Dunn, J. (2014). “Measuring Metaphoricity.” In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL 2014). Stroudsburg, PA: Association for Computational Linguistics. 745-751.
Abstract. This paper presents the first computationally-derived scalar measurement of metaphoricity. Each input sentence is given a value between 0 and 1 which represents how metaphoric that sentence is. This measure achieves a correlation of 0.450 (Pearson’s R, p <0.01) with an experimental measure of metaphoricity involving human participants. While far from perfect, this scalar measure of metaphoricity allows different thresholds for metaphoricity so that metaphor identification can be fitted for specific tasks and datasets. When reduced to a binary classification evaluation using the VU Amsterdam Metaphor Corpus, the system achieves an F-Measure of 0.608, slightly lower than the comparable binary classification system’s 0.638 and competitive with existing approaches.