Abstract
This work presents our recent advances in the field of automatic processing of sign language corpora targeting continuous sign language recognition. We demonstrate how generic annotations at the articulator level, such as HamNoSys, can be exploited to learn subunit classifiers. Specifically, we explore cross-language-subunits of the hand orientation modality, which are trained on isolated signs of publicly available lexicon data sets for Swiss German and Danish Sign Language and are applied to continuous sign language recognition of the challenging RWTH-PHOENIX-Weather corpus featuring German Sign Language. We observe a significant reduction in word error rate using this method.