ThoughtFill

Successful cognitive systems are adaptive systems; their behavior is contigent on the state of the environment. But complex environments can change over time, and previously successful behaviors may become maladaptive. Thus, systems must learn, must perceive changes and modify their behavior accordingly. What's more, learning algorithms can themselves change over time, as systems learn how to learn.

I'm interested in understanding how the dynamics of learning unfold over multiple time scales: from the scale of months and years (development) to the scale of milliseconds (eye movements).

Language Acquisition

Children learning language face a number of difficult problems. First, they must parse individual words from continual speech.

Unlike written language, in which white space indicates individual words, spoken language pauses are not informative about word boundaries. Instead, language learners must use other information to segment speech . My collaborators and I study how the statistical properties of spoken language can bootstrap speech segmentation in adults and children.

Even when the individual words are segmented, they still must be mapped onto their correct referents in the world. As real world environments are filled with a host of objects, it can be quite difficult to determine the referent of a speaker's utterance. We are working to understand how children can solve the word-world mapping problem both in pedagogical settings and from interactions between speakers.