Anchor-based Bilingual Word Embeddings for Low-Resource Languages
This addresses the challenge of poor embedding quality for low-resource languages, which is incremental but beneficial for NLP applications in such contexts.
The paper tackles the problem of building bilingual word embeddings for low-resource languages by using the vector space of a high-resource source language as anchors during training, resulting in improved bilingual lexicon induction and target language monolingual word similarity performance.
Good quality monolingual word embeddings (MWEs) can be built for languages which have large amounts of unlabeled text. MWEs can be aligned to bilingual spaces using only a few thousand word translation pairs. For low resource languages training MWEs monolingually results in MWEs of poor quality, and thus poor bilingual word embeddings (BWEs) as well. This paper proposes a new approach for building BWEs in which the vector space of the high resource source language is used as a starting point for training an embedding space for the low resource target language. By using the source vectors as anchors the vector spaces are automatically aligned during training. We experiment on English-German, English-Hiligaynon and English-Macedonian. We show that our approach results not only in improved BWEs and bilingual lexicon induction performance, but also in improved target language MWE quality as measured using monolingual word similarity.