A Comparative Study on Affective Cues in Text Embeddings Across Psychological Emotion Theories
Fabio Ciani, Harald Schweiger 0, Emilia Parada-Cabaleiro, Markus Schedl · 2026
Read A Comparative Study on Affective Cues in Text Embeddings Across Psychological Emotion Theories by Fabio Ciani, Harald Schweiger 0, Emilia Parada-Cabaleiro, Markus Schedl (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-29068.txt.
Description
Ciani, Schweiger, Parada-Cabaleiro and Schedl probe how much emotion theory lives inside the representations of modern text encoders. Encoders compress text into dense vectors that keep meaning, and they are widely used for sentiment and emotion tasks. Whether those vectors capture well-defined psychological theories of emotion, rather than vague positivity, was unclear.
The team tested twelve recently released encoders. Their embeddings fed regression and classification tasks built on three established emotion frameworks, at both word and sentence level. A semantic leakage-prevention technique guarded the word-level tests against inflated scores. At word level, the newest instruction-aware open-weight encoders held as much or more affective information than proprietary rivals. At sentence level, task-tuned and proprietary encoders scored highest on emotion classification. A qualitative look at the latent spaces rounds out the comparison, showing which affective cues each family of encoders encodes and which it misses.
Catalog record
| CREATOR | Fabio Ciani, Harald Schweiger 0, Emilia Parada-Cabaleiro, Markus Schedl |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 80 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.29068