Terms can have FA or tags equally well, but we don't want to merge
FA with BAI generally, to avoid things like "se fa" and ".i fa bo",
which are nonsense.
I agree that {se fa} has no clear interpretation upon first examination. However, {.i fa bo} can be interpreted like any other {.i TAG bo} construct.
.i broda .i TAG bo brode -> .i broda TAG lo su'u brode
Hence,
.i broda .i fa bo brode -> .i broda fa lo su'u brode
This provides us with another way to do essentially what {la'e di'e} does. For instance,
.i mi pu pensi la'e di'e .i lo mi bruna cu cmalu mutce -> .i mi pu pensi .ifebo lo mi bruna cu cmalu mutce
I was thinking about this: my brother is very short.
Taking this idea to the extreme, we can conceive of a somewhat silly higher-order predicate -- call it {brodrfV} for now -- whose x1 is an arbitrary sumti and whose x2 is a nullary predicate supplied with than fV having the value of the x1. We can define {brodrfV} with the following statement.
.i ko'a brodrfV lo du'u broda <=> broda fV ko'a
We can derive some obvious results from this statement.
.i lo brodrfV be lo du'u fV ko'a broda === ko'a
.i fV ko'a broda === .i fi'o brodrfV ko'a broda
This gives us a way to pick out sumti from du'u-abstractions, an otherwise arduous task for the fancylojban programmer/speaker.
Furthermore, this gives us a way to interpret {se fV}. Since {fV === fi'o brodrfV}, we have {se fV === fi'o se brodrfV}.
For instance
.i lo mi bruna cu cmalu mutce se fe lo du'u mi pu pensi -> mi pu pensi lo du'u lo mi bruna cu cmalu mutce
I've basically hijacked FA to recreate bridi relative clauses.
I'm sure there're plenty of holes in this idea since I cooked it up in just a few minutes. Feel free to come up with weird cases and we can examine them.
Do I want this to be a feature of standard Lojban? Not necessarily. Do I think it's a cool idea? Sure. I hope you do too :)