In Search of Semantic Compositionality in Vector Spaces

  • Authors:
  • Eugenie Giesbrecht

  • Affiliations:
  • FZI Research Center for Information Technology, Karlsruhe, Germany

  • Venue:
  • ICCS '09 Proceedings of the 17th International Conference on Conceptual Structures: Conceptual Structures: Leveraging Semantic Technologies
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

In spite of the widespread usage of geometric models of meaning in computational linguistics and information retrieval research, they have been until recently mostly utilized for modeling lexical meaning. The ability to deal with concept combination, however, is the essential capacity of human language, and any semantic theory should be able to handle it. Making use of Word Space Models (Schütze 1998) and Random Indexing (Sahlgren 2005), we explore the hypothesis that compositional meaning can be captured in such models by adopting a number of mathematical operations for vector composition (summation, component product, tensor product and convolution) to model semantic composition in a multiword unit identification task.