The annotation of the modal particles in the GeWiss corpus A syntactic and semantic-pragmatic analysis of the PTKMA annotation

This article examines the automatic annotation of eight German modal particles in two sub-corpora of the GeWiss Corpus, which originate from the oral examinations of L1 and L2 examinees in a German academic context. Because of the poor reliability of automatic methods for POS tagging of spoken language data, the modal particles were checked manually for correctness using lists of criteria. In addition, the linguistic units ja, eben, halt, einfach, aber, mal, doch and denn which did not have a modal particle annotation, but had been automatically annotated with a different POS tag, were also checked for incorrect annotations, their modal particle properties were examined and the uses of these as modal particles were annotated. The results show that the POS tagging system has a very high error rate of 19,2% in the automatic annotations of the above-mentioned modal particles, and that it annotates the particles with widely varying reliability, ranging from 100% incorrect to 100% correct. Checking the non-PTKMA (modal and modulating particles) types ja, eben, halt, einfach, aber, mal, doch and denn for MP properties shows that several tokens exhibited this property.