---
title: "Small Language Models for Less-Resourced — Saurí 2026"
description: "The paper examines whether Small Language Models (SLMs), models of up to roughly 10–15B parameters, are reliable enough for real-world use in a…"
authors: ["Roser Saurí", "Josep Sànchez-Ferreres", "Lluis Padro", "Josep Carmona"]
year: 2026
venue: "Proceedings of the SIGUL 2026 Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages"
source: "https://aclanthology.org/2026.sigul-1.27/"
doi: "10.63317/4qn4qiyiwuze"
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/sauri-2026-small-language-models-less-resourced/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sauri-2026-small-language-models-less-resourced/)

[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/9/) / Small Language Models for Less-Resourced Languages in a Real-World Scenario: The Case for Catalan

# Small Language Models for Less-Resourced Languages in a Real-World Scenario: The Case for Catalan

**Authors:** [Roser Saurí](https://md.research.elamur.ai/free/en/authors/roser-sauri/), Josep Sànchez-Ferreres, [Lluis Padro](https://md.research.elamur.ai/free/en/authors/lluis-padro/), Josep Carmona · **Year:** 2026 · **Venue:** Proceedings of the SIGUL 2026 Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages · **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.sigul-1.27/) · [PDF](https://aclanthology.org/2026.sigul-1.27.pdf) · [DOI 10.63317/4qn4qiyiwuze](https://doi.org/10.63317/4qn4qiyiwuze) · [OpenAlex](https://openalex.org/W7166525230)

**Citations:** 0 citations (openalex-snapshot · 2026-10-14)

## Summary
The paper examines whether Small Language Models (SLMs), models of up to roughly 10–15B parameters, are reliable enough for real-world use in a minoritized language such as Catalan. In such languages, interference from dominant languages can degrade output and increase bias. Compact models matter for mobile and IoT hosting and for SMEs that need private, affordable, locally run AI. The authors placed several commercial and open-source SLMs inside Bleta, an assistant for older adults developed by Bleta Solutions. They tested the models on a hand-built set of short Catalan dialogues covering five intents. Two skills were assessed: conversing appropriately with the user, including asking for missing details, and producing correct structured function calls. Some SLMs delivered high-quality results in this setting. The two strongest were qwen3 8B and ministral-3 14B, with numeric scores slightly favouring qwen3 8B. However, a Catalan philologist judged qwen3 8B's linguistic errors more damaging to fluency and user trust. The study supports FREC, an initiative to build MCP-based networks of Catalan AI assistants.

## Key points
- **Intents and data:** The five intents are send_message (3 parameters), make_call, get_weather, set_alarm and navigate_to. The handcrafted Catalan dialogues range from single-turn requests to multi-turn exchanges with auxiliary tool calls (e.g., get_contacts) and several function calls per dialogue.
- **Top models:** The best performers were qwen3 8B and ministral-3 14B. Expert qualitative review of their Catalan reversed the slight quantitative advantage of qwen3 8B. The authors call these findings preliminary.
- **Prompt weakness:** User-interaction problems were traced to the instructional prompt. It made models request missing information but did not guide them to give users adequate feedback, which the authors plan to fix.
- **Project context:** The work belongs to FREC (frec.cat), which aims to deploy interconnected, mainly SLM-based Catalan assistants under the Model Context Protocol (MCP).
- **Related resource:** The paper cites the Salamandra family (2B and 7B) from the Barcelona Supercomputing Center's Aina Project as Catalan-focused open models. Detailed per-model scores were not included in the excerpt this summary is based on.

*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](https://md.research.elamur.ai/free/en/shelves/linguistics/)

## Cite

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

```bibtex
@inproceedings{sauri2026small,
  title = {{Small Language Models for Less-Resourced Languages in a Real-World Scenario: The Case for Catalan}},
  author = {Saurí, Roser and Sànchez-Ferreres, Josep and Padro, Lluis and Carmona, Josep},
  booktitle = {Proceedings of the SIGUL 2026 Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages},
  year = {2026},
  doi = {10.63317/4qn4qiyiwuze},
  url = {https://doi.org/10.63317/4qn4qiyiwuze}
}
```

- [← previous SlovKE: A Large-Scale Dataset and LLM Evaluation for Slovak Keyphrase Extraction](https://md.research.elamur.ai/free/en/papers/stevanak-2026-slovke-large-scale-dataset-llm/)
- [next → Small Neural Networks as Models of Cross-Linguistic Speech Perception](https://md.research.elamur.ai/free/en/papers/shankwitz-2026-small-neural-networks-as-models/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/9/)

## References

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

1. AAAI Presidential Panel. 2025. Future of AI research. Technical report, Association for the Advancement of Artificial Intelligence (AAAI). Accessed: 2026-02-19. [find](https://scholar.google.com/scholar?q=AAAI+Presidential+Panel%2E+2025%2E+Future+of+AI+research%2E+Technical+report%2C+Association+for+the+Advancement+of+Artificial+Intelligence+%28AAAI%29%2E+Accessed%3A+2026%2D02%2D19%2E)
2. Misha Bilenko. 2024. Introducing Phi-3: Redefining what’s possible with SLMs. Accessed: 2026-02-19. [find](https://scholar.google.com/scholar?q=Misha+Bilenko%2E+2024%2E+Introducing+Phi%2D3%3A+Redefining+what%E2%80%99s+possible+with+SLMs%2E+Accessed%3A+2026%2D02%2D19%2E)
3. Sascha Brodsky. 2025. Beyond big models: Why AI needs more than just scale to reach AGI. IBM Think News. [find](https://scholar.google.com/scholar?q=Sascha+Brodsky%2E+2025%2E+Beyond+big+models%3A+Why+AI+needs+more+than+just+scale+to+reach+AGI%2E+IBM+Think+News%2E)
4. Flavio Corradini, Matteo Leonesi, and Marco Pian-gerelli. 2025. State of the art and future directions of small language models: A systematic review. Big Data and Cognitive Computing, 9(7). [find](https://scholar.google.com/scholar?q=Flavio+Corradini%2C+Matteo+Leonesi%2C+and+Marco+Pian%2Dgerelli%2E+2025%2E+State+of+the+art+and+future+directions+of+small+language+models%3A+A+systematic+review%2E+Big+Data+and+Cognitive+Computing%2C+9%287%29%2E)
5. Jiawei Gu, Xuhui Jiang, Zhichao Shi, Hexiang Tan, Xuehao Zhai, Chengjin Xu, Wei Li, Yinghan Shen, Shengjie Ma, Honghao Liu, Saizhuo Wang, Kun Zhang, Zhouchi Lin, Bowen Zhang, Lionel Ni, Wen Gao, Yuanzhuo Wang, and Jian Guo. 2026. A survey on LLM-as-a-Judge. The Innovation. [find](https://scholar.google.com/scholar?q=Jiawei+Gu%2C+Xuhui+Jiang%2C+Zhichao+Shi%2C+Hexiang+Tan%2C+Xuehao+Zhai%2C+Chengjin+Xu%2C+Wei+Li%2C+Yinghan+Shen%2C+Shengjie+Ma%2C+Honghao+Liu%2C+Saizhuo+Wang%2C+Kun+Zhang%2C+Zhouchi+Lin%2C+Bowen+Zhang%2C+Lionel+Ni%2C+Wen+Gao%2C+Yuanz)
6. Daniil Gurgurov, Ivan Vykopal, Josef van Genabith, and Simon Ostermann. 2025. Small models, big impact: Efficient corpus and graph-based adaptation of small multilingual language models for low-resource languages. arXiv preprint arXiv:2502.10140. [find](https://scholar.google.com/scholar?q=Daniil+Gurgurov%2C+Ivan+Vykopal%2C+Josef+van+Genabith%2C+and+Simon+Ostermann%2E+2025%2E+Small+models%2C+big+impact%3A+Efficient+corpus+and+graph%2Dbased+adaptation+of+small+multilingual+language+models+for+low%2Dresour)
7. Hansi Hettiarachchi, Tharindu Ranasinghe, Paul Rayson, Ruslan Mitkov, Mohamed Gaber, Damith Premasiri, Fiona Anting Tan, and Lasitha Uyangodage, editors. 2025. Proceedings of the First Workshop on Language Models for Low-Resource Languages. Association for Computational Linguistics, Abu Dhabi, Unite… [find](https://scholar.google.com/scholar?q=Hansi+Hettiarachchi%2C+Tharindu+Ranasinghe%2C+Paul+Rayson%2C+Ruslan+Mitkov%2C+Mohamed+Gaber%2C+Damith+Premasiri%2C+Fiona+Anting+Tan%2C+and+Lasitha+Uyangodage%2C+editors%2E+2025%2E+Proceedings+of+the+First+Workshop+on+Lan)
8. Marissa Ikonomidis, T.J. Alumbaugh, Mark Sherwood, and Cormac Brick. 2025. Gemma 3 on mobile and web with Google AI Edge. [https://developers.googleblog.com/en/gemma-3-on-mobile-and-web-with-google-ai-edge](https://developers.googleblog.com/en/gemma-3-on-mobile-and-web-with-google-ai-edge). Accessed: 2026-02-19. [↗](https://developers.googleblog.com/en/gemma-3-on-mobile-and-web-with-google-ai-edge)
9. Stefanos Laskaridis, Kleomenis Katevas, Lorenzo Minto, and Hamed Haddadi. 2024. MELTing point: Mobile evaluation of language transformers. arXiv preprint arXiv:2403.12844. — [arXiv:2403.12844](https://arxiv.org/abs/2403.12844) [find](https://scholar.google.com/scholar?q=Stefanos+Laskaridis%2C+Kleomenis+Katevas%2C+Lorenzo+Minto%2C+and+Hamed+Haddadi%2E+2024%2E+MELTing+point%3A+Mobile+evaluation+of+language+transformers%2E+arXiv+preprint+arXiv%3A2403%2E12844%2E+%E2%80%94+%5BarXiv%3A2403%2E12844%5D%28https%3A%2F)
10. Dawei Li, Bohan Jiang, Liangjie Huang, Alimohammad Beigi, Chengshuai Zhao, Zhen Tan, Amrita Bhattacharjee, Yuxuan Jiang, Canyu Chen, Tianhao Wu, Kai Shu, Lu Cheng, and Huan Liu. 2025. From generation to judgment: Opportunities and challenges of LLM-as-a-judge. arXiv preprint arXiv:2411.16594. [find](https://scholar.google.com/scholar?q=Dawei+Li%2C+Bohan+Jiang%2C+Liangjie+Huang%2C+Alimohammad+Beigi%2C+Chengshuai+Zhao%2C+Zhen+Tan%2C+Amrita+Bhattacharjee%2C+Yuxuan+Jiang%2C+Canyu+Chen%2C+Tianhao+Wu%2C+Kai+Shu%2C+Lu+Cheng%2C+and+Huan+Liu%2E+2025%2E+From+generation+)
11. Zhenyan Lu, Xiang Li, Dongqi Cai, Rongjie Yi, Fangming Liu, Wei Liu, Jian Luan, Xiwen Zhang, Nicholas D. Lane, and Mengwei Xu. 2025. Demystifying small language models for edge deployment. In Proceedings of the 63rd Annual Meeting of the ACL. ACL. [find](https://scholar.google.com/scholar?q=Zhenyan+Lu%2C+Xiang+Li%2C+Dongqi+Cai%2C+Rongjie+Yi%2C+Fangming+Liu%2C+Wei+Liu%2C+Jian+Luan%2C+Xiwen+Zhang%2C+Nicholas+D%2E+Lane%2C+and+Mengwei+Xu%2E+2025%2E+Demystifying+small+language+models+for+edge+deployment%2E+In+Proceedi)
12. Rithesh Murthy, Liangwei Yang, Juntao Tan, Tulika Manoj Awalgaonkar, Yilun Zhou, Shelby Heinecke, Sachin Desai, Jason Wu, Ran Xu, Sarah Tan, Jianguo Zhang, Zhiwei Liu, Shirley Kokane, Zuxin Liu, Ming Zhu, Huan Wang, Caiming Xiong, and Silvio Savarese. 2024. MobileAIBench: Benchmarking LLMs and LMMs… — [arXiv:2406.10290](https://arxiv.org/abs/2406.10290) [find](https://scholar.google.com/scholar?q=Rithesh+Murthy%2C+Liangwei+Yang%2C+Juntao+Tan%2C+Tulika+Manoj+Awalgaonkar%2C+Yilun+Zhou%2C+Shelby+Heinecke%2C+Sachin+Desai%2C+Jason+Wu%2C+Ran+Xu%2C+Sarah+Tan%2C+Jianguo+Zhang%2C+Zhiwei+Liu%2C+Shirley+Kokane%2C+Zuxin+Liu%2C+Ming+)
13. Lluís Padró and Roser Saurí. 2024. Fine-tuning open access LLMs for high-precision NLU in goal-driven dialog systems. In Proceedings of the Second International Workshop Towards Digital Language Equality (TDLE): Focusing on Sustainability @ LREC-COLING 2024, pages 33–42, Torino, Italia. ELRA and ICC… [find](https://scholar.google.com/scholar?q=Llu%C3%ADs+Padr%C3%B3+and+Roser+Saur%C3%AD%2E+2024%2E+Fine%2Dtuning+open+access+LLMs+for+high%2Dprecision+NLU+in+goal%2Ddriven+dialog+systems%2E+In+Proceedings+of+the+Second+International+Workshop+Towards+Digital+Language+Equal)
14. Mehmood Rashid and Eid Rehman. 2025. The Illusion of Intelligence: Evaluating Large Language Models Against Grounded Criteria of Artificial General Intelligence. Preprints. Publisher: Preprints. [find](https://scholar.google.com/scholar?q=Mehmood+Rashid+and+Eid+Rehman%2E+2025%2E+The+Illusion+of+Intelligence%3A+Evaluating+Large+Language+Models+Against+Grounded+Criteria+of+Artificial+General+Intelligence%2E+Preprints%2E+Publisher%3A+Preprints%2E)
15. Syed Abdul Gaffar Shakhadri, Kruthika KR, and Rakshit Aralimatti. 2025. SHAKTI: A 2.5 billion parameter small language model optimized for edge AI and low-resource environments. arXiv preprint arXiv:2410.11331. — [arXiv:2410.11331](https://arxiv.org/abs/2410.11331) [find](https://scholar.google.com/scholar?q=Syed+Abdul+Gaffar+Shakhadri%2C+Kruthika+KR%2C+and+Rakshit+Aralimatti%2E+2025%2E+SHAKTI%3A+A+2%2E5+billion+parameter+small+language+model+optimized+for+edge+AI+and+low%2Dresource+environments%2E+arXiv+preprint+arXiv%3A2)
16. Guokan Shang, Hadi Abdine, Yousef Khoubrane, Amr Mohamed, Yassine Abbahaddou, Sofiane Ennadir, Imane Momayiz, Xuguang Ren, Eric Moulines, Preslav Nakov, Michalis Vazirgiannis, and Eric Xing. 2025. Atlas-Chat: Adapting large language models for low-resource Moroccan Arabic dialect. In Proceedings of… [find](https://scholar.google.com/scholar?q=Guokan+Shang%2C+Hadi+Abdine%2C+Yousef+Khoubrane%2C+Amr+Mohamed%2C+Yassine+Abbahaddou%2C+Sofiane+Ennadir%2C+Imane+Momayiz%2C+Xuguang+Ren%2C+Eric+Moulines%2C+Preslav+Nakov%2C+Michalis+Vazirgiannis%2C+and+Eric+Xing%2E+2025%2E+Atl)
17. Jiacheng Wang, Yejun Zeng, Jinyang Guo, Yuqing Ma, Aishan Liu, and Xianglong Liu. 2025. SLMQuant: Benchmarking small language model quantization for practical deployment. In Proceedings of the 3rd International Workshop on Rich Media With Generative AI, page 2–10. ACM. [find](https://scholar.google.com/scholar?q=Jiacheng+Wang%2C+Yejun+Zeng%2C+Jinyang+Guo%2C+Yuqing+Ma%2C+Aishan+Liu%2C+and+Xianglong+Liu%2E+2025%2E+SLMQuant%3A+Benchmarking+small+language+model+quantization+for+practical+deployment%2E+In+Proceedings+of+the+3rd+Int)
18. Mingxue Xu, Yao Lei Xu, and Danilo P. Mandic. 2025. TensorSLM: Energy-efficient embedding compression of sub-billion parameter language models on low-end devices. arXiv preprint arXiv:2506.13514. — [arXiv:2506.13514](https://arxiv.org/abs/2506.13514) [find](https://scholar.google.com/scholar?q=Mingxue+Xu%2C+Yao+Lei+Xu%2C+and+Danilo+P%2E+Mandic%2E+2025%2E+TensorSLM%3A+Energy%2Defficient+embedding+compression+of+sub%2Dbillion+parameter+language+models+on+low%2Dend+devices%2E+arXiv+preprint+arXiv%3A2506%2E13514%2E+%E2%80%94+%5Ba)