Wolof language models deliver strong performance with open weights
Wolof-Adapted Industry-Ready Open-Weight Language Models
Computation and LanguageArtificial Intelligence
Summary
Wolof is a language spoken by many people but has few good freely available language tools. The authors created several new open-weight language models tuned especially for Wolof, ranging from smaller to larger sizes. These models can understand and generate Wolof text well, while also working with other languages. They match or come close to much bigger commercial systems and are being tested in real customer service settings. The models are freely accessible online for anyone to use.
What this means in practice
- •For customer-care teams: Use these Wolof-adapted models to improve automated support where Wolof language understanding and generation are critical.
- •For multilingual ai developers: Incorporate these open-weight Wolof models to build language tools that handle Wolof and other languages with better local control and lower cost.
Authors
Gwénolé Lecorvé, Delphine Charlet, Pierre Adam, Floriane Behanzin, Lionel Delphin-Poulat, Johannes Heinecke, Géraldine Damnati, Pelin Kaytmaz, Nourredine Khaous, Charbel Kindji, Meriam Maadi, Abdoulaye Mbaye, Claire Perroux, Ismaël Rousseau, Papa Séga Wadé, Gilles Xiberras, Olivier Zhang, Malek Zitouni
Abstract
Wolof is spoken by 23M people, mostly in Senegal, but remains weakly supported by open-weight language models, while API models raise cost, governance, and sovereignty constraints for local deployments. In this paper, we present 0.6B-14B open-weight models adapted to Wolof through instruction tuning, preference alignment, and model merging. After presenting the pipeline and data, extensive experiments on 63~public and in-house tasks show good performance on Wolof understanding and generation while remaining effective on other languages and cross-lingual skills, with the 8B and 14B models being particularly competitive with substantially larger commercial systems. The models are currently being evaluated in internal customer-care pilots and are publicly available at https://huggingface.co/collections/Orange/wolof-lm .