Themes and focuses extraction system using morphemes and valences

The paper describes an extraction system for themes and focuses in Japanese. Themes and focuses are the most important parameters for representing the discourse structure of a language, and in Japanese are sometimes explicitly shown in a surface sentence with postpositions such as "wa" and "ga". We have already proposed an extraction algorithm for themes and focuses, however, the precondition of the algorithm was such that it was difficult to extract these parameters automatically. The authors show that relaxing the precondition slightly does not degrade the performance of the system significantly. We report a summary of the system, and present a plausible fully automatic extraction system