PROBLEM TO BE SOLVED: To precisely detect an empty category for Japanese.SOLUTION: Based on a phrase structure tree obtained from syntax analysis of an input text by an identity vector generation unit 230, an identity vector is prepared for each deflection phrase node included in the phrase structure tree using child node identity and pass identity. Based on an empty category detection model for detecting an empty category label granted to a deflection phrase node that is learned in advance by an empty category label detection unit 236 and each identity vector of the deflection phrase node, a position and a kind of an empty category in the input text are detected.SELECTED DRAWING: Figure 10
展开▼