柬埔寨语命名实体识别及汉柬双语可比语料库构建方法研究

As the research on Khmer language processing is very weak,research on Khmer named entity recognition plays an important role in the future research of the understanding of Chinese-Khmer bilingual language.Based on the Khmer language characteristics, Firstly, study on the annotation specification of Khmer word segmentation，part-of-speech and named entity to build a 500000-word-sized Khmer tagged corpus,on the basis of which,study on the Khmer named entity recognition methods based on hierarchical hybrid model of conditional random field and Markov logic network according to the characteristic that Khmer linguistic knowledge can be characterized with first-order logic rules,and study on the English-Khmer cross-lingual named entity knowledge transfer methods to get cross-lingual feature for the purpose of improving Khmer named entity recognition which lacks good discriminating features. Finally,study Chinese-Khmer comparable corpora building methods combining bilingual topic model and incremental clustering by means of knowledge dictionary ,according to the characteristic of Khmer news Web pages have the similar topic and content to the Chinese ones.The research achievement of the project will supply basic support and corpus resources to Khmer language processing and Chinese-Khmer machine translation.

柬埔寨语语言信息处理研究非常薄弱，研究柬埔寨语命名实体识别对于今后的汉-柬双语理解研究工作具有重要的支撑作用。汉语-柬埔寨语双语语料资源构建是汉柬双语理解的必备资源。本课题根据柬埔寨语的特点，首先研究柬埔寨语分词、词性标注与命名实体标注规范，构建含50万单词的柬埔寨语词法标注语料库，在此基础上，根据柬埔寨语言学知识能很好地用一阶逻辑规则描述的特点，研究基于条件随机场与马尔科夫逻辑网双层模型的柬埔寨语命名实体识别方法，研究英语-柬埔寨语跨语言命名实体知识转移方法，引入跨语言特征以促进缺乏可区分特征的柬埔寨语命名实体识别工作。最后结合汉柬新闻网页主题相关、内容相近的特点，借助于知识词典，研究通过融合汉柬双语主题建模和增量聚类获取汉-柬双语可比较语料的方法。项目研究成果为柬埔寨语信息处理、汉语-柬埔寨语机器翻译提供底层支持和数据支撑。