Analysis of incorrect POS-tagging in student texts with linguistic errors in German

The electronic learner corpus of student texts in German, the PACT, contains parts-of-speech (POS) tagging. This markup is performed automatically using RFTagger. Since are written by students, they may contain various kinds errors: grammatical, spelling, stylistic, and others. Sentences be formulated incorrectly, without taking into account rules language accepted norms. can affect work programs that process automatic mode, as a result, generate incorrect tagging needs to verified manually. purpose study investigate degree influence errors non-authentic on results part-of-speech Based expert error texts, 11 types were identified tagger quality. For each type error, ten sentences containing an selected from corpus. resulting pool was processed taggers RFTagger TreeTagger. parts speech suggested these compared with determined experts As result comparison, following patterns revealed: mistaken when writing non-declinable form adjective instead declinable; one word separately; absence suffix "-er" possessive adjectives formed geographical names; nouns lowercase letter; verb capital letter. case, article provides analysis forms causes POS-tagging, well differences two taggers. Taking revealed will allow more efficient organization POS-tagging verification German. also useful for developers