Skip to main content
Aller à la page d’accueil de la Commission européenne (s’ouvre dans une nouvelle fenêtre)
français fr
CORDIS - Résultats de la recherche de l’UE
CORDIS

Towards globally accessible language technology and its alignment to cultural contexts

Description du projet

Faire progresser la technologie du traitement multilingue du langage naturel

Les grands modèles linguistiques (LLM), tels que ChatGPT, largement adoptés dans divers secteurs, s’appuient sur la technologie du traitement du langage naturel (TLN) pour générer des réponses fluides à partir d’instructions en langage naturel. Cependant, les LLM et les autres applications basées sur le TLN ont été largement limités à quelques langues en raison des grandes quantités de données linguistiques nécessaires à leur formation. Le projet CulturAL, financé par le CER, vise à surmonter cet obstacle en développant une méthodologie innovante pour le transfert interlinguistique des LLM, permettant leur adaptation à un large éventail de langues, de dialectes et de contextes culturels.

Objectif

My research project focuses on natural language processing (NLP), an area of artificial intelligence concerned with automatic interpretation and generation of human language. NLP is well-known for its widely-used applications, such as machine translation (MT), text mining, question answering or dialogue systems. Much of this technology is now powered by large language models (LLMs), such as ChatGPT, which take a natural language instruction or question as input, and generate a fluent response. Due to their impressive performance in a range of tasks and their ease of use, in less than a year they received a wide adoption in many societal contexts (education, journalism, healthcare and others), and became the central paradigm in the field of NLP.

The development of LLMs, however, requires access to a vast amount of data and resources in a given language, as well as considerable computational infrastructure. As a result, these models are in practice limited to a handful of widely-spoken languages, leaving over 6,000 of the world’s languages and dialects without access to language technology. Furthermore, research on LLM alignment, which aims to ensure the safety of their use, has been almost exclusively directed toward the English-speaking world. Taken together, these problems lead to a major inequity in today’s language technology and artificial intelligence more broadly.

Taking a step towards a more inclusive and equitable language technology, this project will develop a novel methodology for cross-lingual transfer of LLMs to a wide-range of (low-resource, understudied) languages and dialects, and their alignment to diverse cultural contexts. The project will, therefore, advance multilingual NLP technology, extending its reach to populations currently underserved by NLP and making it safe for them to use.

Champ scientifique (EuroSciVoc)

Programme(s)

Thème(s)

Régime de financement

HORIZON-ERC - HORIZON ERC Grants

Voir tous les projets financés dans le cadre de ce programme de financement

Appel à propositions

(s’ouvre dans une nouvelle fenêtre) ERC-2024-COG

Voir tous les projets financés au titre de cet appel

Institution d’accueil

UNIVERSITEIT VAN AMSTERDAM
Contribution nette de l'UE
€ 1 998 926,00
Coût total
€ 1 998 926,25

Bénéficiaires (1)