We read each text at least four times, each pass devoted to a specific annotation task in INCEpTION:

1. In the first pass, we read the text for an overview, without making annotations, concentrating on understanding the content (Fig. 1).
2. The second pass is dedicated to annotating the expressions in the metadata layer, including any relations. We start by identifying words to tag, highlighting them, and selecting the appropriate type from the drop-down menu (Fig. 2). Once all the entities are selected, we connect related entities with the mouse, selecting the appropriate label from the menu for each relation.
3. In the third pass, we annotate expressions in the content layer and their corresponding relations. That is, we again start by identifying words to tag, highlighting them, and selecting the appropriate type from the drop-down menu. Once all the entities are selected, we connect related entities with the mouse, selecting the appropriate label from the menu for each relation (Fig. 3).
4. In the fourth and final pass, we then annotate co-references. As before, we identify potential co-references in the text, highlight them, and connect candidates with the mouse (Fig. 4).