Elamur Research⌕
☰
home / papers / 2026 / Better as Generators Than Classifiers: Leveraging LLMs and Synthetic Data for Low-Resource Multilingual Classification

Better as Generators Than Classifiers: Leveraging LLMs and Synthetic Data for Low-Resource Multilingual Classification

Paper · 2026 · Branislav Pecher, Jan Cegin, Robert Belanec et al.

Summary

The paper asks whether a large multilingual LLM is more useful for low-resource classification as a classifier or as a source of training data for smaller models. The authors treat synthetic data generation as data-driven distillation. One strong multilingual LLM generates labelled examples for 11 typologically diverse languages (low-, medium- and high-resource) across 4 classification tasks. The generator is guided by 10 real examples and filters its own output through self-revision. The synthetic sets are then used in three ways: fine-tuning a multilingual encoder, serving as in-context examples for three compact LLMs, and LoRA instruction-tuning of those LLMs. All variants are compared against the generator's own zero-shot classification. Small numbers of synthetic samples are often enough for the smaller models to equal or beat the generator, and the gains are clearest in low-resource languages. In high-resource languages the gains depend more on the task. This matters for under-resourced languages such as Ukrainian, because a large model is then needed only once, to generate data, and cheap models can handle inference.

Key points

  • Generation setup: the prompt includes 10 randomly drawn human examples. The LLM then judges each candidate, for example checking that it is written only in the target language. A subset is checked manually. Generation continues until each label has 200 samples.
  • Sample efficiency: about 20 synthetic samples on average were enough for smaller models to reach or exceed the generator in low- and medium-resource languages. High-resource languages needed about 100.
  • Comparison with human data: synthetic data performs about as well as human-labelled data only at small sizes (around 50 samples). Beyond that it plateaus, because the generated samples are less diverse and less informative. Fine-tuning on synthetic data was also more sensitive to hyperparameters.
  • Scale and cost: the study ran over 30,000 fine-tuning and instruction-tuning runs. This took roughly 1,000 GPU-hours on A100 80GB cards, with estimated emissions of about 108 kgCO2eq. The shared prompt was engineered on LLaMA-3.1.
  • Code: released at github.com/kinit-sk/multilingual-classifiers-not-generators. The authors say the synthetic samples are included in the supplementary material.

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

🎓Brno University of TechnologyBrno · Czechia11 works · 32,326 citations🤝Kempelen Institute of Intelligent TechnologiesBratislava · Slovakia11 works · 82 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,894 worksMultilingual & cross-lingual NLP · 882 worksLow-resource languages · 287 works

Cite

Cite as Branislav Pecher, Jan Cegin, Robert Belanec, Ivan Srba, Jakub Simko and Maria Bielikova. 2026. Better as Generators Than Classifiers: Leveraging LLMs and Synthetic Data for Low-Resource Multilingual Classification. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.148

@inproceedings{pecher2026better,
  title = {{Better as Generators Than Classifiers: Leveraging LLMs and Synthetic Data for Low-Resource Multilingual Classification}},
  author = {Branislav Pecher, Jan Cegin, Robert Belanec, Ivan Srba, Jakub Simko and Maria Bielikova},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.148},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.148}
}

References

53 references, 28 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. David Ifeoluwa Adelani, Hannah Liu, Xiaoyu Shen, Nikita Vassilyev, Jesujoba O. Alabi, Yanke Mao, Haonan Gao, and En-Shiun Annie Lee. 2024. SIB-200: A simple, inclusive, and big evaluation dataset for topic classification in 200+ languages and dialects. In Proceedings of the 18th Conference of the Eu…
  2. Tatiana Anikina, Jan Cegin, Jakub Simko, and Simon Ostermann. 2025. A rigorous evaluation of llm data generation strategies for low-resource languages. Preprint, arXiv:2506.12158.
  3. Akari Asai, Sneha Kudugunta, Xinyan Yu, Terra Blevins, Hila Gonen, Machel Reid, Yulia Tsvetkov, Sebastian Ruder, and Hannaneh Hajishirzi. 2024. BUFFET: Benchmarking large language models for few-shot cross-lingual transfer. In Proceedings of the 2024 Conference of the North American Chapter of the A…
  4. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. Language models are few-shot learners. Advances in Neural Information Processing Systems, 33:1877–1901.
  5. Jan Cegin, Branislav Pecher, Jakub Simko, Ivan Srba, Maria Bielikova, and Peter Brusilovsky. 2024a. Effects of diversity incentives on sample diversity and downstream model performance in LLM-based text augmentation. In Proceedings of the 62nd Annual Meeting of the Association for Computational Ling… find
  6. Jan Cegin, Branislav Pecher, Jakub Simko, Ivan Srba, Maria Bielikova, and Peter Brusilovsky. 2024b. Use random selection for now: Investigation of few-shot selection strategies in llm-based text augmentation for classification. Preprint, arXiv:2410.10756. find
  7. Jan Cegin, Jakub Simko, and Peter Brusilovsky. 2023. ChatGPT to replace crowdsourcing of paraphrases for intent classification: Higher diversity and comparable model robustness. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 1889–1905, Singapore. Ass… find
  8. Yi-Ling Chung, Aurora Cobo, and Pablo Serna. 2025. Beyond translation: Llm-based data generation for multilingual fact-checking. arXiv preprint arXiv:2502.15419. — [arXiv:2502.15419](https://arxiv.org/abs/2502.15419) find
  9. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association fo…
  10. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Zhiyong Wu, Baobao Chang, Xu Sun, Jingjing Xu, and Zhifang Sui. 2022. A survey for in-context learning. arXiv preprint arXiv:2301.00234.
  11. Luyang Fang, Gyeong-Geon Lee, and Xiaoming Zhai. 2023. Using gpt-4 to augment unbalanced data for automatic scoring. Preprint, arXiv:2310.18365. — [arXiv:2310.18365](https://arxiv.org/abs/2310.18365) find
  12. Jack FitzGerald, Christopher Hench, Charith Peris, Scott Mackie, Kay Rottmann, Ana Sanchez, Aaron Nash, Liam Urbach, Vishesh Kakarala, Richa Singh, Swetha Ranganath, Laurie Crist, Misha Britan, Wouter Leeuwis, Gokhan Tur, and Prem Natarajan. 2023. MASSIVE: A 1M-example multilingual natural language… find
  13. Tianyu Gao, Adam Fisch, and Danqi Chen. 2021. Making pre-trained language models better few-shot learners. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pa…
  14. Parker Glenn, Alolika Gon, Nikhil Kohli, Sihan Zha, Parag Pravin Dakle, and Preethi Raghavan. 2023. Jetsons at the FinNLP-2023: Using synthetic data and transfer learning for multilingual ESG issue classification. In Proceedings of the Fifth Workshop on Financial Technology and Natural Language Proc… find
  15. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  16. Odd Erik Gundersen, Kevin Coakley, and Christine Kirkpatrick. 2022. Sources of irreproducibility in machine learning: A review. arXiv preprint arXiv:2204.07610. find
  17. Yanzhu Guo, Guokan Shang, and Chloé Clavel. 2024. Benchmarking linguistic diversity of large language models. arXiv preprint arXiv:2412.10271.
  18. Daniil Gurgurov, Mareike Hartmann, and Simon Ostermann. 2024. Adapting multilingual LLMs to low-resource languages with knowledge graphs via adapters. In Proceedings of the 1st Workshop on Knowledge Graphs and Large Language Models (KaLLM 2024), pages 63–74, Bangkok, Thailand. Association for Comput…
  19. Daniil Gurgurov, Rishu Kumar, and Simon Ostermann. 2025. GrEmLIn: A repository of green baseline embeddings for 87 low-resource languages injected with multilingual graph knowledge. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 1204–1221, Albuquerque, New Mexico. As… find
  20. Evan Hernandez, Diwakar Mahajan, Jonas Wulff, Micah J Smith, Zachary Ziegler, Daniel Nadler, Peter Szolovits, Alistair Johnson, Emily Alsentzer, and 1 others. 2023. Do we still need clinical language models? In Conference on Health, Inference, and Learning, pages 578–597. PMLR. find
  21. SU Hongjin, Jungo Kasai, Chen Henry Wu, Weijia Shi, Tianlu Wang, Jiayi Xin, Rui Zhang, Mari Ostendorf, Luke Zettlemoyer, Noah A Smith, and 1 others. 2022. Selective annotation makes language models better few-shot learners. In The Eleventh International Conference on Learning Representations. find
  22. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, and 1 others. 2022. Lora: Low-rank adaptation of large language models. ICLR, 1(2):3.
  23. Alina Kramchaninova and Arne Defauw. 2022. Synthetic data generation for multilingual domain-adaptable question answering systems. In Proceedings of the 23rd Annual Conference of the European Association for Machine Translation, pages 151–160, Ghent, Belgium. European Association for Machine Transla… find
  24. Alexandre Lacoste, Alexandra Luccioni, Victor Schmidt, and Thomas Dandres. 2019. Quantifying the carbon emissions of machine learning. arXiv preprint arXiv:1910.09700.
  25. Linlin Liu, Bosheng Ding, Lidong Bing, Shafiq Joty, Luo Si, and Chunyan Miao. 2021. MulDA: A multilingual data augmentation framework for low-resource cross-lingual NER. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Confer… find
  26. Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig. 2023. Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing. ACM Comput. Surv., 55(9).
  27. Qijiong Liu, Nuo Chen, Tetsuya Sakai, and Xiao-Ming Wu. 2024. Once: Boosting content-based recommendation with both open- and closed-source large language models. In Proceedings of the 17th ACM International Conference on Web Search and Data Mining, WSDM ’24, page 452–461, New York, NY, USA. Associa… find
  28. Robert Logan IV, Ivana Balazevic, Eric Wallace, Fabio Petroni, Sameer Singh, and Sebastian Riedel. 2022. Cutting down on prompts and parameters: Simple few-shot learning with language models. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2824–2835, Dublin, Ireland. AC…
  29. Yubo Ma, Yixin Cao, Yong Hong, and Aixin Sun. 2023. Large language model is not a good few-shot information extractor, but a good reranker for hard samples! In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 10572–10601, Singapore. ACL.
  30. J. A. Meaney, Steven Wilson, Luis Chiruzzo, Adam Lopez, and Walid Magdy. 2021. find
  31. SemEval 2021 task 7: HaHackathon, detecting and rating humor and offense. In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 105–119, Online. Association for Computational Linguistics.
  32. Kurt Micallef and Claudia Borg. 2025. Melabenchv1: Benchmarking large language models against smaller fine-tuned models for low-resource maltese nlp. arXiv preprint arXiv:2506.04385.
  33. Sepideh Mollanorozy, Marc Tanti, and Malvina Nissim. 2023. Cross-lingual transfer learning with Persian. In Proceedings of the 5th Workshop on Research in Computational Linguistic Typology and Multilingual NLP, pages 89–95, Dubrovnik, Croatia. Association for Computational Linguistics. find
  34. Marius Mosbach, Tiago Pimentel, Shauli Ravfogel, Dietrich Klakow, and Yanai Elazar. 2023. Few-shot fine-tuning vs. in-context learning: A fair comparison and evaluation. In Findings of the Association for Computational Linguistics: ACL 2023, pages 12284– 12314, Toronto, Canada. ACL.
  35. Amani Namboori, Shivam Sadashiv Mangale, Andy Rosenbaum, and Saleh Soltan. 2023. Gemquad: Generating multilingual question answering datasets from large language models using few shot learning. In NeurIPS 2023 Workshop on Synthetic Data Generation with Generative AI. find
  36. Aytuğ Onan. 2023. Srl-aco: A text augmentation framework based on semantic role labeling and ant colony optimization. Journal of King Saud University - Computer and Information Sciences, 35(7):101611. find
  37. Branislav Pecher, Ivan Srba, and Maria Bielikova. 2024a. Comparing specialised small and general large language models on text classification: 100 labelled samples to achieve break-even performance. arXiv preprint arXiv:2402.12819.
  38. Branislav Pecher, Ivan Srba, and Maria Bielikova. 2024b. A survey on stability of learning with limited labelled data and its sensitivity to the effects of randomness. ACM Computing Surveys, 57(1). find
  39. Frédéric Piedboeuf and Philippe Langlais. 2023. Is ChatGPT the ultimate data augmentation algorithm? In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 15606–15615, Singapore. Association for Computational Linguistics. find
  40. Salsabila Zahirah Pranida, Rifo Ahmad Genadi, and Fajri Koto. 2025. Synthetic data generation for culturally nuanced commonsense reasoning in low-resource languages. Preprint, arXiv:2502.12932.
  41. Raul Puri, Ryan Spring, Mohammad Shoeybi, Mostofa Patwary, and Bryan Catanzaro. 2020. Training question answering models from synthetic data. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 5811–5826, Online. Association for Computational Ling…
  42. Rifki Afina Putri, Faiz Ghifari Haznitrama, Dea Adhista, and Alice Oh. 2024. Can LLM generate culturally relevant commonsense QA data? case study in Indonesian and Sundanese. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 20571–20590, Miami, Florida,…
  43. Chengwei Qin, Aston Zhang, Zhuosheng Zhang, Jiaao Chen, Michihiro Yasunaga, and Diyi Yang. 2023. Is ChatGPT a general-purpose natural language processing task solver? In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 1339–1384, Singapore. ACL.
  44. Alex Reinhart, Ben Markey, Michael Laudenbach, Kachatad Pantusen, Ronald Yurko, Gordon Weinberg, and David West Brown. 2025. Do llms write like humans? variation in grammatical and rhetorical styles. Proceedings of the National Academy of Sciences, 122(8):e2422455122. find
  45. Gaurav Sahu, Pau Rodriguez, Issam Laradji, Parmida Atighehchian, David Vazquez, and Dzmitry Bahdanau. 2022. Data augmentation for intent classification with off-the-shelf large language models. In Proceedings of the 4th Workshop on NLP for Conversational AI, pages 47–57, Dublin, Ireland. Association…
  46. Timo Schick and Hinrich Schütze. 2021. It’s not just size that matters: Small language models are also few-shot learners. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2339–2352, Online. ACL.
  47. Melanie Sclar, Yejin Choi, Yulia Tsvetkov, and Alane Suhr. 2023. Quantifying Language Models’ Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting. find
  48. Indira Sen, Dennis Assenmacher, Mattia Samory, Isabelle Augenstein, Wil Aalst, and Claudia Wagner. 2023. People make better edits: Measuring the efficacy of LLM-generated counterfactually augmented data for harmful language detection. In Proceedings of the 2023 Conference on Empirical Methods in Nat… find
  49. Xiaofei Sun, Linfeng Dong, Xiaoya Li, Zhen Wan, Shuhe Wang, Tianwei Zhang, Jiwei Li, Fei Cheng, Lingjuan Lyu, Fei Wu, and 1 others. 2023. Pushing the limits of chatgpt on nlp tasks. arXiv preprint arXiv:2306.09719.
  50. Qwen Team. 2024. Qwen2.5: A party of foundation models. find
  51. Lucas Weber, Elia Bruni, and Dieuwke Hupkes. 2023. Mind the instructions: a holistic evaluation of consistency and interactions in prompt-based learning. In Proc. of the 27th Conference on Computational Natural Language Learning (CoNLL), pages 294–313, Singapore. Association for Computational Lingui…
  52. Kang Min Yoo, Dongju Park, Jaewook Kang, Sang-Woo Lee, and Woomyoung Park. 2021. GPT3Mix: Leveraging large-scale language models for text augmentation. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 2225–2239, Punta Cana, Dominican Republic. Association for Computati…
  53. Elena Zotova, Rodrigo Agerri, and German Rigau. 2021. Semi-automatic generation of multilingual datasets for stance detection in twitter. Expert Systems with Applications, 170:114547. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pecher-2026-better-as-generators-than-classifiers/