Research On Automatic Recognition Of Uyghur Temporal Expressions

Temporal entity recognition, a part of foundation research of named entityrecognition (NER), is essential to many natural language processing applications(NLP). The temporal dimension of information is fundamental for reasoning abouthow the things change. The recognition and normalization of temporal expressions isnot only an important research topic, but also a very practical challenge. In order toresolve a series of problems including morphologic change, recognition and temporalboundary localization of Uyghur text temporal expressions, a method based onConditional Random Fields combining with Support Vector Machine is proposed,which can be applicable to recognize the Uyghur temporal expression. Uyghurlanguage is a highly inflectional language providing one of the richest and mostchallenging sets of linguistic and statistical features. The recognition recall rate ofConditional Random Fields is not high, the Support Vector Machine modelrecognized results is used for correct. In Uyghur Event-Anchored TemporalExpressions recognition, we replace the lexical features to stem features in order toavoid data sparseness problem. In this case, the effects of different features andtemplate files on temporal expressions recognition were analyzed. The general laws offeatures and template files were formed on experimental data. A multiplecross-validation method was used to evaluate the F-measure of different features andtemplate files. Feature reduction was used to find the optimal set of features.Experimental results show that the proposed method is benefit to Uyghur temporalexpression recognition. The results of study have certain reference significance forother agglutinative languages.