NLP
First systematic review exposes critical NLP gaps across Ghana's 73 indigenous languages
A comprehensive analysis of the Ghanaian NLP landscape reveals severe resource imbalance, with only Twi receiving meaningful attention while 70+ languages remain almost entirely unaddressed.
A new systematic review has mapped the state of natural language processing for Ghana's linguistic landscape, revealing stark disparities in research and resource availability. Researchers examining over 17,000 publications across major academic databases identified just 36 core studies addressing NLP for Ghana's 73 living indigenous languages—a deeply fragmented field marked by severe under-resourcing.
The imbalance: One language dominates#
The research exposes a profound concentration of effort. Twi-centric NLP development has grown modestly over time, supported mainly through religious text alignment projects and crowdsourced initiatives. Meanwhile, the remaining 70+ languages remain almost entirely neglected by the NLP research community. This mismatch means that while some linguistic infrastructure exists for one major language, the vast majority of Ghanaians' native languages have virtually no computational support.
Fragmentation across infrastructure#
Beyond the language-by-language gaps, the field lacks consistency in how work is shared and built upon. Dataset releases, pre-trained model checkpoints, and evaluation methodologies remain scattered and rarely distributed across the research community. This fragmentation prevents researchers from building on each other's work and creates repeated inefficiencies in the field.
A roadmap for low-resource settings#
The review translates its findings into a prioritized research agenda tailored to Ghana's specific challenges: regional linguistic variation, non-standardized writing systems, dialectal differences, and the absence of shared technical infrastructure. The authors frame their approach as a replicable template that could guide systematic review and research planning in other low-resource language contexts, offering a structured pathway for similar assessments worldwide.
The work underscores how computational linguistics has traditionally favored high-resource languages, leaving the majority of the world's linguistic diversity behind.
