MASIVE: Open-Ended Affective State Identification in English and Spanish
This work addresses the need for more nuanced emotion analysis in NLP, particularly for diverse languages and cultures, though it is incremental in expanding from discrete categories to a broader affective scope.
The paper tackles the problem of identifying a broad set of affective states in text, beyond limited emotion categories, by introducing MASIVE, a dataset with over 1,000 unique affective states in English and Spanish, and showing that smaller finetuned multilingual models outperform larger LLMs on this task, with improvements on existing emotion benchmarks.
In the field of emotion analysis, much NLP research focuses on identifying a limited number of discrete emotion categories, often applied across languages. These basic sets, however, are rarely designed with textual data in mind, and culture, language, and dialect can influence how particular emotions are interpreted. In this work, we broaden our scope to a practically unbounded set of \textit{affective states}, which includes any terms that humans use to describe their experiences of feeling. We collect and publish MASIVE, a dataset of Reddit posts in English and Spanish containing over 1,000 unique affective states each. We then define the new problem of \textit{affective state identification} for language generation models framed as a masked span prediction task. On this task, we find that smaller finetuned multilingual models outperform much larger LLMs, even on region-specific Spanish affective states. Additionally, we show that pretraining on MASIVE improves model performance on existing emotion benchmarks. Finally, through machine translation experiments, we find that native speaker-written data is vital to good performance on this task.