Take even a simple _expression_ like "girls' school" (or "girls school" or "girl school") and render it in a loglang, assuming for now it means "school for girls as students". The analytic approach would have this divide into two predicates [girl] and [school] and then join them sententially somehow. The obvious solution is that [school] is a predicate (or family of predicates) which has a place for type of students, in this case girls, so Sv1 [girl]v1 [school]x v1. But there is no guarantee that, for a given girls school, there are in fact girls it is for (it may not be open yet or have very restrictive requirements or...). So, the reference to girls is intensional, possibly to the property of being a girls, whatever that may be:
Tv1 [^girl^] v1 [school] x v1 and the
place of [school] is now "for things characterized by". Of course, [school] has other places as well, for subjects taught, grade range, and probably others which are crucial to being a school. So, we must either distinguish the various predicates with in the cluster [school] or we have to have a lot of vacuous quantifiers (or strictly unquantified variables) to show what is not functioning at the moment. Of course, we might leave [school] alone and have a predicate [for girls], giving just K[for girls]x [school]x. But then there is the issue of how this new predicate is to be related to [girls]. Is it, in fact, merely stressing one place of [girls] -- in which case what do we do with the normal main place? Or is it [for][girls] with some sentential connectives to get that right (and the old problem of there being the right sort of girls and all). Now, of course, we can do all these things and analytic
precision requires us to do so (up to a point, of course).
On the other hand, we can generally just leave the unanalyzed cluster [girl][school] (probably with an connective -- not sentential -- to keep the grammar simple and deal with pretty little girl school teachers and the like), hopefully unpacable on request.