A Comparative Study on Affective Cues in Text Embeddings Across Psychological Emotion Theories

2026-06-27

AI Summary

It remains unclear whether existing text encoders effectively capture the semantic structure of emotion as defined by psychological theories. This work presents the first systematic evaluation of twelve prominent open- and closed-source, general-purpose and task-tuned encoders in their ability to represent lexical- and sentence-level affective semantics under three canonical emotion theory frameworks. Employing both regression and classification probing tasks, the study introduces a novel semantic data leakage mitigation mechanism to enhance the robustness of word-level assessments. Results reveal that state-of-the-art instruction-aware open-source models preserve lexical emotional information on par with or even surpassing closed-source counterparts, while task-tuned and closed-source encoders achieve superior performance in sentence-level emotion classification.

Abstract

Text encoders are known for their utility in natural language processing, as they are able to efficiently compress inputs into dense vectors while preserving semantics. These models have been applied to affective computing, in particular to help with solving sentiment analysis and emotion recognition tasks. Nevertheless, it remains unclear to what extent the latent representations produced by modern text encoders capture well-defined psychological theories of affect. In this work, we investigate the affective capabilities of twelve recently released text encoders by probing their generated embeddings as input features for solving regression and classification tasks across three established emotion frameworks, using both word- and sentence-level data. Additionally, we apply a semantic data-leakage prevention technique to improve robustness in word-level evaluations. Our main findings show that the latent manifolds of the latest instruction-aware open-weight encoders enclose an equal or even a larger amount of affective information in comparison with proprietary counterparts when evaluated at word level. In contrast, embeddings of task-tuned and proprietary encoders reach the highest scores on sentence-level affective classification. Furthermore, a qualitative analysis of latent representations and their encoded affective cues is provided.

Problem

• affective computing
• text embeddings
• emotion recognition
• psychological emotion theories
• latent representations

Innovation

• affective computing
• text embeddings
• emotion theories
• semantic leakage prevention
• instruction-aware encoders