> Generally speaking, predicates which are bound in phase are definite in
> reference, perfective in aspect, and are old information in discourse
> Predicates which are unbound are indefinite, imperfective, and new
> information.

Are these categories (definite-cum-perfective-cum-oldinfo, indef-cum-imperf-
cum-newinfo) your own concoction? If so, then it proves the value of
conlanging as an academic tool for conducting interesting thought-

> The problem is that these three properties don't always
> co-occur; aspect in particular can be easily decoupled from the others
> Phase has turned out to be a delightfully squishy category that I'm
> still trying to think through

I am similarly finding with Livagian that even if you invent the grammar,
the pragmatics can only be discovered. Once you factor hypothetical actual
thinking speakers into the language, the results can be surprising.
(E.g. I recently found quite unexpectedly that an argument that is not
explicitly anaphorically bound is, pragmatically, nigh-on obligatorily
construed as noncoreferential with a potential antecedent in environments
in which the argument could easily have been explicitly marked as bound
by the antecedent. In Livagian, this is.)