Discovering bilingual collocations in parallel corporaA first attempt at using distributional semantics

Resumo

This chapter presents a method that exploits parallel corpora to automatically extract bilingual collocation equivalents. First, we use dependency parsing and statistical measures to identify collocation candidates in corpora. Then, we leverage the parallel corpora to extract bilingual word-embeddings. Finally, we use these distributional models as probabilistic dictionaries in order to identify bilingual collocation equivalents. To evaluate our strategy we carry out a set of experiments in Portuguese and Spanish focusing on verb-object collocations, for example, "reach the maturity" ("atingir a maturidade" in Portuguese, "alcanzar la madurez" in Spanish). The results of our experiments show that this method is useful to automatically identify thousands of bilingual collocation equivalents, achieving a precision of 86%.