This is an old revision of the document!
Segmentation of captions depends on content
Captions can be part of the section containing an image, but may also form their own tile based on content.
Example: the caption in the image below does not belong with the main headline or the subsequent attribution/date line:
Attribution and Date Lines
Attribution (author information), dates, times and other similar 'metadata' information form one tile together.
Example: (single tile, even if we believe these are multiple utterances)
Digressions that clearly form a separate background giving component should not be analyzed in depth hierarchically: background information is seen as an indivisible block.