Data, Licensing and Community Rights: Reflections from the Masakhane Fireside Chat
The Masakhane fireside chat on “Data, Licensing and Community Rights” held during the Deep Learning Indaba 2026 brought together questions that are becoming increasingly important as African language datasets move from community spaces into research, commercial development and AI systems.
The conversation was between Nander Esmeralda Ndam, Legal Practitioner and Researcher at the Data Science Law Lab, and Chris Emezue, a researcher and member of the Masakhane Research Foundation. It was informed by Chris’s 2026 research with GSMA, which audited 300 African language datasets and found that more than 30% did not have identifiable licences.
Rather than treating licensing as a technical or administrative step, the discussion approached it as a question of governance, rights, and responsible dataset design.
Looking beyond the dataset
One of the central ideas emerging from the conversation was that a dataset should not be viewed simply as a collection of information. Particularly where datasets contain speech, oral histories, and other community-generated material, questions of ownership, consent, control, and benefit-sharing can be just as important as the data itself.
This becomes particularly significant in the context of AI. A dataset may be created for research but subsequently used to train commercial models, develop products, or redistributed to third parties. The permissions attached to the dataset, therefore, have consequences far beyond its initial purpose.
The conversation highlighted five fundamental questions that every dataset should be able to answer:
- Who owns or controls the dataset?
- Who may use it?
- For what purposes may it be used? For example, research, commercial development, AI training, or redistribution.
- Which legal framework governs the dataset and its use?
- What, if anything, returns to the contributors or communities whose data is included?
These questions suggest that licensing should be considered at the point of dataset and product design, rather than after the dataset has already been created.
Choosing the right licence
The discussion also considered the limitations of conventional licensing frameworks. Standard open licences can provide useful clarity around permissions and restrictions, but they may not adequately capture the interests, expectations, or cultural considerations of every community.
This is where community-oriented approaches such as NOODL-Nwulite Obodo Open Data Licence become particularly relevant. The discussion explored how creators need to first understand the rights they hold, the rights and expectations of contributors, the uses they wish to permit, and the restrictions or protections that may be necessary before selecting an appropriate licensing framework.
The broader point was that there is no single licence that will necessarily work for every dataset. The licence should reflect the nature of the data, the rights attached to it, and the intended ecosystem of use.
An important misconception: unlicensed does not mean free
One of the most important clarifications from the discussion was that an unlicensed dataset is not necessarily free to use. The absence of a licence does not automatically grant permission. Instead, it creates uncertainty about what users are legally permitted to do. For a researcher, startup or large AI company encountering a valuable dataset without a licence, the questions therefore become:
Can I copy it? Can I use it to train an AI model? Can I share it with collaborators? Can I commercialise products or outputs derived from it? What rights do the people represented in the dataset have?
Where these questions cannot be answered, the resulting uncertainty can translate into legal, ethical and commercial risk.
Questions from the community
The discussion generated practical questions from participants, including how copyright applies to individuals and companies that create or contribute data; how dataset creators should determine the appropriate licence; whether uploading a dataset to GitHub is sufficient to establish or demonstrate ownership; and what the process of licensing actually involves.
These questions reinforced an important point: dataset licensing sits at the intersection of several areas of law and governance. Copyright is only one part of the picture. Ownership, contractual rights, consent, data governance, community interests, and commercialization can all shape what can legitimately be done with a dataset.
Licensing as dataset architecture
The strongest takeaway from the session was that licensing should be treated as part of dataset architecture.
A responsible dataset should not only contain data. From the outset, it should establish who controls it, what uses are permitted, what legal framework applies, what rights contributors and communities retain, whether there are benefit-sharing expectations, and under what terms the dataset may be accessed, reused, or commercialised.
This is particularly important for African language data, where the value of datasets may extend beyond their immediate research purpose. As African language resources increasingly contribute to the development of AI systems, the governance choices made at the dataset level can influence who benefits from that value and who retains control over its use.
The conversation therefore moved the question of licensing away from “Which licence should I put on my dataset?” towards a more fundamental question:
“What rights, responsibilities, and relationships should this dataset recognise—and how should the licence reflect them?”
That shift in perspective may be one of the most important steps towards building more responsible and community-conscious data and AI ecosystems in Africa.
The fireside chat provided a useful space for researchers, practitioners, and members of the broader AI and data community to engage with the legal questions surrounding the creation and use of African datasets.
~Written by Nander Esmeralda Ndam
