Co-occurrence contexts for noun compound interpretation

  • Authors:
  • Diarmuid Ó Séaghdha;Ann Copestake

  • Affiliations:
  • University of Cambridge, Cambridge, United Kingdom;University of Cambridge, Cambridge, United Kingdom

  • Venue:
  • MWE '07 Proceedings of the Workshop on a Broader Perspective on Multiword Expressions
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Contextual information extracted from corpora is frequently used to model semantic similarity. We discuss distinct classes of context types and compare their effectiveness for compound noun interpretation. Contexts corresponding to word-word similarity perform better than contexts corresponding to relation similarity, even when relational co-occurrences are extracted from a much larger corpus. Combining word-similarity and relation-similarity kernels further improves SVM classification performance.