We present a system for computing similaritybetween pairs of words. Our systemis based on Pair Hidden Markov Models,a variation on Hidden Markov Modelsthat has been used successfully for thealignment of biological sequences. Theparameters of the model are automaticallylearned from training data that consistsof word pairs known to be similar. Ourtests focus on the identification of cognates- words of common origin in relatedlanguages. The results show that oursystem outperforms previously proposedtechniques.
展开▼