Here's a brief overview of what the research in my lab is
about. There are traditional questions that people
ask about language: what do we know when we know
language? How is this information acquired? How is
this information used in comprehending and
producing language? What are the brain structures
that support the acquisition and use of language?
Why do people acquire language and not other
species? How does language relate to other
cognitive capacities? What do language disorders
due to neuropathology tell us about the language
system? Questions such as these have dominated
linguistic and psycholinguistic research for many
years, ever since Chomsky formulated them in the
1960s.
Our research attempts to advance the understanding of these issues,
while challenging many broadly-held assumptions about them. For
example, since Chomsky's early work, knowledge of language has been
equated with knowing a grammar. Many consequences followed from this
initial assumption. For example, if the child's problem is to converge
on the grammar of a language, then the problem does seem intractable
unless there are innate constraints on the possible forms of
grammar. What if we abandon the assumption that knowledge of language
is represented as a grammar in favor of, say, neural networks, a more
recently developed way of thinking about knowledge representation,
learning, and processing? Do the same conclusions about the innateness
of linguistic knowledge follow? The answer is: not at all.
Our goal, then, has been to articulate an alternative framework for
thinking about the classic questions listed above. This is not easy:
traditional grammarians have about a 40 year lead on us, and only a
few linguists actually think the alternative approach will
succeed. However, it's a very interesting moment in the study of
language. For many years the study of language was dominating by
theoretical linguistics, particularly syntax. More recently, there
have been important insights coming from outside of traditional
grammatical theory: from computational modeling, from studies of the
brain bases of learning and neurodevelopment, from renewed interest in
the statistical properties of language (which were ignored for many
years following Chomsky's famous observations about the statistical
triviality of sentences such as "Colorless green ideas sleep
furiously.").
Chomsky and his followers (e.g., Pinker) have always had their
critics. However, there was never an alternative theory that could
explain basic facts, such as how children acquire language under the
conditions that they do (of course it's questionable whether the
standard theory did either, but let's leave that issue aside). I
think for the first time we have the major components of such a theory
in hand. And they suggest the remarkable possibility that the
standard conclusions about the nature of language and how it is
acquired are just dead wrong. This would be an incredible turn of
events, a major development in the intellectual history of the study
of language.
That's why it's an interesting moment to be studying language.
My own research focuses on various aspects of language including
reading, which is a particular use of language,
how reading skill is acquired by children, and
forms of dyslexia that occur developmentally or in
adults as a consequence of neuropathology.
This research involves both behavioral studies and
the development of large-scale computational
("neural network") models of normal and
disordered language.
Of course, reading is an interesting topic in its own right, and the
implications of this work concerning educational
policy are important. However, I am mainly
interesting in reading as a domain in which to
study general theoretical principles concerning
knowledge representation, learning, information
processing, and their brain bases. And so the
theoretical framework that was originally
developed in connection with reading is being
applied to many other aspects of language,
including phonology, morphology and lexical
semantics, and its implications concerning the
brain bases of language are beginning to be
studied using neuroimaging. At the moment I am very excited about using the neural network approach to look at language acquisition, where it's clear that statistical learning plays a huge role. Jenny Saffran, one of the world experts in this area, and I are beginning to collaborate on studies that will bring the computational framework together with evidence about how babies learn. The framework is also being applied to the classic question as to whether there is a critical period for language. Some of this research is in progress, but papers are beginning to appear (in the publications archive).
In summary the ultimate goal of the research is to understand the
acquisition and use of language and its brain
bases using computational models as the
theoretical interface between the two.
For details about some of the main current projects, click here.
