Progetti di ricerca e di sviluppo

As part of the CHANGES project, the Treccani Institute has developed three industrial research and experimental development activities that explore innovative ways of enriched digitisation of encyclopaedic heritage through generative artificial intelligence.

Enriched digitisation of encyclopaedic heritage through generative AI

The Treccani Institute, a partner in the CHANGES project under the PNRR, has developed three innovative projects that apply generative artificial intelligence to Italy’s encyclopaedic heritage. These industrial research and experimental development activities aim to radically transform access to knowledge, making it more interactive, inclusive and available to an international audience.

The projects combine the most advanced artificial intelligence technologies – such as Large Language Models (LLM), Retrieval-Augmented Generation (RAG) and hybrid translation systems – with the tradition of scientific and cultural rigour that has always distinguished the Italian Encyclopaedia Institute.

The projects explore various applications of AI; from natural language dialogue with the encyclopaedic database to advanced synthesis and vocalisation features as well as the possibility of accessing content translated into English through innovative methodologies that combine human expertise and intelligent automation.

CONVERSATIONAL RESEARCH

Treccani SOFIA

A conversational search system that allows interaction in natural language with a thematic selection from the Italian Encyclopaedia. Users can ask questions and receive discursive answers based on encyclopaedic content, with references to the sources consulted. The system combines the authority and reliability of Treccani content with the ease of consultation typical of modern conversational interfaces, while ensuring the verification and reliability of the content.

NEW FEATURES

advanced utilization of encyclopedic knowledge

Experimentation with GenAI technologies on Treccani encyclopaedic content, applied to the first volume of the Grande Enciclopedia. The project involves the development of advanced features such as automatic summarisation of entries, translation into English, quiz generation, voice technologies (speech-to-text and text-to-speech), a Q&A system with RAG and the creation of a semantic graph to renew the encyclopaedic consultation experience, offering more immediate access methods in line with contemporary digital habits.

INTERNATIONALISATION

Editorial translation

Innovative methodology for translating a selection of the online encyclopaedia, combining professional translation, translation memories and artificial intelligence. The project uses a hybrid approach: human translation of a training corpus, creation of a Translation Memory and customised Machine Translation, and use of the Human-in-the-Loop system to ensure high-quality translations, making Italian cultural heritage accessible internationally.

Research activities

Evaluation and analysis of Large Language Models (LLMs) for the automatic translation of online encyclopaedia entries from Italian into English. Specifically, two versions of the LLM Lara, recently released by Translated, are being evaluated, as well as one of the mechanisms on which Translated’s translation service is based, called Human in the Loop (HITL).