Proper Noun Recognition and Classification Using Weighted Finite State Transducers

This paper presents a new approach to proper noun recognition and classification in which the knowledge of ambiguities within morphological analyses is used exhaustively in the analysis. Here a proper noun recognizer/classifier is defined by proper noun context patterns on the one hand and by a filter that takes the ambiguity information into account on the other hand. Furthermore, techniques like a lemma based coreference resolution or the softening of the closed world assumption made by the morphology are presented which improve the analysis. The approach is implemented by weighted finite state transducers and tested within the analysis system SynCoP via a hand-written grammar.