The CLIMB project, led by Alessandro Lenci, will receive funding for five years to develop a new generation of cognitively inspired language models
The University of Pisa has secured a major European research grant with the project CLIMB – Cognitively-Inspired Language Models for Investigating Human Language Acquisition and Processing, led by Professor Alessandro Lenci, Full Professor of Linguistics in the Department of Philology, Literature and Linguistics.
The project has been selected for funding by the European Research Council (ERC) under the prestigious ERC Advanced Grant scheme, which supports established researchers with outstanding scientific track records. CLIMB will receive approximately €2.5 million in funding over a period of five years.
The project aims to advance our understanding of the mechanisms underlying human language acquisition and processing by developing a new generation of Language Models that are more closely aligned with human cognition. While today’s Large Language Models achieve remarkable performance, they rely on vast amounts of training data and overlook key aspects of human learning, such as social interaction, multimodal experience, and cognitively grounded learning biases.
“CLIMB seeks to move beyond the assumption that increasingly larger datasets and models are sufficient to achieve human-like language abilities,” said Alessandro Lenci. “Our goal is to develop Language Models that learn more like humans do, integrating insights from linguistics, cognitive science, and developmental psychology. This will not only help us better understand how language is acquired and processed, but also contribute to the development of more efficient, transparent, and sustainable AI systems.”
Over the course of the project, the research team will develop Language Models for multiple languages trained on multimodal datasets that combine text, speech, and visual information, organized according to developmental stages. These models will incorporate theory-driven constraints inspired by linguistic and cognitive research and will be evaluated using innovative benchmarks enriched with human behavioural and biometric data.
CLIMB will also generate a wide range of open scientific resources, including datasets, benchmarks, language models, and analytical tools that will be made publicly available following Open Science principles. The project’s outcomes are expected to advance both fundamental research on language and cognition and the development of more accessible and sustainable language technologies.