Who Builds African Language Technology? Academic Rules, Data Sovereignty, and Epistemic Equity

African language communities are poorly served by language technology, and the datasets and models that would change this are rarely built by the linguists best placed to build them. This article argues that academic promotion policy is part of the reason. We compare how six university and national bodies in Nigeria, South Africa and Kenya define recognised scholarly output. All reward articles, books and proceedings. None names a dataset, software tool or trained model as a recognised output, and several limit credit for the collaborative authorship that dataset work requires. Using decolonial artificial intelligence (AI), data sovereignty and epistemic justice as lenses, we argue that these rules leave the construction of language infrastructure to actors outside African universities and communities, and treat some forms of African linguistic knowledge as lesser scholarship. We propose a five-dimension assessment framework that fits existing scoring scales. One dimension, governance, draws on the CARE (Collective benefit, Authority to control, Responsibility, Ethics) and FAIR (Findable, Accessible, Interoperable, Reusable) principles. The framework replaces percentage author shares with contributor-role statements and guards against metric gaming. A worked example scores a tone-annotated Yorùbá corpus. The framework is untested; a Delphi study with African linguists, natural language processing (NLP) researchers and university administrators is the next step. The article speaks to promotion committees, to linguists and NLP researchers, and to scholars of AI and society who ask who builds language technology and who benefits.