Assumption: Given a speech community, as the language behaviour evolves, the language spoken retains its identity.Thought Experiment: let $C$ be a speech community, all speaking English (as currently understood, in terms of phonology, syntax, semantics and pragmatics). Let $C$ evolve forwards in time, with small shifts in language behaviour until, at a later time, say 100 years later, all speakers in $C$ speak German (as currently understood, in terms of phonology, etc.). By the Assumption above, we conclude,
English = German.The moral of this reductio ad absurdum is, I think, that the Assumption above is not true. Languages should be individuated very finely (both temporally and modally). Differences---even very small ones---in lexicon, phonology, etc., must count as different languages.
Fine-Grained Language Individuation: $L_1 = L_2$ if and only if $L_1$ and $L_2$ have identical phonology, syntax, semantics and pragmatics.Then, the analysis of the Thought Experiment above is what language the speakers speak, or "cognize", shifts.