A Boundary Assembling Method for Chinese Entity-Mention Recognition

A boundary assembling (BA) method is presented for Chinese entity-mention recognition. Given a sentence, instead of recognizing entity mentions in a unitary style, the authors´ BA method first detects boundaries of entity mentions and then assembles detected boundaries into entity-mention candidates. Each candidate is further assessed by a classifier trained on nonlocal features. This method can make better use of nonlocal features and effectively recognize nested entity mentions. Using the ACE 2005 Chinese corpus, the authors´ experimental results show an improvement over state-of-the-art techniques, outperforming existing methods in F-score by 5 percent for entity-mention detection and 4.23 percent for entity-mention recognition.