Patterns of language variation and underlying linguistic features: a new dialectometric approach

For almost forty years quantitative methods have been applied to the analysis of dialect variation: these methods focused mostly on identifying the most important dialectal groups using an aggregate analysis of the linguistic data (Séguy 1973; Goebl 1984; Nerbonne et al. 1999). While viewing dialect differences at an aggregate level certainly gives a more comprehensive view than the analysis of subjectively selected features, the aggregate approach has never fully convinced linguists of its use as it fails to identify the linguistic distinctions among the identified groups. Michele Loporcaro (2009) criticizes that dialectometry measures the structural distances among dialectal varieties without passing through a rationalization of the linguistic structure. Recently, Wieling and Nerbonne (2010, 2011) proposed a promising graph-theoretic method, the spectral partitioning of bipartite graphs, to cluster varieties and simultaneously determine the underlying linguistic basis. Their results for Dutch are promising, and this method can be taken as an answer to Loporcaro‟s criticism, since it permits one to reconstruct which features underlie the aggregate patterns of dialectal variation and the role played by each of them. In this way, the gap between models of linguistic variation based on quantitative analyses and more traditional analyses based on specific linguistic features is significantly reduced. This paper illustrates the application of this technique on the dialectal corpus of the Atlante Lessicale Toscano („Lexical Atlas of Tuscany‟, ALT)1 and discusses the results. The atlas material contains phonetic transcriptions even though the