Deep Case Estimation and Japanese Anaphora Resolution with a Verb-Associative Concept Dictionary
Kiyoaki ITO, Takehiro Teraoka · 2020
In Japanese zero anaphora analysis, it is difficult to identify antecedents in cases where it is necessary to decipher the context and cases where common knowledge information such as common sense needs to be used. The purpose of this study is to improve the accuracy of zero anaphora analysis for these cases. To this end, we propose a method using a verb associative concept dictionary. Experimental results for the analysis of one sentence showed that, in the Top-5 candidate words of the antecedents, the precision of narrowing down the "agent" and the "object" was 91% and 69%, respectively. The results for multiple sentences showed that, in the Top-10 candidate words of the antecedents, the precision of narrowing down the "agent" and the "object" was 86% and 30%, respectively. In the proposed method, only "agent" showed a result close to human judgment. These results demonstrate that the verb associative concept dictionary is effective for anaphora resolution.