Semantic role labeling(SRL) and Chinese sentence similarity computation both are essential tasks and widely used in the Chinese information processing. Recent works on SRL are almost based on supervised method, which need a large labelled corpus. However, this kind of resources are still quite limited for Chinese. In order to deal with this problem, this paper presents a semi-supervised method for labeling the arguments and adjunctions of verbs with their semantic roles, like agentive, recipient,time,locations and so on. Sentence similarity computation is comparing the degree of similarity between sentences, but recent works on this field do not think about the similarity of sentences' semantic frameworks.So, based on the above considerations, this paper is mainly to the following three aspects:Firstly, build a semi-supervised system containing three steps for SRL. By bootstrapping from a small set of labeled corpus, the SRL system achieves the accuracy of 83.32%.Secondly, In order to improve the system's performance, the automatic semantic classification of unknown Nouns is also implemented, and is combined into SRL algorithm. After combined with the automatic classification of unknown terms, the system achieves the accuracy of 83.91%.Thirdly, The SRL system is then applied on the problem of sentence similarity to improve the performance of this task. Experiments show that better perforumance is achieved by combining the results of SRL into sentence similarity computation. |