The Way We Learn

Jury : 
Anne CHRISTOPHE Directeur de recherche Ecole normale supérieure  Directeur de thèse
Chen YU Full professor Indiana University Bloomington Rapporteur
Nuria SEBASTIAN-GALLES Full professor Universitat Pompeu Fabra Rapporteur
Judit GERVAIN Directeur de recherche Université Paris Descartes Examinateur
Luigi RIZZI Full professor Universita di Siena Examinateur
Isabelle DAUTRICHE Chargé de recherche Université Aix-Marseille Examinateur

To learn is to extract and distill pertinent information from a set of evidence. Yet, the evidence a learner has, or could have, at hand may seem insufficient for what she is aiming to acquire. The insufficiency of the evidence is often evoked in respect to language acquisition: the word meanings and grammar that individuals know appear to require more evidence than that to which they have access in their environments. While the brunt of research has focused on identifying overlooked sources of evidence and widening the evidence set, here we switch gears and probe what exactly it is that the evidence set needs to support. We propose that the evidence a learner has may be insufficient to provide her with the knowledge competent language users think they have; however, the very same evidence may be sufficient to provide a learner with the information needed for cognitive processing of language. Very broadly, there may be a gap between what we think we acquire and what we really acquire. 
In the introductory section of this dissertation, we begin by presenting evidence of a gap between what we feel words mean and how meanings are processed in the mind. We frame this gap in the broader context of how minds, with finite access to evidence, make sense the external world. Then, in the first chapter, we investigate how comparably little evidence can fuel acquisition of grammar in an ecologically valid setting. Our results reveal that just a handful of words can spur a virtuous cycle of grammar and vocabulary acquisition in infants. Next, in the second chapter, we examine whether the same set of evidence gives rise to productive knowledge or generalization (i.e., the capacity to use prior knowledge to interpret novel situations) across development, from infancy into childhood and through to adulthood. Our data show that infants and adults generalize a novel grammatical context to new words, but pre-school children do not. We interpret our results within the extant literature, pointing to the live possibility that what counts as knowledge may depend on where an individual places her ‘knowledge threshold’ rather than an immutable ideal. Finally, in the third chapter, we probe whether evidence that reflects a knowledge-state (e.g., explicitly hearing a direct translation: ‘Bamoule’ means ‘cat’) is inherently more informative, or merely more appealing, for a learner. Our results demonstrate that pre-packaged knowledge-state evidence boosts confidence, but has variable effects on performance. Across three chapters and ten experiments, we build up a set of fundamental features about what it is ‘to know’: (1) a little evidence can go a long way, (2) how much evidence is considered to be enough may depend on a modifiable threshold, and (3) the mind may crave certainty. We advance the conclusion that the set of evidence available to an individual can be sufficient to foster knowledge. It may just not be sufficient to foster the kind of knowledge we think we have. Therefore, to better understand the way we learn, we need to investigate what is ‘to know’ from the point of view of the learner.