Computational Metaphor Identification

This project involves the development of computational techniques to identify potential conceptual metaphors in written text. We understand metaphor here in the terms describe by Lakoff and Johnson’s Metaphors We Live By, wherein metaphor is not a linguistic or poetic device, but rather is fundamental to human cognition and is evidenced by systematic language use.

For example, consider the words used to describe having an argument: “He attacked her position.” “Your claims are indefensible.” “She obliterated her opponent.” These words evince images of combat and war. Lakoff and Johnson posit the metaphor that argument is war, wherein our experience of having an argument is structured in terms of the experience of being at war. Such metaphors are nearly ubiquitous in everyday interaction–concepts including time, causality, money, and many others are structured metaphorically–and can be seen based on these patterns of language use.

The goal of this line of research is to use computational metaphor identification to draw potential conceptual metaphors to users’ attention and encourage critical thinking, creativity, and reflection about metaphor. We seek to enable encourage asking the following questions: What aspects of the given situation does this metaphor highlight? What other aspects of the situation does it hide? What is a different metaphor that would highlight or hide different aspects of the situation? In this regard, our work on CMI is a break from traditional AI approaches to natural language processing. Rather than using people as a model for how to make computers think, we seek to users computers as a means of making people think.