Verb Sense Disambiguation Using Selectional Preferences Extracted with a State-of-the-art Semantic Role Labeler

Proceedings of the 2006 Australasian Language Technology Workshop (ALTW2006), pages 139–148. Verb Sense Disambiguation Using Selectional Preferences Extracted with a State-of-the-art Semantic Role Labeler Patrick Ye and Timothy Baldwin Dept. of Computer Science and Software Engineering University of Melbourne Victoria 3010 Australia {jingy,tim}@csse.unimelb.edu.au Abstract This paper investigates whether multi- semantic-role (MSR) based selectional preferences can be used to improve the performance of supervised verb sense dis- ambiguation. Unlike conventional se- lectional preferences which are extracted from parse trees based on hand-crafted rules, and only include the direct subject or the direct object of the verbs, the MSR based selectional preferences to be pre- sented in this paper are extracted from the output of a state-of-the-art semantic role labeler and incorporate a much richer set of semantic roles. The performance of the MSR based selectional preferences is evaluated on two distinct datasets: the verbs from the lexical sample task of SENSEVAL-2, and the verbs from a movie script corpus. We show that the MSR based features can indeed improve the per- formance of verb sense disambiguation. 1 Introduction Verb sense disambiguation (VSD) is the task of examining verbs in a given context and specify- ing exactly which sense of each verb is the most appropriate in that context. VSD is a subtask of word sense disambiguation (WSD). Given a verb sense inventory and a set of verb senses, VSD is essentially a classification task (Yarowsky, 2000). VSD has not received much attention in the lit- erature of WSD until recently. Most of WSD sys- tems disambiguate verbs in the same way as nouns using mostly collocation based features, and this has led to rather poor VSD performance on cor- pora such as SENSEVAL-2. One useful but often ignored source of disambiguation information for verbs is the patterns of verb-argument structures. In this paper, we will attempt to capture these pat- terns through the use of selectional preferences. In general, selectional preferences describe the phenomenon that predicating words such as verbs and adjectives tend to favour a small number of noun classes for each of their arguments. For ex- ample, the verb eat (“take in solid food”) tends to select nouns from to the ANIMATED THING class as its EATER role, and nouns from the EDIBLE class as its EATEE role. However, it is possible to extend the concept of selectional preferences to include nouns which function as adjuncts to predicating words. For ex- ample, the verb hit in the sentence I hit him with my fists stands for “deal a blow to, either with the hand or with an instrument”, but in the sentence I hit him with a car, it stands for “to collide with”, with the only difference between the two instances of hit being their manner modifiers. Intuitively, the inclusion of verb adjuncts can enrich the semantic roles (SRs) and provide additional disambiguation information for verbs. Therefore, in the rest of this paper, the concept of “semantic role” will include both the arguments and adjuncts of verbs. All …

Content truncated to 500 words