Natural Language Processing for Automated Legal Document Analysis in Nigerian Law Firms

Legal document review in Nigerian law firms remains a labour-intensive, time-consuming process that limits the capacity of practitioners to serve clients efficiently, particularly in high-document-volume areas like commercial contract review and property due diligence. This study investigates the application of natural language processing (NLP) for automated legal document analysis in Nigerian law firms, with empirical evaluation of models on a corpus of Nigerian commercial contracts and case law documents. A dataset of 1,200 anonymised commercial contracts obtained from three Abuja-based law firms was preprocessed and annotated for clause classification, obligation extraction, and risk flagging tasks. Three NLP approaches are evaluated: a fine-tuned BERT model, a legal-domain-specific LegalBERT variant, and a GPT-4-based few-shot classification approach. Evaluation uses F1-score for multi-class clause classification and precision-recall for risk flag detection. LegalBERT achieves the highest F1-score of 0.87 for clause classification, while the GPT-4 few-shot approach performs comparably on risk flagging with significantly lower training overhead. The study critically examines dataset bias arising from the predominance of English-language contracts, noting limitations for Pidgin-English documents common in South South and South East contexts. Recommendations include building a Nigerian legal NLP corpus and developing domain-adapted embeddings for Nigerian statutory language. Keywords: natural language processing, legal document analysis, NLP, Nigeria, contract review