Evaluation of cognation judgments undermines computational phylogeny of the Arawakan language family

Abstract

The goal of this paper is to critically examine the linguistic analyses underlying Walker & Ribeiro (2011), a widely cited computational phylogenetic study of the Arawakan language family. To the extent that their claims concerning the internal classification of this vast group of languages hinges on their cognation judgments, and that their more ambitious claims concerning prehistoric migration routes of Arawakan-speaking peoples depend, in turn, on this proposed internal classification, I show that outright rejection of their results is highly commendable. Errors include both false negatives, where cognation relations between comparanda were missed, as well as false positives, where non-cognate material in different languages were judged to be reflexes of single Proto-Arawakan etyma. No clear pattern seems to emerge from their cognation decisions, and the resulting judgments seem to be, in many cases, so strongly detached from even impressionistic assessments of similarity that the resulting distribution of cognation scores could have been produced independently of the data. The paper ends with a plea for greater sobriety in the historical linguistics of native South America, which should focus on clear and well-supported applications of the comparative method before embarking on endeavors that depend on this traditional work for their success. More importantly, though, South Americanists should avoid groundless statements on the supposed uselessness, or exhaustion, of the comparative method as a tool for uncovering the linguistic history of the continent.