presentsUNAIVERSE
The UN AI timeline

    Learn AI /Computer Vision and Language Technology /Natural language processing

    Natural language processingNLP

    Also called: NLP.

    Detail

    In five seconds

    What it is

    A set of computing methods for turning human language into structured signals that software can classify, search, translate, summarise or generate.

    In plain words

    NLP is a group of tools that helps computers work with human language, for example by finding key facts in text, translating it or producing a short summary.

    An everyday example

    A colleague receives a short field report in Arabic and needs to prepare a quick English note for a meeting. First, she highlights place names, dates and organisation names. Next, she underlines the main request and ignores repeated background details. Then she writes a two-sentence English summary for the meeting agenda. The work turns a full page of language into a few structured pieces and a new text suited to a clear task.

    Where the story stops working

    A person can use broad world knowledge, ask clarifying questions and notice sensitive implications. An NLP system usually relies on patterns in data and may not handle ambiguity, irony or missing context well. Also, one system often performs only one narrow task, while the colleague in the story smoothly combines several tasks.

    Remember thisNatural language processing turns text or speech into structured signals for tasks such as search, classification, translation and summarisation. It does not guarantee that the intended meaning has been captured.

    Understand it

    Understand it

    Why it matters

    Non-specialists meet NLP whenever language is processed at scale: search boxes, translation tools, content moderation, help desks, transcription, document sorting and chatbot features. Knowing the term helps you ask better questions about fit for purpose. What language goes in? What task is the system actually doing? What comes out, and who checks it? This matters because language tools can save time, but they can also miss nuance, reproduce bias, perform unevenly across languages and give fluent outputs that still need verification. Good decisions depend on matching the tool to the task, the risk level and the need for human oversight.

    How NLP turns language into task-specific output

    After viewing the diagram, a reader should be able to describe NLP as a process that takes language in, analyses patterns, and produces a task-specific result such as labels, extracted facts, translation or a summary.

    1. 01

      Language input

      The process begins with human language, such as a sentence, report, email or transcript.

      Draw a document icon on the left with sample text: "Floods affected 3 villages near Goma on Monday." Add a caption: input text or speech transcript.

    2. 02

      Break into usable pieces

      Software splits the language into parts such as sentences, words or subwords so it can analyse them.

      Show the sample text entering a box that separates it into tokens: Floods | affected | 3 | villages | near | Goma | on | Monday.

    3. 03

      Find patterns

      The system applies rules, learned patterns, or both to detect language features and relationships.

      Show a middle box with highlights over "3" labelled quantity, "Goma" labelled place, and "Monday" labelled date. Include a small note: learned or rule-based patterns.

    4. 04

      Choose the task

      Different NLP tasks use the analysed text differently, such as classification, extraction, translation or summarisation.

      Create a decision box with four labelled lanes: classify, extract, translate, summarise.

    5. 05

      Produce output

      The final result is shaped for a purpose, not always a full conversation.

      At the right, show four small output cards: topic=disaster update; entities=3 villages, Goma, Monday; translation in another language; summary="Flooding affected three villages near Goma on Monday."

    A left-to-right flow diagram with one sample text entering a series of stages and branching to several possible outputs. Read from input, to text processing, to pattern analysis, to task layer, to output examples. Each stage shows what changes, not decoration.

    Hovering or tapping each stage highlights what information is added or transformed before the next stage.

    Describe this without the diagram

    The diagram shows NLP as a five-stage flow from left to right. First, a text sentence enters the system as input. Second, the sentence is split into smaller pieces such as words and numbers. Third, the system marks useful patterns like quantities, places and dates using rules, learned patterns or both. Fourth, a task layer directs the analysed text toward a specific purpose such as classification, information extraction, translation or summarisation. Fifth, the output appears in a task-shaped form, for example a topic label, a list of extracted facts, a translation or a short summary. The main lesson is that NLP converts human language into structured signals and outputs suited to a particular job.

    With reduced motion: A static left-to-right diagram with the same five stages and labels, using arrows instead of animation to show how input text becomes analysed language features and then a task-specific output.

    Natural language processing, or NLP, covers tools that work with human language in text or speech. The input might be an email, a report, a chat message, a transcript or a question spoken into a device. The output might be a label, a translation, a short summary, an extracted name or a drafted reply.

    A system usually starts by breaking language into smaller pieces, such as words, subwords or sentences. It then looks for patterns learned from many examples, or follows hand-written rules, or combines both. For example, it may detect that "cholera" is a health term, that "Bangkok" is a place name, or that two sentences say similar things in different words.

    NLP is broader than chatbots. Many deployed systems do not hold a conversation at all. They do narrower tasks such as spam filtering, machine translation, sentiment analysis, named entity recognition, search, transcription support or summarisation.

    It is also different from a large language model. A large language model is one kind of system used in some NLP applications. NLP is the wider field that includes older rule-based systems, statistical methods and many task-specific tools.

    Because language depends on context, culture and wording, NLP output can be useful without being fully reliable. A system may miss sarcasm, handle one language better than another, or perform badly on low-resource languages, meaning languages with less digital training material available.

    How “Sorting and rewriting a short field report” maps onto the real thing

    In the storyIn the AI system
    highlighting place names, dates and organisation namesextracting entities such as locations, times and organisations from text
    underlining the main requestidentifying intent or key information
    ignoring repeated background detailsselecting salient content for summarisation
    writing a two-sentence English notetranslating or generating task-shaped output from the original text

    A worked example

    Before informals, a desk officer receives a short French notice and needs a concise English update for the front office.

    Starting from: French text: "La réunion technique commence à Genève mardi matin. Les documents ont déjà été circulés. La salle a changé."

    1. The system splits the text into sentences and words, then identifies the source language as French.
    2. It recognises "Genève" as a place and "mardi matin" as a time expression.
    3. A translation component converts the sentences into English.
    4. A summarisation component selects the meeting time and room change as the points requiring attention, while omitting the less urgent sentence about circulated documents.

    Result: English update: "The technical meeting starts in Geneva on Tuesday morning. Please note that the room has changed."

    In a United Nations workplace

    A humanitarian affairs officer in a country office is assembling the quarterly report to a donor by close of business. An approved language-processing tool classifies narrative updates from implementing partners by activity, extracts place and organisation names, and proposes short summaries for the results section.

    The tool converts varied prose into consistent tags and draft text, reducing manual sorting. The officer still checks each proposed classification and summary against the source before the report goes into clearance.

    Before you try thisPartner submissions may contain personal, protection-sensitive or unpublished operational information. They must not be entered into a tool unless it is approved for that material and access is appropriately controlled. Extracted names, classifications and generated wording must be verified before circulation.

    Where you may hear it

    • At a vendor demo for multilingual document search
    • During an ICT governance board discussion on automated transcription
    • In a procurement meeting reviewing translation and summarisation functions
    • At an inter-agency working group considering automated tagging of reports

    Avoid confusion

    Avoid confusion

    Often confused with

    Common misconceptions

    • NLP means a system can handle language like a person does.

      Not necessarily. Many NLP systems do one narrow task well, such as finding names or translating common phrases, but still fail on ambiguity, sarcasm, domain-specific wording or unusual languages. Useful performance on one task does not mean broad human-like language ability.

    • NLP and chatbots are the same thing.

      A chatbot may use NLP, but NLP is the broader field. Spam filters, machine translation, search ranking, sentiment analysis and entity extraction are all NLP applications even when there is no conversation at all.

    • If the output sounds fluent, it is probably correct.

      Fluency and accuracy are different. A system may produce natural-sounding text while mistranslating a number, dropping a warning or misreading context. Language output should be checked, especially for high-stakes work.

    Quick check

    Which example best shows natural language processing rather than computer vision?

    Go deeper

    Go deeper

    The simple version

    NLP is a group of tools that helps computers work with human language, for example by finding key facts in text, translating it or producing a short summary.

    The technical version

    NLP is the field concerned with computational processing of natural language in text and speech. Typical pipelines include tokenisation, sentence splitting, normalisation, feature extraction or learned representations, followed by task-specific components such as classification, sequence labelling, parsing, retrieval, translation, summarisation or generation. Systems may be rule-based, statistical, neural, or hybrid. The field includes both understanding-oriented tasks, such as named entity recognition and sentiment analysis, and generation-oriented tasks, such as machine translation and summarisation. Performance depends heavily on training data, annotation quality, domain match, language coverage and evaluation method.

    Reviewed external resources

    Where this came from

    Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.

    The one thing to remember

    Natural language processing turns text or speech into structured signals for tasks such as search, classification, translation and summarisation. It does not guarantee that the intended meaning has been captured.