The principal method that has been used to subgroup the Austronesian languages has been the traditional comparative-historical method, largely developed last century in connection with the comparative study of the Indo-European languages. Very briefly, this method systematically compares the regular sound correspondences between the languages compared as a first step towards reconstructing the proto-language from which the daughter languages have descended. Once the reconstruction of the proto-language has been achieved, then individual languages and sets of languages are examined to determine the innovations (phonological, morpho-syntactic and lexical) which they reflect relative to the proto-language. It is upon these innovations that subgrouping depends and proceeds.

Ross (1994) has made two very pertinent observations concerning the way in which these innovations are distributed across languages. He notes that innovations pattern across languages in two different ways reflecting two different developmental sequences. In the first, groups of languages share discrete bundles of innovations. Thus taking, for example languages A to Z, members of the family whose proto-language or putative ancestor language is *AZ, languages A-P may share one bundle of innovations not shared by languages Q-Z and vice versa. These languages would then fall into two distinct subgroups, AP and QZ. This is the distribution of innovations which results when languages have diversified by separation, that is when two or more communities speaking the same language become geographically separated. However, this is not the only way in which languages diversify. They also diversify without physical separation through dialect differentiation in their home territory. In this situation, instead of discrete bundles of innovations there are overlapping bundles which form a chain. Both of these distributional phenomena are observable in the Austronesian family, and will be seen to have important consequences for Austronesian subgrouping today. While the comparative method is a powerful tool, it has some limitations, especially with problems of recognizing contact-induced language change (see the chapter by Tom Dutton in this volume).

Another method used to subgroup languages is lexicostatistics, a method based on the replacement rate of the basic lexicon of a language over time. This method was employed by Dyen (1965) in his well-known lexicostatistical classification of the Austronesian family. While this method is useful as a first approximation, it is most useful with languages which are quite closely related. The major problem with this method, however, is that it is based on a premise that all languages replace vocabulary at a constant rate, which claim is demonstrably erroneous. For this reason the subgrouping hypotheses discussed in this paper are all based on comparative-historical techniques.