Study of Segmentation Strategy on Ambiguous Phrases of Overlap Type

摘要

Ambiguity segmentation is still an open issue in the study of Chinese word segmentation. In corpus, most of the ambiguous phrases of overlap type are pseudo, which can be correctly segmented with only the information within the phrases. While the segmentation for true ambiguity needs grammatical or even semantic information. In this paper, we present a segmentatio strategy on ambiguous phrases of voerlap type with rules based on the independent-wording-ability frequency and collocating of words & parts of speech, which improves the accuracy of segmentation greatly. In an open test of a Chinese corpus with 39,000 characters, the accuracy of segmentation for ambiguous phraes of overlap reached 98