Dr. Franklin Chang was interested in computers and
psychology while he was in high school. When he read about the field of cognitive
science in a college course catalog, it seemed like a good match. This interest
led him to study language modeling and sentence production as an undergraduate
at Carnegie Mellon University. For graduate work, he chose the University of
Illinois at Urbana-Champaign, where the psychology department has a group
focused on language production.
This group included Dr. Gary Dell and Dr. Kathryn Bock, who
were principal investigator and a co-principal investigator, respectively, of
the NSF/KDI-funded project The Role of Experience in Language
Processing. Franklin Chang worked on the project as a graduate student and
was involved in two different lines of research.
The first involved people and a technique called
structural priming. In this, speakers are given a choice between two
sentence structures. That choice is influenced by the presentation of a prime
sentence that is similar in structure to one of the two sentence structures.
For example, a picture of a man showing a woman a dress could be described as
"The man is showing a dress to the woman" or "The man is showing the woman the
dress." When speakers heard the prime sentence "The cheerleader saved a seat
for her boyfriend" beforehand, they were more likely to describe the picture
with the sentence "The man is showing a dress to the woman." That is, they were
primed to use the same structure to describe the picture that they had
experienced as a prime. This suggests that adult syntactic representations are
adapting to their recent experience with language.
The second line of research involved what's called a connectionist
modela computational system that is made up of neuron-like elements
that can learn. For this, the group that Franklin Chang worked with tried to
show that the same connectionist learning mechanism that was used to learn the
language in the first place could be used to model the kinds of changes that
they saw in adult structural priming. So the model represents the link between
an experience-based learning algorithm and adult sentencing processing.
Dr. Chang received his Ph.D. from the University of Illinois
at Urbana-Champaign and today is a postdoctoral staff scientist in the
Psychology Department at Max Planck Institute for Evolutionary Anthropology (in
Leipzig, Germany), where he does research in language acquisition. He continues
to examine the way that learning influences language processingthe
complex system through which we turn thought into grammatical strings of words,
we speak those words, and we understand the speech of others. He does this by
training connectionist models to produce sentences and then comparing their
behavior with that of human speakers.
Traditionally, learning and processing have been thought of
as different problems that require different theories. That premise has led to
complex language processing theories. But, according to Dr.
Chang, "Since we process language by using representations that we have learned
and we learn language by processing it, it seems that we need to have a way of
integrating these two aspects of language. And if we really get this
integration of processing and learning right, then we might find that we don't
need as much innate knowledge."
To conduct research in sentence production, Dr. Chang and
his colleagues typically give people the task of describing a picture or
repeating a sentence from memory. For experimental work with humans, the
researchers change one part of the task or the materials used in a way that is
important for their hypothesis. For computer modeling work, Dr. Chang uses a
software program that allows him to do connectionist simulations.
This work has several practical applications. The structural
priming technique could strengthen language skills in both native and foreign
languages. The modeling work may also help people with language disorders, such as aphasia.
According to Dr. Chang, "If we have a connectionist model of sentence
production, we can simulate different language disorders by lesioning sets of
neurons and then see if learning with different treatment programs can lead to
the recovery of language functions."
To learn more about Dr. Chang's work, visit his Web page at
Back to Top of