---
title: "Greench-v1: distilling SLMs on Greenwashing Detection (2026)"
description: "Checking ESG reports for greenwashing usually requires manual review, which is slow, expensive and inconsistent. 11 references."
authors: ["Federico Raspanti", "Alessandro Pietro Bardelli Bardelli", "Simona Scala", "İrem Demirtaş", "Marilena Di Bari", "Michele Filannino"]
year: 2026
venue: "Proceedings of the 2nd Workshop on Ecology, Environment, and Natural Language Processing"
source: "https://aclanthology.org/2026.nlp4ecology-1.8/"
doi: "10.63317/2ngqn2gr78du"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/raspanti-2026-greench-v1-distilling-slms-greenwashing/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/raspanti-2026-greench-v1-distilling-slms-greenwashing/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/) / Greench-v1: distilling SLMs on Greenwashing Detection

# Greench-v1: distilling SLMs on Greenwashing Detection

**Authors:** Federico Raspanti, Alessandro Pietro Bardelli Bardelli, Simona Scala, İrem Demirtaş, Marilena Di Bari and Michele Filannino · **Year:** 2026 · **Venue:** [Proceedings of the 2nd Workshop on Ecology, Environment, and Natural Language Processing](https://md.research.elamur.ai/free/en/venues/nlp4ecology/) ([Workshop on Ecology, Environment, and Natural Language Processing 2026](https://md.research.elamur.ai/free/en/venues/nlp4ecology/2026/)) · **Publisher:** [European Language Resources Association (ELRA)](https://md.research.elamur.ai/free/en/publishers/european-language-resources-association-elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.nlp4ecology-1.8/) · [PDF](https://aclanthology.org/2026.nlp4ecology-1.8.pdf) · [DOI 10.63317/2ngqn2gr78du](https://doi.org/10.63317/2ngqn2gr78du) · [OpenAlex](https://openalex.org/W7166586257)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary

Checking ESG reports for greenwashing usually requires manual review, which is slow, expensive and inconsistent. Simona Scala, Federico Raspanti and colleagues at Prometeia present Greench-v1, a small Qwen3-4B model that reads ESG documents paragraph by paragraph. For each paragraph it assigns one of three labels (Greenwashing Alert, No Greenwashing, Not Relevant) and gives a short justification tied to that paragraph, so that auditors can prioritise their review. The training data comes from ClimateBERT datasets. The authors balanced the classes by paraphrasing the rare positive examples and used a GPT model to write rationales for every example. They compared four training setups: hard distillation, soft distillation from a domain-tuned Qwen3-14B teacher, GRPO, and GRPO started from the hard-distilled checkpoint. All tuned variants beat zero-shot baselines, and soft distillation and GRPO gave the largest gains. The deployed version uses soft distillation behind a web interface that takes PDFs and exports results. The work shows that a compact, locally hosted model can handle first-pass screening at scale while staying transparent and reproducible, unlike black-box API models.

## Key points

- **Corpus construction:**
  - 1,320 paragraphs were drawn from `climatebert/climate_specificity`.
  - 660 non-claim paragraphs from `climatebert/environmental_claims` were added as the Not Relevant class.
  - Each positive example was paraphrased three times.
  - The final corpus has about 2,000 paragraphs, split 80/20 with stratification (about 1,600 for training and 400 for evaluation).
- **Rationale generation:** Justifications were produced by a GPT model. The abstract says GPT-4o, while the methods section says GPT-4.1.
- **Supplementary data:** A separate set of paragraphs from 30 real ESG reports, labelled by GPT-4.1, was used only for the Hard Distillation + GRPO regime.
- **Main results:** Measured by weighted F1 on the Greenwashing Alert class, against untuned baselines:
  - soft distillation improved the score by 36.7% (relative);
  - GRPO, with an exact-match reward, improved it by 49.0%.
- **Main error type:** False positives, mainly on two kinds of paragraphs:
  - text that combines concrete actions with forward-looking statements;
  - operational disclosures (for example, ESG staffing figures) that give no performance metrics.
- **Planned future work:** rewards based on rationale quality, testing on external benchmarks, and multimodal (image and chart) analysis.

*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: [Prometeia](https://ror.org/02jsytr13).

📚 Shelves (8): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [💼 Business, Management and Accounting](https://md.research.elamur.ai/free/en/shelves/business-management-and-accounting/) · 32 works · [📈 Economics, Econometrics and Finance](https://md.research.elamur.ai/free/en/shelves/economics-econometrics-and-finance/) · 17 works · [📈 Finance](https://md.research.elamur.ai/free/en/shelves/economics-econometrics-and-finance/finance/) · 5 works · [💼 Strategy and Management](https://md.research.elamur.ai/free/en/shelves/business-management-and-accounting/strategy-and-management/) · 4 works

## Cite

Cite as: Federico Raspanti, Alessandro Pietro Bardelli Bardelli, Simona Scala, İrem Demirtaş, Marilena Di Bari and Michele Filannino. 2026. Greench-v1: distilling SLMs on Greenwashing Detection. 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.

```bibtex
@inproceedings{raspanti2026greench,
  title = {{Greench-v1: distilling SLMs on Greenwashing Detection}},
  author = {Federico Raspanti, Alessandro Pietro Bardelli Bardelli, Simona Scala, İrem Demirtaş, Marilena {Di Bari} and Michele Filannino},
  booktitle = {Proceedings of the 2nd Workshop on Ecology, Environment, and Natural Language Processing},
  year = {2026},
  doi = {10.63317/2ngqn2gr78du},
  url = {https://doi.org/10.63317/2ngqn2gr78du}
}
```

- [← previous GreekCommonGen: A Benchmark for Evaluating Generative Commonsense Reasoning in Greek](https://md.research.elamur.ai/free/en/papers/stamopoulos-2026-greekcommongen-benchmark-evaluating-generative-commonsense/)
- [next → GREYC at CRF Filling 2026: Rewrite Before You Extract - Rewriting Clinical Notes for Automated CRF](https://md.research.elamur.ai/free/en/papers/melgarejo-2026-greyc-crf-filling-2026-rewrite/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/)

## References

11 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. ClimateBERT: A pretrained language model for climate-related text. [find](https://scholar.google.com/scholar?q=ClimateBERT%3A+A+pretrained+language+model+for+climate%2Drelated+text%2E)
2. Julia Anna Bingler, Mathias Kraus, Markus Leippold, and Nicolas Webersinke. 2024. How cheap talk in climate disclosures relates to climate initiatives, corporate emissions, and reputation risk. Journal of Banking Finance, 164:107191. [find](https://scholar.google.com/scholar?q=Julia+Anna+Bingler%2C+Mathias+Kraus%2C+Markus+Leippold%2C+and+Nicolas+Webersinke%2E+2024%2E+How+cheap+talk+in+climate+disclosures+relates+to+climate+initiatives%2C+corporate+emissions%2C+and+reputation+risk%2E+Journa)
3. M Birti, F Osborne, and A Maurino. 2025. Optimizing large language models for esg activity detection in financial texts. arxiv. arXiv preprint arXiv:2502.21112. — [arXiv:2502.21112](https://arxiv.org/abs/2502.21112) [find](https://scholar.google.com/scholar?q=M+Birti%2C+F+Osborne%2C+and+A+Maurino%2E+2025%2E+Optimizing+large+language+models+for+esg+activity+detection+in+financial+texts%2E+arxiv%2E+arXiv+preprint+arXiv%3A2502%2E21112%2E+%E2%80%94+%5BarXiv%3A2502%2E21112%5D%28https%3A%2F%2Farxiv%2Eorg%2F)
4. Tom Calamai, Oana Balalau, Théo Le Guenedal, and Fabian M Suchanek. 2025. Corporate greenwashing detection in text–a survey. arXiv e-prints, pages arXiv–2502. [find](https://scholar.google.com/scholar?q=Tom+Calamai%2C+Oana+Balalau%2C+Th%C3%A9o+Le+Guenedal%2C+and+Fabian+M+Suchanek%2E+2025%2E+Corporate+greenwashing+detection+in+text%E2%80%93a+survey%2E+arXiv+e%2Dprints%2C+pages+arXiv%E2%80%932502%2E)
5. Marianne Chuang, Gabriel Chuang, Cheryl Chuang, and John Chuang. 2025. Judging it, washing it: Scoring and greenwashing corporate climate disclosures using large language models. In Proceedings of the 2nd Workshop on Natural Language Processing Meets Climate Change (ClimateNLP 2025), pages 17–31, Vi… [find](https://scholar.google.com/scholar?q=Marianne+Chuang%2C+Gabriel+Chuang%2C+Cheryl+Chuang%2C+and+John+Chuang%2E+2025%2E+Judging+it%2C+washing+it%3A+Scoring+and+greenwashing+corporate+climate+disclosures+using+large+language+models%2E+In+Proceedings+of+the)
6. Nina Gorovaia and Michalis Makrominas. 2025. Identifying greenwashing in corporate-social responsibility reports using natural-language processing. European Financial Management, 31(1):427–462. [find](https://scholar.google.com/scholar?q=Nina+Gorovaia+and+Michalis+Makrominas%2E+2025%2E+Identifying+greenwashing+in+corporate%2Dsocial+responsibility+reports+using+natural%2Dlanguage+processing%2E+European+Financial+Management%2C+31%281%29%3A427%E2%80%93462%2E)
7. Cheng-Yu Hsieh, Chun-Liang Li, Chih-Kuan Yeh, Hootan Nakhost, Yasuhisa Fujii, Alex Ratner, Ranjay Krishna, Chen-Yu Lee, and Tomas Pfister. 2023. [Distilling step-by-step! outperforming larger language models with less training data and smaller model sizes](https://md.research.elamur.ai/free/en/papers/hsieh-2023-distilling-step-step-outperforming-larger/). In Findings of the Association for Computati…
8. Inna Livytska. 2019. The use of hedging in research articles on applied linguistics. Journal of language and cultural education, 7(1):35–53. [find](https://scholar.google.com/scholar?q=Inna+Livytska%2E+2019%2E+The+use+of+hedging+in+research+articles+on+applied+linguistics%2E+Journal+of+language+and+cultural+education%2C+7%281%29%3A35%E2%80%9353%2E)
9. Protima Nomo Sudro and Shreya Mukhopadhyay. 2025. Greenwashing detection with causal explanation: A novel multi-layered approach. In Women in Machine Learning Workshop@ NeurIPS 2025. [find](https://scholar.google.com/scholar?q=Protima+Nomo+Sudro+and+Shreya+Mukhopadhyay%2E+2025%2E+Greenwashing+detection+with+causal+explanation%3A+A+novel+multi%2Dlayered+approach%2E+In+Women+in+Machine+Learning+Workshop%40+NeurIPS+2025%2E)
10. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. Bertscore: Evaluating text generation with bert. 10. [find](https://scholar.google.com/scholar?q=Tianyi+Zhang%2C+Varsha+Kishore%2C+Felix+Wu%2C+Kilian+Q%2E+Weinberger%2C+and+Yoav+Artzi%2E+2020%2E+Bertscore%3A+Evaluating+text+generation+with+bert%2E+10%2E)
11. Language Resource References Webersinke, Nicolas and Kraus, Mathias and Bingler, Julia Anna and Leippold, Markus. 2021. [find](https://scholar.google.com/scholar?q=Language+Resource+References+Webersinke%2C+Nicolas+and+Kraus%2C+Mathias+and+Bingler%2C+Julia+Anna+and+Leippold%2C+Markus%2E+2021%2E)

*Updated 1 November 2026*