TCtract-A Collocation Extraction Approach for Noun Phrases Using Shallow Parsing Rules and Statistic Models

TCtract-A Collocation Extraction Approach for Noun Phrases Using Shallow Parsing Rules and Statistic Models Wan Yin Li, Qin Lu, James Liu Department of Computing The Hong Kong Polytechnic University Hung Hom, Kowloon, Hong Kong {cswyli, csluqin, csnkliu}@comp.polyu.edu.hk Abstract. This paper presents a hybrid method for extracting Chinese noun phrase collocations that combines a statistical model with rule-based linguistic knowledge. The algorithm first extracts all the noun phrase collocations from a shallow parsed corpus by using syntactic knowledge in the form of phrase rules. It then removes pseudo collocations by using a set of statistic-based association measures (AMs) as filters. There are two main purposes for the design of this hybrid algorithm: (1) to maintain a reasonable recall while improving the precision, and (2) to investigate the proposed association measures on Chinese noun phrase collocations. The performance is compared with a pure statistical model and a pure rule-based method on a 60MB PoS tagged corpus. The experiment results show that the proposed hybrid method has a higher precision of 92.65% and recall of 47% based on 29 randomly selected noun headwords compared with the precision of 78.87% and recall of 27.19% of a statistics based extraction system. The F-score improvement is 55.7%. Keywords: Collocation Extraction, Typed Collocations, Phrase Rules, Association Measures. 1 Introduction Statistical approaches in collocation extraction still need improvement in terms of precision ([1], [10], [11]). There are two reasons for this. First, any given set of words has a very large number of possible combinations. One way to make statistical approaches more precise is simply to reduce the context to a fixed small size window of consecutive words. However, this method simply removes collocations beyond the span of the predefined window. The second cause of imprecision is that grammatical analysis itself is imprecise. Candidate collocations which are retrieved using statistical approaches may be “grammatically” related in the extracting system’s terms yet in fact consist of syntactically unrelated items which do not qualify as valid collocations. Ultimately, in the absence of massive computing power or extensive word lists the problems with statistical approaches would appear to be insuperable. For this reason it would appear that it may be useful to introduce other identifying features as hybrids to improve the precision of collocation extraction. On the face of it, linguistic knowledge is one area that would seem to offer an abundance of potentially useful text-related features such as semantic knowledge, morphological knowledge and, as will be discussed in this paper, syntactic knowledge, Studies that have made use of syntactic knowledge mainly fall into two broad categories. The first category makes use of syntactic filters ([1], [21]), which will not be considered in this paper. The second category makes use of collocation patterns ([3]), such as Adjective-Noun, Noun-Noun, Verb-Noun, Verb-Object, Subject-Verb, etc. However, the syntactic knowledge, such as phrase rules in this work, is mainly used for identifying occurrences of particular phenomena of certain given rules. It is not likely to apply the phrase rules individually because both inter-conflicts and the …

Content truncated to 500 words