Automatic Identification of Emotions in Texts: Dealing also with their Indirect Modes of Expression

Anonymous

Abstract: This paper presents a model that predicts whether (A) a sentence contains an emotion or not, (B) according to which mode(s) it is expressed, (C) whether the emotion is basic or complex, and (D) which emotional category it is. The originality of the paper lies in the focus on written texts (encyclopedia, novels, newspapers)---as opposed to the more widely studied conversational (sometimes multi-modal) situation---towards the analysis of text complexity in which emotions are one of the analysis factors according to certain works in psycho-linguistics. Within this particular scope, the major contribution of the paper is to introduce the identification of the modes of expression of the emotions, ranging from a direct lexical mode to the most indirect one where emotions are only suggested. The experiments are carried out on French texts for children. They show that the task is rather difficult but leading to acceptable results in comparison to what human annotators agree on. The results also seem to indicate that the task cannot be simply solved by prompting a large language model and requires a specialized model.

Paper Type: long

Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining

Contribution Types: Model analysis & interpretability, Publicly available software and/or pre-trained models

Languages Studied: French