Any use of language makes (or remakes) meaning as the degree of nonconformity of that particular use of language with generic expectations. Meaning is first the conformation or alteration of expectations: I expect you to obey, you expect the object I mention to be where I said it would be, I expect you to laugh at my joke, etc. The temporality of meaning is anticipatory, and following Lacan, for whom expectation is central to the temporal logic of desire, we could say that meaning is how language executes desire, how it chains together expectations. The perpetuation of desire is the sole marker of successful linguistic communication. This is what it means to communicate and to understand.

One requirement for a sign to mean is that it is finite. What makes a sign finite, that is, not infinitely open to interpretation, is a convention. Because of the absoluteness of the arbitrary relation of a sign to its meaning, that relation appears necessary--since there is no reason a sign should have one particular meaning over another, there is a fortiori no reason it should have any other meaning than the one it has. Thus the arbitrariness of the sign secures the relation to its meaning, and the conventionality of meaning appears as something always already agreed upon.

The anxiety around the decohesion of language (prevalence of foreign speech, the rapid evolution of communication technologies) is based in the otherness of convention (that its establishment precedes one's existence) and consequently upon one's inability to control the mutations of convention. This anxiety is at bottom the fear of not having one's desire perpetuated, of not being understood.

The struggle over convention takes place in the register of significance, which is the partitioning of language into language and metalanguage. Of course, because metalanguage is simply language taken to be metalanguage, it too has a metalanguage, and so on ad infinitum. But that regress can never be realized since a metalanguage is particular to a single language (whose constructed limits it describes). There is no metalanguage of metalanguages (since what a metalanguage describes is always a language), so there is no place from which to speak the absurdity of the regress. The temporality of language (and the finitude of meaning) foreclose the possibility of saying it all and at once. The language that would register the infinite chains of metalinguistic regression would be simply another language. It is the impossibility of such a language that allows for the metalinguistic discourse on conventions and generic expectations, that is, for meaning--the particular, finite, conventional use of language--to have significance.