Horwich's minimalist theory of truth

Horwich's theory of truth is generated by all the unparadoxical instances of:

1. <s> is true if and only if s

together with the assumption that only propositions are true, where "<s>" is the proposition that s, for a sentence s. It is important that the schema (1) be defined purely syntactically. But it is not clear that this can be done. For consider this substitution instance of (1):

2. <This sentence is short> is true if and only if This sentence is short.

The problem, of course, is that the referent of "this sentence" in (2) is liable to be (2), rather than "This sentence is short", and hence we get the wrong truth value. (There is also a minor grammaticality worry because of the capital "T" in the middle of the sentence.) Or take this substitution instance of (1):

3. <A if and only if B> is true if and only if A if and only if B.

This doesn't say what it's intended to say. Fixing up is needed. But it is not clear that one can syntactically specify how the fixing up is to be done in all cases.