A Study of Compositional Semantic Representation Based on Contextual Information

Project Leader:YAMAGUCHI Masaya
Associate Professor, Department of Corpus Studies, NINJAL
Keywords:natural language processing (NLP), semantic representation, contextual information


Contextual information has been used in several NLP tasks (for example, automatic thesaurus construction, word sense disambiguation) based on the distributional hypothesis that "words that occur in the same contexts tend to have similar meanings." The goal of this project is to develop a theory where by the semantic representation of compositional linguistic elements (e.g. compound verbs) can be derived from the contextual information of the relevant components. The research plan is as follows:

  • a) Analysis of the relationship between the contextual information of compositional linguistic elements and that of their constituents,
  • b) Verification of the distributional hypothesis (including the representations of contextual information),
  • c) Assessment of outputs of our NLP systems from a linguistic point of view.