New methodologies for historical linguistics?

Recent research claims that analysis of lexical cognate classes for a basic wordlist can reproduce linguistic subgroups within the Austronesian family (Gray et al. 2009). The analysis is open to question in two respects. Primarily, the lexically-based classification, primed with pre-established cognate classes of the family it seeks to emulate, fails to differentiate shared retentions from shared innovations. Secondly, languages and language families typically disperse through contiguous regions (especially in the Pacific) which means that geography or social distance should be expected crudely to match phylogeny in most cases. The reproduction fails because of local borrowing between branches not closely related to each other. For instance, when we examine disjunct distributions, cases in which the phylogeny does not match a straightforward geographic spread, we can determine which of these (phylogeny or geography) the lexical cognate approach preferentially detects. Where we find a mismatch between geography and phylogeny, Gray et al.’s approach clusters languages based on human geography (that is, social distance), not linguistic subgroup. In all cases of divergence between Gray et al.’s tree and accepted Austronesian trees, the discrepancy is a product of the former representing social distance rather than historical phylogenetic relationships. In summary, the examination of lexical cognate classes is not a valid proxy for the comparative method, though it is a useful heuristic for detecting pairs of languages that are either lexically conservative, or which show the effects of later lexical diffusion (without discriminating between these two outcomes).