Foundations
Natural Language ProcessingNLP
Also known as: computational linguistics
The field of AI focused on getting computers to understand, interpret and produce human language.
Natural language processing, or NLP, is the part of AI that deals with human language, getting computers to read, understand and write the way we speak and type. Every time your phone predicts your next word, a spam filter catches junk mail, or a chatbot answers a question, NLP is at work. Today's chatbots are the most visible result of decades of NLP research.
Natural language processing (NLP) is the field concerned with enabling computers to work with human language: understanding it, extracting meaning, and generating it. It covers tasks such as translation, summarising, sentiment analysis, question answering, speech recognition and text classification.
From rules to learning
Early NLP relied on hand-written grammar rules, which were brittle. It then moved to statistical methods learned from data, and finally to deep learning. The arrival of the transformer and large language models transformed the field: a single model can now handle many language tasks that once needed separate specialised systems.
Why it matters
NLP is the technology layer beneath chatbots, search, voice assistants and much of generative AI for text.
NLP is the study of computational methods for human language, spanning syntax, semantics, pragmatics and discourse across understanding (NLU) and generation (NLG) tasks. Its trajectory moved from symbolic/rule-based systems to statistical NLP (n-grams, HMMs, CRFs) to neural approaches (word embeddings, RNNs/LSTMs) and, decisively, to the transformer, whose pretraining-then-adaptation paradigm (LLMs) collapsed many bespoke pipelines into general foundation models.
Contemporary practice largely reframes discrete tasks (classification, extraction, translation, QA) as text-to-text generation or in-context prompting over a pretrained model, with fine-tuning or retrieval where needed. Enduring challenges include grounding and factuality (hallucination), long-context understanding, low-resource and morphologically rich languages, evaluation beyond fluency, and robustness, many of which persist precisely because scaling improved fluency faster than verifiable understanding.