Chinese Nominal Entity Recognition with Semantic Role Labeling

Abstract

Nominal entity recognition is a fundamental task in natural language processing. Semantic role labeling views a sentence as a predicate-arguments structure, which provides an alternative perspective for the boundary detection and type recognition of nominal entity. In this paper, we propose a nominal entity recognition method with semantic role labeling. First, a Maximum Entropy (ME) model is trained on unlabeled data to address the data sparse problem in acquiring the preferences for each pair of predicate and argument. Then, use the information of semantic role labeling as features in a high quality nominal entity model. The experiments on ACE 2004 Chinese data show that the proposed method improves the performance of the high quality nominal entity recognizer, and achieves higher accuracy and recall rate.