Natural Language Processing is the historical name for the discipline of making computers work with human language. NLP predates the modern LLM era by decades — rule-based parsers, statistical machine translation, named entity recognition, and sentiment analysis were all active subfields throughout the 1990s and 2000s.
The arrival of large language models has largely absorbed NLP as a distinct discipline. Tasks that were once considered separate problems (translation, summarization, classification, question answering) are now all handled by the same foundation models with minor prompting or fine-tuning differences. The specialized algorithmic machinery that defined classical NLP has been displaced by general-purpose language models.
NLP remains a useful term when discussing production systems that solve specific language-handling problems — pipeline architectures that combine retrieval, classification, and generation — especially in domains where dedicated models outperform generalist LLMs or where cost constraints favor smaller task-specific models.