Loss in Translation: Learning Bilingual Word Mapping with a Retrieval Criterion
This work addresses the challenge of cross-lingual word mapping for NLP applications, offering an incremental improvement over existing methods.
The paper tackles the problem of aligning bilingual word embeddings by proposing an end-to-end formulation that directly optimizes a retrieval criterion, resulting in state-of-the-art performance on word translation, with significant improvements for distant language pairs like English-Chinese.
Continuous word representations learned separately on distinct languages can be aligned so that their words become comparable in a common space. Existing works typically solve a least-square regression problem to learn a rotation aligning a small bilingual lexicon, and use a retrieval criterion for inference. In this paper, we propose an unified formulation that directly optimizes a retrieval criterion in an end-to-end fashion. Our experiments on standard benchmarks show that our approach outperforms the state of the art on word translation, with the biggest improvements observed for distant language pairs such as English-Chinese.