Evaluating the factuality of verifiable claims in long-form text generation