Semantics is the study of the meaning of words, constructions, and utterances. The Past few years have seen great progress in the field of semantic analysis through work in different sub-areas such as lexical semantics and semantic parsing. As part of this progress, new knowledge-based resources have been developed in different NLP communities which have opened the field to many new solutions to the problem of understanding language. Continuous development of the WordNet project, the FrameNet Project at UC-Berkeley and the Prop Bank Project at University of Pennsylvania are pivotal examples of these resources.
In this talk, after a brief introduction about some of these semantic resources and relevant semantic concepts, I am going to present the way I have benefited from these resources to minimize my need of hand-labeled data for defining semantic constraints in a NLP application. The application is a classifier which distinguishes between literal and non-literal expressions and uses frame level semantic constraints to classify a sentence. In my pilot study I have chosen 3 sample verbs which are widely used both in literal and non-literal ways. The testing which is done on 200 sentences shows a promising improvement over a baseline that encourages further study on the task.