Generating similes effortlessly like a Pro: A Style Transfer Approach for Simile Generation

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 6455–6469, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 6455 Generating similes effortlessly like a Pro: A Style Transfer Approach for Simile Generation Tuhin Chakrabarty1,2∗, Smaranda Muresan2,4 and Nanyun Peng1,3 1Information Sciences Institute, University of Southern California 2Department of Computer Science, Columbia University 3Computer Science Department, University of California, Los Angeles 4Data Science Institute, Columbia University {tuhin.chakr, smara}@cs.columbia.edu violetpeng@cs.ucla.edu Abstract Literary tropes, from poetry to stories, are at the crux of human imagination and communi- cation. Figurative language, such as a simile, goes beyond plain expressions to give read- ers new insights and inspirations. We tackle the problem of simile generation. Generat- ing a simile requires proper understanding for effective mapping of properties between two concepts. To this end, we first propose a method to automatically construct a parallel corpus by transforming a large number of sim- iles collected from Reddit to their literal coun- terpart using structured common sense knowl- edge. We then fine-tune a pretrained sequence to sequence model, BART (Lewis et al., 2019), on the literal-simile pairs to generate novel similes given a literal sentence. Experiments show that our approach generates 88% novel similes that do not share properties with the training data. Human evaluation on an inde- pendent set of literal statements shows that our model generates similes better than two liter- ary experts 37%1 of the times, and three base- line systems including a recent metaphor gen- eration model 71%2 of the times when com- pared pairwise.3 We also show how replac- ing literal sentences with similes from our best model in machine generated stories improves evocativeness and leads to better acceptance by human judges. 1 Introduction Comparisons are inherent linguistic devices that ex- press the likeness of two entities, concepts or ideas. When used in a figurative sense, these comparisons are called similes. They are a figure of speech that ∗ The research was conducted when the author was at USC/ISI. 1We average 32.6% and 41.3% for 2 humans. 2We average 82% ,63% and 68% for three baselines. 3The simile in the title is generated by our best model. Input: Generating similes effortlessly, output: Generating similes like a Pro. The bottom of the ocean is dark , scary The bottom of the ocean is dark , like a cave The city was beautiful The city was like a painting Literal Input1 GenSimile1 Literal Input2 GenSimile2 TOPIC VEHICLE PROPERTY Figure 1: Examples of two generated similes GenSim- ile1 and GenSimile2 from their literal inputs. compare two different kind of things, usually with the intent to make the description more emphatic or vivid, being often used in literature and poetry to spark the reader’s imagination (Paul et al., 1970). Take the following two examples: “The city was like a painting”, and “If it falls into the wrong hands it would be as catastrophic as a nuclear bomb.” In the first example, the comparison draws on the im- plicit “beauty” property being shared by …

Content truncated to 500 words