Take the sentence “Acme Corp signed a $2 million contract with Jane Lee in Chicago on March 3.” An NER system would tag “Acme Corp” as an organization, “$2 million” as money, “Jane Lee” as a person, “Chicago” as a location and “March 3” as a date. A company could run that over thousands of contracts to build a searchable table of who signed what, where and when.
NER used to be done with hand-built rules and specialized statistical models. Neural networks improved it in the 2010s, and today large language models can often extract entities from a plain-English instruction, though dedicated NER models are still common when speed, cost or consistency matter.
Why it matters
NER is a building block for turning messy text into structured data. It powers features such as search filters, document processing, customer-support routing and redacting personal information before data is shared.
No published Chain of Thought episode discusses NER in depth yet, so this entry relies on the sources below.
Related terms
Natural language processing is the broader field. Semantic analysis covers working out meaning beyond names.
Sources
- Jurafsky and Martin: Speech and Language Processing (3rd ed. draft) — Standard open textbook whose sequence-labeling material defines named entities and the NER task of finding and classifying them; checked October 24, 2026.
- Lample et al.: Neural Architectures for Named Entity Recognition — Neural NER models that tag person, location and organization names without hand-built features, evaluated on standard benchmarks; checked October 24, 2026.