Transformer Based Abstractive Text Summarization for Hausa: A Low Resource NLP Approach

Abstract

The rapid proliferation of digital text across online platforms has created an urgent need for automated summarization systems capable of distilling essential information efficiently. For languages like Hausa, which occupies a prominent position among African languages yet remains significantly underrepresented in natural language processing (NLP) research, this need is especially pronounced. This paper presents an investigation into the application of the mT5_multilingual_XLSum transformer model for abstractive text summarization of Hausa language documents. Drawing on the GEM/xlsum Hausa dataset, this study constructs a comprehensive preprocessing and training pipeline that employs the T5Tokenizer and applies synonym-based data augmentation strategies to enrich the training corpus. Performance is evaluated through ROUGE metrics (ROUGE-1, ROUGE-2, ROUGE-L, and ROUGE-LSUM), with results demonstrating that the transformer-based approach substantially outperforms conventional deep learning baselines, including Long Short-Term Memory (LSTM) and Recurrent Neural Network (RNN) models. The mT5_multilingual_XLSum model achieves ROUGE-1, ROUGE-2, ROUGE-L, and ROUGE-LSUM scores of 63.73%, 41.54%, 51.79%, and 48.81%, respectively. These outcomes underscore the efficacy of transformer architectures in low-resource language summarization and open pathways for improved digital inclusion of Hausa speakers.