Research on Hypothesizing and Sorting the Eg Candidates in Chinese Semantic Parsing

Research on Hypothesizing and Sorting the Eg Candidates in Chinese Semantic Parsing XiangFeng Wei1, Quan Zhang1 1 Institute of Acoustics, Chinese Academy of Science, Beijing, China. 100080 weixiangfeng@tsinghua.org.cn, zhq@mail.ioa.ac.cn Abstract. This paper introduced a semantic parsing model constructed above a symbolic system of concepts for understanding natural language. Based on the model a sentence can be mapped into the corresponding semantic category. In order to obtain the semantic category of a sentence (SC), it is necessary to get the global eigen semantic chunk (Eg) in a sentence. The Eg is like the head verb in a sentence but has its own constitution. It is more difficult to find out the Eg in a Chinese sentence than in English. This paper puts forward an approach and rules to hypothesize and sort the Eg in a Chinese sentence. The approach is named ‘Hypothesis-Test’. The experiments on sentences in which multi-verb appeared showed that the approach was valid and the accuracy rate of the semantic parsing system was 87.2%. Keywords: Semantic Parsing, Hypothesis-Test, Global Eigen Semantic Chunk (Eg), Formalized Rule 1 Introduction The global eigen semantic chunk (abbreviated as ‘Eg’ in the following) is the head dominant words at the top layer of a sentence. Generally the Eg roots in the head verb. However, it is possible that several verbs would appear in the sentence, like ‘have also continued’, ‘seek, reveal and develop’ and ‘guides’ in the example sentence: ‘They have also continued, in the practice of historical activities and by making comparisons, to seek, reveal and develop the truth that guides their advance’. In Chinese sentences, it is more difficult to find out the Eg because of changeless form of Chinese character, flexible parts of speech (POS), lack of case, plural and tense. In order to get the right Eg from all possible verbs in a Chinese sentence, the approach of ‘Hypothesis- Test’ is adopted. When parsing a sentence, the parser firstly hypothesized the Eg according to some special concepts. Then the parser hypothesized the SC of the sentence according to the Eg. Finally, the parser tests the hypothesized SC of the sentence according to the knowledge-base in the computer. If all semantic chunks in the sentence are corresponded with the transcendental concepts in the knowledge- base, the SC is confirmed. Otherwise, the SC is rejected. Therefore, the first important step of parsing the SC of a sentence is to hypothesize the Eg from the possible verbs. It is necessary to find out the right Eg in a sentence by hypothesizing and sorting the Eg candidates. This paper discussed how to establish the rules for hypothesizing and sorting the Eg candidates in Chinese Sentences and how to implement it by rules in computer programs. The results of experiments on the Chinese sentences selected from true articles indicated that the approach of ‘Hypothesis-Test’ is valid. According to the formalized rules, the accuracy rate was 87.2% in which the right Eg was sorted into the first Eg candidate. 250 2 Rules to Sort the …

Content truncated to 500 words