Word and Phrase Translation with word2vec

원문 초록

Word and phrase tables are key inputs to machine translations, but costly to produce. New unsupervised learning methods represent words and phrases in a high-dimensional vector space, and these monolingual embeddings have been shown to encode syntactic and semantic relationships between language elements. The information captured by these embeddings can be exploited for bilingual translation by learning a transformation matrix that allows matching relative positions across two monolingual vector spaces. This method aims to identify high-quality candidates for word and phrase translation more cost-effectively from unlabeled data. This paper expands the scope of previous attempts of bilingual translation to four languages (English, German, Spanish, and French). It shows how to process the source data, train a neural network to learn the high-dimensional embeddings for individual languages and expands the framework for testing their quality beyond the English language. Furthermore, it shows how to learn bilingual transformation matrices and obtain candidates for word and phrase translation, and assess their quality.