An empirically-based spatial segmentation and coreference annotation scheme for comics

There is recent work in applying concepts from discourse analysis to comics in order to formalise visual content across comics sequences, however much of this work lacks empirical verification. We address this gap by assessing inter-annotator agreement on a preliminary annotation scheme for segmenting areas of comic pages, and assigning referents or classifications to these segmentations. A browser-based annotation tool and inter-annotator agreement measures are introduced. We find high agreement for segmentation tasks, while reference tasks show adequate agreement as well as instructive disagreements showing constraints on reader interpretation.