Output details

Data and models for metonymy resolution

<22>Reports the first-ever metonymy resolution competition; presents the first quality-assured corpus-based dataset for this task; clearly demonstrates the inadequacy of the historically widely used selectional-restriction approach for metonymy recognition. The dataset was highly novel for figurative language resolution and an essential step in objective system evaluation. It has become the de facto standard for metonymy resolution, evidenced by (i) the work of 5 international competition-participants from industry and academia; (ii) subsequent work using our dataset (e.g., Nastase, EMNLP-2009, and a follow on metonymy competition organised by Pustejovsky et al., Semeval-2). Appeared in Special Issue on Semantic Evaluations (25% acceptance).