lexically based learning and early grammatical development. Journal of Child Language 24: 187–219. Marshack, Alexander. The roots of civilization: the cognitive beginnings of man’s first art, symbol and notation, revised and expanded edition. Mount Kisco, NY: Moyer Bell Limited. McNeill, David. 2012. How language began: gesture and speech in human evolution. Cambridge: Cambridge University Press. Menninger, Karl. 1969. Number words and number symbols: a cultural history of numbers, translated from the revised German edition by Paul Broneer. Cambridge, MA: MIT Press. Mufwene, Salikoko S. 2001. The ecology of language evolution. Cambridge: Cambridge University Press. Mufwene, Salikoko S. 2013. Language as technology: some questions that evolutionary linguistics should address. In search of Universal Grammar: from Norse to Zoque, ed. Terje Lohndal, 327–58. Amsterdam: John Benjamins. Nissen, Hans, Peter Damerow and Robert Englund. 1993. Archaic bookkeeping: early writing and techniques of economic administration in the ancient Near East. Chicago: University of Chicago Press. Pettersson, John Sören. 1996. Numerical notation. The world’s writing systems, eds. Peter T. Daniels and William Bright, 795–806. Oxford: Oxford University Press. Rudman, Peter Strom. 2007. How mathematics happened: the first 50,000 years. New York: Prometheus Books. Schmandt-Besserat, Denise. 1996. How writing came about. Austin: University of Texas Press. Schoenemann, P. Thomas. 1999. Syntax as an emergent characteristic of the evolution of semantic complexity. Minds and Machines 9:309–46. Stauder, Andréas. 2010. The earliest Egyptian writing. Visible language (Oriental Institute Museum Publications, No. 32), ed. Christopher Wood, 137–47. Chicago: The Oriental Institute of the University of Chicago. Taruskin, Richard. 2005. Music from the earliest notations to the sixteenth century. (Oxford History of Music, 1.) Oxford: Oxford University Press. Tomasello, Michael. 1999. The cultural origins of human cognition. Cambridge, MA: Harvard University Press. Tomasello, Michael. 2003. Constructing a language: a usage-based theory of language acquisition. Cambridge, MA: Harvard University Press. Tomasello, Michael. 2008. Origins of human communication. Cambridge, MA: MIT Press. Tomasello, Michael. 2011. Human culture in evolutionary perspective. Advances in culture and psychology, eds. Michele J. Gelfand, Chi-yue Chiu, and Ying-yi Hong, 5–52. New York: Oxford University Press. Tomasello, Michael, A. C. Kruger, and H. H. Ratner. 1993. Cultural learning. Behavioral and Brain Sciences 16:495–552.

134

William Croft

Tomasello, Michael, Alicia P. Melis, Claudio Tennie, Emily Wyman, and Esther Herrmann. 2012. Two key steps in the evolution of cooperation: the interdependence hypothesis. Current Anthropology 53:673–92. Tomasello, Michael and Amrisha Vaish. 2013. Origins of human cooperation and morality. Annual Review of Anthropology 64:231–55. Trigger, Bruce. 2004. Writing systems: a case study in cultural evolution. Houston 2004a, 39–68. Vihman, Marilyn May. 1996. Phonological development: The origins of language in the child. Oxford: Blackwell. Wenger, Étienne. 1998. Communities of practice: learning, meaning and identity. Cambridge: Cambridge University Press. West, M. L. 1992. Ancient Greek music. Oxford: Clarendon Press. West, M. L. 1994. The Babylonian musical notation and the Hurrian melodic texts. Music & Letters 75:161–79.

6

To What Extent Are Phonological Inventories Complex Systems? Christophe Coupé, Egidio Marsico, and François Pellegrino

1

Theoretical Background

1.1

Phonological Inventories as Complex Systems

Complex systems are often defined on the basis of the interactions that take place between their constituents. This approach echoes the founding principles of structuralism in linguistics more than a century ago. It was indeed Ferdinand de Saussure, in his 1916 Cours de linguistique générale, who defined any language as a system whose building blocks only exist in the structure of their relationships, be they of equivalence or opposition. Structuralism as a theoretical framework in linguistics initially developed more at a synchronic level, before the dynamical phenomena leading to the emergence, preservation, or disruption of structures were addressed. Concepts such as retroaction and self-organization appeared in the wake of cybernetics in the 1950s and 1960s (Ahsby, 1947; Wiener, 1961). Further developments led to notions such as emergence, dynamical equilibrium, chaos, self-organized criticality, and so on. These concepts soon spread to various scientific fields, including the life sciences and the humanities; their ubiquity is one of the pillars of today’s generic “theory of complex systems,” which to some extent succeeded in adding a dynamical perspective to the time-frozen structures of structuralism. In the field of linguistics, considering phonological inventories in the light of the theory of complexity is not new, and they may be the first linguistic structures that benefited from its explanatory power. The notion of self-organization appears in a 1983 paper on phonological universals (Lindblom et al., 1983): on the basis of observed regularities in the inventories of the world’s languages, the location of vowels in the vocalic triangle was accounted for by a selforganized process. According to it, vowels found their optimal positions by maximizing the perceptual distances between them. This maximal dispersion was related to the idea of maximal perceptual contrast: words whose sounds are maximally distant at the perceptual level are easier to discriminate. Lindblom et al.’s approach was further refined along different lines, taking consonants 135

136

Christophe Coupé, Egidio Marsico, and François Pellegrino

into account (Lindblom & Maddieson, 1988), refining the notion of perceptual distance, and shifting from maximal to sufficient perceptual contrast (Vallée, 1994; Schwartz et al., 1997). At the turn of the twenty-first century, the theory of complexity further pervaded the study of phonological systems. Shifting from variation-less inventories at a communal level to collections of speakers and individual inventories was one of the successful paths explored. Borrowing from Steels’ (1996) naming games for the emergence of lexical conventions, de Boer (2000) investigated the formation of individual, yet shareable and “synchronized,” phonological systems in a population of speakers. Multi-agent simulations were the key tool for considering self-organization “in motion,” with the gradual shaping of individual systems through myriads of interactions, in order to reach successful communication, much as one would assume in real situations. First attempts were enriched with more careful models of human perception and production, of language acquisition, of the dynamicity of speech, and of the aggregation of phonemes into higher units of speech (de Boer & Zuidema, 2005; Oudeyer, 2006; Au, 2008). 1.2

Specific Aspects of the Complexity of Phonological Inventories

Complexity theory intersects with fields as diverse as linguistics, epidemiology, and physics. While properties that transcend disciplinary boundaries have been found – such as the ubiquity of power laws – different systems have their own specific properties. Phonological inventories are no exception to the rule, as described next. As highlighted in the previous section, each language is an aggregate of individual idiolects. Each speaker indeed possesses their own phonological inventory. Therefore, speaking of Mandarin, French, or Tzeltal language does not reflect well the geographic distribution and the constantly evolving nature of speakers’ idiolects. That a reductionist approach is needed to characterize a higher-level object is common practice. However, two specific questions arise: (1) Does working on a single phonological inventory for a language make sense? And (2) does comparing different languages on this basis really help us understand the phenomena at hand? A second specific question regards the range of constraints that underlie the organization of phonological inventories and other structures of speech like syllables and words. These constraints refer to at least three domains: (i) speech perception, (ii) speech production, and (iii) speech cognition – which refers to properties of representations in cognitive areas like the mental lexicon. More is known about the first two domains, which exist in partial tension to each other. Indeed, phonological systems are trade-offs between “ease of articulation,” which minimizes articulatory efforts and favors articulatorily similar

To What Extent Are Phonological Inventories Complex Systems?

137

segments, and “perceptual salience,” which promotes maximum or at least sufficient perceptual distance between segments. A last characteristic of phonological inventories when considered as complex systems is the possibility to decompose phonemes into features. A vowel like [i] may be defined by the three features high, front, and unrounded, in relation to the required articulatory configuration to produce it: the tongue close to the palate, in the front of the mouth, with lips unrounded. Descriptive sets of features usually rely more on the articulatory side of speech, with only limited references to perception (e.g., sonorant vs. non-sonorant consonants). The imbrication of features into segments may be seen as an advantage: some phenomena at the level of phonemes may be explained in a simple way by properties of their features. However, it may also be seen as an additional difficulty: the relationship between features and segments may be convoluted and add up to a supplementary layer of complexity. On the one hand, principles like the “maximal use of available features” (MUAF) (Ohala, 1980) and “feature economy” (Clements, 2003b) – the tendency for features to be used parsimoniously to “create” phonemes in an inventory – help to explain observed regularities such as series of segments differing only in one feature, be it voicing, nasality, aspiration and so on. On the other hand, such series often present gaps, systems of the same size include varying features, and the like. Although other complex systems exhibit a layered structure and may be described at different scales1 , features and segments interact in specific ways, which may not be found in other systems and therefore deserve investigation. 1.3

A Quantitative Approach to Phonological Inventories as Complex Systems

A first step toward understanding complexity is to rely on qualitative definitions. The various approaches that Edmonds (1999) illustrates include: – complexity as a holistic feature: some systems are complex and escape traditional reductionist approaches, whereas others are not; or, from a gradualist perspective, some systems are more complex than others; – complexity as relative to the observer of a system is not necessarily inherent to the system itself; – complexity reflects the (meta)language adopted to describe the system. Along these lines, various general concepts such as the size or the internal variety of a system have been analyzed as inadequate when applied to 1

However, many well-known complex systems (e.g., spin glasses, ant colonies, and even human collective behaviors) do not require the consideration of causal interactions between several scales of description, and can be thoroughly analyzed at a single scale.

138

Christophe Coupé, Egidio Marsico, and François Pellegrino

complexity; they have been complemented with more sophisticated mathematical or computational concepts. Although “minimum description size,” “Kolmogorov complexity,” “Bennett’s logical depth,” “algorithmic information content,” or “effective complexity” (ibid; Gell-Mann & Lloyd, 2003) do better when it comes to assessing the complexity of a system, going beyond general considerations and applying them to specific systems appear to be difficult. A recurring problem is therefore to be able to shift from qualitative approaches to quantitative ones. Graph or network theory might today be an interesting framework to this end, given the wide range of fields in which its tools may be used, from statistical physics to epidemiology to sociology (Newman, 2010; for examples in linguistics, see Cancho & Solé, 2001; Dorogovtsev & Mendes, 2001; Ke et al., 2008). The previous considerations have been attested in the field of phonology, and more generally in linguistics (e.g., Dahl, 2004; Miestamo et al., 2008; Sampson, 2009). Efforts to choose how to correctly represent phonological systems in ways that make it easier to discuss complexity span over decades, from early generative phonology (Chomsky & Halle, 1968) to auto-segmental phonology (Goldsmith, 1990) and optimality theory (Prince & Smolensky, 2004). In more recent years, numerous studies have undertaken to revisit phonological systems in the light of complexity theory (e.g., Pellegrino et al., 2009). Among the main approaches, typological studies and interactions between phonology, morphology, and syntax have played a significant role (Plank, 1998; Fenk-Oczlon & Fenk, 2005; Shosted, 2006). Specifically, phonological inventories have been studied in a typological perspective, taking advantage of datasets covering hundreds of languages (Maddieson & Precoda, 1990; Dryer & Haspelmath, 2011). Interestingly, these attempts have often augmented with a quantitative dimension the more qualitative appearance of phonological inventories, which display an organization that escapes either full randomness or full regularity. Scales of complexity for consonantal, vocalic, or tonal systems have been defined, contrasted, and explored geographically (Maddieson, 2006; 2009; 2011); indices of the complexity or economy of phonological inventories have been defined (Marsico et al., 2004), and graph theory has been applied to their structure (Coupé et al., 2009).2 Different descriptive options of the inventories in terms of features have also been explored with computational methods (Coupé et al., 2011). In this chapter, we try to complement the previous approaches by questioning the degree of complexity of phonological inventories. The motives behind this attempt are as follows: without evading a cross-linguistic framework that has proven fruitful in recent years, we would like to come back to a more primary definition of complex systems: complex systems are first of all made of interacting elements, and their complexity arises from the nature of these 2

The statistical study presented hereafter partly draws on this contribution.

To What Extent Are Phonological Inventories Complex Systems?

139

interactions. Linguists familiar with computational studies in the emergence of language may refer here to Steels (1997)’s definition: …a complex dynamical system [ …] consists of a set of interacting elements where the behavior of the total is an indirect, non-hierarchical consequence of the behavior of the different parts. (p. 2)

This definition is appealing because we wish to rely on a basic approach that, without ignoring the specificities of phonological inventories in terms of features and segments, could also be applied to other inventories of linguistic elements. More precisely, we wish to propose a deductive methodology that can quantify the degree of relatedness of the building blocks of a collection of inventories. This would hopefully open the door to new investigations of today’s available digital linguistic datasets. In particular, we will see that this approach helps assess simultaneously the strengths of individual constraints. The previous notion of “indirect, non-hierarchical consequence” has also been expressed as non-linearity: in a complex system, the behavior of the whole cannot be linearly derived – that is, by way of superposition or addition – from the behavior of the different parts. In section 2 of this chapter, we propose a simple mathematical way to partially estimate the degree of non-linearity of phonological inventories. To be clear, our way to answer the questions “Are phonological inventories complex systems?” or “To what extent may phonological inventories be considered as complex systems?” will therefore be: “Can the compositional properties of phonological inventories be linearly derived from the properties of the segments and features that compose them?” 1.4

Relying on Available Data

We wish our approach to be faithful to available data on languages. Typologists have made clear that studying languages from a wide range of families leads to patterns that may be quite different from extrapolations from languages exhibiting a narrow range of structural diversity (Evans & Levinson, 2009). This statement should serve as a caveat when it comes to applying tools and notions of the theory of complexity. Many of them indeed come from statistical physics, where a simplification of the properties of the real phenomena at hand is often accepted as a first step of the analysis. This usually involves reducing the number of dimensions, considering open systems rather than systems with limits, and adopting a “mean field” perspective. This approach is often necessary to turn a mathematically intractable problem into a solvable one. However, one should beware of not neglecting the reality of available data for the sake of proposing a unifying framework. Regarding phonological inventories, this translates into the issue of available data, and how general principles of organization may be derived from an analysis of the observable diversity based on sound statistical principles. We rely on

140

Christophe Coupé, Egidio Marsico, and François Pellegrino

the UPSID database, compiled over the course of many years by Ian Maddieson (Maddieson, 1984; Maddieson & Precoda, 1990), a version of which was integrated in the World Atlas of Linguistic Structures (Dryer & Haspelmath, 2001). In our specific version of the database, 451 inventories were available in a balanced way across linguistic families and geographic areas. This is less than one tenth of all languages spoken on Earth today; however, given the sampling methods adopted, it is likely to be representative of the worldwide diversity. The sustained efforts required to collect and compile the data are sometimes met with suspicion regarding their correctness (see Clements, 2003a, for some of the problems met). One also has to keep in mind that such synchronic collections are snapshots of diachronic flows that often carry ongoing sound changes and shape the dynamical equilibriums observed at any moment in time. Nevertheless, large-scale typological studies likely compensate for errors at the level of individual languages. In section 2, we will present our mathematical approach to interactions between segments and features in phonological inventories, and explain how we take into account the limited amount of evidence offered by the available datasets. 1.5

Objectives

Given the background introduced in the previous sections, we can summarize our approach and its objectives as follows: – we adopt a deductive data mining approach with the UPSID dataset; – we focus on interactions between constituents of phonological inventories; – we quantify the degree of non-linearity of these interactions to estimate the degree of complexity of the inventories; – we further analyze the interactions to confirm or infirm current hypotheses on the structuration of the inventories; and – we hope to discover new facts and formulate new hypotheses on their basis. In section 2, we describe our methodology and the choice we made in terms of statistical analysis. We then provide the results of our computations and how we interpret them. Finally, conclusions and perspectives are given in section 4. 2

Methodology

2.1

Defining Simple Phonological Inventories and a First Order of Complexity

For the sake of clarity, we start by defining a case where the complexity of phonological inventories could be considered as minimal, or at least very low.

To What Extent Are Phonological Inventories Complex Systems?

141

A first step towards understanding how the world’s languages build their words consists in paying attention to phonemes independently from each other, and estimating their probability of occurrence. It is well known that while some sounds like [i], [u], [p], or [k] are very frequent, others only appear in small percentages of languages – for example around 5 percent for [y]. From the UPSID dataset, it is easy to compute the individual frequencies of the approximately 900 segments and 100 features that compose the inventories. Let us assume that we can infer the composition of inventories on the sole basis of this heterogeneous distribution of segments. The presence or absence of any given phoneme in any inventory would be the reflection of its overall frequency in the whole set of inventories; it would not depend on other segments being present or absent in this inventory. In such a case, the compositional properties of the inventories would directly derive from the individual properties of segments that compose them. If such a situation were true, we could refute the idea that phonological inventories are complex systems. On the contrary, estimating the extent to which the composition of an inventory departs from what can be expected from the overall frequencies of its segments would provide a measure of its complexity. Our measure only partially encompasses the notion of complexity for inventories. Indeed, it relies on pairwise interactions between segments or features, and therefore decomposes potentially very complex relationships into much simpler ones. Such a reductionist approach directly derives from models in physics or sociophysics (e.g., Jensen, 2006), and echoes both the basic structure of graphs with nodes connected by vertices, and interactions in multi-agent models of language emergence and evolution. We may say that it relates to a first order of complexity; a “0-order” would relate to the absence of interactions between the components of the inventories, an order higher than one to complex interactions between more than two components. We may now turn to the actual evaluation of pairwise interactions. 2.2

Evaluating the Interaction Between Two Components of Inventories

Considering a pair of two components, how can we estimate their degree of interaction based on their frequencies of appearance in the available inventories? We have adopted Clements’ (2003a; 2003b) assessment of feature economy in the UPSID inventory. Assuming that speech sounds tend to be composed of features that are used elsewhere in the system, Clements tested two complementary hypotheses: (i) a given sound will occur more frequently in systems in which all of its features are distinctively present in other sounds (mutual attraction); and (ii) a given sound will have a lower than expected frequency in systems in which one or more of its features are not distinctively present in other sounds. To this end, he set up a number of contingency tables (see

142

Christophe Coupé, Egidio Marsico, and François Pellegrino

Table 6.1 Observed frequencies of voiced labial fricatives and voiced coronal fricatives across languages in UPSID. Expected frequencies are shown in parentheses. Reproduced from Clements, 2003a Voiced Coronal Fricatives

Voiced labial fricatives

Present Absent Total

Present

Absent

Total

110 (57) 65 (118) 175

37 (90) 239 (186) 276

147 304 451

Table 6.1) to investigate feature economy for various categories of sounds and features. To evaluate the significance of the differences between observed and expected distributions, Clements used the χ² test.3 For example, Table 6.1 led him to the conclusion that the association between voiced labial fricatives and voiced coronal fricatives in inventories was significantly positive (χ² = 119.203, p < 0.0001). This was an argument for the economy of the features [+continuant] and [+voiced] shared by these consonants.4 We have retained from Clements’ studies both the statistical approach and the following rationale: if two components appear in inventories together significantly more or significantly less than what their individual frequencies would predict, they may be said to interact, and the deviation from an absence of interaction reflects the strength of this interaction. In other words, our null hypothesis (H0) is the following one: there is no interaction between two components, that is, the frequency of their co-occurrence in inventories can be deduced from their individual frequencies – more precisely the frequency of co-occurrence is the product of the two individual frequencies. We investigated cases for which the null hypothesis could be rejected, and further estimated the confidence one could have in these rejections. We tested interactions for all pairs of segments and all pairs of features. Our findings are summarized in section 3. The test was conducted to evaluate the first-order complexity of inventories on the basis of the number of significant interactions. The χ² is actually only an approximate test of the statistical significance of a deviation from a theoretically expected distribution of observations (with two categories like “present” or “absent”). Its low computational cost 3 4

With Yates’ correction for tables with cells containing values of seven or less. Clements also tested for genetic or areal skewing by examining distributions in different groups of languages.

To What Extent Are Phonological Inventories Complex Systems?

143

Table 6.2 Observed Frequencies of /a/ and /ã/ Across Languages in UPSID (Expected Frequencies in Parentheses) /ã/ Present

/a/

Present Absent Total

82 (72) 1 (11) 83

Absent

Total

310 (320) 58 (48) 368

392 59 451

and its adequacy for moderately large values explain why it is often preferred to the binomial test, although the latter is more exact. However, because higher computational costs were not a serious issue for us, and because we were willing to address cases where very frequent or very rare components were considered, we chose to rely on the binomial test anyway. We also considered not only the associations of two components A and B both present in inventories, but more generally the four possibilities A & B, !A & B, A & !B, !A & !B, where “!” stands for the absence of a component. This allowed us to consider additional relevant interactions, as exemplified in Table 6.2. Out of the four associations /a/ & /ã/, !/a/ & /ã/, /a/ & !/ã/, !/a/ & !/ã/, only the second one departs significantly from what can be expected given the individual frequencies of /a/ and /ã/ (p < 0.001, when p > 0.05 for all other associations). Especially, looking at the positive association of /a/ and /ã/ does not reveal any significant trend, likely because of the high frequency of /a/. After other pairs of oral vowels and their nasalized counterparts have been tested, the following conclusion can be drawn – and only this one: inventories where a nasalized vowel is present without its oral counterpart are significantly disfavored. This may be explained by the mechanism of transphonologization through which nasalized vowels derive from their oral counterparts by acquiring the nasal feature of an adjacent consonant. Without the oral vowel, the nasalized counterpart cannot appear, and rare cases like !/a/&/ã/ only occur when the oral vowel subsequently changes or disappears – usually by a change in quality. This example illustrates why looking at absent components may be interesting. It also shows how synchronic inventories may reveal relevant diachronic processes, although in an implicit manner. All in all, we took a binary approach to inventories, as they were not only described by the set of phonemes they contained, but also by the set of all the others they did not. An association of two components may occur significantly more or less than expected (in terms of individual frequencies). Additionally, several associations among the four possibilities may be shown to be statistically significant. This makes sense when one thinks of the redundancy of the four possible

144

Christophe Coupé, Egidio Marsico, and François Pellegrino

cases. Three combinations frequently found can be highlighted. Assuming that “> 0” means a frequency greater than expected and “< 0” a frequency lower than expected: – As in the previous example, !A&B < 0 means that the presence of B implies the presence of A. This association may appear or not with other associations; – A&B > 0, !A&!B > 0, !A&B < 0, A&!B < 0 indicate that the two components A and B mutually attract each other; – A&B < 0, !A&!B < 0, !A&B > 0, A&!B > 0 indicate that the two components A and B mutually repel each other. A limitation was the difficulty to automatically take into account what Clements (2003a) called covert attractors and subset effects. Covert attractors referred to situations where a significant association between A and B is due to the covert influence of a third component C, rather than to a direct relation between A and B. Subset effects pointed to positive associations between A and B due to a subset of A rather than to A as a whole. These two cases ultimately arise from decomposing structural complexity into pairwise interactions only. Because we analyzed all possible pairs of components, it was difficult to detect and test such situations, where Clements was guided by its theoretical framework and its restriction to a limited number of cases. We therefore could only keep in mind that the number of significant associations was likely an overestimation of the real interactions at play. 2.3

Evaluating the Interaction Between Multiple Pairs of Components

Our approach relied on the application of the previous statistical test to all possible pairs of components composing the UPSID inventories. We therefore had to face the problem of multiple comparisons (Hochberg & Tamhane, 1987), which is known to inflate the type I error rate (i.e., the detection of false positives). In our case, this meant some p-values for the binomial test smaller than the significance threshold (the standard 0.05 value) falsely leading to the rejection of the null hypothesis of no interaction between the two components. We considered different corrections of the standard test to address this issue, often referred to as controlling the family-wise error rate (FWER). Among the better known methods is the Bonferroni correction (Dunn, 1961). It states that for m tests, rejecting p-values of tests smaller than α/m guarantees a FWER lower than α. This correction is however very conservative. Although more powerful and still controlling for the FWER, the Šidák correction and Hochberg’s “step-up” procedure could not be considered, because they required the tests to be independent of each other (Šidák, 1979; Hochberg, 1988). On the other hand, Holm’s so called “step-down” procedure, also called the HolmBonferroni method, does not require independence (Holm, 1979), and guarantees control for the FWER. Finally, because of various theoretical issues, some

To What Extent Are Phonological Inventories Complex Systems?

145

recommend to stick to the per comparison error rate (PCER) and ignore the multiplicity problem (Saville, 1990). Benjamini and Hochberg (1995) provided an analysis of the previous tests and of their contexts of use which resonated with our own approach: – The previous procedures control the FWER in a strong sense. “At levels conventional in single-comparison problems,” e.g., 0.05, they “tend to have substantially less power than the per comparison procedure of the same levels” (ibid, p. 290) – “Often the control of the FWER is not quite needed. The control of the FWER is important when a conclusion from the various individual inferences is likely to be erroneous when at least one of them is” (ibid, p. 290). In our own case, we did not intend to draw conclusions on the basis of at least one significant association between two components of the inventories. We also did not want to increase the number of false negatives too much for the sake of avoiding false positives at all costs. Instead, we tried to optimize the statistical power overall. We therefore eventually followed Benjamini and Hochberg’s approach, with controls for a false detection rate (FDR) rather than the FWER. Rather than taking into account the question whether any error was made, the authors focused on the number of erroneous rejections. Controlling for the FDR amounts to controlling the FWER in a weaker sense; indeed, what is contained is the expected proportion of errors among the rejected hypotheses. Concretely in our case, the procedure guaranteed that the number of false positives in the discovered significant associations was lower than a chosen threshold, which we took to be equal to 5 percent. The algorithm applied to our study unfolded as follows: if m associations of components were considered and m p-values p1 , p2 , …pm obtained with the binomial test, – The associations A1 , A2 , … Am were ordered in ascending order of their p-values. – Starting from the first, that is, smallest p-value, for a given threshold α (which we considered equal to 0.05), K was computed as the largest k such that pk ࣘ α x k / m. – The null hypothesis was rejected for, and only for, the associations A1 , A2 , … AK , that is, these associations were accepted as expressing true interactions between the components each was made of. 2.4

Dealing with Limited Amounts of Data

Relying on the limits of statistical tests like the χ² or the binomial test and applying relevant corrections to multiple tests were the first ways to deal with the limited amount of data offered by the UPSID database, and to draw conclusions from them and only from them.

146

Christophe Coupé, Egidio Marsico, and François Pellegrino

Table 6.3 Categorization of the Different Pairwise Associations of Components of Phonological Inventories According to our Statistical Approach Meaningful associations

Significant associations Non-significant associations

Meaningless associations

Non-significant associations

In cases where the overall frequency of a component in the inventories was either very low or very high, checking associations with other components was likely to result in non-significant results (except in the case of rare segments most often occurring together in inventories). This explained why very frequent segments like /a/, /i/, or /u/, overall did not strongly interact with other segments according to the binomial test. One may see this as frustrating, because not much can then be said about the relations these very frequent segments have with others. On the contrary, we see this as an opportunity to reflect the available amount of information in a neutral way. This refers to the idea of the aforementioned deductive approach. To further understand the extent to which our limited information constrained what could be found, we posited a distinction between two categories of associations: meaningful and meaningless ones. On the one hand, all statistically significant associations were meaningful. On the other hand, non-significant associations were either meaningful or not, depending on whether the individual frequencies of the two components could have led to a significant association, or whether their values were too extreme for this. In the first case, the dataset contained enough information to potentially detect a significant association, and it turned out not to be the case. In the second case, nothing could be expected, therefore the term “meaningless.” The distinction allowed us to compute percentages of significant associations on the basis of meaningful associations, rather than on the basis of tested associations (see section 3). This approach is summarized in Table 6.3. Detecting meaningful non-significant associations once again implied multiple comparisons. For each non-significant association, we computed the smallest p-value that could be obtained given the individual frequencies of the components. We then applied Benjamini and Hochberg’s procedure on all these p-values to estimate which non-significant associations could be considered meaningful while controlling for the FDR with a 5 percent error rate. The approach we considered with pairs of components could potentially be applied to n-uplets with n > 2. However, in practice, the size of the UPSID database was too small to correctly estimate meaningful associations but for a limited number of n-uplets. This would have seriously hampered our approach.

To What Extent Are Phonological Inventories Complex Systems?

147

The explosion of the number of combinations would also have made the analysis and presentation of the results significantly more difficult.

3

Results and Interpretations

3.1

Overview

We considered two levels of analysis for UPSID: segments and features. Our approach gave quantitative measures and an estimation of first-degree complexity for each level. Each time, we considered vowels and consonants separately, but also gave figures for all segments or features considered together. This included the addition of diphthongs, although we chose to exclude them from our analyses for the sake of clarity and because they are overall less studied and understood than monophthongs and consonants. To be clear, a feature was present in a phonological inventory if at least one segment containing it was part of that inventory. Two features could therefore co-occur in an inventory whether or not they co-occurred in the same segment. In each case, we computed, along with a few other figures, the number of tested associations,5 the number of significant associations, and non-significant meaningful and non-significant meaningless associations. The percentage of significant associations in meaningful associations revealed the degree of first-order complexity of the inventories at the chosen level of description and for the set of components – vowels, consonants, vocalic features, consonantal features, all segments or all features – considered. Knowing the significant associations, we could also compute the number of pairs that presented at least one significant association among the four possible configurations A&B, !A&B, A&!B, and !A&!B. The percentage of “significant pairs” – with respect to the total number of pairs of components – could then be computed, and give another estimation of the first-order complexity. We also investigated the distribution of significant associations according to their p-values, in order to go beyond the single figure of their numbers and possibly observe different distributional configurations. More precisely, pK being the threshold for single comparisons as defined in section 2.3, we divided the associations into groups defined by the limits [pK /10n+1 , pK /10n ] (n being an integer). Finally, we reported the most significant associations sorted by p-values. This helped to shift from our somewhat “a-theoretical” approach to phonological

5

The number of tested associations is equal to four times the number of possible pairs of components. This value corresponds to the four interactions A&B,A&!B, !A&B, !A&!B; n being the number of components, the number of possible pairs is equal to C(n,2) = n.(n − 1)/2.

148

Christophe Coupé, Egidio Marsico, and François Pellegrino

Table 6.4 Comparison of Four Different Approaches to Multiple Testing for Vocalic Features PCER # of components # of tested associations # of significant associations # of non-significant associations

35 1,265

B

3 1,297

HB

FDR

3 1,297

26 1300 5 1,295

theory and to see if the first one could bring new arguments to the second. Associations can be partitioned into two main categories, according to whether the observed number of occurrences is smaller or greater than what was expected. 3.2

Comparing Corrections for Multiple Testing

In connection with the discussion in section 2.3, we compared the figures obtained with different approaches to the problem of inflated type I error rate in multiple testing: – no correction, that is, control for the per comparison error rate (PCER) – The Bonferroni correction (B) – The Holm-Bonferroni procedure (HB) – Benjamini and Hochberg’s control for the false discovery rate (FDR) rather than for the family-wise error rate (FWER). Tables 6.4 and 6.5 offer data for these four procedures in the case of vocalic features and consonants.6 The choice of these two cases was guided by their clear distinction in terms of number of components and number of tested associations. The results are in agreement with what could be expected in terms of conservatism: the order of the four procedures, ranked according to the number of detected significant associations, is PCER > FDR > HB > = B. Note also that even with the less conservative approach, PCER, the number of significant associations is but a small fraction of the total number of tested associations. In both cases, the Bonferroni and Holm-Bonferroni corrections provided identical results; the Benjamini-Hochberg procedure returned more associations, especially in the case of consonants, but was still very far from the low degree of conservatism of the control for PCER. In the following paragraphs, all data given are based on Benjamini and Hochberg’s method for estimating significant associations and meaningful as opposed to meaningless ones. 6

For the sake of clarity, we did not provide data here about meaningful associations, as we always used Benjamini and Hochberg’s procedure for them.

To What Extent Are Phonological Inventories Complex Systems?

149

Table 6.5 Comparison of Four Different Approaches to the Statistical Issue of Multiple Testing for Consonants PCER # of components # of tested associations # of significant associations # of non-significant associations

3.3

10,754 665,530

B

249 676,035

HB

FDR 582 676,284 731 675,553

249 676,035

Complexity of the Interactions Between Segments

Tables 6.6, 6.7, 6.8, and Figure 6.1 provide the main results for the interactions between vowels, between consonants, and between all segments in UPSID phonological inventories. It first appears that the number of significant associations is very small compared to the number of meaningful associations – the maximum is 1.8 percent for vowels – and therefore even smaller compared to the number of tested associations. Being careful with what the UPSID database may reliably tell us about associations of segments, our conclusion is as follows: it is not possible to rule out that segments interact very little with each other, to the exception of a few strong interactions that we will further describe later. The percentage of significant pairs is also very small, always less than 1.5 percent, which corroborates the previous idea. Once again, it is important to remember here and in what follows that there might be a significant number of false negatives. We can note here that even with a more relaxed approach, which is controlling only

Table 6.6 Estimation of Pairwise Interactions at the Segmental Level

# of elements # of pairs # of tested associations # of significant associations # of meaningful pairs # of non-significant associations # of meaningful non-significant associations % of meaningful associations (w.r.t. tested ones) % of significant associations (w.r.t. meaningful ones) % of significant pairs (w.r.t. total number of pairs)

Vowels

Consonants

All segments

162 13,041 52,164 292 200 51,872 15,621 30.5% 1.8% 1.5%

582 169,071 676,284 731 554 675,553 180,200 26.8% 0.4% 0.3%

833 346,528 1,386,112 851 617 1,385,261 370,611 26.8% 0.2% 0.2%

150

Christophe Coupé, Egidio Marsico, and François Pellegrino

Table 6.7 The 20 Vowel-Vowel Associations with the Smallest p-values

ID

Seg. 1

Seg. 2

Assoc.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

u˜ ã ã uː oː õ ʊ u˜ ã õ ɔ ã ã ɔ̃ ã o oː u ɪ o

˜ı ˜ı u˜ iː eː e˜ ɪ ˜ı õ ˜ı ɛ ˜ı u˜ ɛ̃ ˜ı ɘ iː i i e

A&B A&B A&B A&B A&B A&B A&B A & !B A&B A&B A & !B !A & B !A & B A&B A & !B !A & B A&B !A & !B A & !B !A & !B

300

Observed

Expected

Log 10 (p-value)

69 72 67 36 32 47 55 5 54 53 26 10 7 27 11 25 29 49 46 116

13 15 13 3 2 7 10 60 11 11 95 66 60 2 67 90 3 10 9 50

−27,7 −27,0 −25,8 −25,3 −23,9 −23,4 −21,8 −21,1 −20,0 −19,5 −19,2 −19,0 −18,9 −18,7 −18,6 −18,1 −18,0 −17,9 −17,4 −17,4

279

Distributions of Segmental Interactions 250

200

150

100

130 86

81 64

50

0 pK.10-1

45 42 39 32 25 22 20 18 10 8 5 514 13 5 7 9 65 4 3 2 2 33 6 2 1 3 1 3 1 2 1 3 11 1 2 1 2 11 1 pK.10-5

pK.10-9

pK.10-13 Vowels

pK.10-17

pK.10-21

1

1

pK.10-25

Consonants

Figure 6.1 Distributions of segmental associations according to the value of the binomial test.

To What Extent Are Phonological Inventories Complex Systems?

151

Table 6.8 The 20 Consonant-Consonant Associations with the Smallest p-values

ID

Seg. 1

Seg. 2

Assoc.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

ɡ kʰ d kʰ ŋɡ nd kʰ ŋɡ  gb

b pʰ b pʰ mb mb pʰ nd  kp

tʰ kʰ k n   t 

pʰ tʰ t n̪ k t̪ b p b ɓ pʰ

A & !B A&B A & !B !A & B A&B A&B A & !B A&B A&B A&B A&B A&B !A & B A&B A&B !A & B A&B !A & !B A&B A & !B

n̪ ɡ k ɡ ɗ tʰ

Observed

Expected

Log 10 (p-value)

9 94 6 7 43 42 9 39 34 75 75 48 69 45 73 43 45 155 35 8

92 23 76 77 4 4 79 4 3 18 18 7 16 6 19 126 7 72 4 64

−31,7 −30,5 −27,6 −27,0 −26,7 −26,0 −25,9 −24,4 −23,9 −23,8 −23,3 −23,0 −22,9 −21,9 −21,3 −21,3 −21,2 −21,1 −21,0 −19,8

for PCER and not for FDR, the percentage of significant pairs and significant associations would still be low, in the order of 5 percent. The percentage of meaningful associations, whether significant or not, is around a quarter or a third of tested associations, and this even after controlling for FDR. This means that the percentage of significant associations and significant pairs is established upon a relatively high number of meaningful associations, which underscores the assumption that there is a significant amount of information in the inventories regarding segment-segment associations. A very low percentage of meaningful associations would have undermined the credibility of the percentages given in the previous paragraph. As could have been expected, Figure 6.1 reveals an exponential-like decay of the number of significant associations as the value of the binomial test becomes smaller. The long tails of consonant and vowel distributions indicate a number of very strongly significant associations, which therefore deserve further examination because they may tell us something important about the structure of phonological inventories, and potentially what the strongest forces are at play in their organizations. The exponential-like decay also once again

152 i

Christophe Coupé, Egidio Marsico, and François Pellegrino u ʊ o

ɪ e ɛ

ɔ a+

a

ũ

ı̃

õ

ẽ

u:

i: e:

o:

ɔ̃

ɛ̃ ã

a:

Figure 6.2 Graphical representation of the relations between vowels. Solid lines correspond to relations of mutual attraction, dotted lines to relations of repulsion.

reinforces the idea that the interactions between segments are very limited, because among significant associations, many do not depart much from the threshold of the statistical test. For vowels, the most significant associations involve, 14 times out of 20, a secondary feature: nasality (11) or length (3). The six others include height and frontness. If one looks more broadly at all significant associations, these tendencies are confirmed.7 All in all, the main results are the following: – For most pairs of either nasalized vowels or long vowels, the two phonemes tend (i) to occur together more than expected and (ii) to be both absent more than expected; and (iii) situations where one is present and the other one is absent are much less frequent than expected. These patterns clearly suggest mutual attraction between nasal vowels and between long vowels.8 In other words, a vowel with a secondary articulation is seldom isolated in a vocalic system. This result is in line with the MUAF principle and the notion of feature economy (see section 1.2). – As far as height and frontness are concerned, our results confirm the wellknown dispersion theory (Liljencrants & Lindblom, 1972), according to which front and back dimensions are balanced for a given height, and neighboring height degrees tend to repeal each other. Figure 6.2 summarizes the relations of mutual attraction or repulsion for primary vowels as well as nasalized and long vowels. Mutual attraction and reinforcement are particularly visible for nasalized vowels, and the repulsion between high and lowered-high primary vowels clearly illustrate the avoidance of neighboring height degrees in a system. Interestingly, both MUAF and dispersion theory are systemic principles recovered here via one-to-one segments interactions, meaning that our approach can reach the “collective” level, although in a non-straightforward 7 8

Other secondary features appear, although patterns of mutual attraction or repulsion only clearly appear for nasality and length. One of the four possible cases is missing for some pairs, but there is no case where mutual attraction seems contradicted.

To What Extent Are Phonological Inventories Complex Systems?

153

way. It adds the benefice of being able to rank these principles according to their prominence in inventories. The results are similar for consonants, where most of the twenty top significant associations (with the exception of #9, #13, and #15) involve series of stop consonants. They show attraction between pairs of series of elements based on some shared feature (voiced, aspirated, ejective or pre-nasalized), and confirm that phonological systems have a highly productive use of features. Associations #13 and #15 more directly involve place of articulation, particularly dental consonants. Association #13 suggests repulsion between the alveolar nasal /n/ and its dental counterpart, which is confirmed by three further associations ranked #36, #50, and #236; /n/ and its dental counterpart tend to avoid each other. Association #15, on the contrary, suggests that we find many more systems containing dental /n/ and dental /t/ than expected, and further associations #26, #76, and #240 confirm that there is a mutual attraction between them. Our analysis is that both associations relate to the fact that the coronal place of articulation is somehow unspecified (Maddieson, 1984) in systems; the most frequent choice is alveolar, but some languages choose a dental articulation, and when they do so, it happens across manner of articulation, hence association #15. Very rarely, languages have both place of articulation (especially with the same manner); in UPSID, only fourteen languages have both /n/ and its dental counterpart (which is much less than expected, as highlighted by association #36), and among them nine are Australian languages, which are well known for contrasting consonants on more places of articulation than usual, especially in the coronal region. The other five languages are scattered over five different families. When we examine the associations for segments considered altogether, the most significant ones only concern vowel-to-vowel or consonant-to-consonant associations. As shown in Table 6.9, only twelve associations involve a vowel and a consonant, which are ranked moreover between #331 and #815 out of the 851 total significant associations. Strangely enough, except for the 11th one, they all involve a labial-velar stop ( kp or  gb ) and a lower-mid vowel (ɛ or ɔ), either oral or nasal. Any perceptual or articulatory explanation for these relationships would probably seem too farfetched. In fact, a closer look at the data reveals a family effect. Table 6.10 gives the distribution of the various segments in both the whole UPSID database and the Niger-Congo family (which contains 55 languages). The second column of Table 6.10 gives the numbers and percentages of occurrence of the segments in the UPSID database. Looking at the third column, a strong bias in the distribution of labial-velars in favor of the NigerCongo family appears. As a matter of fact, around 90 percent of the languages having one or the other is in the Niger-Congo family. This isolates this family as the relevant subset of languages to investigate. As for the more frequent

154

Christophe Coupé, Egidio Marsico, and François Pellegrino

Table 6.9 Vowel-Consonant Significant Associations in the Inventories

ID 1 2 3 4 5 6 7 8 9 10 11 12

Seg. 1

Seg. 2

Assoc.

Observed

Expected

Log 10 (p-value)

 kp  kp  gb  gb  gb  gb  gb  kp  kp  kp

ɔ̃ ɛ̃ ɔ̃ ɛ̃ ɔ ɛ ɔ ɔ ɛ ɔ õ ɛ

A&B A&B A&B A&B A & !B A & !B A&B A & !B A & !B A&B !A & B A&B

15 15 15 15 5 4 34 5 4 30 14 35

2 2 2 3 24 22 14 22 20 12 3 16

−7.7 −6.8 −6.7 −6.2 −6.0 −6.0 −5.5 −5.1 −5.1 −4.8 −4.7 −4.7

m  gb

higher-mid and lower-mid oral vowels, their distribution across linguistic families is broader; they however appear in a large majority of Niger-Congo languages. Their nasalized counterparts are rarer in UPSID, but still appear in 19 Niger-Congo languages. These figures, when compared to the total size of the family (55 languages), lead to percentages of occurrence in the family all much higher than those found in the whole UPSID database. The Niger-Congo languages thus form a subset where nearly all labial-velar stops ( kp or  gb ) co-occur with more numerous than average higher-mid or lower-mid vowels, whether oral or nasal. It is this peculiar distribution in the family that makes the associations significant for the whole sample. Table 6.10 Distribution of the Segments Involved in the Significant Associations Between a Vowel and a Consonant (N-C stands for Niger-Congo, # for Number, % for Percentage)

Segment  kp  gb ɛ ɔ ɛ̃ ɔ̃

# and % of languages containing the segment (in UPSID)

# of N-C languages containing the segment

% of the N-C languages containing the segment

35 (8%) 39 (9%) 187 (41%) 163 (36%) 35 (8%) 32 (8%)

32 34 44 45 19 19

58% 62% 80% 82% 35% 35%

To What Extent Are Phonological Inventories Complex Systems?

155

Table 6.11 Estimation of Pairwise Interactions at the Feature Level

# of elements # of pairs # of tested associations # of significant associations # of meaningful pairs # of non-significant associations # of meaningful non-significant associations % of meaningful associations (w.r.t. tested ones) % of significant associations (w.r.t. meaningful ones) % of significant pairs (w.r.t. total number of pairs)

Vocalic Features

Consonantal Features

All Features

26 325 1,300 5 4 1,295 513 39.9% 1.2% 1.5%

55 1,485 5,940 77 53 5,863 2,679 46.4% 2.8% 3.6%

97 4,656 18,624 145 104 18,479 8,017 43.8% 1.8% 2.2%

This result indicates that our approach makes it possible to find genetic (and also areal) traits in the organization of phonological inventories, although again not straightforwardly. This suggests that a more systematic exploration of these parameters should be done, to distinguish between broadly distributed significant co-occurrences and more localized ones. This will be investigated more thoroughly in future work. 3.4

Complexity of the Interactions Between Features

This section reports our findings at the level of features. As in the previous section, Tables 6.11, 6.12, and 6.13 and Figure 6.3 provide our main findings for the interactions between vocalic features, consonant features, and all features (including diphthong features) in UPSID phonological inventories. Given the smaller number of features, the number of elements and associations tested were smaller than for segments. However, the different percentages used to assess first-order complexity, despite being slightly higher, are of the same magnitude, with values between 1.2 and 3.6 for the percentages of significant associations and significant pairs. Once again, based on information that can reliably be extracted from the UPSID database for associations of features, it cannot be concluded that features interact with each other, except for a few pairs. It is tempting to claim that at least for vowels, the features that compose the segments of the inventories occur predominantly on the basis of their individual frequencies, owing to their intrinsic articulatory or perceptual properties. This statement is limited by the potential existence of a high number of false negatives. However, the percentages of meaningful associations are higher than for segments, between 40 and 50 percent. These values hence favor the reliability of the previous percentages.

156

Christophe Coupé, Egidio Marsico, and François Pellegrino

Table 6.12 The 20 Associations of Vocalic Features with the Smallest p-values. Only the first five are significant

ID

Feature 1

Feature 2

Assoc.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

lowered-high9 long lowered-high low lip-compressed nasalized nasalized retracted central long low higher-mid central creaky-voiced lowered-high breathy-voiced lower-mid retroflexed raised-low10 rounded

high velarized high raised-low raised velarized velarized advanced low velarized raised-low nasalized mid voiced retracted voiced lowered-high lip-compressed long high

A & !B A&B !A & !B !A & B A&B A&B !A & B A&B !A & !B !A & B !A & !B !A & B !A & B A & !B A&B A & !B !A & B A&B A&B !A & !B

Observed

Expected

Log 10 (p–value)

21 8 4 8 2 9 0 2 4 1 1 9 0 1 4 1 31 1 14 2

5 1 19 1 0 2 6 0 0 7 7 19 4 0 0 0 44 0 7 0

−7,1 −5,0 −4,9 −4,2 −4,1 −3,6 −3,0 −3,0 −2,7 −2,5 −2,4 −2,3 −2,2 −2,1 −2,0 −2,0 −1,8 −1,8 −1,8 −1,7

The number of significant associations for vocalic features was extremely small: only 5 out of 1,300 were detected by the statistical approach. By comparison, there were 35 of them when controlling only for the PCER (305 and 649, respectively, for consonantal features and all features; controlling for FDR gives 77 and 145). The distributions of significant associations according to the value of the binomial test (Figure 6.3) once again suggest exponential-like decay, at least for consonantal features (and also for all features – not shown). At the segmental level, the principle of feature economy could be advocated as a likely explanation for many of the observed interactions. This is no longer the case with features, and significant associations may therefore tell us something different about phonological inventories. For vowels, close to nothing could be said on the basis of the five significant associations. For the sake of analyzing the data a bit further, we looked at the 35 interactions returned by controlling for PCER. The following proposals 9 10

This feature is equivalent to near-close. This feature is equivalent to near-open.

To What Extent Are Phonological Inventories Complex Systems?

157

Table 6.13 The 20 Associations of Consonantal Features with the Smallest p-values

ID

Feature 1

Feature 2

Assoc.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

labial-velar labial-velar palatal non-sibilant-fricative labiodental lateral-fricative postalveolar ejective glottal postalveolar postalveolar click ejective glottal ejective trill-or-unspecified nasalized nasalized glottalized glottalized

approximant approximant approximant sibilant-fricative non-sibilant-fricative affricate-lateral sibilant-affricate affricate-lateral non-sibilant-fricative sibilant-affricate sibilant-affricate affricate lateral-fricative non-sibilant-fricative affricate-lateral flap affricate click affricate click

A & !B !A & !B !A & !B !A & !B A & !B A&B !A & !B A&B !A & !B A & !B !A & B A&B A&B A & !B !A & B A&B A&B A&B A&B A&B

Observed

Expected

Log 10 (p-value)

0 45 29 32 0 21 109 24 45 51 52 6 32 12 2 26 5 5 4 4

33 11 4 6 25 3 57 4 14 102 103 0 8 42 21 61 0 0 0 0

−15,3 −14,2 −14,0 −11,9 −11,5 −10,9 −10,7 −10,6 −10,4 −9,6 −9,5 −9,5 −9,3 −7,9 −7,2 −7,2 −6,7 −6,7 −6,5 −6,5

30 27

Distributions of Feature Interactions 25

20 16 14

15

10 6 5

4

4 1

4 2

1

2

1

0 pK.10-1

pK.10-2

pK.10-3

pK.10-4

pK.10-5

pK.10-6

Vocalic features

pK.10-7

pK.10-8

pK.10-9 pK.10-10 pK.10-11 pK.10-12 pK.10-13

Consonantal features

Figure 6.3 Distributions of feature associations according to the value of the binomial test.

158

Christophe Coupé, Egidio Marsico, and François Pellegrino

Table 6.14 Pairs of Consonantal Features Characterized by Mutual Attraction and Represented in the 20 Most Significant Associations Feature 1 postalveolar approximant palatal sibilant-fricative non-sibilant-fricative lateral-fricative ejective glottal click

Feature 2 Place Mode Place Mode Mode Mode Secondary Place Mode

sibilant-affricate labial-velar approximant non-sibilant-fricative labiodental lateral-affricate lateral-affricate non-sibilant-fricative affricate

Mode Place Mode Mode Place Mode Mode Mode Mode

must therefore be treated with great caution: few clear patterns could be distinguished overall except for the mutual repulsion between the features high and lowered-high (near-close), which echoed the idea of sufficient perceptual contrast already mentioned at the segmental level. The presence of velarized vowels seemed to imply those of nasalized vowels as well as those of long vowels. Raised-low (near-open) vowels seemed to appear when low vowels were absent. Regarding consonantal features, many of the interactions turned out to be mutual attraction. Analyzing all significant associations while controlling for PCER, we found in particular that the 16 most significant associations presented in Table 6.13 were all complemented by other associations supporting mutual attraction as depicted in section 2.2. Looking only at significant associations with control for FDR, mutual attraction was clearly suggested, but without all cues being available: three or four associations were found among the 77 significant associations for most of the 16 most significant associations. On the other hand, only one relation of mutual repulsion was found among the 77 “FDR-significant” associations, with all four associations confirming it: it was for the pair trill-or-unspecified and flap. Table 6.14 shows that different pairs of mutually attracted consonantal features may be composed of features relating to: (i) two modes of articulation, (ii) a mode and a place of articulation, or (iii) a secondary feature and a mode. None of these pairs can be explained straightforwardly, but they give rise to interesting hypotheses in many directions, whether at the articulatory, perceptual, historical, genetic, or areal level. All these questions are worth investigating but more work is required to firmly ground any proposal. Finally, we found very few mutual attractions between pairs of features related to two places of articulation, one exception being between palatal and labial-velar.

To What Extent Are Phonological Inventories Complex Systems?

159

Our study of associations between all features revealed many significant interactions between features appearing exclusively in diphthongs. Interestingly too, leaving aside secondary features carried by either vowels or consonants, no significant associations were found between consonantal and vocalic features, or between diphthong features and either consonantal or vocalic features. In line with what was found in the study of segments, this argues for independence between the separate domains of vowels, consonants, and diphthongs. 4

Conclusions and Perspectives

We proposed a generic approach to estimate the complexity of phonological inventories relying on the information provided by a dataset of 451 actual inventories. This method derived from a simple yet central definition of complex systems stating that they are usually composed of elements whose global behavior cannot be linearly deduced from their individual properties. We considered individual frequencies of segments and features to be such individual properties and developed a statistical model to test the significance of pairwise interactions, trying as much as possible not to extrapolate from the information provided by the dataset. Analyses at both the segment and feature levels revealed that under a mildly conservative approach (the control for FDR), only a very small number of significant pairwise associations could be deduced from the available data, and that it was therefore difficult to conclude in favor of strongly non-linear interactions between either features or segments. On account of this, our tentative conclusion is that the degree of complexity of phonological inventories is rather low. However, one should consider this position with caution, given the likelihood of a significant number of false negatives, which were not prevented by the statistical test used. Detailed analyses of the most significant associations supported classical hypotheses about the structuration of inventories like feature economy (Clements, 2003b) or maximal use of available features (Ohala, 1980). It also suggested a ranking of some organizational constraints and specific interactions, including the previous principle and interactions between nasalized vowels or between long vowels. Some additional phenomena were revealed, like the very limited number of interactions between consonants and vowels, whether at the segment or feature level. Other paths could be followed to better consider the knowledge that can be extracted from available datasets like UPSID. Rather than focusing on pairwise interactions, more powerful methods to learn the regularities of the inventories could be used. Artificial neural networks (ANN) would be one of these methods.

160

Christophe Coupé, Egidio Marsico, and François Pellegrino

It would also be interesting to apply our method to collections of inventories from other unrelated domains, and compare the percentages of significant associations with those obtained in this chapter. Could we find situations with much higher percentages, or would we always observe weak values, even for systems that can without doubt be considered as complex? Such comparative findings would provide a stronger basis for our conclusion regarding the limited complexity of phonological inventories. Evolutionary models have also been proposed on the basis of the synchronic information provided by UPSID (Coupé et al., 2009). Although these models are built on statistical tests like those presented in this chapter, they did not include corrections of multiple testing. Fixing this shortcoming would represent a possible next step. Our main findings are limited in relation to established theories. However, two comments can be made: first of all, the statistical tests failed to reveal that phonological inventories were strongly non-linear in their organization. This inflects the issues at play, prompting investigators to prove why they think that phonological inventories are complex systems rather than simple ones. It can be conceded though that looking only at pairwise interactions is a reductionist approach, and that more sophisticated structures could be considered. Clements’ approach was to look for arguments in favor of feature economy. This led him to consider specific classes of segments to apply his statistical test. Such an approach can be termed “theory-driven,” when our own initially follows a theory-less, bottom-up approach. This does not mean that theoretical phonological questions may not be addressed and answered, but that they do not play a role in the design of the method. Both approaches are of course interesting in their own sake, and are partially derived from the Kantian distinction between nomothetic and idiographic tendencies. What is gained in genericity is lost in terms of specific validations, and vice versa. Secondly, we want to stress that the main interest of our approach lies in its capacity to highlight the gap between the concepts and tools commonly used in complexity theory, especially when applied to physical or biological systems, and in the theories proposed by scholars in linguistics. It seems that, until further improvement, these concepts and tools still often lack the power and resolution to address specific questions of the field, and may therefore be considered as irrelevant by some. Furthermore, defining the “complexity of language” has turned out to be a difficult enterprise, especially when one assesses the multiple conceptions and measures of complexity. Definitions change depending on whether the investigator deals with the acquisition of specific structures, the amount of covert versus overt information in communication, the respective complexity of syntax versus morphology versus phonology, and so on. This difficulty may once again be explained in part by the divide between

To What Extent Are Phonological Inventories Complex Systems?

161

nomothetic and idiographic approaches. The former apply better to natural sciences, where most of the tools of complexity theory have been developed, and the latter to the humanities. Echoes of this argument can be heard in the heated debates that have followed recent publications that relied on nomothetic approaches to investigate the distribution of language structures in large samples of languages, for example, Atkinson’s approach to the worldwide distribution of the size of phonological inventories (Atkinson, 2011a, 2011b; Bybee, 2011), or Lupyan & Dale’s (2010) study of the links between morphological complexity and social structures. REFERENCES Ashby, W. Ross. 1947. Principles of the Self-Organizing Dynamic System. Journal of General Psychology 37: 125–128. Atkinson, Quentin D. 2011a. Phonemic Diversity Supports Serial Founder Effect Model of Language Expansion from Africa. Science 332: 346–349. Atkinson, Quentin D. 2011b. Linking Spatial Patterns of Language Variation to Ancient Demography and Population Migrations. Linguistic Typology 15(2): 321–332. Au, Ching-Pong. 2008. Acquisition and Evolution of Phonological Systems. Taipei: Institute of Linguistics, Academia Sinica. Benjamini, Yoav, and Yosef Hochberg. 1995. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society Series B (Methodological) 57(1): 289–300. Bybee, Joan. 2011. How Plausible Is the Hypothesis that Population Size and Dispersal are Related to Phoneme Inventory Size? Introducing and Commenting on a Debate. Linguistic Typology 15(2): 147–153. Cancho, Ramon Ferrer I., and Ricard V. Solé. 2001. The Small-World of Human Language. Santa Fe Institute Working Paper 268: 2261–2266. Chomsky, Noam, and Morris Halle. 1968. The Sound Pattern of English. New York: Harper & Row. Clements, G. Nick. 2003a. Feature Economy in Sound Systems. Phonology 20: 287– 333. Clements, G. Nick. 2003b. Feature Economy as a Phonological Universal. Proceedings of the 15th International Congress of Phonetic Sciences, Barcelona, Spain, 371– 374. Coupé, Christophe, Egidio Marsico, and François Pellegrino. 2009. Structural Complexity of Phonological Systems. Approaches to Phonological Complexity, Phonology & Phonetics Series vol. 16, ed. by François Pellegrino et al., 141–169. Berlin, New York: Mouton de Gruyter. Coupé, Christophe, Egidio Marsico, and Gérard Philippson. 2011. How Economical are Phonological Inventories? Proceedings of the 17th International Congress of Phonetic Sciences, Hong Kong, China. Dahl, Osten. 2004. The Growth and Maintenance of Linguistic Complexity. Amsterdam: Benjamins.

162

Christophe Coupé, Egidio Marsico, and François Pellegrino

de Boer, Bart. 2000. Self-Organization in Vowel Systems. Journal of Phonetics 28(4): 441–465. de Boer, Bart, and Willem H. Zuidema. 2005. Multi-Agent Simulations of the Evolution of Combinatorial Phonology. BNAIC 2005 – Proceedings of the 17th Belgium-Netherlands Conference on Artificial Intelligence, Brussels, Belgium, 66–73. De Saussure, Ferdinand. 1916. Course in General Linguistics. Edited by Charles Bally and Albert Sechehaye in collaboration with Albert Riedlinger, translated by Wade Baskin. New York: McGraw-Hill Book Company, 1966. Dorogovtsev, Sergey N., and José F. F. Mendes. 2001. Language as an Evolving Word Web. Proceedings of The Royal Society of London Series B, Biological Sciences 268: 2603–2606. Dryer, Matthew S., and Martin Haspelmath (eds.). 2011. The World Atlas of Language Structures Online. Munich: Max Planck Digital Library. Available at http://wals. info/. Dunn, Olive J. 1961. Multiple Comparisons among Means. Journal of the American Statistical Association 56: 52–64. Edmonds, Bruce. 1999. What is Complexity? The Philosophy of Complexity per se with Application to Some Examples in Evolution. The Evolution of Complexity, Dordrecht: Kluwer. Evans, Nicholas, and Stephen C. Levinson. 2009. The Myth of Language Universals: Language Diversity and its Importance for Cognitive Science. Behavioral and Brain Sciences 32: 429–48. Fenk-Oczlon, Gertraud, and August Fenk. 2005. Cross-linguistic Correlations Between Size of Syllables, Number of Cases, and Adposition Order. Sprache und Natürlichkeit: Gedenkband für Willi Mayerthaler, ed. by Gertraud Fenk-Oczlon and Christian Winkler, 75–86. Tübingen: Narr. Gell-Mann, Murray, and Seth Lloyd. 2003. Effective Complexity. Nonextensive Entropy – Interdisciplinary Applications, ed. by Murray Gell-Mann, & Constantino Tsallis, 387–398. Oxford: Oxford University Press. Goldsmith, John. 1990. Autosegmental and Metrical Phonology. Oxford: Blackwell. Hochberg, Yosef. 1988. A Sharper Bonferroni Procedure for Multiple Tests of Significance. Biometrika 75(4): 800–802. Hochberg, Yosef, and Ajit C. Tamhane. 1987. Multiple Comparison Procedures. New York: Wiley. Holm, Sture. 1979. A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics 6(2): 65–70. Jensen, Pablo. 2006. Network-Based Predictions of Retail Store Commercial Categories and Optimal Locations. Phys. Rev. E 74: 035101. Ke, Jinyun, Tao Gong, and William Shi-Yuan Wang. 2008. Language Change and Social Networks. Communications in Computational Physics 3(4): 935–949. Liljencrants, Johan, and Bjorn Lindblom. 1972. Numerical Simulation of Vowel Quality Systems: The Role of Perceptual Contrast. Language 48: 839–862. Lindblom, Bjorn, Peter MacNeilage, and Michael Studdert-Kennedy. 1983. SelfOrganizing Processes and the Explanation of Phonological Universals. Explanations for Language Universals, ed. by Brian Butterworth, Bernard, Comrie, and Östen Dahl, 181–203. Berlin: Mouton.

To What Extent Are Phonological Inventories Complex Systems?

163

Lindblom, Bjorn, and Ian Maddieson. 1988. Phonetic Universals in Consonant Systems. Language, Speech and Mind, ed. by Charles Li and Larry Hyman, 62–78. London: Routledge. Lupyan, Gary, and Rick Dale. 2010. Language Structure is Partly Determined by Social Structure. PLoS ONE 5(1): 1–10. Maddieson, Ian. 1984. Patterns of Sounds. Cambridge: Cambridge University Press. Maddieson, Ian. 2006. Correlating Phonological Complexity: Data and Validation. Linguistic Typology 10(1): 106–23. Maddieson, Ian. 2009. Calculating Phonological Complexity. Approaches to Phonological Complexity, Phonology and Phonetics Series, vol. 16, ed. by François Pellegrino et al., 85–110. Berlin, New York: Mouton de Gruyter. Maddieson, Ian. 2011. Phonological Complexity and Linguistic Patterning. Proceedings of the 17th International Congress of Phonetic Sciences, Hong Kong, China. Maddieson, Ian, and Kristin Precoda. 1990. Updating UPSID. UCLA Working Papers in Phonetics 74: 104–111. Marsico, Egidio, Ian Maddieson, Christophe Coupé, and François Pellegrino. 2004. Investigating the ‘Hidden’ Structure of Phonological Systems. Berkeley Linguistics Society 30: 256–67. Miestamo, Matti, Kaius Sinnemäki, and Fred Karlsson (eds.). 2008. Language Complexity: Typology, Contact, Change. Amsterdam: Benjamins. Newman, Mark. 2010. Networks: An Introduction. Oxford: Oxford University Press. Ohala, John J. 1980. Chairman’s Introduction to Symposium on Phonetic Universals in Phonological Systems and Their Explanation. Proceedings of the 9th International Congress of Phonetic Sciences, 1979, 184–185. Oudeyer, Pierre-Yves. 2006. Self-Organization in the Evolution of Speech. Oxford: Oxford University Press. Pellegrino, François, Egidio Marsico, Ioana Chitoran, and Christophe Coupé (eds.). 2009. Approaches to Phonological Complexity. Phonology & Phonetics Series vol. 16. Berlin, New York: Mouton de Gruyter. Plank, Franz. 1998. The Co-Variation of Phonology with Morphology and Syntax: A Hopeful History. Linguistic Typology 2(2): 195–230. Prince, Alan, and Paul Smolensky. 2004. Optimality Theory: Constraint Interaction in Generative Grammar. Oxford: Blackwell. Sampson, Geoffrey. 2009. Language Complexity as an Evolving Variable. Oxford: Oxford University Press. Saville, Dave J. 1990. Multiple Comparison Procedures: The Practical Solution. American Statistician 44: 174–180. Schwartz, Jean-Luc, Louis-Jean Boë, Nathalie Vallée, and Christian Abry. 1997. The Dispersion-Focalization Theory of Vowel Systems. Journal of Phonetics 25(3): 255–286. Shosted, Ryan K. 2006. Correlating Complexity: A Typological Approach. Linguistic Typology 10(1): 1–40. Šidák, Zbynˇek. 1967. Rectangular Confidence Regions for the Means of Multivariate Normal Distributions. Journal of the American Statistical Association 62(31): 626– 633. Steels, Luc. 1996. Self-Organising Vocabularies. Cambridge: MIT Press.

164

Christophe Coupé, Egidio Marsico, and François Pellegrino

Steels, Luc. 1997. The Synthetic Modeling of Language Origins. Evolution of Communication 1(1): 1–34. Vallée, Nathalie. 1994. Systèmes Vocaliques: De la Typologie aux Prédictions. Thèse de doctorat. Grenoble: Université Stendhal. Wiener, Norbert. 1961. Cybernetics: or Control and Communication in the Animal and the Machine, 2nd edition. Cambridge: MIT Press.

7

A Complexity View of Ontogeny as a Window on Phylogeny Barbara L. Davis

1

Introduction

One of the goals of this book is to consider the strength of a complexity perspective on the long time scale origins of language in human speakers and listeners. In this chapter, a complexity view of the ontogeny of speech acquisition in human infants will be addressed as it potentially illuminates the phylogenetic process of language evolution. In the broader context of language origins, the ethnographer Nikko Tinbergen (1963) has suggested four questions that should be addressed to characterize the evolution of the human language capacity. These questions include: (1) What are the stimuli that produce the response in the organism? (2) How does the behavior contribute to the organism’s survival and reproductive success? (3) How does the behavior develop in the organism’s lifetime? and last, (4) How did the behavior arise in the organism’s species? (Tinbergen, 1963). Cogent to the topic addressed here, the last question concerns ontogeny in organisms, in our case, human infants who are learning language. In Tinbergen’s perspective, the ethnographer needs to ask how the young organism’s behavior changes with age, what early experiences are necessary for the behavior to be manifest, and which developmental steps and environmental factors play a role in ontogenesis. In the spirit of Tinbergen’s ontogenetic question, the phonological component of language will be evaluated as it relates to examining the value of ontogeny for considerations of phylogenetic origins. Phonology uniquely concerns two dimensions of capacity that must be acquired in human infants relative to mastery of language structure and function. First, it is critical to consider the child’s acquisition of an ambient language phonological knowledge base. In linguistics, this dimension has often fallen within classical (Chomsky & Halle, 1968) and contemporary (Prince & Smolensky, 2004) views of phonological acquisition. A second dimension of phonological acquisition that has been included in understanding phonological acquisition includes the perception and action components of the infant’s suite of biological capacities available for acquiring phonology. Inclusion of this “embodied” aspect into

165

166

Barbara L. Davis

understanding of the nature of phonological acquisition implies that both the phonological knowledge base and the use of bodily capacities for implementation of this knowledge base for communicative function within the environment are necessary for specifying the nature of the ontogenetic acquisition process in human infants. In this view, both the neural-cognitive knowledge base for phonology and the perception-action capacities supporting use of this knowledge base are necessary to specify the acquisition process. Neither alone is sufficient. This multifaceted view of phonological acquisition is congruent with complexity science, where multiple and diverse components are seen as cocreating the eventual behavioral product. In an extended consideration of complexity science as a theoretical base for specifying acquisition of phonological capacities in contemporary humans, this perspective has recently been termed “Knowing and Doing” (Davis & Bedore, 2013). Modern children’s acquisition of phonology potentially enables a short time scale view across ontogeny of the phylogeny of human language capacities within Tinbergen’s conceptualization. It also supports a consideration of the value of complexity science in understanding ontogenetic processes and how they may illustrate important aspects of phylogenetic origins of language. The fundamental principles of complexity science are compatible with responses to Tinbergen’s ethnographic questions about ontogeny. Human infants’ intrinsic biological capacities and general-purpose enabling mechanisms mature on an extended developmental timetable when compared with other species. This extended timetable for development of adult complexity has been characterized as neoteny. Because the process of development is extended in human infants relative to other modern species, the human ontogenetic process affords an opportunity for examining the emergence of a complex system when it is in its simplest early phases. 2

Historical Background

One fundamental aspect of the acquisition of language in humans is important to illuminating the dialogue on the relevance of complexity constructs to understanding the acquisition of knowledge as it might relate to phylogenic considerations. The nature versus nurture debate has been a core aspect of the continuing controversy regarding the origins of complex capacities in humans for centuries (viz. Descartes, 1637; Kant, 1924). It has formed a fundamental area of difference in the epistemological study of the origins of children’s acquisition of knowledge across multiple areas of description, not limited to phonology or even to language (see Woolhouse, 1988). Motor, social, perceptual and cognitive domains of acquisition have been sites of study for this issue. A brief overview of nature versus nurture as a continuing question in understanding the origins of complex capacities provides a conceptual background for

A Complexity View of Ontogeny as a Window on Phylogeny

167

considering a complexity perspective on phonological acquisition in the present context. The earliest competing hypotheses about the origin of knowledge in humans, particularly human infants, are to be found in antiquity. They formed the basis for later focus on the origins of language knowledge. Plato was the first philosopher to propose innateness as a solution to the question of the origins of knowledge. He proposed mental “Essences” as necessary building blocks for the later knowledge a child acquires. Platonic philosophy asserted that humans are the only living creatures who possess innate ideas. In contrast, Locke, a seventeenth-century empiricist, argued that the child is a tabula rasa or blank slate on which knowledge is written. Diverging from Plato’s philosophical perspective, Locke emphasized the role of experience in the acquisition of complex knowledge. Descartes (1637), following Platonic philosophy, asserted that humans are the only living creatures who possess innate ideas. However, Descartes saw the source of innate human ideas as divine, albeit as an endowment from the God of Christian divinity rather than the Platonic Greek gods. These early competing hypotheses about the origins of knowledge formed the basis for the later focus on language knowledge. In later centuries, de Saussure (1916) focused the debate about the origins of knowledge in young children on language among the other dimensions of knowledge that the child was acquiring at the same time. He conceived of language as a cultural phenomenon that was the result of input from cultural inventions unrelated to biology. The origin of these complex behaviors resided outside of the individual, compatible with the empiricist position espoused by Locke much earlier. De Saussure’s view of language diversity was in contrast to Platonic philosophy, which placed the language capacity within the child. Roman Jakobson (1941) focused on children’s phonological acquisition as a manifestation of mental representations of perceptual contrasts in human speakers. He proposed a deterministic order of acquisition by children that was universal across languages and individuals. Following de Saussure, children were supposed to learn universal feature or contrast hierarchies unrelated to biological constraints. Jakobson’s proposal was not based on actual data from children. Chomsky and Halle’s (1968) early work on linear generative phonology expressed in The Sound Pattern of English (SPE) formed the first driving force for proposals of knowledge-based phonological essences in the philosophical tradition of Plato and Descartes. Chomsky and Halle proposed knowledgebased essences (termed “distinctive features”) that were present at birth to guide acquisition and revealed by maturation. More modern constructions of phonological theory related to the origins of complex knowledge in human children can be found more recently in Optimality Theory (OT; Prince & Smolensky, 2004). The Handbook of Phonology (edited by DeLacy, 2007) provides a

168

Barbara L. Davis

comprehensive overview of these perspectives. In these constructions of the acquisition of phonology, the child learner is seen as moving toward adult language competence by re-ranking a set of mentally available constraints. Observed “surface” forms in child output and in languages arise from the resolution of conflicts between these ranked grammatical constraints. In some OT perspectives, constraints are considered innate and universal (e.g., Dekkers, 2000). Other OT approaches do not assume innate or universal status (e.g., Boersma, 1998). But the origin of knowledge is seen in a priori Platonic essences that are revealed by maturation and parameterized by perceptual input from the environment about ambient phonology. While Chomsky and Halle’s proposals formed the first basis for the modern phonological paradigm in which language is considered modular and “special” in humans, their basis for understanding acquisition of phonology has much in common with contemporary phonological theory (e.g., OT). In both conceptualizations, phonology is seen as a symbol manipulation system based on computational properties shared by speakers and listeners. The speaker and listener are viewed as computational devices utilizing a modular system to communicate with formal linguistic symbols (see Scott-Phillips, 2010, for a review of this perspective). Critically, language is viewed as a computational system that operates independently of function. Evidence for computational operations is found in frequency counts of language forms. Description of these forms and their frequency is equated with explanation of their origins within the underlying grammar. Assertions of the “psychological reality” of speech elements rather than a neural basis for system properties are the norm in this perspective on the origins of complex language knowledge in human children. Actions of the body in perceiving and producing speech are not intrinsic to the computational system. Function of the system in achieving the young child’s social needs for survival and growth in knowledge is also outside the scope of the computational system envisioned by classic and contemporary phonological theories. 3

Alternative Theoretical Proposals

An alternative theoretical model for understanding the origins of phonological knowledge in human infants can be found in complexity theory (Davis & Bedore, 2013; Pellegrino et al., 2009). Complexity science refers to a view of complex phenomena where local components of a system interact and complex behavior emerges from those interactions (Holland, 1998). Local interactions create interconnectivity. Complexity theory has been employed to characterize complex systems as diverse as weather systems (Prigogine & Stengers, 1984), developmental biology (Kauffman, 1995), and comparative studies of the biological structures and functions of diverse organisms (Camazine et al., 2001), phenomena that had long resisted linear explanations.

A Complexity View of Ontogeny as a Window on Phylogeny

169

It is useful to take into account some of the claims of complexity science as they are illustrated by emergence of complexity in children’s acquisition of the phonological component of language. Within the tenets of complexity science, phonological knowledge and behavioral patterns can be seen as emerging from connections enabled by general-purpose child capacities such as learning and cognition as opposed to language-dedicated modular mechanisms. Heterogeneous inputs to the system from child intrinsic production, perception, neural cognitive and social interaction capacities are tuned by feedback from parent input available to the child. These interactions among system components are context-dependent in application during social routines across acquisition. Components operate in a hierarchy (Wilson & Holldöbler, 1988) in which “higher” (e.g., neural-cognitively instantiated knowledge) and “lower” (e.g., perceptual and production system processing) levels of the overall complex system are interactive systems that develop interconnectivity in support of growth in complexity of phonological knowledge and output patterns. Language – in particular, the phonological component of language that is the focus here – can be seen as a communal communication system. This system includes both the micro (i.e., caregiver interactions) and macro (i.e., the larger cultural community of the child) dimensions of the social world of the child as well as the structure(s) that specify spoken language forms used in communicative interactions. As an example, local social interactions between mothers and children create mutually understood communications. These communications are coded with recognizable sounds as the child utilizes her perception-cognition-production system abilities in salient social routines with her mother. “Peek-a-boo” routines in the first year of life require participation in formulaic exchanges using verbal sequences that have meaning to the participants. Across development, growth in language capacities must be accompanied by growth in the phonological system to convey those ideas to others (i.e., the /r/ and /s/ sounds in “rhinoceros” for a trip to the zoo). Perceptual capacities are functional in enabling the child to differentiate between salient and non-essential incoming stimuli. Cognitive-neural capacities support the young child in gaining and storing increasingly precise phonological knowledge structures. Production system capacities enable the child to perform more diverse goal-directed speech movements related to her ambient language requirements. All of these strands cooperate in enabling the system to gain complexity. Again, none are considered sufficient alone to the task of acquisition of phonology; implying a complex system approach to understanding the process. A complementary theoretical perspective to complexity theory can be found in embodiment theory. Prominent representatives of the embodiment perspective are found in philosophy (M. H. Johnson, 1987), linguistics (Lakoff & Johnson, 1999), cognitive science (Varela, Thompson, & Rosch, 1991),

170

Barbara L. Davis

neuroscience (Damasio, 1994; Edelman, 1992), artificial intelligence (Clark, 1997), and developmental cognition (Smith & Breazeal, 2007). In embodiment perspectives, mental activity and underlying brain activity cannot be understood outside the context of bodily activities. According to Clark (1997), the goal is to learn to manipulate a mental model in the same way as we originally manipulated the real world … to internalize cognitive competencies rooted in manipulations in the external world. (See Iverson, 2010, for an extended treatment of the importance of movement in the development of cognition.) Applied to phonological acquisition, embodiment approaches support the assertion that the human speech production and perception systems may form a primary limiting factor on acquisition of speech output as well as, potentially, the structure and processing of phonological knowledge (Davis & MacNeilage, 2000). As a biological system, the speech production-perception-neural apparatus actually operates to mold or sculpt the set of complex phonological patterns used for linguistic communication by humans. Sound patterns that utilize the human production and perception systems in achieving maximal perceptual distinctiveness for minimal production difficulty are valuable assets for efficient and rapid human communication. Those attributes spring from the tandem operation of the peripheral and physically embodied perception and production systems in speaking and listening. They are stored in neural tissue based on repeated use in speaker-listener interactions (Lindblom, 2008). Critically, acquisition of the complex system of phonological knowledge and behavior is embodied in the sense that the child’s body operates in the world as a critical factor in the complex system underlying acquisition of her phonological system. A second complementary theoretical perspective that supports the proposals of complexity science in understanding phonological acquisition is found in dynamic systems theory (Thelen & Smith, 1994). A dynamic system is a system that is stable, yet far from thermodynamic equilibrium. Equilibrium conditions can only be maintained by a continuous flow of free energy into and out of the system according to dynamic system perspectives. Biological systems have been evaluated as prime examples of open systems within this paradigm. Over time their order and complexity are not only maintained but may actually increase, as in development. Researchers have employed dynamic systems to model early acquisition of complex action in the locomotor domain (Thelen, 1995; Thelen & Smith, 1994), in neuro-physiological structures, and in vocal tract biomechanics in neonates (Steven M. Barlow, 1998). Thelen and Smith (1994) tested dynamic systems in the domain of infant acquisition of crawling and walking. Based on this body of work, Thelen and colleagues propose that the infant’s own system components, including production, perception, and cognitive capacities, are mutually interactive across development. None can individually specify the nature of the complex system.

A Complexity View of Ontogeny as a Window on Phylogeny

171

Dynamic systems emphasize the changing nature of a complex system across time. Diverse inputs from the environment interacting with child output create this change. Dynamic and qualitative changes occur in the complex system across the process of acquisition. This construction of the process of acquiring complexity in phonological form is in contrast with linguistic perspectives (Prince & Smolensky, 2004). These perspectives propose a progressive revelation of child internal preexisting, static phonological knowledge structures via maturation guided by external perceptual triggering from input. Multiple causes of observable changes lie in the heterogeneous sources of input into the complex system. Oyama (2000) has suggested the term constructivist interaction, implying that interpenetration between organism and environment is integral to the acquisition of a complex behavioral and knowledge repertoire. Thelen (1991) asserted that perceptual-motor activities resulting in a complex vocal behavior require coordination of multiple subsystem elements. This coordination produces spatially and temporally organized behavior with a cohesive structure, the essential defining quality of human speech production. Operation of the elements of the production system, including respiratory, phonatory, and articulatory capacities, and the observable perceptual biases toward sensitivity to salient acoustic and visual features embody two critical dimensions of this system. The system is open. The status of the infant’s production system, as well as sensory processes, externally available adult speech and language models, and the socio-cultural milieu requiring function from the child are crucial inputs driving the nature of the overall system. The system is dynamically open to inputs from each of these subcomponents. The confluence of principles available from these complementary theoretical perspectives enables a proposal that can be termed a biological-functional approach to phonological acquisition. In this view, the heterogeneously determined, embodied, and dynamically changing complex system found in the young human child acquiring phonology supports the proposal that ontogeny serves as a window on phylogenetic explanation. These theoretical perspectives ask that we encompass consideration of both biological and social dimensions of function in understanding phonological acquisition. Acquisition of increasingly mature behavioral patterns and underlying knowledge about phonology are viewed as being accomplished by interactions of biological and social components of a complex system across the process to achieve maximal function for the young child. In acquiring knowledge patterns, the child is internalizing neural/cognitive competencies rooted in social manipulations in the external world. To master speech-related behavioral patterns, the child is assembling functionally adaptive behavioral patterns that exploit intrinsic dynamics of the production and perception systems to respond to local social contexts. Both growing complexity in knowledge and growing

172

Barbara L. Davis

complexity in the behavioral repertoire enable the young child to function in the world around him or her to acquire relevant information for long-term survival in the world. Thus, in a biological-functional approach, outcomes of phonological acquisition result from multiple interactions among heterogeneous aspects of a complex system. These explanatory principles from diverse theoretical perspectives provide a potential window on explanation for phylogeny of human language. They add the components of function and change over time in an open system that is heterogeneous and enabled by feedback as tools for explanation of patterns observed. In terms of Tinbergen’s question about ontogeny, these principles enable access to “How did the behavior arise in the organism’s species?” 4

Considerations from Data on Acquisition

Data for evaluating a behavioral–functional-based complexity perspective can be found in the earliest phases of speech-like behavioral output in human infants; the pre-linguistic babbling period. Perceptually rhythmic speech-like syllables are apparent to listeners between seven and nine months in infants who are developing typically. These robustly apparent rhythmically speechlike vocal outputs give evidence of aspects of the infant movement system that are available for initial approximation of the serial organization of adult speech (Davis & MacNeilage, 1995). They do not appear reliably in infants with diminished perceptual capacities, emphasizing the importance of auditory perceptual capacities in onset of babbling (Davis et al., 2005; Von Hapsburg & Davis, 2006). They are also seen as reflecting neural integrity for supporting rhythmic behaviors across domains, including other gross and fine-motor aspects of infant behavior in this period (Thelen & Smith, 1994), And they appear in reliably more mature form in social conditions where there is maternal contingency (Gros-Lewis et al., 2006) supporting the importance of a social component for developmental processes. These diverse foundations support a multifaceted basis for the emergence of complexity in infant speech-like output related to the mature patterns of the language community. This early foundation includes embodied biological perception, production, and neural capacities as well as social contextual components. The sounds and sequential structures in this earliest period of speech-like behavior appear quite restricted. A relevant question of interest in considering a complexity perspective on acquisition concerns the relationships between early patterns observable in infants during canonical babbling and overall phonological patterns that must be mastered for intelligible speaking and listening in their ambient language environment. The process of acquisition from the earliest beginnings to the eventual outcome of a mature complex phonology highlights the resonance of questions of ontogeny with phylogenetic questions.

A Complexity View of Ontogeny as a Window on Phylogeny

173

Both the process of acquisition and the structural sound system products of that process are of interest in understanding the emergence of complexity in this context. A cross-linguistic approach that includes analysis of diverse languages is of value to understanding common patterns as well as underscoring the diversity of language complexity that must be learned by infants to achieve intelligible message transmission in their own unique language. Cross-language analyses of output patterns have a long history in linguistic study (e.g., Maddieson, 1984). Cross-linguistic profiles of child acquisition patterns are a newer aspect of study relative to understanding the nature of child-adult similarities and differences (Kern & Davis, 2009). Comparison of patterns in infant output and in languages provides a cornerstone for understanding the ambient language targets to be achieved by young children in diverse language environments. It shows, as well, the numerous ways in which languages have exploited the speech production mechanism for achieving message distinctiveness. As noted earlier, aspects of the complex system include the physical implementation mechanisms of the production, perceptual, and neural sub-systems. These sub-systems are common across child and adult speakers where phonological implementation is overlaid on the primary and basic life functions of these bodily systems. But the young child presents a dynamic and changing system at the levels of neural-cognitively instantiated knowledge and implementation of that knowledge through speech-related actions. In contrast, adult speakers and listeners could be viewed as embodying a relatively static complexity where stable properties of a complex system are used in rapid message transmission. However, the congruence of adults and infants enables the child to begin the process of mastery of phonology within social interchanges as their growing system functions to enable communication of meanings using vocal means. As well, the behavioral phenotypes of consonants, vowels, serial properties of syllable types and within and across syllable patterns must be accounted for as another dimension of complexity reflected in outcomes of operation of bodily systems. In a complexity perspective these observable structures that serve functional purposes emerge from the operation of the body in adults and in children as they interact in social contexts. 4.1

Stability and Change

The behavioral sound patterns that are found to be in common in infant output and in languages within and across diverse ambient language environments suggest the concept of stability. Stable properties in phonological components can be considered as those that are available early to infants in the first year of life who are producing canonical babbling and are also retained by adult speakers using phonological inventories who are fully mature speakers. Within a

174

Barbara L. Davis

dynamic system approach, these stable output properties can be seen as “stable states” (Thelen & Smith, 1994). In contrast, properties that are modified over time in acquisition reflect needed change in the system. Change is necessary in order to maximize the sound properties available to produce rapid message transmission with the phonological elements available. If some aspects of infant systems in common with adults connote stability, others change with maturity and social input to enable increases in message complexity. Again, both stability and change are necessary components of the system to enable children and adults to use the system for communication with one another as the child is simultaneously acquiring new aspects of the system for communicating more and diverse ideas about the world. 4.2

Stability: Data

First, let us consider data that would connote stability in common across infants and adult speakers. In this regard, it has been observed that some robust patterns in babbling preference are retained in adult users of contemporary languages (MacNeilage et al., 2000a; 2000b). According to Bell and Hooper (1978), there is a tendency for words in the world’s languages as well as in early infant productions to begin with a consonant and end with a vowel. The open syllable (CV) is both the only universal syllable type in the world’s languages (Maddieson, 1984) and the favored syllable type of babbling across a number of languages studied to date (e.g., Davis & MacNeilage, 1995; Kern & Davis, 2009). Some sound types occur more frequently both in babbling and in languages. Simple stop consonants and nasals favored in babbling are highly frequent in languages of the world (Maddieson, 1984). These sound types tend to dominate the repertoire of languages with small systems (< 15 phonemes) containing a few segments characterized as articulatorily “simple” (Lindblom & Maddieson, 1988; Lindblom, Krull, & Stark, 1993). Maddieson (1984) showed that all 317 observed languages he analyzed exhibited a version of [p] and [t], at least one nasal consonant, and a high frequency of [w] and [j]. Further, 88 percent of language phonologies included the vowel [a]. Certain within-syllable CV co-occurrence patterns favored in babbling within and across languages are also retained in mature language users. Studies of babbling patterns in widely varied languages indicate many infants prefer within-syllable patterns that show lack of movement between the consonant and the vowel portion of the syllable (i.e., labials and central vowels, coronals and front vowels and dorsals and back vowels). Studies include children speaking English (Davis & MacNeilage, 1995; Oller & Steffans, 1993; Giulivi et al., 2011), Korean (Lee, Davis, & MacNeilage, 2008), French (Vihman, 1992), Mandarin Chinese (Chen & Kent, 2005), and Equadorean-Quichua

A Complexity View of Ontogeny as a Window on Phylogeny

175

(Gildersleeve-Neumann, Davis, & MacNeilage, 2013) as well as in EnglishSerbian bilinguals (Zlatic et al., 1997). In these languages, children most frequently (although not universally) use their vocal mechanisms for producing rhythmic syllables without articulatory movement between consonant and vowel within syllables. This relationship has been termed the Frame-Content Theory (FC; MacNeilage & Davis, 1990). The FC theory proposes that the aspect of the infant movement system available for replicating speech like output at the onset of canonical babbling is the jaw. Rhythmic open-close jaw oscillations without independent movement of the tongue within syllables result in a listener percept of consonant and vowel properties dominated by least movement. Relative to within-syllable patterns retained in languages that are often present in infant syllable-level inventories, Janson (1986) studied consonantvowel relationships derived from written texts of five languages (Finnish, Turkish, Latin, Latvian, and Setswana). Maddieson and Precoda (1991) studied consonant-vowel relationships derived from dictionary counts in five additional languages (viz., Hawaiian, Rotokas, Piraha, Kadazan, and Shipibo). In both studies, there was a significant tendency for coronals to favor front vowels and disfavor back vowels and for dorsals to disfavor back vowels. There was no tendency shown in these languages for labial central vowel co-occurrences. MacNeilage and Davis (1993) also analyzed intrasyllabic trends involving stop consonants, nasals, and vowels in ten languages other than those studied previously. Their analysis of English, Estonian, French, German, Hebrew, Japanese, New Zealand Maori, Quichua, Spanish, and Swahili showed a significant tendency for dentals/coronals to favor front vowels. This front closure portion was also found to disfavor back vowels. A significant tendency for dorsals to disfavor front vowels was also observed in the data. Additionally, there was a nonsignificant trend for dorsals to favor back vowels. In seven languages, labials were predominantly associated with central vowels, coronals to front vowels. In eight languages, dorsals were associated with back vowels. Thus, there appear to be commonalities between infants and adult speakers in both basic sound and syllable properties that appear with regularity in output inventories. Questions that should be posed in this regard include the following: 1) what is the relationship of child patterns in pre-linguistic babbling to ambient language patterns across languages? and 2) Do child patterns vary crosslinguistically in the babbling period? Study of babbling, as previously noted, enables consideration of the complex system embodied in perceptually speechlike behaviors that occur and the early period of onset of these behaviors. Statistical evaluation of relationships between infants and languages has been pursued using an infant database available through PhonBank (MacWhinney & Rose, 2005). Participants in the presently reported data include twentysix children in Turkish, Dutch, Tunisian, French, Romanian, and English

176

Barbara L. Davis

Table 7.1 Comparison of Consonant Vowel Association Levels Within Syllables Co-occurrence

Finding

Languages

Other

Languages

Coronal/front Labial/front Dorsal/front Coronal/back Labial/back Dorsal/back Coronal/central Labial/central Dorsal/central

=/+ = = − = − + + /= − +

5 5 5 5 4 5 6 6 3 Dutch, Tunisian, English 3 Romanian, French, Turkish

− − + = − =

Romanian 1 Romanian 1 Romanian 1 A-Eng. 2 – 1 French, 1 A-Eng 1 French

+ Sig. greater than AL = No sig. diff. from AL; – Sig. less than AL

language environments. Infants were typically developing and between 7 and 12 months of age. These data are part of a larger study (Kern & Davis, 2009). For the larger study, data collection encompassed the canonical babbling period. Canonical babbling is defined as rhythmic alternations between consonant and vowel-like properties, giving a percept of rhythmic speech that simulates adult output without conveying meaning (Davis & MacNeilage, 1995; Oller, 1980). Spontaneous comfort state vocalizations that were perceptually rhythmic and contained a consonant-like closure phase + vowel-like open phase within a single utterance string (viz., CV, CVCV, CVCVCVCV) were audio- and video-recorded bimonthly for one hour from infants in each language environment. Language data for comparison included 1,000 words randomly selected from dictionaries on computers in each of the ambient languages. Data analyzed that will be reported here from the larger study by Kern and Davis were sequential consonant-vowel patterns within syllables (viz., consonantvowel co-occurrences of consonant labial, coronal, and dorsal place of articulation). Also reported are across-syllables patterns for reduplication versus variegation and changes in (1) place versus manner for consonants and (2) height versus backness for vowels. The standardized residual [observedexpected/SQRT(expected)] was calculated for statistical comparison of infant and language properties. Results for intrasyllabic consonant-vowel co-occurrences are shown in Table 7.1. Based on FC predictions of lack of independent movement of articulators separate from the jaw cycle, associations are predicted to be found between labial closures and central vowels, coronal closures and front vowels, and dorsal closures with back vowels. As noted, a + in the Finding

A Complexity View of Ontogeny as a Window on Phylogeny

177

Table 7.2 Intersyllabic Infant-Language Matching Results for Intrasyllabic Properties

CVCV

Sequence

Value

Languages

Other Patts.

Languages

Reduplication Variegation Height Backness Place Manner

= = = = = =

6 6 5 5 6 6

− +

Tunisian Tunisian

+ Sig. greater than AL = No sig. diff. from AL; – Sig. less than AL

column indicates that the infants showed significantly more of that CV cooccurrence than was found in their ambient language environment. An = indicates that the frequency of occurrence was not significantly different than that in the ambient language. A – displays comparisons where the infants produced significantly fewer of that CV combination than was found in their ambient language. As shown, in 5.5 of 9 potential CV co-occurrence environments, the infant’s frequency of occurrence was equal to the frequency in their own ambient language. Significantly less frequently occurring in the infants were coronal-back, dorsal-back, and dorsal-central associations. In every case except dorsal-central vowel associations, four to six of the language groups followed the significance trends. Exception language groups included two languages where the children’s frequency of occurrence was equal to their ambient language frequencies, one language where the children produced more than their ambient language, and three languages where the children produced fewer of that CV co-occurrence than their ambient language. Importantly, the predicted CV co-occurrences of labial central, coronal front, and dorsal back were equivalent for six, five, and five of the languages, respectively. While the infants were not universally producing the within-syllable patterns found in their languages, in particular, for the patterns predicted by the FC hypothesis, they were predominantly doing so. There were no consistent counter-trends in the picture where one type of CV co-occurrence was found to be predominantly different from the ambient language pattern. The numbers in the final two columns indicate findings that were not consistent with the broader trend. In each case, the number of children in that language who displayed the diverse trend is shown. Six of 27 children showed individual patterns that were different from the group trends overall. Intersyllabic infant-language matching results for intrasyllabic properties are displayed in Table 7.2. The use of the +, =, and – signs are the same as for the within syllable CV co-occurrence results. For reduplication and variegation

178

Barbara L. Davis

frequency comparisons between infants and ambient languages, six out of six infant groups showed no statistically significant difference from their own ambient language properties. Thus, these infants can be seen as anchoring early in their language regarding the frequency of variegation relative to reduplication shown across syllables. For the comparison of variegation in vowel height versus vowel backness, infants in five out of six languages showed the same pattern as their own ambient language. The Tunisian infants were the only group that showed diversity in the patterns of vowel variegation. For consonant variegation, six out of six languages showed no statistically significant difference from the frequency of variegation in either place or manner dimensions from their own ambient language patterns for frequency of occurrence. In summary, there is abundant evidence of stability in these data. Stability is indicated when phonological patterns that infants are producing in canonical babbling in their first year of life are also found in mature adult speakers in their own ambient language. In the case of within-syllable CV cooccurrences, infant participants across the six languages showed predominant matching of the within-syllable characteristics of their ambient language. It should be noted that only the Dutch language does not show predicted CV cooccurrences (MacNeilage et al., 2000; Kern & Davis, 2009). Relative to intersyllabic reduplication-variegation patterns, there was matching of ambient language variegation levels across syllables in all six languages analyzed. Types of variegation for consonants and for vowels were also remarkably similar across these languages. These findings for within- and across-syllable comparisons for infants relative to their ambient language patterns in these contemporary languages show remarkable levels of similarity between infants and mature language patterns. This finding implies that these types of patterns characteristic of mature language users in these six languages are available to the movement system of beginning users of rhythmic syllables in those languages; an early “anchor” in the language if you will. Although they are also characteristic of these languages, they are not found in all of the languages. This finding of relative stability between infants at the onset of syllable use and their language suggests that these patterns could indicate a route to syllable-like organization of serial output in early hominids initiating vocal communication based on the jaw cycle and accompanying phonation co-opted from earlier functional routines. Evolution may likely have preserved functional aspects of the internal structure of the jaw cycle (viz., intrasyllabic regularities) elaborating over time with additional adaptations seen in increase of independent articulator movements within syllables in the service of increase in message complexity. The lack of universal occurrence of these patterns across languages suggests that individual paths to language complexity are potentially unique within a landscape of dramatic regularity in sequential movement patterning.

A Complexity View of Ontogeny as a Window on Phylogeny

5

179

Change

We have reviewed some sites of early occurring stability in matching by infants in their own output to their ambient language characteristics. Results reported have been interpreted as early anchors of stability between infants and languages in the complex system observable in production output across acquisition. Understanding of acquisition from a complexity perspective must include progressive diversification across acquisition or “change” in infant output capacities as well. This progressive diversification in the inventory of sound types and how they are produced in sequences relative to ambient language patterns is usually considered a critical index of increasing complexity toward mature phonological capacities. a

Change: Data

One index of change in complexity across acquisition is in the area of consonant assimilation. Consonant assimilation is a type of simplification that is observed in early child inventories where consonants in a word are changed to become more like one another (for instance, dog is produced either as “dod” or as “gog”). Children who are developing typically most often resolve these kinds of simplification patterns by 36–48 months of age as they progress toward diversified sound inventories and accuracy using those inventories relative to ambient language targets they are attempting. Kim (Kim, 2010; Kim & Davis, 2009) conducted a longitudinal study of consonant assimilation as an index of required change in phonological complexity in ten children learning English. The children’s word-based output was analyzed between 12 and 36 months of age. In addition, data were divided into age time periods to examine change over time: 12–18; 19–24; 25–30; and 31– 36 months. Audio and video data were recorded bimonthly for one hour from each child. Data gathered were part of the Texas Davis database on PhonBank (MacWhinney & Rose, 2005). Spontaneously occurring words that could be identified as such by listeners were analyzed. CV and CVCV forms were analyzed. This included 1467 assimilated word forms (viz. 1058 CVC and 409 CVCV). The forms that contained consonant assimilations accounted for 7 percent of the entire corpus studied for the ten children. There were 20,522 words in the corpus overall during the entire period. The results reported here are illustrative of the data from the larger study (Kim, 2010). Results for consonant place assimilation showed that assimilations occurred in the following levels of frequency: Labial > Coronal > Dorsal (see Figure 7.1). The relative frequencies were calculated in ratios of assimilation to target words. For example, the ratio of 271 labial assimilations to 335 target words containing labials accounted for .81 overall. The same trend was found in both CVC and CVCV word forms. This finding indicates that word forms

180

Barbara L. Davis

1.00 0.81

Ratios

0.80 0.60

0.54

Labial Assimilation Coronal Assimilation

0.40

0.21

Dorsal Assimilation

0.20 0.00 Place Assimilation

Figure 7.1 Consonant Place Assimilation Patterns.

were most likely to change to labials (e.g., bag → bab) when there were two different consonant places in the CVC or CVCV word target, next most likely to change to coronals (e.g., dog → dod), and least likely to change to dorsals (e.g., pig→ gig). For consonant manner of articulation, the pattern of assimilations in CVC and CVCV word forms (Figure 7.2) was as follows: Stop ࣙ Nasal > Fricative assimilation. Ratios were computed in the same way as previously described for consonant place. This result indicates a dominant propensity to change toward oral and nasal stop manner, followed at very low levels by changes toward fricatives. Here the same trend was found in any consonant sequences in targets except for the nasal-stop sequence. Developmental patterns in resolution of consonant assimilation were also analyzed. The data for each participant were divided into the following time periods: 12–18; 19–24; 25–30; and 31–36 months. Results of this analysis are displayed in Figure 7.3 as raw frequencies. In the figure, it can be seen that frequency of assimilation was the highest at Time 2 (18–24 months), and then decreased across the remaining period of analysis. Labial and coronal assimilation persisted as the highest frequency patterns produced by the children in assimilated forms.

1.00

Ratios

0.80

0.91

0.79

0.60

Stop Assimilation Nasal Assimilation Fricative Assimilation

0.40

Other Assimilation

0.20 0.06

0.05

0.00

Manner Assimilation

Figure 7.2 Consonant Manner Assimilation Patterns.

A Complexity View of Ontogeny as a Window on Phylogeny

181

Frequency of Place Assimilation

350 300

48

250 19

200 191

150

114

0

25 19 41 Time1

70 Time2

100

Time3 Time periods

Dorsal Assimilation Coronal Assimilation

95

100 50

9

Labial Assimilation

66 Time4

Figure 7.3 Labial, Coronal, and Dorsal Assimilation Pattern Changes over Time.

Overall, these typically developing children followed a Labial > Coronal > Dorsal pattern in their assimilations of consonants within word forms in this early period of development. However, in the second time period where assimilation frequency peaked, coronals dominated, indicating a change in relative frequencies as discrete time periods are considered. As noted, this perspective aims to take a movement-based approach to understanding the acquisition of phonological capacities in young children. Results can be conceptualized as a movement-based hierarchy relative to movement patterns that are available to young children. Labials are the most available for children (Davis et al., 2002). They involve mandibular movement only without engagement of the tongue. Coronals are somewhat less frequent but still high in frequency relative to dorsals. They involve complete tongue tip contact. Within this movement-based approach, dorsals are the least available (Locke, 1983). They involve the back of the tongue, which is much less flexible than the tongue tip relative to movement patterning. It should be observed that these relative frequencies found in assimilated word forms match the L > C > D frequencies found in early words overall (Davis, MacNeilage, & Matyear, 2002). In contrast, languages show a pattern of more frequent coronals than labials (Maddieson, 1984). Relative to consonant manner, the hierarchy of frequency found in assimilated word forms was: Stop ࣙ Nasal > Fricative. Again, this pattern can be interpreted within the framework of a movement-based hierarchy. Stops and nasals are also more available to the young child’s movement system. Both involve complete closure of the oral tract followed by release during movement

182

Barbara L. Davis

sequences of close-open syllables (Davis et al., 2002). In contrast, fricatives involve fine adjustments of varied degrees of closure and far more precision; they are found at very low frequencies in the inventories of young children at the onset of word use (Gildersleeve-Neumann, Davis, & MacNeilage, 2000). b

Summary

Results from analysis of a frequently attested pattern in early child speech indicate that these children continue to favor labial and coronal forms available to the movement system from the onset of word use when they produce assimilations. Thus, acquisition of serial complexity in frequently occurring word forms begins with assimilation to available movements. Resolution of these constraints guided by perceptual input from ambient language is a process of overcoming movement constraints to match ambient language word target complexity. Overall, results support a proposal of movement-based principles guiding output patterns in formative periods of phonological development guided by socially mediated perceptual input to meet functional goals within the environment. 6

Ontogeny and Phylogeny

Tinbergen (1963) suggests that consideration of the processes and products of early speech acquisition in human infants might provide a window into phylogenetic processes in early users of the vocal medium for communication. In considering Tinbergen’s questions, contemporary theoretical conceptualizations found in complexity science, embodiment theory, and dynamic systems theory have been proposed as an alternative to the dominant linguistic paradigm for considering acquisition of the phonological component of language. In these theoretical approaches, phonological acquisition is seen as founded on a complex and embodied dynamic system that changes over time to support social function in the environment. The complex system proposal considered here is founded on multiple heterogeneous inputs from the child’s biological makeup (viz., neural-cognitive, production, and perception systems). It is enabled by socially available phonological input from members of the child’s speech community who “show the way” toward using the sounds necessary to communication of precise linguistic meanings. The social enabling system is embedded in function, as the forms illustrated by adults in input and the forms produced by children in output are in service of functional interactions. These functional interactions help the child construct and use his or her growing knowledge about the world for survival and for scaffolding myriad types of learning about relevant world knowledge. In Tinbergen’s sense, this complex system provides a picture of the putative

A Complexity View of Ontogeny as a Window on Phylogeny

183

process that could have been implemented over historical time in early speakers; emergence of a complex and embodied dynamic system implemented socially for functional purposes of supporting the organism’s survival and growth in knowledge. In considering closely the kinds of processes that might have been apparent over historical time, we have viewed the products of the complex dynamic system of the contemporary human infant as characterized by both stability and change. Early stability in the system creates a background for needed growth in complexity to produce diverse messages. We have reviewed data indicating that within-syllable patterns related to lack of movement of active articulators independent of the jaw in rhythmic syllables are found in common in young children and adult speakers across a number of languages. These patterns have also been found in a putative corpus of early speakers (MacNeilage & Davis, 2000c), supporting a proposal regarding their fundamental status as a base for growth in production system complexity in service of perceptual distinctiveness in phylogeny. In addition to stability in the output system as a foundation for rapid message transmission, ontogenetic and phylogenetic processes must encompass change to accomplish the functional goal of perceptual distinctiveness. In ontogeny, children must develop from highly constrained movement types and movement sequences to ambient language-specific levels of movement variegation that accurately match their growing repertoire of word targets. Early on, children depend on available movement properties for early productions of word-based sequences. Consonant assimilation is an early structural property of word forms produced by young children. Data reviewed on resolution of assimilations suggests that children rely on available movement patterns at the onset of word use in assimilations and emerge into word-target-related movement variegation. The process of change toward language-like variegation for word targets can be seen as based on the child’s complex, embodied dynamic biological system tuned by social input for functional purposes. Consistent with phylogenetic conceptualizations, the process of change is for functional purposes. REFERENCES Barlow, Steven M. (1998). Real time modulation of speech-orofacial motor performance by means of motion sense. Journal of Communication Disorders, 31(6), 511–534. Bell, Allen & Hooper, Joan B. (eds.) (1978). Syllables and segments. Amsterdam: North-Holland. Boersma, Paul (1998). Functional Phonology: Formalizing the Interactions Between Articulatory and Perceptual Drives. The Hague: Holland Academic Graphics. Camazine, Scott, Deneubourg, Jean-Louis, Franks, Nigel, Sneyd, James, Theraulaz, Guy, & Bonabeau, Eric (2001). Self-Organization in Biological Systems. Princeton: Princeton University Press.

184

Barbara L. Davis

Chen, Lei-Mei & Kent, Raymond D. (2005). Consonant-vowel co-occurrence patterns in Mandarin-learning infants, J. Child Lang. 32, 507–534. Chomsky, Noam & Halle, Morris (1968). The Sound Patterns of English. New York: Harper & Row. Clark, Andy (1997). Being There: Putting Brain, Body, and World Together Again. Cambridge, MA: The MIT Press. Damasio, Antonio R. (1994). Descartes’Error: Emotion, Reason, and the Human Brain. New York: G. P. Putnam. Davis, Barbara L. & Bedore, Lisa M. (April, 2013). Knowing and Doing: An Emergence Approach to Speech Acquisition, New York: Routledge, Psychology Press. Davis, Barbara L., McCaffrey, Helen, Von Hapsburg, Deborah, & Warner-Czyz, Andrea (2005). Perceptual Influences on Motor Control: Infants with Varied Hearing Levels, Volta Review, Vol. 105(1), 7–27. Davis, Barbara L. & MacNeilage, Peter F. (1995). The Articulatory Basis of Babbling. Journal of Speech and Hearing Research, 38, 1199–1211. Davis, Barbara L. & MacNeilage, Peter F. (2000). An Embodiment Perspective on the Acquisition of Speech Perception. Phonetica, 57(Special Issue), 229–241. Davis, Barbara L., MacNeilage, Peter F., & Matyear, Christine (2002). Acquisition of Serial Complexity in Speech Production: A Comparison of Phonetic and Phonological Approaches to First Word Production. Phonetica, 59, 75–107. Dekkers, Joost (2000). Optimality Theory: Phonology, Syntax and Acquisition. Oxford: Oxford University Press. de Saussure, Ferdinand (1916). Course in General Linguistics (R. Harris, Trans.). Peru, IL: Carus Publishing Company. Descartes, Rene (1637). The Philosophical Works of Descartes (E. S. Haldane & G. T. R. Ross, Trans. Vol. Volume 1). New York: Cambridge University Press. Edelman, Gerald M. (1992). Bright Air, Brilliant Fire: On the Matter of the Mind. New York: Basic Books. Gildersleeve-Neumann, Christina, Davis, Barbara L., & MacNeilage, Peter F. (2000). Contingencies governing production of fricatives, affricates and liquids in babbling. Applied Psycholinguistics, (21), 341–363. Giulivi, Sara, Whalen, Douglas H., Goldstein, Louis M., Nam, Hosung, & Levitt, Andrea G. (2011). An Articulatory Phonology Account of Preferred ConsonantVowel Combinations. Language Learning and Development, 7 (3), 202–225. Kauffman, Stuart (1995). At Home in the Universe: The Search for the Laws of SelfOrganization and Complexity. New York: Oxford University Press. Kern, Sophie & Davis, Barbara L. (2009). Emergent complexity in early vocal acquisition: Cross-linguistic comparisons of canonical babbling. In Chitoran, Iona, Coupé, Christophe, Marsico, Egideo & Pellegrino, Francoise (eds.), Approaches to Phonological Complexity. Phonology and Phonetics Series, Berlin: Mouton de Gruyter. 353–376. Kim, Namhee (2010). Consonant Assimilation in Early Phonological Development: A Phonetic Perspective, Unpublished doctoral dissertation, The University of Texas at Austin, Austin, Texas. Kim, Namhee & Davis, Barbara L. (April 2009). Consonant Assimilation in early words, A Phonetic Perspective, Society for Research in Child Language Development, Denver: Colorado

A Complexity View of Ontogeny as a Window on Phylogeny

185

Lakoff, George, & Johnson, Mark (1999). Philosophy in the Flesh: The Embodied Mind and Its Challenge to Western Thought. New York: Basic Books. Lee, Suyoung, Davis, Barbara L., & MacNeilage, Peter F. (2008). Segmental properties of input to infants: A study of Korean. Journal of Child Language, 35(3), 591– 617. Lindblom, Bjorn, & Maddieson, Ian (1988). Phonetic Universals in Consonant Systems. In L. M. Hyman & C. N. Li (eds.), Language, Speech, and Mind (62–78). London: Routledge. Lindblom, Bjorn, Krull, Diana, Stark, Johan (1993). Phonetic Systems and Phonological Development. In Boysson Bardies, Benedicte, de Schoen, Jusczyk, Peter, MacNeilage, Peter, Morton (eds.), Developmental Neurocognition: Speech and Face Processing in the First Year of Life. Kluwer: Dordrecht, 399–409. Lindblom, Bjorn (2008). The target hypothesis, dynamic specification, and segmental independence. In Barbara L. Davis & Krisztina Zajdo (eds.), Syllable Development: The Frame Content Theory and Beyond (327–354). New York: Routlege/Taylor & Francis. MacNeilage, Peter F. & Davis Barbara L. (1990). Acquisition of Speech Production: Frames, then Content. In Jeannerod, Marc (ed.), Attention and Performance XIII; Motor Representation and Control, Hillsdale, NJ: LEA, 452–468. MacNeilage, Peter F., Davis, Barbara L., Kinney, Ashlyn & Matyear, Christine (2000a). The Motor Core of Speech: A Comparison of Serial Organization Patterns in Infants and Languages. Invited submission to Special Millenium Issue. Child Development, 71(1), 153–163. MacNeilage, Peter F., Davis, Barbara L., Matyear, Christine & Kinney, Ashlyn (2000b). Origin of Speech Output Complexity in Infants and in Languages. Psychological Science, 10(5), 459–460. MacNeilage, Peter F. & Davis, Barbara L. (2000c). Origin of the Internal Structure of Words. Science, 288, 527–531. MacWhinney, Brian & Rose, Yvan (2005). A Shared Database for the Study of Phonological Development, PhonBank, available at http://childes.psy.cmu.edu/phon/. Maddieson, Ian (1984). Patterns of Sounds. Cambridge, New York: Cambridge University Press. Maddieson, Ian & Precoda, Kristin 1991. Syllable Structure and Phonetic Models. UCLA Working Papers Phonetica, 78, 38–49. Oyama, Susan (2000). Evolution’s Eye: A System’s View of the Biology-Culture Divide. Durham, SC: Duke University Press. Oller, D. K. & Steffans, M. L. (1994). “Syllables and Segments in Infant Vocalizations and Young Children” (speech). In M. Yavas, (ed.), First and Second Language Phonology 45–61, San Diego: Singular Publishing Group. Scott-Phillips, Thomas C. (2010). Evolutionary Psychology and the Origins of Language. Journal of Evolutionary Psychology, 8(4), 289–307. Oller, D. K. (1980). The emergence of the sounds of speech in infancy. In G. H. YeniKomshian, J. F. Kavanagh & C. A. Ferguson (eds.), Child Phonology 1: Production. New York, NY: Academic Press. Pellegrino, Francoise, Marisco, Egidio, Chitoran, Ioana, & Coupe, Christophe (eds.) (2009). Approaches to Phonological Complexity, Phonology and Phonetics Series, Berlin: Mouton de Gruyter.

186

Barbara L. Davis

Prigogine, Ilya, & Stengers, Isabelle (1984). Order Out of Chaos: Man’s New Dialogue with Nature. New York: Bantam. Prince, Alan, & Smolensky, Paul (2004). Optimality Theory: Constraint Interaction in Generative Grammar. Malden, MA: Blackwell Publishing. Smith, Linda B., & Breazeal, Cynthia (2007). The Dynamic Lift of Developmental Processes. Developmental Science, 10(1), 61–68. Thelen, Esther (1991). Motor Aspects of Emergent Speech: A Dynamic Approach. In N. Krasnegor, D. Rumbaugh, & M. Studdert-Kennedy (eds.), Biological and Behavioral Determinants of Language Development (339–362). Hillsdale, NJ: Lawrence Erlbaum. Thelen, Esther (1995). Motor Development: A New Synthesis. American Psychologist, 50(2), 79–95. Tinbergen, Nicolas (1963). On Aims and Methods of Ethology. Zeitschrift für Tierpsychologie, 20, 410–433. Von Hapsburg, Deborah & Davis, Barbara L. (2006). Exploring the Effect of Auditory Sensitivity on Canonical Babbling, Journal of Speech, Language, and Hearing Research, 49, 809–822. Varela, Francisco J., Thompson, Evan, & Rosch, Eleanor (1991). The Embodied Mind. Cambridge, MA: MIT Press. Vihman, Marilyn May (1992). Early Syllables and the Construction of Phonology in Ferguson, Menn, Stoel-Gammon (eds.), Phonological Development: Models, Research, Implications, 393–422, Timonium, MD: York Press. Wilson, E. O., & Holldöbler, Bert (1988). Dense heterarchies and mass communication as the basis of organization in ant colonies. Trends in Ecology and Evolution, 3, 65–68. Woolhouse, R. S. (1988). The Empiricists. Oxford: Oxford University Press. Zlatic, Larissa, MacNeilage, Peter F., Matyear, Christine, & Davis, Barbara L. (1997). Babbling of twins in a bilingual environment. Applied Psycholinguistics, 18(4), 453–469.

8

Language Choice in a Multilingual Society: A View from Complexity Science1 Lucía Loureiro-Porto and Maxi San Miguel

1

Introduction

Complexity Science, which can be defined as the science that studies complex systems and emergent phenomena through the construction of computational models, has for decades proved to be useful to the understanding and description of quite diverse scientific phenomena, such as flocks of birds or fish, as well as the behaviour of the brain as an aggregate of neurons and traffic jams (Mitchell 2009). Those working in the social sciences, from Thomas Hobbes (seventeenth c.) onwards (Ball 2004) have felt attracted to the shaping of a statistical view of social phenomena. Yet despite these early attempts, it is only recently that Complexity Science has begun to be used as a framework for the study of social phenomena. Indeed, its relevance is now sometimes said to be greater than any other scientific perspective, given that the boundaries between disciplines are fuzzier than ever. As noted by Ball (2012: vii), ‘The major challenges of the twenty-first century are not ones that can be understood, let alone solved, from a particular academic perspective.’ This chapter approaches the study of complexity in language from a different perspective from the majority of other contributions in this volume. It aims to provide a model of language choice in a multilingual society by examining the different factors that might play a role in the survival of one (or all) of the available languages, and does this by adopting an approach based on Complexity Science. It contributes to the study of the sociology of language, discussing the possible role played by the interaction of agents (i.e. autonomous entities 1

The authors wish to thank Xavier Castelló, Víctor M. Eguíluz and Federico Vazquez for their indispensable contributions to earlier studies which made the present chapter possible. In addition, for financial support, Lucía Loureiro-Porto thanks the Spanish Ministry for Science and Innovation and the European Regional Development Fund (grant FFI2011–26693-C02–02) and the Spanish Ministry of Economy and Competitiveness (grants FFI2014-53930-P and FFI201451873-REDT); Maxi San Miguel acknowledges financial support from FEDER and MINECO (Spain) under project FIS2015-63628-C2-2-R. Thanks are also due to Riitta Toivonen, Jari Saramäki and Kimmo Kaski for fruitful discussions of the topic, and to an anonymous reviewer for very insightful suggestions.

187

188

Lucía Loureiro-Porto and Maxi San Miguel

which can make decisions) and the structure of the social network in language choice. It also presents a model for language choice that encompasses parameters such as the role of bilingual speakers and the effects of the kind of network in which interaction takes place. 2

Complexity Science, Complex Systems, and the Social Sciences

2.1

Complexity

Some 31 different definitions of a complex system can be found in the literature, notably in earlier works (Horgan 1995). However, Complexity Science is now a mature, settled discipline with a central paradigm that reflects the triumph of emergence over reductionism. According to Anderson (1972) and Gallagher and Appenzeller (1999), for example, complex systems are those in which the collective behaviour of the constituent parts exhibits properties that cannot be inferred from their separate individual properties. In a way, it is a modern phrasing of the Aristotelian quote: ‘The whole is greater than the sum of its parts’ and it basically implies that individual behaviour is not a guide for collective behaviour. This behaviour is said to be ‘emergent’ when an effective theory of the system at some scale, or level of organisation, is qualitatively different from a lower-level theory. In other words, the emergent behaviour of the system cannot be understood in terms of the simple extrapolation of the properties of the parts. Thus, for instance, sociology is not to be reduced to the study of the psychology of individuals. Instead of focusing on the individual features of the elements of the system, attention should be paid to their interactions, as Schelling (1978) claimed in his pioneering work exploring self-organisation in societies. We usually have to look at the system of interaction between individuals and their environment, that is, between individuals and other individuals or between individuals and the collectivity. And sometimes the results are surprising. Sometimes they are not easily guessed. Sometimes the analysis is difficult. Sometimes it is inconclusive. But even inconclusive analysis can warn against jumping to conclusions about individual intentions from observations of aggregates, or jumping to conclusions about the behavior of aggregates from what one knows or can guess about individual intentions (Schelling 1978: 13–14).

Hence, the collective behaviour of elements in complex systems entails emergence of properties that can hardly, if at all, be expected from properties of those parts. Examples of complex systems include anthills, ants themselves, human economies, climate, nervous systems, cells and living things, including human beings, as well as modern energy and telecommunication infrastructures (San Miguel et al. 2012). When in a system composed of many parts,

Language Choice in a Multilingual Society

189

Table 8.1 Typical Features of Complex Systems (adapted from San Miguel et al. 2012) Feature

Example of Complex System

Many heterogeneous parts Complicated transition laws Unexpected or unpredictable emergence Sensitive dependence on initial conditions Networked hierarchical connectivities Interactions of autonomous agents Self-organisation of collective shifts Adaptivity to changing environments Co-evolving subsystems Ill-defined boundaries Multilevel dynamics

Cities, companies, climates Markets, disease transmission Chemical systems, accidents Weather systems, investments Social networks, ecosystems Road traffic, dinner parties Revolutions, fashion Biological systems, manufacturing design Computer virus software Pollution, terrorism Armies, governments, internet

each part has a clear, identifiable function which makes prediction possible, such systems are said to be ‘complicated systems’ (San Miguel et al. 2012). For example, an airplane is a ‘complicated system’ in this sense. Complex systems, by contrast, are characterised by the emergence of collective behaviour which cannot be reduced to the individual behaviour of each part. Instead, the properties of its behaviour emerge from the interaction between the parts, as in the case of the brain. The many constituent parts may be heterogeneous, as in the case of cities or climates, or homogeneous, as with water molecules forming waves in the ocean. These parts are also sensitive to network connectivity, which may guarantee the survival or cause the death of an ecosystem. They may be the result of self-organisation, as in the case of revolutions or fashion shifts, and they may have co-evolving systems, as with computer virus software, which evolves as anti-virus software is developed. Further characteristics of complex systems are included in Table 8.1 (adapted from San Miguel et al. 2012). Several steps are considered necessary for a thorough analysis of a complex system (from San Miguel et al. 2012): i) Exploration, observation and basic data acquisition ii) Identification of mechanisms iii) Modelling iv) Model validation, implementation and prediction v) Construction of a theory Needless to say, not all studies of complex systems achieve the final level here, and this will not be our intention in this chapter either. For the purposes of this chapter, we will focus on the first three levels: observation of the problem (language choice in multilingual societies), identification of mechanisms (based on

190

Lucía Loureiro-Porto and Maxi San Miguel

observation and on previous studies), and mathematical modeling (which will be qualitatively compared to several well-known linguistic situations), as will be seen in section 3.2 2.2

Social Complexity

Human societies are a prototypical complex system. Made of a high number of individuals whose interactions determine the evolution of the system (war, trade, arts, linguistic change, etc.), society is claimed to share many features with other emergent phenomena occurring in other systems composed of a large number of interacting individuals. In fact, very early in the history of sociology scholars tried to devise a mechanistic view of society, from Thomas Hobbes to David Hume, John Locke, Jeremy Bentham, John Stuart Mill and Karl Marx. To different degrees, they all tried to apply statistical laws of the natural sciences to societies (Ball 2004). Although their approaches remain largely part of the history of the field, the truth is that society conforms broadly to all the features outlined in Table 8.1. For example, it is made of many heterogeneous parts, it is network-based, it exhibits ill-defined boundaries, it includes co-evolving subsystems, and it is characterised by non-equilibrium dynamics (which explains why societies never stop developing new features). In fact, many current social changes can only be properly described in terms of Complexity Science, such as the Arab Spring revolutions, the 15M movement in Spain (Borge-Holthoefer et al. 2011),3 the Occupy Movement in the USA, the world financial crisis, international terrorism, and cyber-crime. All of these can be considered emergent phenomena whose very explanation is a matter of the interactions between individuals rather than the particular psychology of each human being involved. Nevertheless, in addition to the prototypical features of complex systems, society is also characterised by a high number of variables which differentiate humans from, say, gas particles, including ‘memory, anticipation, emotion, creativity and intention’ (Ball 2012: XI). Thus, when modelling society, it is also necessary to consider a variety of technical, financial, ethical and cultural dimensions. This chapter, then, is a contribution to the development of this new approach to social phenomena, given that the increasingly connected world in which we live will undoubtedly have consequences for the survival and disappearance of languages. 2

3

Other scholars have indeed dealt with empirical data as a means of validating their models (step (iv) in the previous list), such as the pioneering work of Abrams and Strogatz (2003), who do so with data on Scottish Gaelic and Quechua in Peru, and Kandler (2010), who studies the survival of Celtic languages. See also the 15M Movement video available online at BIFI (Instituto Universitario de Investigación Biocomputación y Física de Sistemas Complejos): http://15m.bifi.es/index_en.php.

Language Choice in a Multilingual Society

2.3

191

Modelling Society: Bases and Challenges

Mathematical models used within the framework of Complexity Science are usually considered simple, because they involve few components and thus avoid dealing with the myriad of variables that general sociological or humanistic studies typically grapple with. It is not uncommon that these ‘simple’ models end up being used in fields other than the one for which they were designed. One well-known example of this is the Ising model, developed in 1925 to model magnetism, but has gone on to be used in fields as diverse as DNA studies and the study of political elections (San Miguel 2012). According to Schelling (1978, chapter 3), simplicity should therefore not be seen as a drawback but as an advantage, in that the descriptions of different physical, social or animal phenomena may shed light on the notion, traditionally concealed or overlooked, that different phenomena may behave according to recurrent patterns. In this sense, one might ask the question whether, and to what extent, a model might predict the outcome of such disparate phenomena? The answer is that our interest in modelling, rather than being aimed exclusively at a quantitative prediction, can be more about understanding the behaviour of the system, describing mechanisms involved, and developing concepts which may lead to forecasting probable scenarios.4 Our approach, thus, is very much in Nettle’s (1999a: 103) line, who claims that ‘A simulation, however rudimentary, is thus an improvement over a merely verbal argument, in deciding what general conditions must obtain for languages to evolve in the way that they do.’ The basic elements upon which the modelling of complex collective social phenomena is based are the following: a) Characterisation of the agents. As mentioned earlier, the agents are autonomous entities that can make decisions and which are the mathematical equivalent of social individuals. Their characterisation includes the different states which describe the agents and, therefore, they may exhibit some degree of heterogeneity, as in the case of Nettle’s (1999a, 1999b) agents, which differ according to age and degree of influence on society, and the case of Ke et al.’s (2008) paper on language change in which agents differ in two aspects, namely age and type of learner. Regarding language choice, the focus of this chapter, agents are nevertheless homogeneous and will be attributed one of three possible values: they may be speakers of language A, of language B, or of both A and B. b) Interaction rules. These rules determine the way in which agents change their state through interaction among themselves. Different models include 4

In current complex systems research an important distinction is made between prediction and forecasting. Prediction is considered to be quantitatively precise, while forecasting describes possible future general scenarios.

192

Lucía Loureiro-Porto and Maxi San Miguel

different interaction rules. For example, the Voter Model is based on the agent’s imitation of another randomly chosen agent, while in Granovetter’s (1978) Threshold Model an agent will change its status only when the majority of the agents which interact with it share that different status. On the other hand, Axelrod’s Model of cultural dynamics (Axelrod 1997) takes into account homophily as a factor which determines the degree to which an agent may influence another (and make it change a given feature). c) Patterns of activity. Normally, computational approaches and simulation models update the state of the agent at a constant rate. The underlying assumption is that there is a well-defined mean inter-event time in human interactions. However, human activity patterns are not characterised by a constant rate of interactions: Beyond circadian rhythms, there is now ample evidence in data of communication technologies that interactions between two people occur by bursts of activity (Wu et al. 2010), so that the distribution of inter-event times is heavy tailed with no characteristic time. The distribution of time intervals between two interactions becomes an important element in determining collective phenomena (Barabási 2005). In terms of modelling, the updating rule must be considered as part of the model itself and not as a technical simulation detail (Fernández-Gracia et al. 2011). d) Complex networks. Determining which agents interact with which others will be the key to configuring the network. In a fully connected society, each agent interacts with all other agents. However, because this rarely happens in real situations, further types of networks have been modelled, such as small world networks, random networks, or networks with community structure. All of these networks are described in section 2.4. There are many examples of simple models that have contributed to the understanding of the mechanisms underlying social phenomena. For reasons of space, we will focus on some of the studies arising from Axelrod’s Model of cultural diffusion (Axelrod 1997). This model was born with the intention of providing an answer to the following question: If people tend to share more and more cultural features as they interact, why is it that cultural diversity persists? The model, which defines culture as a series of individual features that may change owing to social interactions, is based on two principles. The first is the homophily principle, according to which interaction is more likely to take place between similar agents (i.e. agents which share several cultural features). The second principle is the principle of social influence, according to which social interaction favours cultural similarity. The model illustrates how an interaction mechanism of local convergence can generate global polarisation (i.e. persistence of cultural diversity). It also explains the transition from cultural globalisation to cultural polarisation, depending on the number of cultural traits allowed for each cultural feature. In addition, culturally polarised states

Language Choice in a Multilingual Society

193

are shown to be unstable against cultural drift in a fixed social network, but cultural diversity can be stabilised by a dynamic co-evolution of the state of the agents and the network of interaction; the social network evolves in tandem with the collective action that it makes possible, that is, circumstance makes men as much as men make circumstance (Centola et al. 2007). This model has also been used to describe the effects of propaganda and mass media effects (e.g. González-Avella 2007, 2010). These approaches to the model describe two interesting mechanisms: (i) rather than leading to cultural homogeneisation, excessively direct propaganda causes social polarisation, as interactions among agents sharing few features tend not to take place; and (ii) a given social group may converge in a state of cultural globalisation different from the message of the propaganda. 2.4

Complex Networks and Social Networks

The use of networks in the study of social behaviour has been reported for the last eight decades (Borgatti et al. 2009). Thus, the psychiatrist Moreno (1934) explained the epidemic of runaways at the Hudson School for Girls in upstate New York by using sociometry, a technique that allowed him to graphically represent individuals’ feelings for each other by means of networks. Another example is Travers and Milgram’s (1969) often-quoted study that illustrates the well-known small world phenomenon. By randomly selecting some 300 individuals in the United States, all of whom were asked to direct letters to a certain target person, the authors showed that the average distance between two randomly chosen individuals was about six (the well-known ‘six degrees of separation’). The experiment, which was criticised on the grounds that relatively few letters actually reached their intended recipients, was later successfully replicated in two studies by Leskovec and Horvitz (2007, 2008), who used hundreds of thousands of individuals and replaced letters with online instant messages. An extensive network-based study is that by Onnela et al. (2007), who analysed a mobile phone network with more than four million users and substantiated the theory of weak ties formulated by Granovetter (1973). With these examples we can see how, although networks have been studied both from mathematical and social perspectives for several decades, ‘Network Science’ can only be said to have emerged as a discipline in the 1990s (see, as general references Barabási 2002 and Watts 2003), when, owing to the advances in statistical physics, large networks came to be studied for different purposes. Thus, network science has allowed for the comparison of natural ecosystems, social networks, electric systems, the internet, and so on, identifying universal common properties of all these. This universality and the ability to compare different systems by means of mathematically created models allows for current network science to now be considered ‘one of the most vibrant,

194

Lucía Loureiro-Porto and Maxi San Miguel

diverse and insightful areas of complex-systems research, which has cast a new light on many areas of the social, natural and engineering sciences’ (Ball 2012: 23). Complex networks (or graphs, a term preferred in mathematics) can be said to be the backbone of complex systems (see, as overviews of this topic, Albert and Barabási 2002; Dorogovtsev and Mendes 2002; Bocaletti et al. 2006; Dorogotsev et al. 2008). In such studies of society, networks have come to be modelled by nodes (representing individuals) linked by representations of social ties, very much in the line with studies carried out in the humanities following Milroy’s (1980) pioneering work on interaction networks (see, for example, the special issue of the International Journal of the Sociology of Language on the subject matter, edited by de Bot and Stoessel 2002, and Ke et al. 2008). Despite this common basis, different types of networks have been proposed from network science, some of which are described in what follows. Firstly, in random networks (Erdös and Rényi 1959) every node is randomly connected to another. This obviously does not exactly match reality, in which individuals choose their links on the basis of their common interests, geographical situation, and so on. Nonetheless, random networks yield interesting results, drawing them closer to real networks, such as the fact that the distance between connected nodes is small and grows slowly with the system’s size. Secondly, small world networks (Watts and Strogatz 1998) came to be modelled in such a way that they reflected the short average path length between two randomly chosen individuals (which had been empirically observed in Travers and Milgram 1969). In addition, small world networks also introduce a large clustering coefficient, something which is also observed in real networks, as seen in human society where new acquaintances often already share a friend or a friend of a friend. This has been seen empirically on many occasions, such as in the collaboration networks of movie actors, as in the so-called Kevin Bacon Game at www.oracleofbacon.org, which shows different ways in which movie actors are connected. Thirdly, when all the nodes of a network are connected all to all (i.e. when all the nodes of a networks have only one degree of separation between them), the network is said to be fully connected. We will also highlight in the present chapter networks with community structure, that is, those in which some groups of nodes are much more connected with one another than with other nodes in the network, leading to a modular structure which very much resembles the type of topology we find in real networks. Several modellers from statistical physics have recently tried to find a satisfying algorithm for the modelling of networks with community structure (e.g. Girvan and Newman 2002), but the network used in the current chapter is based on the algorithm proposed by Toivonen et al. (2006). These four types of complex networks have been analysed in detail so as to assess their role in language choice in a multilingual community, and results are summarised in section 3.4.

Language Choice in a Multilingual Society

2.5

195

Complexity Science and Linguistic Studies

The study of language dynamics from the point of view of Complexity Science is a growing academic field, which tries to answer questions related to language evolution, language and cognition, and language competition (or the dynamics of language use in multilingual communities), as shown by Castelló et al. (2011). This approach to language sets itself apart from Saussurean’s structural linguistics and from the Chomskyan tradition of generative grammar in that Complexity Science considers language to be a complex adaptive system (Ellis and Larsen-Freeman 2009). As such, language, with its major social function, is a system consisting of multiple agents interacting with one another (The “Five Graces” Group 2009). In this approach, Complexity Science has a lot to contribute, because interactions among individuals give rise to global emergent properties, such as consensus or fragmentation (Loreto and Steels 2007). Complexity Science tries to identify the mechanisms that favour consensus or fragmentation among the members of a social group, how opinion is formed in societies, how a shared language emerges (Naming Game, for example, in Steels 1995), and how network structure affects interactions among individuals in addition to all the previous phenomena, as well as others. The applications are so varied and promising that scholars from different disciplines have been attracted by the fresh air of Complexity Science, as seen, for example, in the special issue of the journal Advances in Complex Systems, edited by Baronchelli et al. (2012), where language dynamics are studied from the perspective of Complexity Science by linguists, cognitive scientists, psychologists, philosophers, physicists, mathematicians, computer scientists and Artificial Intelligence (AI) researchers. A main tool used in Complexity Science is computer simulations, one of whose advocates in linguistic studies is Nettle (1999a, 1999b). He makes clear that computer simulations are not aimed at formulating predictions, but at exploring the general effects of the main relationships of a real situation across a range of conditions (Nettle 1999a: 103). His work had quite an impact on studies of language change and of the spread of linguistic innovations; it inspired, for example, the work by Ke et al. (2008), who address his notion of threshold problem. This notion intends to account for an important difference between biological and cultural evolution: genetic mutations and cultural changes do not spread in the same way. Accordingly, Ke et al. (2008) propose a model for the spread of language change in different networks (regular, small-world, random and scale-free networks). A precedent of such ideas of threshold and the spread of innovations in the sociological literature is the classical paper by Granovetter (1978). In addition to Nettle (1999a, 1999b) and Ke et al.’s (2008), other relevant computational studies of language dynamics include Schulze and Stauffer

196

Lucía Loureiro-Porto and Maxi San Miguel

(2006) and Schulze, Stauffer and Wichmann (2008). We follow this trend of computational studies. However, while Nettle (1999a, 1999b) and Ke et al. (2008) study language change, we focus on social competition between two languages. We do this from a cross-disciplinary perspective, because, like Blythe and Croft (2010: 19), we also believe that ‘there are benefits to collaboration between empirical researchers in the humanities, who generally have not been trained in mathematical analysis of quantitative data, and modelers from physics or other strongly mathematical fields, whose models are not always based on empirical data from the domains being modeled’. Our findings are qualitatively compared with real-life situations in section 3. Models of Language Choice in Multilingual Societies5

3

Language contact is a phenomenon which has long been of interest to linguists from different perspectives. Some focus on language contact and its consequences on a language’s structure or on linguistic ecology (e.g. Weinreich 1953; Hamers and Blanc 1989; Romaine 1989; Grenoble and Whaley 1998, 2006; Nettle and Romaine 2000; Mufwene 2001, 2008), while others look at the way in which speakers might favour one available language over another, leading to the potential endangerment of the other language or languages (e.g. Fishman 1991, 2001; Grenoble and Whaley 1998, 2006; Crystal 2000; Nettle and Romaine 2000; Hinton and Hale 2001; Bradley and Bradley 2002; UNESCO 2003; Wölck 2004; Tsunoda 2005; Mufwene 2003, 2004, 2008; Ehala and Niglas 2007; Ehala 2010). Indeed, this is our own area of interest. We build on two models of language change (Abrams-Strogatz 2003, and the Bilinguals Model by Wang and Minett 2005 and Minett and Wang 2008) and study the different phenomena that may emerge from projecting these models on different complex networks, as well as by changing the value of the parameters. In a way, our study could be compared to that of Ke et al. (2008), although our focus is language competition, rather than language change. The following sections summarise the models we rely on and our results. 3.1

Models

As noted in section 2.1, complex systems need not be complicated, but the truth is that in real life they often are. That is the case with a multilingual society, and for this reason sociolinguistic studies from Labov (1972) onwards have paid attention to as many variables as possible (density of the population, age, 5

This section summarises the main results obtained from a six-year research project which has been partially published elsewhere (Castelló et al. 2006, 2007a, 2007b, 2008, Toivonen et al. 2009, Vazquez et al. 2010 and Castelló et al. 2013). We hereby acknowledge the work of all the authors of the previous publications.

Language Choice in a Multilingual Society

197

gender, social class, etc.). On the contrary, the virtue of modelling, in general, is that models are designed to be as simple as possible, because, in San Miguel et al.’s (2012) words: [t]he aim of simple models is to get [a] better understanding of the so-called ‘stylized facts’ of the system, i.e. to make simplified, abstracted, or typical observations – in other word[s, to] capture some ‘essence’ of the real system (San Miguel et al. 2012: 7).6

Then, with the aim of capturing the ‘essence’ of language choice in a multilingual society, mathematical models of language competition designed in Complexity Science include few variables. That is the case with both AbramsStrogatz Model (2003) and the Bilinguals Model (Wang and Minett 2005, Minett and Wang 2008), on which we base our work here. In the model proposed by Abrams and Strogatz (2003), agents may be speakers of either of the two languages available: A or B. The main factors that condition the probability (P) that one agent changes its choice of language from B to A (PB→A )7 are the prestige (s) of the languages and the relative number of neighbouring speakers (σ) using each language in the community (σA and σB ), as seen in the following equations: pB−A = s · (σA )a pA−B = (1 − s) · (σB )a (1) Abrams-Strogatz (2003) Model of Language Competition As (1) shows, in addition to s (prestige) and σ (density of the population), there is a third variable that conditions the probability that a speaker changes their status, namely a, which we label ‘volatility’ and which represents the speaker’s willingness to shift language. With these simple equations Abrams and Strogatz (2003: 900) manage to fit their model to the data found for Scottish Gaelic in Sutherland and Quechua in Huanuco. 6

7

The dilemma regarding the ideal number of variables and parameters to be included in a model parallels that of the ideal size of a map, as found in Lewis Carroll’s Silvie and Bruno Concluded, illustrated with this passage: ‘What do you consider the largest map that would be really useful?’ ‘About six inches to the mile.’ ‘Only six inches!’ exclaimed Mein Herr. ‘We very soon got six yards to the mile. Then we tried a hundred yards to the mile. And then came the grandest idea of all! We actually made a map of the country, on the scale of a mile to the mile!’ ‘Have you used it much?’ I enquired. ‘It has never been spread out, yet,’ said Mein Herr: ‘The farmers objected: they said it would cover the whole country, and shut out the sunlight! So now we use the country itself, as its own map, and I assure you it does nearly as well.’ Note that in these models speakers may switch from A to B and vice versa (both A > B and B > A are possible). One important difference regarding our study and Ke et al.’s (2008) pioneering model of language change is that in their model the change is unidirectional (only U > C is possible, namely Unchanged > Changed or adoption of an innovation in a process of innovation spreading).

198

Lucía Loureiro-Porto and Maxi San Miguel

The simplicity of the Abrams-Strogatz Model, however, results in one important aspect being missed: the fact that speakers may be users of both the languages available in the community, which proves to be a key element of language modelling. In Wang and Minett (2005) (later published with some mathematical modifications as Minett and Wang 2008), bilinguals are given an important role since they represent the bridge between two monolingual stages as a necessary step for a monolingual speaker of language A to become a monolingual speaker of language B, or vice versa. The mathematical formulation of the Bilinguals Model departs from (1), including transitions from using language A or B to become bilingual (AB), so that there is no direct change from one monolingual state to the other, as seen in (2): pA→AB = (1 − s) · (σB )a pB→AB = s · (σA )a pAB→A = s · (1 − σB )a pAB→B = (1 − s) · (1 − σA )a (2) Bilinguals Model (Minett and Wang 2005) Whenever there is more than one model designed to account for the same phenomenon, the best model is identified by testing the results obtained from model-based simulations against real data: the higher the degree to which model results and data match, the higher the validity of the model. This, for example, is the procedure followed in the natural sciences, but things become more complex (or complicated) when modelling social phenomena, since, as Ball (2012: xi) observes: It is extremely difficult – practically, financially and ethically – to conduct experiments on human social systems. And even when this is possible, the number of parameters of variables describing the system is commonly very large, and some of them may be hard or impossible to quantify. [ …] In such cases, we need not despair of the value of models or theories. Rather, it might be necessary to accept that a particular phenomenon has no unique explanation, no ‘best’ model that accounts for it (Ball 2012: xi).

Therefore, in this chapter we will not strive to arrive at firm conclusions regarding which of the two models is better: Abrams-Strogatz’s or Minett and Wang’s. We will simply compare them with the aim of exploring the different parameters they include, namely monolingualism versus bilingualism (section 3.2) and the roles of prestige (s) and volatility (a) (section 3.3), as well as a very important element in Complexity Science, namely the role played by agent interaction in different topological networks (section 3.4). 3.2

Effects of Bilingualism (Versus Monolingualism)

The effects of monolingualism versus bilingualism have been the subject of different studies which we have published elsewhere (Castelló et al. 2007a, 2013;

Language Choice in a Multilingual Society

199

Figure 8.1 One-dimensional regular lattice of degree k = 2.

Castelló 2010) by comparing the Abrams-Strogatz Model and the Bilinguals Model in two kinds of networks: regular networks and small-world networks. A regular network is one in which each agent has the same number of immediate neighbours (degree, k) and is connected only to them (see Figure 8.1 and Figure 8.2). In a simulation tool available online (http://ifisc.uib.es/research/complex/ APPLET_LANGDYN.html) we have used the network represented in Figure 8.2 to compare both models, as can be seen in Figure 8.3.

Lattice Network

Figure 8.2 Square regular lattice of degree k = 4.

200

Lucía Loureiro-Porto and Maxi San Miguel

Figure 8.3 Simulation tool designed to visualise the effects of bilingualism, prestige and volatility.

The tool shows that the default value for s (prestige) is 0.5, which means that both languages are socially equivalent. Likewise, the default value for a (which we have termed volatility) is 1. The population density of each option (monolinguals A and B in Abrams-Strogatz Model, and monolinguals A and B and bilinguals in the Bilinguals Model) is proportionally given (notice that the sum of the density of speakers is 1 and that bilingual agents, only found in the right-hand square, are white dots). With the aim of assessing the role of bilingualism, the values of s and a have not been altered for either the regular lattice (simulation tool) or small-world network (see Castelló et al. 2007a, 2013, section 3.2). The qualitative results we have obtained have the following implications: (a) The presence of bilinguals (right-hand part of the tool) gives rise to welldefined borders between the A and B domains, with the bilinguals forming these boundaries. These well-defined domains grow in time much faster than when there are no bilingual agents (Abrams-Strogatz Model, lefthand part of the tool). On the other hand, except for a minority of realisations leading to dynamical traps, the presence of bilinguals accelerates language death, that is, in the Bilinguals Model one of the languages disappears earlier than in the Abrams-Strogatz Model (Castelló et al. 2006).

Language Choice in a Multilingual Society

201

Figure 8.4 Bilingual speakers on the borders of monolingual domains (regular lattice).

(b) Once the interaction among individuals starts (‘start’ button on Figure 8.3), the number of bilinguals, which was one third in the initial conditions, decreases dramatically to ca. 3 percent. (c) The reduced numbers of bilingual speakers are located on the borders of the monolingual domains, as seen in Figure 8.4. This can be interpreted as bilinguals choosing language A or language B depending on the interlocutor. They could be considered bridging agents. In fact, a recent study of a Belgian mobile phone network (Blondel et al. 2008) shows that this is indeed the location of bilinguals, as shown in Figure 8.5. (d) Finally, the dynamics lead inevitably to language death, both in the case of Abrams-Strogatz and the Bilinguals Model. It is, however, in this third finding that we observe the most pronounced differences between the synergistic effects of bilingualism and the type of network. Although language death is the final scenario in both regular lattices and small-world networks, the latter accelerate the process (see Castelló et al. 2007a, and also section 3.4). Qualitatively this is a highly relevant result in an increasingly globalised world, where distance among speakers is much less than one would expect. Perhaps the rapid erosion of grammatical, lexical and phonological features that gives way to the emergence of linguistic varieties such as world standard spoken English (Crystal 1997) can be explained in terms of a small-world effect, just like the disappearance of language A or B in the simulations here.

202

Lucía Loureiro-Porto and Maxi San Miguel

Figure 8.5 Islands of monolingual speakers linked by a small number of bilinguals. [Figure from Blondel et al. 2008]

3.3

Prestige and Volatility: Different Values and Effects

The simulation tool described earlier (see Figure 8.3) allows the user to alter the values of the parameters s (prestige) and a (volatility) in real time. Let us first consider the effects of prestige. As described in detail in Castelló et al. (2013; section 3.3), the values of s have two basic implications: (a) languages may be socially equivalent (s = 0.5) and (b) one of the languages is socially more prestigious than the other (bearing any other value). The results obtained in the simulation tool confirm

Language Choice in a Multilingual Society

203

this: a situation in which the two languages exhibit different degrees of prestige accelerates the disappearance of the less prestigious language. There are hundreds of examples of the effects of prestige on language communities around the world (e.g. the threat to Native American languages in America). Likewise, there are many real examples of languages which have increased their probability of survival once the prestige associated with them has been actively enhanced, as is the case of Quechua in Cuzco (see Manley 2008). Nevertheless, although prestige is traditionally considered a powerful factor in the survival or death of a language, dissenting voices have also been heard against the overestimation of prestige (e.g. Mufwene 2003). For this reason, we have modelled another important factor in the survival or death of a language, namely speakers’ willingness to shift languages, parameter a in the model. As far as parameter a is concerned, although it is not often used in linguistics, it has its precedent in social impact theory (Nowak et al. 1990) and it is also a main ingredient of the work on language change by Nettle (1999a, 1999b). In Nowak et al.’s (1990) theory, the impact of a group of agents on a given individual is proportional to a power a of their number. The departure from a linear relation (a = 1) is argued to allow for the fact that persuasiveness is not just proportional to the size of the group. While for matters of public opinion empirical studies show that a is ca. 0.5 (Nowak et al. 1990), Nettle (1999a, 1999b) fixed a = 0.8 in his simulations. However, Abrams and Strogatz (2003), with no reference to social impact theory, fitted a = 1.3 to reproduce data on language competition between Spanish and Quechua, English and Welsh, and English and Scottish Gaelic. We will use a as a free parameter to study different scenarios of language competition and finding different qualitative behaviour for a < 1 and a > 1. Thus, in our simulation tool (which, it must be remembered, is based on a regular lattice), parameter a (or volatility) has a default value 1 which can be changed to any value between 0.1 and 5. The lower the value of a, the higher the degree of volatility, and vice versa. This means that the most volatile system is that in which a = 0.1, and the least volatile one, that in which a = 5. The simulation tool shows how the model responds to the expected result: the effects of a low prestige will be minimised by a system of low volatility.8 In other words, if speakers are unwilling to shift languages (for instance, because it functions as a strong ethnic identity marker), they are less influenced by prestige. This explains, for example, the survival of minority languages for centuries, even if they are apparently despised by socially more prestigious languages (H-languages). In addition, the comparison between the Abrams-Strogatz Model and the Bilinguals Model shows how 8

Even if one might be tempted to think that prestige is one of the factors underlying volatility (i.e. that speakers will be more willing to shift to the most prestigious language), it must be noted that in the model the two parameters work independently.

204

Lucía Loureiro-Porto and Maxi San Miguel

Figure 8.6 Abrams-Strogatz Model vs. Bilinguals Model, when a = 0.1 (high volatility).

the presence of bilinguals constitutes a factor slowing down the death of the less prestigious language. Elsewhere Castelló et al. (2013: section 3.3) established a qualitative comparison between this computer simulation and a real social situation, namely that exhibited by the ‘competition’ between Spanish and Galician in NW Spain under dictator Franco’s regime (1936–1975). In this period, the overt prestige of Spanish was clearly higher than that of Galician, the L-language, but the latter was not threatened by the former, because of an attested lack of volatility of its speakers, both in linguistic and social terms. In a nearly feudal society in which class-shifting was virtually impossible, language shift was pointless, as it would isolate speakers from their home group, while integration in the community speaking the other language was prevented by the social rules. Speakers of Galician were bilingual; they used Spanish for specific purposes (Ayestaran and de la Cueva 1974), but the diglossic structure in which languages coexisted (without competition) appears to have delayed the possible threat from Spanish, the more prestigious language. The last interesting effect of volatility can be seen if a is lowered to 0.1, its lowest possible value in the simulation tool, which represents a highly volatile system, that is, one in which all speakers are willing to shift languages (as might be the case, for example, at a party with Erasmus students anywhere in Europe). The results are worth noting, because, counterintuitively, this does not lead to language death or to the imposition of one language over the other. Rather, the situation obtained is quite the opposite: no language ever dies. Irrespective of the number of interactions which are allowed (note that in Figure 8.6 more than 8,000 interactions have taken place), the density of speakers of both languages available remains without significant alteration. Even more interesting is the result obtained for the Bilinguals Model. Contrary to the situation previously described (in which, once interactions take place, the number of bilinguals becomes very low because they are restricted to the borderline areas),

Language Choice in a Multilingual Society

205

bilinguals persist in representing nearly a third of the agents in the lattice and are placed throughout the network (with no clear domain of monolingual speakers emerging), as can be seen in Figure 8.6. High volatility, as we have argued elsewhere (Castelló et al. 2013; section 3.3), may be responsible for the emergence of contact varieties in which speakers master both languages available and switch often and comfortably. Evidence for this claim may be found in code-switching varieties such as Spanglish in the United States and Yanito (or Llanito) in Gibraltar, a British colony on the south of the Iberian Peninsula, which has for 300 years experienced constant contact between Spanish and English. In the words of a Gibraltarian interviewed by Levey (2008): ‘I’ve got about two or three different groups of friends and some of them speak just English and some of them speak a mixture but I think that does tend to happen a lot [ …] I have no problems with Spanish or English. I just adapt’ (2008: 80). This natural ability to adapt is what we call volatility. From a general complexity perspective the change from low to high volatility at a = 1 is a transition point. For high volatility (a < 1) the system does not evolve to an ordered state and there is dynamical coexistence of two languages, while for low volatility (a > 1) the system orders with growing spatial domains of the two languages until one of them takes over in what is known as spontaneous symmetry breaking (for s = 0.5). The parameter a is the analogue of temperature in thermodynamic phase transitions, with a = 1 being the analogue of the critical point. A main difference in that analogy is that here the individual changes of state (language shift by an agent) only occur at the spatial borders between domains, not in the bulk of the domains, with these borders being only well defined for a > 1. 3.4

Effects of the Topological Structure: Different Complex Networks

Having now qualitatively addressed the roles of bilingualism, prestige and volatility in regular lattices, in this section we compare the results obtained in the simulation tool with those from mathematical approximations to the types of networks described in section 2.4, namely, fully connected networks, random networks, small-world networks and networks with community structure. Our aim is to assess whether the topological structure of the network has any effect on the dynamics and, therefore, on the survival or disappearance of the languages in competition. Various previous studies have included a quantitative approach to the differences among the networks (Castelló et al. 2006, 2007b, 2011; Vazquez et al. 2010). Here, for the sake of clarity and brevity, we will start by comparing fully connected networks and random networks on the one hand, and small-world networks and networks with community structure on the other.

206

Lucía Loureiro-Porto and Maxi San Miguel

Fully connected networks, it must be recalled, are those in which each agent is connected to all other agents in the network. Random networks, in turn, are those in which each agent is randomly connected to more than one other agent, with some agents being connected to more agents than some others, which have only a handful of neighbours. In fully connected networks we have observed an abrupt behaviour in both the Abrams-Strogatz Model and the Bilinguals Model. That is, in both models the situation shifts from language maintenance to language death discretely, rather than smoothly for a threshold value of a. In addition, in neither model does parameter s (prestige) have any effect on language maintenance, as the only factor that conditions the transition from language coexistence to language extinction is parameter a (volatility). However, an important difference is found between both models: for the two languages to survive in the Bilinguals Model, agents need to be more volatile than in the Abrams-Strogatz alternative. That is, in the presence of bilinguals language coexistence is possible only if speakers are highly volatile, if they are willing to shift languages.9 If we compare these results from fully connected networks with those obtained for random networks, we observe both a qualitative and a quantitative difference between them, as in the latter prestige does play a role in language maintenance or death. While in fully connected networks the value of parameter s is irrelevant, in random networks both a and s condition the survival of one or the two languages. Obviously, the most prestigious language has more chances to survive, but, one may wonder, which difference between these two types of networks puts prestige in such a relevant position? The answer lies in the effects of local interactions. While in fully connected networks there are no local interactions at all (since agents are connected all to all); in random networks each agent has a finite, small number of neighbours. The number of neighbours of each agent is called k (degree), and random networks tend to become fully connected networks as the mean degree increases. Thus, if the mean degree of a random network is low, local interactions take place and the effect of prestige is higher. At the same time, as the mean degree increases (i.e. as the random network becomes more and more similar to a fully connected one), interactions become global and the effect of prestige is diminished. To sum up the comparison between fully connected and random networks, we could say that the lower the mean degree of a random network, the higher the number of local interactions, and, therefore, the stronger the role of prestige. The role of local interactions, then, must be considered as a relevant factor in the modelling of language choice, as it reduces the chances of language coexistence. 9

For the Abrams-Strogatz Model the value of a in which this transition takes place is 1, while for the Bilinguals Model it is 0.63.

Language Choice in a Multilingual Society

207

Figure 8.7 Snapshots of the Abrams-Strogatz Model (top) and the Bilinguals Model (bottom) in a small-world network (bilinguals = white). (Figure from Castelló et al. 2013)

Having assessed the role of local interactions, let us now focus on two types of networks which do allow for local interactions, namely small-world networks and networks with community structure. Figure 8.7 illustrates the comparison between the Abrams-Strogatz Model and the Bilinguals Model in a small-world network (for a = 1, s = 0.5) and reveals another important piece of information regarding modelling language maintenance: the presence of bilinguals accelerates language death in this type of network, as, after the same number of interactions the minority language is about to disappear in the Bilinguals Model, while it is not still in danger of extinction in the Abrams-Strogatz Model. One could think, then, that bilinguals play a negative role in language maintenance, but the comparison between small-world networks and networks with community structure prevents us from jumping to conclusions. Figure 8.8 compares the Abrams-Strogatz Model and the Bilinguals Model in a network with community structure and shows how the presence of bilinguals prevents language death. After 50 interactions (t = 50) in the Abrams-Strogatz Model, the number of black agents is much smaller than after 100 interactions (t = 100) in the Bilinguals Model. This difference between both models in this type of network goes on for different time values, until the minority language

208

Lucía Loureiro-Porto and Maxi San Miguel

Figure 8.8 Abrams-Strogatz Model (left) and Bilinguals Model (right) in a network with community structure (t = time). (Figure from Castelló et al. 2007b)

Language Choice in a Multilingual Society

209

dies out in the Abrams-Strogatz Model, while in the Bilinguals Model the path to extinction may be simply blocked by a bilingual agent being the single connection between an isolated community and the remainder of the network (bottom, right-hand side of Figure 8.8). After 1,000 interactions among the agents (t = 1000) an isolated community persists nonetheless, owing to the communication held with the other communities by means of the (white) bilingual agent, which may be considered an agent with a privileged situation in the network and which plays the role of a social bridge. The situation in Figure 8.8 is not counterintuitive at all, and there are plenty of examples of this sort in the real world of language contact, such as the often-cited case of the Pennsylvania Dutch, an old variety of German which has survived for centuries, owing to its use in a community that is not just geographically but also socially isolated. A final finding regarding the role played by the topological structure of the network concerns the different times needed for extinction in the presence and in the absence of bilinguals. We have quantitatively measured this time for regular lattices, small-world networks and networks with community structure when a = 1 and s = 0.5 (socially equivalent languages). Our findings show that the extinction speed of one of the languages in the Abrams-Strogatz Model is similar in the three types of networks, while important differences are observed for the Bilinguals Model. In the latter, the extinction of one of the languages is faster in small-world networks, slower in regular lattices and even slower in networks with community structure (small-world networks > regular lattices > networks with community structure). Summing up this section, it can be said that, without question, the kind of network in which interactions take place is a strong influencing factor on language dynamics, as it plays a central role in the potential survival or disappearance of one of the languages in competition. 4

Summary and Conclusions

This chapter represents an approach to the study of language competition and selection in a multilingual population (Mufwene 2001, 2008) from the point of view of Complexity Science. Such a framework departs from traditional sociolinguistic methods that consider many individual features of the potential speakers of a language, focusing instead on interactions among individuals and the different possible outcomes of this. With the aim of guiding the reader through the basic aims and methods of Complexity Science, section 2 introduced the topic (2.1) and discussed the various ways in which this interdisciplinary framework has proved useful to understanding different social phenomena (2.2 and 2.3). It also concentrated on the crucial concept of network (section 2.4) and described the different complex networks used in various proposals for modelling society. Finally, it provided a brief summary of the

210

Lucía Loureiro-Porto and Maxi San Miguel

different approaches to language which have been made from the point of view of Complexity Science (section 2.5). Section 3 included the presentation of two models of language choice Abrams-Strogatz (2003) Model and the Bilinguals Model by Wang and Minett (2005; Minett and Wang 2008). As stated in section 3.1, our aim was not to determine which of the two models is better. We find both of them useful as points of departure for our study of the different factors that appear to play the most central role in the survival or disappearance of a language: (a) bilingual speakers, (b) prestige of the languages and volatility of the speakers, and (c) topological structure of the network. The following are our main findings: (1) Bilinguals have been found to play a central role in language competition; the results for the Abrams-Strogatz and the Bilinguals Models show significant differences. Firstly, the presence of bilingual agents reduces the possibility of language coexistence in fully connected networks, because a higher degree of volatility is required for both languages to survive than in the Abrams-Strogatz Model. Secondly, bilinguals accelerate language death in small-world networks (Figure 8.7). Thirdly, bilinguals prevent language death in networks with community structure (Figure 8.8). The role of bilinguals, then, is highly dependent on the topological structure of the network. (2) The prestige of the languages (parameter s) has usually been claimed to be determinant in the survival of a language, although dissenting voices have been heard, such as Mufwene (2003). The volatility of the speakers (parameter a) is probably not so often used in sociolinguistic studies, but, as we have shown, it has its origins in social impact theory (Nowak et al. 1990) and it has been included in pioneering attempts to model language change (Nettle 1999a,1999b). Both parameters have been studied in detail for regular lattices (section 3.3) and have also been qualitatively approached for fully connected and random networks (section 3.4). Our findings are in line with Mufwene’s (2003) as they question the overwhelming weight accorded to prestige as a factor affecting language maintenance. The volatility of the speakers proves itself to be a more determining feature: low volatility slows down language death in regular lattices (for socially inequivalent languages), whereas high volatility prevents it from taking place (for s = 0.5, as seen in Figure 8.6). In fully connected networks we have observed that volatility is the only factor that conditions the transition from language coexistence to language extinction, regardless of the value of s (i.e. regardless the prestige of the two languages). Finally, in both fully connected and random networks a high volatility of the speakers is required for both languages to survive. Thus, the models presented here question the almighty role played by prestige and militate instead for volatility as a factor of paramount importance in resolving language competition.

Language Choice in a Multilingual Society

211

(3) The topological structure of the network has been proved to play a central role in language competition too. In the first place, we have found that in networks that do not give way to local interactions (fully connected networks), the role of prestige is simply irrelevant, while it does interact with volatility in networks that give way to local interactions, that is, those in which agents have a finite, small number of neighbours (random networks, small-world networks and networks with community structure). Thus, local interactions have been considered responsible for an increase in the role played by prestige. Secondly, the topological structure of the network also has been found to play a role in the speed of extinction when bilingual agents are present: extinction takes place earlier in small-world networks than in regular lattices, and earlier in regular lattices than in networks with community structure. Apart from these general results, it should be noted that the combination of bilingual speakers and networks with community structure (the most likely in real situations) is a powerful factor in the survival of threatened languages (Figure 8.8) and it actually helps to understand real-life situations of language maintenance. This leads us to two important conclusions. Firstly, even if our intention was not to assess the degree of representativeness of the models provided by Abrams and Strogatz (2003) and by Wang and Minett (2005), we can affirm that the introduction of bilinguals provides a more accurate approach to reality. Secondly, the detailed assessment of different complex networks allows us to state that those with community structure (based on the algorithm by Toivonen et al. 2006) do provide a closer representation of many social networks. Our study, then, goes beyond the results by Ke et al. (2008), who do not consider this type of topological structure, although they recognise that their modelling is necessary and desirable (2008: 947). These two general conclusions constitute a relevant contribution to the study of language dynamics, with special reference to language competition. Nevertheless, we are aware that further research must include improvements in the models, such as the possibility of modelling the use of two languages by the same speaker depending on the interlocutor. That is, in the models, language could be considered a feature of each link between agents rather than a feature of each agent, in line with what has been generally modelled by FernándezGracia et al. (2012). A second improvement of the model could be to include dynamical networks, in which links are not fixed, but change over time (just like human relationships), as studied by Carro et al. (2014). Following these two general studies, one could derive dynamical models of language competition considering, for example, that the preference to use a language (including its knowledge) is a property of the agent, while the use of a language is a property of the link. In a different line, future developments of the work could also involve the modelling of heterogeneous agents, by introducing parameters such

212

Lucía Loureiro-Porto and Maxi San Miguel

as the age or the progressive acquisition of a language, as done in other studies (e.g. Ke et al. 2008). As for the contribution to Complexity Science, our approach to the study of language dynamics, as presented in this chapter, fulfils the three initial steps acknowledged to be taken in this field, following San Miguel et al. (2012), described and discussed in section 2.1. Thus, we have observed reality about language coexistence and possible competition and reviewed some literature on language endangerment in order to identify some of the mechanisms involved in language choice. What we learned enabled us to design models that may contribute to a better understanding of the dynamics involved in language shift or maintenance. We are aware, nonetheless, that there are still two further necessary steps in order to meet the aims of Complexity Science: the validation of the model and the construction of a theory. The validation of the model is only possible through comparison with real ethnographic data. Although prior attempts to model language shift have resorted to real data (e.g. Abrams and Strogatz 2003; Blondel et al. 2008; Kandler et al. 2010), this is a step we have not yet taken, given that our primary aim has not been to predict but to forecast by trying to understand the mechanisms that underlie language shift from a strictly theoretical point of view, leaving aside real data. This chapter shows that this can indeed be done and reveals the role played by different factors in language competition, such as bilingual speakers, prestige of the languages, speakers’ volatility and topographical structure of the network in a way that the mere analysis of data would not allow. In the same vein, we would like to point out that blind data analysis need not lead us to the formulation of a theory, but to a simple correlation among variables without establishing cause-relation implications. In this sense, we believe that non-data-based theoretical models may contribute to questioning the observer’s commonsense beliefs (Watts 2011) at the same time as new questions open up in a given field. To sum up, then, even if future data-based studies may help calibrate and validate the models presented here, we think that the approach developed in this chapter reveals important information about the factors that play a role in language choice and which might have gone unnoticed in a study employing a discursive or a data analysis approach. REFERENCES Abrams, Daniel M., and Steven H. Strogatz. 2003. Modeling the dynamics of language death. Nature 424, 900. Albert, Réka, and Albert-Lászlo Barabási. 2002. Statistical mechanics of complex networks. Review of Modern Physics 74, 47–97. Anderson, Philip. 1972. More is different. Science 177, 393–6. Axelrod, Robert. 1997. The dissemination of culture: A model with local convergence and global polarization. The Journal of Conflict Resolution 41(2), 203–26.

Language Choice in a Multilingual Society

213

Ayestaran Aranaz, Margarita, and Justo de la Cueva Alonso. 1974. Las familias de la provincia de Pontevedra (Galleguidad y conflicto lingüístico). Sevilla: Instituto de Ciencias de la Familia. Ball, Philip. 2004. Critical Mass: How One Thing Leads to Another. London: Heinemann. Ball, Philip. 2012. Why Society is a Complex Matter. Meeting Twenty-First Century Challenges with a New Kind of Science. Berlin: Springer Verlag. Barabási, Albert-Lászlo. 2002. Linked: The new science of networks. Cambridge: Perseus. Barabási, Albert-Lászlo. 2005. The origin of bursts and heavy tails in human dynamics. Nature 435, 207–11. Baronchelli, Andrea; Vittorio Loreto; and Francesca Tria. 2012. Language dynamics. Advances in Complex Sytems 15(3–4), 1203002. Blondel, Vincent; Jean-Loup Guillaume; Renaud Lambiotte; and Etienne LeFebvre. 2008. Fast unfolding of communities in large networks. Journal of Statistical Mechanics P10008. Blythe, Richard and William Croft. 2010. Can a Science–Humanities Collaboration Be Successful? Adaptive Behavior 18, 12–20. Boccaletti, Stefano; Vito Latora; Yamir Moreno; Martin Chavez and D.-U. Hwang. 2006. Complex Networks: Structure and Dynamics. Physics Reports 424, 175– 308. Borgatti, Stephen P.; Ajay Mehra; Daniel J. Brass; and Giuseppe Labianca. 2009. Network analysis in the social sciences. Science 323, 892–5. Borge-Holthoefer Javier; Alejandro Rivero; Íñigo García; Elisa Cauhé; Alfredo Ferrer; Darío Ferrer; David Francos; David Iñíguez; María Pilar Pérez; Gonzalo Ruiz; Franciso Sanz; Fermín Serrano; Cristina Viñas; Alfonso Tarancón; and Yamir Moreno. 2011. Structural and Dynamical Patterns on Online Social Networks: The Spanish May 15th Movement as a Case Study. PLoS ONE 6(8), e23883. doi:10.1371/journal.pone.0023883. de Bot, Kees, and Saskia Stoessel (eds.). 2002. Language change and social networks. [Special issue]. International Journal of the Sociology of Language 153. Bradley, David, and Maya Bradley (eds.). 2002. Language Endangerment and Language Maintenance. London/New York: Routledge Curzon. Carro, Adrian; Federico Vazquez; Raúl Toral; and Maxi San Miguel. 2014. Fragmentation transition in a coevolving network with link-state dynamics. Physical Review E, 89(6): 062802. Castelló, Xavier. 2010. Collective Phenomena in Social Dynamics: Consensus problems, ordering dynamics and language competition. Ph.D. dissertation. IFISC (Institute for Cross-Disciplinary Physics and Complex Systems, CSIC-UIB), Palma de Mallorca, Spain. Available online at http://ifisc.uib-csic.es/publications/ publication-detail.php?indice=2103, accessed 8 November 2012. Castelló, Xavier; Víctor M. Eguíluz; and Maxi San Miguel. 2006. Ordering dynamics with two non-excluding options: Bilingualism in language competition. New Journal of Physics 8, 308–22. Castelló, Xavier; Lucía Loureiro-Porto; Víctor M. Eguíluz; and Maxi San Miguel. 2007a. The fate of bilingualism in a model of language competition. Advancing social simulation. The First World Congress, ed. by Shingo Takahashi, David Sallach and Juliette Rouchier, 83–94. Berlin: Springer Verlag.

214

Lucía Loureiro-Porto and Maxi San Miguel

Castelló, Xavier; Riitta Toivonen; Víctor M. Eguíluz; Jari Saramäki; Kimmo Kaski; and Maxi San Miguel. 2007b. Anomalous lifetime distributions and topological traps in ordering dynamics. Europhysics Letters 79, 66006–1–6. Castelló, Xavier; Riitta Toivonen; Víctor M. Eguíluz; Lucía Loureiro-Porto; Jari Sarmäki; Kimmo Kaski; and Maxi San Miguel. 2008. Modelling language competition: Bilingualism and complex social networks. The Evolution of Language. Proceedings of the 7th International Conference (EVOLANG7), ed. by Andrew D. M. Smith, Kenny Smith and Ramon Ferrer i Cancho, 59–66. Singapore: World Scientific Publishing Co. Castelló, Xavier; Federico Vazquez; Víctor M. Eguíluz; Lucía Loureiro-Porto; Maxi San Miguel; Laetitia Chapel; and Guillaume Deffuant. 2011. Viability and Resilience in the Dynamics of Language Competition. Viability and Resilience of Complex Systems. Concepts, Methods and Case Studies from Ecology and Society, ed. by Gillaume Deffuant and Nigel Gilbert, 49–84. Boston/Dordrecht/London: Kluwer Academic Publishers. Castelló, Xavier; Lucía Loureiro-Porto; and Maxi San Miguel. 2013. Agent-based models of language competition. International Journal of the Sociology of Language 221, 21–51. Centola, Damon; Juan Carlos González-Avella; Víctor M. Eguíluz; and Maxi San Miguel. 2007. Homophily, Cultural Drift and the Co-Evolution of Cultural Groups. Journal of Conflict Resolution 51, 905–29. Crystal, David. 1997. English as a Global Language. Cambridge: Cambridge University Press. Crystal, David. 2000. Language Death. Cambridge: Cambridge University Press. Dorogovtsev, Sergey, and José Fernando Ferreira Mendes. 2002. Evolution of networks. Advances in Physics 51, 1079. Dorogovtsev, Sergey N.; Alexander V. Goltsev; and José Fernando Ferreira, Mendes. 2008. Critical phenomena in complex networks. Review of Modern Physics 80, 1275. Ehala, Martin. 2010. Ethnolinguistic vitality and intergroup processes. Multilingua 29, 203–21. Ehala, Martin, and Katrin Niglas. 2007. Empirical evaluation of a mathematical model of ethnolinguistic vitality: The case of Võro. Journal of Multilingual and Cultural Development 28(6), 427–44. Ellis, Nick C., and Diane Larsen-Freeman. 2009. Language as a complex adaptive system. Oxford: Wiley-Blackwell. Erd˝os, Paull, and Alfréd Rényi. 1959. On Random Graphs. I. Publicationes Mathematicae 6, 290–97. Fernández-Gracia, Juan; Víctor M. Eguíluz; and Maxi San Miguel. 2011. Update rules and interevent time distributions: Slow ordering vs. no ordering in the Voter Model. Physical Review E 84, 015103. Fernández-Gracia, Juan; Xavi Castelló; Víctor M. Eguíluz; and Maxi San Miguel. 2012. Dynamics of link states in complex networks: The case of a majority rule. Physical Review E 86.066113(1–8). Fishman, Joshua A. 1991. Reversing language shift: Theoretical and empirical foundations of assistance to threatened languages. Clevedon: Multilingual Matters. Fishman, Joshua A. (ed.). 2001. Can Threatened Languages Be Saved? Reversing Language Shift Revisited: A 21st-Century Perspective. Clevedon: Multilingual Matters.

Language Choice in a Multilingual Society

215

Gallagher, Richard, and Tim Appenzeller. 1999. Beyond reductionism. Science 284, 79. Girvan, M. and M. E. J. Newman. 2002. Community structure in social and biological networks. Proceedings of the National Academy of Science 99(12), 7821–6. González-Avella, Juan Carlos; Mario G. Cosenza; Konstantin Klemm; Víctor M. Eguíluz; and Maxi San Miguel. 2007. Information Feedback and Mass Media Effects in Cultural Dynamics. Journal of Artificial Societies and Social Simulation 10, 1–17. González-Avella, Juan Carlos; Mario G. Cosenza; Víctor M. Eguíluz; and Maxi San Miguel. 2010. Spontaneous ordering against an external field in nonequilibrium systems. New Journal of Physics 12, 013010. Granovetter, Mark S. 1973. The strength of weak ties. The American Journal of Sociology 78, 1360–80. Granovetter, Mark S. 1978. Threshold Models of Collective Behavior. The American Journal of Sociology 83–1420–43. Grenoble, Lenore A., and Lindsay J. Whaley (eds.). 1998. Endangered Languages: Current Issues and Future Prospects. Cambridge: Cambridge University Press. Grenoble, Lenore A., and Lindsay J. Whaley. 2006. Saving Languages: An Introduction to Language Revitalization. Cambridge: Cambridge University Press. Hamers, Josiane F., and Michel H. A. Blanc. 1989. Bilinguality and Bilingualism. Cambridge: Cambridge University Press. Hinton, Leanne, and Kenneth Locke Hale. 2001. The green book of language revitalization in practice. San Diego: Academic Press. Horgan, John. 1995. From complexity to perplexity. Scientific American 272(6), 104–9. Kandler, Anne; Roman Unger; and James Steele. 2010. Language shift, bilingualism and the future of Britain’s Celtic languages. Philological Transactions of the Royal Society B 365, 3855–64. Ke, Jinyun; Tao Gong; and William S-Y Wang. 2008. Language Change and Social Networks. Communications in Computational Physics 3(4), 935–49. Labov, William. 1972. Sociolinguistic Matters. Philadelphia: University of Pennsylvania Press. Leskovec, Jure, and Eric Horvitz. 2007. Worldwide Buzz: Planetary-Scale Views on an Instant-Messaging Network. Microsoft Research Technical Report MSR-TR-2006– 186, Microsoft Research, June 2007. Available online at http://research.microsoft .com/apps/pubs/default.aspx?id=70389, accessed 8 November 2012. Leskovec, Jure, and Eric Horvitz. 2008. Planetary-Scale Views on a Large InstantMessaging Network. Proceedings of WWW 2008 Beijing, China, ed. by Jinpeng Huai and Robin Chen 695–704. Available online at wwwconference.org/ www2008/papers/Proceedings.html, accessed 8 November 2012. Levey, David. 2008. Language Change and Variation in Gibraltar. Amsterdam and Philadelphia: John Benjamins. Loreto, Vittorio, and Luc Steels. 2007. Emergence of language. Nature Physics 3, 758– 60. Manley, Marilyn S. 2008. Quechua language attitudes and maintenance in Cuzco, Peru. Language Policy 7, 323–44. Milroy, Lesley. 1980. Language and Social Networks. Oxford: Blackwell. Minett, James W., and William S-Y. Wang. 2008. Modelling endangered languages: The effects of bilingualism and social structure. Lingua 118, 19–45.

216

Lucía Loureiro-Porto and Maxi San Miguel

Mitchell, Melanie. 2009. Complexity: A Guided Tour. Oxford: Oxford University Press. Moreno, Jacob L. 1934. Who shall survive? Washington DC: Mental Disease Publishing Company. Mufwene, Salikoko. 2001. The Ecology of Language Evolution (Cambridge Approaches to Language Contact 1). Cambridge: Cambridge University Press. Mufwene, Salikoko. 2003. Language Endangerment: What Have Pride and Prestige Got to Do With It? When languages collide. Perspectives on Language Conflict, Language Competition and Language Coexistence, ed. by Brian Joseph, 324–46. Ohio: Ohio State University Press. Mufwene, Salikoko. 2004. Language birth and death. Annual Review of Anthropology 33; 201–22. Mufwene, Salikoko. 2008. Language Evolution. Contact, Competition and Change. London/New York: Continuum International Publishing Group. Nettle, Daniel. 1999a. Using social impact theory to simulate language change. Lingua 108, 95–117. Nettle, Daniel. 1999b. Is the rate of linguistic change constant? Lingua 108, 119–36. Nettle, Daniel, and Suzanne Romaine. 2000. Vanishing Voices: The Extinction of the World’s Languages. Oxford: Oxford University Press. Nowak, Andrzej; Jacek Szamrej; and Bibb Latané. 1990. From private attitude to public opinion: A dynamical theory of social impact. Psychological Review 97, 362–76. Onnela, Jukka-Pekka; Jari Saramäki; Jörkki Hyvönen; Gábor Szabó; David Lazer; Kimmo Kaski; Janos Kertész; and Albert-Lászlo Barabási. 2007. Structure and tie strengths in mobile communication networks. Proceedings of the National Academy of Science 104, 7332–36. Romaine, Suzanne. 1989. Bilingualism. Oxford/New York: Blackwell. San Miguel, Maxi. 2012. Fenómenos colectivos sociales. Revista Española de Física 26, 56–63. San Miguel, Maxi; Jeffrey H. Johnson; Janos Kertesz; Kimmo Kaski; Albert DíazGuilera; Robert S. MacKay; Vittorio Loreto; Peter Erdi; and Dirk Helbing. 2012. Challenges in Complex Systems Science. European Physical Journal Special Topics 214, 245–71. Schelling, Thomas. 1978. Micromotives and Macrobehaviour. New York: Norton. Schulze, Christian, and Dietrich Stauffer. 2006. Recent developments in computer simulations of language competition. Computing in Science and Engineering 8(3), 60– 7. Schulze, Christian; Dieter Stauffer; and Soeren Wichmann. 2008. Birth, survival and death of languages by Monte Carlo simulation. Communications in Computational Physics 3, 271–94. Steels, Luc. 1995. A self-organizing spatial vocabulary. Artificial Life 2, 319–32. The “Five Graces” Group (Clay Beckner, Richard Blythe, Joan Bybee, Morten H. Christiansen, William Croft, Nick C. Ellis, John Holland, Jinyun Ke, Diane LarsenFreeman and Tom Schoenemann). 2009. Language is a complex adaptive system: A position paper. In Ellis & Larsen-Freeman, 1–26. Toivonen, Riitta; Jukka-Pekka Onnela; Jari Saramäki; Jörkki Hyvönen; and Kimmo Kaski. 2006. Model for social networks. Physica A 371, 851–60. Travers, Jeffrey, and Stanley Milgram. 1969. An experimental study of the Small World problem. Sociometry 32, 425–43.

Language Choice in a Multilingual Society

217

Tsunoda, Tasaku. 2005. Language endangerment and language revitalisation. Berlin/New York: Mouton de Gruyter. UNESCO. 2003. Language vitality and language endangerment. UNESCO (Ad Hoc Expert Group on Endangered Languages). Available online at www.unesco.org/ culture/ich/doc/src/00120-EN.pdf, accessed 7 July 2014. Vazquez, Federico; Xavier Castelló; and Maxi San Miguel. 2010. Agent-based models of language competition: Macroscopic descriptions and order-disorder transitions. Journal of Statistical Mechanics: Theory and Experiment, P04007. Wang, William S.-Y., and James W. Minett. 2005. The invasion of language: emergence, change and death. Trends in Ecology and Evolution 20(5), 263–9. Watts, Duncan. 2003. Six degrees: The science of a connected age. New York: Norton. Watts, Duncan. 2011. Everything Is Obvious: How Common Sense Fails Us. New York: Crown Business. Watts, Duncan J., and Steven H. Strogatz. 1998. Collective dynamics of ‘small-world’ networks. Nature 393(6684), 409–10. Weinreich, Uriel. 1953. Languages in contact. Findings and Problems. New York: Linguistic Circle. Wölck, Wolfgang. 2004. Universals of language maintenance, shift and change. Collegium Antropologicum 28(1), 5–12. Wu, Ye; Changsong Zhou; Jinghua Xiao; Jürgen Kurths; and Hans Joachim Schellnhuber. 2010 Evidence for a bimodal distribution in human communication. Proceedings National Academy of Sciences (USA), 107, 18803–8.

9

Complexity and Language Contact: A Socio-Cognitive Framework Albert Bastardas-Boada

1

Introduction

Throughout most of the twentieth century, analytical and reductionist approaches have dominated in biological, social, and humanistic sciences, including linguistics and communication. We generally believed we could account for fundamental phenomena in invoking basic elemental units. Although the amount of knowledge generated was certainly impressive, we have also seen limitations of this approach. Discovering the sound formants of human languages, for example, has allowed us to know vital aspects of the ‘material’ plane of verbal codes, but it tells us little about significant aspects of their social functions. I firmly believe, therefore, that alongside a linguistics that looks ‘inward’ there should also be a linguistics that looks ‘outward’, or even one that is constructed ‘from the outside’, a linguistics that I refer to elsewhere as ‘holistic’ (Bastardas 1995), though it could be identified by a different name. My current vision is to promote simultaneously the perspective that goes from the part to the whole and goes from the whole to the parts, that is, both from the top down and from the bottom up (see Bastardas 2013). This goal is shared with other disciplines which recognise that many phenomena related to life are interwoven, self-organising, emergent and processual. Thus, we need to re-examine how we have conceived of reality, both the way we have looked at it and the images we have used to talk about it. Several approaches now grouped under the label of complexity have been elaborated towards this objective of finding new concepts and ways of thinking that better fit the complex organisation of facts and events. 1.1

Complexity/Complexities

The use of the term complexity in science poses serious difficulties if we do not first clarify the sense in which we are using it. The reason is that this label has been taken up by a variety of disciplines and schools of thought. Thus, it has been attached to perspectives, phenomena and aspects of reality that do not dovetail very well with one another, which creates a good deal of confusion that must be dispelled at the outset. 218

Complexity and Language Contact: A Socio-Cognitive Framework

219

The most frequent initial confusion is the use of the term complexity to refer not to a given scientific approach, but to an intrinsic quality of many of the phenomena of reality. In layman’s language, complexity or complex is customarily used to signify complication, confusion, intricacy, diversification or a large number of units and rules in play. For instance, we speak of the complexity of a country or of a society, and the complexity of the human body or brain, or of a specific language. Complexity, in these sorts of uses, does not refer so much to a particular way of conceiving reality as to a feature of specific phenomena of the world that we wish to understand. In linguistics, the application of the term complexity has largely focused on the structural and grammatical features of human languages, particularly the comparative study of their grammatical systems. In other words, the focus has been on the existing diversity in the number of units used and their characteristics and combinations at different levels of verbal codes as well as how they change and develop over time (see, for example, Sampson et al. 2009; Nichols 2009; Emmert-Streib F. 2010; McWorther 2011). As previously observed, from a scientific point of view, and starting from different fields and distinct lines of research, several authors have been constructing a paradigm that has also come to be known as the sciences (or theory) of complexity (e.g. Morin 1980, 1991, 1992, 1994, 1999, 2001; Wagensberg 1985; Gell-Mann 1996; Cilliers 1998; Gershenson & Heylighen 2005; Castellani & Hafferty 2009; Mitchell 2009; Jörg 2011). This is only one and perhaps the most appropriate name from among others currently available, including cybernetics (Wiener 1948; Ashby 1956; Bateson 1972; Heylighen 2001), systemics (Von Bertalanffy 1981; Capra 1982), ecology (Margalef 1991), chaosology (Flos 1995; Bernárdez 1995), autopoiesis or self-organisation (Maturana & Varela 1996; Maturana 2002; Solé & Bascompte 2006), emergentism (Holland 1998; Johnson 2002), and networks science (Newman et al. 2006; Solé 2009). In this approach, the terms complexity and complex are used to refer in general to a characteristic typical of a great number of elements and phenomena of reality: an organisation comprised of interwoven units that give rise to new and emergent levels of organisation and (inter)actions, with properties and capabilities that are distinct from those of the initial constituent elements. With this type of phenomena, there is a high degree of awareness that any application of a more or less mechanistic picture is fraught with difficulty, because of the vast number of dynamic inter-retro-actions produced, which prompt the emergence of new organisational levels that have different functions and, at the same time, integrate functionally with other coexisting phenomena. However, even within the theory or science of complexity approach itself, there are important distinctions to make. On the one hand, we have a perspective that focuses more on modelling and computation. On the other, the focus is

220

Albert Bastardas-Boada

more epistemological and philosophical, that of the pensée complexe, inspired primarily by Edgar Morin. Morin (2007) called these two sets of approaches, respectively, ‘restricted’ (complexité restreinte) and ‘general’ complexity. Turning now to the more methodological and formalist approaches, the science of complexity has made important contributions connected to the potentialities of computing and to the appearance of new forms of mathematical reasoning that better suit complex and dynamic phenomena with a high degree of interactivity and mutual emergent feedback. In recent decades, physicists, mathematicians, computer scientists and some biologists and sociologists, principally, have been the driving force in important lines of research and thought devoted to studying the formal properties, potentialities and characteristics of ‘complex systems’. This ‘synthetic’ method – as it has been called by Luc Steels (1995) to distinguish it from inductive and deductive methods – offers us something different from what we have seen so far: a chance to understand the genesis and development of phenomena (see e.g. Abrams & Strogatz 2003; Ball 2005; Newman et al. 2006). This method involves simulating complex processes with agent-based programs that model behaviours generally governed by simple rules that are themselves usually based on ‘if stimulus, then response’ formulations (cf. Wolfram 1983, 2002; Epstein 2006; Castelló et al. 2011). Such ‘complex adaptive systems’ (CAS)1 can also learn from their relationships with the context in order to adapt better (by more adequately making use of the environment in order to take full advantage of it for their own purposes). 1.2

Agents and Models in Language Contact

Some researchers have already started to apply the computational tools to language contact and bilingualism (e.g., Loureiro et al., chapter 8; Castelló 2010; Castelló et al. 2013). The model typically uses squares in a computer screen to depict agents governed by simple rules of conduct they apply in accordance with any other types of agents with which they come into contact. After a given number of iterations, the result at the level of language will be the greater or lesser use of one or another of the coexistent languages. For instance, if one of the groups of agents is more predisposed to use its second language to speak with members of the other group than to use its first language, we can see on screen how such a situation will evolve. One of the contributions of these types of methodologies is that they enable us to clearly visualise the bottom-up phenomenon of sociocultural organisation that emerges from interactions among agents (Holland 1998). We can clearly 1

Although the term has grown in popularity due to the efforts of the Santa Fe Institute, it should be remembered that the sociologist Walter Buckley was using it as early as 1967.

Complexity and Language Contact: A Socio-Cognitive Framework

221

see how, based on the application of a few stable rules, social behaviours can shape a society’s customs and other cultural aspects. The complex concept of self-organisation (Ashby 1962; Solé & Bascompte 2006) is highly useful for understanding historical developments, including the evolution of language. This is also true for the evolution of networks (Solé 2009), which enable us to account for cultural variations that can be produced in accordance with the configuration of the web of relations that individuals maintain with one another. Thus, this type of modelling lets us literally see, on screen, in silico, the process of reduction in the use of a language, showing agents who are applying rules that they do not suspect will potentially generate negative consequences over time and ultimately the practical extinction of what was once their first language. The strategy of simulation, therefore, is a productive one, surpassing classical statistical tools to the extent that it enables us to control the parameters of an evolving situation and what emerges out of that situation.2 As for the second set of approaches, ‘general’ complexity (Morin 1992, 1994, 2008) is more heavily committed to an epistemological, multidimensional, integrated and dynamic view of reality: the world is constituted by the ‘emergent’ overlap of different elements that produce new properties or organisations as they complexify at higher levels. And this may go on from initial physical and genetic elements all the way up to human societies and cultures. It postulates that in order to gain an adequate understanding of the interwoven fabric of all these domains in motion, we need to go beyond a way of thinking that tends to separate the elements of reality and treat them in isolation. It pushes us beyond reductionist thinking that prioritises the elementary units and quantitative aspects of phenomena. It calls for us to think in terms of ‘both/and’ and not ‘either/or’, applying fuzzy logic (Munné 2013) rather than Aristotelian logic. It demands that we sidestep the pitfalls of dichotomies, and it builds from the fact that complex thinking is not the ‘opposite’ of simple thinking, but rather incorporates it3 (Bastardas 2002). In linguistics, there are also contributions that have arisen between these two major positions, building on the viewpoint of ‘complex adaptive systems’ from the Santa Fe Institute but seeking to go beyond computational modelling, 2

3

Not only simulations, but also programmes of this type using real data have been run to validate a theoretical model. One example is the use of cellular automata to examine the processes of language shift in Spain in a study devised by the group led by Francesc S. Beltran, using data from the autonomous community of Valencia (2009 and 2011). The model assumes social pressure – the number of people in the neighbourhood who encourage one behaviour or another – to be one of the fundamental variables in the evolution of the sociolinguistic situation, and this allows us to view the evolution of intergenerational language transmission. It must be conceded that the use of the terms complex and complexity in the vast majority of publications appearing in English – the most widespread language of science – corresponds much more to ‘restricted’ complexity than to the more general perspective. For instance, the activity of researchers at the Santa Fe Institute (Gell-Mann, Holland, etc.) has been immense and extremely interesting. At present, this approach is also seeing a generous crop of developments in Europe.

222

Albert Bastardas-Boada

toward interesting theoretical propositions like the ones put forward by Holland (2005), The ‘Five Graces’ Group (2009) or Massip-Bonet (2013). Similarly, the ecology-inspired contributions of Mufwene (2001, 2008, 2013) take a broad perspective consistent with the approaches of general complexity. In economics, for example, the work of Brian W. Arthur (2013) takes a very open view also. The general-complexity perspective aims not so much to devise precise theories about specific phenomena as it sets out to take a comprehensive view of reality, a view that is holistic and, at the same time, mindful of the autonomous parts. From this viewpoint, in the words of Castellani and Hafferty: Social complexity theory is more a conceptual framework than a traditional theory. Traditional theories, particularly scientific ones, try to explain things. They provide concepts and causal connections (particularly when mathematicised) that offer insight into social phenomenon ( …) Scientific frameworks, by contrast, are less interested in explanation. They provide researchers effective ways to organize the world; logical structures to arrange their topics of study; scaffolds to assemble the models they construct (2009:34).

Certainly, the two perspectives of complexity go together, but they correspond to distinct levels and emphases, and they need to be complementary and integrated. Recently, authors have taken this task in hand and are able to offer their reflections to us. This is the case, for example, with Malaina (2012), Rodríguez Zoya (2013), Roggero (2013), Ruiz Ballesteros (2013), Solana (2013), and Byrne & Callaghan (2014), who deeply take both traditions into consideration, integrating and evaluating them. Although computational methods and strategies are useful to illuminate how human situations evolve, including situations of language contact, it would be hard to go so far as to say that restricted complexity needed to be the foundation on which to build a broad and complex vision of sociocultural reality. In any event, it should be the other way round. Our perspective ought to be built on the epistemological and representational foundations of general complexity, and we need to use the methods, tools and concepts of restricted complexity on the phenomena and processes that they can best shed light on.4 Based on general complexity, my approach explores the world in a way that helps us understand how language contact phenomena unfold. What follows is my personal synthesis of the main principles of the complexity perspectives in contrast with the more traditional scientific ones. The concepts listed in the following two columns are not necessarily opposites. 4

“[W]e do not consider that mathematical representations of the social represent some ideal towards which social science should be aiming. ( …) Mathematics can be a useful tool for describing the reality but reality is its messy self, not a higher abstract order existing in mathematical form. ( …) When we approach the complex social we need methods which can take accounts of context, agency and temporality” (Byrne & Gallaghan 2014:257).

Complexity and Language Contact: A Socio-Cognitive Framework

223

Traditional Scientific Perspectives conceptual reification

General Complexity Perspectives there is no science without an observer (centrality of brain/mind)

territory

maps (we see by means of concepts and words)

scientific truth

provisional theories

elements

elements-and-contexts, interweaving, figurations, interdependences, networks

objects

events and processes

steady-state

dynamic flux, change, evolution, development

classical logic

fuzzy logic

linear causality

circular, retroactive and nonlinear causality, recursivity

either/or dichotomies

both/and; integration and complementarity

top-down and planned creation

bottom-up, self-organisation and emergence

unidimensionality

inter-influential multidimensionality

‘explicate order’ (things are unfolded and each thing lies only in its own particular region of space)

‘implicate order’ (everything is folded into everything; a hologram: the parts contain information on the entire object)

fragmentation of disciplines

inter- and transdisciplinarity

structure, code

meaningful and emotional interaction

Figure 9.1 Main principles of the complexity perspectives in contrast with the more traditional scientific ones.

2

A Socio-Cognitive Complexity Perspective on Language Contact5

In principle, it is not easy to apply the perspective of general complexity to understanding the co-determinants and evolution of language behaviours in situations of intense language contact. Quite often, the study of these cases draws on a sociolinguistic tradition that focuses more on fragmentary aspects – for example, bilingual competences, code-switching, identities, policies, and so on – than on a comprehensive view that is dynamic and transdisciplinary. By contrast, our proposal aims to provide the basis of an integrative focus, from a perspective of human socio-complexity which draws on the contributions of traditional approaches to the study of language systems, but goes beyond them. One possible way to envisage an integrated study of the complexity of evolving conditions in situations of language contact is to conceive of them from an 5

In this chapter, I will not go into the specific effects of contact over and above habitual language forms, such as the phenomenon of interference, language borrowing or change.

224

Albert Bastardas-Boada

ecological point of view (Haugen 1972; Mackey 1979, 1980, 1994). Margalef (1991:80) calls an ecosystem: a level of reference formed by individuals together with the materials that are produced by their activity ( …) and the matrix or physical surroundings in which they are included and in which they act.

The assumption of the ecosystem concept is that the fate of a particular linguistic variety – that is, its survival, its alteration, or its extinction – depends basically on the evolution of the sociocultural factors that are involved in its production. Its structure, then, is basically governed by the social functions that it is required to perform. This is, in fact, a characteristic of the complex adaptive systems. As Holland states, “the context in which a persistent emergent pattern is embedded determines its function” (1998: 226). Consequently, this approach sees the relation between languages and linguistic groups as a three-way (rather than a two-way) phenomenon. In conceiving the relation between two species, for example, the ecological perspective bears in mind at all times the milieu in which the relation develops. This perspective is vitally important to understand the impact of migration on language contact, or the integration of a politically minoritised group in a state, as it underscores the need to take into account the structure of the broad sociopolitical environment as well as the groups in question. Another principle on which this approach is based is that the different orders and phenomena of the reality make up an interrelated whole, in which there are not only circular, mutual influences between two variables but also a set of dynamic interactions that make up the reality, as explained later. Thus mental, interactional, collective, economic, political, and linguistic phenomena coexist in such a way that one constitutes the other and vice versa. To express the image, I use the metaphor of the musical score which enables us to visualise different planes of the same unitary phenomenon and which exists sequentially, that is, in time. The static image of reality is also challenged. Contrary to the traditional approach, time is an essential, continually present variable. Apparent stability is always the result of a dynamic equilibrium that allows the conservation of the identity of the units even if their elements are changed. More than as a structure, reality should be seen as a set of events, or, to quote Bohm, as a “universal flux of events and processes” (1987:31). From this perspective the fragmentation into disciplines is also questioned. As reality is multidimensional, an inter- and transdisciplinary focus is necessary, especially in the sociocultural sciences. The new conceptual landscapes must then allow the integration of perspectives of the different approaches in a global theorisation which considers simultaneously all the necessary levels of human beings in an integrated, coordinated way (see Capra 1982, 1996,

Complexity and Language Contact: A Socio-Cognitive Framework

225

2002). In fact, the inspiration for this approach comes not only from outside but also from inside the sociocultural sciences. According to Bastide (1971:8), Auguste Comte himself seems to have suggested as much in an era that lacked the conceptual instruments required for an ecological or complex approach: ‘In the natural sciences the elements exist before the whole; in the human sciences, the whole exists before the parts’ (contemporary physicists make the same claim for quantum physics). Twentieth-century authors, such as Gregory Bateson (1972), Norbert Elias (1982, 1990, 1991), Kurt Lewin (1978), and Walter Buckley (1967, 1968), declare their support for an approach of this kind, albeit from different angles. The report of the Gulbenkian Commission on the reorganisation of the social sciences (1996), chaired by Immanuel Wallerstein, clearly pointed in the same direction, as well as all the works by Edgar Morin. Uriel Weinreich would agree: It is in a broad psychological and sociocultural setting that language contact can best be understood. ( …) On an interdisciplinary basis research into language contact achieves increased depth and validity (Weinreich 1968: 4).

Noteworthy is also the fact that the orchestral complexity metaphor enables us to understand and to organise in separate and yet interrelated ways the dimensions which are most relevant to determining the behaviour of humans in situations of linguistic diversity and contact. So, as an exploratory example, we will construct a pentagram for each of the voices or instruments without forgetting their interrelation with the other pentagrams among which relations of harmonic interdependence arise.6 For the moment, and in a brief and simplified image, our score will comprise the following emergent and superposing basic parts: the minds, social interaction, human groups, and political power (Bastardas 1996). Language varieties ‘live’ and interact with these dimensions. 3

The Co-Environment of Linguistic Varieties

3.1

The Brain/Mind Complex

As John Holland says, “If we are to understand the interactions of a large number of agents, we must first be able to describe the capabilities of individual agents” (1996:7). Consequently, little can be understood about human 6

In the orchestral score one can see the evolutions of each of the instruments or voices and of the whole that results from the superimposition of one on the other in the interpretive sequence of the work. This is no more than applying the vision of systems, where each level forms part of a whole of multiple interrelated levels, the cooperation of all of which produces the emergence of a specific behaviour or global product able to be perceived by and to influence another human being.

226

Albert Bastardas-Boada

behaviour if we do not begin our examination of the metaphorical score by looking at the ‘brain/mind complex’, since it is at this level that the ultimate control over human action and understanding lies. Let us, therefore, take the human being as a bio-social product endowed with a brain/mind that will enable and regulate the individual’s relationship with the world7 (see Maturana & Varela 1996). The usual development of the neurocognitive complex occurs – and out of necessity must occur – in close interrelation with the sociocultural context, in other words, through interaction with other human predecessors and their products. Without this requisite activation, during the optimum phase, of the genetic programming by the stimuli of social activity, no brain can develop properly or have any chance of recovery during its lifetime. In all probability, the interrelationship between the developing brain/mind and the sociocultural phenomenon can be seen more accurately in terms of self-organising systems rather than in terms of the computer metaphor, with its traditional inputs and outputs (Varela et al. 1992:157). The fruit of this functional autonomy is that the brain/mind will construct itself from the perceived cultural artefacts derived from the social interactions, so that, as Morin suggests, it is a complex phenomenon formed from inseparable elements – the brain/mind, the individual and the society/culture – each of which, in its own way, contains the others (1986:84). This approach also helps to advance our anthropological understanding, given that our cultural knowledge is to be found – as it would appear to be in reality – at the interface of these elements rather than in one or in all of them (Varela et al. 1992:178). In socialisation, the structure of the sociocultural contexts represents a factor of considerable influence on the final results of this process. If someone comes across different ways of speaking in the set of contexts in which he or she lives, the degree and quality of development in each of these linguistic varieties might differ according to the type and the intensity of exposure or use. The variety or varieties used in the family setting – in particular that of the mother or person(s) who spend(s) the most time with the child in the first year of his or her life – will supply the initial elements for the development of comprehension and expression. It/They will tend to become the code or codes that will form the base for the conceptual structuring of reality, for the development of the collective identity of the individual and – if the social context does not impede it – for the informal linguistic communication ability of the person. If the remaining social contexts – neighbours, networks of friends, pupils at school, teachers, the mass media, and so on – confirm the way of speaking the individual has acquired in the family – and/or at the kindergarten, which 7

Unlike the physical sciences, at the level of human phenomena it is not only the mind of the observer that we should take into account, but the minds of the subjects of the observation as well. We should take account of the mind not merely because of its intrinsic importance, but because it is inside the mind that the great majority of the courses of action of humans are determined.

Complexity and Language Contact: A Socio-Cognitive Framework

227

these days acts in part as a substitute for these contexts – the individual will gradually expand his competence in this code. He will acquire the registers corresponding to the different functions and situations and then will use them as a matter of course in his daily communicative acts with no problem whatsoever. If, on the other hand, the individual finds other distinct varieties outside the initial context of socialisation, then he will be presented with a problem of a different nature with complex causes and consequences. Thus, if, for example, the variety developed in the family setting is neither spoken nor understood in all the other contexts, the individual will be obliged to acquire as quickly as possible the other way(s) of speaking and will become bi- or multilingual – or bi- or polylectal if they are varieties of the same language.8 In this type of situation the individual might finally develop a greater degree of competence in his second variety than in his first – in particular, if exposure to the latter occurs during the critical period and, if, in addition to the informal contexts, the new variety is also that of the formal public contexts – the medium of formal education, of the general street signs, of the media, and so on. In terms of understanding linguistic behaviour, there are two main interrelated functions of the brain/mind complex that would appear to be of particular relevance: the representation of reality and control over behaviour. It is in the brain/mind complex where we construct and sustain ideas about the reality that we experience, and from where we activate our motor organs to carry out specific actions – determined in accordance with the discursive representations and interpretations of the reality that we make (Van Dijk 2010). And this we can do, as we shall see, either from the conscience or the ‘subconscience’. We can hold certain definitions of reality without being conscious of so doing, and similarly we can undertake certain actions without having been conscious before, or at the time, of having done so. The conscience, therefore, does not exhaust the mind. Many of our mental acts are not directly accessible from the conscience.9 8

9

Individuals might also find two (or more) systems of linguistic communication within the family domain. The most typical case is that which occurs where each progenitor uses a different language to address the child. In such situations, and if the person-language norm is consistent, the child should be able to develop more or less equally the bases of two mother tongues (Weinreich 1968:77), and will use them appropriately according to the situation (Fantini 1982:63). In fact, as Popper and Lorenz (1992:30) pointed out, learning includes the effort of consigning what one has just learnt to the subconscience. Thus, a large part of our behavioural and cognitive activity is subconscious. The high degree of consciousness that we maintain over each action when learning to drive a car becomes part of a routine and our subconscience when we have some experience and we wish to centre our attention on the road. We must conclude, therefore, that the phenomenon also affects linguistic behaviour and all other human activities. Indeed, Bateson believes that “the conscience must always be limited to a rather small fraction of mental process. ( …) The unconsciousness associated with habit is an economy both of thought and of consciousness; and the same is true of the inaccessibility of the processes of perception. The conscious organism does not require (for pragmatic purposes) to know how it perceives – only to know what it perceives” (1972:136).

228

Albert Bastardas-Boada

At the heart of this conception of the human being as a cognitive-interpretive being is, as maintained by the perspective of symbolic interactionism, the view that “the meaning does not emanate from the intrinsic makeup of the thing that has meaning but rather from and through the defining activities of the people as they interact” (Blumer 1969:4). Things do not have meaning on their own; rather, it is human beings that attribute meaning to things, be they physical objects, words or actions, through the cognitive processing of apprehended information and internalised interpretive procedures. Facing any perception and, frequently, from our subconscience, the world is processed and understood drawing on the available cognitive depository. Any perception that cannot be recognised and interpreted from the knowledge available at that time will activate the conscience in order to produce a hypothesis that makes sense, that might explain what it is that we are perceiving, how it relates with our other perceptions, what function it performs, for whom, and the like. As Schutz said, I cannot understand a cultural object without referring it to the human activity in which it originates. For example, I do not understand a tool without knowing the purpose for which it was designed, a sign or symbol without knowing what it stands for in the mind of the person who uses it, an institution, without understanding what it means for the individuals who orient their behaviour with regard to its existence (1974:41).

While human beings develop direct referential interpretive capacities in relation to the linguistic structures perceived in their social interactions, they also develop social evaluative interpretations of these same linguistic structures, in particular in situations of diversity of ways of talking. Therefore, speaking in one or another variety, or using each other’s linguistic form, might be socially significant and have major repercussions on the interaction that develops. Just as we can assign meanings to the social status of the clothes that we wear, the linguistic varieties used can also be associated with specific social meanings. When we interpret our perceptions, we do so polyphonically, multidimensionally. Virtually never do we consider one level of meaning in isolation; we integrate within our pertinent perceptions or information, and, what is more, from within a hierarchical organisation. The social meanings of the linguistic forms are part of the individual’s cognitive-interpretive stock and can thus influence the action both of the potential user and the interlocutor. The latter might, for example, not offer room to someone who speaks in a way that is considered negative socially. Therefore, the individual that is perceived negatively might decide, for example, to abandon the linguistic variety that is disadvantageous, perhaps especially when it is also derided. From out of this context of competences, meanings, habitus and unconscious routines, the individual will choose specific language forms to use in communication with other people. As we can see, for example, individuals in

Complexity and Language Contact: A Socio-Cognitive Framework

229

situations of social bilingualism will use one code or another in line with the representation they have of their interlocutor, the available competences (of the two individuals), the social meanings of the varieties to hand, their previously established custom, and their cognitive interpretation of the context in which the interaction takes place. The sum total of behavioural decisions taken by the individuals of a given society will mark the evolving dynamic of the situation. 3.2

The Level of the Social Interaction of Brains/Minds

Individual brains/minds are also, at the same time, the building blocks of a higher level of complexity, which emerges naturally from the properties of its constituents but adds new features that are typical and characteristic of the new interactional dimension and of the social situations in which this dimension is produced. Thus, at this level, there are all the elements that we have identified as belonging to the brain/mind complex as well as all those that arise from the need for organisation at the level of interaction. Social interaction must be viewed as an inextricably socio-mental relation and therefore cognitive and interpretive in nature. If, in normal circumstances, human action is given meaning by a subject who can actively form interpretations through observation and perception, then any interaction is necessarily mutually significant. Actions, movements, gestures, verbalisations, paraverbal elements, the language forms and varieties that are being used, the situations in which these occur, the biographical precedents of the relation, expected intentions, and other factors will be a constant source of conscious or unconscious interpretations processed holistically between interacting individuals (Serrano 1993). Understanding the social organisation of interpersonal relations becomes crucial to understanding enormously important aspects of language behaviour. In many cases, human interactions are well organised, quite often predictable within given limits, and meaningful. Daily encounters tend to unfold according to socially established norms and rituals that coordinate social life so that we do not have to improvise behaviour each time we come into contact with another human being and so that we can adequately interpret our interaction. Though self-organisation constantly applies in daily interactions, it is not a simple fact or at least a fact easy to describe in detail. The entire complex of behaviours will be interpreted holistically by the interlocutors in terms of the instructions of ‘scripts’ for various social settings, which determine the extent to which the behaviours fit socially habitual expectations. If the ‘script’ for a given ceremony calls for a given level of formality of apparel, for example, a person wearing clothing categorised by the social majority as ‘informal’ may be judged negatively. This negative assessment, however, will be attenuated or even changed if other significant aspects of the

230

Albert Bastardas-Boada

same person are valued positively, such as his or her way of talking, gestures or accessories, or if the person has a convincing reason for dressing in this way. Once adopted and established sociocognitively, the most daily, regular and repetitive norms of action generally become subconscious and are followed routinely, almost automatically in most cases (Nisbet 1977). Such actions include how to greet someone, what to say and what to do when departing a place, how to structure a conversation, and also what language or variety to use when speaking with a particular person. A relation between individuals typically tends to establish fixed patterns of behaviour between the two participants. For example, if at some time we adopt the custom of kissing at each meeting and we repeat this behaviour for a certain number of days, it is highly likely to become a norm and, therefore, an expectation that must be satisfied at each meeting in the future. Similarly, once we have, by mutual agreement, adopted a given language behaviour and more or less confirmed it by periodic repetition, the selection of the variety or language becomes subconscious and routine; and it will tend to be perpetuated. Indeed, at some point, changing the variety or language will become extremely difficult. In social situations involving language diversity, selecting the variety or language to use is not a simple action. An initial factor that can influence this selection is the language competence of the individuals involved. If two individuals can only understand and speak one variety, they will use it in all likelihood. However, if they also understand and speak another one, the choice is more complex. They may choose to use either one. That is, the two individuals are likely to negotiate the variety to be used, because it is common in communicative relations to prefer the use of one and only one variety by both individuals, provided that their mastery is sufficient to make this possible (Hamers & Blanc 2000; Hamers 2004). Presumably, if there is a discrepancy, the negotiation is won by the interlocutor who is more persuasive. If each interlocutor remains firm in his own position, the ensuing interactions will involve the two interlocutors speaking different languages – what has been called ‘bilingual conversation’ – or the interlocutors will tend to avoid interactions so as not to reproduce the conflict each time they communicate. Although language behaviour tends to be decided subconsciously and routine, the possibility always exists to bring it back to the conscious level and control it directly and reflectively by the individual, overcoming the constraints of competence and habitual behaviour, if desired (Bastardas 1995). Of course, this also entails social consequences, negative and positive, that may arise from the individual’s decision. For example, individuals may decide that it is better for them to change their manner of speaking with specific people or in general, using individual words or constructions, a different variety of the same language or a different language, rather than speaking as they always have.

Complexity and Language Contact: A Socio-Cognitive Framework

231

This fact lies behind many of the ways in which sociolinguistic situations evolve. As we shall see next, factors belonging to higher levels of organisation (e.g., the group or political level) exert an influence at the interactional level as a result of the complex interdependence across the various domains of the sociolinguistic ecosystem. Thus, an individual’s language behaviour can be affected not only by elements at the interactional level such as the necessary organisation of the use of varieties in conversation, but also by elements such as the issue of identity in relation to the language groups in contact or the regulation of public uses implemented by the political authorities. Both domains can influence the language choices made by individuals in their interactions, in accordance with their representations of reality. 3.3

The Social Group Dimension

3.3.1 The Emergence of Networks and Groups Not only are human societies organised by interactional pairs but also their members interact with a greater number of people with whom they co-construct in complex ways a new level of reality, a ‘groupness’ that shapes many of the aspects of what we call ‘culture’ as opposed to ‘nature’. These individuals-in-society, as Norbert Elias would say, take advantage of the potentialities of the brain to build networks and together consolidate forms of sociocultural and communicative organisation that will adapt to the changing nature of their collective historical experiences. The group, therefore, will usually be the basic unit of survival and social control, setting standards that will constrain individual and collective action. In the group, Morin’s idea of recursiveness is at work; that is, the individual makes up the group that makes up the individual. Once it has formed, a group tends to persist if it is a functional organisation and it benefits individual members. Thus, cause and product maintain and change one another. The existence of groups and limited networks of intense interaction gives rise to the possibility of cultural diversity. Each network can autonomously create representations and establish forms and norms of conduct that differ to varying degrees from those adopted by other collectives. More specifically, groups differ in the degree of importance they give to specific elements of daily life, in the behaviours they deem appropriate in various situations, in the language forms they use or prefer to use, and so on. Cultural and linguistic diversity is a real, well-established fact. Sociocultural categorisation plays a significant role in decisions that individuals make to take joint action. As members of socially, economically and culturally stratified urban societies, we always interpret others as members of some social group or category that is the same as or different from our own, with the ensuing normative and evaluative associations, and we make decisions

232

Albert Bastardas-Boada

about our actions based on such associations. In ethno-linguistic conflicts it is crucial to distinguish situations characterised by political subordination from conflicts produced by mere territorial co-existence. In the first case, the conflict is typically based on the expansionism of a demographically or militarily stronger group into a neighbouring territory inhabited historically by other collectives with different cultures. In the second case, groups cohabit regularly in the same territory and some discord arises from some reason within or outside the groups. Both cases can generate a high level of awareness of ethnic identity within the collectives in conflict, and this awareness can have an effect on any possible inter-group behaviours, because ethnic identity, according to Barth, “is similar to sex and rank, in that it constrains the incumbent in all his activities, not only in some defined social situations” (1976:18). In inter-group relations, the system of linguistic communication that is used may become highly significant. It may come to act as an ethnic identifier, with the consequences that that identification entails. Similarly, the overall configuration of ethnicity and inter-group relations may have an effect at the level of language. For subordinate groups, a positive ethnic consciousness can be the reason for maintaining their language, while a negative one is the most common cause for them to abandon it. The differences, for example, in the evolution of the sociolinguistic situation of Catalonia and the Valencian Community, in Spain, are probably based on that factor, which itself is due to other elements, like differences in the history of their economic development (Ninyoles 1978; Aracil 1982). Many ethnolinguistic groups that have been incorporated in larger political units, often forced by historical events, are typically structured according to the environment of another ethnolinguistic group which can control and ‘patrimonialise’ the political power by virtue of its demography or some other strength. These groups wind up accepting or rejecting their involvement at the level of identity in a larger body they are part of, on which they depend economically and politically, and yet which they perceive as alien to their own self-defined cultural characteristics. A more or less significant segment of the population may move toward full acceptance of the superordinate identity, accepting it as basic, while the original ethnic identity comes to be viewed as secondary. At the other extreme, another segment may see the matter in the opposite manner: native cultural traits are fundamental and primary; given that these traits are typically in decline in the face of traits associated with the polity in which the ethnic group is found, the cultural forms represented by the state and the state itself lose legitimacy and are rejected. A third position may also arise in which individuals contrive a combination of identities and find greater compatibility between the group categories in conflict. We can see all these evolutions in the current cases of language contact in Spain (Siguan 1993; Coller 2006; Strubell & Boix-Fuster 2011).

Complexity and Language Contact: A Socio-Cognitive Framework

233

3.3.2 Groups and the Macrosocial Order The dimension of power, like that of social category, is constantly present in social life and is an absolutely central element in the changing fortunes of groups. Typically, human groups do not have identical controls over economic, informational, demographic, or political resources. An awareness of the differences often comes with inter-group comparisons. A habitual way of speaking that is associated with a particular group of high socioeconomic status, for example, may be admired or rejected according to how it is perceived by other groups it coexists with. Certain ways of speaking among the upper classes can be ridiculed by members of other social groups and vice versa. In other situations, members of social groups at the lower end of the social scale can have self-negating representations that spur in them the desire to adopt the language traits of economically higher social groups in order to raise their own personal prestige and improve their self-image. Indeed, as Pierre Bourdieu (1980, 1984) points out very effectively, social positions are closely related to the predispositions – the representational and behavioural habitus – of the individuals who occupy those positions. This fact can often be the cause of more or less large-scale shifts in the language behaviour of individuals aiming to emulate the language forms that are more closely identified across the entirety of the society as belonging to affluent, socially dominant groups. In today’s information society, the greater or lesser possession of cultural or symbolic goods (e.g. academic training, skills in the arts or high technology or specialised knowledge of diverse types) is also a factor of social differentiation that is not purely economic in nature. Intellectual elites can also constitute social groups that are perceived and evaluated as having prestige by a large part of the population, particularly by the middle and working classes. As a result, they can have a significant influence on the evaluation and use of specific language forms. Through institutionalised roles occupied by virtue of political power or social structure, they convey to the population whose forms shall be considered legitimate and valuable in public discourse. A minority population in a subordinate position can easily shift cultures, a phenomenon that has occurred frequently in all periods of history. 3.4

Political Power

We have seen how human beings construct new sociocognitive realities basically in a self-organised manner. They hold representations based on facts of the world, reaching a consensus on the forms of their interactions and forming cultural and economic groupings. The culmination of contemporary developed societies is their organisation into states, which are not based specifically on phenomena of self-organisation, but rather on an explicit, formal act of establishment. At this level, particular individuals – sometimes elected

234

Albert Bastardas-Boada

democratically and sometimes not – are able to exercise significant power over many aspects of social life. It is a fact that the contemporary state intervenes more than ever in the life of human communities and exercises an enormous influence. The level of language not only fails to escape its influence but rather, contrary to what one may think, can be highly controlled and determined by political power. Specifically in the area of language, the impact of political power is both direct and indirect. Because the state can require the compulsory fulfilment of its provisions, the explicit or implicit declaration of an ‘official language’ will result in the codification of the selected variety or varieties, assuming they were not previously codified. It will also extend their knowledge and use to public functions across the entire territory where they are named. Typically, even without explicit regulation, the variety selected as an official language will also tend to be adopted in the remainder of public communication that is not dependent on the state government. Quite often, it will be the only variety that citizens consciously and reflectively learn and the only one readily available for them to use in formal speaking and writing. As a result, it will de facto become the language variety that can be used comprehensively in institutionalised communications (Corbeil 1983) within the area over which the state exercises sovereignty. It may even be used in private writing to a great extent. Moreover, as we shall see, the linguistic characteristics of the official variety may eventually be adopted even in informal spoken communication, particularly in cities, where the process of urbanisation also entails both the destruction of local sociocultural ecosystems and the need to adopt new norms of communication in the complex urban environment. In phenomena of language contact, the state can be particularly critical to how situations evolve, because the linguistic pressures brought to bear by the regulation of the official status of languages can play a crucial role in the maintenance or abandonment of the varieties in contact. When a state with a multiethnic population identifies itself solely with one of its ethnolinguistic groups, the situation is a source of potential conflicts between the state and the dominant ethnolinguistic group, on the one hand, and smaller ethnolinguistic groups, on the other hand, like in Spain during General Franco’s rule. The political will of the state to unify its citizens linguistically can crash head on into communities that often prefer to stick to their own language varieties, varieties whose structures may differ sharply from those of the official language. Such communities, which may have a historical awareness of collective differentiation, may not be ready to accept a homogenising policy. If the history also includes forced annexation, economic or religious differentiation and a policy of national uniformity pursued at the expense of the languages and cultures of smaller groups, the conflict can be prolonged and acute. Under these conditions, the politicisation of the ethnolinguistic reality is inevitable, because the state is the

Complexity and Language Contact: A Socio-Cognitive Framework

235

instrument needed by an ethnolinguistic group wishing to become a ‘nation’ or even simply to protect itself from assimilation. The state’s political subordination of some of its constituent populations is the fundamental cause of many processes of language shift, which are nothing more, in such cases, than the displacement of the traditional language varieties of smaller groups toward disuse, privileging the official standard varieties sponsored by the state. 4

Ecosystem Dynamics

So far we have drawn the lines of the orchestral or polyphonic score that can be used metaphorically to develop a dynamic view of the processes of language contact from the perspective of complexity. This means that we will understand the evolving outcome of these types of processes much better if we see them as a consequence of the mutual influences of the various lines in the score, that is, of the conflict or reconciliation of different pressures present in the sociocultural ecosystem (Terborg & García-Landa 2013). Thus, for example, if we want to gain a holistic understanding of the development of sociolinguistic situations such as in Catalonia, we should simultaneously consider all levels: their complex interdependences and the changes occurring over time. We can clearly observe interrelations among the distinct domains, for example, in the process of restoring the Catalan language undertaken by the government of Catalonia. While the public school system seeks to expand knowledge of Catalan (and Spanish) among a population in which one segment has Catalan as L1, another segment has Spanish as L1, and a further growing segment has a mother tongue other than these two, the degree of success in increasing the interpersonal use of the autochthonous language is relative. The reason is that there are complex factors at play that have their own, hard-to-change dynamics. On the interactional level, for example, there is a very widespread habit among L1 Catalan speakers to accommodate L1 Spanish speakers and not vice versa – a behaviour that is also being taken up by a large proportion of the younger generations and becoming readily automatic and unconscious. What is occurring at this level would seem to contradict what is happening at the group level, where there is a growing awareness of Catalan identity, at least in the group of L1 Catalan speakers. Contrary to what might be expected, however, this is not carrying over to the level of language behaviour adopted with L1 Spanish speakers who have already become bilingual – at least within the school context – in Catalan. The brain’s mechanisms of memory and routine formation appear to play such a fundamental role that once a language choice for speaking with a person has been made, the choice is typically repeated in new encounters, even throughout an entire lifetime and sometimes in contexts that were considered inappropriate. One example of this

236

Albert Bastardas-Boada

can be seen in partners of mixed linguistic origins who get to know each other in Spanish and continue using Spanish with one another throughout their lives, even though they speak Catalan to their children. The opposite may also occur, as we more often see in the Valencian case, when partners speak Catalan to each other, but speak Spanish to their children (Querol 1990; Conill 2003; Montoya & Mas 2012). A complex, multidimensional perspective enables us to grasp much better why there are some processes in which state policies prevail and populations adopt the code that has been declared official, while there are other cases in which this does not occur so easily. Apart from the previously cited elements of routinisation and the subconscious, emotional collective identifications and cognitive self-images also come into play in the acceptance or rejection of a state language. If a language group views and experiences a state’s language policy as illegitimate, for example, because its own language is not given official recognition, it can tend to reject the policy and prefer not to use the state language. On the other hand, if the group sees an opportunity in the state language to make social and economic progress and it has negative representations of its own code, it will tend to adopt this language much more readily. The emphasis placed by the perspective of complexity on the time and processuality of phenomena is also highly useful in grasping how different situations evolve. In the case of Catalonia, for example, it is evident that current difficulties in further expanding the use of Catalan have their roots in the path dependence on earlier events. As the Franco dictatorship achieved a high level of bilingualisation in Spanish among L1 Catalan speakers in a context in which the public use of Catalan was banned, Catalan speakers largely grew accustomed to using Spanish with L1 Spanish speakers coming to reside in Catalonia, particularly in the city of Barcelona. Now, the established habit is deeply ingrained and much more difficult to change, despite explicit campaigns with coherent arguments about the danger posed for the future of Catalan. In addition, a complex, systemic view can more adequately explain why the group of L1 Spanish speakers increases their interpersonal use of Catalan less, given the fact that the earlier adaptation of L1 Catalan speakers leaves them no opportunities to use Catalan, despite government efforts (Boix-Fuster & Farràs 2013). The biological replacement of populations through generational change is also of major importance in how processes of contact evolve. New individuals who replace those who came before can be socialised in a context distinct from their parents and cognitively adopt new representations that prompt them to shift to the prevailing behaviours. As complex adaptive systems, the younger generation may already be more habituated to contact with another group than their own parents, for example, and they may develop the language skills needed in a situation of contact earlier on, at a time that is more biocognitively

Complexity and Language Contact: A Socio-Cognitive Framework

237

optimal. Economic contexts, too, can evolve and come to exert influence on the social meanings and attitudes ascribed to language varieties, which can in turn lead to the adoption of new behaviours. Viewing these issues from a complex perspective is not only useful but necessary. Consequently, it is of special importance to stress that language contact must be understood as a historical and, therefore, temporal phenomenon, with earlier events playing a major role in how the phenomenon evolves. In other words, we need to pay attention not only to the synchronic elements, but also to the diachronic ones, because the latter may determine the future development of the phenomenon (Elias 1982). 5

Conclusion

We have seen some examples of how restricted and general complexity perspectives can help understand the interwoven mechanisms of sociolinguistic dynamics at work in cases of contact. Given its holistic view and its mindfulness of its own parts, the suggested image of the orchestra that we have used as a metaphor can enable us to integrate the micro and macro (i.e., respectively, inter-individual and institutional) dimensions present in human experience and reflect the mutual interdependencies of the participating levels. What this can encompass ranges widely from genetic to sociopolitical constraints, and simultaneously, it takes into consideration the presence of time as an inescapable context for the emergence and existence of language forms and varieties. From the general complexity perspective we must emphasise the cognitive and emotional uniqueness of human agents and not lose sight of the importance of this factor in their (inter)actions. In addition, the human social organisation itself makes linguistic/communicative activity ever present in a variety of areas and specific contexts that also exert a reverse influence on the linguistic level. The human linguistic phenomenon is at one and the same time an individual, social and political fact. As such, its study should bear in mind these complex interrelations, produced inside the framework of the sociocultural and historical ecosystem of each human community. The complexity of these interrelationships, their multidimensionality and the numerous factors that can affect their evolutionary dynamics will not make it easy to apply tools in the field of linguistics that are also valid for other, less complex phenomena. For example, the methods and concepts of restricted complexity can be used as supplementary strategies that are highly useful in studying certain characteristics, the stages and speeds of processes of language contact, but always within the frame of the broader view offered by general complexity. Methodologically, this use of computational instruments specific to restricted complexity ideally needs to be accompanied by the combined use of

238

Albert Bastardas-Boada

qualitative strategies, because without such strategies we cannot gain access to the sociocultural meanings that individuals confer to their language forms and to the representations by which they experience the events of their lives (Mead 1934; Bruner 1990). From the perspective of general complexity, conducting comparative case studies is an interesting strategy in this regard and it also complements approaches that are based more on statistics and formal modelling (Byrne & Callaghan, 2014). Complexity aims to overcome fragmentary methodological views. It postulates their harmonious integration in the service of attaining the deepest possible understanding of phenomena as a whole. As stated by Byrne and Callaghan: [W]e see complexity as providing a framing for the unifying of a whole set of opposites in scientific practice, of quantitative and qualitative research, of analysis and holism as modes of understanding, and of relativism and hard realism as epistemological position (Byrne & Callaghan, 2014:255).

In order to develop paradigms, the two major complexity approaches need to find more common ground and take steps toward a mutual integration based on the acceptance of the shortcomings of each approach, achieving progress through a non-contradictory complementarity of perspectives (Heylighen, Cilliers & Gershenson 2007; Bastardas 2014). It must be conceded that the practical and methodological applications of basic complexity ideas need to be developed much farther in order to apply them to specific research. As Roggero has noted: [t]oday, there are more experts in formal disciplines taking an interest in the social sphere than there are sociologists borrowing the techniques of the formal disciplines. If a meeting of minds takes place, it will turn out to be hugely beneficial for both groups. The first will need to learn sociology’s language and ways of thinking, including the sociological culture; the second will have to contend with the formal rigour, the methodological demands and the utilisation of useful computer tools found in the formal disciplines (2013:116).

At the same time, the limits of complex adaptive systems as computational strategies must be accepted in the pursuit of a better understanding of the evolutionary processes typical of human beings. In the final analysis, models always have a narrative running behind them that reflects the attempts of a human being to understand the world, and models are always interpreted on that basis. This is precisely what Allen and Hoekstra have recognised in the field of ecology: Narratives are the bottom line in science. Yes, there are hypotheses, predictions, theories and models, but all of these devices are in the service of achieving compelling narratives. ( …) The end product of science is a story improved by models and made convincing by predictions (2015: 310).

Complexity and Language Contact: A Socio-Cognitive Framework

239

REFERENCES Abrams, Daniel M. & Steven H. Strogatz. 2003. Modelling the dynamics of language death. Nature 424, 900. Allen, Timothy F. H. & Thomas W. Hoekstra. 2015. Toward a unified ecology, 2nd edn. New York: Columbia University Press. Aracil, Lluís V. 1982. Papers de sociolingüística [Sociolinguistic papers]. Barcelona: La Magrana. Arthur, W. Brian. 2013. Complexity economics: A different framework for economic thought. Available at www.santafe.edu/media/workingpapers/13-04-012.pdf. Ashby, W. Ross. 1956. An introduction to cybernetics. London: Chapman & Hall. Ashby, W. Ross. 1962. Principles of the self-organizing system. In Heinz Von Foerster & George W., Jr. Zopf (eds.), Principles of self-organization: Transactions of the University of Illinois Symposium, 255–278. London: Pergamon Press. Ball, Philip. 2005. Critical mass: how one thing leads to another. London: Random House, Arrow Books. Barth, Fredrik (comp.). 1976. Los grupos étnicos y sus fronteras. México: Fondo de Cultura Económica. Bastardas-Boada, Albert. 1995. Language management and language behavior change: policies and social persistence. Catalan Review 9 (2), 15–38. Bastardas-Boada, Albert. 1995b. Cap a un enfocament holístic per a la lingüística [Towards a holistic approach for linguistics]. In Rosa Artigas et alii (eds.), El significat textual [Textual meaning], 15–20. Barcelona: Generalitat de Catalunya, Departament de Cultura. Bastardas-Boada, Albert. 1996. Ecologia de les llengües. Medi, contacte i dinàmica sociolingüística [Ecology of languages. Sociolinguistic environment, contact and dynamics]. Barcelona: Proa. Bastardas-Boada, Albert. 2002. World Language Policy in the Era of Globalization: Diversity and Intercommunication from the Perspective of ‘Complexity’, Noves SL. Revista de Sociolingüística, 1–9. Bastardas-Boada, Albert. 2013. General linguistics and communication sciences: sociocomplexity as an integrative perspective. In Massip-Bonet & Bastardas-Boada (eds.), Complexity perspectives on language, communication and society. Berlin: Springer, 151–173. Bastardas-Boada, Albert. 2014. Towards a complex-figurational socio-linguistics: Some contributions from physics, ecology and the sciences of complexity, History of the Human Sciences 27 (3), 55–75. DOI: 10.1177/0952695114534425. Bastide, Roger. 1972. Antropología aplicada. Buenos Aires: Amorrortu. Bateson, Gregory. 1972. Steps to an ecology of mind. New York: Ballantine Books. Beltran, Francesc S., Salvador Herrando, Doris Ferreres, Marc-Antoni Adell, Violant Estrader & Marcos Ruiz-Soler. 2009. Forecasting a language shift based on cellular automata. Journal of Artificial Societies and Social Simulation 12(3), 1–8. Beltran, Francesc S., Salvador Herrando, Violant Estrader, Doris Ferreres, Marc-Antoni Adell, & Marcos Ruiz-Soler. 2011. A language shift simulation based on cellular automata. In Emmanuel G. Blanchard & Danièle Allard (eds.), Handbook of research on culturally-aware information technology: perspectives and models, 136–151. Hershey / New York: Information Science Reference.

240

Albert Bastardas-Boada

Bernárdez, Enrique. 1995. Teoría y epistemología del texto. Madrid: Cátedra. Bertalanffy, Ludwig von. 1969. General system theory. New York: George Braziller. Blumer, Herbert. 1969. Symbolic interactionism: Perspective and method. Englewood Cliffs, NJ: Prentice-Hall. Bohm, David. 1987. La totalidad y el orden implicado. [Spanish translation of Wholeness and the Implicate Order. London: Routledge & Kegan Paul, 1980]. Barcelona: Kairós. Boix-Fuster, Emili & Jaume Farràs. 2013. Is Catalan a medium-sized language community too? In F. Xavier Vila (ed.), 157–178. Bourdieu, Pierre. 1980. Le sens pratique. Paris: Éditions de Minuit. Bourdieu, Pierre. 1984. Distinction: A social critique of the judgement of taste. Cambridge: Harvard University Press. Bruner, Jerome. 1990. Acts of meaning. Cambridge, MA: Harvard University Press. Buckley, Walter. 1967. Sociology and modern systems theory. Upper Saddle River, NJ: Prentice Hall. Byrne, David & Gill Callaghan. 2014. Complexity theory and the social sciences. The state of the art. London: Routledge. Capra, Fritjof. 1982. The Turning Point. New York: Simon & Schuster. Capra, Fritjof. 1996. The web of life. A new synthesis of mind and matter. Hammersmith, London: HarperCollins, 1996. Capra, Fritjof. 2002. The hidden connections. New York: Doubleday. Castellani, Brian & Frederic William Hafferty. 2009. Sociology and complexity science. Berlin: Springer-Verlag. Castelló Llobet, Xavier. 2010. Collective phenomena in social dynamics: Consensus problems, ordering dynamics and language competition, doctoral dissertation. Palma de Mallorca: University of the Balearic Islands. Castelló, Xavier, Federico Vazquez, Víctor M. Eguíluz, Lucía Loureiro-Porto, Maxi San Miguel, Laetitia Chapel & Guillaume Deffuant. 2011. Viability and resilience in the dynamics of language competition. In Guillaume Deffuant & Nigel Gilbert (eds.), Viability and resilience of complex systems, 39–74. Heidelberg: Springer. Castelló, Xavier, Lucía Loureiro-Porto & Maxi San Miguel. 2013. Agent-based models of language competition. International Journal of the Sociology of Language 221, 21–51. Cilliers, Paul. 1998. Complexity and postmodernism. Understanding complex systems. London: Routledge. Coller, Xavier. 2006. Collective identities and failed nationalism. The case of Valencia in Spain. Pôle Sud 25, 107–136. Conill, Josep J. 2003. The situation of Valencian as reported in non-institutional sociolinguistic research (1998–2002). Noves SL. Revista de Sociolingüística, 1–9. Corbeil, Jean-Claude. 1983. Élements d’une theorie de la regulation linguistique. In Édith Bedard & Jacques Maurais (eds.), La norme linguistique, 281–303. Quebec & Paris: Conseil de la Langue Française. Elias, Norbert. 1982. Sociología fundamental. [Spanish translation of Was ist Soziologie? Munich, Juventa Verlag, 1970]. Barcelona: Gedisa. Elias, Norbert. 1990. La sociedad de los individuos [Spanish translation of Die Gesellschaft der Individuen, 1987]. Barcelona: Península. Elias, Norbert. 1991. The symbol theory. London: SAGE.

Complexity and Language Contact: A Socio-Cognitive Framework

241

Emmert-Streib, Frank. 2010. Statistic complexity: Combining Kolmogorov complexity with an ensemble approach. PLoS ONE 5(8), 1. Epstein, Joshua M. 2006. Generative social science: Studies in agent-based computational modeling. Princeton, NJ: Princeton University Press. Fantini, Alvino E. 1982. La adquisición del lenguaje en un niño bilingüe, Barcelona: Herder. Flos, Jordi (ed.). 1995. Ordre i caos en ecología [Order and chaos in ecology]. Barcelona: Publicacions de la Universitat de Barcelona. Gell-Mann, Murray. 1994. The Quark and the Jaguar: Adventures in the Simple and the Complex. New York: Henry Holt and Company. Hamers, Josiane F. & Michel H. A. Blanc. 2000. Bilinguality and bilingualism. Cambridge: Cambridge University Press. Hamers, Josiane. F. 2004. A sociocognitive model of bilingual development. Journal of Language and Social Psychology 23 (1), 70–98, DOI: 10.1177/ 0261927X03261615. Gershenson, Carlos & Francis Heylighen. 2005. How can we think the complex? In Richardson, K. (ed.). Managing organizational complexity: Philosophy, theory and application, vol. 3, 47–62. Charlotte, NC: Information Age Publishing. Haugen, Einar. 1972. The Ecology of Language. In: Anwar S. Dil (ed.), The Ecology of Language, 325–339. Stanford: Stanford University Press. Heylighen, Francis. 2001. The science of self-organization and adaptivity. In L. Douglas Kiel (ed.), Knowledge management, organizational intelligence and learning, and complexity. Encyclopedia of life supports systems. Oxford: Eolss Publishers. Heylighen, Francis, Paul Cilliers & Carlos Gershenson. 2007. Complexity and philosophy. In Jan Bogg & Robert Geyher (eds.), Complexity, science and society, 117– 134. Oxford: Radcliffe Publications. Holland, John H. 1996. Hidden order. How adaptation builds complexity. New York: Basic Books. Holland, John H. 1998. Emergence. From chaos to order. Cambridge, MA: Perseus Publishing. Holland, John H. 2005. Language acquisition as a complex adaptive System. In James W. Minett & William S-Y. Wang (eds.), Language acquisition, change, and emergence, 411–436. Hong Kong: City University Hong Kong Press. Johnson, Steven. 2002. Emergence. London: Penguin Books. Jörg, Ton. 2011. New thinking in complexity for the social sciences and humanities. Heidelberg: Springer. Mackey, William F. 1979. Toward an ecology of language contact. In William F. Mackey & Jacob Ornstein (eds.), Sociolinguistic studies in language contact, 453–460. La Haya: Mouton. Mackey, William F. 1980. The ecology of language shift. In Peter Hans Nelde (ed.), Sprachkontakt und Sprachkonflikt, 35–41. Wiesbaden: Franz Steiner Verlag. Mackey, William F. 1994. La ecología de las sociedades plurilingües. In Albert Bastardas Boada & Emili Boix-Fuster (eds.), ?‘Un estado, una lengua? La organización política de la diversidad lingüística, 25–54. Barcelona: Octaedro. Malaina, Álvaro. 2012. Le paradigme de la complexité et la sociologie. Possibilité et limites d’une sociologie complexe. Paris: L’Harmattan. Margalef, Ramon. 1991. Teoría de los sistemas ecológicos. Barcelona: Publicacions de la Universitat de Barcelona.

242

Albert Bastardas-Boada

Massip-Bonet, Àngels. 2013. Language as a complex adaptive system: Towards an integrative lingüístics. In Massip-Bonet & Bastardas-Boada (eds.), 35–60. Massip-Bonet, Àngels & Albert Bastardas-Boada (eds.). 2013. Complexity perspectives on language, communication and society. Heidelberg: Springer. Maturana, Humberto. 2002. Autopoiesis, structural coupling and cognition. Cybernetics & Human Knowing 9(3–4), 5–34. Maturana, Humberto & Francisco J. Varela. 1996. El árbol del conocimiento. Las bases biológicas del conocimiento humano. Madrid: Editorial Debate. McWhorter, John H. 2011. Linguistic simplicity and complexity. Why do languages undress? Berlin: De Gruyter Mouton. Mead, George H. 1934. Mind, self and society (Charles W. Morris, ed.). Chicago: The University of Chicago Press. Mitchell, Melanie. 2009. Complexity: A guided tour. Oxford: Oxford University Press. Montoya Abat, Brauli, & Antoni Mas i Miralles. 2012. Language teaching and family linguistic transmission: Two correlative factors in the Valencian Region (Catalan vs. Spanish)? Sociolinguistic Studies 6(1), 45–63. Morin, Edgar. 1980. La méthode. 2. La vie de la vie. Paris: Éditions du Seuil. Morin, Edgar. 1986. La méthode. 3. La connaissance de la connaissance. París: Éditions du Seuil. Morin, Edgar. 1991. La méthode. 4. Les idées. Leur habitat, leur vie, leurs moeurs, leur organisation. Paris: Éditions du Seuil. Morin, Edgar. 1992. Introduction à la pensée complexe. Paris: ESF. Morin, Edgar. 1994. La complexité humaine. Paris: Flammarion. Morin, Edgar. 1999. La tête bien faite. Repenser la réforme, réformer la pensée. París: Éditions du Seuil. Morin, Edgar. 2001. La méthode. 5. L’humanité de l’humanite. L’identité humaine. Paris: Éditions du Seuil. Morin, Edgar. 2007. Restricted complexity, general complexity. (Translated from French by Carlos Gershenson.) In Carlos Gershenson, Diederik Aerts, & Bruce Edmonds (eds.), World views, science and us. Philosophy and complexity, 5–29. Singapore: World Scientific. Morin, Edgar. 2008. On complexity. Advances in systems theory, complexity, and the human sciences. New York: Hampton Press. Mufwene, Salikoko S. 2001. The ecology of language evolution. Cambridge: Cambridge University Press. Mufwene, Salikoko S. 2008. Language evolution. Contact, competition and change. New York: Continuum International. Mufwene, Salikoko S. 2013. The emergence of complexity in language: An evolutionary perspective. In Massip-Bonet & Bastardas-Boada (eds.), 197–218. Heidelberg: Springer. Munné, Frederic. 2013. The fuzzy complexity of language. In Massip-Bonet & Bastardas-Boada (eds.), 175–196. Newman, Marc, Albert-László Barabási & Duncan J. Watts. 2006. The structure and dynamics of networks. Princeton, NJ: Princeton University Press. Nichols, Johanna. 2009. Linguistic complexity: A comprehensive definition and survey. In Geoffrey Sampson, David Gil & Peter Trudgill (eds.), Language complexity as an evolving variable, 110–125. Oxford: Oxford University Press.

Complexity and Language Contact: A Socio-Cognitive Framework

243

Ninyoles, Rafael L. 1978. Conflicte linguistic valencià [Valencian linguistic conflict]. València: Eliseu Climent ed. Nisbet, Robert. 1977. The social bond. New York: Random House. Popper, Karl & Konrad Lorenz. 1992. El porvenir está abierto. Barcelona: Tusquets Ed. (Spanish translation of Die Zukunft ist offen. Munic: E. Piper GmbH, 1985.) Querol, Ernest. 1990. El procés de substitució lingüística: la comarca dels Ports com a exemple [Language shift process: Els Ports county as an example]. Miscel·lània 89, 87–196. València: Generalitat Valenciana. Rodríguez Zoya, Leonardo G. 2013. Le modèle épistémologique de la pensée complexe. Analyse critique de la construction de la connaissance en systèmes complexes. Unpublished Ph. D. Thesis, Université de Toulouse. Roggero, Pascal. 2013. Para una sociología según El método. In Ruiz Ballesteros & Solana Ruiz (eds.), 103–123. Ruiz Ballesteros, Esteban. 2013. Hacia la operativización de la complejidad en ciencias sociales. In Ruiz Ballesteros & Solana Ruiz (eds), 137–172. Ruiz Ballesteros, Esteban & José Luis Solana Ruiz (eds.). 2013. Complejidad y ciencias sociales. Seville: Universidad Internacional de Andalucía. Sampson, Geoffrey, David Gil & Peter Trudgill (eds.). 2009. Language complexity as an evolving variable. Oxford: Oxford University Press. Schutz, Alfred. 1974. El problema de la realidad social [Spanish translation of Collected papers: I. The problem of social reality. The Hague: Martinus Nijhoff, 1962]. Buenos Aires: Amorrortu. Schutz, Alfred & Thomas Luckmann. 1977. Las estructuras del mundo de la vida. Buenos Aires: Amorrortu. (Spanish translation of The Structures of the Life-World. Evanston, IL: Northwestern University Press, 1973). Searle, John R. 1985. Mentes, cerebros y ciencia. Madrid: Ed. Cátedra. (Spanish translation of Minds, Brains and Science. The 1984 Reith Lectures, Cambridge, MA: Harvard University Press.) Serrano, Sebastià. 1993. Comunicació, societat i llenguatge [Communication, society and language]. Barcelona: Empúries. Siguan, Miquel. 1993. Multilingual Spain. Amsterdam: Swets-Zeitlinger. Simon, Herbert A. 1962. The architecture of complexity. Proceedings of the American Philosophical Society 106 (6), 467–482. Solana Ruiz, José Luis. 2013. El concepto de complejidad y su constelación semántica. In Esteban Ruiz Ballesteros & José Luis Solana Ruiz (eds.), Complejidad y ciencias sociales, 19–101. Seville: Universidad Internacional de Andalucía. Solé, Ricard. 2009. Redes complejas. Del genoma a internet. Barcelona: Editorial Tusquets. Solé, Ricard V. & Jordi Bascompte. 2006. Self-organization in complex systems. Princeton, NJ: Princeton University Press. Steels, Luc & Rodney Brooks (eds.). 1995. The artificial life route to artificial intelligence. New Haven, CT: Lawrence Earlbaum. Strubell, Miquel & Emili Boix-Fuster (eds.). 2011. Democratic policies for language revitalisation: the case of Catalan (Palgrave studies in minority languages and communities). Basingstoke: Palgrave Macmillan. Terborg, Roland & Laura García-Landa. 2013. The ecology of pressures: Towards a tool to analyze the complex process of language shift and maintenance. In MassipBonet & Bastardas-Boada (eds.), 219–239.

244

Albert Bastardas-Boada

The ‘Five Graces’ Group: Beckner, Clay, Richard Blythe, Joan Bybee, Morten H. Christiansen, William Croft, Nick C. Ellis, John Holland, Jinyun Ke, Diane LarsenFreeman & Tom Schoenemann. 2009. Language is a complex adaptive system. Language Learning 59, 1–26. Van Dijk, Teun V. 2010. Discourse and context. A sociocognitive approach. Cambridge: Cambridge University Press. Varela, Francisco J., Evan Thompson & Eleanor Rosch. 1992. The Embodied Mind. Cognitive Science and Human Experience. Cambridge, MA: The MIT Press. Vila, F. Xavier (ed.). 2013. Survival and development of language communities. Prospects and challenges. Clevedon, UK: Multilingual Matters. Wagensberg, Jorge. 1985. Ideas sobre la complejidad del mundo. Barcelona: Tusquets editores. Wallerstein, Immamnuel (ed.). 1996. Ouvrir les sciences sociales. Rapport de la Commission Gulbenkian. Paris: Descartes & Cie. Watzlawick, Paul, John H. Weakland & Richard Fisch. 1974. Change. Principles of Problem Formation and Problem Resolution. New York: Norton. Weinreich, Uriel. 1968. Languages in contact. The Hague: Mouton. Wiener, Norbert. 1948. Cybernetics or control and communication in the animal and the machine. Cambridge & Massachusetts: The MIT Press. Wolfram, Stephen. 1983. Statistical mechanics of cellular automata. Reviews of Modern Physics 55(3), 601–644. Wolfram, Stephen. 2002. A new kind of science. Champaign: Wolfram Media.

Index

Abrams, Daniel M., 190, 196–212, 220 acquisition language, 8, 12, 23, 34, 55, 91, 110, 124–125, 136, 160, 165–183, 212 speech, 23, 165–183 actional basis, 107, 126, 130 Allen, Timothy F. H., 238 Anderson, Philip, 188 ape, 74–88, 91, 93, 104, 128 Apel, Willi, 119 approach analytical, 218 biological-functional, 23, 171–172 reductionist, 136, 137, 141, 160, 218, 221 Arthur, W. Brian, 222 Australian languages, 153 autopoiesis, 219 Bagley, Robert W., 113, 114 Baines, John, 112, 115 Ball, Philip, 187, 190, 194, 198, 220 Barabási, Albert-Lászlo, 192, 193–194, 219, 220 Barsalou, Lawrence W., 72, 73 Bastardas-Boada, Albert, 1, 3, 10, 13, 25 Bates, Elizabeth, 90, 91 BEAGLE method, 84–87 Beckner, Clay, 1, 10, 13, 53, 67, 93, 195, 222 Bedore, Lisa M., 166, 168 Bell, Allen, 174 Berwick, Robert C., 52, 87, 88, 89, 91 Beuls, Katrien, 18–19, 20, 36, 42, 44 bilingualisation, 236 bilingualism, 23–24, 73, 175, 196–212, 220, 223, 229, 230, 235, 236 biocultural evolutionary process, 68 biological capacity, 165, 166 Blumer, Herbert, 228 Blythe, Richard, 1, 9, 10, 13, 53, 67, 93, 195, 196, 222 Boccaletti, Stefano, 194 Bohm, David, 224

Bolhuis, Johan J., 87, 88, 89, 91 Bolinger, Dwight, 124 Bonabeau, Eric., 39, 168 Boone, Elizabeth Hill, 110 bootstrapping, 33, 90 borrowing, 59, 60, 61, 223 Bourdieu, Pierre, 233 brain network, 73, 74 size, 74, 77 temporal lobe, 71, 73, 78 Bratman, Michael, 104, 127 Brighton, Henry, 33, 53, 92 Buccellati, Giorgio, 113 Burkart, Judith Maria, 74 Bybee, Joan, 1, 10, 13, 53, 67, 93, 94, 102, 161, 195, 222 Byrne, David, 222, 238 Callaghan, Gill, 222, 238 Camazine, Scott, 39, 168 capacity cognitive, 11, 12, 13, 15, 20, 23, 82, 102, 125, 170 language, 12, 23, 51, 101, 121, 165–166, 167, 169. See also production system neural, 23, 166, 169, 172, 173 perceptual, 169, 170, 172 Capra, Fritjof, 219, 224 Cassidy, Kimberly, 90 Castelló, Xavier, 195, 196, 198–209, 211, 220 Catalonia, 232, 235–236. See also Spanish language causality, 34, 35, 73, 110, 137, 222, 223 Chafe, Wallace, 101 Changizi, Mark A., 74–77 chaos, 2, 7, 12, 14, 70, 135, 219 Chater, Nick, 52–53, 68, 94 Cheney, Dorothy L., 79, 88 chimpanzee, 21, 50, 74, 75, 77, 79, 83 Chomsky, Noam, 12, 49, 50, 51, 52, 55, 68, 87, 88, 89, 91, 138, 165, 167–168, 195

245

246

Index

Christiansen, Morten H., 1, 10, 13, 52–53, 67, 68, 93, 94, 195, 222 Clark, Andy, 170 Clark, Herbert H., 104–107 class upper, 233 working, 233 Clements, George N., 55, 58, 137, 140, 141–142, 144, 159, 160 code-switching, 205, 223. See also bilingualism cognition domain-general, 51–53, 58, 61, 62, 93 grounded, 72–73 language-specific, 49, 51–53, 58, 61, 62 social, 104–110, 121, 125, 126–131, 223–225, 230, 233 cognitive factor, 51, 53, 56, 58 combinability, 101, 111 combinatorial structure. See structure common ground, 21, 104–108, 117–122, 126–131 communication domain-specific, 102 general-purpose, 22, 102–103, 117, 122, 124 institutionalized, 234 communicative needs, 13, 14, 15, 44 pressure, 7, 10, 13, 17, 23 technology, 12 competence, 12, 13, 168, 223, 226–227, 228–229, 230 complex adaptive system, 1, 10, 14, 17, 67–68, 70, 93, 94, 195, 220, 221, 224, 236, 238 network, 24, 138, 192, 193–196, 198–201, 205–212 complexity absolute, 11 algorithmic, 5, 11 bit, 4, 16 conceptual, 67, 68, 70–71, 73, 74–78, 83–94, 102 effective, 5, 13, 138 form, 30, 31, 32, 41, 42–43 general, 219–225, 237–238 hidden, 14 interactional, 2, 6 inventory, 30, 31, 44 Kolmogorov, 5, 138 learnability, 11 learning, 30, 32–33, 40–41 linguistic, 4, 5, 6, 8, 10, 11, 12, 13, 14, 16, 20, 21, 23, 25, 48, 68, 101, 103

minimum description length, 5, 138 morphosyntactic, 102 orchestral, 225 population-level, 30, 33, 38–40 processing, 30, 32, 38 relative, 11–12, 48 restricted, 219–225, 237–238 science, 1–4, 7, 9, 11, 12, 14, 20, 21, 22–23, 166, 168–169, 170, 182, 187–188, 219–220 serial, 182 social, 190, 222 social-cognitive, 21, 101, 103 theory, 2, 18, 22, 69, 135–136, 138, 139, 160, 168, 169, 219, 222. See also complexity, science usage, 15 compositionality, 17, 92, 102 computational properties, 168 computer simulation. See simulation, computer conceptualization, 20–21, 67, 69, 70, 71–74, 75, 78–94, 102, 110, 125 constraint, 4, 7, 8, 9, 10, 11, 33, 48, 49, 50, 53–55, 69, 70, 91, 93, 136–137, 139, 146, 159, 167, 168, 182, 183, 230, 231, 232, 237 convention, 21, 44, 46, 85, 89, 93, 105–108, 110, 113, 126–127, 128, 130, 136 co-occurrence pattern, consonant-vowel, 174–178 phoneme co-occurrence, 4, 141–144, 145–155, 159–160 Cooper, Jerrold S., 113–114, 115, 124 cooperation, 3, 6, 22, 104, 225 coordination, 2, 22, 56, 57, 104–108, 122, 124, 126–131, 171 problem, 105–106 cortical area, 71, 74, 75–78 creole, 7, 9, 16, 18, 19, 35 Croft, William, 1, 9, 10, 13, 14, 20, 21–22, 53, 67, 93, 101, 102, 123, 130, 195, 196, 222 cybernetics, 22, 135, 219 Dale, Rick, 9, 161 Damasio, Antonio R., 72, 73, 78, 170 Damasio, Hanna, 72, 73, 78 Damerow, Peter, 111, 113, 114, 124 Davis, Barbara L., 23, 166, 168, 170, 172, 173, 174–178, 179, 181–182, 183 de Boer, Bart, 3, 8, 20, 53, 54, 55, 58–60, 136 De Landa, Manuel, 35

Index de Waal, Frans, 79, 88 Deacon, Terrence William, 85, 87 Deaner, Robert O., 74 DeGraff, Michel, 4, 7 Deneubourg, Jean-Louis, 39, 168 Descartes, René, 166–167 Diessel, Holger, 123, 129 Dobzhansky, Theodoszius, 49 Dorogovtsev, Sergey, 138, 194 duality of patterning, 101, 102–103, 124, 125, 130 dynamic systems theory, 170–171, 182 ecology, 3, 9, 21, 25, 219, 222, 238 language ecology, 8–9, 13, 15, 17, 196 ecosystem, 24, 189, 193, 224, 231, 234, 235, 237 Edelman, Gerald M., 170 Elias, Norbert, 225, 231, 237 Ellis, Nick C., 1, 10, 13, 53, 67, 93, 195, 222 embodiment theory, 34, 165, 169–170, 171, 172, 175, 182–183 emergence, 2, 3, 7–8, 9, 10, 12–13, 14, 15, 17–23, 34–36, 38, 50, 52, 54–62, 67–68, 70, 87–94, 104, 107–110, 112, 113, 116–117, 119, 124, 126–131, 135, 136, 139, 141, 166, 169, 172, 173, 183, 188–189, 201, 205, 219, 223, 225, 237 emergent pattern, 2, 3, 9, 93, 224 emergent phenomenon, 10, 17, 20, 23, 187, 190 emergentism, 18, 219 emotion, 70, 73, 190, 223, 236, 237 Englund, Robert, 111–112, 113–114, 124 epistemology, 11, 166, 220, 221, 222, 238 equilibrium, 3, 6, 7, 10, 135, 140, 170, 190, 224 essence, 167–168, 197 ethnic consciousness, 232 Evans, Nicholas, 51–53, 68, 139 Everett, Daniel L., 101 evolution, 2, 10, 12, 14, 18, 20, 21, 34, 36, 49, 52, 53, 54, 67, 78, 89, 92, 94, 103, 106, 108–131, 190, 221, 223, 224, 232 biological, 20–21, 50, 52–54, 62, 68 co-evolution, 53, 67, 193 cultural, 20–21, 22, 50, 51–52, 53–54, 55–62, 67, 68, 70, 89–90, 92–94, 195 language, 8, 9, 13, 18, 19, 20, 21, 23, 24, 35–36, 42, 44, 56, 61, 67, 70, 75, 78, 87, 92–94, 101, 102, 110, 122–131, 141, 165, 195, 221, 223 exaptation, 12, 13, 21–22, 32, 52, 123, 124 expansionism, 232 experimental cultural learning. See learning

247 Fauconnier, Gilles, 102 feature economy, 50, 55–56, 58, 61, 137, 141–142, 152, 156, 159, 160 maximal use of available features, 137, 152, 159 Fishman, Joshua A., 196 Fitch, W. Tecumseh, 50, 52, 53 Five Graces Group, The, 1, 10, 13, 53, 67, 93, 195, 222 frame-content theory, 175 Franks, Nigel, 39, 168 Friederici, Angela D., 87, 88, 89, 91 fuzzy logic, 221, 223 Gahl, Susanne, 91 Galantucci, Bruno, 35, 50, 56–58, 62 Ganis, Giorgio, 73 Garcia Casademont, Emilia, 38 García-Landa, Laura, 235 Gardner, Beatrix T., 79 Gardner, R. Allen, 79 Gell-Mann, Murray, 5, 13, 138, 219, 221 Gershenson, Carlos, 2, 219, 238 Gibson, Kathleen R., 74 Gildea, Patricia M., 84 Gleitman, Lila, 90–91 Goldberg, Adele E., 48, 91 Gong, Tao, 3, 92, 138, 191, 194, 195–196, 197, 211, 212 Goodman, Judith C., 90, 91 grammatical system, 18, 21, 30, 31, 32, 36, 41, 88, 89–90, 92, 93, 94, 101, 116, 122, 124, 125, 130, 219 grammaticalization, 8, 19, 22, 41, 123 Granovetter, Mark S., 192, 193, 195 Greenfield, Patricia M., 91 Grenoble, Lenore, 196 Grice, Paul, 105 Griffiths, Thomas L., 51, 56, 58 group minoritized, 224, 233 social, 25, 48, 50, 56, 78, 79, 81, 110, 126, 128–130, 193, 195, 203, 204, 205, 220, 224, 225, 231–237 Guest, Ann Hutchinson, 120–121 Gullah, 8 habitus, 228, 233 Haiman, John, 88, 102 Halle, Morris, 54–55, 138, 165, 167–168 Hawkins, John A., 1, 13–14, 16, 19, 31 Heine, Bernd, 8, 35 helpfulness, 108, 127, 128 Heylighen, Francis, 219, 238

248

Index

Hockett, Charles F., 22, 101 Hoekstra, Thomas W., 238 holism, 90, 137, 218, 222, 235, 237, 238 Holland, John, 1, 10, 13, 53, 67, 93, 168, 195, 219, 220, 221, 222, 224, 225 hologram, 223 homunculus, 77 Hooper, Joan B., 174 Houston, Stephen D., 115 human infant, 13, 23, 56, 110, 165–183 icon, 21, 85, 102, 111, 113–117, 119, 120–121, 123, 125, 129 iconic gesture, 123, 129 iconicity, 102, 115, 123 identity collective, 226 ethnic, 203, 232 idiolect, 10, 11, 19, 22, 33, 136 index, 85–87, 114, 116, 123, 125, 129 interaction between components, 12, 16 constructivist, 171 functional, 182 social, 10, 20, 21, 24, 103–104, 108, 128, 169, 192, 225, 226, 228, 229–231 interdisciplinarity, 3, 209, 225 interpretive procedure, 228–230 intersyllabic property, 177–178 invisible hand, 2 Isler, Karin, 74 Jackendoff, Ray, 68, 70, 89 jaw oscillation, 175, 176, 178, 183 Jerison, Harry J., 74 Johnson, Mark, 73, 169 joint action, 21, 22, 103–108, 110, 117, 121, 122, 125–131, 231 attention, 106, 110, 126, 127, 130 salience, 106–108, 126–127, 129 Jones, Michael N., 84 Kanzi, 79–87, 88, 91 Kauffman, Stuart, 168 Ke, Jinyun, 1, 3, 7, 10, 13, 53, 67, 93, 138, 191, 194, 195–196, 197, 211, 212, 222 Kern, Sophie, 173, 174, 176, 178 Kilmer, Anne Draffkorn, 117 Kinney, Ashlyn, 174 Kirby, Simon, 3, 33, 50, 53–54, 56–59, 62, 92, 102 Kuteva, Tanja, 8, 35 Lakoff, George, 73, 169

Langacker, Ronald W., 88, 102 language as technology, 14 choice, 23–24, 187–188, 189, 191, 194, 196–212, 231, 235 coexistence, 24, 196–212. See also bilingualism competition, 9, 12, 23–24, 195–212 death, 196–212, 224 development, 7, 9, 15, 17, 18, 20, 21, 23, 68, 88, 90, 91, 92, 94, 108, 110, 111, 112, 116, 119, 129, 166, 169, 170, 181, 182, 190, 220, 223, 226, 232, 235, 237 diversity, 33, 51, 139–140, 167, 225, 230, 231 dynamics, 4, 8, 18, 24, 33, 36, 67, 94, 171, 195–196, 201, 205, 209, 211–212, 237 extinction, 196–212, 221, 224 game, 34–38, 56 maintenance, 196–212, 234 minoritized, 203–204, 207–209 official, 234–236 policy, 25, 223, 234, 236 survival, 24, 187, 190, 196–212, 224 universal, 1, 45, 48, 51–52, 53, 68, 88–89, 93, 135, 167, 168, 174, 178 variation, 1, 10, 14, 19, 20, 22, 31, 33, 38–40, 44, 62, 136 variety, 7, 9, 19, 21, 209, 224, 226–227, 228–229, 230–231, 234–235, 237 Lass, Roger, 123 lateral inhibition, 33, 38–40, 44 learning efficiency, 41, 42 experimental cultural, 50, 54, 56–62 iterated, 56–62, 92, 102 Lewis, David, 105–106 lexigram, 79 Lieven, Elena V. M., 124 Liljencrants, Johan, 55, 152 Lindblom, Björn, 55, 135, 136, 152, 170, 174 linear generative phonology, 167 Lupyan, Gary, 9, 161 Mackey, William F., 224 MacNeilage, Peter, 12, 135, 170, 172, 174–178, 181–182, 183 Maddieson, Ian, 6, 48, 136, 138, 140, 153, 173, 174, 175, 181 Malaina, Álvaro, 222 Margalef, Ramon, 219, 224 Marshack, Alexander, 111 Massip-Bonet, Àngels, 1, 3, 10, 13, 222 mathematics, 22, 121, 194, 222

Index Maturana, Humberto, 219, 226 Matyear, Christine, 174, 175, 181–182 maximal perceptual distinctiveness, 55, 170, 183 McWhorter, John H., 1, 7, 16 Mead, George H., 238 Menninger, Karl, 112 Milroy, Lesley, 194 Minett, James W., 7, 13, 196–198, 210, 211 minimal production difficulty, 170 model, 8, 11, 22, 34, 35, 36, 37, 49, 51, 55, 56, 60, 61, 68, 69, 70, 74, 76, 88, 91, 92, 94, 102, 110, 136, 141, 159, 160, 168, 170, 171, 187, 190, 191, 192, 193, 195, 196, 197, 198, 199, 203, 206, 207, 210, 211, 212, 220, 221, 222, 238 Abrams-Strogatz, 196–212 agent-based, 13, 18, 33–46, 55, 92–94, 136, 141, 191–193, 195–212, 220–221 computational, 35, 56–57, 94, 101, 168, 187, 192, 195–212, 220–221 modeling, 3, 8, 13, 18, 23–24, 33–46, 56, 131, 136, 170, 187–212, 219, 220–221, 238 module, 23, 169, 194 language, 1, 6–7, 12, 16–17, 19, 24, 168 monkey, 73, 78–79, 88 morality great ape, 128 norm-based, 128 second-person, 128–129 Morin, Edgar, 219, 220, 221, 225, 226, 231 Mufwene, Salikoko, 3, 7–10, 13–15, 17, 19, 20, 21, 22, 31, 35, 44, 50, 53, 102, 123, 130, 196, 203, 209, 210, 222 multimodality, 115, 116, 117, 124, 125 nature versus nurture, 51, 166–168 neoteny, 166 Nettle, Daniel, 191, 195–196, 203, 210 network node degree, 194, 199, 206 random, 192, 194, 205–206, 210–211 regular lattice, 198–201, 205–209, 210–211 small-world, 24, 195, 198–201, 205–209, 210, 211 neume, 118–119 staffless, 118–119 Newmeyer, Frederick J., 1, 5, 14, 19 Niger-Congo languages, 153–154 Nissen, Hans, 111, 113, 114, 124 non-linearity, 23, 139, 140, 159, 160 notation, 109, 110–124 dance, 110, 120–124 Franconian, 119

249 musical, 22, 109, 110, 117–119, 120–122 stick-figure, 120 Nowak, Andrzej, 203, 210 number, 110, 111–112, 117, 121, 123 system, 111–112 Ohala, John J., 55, 58, 137, 159 ontogeny, 8, 18, 124, 165–166, 171–173, 182–183. See also phylogeny open system, 31, 139, 170, 172 optimality theory, 138, 167 order, 2, 3, 4, 5, 7, 34, 36, 37, 48, 50, 53, 54, 59, 61, 69, 75, 79, 84, 92, 104, 105, 112, 114, 120, 128, 130, 136, 141, 142, 145, 147, 148, 151, 155, 167, 170, 174, 205, 212, 220, 221, 222, 224, 228, 233, 238 explicate, 223 implicate, 223 organisation, 5, 7, 12, 13, 50, 55, 61, 135, 136, 138, 139, 151, 155, 160, 172, 178, 188, 189, 218, 219, 220, 221, 228, 229, 231, 233, 237 Oyama, Susan, 171 pairwise interaction, 5, 141, 144, 149, 155, 159–160 perceptual basis, 126 Pettersson, John Sören, 111–112 PhonBank, 175, 179 phoneme inventory. See phonological, inventory phonological feature, 23, 54–56, 58, 61, 137, 138–139, 140–143, 147–160, 201 inventory, 7, 8, 14, 20, 23, 48–49, 135–161, 173 phylogeny, 7–8, 9, 12, 13, 15, 17–18, 21, 23, 124, 165–167, 171–172, 182–183 pidgin, 17, 18 pigeon, 78–79 political power, 25, 225, 232, 233–235 precedent, 106, 108, 126–127, 128, 129, 130 prefrontal lobe, 73, 77, 78. See also brain pre-language, 101, 103, 110, 122–125, 126, 129 prestige, 197, 198, 200, 202–204, 205–206, 210–211, 212, 233 Prigogine, Ilya, 168 Prince, Allen, 138, 165, 167, 171 processing cognitive, 14, 30, 32, 228 language, 30, 31, 35, 37, 78 neural, 71–78

250

Index

production system, 23, 169, 170, 171, 172 psychological reality, 168 recursion, 52, 101, 102, 223, 231 reductionism, 188 regular lattice, 24, 203 representation, 21, 35, 72–73, 77, 79, 84, 102, 109, 111–124, 136, 138, 167, 193, 194, 197, 211, 222, 227, 229, 231, 233, 236, 238 representational art, 109 rhythm, 117–119, 122, 172, 175, 176, 178, 183, 192 Rissanen, Jorma, 5 Roggero, Pascal, 222, 238 Romaine, Suzanne, 196 Rosch, Eleanor, 169, 226 Rudman, Peter Strom, 111, 112 Ruiz Ballesteros, Esteban, 222 rule, 4, 9, 10, 11, 14, 15, 25, 33, 58, 61, 88–90, 107, 191–192, 204, 219, 220–221 San Miguel, Maxi, 3, 6, 23–24, 188–189, 191, 192, 193, 195, 196, 197, 198–209, 211–212, 220 Saussure, Ferdinand de, 1, 7, 10, 135, 167, 195 scale, 39, 53, 109, 128, 129, 137, 140, 188, 195, 233 time, 8, 53–54, 165, 166 Schelling, Thomas, 188, 191 Schmandt-Besserat, Denise, 113 Schutz, Alfred, 228 Scott-Phillips, Thomas C., 50, 56, 58, 62, 168 selectional pressure, 102–103, 109–110, 124, 127–128 self-organisation, 2, 5, 7, 12, 39, 50, 55, 61, 135–136, 188, 189, 218, 219, 221, 223, 226, 229, 233 semasiographic system, 21–22, 103, 108–125, 129 shared expertise, 107–108 practice, 107–108, 126 signaling system, 130 simulation, 73, 93, 94, 191 computer, 34–46, 55, 56–57, 60, 92, 136, 192, 195–212, 220–221 Smith, Kenny, 33, 50, 53, 56–58, 62, 92 Smolensky, Paul, 138, 165, 167, 171 Sneyd, James, 39, 168 social context, 10, 13, 15, 21, 67, 78, 79, 84, 87, 88, 90, 93, 103, 104, 171, 172, 173, 226–227, 229, 235–237

function, 182, 195, 218, 224 institution, 93, 128 meaning, 228–229, 237 Solana Ruiz, José Luis, 222 Spanish language, 16, 175, 203, 204, 205, 235–236 speech community, 10, 103, 108, 123, 182 corpus, 84 stability, 39, 42, 52, 53, 62, 129, 170, 173–179, 183, 193, 221, 224 statistical control, 144–145 Stauder, Andréas, 115 Steels, Luc, 3, 8, 14, 18–19, 20, 31, 33–35, 36, 38, 44–45, 53, 56, 92, 136, 139, 195, 220 Stengers, Isabelle, 168 Strogatz, Steven Henry, 2, 190, 194, 196–212, 220 structure argument, 90–91 combinatorial, 58, 61 conceptual, 21, 87–88, 102–103, 121 phonotactic, 58 structuralism, 135 syntactic, 8, 30, 32–33, 90–91, 102, 105, 123 syllable, open, 174, 182 symbol, 40, 41, 57, 85–87, 102, 111, 112, 114, 118, 120, 122, 123–124, 125, 129, 168, 228, 233 manipulation, 168 symbolic notation, 114, 120 synchronisation, 2, 136 syntactic embedding, 102 tablet numerical, 113 numero-ideographic, 113, 115, 116 protocuneiform, 111, 113–114, 115 Taruskin, Richard, 118–119 Terborg, Roland, 235 Thelen, Esther, 170–171, 172, 174 Theraulaz, Guy, 39, 168 Thompson, Evan, 169, 226 Tinbergen, Nicolas, 165–166, 172, 182–183 Toivonen, Riitta, 194, 196, 205, 210, 211 Tomasello, Michael, 48, 51, 52, 90, 91, 104, 106, 110, 124, 126, 127–128, 129 transdisciplinarity, 223 Universal Grammar, 52, 68, 88–91 UPSID, 48–49, 139–160

Index

251

Valencian community, 232, 236 Van Trijp, Remi, 31, 36, 45 Varela, Francisco J., 169, 219, 226 Verhoef, Tessa, 54, 58–60 Vihman, Marilyn May, 124, 174 volatility, 197, 198, 200, 202–205, 206, 210–211, 212

Wang, William Shi-Yuan, 3, 7, 13, 77, 138, 191, 194, 195–198, 210, 211–212 Watts, Duncan, 193, 194, 212, 219, 220 Weinreich, Uriel, 9, 196, 225, 227 Wenger, Étienne, 107 West, Martin Litchfield, 117–118 Wolfram, Stephen, 220

Wagensberg, Jorge, 219

Zuidema, Willem, 3, 58, 59, 60, 136