Evaluating Large Language Models for Character Identification in Italian Renaissance Epics A Case Study on Orlando Furioso

In this exploratory work, we evaluate the applicability of Large Language Models to character identification in the low-resource narrative domain of Italian Renaissance epic poetry, using Ludovico Ariosto’s Orlando Furioso as a case study. To mitigate contextual fragmentation, our pipeline integrates recursive narrative memory generation and character extraction. Experimental results are promising (F1 score of 0.758 with Qwen-2.5-72B), but qualitative error analysis reveals that performance remains constrained, in particular, by cataphoric resolution failures where character identities are disclosed non-linearly. We conclude by reflecting on new research directions in light of these findings.