Developing a Setswana Grammar Checker: Advancing Language Practice and Applied Linguistics via NLP

Abstract: This paper presents the development of a grammar checking system for Setswana, a Southern Bantu language widely spoken in Southern Africa but lacking foundational natural language processing (NLP) tools. Despite its official status in Botswana and South Africa and substantial speaker population, Setswana remains underrepresented in digital language technologies. Addressing this gap, we introduce a grammar checker based on a Long Short-Term Memory (LSTM), designed to classify declarative sentences as grammatically correct or incorrect. The system was trained on a manually curated and annotated corpus of 1,700 Setswana sentences, capturing key morphosyntactic patterns characteristic of the language. Achieving a classification accuracy of 96%, the model demonstrates the viability of neural approaches for applied linguistic tasks in low-resource settings. This work contributes to applied language practice by supporting grammar teaching, educational technologies, and digital literacy for Setswana, and offers a scalable framework for enhancing language resources across the Southern African linguistic landscape