In natural languages, elements that can be easily deduced by a reader are frequently omitted from expressions in texts. This phenomenon causes considerable problems in natural language processing systems. In particular, the subject and object are often omitted in Japanese, whereas they are normally obligatory in English. In Japanese-to-English machine translation systems, therefore, it is necessary to identify case elements omitted from the original Japanese (zero pronouns) for their accurate translation into English expressions. We are investigating a technique for anaphora resolution of Japanese zero pronouns that is suitable for applications using a practical machine translation system with an unlimited translation target area. This method employs anaphora resolution conditions specially designed for zero pronouns based on the location of their antecedents : antecedents in the same sentence (intra-sentential anaphora), antecedents in a different sentence of the document (inter-sentential anaphora), and antecedents that do not explicitly exist in documents (deictic anaphora). Furthermore, we are also investigating a method for automatically extracting rules of Japanese zero-pronoun resolution from Japanese and English aligned sentence pairs. This method takes into account the practical situation for making resolution rules of zero pronouns in specific domain texts. Here, the types of usable corpora (monolingual and/or bilingual) for examining the extraction of resolution rules change depending on the type of NLP system using extracted resolution rules.

Natural language understanding, including Japanese zero-pronoun resolution, commonly requires an enormous volume of knowledge of word meaning that can be used to determine the semantic relationship between one sentence and another. To avoid an explosion in the volume of knowledge, we propose a system of 107 verbal semantic attributes of Japanese that considers both dynamic characteristics and the relationship of verbs to cases such as subjects and objects.

Machine translation also requires discourse analysis for selecting translation candidates suitable for a specific discourse. We are investigating a method for translating anaphoric expressions such as zero pronouns in Japanese into natural English based on their context.