Space traveling : assessing the "soundness" of class labels in memory-based learning and the case of Middle Dutch spelling variation

Abstract: In this paper we highlight an aspect of previous research into lemmatization for Middle Dutch, a medieval language characterized by a lot of spelling variation. We briefly present a novel, memory-based learning method that assigns a similarity score to pairs of tokens. This method is based on assessing the soundness of a given class label, an untypical question in a kNN setting.