摘要

Semantic role labeling (SRL) is a kind of shallow semantic analysis. The full parsing based SRL works well on handcrafted tree already. Yet, when a real full parser is used, its performance drops sharply. Shallow parsing base SRL does not drop like it under the same condition. However, shallow parsing produces too little information so that the promotion in performance is less even using handcrafted syntactic parsing. In this paper, a shallow parsing based SRL integrated with full parsing is proposed. More information yielded in full parsing is added into shallow parsing based SRL to enhance the performance. And a result of 76.2% (F1 score) is obtained from the automatic syntactic parsing.

全文