The aim of the LACUNAE Greek project is to apply artificial intelligence to support the process of restoring gaps in archaic and classical Greek epigraphic texts written in different dialects and alphabets, i.e. documents whose small number and peculiar linguistic features pose specific methodological problems. Given the characteristics of the selected documents, the project aims to develop a tool capable of taking into account literary sources in addition to epigraphic ones in order to fill in the gaps. Our starting point for this is, first and foremost, the models that represent the state-of-the-art tools used today, i.e. Ithaca and Ancient Greek BERT. In our paper, we present the preliminary results of the work carried out on archaic inscriptions in Greek dialects, while also discussing the methodological criticisms of the work and the strengths and weaknesses of the approaches used. Our discussion will focus on evaluating the performance of the models across different dialects at different levels of granularity and on testing new examples not yet seen by them. Additionally, given the scarcity of archaic Greek inscriptions, we investigate the potential of employing large language models—both general-purpose and instruction-tuned—to leverage their broad knowledge. Finally, we will share the lessons learned and discuss potential improvements for the critical aspects of the models. The code and datasets developed within the LACUNAE project are available in the project repository: https://github.com/AI4CH-UniUD/The_LACUNAE_Project-Performing_AI_on_Archaic_Greek_Epigraphy.
The LACUNAE project: evaluating NLP models for the restoration of archaic Greek inscriptions
Zunino, Maddalena L;Mignosa, Valentina;Serra, Giuseppe
2026
Abstract
The aim of the LACUNAE Greek project is to apply artificial intelligence to support the process of restoring gaps in archaic and classical Greek epigraphic texts written in different dialects and alphabets, i.e. documents whose small number and peculiar linguistic features pose specific methodological problems. Given the characteristics of the selected documents, the project aims to develop a tool capable of taking into account literary sources in addition to epigraphic ones in order to fill in the gaps. Our starting point for this is, first and foremost, the models that represent the state-of-the-art tools used today, i.e. Ithaca and Ancient Greek BERT. In our paper, we present the preliminary results of the work carried out on archaic inscriptions in Greek dialects, while also discussing the methodological criticisms of the work and the strengths and weaknesses of the approaches used. Our discussion will focus on evaluating the performance of the models across different dialects at different levels of granularity and on testing new examples not yet seen by them. Additionally, given the scarcity of archaic Greek inscriptions, we investigate the potential of employing large language models—both general-purpose and instruction-tuned—to leverage their broad knowledge. Finally, we will share the lessons learned and discuss potential improvements for the critical aspects of the models. The code and datasets developed within the LACUNAE project are available in the project repository: https://github.com/AI4CH-UniUD/The_LACUNAE_Project-Performing_AI_on_Archaic_Greek_Epigraphy.| File | Dimensione | Formato | |
|---|---|---|---|
|
20261012 LACUNAEproject (DSHadvance).pdf
non disponibili
Tipologia:
Versione dell'editore
Licenza:
Accesso chiuso-personale
Dimensione
1.71 MB
Formato
Adobe PDF
|
1.71 MB | Adobe PDF | Visualizza/Apri |
I documenti in ARCA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



