# What is Named Entity Recognition?

Also called NER.

Named entity recognition is a natural language processing task that locates spans of text referring to specific things and assigns each a type, such as person, organization, location, date, or monetary amount. It converts unstructured text into structured fields that other systems can filter, index, or store. Type schemes are defined per application rather than being universally fixed.

Output is normally represented as character offsets plus a label, often encoded with a tagging scheme that marks the beginning and the continuation of each span. This span level representation matters because entities can be several words long, can nest inside one another, and can overlap with ordinary vocabulary, so simple dictionary matching misses many real cases.

Approaches progressed from name lists and hand written patterns, to statistical sequence models such as conditional random fields, to fine tuned neural encoders, and more recently to prompting general purpose language models with a description of the type scheme. Each step reduced task specific engineering while increasing dependence on how representative the training or prompting examples are.

Recognition is usually followed by linking, in which a detected mention is resolved to a specific record in a knowledge base or database. Recognizing that a string denotes an organization is not the same as knowing which organization it is, and ambiguity, abbreviation, and duplicate names make that second step the harder one in most pipelines.

Difficulty varies with domain and language. Clinical, legal, and financial text uses vocabulary and abbreviations absent from general corpora, and languages without capitalization cues or with rich morphology require different modeling. Evaluation reports precision, recall, and their harmonic mean at the span level, where a partially overlapping prediction normally counts as an error.

## Key points

- Finds and types entity mentions in text.
- Typical types include person, organization, location, and date.
- Output is labeled spans, not just matched keywords.
- Often paired with linking to a knowledge base.
- Domain vocabulary and language features change difficulty sharply.

## In practice

An invoice arrives as plain text. Named entity recognition marks the supplier name as an organization, the issue date as a date, the total as a monetary amount, and the signatory as a person. Those fields populate an accounting record without manual typing. A supplier written in an unusual abbreviation is still detected, then resolved by a separate linking step.

## Related terms

- [Natural Language Processing](/en/glossary/natural-language-processing)
- [Entity Resolution](/en/glossary/entity-resolution)
- [Intent Classification](/en/glossary/intent-classification)
- [Knowledge Graph](/en/glossary/knowledge-graph)
- [Sentiment Analysis](/en/glossary/sentiment-analysis)

[Back to the AI Glossary](/en/glossary)