Inducing Positive Perspectives with Text Reframing

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 3682 - 3700 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Inducing Positive Perspectives with Text Reframing Caleb Ziems ⋆† Minzhi Li ⋆⋄ Anthony Zhang † Diyi Yang † †Georgia Institute of Technology {cziems, azhang305, dyang888}@gatech.edu ⋄National University of Singapore li.minzhi@u.nus.edu Abstract Sentiment transfer is one popular example of a text style transfer task, where the goal is to reverse the sentiment polarity of a text. With a sentiment reversal comes also a reversal in meaning. We introduce a different but related task called positive reframing in which we neu- tralize a negative point of view and generate a more positive perspective for the author with- out contradicting the original meaning. Our in- sistence on meaning preservation makes posi- tive reframing a challenging and semantically rich task. To facilitate rapid progress, we intro- duce a large-scale benchmark, POSITIVE PSY- CHOLOGY FRAMES, with 8,349 sentence pairs and 12,755 structured annotations to explain positive reframing in terms of six theoretically- motivated reframing strategies. Then we eval- uate a set of state-of-the-art text style trans- fer models, and conclude by discussing key challenges and directions for future work. To download the data, see https://github. com/GT-SALT/positive-frames 1 Introduction Gratitude is not only the greatest of virtues, but the parent of all the others. — Marcus Tullius Cicero Text style transfer (TST) has received much at- tention from the language technologies community (Hovy, 1987; Jin et al., 2020), where the goal is to change some attribute, like the sentiment of the text, without changing any attribute-independent content (Mir et al., 2019; Fu et al., 2018; Logeswaran et al., 2018). Some TST applications such as de-biasing (Pryzant et al., 2020; Ma et al., 2020) and para- phrasing (den Bercken et al., 2019; Xu et al., 2012) require meaning-preserving transformations, while political leaning (Prabhumoye et al., 2018), senti- ment (Shen et al., 2017; Hu et al., 2017), and topi- cal transfer (Huang et al., 2020) allow for a change ⋆Equal contribution. Figure 1: Positive reframing vs. negative-to-positive sentiment style transfer. in the underlying meaning. For instance, for a negative review, “this was a bland dish,” we can use a sentiment TST model to create a more posi- tive “this was a tasty dish,” by swapping the word bland with tasty. Although the input’s structure and attribute-independent content are preserved, the truth-conditional meaning is clearly altered. In this work, we introduce a closely related task— positive reframing—that differs from sentiment TST in important ways. We effectively reframe negative text by inducing a complementary pos- itive viewpoint (e.g. glass-half-full), which nev- ertheless supports the underlying content of the original sentence. The reframe should implicate rather than contradict the source (see Figure 1), and the transformation should be motivated by the- oretically justified strategies from from positive psychology (Harris et al. 2007; see Section 3). To use the example from before, we could re- frame “this was a bland dish” with the self-affir- …

Content truncated to 500 words