Retour vers liste

Détail de la contribution

Auteur: Michel DEGRAFF

Co-Auteur(s): Trevor Bass (Litle & Co) Robert Berwick (MIT, Electrical Engineering & Computer Science)

Computational Phylogenetics, Creole Languages and Family Values

Abstract/Résumé: Creole languages have traditionally been excluded from the Comparative Method. Bakker & al 2011 & Daval-Markussen/Baker 2012 (“B/D-M&al”) apply novel phylogenetic tools (SplitsTree) that construct networks rather than trees to look for “irrefutable” evidence that: (i) Creoles are typologically distinct from non-Creoles; (ii) Creoles emerged from pidgins qua “simplified forms of interethnic makeshift languages,” thus their extreme simplicity. In contrast we show that SplitsTree cannot reconstruct evolutionary history. SplitsTree cannot reveal Creoles’ ancestral states; nor do the horizontal lines in the displayed network denote putative contact events. What the network actually does is to display potentially alternative clusters (“splits”) for languages. While such clusters seem relevant for typological classification, their reliability depends, inter alia, on statistically and empirically adequate samplings of languages, features and feature values. However B/D-M&al’s selection of features and languages is subject to severe ascertainment biases. Furthermore, their analyses do not adequately control for the fact that most of the sampled Creoles originate from contact among subsets of Germanic, Romance and Niger-Congo. Additional flaws are factual, logical, conceptual and statistical, including: (i) inaccurate descriptions: e.g., the assignment of “anti-passive” and “person marking on adpositions” to English, French, Spanish and German; the claim that only two Creoles are known to have “productive reduplication” though reduplication is common among Creoles (Kouwenberg); (ii) inaccurate generalizations: e.g., “no tense/aspect affixes” as a pan-Creole feature is contradicted by such affixes in certain Portuguese Creoles (Luís/Holm); (iii) massively interdependent features: this gives unwarranted extra weight to certain features (Kouwenberg); (iv) logical contradictions: e.g., co-occurrence of “double object marking” and “no object marking” in individual languages; (v) typological contradictions: e.g., both “no indefinite article” and “indefinite articles = one” as pan-Creole features; (vi) statistical flaws: too few features applied to too many languages inevitably creates many spurious “typologies.” Such errors undermine B/D-M&al’s claims about Creole formation. We conclude that computational phylogenetics methods must be reviewed with great care, as we do here. Then we show that Creole formation reduces to language change when phylogenetic methods are adequately applied. Caribbean Creoles, e.g., descend from European languages, with Sprachbund-like Niger-Congo influence (Aboh/DeGraff; cf. Mufwene).