annotation

Named Entity Recognition (NER)

Labeling spans of text as people, organizations, locations, or other defined entity types.

What this means for trainers

Boundary precision matters as much as category choice — get the span exactly right, not just approximately right, since downstream systems depend on exact character offsets.

Named Entity Recognition tags specific spans within a text, not whole documents or sentences, as instances of predefined categories like person, organization, location, or date. Unlike document classification, which labels a whole piece of text once, NER requires finding and bounding every relevant mention within it, which can mean dozens of separate spans in a single document.

NER is foundational infrastructure for a lot of downstream NLP work. It is often the first pass before entity linking, which maps a mention to a specific real-world entity, or relation extraction, which labels how two entities relate to each other in the text.

Ambiguity in NER usually comes from entity boundaries, such as whether a title counts as part of a person's name, and from overlapping categories, such as whether a company name is a location, an organization, or both depending on context. Both are resolved through clear guidelines that spell out exact rules rather than relying on annotator intuition.