Bahar Salehi, Afsaneh Fazly, and Mansoor Zolghadri Jahromi
Multiword expressions, non-compositionality, bilingual corpus, word alignment
We investigate the use of a bilingual parallel corpus for automatically extracting non-compositional English Verb Particle Constructions (VPCs). Previous studies only focus on compositionality of VPCs. However, we believe that detecting non-compositionality of VPCs is more important due to their idiomatic meaning. Because of relative sparsity of non-compositional VPCs in the existing datasets, detecting non-compositional VPCs becomes more challenging. We examine the usefulness of five simple features, for our task, using a nearest neighbor classifier, and got the accuracy of 76.7%, and the recall score of 78.6% in detecting non-compositional VPCs.
Important Links:
Go Back
