Abstract
Verb frames are very important for language learners, since they capture the semantics and word usages associated with verbs. Unfortunately, most online dictionaries such as Longman Dictionary show verb frames with broad semantic categories (i.e., something and somebody) which are not very informative. In this work, we introduce a method for automatically generating more comprehensive verb frames. The method involves extracting verb argument tuples based on grammatical relations acquired from a parsed corpus, obtaining intended semantic categories for each argument based on a knowledge base, estimating the probabilities of each semantically labeled tuples, and finally generating verb frames. We present a prototype system, FrameFinder, that applies the method to generate verb frames automatically. Evaluation on a set of verbs with manually compiled semantic patterns shows that the method is able to extract with high accuracy for the important high frequnecies verbs for language learning.