Verb Classification – Machine Learning Experiments in Classifying Verbs into Semantic Classes

This paper presents the results of our machine learning experiments in verb classification. Using Beth Levin’s semantic classification of the English verbs as a gold standard, we (i) test the hypothesis that the syntactic behavior of a verb can be used to predict its semantic class, and (ii) investigate whether a robust shallow parser can provide the necessary syntactic information. With 277 verbs belonging to six of Levin’s classes, we do type classification experiments using RIPPER, an inductive rule learner. Having only a set of n most likely subjects or objects as features, this machine learning algorithm is able to predict the correct class with ± 58% accuracy. This result is comparable with results from other researchers, like Merlo and Stevenson, Stevenson and Joanis, and Schulte im Walde.