Towards Label-Agnostic Emotion Embeddings

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9231–9249 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 9231 Towards Label-Agnostic Emotion Embeddings Sven Buechel Luise Modersohn Udo Hahn Jena University Language & Information Engineering (JULIE) Lab Friedrich-Schiller-Universität Jena, Fürstengraben 27, 07743 Jena, Germany firstname.lastname@uni-jena.de https://julielab.de Abstract Research in emotion analysis is scattered across different label formats (e.g., polarity types, basic emotion categories, and affective dimensions), linguistic levels (word vs. sen- tence vs. discourse), and, of course, (few well- resourced but much more under-resourced) natural languages and text genres (e.g., prod- uct reviews, tweets, news). The resulting het- erogeneity makes data and software devel- oped under these conflicting constraints hard to compare and challenging to integrate. To resolve this unsatisfactory state of affairs we here propose a training scheme that learns a shared latent representation of emotion inde- pendent from different label formats, natural languages, and even disparate model architec- tures. Experiments on a wide range of datasets indicate that this approach yields the desired interoperability without penalizing prediction quality. Code and data are archived under DOI 10.5281/zenodo.5466068. 1 Introduction Emotion analysis in the field of NLP1 has expe- rienced a remarkable evolution of representation schemes. Starting from the early focus on polar- ity, i.e., the main distinction between positive and negative feelings emerging from natural language utterances (Hatzivassiloglou and McKeown, 1997; Turney and Littman, 2003), the number and variety of label formats, i.e., groups of emotional target variables and their associated value ranges, has been growing rapidly (Bostan and Klinger, 2018; De Bruyne et al., 2020). This development is a double-edged sword though. On the one hand, the wide variety of available label formats allows NLP models to become more informative and richer in expressive power. This gain is because many of the newer representation 1We use “emotion” as an umbrella term for phenomena such as polarity, sentiment, feelings, or affective states. schemes follow well-researched branches of psy- chological theory, such as basic emotion categories or affective dimensions (Ekman, 1992; Russell and Mehrabian, 1977), which offer information comple- mentary to each other (Stevenson et al., 2007). Oth- ers argue that different emotional nuances turn out to be particularly useful for specific targeted down- stream applications (Bollen et al., 2011; Desmet and Hoste, 2013). On the other hand, this proliferation of label formats has led to a severe loss in cross-data com- parability. As Tab. 1 illustrates, the total volume of available gold data is spread not only over dis- tinct languages but also a huge number of emotion annotation schemes. Consequently, comparing or even merging data from different rating studies is often impossible. This, in turn, contributes to the development of an unnecessarily large number of prediction models, each with limited coverage of the full range of human emotion. To escape from these dilemmata, we propose a method that mediates between such different rep- resentation schemes. In contrast to previous work which unified some sources of heterogeneity (see §2), to the best of our knowledge, …

Content truncated to 500 words