Improving Japanese-English Bilingual Mapping of Word Embeddings based on Language Specificity

Abstract

Recently, cross-lingual word embeddings have attracted a lot of attention, because they can capture semantic meaning of words across languages, which can be applied to cross-lingual tasks. Most methods learn a single mapping (e.g., a linear mapping) to transform word embeddings space from one language to another. In this paper, we propose an advanced method for improving bilingual word embeddings by adding a language-specific mapping. We focus on learning Japanese-English bilingual word embedding mapping by considering the specificity of Japanese language. On a benchmark data set of JapaneseEnglish bilingual lexicon induction, the proposed method achieved competitive performance compared to the method using a single mapping, with better results being found on original Japanese words.