Elamur Research⌕
☰
home / papers / 2026 / Can Emotions Signal Gender? Investigating Implicit Cues in Human and LLM Translations of Amazon Reviews

Can Emotions Signal Gender? Investigating Implicit Cues in Human and LLM Translations of Amazon Reviews

Summary

Russian marks the speaker's gender on past-tense verbs, adjectives and participles, so translators must choose a gender for first-person English text even when the source gives none. Earlier work on this data found that product category nudges such choices. This study asks whether the emotion expressed in a source review does the same. The authors use the DiHuTra corpus of English Amazon reviews and their Russian translations by professionals, translation students, GPT-4, Llama and Mistral. They label each source review with a GoEmotions-based RoBERTa classifier and manually annotate first-person gender in each translation. They then test for associations, considering product category as a possible confound.

Reviews tagged as love were rendered with feminine forms more often by both human groups, with small-to-moderate effects. GPT-4 and Mistral showed no such link, but Llama did. Other frequent emotions showed no comparable positive pattern. The authors frame this as exploratory, correlational evidence that particular emotions may act as implicit gender cues. They also present it as a proof of concept for combining emotion classification with gender annotation.

Key points

  • Data: DiHuTra contains 196 English reviews, 14 in each of 14 product categories, with 3-star reviews excluded. LLMs received only a plain "translate into Russian" prompt. Human translators saw no metadata and got no instructions on gender.
  • Masculine default: Masculine labels dominate in every group: 113 for GPT-4, 101 for Llama, 113 for Mistral, 88 for professionals and 91 for students. Feminine labels range from 21 (GPT-4) to 36 (professionals). Only clear m/f reviews were kept, leaving 118–136 per group.
  • Emotion labels: Labels come from SamLowe/roberta-base-go_emotions (27 emotions plus neutral). Masculine-coded emotions such as anger and pride never appeared among the top-1 or top-2 labels, so they were excluded from the analysis.
  • Love versus product category: Professionals and students each gave feminine forms to 9 love-labelled reviews. In both groups, 6 of these came from categories not coded as feminine, which suggests the effect is not simply driven by product category.
  • Limitations:
    • The corpus is very small; for example, Llama's result rests on only 17 love-labelled reviews.
    • Emotion labels are not ground truth: the classifier family reaches only about 0.45 macro-F1 on GoEmotions, and a GPT-4o re-annotation served merely as a plausibility check.
    • The study covers one language pair and one domain.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages · Machine translation

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{manakhimova2026can,
  title = {{Can Emotions Signal Gender? Investigating Implicit Cues in Human and LLM Translations of Amazon Reviews}},
  author = {Manakhimova, Shushen and Lapshinova-Koltunski, Ekaterina},
  booktitle = {Proceedings of the 4th Workshop on Gender-Inclusive Translation Technologies (GITT 2026)},
  year = {2026},
  url = {https://aclanthology.org/2026.gitt-1.6/}
}

References

17 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Agresti, Alan. 1990. Categorical data analysis. A Wiley-Interscience publication. Wiley, New York [u.a.]. find
  2. Brody, Leslie R. and Judith A. Hall. 2008. Gender and emotion in context. In Lewis, Michael, Jeannette M. Haviland-Jones, and Lisa Feldman Barrett, editors, Handbook of Emotions, pages 395–408. The Guilford Press, New York, 3 edition. find
  3. Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates. find
  4. Demszky, Dorottya, Dana Movshovitz-Attias, Jeongwoo Ko, Alan Cowen, Gaurav Nemade, and Sujith Ravi. 2020. GoEmotions: A dataset of fine-grained emotions. In Jurafsky, Dan, Joyce Chai, Natalie Schluter, and Joel Tetreault, editors, Proceedings of the 58th Annual Meeting of the Association for Computa… find
  5. Hackenbuchner, Janiça, Joke Daems, Arda Tezcan, and Aaron Maladry. 2024. You shall know a word’s gender by the company it keeps: Comparing the role of context in human gender assumptions with MT. In Savoldi, Beatrice, Janiça Hackenbuchner, Luisa Bentivogli, Joke Daems, Eva Vanmassenhove, and Jasmi… find
  6. Lapshinova-Koltunski, Ekaterina, Maja Popović, and Maarit Koponen. 2022. DiHuTra: a parallel corpus to analyse differences between human translations. In Moniz, Helena, Lieve Macken, Andrew Rufener, Loı̈c Barrault, Marta R. Costa-jussà, Christophe Declercq, Maarit Koponen, Ellie Kemp, Spyridon Pil… find
  7. Lowe, Sam. 2022. roberta-base-go emotions. https://huggingface.co/SamLowe/roberta -base-go_emotions. ↗
  8. Nesbitt, Muriel N. and Nolan E. Penn. 2000. Gender stereotypes after thirty years: A replication of rosenkrantz, et al. (1968). Psychological Reports, 87:493 – 511. find
  9. Pangakis, Nicholas, Samuel Wolken, and Neil Fasching. 2023. Automated annotation with generative ai requires validation. ArXiv, abs/2306.00176. find
  10. Plant, Ashby, Janet Hyde, Dacher Keltner, and Patricia Devine. 2000. The gender stereotyping of emotions. Psychology of Women Quarterly, 24:81 – 92, 03. find
  11. Plaza-del-Arco, Flor Miriam, Amanda Cercas Curry, Alba Curry, Gavin Abercrombie, and Dirk Hovy. 2024. Angry men, sad women: Large language models reflect gendered stereotypes in emotion attribution. In Ku, Lun-Wei, Andre Martins, and Vivek Srikumar, editors, Proceedings of the 62nd Annual Meeting of… find
  12. Popovic, Maja and Ekaterina Lapshinova-Koltunski. 2024. Gender and bias in Amazon review translations: by humans, MT systems and ChatGPT. In Savoldi, Beatrice, Janiça Hackenbuchner, Luisa Bentivogli, Joke Daems, Eva Vanmassenhove, and Jasmijn Bastings, editors, Proceedings of the 2nd International… find
  13. Rosenkrantz, Paul, Susan Vogel, Helen Bee, Inge Broverman, and Donald Broverman. 1968. Sex-role stereotypes and self-concepts in college students. Journal of Consulting and Clinical Psychology, 32:287–295, 06. find
  14. Savoldi, Beatrice, Marco Gaido, Luisa Bentivogli, Matteo Negri, and Marco Turchi. 2021. Gender bias in machine translation. Transactions of the Association for Computational Linguistics, 9:845–874. find our card
  15. Stanovsky, Gabriel, Noah A. Smith, and Luke Zettlemoyer. 2019. Evaluating gender bias in machine translation. In Korhonen, Anna, David Traum, and Lluı́s Màrquez, editors, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1679–1684, Florence, Italy, July.… find our card
  16. Vanmassenhove, Eva. 2024. Gender bias in machine translation and the era of large language models. ArXiv, abs/2401.10016. — [arXiv:2401.10016](https://arxiv.org/abs/2401.10016) find
  17. Weber, Sabine, Lynn Greschner, and Roman Klinger. 2026. Less is more? the role of demographic author information in emotion classification of ambiguous text. In Piperidis, Stelios, Núria Bel, Henk van den Heuvel, Nancy Ide, Simon Krek, and Antonio Toral, editors, Proceedings of the Fifteenth Langua… find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/manakhimova-2026-can-emotions-signal-gender-investigating/