Lost in Translations? Building Sentiment Lexicons using Context Based Machine Translation

In this paper, we propose a simple yet efective approach to automatically building sentiment lexicons from English sentiment lexicons using publicly available online machine translation services. The method does not rely on any semantic resources or bilingual dictionaries, and can be applied to many languages. We propose to overcome the low coverage problem through putting each English sentiment word into diferent contexts to generate diferent phrases, which efectively prompts the machine translation engine to return diferent translations for the same English sentiment word. Experiment results on building a Chinese sentiment lexicon (available at https://github.com/fannix/ChineseSentiment-Lexicon) show that the proposed approach signiicantly improves the coverage of the sentiment lexicon while achieving relatively high precision.