Non-Parametric Bayesian Areal Linguistics

Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the ACL, pages 593–601, Boulder, Colorado, June 2009. c⃝2009 Association for Computational Linguistics Non-Parametric Bayesian Areal Linguistics Hal Daum´e III School of Computing University of Utah Salt Lake City, UT 84112 me@hal3.name Abstract We describe a statistical model over linguis- tic areas and phylogeny. Our model recov- ers known areas and identifies a plausible hi- erarchy of areal features. The use of areas improves genetic reconstruction of languages both qualitatively and quantitatively according to a variety of metrics. We model linguistic areas by a Pitman-Yor process and linguistic phylogeny by Kingman’s coalescent. 1 Introduction Why are some languages more alike than others? This question is one of the most central issues in his- torical linguistics. Typically, one of three answers is given (Aikhenvald and Dixon, 2001; Campbell, 2006). First, the languages may be related “genet- ically.” That is, they may have all derived from a common ancestor language. Second, the similarities may be due to chance. Some language properties are simply more common than others, which is of- ten attributed to be mostly due to linguistic univer- sals (Greenberg, 1963). Third, the languages may be related areally. Languages that occupy the same geographic area often exhibit similar characteristics, not due to genetic relatedness, but due to sharing. Regions (and the languages contained within them) that exhibit sharing are called linguistic areas and the features that are shared are called areal features. Much is not understood or agreed upon in the field of areal linguistics. Different linguists favor differ- ent defintions of what it means to be a linguistic area (are two languages sufficient to describe an area or do you need three (Thomason, 2001; Katz, 1975)?), what areal features are (is there a linear ordering of “borrowability” (Katz, 1975; Curnow, 2001) or is that too prescriptive?), and what causes sharing to take place (does social status or number of speakers play a role (Thomason, 2001)?). In this paper, we attempt to provide a statistical answer to some of these questions. In particular, we develop a Bayesian model of typology that al- lows for, but does not force, the existence of linguis- tic areas. Our model also allows for, but does not force, preference for some feature to be shared are- ally. When applied to a large typological database of linguistic features (Haspelmath et al., 2005), we find that it discovers linguistic areas that are well documented in the literature (see Campbell (2005) for an overview), and a small preference for cer- tain features to be shared areally. This latter agrees, to a lesser degree, with some of the published hi- erarchies of borrowability (Curnow, 2001). Finally, we show that reconstructing language family trees is significantly aided by knowledge of areal features. We note that Warnow et al. (2005) have indepen- dently proposed a model for phonological change in Indo-European (based on the Dyen dataset (Dyen et al., 1992)) that includes notions of borrowing. Our model is different …

Content truncated to 500 words