What Tarski’s Theorem shows is that interpreted formal languages that are interesting (i.e., with enough expressive machinery to represent arithmetic or fragments thereof) cannot contain a predicate whose extension is the set of code numbers (e.g., 
) of sentences true in the interpretation. The extension of any proposed truth predicate in such a system escapes the definitional machinery of the system. Of course, the truth predicate for first-order arithmetic can be defined with appeal to a stronger system, like second-order arithmetic, in the case of the Peano Axioms, etc.
Hellman’s first example is the following. It is a corollary of Tarski’s theorem that a theory in the language, 
, of arithmetic (e.g., an axiom system 
containing Robinson Arithmetic (
)) with symbols for zero, successor, addition, and multiplication, when extended with a one place predicate, 
(read “true in arithmetic” such that for each closed sentence 
in a new axiom of the form 
(where 
is the numeral for a code number for the sentence ), the resulting theory 
contains no explicit definition of in .
Connecting this to our ongoing discussion of determination of truth and reference in special collections of models, suppose that 
is the class of standard 
-models of . Then we have:
- In -structures -truth determines
-truth.
- In -structures -reference determines -reference
Which means that once you have the arithmetical truths in the class , then so are the ‘true-in-arithmetic’ truths and the same goes for the reference of the vocabularies. To avoid collapsing to reductionism via Beth’s theorem, note that there is no first-order theory (like those under discussion) in a language with finitely many non-logical symbols has as it’s models just the models in in .
If you extend to 
containing all models of , then you do get reductionism, since determination of reference in amounts to implicit definability in -thus showing that there exist non-standard models of arithmetic.
This is a good example because it is clear, based on popular, well established results and firmly shows how determination of truth and reference in one core theory carry over to it’s extension, without thereby reducing the extension to the core.
In the next update I’ll discuss Hellman’s second definability example.
