09/03 Eva-Maria Steiger - Abstract
09/10 John Kruschke - Abstract
09/17 Mary Peterson - Abstract
09/24 David Stringer, Department of Second Language Studies, Indiana University - Abstract
10/01 Dan Little - Abstract
10/15 Shohei Hidaka, Linda B. Smith - Abstract
10/22 Woojae Kim, PBS - Abstract
10/29 Chris Honey, PBS - Abstract
11/05 Tom Busey & Chen Yu - Abstract
11/12 Geoff Bingham, PBS & Qin "Arthur" Zhu - Division of Kinesiology and Health, University of Wyoming - Abstract
11/19 Andrew D. Wilson, University of Warwick - Abstract
12/03 Josh Goldberg, Indiana University - Abstract
12/10 N/A - Abstract

Abstract

9/3: Eva-Maria Steiger
TBA

TBA

9/10: John Kruschke
TBA

TBA

9/17: Mary Peterson
Context Affects Figure-Ground Perception: Global Effects on Local Competition

We have proposed that the perception of figures versus grounds results from inhibitory competition between candidate objects on opposite sides of an edge. Is the inhibitory competition at a single edge affected by competitions occurring at nearby edges; in other words, are context effects observed? I will report experiments demonstrating that context affects figure-ground perception but only when (1) the contending objects can be detected in parallel across the visual field, (2) one contender is strong than the other; and (2) the weaker contenders are homogeneously colored; the homogeneity of the stronger contenders is irrelevant. We propose that spreading suppression underlies the context effects and interpret these results as support for a competitive account of figure-ground perception.

9/24: David Stringer, Department of Second Language Studies, Indiana University
Lexical Concepts in Language Development: Empirical Evidence Against Atomism

In contrast to the core assumption in most approaches to lexical semantics, Fodor (1998, 2000, 2005) and Fodor and Lepore (1998, 2002, forthcoming) argue that lexical concepts are atomic rather than compositional, and call into question the validity of lexical semantic research. Although there have been several recently aired conceptual arguments against atomism (e.g. Bach, 2000; Keil and Wilson, 2000; Pinker, 2005, 2007), the prevailing wisdom seems to side with Pulman (2005) when he states with some regret that there is no empirical evidence for decomposition. In contrast, I argue that research on language acquisition provides robust empirical support for the componential approach, and discuss two experiments which show that children split the lexical atom as they acquire verbal and prepositional predicates. The first was a partial replication of an experiment by Gropen et al. (1989) with younger test subjects, in which syntactic behaviour was reliably generalized to nonce locative verbs on the basis of semantic components. The second involved elicited descriptions of motion events, in which verb classes defined by meaning components constrained complement selection throughout language development. Crucially, the semantic components hypothesized are not primitives invoked to compute definitions of the type attacked by Fodor and Lepore, but are those elements of meaning which have observable (and predictable) effects in syntax. It is argued that both first and second language learning provide ideal testing grounds for theories of conceptual semantics, as lexical representations are subject to developmental change.

10/1: Dan Little
Function Effects in Probabilistic Category Learning

Probabilistic category learning differs from traditional, deterministic category learning in that items are assigned to categories based on some underlying probability. One result of probabilistic assignment is that perfect performance is usually unobtainable, and participants typically respond by probability matching. That is, participants match their response proportions to the underlying objective probabilities for each category. If response proportion and the underlying objective probability are taken to represent continuous magnitudes, then probabilistic categorization can be seen as an analogue of function learning. Here we report a series of experiments which show that several effects which are prominent in function learning, such as the relative difficulty of learning different types of functions, extrapolation, and knowledge partitioning, also arise in probabilistic category learning. These findings suggest the possibility that function-based representations, rather than rule-based or exemplar-based representations, guide responding during category learning.

10/15: Shohei Hidaka, Linda B. Smith
A Geometrical Analysis of Categories and Kinds

Given a single instance of a novel category, two- and three year- old children systematically generalize its name to other novel things based on appropriate feature dimensions. We explain this in terms of a prediction of the probabilistic density (category likelihood) in feature space from a single novel instance. In principle, observing more instances from a particular probabilistic density, one can estimate the probabilistic density more accurately. In this sense, children's success in generalization from a single instance seems to go beyond the theoretical limit. We provide a theoretical account for the phenomenon. In our theory, these kind of kind specific generalizations, a fast mapping from a single instance to a whole category is due to the structure of the system of learned categories and a sort of optimization of the category organization.

10/22: Woojae Kim, PBS
Bayesian Analysis of Cocaine Abusers' Gambling Data Using the Expectancy Valence Learning Model

Bayesian data analysis has become popular in statistics as an alternative method to the classical, frequentist counterpart. Among many advantages of this approach, extension to hierarchical models and the Bayes factor method for model evaluation are useful tools for modeling. In the present study, these tools are applied to the analysis of the Expectancy Valence Learning (EVL) model (Busemeyer and Stout, 2002) with data from a clinical population. The use of the Bayes factor as a model selection method provides not only evidence for the EVL model against baseline models among both healthy and patient groups, but also grounds for discriminating the patient group from the healthy group in their EVL parameters. The use of a hierarchical model accounts for individual differences within the groups as well as makes more accurate parameter estimation possible.

10/29: Chris Honey, PBS
Anatomical and Functional Networks in the Human Cerebral Cortex

In the human cerebral cortex, the activity levels of neuronal populations are continuously fluctuating. When neuronal activity, as measured using functional magnetic resonance imaging (fMRI), is temporally coherent across two populations, those populations are said to be functionally connected. The patterns of functional connectivity across the brain are presumed to reflect its underlying structural (anatomical) architecture. Our research group has recently obtained measurements of resting state functional connection patterns (using fMRI) and structural connection patterns (using diffusion spectrum imaging (DSI) tractography) in the same individuals. In this talk I will describe some of our ongoing investigations into this structure-function relationship, and in particular will touch on the spatial inhomogeneity of the relationship and the functional effects of individual differences in anatomy. I will also describe some of our recent computational modelling efforts, and will outline what I see as the key remaining questions regarding anatomical and functional networks in the brain.

11/5: Tom Busey & Chen Yu
Machine learning approaches applied to eyetracking data acquired from latent print examiners.

In this talk we describe an eyetracking dataset acquired from fingerprint experts and novices. Our goal is to determine what features experts rely on when matching latent and inked fingerprints, and how they might approach this task in a different way than novices. Recent court challenges have argued that juries should be shown the physical evidence directly, and that latent print examiners do not add anything to the legal proceedings. Indeed, this is a difficult problem, since a latent print identification is basically a similarity judgment based on fundamentally incomplete evidence (no latent print can be compared against all fingers in the world). Nonetheless, we find some conditions that demonstrate differences between experts and novices, and we use machine learning approaches such as machine translation to determine which features experts use when making matches and how their strategies might differ from novices. Tom Busey will provide a brief introduction to our portable and open-source eyetracking system, along with a summary of basic statistics of expert and novice behavior, and then Chen Yu will describe the results of the machine translation approaches.

11/12: Geoff Bingham, PBS & Qin "Arthur" Zhu - Division of Kinesiology and Health, University of Wyoming
Hefting to Perceive the Affordance for Maximum Throwing Distance is a Smart Perceptual Mechanism

Bingham et al (1989) showed that people skilled in over arm throwing are able to choose optimal objects for throwing to a maximum distance. They choose the optimal weight for each graspable sized (spherical) object, a different weight for each size. Objects were chosen from a set of 8 weights in each of 6 different sizes (from 1" to 6" in diameter). Subsequently, they threw all 48 objects to maximum distances multiple times and the objects they had chosen were thrown the farthest. Bingham et al hypothesized that hefting acted as a smart mechanism to provide information about the objects in the context of throwing. Zhu and Bingham (2008) found evidence to undermine the smart mechanism hypothesis. We now return to this question in the context of learning to throwing. We investigated two alternative hypothesis about how one might learn to perceive the affordance property, namely, by function learning or by a smart mechanism. Adult participants learned to throw long distance over a month of practice with restricted sets of objects. Their ability to perceive the affordance was tested before and after learning. The results supported a smart mechanism hypothesis.

11/19: Andrew D. Wilson, University of Warwick
Learning a Novel Coordination

Rhythmic movement coordination is a classic perception-action task that has been studied and modeled extensively over the years. The key phenomena of the task are that there are only two intrinsically stable states (0 and 180 mean relative phase), 0 is more stable than 180, and other states can (but must) be learned. I'll first review some data from experiments that investigated exactly what changes when a person learns a novel coordination (90 mean relative phase). Several other experiments have looked at the way in which this acquisition varies with schedule of learning and sleep deprivation. In general, people quickly learn to detect a new information variable which allows for stable performance at 90 (revealed by perturbing trained performance), and there is some evidence that sleep, not time, is required for the improvement to reveal itself.
This informational aspect to learning has one interesting consequence - learning 90 generalizes in highly specific ways and improvements at 90 are effectively encapsulated there. This makes this task ideal for studying processes of motor learning, especially in clinical populations (such as people recovering from strokes). The key is that 90 degrees is genuinely novel and changes at 90 are due solely to learning, rather than simple recovery of function. I'll review some ongoing work that is assessing learning capacity in stroke patients (with and without chemical or electrical intervention), and how changes in self-efficacy affects the rate of learning.

12/3: Josh Goldberg, Indiana University
Infant Looking: Peering Underneath the Preference Score

In a preferential looking experiment, infants are shown repeated pairings of a stimulus with a variety of novel stimuli. Reduced looking at the repeated (familiar) stimulus, relative to the changing (novel) one is taken as an indication of learning about the familiar stimulus. This is typically reported as a preference score, the ratio of time spent looking at the novel to total looking time in a trial. What can we learn about that learning, though, by teasing out the individual fixations that play into the preference score? It turns out quite a lot.

12/10: N/A
No Cog Lunch