This paper investigates the performance of two models on Cavalcanti’s Rhymes: a supervised neural model (Stanza) trained on the Italian-Old treebank (comprising Dante’s Divine Comedy), and an unsupervised generative Large Language Model (LLM) accessed via the ChatGPT API (o3 version). This study highlights the crucial role of textual edition in processing historical texts, illustrating this through examples from different editions. It also presents a manual error analysis of the models’ outputs, focusing on both the most frequent and the most linguistically nuanced errors.
Corbetta, C., Colombi Anna, E., Moretti, G., Passarotti, M. C., What Is Better for Syntactic Parsing? A Comparison between Supervised and Unsupervised Models on Dante and Cavalcanti, in Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025), (Cagliari, 24-26 September 2025), CEUR Workshop Proceedings, Cagliari 2025: 292-303 [https://hdl.handle.net/10807/331456]
What Is Better for Syntactic Parsing? A Comparison between Supervised and Unsupervised Models on Dante and Cavalcanti
Moretti, Giovanni;Passarotti, Marco Carlo
2025
Abstract
This paper investigates the performance of two models on Cavalcanti’s Rhymes: a supervised neural model (Stanza) trained on the Italian-Old treebank (comprising Dante’s Divine Comedy), and an unsupervised generative Large Language Model (LLM) accessed via the ChatGPT API (o3 version). This study highlights the crucial role of textual edition in processing historical texts, illustrating this through examples from different editions. It also presents a manual error analysis of the models’ outputs, focusing on both the most frequent and the most linguistically nuanced errors.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
