 |
Research Activities
It is often claimed that language is a major component of what makes us human. Consistent with this notion, the study of language is central to Cognitive Science. The broad objective of our research program is to understand how people comprehend and produce language. To that end, our research program involves two related streams.
There are a lot of interesting and intriguing issues regarding how people come up with the meaning of object nouns such as "dog" and "chair" when they read or hear these words. For the past number of years, we have been working on some of these. For example, we have used human experiments and neural network simulations to understand the role of feature correlations (and causal feature relations) in the computation of word meaning (McRae, de Sa, & Seidenberg, 1997; McRae, Cree, Westmacott, & de Sa, 1999; McNorgan, Kotack, Meehan, & McRae, in press). We have also used human semantic priming experiments and neural network simulations to investigate the organization of semantic memory (McRae & Boisvert, 1998; Cree, McRae, & McNorgan, 1999). This research has been based to a large extent on a huge set of semantic feature production norms that we have collected over the years that presently includes 541 living and nonliving things (McRae, Cree, Seidenberg, & McNorgan, 2005). We have also used these norms to provide insight into the factors underlying intriguing phenomena concerning brain-injured patients who show category-specific semantic deficits; differential impairment in their knowledge of various types of things, such as animals vs. fruits and vegetables vs. nonliving things (Cree & McRae, 2003). In addition, in collaboration with John Kounios, Larry Barsalou, and Alex Martin, we are using ERP and fMRI neural imaging techniques to understand how all of these types of information are represented in people's brains. An theoretical and empirical overview of a bunch of this research is presented in McRae (2004).
Our second research stream focuses on interrelated issues dealing with sentence processing, thematic roles, verb meaning, and word-specific syntactic knowledge. A thematic role refers to the part played by an entity (as denoted by a noun) in an event (as denoted by a verb). For example, understanding "The customer cut her steak with her knife." requires using knowledge about cutting events, customers, steaks, knives, and their interrelationships to know that customer is the agent (the cutter), steak is the patient (the thing being cut), and knife is the instrument. We have focused on thematic roles because they are central to people's ability to understand sentences and to people's mental representation of verbs. We have advanced a theory of thematic roles as verb-specific concepts. McRae, Ferretti, and Amyote (1997) presented the theory and tested some of its predictions in conceptual and on-line sentence comprehension tasks. Ferretti, McRae, and Hatherell (2001) and McRae, Hare, Elman, and Ferretti (2005) used single-word and cross-modal sentence priming techniques to provide evidence that thematic roles are tied to event representations associated with verbs and other components of events. McRae, Spivey-Knowlton, and Tanenhaus (1998) and Elman, Hare, and McRae (2004) provide a detailed understanding of sentence processing by combining close scrutiny of the relevant information provided in a sentence with specifying a mechanism that integrates it.
In collaboration with Mary Hare and Jeff Elman, we have investigated how meaning and structure are co-dependent (Hare, McRae, & Elman, 2003, 2004). A number of researchers have shown that people use word-specific knowledge when understanding language, such as the probability that a specific verb is followed by various structures. Our research demonstrates that people are sensitive to the fact that this type of structural information depends on verb meaning. Currently, we are focusing on the role of event representations and event-based expectancy generation in language comprehension.
Finally, an important and innovative aspect of our research is combining theories and methods from a number of areas, namely: word recognition, concepts and categorization, and sentence processing (which necessarily involves linguistics), all in combination with neural network modeling, neuroimaging and patient studies. Combining areas produces novel and interesting theories and experiments, and we believe it is critical to a comprehensive account of language processing.
Grant support
Natural Sciences and Engineering Research Council of Canada
Representation and computation of word meaning
$27,000 per year for April 2002 - March 2006
National Institutes of Health (United States)
Expectancy generation in sentence processing
Total direct costs = $1,066,500
My portion = $55,000 per year for July 2001 - June 2006
In collaboration with Jeffrey L. Elman and Mary Hare
