Data Science for Social Impact

For those at MBZUAI in Abu Dhabi, I will be giving a seminar talk tomorrow (13 Oct 2025)

Introducing Kesego Mokgosi (visiting PhD Student) and an update on "Marito -> Mafoko" Introducing Kesego Mokgosi: Visiting PhD Researcher with the AI4D African Languages Lab

Introducing Kesego Mokgosi: Visiting PhD Researcher with the AI4D African Languages Lab open.substack.com/pub/dsup/p/i...

Are data scientists in Africa allowed to use copyrighted news broadcasts to train AI and develop technologies for local languages? This is a critical question for innovation in African Natural… Can African Data Scientists Use Copyrighted News for AI Training? 🌍🤖

New work out of the Data Science Law Lab at UP - @DataLawAfrica

Working to answer -> Are data scientists in Africa allowed to use copyrighted news broadcasts to train AI and develop technologies for local languages?

Watch -> youtu.be/p-zFzlbhUw4
Paper -> doi.org/10.1080/1360...

Dr Mpho Mokoatle: a beacon of hope in cancer research through her PhD in Computer Science Dr Mpho Mokoatle has emerged as a beacon of hope following her recent PhD graduation in Computer Science from the University of Pretoria. Her groundbreaking research aims to harness advanced…

Huge congratulations again to Dr. Mpho Mokoatle on her PhD in Computer Science from UP! 🎓 So proud her groundbreaking cancer research and inspiring journey are featured by the Weekend Argus. Her work in DNA analysis for early detection is truly life-changing!

iol.co.za/weekend-argu...

My Inaugural Lecture is now online 🎥
Beyond the Symbols: Natural Language Processing as an Adaptive Problem

I reflect on my journey in advancing AI & NLP for African languages, tackling data scarcity, building communities & imagining inclusive futures.

👉 www.youtube.com/watch?v=hwef...

YouTube video by Data Science for Social Impact (DSFSI) Beyond the Symbols: Natural Language Processing as an Adaptive Problem - Prof. Vukosi Marivate

Next was an excellent talk by @vukosi.bsky.social on natural language processing for low resource languages and the importance of Africans developing their own solutions at @dsfsi.bsky.social youtu.be/hwefANMnSxo?... (5/8)

Key message: AI in education must reflect the plurality of learners – their tongues, their stories, their right to be heard.

Computer science + education can (and must) work together to make this happen.

#AI #Education #UNESCO #AfricanLanguages @unesco.org

Our chapter – Ensuring Inclusive, Contextualized AI in Education – looks at:

* The promises and pitfalls of AI in classrooms
* Risks of excluding African languages & knowledge systems
* The need for grounded, community-led innovations

We’re excited to share that members of our Data Science for Social Impact Lab at UP contributed to UNESCO’s new global report: AI and the Future of Education: Disruptions, Dilemmas and Directions.

🔗 unesdoc.unesco.org/ark:/48223/p...

@unesco.org

The day began with the AI Policy Breakfast (Partnership on AI), connecting policymakers and researchers.

Day 4 proved once again: African AI is not only technically vibrant — it is leading on global conversations around policy, ethics, and equity. 🚀

Resources for those who want to go deeper:
📄 AUDA-NEPAD White Paper on AI in Africa: 1drv.ms/f/s!At9WNpyX...
🔑 NOODL License: licensingafricandatasets.com
🔑 Esethu License: aclanthology.org/2025.acl-lon...

Another highlight: the AI Policy panel with Jane Munga, Virginia Dignum & Barbara Glover.
We explored how Africa can chart its own path in AI governance, infrastructure, and inclusion.

The AI Policy in Africa Panel at DLI 2025 in Kigali, Rwanda featured a special message from African Union Commissioner for Infrastructure and Energy, H.E. Ms. Lerato D. Matabogee. She shares the AU’s… African Union Commissioner Message at DLI 2025

We were honoured to have @vukosi serve as a keynote panelist in the AI Policy for Africa session, opened by AU Commissioner H.E. Lerato Mataboge.

Her message on Urunana (hand in hand) was powerful: solidarity & responsibility in advancing AI.
🎥 Watch: www.youtube.com/watch?v=Nu-6...

The poster session was electric ⚡ — showcasing the vibrancy of the African AI community.
DSFSI contributions came from:
• Thapelo
• Nontokozo
• Idris
• Takura
(plus work from Seani & Dawit )

Day 4 at the #DeepLearningIndaba was an inspiring mix of research, policy, and community building. 🌍 #DLI2025

We also had a faculty panel discussing opportunities & challenges facing academics in AI across the continent, plus breakout sessions on Data, Startups, and community-building. A busy and inspiring day, showcasing the depth, ambition, and impact of Africa’s AI research community. 🚀 #DLI2025 5/5

This achievement was led by @ANKahira with contributions from UP’s own Lesego Makhafola — building on mentorship from Prof @Vukosi who has coordinated poster sessions and non-archival proceedings in recent years. 🙌 4/n

Highlight of the day: the Indaba’s first-ever conference proceedings 🎉

* 100 submissions
* 19 accepted papers
* 7 presented on Day 3

A major step in growing African-led AI research.

3/n

Lecture Notes: From Benchmarks to Responsible AI Evaluation Lecture Notes: Beyond Benchmarks: Building a Science of AI Measurement Presenter: Samni Koyejo (Keynote)Focus: Building trustworthy AI through validity-centered evaluation, cost-efficient tools, and…

We began with an insightful keynote by Prof. @SamniKoyejo: Beyond Benchmarks: Building a Science of AI Measurement.
📝 Notes here: docs.google.com/document/d/1... 2/n

Day 3 at the #DeepLearningIndaba was another milestone moment for African AI. 🌍 #DLI2025 @deepindaba 1/n

This seminar was made possible by the Carnegie African Diaspora Fellowship Program (CADFP), and hosted by DSFSI, AfriDSAI, the Dept. of Computer Science, and the Faculty of Health Sciences @ University of Pretoria
#AI4Good #HealthEquity #AfricanInnovation

Dr. Gichoya explores how Africa can go beyond basic chatbots—focusing on context-specific AI, building diverse datasets, addressing bias, and nurturing local data science talent. Vital for anyone interested in #AI, #healthcare, and #Africa.

YouTube video by Data Science for Social Impact AI for African Healthcare: Beyond Chatbots Toward Contextually Informed Innovation - Dr Judy Gichoya

🌍 Excited to share a powerful seminar by Dr. Judy Gichoya (Emory Univ): “AI for African Healthcare: Beyond Chatbots Toward Contextually Informed Innovation.”
🎥 Watch: youtu.be/c8b7JjszYi0?...

UP is breaking the language barrier in academia!

Join the launch of the Translation of Abstracts Project—bringing research to Sepedi, isiZulu, and Afrikaans.

12 Aug, 09:00–12:00, Javett Auditorium or online.

Info: www.dsfsi.co.za/blog/library...

RSVP: up-za.libcal.com/event/14796230

Cross-lingual embedding methods and applications: A systematic review for low-resourced scenarios The field of Natural Language Processing (NLP) has achieved significant success in various areas, such as developing large-scale datasets, algorithmic…

Updated paper link 🖇️ www.sciencedirect.com/science/arti...

🎧 Also, check out the NotebookLM podcast for broader reflections on language and inclusion in tech.
🔗 notebooklm.google.com/notebook/0f0...

We argue:
👉 Start with small projection models
👉 Build quality monolingual corpora
👉 Use bilingual lexicons smartly
👉 Explore language similarity to guide transfer

Essential reading for anyone working on #LowResourceNLP. #AI4Africa

Highlights:

Deep dive into CLTL model types & architectures

Taxonomy of projection & joint models

Evaluation metrics

Use in real-world tasks like NER, classification, and parsing

Gaps + future research paths

We review the landscape of cross-lingual transfer learning (CLTL) to tackle the massive imbalance in NLP resources across languages—especially African & other low-resourced ones.

This work helps us understand and build inclusive language tech. 🌍📚 #AfriNLP #CLTL

🚨 New Paper Out!
“Cross-lingual embedding methods and applications: A systematic review for low-resourced scenarios”
by Thapelo Sindane, @vukosi.bsky.social, and Abiodun Modupe.
Published in Natural Language Processing Journal.
🔗 doi.org/10.1016/j.nl...