This seems to be a callout to a point Alan Kay has been making for a while. RDF might be a good way to dip your toes in the water, but Alan Kay has gone further and called out languages such as "Linda" and related "tuple-space" research as directions in automatically figuring this stuff out.
For my own part I would also look into metaphor based research. By this I mean that looking to Biology for metaphors worked really well for OO programming so do the same thing here. Humans have been dealing with the mapping of their native language to that of another language for centuries now. I am sure that amongst anthropologists and linguists there is a pretty good body of research on how "first-contact" communication has been accomplished in the past and probably people have attempted to figure out principles to apply as well. There is probably a lot of fertile field here to till from a computer science perspective. NASA might have even sponsored some interesting research in this field, how did we design the voyager plaque, for instance.
APIs are currently the core fabric by which communication occurs within a computing system. As long as we use them, we end up with specialization of communication between computing systems.
This specialization, in my opinion, is the root cause problem in programming computing systems.
Bret Victor had this to say "The only way it (communication between systems) can scale, they (computers) have to figure out (dynamically), a common language".
Here I feel he is missing a key point. It is not a common language we are looking for, but a common architecture by which information is communicated between systems. Or, in this case, a non-architecture or anti-API by which communication takes place between systems.
