Measuring the compositionality of collocations via word co-occurrence vectors: shared task system description

  • Authors:
  • Alfredo Maldonado-Guerra;Martin Emms

  • Affiliations:
  • Trinity College Dublin, Ireland;Trinity College Dublin, Ireland

  • Venue:
  • DiSCo '11 Proceedings of the Workshop on Distributional Semantics and Compositionality
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

A description of a system for measuring the compositionality of collocations within the framework of the shared task of the Distributional Semantics and Compositionality workshop (DISCo 2011) is presented. The system exploits the intuition that a highly compositional collocation would tend to have a considerable semantic overlap with its constituents (headword and modifier) whereas a collocation with low compositionality would share little semantic content with its constituents. This intuition is operationalised via three configurations that exploit cosine similarity measures to detect the semantic overlap between the collocation and its constituents. The system performs competitively in the task.