When GL meets the corpus: a data-driven investigation of semantic types and coercion phenomena

In this paper we present an analysis of corpus-derived V-Object combinations aiming to provide a data-driven characterization of Semantic Types (STs) and improve our understanding of how types behave compositionally, i.e. how they enter compositional processes and are modulated by them. As a theoretical framework, we adopt the enriched compositional rules and the type system as presented in Pustejovsky (2006). Our main concerns are twofold: i.) first of all, we will show with a specific case-study how a data-driven investigation can shed light on the organization of the type system and on semantic compositional operations affecting types; ii.) starting from the results of this investigation, we intend to propose a general methodology for lexical modeling in which the Generative Lexicon (GL) theory and corpus analysis are deeply interwoven in a process of mutual feeding. In fact, we argue that, if on the one hand corpus data can help to anchor the study of lexical dynamics and type system on empirical evidence, on the other hand GL can provide the crucial interpretative key for corpus data.