Entity Extraction Without Language-Speci c Resources

We describe a named-entity tagging system that requires minimal linguistic knowledge and thus may be applied to new target languages without signiicant adaptation. To maintain a language-neutral posture, the system is linguistically na ve, and in fact, reduces the tagging problem to supervised machine learning. A large number of binary features are extracted from labeled data to train classiiers and compu-tationally expensive features are eschewed. We have initially focused our attention on linear support vectors machines (SVMs); SVMs are known to work well when a large number of features is used as long as the individual vectors are sparse. We call our system SNOOD (Hop-kins APL Inductive Retargetable Named Entity Tagger).