Why do infants learn language so fast? A reverse engineering approach

Introduction

How do infants learn their first language(s)? The popular yet controversial 'statistical learning hypothesis' posits that they learn by gradually collecting statistics over their language inputs. This is strikingly similar to how current AI's Large Language Models (LLMs) learn and shows that simple statistical mechanisms may be sufficient to attain adult-like language competence.

Language Input Estimates

But does it? Estimates of language inputs to children show that by age 3, they have received 2 or 3 orders of magnitude less data than LLMs of similar performance. The gap grows exponentially larger with children's age. Worse, when models are fed with speech instead of text, they learn even slower.

Research Question

How are infants such efficient learners? This project tests the hypothesis that in addition to statistical learning, infants benefit from three mechanisms that accelerate their learning rate:

1. They are born with a vocal tract which helps them understand the link between abstract motor commands and speech sounds, and decode noisy speech inputs more efficiently.
2. They have an episodic memory enabling them to learn from unique events, instead of gradually learning from thousands of repetitions.
3. They start with an evolved learning architecture optimized for generalization from few and noisy inputs.

Methodology

Our approach is to build a computational model of the learner (an infant simulator), which when fed by realistic language input produces outcome measures comparable to children's (laboratory experiments, vocabulary estimates). This gives a quantitative estimate of the efficiency of each of the three mechanisms, as well as new testable predictions.

Language Focus

We start with English and French that have both accessible large annotated speech corpora and documented acquisition landmarks and focus on the first three years of life.

Community Building

We then help build similar resources across a larger set of languages by fostering a cross-disciplinary community that shares tools, data, and analysis methods.