EMOTION DETECTION AND SEMANTICS OF EMOTIVES: DISTRESS AND ANGER IN ANNOTATED TEXT DATASET

Abstract

The article explores the ways of making emotional lexemes semantic description consistent with interpretative intuition of the ordinary language speaker. The research novelty is determined by the fact that it is based on the data retrieved from the emotional assessment of 3920 internet-texts in Russian made by informants via using a specially designed computer interface. When applied this interface, we can aggregate the weight of 8 emotions (distress, enjoyment, anger, surprise, shame, excitement, disgust, fear) in text. Thus, the data we have used for this publication includes two sets of 150 internet-texts assessed by 2000 informants with the highest score of emotions of distress or anger. The scope of the study covers the semantics of two mentioned above lexemes (grust’ and gnev) analyzed through the prism of collective introspection of informants. The article purpose is to discuss the case when a semantic description of emotives is given by an expert, which largely uses “the best texts” of corresponding emotions, according to the collective opinion of informants. Our methods include psycholinguistic experiment, corpus and semantic analysis. The research led us to three main conclusions. Firstly, the semantic descriptions of emotives grust’ and gnev obtained in proposed way represent prototypical scenarios of living an emotion in social context and take into account not only the introspective sensations of an expert-linguist, but the interpretative strategies of language users. Secondly, such semantic explanation provides us with keys for explaining, why machine learning technologies are better at detecting anger than sadness in text. Finally, it creates a precedent in using new technologies for making an ecological semantic description of emotive vocabulary. The research results can find application in emotiology, lexicographic practice and didactics.