In Kaplan's (currently unpublished) "The Meaning of 'Oops' and 'Ouch'," he discusses the meaning (surprise!) of expressive terms like "goodbye" and "ouch." One of the conclusions he comes to is that the metalanguage for a language L containing expressive terms does not need to be translational. Like the metalanguage for a language containing indexicals, the metalanguage M for L does not need to contain expressive terms to translate the expressives of L. He thinks we can give the semantic content of expressives in L by using purely descriptive terms in M. He brings this out by saying that M doesn't need to contain honorificsc to explain the meaning of honorifics. We can't use expressives in M, but we can describe what they do. There is a surprising amount of resonance here with what Brandom says in his Locke lectures. Brandom says that you can have a vocabulary V that allows you to say what you are doing in a certain practice P, even though the practice of deploying V cannot do what is done in P. To rephrase it, you can have a metalanguage that can describe the use in the language without containing some of the described phenomena, i.e. expressives. Brandom points out the same point in the case of indexicals in the appendix to his second lecture. He brings it out by contrasting what he calls the Kaplan-Stalnaker semantics with the Perry-Anscombe pragmatics. I'll have to go back through the two papers to see what other points of contact there are.