Evolved structure of language shows lineage-specifictrends

[Orakulli]

Copyright © Evolved structure of language shows lineage-specifictrends in word-order universals Michael Dunn1,2, Simon J. Greenhill3,4, Stephen C. Levinson1,2 & Russell D. Gray3 Languages vary widely but not without limit. The central goal of linguistics is to describe the diversity of human languages and explain the constraints on that diversity. Generative linguists following Chomsky have claimed that linguistic diversity must be constrained by innate parameters that are set as a child learns a language1,2. In contrast, other linguists following Greenberg have claimed that there are statistical tendencies for co-occurrence of traits reflecting universal systems biases3–5, rather than absolute constraints or parametric variation. Here we use computational phylogenetic methods to address the nature of constraints on linguistic diversity in an evolutionary framework6. First, contrary to the generative account of parameter setting, we show that the evolution of only a few word-order features of languages are strongly correlated. Second, contrary to the Greenbergian generalizations, we show that most observed functional dependencies between traits are lineage-specific rather than universal tendencies. These findings support the view that—at least with respect to word order—cultural evolution is the primary factor that determines linguistic structure, with the current state of a linguistic system shaping and constraining future states. Human language is unique amongst animal communication systems not only for its structural complexity but also for its diversity at every level of structure and meaning. There are about 7,000 extant languages, some with just a dozen contrastive sounds, others with more than 100, some with complex patterns of word formation, others with simple words only, some with the verb at the beginning of the sentence, some in the middle, and some at the end. Understanding this diversity and the systematic constraints on it is the central goal of linguistics. The generative approach to linguistic variation has held that linguistic diversity can be explained by changes in parameter settings. Each of these parameters controls a number of specific linguistic traits. For example, the setting ‘heads first’ will cause a language both to place verbs before objects (‘kick the ball’), and prepositions before nouns (‘into the goal’)1,7. According to this account, language change occurs when child learners simplify or regularize by choosing parameter settings other than those of the parental generation. Across a few generations such changes might work through a population, effecting language change across all the associated traits. Language change should therefore be relatively fast, and the traits set by one parameter must co-vary8. In contrast, the statistical approach adopted by Greenbergian linguists samples languages to find empirically co-occurring traits. These co-occurring traits are expected to be statistical tendencies attributable to universal cognitive or systems biases. Among the most robust of these tendencies are the so-called ‘‘word order universals’’3 linking the order of elements in a clause. Dryer has tested these generalizations on a worldwide sample of 625 languages and finds evidence for some of these expected linkages between word orders9. According to Dryer’s reformulation of the word-order universals, dominant verb object ordering correlates with prepositions, as well as relative clauses and genitives after the noun, whereas dominant object–verb ordering predicts postpositions, relative clauses and genitives before the noun4. One general explanation for these observations is that languages tend to be consistent (‘harmonic’) in their order of the most important element or ‘head’ of a phrase relative to its ‘complement’ or ‘modifier’3, and so if the verb is first before its object, the adposition (here preposition) precedes the noun, while if the verb is last after its object, the adposition follows the noun (a ‘postposition’). Other functionally motivated explanations emphasize consistent direction of branching within the syntactic structure of a sentence9 or information structure and processing efficiency5. To demonstrate that these correlations reflect underlying cognitive or systems biases, the languages must be sampled in a way that controls for features linked only by direct inheritance from a common ancestor10.However, efforts to obtain a statistically independent sample of languages confront several practical problems. First, our knowledge of language relationships is incomplete: specialists disagree about highlevel groupings of languages and many languages are only tentatively assigned to language families. Second, a few large language families contain the bulk of global linguistic variation, making sampling purely from unrelated languages impractical. Some balance of related, unrelated and areally distributed languages has usually been aimed for in practice11,12. The approach we adopt here controls for shared inheritance by examining correlation in the evolution of traits within well-established family trees13. Drawing on the powerful methods developed in evolutionary biology, we can then track correlated changes during the historical processes of language evolution as languages split and diversify. Large language families, a problemfor the sampling method described above, now become an essential resource, because they permit the identification of coupling between character state changes over long time periods. We selected four large language families for which quantitative phylogenies are available: Austronesian (with about 1,268 languages14 and a time depth of about 5,200 years15), Indo-European (about 449 anguages14, time depth of about 8,700 years16), Bantu (about 668 or 522 for Narrow Bantu17, time depth about 4,000 years18) and Uto-Aztecan (about 61 languages19, time-depth about 5,000 years20). Between them these language families encompass well over a third of the world’s approximately 7,000 languages. We focused our analyses on the ‘word-order universals’ because these are the most frequently cited exemplary candidates for strongly correlated linguistic features, with plausible motivations for interdependencies rooted in prominent formal and functional theories of grammar. To test the extent of functional dependencies between word-order variables, we used a Bayesian phylogenetic method implemented in the software BayesTraits21. For eight word-order features we compared correlated and uncorrelated evolutionary models. Thus, for each pair of features, we calculated the likelihood that the observed states of the characters were the result of the two features evolving independently, and compared this to the likelihood that the observed states were the result of coupled evolutionary change. This likelihood calculation was conducted over a posterior probability distribution of phylogenetic trees constructed using basic vocabulary data from each of the language families: 79 Indo-European languages16,22, 130 Austronesian languages15,23, 66 Bantu languages24 and 26 Uto-Aztecan languages (R. Ross & R.D.G., manuscript in preparation). Information on wordorder typology was derived partly from the World Atlas of Language Structure database25 and expanded with additional coding from grammatical descriptions (Supplementary Information section 1.3 and 2). As an illustration, the states of two of these features mapped against a summary of the posterior tree samples for all four language families are shown in Fig. 1. In this case, visual inspection shows that these characters appear to be linked in some families. However, the Bayesian phylogenetic approach allows us to assess this formally by quantifying the relative fits of dependent and independent models of character evolution across all trees in the posterior probability distribution. This method incorporates the uncertainty in the estimates of the tree topology, the rates of change and the branch lengths. The extent to which a dependent model of evolution provides a superior explanation of the variation of word-order features to an independent model is measured using Bayes factors (BF) calculated from the marginal likelihoods over the posterior tree distribution. BF.5 are conventionally taken as strong evidence that the dependent model is preferred over the independent model13,26. The results of the BayesTraits analysis of correlated trait evolution are summarized in Fig. 2. These differ considerably from the expectations derived from both universal approaches. The Greenbergian approach suggests robust tendencies towards linkages due to intrinsic system biases, while the generative approach assumes these will be ‘hard’ systems constraints set by discrete choices over a small innate parameter set1,27. Instead, our major finding is that, although there are linkages or dependencies between word-order characters within language families, these are largely lineage-specific, that is, they do not hold across language families in the way the two universals approaches predict. Dryer’s study of the Greenberg word-order universals4 across a world-wide sample of related and unrelated languages found a set of dependent word-order relations that show correlations with the order of verb and object, and another set of word-order relations that were independent of this. We extracted from his analyses two predictions of strong tendencies across all languages. First, all the word-order relations in the dependent set should be correlated: these are verb– object order, adposition–noun order, genitive–noun order, relativeclause– noun order. Second, no dependencies are expected between the dependent set and the independent set (including demonstrative– noun, numeral–noun, adjective–noun and subject–verb orders). 2008. All Rights Reserved.

Copyright © http://www.mpi.nl/publications/escidoc-95245/ 2008. All Rights Reserved.

[Phoenix]