Automatic metaphor detection using constructions and frames

Abstract

This paper presents MetaNet’s automatic metaphor detection system that applies theoretical principles from construction grammar, frame semantics, and recent developments in conceptual metaphor theory, including the theory of cascades (Lakoff 2014). The system has achieved relative success in identifying metaphorical expressions for a range of target domains from large corpora and holds promise as a useful tool for corpus-based study of metaphor. The detection system relies on MetaNet’s conceptual network of frames and metaphors as a computational resource for its functionality, and improves automatically as the representations stored in the network are built up. In addition, because of its theoretically principled design the system’s level of accuracy at identifying metaphorical expressions provides feedback to linguists about the accuracy of the frame and metaphor analyses in the network.