Rise from your grave, O Rudolph

There is no such thing as a natural language, as in a natural system of semantics that is specifiable by satisfaction conditions and T-sentences. Even the sentences within our so-called idiolects are higgledy-piggledy and under constant revision due to the freedom of interpretation one can make of sentences in context (think of the Lewisian semantic contextualist insight), thus failing to meet the expectations of classical conceptual theories. If there is an idiosyncratic lexicon, it consists of conceptual atoms (i.e., whatever patterns we consider it useful to store in memory as primitives) that have large parameters of application with respect to the kinds of inferential roles they might play, and more importantly, the kinds of questions they might answer. These patterns are not tractably specifiable, and do not directly enter into sentences (thus lack compositionality), but they are all we’ve got. When it comes to the kind of systematic, inference-supporting sort of thing that is connected to determinate beliefs, there are only drafts of languages. When we draft languages, we specify (often arbitrarily) the boundaries of a concept by construing discrete conceptions. Importantly, when we consult our intuitions about meanings, we are exercising a kind of contextually inspired judgment-call that ignores the regularities behind our use of both words and sentences — this is why it is inappropriate to say there is an idiolect, because whatever is rigid is fleeting, and whatever can be sustained is innocent of our expectations.

This view, I think the correct view, holds the following mish-mash of positions:
a) Descriptive lexical atomism. Words can be identified with concepts, and concepts are the only feature of our idiolects that last long enough to support interpretations of linguistic activity across time. This lends itself to a conceptual role interpretation of foundational holism.
b) Inferential relations and compositionality only hold with respect to our conceptions, which are artificial and fleeting. An inferential role interpretation of foundational holism is unstable across time and only works in making sense of drafts of languages. However, it is internally stable to the nth degree, and so supports a kind of imaginary and artificial analytic-synthetic distinction (among other things).

All this should reek of Carnap. Rise from your grave, O Rudolph…