Research Vision

AI should not simply become more capable. It should become more trustworthy, multilingual, culturally aware and human-centred.

Dr Afli's research pursues language technologies that work for people and communities who are under-served by current AI — across languages, cultures, health and public life. The programme joins four commitments:

Multilingual and low-resource languages

Most of the world's languages lack the data and tools that high-resource languages enjoy. Work on multilingual and culturally aware AI — including Irish, Arabic and other low-resource languages, and open-source tooling such as adaptNMT and adaptMLLM — aims to narrow that gap.

Cultural reasoning

Language models interpret culturally specific knowledge and values unevenly. Research on culturally grounded question answering (for example the Arabic PalmX and QIAS systems) studies how models handle cultural context.

Trustworthy evaluation

Claims of progress are only as sound as the evaluation behind them. Work on evaluation science — robust text-embedding evaluation (PTEB) and the limits of automated judges — treats evaluation as a first-class research problem.

Human-centred and trustworthy AI

Privacy-preserving and explainable methods — including the European patent on split federated learning — keep people and their data at the centre. See Human-Centred AI and Trustworthy AI.

How the centres contribute

The Human-Centred AI Research Group is the group home; the Rinn Artificial Intelligence national centre and the ADAPT Centre provide national research platforms; and Horizon Europe projects such as GenDAI extend the work in AI for health.

Future directions

The current agenda centres on inclusive multilingual language models, cultural reasoning in generative AI, and reliable multilingual evaluation — advancing AI that is dependable, culturally aware and human-centred across language, health and society.

The next five years

Building on the current programme and existing collaborations, the work over the coming years is expected to develop along several connected lines:

• Inclusive multilingual models. Extending language models and translation to more low-resource languages, continuing the direction of adaptNMT and adaptMLLM.
• Trustworthy evaluation. Strengthening evaluation methods for language models and embeddings, building on PTEB and work on the limits of automated judges.
• Trustworthy and safe AI. Advancing privacy-preserving and explainable methods, continuing the split-federated-learning line protected by the 2024 patent.
• AI for health. Deepening the genomics-and-language work through Horizon Europe (GenDAI) and related collaborations.
• Public-sector and societal AI. Continuing the PoliticalNLP direction on NLP for governance and public discourse.
• European collaboration and doctoral research. Growing the Human-Centred AI Research Group and its role within Rinn AI and ADAPT, and supervising doctoral researchers across these themes.

These directions build on existing projects, publications and partnerships rather than new commitments; specific projects and funding are confirmed as they are announced.

Explore the research journey, the research themes, and the publications behind this vision.