This dissertation contributes to the growing body of research on non-canonical questions by investigating meaning variations in German question tags from empirical, theoretical and computational perspectives. German exhibits a wide range of invariant tags, which are typically interchangeable across their contexts of occurrence but are nevertheless subject to certain constraints. The felicity conditions of these tags have not yet been fully explained, as previous research is limited to several small-scale empirical studies focusing on individual tag forms rather than examining the full array of tags used in this language. To address this gap, the usage patterns of the tags “gell”, “ja”, “ne”, “nicht” and “oder” are investigated by means of corpus-based and experimental methods. Based on the collected evidence, the felicity conditions of these tags are identified, and the cases of their (non-)interchangeability across different types of contexts are determined. The meaning contributions of these tags are then formalized within the scoreboard model of discourse (Malamud & Stephenson, 2015). Furthermore, a computational approach is developed to test the ability of language models, exemplified by BERT (Devlin et al., 2019), to predict discourse-based variation in the use of these tags, with the aim of laying the groundwork for computational modeling of tag questions and related utterances.
Yulia Clausen (Thu,) studied this question.