Event Categorization beyond Verb Senses

Verb senses are often assumed to distinguish among different conceptual event categories. However, senses misrepresent the number of event categories expressed both within and across languages and event categories may be “named” by more than a word, i.e. a multi-word expression. Determining the nature and number of event categories in an event description requires an understanding of the parameters relevant for categorization. We propose a set of parameters for use in creating a Gold Standard of event categories and apply them to a corpus sample of 2000 sentences across 10 verbs. In doing so, we find an asymmetry between subjects and direct objects in their contributions to distinguishing event categories. We then explore methods of automating event categorization to approximate our Gold Standard through the use of hierarchical clustering and Latent Semantic Analysis (Deerwester et al., 1990).