Files are of type 0x808097B8
They contain similar/related dialogue. e.g. All of a strike's dialogue will be in one file.
The file has a top-level array of 0x80809729
blocks (which I call a "DialogueTree") which represents all the different dialogue groups of dialogue. e.g. Each "trigger" of dialogue in a strike will be one of these top level entries.
There is roughly a nested tree of three types of data:
0x8080972d
Branch, that describes random dialogue or condition-specific dialogue0x8080972a
Sequence, for a sequence of multiple lines of dialogue0x80809733
Line, an actual line of dialogue. caption, narrator, audio file
All these classes have various hashes which seem to be unique to the file, but not globally unique. I think they're closer to a "type" hash (or maybe there's just no uniqueness guarentee/requirement). e.g. Zavala and Ikora's idle tower dialogue share the same hash.