The research collaboration between the Cantonal University Library of the Canton of Fribourg (BCU), EPFL+ECAL Lab, and CVLab has led to the creation of an interactive device named ‘Èvokâ’. This experimental installation explores the potential of using the heritage collection of the BCU to foster a sense of collective identity and strengthen social cohesion among citizens. Èvokâ utilizes generative AI technology, similar to ChatGPT, to craft unique stories for each document. It does so by analyzing visual information extracted from the documents and integrating it with their metadata. While the installation has been successful in generating narrative engagement, it also highlights certain limitations when dealing with ancient and local documents. For instance, current AI models, including ChatGPT, are not trained to recognize region-specific artifacts such as the iconic Fribourg milk pot. Additionally, these AI models lack training in creating sentences that incorporate the linguistic nuances of native Fribourg Patois.
The project presented here aims to address these limitations by fine-tuning a Large Language Model (LLM) to the specific cultural and linguistic characteristics of the Canton de Fribourg. Leveraging the extensive digitalization and availability of Fribourg newspapers, the goal is to develop an AI model capable of generating narrative content using the unique vocabulary and idiomatic expressions specific to the Fribourg region.
The involved student will be part of the EPFL+ECAL Lab team conducting the design research project composed of engineers, designers, and psychologists. Ultimately, the created model will be integrated within the Èvokâ prototype.
Key Domains
- LLM
Prerequisites
- Native French speaker
- Some knowledge of machine learning, deep-learning, fine-tuning LLM.
- Proficiency in Python.