Gestures are an integral part of human communication. Researchers have long identified different kinds of gestures, such as iconic gestures (e.g. swinging one's arms up and down to indicate running), beat gestures (e.g. tapping one's hands for emphasis), and deitic gestures (e.g. pointing to a table to indicate location). There is substantial behavioral literature on the production and comprehension of different kinds of gestures, but little is known about how gestures are processed by the brain. This is thus the subject of one of my current fMRI studies.

Not much is known about the changes in the brain that accompany the development of language abilities in children. As children learn to become more proficient readers and more sophisticated users of language, we expect that different brain regions to become more efficient in working together. We are currently running a longitudinal fMRI study to better understand the connectivity changes in brain networks under different language tasks.

As has long been argued in the field of first language acquisition, the input to grammar learning is impoverished: the grammatical structures of utterances are not present in the input and explicit corrections are rare. Additionally, in the case of pro-drop languages such as Mandarin Chinese, verb arguments are often omitted in the utterances. That all typically-developing children somehow manage to acquire language given such impoverished input has led to postulations that language is innate. In recent years, however, researchers have begun to pay more attention to the abilities that children bring to the task of language learning.

I took the view that grammars consist of conventionalized mappings between form and meaning, and modeled how a child could learn grammar. My computational model demonstrates how the problem of impoverished input can be alleviated by bootstrapping from context. By tracking situational and discourse events with a dynamically updated model of context, my model is able to partially infer meaning structures of utterances, which in turn provides leverage for hypothesizing new grammatical structures. Using Embodied Construction Grammar, a unification-grammar based construction grammar formalism, my model learns early grammatical constructions from situated language input.

Language is used in context, thus a system for language understanding must take contextual information into account. As part of the Neural Theory of Language (NTL) project , I worked on an improved grammar representation and the supporting processing machinery to deal with omissible and optional grammatical constituents. This allows for patterns of argument omission in pro-drop languages to be succinctly represented in the grammar without losing the generalizations across the patterns. The backbone support is a probabilistic best-fit constructional analyzer, developed by my colleague John Bryant, along with a model of context that is dynamically updated through simulation.