Extracting Volcanological Knowledge from Historical Texts: A Language-Technology Pipeline for Diachronic Geovisualization

This paper presents the first results of the CorVo project, a transdisciplinary project combining volcanology and computational linguistics to extract and structure volcanological knowledge from historical documents concerning Mount Vesuvius. We introduce the CorVo corpus, a multilingual diachronic corpus of 180 digitized texts (16th–20th centuries), selected to represent the main eruptive scenarios of the volcano. The digitization workflow integrates image pre-processing, OCR, and LLM-based post-correction to address challenges posed by degraded pages, historical typefaces, and orthographic variation. A domain-aware information extraction pipeline was developed to identify both standard toponyms and fine-grained spatial entities, which are typically overlooked by traditional NER systems. Extracted entities undergo human-in-the-loop validation and georeferencing through a dedicated annotation interface supporting multiple spatial geometries. The resulting dataset enables temporally normalized diachronic geovisualization of the textual-spatial footprint of Vesuvian eruptions across centuries.