The key to the concept of a linguistic rule is its complex relation to pattern-governed linguistic behavior. The general concept of pattern governed behavior is a familiar one. Roughly it is the concept of behavior which exhibits a pattern, not because it is brought about by the intention that it exhibit this pattern, but because the propensity to emit behavior of the pattern has been selectively reinforced, and the propensity to emit behavior which does not conform to this pattern selectively extinguished.
“Meaning as a Functional Classification” (1974)
Sellars’ main point that following a rule does not require intentionality is very much his own (I guess it owes something to Peirce’s notions of conceptual acquisition), but this is a very lucid statement of what I believe Wittgenstein himself to be saying about following linguistic rules: i.e., that it is a genuinely evolutionary process in which various linguistic patterns thrive or die off, and it is the very act of their linguistic usages in a particular pattern that legislates their continued use in that pattern.
But I’ll also, somewhat grudgingly, admit that I see some Hegel in here too. As all legislative usage has the potential to be transgressive against some dominant propensity, perhaps I can draw the analogy to the very end of the Phenomenology and its two antagonists, Acting Consciousness and Judging Consciousness. AC transgresses, JC condemns. AC confesses, JC forgives, and thus in that reconciliation we reach Absolute Knowing. Okay, that was the quick version. But linguistic usage brushes up against two opposing walls that are somewhat analogous to AC and JC: behavioral dissuasion and behavioral reinforcement, respectively. It is all conditioning, but it is a process of reconciliation too, in the same way that evolution reconciles mutation with fitness.