Identifying senses of particles in verb-particle constructions

Archna Bhatia, Choh Man Teng, James Allen

Abstract: In order to attain broad coverage understanding, a system need not only identify multiword expressions such as verb-particle constructions (VPCs), but must com- pute their meaning. It is not plausible to hand enumerate all possible combinations, although WordNet is an admirable start. This chapter focuses on the identification of senses of particles in VPCs in order to compute the meanings of VPCs – using information obtained from existing lexical resources such as WordNet, and aug- menting it with additional knowledge based on linguistic investigation of VPCs identified in terms of generalizations encoded in the TRIPS ontology. The approach consists of first determining compositionality of a VPC based on the information present in WordNet, and then assigning a relevant sense to the particle in a com- positional VPC based on the sense classes we have identified and encoded in the TRIPS’ computational lexicon. Contributions of the described work are twofold: (1) A discussion of senses of particles in VPCs and corresponding generalizations makes a linguistic contribution. (2) We show how linguistic knowledge can be used to automatically parse sentences containing VPCs and obtain a semantic represen- tation of them. An advantage of the described approach is that VPCs not explicitly found in lexica can be identified and semantically interpreted.