Collocation Extraction Using Web Feedback Data

As an important linguistic resource, collocation represents a significant relation between words.Automatic collocation extraction is very important formany natural language processing applications such asmachine translation, information extraction, and information retrieval. While traditional collocation extraction approaches are based on linguistic corpus, we propose to acquire collocations from the Web. Three classical lexical association measures (co-occurrence frequency, mutual information and t-test) are used to automatically extract collocation. Based on the experimental results, the benchmarksindicate that superior performance of this new Web-basedapproach in both high precision and recall.