In [5], Hjorth proved that for every countable ordinal $alpha$, there exists a complete $mathcal{L}_{omega_1,omega}$-sentence $phi_alpha$ that has models of all cardinalities less than or equal to $aleph_alpha$, but no models of cardinality $aleph_{alpha+1}$. Unfortunately, his solution does not yield a single $mathcal{L}_{omega_1,omega}$-sentence $phi_alpha$, but a set of $mathcal{L}_{omega_1,omega}$-sentences, one of which is guaranteed to work. It was conjectured in [9] that it is independent of the axioms of ZFC which of these sentences has the desired property. In the present paper, we prove that this conjecture is true. More specifically, we isolate a diagonalization principle for functions from $omega_1$ to $omega_1$ which is a consequence of the Bounded Proper Forcing Axiom (BPFA) and then we use this principle to prove that Hjorths solution to characterizing $aleph_2$ in models of BPFA is different than in models of CH. In addition, we show that large cardinals are not needed to obtain this independence result by proving that our diagonalization principle can be forced over models of CH.