Natural language processing (NLP)
The aim is to let computers organize, search, transform or produce human language usefully. That includes identifying the intent of a question, comparing documents, summarizing, translating and assisting writing. Quality must be measured for the task, language and audience involved.
Text reviewed on
← Back to the 7 pillars
When it emerged
Machine translation was an early goal of computer language processing. In January 1954, IBM and Georgetown University publicly demonstrated a Russian-to-English translation system. Its vocabulary and rules were limited, but the experiment made visible a problem that would remain open for decades.
What it is
Studies methods for working with human language in tasks such as classification, search, translation, information extraction and text generation. LLMs are one family of models used for some of these tasks, not the whole field.
How it works
A system takes text, divides it into units and relates form, context and meaning to a task. Depending on its design, it may use linguistic rules, statistical models or models trained on large text collections. Translation, entity extraction and question answering need different outputs and evaluations; an LLM is only one class of tool in the field.
What it is for
The aim is to let computers organize, search, transform or produce human language usefully. That includes identifying the intent of a question, comparing documents, summarizing, translating and assisting writing. Quality must be measured for the task, language and audience involved.
Where it is used
Document search, translation, request classification and extracting names from contracts are typical applications. In customer support, for example, a system may identify a message's topic and retrieve the relevant policy before suggesting a response for review.
Limits and safeguards
Fluent text does not guarantee understanding or factual accuracy. A grammatically convincing sentence may still be wrong. Ambiguity, irony, regional variation and missing context make the task harder, so factual answers need sources and review in proportion to the risk.