From jimmyd at cc.gatech.edu Wed Nov 12 16:22:22 1997 From: jimmyd at cc.gatech.edu (Jim Davies) Date: Wed, 12 Nov 1997 16:22:22 -0500 (EST) Subject: Modeling Orangutans with ACT-R Message-ID: I may want to model an Orangutan that knows some sign language, and I would love to do it in ACT-R. * Has there been any ACT modeling done of non-human primates? * What parameters, if any, would it be a good idea to change? If I try to do this, you all will be hearing more about it on this list. JimDavies jimmyd at cc.gatech.edu home: 404/370-1762 http://nis-www.lanl.gov/~jimmyd/ work: 404.894.6710 The orangutan would like to add something: ps: me me me banana eat you me eat banana banana you banana me eat eat From ericscot at saturn.sdsu.edu Wed Nov 5 17:50:39 1997 From: ericscot at saturn.sdsu.edu (Eric Scott) Date: Wed, 5 Nov 1997 14:50:39 -0800 (PST) Subject: Request for cog. model evaluation ref's. Message-ID: <199711052250.OAA26368@saturn.sdsu.edu> I'm involved in a project which takes as input the recorded behavior of subjects in batch form, serving as a kind of script. We're modeling their behavior by selectively telling certain productions not to fire at times when to do so would be inconsistent with subject's responses, thus allowing a less-than-perfect model of the subject to stay in synch with the script. At the end of processing such a script, we have a 'score' for each of these productions. So a production which was consistent with the data 90% of the time would be a 'pretty good' production, while a production which didn't do so well would show room for improvement. So I'm interested in literature on the subject of evaluating the performance of cognitive models, especially with respect to the sequential predictions they make in comparison to the responses actually made by subjects. If anyone could share any leads, I'd be obliged. Thanks, Eric Scott From klahr+ at andrew.cmu.edu Wed Nov 5 18:04:52 1997 From: klahr+ at andrew.cmu.edu (David Klahr) Date: Wed, 5 Nov 1997 18:04:52 -0500 (EST) Subject: Request for cog. model evaluation ref's. In-Reply-To: <199711052250.OAA26368@saturn.sdsu.edu> References: <199711052250.OAA26368@saturn.sdsu.edu> Message-ID: Excerpts from mail: 5-Nov-97 Request for cog. model eval.. by Eric Scott at saturn.sdsu.e > So I'm interested in literature on the subject of evaluating > the performance of cognitive models, especially with respect to > the sequential predictions they make in comparison to the responses > actually made by subjects. If anyone could share any leads, I'd be > obliged. > See fig 6.8, Newell * Simon (72). They did it the hard way. At a more generic level (non-PS models) you might find the "rule assessment" literature relevant. See Siegler(81) SRCD monographs, 46 (2, #189) Klahr & Robinson (81) cog psych, 13, 113-148 Fay & Klahr (96) Child development, 67, 689-716 (esp pp 707-709) dk From pirolli at parc.xerox.com Wed Nov 5 18:49:27 1997 From: pirolli at parc.xerox.com (Peter Pirolli) Date: Wed, 5 Nov 1997 15:49:27 PST Subject: Request for cog. model evaluation ref's. In-Reply-To: References: <199711052250.OAA26368@saturn.sdsu.edu> <199711052250.OAA26368@saturn.sdsu.edu> Message-ID: Also, check out the Appendix to Chapter 5 of Card, Moran, & Newell (1983) "The Psychology of Human-Computer Interaction." The approach is basically a dynamic programming algorithm that attempts to minimize a cost function that grows with the degree of sequence mismatching. The idea is very general. From Dario_Salvucci at inbox.fox.cs.cmu.edu Wed Nov 5 23:50:16 1997 From: Dario_Salvucci at inbox.fox.cs.cmu.edu (Dario Salvucci) Date: Wed, 05 Nov 1997 23:50:16 -0500 Subject: Request for cog. model evaluation ref's. In-Reply-To: Your message of "Wed, 05 Nov 1997 14:50:39 PST." <199711052250.OAA26368@saturn.sdsu.edu> Message-ID: <30653.878791816@inbox.fox.cs.cmu.edu> > So I'm interested in literature on the subject of evaluating > the performance of cognitive models, especially with respect to > the sequential predictions they make in comparison to the responses > actually made by subjects. If anyone could share any leads, I'd be > obliged. You may also be interested in the following work, which discusses this topic in relation to Soar. The work includes an application of (among other techniques) the Card, Moran, & Newell (1983) sequence matching algorithm. Ritter, F. E. (1992). A methodology and software environment for testing process model's sequential predictions with protocols. Doctoral Dissertation, Department of Computer Science, Carnegie Mellon University. Ritter, F. E., & Larkin, J. H. (1994). Developing process models as summaries of HCI action sequences. Human-Computer Interaction, 9, 345-383. - Dario From cl+ at andrew.cmu.edu Mon Nov 10 17:02:26 1997 From: cl+ at andrew.cmu.edu (Christian J Lebiere) Date: Mon, 10 Nov 1997 17:02:26 -0500 (EST) Subject: 1998 ACT-R Summer School and Workshop Message-ID: Thank you all for your feedback regarding the most convenient scheduling of the 1998 ACT-R Summer School and Workshop. It is of course impossible to please all the people all the time, and this is no exception. Some people, in particular Europeans, prefered the latest possible date. Many people, however, objected strenuously to scheduling the workshop immediately before the Cognitive Science conference, as we did last year. We tried to reach the best possible compromise by holding the workshop from Saturday July 18 to Tuesday July 21. As usual, the summer school will take place the preceding two weeks, from Monday July 6 to Friday July 17, with the intervening weekend free. A detailed formal announcement will be released in early December. Note: AAAI-98 will be held in Madison, Wisconsin from July 26 to 30 (see http://aaai.org). This may be of interest to those coming from afar and wondering what to do in the US between the end the Workshop and the start of CogSci98. From gray at gmu.edu Fri Nov 14 17:20:55 1997 From: gray at gmu.edu (Wayne Gray) Date: Fri, 14 Nov 1997 17:20:55 -0500 Subject: Schema thy from an act-r perspective: Comments? Message-ID: Taken from a work-in-progress. Reflections upon an act-r account of schema theory. First two paragraphs introduce schemas and what they are used for. "AOs" are submarine approach officers. Third paragraph provides the ACT-R explanation of schemas. I hope the first two paragraphs are coherent out of context. Question for the group -- have I missed anything? Are schemas really simply an "intervening variable" or is there more? Are they a hypothetical construct? (If you have no idea what the distinction is and if it is not obvious from the context, check out: MacCorquodale, K. and P. E. Meehl (1948). "On a distinction between hypothetical constructs and intervening variables." Psychological Review 55: 95-107. A truely important paper.) Wayne *********************************** We alluded above to the AO's ability to see patterns among attributes of the solution. This ability to recognize a well-formed solution by identifying the interrelationships and constraints present in the data, implies the existence of certain types of data-structures in the AO's long-term memory. In the psychological literature, such data-structures are often referred to as schemas (Brewer & Nakamura, 1984; VanLehn, 1989). They function to organize and represent large amounts of related information. We postulate that AOs have such data-structures and use them in two ways. First, at any given time, the schema provides a summary of the current state of the situation assessment. This summary guides the AO in deciding what to do next; i.e., what piece of information (attribute) is currently the most important to obtain. Second, the schema organizes the data in such a way as to make a good solution apparent. Whether the schema triggers a recognition-like process ("ah-hah, I have it!") or merely supports a more deliberate assessment of the current information are alternatives that will not be addressed here. The important point is that (a) schemas provide a pre-existing organizational structure into which the data may be placed as it is received, and (b) the schema is what is interrogated (either by pattern recognition type processes or more deliberate processes) to determine if a good solution has been found. At the theoretical level, we view schema as an intervening variable not as a hypothetical construct (MacCorquodale & Meehl, 1948). In our view, a schema is simply a specialized set of declarative memory elements (DMEs) and their associated procedural memory elements (productions). The DMEs that compose "a schema" differ from other DMEs only in that they constitute specialized types of DMEs (i.e. task specific) with preexisting (pre-scenario) attributes and values, and strong preexisting associative connections to other task-specific DMEs. Likewise, the schema productions differ from other productions only in that they are specialized to operate upon the schema DMEs during performance of a very specialized task. To say that our AOs have schemas for situation assessment, is a simple claim that they possess a set of interrelated DMEs and productions that are specialized for the situation assessment task. #=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#= Wayne D. Gray HUMAN FACTORS & APPLIED COGNITIVE PROGRAM SNAIL-MAIL ADDRESS (FedX et al) VOICE: +1 (703) 993-1357 George Mason University FAX: +1 (703) 993-1330 ARCH Lab/HFAC Program ********************* MSN 2e5 * Work is infinite, * Fairfax, VA 22030-4444 * time is finite, * http://www.hfac.gmu.edu/People/WGray/Wgray.html * plan accordingly. * #=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#=#= From ritter at psychology.nottingham.ac.uk Mon Nov 17 14:27:43 1997 From: ritter at psychology.nottingham.ac.uk (ritter at psychology.nottingham.ac.uk) Date: Mon, 17 Nov 1997 19:27:43 GMT Subject: 5 jobs going at Nottingham Message-ID: <199711171927.TAA01706@vpsyc.psyc.nott.ac.uk> This is not up to date, but I thought I would pass on to you that there are several posts going at Nottingham, which has traditionally been very sympathetic to cognitive modelling. The deadline for applications I think is the 21st or 29th of November, so I have been remiss in sending this out, but it has slipped several times, so a prompt response or enquiry is likely to be well received even if you are a few days late. Cheers, Frank **************************************************************** UNIVERSITY OF NOTTINGHAM, UK DEPARTMENT OF PSYCHOLOGY The University of Nottingham is continuing its commitment to excellence in both research and teaching in the Department of Psychology. Applications are invited for the following posts from outstanding researchers with demonstrable teaching skills. Research interests in any area of psychology will be considered, although the Department wishes to develop its known strengths in applied psychology, cognitive psychology and cognitive neuroscience. One post will be in developmental psychology. Candidates should be establishing, or have established, a strong track record in research. They will also have good teaching and dissemination skills, and a strong "collegial spirit". An international outlook would be an advantage. All members of academic staff in the Department have independent research programmes, although many choose to work in collaboration with one of the Department's successful research groups. They also contribute to administration and to teaching, either through the Department's Undergraduate Programme or through its Postgraduate School. All are expected to supervise student research projects. Informal enquiries may be addressed to Professor T Cox, E-mail: HoD at psychology.nottingham.ac.uk or to Professor G Underwood, E-mail: gju at psychology.nottingham.ac.uk CHAIR/READER The Department is seeking to appoint to two posts of Reader, although an appointment at Chair level may be considered. Candidates will be expected to provide leadership in their areas of research and teaching. Salary will be within either the Professorial scale, minimum 36,050 per annum, or within the Reader scale, 29,380 - 33,202 per annum, depending on experience. Further details and application forms are available from the Personnel Office, Highfield House, The University of Nottingham, University Park, Nottingham, NG7 2RD. Tel:0115 951 5205. Please quote ref. MCM/051. Closing date: 7 November 1997. LECTURERS (TWO POSTS) Salary will be within the range 16,045 - 27,985 per annum, depending on qualifications and experience. Please quote ref. MCM/051. Closing date: 7 November 1997 TEACHING FELLOW Salary will be within the range 16,045 - 17,606 per annum, depending on qualifications and experience. This post will be offered on a fixed-term contract for a period of three years. Please quote ref. SMM/181. Closing date: 7 November 1997 The lectureships and teaching fellowships could be decided in the light of the senior appointments. Further details and application forms are available from the Personnnel Office, Highfield House, The University of Nottingham, University Park, Nottingham, NG7 2RD. Tel: 0115 951 3262, Fax: 0115 951 5205. Please quote relevant reference number. ______________________________________________________ Geoffrey Underwood Professor of Cognitive Psychology Department of Psychology, University of Nottingham, Nottingham NG7 2RD, United Kingdom Phone: +44 115 951 5313 Fax: +44 115 951 5324 e-mail: gju at psychology.nottingham.ac.uk You want more? Then try: http://www.psychology.nottingham.ac.uk/staff/gju/gju.html ______________________________________________________ > From plb at cs.cmu.edu Thu Nov 20 15:17:18 1997 From: plb at cs.cmu.edu (Peter Brusilovsky) Date: Thu, 20 Nov 1997 15:17:18 -0500 Subject: CIRCLE seminars: a new CMU-LRDC seminar series on tutoring and related topics Message-ID: A new seminar series has begun, and you may want to subscribe to the distribution list in order to recieve announcements. The seminar series is sponsored by CIRCLE and involves talks on tutoring and related topics. The first speaker was Ray Mooney from the University of Texas, who spoke last week about a machine learning program that induced student bugs from their errors, and thus provided more focused tutoring that raised their gain scores relative to tutoring that was not guided by a student model. The next speaker is Reva Freedman from the University of Illinois, who will present results from a natural-language based tutoring system (see announcement below). If you would like to subscribe to the distribution list for announcements, send a message to Majordomo at list.pitt.edu with the following lines in the body: subscribe circle-announce help To find out more about CIRCLE (Center for Interdisciplinary Research on Constructive Learning Environments) see http://www.pitt.edu/~circle or contact Kurt VanLehn (vanlehn at cs.pitt.edu). -- Kurt VanLehn, Peter Brusilovsky, Dan Suthers LEARNING FROM TUTORS, LEARNING FROM TUTORING SYSTEMS Reva Freedman Research Associate Computer Science and Applied Mathematics Illinois Institute of Technology Nov. 24, 1997, noon, LRDC 2nd floor auditorium. The Circsim-Tutor project builds natural-language based tutoring systems for a causal domain (cardiovascular physiology). One of the goals of the Circsim-Tutor project is to generate conversations similar in both pedagogical strategy and language use to tutoring sessions conducted by expert human tutors. For this reason we have spent considerable time and energy on the analysis of transcripts, including a multi-level SGML-based markup of conversational goals. In this talk I will describe some of our recent results in this area. I will concentrate on the tutor's responses to unexpected utterances on the part of the student, including wrong answers, "near-misses", and "don't know" answers. These responses include several types of changes to the tutoring plan as well as a possible change in the level of the domain model being employed. Human tutors are generally not aware of the dialogue strategies they use at this level of detail. These results have influenced the design of the current version of Circsim-Tutor. Of course, the influence is bidirectional, since the model of mechanized tutoring used affects the phenomena one sees in the transcripts. I will discuss some issues in the design of dialogue-based tutoring systems, involving both the representation of pedagogical strategy and surface text, which this work has brought to light. The its-www list is handles by a common majordomo server at the U. of Pittsburgh. Here is the information how to work with it. To subscribe: ** DO IT NOW IF YOU DO NOT WANT TO MISS IMPORTANT INFO ** To send a message to its-www list: Send it to the address: its-www at list.pitt.edu To get help about using Majordomo server: send a message to Majordomo at list.pitt.edu with the following line in the body: help Dr. Peter Brusilovsky School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213, USA Phone 412 268 56 84 Fax 412 268 55 76 WWW http://www.contrib.andrew.cmu.edu/~plb/home.html From plb at cs.cmu.edu Tue Nov 25 18:49:34 1997 From: plb at cs.cmu.edu (Peter Brusilovsky) Date: Tue, 25 Nov 1997 18:49:34 -0500 Subject: CIRCLE seminar schedule Message-ID: Here is the CIRCLE seminar schedule for the end of 1997: Monday, December 8, 1997 Steven Ritter and Olga Medvedeva Department of Psychology Carnegie Mellon University "Reading the tutor's mind: Content authoring for an intelligent tutoring system" 12pm, LRDC 2nd floor auditorium (3939 O'Hara Street). Wednesday (!), December 17, 1997 Haruki Ueno Tokyo Denki University, Japan Knowledge-Based Program Understander: ALPUS - Concepts, Methodology and Applications. 12pm, Wean Hall 4625 (Carnegie Mellon Campus) (!) Regular seminar's day is Monday. This is a special arrangement. The next CIRCLE seminar speaker is Steven Ritter from the Department of Psychology, Carnegie Mellon University, who will discuss some issues of content authoring for intelligent tutoring systems and present pSAT, a WWW/Java-based content authoring tool (see announcement below). *** All further announcement of CIRCLE seminars will be distributed via circle-announce distribution list. *** If you would like to subscribe to the list, send email: To: majordomo at list.pitt.edu The subject line can be anything. The body of the message should begin with: subscribe circle-announce end The email address in must be *exactly* the same as the address you are sending from. This is for security (so that people can normally only subscribe themselves). The "end" is not needed if no signature is appended. To unsubscribe merely replace "subscribe" with "unsubscribe". To find out more about CIRCLE see http://www.pitt.edu/~circle or contact Kurt VanLehn (vanlehn at cs.pitt.edu). -------------------------------------------------- Reading the tutor's mind: Content authoring for an intelligent tutoring system Steven Ritter and Olga Medvedeva Department of Psychology Carnegie Mellon University December 8, 1997, 12 noon, LRDC 2nd floor auditorium. Abstract Intelligent tutoring systems are generative. Their instructional abilities can apply to a whole range of content, beyond that developed for the initial version of the system. For simple domains, where the system can completely and reliably solve all problems in the domain, this allows content to be added easily. But there are substantial challenges with domains in which the system cannot solve the problem on its own. The PAT Algebra system operates in such a domain: students are asked to read and understand word problems, but the system cannot be expected to possess natural language capabilities sufficient for the task. In such cases, the problem representation needs to be coded so that natural language understanding does not take place at runtime. The author's task is to enter the problem text along with a computer-understandable representation of the problem. New questions arise in the designing an authoring tool in this kind of domain: How can we ensure that the text and underlying problem representation remain "in sync"? How can we allow authors to create the underlying representation without this becoming a programming task? How can we communicate the meaning and consequences of the underlying representation to the author? How can the author correct misinterpretations? In this talk, we describe our approach to answering these questions for the pSAT (Problem Situation Authoring Tool) system. In addition, we discuss ideas for how exposing the tutor's representation (opening its "mind") can help teachers better understand the system, the domain and their students. Contact information Please, contact Steven Ritter directly (268-3498 or sritter+ at andrew.cmu.edu) Dr. Peter Brusilovsky School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213, USA Phone 412 268 56 84 Fax 412 268 55 76 WWW http://www.contrib.andrew.cmu.edu/~plb/home.html