REVERINO: REgesta generation VERsus latIN summarizatiOn

Abstract

The contribution presents REVERINO, a dataset composed of pairs of regesta and full texts of medieval papal documents, designed to support the automatic generation of regesta through artificial intelligence techniques. After describing the process of building the corpus from printed collections through a pipeline of annotation, extraction, and post-processing, the study uses the dataset as a benchmark to evaluate the ability of language models to summarize Latin texts. The results show significant limitations in the automatic generation of regesta, while highlighting performance differences among models and suggesting possible strategies for improvement.