Computing Word Similarity and Identifying Cognates with Pair Hidden Markov Models

Proceedings of the 9th Conference on Computational Natural Language Learning (CoNLL), pages 40–47, Ann Arbor, June 2005. c⃝2005 Association for Computational Linguistics Computing Word Similarity and Identifying Cognates with Pair Hidden Markov Models Wesley Mackay and Grzegorz Kondrak Department of Computing Science University of Alberta Edmonton, Alberta, Canada T6G 2E8 {wesleym,kondrak}@cs.ualberta.ca Abstract We present a system for computing sim- ilarity between pairs of words. Our sys- tem is based on Pair Hidden Markov Mod- els, a variation on Hidden Markov Mod- els that has been used successfully for the alignment of biological sequences. The parameters of the model are automatically learned from training data that consists of word pairs known to be similar. Our tests focus on the identification of cog- nates — words of common origin in re- lated languages. The results show that our system outperforms previously proposed techniques. 1 Introduction The computation of surface similarity between pairs of words is an important task in many areas of nat- ural language processing. In historical linguistics phonetic similarity is one of the clues for identi- fying cognates, that is, words that share a com- mon origin (Oakes, 2000). In statistical machine translation, cognates are helpful in inducing transla- tion lexicons (Koehn and Knight, 2001; Mann and Yarowsky, 2001), sentence alignment (Melamed, 1999), and word alignment (Tiedemann, 2003). In dialectology, similarity is used for estimating dis- tance between dialects (Nerbonne, 2003). Other applications include cross-lingual information re- trieval (Pirkola et al., 2003), detection of confusable drug names (Kondrak and Dorr, 2004), and lexicog- raphy (Brew and McKelvie, 1996). Depending on the context, strong word similarity may indicate either that words share a common ori- gin (cognates), a common meaning (synonyms), or are related in some way (e.g. spelling variants). In this paper, we focus on cognates. Genetic cognates are well-suited for testing measures of word simi- larity because they arise by evolving from a single word in a proto-language. Unlike rather indefinite concepts like synonymy or confusability, cognation is a binary notion, which in most cases can be reli- ably determined. Methods that are normally used for computing word similarity can be divided into orthographic and phonetic. The former includes string edit dis- tance (Wagner and Fischer, 1974), longest common subsequence ratio (Melamed, 1999), and measures based on shared character n-grams (Brew and Mc- Kelvie, 1996). These usually employ a binary iden- tity function on the level of character comparison. The phonetic approaches, such as Soundex (Hall and Dowling, 1980) and Editex (Zobel and Dart, 1996), attempt to take advantage of the phonetic characteristics of individual characters in order to estimate their similarity. All of the above meth- ods are static, in the sense of having a fixed defi- nition that leaves little room for adaptation to a spe- cific context. In contrast, the methods proposed by Tiedemann (1999) automatically construct weighted string similarity measures on the basis of string seg- mentation and bitext co-occurrence statistics. We have created a system for determining word similarity based on a Pair Hidden …

Content truncated to 500 words