Introduction to the special issue on computational cognitive modeling

th This special issue of Cognitive Systems Research features the top twelve papers from the 4 International Conference on Cognitive Modeling (ICCM-2001). These papers represent the current trends in computational cognitive modeling, displaying both the diversity and commonalities of the field. This introduction to the special issue describes the motivations for the conference and the special issue, and overviews the diversity and commonalities found in the papers.  2002 Elsevier Science B.V. All rights reserved. There are many forms of modeling that serve an a central but sometimes fractionated theme in reimportant role in cognitive science. AI models speak search on cognition. The centrality of the computato the difficulty of the task being approached and the tional approach to cognition is suggested by two nature of computations that effectively solve the facts. First, computational cognitive modeling is not task. Verbal frameworks, often called models in an isolated enterprise limited to its own world of psychology, describe a set of mechanisms at an specialist conferences and journals. Although abstract level without prematurely committing to specialist conferences do exist, work in computationspecifics (Lovett & Schunn, 2000). Mathematical al cognitive modeling is presented at core cognitive models provide concise closed-form descriptions of conferences and published in mainstream cognitive regularities in data. By contrast, computational journals. Second, it is now possible to publish purely models provide mechanistic explanations and the computational work in experimental psychology means for testing the sufficiency of those explanajournals (e.g., Ellis & Ralph, 2000; Farrell & Lewantions. dowsky, 2000). This unheralded breakthrough sugIn recent years, the popularity of the computationgests that experimentalists have begun to value al cognitive modeling approach has grown to become modeling in its own right rather than simply tolerating it as an addendum to experimental work. Despite its increasing centrality, the modeling *Corresponding author. Tel.: 1 1-412-624-8807; fax: 1 1-412world seems fractionated. Our general face is divided 624-7439. E-mail address: [email protected] (C.D. Schunn). between meetings that are either too general (such as 1389-0417/02/$ – see front matter  2002 Elsevier Science B.V. All rights reserved. PI I : S1389-0417( 01 )00038-9 2 C.D. Schunn, W.D. Gray / Cognitive Systems Research 3 (2002) 1 –3 the annual Cognitive Science conference) or too knowledge representation as a common denominator specific (such as the annual neural nets, ACT-R, and in all modeling approaches became a central theme Soar meetings) to address issues that transcend in formal and informal discussions held throughout modeling approaches. Such issues include statistics ICCM-2001. and methodologies for testing and validating compuThe 35 papers presented at ICCM-2001 were tational cognitive models, detailed comparisons of diverse along many dimensions. Talks by 22 estabdifferent approaches, as well as understanding the lished researchers were interspersed with talks by 13 contribution that different approaches can make to of the next generation of researchers. The disciplines understanding the same phenomena. It was to fill this contributing to ICCM-2001 were likewise diverse. void that the International Conferences on Cognitive Psychology had the largest representation with 14 Modeling (ICCM) came into being. papers. The next largest group was 13 papers from th The 4 ICCM, ICCM-2001 took place at George the hybrid disciplines of Cognitive Science, Artificial Mason University in late July of 2001 (see http: / Intelligence, Human-Computer Interaction, and In1 /hfac.gmu.edu/ | iccm). The Newell Award for best formatics. Our tally of disciplines is completed by student paper was awarded to Michael Fleetwood for Computer Science (4), industry and government his paper with Michael Byrne, Modeling Icon Search laboratories (3), and Industrial Engineering (1). in ACT-R /PM. Dario Salvucci and Kristan Macuga ICCM-2001 was geographically diverse as well. As won the Siegel-Wolf Award for the best applied might be expected, the host country provided the research paper. (Both of these papers are included in largest single contingent with 20 papers from the this special issue.) During the Doctoral Consortium United States. Fourteen papers were from Europe (5 meeting the day prior to the conference, seven UK, 2 from each of Germany, Portugal and The doctoral students from universities around the world Netherlands; 1 from each of Italy, Bulgaria and met with three faculty researchers. The Doctoral Sweden) and one was from Japan. Consortium provided an opportunity for the students Despite the stimulating effects of diversity in to meet their peers and mentors and to explore their experience, discipline and geography, the most imdissertation work in an intense but friendly, multiportant goal for ICCM-2001 was diversity in apapproach environment. The work presented at the proaches to modeling cognitive data. ICCM scored Doctoral Consortium was also presented during the high on this measure as well. The largest block of ICCM-2001 poster session and is documented in the papers (10) was in the neural net tradition. The next proceedings. The next ICCM is expected to be in largest (9) used the hybrid architecture of ACT-R Bamberg, Germany in the Spring of 2003. (Anderson & Lebiere, 1998). Pure symbol system Arthur Markman opened ICCM-2001 with a plenarchitectures were represented by CHREST (Gobet ary talk. The birth of new modeling approaches & Simon, 2000), inC (Guhe & Habel, 2001), Soar seems accompanied by hyperinflated claims for the (Newell, 1990) and CogNet (Zachary, Ryder & new approach and against the old approaches. In Hicinbothom, 1998) (7 papers). The conference also recent years, Markman and his colleague Eric Dietincluded two applications to modeling of machine rich have done much to put in perspective the learning techniques, three mathematical modeling or complementary strengths and weaknesses of extant Bayesian modeling papers, and four papers that modeling approaches (e.g., Markman, 1999; Markfocused on issues that cut across modeling apman & Dietrich, 2000a,b). Markman’s focus on proaches. The 12 papers chosen for this special issue of Cognitive Systems Research reflect the emerging While the 2001 meeting was called the 4th ICCM, in fact the commonalities in the field of computational cognitive first two meetings did not use the ICCM title. The first meeting, in modeling as well as the diversity of approaches 1996, was called the European Workshop on Cognitive Modeling. found at ICCM-2001. In selecting this dozen we The second meeting, in 1998, was called the European Conference were guided by a number of criteria. First, we used on Cognitive Modeling. The third meeting, held at Groningen in the ratings and comments provided by the original The Netherlands in 2000, was the first to be called an International Conference on Cognitive Modeling. reviewers of ICCM-2001 submissions. Authors had C.D. Schunn, W.D. Gray / Cognitive Systems Research 3 (2002) 1 –33 submitted full papers to the conference, and several DASW01-00-K-0017 and the U.S. Office of Navalreviewers carefully reviewed each submission. SecResearch Grant N00014-01-1-0321. Wayne D.ond, for each member of the set of highly rated Gray’s work on this introduction and ICCM-2001papers, we used our knowledge of the differences was supported in part by a grant through the U.S. Airbetween the paper submitted and the paper published Force (USAF) Office of Scientific Research byto judge how responsive the authors had been to Department of Defense Grant F49620-97-1-0353.reviewer comments. Finally, our experience as atten-dees of ICCM-2001 allowed us to use the excitementgenerated by the talk as a gauge of the work’s Referencesimportance.The results of this selection process are 12 papers Anderson, J. R., & Lebiere, C. (Eds.) (1998). Atomic components of thought. Mahwah, NJ: Erlbaum.that illustrate the diversity of computational cogni-Ellis, A. W., & Ralph, M. A. L. (2000). Age of acquisition effectstive modeling. Some of the papers focus on fre-in adult lexical processing reflect loss of plasticity in maturingquently-modeled phenomena, whereas others focus systems: Insights from connectionist networks. Journal ofon phenomena modeled for the first time. Some of Experimental Psychology-Learning Memory and Cognition,the models used results from neurobiology to con26(5), 1103–1123. Farrell, S., & Lewandowsky, S. (2000). A connectionist model ofstrain their implementation details, whereas otherscomplacency and adaptive recovery under automation. Journalremain purely at Marr’s (1982) computational level.of Experimental Psychology-Learning Memory and Cognition,Some of the models were built within general 26(2), 395–410.cognitive architectures (e.g., ACT-R), whereas other Gobet, F., & Simon, H. A. (2000). Five seconds or sixty?models were built within a framework developed for Presentation time in expert memory. Cognitive Science, 24(4), 651–682.a more narrow range of phenomena. Finally, some ofGuhe, M., & Habel, C. (2001). The influence of resourcethe models were developed from a pure researchparameters on incremental conceptualization. In Altmann, E.perspective, whereas other models were developed to M., Cleeremans, A., Schunn, C. D., & Gray, W. D. (Eds.),solve applied problems.Fourth International Conference on Cognitive Modeling (pp.What bind these diverse papers together are two 103–108). Mahwah, NJ: Erlbaum. Lovett, M. C., & Schunn, C. D. (2000). The importance offeatures that reflect the core principles of computa-frameworks for directing empirical questions: Reply to Goodietional cognitive modeling. First, these papers goand Fantino (2000). Journal of Experimental Psychology:beyond verbal or mathematical descriptions of General, 129(4), 453–456.theories or phenomena to provide mechanistic acMarkman, A. B. (1999). Knowledge representation: Mahwah, NJ:counts that can produce the phenomena being exErlbaum. Markman, A. B., & Dietrich, E. (2000a). Extending the classicalplained. Second, the papers pay serious attention toview of representation. Trends in Cognitive Sciences, 4(12),details of the empirical data. What differentiates470–475.computational cognitive modeling from AI apMarkman, A. B., & Dietrich, E. (2000b). In defense of representa-proaches and verbal models is our attempt to retion. Cognitive Psychology, 40(2), 138–171.produce the behavioral consequences of cognitive Marr, D. (1982). Vision. San Francisco: W. H. Freeman. Newell, A. (1990). Unified theories of cognition. Cambridge, MA:processes in as great a detail as possible.Harvard University Press. Zachary, W. W., Ryder, J. M., & Hicinbothom, J. H. (1998). Cognitive task analysis and modeling of decision making inAcknowledgementscomplex environments. In Cannon-Bowers, J. A., & Salas, E. (Eds.), Making decisions under stress: Implications for individual and team training (pp. 315–344). Washington, DC:Christian D. Schunn’s work on the introductionAmerican Psychological Association.and ICCM-2001 was supported in part by grantsfrom the U.S. Army Research Institute Grant