In order to approach the question of automatization, an online annotation experiment was set up which gathered (« crowd-sourced ») larger amounts of segmentation decisions for the single phenomenon of segment boundaries at gaps and pauses in the interaction. Evaluation of the experiment showed that a clear correlation between properties of the gaps (duration, filled or not) and their classification into segment boundaries exists, but is in itself not sufficient as a statistical basis for a reliable automatic segmentation method (Schmidt/Westpfahl 2018a).
