Glossary
What is natural language processing?
In short
Natural language processing (NLP) is the field of AI that teaches computers to understand, analyze and create human text.
AI basics · NLP
MMOOZZGG.com editorsChecked: 1 min read
For a child
People talk in different ways: they joke, mix up the word order and make spelling mistakes. A computer understands strict commands better, like in the game “robot, step forward”. Scientists spent a long time teaching machines to understand ordinary speech the way a friend does: who did what, what the request is, whether a text is sad or happy. Now a computer can translate a fairy tale into another language or retell a long book in a few lines.
In plain words
A natural language is a language that people speak, as opposed to a programming language. A computer first cuts text into words and pieces of words, then turns them into numbers, and a model uses these numbers to solve a task. The tasks vary: understand what an email is about and which department should get it, find dates and amounts in a contract, translate, shorten a long report, answer a question. Translation apps, spell check in email and the voice assistant in your phone all run on these methods.
For business
Every company has texts: customer emails, transcribed calls, contracts, reviews, requests. NLP methods sort incoming mail and chats, pull company details from documents, find answers in a knowledge base and rate the tone of reviews. Today large language models solve most of these tasks. So the cost depends on the volume of text in tokens, the choice between a cloud and your own server, and the accuracy you need. For any language other than English, test quality on your own texts: models confuse industry terms and abbreviations most often. Personal data in messages needs a separate decision about where the model runs. This review is part of an AI assistant pilot.
For a specialist
Natural language processing covers natural language understanding and natural language generation. Typical tasks: tokenization, morphological and syntactic analysis, named entity recognition, classification, machine translation, summarization and question answering. Methods moved from rules and statistical models to neural networks. The Transformer architecture (Vaswani et al., 2017) dropped recurrence and convolution and relies only on the attention mechanism. BERT (Devlin et al., 2018) set the pattern “pretraining on unlabeled text, then fine-tuning with one output layer” and beat earlier results on 11 tasks. Modern large language models solve most NLP tasks with one model that follows instructions. For Russian, rich morphology and free word order matter: they affect tokenization and the quality of entity extraction.
Example
A legal department uploads a hundred supply contracts. The model finds the payment term, penalties and termination clause in each one and puts them in a table, and a lawyer checks only the disputed points.
Related terms
Questions and answers
Sources
Where companies use it
Not sure whether your company needs this?
AI solutions map