Learnability and representation
One of the major questions in cognitive science focuses on the nature and structure of our mental representations, and how that relates to learnability in those areas. As representations grow more structured or more complex, it may seem increasingly unlikely that the ability to use and manipulate them could be learned without strong prior biases of some sort; conversely, learning without such complex representations or strong biases would itself presume an extremely complicated learning mechanism. I’m interested in exploring both sides of this question, and find computational modelling to be a very useful methodology for doing so: it allows us to precisely manipulate both representation and learning mechanism, and see how these factors interact with the amount and type of data to produce different behaviours. Some of my work in this area (along with that of various collaborators) focuses on the acquisition of abstract syntactic principles [1 2 3], the no negative evidence problem [1], recursion, the structure and learnability of dynamic categories [1 2], and why our representations are sparse [1]. More recently, I am also investigating details about the structure of our semantic networks [1] and how assumptions about data drive what is learned [1].
Child and adult differences in language acquisition
Children and adults differ widely in their language learning abilities. Children take longer to acquire the rudiments of a language but ultimately attain native-level proficiency without explicit study, while most adults never reach that level despite years or decades of active effort. Why this discrepancy? There are many possibilities, and I investigate a number of them. One is that other cognitive differences, like speed of processing or memory, may drive linguistic differences; for instance, in some recent work I explore whether memory limitations drive linguistic reglarization [1] [2], and in other work I have studied the relationship between cognitive abilities and language acquisition in infants [1]. Another possibility is how differences in lower-level linguistic abilities like phonetic perception differ between individuals [1] or might drive higher-level differences in aspects of language like word or grammar learning [1]. I am currently running several projects that investigate both of these directions further.
Language evolution
Within the past decade, research into language evolution has exploded, thanks to the increasing relevance and sophistication of cognitive modelling. I am interested in exploring questions about how the structure of the environment [1] and the dynamics of communication [2] interact with our cognitive biases to shape the way that languages evolve, and the characteristics of the resulting languages.
Hierarchical learning
One of the most powerful abilities human learners have is to learn simultaneously on many levels: to learn not just specific information (e.g., that things called balls tend to be round) but also to form abstract inferences (e.g., that nouns with the same label often have similar shapes). Some of my work explores how this sort of learning is possible [1] and under what conditions adults [1 2] and children [1] can do it. Current studies are focused on establishing the bounds of people’s abilities, and how they depend on cognitive factors like memory or attention.
The role of computational modelling in cognitive science
As a computational modeller, I am very interested in abstract questions about what modelling — particularly Bayesian modelling — can tell us about the mind. How does it work [1]? What implications does it have for innateness [1]? What can it tell us about statistical learning or language acquisition, and why? How does it compare to other computational modeling frameworks [1]?
Other questions
I tend to be interested in a lot of things, as you may have noticed! Other interests, which may expand or contract in future, include what assumptions people make about labels when learning new categories or words [1 2], how people search hypothesis spaces [1 2], how humans learn to lie and recognise liars (or lack of expertise in general) [1 2], and how people learn about sequences in different domains [1]. This list will no doubt change in response to new research and new ideas.