FRBR's treatment of a translation as creating a new FRBR expression of a FRBR work is indeed at variance with the use of the notion of work in copyright law (where a translation is the canonical example of a derivative work in 17 USC 101 and Berne convention).

There is a good case to be made that translations are composite works consisting of the original content together with the intellectual work of the translator; however there is a case to be made that this approach is not optimal for grouping or discovery.

The boundary of work and expression became blurred as the IFLA process ground on. It is probably better to unpack and formalize the concepts rigorously, and only then define any simplified models (which is what I think is meant by indecs abstractions being fuzzy and context sensitive).