Natural language processing (NLP)
Natural language processing (NLP) is the AI subfield concerned with giving computers the ability to read, understand, and generate human language. It spans both text and speech, from speech recognition that turns spoken audio into written text to the analysis and generation of prose.
The NLP pipeline reflects this dual nature. Acoustic models recognise the individual sounds, or phonemes, in an audio signal, converting them into a raw sequence of candidate words. A language model then interprets meaning from that sequence and corrects errors using surrounding context — the same pipeline described in voice-to-text.
Modern NLP is dominated by neural language models. Large language models learn grammar, word usage in context, and even reasoning patterns from vast text corpora without being explicitly given any rules. They compare billions of word and phrase usages to predict and generate sentences, like predictive text at massive scale.
The field has moved through three broad phases. The earliest work, from the 1950s through the 1970s, relied on hand-crafted rule-based systems that encoded linguistic knowledge as symbolic programs. These symbolic approaches could handle narrow, well-structured tasks but were brittle in the face of real-world language. The 2000s brought a shift to statistical methods — spam filters, recommenders, and early machine translation systems that learned probabilistic patterns from data. From the 2010s onward, the transformer architecture and deep learning took over, enabling models trained on internet-scale text that far surpassed earlier approaches.
NLP underpins a wide range of practical applications: chatbots, machine translation, automatic summarization, sentiment analysis, and AI assistants such as ChatGPT and Claude. It is also the foundation of AI coding assistants, which apply the same text-modeling techniques to source code.