S-clause segmentation for efficient syntactic analysis using decision trees

S-clause Segmentation for Efficient Syntactic Analysis Using Decision Trees Mi-Young Kim Jong-Hyeok Lee Div. of Electrical and Computer Engineering Pohang University of Science and Technology (POSTECH) and Advanced Informa- tion Technology Research Center(AlTrc) San 31 Hyoja-dong, Nam-gu, Pohang 790-784, R. of Korea colorful@postech.ac.kr jhlee@postech.ac.kr fax: +82-54-279-5699 fax: +82-54-279-5699 Keywords : S-clause segmentation, Decision trees, Syntactic analysis, Long sentence analysis Abstract In dependency parsing of long sentences with fewer subjects than predicates, it is difficult to recognize which predicate governs which subject. To handle such syntactic ambiguity between subjects and predicates, this paper proposes an “S- clause” segmentation method, where an S(ubject)- clause is defined as a group of words containing several predicates and their common subject. We propose an automatic S-clause segmentation method using decision trees. The S-clause information was shown to be very effective in analyzing long sentences, with an improved performance of 5 percent. 1 Introduction The longer the input sentences, the worse the parsing results are, since problems with syntactic ambiguity increase drastically. In our parser, subject errors form the second largest error portion, as 24.15% of syntactic parsing errors (see Table 1). Although the dependency errors in NP form the largest error portion, these errors are not significant since many applications (e.g. MT sys- tems) using parsers deal with the NP structure as a one unit and do not analyze the syntactic relations within NP. So, this paper proposes a method to resolve subject dependency error problems. To improve the dependency parsing performance, we need to determine the correct dependency relations of subjects. In most cases, a long sentence has fewer subjects than predicates. The reason is that several predicates can share one subject if they require the same word as their subject, or that the subject of a predicate is often omitted in a Korean sentence. So, in a long sentence, it is difficult to recognize the correct subject of some subjectless VPs. This paper proposes an S(ubject)-clause segmentation method to reduce ambiguity in determining the governor of a subject in dependency parsing. An S(ubject)-clause is defined as a group of words containing several predicates and their common subject. An S-clause includes one subject and several predicates which share the subject. The S- clause segmentation algorithm detects the boundary of predicates which share a common subjective word. We employ the C4.5 decision tree learning algorithm for this task. The next section presents the background of previous work on sentence segmentation and clause detection. Next, dependency analysis procedure using S-clauses in Korean will be described. Afterwards, the features for decision tree learning to detect S-clauses will be explained, and some experimental results will show that the proposed S-clause segmentation method is effective in dependency parsing. Finally, a conclusion will be given. 2 dependency tree errors subject- predicate dependency errors predicate- predicate dependency errors adjunct- predicate dependency errors complement- predicate dependency errors dependency errors within NP dependency errors resulting from POS- tag errors error % 24.15% 14.29% 17.35% 8.84% 27.55% 7.48% Table 1. Dependency Tree Errors for …

Content truncated to 500 words