NLP covers tasks such as translation, summarising, sentiment analysis, search, speech recognition and chatbots. It has moved from hand-written grammar rules, through statistical methods, to today's neural networks and large language models.
How it works
Computers cannot read words directly, so NLP turns language into numbers. The first step is tokenisation, splitting text into words or sub-word pieces. Classic NLP then applied a pipeline of separate tasks: tagging parts of speech, parsing sentence structure, and named entity recognition, which finds people, places and organisations. Early systems relied on linguists writing rules. From the 1990s, statistical methods learned patterns from large text collections, such as which words tend to appear together. In the 2010s, word embeddings represented each word as a vector, so words with similar meanings sat close together. The big shift came with the transformer architecture in 2017, whose attention mechanism reads each word in light of all the others. Models such as BERT, released by Google in 2018, were pre-trained on huge amounts of text and then adapted to specific tasks. Large language models extended this, handling many NLP tasks with a single model and just a prompt. Speech recognition and text-to-speech connect NLP to spoken language.
Why it matters
Almost every business runs on language: emails, reviews, contracts, support tickets, captions and calls. NLP lets software sort, search, summarise and respond to that language at scale. Creators benefit through auto-captions, translation, transcription and comment moderation. Businesses use it to spot customer sentiment, route support requests and pull key facts out of documents. Knowing the field helps you see that chatbots are one part of a much wider toolkit.
Where you will see it
- Analysing thousands of product reviews to find the most common complaints.
- Automatically routing support emails to billing, technical or sales teams.
- Generating captions and translations for videos.
- Extracting names, dates and amounts from contracts or invoices.
Limits and common misconceptions
- Sarcasm, slang, dialects and mixed languages still trip up many systems, especially for less widely spoken languages.
- Models learn from human text, so they can reproduce stereotypes and biased associations.
- Fluent output is not the same as understanding; systems can misread meaning while sounding correct.
How it differs from related terms
- Natural language processing (NLP) vs Large language model (LLM) NLP is the whole field and its many techniques; an LLM is one powerful kind of model used within NLP today.
- Natural language processing (NLP) vs Speech to text Speech-to-text converts spoken audio into written words, a step that feeds into NLP tasks such as summarising or analysing what was said.
Tools that use it
Frequently asked questions
What is NLP with an example?
NLP is technology that helps computers work with human language. A common example is your email app's spam filter, which reads messages and decides whether they are junk. Others include voice assistants understanding spoken commands, auto-generated video captions, and translation apps converting a menu into your language.
Is NLP the same as AI?
No. NLP is a subfield of artificial intelligence that focuses specifically on language. AI also includes computer vision, robotics, planning and more. Modern NLP relies heavily on machine learning, especially deep learning, and large language models have made NLP one of the most visible parts of AI.
Is NLP still relevant now that we have LLMs?
Yes. LLMs are themselves a product of NLP research, and the field still studies tokenisation, evaluation, multilingual performance, bias and efficiency. Smaller, task-specific NLP models also remain popular because they are cheaper, faster and more predictable for jobs like classification or entity extraction.
Related terms
- Model In AI, a model is a trained mathematical system, defined by its architecture and learned parameters, that turns an input into an output.
- Multimodal AI AI that can take in or produce more than one type of data, such as text, images, audio and video, and connect information across them.
- Neural network A machine learning model made of layers of simple connected units, artificial neurons, whose connection strengths are adjusted during training to learn patterns from data.
- No-code Building apps, websites or automations through visual editors, templates and settings instead of writing programming code.
- Open-weight model An AI model whose trained weights are publicly released, so anyone can download, run and often fine-tune it on their own hardware.