CSJ Archive > Contents > Conference Proceedings: 2008 (30th Annual Conference; Washington, DC)

CogSci 2008

30th Annual Meeting of the Cognitive Science Society

July 23-26, 2008    Washington, DC, USA

Poster Session I -- Attention & Implicit Learning

(Thursday, July 24, 2008, 5:30-7:00 PM, Exhibit Hall)


Eye-Tracking Analysis of Cue Competition Effects Reveals Learned Inattention

Kelly M. Goedert   (goederke@shu.edu)
Seton Hall University

Brianna M. Eiter   (psybme@hofstra.edu)
Hofstra University

When simultaneously learning of two potential causes of a common outcome individuals underestimate the effectiveness of a moderately contingent cause in the presence of a highly contingent alternative (i.e., cue competition; Baker et al., 1993). Here we investigated whether cue competition effects observed in simultaneous blocking could be attributed to learned inattention (Kruschke, 2001). We recorded eye movements of participants while they acquired contingency information about two potential causes of an outcome in three contingency conditions designed to detect two forms of cue competition: discounting and conditionalization (Goedert, Spellman, & Harsch, 2005). Regardless of the form of cue competition, total fixation duration and the number of fixations for a weakly contingent cause decreased across learning blocks when that cause was learned about in the presence of a highly contingent alternative. These results are consistent with the learned inattention effects predicted by Kruschke's EXIT model of associative learning.


(Return to Top)

Assessing the Role of Estimation Strategy and Representation Format in Statistical Learning

Mirta Galesic   (galesic@mpib-berlin.mpg.de)
Max-Planck Institute for Human Development

Aron K. Barbey   (BarbeyA@ninds.nih.gov)
National Institute of Neurological Disorders and Stroke

Frank Krueger   (KrugerF@ninds.nih.gov)
National Institute of Neurological Disorders and Stroke

Jordan Grafman   (GrafmanJ@ninds.nih.gov)
National Institute of Neurological Disorders and Stroke

Gerd Gigerenzer   (gigerenzer@mpib-berlin.mpg.de)
Max-Planck Institute for Human Development

Encoding statistical regularities in the environment provides the basis for adaptive behavior, enabling predictive inferences that are essential for decision making in a dynamic world of risk and uncertainty. Accumulating evidence from the statistical learning literature demonstrates that frequency judgment reflects multiple cognitive strategies broadly classified into enumeration versus non-enumeration approaches (the encoding of event frequencies versus facts about the entire class of events; Brown 1995). This distinction is further supported by the statistical reasoning literature, which demonstrates alternative patterns of inference derived from statistical representations encoded on the basis of enumeration versus non-enumeration strategies (natural frequency versus relative frequency formats; Gigerenzer & Hoffrage, 1995). In this work, we investigate the role of estimation strategy and statistical format in frequency judgment within an experiment in which we manipulate frequency encoding context - promoting either enumeration or non-enumeration strategy ? as a function of the test format - natural frequency versus chances.


(Return to Top)

What to Infer From an Inference: Prior Knowledge and Task Demands Influence Contingency Learning

Daniel A. Sternberg   (sternberg@stanford.edu)
Stanford University

James L. Mcclelland   (jlm@psych.stanford.edu)
Stanford University

How do we make inferences about cue-outcome relationships? While the literature on human contingency learning has often centered on whether error-driven or normative models provide better accounts, the learning tasks themselves are often slow and deliberative, using stimuli from domains with familiar causal structures (e.g., symptoms and diseases). We designed two experiments to test the influences of the learning task and background knowledge in learning cue-outcome relationships. Experiment 1 used a fast-paced go/no-go task while Experiment 2 used an untimed predictive learning task with the same stimuli. In both studies, half of the subjects were given a causal cover story before beginning the experiment. RTs in the go/no-go task showed robust learning of the trained relationships, but failed to uncover cue competition effects in either instruction condition. The prediction task did evoke cue competition and hints at an interaction between background knowledge and predictive learning in producing these effects.


(Return to Top)

An Iterated Learning Model of the Emergence of Vowel Harmony

Frédéric Mailhot   (fmailhot@connect.carleton.ca)
Carleton University

It is well-known that speakers coarticulate vowel properties across an intervening consonant in producing VCV sequences in running speech (Hardcastle & Hewlett 1999). John Ohala (1994) has claimed that vowel harmony is a result of the phonologization of this vowel-to-vowel coarticulation. I will present an iterated learning" model (Kirby 1999) with the goal of investigating this claim. The aim is to clarify the circumstances under which vowel harmony may emerge in the grammars of individuals and in the shared languages of populations of speakers, as well as examining how transparent vowels can arise. In each generation of the model, the learners use a simple learning algorithm (clustering and analysis-by-synthesis) to infer structured representations of lexical items from the externalized output of the speakers."


(Return to Top)

Iterated Learning As a Model for the Spatial Distribution of Linguistic Hypotheses

Michael L. Kalish   (kalish@louisiana.edu)
University of Louisiana At Lafayette

The transmission of knowledge across generations of learners may account for both linguistic universals and the spatial distribution of specific languages. Iterated learning has been examined only in the case of a large population of well-mixed individuals reproducing without constraint by fitness (e.g., Griffiths & Kalish, 2007). For this paper I examined two processes, one Moran-like and the other deterministic, where selection was based on communicative fitness and mutation was based on Bayesian learning. The results of both processes differ from the infinite case, but neither produces spatially segregated stationary distributions of hypotheses. Thus, either (1) languages are indeed homogeneous (that is, they represent a single universal hypothesis), or the homogeneity of the model is not seen in nature because (2a) our hypothesis space is still in transition or (2b) more complex processes govern the space of learners. Distinguishing these three possibilities remains a target for this research.


(Return to Top)

Cognitive Models of Strategy Shifts in Interactive Behavior

Christian P. Janssen   (cjanssen@ai.rug.nl)
University of Groningen

Wayne D. Gray   (grayw@rpi.edu)
Rensselaer Polytechnic Institute

Michael J. Schoelles   (schoem@rpi.edu)
Rensselaer Polytechnic Institute

Many tasks can be solved using different strategies. Based on previous experiences with those strategies, humans develop preferences for specific strategies in specific contexts. For example, Gray, Sims, Fu and Schoelles (2006) studied a task in which participants showed a shift in strategy, to strategies that were mostly based either on interaction with the environment or on memory. However, ACT-R models were unable to show this effect. This was ascribed to the binary choice mechanism of ACT-R's utility learning mechanism (Gray, Schoelles & Sims, 2005). Our latest effort uses ACT-R's new utility learning mechanism, which is not based on binary choice, but on reinforcement learning (Anderson, 2007, How can the human mind occur in the physical universe?). Our study therefore serves as a case study for the success of this reinforcement learning mechanism, but mostly as a way to gain more insight in the cognitive processes involved in strategy selection.


(Return to Top)

Reinforcement Learning in Dynamic Environments

Todd M. Gureckis   (todd.gureckis@nyu.edu)
New York University

Bradley C. Love   (brad_love@mail.utexas.edu)
University of Texas

Successful investors seeking returns, animals foraging for food, and pilots controlling aircraft all must take into account how their current decisions will impact their future standing. However, one challenge facing decision makers is that what appear to be attractive options in the short-term do not always turn out best in the long run. We present recent work looking at how people learn effective control strategies in a dynamic task which places short- and long-term rewards in conflict. We find that perceptual cues that readily align with the underlying state of the task environment may help people overcome the impulsive appeal of short-term rewards and that such cues are differentially weighted in the decision process. Our work is motivated by current research in reinforcement learning which details how learners discount future rewards, the importance of state representations, and the role that exploration and exploitation play in effective learning.


(Return to Top)

Issues in Acquiring Interactive Routines

Bella Z. Veksler   (zafrib@rpi.edu)
Rensselaer Polytechnic Institute

Michael J. Schoelles   (schoem@rpi.edu)
Rensselaer Polytechnic Institute

Wayne D. Gray   (grayw@rpi.edu)
Rensselaer Polytechnic Institute

Prior work has shown that improvement with practice on a simple decision making task may not be due to changes in decision-making strategies, but may result from the adaptation of primarily perceptual-motor routines to the design of the task environment. This hypothesis was supported by a series of computational cognitive models that produced performance speedups due to an increased interleaving of cognitive, perceptual, and motor operations for subtask performance. Our attention has now turned to how these speedups may be acquired with practice. Using a reinforcement learning based approach, attempts to model the speedup have focused our attention on the span of effort over which reinforcement should be calculated.


(Return to Top)

Regulatory Fit Effects in a Dynamic Decision-Making Task

A. Ross Otto   (rotto@mail.utexas.edu)
University of Texas At Austin

Arthur B. Markman   (markman@psy.utexas.edu)
University of Texas At Austin

Bradley C. Love   (brad_love@mail.utexas.edu)
University of Texas At Austin

This work explores the influence of motivation on maximizing behavior in a dynamic choice task. We explore whether a fit between a situational regulatory focus and the reward structure of the task leads to greater cognitive flexibility than a mismatch between situational focus and reward structure. Participants made repeated choices in which the rewards from each response varied depending on other recent responses, placing short- and long-term rewards in conflict. We find that while participants in both fit and mismatch show equivalent proportions of optimal, long-term responses, fit and mismatch conditions elicit differential levels of sensitivity to changes in reward contingencies, as revealed by different patterns in response dynamics. Our analyses are also driven by current work in Reinforcement Learning that models how learners interpolate state signals, discount future rewards, and vary their degree of exploration and exploitation in effective learning across blocks.


(Return to Top)

Modelling Attentional Networks: The Modulation Effects and Simulation of Alzheimer's Disease

Fehmida Hussain   (f.hussain@sussex.ac.uk)
University of Sussex

Sharon Wood   (s.wood@sussex.ac.uk)
University of Sussex

This paper is an effort towards realization of psychologically plausible models of attentional networks to assess the modulation effects of attentional networks and to simulate attention related disorders like Alzheimer's disease. The work presented here computationally models the attentional networks task (ANT) which can be used to assess the efficiency and interactions of these disparate networks, collectively responsible for different functions related to attention mechanisms. The present research builds upon the model of ANT to show the modulation effects of one network on the other and suggests how the model can be used to simulate neglect conditions related to attention. The model is evaluated against data sets from experimental studies and the model's fit to data is assessed statistically. Building such models of attention will enable us to understand these networks better and empower us to make predictions about how attention will behave in various normal and abnormal conditions.


(Return to Top)

Multimodal Distance Perception is a Function of Multimodally Specified Effort

Eliah White   (whiteej@email.uc.edu)
University of Cincinati

Matthew Streit   (streitms@email.uc.edu)
University of Cincinnati

Kevin Shockley   (kevin.shockley@uc.edu)
University of Cincinnati

Michael A. Riley   (michael.riley@uc.edu)
University of Cincinnati

Prior research has demonstrated that distance perception is a function of the effort required to traverse a distance (e.g., Proffitt, Stefanucci, Banton, & Epstein [2003] demonstrated that wearing a heavy backpack increases perceived distance) and is influenced systematically by visual-motor recalibration (e.g., Mohler, Thompson, Creem-Regehr, Pick, Warren, Rieser, & Willemsen, 2004; Rieser, Pick, Ashmead, & Garing, 1995). We propose a quantitative model that accounts for these findings in terms of the relation between perceived distance and multimodally specified effort (MSE) the effort (i.e., metabolic cost [cal]) required to traverse an apparent [i.e., visually specified] distance (m). Actual effort and apparent walking speed were manipulated independently to achieve identical changes to MSE. If distance perception is specific to MSE then the respective influence of each modality (effort [proprioception] and apparent speed [vision]) on perceived distance should be equivalent. Reported distance increased with MSE for each modality with no differences between modalities.


(Return to Top)

Slow Saccade Lines in Eye-Track As Units of Semantic Cognition

Yukio Ohsawa   (ohsawa@sys.t.u-tokyo.ac.jp)
Dept. Systems Innovation, School of Engineering, the University of Tokyo

Yusuke Maeda   (yusa_maeda@yahoo.co.jp)
School of Engineering, the University of Tokyo

This study aims at finding a typical pattern in the eye movement when human obtains an interpretation of what a presented image means, i.e., the eye-track pattern in semantic cognition. We conducted experiments hiring 21 subjects looking at abstract artworks and commercial posters, and compared the eye-track lines with the statements about the subject's interpretation of the image. The results externalized a tendency that eyes draw lines in the velocity between ordinary saccade and slow drift, following links from/to objects in the image corresponding to fragments (sentences) of interpretations the subjects mentioned after looking at images. We assume these fragments reflect to the units of semantic cognition. Furthermore, when they reached an insight, i.e., when they obtained a new interpretation after feeling difficulty in understanding the image, the eye track changed into the middle-velocity line even though eyes had been drifting in the velocity of fast saccade.


(Return to Top)

Poster Session I -- Causal Cognition

(Thursday, July 24, 2008, 5:00-7:00 PM, Exhibit Hall)


Shared Features in Causal Induction

Adam Darlow   (Adam_Darlow@brown.edu)
Brown University

Shared perceptual features are strong cues for identifying causal relations among objects and events. These cues are central to causal induction in pre-scientific thinking (Rozin & Nemeroff, 1990), but theories of causal induction based on covariation (Cheng, 1997; Spirtes et al., 1993) implicitly assume people ignore them. These theories treat events as binary, either present or absent with no other features relevant to induction. This study investigated whether people use shared perceptual features when identifying the cause of an event. In an experiment where adult participants judged which blocks caused a target block to change color (Gopnik et al., 2004), they judged a block more likely to be a cause if its color matched the target block's resulting color. An extension of the Bayesian causal support model (Tenenbaum & Griffiths, 2001) demonstrates that this behavior is justified rationally when features are potentially propagated along causal links.


(Return to Top)

Are Self-Explanations Always Beneficial?

Jared B. Katz   (jbk2006@columbia.edu)
Teachers College Columbia University

Deanna Kuhn   (dk100@columbia.edu)
Teachers College Columbia University

We explore ways in which children become convinced of a theory's truth. Specifically, we examine the possibility that generating self-explanations during the learning process, generally thought to be beneficial, may in fact hinder causal learning in some situations. Thirty fourth-grade students were given eight sessions in which to investigate a multivariable database and identify the causal effects of five dichotomous variables. Students in the experimental condition also engaged in a related task where they generated self-explanations of the mechanisms underlying the causal effects they claimed to be present in the database. The control group did not generate these additional explanations. Non-explainers generated significantly more valid causal claims than self-explainers thereby supporting the hypothesis of superior mastery (of the system's correct causal structure) in the control condition. Discussion focuses on how self-explaining may hinder the development of a data-reading" skill used to evaluate information with respect to a theory's perceived truth. "


(Return to Top)

People Distinguish Causes That Occur to One or Multiple Entities

Benjamin Rottman   (benjamin.rottman@yale.edu)
Yale University

Woo-Kyoung Ahn   (woo-kyoung.ahn@yale.edu)
Yale University

The current models and theories of causal learning do not distinguish between causal events occurring within the same entity versus across different entities, but people may be sensitive to this distinction. For example, decreased effectiveness observed across multiple entities, is normally interpreted as lack of a generative cause. Yet, a single person's repeated intake of caffeine leading to decreased effects of caffeine on alertness may be interpreted as the person becoming tolerant to caffeine, and the reasoner may still ascribe generative causal efficacy in this situation. Similarly, when a person shows no reaction to an antidepressant, this lack of covariation may not necessarily be interpreted as lack of causality, since the effect of an antidepressant may need to build up. The current experiments demonstrate, using completely novel causal relations, that lay-people are indeed sensitive to single-case versus multiple-case learning situations and induce different causal relations.


(Return to Top)

This Stat Seems Bogus! Perspectives on Causality in Determining Veracity

Edward Munnich   (emunnich@usfca.edu)
University of San Francisco

Michael Ranney   (ranney@berkeley.edu)
University of California, Berkeley

Myles Crain   (mylestone@berkeley.edu)
University of California, Berkeley

Luke Rinne   (lrinne@uclink.berkeley.edu)
University of California, Berkeley

Luke Miratrix   (luke@vzvz.org)
University of California, Berkeley

Andrew Galpern   (AndrewGalpern@berkeley.edu)
University of California, Berkeley

 Past research indicates that a single, veridical statistic can catalyze individuals numerical understandings. But how do people assess potentially false numbers? Journalism students read a fictional colleague s numerical statements (e.g., 20% of U.S. power is nuclear ; with the number being high, low, or veridical). They first listed reasons why each number might be too high or low, and then indicated whether each number seemed too high, too low, or accurate. Considering underlying causal mechanisms (e.g., regulations limiting nuclear reactors) can help in discerning faulty numbers, but (a) participants causal reasons varied widely in quality, and (b) although the direction for which one listed more reasons reliably predicted the direction one indicated, this relationship did not obtain when we considered only causal reasons. We discuss expert-novice differences regarding the kinds of information applied to numerical reasoning, and outline ways in which awareness of causal factors may facilitate estimation.


(Return to Top)

Drawing Inferences from Causal Analogies

Julie Colhoun   (colhoun@northwestern.edu)
Northwestern University

Dedre Gentner   (gentner@northwestern.edu)
Northwestern University

Theories of analogical reasoning (e.g., Gentner, 1983) suggest that greater structural similarity between two analogs increases the likelihood of making candidate inferences from base to target (Lassaline, 1996). However, using a causal structure, where the base had a preventative cause and two generative causes all linked to an effect (E), Lee & Holyoak (2007) found a dissociation between similarity and the inference of E, casting doubt on the applicability of analogy theory to causal structures. This argument does not distinguish between the likelihood of bringing across candidate inferences and the plausibility of the effect inference. Our studies replicate L&H's data, but show that inference projection is not dissociated from similarity. We argue that candidate inferences (PROMOTES(G,E) and PREVENTS(P,E)) are projected from base to target, and E's plausibility is subsequently computed. That is, rather than running a causal model during mapping as L&H suggest, participants evaluate the inferences after the alignment-and-inference process.


(Return to Top)

Causal Explanations and Backward Counterfactuals

Morteza Dehghani   (morteza@northwestern.edu)
Northwestern University

Rumen Iliev   (r-iliev@northwestern.edu)
Northwestern University

Stefan Kaufmann   (kaufmann@northwestern.edu)
Northwestern University

We examine the relationship between counterfactual conditionals, causal explanation, and fact mutability. Causal Bayesian networks are widely used in modeling counterfactual reasoning about 'what would be if A were true'. Typically, such reasoning is modeled by intervention, suppressing all inferences about facts that are causally independent of A. Previous experimental results suggest that people's interpretation of conditionals 'if had been A, would have been C' does not uniformly conform to this rule: Speakers hold some of A's non-descendants constant, but revise others. How exactly this selection is made remains an open question. We examine two possible criteria for choosing between multiple direct parents of A: their comparative goodness as explanations for the hypothetical truth of A, and their comparative mutability in the sense of Norm Theory. Our experiments suggest that both play a significant role and that they are not independent. We propose a formal account for modeling this interaction.


(Return to Top)

Poster Session I -- Concepts & Categories

(Thursday, July 24, 2008, 5:30-7:00 PM, Exhibit Hall)


Rule Emergence From an Unsupervised, Dual-Network Connectionist Model of Category Learning

Rosemary A. Cowell   (r.a.cowell@kent.ac.uk)
Centre for Cognitive Neuroscience and Cognitive Systems, University of Kent, Canter

Robert M. French   (robert.french@u-bourgogne.fr)
Lead-Cnrs, U. of Burgundy,  Dijon, France

We present a dual-network, connectionist model of category learning, in which rules gradually emerge from knowledge of the statistical distribution of stimulus attributes. We define a rule for categorization as having emerged when the observer -- in this case, the network -- disregards a significant subset of an object s features and focuses only on a reduced subset of features in order to determine the category. The model's architecture is based on the interaction of a statistical-learning (Kohonen) network and a competitive-learning rule network. The statistical-learning network is implemented with a neurobiologically plausible Hebbian learning mechanism and forms category representations on the basis of perceptual similarity. The competitive-learning rule network extracts rules from the statistical-learning network, by exploiting noise in the system. The model can categorize stimuli that are misclassified by a standard Kohonen network that relies only on perceptual similarity. In addition, we present predictions for reaction times in categorization.


(Return to Top)

Multi-Sensory Statistical Learning: Evidence for Modality-General Mechanisms

Aaron D. Mitchel   (adm241@psu.edu)
The Pennsylvania State University

Daniel J. Weiss   (djw21@psu.edu)
The Pennsylvania State University

 Statistical learning has been shown to operate in a number of sensory modalities, including auditory (Saffran et al., 1999), visual (Fiser & Aslin, 2002), and motor (Hunt & Aslin, 2001) domains, suggesting that statistical learning is governed by a domain general mechanism. However, recent research suggests that statistical learning may be subject to modality constraints, and thus the mechanism(s) underlying this process may be governed by domain-specific subsystems (Conway & Christiansen, 2005; Turesson & Ghazanfar, in review). Here we test this claim by simultaneously presenting learners with auditory tone streams and visual shape streams, varying the amount of audio-visual synchrony across conditions. Both auditory and visual streams were equally learned, a pattern inconsistent with modality constraint theories. Further, neither streams were learned when the audio-visual synchrony was disrupted. This implies that amodal information is critical for tracking simultaneous streams across senses, consistent with modality general accounts of statistical learning.


(Return to Top)

Effects of Category Learning on Similarity of Line Stimuli Representing Social Groups

Janet Andrews   (andrewsj@vassar.edu)
Vassar College

Kenneth Livingston   (livingst@vassar.edu)
Vassar College

Daniel Bliss   (dabliss@vassar.edu)
Vassar College

Tatiana Vlahovic   (tavlahovic@vassar.edu)
Vassar College

There is considerable evidence that learning to classify objects into categories influences their perceived similarity. After category learning, objects classified together may appear more similar (?compression?) and/or objects classified differently may appear more different (?expansion?). However, these effects do not seem to occur when objects vary in only one dimension. A notable counterexample is the expansion effect reported by Tajfel and Wilkes in 1963 (which they called accentuation of category differences) using single lines varying only in length, a result that is widely cited in the social categorization/stereotyping literature. Their result has been found by many (including us) to be very difficult to replicate. New preliminary results using stimuli composed of one line vs. two lines, where the lines are given an interpretation involving the personality traits of hypothetical fraternity/sorority type social groups, suggest compression for the two-dimensional case but no effects for one dimension, contrary to Tajfel and Wilkes.


(Return to Top)

Mode Shifts During Category Learning

Kimery R. Levering   (kleveri1@binghamton.edu)
State University of New York: Binghamton

Kenneth J. Kurtz   (kkurtz@binghamton.edu)
State University of New York: Binghamton

In light of recent findings on the role of selective attention in category learning, we propose that learners progress systematically through three distinct modes during the course of classification learning. Learners begin in an exploration mode in which they determine the nature of the stimuli and test unidimensional hypotheses. For complex category structures, this is followed by an acquisition mode in which a generative or discriminative basis for categorization is developed. After perceived success, learners enter an application mode in which the achieved solution is applied with consistency and efficiency and attention is selectively distributed. To evaluate this, we developed a novel procedure that makes feature viewing choice and self-assessment into explicit components of the learning trial. A comparison of the points at which qualitative shifts occurred showed converging evidence of consistent mode progressions. Overall, perceived task mastery occurred first, followed by perfect accuracy, and then efficient feature selection.


(Return to Top)

Modeling Unsupervised Perceptual Category Learning

Brenden M. Lake   (brenden@stanford.edu)
Stanford University

Gautam K. Vallabha   (gautam.vallabha@gmail.com)
Stanford University

James L. Mcclelland   (mcclelland@stanford.edu)
Stanford University

During the learning of speech sounds and other perceptual categories, (1) category labels are not provided, (2) the number of categories is unknown, and (3) the stimuli are encountered incrementally. These three constraints provide a challenge for models, but they have been recently addressed in the Online Mixture Estimation (OME) model of unsupervised vowel category learning [Vallabha et al., PNAS, 2007, 104:13273-13278]. The model treats categories as Gaussian distributions, gradually estimating both the number and parameters of the categories. While the model has been shown to successfully learn vowel categories, it has not been evaluated as a model of the learning process. In this work, we applied the OME model to an unsupervised learning experiment with simple visual stimuli [Rosenthal et al., PNAS, 2001, 98:4265-4270]. The model accounts for why subjects place category centers near frequent stimuli and how this trend develops throughout learning.


(Return to Top)

Learned Attention to Analogical Matches as an Alternative to Re-Representation

Marc T. Tomlinson   (mtomlinson@love.psy.utexas.edu)
University of Texas At Austin

Bradley C. Love   (brad_love@mail.utexas.edu)
University of Texas At Austin

Building Relations through Instance Driven Gradient Error Shifting (BRIDGES) is an exemplar-based connectionist model of human category learning that can appreciate analogical relationships between stimuli and stored predicate representations of exemplars. BRIDGES learns to shift attention over the course of learning to reduce error and, in the process, alters its notion of similarity. A shift toward relational sources of similarity allows BRIDGES to display what appears to be an understanding of abstract domains, when in fact performance is driven by similarity-based structural alignment (i.e., analogy) to stored exemplars. In this contribution, we simulate Kotovsky and Gentner's (1996) development studies. Their studies demonstrate that children come to understand higher-order relations following experience with similar lower-order relations. Kotovsky and Gentner suggest that children come to re-represent domains in terms of higher-order relations. In contrast, BRIDGES suggests that shifts in attention are sufficient to explain performance.


(Return to Top)

Cooperative Categorization: Coordination of Reference and Categories in Learning a Joint Prediction Task

John Voiklis   (jv37@columbia.edu)
Teachers College, Columbia University

James E. Corter   (corter@exchange.tc.columbia.edu)
Teachers College, Columbia University

We investigated the interaction of structure and convention in the emergence of schemes for joint reference in the context of indirect category learning. Participants worked individually or in dyads to learn a set of functionally-defined categories, instantiated as supposed alien creatures. The perceptual structure of these categories was complex: one function could be predicted by a unidimensional rule but the other was defined by a family-resemblance substructure. In addition to the main function-prediction task, each learner worked individually to sort the exemplars (pre- and post-function prediction) and in an individual prediction test that yielded selective attention data. Dyadic learners predicted the functional features with significantly greater accuracy compared to individual learners. This dyadic advantage was even greater for predicting the simple rule-based function compared to the FR function. Also, the post-task sorts produced by dyadic learners correlated more closely to the true categories than did those of individual learners.


(Return to Top)

To Find Fault is Easy? The Role of Comparison in Learning a Geological Structure

Benjamin Jee   (b-jee@northwestern.edu)
Northwestern University

David H. Uttal   (duttal@northwestern.edu)
Northwestern University

Dedre Gentner   (gentner@northwestern.edu)
Northwestern University

Geoscience education involves learning to identify geological structures. We examined the role of comparison in learning to identify faults (rock fractures showing evidence of movement). Comparison involves structurally aligning two representations (Gentner & Markman, 1997). Learning novel structures may be facilitated by progressive alignment comparing similar scenes before more distant exemplars (Gentner, Loewenstein, & Hung, 2007). Highly similar items are easier to align structurally, thereby highlighting their common relational structure, and any alignable differences (readily comparable but differing features). In the present study, participants with varying levels of geology course experience were presented with 20 pairs of geological photos, and had to indicate which photo in each pair displayed a fault. We contrasted a similar-first condition, in which perceptually similar pairs preceded dissimilar pairs, and a dissimilar-first condition, which reversed this ordering. Consistent with progressive alignment, similar-first learners were more accurate, especially if they had some background geology experience.


(Return to Top)

Intuitive Conceptions of Innateness in Cognitive Science Undergraduates

Otto Lappi   (otto.lappi@helsinki.fi)
University of Helsinki

Innateness is a core concept in cognitive science, yet controversy and heated theoretical debate is rife wherever innateness is discussed. Several hundred years of philosophical argumentation and theoretical debate has failed to resolve even the issue of how best to define innateness for scientific purposes. But however vague or confused the concept of innateness might be for scientific purposes, students nevertheless enter cognitive science education with well established commonsense conceptions about what counts as innate and what does not. This poster reports on a pilot study that probed cognitive science students conceptions on innateness. In the present study we address the following questions: (a) How mutually consistent are commonsense intuitions on innateness across students? (b) Do the putative commonsense intuitions put forward by philosophers and scientists in theoretical argumentation actually hold up against data on commonsense conceptions.


(Return to Top)

Concept Creation Derived From Vacillation

Jun Nakamura   (jyulis@gmail.com)
Department of Technology Management for Innovation, University of Tokyo

Yukio Ohsawa   (ohsawa@sys.t.u-tokyo.ac.jp)
School of Engineering, University of Tokyo

We aim at clarifying the effect of vacillation, where categorization of words as an action provides with an implication for human s concept creation. We hypothesize that vacillation and creation are induced by an uncertain interpretation of presented information. We evaluated this hypothesis on the playing logs of a developed Web based tool to enjoy categorization of given 20 words to make at most 5 groups corresponding to players original concepts. We defined vacillation as player s suspending (decreasing the frequency of) word grouping/regrouping actions. As a result for 12 players we found (1) players concentrate on specific words with writing, reviewing, and changing concepts in their own interpretation of the words during vacillation time, (2) vacillation tends to be induced by words difficult to group, i.e., words left uncategorized, and (3) the reconstruction of groups for including the uncategorized words into some group accelerates the creation of new concepts.


(Return to Top)

Poster Session I -- Decision Making

(Thursday, July 24, 2008, 5:30-7:00 PM, Exhibit Hall)


Polarized Correlation-Based Beliefs: a Computational Genesis

Richard B. Anderson   (randers@bgnet.bgsu.edu)
Bowling Green State University

Michael E. Doherty   (mdoher2@bgnet.bgsu.edu)
Bowling Green State University

A series of simulation studies examined situations in which each of many cognitive agents extracts a small (n = 3), random sample of continuous x, y data from a large population in which the correlation between x and y is zero. On the assumption that each sample correlation represents an agent's belief about the value of the population correlation, the distribution of agents' beliefs was found to be bimodal, with clustering near +1 and -1 (despite a zero population correlation). Furthermore, the shape of the distribution depended on the number of data pairs per sample and on the measurement scale (e.g., the Pearson r versus the signed square of the Pearson r). Thus, the results suggest an abstract computational mechanism, rooted in the information environment, for producing polarized correlation-based beliefs. Behavioral research, in progress, assesses the operation of such a mechanism in human judgment.


(Return to Top)

Language Affects the Side Effect Effect

Stephen Flusberg   (sflus@stanford.edu)
Stanford University

Caitlin Fausey   (cmfausey@psych.stanford.edu)
Stanford University

Lera Boroditsky   (lera@psych.stanford.edu)
Stanford University

How do we decide whether an action is intentional and how responsible an agent is for the consequences of the action? This question becomes especially tricky when we reason about complex actions such as accidents and the side effects of intentional actions. Both the language that is used to describe an accident (Fausey, in prep) as well as the moral valence of the side effect of an intentional action, influence attributions of responsibility to the agent (Knobe, 2003). Here we examined whether the same language manipulations that influence attributions of responsibility for accidents also influence attributions for side effects of intentional actions. Using stimuli modified from Knobe (2003), we found that describing a side effect using agentive language like ?He harmed the environment? led to stronger attributions of intentionality and responsibility than using non-agentive language like ?The environment worsened?, but only when the side effect had a negative moral valence.


(Return to Top)

Ambiguity Preference

Liema Davidovitz   (liema@ruppin.ac.il)
Ruppin Academic Center

Yossi Yassour   (yassour@ruppin.ac.il)
Ruppin Academic Center

Ambiguity Preference~~Liema Davidovitz, Yossi Yassour~~Purpose~~Ambiguity refers to uncertain situations where the probability of the possible events is unknown. The purpose of this paper is to investigate under which conditions people prefer ambiguity and ignorance over clarity and knowledge. Ellsberg (1961), has shown that people, when faced with positive rewards, present aversion to ambiguity. Similar results were shown by Ganzach (2000). On the other hand, recent research shows that on some occasions people prefer ignorance on knowledge (Ehrich and Irwin, 2005, Botti and Iyengar, 2006). Design In an experimental framework the attitude towards ambiguity was measured using 4 scenarios (monetary, health, survival, vacation), all in positive and negative framing. Findings We found that people show aversion to ambiguity in the positive domain but preference to ambiguity in the negative domain.  These findings show similarity to Prospect Theory, namely, attitude towards ambiguity, like risk, depends on the positive or negative domain.


(Return to Top)

Inferring Population Correlations From Small Samples

Justin M. Gilkey   (jgilkey@bgsu.edu)
Bowling Green State University

Richard B. Anderson   (randers@bgsu.edu)
Bowling Green State University

Accurate inference of population correlations from small samples is difficult, in part because of the skewed distribution of Pearson s r. A series of simulations attempted to train decision makers to compensate for the sampling distribution s skew. Twenty-five simulated decision makers, implemented as neural networks, were presented with samples of 3 to 24 data-pairs, drawn from populations exhibiting a correlation (between two binary variables) ranging from .2 to .8. The decision makers were first trained to classify each sample as belonging to one of the four populations, and then were tested with novel samples. Classification performance was biased in a manner consistent with the sampling distribution s skew, but achieved above-chance accuracy, at each level of sample size, in inferring the correct population correlation. These results suggests, that through experience, decision makers extracted useful information about the population correlation (given their above-chance accuracy), but were unable to completely compensate for the distribution s skew.


(Return to Top)

Tri-Part Affect: Its Structure, Its Biological Basis and Its Role in Decision Making

Ken Gunnells   (Ken.Gunnells@ua.edu)
The University of Alabama

Over the years and particularly in more recent times there has been a growing acknowledgement of the role of affect in decision making. Theory and evidence is presented from a biofunctional perspective to give further understanding of the biological embodiment of affect and its role in decision making. Three biologically embodied sources of contribution come together in biofunctional activity; namely emotion, mood and temperament. Each of the three is uniquely characterizable in biofunctional, experiential, and psychological terms. Emotion, based on active brain functioning, is of-the-moment, episodic, relatively intense, and typically target specific. Mood, based on dynamic brain functioning, is of a longer period of time, is less intense and more diffuse, and typically is target nonspecific. Temperament, also biologically based, sets the baseline and range for emotion and mood. These three affective manifestations, functioning as a system deliver the human affective experience in decision making.


(Return to Top)

Strategy Selection At the Technological Interface

Matthew M. Walsh   (mmw187@andrew.cmu.edu)
Carnegie Mellon University

John Anderson   (ja@cmu.edu)
Carnegie Mellon University

We studied how people select between mental and technological strategies. To explore this topic, we created a computer-based paradigm where participants solved multiplication problems. On some trials, participants could solve the problem mentally or with a calculator built into the interface. On others, they were required to use a mental or a calculator solution. Besides gathering timing data and error rates, we recorded mouse cursor movement. Participants? motor behavior was revealing. Although people initiated movement to the calculator or answer box within 230 ms, they frequently redirected motions mid-execution. Thus, some trajectories were direct while others first approached the non-selected box. Movement initiation direction depended on problem difficulty and calculator responsiveness. Ultimate strategy selection also depended on these factors, but was further influenced by movement initiation direction. We conclude that strategy selection is iterative, as revealed by redirections in cursor movements. In addition, early motor behaviors bias ultimate selections.


(Return to Top)

Cognitive Representations, Acculturation, and Adolescent Risky Decision-Making.

Wanda D. Casillas   (wdc23@cornell.edu)
Cornell University

Valerie F. Reyna   (vr53@cornell.edu)
Cornell University

Britain Mills   (bam64@cornell.edu)
Cornell University

Steven Estrada   (sme27@cornell.edu)
Cornell University

We investigate the relationship between acculturation of adolescents and their engagement in sexual activity, where acculturation refers to a departure from traditional Hispanic values and into the majority American culture. Five hundred forty-two participants completed surveys about acculturation, immigration, and values related to risk. Hispanics, who were less sexually active, were also less acculturated. A planned comparison conducted to include mixed Hispanics showed that mixed Hispanics, of intermediary acculturation level, were also intermediately likely to be sexually active. Subsequent analyses highlighted underlying reasons for ethnic differences in sexual behaviors and revealed differences in mental representations of risk and associated values. Less acculturated adolescents endorsed values associated with family and relationships, as expected based on notions of traditional culture, which acted as a protective factor against adolescent sexual initiation; more acculturated adolescents endorsed a calculative approach to risk, which encouraged risk-taking, a finding consistent with fuzzy-trace theory.


(Return to Top)

Poster Session I -- Language & Speech

(Thursday, July 24, 2008, 5:30-7:00 PM, Exhibit Hall)


Escaping Snakes and Acquiring Cash:: Categorical Processing of Affective Words

Zachary Estes   (z.estes@warwick.ac.uk)
University of Warwick

James S. Adelman   (j.s.adelman@warwick.ac.uk)
University of Warwick

Sabrina Simmons   (s.g.simmons@warwick.ac.uk)
University of Warwick

We demonstrate that negative words (e.g., "snake," "coffin") tend to elicit slower responses than positive words (e.g., "cash," "heaven"). This effect of valence on response times was not mediated by arousal, nor was it attributable to word length, word frequency, orthographic neighborhood size, contextual diversity, or initial phoneme. The relation between valence and response times was categorical rather than linear; extremely negative words (e.g., "poison") and slightly negative words (e.g., "needle") elicited responses that were equally slow. Moreover, despite the fact that stimulus valence becomes less extreme across repetitions, response times to negative words and positive words did not converge across repetitions. Thus, affective words appear to be processed categorically. Such categorical processing of affective stimuli may be adaptive for minimizing the likelihood of catastrophic error: Essentially, the benefit of quickly averting an extremely dangerous stimulus outweighs the cost of overreacting to a mildly threatening stimulus. "


(Return to Top)

Biddies, Crones and Codgers: Adults?' Connotative Understanding of Gender-Specific Vocabulary

Wendelyn Shore   (shorewj@plu.edu)
Pacific Lutheran University

Marianne G. Taylor   (taylormg@plu.edu)
Pacific Lutheran University

What does it mean to have partial knowledge about a word? Shore and colleagues (e.g., Durso & Shore, 1991) reported that adults know some semantic constraints on a word?'s meaning even when they deny its existence in their language (unknown words). When they claim familiarity with the word but deny knowing its meaning (frontier words), they nevertheless demonstrate some knowledge about its definition. We found that adults correctly identify gender category membership for frontier and unknown words, although accuracy for partially known words was higher when those words referred to males rather than females. The current study investigates adults'? connotative understanding of gender-specific vocabulary at three levels of word knowledge. Results indicate that when the words?' meanings are known, words referring to males are rated more favorably on all scales (e.g., more powerful, more important) than are words referring to females. This difference is reduced when words are partially known.


(Return to Top)

Dynamic Lexical Processing Evident in Wiimote Trajectories

Nicholas Duran   (nduran@memphis.edu)
University of Memphis

Rick Dale   (radale@memphis.edu)
University of Memphis

Results of an experiment tracking motor output during semantic interpretations demonstrated that increasing exposure to one interpretation leads to a greater commitment for that interpretation, evident in the dynamics of arm movements. We used a Nintendo Wiimote to capture relatedness evaluations between homonyms and a preceding set of 1, 3, or 5 word pairs biasing the subordinate sense of the homonym. After each homonym, a target word was presented to activate the dominant interpretation (e.g., 5-word sequence about trees, followed by relating ?bark? to ?growl?). Wiimote movement trajectories revealed increased motion time for dominant interpretations after longer processing of subordinate word pairs. This entrenchment in subordinate interpretation also took longer to escape when the entrenchment was allowed to proceed for 5-word pairs, but not for the 1-word pair conditions. These data recommend a dynamical systems account of semantic processing, where a strongly established attractor basin takes longer to undo.


(Return to Top)

Bilingual Sentence Comprehension in Conditions of Perceptual and Attentional Stress

Eileen R. Cardillo   (eica@mail.med.upenn.edu)
University of Pennsylvania

Jennifer Aydelott   (j.aydelott@bbk.ac.uk)
Birkbeck College, University of London

Although the sentence comprehension abilities of highly proficient bilinguals are often indistinguishable from those of native speakers in quiet, they are significantly impaired in more difficult listening conditions. The nature of this disadvantage for non-native speakers was explored in four auditory sentence-priming experiments with Italian-English bilinguals. In all cases, participants made lexical decisions on the final words of semantically congruent, neutral, or incongruent sentences in their non-dominant language (English) while ignoring an occasionally present competing talker. Across experiments, competing speech varied in its spatial isolability (same or different ear as attended speech) and dominance (Italian/dominant or English/non-dominant). Results indicated that the ability to take advantage of semantic context was disrupted more when competing and attended speech were not spatially isolable. Further, competing speech in the same language as attended speech produced greater interference than competing speech in the dominant language.


(Return to Top)

"A Door" Or "Adore"? Word Segmentation in Semi-Spontaneous Speech"

Dahee Kim   (daheekim@ling.ohio-state.edu)
OSU

Anouschka Bergmann   (anouschka@ling.ohio-state.edu)
OSU

Christine Szostak   (szostak.1@osu.edu)
OSU

Mark A. Pitt   (pitt.2@osu.edu)
OSU

Comprehension of spoken language requires segmentation of continuous speech into discrete words. An understanding of the process of  segmentation can be gained from experiments using semi-spontaneous speech. Stimuli were sentences with an ambiguous string of phonemes  that could be segmented in two ways (e.g., a door vs. adore ).  Acoustic analyses of the productions of these sentences showed few local cues were present for disambiguation. Perception experiments were then  performed to determine what information is necessary for correct segmentation.  The results indicate that local acoustic cues at the word boundary are insufficient. Rather, information from the larger sentential context is essential, showing that word segmentation can be a heavily (perhaps solely) top-down process.


(Return to Top)

Towards a Unified Model of Word Pronunciation

T. Florian Jaeger   (fjaeger@bcs.rochester.edu)
University of Rochester

Celeste Kidd   (ckidd@bcs.rochester.edu)
University of Rochester

Speakers pronounce predictable words shorter and with less articulatory detail on average (e.g. Bell et al., 2003). Other studies have found that speakers also modulate the duration of a word, if the next words are not readily available for pronunciation to buy more time (e.g. Clark & Fox Tree, 2002). We present the first studies that directly compare and integrate these two principles (redundancy avoidance and strategic lengthening) into a unified model. We model the duration of 65,000 determiners from a time-aligned corpus of spontaneous American English speech. We present cases studies on individual determiner types (e.g. 15,0000 cases of a; 20,0000 cases of the) as well as combined a study on all determiner types (including controls for differences in phonological complexity). Our studies provide evidence that speakers' modulation of speech rate is due to three underlying causes: redundancy avoidance (predictable words are reduced), strategic lengthening before unavailable material, and joint retrieval of collocations.


(Return to Top)

Comparative Neuroimaging of Morphological Regularity

So-Hee Kim   (skim32@mail.utexas.edu)
The University of Texas At Austin

The aim of this study is to investigate whether second language (L2) speakers of English show neuronal responses similar to those exhibited by native (L1) speakers of English. Specifically, by using an event-related functional magnetic resonance imaging (fMRI) technique, this study compares the neural patterns of Korean speakers with those of native English speakers as they both process the morphologies of regular and irregular past tense forms in English. Because the Korean language does not have two different, relatively complex inflectional morphological systems as does English, the output from Korean speakers who also have a high command of English can help answer the question whether L2 speakers use the similar neural systems that L1 speakers use. Answers to that research question will contribute to our understanding of the mental representation system by which the brain processes cross-linguistic differences and similarities in morphology. \ \


(Return to Top)

Online Processing of Gender-Marked Articles by Spanish-Speaking Children, Adults, and Second Language Learners

Casey Lew-Williams   (casey@psych.stanford.edu)
Stanford University

Anne Fernald   (afernald@stanford.edu)
Stanford University

This research explored whether Spanish-speaking children, adults, and adult L2-learners would use gender-marked articles (la, el) to more rapidly identify nouns. In Experiment 1, participants eye movements were monitored as they viewed familiar objects with names of either the same (pelota, ball[fem], galleta, cookie[fem] ) or different grammatical gender (pelota, zapato, shoe[masc] ) while listening to a Spanish sentence (Encuentra la pelota, Find the ball ). L1-children and L1-adults, but not L2-learners, were faster to orient to targets on different-gender trials, when the article was informative. Experiments 2 and 3 explored whether L1-adults and L2-learners would show efficiency in processing novel nouns. Participants were tested on article-noun sequences that they either did (Exp. 2) or did not (Exp. 3) hear together during a teaching phase. L2-learners took advantage of informative articles only when tested on the same article-noun sequences heard during teaching. Implications for distributional and grammatical accounts of language processing are discussed.


(Return to Top)

Semantic Cognitive Mapping of Natural Language

Alexei V. Samsonovich   (asamsono@gmu.edu)
George Mason University

Giorgio A. Ascoli   (ascoli@gmu.edu)
George Mason University

Rebecca F. Goldin   (rgoldin@gmu.edu)
George Mason University

Methods of topology and geometry become increasingly popular in semantic analysis. The challenge of mapping all semantics of English texts can be compared to the human genome project by its significance. A semantic cognitive map (SCM) is understood here as an abstract manifold X in which documents are allocated as points, such that geometric relations reflect semantic relations. This notion underlies the approach of the present study, the objective of which is to determine general topological and geometrical characteristics of X. For this purpose, we use another SCM that we previously constructed from dictionary words allocated in a vector space V, where synonyms/antonyms have positive/negative dot products as vectors (Samsonovich & Ascoli 2007). We show how V can be used to determine some aspects of the local structure of X and to suggest a plausible model of X. The study follows a trajectory from initial observations through experimental hypothesis testing toward a mathematical theory.


(Return to Top)

Poster Session I -- Learning and Development

(Thursday, July 24, 2008, 5:30-7:00 PM, Exhibit Hall)


The Role of Mental Simulation and Embodied Instruction in Promoting Understanding of Robotics Systems

Margaret S. Chan   (msc29@columbia.edu)
Teachers College, Columbia University

John B. Black   (jb21@columbia.edu)
Teachers College, Columbia University

How we can help young children understand complex systems? Individuals construct knowledge through activity. Designing and programming robotics systems is a promising vehicle to help students understand how systems work. To reason about how systems functions, individuals need to construct a mental model of these systems (Gentner & Stevens, 1983; Wilensky & Resnick, 1999). We believe understanding the system?s structural configuration and the functional relations among systemic components are equally critical. Recent cognitive science research has contended that human cognition is embodied when the sensory areas of the brain become activated along with its symbolic areas during the performance of a symbolic task, such as solving math problems. (Barsalou, 2005; Smith, 2005). The study extends previous research by investigating the effectiveness of embodied instruction and mental simulation to help children understand programming and robots as dynamic systems. Results indicated embodied instruction interacts with programming experience in children's understanding of robotics.


(Return to Top)

Parent-Child Interactions and Note-Keeping During Science Play

Elizabeth L. Echeveste   (elizabethmcleod@csufresno.edu)
California State University, Fresno

Lara M. Triona   (ltriona@csufresno.edu)
California State University, Fresno

The current study examines the way that parent-child interactions influence children's science learning during informal science activities. Previous research suggests that when parents and children work together on a problem solving activity the child often takes on the role of manipulating the materials while the parent records the results (Gleason & Schauble, 1999). The current study examines informal interactions during a museum exhibit activity. The exhibit allowed children to test cars with different wheels and record their trials on a chart. We expected families that were rated as having didactic patterns of interactions would engage in more systematic testing and note taking while families who used modeling or promoted self-discovery would be more playful and less likely to take notes. An analysis of 18 interactions (ranging from 45 seconds to 10 minutes) confirmed this pattern of results. These findings corroborate prior work emphasizing the social nature of learning.


(Return to Top)

Analogies Can Facilitate the Understanding of Counter-Intuitive Expository Texts

Stella Vosniadou   (svosniad@phs.uoa.gr)
University of Athens

Irini Skopeliti   (eskopel@phs.uoa.gr)
University of Athens

Svetlana-Lito Gerakakis   (sgerak@phs.uoa.gr)
University of Athens

Panagiotis Blitsas   (panayotisb@yahoo.com)
University of Athens

An experiment was conducted investigating the hypothesis that analogies can facilitate the understanding of counter-intuitive expository texts in elementary school children. The expository texts presented the scientific explanation of day/night cycle either with or without the use of analogies. Text understanding required children to replace a familiar explanation/mechanism (e.g. Sun's movement and occlusion) with an unfamiliar and counter-intuitive one (Earth's axis rotation). We hypothesized that analogy would help in this process because it would present the new mechanism through a familiar domain (analogy to gyros, meat roasted on vertical rotisserie). We expected children from the analogy condition to (a) recall more ideas, especially those referring to the description of the mechanism, (b) create less high-level distortions related to the mechanism, and (c) achieve more significant changes in their original explanatory structures. The results confirmed all the above hypotheses, supporting the claim that analogies can facilitate the understanding of counter-intuitive information. Acknowledgment This work is supported by the project ANALOGY: Human-The Analogy Making Species, financed by the FP6 NEST Program of the European Commission. (STREP Contr. 029088).


(Return to Top)

Children's Computation and Representation of Past Tense

Cristina Dye   (cdd24@georgetown.edu)
Georgetown University

Matthew Walenski   (mwalenski@ucsd.edu)
University of California At San Diego

Stewart Mostofsky   (mostofsky@kennedykrieger.com)
Kennedy Krieger Institute \ Johns Hopkins University

Michael Ullman   (michael@georgetown.edu)
Georgetown University

Language requires both storage and composition. However, exactly what is memorized and what is assembled remains controversial. Inflectional morphology and particularly examination of regular and irregular past tense forms has been a fertile terrain for investigating this issue. Recent work showed that in adults the storage vs. composition of past tense forms is influenced by factors such as frequency and imagebility, with frequency being the most important. The aim of the present study was to examine how such factors might affect storage vs. composition in children. Fifty-three normally developing children with ages ranging from 8 to 12 were tested on a past tense production task which involved 32 regular forms (e.g., fail-failed) and 32 irregular forms (e.g., hold-held). Results indicate that children generally resemble adults, but that in children imageability seems to play a more important role. Details of the analysis and results are presented, along with discussion and implications.


(Return to Top)

Process of Comparison: Structural Alignment in Everyday Learning Experience

Stella Christie   (christie@northwestern.edu)
Northwestern University

Dedre Gentner   (gentner@northwestern.edu)
Northwestern University

Vladimir Sloutsky   (VSloutsky@ehe.osu.edu)
Ohio State University

Prior studies have shown that comparison between two exemplars can help children form relational abstractions. Much of this research has utilized a fairly strong ?invitation to compare.? The exemplars are often given a common label (e.g. blicket), on the theory that common labels invite comparison (Gentner & Namy, 1999). Further, children are explicitly invited to compare (?Can you see why these are both blickets??). We tested whether comparison benefits would occur with a less strenuous ?invitation to compare,? such as might occur in everyday experience. We taught 4-year-olds new spatial relational patterns (Christie & Gentner, 2007). In Experiment 1 we removed common labels, and tested whether another commonality (both relational patterns had a ?surprise? hidden behind them) would also engage structural alignment processes. In Experiment 2, we retained common labels, but removed the explicit encouragement to compare. The results showed relational learning benefits (though somewhat attenuated) even from reduced comparison conditions. \ \


(Return to Top)

Representing Part-Whole Structure in Diagrams

Reality S. Canty   (rcanty1@uic.edu)
University of Illinois At Chicago

Alison Castro-Superfine   (amcastro@uic.edu)
University of Illinois At Chicago

Anne Marie Marshall   (anne.m.marshall@gmail.com)
University of Illinois At Chicago

We examined the impact of knowledge on children s ability to represent part-whole structure in three diagram classes: hierarchy (box-diagram), matrix (ten-frame), and network (number line) (Novick, 2006). Thirteen first-graders of low (LC), moderate (MC), and high (HC) additive competence were asked to show how many ways they could break 10 into 2 parts using each diagram. The task was decomposed into the necessary inferences for representing part-whole structures in the problem statement and in each diagram. Competence and Diagram class impacted children s capacities to meet inferential demands per solution attempt and to generate valid part-whole structures. Competence alone influenced the form of reasoning used on the task. Finally, an analysis of the children s committed errors found that fractured part-whole thinking was associated with additive competence and the differential distribution of missed inferences across the diagrams. The study highlighted the importance of mathematics and diagrammatic knowledge in children s part-whole reasoning.


(Return to Top)

Science Play: Comparing Children Alone, with Peers, and with Adults

Lara M. Triona   (ltriona@csufresno.edu)
California State University

Maureen A. Callanan   (callanan@ucsc.edu)
University of California

Children's curiosity while playing can be compared with scientists' curiosity when building explanations (e.g., Gopnik, Meltzoff, & Kuhl, 2001; Simon, 2001). At a museum exhibit children's engagement is sustained longer when with an adult than by themselves (Dierking & Falk, 1994). In the current study, museum visitors compared several cars that had different shaped front-wheels. Prior findings were replicated and additionally it was found that interacting with other children was not related to the amount of time spent at the exhibit. However, children conducted about twice as many trials with the cars when other children were present than without. Interestingly, children with an adult but without other children conducted the fewest tests. Whether with an adult or additional children, visitors' experience is likely useful: parents potentially encourage children to reflect more on the meaning or interpretation, while with other children they get to see more trials of cars rolling.


(Return to Top)

Investigating the Building Blocks of Numerical Representations: Subitizing and Finger Gnosis

Marcie Penner-Wilger   (mpwilger@connect.carleton.ca)
Carleton University

Lisa Fast   (lisa@hume.ca)
Carleton University

Jo-Anne Lefevre   (jlefevre@connect.carleton.ca)
Carleton University

Brenda L. Smith-Chant   (bresmith@trentu.ca)
Trent University

Sheri-Lynn Skwarchuk   (s.skwarchuk@uwinnipeg.ca)
University of Winnipeg

Deepthi Kamawar   (dkamawar@ccs.carleton.ca)
Carleton University

Jeffrey Bisanz   (jeff.bisanz@ualberta.ca)
University of Alberta

Wendy Ann Deslauriers  (wadeslau@connect.carleton.ca)
Carleton University

What precursor abilities form the building blocks of numerical representations? Two abilities were investigated: the ability to mentally represent small numerosities, indexed by subitizing speed (Butterworth, 1999; Dehaene, 1997), and the ability to mentally represent one'?s fingers, indexed by finger gnosis (Anderson & Penner-Wilger, 2007; Butterworth, 1999). We examined the concurrent relation between these abilities and a task assessing numerical representation?--number-line estimation (Siegler & Booth, 2004). Canadian children in Grades 1-5 (N=222) participated as part of the Count Me In longitudinal study. Subitizing and finger gnosis were both significantly correlated with the linearity of children's number-line estimates (controlling for grade and processing speed). Subitizing and finger gnosis each predicted significant unique variance in children's estimates (semi-partials rs .21 and .15, respectively; R2 = .28). Our results are consistent with the view that the ability to mentally represent both small numerosities and one's fingers are building blocks for numerical representations.


(Return to Top)

Children's Acquisition of Knowledge about Creatures on a Continuum: Land-Amphibian-Water

Robyn Kondrad   (rkondrad@virginia.edu)
Arizona State University and University of Virginia

Susan C Somerville   (susan.somerville@asu.edu)
Arizona State University and Flinders University of South Australia

Forty-nine children (mean age 5;2) learned either biological or psychological information about 15 novel creatures. These creatures were identical in gross morphology except for their 4 features (nose, mouth, skin, and feet), adapted well to land or water, or moderately well to both. Each unique set of features, varied in such a way as to define relative positions of creatures along a land-amphibious-water continuum. Children s performance on two tasks was compared between those who learned to either associate structured biological or arbitrary psychological information with particular features. All children demonstrated memory for these associations immediately following learning and after a delay. Judgments of similarity demonstrated that if children had learned biological information, the number of dissimilar judgments provided was significantly greater than if children had learned psychological information. This novel finding suggests that earlier processes may occur, paving the way for the type of knowledge restructurings revealed in past studies.


(Return to Top)

Eye-Movement Patterns of Children with Dyslexia: Length and Frequency Effects

Evgenia Hristova   (ehristova@cogs.nbu.bg)
New Bulgarian University

Alexander Gerganov   (agerganov@cogs.nbu.bg)
New Bulgarian University

Ekaterina Todorova   (e.todorova@nbu.bg)
New Bulgarian University

Dyslexia is a specific learning disability that is related to difficulties in written language processing. The study explores the length and frequency effects on eye-movement patterns in Bulgarian. Eye movements of dyslexic children (between 8 and 11 years of age) with Bulgarian as a native language are studied. Eye movements were recorded while children read sentences that appeared one by one on the screen. The sentences are constructed in such a way that they contain a short (5-letters) or a long (8-letters) word that could be either low- or high-frequency. First-pass fixation duration and total time on the words are compared between dyslexic and normal readers. This is the first study investigating these effects in Bulgarian ? a language with Cyrillic alphabet and regular orthography. The data are interpreted in the light of various theories of developmental dyslexia and possible applications in the dyslexia treatment are discussed.


(Return to Top)

Probabilistic Learning With and Without Feedback: Preschoolers' Performance in the Face of Errors

Hanna Muenke   (hmuenke@stanford.edu)
Stanford University

Daphna Shohamy   (shohamy@psych.colombia.edu)
Columbia University

Natasha Kirkham   (n.kirkham@bbk.ac.uk)
Birkbeck College

Probabilistic learning was investigated in 3- and 4-year-old children. Children participated in a computerized learning game in which they were required to predict which color balloon a Mr. Potatohead would prefer. Color preferences were probabilistically defined, and children received immediate response-contingent feedback regarding their prediction. A separate group of 4-year-olds participated in an observational condition in which they learned the same series of associations without feedback, by observing a sequence of Mr. Potatoheads holding their balloons, and subsequently were tested on their knowledge of the associations without receiving any feedback. Results show that younger children performed worse than older children on the Feedback condition, and that older children performed poorly on the Observational condition relative to the Feedback-based condition. The results are discussed in relation to developmental theories of probabilistic learning.


(Return to Top)

They Were Trained, But They Did Not All Learn: Individual Differences in Uptake of Learning Strategy Training

Jarrod Moss   (jarrodm@pitt.edu)
University of Pittsburgh

Christian D. Schunn   (schunn@pitt.edu)
University of Pittsburgh

Kurt Vanlehn   (vanlehn@cs.pitt.edu)
University of Pittsburgh

Walter Schneider   (wws@pitt.edu)
University of Pittsburgh

Danielle S. Mcnamara   (d.mcnamara@mail.psyc.memphis.edu)
University of Memphis

Kevin Jarbo   (kej8@pitt.edu)
University of Pittsburgh

Self-explanation is an effective learning strategy in which students explain expository text and examples to themselves. Training students to self-explain has been shown to increase the rate at which they learn relative to a control group, and the quality of self-explanations that student produce is associated with learning. This study was designed to examine individual differences associated with good versus poor self-explainers following extensive training in how to self-explain. Several possible explanations for failure to implement the learning strategy were examined. The individual difference with the strongest association with self-explanation quality was a tendency to zone out and think about other things while reading the text. These results have implications for improving the design of tutoring systems.


(Return to Top)

Perception for Action: Dramatic Changes Between 18 and 24 Months

Sandra Y. Street   (systreet@indiana.edu)
Indiana University

Linda B. Smith   (smith4@indiana.edu)
Indiana University

 The ability to insert objects into openings of similar size and shape has commonly been used as a task of developmental assessment and can be a marker of a developmental delay. However, there has been very little research done on this developmental process. The current study determined that this ability is acquired between 18 and 24 months of age. By using shape sorting and posting tasks with the two groups of children we found the tasks to be difficult at 18 months but extremely easy at 24 months. Our next question addresses the mechanism behind this developmental process. We hypothesize that there is a visual motor connection developing during this short period of time. By using tasks which separate motor and visual abilities, we expect that 18 month old children have the skills to perform the separate tasks, yet are unable to complete a task requiring the integration of the two components. 


(Return to Top)

Analysis of Infant Vocalizations Using a Self-Organizing Map

Anne Sanda Warlaumont   (awarlmnt@memphis.edu)
The University of Memphis

D. Kimbrough Oller   (koller@memphis.edu)
The University of Memphis

Eugene H. Buder   (ehbuder@memphis.edu)
The University of Memphis

Robert Kozma   (rkozma@memphis.edu)
The University of Memphis

Rick Dale   (radale@memphis.edu)
The University of Memphis

This study represents a first attempt to apply neural networks to the acoustics of prelinguistic non-cry vocalizations produced by human infants. Results suggest that neural networks may be useful in describing very early vocal development, a foundation for speech; this study also lays groundwork for future attempts at modeling the perception of infant sounds. Spectrographic representations of vocalizations produced by three infants at three ages (3-4, 6-7, and 10-11 months) were classified using a self-organizing neural network map (SOM) combined with a single-layer perceptron. The trained network classified utterances based on abstract acoustic features. These features were predictive of whether an utterance was judged a Vowel, Squeal, or Growl by adult coders. The network also predicted the age at which an utterance was produced and the child who produced the utterance. The SOM s sensitivity to individual differences appeared greater than its sensitivity to age differences.


(Return to Top)

Poster Session I -- Modeling and Cognitive Architecture

(Thursday, July 24, 2008, 5:30-7:00 PM, Exhibit Hall)


Mindmodeling@Home

Kevin Gluck   (kevin.gluck@mesa.afmc.af.mil)
Air Force Research Laboratory

Jack Harris   (jack.harris@mesa.afmc.af.mil)
Air Force Research Laboratory

Research efforts with large computational demands have begun to leverage volunteer computing nodes to vastly increase their access to computational resources. This approach has been adopted in scientific fields as diverse as astronomy, biology, climatology, mathematics, and physics. Until now cognitive science has not had a presence in the volunteer computing community. This poster reports the creation and current status of MindModeling@Home, the first volunteer computing project dedicated to advancements in computational cognitive process modeling (http://mindmodeling.org).


(Return to Top)

The Model Brain: Brain Information Hydrodynamics (Bih)

Muneo Kitajima   (kitajima@ni.aist.go.jp)
National Institute of Advanced Industrial Science and Technology

Makoto Toyota   (mako1950@mac.com)
T-Method

Hideaki Shimada   (hshimada@shinshu-u.ac.jp)
Shinshu University

We propose Brain Information Hydrodynamics as a theory for constructing the Model Brain, traditionally conceived as electronically based neuronal networks and/or chemically based hormone field. In BIH, the influx of environmental information is first filtered to reduce the amount of information to a tractable number of chunks. The influx flows along the terrain, originally shaped by genes, and then transformed through experience. A number of flow modes exist, each corresponding to a band in Newell's time scale of human action. Immediate behavior is generated when the influx reaches the cerebellum directly. Deliberate behavior is generated when it is trapped midway in the cerebrum where a number of vortices are created to transform the values of attributes of the information that the influx conveys successively to the ones finally exerted. Real-time constraints of behavior are satisfied by creating emotional vortices, forcing the flow to reach the cerebellum in a timely manner.


(Return to Top)

A Functional, Evolutionary, and Developmental Model of Neocortex

Derek James   (djames@gmail.com)
University of Louisiana At Lafayette

Anthony Maida   (maida@cacs.louisiana.edu)
University of Louisiana At Lafayette

We present a model of mammalian neocortex. The model posits that neocortex is highly uniform and modular, and that the fundamental functional unit is the neocortical minicolumn. Each minicolumn is situated in a hierarchy and functions as an inference/prediction unit. The model also proposes that large-scale structure and organization of neocortex is encoded and optimized via evolutionary processes, while local, fine-grained structure is a product of developmental dynamics. This paper describes the model and presents an account of how concepts are acquired through Hebbian mechanisms and processed in the model once those concepts are learned.


(Return to Top)

How Many Exemplars Do We Need? Explorations with the 'Rex Leopold II' Model

Maarten De Schryver   (Maarten.DeSchryver@Ugent.be)
Ghent University

Katleen Vandist   (Katleen.Vandist@Ugent.be)
Ghent University

Yves Rosseel   (Yves.Rosseel@Ugent.be)
Ghent University

Despite the dominance of exemplar models, there is still ambiguity about the number of exemplars stored in memory. It seems implausible that every exemplar we have encountered would be a separate part of the representation. In this poster, we introduce the Rex Leopold II model. This model is designed to be identical to the General Context Model (Nosofsky, 1986) with the exception that the full set of exemplars can be replaced by a reduced set of exemplars. The model tries to identify exemplars that can be omitted from the full set. Importantly, Rex LII is designed in such a way that omitting those exemplars may not affect categorization performance. The aim of Rex LII is not to outperform the GCM, but to fit at least as well using only a reduced set of exemplars. Preliminary results confirm our belief that only a subset of exemplars may be sufficient for category representation.


(Return to Top)

Preference Aggregation Based Cognitive Modeling: An Alternative Explanation of the Wason Selection Task

Kenryo Indo   (kindo@kanto-gakuen.ac.jp)
Kanto Gakuen University

The literature on the Wason Selection Task (WST) for the rule if p then q" reveals that most subjects fail to choose a combination of "not q" and "p" cards. In this paper, an alternative modeling for the WST based on preference aggregation with its simulation by PROLOG is proposed. Let A be the set of cards and a threshold of inspection. First, a set of "more suspicious than" relations on A, as PROLOG clauses, is generated. Then, the choice of cards is decided by a pairwise majority vote. Assuming a suspicious stance for that rule, a subject chooses cards correctly only when the cyclic voting occurs. However, barring this exception, pairwise majority is the one and only non-dictatorial aggregation which is also proof against varying a relation. Therefore, the WST is difficult because the subject, who is not likely decisive, should stick to the self-deceptive cycle."


(Return to Top)

Dynamic Field Theory of Sequential Action: A Model and Its Implementation on an Embodied Agent

Yulia Sandamirskaya   (sandayci@neuroinformatik.rub.de)
Institut Für Neuroninformatik, Ruhr-Universität Bochum

Gregor Schöner   (gregor.schoener@neuroinformatik.rub.de)
Institut Für Neuroninformatik, Ruhr-Universität Bochum

 How sequences of actions are learned, remembered, and generated is a core problem of cognition. Sequence generation is also a challenge for Dynamical Systems approaches to cognition, because the attractor concept seems to be in conflict with the need to destabilize a state in order to switch to the next one. Here we examine how Dynamic Field Theory (DFT), a neuronally grounded dynamical systems approach to embodied cognition, may address sequence learning, sequence memory, and sequential action. To demonstrate that the approach solves the key problem of stabilizing sequences against variable timing of individual actions and noisy sensory feedback about the state of the action system, we implement the approach on a simple autonomous robot. We demonstrate how the robot acquires sequences from experiencing the associated sensory information in serial order and how the robot generates sequences based on visual information from its environment using low-level visual features.


(Return to Top)

Grounding Classical Cognitive Science

Bernadette Guimberteau   (bguimberteau@cal.berkeley.edu)
University of California At Berkeley

What happens when one grounds classical cognitive science research? Do conceptual insights emerge from the exercise or do they remain the same? Do well-known empirical results lose their robust status, or do they hold firmly against the inquiry? This paper grounds cognition with the Tower of Hanoi task. Conceptually, a new class of strategies emerges that is compared to known classical classes. Empirically, the paper points out that a number of classical results are revised. In other words, grounding problem solving with the task is not a trivial exercise.


(Return to Top)

The Conscious Structure of the Brain

Phil Maguire 
Nui Maynooth

Kyle O' Connor 
University College Dublin

Despite the fact that phenomenal consciousness appears undeniable from the first person perspective, reductionist accounts of brain function fail to provide a satisfactory explanation. We propose a novel perspective of consciousness which can resolve this problem. The brain has been shaped by evolution to produce optimised behaviour for an organism. The resulting structure is somehow capable of supporting consciousness. Clearly then, the conscious structure of the brain must be related to its potential for action. We propose that a system is conscious if it is somehow isomorphic with a singularity. The brain meets this criterion because the optimal behaviour for an organism is that which reflects the actions of a singular entity in space and time. According to this account, consciousness is a global systemic property of the brain. It arises from the manner in which the brain constantly re-arranges itself to reflect the behavioural disposition of a singular entity.


(Return to Top)

Investigating Functional Cooperation in the Human Cortex with Graph-Theoretic Methods

Michael Anderson   (michael.anderson@fandm.edu)
Franklin & Marshall College; University of Maryland

Joan Brumbaugh   (joan.brumbaugh@fandm.edu)
Franklin & Marshall College

Aysu Suben   (aysu.suben@fandm.edu)
Franklin & Marshall College

This paper introduces a very simple analytic method for mining large numbers of brain imaging experiments to discover functional cooperation between regions. We then report some preliminary results of its application, illustrate some of the many future projects in which we expect the technique will be of considerable use, and describe a research resource for investigating functional cooperation in the cortex that will be made publicly available through the lab website. One significant finding is that differences between cognitive domains appear to be attributable more to differences in patterns of cooperation between brain regions, rather than to differences in which brain regions are used in each domain. This is not a result that is predicted by prevailing localization-based and modular accounts of the organization of the cortex.


To Poster Session II Member Abstacts, Friday, July 25, 2008, 5:00-7:00 PM, Exhibit Hall

To Poster Session III Member Abstracts, Saturday, July 26, 2008, 5:00-7:00 PM, Exhibit Hall

 

Cognitive Science Journal Archive

The Cognitive Science Journal Archive currently contains electronic versions of 459 articles (of 98 issues and 24 years) of the Cognitive Science Journal and collects materials published in the Proceedings of the Annual Cognitive Science Conference. It is maintained by the CogWorks Laboratories of RPI's Cognitive Science Department.

CogWorks Labs, RPI