Natural Language Processing

Natural language processing (NLP) is a branch of artificial intelligence that enables computers to process, analyse and generate human language. It combines linguistic knowledge with computational methods so that machines can work with unstructured language in text or speech-derived form.

How NLP works

An NLP system converts language into representations that algorithms can process, identifies relevant patterns or meaning, and produces an output such as a label, summary or response. Modern systems commonly use machine learning, including deep learning and large language models, alongside linguistic rules and statistical methods.

• Pre-processing may divide text into words or other units, normalise it and identify grammatical structure.
• The system uses context to resolve, with varying success, ambiguity in words, sentences and user intentions.
• Speech recognition can first convert spoken language into text, after which NLP methods process the content.

Major tasks and applications

NLP supports both natural language understanding, which extracts information or intent, and natural language generation, which produces human-readable language.

• Core tasks include text classification, information extraction, machine translation, question answering, summarisation and conversational assistance.
• Named-entity recognition identifies categories such as persons, places, organisations and dates within text.
• In disaster management, NLP can classify distress messages, extract locations and needs, translate multilingual communications and summarise situation reports.

Limitations and safeguards

Human language is context-dependent, ambiguous and culturally varied, so NLP outputs remain probabilistic rather than infallible. Systems may reproduce training-data bias, misunderstand low-resource languages or generate plausible but incorrect content.

• NLP cannot independently establish whether a statement is true, so critical outputs require human verification.
• Disaster applications require attention to privacy, data security, misinformation and unequal performance across languages.
• Performance depends on representative data, clear evaluation and integration with reliable official information systems.