Approaches: Computational Cladistics

Abstract Before turning to attempts to construct and evaluate linguistic family trees, which will be the main topic of this chapter and the next, we return briefiy to lexicostatistics, and to disturbing similarities of this method with mass comparison. In the chapters that follow we will be using elements of lexicostatistics, for instance in generating trees based on cognacy scores over standard Swadesh-type lists. We have also said that, in common with many comparative historical linguists, we regard mass comparison as inherently problematic, and as crucially compromised by a lack of rigour in determining what can count as a phonetic or semantic match. However, there remain areas of overlap between the two methods, since both are predominantly lexically based, and involve the comparison of lists. True, those lists are set in lexicostatistics, while there is no basic vocabulary criterion in mass comparison; lexicostatistics also tends to be applied step by step, in a series of pairwise comparisons between languages, whereas in mass comparison the more lists, the merrier. But, at least at first glance, these differences do not decisively favour lexicostatistics. For one thing, we made a series of suggestions in Chapter 2 for a slackening of the constraints on the composition of test lists, so that many-to-one or one-to-many matches between items and meanings might be allowed, and culture-specific meanings might be included in locally variable lists. Worse still, Greenberg (1987: 27) clearly sees the simultaneous comparison of multiple lists as strongly advantageous: ‘To inspect languages pairwise, or at a half-guess, is a different thing from a multilateral comparison undertaken with a consciousness of the types of resemblances that are likely to bespeak common origin.’ Moreover, Greenberg argues that mass comparison is free from the restrictions of time depth (see Ch. 7 below) which affect lexicostatistics and the comparative method, since ‘there is no theoretical limit to the depth at which classification can be carried out when the number of languages examined is large’ (ibid. 28—9). It would appear that mass comparison and lexicostatistics emerge as approximately equal on the issue of list composition, with mass comparison edging ahead on the number of lists to be compared.