The innovation of the Polyglot project lies in its commitment to making foundation models accessible and effective for low-resource languages, which have historically been excluded from the major advances in generative AI. Rather than scaling existing models indiscriminately, Polyglot takes a targeted, sustainable, and open-source approach by developing tailored tools, datasets, and models that can be adapted to the unique linguistic and cultural contexts of underrepresented communities. This not only democratizes access to AI technologies but also empowers local researchers and speakers to actively shape AI systems in ways that reflect their values and identities.The innovation of the Polyglot project lies in its commitment to making foundation models accessible and effective for low-resource languages, which have historically been excluded from the major advances in generative AI. Rather than scaling existing models indiscriminately, Polyglot takes a targeted, sustainable, and open-source approach by developing tailored tools, datasets, and models that can be adapted to the unique linguistic and cultural contexts of underrepresented communities. This not only democratizes access to AI technologies but also empowers local researchers and speakers to actively shape AI systems in ways that reflect their values and identities.

An interdisciplinary project is one that brings together researchers and methods from different academic fields to collaboratively address complex problems that cannot be fully understood through a single disciplinary lens. In the case of Polyglot, for instance, deep learning specialists, high-performance computing experts, linguists, and philosophers work side by side—not only to build language models for underrepresented languages, but also to ensure that these technologies are developed ethically, sustainably, and with cultural sensitivity. This kind of collaboration allows for richer insights and more responsible innovation, blending technical excellence with societal awareness.

The Polyglot project is expected to generate a diverse set of high-impact outputs that contribute both academically and practically to the NLP community. These include a comprehensive suite of datasets, monolingual large language models, and evaluation benchmarks tailored to low-resource languages, all made openly available. An open-source repository will ensure transparency and reproducibility, allowing others to build upon our work. We also aim to produce several peer-reviewed publications—at least one per target language—covering key aspects of the project, such as dataset creation and model performance. Beyond that, the project supports graduate research, offering fertile ground for Master’s and PhD theses. Finally, we plan to host an international, interdisciplinary workshop to foster dialogue around the development of LLMs for underrepresented languages.