Mnduong writeup of Borthwick et al.

This is a review of Borthwick_1998_exploiting_diverse_knowledge_sources_via_maximum_entropy_in_named_entity_recognition by user:mnduong.

• This paper proposes a named entity recognition system using the maximum entropy framework.
• The task is to identify names from 7 categories, from the MUC-7 testbed.
• The method uses features such as capitalization, vocabulary words that appeared at least 3 times, section identifications, dictionary features. Furthermore, each feature is selected only if it's fired at least 3 times in the training data.
• The method yields state-of-the-art results (among statistical systems) when used by itself and highest results ever reported when used in combination with handcrafted systems.