
How does language learning emerge from everyday experience? Our new preprint, led by @HRaviv830 offers a mechanistic answer.

The latest news from Uri Hasson's cognitive neuroscience research group at Princeton University

How does language learning emerge from everyday experience? Our new preprint, led by @HRaviv830 offers a mechanistic answer.

An AI model fed only what one child heard, day-by-day for the first 3 years of its life, learned no words. But with nightly replay (a bit like sleep), it learned thousands. New findings in @ScienceMagazine from @Princeton researchers provide a clue to how kids learn language and how to train AI on less data. pni.princeton.edu/news/2026/…


Deep gratitude to all fifteen families and everyone who worked tirelessly for nearly half a decade to make this study possible. Thank you all!!

Clinical science may ultimately benefit from mechanistic models of individual development to support personalized approaches.

The implications span disciplines. Psychology and neuroscience can test developmental hypotheses against real-life experiences and behavior across children and years. AI can explore biologically and cognitively grounded approaches to understanding human intelligence.

Together, these findings highlight how individual experience and learning mechanisms interact to shape development.

Replay matters. Removing nightly replay dramatically impaired learning. Exposure alone wasn’t enough to consolidate what models encountered.

Two findings stood out: Input density matters. With only a fraction of daily recordings, models learned far fewer words. Rich experience supported learning both frequent and rare words.

All eight models learned, developing large lexicons of word forms on a timescale closely aligned with real-life development. The models captured individual differences in learning pace by varying just three cognitively interpretable learning parameters.

Each model learned only from the speech that child heard, day by day. There was no pre-built vocabulary. Daily exposure increased as children spent more time awake, while nightly replay revisited recent speech to approximate sleep-related memory consolidation.

Fifteen families generously opened their homes, allowing us to document their children's everyday lives across their first 1,000 days. Using speech recordings from eight children, we built one AI model per child.

What does it take to learn language from everyday experience? Our new study, published today in Science, offers a new way to answer this question.

Read full paper here: biorxiv.org/content/10.64898… with @MiriamHavin @MeshulamMeir @R_Tikochinski @HassonLab @GoldsteinYAriel @TAILIN52261306

We’ve spent years showing that people hearing the same story share neural responses in high-level cortex. Now: instrumental music—no words—does too, converging on the story listeners imagine. New in @NatureComms, led by @ItamarJalon with @LisaMargulis: nature.com/articles/s41467-0…

New perspective piece out in @NeuroCellPress with @HassonLab, @zaidzada_, and Adele Goldberg! We try to articulate some of our excitement about LLMs and discuss what kinds of insights they might provide into the neural computations supporting natural language in the human brain.