Entity Disambiguation and Linking in Catullan Poetry: Toward an Enriched Digital Edition and a Knowledge Graph
Bibliography Reference
Publication Abstract
This work forms part of a broader project aimed at studying love discourse through computational methods. In this paper, we focus on a corpus of Classical Latin poetry, namely the works of Gaius Valerius Catullus, whose semantic and historical density makes the identification and disambiguation of references to people, places, and groups especially relevant. Given the nature of poetic texts, such entities often appear in highly allusive contexts, with morphological variation, semantic ambiguity, and stylistic devices such as hyperbaton, which make their automatic processing particularly challenging. To address this problem, we comparatively evaluate two entity linking approaches over Wikidata: a modular pipeline based on named entity recognition, candidate generation, and semantic ranking, and a generative approach based on GPT-4.1, used both for mention detection and for final entity selection. The evaluation relies on a manually curated ground truth and distinguishes between named entity recognition and semantic linking. The results show complementary behaviors: the traditional pipeline offers greater traceability and broader candidate-space coverage, whereas the generative approach achieves higher precision in mention detection and better performance in final entity selection, albeit with a more conservative strategy. Overall, the paper provides a comparative evaluation of methods for Latin poetry and lays the groundwork for enriching an XML-TEI digital edition with auditable identifiers and for Tdesigning a literary knowledge graph.
BibTeX Source Entry
@inproceedings{entity-disambiguation-and-linking-in-catullan-poetry-toward-an-enriched-digital-edition-and-a-knowledge-graph,
year = {2026},
title = {Entity Disambiguation and Linking in Catullan Poetry: Toward an Enriched Digital Edition and a Knowledge Graph},
author = {Carlos Javier Nusch},
abstract = {This work forms part of a broader project aimed at studying love discourse through computational methods. In this paper, we focus on a corpus of Classical Latin poetry, namely the works of Gaius Valerius Catullus, whose semantic and historical density makes the identification and disambiguation of references to people, places, and groups especially relevant. Given the nature of poetic texts, such entities often appear in highly allusive contexts, with morphological variation, semantic ambiguity, and stylistic devices such as hyperbaton, which make their automatic processing particularly challenging. To address this problem, we comparatively evaluate two entity linking approaches over Wikidata: a modular pipeline based on named entity recognition, candidate generation, and semantic ranking, and a generative approach based on GPT-4.1, used both for mention detection and for final entity selection. The evaluation relies on a manually curated ground truth and distinguishes between named entity recognition and semantic linking. The results show complementary behaviors: the traditional pipeline offers greater traceability and broader candidate-space coverage, whereas the generative approach achieves higher precision in mention detection and better performance in final entity selection, albeit with a more conservative strategy. Overall, the paper provides a comparative evaluation of methods for Latin poetry and lays the groundwork for enriching an XML-TEI digital edition with auditable identifiers and for Tdesigning a literary knowledge graph.},
booktitle = {Simposio Argentino de Inteligencia Artificial y Ciencia de Datos (ASAID 2026) - JAIIO 55 (Universidad Tecnológica Nacional - Facultad Regional La Plata, 10 al 13 de agosto de 2026)},
}
