Applying Machine Translation to Psychology: Automatic Translation of Personality Adjectives

摘要

We introduce our approach to apply machine translation to psychology, especially to translate English adjectives in a psychological personality questionnaire. We first extend seed English personality adjectives with a word2vec model trained with web sentences, and then feed the acquired words to a phrase-based machine translation model. We use Moses trained with bilingual corpora that consist of TED subtitles, movie' subtitles and Wikipedia. We collect Japanese translations whose translation probabilities are higher than .01 and filter them based on human evaluations. This resulted in 507 Japanese personality descriptors. We conducted a web-survey (N=17,751) and finalized a personality questionnaire. Statistical analyses supported the five-factor structure, reliability and criterion-validity of the newly developed questionnaire. This shows the potential applicability of machine translation to psychology. We discuss further issues related to machine translation application to psychology.