the Person-Case Constraint as feature-structural-level crossover effects

Here's a new idea recently tweaked off of my dissertation work: that Person-Case Constraint phenomena are driven by crossover effects operating not at the phrasal level, but at the feature-structural level.

That is, in the model I developed back in 2006, Person contrasts are built up out of referential dependencies---the most salient of which is that non-SAP Person status (i.e. 3rd person) is in a relevant sense referentially dependent upon the establishment of SAP referents. There are a number of ways to conceptualize the reasons for this; one of the simpler ones is simply that 3rd person is defined negatively as the other, as the non-SAP---even as the reverse relation is NOT the case---and so is in that sense dependent upon SAP referent establishment to be defined at all.

With this basic idea, we can see that, essentially, there is a little SAP component implicit in 3rd person status. Such that if 3rd person is introduced above SAP elements within the relevant locality domain---e.g. the Goal-Theme structure of a ditransitive, among others---a crossover effect results, because you are essentially demanding the representation to interpretation of a SAP-derivative feature structure before establishing the SAP feature structure itself. In short, [3] scoping over [1|2] in the relevant locality domain is bad for the same reasons that a pronoun scoping over its referential source (i.e. its antecedent R-expression) in the relevant locality domain is also uninterpretable.

That's pretty much it for the idea. What I've been struggling with for the past four years or so are, among others, two questions:

-Why is the Goal-Theme construction and its ilk the locality domain for feature-structure-level crossover effects? All I can think here is that these are co-phasal, even as other possible relations involving, say, external arguments, engage enough extra structure to make possible PCC candidates no longer relevantly local. But I need more than that.

-How, how, how, to properly formalize this sense of referential derivativity? All along I've known that it's not exactly the same as referential dependency as it's traditionally discussed, but still has all the same basic properties, and indeed, with this new idea of equating it to crossover configurations, is even more so now.

And indeed, it's what I really like about this idea: a nice example of the kind of fractal self-similarity---in structure, and in interpretational constraints thereon/therefrom---at each iteration of representational structure-building that Boeckx 2008 talks about, and that I find so compelling.

So, any ideas?

[By the way, the general idea here was inspired by something anyone who works in PCC phenomena would be interested to know about: that the Algonquian split between "proximate" and "obviative" 3rd person feeds into this system: obviative 3rd person over proximate 3rd person gives rise to the same PCC effects as 3rd person in general does over SAP persons. It is realizing that the proximate/obviative contrast is just a further iteration within the 3rd person domain of the same "referential derivativity" structure-building that creates the SAP/non-SAP contrast that pushed me towards this whole model of Person feature contrasts in the first place.]