Mapping Senses In Babelnet To Chinese Based On Word Embedding

Along with language evolution, different languages have formed self-contained concept architecture. The same concept may be described with different semantic relations in different languages. The mismatch among concepts hinders synergetic construction of language resources. Since the concepts are identical in nature, if they could be aligned, the concept architecture of different language would complement with each other. Aiming at the problem, this paper proposes a method to automatically align the concepts in different languages based on the sentence embedding with the help of a bilingual dictionary. Firstly, a word embedding is trained on a monolingual corpus; Secondly, based on the word embedding, the sentence embedding of concept definition and examples are generated; Thirdly, for each concept in the target language, its similarity with candidate concepts in the source language are computed based on the sentence embedding; Then, it would be aligned to the concept in the source language with the maximum similarity. We have developed a test dataset to evaluate our methods. The experiment on the dataset has demonstrated the best performance of our method, which has achieved 75.75% in F-1-measure.