One of the concepts of linguistic theory is the meta-language. If a statement is made about a language, that statement (this concept suggests) is not itself in that language, however much it may appear to be so. It cannot be because if it were in that language, it could not be about it, it cannot be about something of which it is itself a part. There are problems here which we need not take up at present; our subject is the difficulty which ensues if we agree that a statement about a language is necessarily in a meta-language.
This difficulty begins to appear when we recognise that in saying “any statement about a language is necessarily in a meta-language” we have made a statement about a meta-language, namely the statement that a meta-language is what any statement about a language must be in. For that statement applies to all languages, including meta-languages. If any statement about any language must be meta in relation to that language this will be none the less so if the language in question is itself a meta-language. A statement about at meta-language must therefore be in a meta-meta-language. And that statement, “any statement about a meta-language must be in a meta-meta-language” must itself be in a meta-meta-meta-language. And so on. To accept the concept of meta-language is to become involved in an infinite regression. At least, so it appears, and so it is widely believed among the people interested in such things. Here we suggest that the difficulty is an artificial one arising from the use of rigid, static, non-dialectical thinking.
The argument presented above rests on the assumption that X is X and evermore shall be so; that it is not, and can never become, non-X. In this instance, that a statement made in a meta-language must always remain a statement made in a meta-language, that it can never become a statement made in direct language.
But a meta-language is, by definition, the language in which statements about another language are made. When it ceases to be that, it ceases to be a meta-language. So long as we make statements about the direct language those statements are in a meta-language. But when those statements themselves become that about which statements are made, then they cease to be statements about another language and become statements about which statements are made. As such, they are not in meta- but direct language.
Statements made about them are in meta-language, but as soon as those statements themselves have statements made about them then they, in turn, cease to be statements made about language and become statements about which statements are made, and as such they are not in, meta- but in direct language.
This view seems to be at least as much in harmony with the concept of meta-language as the view first presented above, and it does not lead into an infinite regression.
from Ideological Commentary 17, March 1985.