Memorybased onestep namedentity recognition Effects of seed list features, classifier st.pdfVIP

Memory-based one-step named-entity recognition: Effects of seed list features, classifier stacking, and unannotated data Iris Hendrickx and Antal van den Bosch ILK / Computational Linguistics Tilburg University, The Netherlands {I.H.E.Hendrickx,Antal.vdnBosch}@uvt.nl 1 Outline For example, for the English word Indian the following feature representation is made: Indian NNP I-NP We present a memory-based named-entity recognition 1 0 0 0 0 0 0 0 I i a n AJ0-NN1 N-s system that chunks and labels named entities in a one- I-MISC, where NNP is the provided PoS tag, I-NPthe shot task. Training and testing on CoNLL-2003 shared chunk tag; the binary features represent the orthographic task data, we measure the effects of three extensions. features (where in this case only Begin cap is positive); First, we incorporate features that signal the presence of AJO-NN1 is the PoS tag of the BNC-trained-tagger; wordforms in external, language-specific seed (gazetteer) N-s is the lemmatizer output for noun-singular; the last lists. Second, we build a second-stage stacked classifier element, I-MISC, is the annotated class label. that corrects first-stage output errors.