Hi,
For the core/minimal XLIFF I would go for an even simpler model than Rodolfo's:
-- I think segmentation is too important to not be part of the core structure of XLIFF, and the representation as extra info like it is the case in 1.2 (because of the need to be backward compatible with 1.1) is not adequate. It does not mean a file must always be segmented, just that representing a segmented content must be simple and the processing of segmented vs un-segmented content should be seamless. For simplicity I've represented this as