欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
基于汉英双语命名实体的识别与对齐特性，文中提出了一种双语命名实体交互式对齐模型，其中的修正对齐计算体现了汉英实体识别与对齐的密切结合：一方面，利用双语对齐信息帮助实体识别；另一方面，实体的对齐过程对实体的识别结果又具有一定的修正作用，两方面的结合实现了双语实体识别与对齐之间的交互式互助过程．实验证明，这种交互式对齐模型不仅显著提高了汉英实体对齐的性能（F值从74．4％提高到81．2％），而且有效地提高了汉英实体识别的正确率和召回率．
英文摘要：
Named entity （NE） recognition is an essential early stage and has long been a thorny problem in many natural language processing pipelines. Generally, bilingual named entity recog nition and alignment are processed sequentially and independently, regardless of interactions between the two stages. Therefore, NE recognition errors are propagated and compounded in NE alignment stage. Actually, bilingual alignment information, other than monolingual information, provides further indications for NE recognition. It is necessary to capture the interactions be- tween NE recognition and alignment. Accordingly, based on the characteristics of NE recognition and alignment, this paper proposes an interactive bilingual NE alignment model, which combines basic alignment and corrective alignment. Basic alignment is achieved as traditional alignment approach, while the corrective alignment realizes the joint of NE recognition and alignment. On the one hand, bilingual alignment information is utilized for NE recognition; on the other hand, NE recognition errors can be recovered in the NE alignment stage. Both the NE boundaries and type can be corrected in such interactive NE alignment model. The experiments show that this novel model not only achieves a significant improvement of the Chinese-English NE alignment quality （F-score from 74.4% to 81.2%）, but also improves the performance of NE recognition.
同期刊论文项目
同项目期刊论文
期刊信息
位置：