Textual entailment as an evaluation framework for Metaphor resolution: A proposal

Abstract

We aim to address two complementary deficiencies in Natural Language Processing (NLP) research: (i) Despite the importance and prevalence of metaphor across many discourse genres, and metaphor's many functions, applied NLP has mostly not addressed metaphor understanding. But, conversely, (ii) difficult issues in metaphor understanding have hindered large-scale application, extensive empirical evaluation, and the handling of the true breadth of metaphor types and interactionswith other language phenomena. In this paper, abstracted from a recent grant proposal, a new avenue for addressing both deficiencies and for inspiring new basic research on metaphor is investigated: namely, placing metaphor research within the "Recognizing Textual Entailment" (RTE) task framework for evaluation of semantic processing systems.