Zusammenfassung
Authors: Michael Dunn, Simon Greenhill
The Evolution of Phonological Complexity in Austronesian.
Common sense tells us that there must be correlations between aspects of the phonological and phonotactic organisation of languages, and that increased complexity in one area of a language correlate with reduced complexity in another. It has been suggested, for example, that mean word length might correlate (negatively) with the size of the phonological inventory, or that the number of consonants correlates (negatively again) with the number of vowels. These have a functional motivation: languages with many phonemes can make many lexical distinctions with short words, whereas languages with few phonemes must have longer words in order to make the same number of distinctions. This particular correlation has been proven (e.g Nettle 1995) and disproven (e.g. Maddieson 1984) several times, depending on the database and statistical approach used. More recently, Trudgill has made another proposal (with particular reference to Polynesian; Trudgill 2004) that sociolinguistic parameters, such as speech community size, might provide a motivating factor underlying some of this variation. In any case, there are good reasons to reserve judgement on the question of functional motivation for variation in phonological and phonotactic complexity.
Outline
This study approaches this question from another angle, measuring the coevolution of different linguistic features within the Austronesian language family. Genealogical issues form a confound in any attempt to formulate statistical universals. Current best practice to deal with this is to appeal to a genealogically independent cross linguistic sample. While in general independent sampling is a good solution to the problem of genealogical dependencies, it has weaknesses too, especially in terms of statistical power. But more importantly in our case, it overrepresents linguistic isolates and languages from small families, the very languages which tend to embody the sociolingistic parameters that Trudgill has predicted will drive particular patterns of linguistic complexity. Instead of trying to remove the phylogenetic structure of the data, the coevolutionary approach builds known phylogenetic structure into the analysis. If phonological complexity is negatively correlated with word length, then over the entire phylogenetic tree of Austronesian we would expect that a change in complexity of the reconstructed phonological system at any node of the tree would be (probabilistically) coupled with a change in reconstructed mean word length. If the hypothesis is false, then changes in complexity and word length should vary independently. With a rich database, as is available from the Austronesian Basic Vocabulary Database, each of these states of affairs can be modelled and the relative likelihood of each can be calculated.
