German deverbal nominalizations in -ung denote a broad variety of sortal types, including events, result states and different kinds of objects, thus being a typical instance of systematic polysemy, or more precisely an instance of Nunberg’s (1995) ‘dense metonymy’. Brandtner & von Heusinger (2010) argue that these nominalizations do not have a complex lexical representation such as Pustejowsky’s dot objects, but rather an underspecified representation that is specified in the compositional process, pointing out the unacceptability of copredications involving predicates with different selectional restrictions. However, there are such constructions that are acceptable, counter to their predictions. To account for these they apply a second, coercion operation, namely Nunberg’s (1995) pragmatic process of predicate transfer; they thus explain copredication of nominalizations by two different processes: the first predicate semantically restricts the nominalization to one particular type, while the second predicate is subject to predicate transfer, such that it then can be applied to the determined type of the nominalization. According to Nunberg, this predicate transfer process is restricted by two pragmatic conditions: (i) there must be a salient functional relation between the original and the shifted meaning, and (ii) there must be ‘noteworthy’ information expressed by the sentence. In this paper we investigate Brandtner & von Heusinger’s account and the predictions made by Nunberg experimentally by testing the acceptability of reading shifts using judgement experiments, carefully controlling for contextual and lexical variables. The results reveal that the effect of the salience and noteworthiness conditions, which we reformulate as one single condition on relatedness, is psychologically real and accounts for the contrasts in readings in Nunberg’s specific examples. However, the results also show that the effect is not specific to the reading shift environments, but is rather a background coherence effect of priming or plausibility. The analysis as predicate transfer would thus corroborate the claim that there is complex representation for nominalization, but rather an incremental specification process.