General pUrpose dIalogue-based DigitAl iNformation acCEss – GUIDANCE

This project takes place in the context of large language models (LLMs)
and conversational systems (e.g. ChatGPT, WebGPT), which have
experienced tremendous practical progress in the last few months. The
project Guidance aims to conduct research on General Purpose
Dialogue-assisted Digital Information Access, specifically how to enable
users to access digital information, with the goal of overcoming several
limitations of current LLMs:

1. LLMs were not designed with Information Access – whether at the
level of pre-training tasks or fine-tuning ones
2. LLMs have limited generalization abilities to new domains and/or
languages;
3. The veracity and truthfulness of the output are questionable.
4. Potentially state-of-the-art LLMs models are not open access and the
scientific methodology and proper evaluation are barely described in
the scientific literature.

From a community building perspective, Guidance project aims at
federating the Information Retrieval (IR) French Community project, by
bringing together experts of the field to advance the development of
Dialogue-based Information Access (DbIA) models leveraging LLMs.
Guidance is backed up by partners belonging to the ARIA association and
gathers 18 researchers from 6 IR and NLP-related groups within 4
research laboratories. The partners furthermore commit to producing
open-access annotated resources, both at a national and international
level. These resources will be used to evaluate and develop models for
DbIA, and will constitute a precious resource for releasing open access
DbIA systems.

From a research perspective, Guidance addresses four challenges
associated with this project:

1. How to design new LLMs or re-use LLMs for DbIA;
2. how to leverage retrieval-Enhanced Machine Learning (ReML)
techniques to improve the accuracy and efficiency of information
retrieval systems;
3. Adapt LLMs and develop new architectures (for DbIA models) to deal
with low resource and domain adaptation – with special attention
paid to the low/medium-resource languages (e.g. Occitan, French);
4. Design DbIA models that can ensure the veracity, explainability of
retrieved and synthesized information, while preserving the user’s
subjectivity