Predicting the Effects of Cell-Phone Dialing on Driver Performance
Dario D. Salvucci (dario@cbr.com)
Kristen L. Macuga (kristen@cbr.com)
Nissan Cambridge Basic Research, Four Cambridge Center
Cambridge, MA 02142 USA
Legislators, journalists, and researchers alike have recently directed a great deal of attention to the effects of cellular telephone ("cell phone") use on driver behavior and performance. This paper demonstrates how cognitive modeling can aid in understanding these effects by predicting the impact of cell-phone dialing in a naturalistic driving task. We developed models of four methods of cell-phone dialing and integrated these models with an existing driver model of steering and speed control. By running this integrated model, we generated a priori predictions for how each dialing method affects the accuracy of steering and speed control with respect to an accelerating and braking lead vehicle. The model predicted that the largest effects on driver performance arose for dialing methods with high visual demand rather than methods with long dialing times. We validated several of the model's predictions with an empirical study in a fixed-based driving simulator.
Modeling Task Performance Using the Queuing Network Model Human Processor (QNMHP)
Robert Feyen (rfeyen@umich.edu)
Department of Industrial and Operations Engineering, The University of Michigan
1205 Beal Ave., Ann Arbor, MI 48109-2117 USA
Yili Liu (yililiu@umich.edu)
Department of Industrial and Operations Engineering, The University of Michigan
1205 Beal Ave., Ann Arbor, MI 48109-2117 USA
Human performance modeling approaches (HPMAs) that are comprehensive and computational are particularly useful for engineering design. Current approaches have strengths in modeling a person's actions, but lack underlying mathematical foundations on which to base predictions of time and capacity related performance measures. This paper presents a complementary approach that combines elements of the GOMS/Model Human Processor approach with the mathematical concepts and methods of queuing networks. Called the Queuing Network Model Human Processor (QNMHP), the approach provides quantitative predictions and theoretical insights regarding a person’s performance. The general queuing network and the approach are discussed with respect to human performance and neuroscience findings. The QNMHP is used to model reaction time tasks and the results compare favorably to prior literature; these findings are discussed briefly along with current efforts to model a driving task.
Modelling Taxi Drivers' Learning and Exceptional Memory of Street Names
Tei Laine
Indiana University: Computer Science Department and the Cognitive Science Program
Virpi Kalakoski
University of Helsinki: Department of Psychology
A computer simulation was designed to model taxi drivers' learning and memory performance, and predict experimental results in a memory test in which the stimuli are lists of street names ordered according to varying degrees of meaningfulness. The objectives of the study are, firstly to explicate the quantitative and qualitative differences between performance outcomes observed in expert and novice drivers in memory tests, and secondly to formalise the behavioural traits assumed to constitute the essence of expertise, and finally to test the adequacy of these assumptions with a computer simulation.
A Modular Neural-Network Model of the Basal Ganglia's Role in Learning and Selecting Motor Behaviours
Gianluca Baldassarre (gbalda@essex.ac.uk)
Department of Computer Science, University of Essex
CO4 3SQ Colchester, UK
This work presents a modular neural-network model (based on reinforcement-learning actor-critic methods) that tries to capture some of the most-relevant known aspects of the role that basal ganglia play in learning and selecting motor behavior related to different goals. In particular some simulations with the model show that basal ganglia selects "chunks" of behaviour whose "details" are specified by direct sensory-motor pathways, and how emergent modularity can help to deal with multiple behavioral tasks. A "top-down" approach is adopted. The starting point is the adaptive interaction of a (simulated) organism with the environment, and its capacity to learn. Then an attempt is made to implement these functions with neural architectures and mechanisms that have a neuroanatomical and neurophysiological empirical foundation.
The Role of Computational Modeling in Understanding Hemispheric Interactions and Specialization
James A Reggia and Reiner Schutz
University of Maryland
We describe results from three models of paired left and right cerebral regions communicating via a simulated corpus callosum. Conditions are identified under which functional lateralization emerges during learning, and under which an intact hemispheric region contributes to recovery when the contralateral one is damaged. It proved easy to demonstrate hemispheric specialization in the context of a variety of underlying cortical asymmetries, consistent with past arguments that lateralization of cognitive functions is a multi-factorial process. However, no single assumption about transcallosal influences was adequate to account for existing experimental data. A possible solution to this "callosal dilemma" is suggested.
Hippocampal Cognitive Maps: An Alternative View
Alexei Samsonovich (asamsono@gmu.edu)
Krasnow Institute for Advanced Study at George Mason University
2A1 Rockfish Creek Lane, Fairfax, VA 22030-4444 USA
Giorgio A. Ascoli (ascoli@gmu.edu)
Krasnow Institute for Advanced Study and Department of Psychology, George Mason University
2A1 Rockfish Creek Lane, Fairfax, VA 22030-4444 USA
Hippocampal place cells in rodents each selectively fire at a high rate when the animal is in a particular location of an environment; however, the firing rate never reaches zero. The present work is based on a hypothesis that the gradient of a firing rate distribution of each place cell encodes directions to an arbitrarily selected potential goal, located at the maximum of the firing rate. It is found that in a simple connectionist model this property of place cells can result from associative learning. After extensive, random exploration of a complicated maze, a simulated virtual robot finds quasi-optimal paths to/from any given locations in the maze and exploits shortcuts when new doorways are opened. The model's place cell dynamics are compatible with hippocampal multiunit data recorded from freely behaving rats. Interestingly, the same model can also solve non-spatial tasks.
Modeling Icon Search in ACT-R/PM
Michael D. Fleetwood and Michael D. Bryne
{fleet, byrne}@rice.edu)
Rice University, Department
of Psychology
6100 Main Street MS-25,
Houston, TX 77005 USA
As the use of graphical user interfaces expands into new
areas, icons are becoming an increasingly important aspect of GUIs. Oddly,
little research has been done into the the costs and benefits associated with
using icons. One aspect of icons, icon borders, has been proposed as a means of
adding information to icons. An experiment was conducted in which the potential
cost in response time of using simple icon borders was investigated. Two models
were then constructed in ACT-R/PM to carry out the same icon search task as in
the experiment. The results of the modeling effort indicated an area where the
design of the experiment could be improved. A second, “improved”
experiment was carried out, the results of which suggest areas for further
improvement in the ACT-R/ PM models.
Extending Task Analytic Models of Graph-based Reasoning: A Cognitive
Model of Problem Solving with Cartesian Graphs in ACT-R/PM
David Peebles (djp@psychology.nottingham.ac.uk)
Peter C.-H. Cheng (pcc@psychology.nottingham.ac.uk)
ESRC Centre for Research in Development, Instruction and Training,
Department of Psychology, University of Nottingham, Nottingham, NG7 2RD, U.K.
Models of graph-based reasoning have typically accounted for the variation in problem solving performance with different graph types in terms of a task analysis of the problem relative to the particular visual properties of each graph type (e.g. Lohse, 1993; Peebles, Cheng & Shadbolt 1999, submitted). This approach has been used to explain response time and accuracy differences in experimental situations where data are averaged over experimental conditions. An experiment is reported in which participants’ eye movements were recorded while they were solving various problems with different graph types. The eye movement data revealed fine grained fixation patterns that are not captured by current analyses based on optimal fixation sequences. It is argued that these patterns reveal the effects of working memory limitations during the time course of problem solving. An ACT-R/PM model of the experiment is described in which a similar pattern of eye fixations is produced as a natural consequence of the decay in activation of perceptual chunks over time.
A Model of Individual Differences in Learning Air Traffic Control
Niels A. Taatgen (niels@ai.rug.nl)
Artificial Intelligence, University of Groningen
Grote Kruisstraat 2/1, 9712 TS Groningen, the Netherlands
Individual differences in skill acquisition are influenced by several architectural factors. According to Ackerman’s theory, general intelligence, speed of proceduralization and psycho-motor speed influence different stages of skill acquisition. Ackerman tested this theory by correlating performance on an Air Traffic Controller (ATC) task with tests on specific abilities. The present study discusses an ACT-R model of the ATC task in which the relevant abilities can be manipulated directly, providing additional support for the theory. Keywords: Skill acquisition, Air Traffic Control, Individual differences, ACT-R
Validating a Tool for Simulating User Interaction
Jan Misker (jan@misker.nl)
Artificial Intelligence, University of Groningen,
Grote Kruisstraat 2/1, 9712 TS Groningen, The Netherlands
KPN Research,
Sint Paulusstraat 4, 2264 XZ Leidschendam, The Netherlands
Niels A. Taatgen (niels@ai.rug.nl)
Artificial Intelligence, University of Groningen
Grote Kruisstraat 2/1, 9712 TS Groningen, The Netherlands
Jans Aasman (J.Aasman@kpn.com)
KPN Research,
Sint Paulusstraat 4, 2264 XZ Leidschendam, The Netherlands
Industrial Design Engineering, University of Delft,
Jaffalaan 9, 2628 BX, Delft
In this paper a tool will be presented that simulates human perception and motor behavior in interaction with graphical user interfaces in the Microsoft Windows environment. The simulated hand and eye tool can be used in combination with any cognitive architecture. In order to validate the simulated hand and eye an experiment has been conducted in which human subjects showed simple, low-level interaction behavior. Eye movements and finger presses were measured and recorded. This allowed for basic validation of the simulated hand and eye. Furthermore, based on the collected data, an adjustment of the EMMA theory is suggested. To demonstrate the full usage if the tool, ACT-R has been used to model basic skill acquisition in instructed interaction behavior.
Modeling Children’s Case Marking Errors with MOSIAC
Steve Croker, Julian Pine, & Fernand Gobet
University of Nottingham
We present a computational model of early grammatical development which simulates case-marking errors in children’s early multi-word speech as a function of the interaction between a performance-limited distributional analyser and the statistical properties of the input. The model is presented with a corpus of maternal speech from which it constructs a network consisting of nodes which represent words or sequences of words present in the input. It is sensitive to the distributional properties of items occurring in the input and is able to create ‘generative’ links between words which occur frequently in similar contexts, building pseudo-categories. The only information received by the model is that present in the input corpus. After training, the model is able to produce child-like utterances, including case-marking errors, of which a proportion are rote-learned, but the majority are not present in the maternal corpus. The latter are generated by traversing the generative links formed between items in the network.
Bootstrapping in Miniature Language Acquisition
Rutvik Desai (RUDESAI@Indiana.Edu)
Computer Science Department and The Cognitive Science Program,
Indiana University, Bloomington, IN 47405 USA
Given the difficulties in learning meanings of words by observing the referent, it has been suggested that children use the syntactic context of the word to predict part of its meaning, a hypothesis known as syntactic bootstrapping. Semantic bootstrapping is the opposite theory that the knowledge of semantics helps in acquiring syntax. While there is evidence that children can apply their knowledge of correlations between syntax and semantics to perform bootstrapping, it is not clear how they come to know about these correlations in the first place. Here, a connectionist network is presented that learns to comprehend a miniature language by associating sentences with the corresponding scenes. In doing so, it learns the syntactic/semantic correlations and exhibits bootstrapping behavior. It is argued that such specialized phenomena can emerge when general mechanisms are applied to a specific task, and it is not always necessary to endow the learner with pre-existing specialized mechanisms.
A Self-Organizing Neural Network Model of the Acquisition of Word Meaning
Igor Farkas & Ping Li
University of Richmond
In this paper we present a self-organizing neural network model of the acquisition of word meaning. Our model consists of two sub-networks and builds on the basic concepts of Hebbian learning and self-organization. One network learns to approximate word transition probabilities, which are used for lexical representation, and the other network, a self-organizing map, is trained on these representations, projecting them onto a 2D space. The model relies on lexical co-occurrence information to represent word meanings in the lexicon. The results show that our network is able to acquire semantic representations from both artificial data and real corpus of language use. In addition, the network demonstrates ability to develop rather accurate word representations even with sparse training data set.
Modeling the optimal infinite stage in MOSIAC: a generalization to Dutch
Daniel Freudenthal, Julian Pine, & Fernand Gobet
University of Nottingham
This paper presents a model of a stage in children’s language development known as the optional infinitive stage. The model is an adaptation of one which was originaly developed for English, where it was shown to provide a good account of several phenomena. The model, which is a discrimination network, analyzes the transitional probabilities of words in the input, and derives word classes from them by linking words that are similar. Where the earlier version of the model is sensitive only to characteristics of phrases that follow target words, the present version also takes preceding input into consideration. Also, the present version uses a probabilistic rather than an absolute learning mechanism. Generalisation of the model to Dutch is considered a strong test of the model, since Dutch displays the optional infinitive phenomenon, while it differs in its syntactic characteristics. The model was presented with child directed input from two Dutch mothers, and its output was compared to that of the respective children. Despite the fact that the model was developed for a different language, it captures the optional infinitive phenomenon in Dutch as it does in English, while showing sensitivity to Dutch syntax. These results suggest that a simple distributional analyzer can capture the regularities of different languages despite the apparent differences in their syntax.
The Influence of Resource Parameters on Incremental Conceptualization
Markus Guhe (guhe@informatik.uni-hamburg.de)
Research Group Knowledge and Language Processing (WSV),
Department for Informatics, University of Hamburg,
Vogt-Kölln-Straße 30, D-22527 Hamburg, Germany
Christopher Habel (habel@informatik.uni-hamburg.de)
Research Group Knowledge and Language Processing (WSV),
Department for Informatics, University of Hamburg,
Vogt-Kölln-Straße 30, D-22527 Hamburg, Germany
INC (incremental conceptualizer) is a cognitively motivated model of the first component of the language production process, conceptualization. It produces preverbal output for the domain of giving on-line descriptions of events that has the same structure as verbalizations of humans given the same task. The architecture and processing mechanisms of INC are based on simple, cognitively motivated principles. Its behavior is controlled by parameters that are set according to the available resources. Thus, INC requires no explicit and complex instructions about how to perform a given verbalization task, e.g. with respect to the level of detail, because the level of detail is an effect that depends on resources.
Modeling Counteroffer Behavior in Dyadic Distributive Negotiation
Danilo Fum * (fum@univ.trieste.it)
Department of Psychology; via S. Anastasio 12
Trieste, I-34134 Italy
Fabio Del Missier (delmisfa@univ.trieste.it)
Department of Psychology; via S. Anastasio 12
Trieste, I-34134 Italy
An experiment on dyadic distributive negotiation is presented that analyzes the role of the market price as a credible reference point in a bargain between a human buyer and a computerized seller implementing a contingent negotiation strategy. The market price had strong effects on the initial reservation and aspiration prices, and indirectly affected the settlement price and the number of negotiation cycles, but not the agreement likelihood. An explicit frame-related manipulation, induced by the instructions, did not yield significant effects. Two simulative models of the offer formation process, grounded on the behavioral decision approach, were proposed and evaluated. The results support the view of the negotiator as a limited information-processing decision-maker, and suggest the possibility of contingent selection of reference points.
Fitting the ANCHOR Model to Individual Data: A Case Study in Bayesian Methodology
Alexander A. Petrov (apetrov@andrew.cmu.edu)
Department of Psychology; Carnegie Mellon University
Pittsburgh, PA 15213 USA
This paper presents a memory-based model of direct psychophysical scaling. The model is based on an extension of the cognitive architecture ACT-R and uses anchors that serve as prototypes for the stimuli classified within each response category. Using the ANCHOR model as a specific example, a general Bayesian framework is introduced. It provides principled methods for making data-based inferences about models of this kind. The internal representations in the model are analyzed as hidden variables that are constructed from the stimuli according to probabilistic representation rules. In turn, the hidden representations produce overt responses via probabilistic performance rules. Incremental learning rules transform the model into a dynamic system. A parameter-fitting algorithm is formulated and tested on experimental data.
A Bayesian Model for the Time Course of Lexical Processing
Mark Steyvers (msteyver@psych.stanford.edu)
Department of Psychology, Stanford University
Stanford, CA 94305-2130
Eric-Jan Wagenmakers (pn_wagenmakers@macmail.psy.uva.nl)
Department of Psychology, University of Amsterdam
Amsterdam, The Netherlands.
Richard Shiffrin (shiffrin@indiana.edu)
Indiana University,
Bloomington, IN 47405-7007
René Zeelenberg (pn_zeelenberg@macmail.psy.uva.nl)
Department of Psychology, University of Amsterdam
Amsterdam, The Netherlands
Jeroen Raaijmakers (raaijmakers@psy.uva.nl)
Department of Psychology, University of Amsterdam
Amsterdam, The Netherlands.
A Bayesian-based model for lexical decision, REM-LD, is fit to data from a novel version of a signal-to-respond paradigm. REM-LD calculates the odds that a test item is a word, by accumulating likelihood ratios for each lexical entry in a small neighborhood of similar words. The new model predicts the time course of observed effects of nonword lexicality, word frequency and repetition priming. It can also make qualitative predictions for the response time distributions in tasks with subject paced responding.
An Attractor Network Model of Serial Recall
Matt Jones (mattj@umich.edu) and Thad A. Polk (tpolk@umich.edu)
Department of Psychology, 525 E. University
Ann Arbor, MI 48109 USA
We present a neural network model of verbal working memory which attempts to illustrate how a few simple assumptions about neural computation can shed light on cognitive phenomena associated with the serial recall of verbal material. We assume that neural representations are distributed, that neural connectivity is massively recurrent, and that synaptic efficiency is modified based on the correlation between pre- and post-synaptic activity (Hebbian learning). Together these assumptions give rise to emergent computational properties that are relevant to working memory, including short-term maintenance of information, time-based decay, and similarity-based interference. We instantiate these principles in a specific model of serial recall and show how it can both simulate and explain a number of standard cognitive phenomena associated with the task, including the effects of serial position, word length, articulatory suppression (and its interaction with word length), and phonological similarity.
The role of abstract patterns in implicit learning
Volodymyr V. Ivanchenko (vlad@cogs.nbu.acad.bg)
Central and Eastern European Center for Cognitive Science
Department of Cognitive Science and Psychology,
New Bulgarian University,
21 Montevideo Street, Sofia 1635, Bulgaria
Implicit learning is thought to underlie language acquisition, acquisition of reading and writing abilities and many other phenomena central to cognition. The main finding in this field is that humans exposed to the stimulus material, which comprises some regularities, unintentionally acquire an ability to discriminate between stimuli with and without these regularities. Moreover, when these regularities are instantiated with a symbol set different from the training one participants still succeed in the task (so-called transfer effect, Reber,1967). I hypothesized an existence of one general mechanism capable of explaining both the discrimination and the transfer phenomenon. This mechanism is sensitive to the existence of a symbol repetition pattern in a stimulus string. I called such a repetition an abstract pattern (AP) since a fact of symbol repetition is instantiation independent. Two experiments were conducted in order to check whether participants base their grammaticality judgment on the fact of presence of AP in the test material during implicit learning of artificial grammar (AG). I found a tendency to recognize as grammatical, items with an arbitrary AP rather than those without any AP and items with an AP seen during training rather then those with an unseen AP. A computer simulation in the form of a three layer autoassociator was run and found useful for explaining participants' overall performance as well as their responses to the particular stimuli.
Intention superiority effect: A context-sensitivity account
Christian Lebiere (cl@cmu.edu)
Human Computer Interaction Institute; Carnegie Mellon University
Pittsburgh, PA 15213 USA
Frank J. Lee (fjl@cmu.edu)
Department of Psychology; Carnegie Mellon University
Pittsburgh, PA 15213 USA
Intention superiority effect (Goschke & Kuhl, 1993;
Marsh, Hicks, & Bink, 1998) is the finding that the times to retrieve
memory items related to uncompleted or partially completed intentions are
faster than for those with no associated intentions. However, this relationship
reverses when the intended tasks are completed (Marsh, Hicks, & Bink, 1998;
Marsh, Hicks, & Bryan, 1999). That is, the times to retrieve memory items
related to completed intentions are slower than for those with no associated
intentions. In this paper, we present a computational account of the intention
superiority effect using the ACT-R (Anderson & Lebiere, 1998) cognitive
architecture. Our modeling approach is based on the idea that uncompleted or
partially completed intentions are available as context in the current goal,
and they prime related memory items while inhibiting unrelated memory items.
However, once the intended tasks are completed, they are removed from the
current goal, which produces an inhibitory effect on memory items associated
with them. We describe an ACT-R model that is able to reproduce all of the
effects reported in Marsh, Hicks, and Bink (1998).
Modeling Selective Attention: Not Just Another Model of Stroop (NJAMOS)
Marsha C. Lovett (Lovett@CMU.EDU)
Department of Psychology, Carnegie Mellon University
Pittsburgh, PA 15213 USA
The Stroop effect has been studied for more than sixty years, and yet it still defies a complete theoretical account. The model NJAMOS offers a new theoretical account that integrates several explanations of the Stroop phenomenon into a hybrid model. NJAMOS is built within the ACT-R cognitive architecture (Anderson & Lebiere, 1998). Besides fitting a variety of experimental results, NJAMOS offers the potential to capture strategic variation in what is typically considered a low-level attentional phenomenon.
Inducing models of human control skills
Rui Camacho, (rcamacho@fe.up.pt)
Laboratorio de Intellgencia Artificial e Ciencias da Computacao
Rua do Campo Allegre, 823, 4150 Porto, Portugal
We propose a new model, called Incremental Correction (IC) model to address the problem of reverse engineering human control skills using the {\it Behavioural Cloning} methodology. The proposed model is based on the concept of closed loop or feedback control. Traditional knowledge acquisition methodologies like interview methods are not applicable since control skills are usually performed in humans at a sub-cognitive level and are therefore tacit. Controllers using the IC model exhibit an increase in controller robustness and a reduction in encoding complexity. The controllers were empirically evaluated in the complex control application of controlling a simulation of a F-16 aircraft.
The Application of Mathematical Techniques for Modeling Decision-Making: Lessons Learned from a Preliminary Study
Gwendolyn E. Campbell (CampbellGE@navair.navy.mil)
Wendi L. Buff (BuffWL@navair.navy.mil)
Amy E. Bolton (BoltonAE@navair.navy.mil)
David O. Holness (HolnessDO@navair.navy.mil)
Naval Air Warfare Center Training Systems Division AIR-4961
12350 Research Parkway
Orlando, Florida 32826
We are conducting a series of research studies to investigate the application of mathematical modeling for training decision-making. Along the way, we have learned many lessons in trying to apply math modeling to training such as choosing the right domain, including a sufficient number of observations, adding the right predictors to the mathematical equations, and selecting the right model on which to base training feedback. We hope these lessons learned will help other researchers who are interested in applying mathematical modeling techniques.
Factorial Modeling: A Method for Enhancing the Explanatory and Predictive Power of Cognitive Models
Rita Kovordanyi (ritko@ida.liu.se)
Department of Computer and Information Science
Linköpings Universitet, SE-581 83 Linköping, Sweden
The construction and evaluation of cognitive models can, and often do, lead to NOVEL insights into what might constitute a valid account for an empirical phenomenon. These insights constrain the space of viable models, and could be useful also on a theoretical plane, by promoting a deeper understanding of the studied phenomenon. We propose the factorial method for deriving novel, that is, not theory–based constraints in a principled way during model development. The method is based on a systematic comparison of alternative models, realized through a cross–combination of model components in a generic cognitive model. We illustrate the method by describing an application in the area of mental imagery. We conclude by discussing ways to increase the generalizability of results that can be obtained using the factorial method.
Infinite RAAM: A Principled Connectionist Substrate for Cognitive Modeling
Simon Levy & Jordan Pollack
Dynamical and Evolutionary Machine Organization
Volen Center for Complex Systems,
Brandeis University, Waltham, MA 02454, USA
Unification-based approaches have come to play an important role in both theoretical and applied modeling of cognitive processes, most notably natural language. Attempts to model such processes using neural networks have met with some success, but have faced serious hurdles caused by the limitations of standard connectionist coding schemes. As a contribution to this effort, this paper presents recent work in Infinite RAAM (IRAAM), a new connectionist unification model. Based on a fusion of recurrent neural networks with fractal geometry, IRAAM allows us to understand the behavior of these networks as dynamical systems. Using a logical programming language as our modeling domain, we show how this dynamical-systems approach solves many of the problems faced by earlier connectionist models, supporting unification over arbitrarily large sets of recursive expressions. We conclude that IRAAM can provide a principled connectionist substrate for unification in a variety of cognitive modeling domains.
Towards a Technology for Computational Experimentation
Peter Yule (p.yule@bbk.ac.uk)
School of Psychology, Birkbeck College, University of London, Malet St.,
London, WC1E 7HX
Richard Cooper (r.cooper@bbk.ac.uk)
School of Psychology, Birkbeck College, University of London, Malet St.,
London, WC1E 7HX
The evaluation of cognitive models is sometimes compromised by inadequate comparison of model behaviour with human behaviour. Inadequacies may arise through unconstrained setting of model parameters to yield the target behaviours and/or through incomplete replication of the human participant’s environment when running the model. This paper reviews methodologies for the evaluation of cognitive models that address (to varying extents) these difficulties. It is proposed that standard empirical psychology provides the basis for an adequate methodology, in which within-model variations are treated as between-subjects variables. The control of large-scale between-subjects designs requires substantial computational infrastructure, and the second half of the paper presents details of a general web-based client-server system that embodies this infrastructure. The system may be used to support the execution of computational experiments, human experiments, and computational parameter studies. Three case studies illustrate these different uses.
Fernand Gobet (frg@psyc.nott.ac.uk)
Samuel Jackson (jacksonnumber5@hotmail.com)
School of Psychology
University of Nottingham
Nottingham NG7 2RD, UK
This study reflects a recent shift towards the study of early stages of expert memory acquisition for chess positions. Over the course of fifteen sessions, two subjects who knew virtually nothing about the game of chess were trained to memorise positions. Increase in recall performance and chunk size was captured by power functions, confirming predictions made by the template theory (Gobet & Simon, 1996, 1998, 2000). The human data was compared to that of a computer simulation run on CHREST (Chunk Hierarchy and REtrieval STructures), an implementation of the template theory. The model accounts for the pattern of results in the human data, although it underestimates the size of the largest chunks and the rate of learning. Evidence for the presence of templates in human subjects was found.
An ACT-R Model of the Evolution of Strategy Use and Problem Difficulty
Glenn Gunzelmann (glenng@andrew.cmu.edu)
John R. Anderson (ja+@cmu.edu)
Department of Psychology, Carnegie Mellon University
Pittsburgh, PA 15213
Research has shown the importance of strategies in guiding problem solving behavior. The experiment and model presented here provide further specification of how more optimal strategies come to be adopted with experience. Isomorphs of the Tower of Hanoi were used to allow participants to develop a degree of expertise with a novel task. In the solutions, evidence for at least two strategies is apparent. The results suggest that when strategies are not successful in achieving the goal, other strategies may emerge and eventually come to dominate performance in a task. The ACT-R model of this task captures participant performance by using the same strategies to solve the problems and by gradually switching to more effective ones as simple strategies fail in solving the problems.
Modeling How and When Learning Happens in a Simple Fault-Finding Task
Frank E. Ritter (ritter@ist.psu.edu)
School of IST, 504 Rider, 120 Burrowes St.,
The Pennsylvania State University
State College, PA 16801 USA
Peter Bibby (peter.bibby@nottingham.ac.uk)
School of Psychology,
The University of Nottingham
Nottingham, NG7 2RD UK
We have developed a process model that learns in multiple ways using the Soar chunking mechanism while finding faults in a simple control panel device. The model accounts very well for measures such as problem solving strategy, the relative difficulty of faults, average fault-finding time, and, because the model learns as well, the speed up due to learning when examined across subjects, faults, and even series of trials for individuals. However, subjects tended to take longer than predicted to find a fault the second time they completed a task. To examine this effect, we compared the model's sequential predictions—the order and relative speed that it examined interface objects—with a subject's performance. We found that (a) the model's operators and subject's actions were applied in basically the same order; (b) during the initial learning phase there was greater variation in the time taken to apply operators than the model predicted; (c) the subject appeared to spend time checking their work after completing the task (which the model did not). The failure to match times on the second time seeing a fault may be accounted for by the subject spent checking their work whilst they learn to solve the fault-finding problems. The sequential analysis reminds us that though aggregate measures can be well matched by a model, the underlying processes that generate
these predictions can differ.
Learning of Joint Visual Attention by Reinforcement Learning
Goh Matsuda
Graduate School of Arts and Sciences,The University of Tokyo
3-8-1 Komaba,Meguro-ku,Tokyo,153-8902 Japan
Takashi Omori
Graduate School of Engineering,Hokkaido University
Kita-ku,Kita 13 Jou Nishi 8 Cho,Sapporo,060-8628 Japan
In this paper, we propose a neural network model of joint visual attention learning that plays an important role in infant development, and we discuss previous studies of experimental psychology on joint visual attention based on simulation results using the model. We assumed an imaginary experiment to develop the model. A mother and an infant are sitting face to face with a table between them. Some objects familiar to the infant are placed on the table, and toys operated by remote control are put outside of the infant’s view. The infant is given a reward of seeing something interesting only when the infant follows the mother’s gaze after eye contact. We constructed the model of this experiment with a reinforcement learning algorithm, and simulated the experiment on a computer. As a result, it was revealed that the infant could learn a series of joint-visual-attention-like actions by receiving rewards from an environment, although it initially has little knowledge of the environment. This result suggests that infants can acquire joint visual attention without comprehension of the nature of joint attention, i.e., ”I’m looking at the same thing that others are looking at.”
The Anatomy of Human Personality: A Computational Implementation
Harald Schaub (harald.schaub@ppp.uni-bamberg.de)
Institute of Theoretical Psychology, Otto-Friedrich-University Bamberg
D-96045 Bamberg, Germany
This work focuses on modeling of personality-specific behavior in complex situations. For that purpose we describe the basic assumptions of the underlying PSI –theory and point out which cognitive, motivational and emotional parameters seem to be suitable for modeling characteristics of different personality types. The modulation of parameter settings in the PSI-model generates different personalities patterns, which differ in the way of coping with specific situations. We will show the accordance of behavior pattern produced by the computer model with empirical data from human subjects.
Generating Subjective Workload Self-Assessment from a Cognitive Model
Wayne W. Zachary (Wayne_Zachary@chiinc.com)
Jean-Christophe Le Mentec (j-c_le_mentec@chiinc.com)
Vassil Iordonav (vassil_iordanov@chiinc.com)
CHI Systems,Inc.716 N.Bethlehem Pike
Lower Gwynedd,PA 19002 USA
Cognitive modeling has paid little attention to workload assessment, particularly subjective self-assessment. This paper demonstrates an approach by which a computational cognitive model can make self-assessments of its subjective workload.Comparisons with human subject data show a good degree of correspondence. The methods used are generalizable to modeling other introspective and self-assessment processes.
Learning Relational Correlations
Michael Gasser (Gasser@Indiana.Edu)
Department of Computer Science; Indiana University
Bloomington, IN 47405 USA
Eliana Colunga (EColunga@Indiana.Edu)
Department of Computer Science; Indiana University
Bloomington, IN 47405 USA
A conventional view of object categories is that they represent correlations among sets of object features. In this paper we present an analogous view of relational categories, the Micro-Relation Theory. On this view, relational categories such as ON and HIT are built up out of correlations among primitive relational features, which we call micro-relations. The process of learning relational categories involves three phases, the learning of the micro-relations within object dimensions, the learning of correlations between the micro-relations across dimensions, and the generalization from absolute to relative relations within dimensions. This paper focuses on the first two phases. We describe an experiment demonstrating the first phase of relational learning and a neural network simulation of the experiment. We conclude with a discussion of future work on the second and third phases of relational learning predicted by the theory.
An Explanation of the Length Effect for Rotated Words
Carol Whitney (cwhitney@cs.umd.edu)
Neural and Cognitive Sciences Program
Philosophy Department
University of Maryland
College Park, MD 20742
We review a model of letter-position encoding, wherein position is tagged by timing of firing relative to an underlying oscillatory cycle. We show how this model can account for data concerning reaction times for lexical decision on rotated letter strings.