Contact and Phylogeny in Island Melanesia

Due to the continual process of erosion of the lexical material in a language,the phylogenetic signal
detectable by the comparative method inevitably
disappears into noise at some point in the past. Dunn et al. (2005) presents some results of a project investigating deep time relationships between the non-Austronesian group of languages (the so-called ‘Papuan’ languages) in Island Melanesia, a group of languages which (i) could plausibly be related (especially when considering recurrent typological similarities across most of the group), but which (ii) cannot be shown to be related using standard linguistic methods. Methods from computational biology
have been adapted to address questions of language
history, treating structural features of language as
genealogically transmitted traits.
Island Melanesia is a fascinating natural laboratory for the study of language change. From 30000 to 3000 years before present these islands (the Bismarck Archipelago,
Bougainville, and the Central Solomons) formed the
furthermost limit of human dispersal into the south
Pacific. In the last 3000 years the Austronesian expansion
encompassed near Island Melanesia, and continued
eastwards into the Pacific. Currently 90% of the languages
in near Island Melanesia are members of the Oceanic
subgroup of the Austronesian family. With a few interesting
exceptions, these languages fall into reconstructable genealogical relationships using standard comparative
methods. The remaining 10% of the languages of the
region are the hitherto unrelatable Papuan remnants of the pre-Austronesian linguistic diversity. The Papuan languages are largely out of contact with each other.
The stability of individual structural features relative to the lexicon is known to be variable, yet it takes rather special conditions of prolonged contact for extensive
exchange of structure to take place. While we do not have
evidence to support a claim that language structure is
universally more conservative than lexical form based
relationships, it is plausible that in some cases it would be so, and the languages of Island Melanesia seems likely candidates. This is in any case an empirical question.
The presence of the Oceanic Austronesian languages in the region gives us a control group of languages with known phylogeny from standard methods. Structural phylogenetic methods use an independent set of data to the lexical data of the comparative method, so once a computational tool can be shown to detect a phylogenetic signal that matches the known signal of the control, it becomes scientifically interesting to apply the tool to the target group of lexically unrelatable Papuan languages. The set of relationships shown by the linguistic structural
data are geographically plausible, and have provided
hypotheses for testing in human genetics.
To date the main method we have used to investigate structural relationships between languages is maximum parsimony, which emphasises the phylogenetic component of the historical signal carried by structural traits. There is no doubt that that a process of reticulation through
borrowing and contact must also explain some of the variation. Any set of linguistic structural data will show some reticulation (in linguistic terms, homology due to contact-induced change and chance convergence). In
biological systems these phenomena are relatively rare, and the methods for investigating them are newer. The NeighborNet method implemented in the SplitsTree package (Bryant and Moulton, 2004) is emerging as the current standard (used e.g. by Bryant, Filimon and Gray 2005; Cysouw and coworkers with the WALS data at MPI EVA, Leipzig). These networks concur with the
observations of descriptive linguists that Oceanic-Papuan
ontact has had a significant influence on some groups of languages within Island Melanesia. Deeper exploration of the typological database used shows--not unexpectedly--
that traits are differentially involved in horizontal
transfer between genera, leading to the possibility of
developingdata-driven methods for statistically
compensating for the conctact signal and amplifying the signal of phylogeny.
While for a long period strict adherence to the comparative method has provided a needed rigor to counterbalance other more dramatic but ultimately unverifiable methods
(e.g. the ‘megalo-comparativists’; Matisoff 1990),
advances in areal linguistics and statistical approaches now allow us to investigate linguistic change outside the scope of the comparative method.
References
Bryant, D. & Moulton, V. 2004. NeighborNet: an
agglomerative algorithm for theconstruction of
phylogenetic networks. Molecular iology and
Evolution 21(2):255-265. Bryant, D., Filimon, F. & Gray, R. 2005. Untangling our past: Languages, Trees, Splits and Networks. In: The Evolution of Cultural Diversity: Phylogenetic Approaches. ed. by R. Mace, C. Holden, S. Shennan. UCL Press.
Dunn M., Terrill A., Reesink G., Foley R. & Levinson
S.C. 2005. Structural Phylogenetics and the
Reconstruction of Ancient Language History. Science,
309:2072-2075 .
Matisoff, J.A. 1990. On megalo-comparison: a discussion note. Language 66:106-20.