Research On The Automatic Acquisition Of Preferred Semantic Classes In Chinese

Selectional preference is an important lexical knowledge which can be widely used in the field of natural language processing,such as metaphorical computing,syntactic analysis,semantic role labeling,word sense disambiguation,anaphora resolution,machine translation,etc.The manually constructed knowledge base of selectional preference is hard to meet the needs of natural language processing.Therefore,automatic acquisition of preferred semantic classes from large-scale corpus is needed.This thesis focuses on the Chinese selectional preference of semantic categories,the Chinese preferred semantic classes.We propose a model for automatic selectional preference acquisition based on the minimum description length principle and semantic taxonomy,and we construct the knowledge base of selectional preference.The main contents of this thesis are as follows:(1)Extracting and analyzing selectional preference knowledge in existing knowledge base.First,we systematically extract the selectional preference knowledge of HowNet,and summarize the selectional preference knowledge in VerbNet and SKCC.Then,by analyzing and comparing the three above,it is found that the selectional preference of VerbNet and HowNet are basically consistent,which is more perfect for each other as supplement,and the selectional preference on the subject and object of the verb in SKCC is stricter.(2)Obtaining selectional preference knowledge from corpus.We propose a model for automatic acquisition of Chinese selectional preference based on the principle of minimum description length and the semantic taxonomy.We modify the existing noun semantic taxonomy of HowNet,and then get selectional preference knowledge based on a large-scale corpus.The pseudo-disambiguation experiment shows that the proposed method is better than the method based on KL distance.And the selectional preference in SKCC is introduced as the source of gold standard test set to evaluate selectional preference acquisition performance which is based on the adjusted noun semantic taxonomy of SKCC,and the relaxed accuracy reaches 75.26%.(3)Building a knowledge base of selectional preference.Firstly,we match seed words for each sense of polysemous verb through the fifth edition of Modern Chinese Dictionary.Then the cosine similarity between the word vector of the target word and the word vector of the seed word is calculated.The target words are thus divided into different meanings,and the purpose of word sense disambiguation is achieved.In the end,we constructed a knowledge base of verb-object selectional preference.Focusing on the construction of semantic knowledge base,we extract and analyze the selectional preference knowledge in existing knowledge base,and propose a model for automatic acquisition of Chinese selectional preference based on the principle of minimum description length and the semantic taxonomy.The comparison experiment shows the effectiveness of this method.