Yes, I have been discussing this very notion with Guha quite a bit, and even though we have a certain disagreement on certain details, we agreed on the overall framework. I completely agree with you that the wiki of functions will likely become a major puzzle piece on this issue.
Roughly, my assumption is that noun phrases and references by description are more or less the same thing for certain purposes, and that some of the constructors we will be creating for Abstract Wikipedia will be able to serve a double function of providing not only building blocks for the natural language generation part, but also can be mapped to a formal model that can propagate identity, and that has the additional advantage that it can be rendered in different natural languages (that's stronger than what Guha is suggesting, but should cover his use cases as well).
Or, put differently, we will be able to capture identity by description because descriptions such as "Andrew McCallum's right foot" are expressible in our system, and can be handed around and reasoned over, either within the wiki of functions or within a wider framework, and there will be a sufficiently shared understanding of the content of these descriptions. grounded in the same shared understanding we will need to achieve for the Abstract Wikipedia goal anyway.
So, yes, I totally agree with your observation, and I think that will be a rather neat additional result of our work.