Towards Automatic Verb Acquisition from VerbNet for Spoken Dialog Processing

This paper presents experiments on using VerbNet as a resource for understanding unknown verbs encountered by a spoken dialog system. Coverage of unknown verbs in a corpus of spoken dialogs about computer purchasing is assessed, and two methods for automatically integrating representations of verbs found in VerbNet are explored. The first identifies VerbNet classes containing verbs already defined in the system, and generates representations for unknown verbs in those classes, modelled after the existing system representation. The second method generates representations based on VerbNet alone. The second method performs better, but gaps in coverage and differences between the two verb representation systems limit the success of automatic acquisition.