Performative Arts constitute an ideal application domain for Generative AI technologies as for the complexity in the conceptual and intellectual background characterizing the nature of the field.
The goal of this project is to demonstrate that a GenAI model, trained on Performative Arts data can achieve more accurate results on typical domain specific tasks, such as the querying of historical repositories in natural language or the interactive dialogue with different types of users on artistical issues.
The project focuses on three main steps: ArTBase (Archive and Theatre Database), a national database for italian historical theatres and theatre archives; ArTLLaMA, an adaptation of LLaMA, a Large Language Model, to the field of performative arts to query in natural language ArTBase; and, finally, ArTChat (Archive and Theatre Chat). This last step will focus on the creation of an interactive portal: a conversational agent, expert in the field of history of theatre, that will be able to distinguish the different types of users and select information for them, according to their needing. ArTChat can achieve better performaces with respect to the existing generalistic models, because it will provide certified information, validated by expert in the field, extracting them from ArTBase and from a corpus of the domain, used to train the model.
The portal will also link the user to the Hyperstage platform for more information about the reconstruction of each spectacle, guaranteeing a network of exchange of knowledge and promoting both research and the enhancement of the Italian Cultural Heritage.
This training resource will be released on a dedicated GitHub page and provides a valuable resource for AI innovation in Cultural Heritage studies. The purpose is also to demonstrate that an interdisciplinary approach is crucial, and that AI and Humanities can favor each other. We hope also that it will advance the Gen AI contribution to Digital Heritage research.
Elisa Passone
PhD Student in Data Science
Pillar Culture, Arts and Society
University of Rome Tor Vergata
passone@ing.uniroma2.it