The alignment dictionary 23 includes correspondence relationships between words for representations in the reference language and representations in the target language. For example, when the reference language is English and the target language is Japanese, the alignment dictionary 23 includes correspondence relationships between words for representations in English and representations in Japanese. An example of the alignment dictionary 23 is Weblio that is an online English dictionary service in Japan.

The distributed representation learning section 11 uses a technique for producing a distributed representation of a word to learn a distributed representation of a word included in a word space of the reference language. For example, the distributed representation learning section 11 receives the reference language learning corpus 21 and uses, for example, the Skip-gram model of Word2Vec to learn distributed representations of words included in the reference language learning corpus 21. The learning using the Skip-gram model of Word2Vec is executed using an existing technique. The learning may be executed using a technique disclosed in “Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. Efficient Estimation of Word Representations in Vector Space. In Proceedings of Workshop at ICLR, 2013”. Alternatively, the learning may be executed using a technique disclosed in “Xin Rong. word2vec Parameter Learning Explained”.