Evaluation of Sparse Binary Word Vectors in Chinese-English Machine Translations

We study the problem of language representation for automatic processing. Based on the idea of sparse lifting, our work projects dense word vectors into a higher-dimensional space and then generates sparse binary representations for words. In this way, our model combines the benefits from both one-hot representations and the distributional word vectors and provides a general representation scheme for automatic natural language processing. In an empirical study on Chinese to English bidirectional machine translations, the proposed model clearly demonstrates its advantage in downstream applications.