The speaker discusses the evolution of natural language processing techniques, starting from rule-based approaches to the current corpus-based techniques that utilize statistics and prediction models, with fewer references to linguistics.