What specifically do you see as the fit between GPUs and array languages?
This isn't really a new idea: http://portal.acm.org/citation.cfm?id=579.357248&coll=GU...
...(and gpus / larrabee etc. aren't solely vector processors, but the idea is apparent).
Most of the bulk numeric actions in an array language map pretty nicely to the data-parallel approach you need to use to take advantage of a gpu or larrabee (if it ever shows up); in particular take a look through this:
http://www.ncsa.illinois.edu/~gshi/LRBni_cheatsheet.pdf
...and see how much more straightforward it'd be to take advantage of (compared to SSE and so on). Your interpreter has to be a little more sophisticated (work has to be kept in units of 512 bytes) but seems much more tractable than previously.
Since this isn't a new idea there's history to learn from; it was previously the case that you'd get a speedup from offloading work to the vector units but not really a cost-proportionate one. But now if you look at the performance differential between cpus and gpus and their relative costs it starts making sense again.
I think it'd be easier to untangle if you already had a working knowledge of the language but given the line-noisy look of J I can't blame you for trying the shortcut.
The interactive J labs are very good. You might also find the J for C programmers discussion of rank particularly helpful if you are looking at learning by understanding an implementation.
I figure the array languages will make a minor comeback soon what with gpus / larrabee / etc showing up on the horizon and what with more-versatile input devices showing up (making it easier to go back to funky symbols instead of line noise).