Transformer-Based Model for Named Entity Recognition in Hausa Language Texts

Named Entity Recognition (NER) is a fundamental problem in Natural Language Processing (NLP) that focuses on identifying and classifying textual entities such as people, places, organizations, dates, and numerical expressions. Although transformer-based architectures have greatly improved NER performance across major global languages, there has been little gain for low-resource African languages like Hausa. The lack of annotated corpora, language variety, orthographic irregularities, and limited computational resources continue to impede the development of strong Hausa NLP systems. This paper presents a transformer-based framework for Named Entity Recognition in Hausa language texts that uses recent deep learning architectures for low-resource language processing. The proposed framework combines Hausa textual data collection, preprocessing, manual annotation, transformer fine-tuning, and model evaluation. Multiple transformer models, such as Multilingual Bidirectional Encoder Representations from Transformers (mBERT), XLM-RoBERTa, and AfriBERTa, are being investigated for their effectiveness in recognizing named entities in Hausa texts sourced from news articles, social media platforms, educational materials, and online repositories. The study uses entity annotation schemas, standardized NLP preprocessing methods, and performance evaluation metrics like Precision, Recall, Accuracy, and F1-Score in a supervised learning approach. The study intends to provide a transformer-based entity recognition system, a reusable Hausa NER dataset, and empirical insights into deep learning applications for native African languages. The findings demonstrated the feasibility of transformer architectures in improving Hausa language entity extraction and advancing the broader ecosystem of low-resource language technologies.