Recognition Of Temporal, Event Expressions And Their Attributes In Chinese Texts

In recent years, time and event recognition task has been received wide attention and rapid development. As the basis of the study of the temporal relation, temporal expression recognition, the identification of the event and its attributes has also become an important task in natural language processing. In2010, Time and event recognition as two separate sub-tasks included in the Temp-Eval evaluation meetings. The evaluation of the meeting involved six languages:Chinese, English, Italian, French, Korean and Spanish.This paper mainly studies two tasks of TempEval-2evaluation task, the identification of time and events and their associated attributes. Experimental corpus is from the meeting of the Chinese evaluation corpus, and involving marked TimeML standard. Now, Specific research works are stated as follows:1The analysis of the research questionsDetailed definition and analysis of the identification problem of time and events, and analyzed to identify the specific content of the work and identify the difficulties. This part of the analysis is the basis of the entire article, it is a preparatory work for proposing one recognition method.2Recognition of temporal and its typeFor recognition work of temporal expression and its type, the article made specific and comprehensive introduction, including the identification of ideas and recognition process. The main work of the section is time expression recognition and identification of the type of time. This part proposed a method based on rule base, which is generated by "Unit of time" based POS tagging to indentify the time expression, and a method based on ME (maximum entropy) model to label the type of time expression. Time expression recognition yield the precision, recall and F-measure values of85.16%,83.16%and84.17%, respectively, the time type correct rate is93.02%. Finally, this section makes a in-depth analysis and summary of experimental results and error.3Recognition of event and its attributesThe tasks of event identifies was to determine the extent of the events in a text as defined by the TimeML event tag. In addition, the values of the features tense had to be determined. The recognition of event based on dependency analysis and rule-based approach to identify, and event tense used the rules and semi-supervised learning method to identify. Event recognition yield accuracy, recall and F-measure values of89.2%,82.8%,85.9%. The correct rate of event tense identification is76.9%. Article analyzed shortcomings and errors by the comparison between experimental results and others, but also conducted in-depth analysis of the problems in the recognition process.