Linguistic Variation and Computation (Invited talk)

Linguistic Variation and Computation John Nerbonne Alfa-informatica, BCN, University of Groningen 9700 AS Groningen, The Netherlands nerbonne@let.rug.n1 Abstract Language variationists study how lan- guages vary along geographical or so- cial lines or along lines of age and gender. Variationist data is available and challenging, in particular for DI- ALECTOLOGY, the study of geograph- ical variation, which will be the focus of this paper, although we present ap- proaches we expect to transfer smoothly to the study of variation correlating with other extralinguistic variables. Tech- niques from computational linguistics on the one hand, and standard statis- tical data reduction techniques on the other, not only shed light on this clas- sic linguistic problem, but they also sug- gest avenues for exploring the question at more abstract levels, and perhaps for seeking the determinants of variation. 1 Introduction The study of language variation has always been an important aspect of linguistic research. It pro- vides insights into historical, social and geograph- ical factors of language use in society. GilHeron, the father of French dialectology, was, for exam- ple, famous for showing that several linguistic di- visions, running roughly East-West across French, corresponded closely with well established cul- tural divisions, in particular the ethnic split be- tween slightly Romanized Celts in the North, and thoroughly Romanized non-Celts in the South, the legal division between the common law North and the Roman law South, and patterns of agricul- ture and architecture (see Chambers and Trudg- ill (1980, pp.111-123)). In recent years theoreti- cians have also turned increasingly to the study of dialects as a means of demarcating the possible range of human language in more detail (Beninca, 1987). The present paper sketches some ways in which techniques from computational linguistics (CL) can be put to use in the study of variation. Language variationists study how languages vary along geographical or social lines or along lines of age and gender. Variationist data is available and challenging, in particular for DI- ALECTOLOGY, the study of geographical varia- tion, which will be the focus of this paper, al- though we present approaches we expect to trans- fer smoothly to the study of variation correlat- ing with other extralinguistic variables. Most non-computational studies focus on a small num- ber of features and cannot characterize AGGRE- GATE levels, e.g., the Bavarian dialect or the lan- guage of London teenagers, using these few char- acteristics. Aggregate characterizations are elu- sive because large data sets invariably contain counter-indicating tendencies leading to the ana- lytical challenge of characterizing notions of ag- gregate levels without simply insisting on the im- portance of one's favorite features. Techniques from computational linguistics on the one hand, and standard statistical data reduction techniques on the other, not only shed light on this classic lin- guistic problem, but they also suggest avenues for exploring the question at more abstract levels, and 3 Figure 1: Bloomfield's (1933:328) classical discussion of the problems of determining dialect areas. The vowels in Dutch huis, muis ('house', 'mouse') were the same historically, but …

Content truncated to 500 words