Strategies for building high quality bilingual lexicons from comparable corpora

This chapter outlines two strategies to automatically build bilingual dictionaries: One is based on the use of a pivot language and existing bilingual dictionaries, while the other relies on string similarity and cognate extraction. Both strategies have in common the use of translation equivalents extracted from comparable corpora to filter out odd bilingual pairs and validate the correct ones. The correctness of the entries validated with comparable corpora is very high, close to that achieved by using parallel corpora. The chapter reports several case studies describing how to build new high-quality bilingual lexicons, namely English-Galician, English-Portuguese, and Portuguese-Spanish dictionaries with more than 90% precision. This outperforms state-of-the-art systems on bilingual extraction from comparable corpora, whose best scores hardly reach 70 or 80%.

keywords: Information Extraction, Bilingual Dictionaries, Comparable Corpora