Modeling Label Semantics for Predicting Emotional Reactions

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4687–4692 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 4687 Modeling Label Semantics for Predicting Emotional Reactions Radhika Gaonkar, Heeyoung Kwon, Mohaddeseh Bastan, Niranjan Balasubramanian Stony Brook University, Stony Brook, New York {rgaonkar, heekwon, mbastan, niranjan}@cs.stonybrook.edu Nathanael Chambers US Naval Academy, Annapolis, MD nchamber@usna.edu Abstract Predicting how events induce emotions in the characters of a story is typically seen as a stan- dard multi-label classification task, which usu- ally treats labels as anonymous classes to pre- dict. They ignore information that may be con- veyed by the emotion labels themselves. We propose that the semantics of emotion labels can guide a model’s attention when represent- ing the input story. Further, we observe that the emotions evoked by an event are often re- lated: an event that evokes joy is unlikely to also evoke sadness. In this work, we explicitly model label classes via label embeddings, and add mechanisms that track label-label correla- tions both during training and inference. We also introduce a new semi-supervision strategy that regularizes for the correlations on unla- beled data. Our empirical evaluations show that modeling label semantics yields consis- tent benefits, and we advance the state-of-the- art on an emotion inference task. 1 Introduction Understanding how events in a story affect the char- acters involved is an integral part of narrative un- derstanding. Rashkin et al. (2018) introduced an emotion inference task on a subset of the ROCSto- ries dataset (Mostafazadeh et al., 2016), labeling entities with the emotions they experience from the short story contexts. Previous work on this and re- lated tasks typically frame them as multi-label clas- sification problems. The standard approach uses an encoder that produces a representation of the tar- get event along with the surrounding story events, and then pushes it through a classification layer to predict the possible emotion labels (Rashkin et al., 2018; Wang et al., 2018). This classification framework ignores the seman- tics of the emotions themselves. Each emotion label (e.g., joy) is just a binary prediction. How- ever, consider the sentence, “Danielle was really short on money”. The emotional reaction is FEAR of being short on money. First, if a model had lexi- cal foreknowledge of “fear”, we should expect an improved ability to decide if a target event evokes FEAR. Second, such a model might represent rela- tionships between the emotions themselves. For ex- ample, an event that evokes FEAR is likely to evoke SADNESS and unlikely to evoke JOY. When pre- vious models frame this as binary label prediction, they miss out on ways to leverage label semantics. In this work, we show that explicitly modeling label semantics improves emotion inference. We describe three main contributions1. First, we show how to use embeddings as the label semantics rep- resentation. We then propose a label attention net- work that produces label-informed representations of the event and the story context to improve pre- diction accuracy. Second, we add mechanisms that …

Content truncated to 500 words