Universal typological dependencies should be detectable in the history of language families

Abstract

1. Introduction We claim that making sense of the typological diversity of languages demands a historical/evolutionary approach.We are pleased that the target paper (Dunn et al. 2011a) has served to bring discussion of this claim into prominence, and are grateful that leading typologists have taken the time to respond (commentaries denoted by boldface). It is unfortunate though that a number of the commentaries in this issue of LT show significant misunderstandings of our paper. Donohue thinks we were out to show the stability of typological features, but that was not our target at all (although related methods can be used to do that: see, e.g., Greenhill et al. 2010a, Dediu 2011a). Plank seems to think we were arguing against universals of any type, but our target was in fact just the implicational universals of word order that have been the bread and butter of typology. He also seems to think we ignore diachrony, whereas in fact the method introduces diachrony centrally into typological reasoning, thereby potentially revolutionising typology (see Cysouw’s commentary). Levy & Daumé think we were testing for lineage-specificity, whereas that was in fact an outcome (the main finding) of our testing for correlated evolution. Dryer thinks we must account for the distribution of language types around the world, but that was not our aim: our aim was to test the causal connection between linguistic variables by taking the perspective of language evolution (diversification and change). Longobardi & Roberts seem to think we set out to extract family trees from syntactic features, but our goal was in fact to use trees based on lexical cognates and hang reconstructed syntactic states on each node of these trees, thereby reconstructing the processes of language change

Cognitive processing, language typology, and variation

“…Despite such differences, it is clear that what links processing and typological distributions is language change. This connection between diachrony and typology has been entertained since Greenberg's early work, but typologists have been able to implement it only recently to modeling typological distributions …”

“…However, genealogically balanced sampling produces mere synchronic snapshots of the type‐distributions and does not allow making diachronic interpretations from those distributions . The second approach tries to solve this problem by taking an essentially diachronic approach to universals, whereby universal are understood as structural pressure that changes languages over time (see Refs and references therein). Languages are then sampled both across and within families to assess the probability of a language shifting from one type to another.…”

“…This connection between diachrony and typology has been entertained since Greenberg's early work, but typologists have been able to implement it only recently to modeling typological distributions. 3,[17][18][19]65 In this review, I have surveyed experimental evidence for several statistical universals. As the body of relevant psycholinguistic work is increasing, the more evidence it seems to provide that typological generalizations may be reflected in cognitive processing preferences.…”

“…First, the effects of shared inheritance and horizontal transmission (language contact) may be controlled via genealogically balanced sampling, by controlling how many languages are chosen from each language family, 2,14,15 and the remaining effect is then attributed to cognitive bias. [16][17][18] However, genealogically balanced sampling produces mere synchronic snapshots of the type-distributions and does not allow making diachronic interpretations from those distributions. 17 The second approach tries to solve this problem by taking an essentially diachronic approach to universals, whereby universal are understood as structural pressure that changes languages over time (see Refs 16-21 and references therein).…”

Correlational Studies in Typological and Historical Linguistics

“…The correlated evolution method estimates the probabilities with which pairs of features change values ( Figure 3). If changes in one feature strongly tend to trigger changes in the other, we may be entitled to draw an inference of causality (for a detailed explanation and clarification, see Levinson et al 2011). We may do so because time can be inferred on the family tree, flowing from the past (the protolanguage, from which relatedness and similarity ultimately emanate) toward the future (observed languages, which tend to diverge from each other due to language change), allowing the detection of the temporal ordering of changes in feature values.…”