Brighter than Gold: Figurative Language in User Generated Comparisons

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2008–2018, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Brighter than Gold: Figurative Language in User Generated Comparisons Vlad Niculae and Cristian Danescu-Niculescu-Mizil MPI-SWS Cornell University vniculae@mpi-sws.org, cristian@mpi-sws.org Abstract Comparisons are common linguistic de- vices used to indicate the likeness of two things. Often, this likeness is not meant in the literal sense—for example, “I slept like a log” does not imply that logs ac- tually sleep. In this paper we propose a computational study of figurative compar- isons, or similes. Our starting point is a new large dataset of comparisons extracted from product reviews and annotated for figurativeness. We use this dataset to char- acterize figurative language in naturally occurring comparisons and reveal linguis- tic patterns indicative of this phenomenon. We operationalize these insights and ap- ply them to a new task with high relevance to text understanding: distinguishing be- tween figurative and literal comparisons. Finally, we apply this framework to ex- plore the social context in which figurative language is produced, showing that simi- les are more likely to accompany opinions showing extreme sentiment, and that they are uncommon in reviews deemed helpful. 1 Introduction In argument similes are like songs in love; they describe much, but prove nothing. — Franz Kafka Comparisons are fundamental linguistic devices that express the likeness of two things—be it en- tities, concepts or ideas. Given that their work- ing principle is to emphasize the relation between the shared properties of two arguments (Bredin, 1998), comparisons can synthesize important se- mantic knowledge. Often, comparisons are not meant to be under- stood literally. Figurative comparisons are an im- portant figure of speech called simile. Consider the following two examples paraphrased from Ama- zon product reviews: (1) Sterling is much cheaper than gold. (2) Her voice makes this song shine brighter than gold. In (1) the comparison draws on the relation be- tween the price property shared by the two metals, sterling and gold. While (2) also draws on a com- mon property (brightness), the polysemantic use (vocal timbre vs. light reflection) makes the com- parison figurative. Importantly, there is no general rule separating literal from figurative comparisons. More gen- erally, the distinction between figurative and lit- eral language is blurred and subjective (Hanks, 2006). Multiple criteria for delimiting the two have been proposed in the linguistic and philo- sophical literature—for a comprehensive review, see Shutova (2010)—but they are not without ex- ceptions, and are often hard to operationalize in a computational framework. When considering the specific case of comparisons, such criteria cannot be directly applied. Recently, the simile has received increasing at- tention from linguists and lexicographers (Moon, 2008; Moon, 2011; Hanks, 2013) as it became clearer that similes need to be treated separately from metaphors since they operate on funda- mentally different principles (Bethlehem, 1996). Metaphors are linguistically simple structures hid- ing a complex mapping between two domains, through which many properties are transferred. For example the …

Content truncated to 500 words