Developing a new generation of artificial intelligence that learns to use language in the same way humans do from early childhood: this is the goal of CLIMB, a project coordinated by Alessandro Lenci, Full Professor of Linguistics in the Department of Philology, Literature and Linguistics at the University of Pisa. The proposal has just been awarded an ERC Advanced Grant, the prestigious funding scheme of the European Research Council reserved for leading researchers. CLIMB (Cognitively-Inspired Language Models for Investigating Human Language Acquisition and Processing) will receive approximately €2.5 million over five years.
At the centre of the research are language models, the systems that form the core of the best-known generative AI applications, such as ChatGPT and Claude. These computational systems can generate text, answer questions, translate content and interact with people in increasingly natural ways.
In recent years, these systems have achieved remarkable results, but they operate very differently from human beings. To learn a language, current models require training on enormous quantities of data, whereas children acquire language through much more limited experience based on social interaction, communication, observation of the world and cognitive abilities that guide learning. This is the challenge at the heart of the project.
“With CLIMB, we aim to move beyond the paradigm that building intelligent language systems simply requires increasing the amount of data and the size of models,” explains Alessandro Lenci. “Our goal is to develop systems that learn language in a way that more closely resembles human learning, integrating insights from linguistics, cognitive science and developmental psychology. This will allow us not only to better understand how human linguistic competence develops, but also to design artificial intelligence systems that are more efficient, transparent and sustainable.”
The project will develop language models for six languages: Italian, English, French, German, Norwegian and Chinese. Training will be based on data combining text, speech and visual information, organised according to the different stages of human language development.
The results will contribute both to research on the processes underlying human language and to the development of AI-based technologies — including virtual assistants, translation systems and language-processing tools — that are more accessible and sustainable.
“This result confirms the University of Pisa’s role as a leading centre for research in artificial intelligence, computational linguistics and cognitive science. The preparation of the CLIMB project benefited from the support of the Advanced@UNIPI 2025 programme, which proved fundamental in building a proposal capable of competing successfully at the European level,” concludes Lenci.
Alessandro Lenci is Full Professor of Linguistics at the University of Pisa, Director of the Department of Philology, Literature and Linguistics, and founder of the Computational Linguistics Laboratory (CoLing Lab). He is internationally recognised for his contributions to the fields of distributional semantics, language models and the intersection of linguistics, artificial intelligence and cognitive science.
