Lewis on Meaning and Revelation
In §7 of "Naming the Colours", David Lewis considers the view that colour terms can be analysed in terms of colour experiences which in turn are identified by "a simple, ineffable, unique essence that is instantly revealed to anyone who has that experience".
Then if it were also common knowledge that everyone in the community becomes acquainted with magenta early in life (and if the community were properly dismissive of sceptical doubts about inverted spectra, etc.), it would be common knowledge throughout the community that magenta is the colour that typically causes experiences with essence E.
Lewis goes on to reject this porposal because it contradicts (type-A) materialism. But he doesn't reject the general idea itself: "[The doctrine of revelation] is false for colour experiences. [Footnote:] Maybe revelation is true in some other cases -- as it might be for the part-whole relation."
Does he mean that we identify the part-whole relation by its ineffable essence? He can't really believe that, if only because he doesn't believe in the part-whole relation. That's because on his view relations are classes of ordered pairs; but since everything without exception is a part of itself, including proper classes, the part-whole relation would have to contain pairs containing proper classes, which is impossible. Lewis was aware of this, see "Tensing the Copula" for a similar argument against the existence of the membership relation.
Anyway, does he mean that, contrary to what I said here recently, the term "part" could get its meaning somehow directly, in a way inscrutable to physics? No. That would go against his materialism and his "a priori reductionism about everything" ("Debugged", p.292).
I think what he means is that perhaps the notion of parthood can't be further analysed except in other mereological terms. Parthood would then be conceptually fundamental, like maybe identity, negation, and instantiation (on the latter see again "Tensing the copula"). That's not too implausible. In particular, parthood is not implicitly defined by the axioms of extensional mereology. These axioms merely specify a boolean algebra without zero. So it is trivial to find models in which "parthood" relates things that are not at all parts of one another.
If "part" is not analysable, then it seems that descriptivism as a theory of meaning fails for this term: "part" does not have its meaning in virtue of what the relevant people believe about parthood. But descriptivism is not the complete theory of meaning. Obviously it presupposes that the meaning of 'old' terms used in the description are somehow given. The simplest answer is Global Descriptivism: the old terms themselves get their meaning in virtue of their place in total theory. But this still leaves the meanings completely undetermined, even if we add the (unreasonably strong) constraint that total theory must be interpretated so that it comes out true. Further constraints besides descriptive role and truth are needed.
Lewis' own view is a bit more complicated. On first sight, he seems to offer three quite different accounts of meaning and reference: 1) meanings can be specified descriptively a la Ramsey-Carnap-Lewis; 2) meanings are whatever the best theory about the linguistic conventions prevailling in the relevant community says they are; 3) mental content is determined by (largely bayesian) rules of interpretation. But these accounts are not really at odds with each other: If the theory used in the Ramsey-Carnap-Lewis account is 'common knowledge' in the community, (1) and (2) coincide. (If not, descriptivism can't give an account of public meaning, and hence can't conflict with (2).) Both (1) and (2) ultimately rely on (3). Linguistic conventions for instance prevail in virtue of certain beliefs and expectations in the community. On Lewis' view, (3), i.e. the interpretation of mental content, does not deliver a compositional semantics for mental states, and certainly not for words in public language. Instead, it directly assigns a (weighted) space of possibilities to the individual's total belief state (similarly for other states). Derivatively and very roughly, the individual believes a specific proposition iff the proposition is true at the possibilities in his belief space. Among these propositions are propositions concerning linguistic expressions. That's how (3) supports (2).
At any rate, the fact that a term can't be reductively analysed doesn't mean that it gets is meaning by magic. My beliefs about parthood are beliefs about parthood roughly in virtue of that being the best interpretation of my belief state. And the English word "part" has its meaning in virtue of a convention to use "part" in such a way that the best semantics interprets it as meaning parthood.