In recent years, Natural Language Processing (NLP) has rapidly advanced, with sophisticated language models paving the way toward improved information extraction (IE) methods. Despite these advancements, the application of modern NLP techniques within the Mexican legal system (Spanish-Mexico) remains underexplored. This study investigates whether BERT-based models can effectively and dynamically extract legal references within this challenging framework. A comprehensive comparison was conducted between traditional methods—regular expressions and rule-based approaches—and contemporary transformer-based techniques. The experiments encompass four critical tasks: Article Segmentation, Law Reference Extraction, Article Mention Detection, and Entity Linking. Results indicate that while regular expressions and rule-based methods achieve high precision in structured texts, BERT-based NER models demonstrate superior adaptability and recall, particularly in handling unstructured and diverse legal documents. However, for tasks demanding stringent accuracy, such as Entity Linking, traditional rule-based approaches outperform Relation Extraction (RE) models due to their higher precision. This duality underscores the potential of integrating both traditional and modern NLP techniques to optimize legal text processing in Mexico. Our findings contribute to the expanding field of legal NLP by addressing unique challenges within the Mexican legal system and providing a foundational framework for future research efforts aimed at enhancing IE processes in complex legal environments.