This empirical submission reports on two original methods geared towards producing semantic annotations for the decompositional marker again. The two methods are (i) expert annotation based on a comprehensive set of guidelines and (ii) quality-controlled crowdsourcing with ensuing evaluation on the basis of the expert annotation. We report on a number of strategies for yielding a ‘crowd winner’ and present as the most promising candidate KMeans clustering of annotation vectors which are supplemented with corpus and annotational features. We report an observed accuracy of 85.54% with Cohen’s \(\kappa \) at 0.73.