Named entity recognition without domain-knowledge using conditional random fields

This paper addresses the problem of not us-ing any domain-knowledge in named entity recognition (NER) tasks. Experiments on two well-known datasets show that the cur-rently mostly used technique – conditional random fields (CRF) – achieves results which are respectable. It is discussed if it is accept-able to pass on better results to get results in a faster and modular way. 1. Conditional Random Fields CRFs are undirected graphical models (Lafferty et al., 2001). In theory of CRFs a set of states Y is glob-ally conditioned by an observation sequence X. In the case of NER the states are the alphabet used fo