Full dataset and scripts on GitHub. (To be released after paper has been accepted.)
In this paper, a case alternation in German measure noun phrases is examined from the perspective of cognitively oriented corpus linguistics. In certain pseudo-partitive constructions, the embedded kind-denoting noun either agrees in its case with the measure head noun (eine Tasse guter Kaffee ‘a cup of good coffee’) or it stands in the genitive (eine Tasse guten Kaffees). My analysis is formulated in terms of Prototype Theory. I assume that the choice of alternants is influenced by non-alternating neighbouring constructions representing the prototypes more directly. I argue that the frequencies with which individual lemmas occur in these prototypical non-alternating constructions partially predict which alternant is chosen and that the genitive alternant is prototypical of strongly grammaticalised measure nouns. I present a large-scale corpus study using the DECOW corpus to support this theory. In the statistical analysis, I compare Maximum Likelihood estimators to Bayesian estimators (recently proposed for similar studies), showing that results predictably do not differ. Finally, two experiments (forced choice and self-paced reading) are reported. The usage-based findings show a clear correlation with the behaviour of cognitive agents in both experiments. The present study therefore contributes to the well-established field of research into alternations using corpus and experimental methods.