Learning to Understand Figurative Language: From Similes to Metaphors to Irony

Learning to Understand Figurative Language: From Similes to Metaphors to Irony Tony Veale (TONY.VEALE@UCD.ie) School of Computer Science and Informatics, University College Dublin Belfield, Dublin 4, Ireland Yanfen Hao (YANFEN.HAO@UCD.ie) School of Computer Science and Informatics, University College Dublin Belfield, Dublin 4, Ireland equivalent categorization. Indeed, Roncero et al. (2006) note that similes found on the internet are far more likely than the equivalent metaphors to be accompanied by an explicit explanation, which suggests that similes are less constrained by norms of category structure, and thus less likely than metaphors to be implicitly explained by these norms. Hanks (2004) goes as far as to argue that this non- reliance on category norms makes simile a freer and more creative form of expression than metaphor, since similes can serve as dynamic “triggers for the imagination” without having to appeal either to linguistic conventions or experiential gestalts. Chiappe et al. (2003) demonstrate that metaphoric expressions of a relationship are preferred when the relationship is an apt one, which suggests that it should follow more obviously from the corresponding categorization. These authors also find that aptness correlates strongly with ease of comprehension, and indeed, similes can enhance both their aptness and their comprehensibility by opting for explicit self-explanation: when one says “my left tire is as bald as a bowling ball”, there is no ambiguity whatsoever as to the property that is shared by topic and vehicle, even if baldness is not a literally sensible property of artifacts. Bowdle and Gentner (2005) argue, in a hypothesis they call the career of metaphor, that as metaphors become more conventionalized, they are more likely to be processed as categorizations than as comparisons. This suggests that increased familiarity with a particular metaphoric vehicle allows for greater competence in how the vehicle category is applied and extended to include new members (following Glucksberg, 2001). If, as the career of metaphor hypothesis suggests, there is an “evolutionary path …from comparison to categorization”, it is consistent to also argue for an evolutionary path between simile and metaphor. Certainly, explicit similes of the form “X is as P as Y” indicate that P is a highly salient property of Y, salient enough that Y can be used to exemplify P-ness. If exposed to enough similes of this form, or similes with accompanying explanations (like those reported by Roncero et al., 2006), a cognitive agent can build a detailed conceptual picture of the features P i that define a category Y. Since these will be the most salient and diagnostic features of Y, they can be used to build a category representation of Y that can subsequently be used Abstract Simile is widely viewed as a less sophisticated conceptual device than metaphor, not least because similes are explicitly marked and are frequently more obvious about the meanings they carry. Nonetheless, this lack of sophistication makes simile an ideal basis for acquiring the category-specific knowledge required to understand metaphor. In this paper we describe a computational approach to simile and metaphor that takes the career-of-metaphor hypothesis of Bowdle and Gentner (2005) as its starting point. We describe how the category-defining knowledge required by metaphor can be acquired from exposure to explicit similes, and demonstrate that this knowledge offers a richer and more diagnostic picture of category structure than that acquired from alternate sources. Keywords: metaphor; simile; irony; salient property; category representation. Introduction Figurative language can range from the sublime and the enigmatic to the banal and the obvious. Metaphor, for instance, is widely considered to be the epitome of creative expression, for metaphors often transcend the merely descriptive to yield profoundly enlightening insights; metaphors can be richly allusive, playful and challenging, and open to constant re-interpretation by new readers in new contexts (e.g., see Gibbs, 1994). Because metaphor allows us to view one concept through the prism of another, it is an inherently asymmetric device in which the meaning of a juxtaposition depends crucially on the direction of the information flow (Ortony, 1979). Similes, in contrast, seem an altogether more humble form of expression. The use of the hedge words “like” or “as” marks simile as a diffident figure of speech, which (unlike metaphor) stops short of ascribing category membership to merely draw attention to certain shared properties. So while the metaphor “drug dealers are vampires” challenges our conception of vampires and the criteria needed for membership in this category (see Glucksberg, 2001), the corresponding simile, “drug dealers are like vampires” merely enjoins us to look for common properties which in themselves may be insufficient to support category inclusion. This reluctance to categorize marks simile as a symmetric form of comparison. The hedging and diffident nature of similes might also be said to signal a lack of confidence in the aptness of the