Named Entity Recognition for Danish

Named entity recognition is a complex but rewarding task with a number of obvious applications- semantic corpus annotation, information retrieval, text indexing, to name but a few. As in other areas of NLP, more work has been done on English than on lesser languages, and Danish, on the brink of extinction as a language of science and research, seemed a particularly worthwhile target, and after doing independent investigations for some time, I joined the nordic Nomen Nescio project (Johannesen, 2002) about a year ago. The work described in this article is based on corpus data from the Danish twin corpora Korpus90 and Koprus2000 (about 56 million words), which have recently been made available in grammatically annotated form. The annotation was a joint venture between the VISL project at Southern Denmark University and the Danish Society for Language and Literature (DSL), who had compiled the corpora from a variety of sources, including both literature, news and science texts. Word based morphosyntactic information was added with a multi-level Constraint Grammar parser (DanGram, Bick 2001), a method that allows the annotation of running text with a high degree of robusticity and a comparatively low percentage of errors (less than 1 % for word class). Tag types covered part of speech, inflexion and syntactic function (e.g