Rbalasub writeup of Borthwick et al.

A review of borthwick_1998_exploiting_diverse_knowledge_sources_via_maximum_entropy_in_named_entity_recognition by user:rbalasub

This paper uses a maximum entropy approach to do NER using lexical features, dictionaries, sectional information and other token specific features. The approach works well because of the flexibility available to add virtually any kind of feature. As is common, viterbi search is used during decoding. The task of doing NER is framed as assigning tags to each token indicating if the token is in the beginning, middle, end of a named entity or if it's not in any NE at all. The system does even better when tags assigned by another system is added as an feature.

In general, the approach is solid and stands true even now 10 yrs after it was published.