 |
Laboratory
Links
|
|
|
|
James T. Townsend Lab
Department of Psychology
Student Research
Leslie Blaha
Configural Learning
Configural learning is the process of developing perceptual representations and mechanisms in which complex sets of elements are identified as unitary whole objects. We have developed the first complete human information processing definition of configural learning: the development of a facilitatory parallel architecture exhibiting super capacity efficiency in tasks requiring exhaustive feature processing. Empirically, we have investigated perceptual unitization as one mechanism of configural learning, and results characterize unitization by a shift from extreme limited to extreme super capacity processing, consistent with the proposed definition. We are developing a Hebbian-style dynamic systems model to capture the learning data. The model builds on Townsend and Wenger’s (2004) linear accumulator parallel interactive processing model.
Face Adaptation Aftereffects
Adapting to a face can alter the perception of the faces immediately following. For example, adaptation to a female face will make an androgynous face appear more male, and adaptation to a male face will make the same androgynous face appear more female. Aftereffects have been identified along the dimensions of gender, race, emotion, identity and face feature distortion.
The Townsend Lab is currently applying our Systems Factorial Technology (SFT) and our General Recognition Theory (GRT) methodologies to model face adaptation aftereffects. In particular, GRT will enable us to systematically map changes in the underlying perceptual representation space before and after adaptation, to see if the aftereffects are related to perceptual independence or separability of face dimensions, or even decisional interactions in the aftereffect judgments.
Nicholas Altieri
Project Title: Audiovisual Speech Integration
Primary Funding Agency: National Institute of Health Speech Training Grant (No. DC-00012)
Personal Webpage: mypage.iu.edu/~naltieri
The primary focus of my research program carried out in David Pisoni’s and James Townsend’s laboratories at Indiana University, involves investigating how the “brain” integrates auditory and visual speech inputs. I developed an interest in examining, at a theoretical level, how the multimodal components of the signal are “integrated” in real time. I have employed mathematical and statistical tools in several projects to develop a keener understanding of how neural circuits in the brain combine unimodal sources of speech information to form a unified “gestalt” percept, as well as an ecological and useful one.
A considerable degree of my dissertation work consists of constructing a unified model-theoretic approach toward understanding audiovisual speech integration; an approach that can account for aspects of the system related to architecture and processing efficiency (aka workload capacity; see Townsend & Nozawa, 1995). Architecture refers to how inputs are operated upon during the recognition process. Two broadly defined types of audiovisual integration under consideration include: parallel architecture, in which auditory and visual information are processed in separate channels, and separate decisions are made on each channel; and coactive architecture, in which auditory and visual information are coalesced into a unified code prior to the decision stage. The issue of processing efficiency or workload capacity is concerned with whether the addition of visual (or auditory) information to a unimodal speech signal produces more efficient processing in the time domain.
Answering architecture and capacity-related questions requires the implementation of statistical methodology such as the Double Factorial Paradigm (DFP: Townsend & Nozawa, 1995; Townsend & Wenger, 2004 for details). The DFP is an experimental paradigm in which the saliency of the auditory and visual signals (Sternberg, 1969), and “presence versus absence” (audiovisual stimuli vs. auditory-only and visual-only) are manipulated. The dependent variable is typically reaction time rather than accuracy. Data analysis involves estimating empirical survivor functions from the different factorial conditions and computing the “survivor interaction contrast”. The shape of the survivor interaction contrast is informative with regard to architecture since parallel and coactive models predict different shapes for the contrast. Data analysis also involves computing the Capacity Coefficient, C(t), (Townsend & Nowawa, 1995). This measure assesses efficiency by calculating the integrated hazard function ratio between the audiovisual condition, and the sum of the integrated hazard functions of auditory-only and visual-only conditions. A higher ratio implies a greater audiovisual redundancy gain.
Research Program
My current research program consists of applying the DFP in different experimental settings and using different stimulus materials. Architecture and capacity have been assessed in word discrimination tasks, experiments involving different word-set sizes and signal to noise ratios, as well as tasks with stimulus sets containing McGurk stimuli (i.e., auditory “b” and visual “g”). Overall, the results from these studies have shown that: a) audiovisual speech processing occurs in parallel, and b) that integration is inefficient, measured by C(t), unless the auditory signal-to-noise ratio is low.
Haiyuan Yang
My research interests are in the areas of configural information processing and human decision. I am currently working on two projects: One is generalizing Townsend’s Systems Factorial Technology (SFT) to make it include both the information of reaction time and accuracy when predicting the architecture of information processing (in collaboration with Joe Houpt). The other project is exploring the dependency between the processing of categorization and decision- making. One can think of dependency in this case as a form of cognitive configurality among internal processes or judgments (in collaboration with Noah Silbert).
Devin Burns
I am interested in applying the intricate mathematics of differential geometry and dynamic systems in the study of human cognition. My current projects include work on configural processing through the example of face recognition. I am also interested in the ecological and embodied view points of cognition.
Joe Houpt
I am a fourth year graduate student pursuing a joint PhD between Cognitive Science and Psychology. In my academic career at Indiana University so far I have worked on three lines of research. The first is on applying Systems Factorial Technology (SFT) to visual word perception. The second is on developing statistical tools for SFT. The third is focused on adapting SFT for use with clinical populations.
In regards to visual word perception, I have worked to measure the efficiency of letter perception in word contexts, pronounceable non-word contexts, and unpronounceable non-word contexts using response time based measures. The increase in letter perception efficiency due to either word or pronounceable non-word context has been well established in the accuracy domain, known as the 'word superiority effect' and 'pseudoword superiority effect' respectively. Despite that fact, when the stimuli are as well controlled as in the accuracy based tasks, there has been little evidence for these effects using response times. I developed a task that allowed us to employ the workload capacity coefficient, a general response time based measure of efficiency developed by my advisor, in this domain. Using this procedure we found evidence for a response time based word superiority effect and pseudoword superiority effect, as well as inefficient processing of letters in unpronounceable non-words.
My work on statistical tools for Systems Factorial Technology, the measurement tools developed by Dr. Townsend, has been to develop frequentist tests of the survivor interaction contrast (SIC). The SIC is a contrast between functions describing the response time distributions under related conditions in a cognitive task. Using the right within subject manipulations, the SIC can be used to determine many important properties of the cognitive system. Although mathematically well founded, the SIC lacked an apparatus for statistical testing. I developed the necessary theory and set of statistical tests for the SIC. I am currently involved in a collaborative project with Dr. Andrew Heathcote and Dr. Ami Eidels at University of Newcastle to develop Bayesian tests for use with the SIC. Our intent is to first replicate the frequentist tests I have already developed within the Bayesian framework, and then generalize these tests to group level analysis using hierarchical Bayesian methods.
I have completed one project translating the tools of Systems Factorial Technology for use in the clinical domain and begun another. In the first project I collaborated with Dr. Shannon Johnson at Dalhousie University and others in the Townsend lab to test existing hypotheses about disruptions in global processing in autistic children. Although we found clear distinctions between matched controls and autistic participants, we did not find clear evidence of deficits in global processing in this task. I have also begun to work with Michael Enders, a graduate student in Dr. Peter Finn's lab on applying SFT to study working memory deficits and the relationship between those deficits and a variety of clinical disorders. Thus far, we have developed a task that we have shown works well for using SFT to probe the working memory constructs in which we are most interested, using pilot studies of normal individuals.
|
