MBZUAI's Push for Linguistic Diversity in AI

The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), the world's first dedicated graduate-level AI research university in Abu Dhabi, United Arab Emirates (UAE), is at the forefront of making artificial intelligence (AI) more inclusive. A recent milestone highlights this commitment: former MBZUAI postdoctoral researcher Atnafu Lambebo Tonja has been appointed as a Google DeepMind Academic Fellow at University College London (UCL). This three-year, fully funded fellowship allows Tonja to advance his work on multilingual AI for low-resource languages, building directly on his contributions at MBZUAI.

Tonja's journey underscores MBZUAI's role in nurturing talent for global challenges. Originating from Ethiopia—a country with over 2,000 languages across Africa where few are represented in AI research—he focused on natural language processing (NLP) for underrepresented tongues during his PhD in Mexico and visiting stints. At MBZUAI in 2024, collaborating with Professor Thamar Solorio, Vice Provost of Faculty Excellence and Advancement, he co-authored the award-winning paper 'The Zeno’s Paradox of Low-Resource Languages,' recognized at EMNLP 2024. This work exposed how AI progress slows for scarce-data languages, much like Zeno's paradoxes hinder motion perception.

MBZUAI's Natural Language Processing Department emphasizes Arabic dialects, MENA (Middle East and North Africa) languages, and Global South tongues, addressing a critical gap: of the world's 7,000+ languages, 90% are in the Global South, yet AI models skew toward high-resource ones like English.

Key Contributions from MBZUAI Researchers

Tonja led the Afri-MCQA benchmark, a multimodal dataset with 7,500+ question-answer pairs in 15 African languages from 12 countries. Testing large language models (LLMs) on cultural knowledge—like foods, traditions, and daily life via text, speech, and images—it revealed LLMs' deficiencies in African contexts. Accepted at ACL 2026 (top 5% selectivity), this benchmark pushes for culturally attuned AI, essential for education and healthcare in diverse regions.

Complementing this, Professor Solorio secured $1 million from Google.org in February 2026 for the MENA AI Opportunity Initiative. The project aims to bridge the 'data divide' by creating 'resource-lean' AI models needing less annotated data and compute power. Goals include native frameworks rooted in MENA's sociocultural realities, talent training for postdocs, and applications in education, cultural preservation, and digital communication. Solorio emphasized: 'This funding redefines the field, moving beyond adapting high-resource models to linguistically grounded AI for MENA languages.'

These efforts align with MBZUAI's mission since its 2019 founding, rapidly ascending to global top-10 AI rankings in computer vision and NLP by 2025. The university's focus on underrepresented languages positions UAE as a hub for equitable AI.

The Google DeepMind Academic Fellowship Explained

Google DeepMind's Academic Fellowship Programme supports early-career AI researchers with funding, mentorship, and leadership training. Tonja's role at UCL's Centre for Artificial Intelligence involves developing AI systems and benchmarks for under-resourced languages, including cross-lingual transfer, multimodal models (speech-text-vision), and culturally relevant evaluations. Starting March 2026, it builds on his MBZUAI work, filling gaps where AI fails diverse communities.

Tonja noted: 'LLMs excel in general knowledge but gap in specific countries or communities... How can we trust them in real-world apps?' His fellowship addresses this, promoting equitable tech. UCL's prior fellow, Dr. David Adelani, advanced to McGill professorship, showing the programme's impact.

In UAE context, such fellowships highlight MBZUAI's pipeline for international leadership, attracting global talent via full scholarships for MSc/PhD programs.

UAE's Strategic Vision for AI Supremacy

The UAE's National AI Strategy 2031 invests billions to lead globally, with MBZUAI central. Initiatives like Ruwwad AI Scholars Fellowship train Emirati PhDs for faculty roles abroad, while partnerships (MIT, Google) boost research. Abu Dhabi's AI ecosystem, including G42 and TII, deploys supercomputers and sovereign AI infrastructure.

Language inclusivity fits UAE's multicultural fabric—home to 200+ nationalities—and bridges Global South gaps. MBZUAI's report 'AI for the Global South,' launched at India AI Impact Summit 2026, outlines 12 research priorities for equitable development, emphasizing low-resource languages.

This positions UAE universities as innovators, fostering jobs in NLP, AI ethics, and multilingual tech.

Challenges and Solutions in Low-Resource Language AI

• Data Scarcity: Low-resource languages lack annotated corpora; solutions include synthetic data generation and transfer learning from high-resource kin.
• Cultural Bias: Models misinterpret nuances; benchmarks like Afri-MCQA test cultural reasoning.
• Compute Barriers: Resource-lean models lower entry, enabling startups in MENA/Africa.
• Evaluation Gaps: Zeno’s Paradox shows stalled progress; new metrics track true advancement.

Solorio's initiative trains postdocs, while Tonja's work scales multimodal benchmarks. Real-world: AI translators for Ethiopian markets, preserving indigenous knowledge.

Case Studies: Benchmarks Revolutionizing AI

Afri-MCQA reveals LLMs' cultural blind spots—e.g., failing African proverbs or rituals—forcing model redesigns. InkubaLM, an African LLM, demonstrates feasible small models rivaling giants.

MBZUAI's Hindi model tests cultural transfer, showing AI must encode societal norms. These advance step-by-step: data collection → model training → benchmark validation → deployment.

In UAE, similar for Arabic dialects ensures sentiment analysis captures regional variances.

Stakeholder Perspectives and Collaborations

Google's Yossi Matias: 'Aligns with accelerating discovery via cooperation.' Tonja credits MBZUAI's 'collaborative environment' for freedom on ambitious problems.

UAE leaders view this as soft power: exporting AI talent via fellowships. Partners like Google DeepMind, UCL amplify reach. Multi-perspective: African speakers gain voice; MENA startups innovate; global AI firms access diverse data ethically.

Implications for UAE Higher Education and Careers

MBZUAI's feats boost UAE's allure for AI PhDs—full scholarships, stipends, housing. Faculty roles surge; explore UAE academic jobs. Research jobs in NLP exploding, with salaries competitive globally.

Programs like Global AI Leadership train 36+ UAE execs yearly. Future: more fellowships, sovereign AI supercomputers (e.g., UAE-India collab).

Photo by Emin Huric on Unsplash

Future Outlook: Equitable AI Horizons

By 2030, UAE aims AI GDP contribution of AED 110B. MBZUAI leads with reports guiding Global South AI. Actionable: students pursue NLP MScs; unis invest in benchmarks; policymakers fund data sovereignty.

Tonja: 'So much to do... looking forward to filling gaps.' UAE's model—research excellence + inclusivity—sets benchmark for higher ed worldwide.

Stakeholders predict: 50% low-resource lang coverage by 2035, boosting education access, preserving heritage.