if
{ko'a poi broda} is a referring _expression_ in itself, one might expect
{ro ko'a poi broda} to mean
(1) {ro da poi me ko'a poi broda},
but instead we have a separate rule which makes it
(2) {ro da poi me ko'a gi'e broda},
and (1) and (2) agree only in case (iv).
They almost agree in case (v), and I'm tempted to amend case (v)
to have it give {ro ko'a poi broda} meaning (1) rather than meaning (2).
> PA broda noi brode
> PA da (to ri broda toi) poi ke'a broda
>
> Some of this depends on {ri}'s ability to repeat {da} and quantified
> terms. Writing {PA da (to da broda toi)} would be weird, as the {da}
> could just as well be a new variable.
Yeah, I don't think that's really legitimate. Pretending that {ri} can
pick up the {da} at all (and actually I believe it just skips over it to
the previous sumti), I'd say the bracketed phrase there has an unbound
("donkey") variable, and we should either consider it an error or
universally quantify it out to give
{PA broda noi brode} -> {PA da poi broda zi'e noi brode}
-> {to da brode toi PA da poi broda} *
-> {to ro da brode toi PA da poi broda}
or, perhaps, remembering the domain of the variable and only universally
quantifying over that, giving
{to ro da poi broda brode toi PA da poi broda} ;
but that's a bit of a pain in practice, since the clause giving the
domain could itself mention unbound variables, and you have to recurse.