Udaras na Gaeltachta Pilots Irish-Language AI Chatbot

Udaras na Gaeltachta, working with the Department of Tourism, Culture, Arts, Gaeltacht, Sport and Media and the Department of Public Expenditure, has launched a European tender and open market consultation to develop an AI speech platform called Ard-Intleacht na Gaeilge. The program aims to pilot Irish-to-Irish speech services and an Irish-fluent public-service chatbot by the end of 2026, with a dedicated company (a DAC) to manage development, deployment, and training.

The tender frames a multi-component solution that must work with existing commercial AI products and support the range of Irish dialects. Key technical building blocks likely to be required include ASR for Irish dialects, TTS with dialect-aware voices, conversational state tracking for dialogue, and models to convert between English and Irish where needed. The program explicitly plans to leverage the State's archival audio and video and existing Irish-language corpora for training and evaluation. The procurement process includes a Prior Information Notice and Open Market Consultation, signaling requirements for interoperability and production-grade support.

• •Platform must deliver high-quality public-service responses in Irish across dialects
• •Tool must interoperate with commercial AI systems and be usable by multiple State agencies
• •A DAC will provide services and training

This is a national-scale, government-backed attempt to operationalize language technology for a lesser-used language. For practitioners the project matters for three reasons. First, it aggregates otherwise fragmented, high-value training data (archive audio and video) that can enable stronger ASR and TTS baselines for Irish. Second, explicit procurement language favoring interoperability increases the chance the solution will use modular components-fine-tuned acoustic models, lightweight neural TTS voices, and an LLM or dialogue manager tuned for Irish conversational norms-rather than a single proprietary black box. Third, the program is framed as a model for other minority-language initiatives across Europe; a successful rollout would set procurement, evaluation, and data-governance precedents.

Expect classic low-resource-language challenges: uneven dialect coverage, limited labelled transcriptions, speaker imbalance, and licensing constraints for archive material. Solutions will likely combine transfer learning from larger multilingual speech models, self-supervised pretraining on raw audio, targeted data augmentation, and active learning cycles with human-in-the-loop transcription and validation. Privacy and consent for archival voices will require careful legal and annotation workflows.

Procurement milestones through the tender and OMC, selection of commercial partners or open-source stacks, published evaluation metrics for ASR/TTS by dialect, and the formation and remit of the DAC. Also watch how training and governance around archival data and voice consent are handled; these choices will determine reusability and research impact.

"It is essential that the emerging AI technology be available to the next generation of Irish speakers," said Tomas O Siochain, CEO of Udaras na Gaeltachta. "By investing in technology, we are fostering a sustainable future for the language," added Siobhan Ni Ghadhra, Chairperson.

This is a technically ambitious, policy-driven effort that will matter to practitioners building low-resource speech systems and to public-sector teams planning bilingual digital services.