Generalizing Subcategorization Frames Acquired from Corpora Using Lexicalized Grammars

This paper presents a method of improving the quality of subcategorization frames (SCFs) acquired from corpora in order to augment a lexicon of a lexicalized grammar. We first estimate a confidence value that a word can have each SCF, and create an SCF confidence-value vector for each word. Since the SCF confidence vectors obtained from the lexicon of the target grammar involve co-occurrence tendency among SCFs for words, we can improve the quality of the acquired SCFs by clustering vectors obtained from the acquired SCF lexicon and the lexicon of the target grammar. We apply our method to SCFs acquired from corpora by using a subset of the SCF lexicon of the XTAG English grammar. A comparison between the resulting SCF lexicon and the rest of the lexicon of the XTAG English grammar reveals that we can achieve higher precision and recall compared to naive frequency cut-off.