Emotion-Conveying Words in Polish Social Media

Abstract

A growing body of research has attempted to categorize emotions in social media text. However, emphasis on macro-scale trends does not provide a nuanced view of how those classifications are drawn. This article builds on Oberländer’s work on semantic role labeling in sentiment analysis, using their 2020 schema of cue word, target, cause, and experiencer to examine semantic roles in social media posts. Using a corpus of geopolitical Polish-language Facebook data annotated for the presence and intensity of 23 distinct emotions, we generate three hypotheses regarding the actors and emotions in our data. We use two subcorpora of posts containing contempt and admiration, emotions that are roughly bivalent and under-researched in the current literature. Our findings suggest that part-of-speech is not a relevant consideration, and that emotion-conveying words are monovalent–that is, they do not signal multiple emotions in different contexts. We also find differences in the semantic roles towards which our two bivalent emotions are directed, as well as the relative intensity with which they are expressed. We hope this exploratory study can inform future research on the integration of semantic role labeling and sentiment analysis.

Keywords: sentiment analysis; semantic role labeling; emotion; social media