Establishing cross-linguistic semantic relatedness through monolingual corpora

Each instance of language comparison requires observations on semantic equivalence. Meaning is by far the most popular tertium comparationis in contrastive and typological research. However, the question of how semantic equivalence is to be determined remains extremely difficult to solve. This paper presents an approach to detect semantic relatedness between a limited range of lexical items from different languages on the basis of monolingual data. Applying distributional similarity (Dagan et al. 1999) cross-linguistically, it identifies semantically related verbs governing embedded interrogatives by looking at the frequency of the question words (i.e. wh-items) that are used in the embedded interrogatives in monolingual corpora. Convincing results are obtained for six different language pairs: English-French, English-Dutch, English-Spanish, French-Dutch, French-Spanish and Dutch-Spanish.