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

Cognitively-Inspired Language Models for Investigating Human Language Acquisition and Processing

Objectif

The CLIMB project aims to develop Language Models that are more cognitively grounded, with the goal of investigating the mechanisms underlying human language acquisition and processing. Large Language Models demonstrate near “superhuman” performance, but require equally “superhuman” amounts of training data and ignore key aspects of human design, such as spoken and visual input, social interaction, and cognitively grounded inductive biases. This misalignment limits their scientific relevance as models of language learning and their suitability as experimental testbeds for linguistic theories.

CLIMB introduces an integrated framework that draws on findings from linguistics, cognitive science, and developmental psychology. It will construct Language Models for multiple languages trained on ecologically valid, multimodal datasets structured by developmental stage, and with a theory-driven design including inductive biases informed by linguistic and cognitive research. These models will be tested on new challenging benchmarks enriched with behavioural and biometric data, and will serve as controlled experimental platforms for validating hypotheses about the structures of language, learning trajectories, and domain interactions, such as the interplay between syntactic and semantic processing. Interpretability analyses will track the emergence of the models’ representations and establish an explanatory framework to link them to linguistic theory.

CLIMB also contributes to artificial intelligence by demonstrating how a linguistically and cognitively motivated design can enhance efficiency, representational interpretability, and behavioural transparency. By focusing on smaller and accessible models, CLIMB will support the development of more sustainable, open language technologies. The project provides a rigorous framework for evaluating the linguistic validity of artificial learners, bridging the gap between computational modelling and human cognition.

Champ scientifique (EuroSciVoc)

Mots‑clés

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-2025-ADG

Voir tous les projets financés au titre de cet appel

Institution d’accueil

UNIVERSITA DI PISA
Contribution nette de l'UE
€ 2 476 572,00
Coût total
Aucune donnée

Bénéficiaires (1)