Black October, explainedAppsGuidesAI DictionarySmall FixesOn RecordAboutFollow the lab

Natural language processing

Also called NLP

Natural language processing, or NLP, is the part of AI that helps computers read, understand and produce everyday human language, written or spoken.

Inside language modelsCurious

In plain words

Computers are happiest with tidy data: numbers, forms, tables. People talk in messy sentences full of slang, typos and hidden meaning. NLP is the field that bridges the two.

When we type "book a cab to the airport tomorrow at 6 am", an app has to work out what we want (a cab), where (the airport) and when (tomorrow, 6 am). Pulling that meaning out of ordinary words is NLP.

We use it all day without noticing: spam filters, autocorrect and keyboard suggestions, voice assistants, translation apps, search engines that understand questions, and chatbots.

NLP is much older than ChatGPT. It began with hand-written grammar rules in the 1950s, moved to statistics in the 1990s, and runs mostly on neural networks today. Large language models are its most powerful tool so far.

Everyday example

A new phone operator at a busy restaurant

Callers say "two butter chicken, one dal, less spicy, to the blue gate near the temple". A good operator turns that into a clean order slip: items, quantities, notes, address. She handles accents, half-finished sentences and "actually, make that three". NLP is the computer's version of that operator: from loose human talk to something it can act on.

Where the example breaks: The operator understands food, places and people. An NLP system only maps patterns of words to outputs, so an unusual phrasing or a joke can trip it up in ways that would never fool a person.

Go deeper

How it actually works

NLP sits where computer science, linguistics and machine learning meet. It covers both text and speech.

  1. Rules era

    Early systems used rules written by hand. In January 1954, the Georgetown-IBM experiment translated just over 60 Russian sentences into English using a 250-word vocabulary and six grammar rules. Rules were precise but broke on anything unexpected.

  2. Statistics era

    From around 1990, researchers at IBM and elsewhere let computers learn probabilities from large collections of real text, for example which English phrase usually matches a French one. Systems became sturdier, but each task still needed its own model.

  3. Neural era

    Neural networks learned word meanings as vectors (embeddings) and handled whole sentences. In 2016, Google switched its Chinese to English translation to a neural system that cut errors by 55 to 85% on major language pairs, it said.

  4. Foundation models

    Since 2018, one large pre-trained model, such as BERT or GPT, is adapted to many tasks at once. Today a single LLM can translate, summarise, classify and chat, jobs that used to need separate systems.

  5. Classic tasks

    Splitting text into tokens, tagging parts of speech, spotting names and places (named entity recognition), working out what "it" refers to, judging sentiment, translating, summarising, answering questions and turning speech into text.

First public demo
Georgetown-IBM machine translation experiment, January 1954
Statistical turn
"A Statistical Approach to Machine Translation", IBM researchers, 1990
Neural switch
Google Translate's neural system, September 2016
Standard textbook
Jurafsky and Martin, "Speech and Language Processing", free online, latest draft August 2026

Questions people ask

What is the difference between NLP and LLMs?

NLP is the whole field of getting computers to work with human language. A large language model is one powerful tool inside that field. Think of NLP as medicine and an LLM as one very effective treatment.

Is ChatGPT an example of NLP?

Yes. ChatGPT is a chatbot app built on large language models, which are the most advanced NLP technology today. It reads our language and writes language back, which is exactly NLP's job.

What are examples of NLP in daily life?

Email spam filters, autocorrect and next-word suggestions, voice assistants, Google Translate, search engines, customer service chatbots, and apps that summarise reviews or sort messages. Most of it runs quietly in the background.

Is NLP the same as neuro-linguistic programming?

No. Neuro-linguistic programming is a self-help and communication method from the 1970s that shares the initials. In AI and computer science, NLP always means natural language processing.

Does NLP work for Indian languages?

Increasingly well, though often less well than for English. Tokenizers split many Indian scripts into more pieces, which makes processing slower and costlier (tokenization). Newer models train on far more languages, and newer tokenizers are narrowing the gap.

Further reading

Prepared by Mundhra Labs. Reviewed by Aditya Mundhra. Checked October 2026.