A Transformer-Based Language Model for Sentiment Classification and Cross-Linguistic Generalization: Empowering Low-Resource African Languages

Sentiment analysis is a widely studied area of natural language processing (NLP), however, a lot of attention has been given to languages with higher resources, like English. By concentrating on low-resource African languages including Hausa, Nigerian Pidgin, Igbo, Yoruba, Swahili, and Kinyarwanda for which there is a lack of curated datasets for sophisticated AI applications. This research intends to close the gap by examining the viability of current NLP strategies and creating new ones to meet this challenge. Sentiment analysis enables us to comprehend the feelings and viewpoints expressed in textual information. A transformer-based language model is presented in this research for low-resource African languages in order to classify sentiment and generalize across languages. We fine-tune the XLM-RoBERTa model on Hausa, Nigerian Pidgin, Igbo, and Yoruba, and then assess its performance in these languages as well as Swahili and Kinyarwanda, which the model had not previously seen. Our findings show that, even in languages with few linguistic resources, the model is effective at performing sentiment analysis tasks for known and unknown languages. This research helps close the digital language gap by extending natural language processing capabilities to under-resourced African languages.