Elamur Research⌕
☰
home / papers / 2026 / Sentiment and Stance in EFL Responses to AI-Generated Environmental Content

Sentiment and Stance in EFL Responses to AI-Generated Environmental Content

Paper · 2026 · Andry Sophocleous

Summary

Most environmental sentiment analysis relies on large-scale automated processing of social media or public data. Far less is known about how people actually interpret and judge environmental content produced by generative AI. This qualitative study addresses that gap with 81 multilingual secondary-school EFL learners in Nicosia, Cyprus. Over several classroom sessions, the learners worked with DALL·E images of four ecological scenes. They described the images, wrote empathic human–animal dialogues, and assessed image descriptions written by ChatGPT 5.2. The author coded the learners' texts using discourse analysis grounded in appraisal and stance theory (Martin & White; Hunston & Thompson) and Stibbe's ecolinguistics. Coding focused on sentiment, stance, responsibility, and how far learners accepted or resisted the AI framing. The learners went well beyond literal description. They voiced emotions, often mixed ones, attributed blame and agency, and pointed out what the AI texts left out about human–environment relations. A comparison with a second model, Theophrastus 5.2, suggests that different systems frame nature differently. The study argues for building qualitative human evaluation into the assessment of generative models.

Key points

  • Participants: 81 learners aged 15–17, drawn from five EFL classes at CEFR B1–B2. Their first languages were Greek, Cypriot Greek, Bulgarian, Romanian, and Slovak. Data were collected between November 2025 and February 2026.
  • Materials: four DALL·E images, depicting a polar bear marooned on a patch of ice, a contaminated river, a flourishing tropical forest, and a garden on a city rooftop. ChatGPT 5.2 produced the descriptions using the same simple prompt that learners received.
  • Task design: an unanalysed ecolinguistic warm-up on nature metaphors in Greek and English, followed by three tasks:
    • image description;
    • dialogues prompted by a whale soundtrack;
    • evaluation of AI descriptions of images the pairs had not seen before.
  • Coding: combined deductive categories (sentiment, stance, responsibility) with inductive ones, including anthropocentric vs. ecocentric framing and care vs. catastrophe narratives. The analysis did not quantify linguistic features.
  • Findings and limitations: responses to the forest image often blended appreciation with worry about future loss. The author acknowledges that the study covers a single school, uses only written data, and is not generalizable.

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

 Institutions

Also on the byline: University of Nicosia.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksLiterature and Literary Theory · 519 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 972 worksGrammatical error correction & learner language · 156 works

Cite

Cite as Andry Sophocleous. 2026. Sentiment and Stance in EFL Responses to AI-Generated Environmental Content. Proceedings of the 2nd Workshop on Ecology, Environment, and Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/42nje3826fwz

@inproceedings{sophocleous2026sentiment,
  title = {{Sentiment and Stance in EFL Responses to AI-Generated Environmental Content}},
  author = {Andry Sophocleous},
  booktitle = {Proceedings of the 2nd Workshop on Ecology, Environment, and Natural Language Processing},
  year = {2026},
  doi = {10.63317/42nje3826fwz},
  url = {https://doi.org/10.63317/42nje3826fwz}
}

References

16 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ibrohim, M. O., Bosco, C., & Basile, V. (2023). Sentiment analysis for the natural environment: A systematic review. ACM Computing Surveys, 56(4), Article 88. https://doi.org/10.1145/3604605 — [doi:10.1145/3604605](https://doi.org/10.1145/3604605) ↗
  2. Bachleda, S., Neuner, F. G., Soroka, S., Guggenheim, L., Fournier, P., & Naurin, E. (2020). Individual-level differences in negativity biases in news selection. Personality and Individual Differences, 155, 109675. https://doi.org/10.1016/j.paid.2019.109675 — [doi:10.1016/j.paid.2019.109675](https://doi.org/10.1016/j.paid.2019.109675) ↗
  3. Martin, J. R., & White, P. R. R. (2005). The language of evaluation: Appraisal in English. Palgrave Macmillan. find
  4. Brader, T. (2006). Campaigning for hearts and minds: How emotional appeals in political ads work. University Of Chicago Press. find
  5. Cook, G. (2015). ‘A pig is a person’ or ‘You can love a fox and hunt it’: Innovation and tradition in the discursive representation of animals. Discourse & Society, 26(5), 587–607. https://doi.org/10.1177/0957926515576639 — [doi:10.1177/0957926515576639](https://doi.org/10.1177/0957926515576639) ↗
  6. Soroka, S., Fournier, P., & Nir, L. (2019). Cross-national evidence of a negativity bias in psychophysiological reactions to news. Proceedings of the National Academy of Sciences, 116(38), 18888–18892. https://doi.org/10.1073/pnas. 1908369116 — [doi:10.1073/pnas](https://doi.org/10.1073/pnas) ↗
  7. Du, X., Kowalski, M., Varde, A. S., de Melo, G., & Taylor, R. W. (2019). Public opinion matters: Mining social media text for environmental management. SIGWEB Newsletter (Autumn), Article 5. find
  8. Stede, M., & Patz, R. (2021). The climate change debate and natural language processing. In Proceedings of the 1st Workshop on NLP for Positive Impact (pp. 8-18). Association for Computational Linguistics. find
  9. Grasso, F., Locci, S., Siragusa, G., & Di Caro, L. (2024). EcoVerse: An annotated Twitter dataset for eco-relevance classification, environmental impact analysis, and stance detection. In N. Calzolari, M.-Y. Kan, V. Hoste, A. Lenci, S. Sakti, & N. Xue (Eds.), Proceedings of the 2024 Joint Internatio… find
  10. Stibbe, A. (2012). Animals erased: Discourse, ecology, and reconnection with the natural world. Wesleyan University Press. find
  11. Stibbe, A. (2015). Ecolinguistics: Language, ecology and the stories we live by. Routledge. find
  12. Thomashow, M. (1995). Ecological identity: Becoming a reflective environmentalist. MIT Press. find
  13. Stibbe, A. (2021). Ecolinguistics: Language, ecology and the stories we live by (2nd ed.). Routledge. find
  14. Hart, P. S., Soroka, S. N., & Hiaeshutter-Rice, D. (2024). Go negative for clicks: Negative sentiment in environmental advocacy emails is associated with increased public engagement. Environmental Communication, 1–7. https://www.tandfonline.com/doi/10.1080/1752 4032.2024.2372579 — [doi:10.1080/1752](https://doi.org/10.1080/1752) ↗
  15. Wong-Parodi, G., & Feygina, I. (2021). Engaging people on climate change: The role of emotional responses. Environmental Communication, 15(5), 571–593. https://doi.org/10.1080/17524032.2020.187105 1 — [doi:10.1080/17524032.2020.187105](https://doi.org/10.1080/17524032.2020.187105) ↗
  16. Hoewe, J., & Ahern, L. (2017). First-person effects of emotional and informational messages in strategic environmental communications campaigns. Environmental Communication, 11(6), 810–820. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sophocleous-2026-sentiment-stance-efl-responses-ai/