<p>Prior’s invented connective ‘tonk’ is sometimes taken to reveal a problem for certain inferentialist approaches to metasemantics: according to such approaches, the truth-theoretic features of our expressions are fully determined by the rules of inference we’re disposed to follow, but admitting the ‘tonk’ rules into a language would lead to intuitively absurd results. Inferentialists tend to insist that they can avoid these results: there are constraints on what sets of inference rules can legitimately be admitted into a language, the story goes, and the rules governing disruptive expressions like ‘tonk’ are defective and so inadmissible. I argue, though, that from an inferentialist perspective, there’s no genuine sense in which rules like the ‘tonk’ rules are defective; in fact, anyone who endorses the relevant sort of inferentialism turns out to be committed to Carnap’s principle of tolerance. I then explain why this, despite appearances, isn’t a problem for inferentialism.</p>