Recent years have seen much discussion on the use and meaning of typological argumentation when reconstructing language history and language relations. We address the conclusions and methodology of a paper "Structural phylogenetics and the reconstruction of ancient language history" (Science, Sept. 23, 2005), which claims that, on the basis of a typological comparison, the non-Austronesian languages and (Austronesian) Oceanic spoken to the immediate east of New Guinea can be shown to belong to two unrelated genetic entities. We argue that the data and discussion in this paper do not allow us to conclude that the non-Austronesian languages in the study form a valid linguistic group in any historical sense, or that the methods they apply can be used to make claims about linguistic relatedness.
1. TYPOLOGY, STATISTICS, AND HISTORICAL LINGUISTICS. In this paper we intend to provide a qualitative critique of a recent well-publicized paper dealing with the historical interaction of Oceanic languages and non-Austronesian languages spoken in the east of mainland New Guinea and the islands extending east to the middle Solomons, an area its authors call "Island Melanesia." (1) The paper, Dunn et al. (2005), makes the claim that structural ("typological') features of a group of languages can be used to determine the genetic affiliations of those languages. This work has been popularized both within and without the linguistic community, and has been heavily cited as an example of the introduction of new (with the unspoken assumption: better) methods into historical linguistics. We show that the work in this paper does not establish this claim, and, moreover, does not establish any firm factual claims concerning their study area either.
Dunn et al. (2005) apply a computational tool used in (biological) genetic classification to language data, using a dataset based on structural features of language rather than one containing comparative lexical data. (2) The authors apply this method to the question of the relationships between non-Austronesian languages in Island Melanesia. The Oceanic population arrived in this region approximately 3,500 years ago (e.g., Lynch, Ross, and Crowley 2002; Pawley forthcoming), and the relationships between these languages are therefore accessible to the methods of comparative linguistics. These relationships have indeed been studied, and the existence of an Oceanic subgroup within Austronesian is beyond question; further, there is general agreement on the subgrouping within Oceanic (Lynch, Ross, and Crowley 2002). However, it has proved impossible to establish any similar relationships between the non-Austronesian languages in the region, which share almost no lexical cognates (see, e.g., Ross, 2001). As Dunn et al. note, there is archaeological evidence for human presence in Island Melanesia going back 35,000 years; if linguistic lineages are even a third as long as that, any relationships that might have existed in the lexical data would be expected to have been lost. As an alternative, Dunn et al. use structural data to try to recover relationships. Their dataset consists of 125 binary characters (relating to phonological inventories, morphological patterns, word order patterns, and other syntactic features) coded for 31 languages, 16 Oceanic and 15 non-Austronesian. As a first step, they carved out a computational analysis of the Oceanic data, and the result they present is a tree that agrees well with the consensus picture obtained from the application of "standard" comparative linguistics methods, as mentioned previously. Dunn et al. take this as evidence that historical relationships between languages can be detected using such typological data, and that they are therefore justified in applying the same technique to the non-Austronesian data. The tree that results from this second analysis defines three groups of languages that match the geography of the region. …