It's hard to not mention ColorForth here as a primarily-text-based UI, although it ran in graphical mode and did its own text rendering. There were a few language-based innovations that tied closely in to the UI, including the use of color tags on pre-tokenized source code to both select display color, but also to select semantic category of the symbol (e.g. execute at compile time like a macro, execute at run time like a function call, define a new symbol). Probably more interesting is its use of a modified dvorak keyboard layout (later with qwerty support as well), with a consistent onscreen display of the keyboard mapping, indicating what each press will do at any time. I can see this concept generalizing to something like vi where the key-to-meaning mapping is equally dynamic and modal. Lots of other info at https://www.greenarraychips.com/home/documents/greg/cf-humin... and elsewhere.