Laurel Perkins

Research

What we can learn depends on what we already know; a child who can’t count cannot learn arithmetic. Just as in other domains of learning, language acquisition is incremental. Children gradually grow their grammars over the course of development by building on prior knowledge in order to learn from their linguistic input, as they currently represent it. My research investigates how children at the earliest stages of syntax and semantics acquisition represent their input and how they learn from it. Using a combination of behavioral and computational methods, I ask how linguistic, cognitive, and conceptual development interact throughout this process.

Noise-tolerant syntax learning

At early stages of grammatical development, children’s representations of their linguistic input are immature, incomplete, and sometimes inaccurate. Acquiring any piece of grammatical knowledge therefore requires mechanisms for abstracting away from messiness in the data, as a child perceives it. One strand of my research investigates these mechanisms computationally, focusing on the case study of early syntax learning. We model learning as selection among restrictive hypotheses, embedded within a system that also produces “noise” relative to the phenomenon being acquired.

Acquiring non-local syntactic dependencies

Another strand of my research investigates how infants identify syntactic dependencies between non-local expressions, such as wh-dependencies, and how this acquisition process interacts with their acquisition of verb argument structure. These dependencies take various forms cross-linguistically, so learners must discover their form in the specific language that they are exposed to. I use behavioral methods to diagnose infants’ early dependency representations, and computational methods to model how they acquire them.

Relating scene and sentence percepts in verb learning

If syntactic categories are related in a principled way to conceptual categories, then learners might be able to leverage these correlations as an initial bootstrap into the target grammar. Another collection of projects studies how these correlations inform children’s early inferences about verb meaning. Specifically, I ask how children expect their representations of the argument structure of sentences to correspond to their representations of the participant structure of events.