【作者】 张晓甜； 赵海；

【Author】 Zhang Xiaotian,Zhao Hai MOE-Microsoft Key Laboratory of Intelligent Computing and Intelligent System Department of Computer Science and Engineering,Shanghai Jiao Tong University,Shanghai 200240

【机构】 教育部-微软智能计算与智能系统重点实验室； 上海交通大学计算机科学与工程系；

【摘要】 本文研究了非监督的中文短语结构句法分析。我们首次精确重现了Rens Bod在[2]中阐述的非监督数据驱动模型U-DOP。应用U-DOP方法在CTB[14]上达到了提出该方法的原始文献所报道的结果,同时,按照已有文献的评测策略,在已知的基于词性串分析的非监督短语结构句法分析系统中,本文报道了在可比较实验条件下的最高性能。进一步地,所实现的U-DOP的结果与另一个基于词的非监督句法分析CCL[13]的结果进行了经验性的比对和分析。更多还原

【Abstract】 This paper investigates unsupervised phrase parsing for Chinese.It is the first time according to our best knowledge that unsupervised data oriented(U-DOP) model described in[2]is fully and exactly re-implemented Our U-DOP implementation achieves the similar results on CTB as reported in[2].Moreover,using the evaluation measure in[3],our system achieves the highest F1 score among all the existing POS-sequence based unsupervised phrase parsing models.We also give a detailed comparison between the performance of our U-DOP system and the unsupervised CCL parser[13]in terms of prediction accuracy on different kinds of phrases.更多还原