Proper Noun Recognition With Transformation-based Learning

Word segmentation is the foundation of Chinese information processing, and its performance is influenced extremely by the recognition of unknown words, especially the recognition of proper nouns. According to proper nouns'internal structures and their context, this thesis proposes a novely approach to recognize proper nouns, namely using rules primarily to recognize proper nouns.In this novelty approach, proper nouns' term information together with rules attained by transformed-based error-driven learning is used to label properties of segmented text, so as to recognize proper nouns. Not only the extraction of term information and rules is automatic, but also the system will be fit for new condition just by changing training corpus without any manual intervention.Firstly, the issues of recognizing proper nouns are discussed as well as the popular approaches are presented and compared, and according to the analysis, a new approach, namely using rules primarily to recognize proper nouns, is evoked for the purpose. Then, after analyzing the main idea and presenting the design, it shows the system design in detail, including defining basis concepts, pre-processing text, collecting proper nouns'information, extracting rules and labeling properties. Finally, the system realization is divided into two parts and detailed illuminate is presented respectively.This thesis proposes a novely approach to recognize proper nouns and the recognition system is realized based on the approach. The system is tested on several groups of data, and the results show that the performance is satisfactory, and the open test results also show that the system's capability of recognizing proper nouns is of great...