From tkelley at arl.mil Mon Jun 7 16:43:52 1999 From: tkelley at arl.mil (Troy Kelley) Date: Mon, 07 Jun 1999 15:43:52 -0500 Subject: Error modeling Message-ID: I would like to hear the group's thoughts on this. Don Norman has noted that there are basically two types of errors. Slips and mistakes. Slips are errors where you, "intend to do one action, but do another" and mistakes, "result from the choice of inappropriate goals". Norman notes that there are many types of slips: capture errors, description errors, data-driven errors, associative activation errors, loss-of-activation errors, and mode errors. In ACT-R there seem to be two primary ways to produce errors: errors of commission and errors of omission. It would seem that both of these map nicely onto what Norman has called "slips". An error of commission maps nicely onto a type of slip called associative activation errors and errors of omission map somewhat nicely onto loss-of-activation errors. I am wondering how what Norman calls "mistakes" map to the types of errors that can be simulated in ACT-R? It would seem like "mistakes" would result from having inappropriate chunks in declarative memory, but I would like to hear some other thoughts on this. Also, has there been any work in representing schemes in ACT-R? An inappropriate scheme might produce the type of errors Norman has identified as "mistakes". Troy From cl at andrew.cmu.edu Mon Jun 7 17:00:21 1999 From: cl at andrew.cmu.edu (Christian Lebiere) Date: Mon, 07 Jun 1999 17:00:21 -0400 Subject: Error modeling Message-ID: > I am wondering how what Norman calls "mistakes" map to the types of > errors that can be simulated in ACT-R? It would seem like "mistakes" > would result from having inappropriate chunks in declarative memory, but > I would like to hear some other thoughts on this. I guess that "slips" have preoccupied us more because they are really architectural errors. "Mistakes", on the other hand, are knowledge errors whic are primarily task- and subject-dependent. Of course, incorrect knowledge can be represented and used in ACT-R, but it usually doesn't have much bearing on the architecture. One relation between "slips" and "mistakes" is developed in chapter 9 of the book (pp. 322 and following) and at further lengths in my thesis: http://reports-archive.adm.cs.cmu.edu/anon/1998/abstracts/98-186.html A "slip" resulting from the partial matching of a correct fact in memory (e.g. retrieving 3+5=8 for 3+4=?) results in an erroneous fact being added to declarative memory (3+4=8) which can in turn be retrieved as a "mistake". > Also, has there been any work in representing schemes in ACT-R? An > inappropriate scheme might produce the type of errors Norman has > identified as "mistakes". Eric Scott presented at the 1998 ACT-R workshop a clever implementation of schemas in ACT-R: http://act.psy.cmu.edu/ACT/ftp/workshop/Workshop-98/Scott/quick_index.html Although it does not explicitely mention errors, it seems to have some potential for your application. Christian From maclaren+ at andrew.cmu.edu Mon Jun 7 17:41:56 1999 From: maclaren+ at andrew.cmu.edu (Benjamin A Maclaren) Date: Mon, 7 Jun 1999 17:41:56 -0400 (EDT) Subject: Error modeling Message-ID: > It would seem like "mistakes" would result from having inappropriate chunks in declarative > memory, but I would like to hear some other thoughts on this. There are two ways you might want to think about mistakes. Comprehension and execution "mistakes." You seem to be referring to comprehension problems. But you could also have productions that perform incorrect actions from the right declarative memory, either because the production has the wrong action side, or possibly an overly general lefthand side (this of course could also be modeled with partial matching). In our model of early algebra we have made all the mistakes the result of overly general lefthand sides or productions failing to fire. Ben From gray at gmu.edu Mon Jun 7 17:43:12 1999 From: gray at gmu.edu (Wayne Gray) Date: Mon, 7 Jun 1999 17:43:12 -0400 Subject: Error modeling Message-ID: I'm not convinced that most of the issues surrounding errors are act-r issues. There are definitely act-r issues there; however, you need a fairly deep cognitive task analysis before you can separate these out. You might be interested in a recent paper of mine: Gray, W. D. (in press). The nature and processing of errors in interactive behavior. Cognitive Science. Preprints are available via: http://hfac.gmu.edu/Publications/pubs.html _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/ 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 3f5 * Work is infinite, * Fairfax, VA 22030-4444 * time is finite, * http://hfac.gmu.edu * plan accordingly. * _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/ From altmann at osf1.gmu.edu Tue Jun 8 10:45:01 1999 From: altmann at osf1.gmu.edu (ERIK M. ALTMANN) Date: Tue, 8 Jun 1999 10:45:01 -0400 (EDT) Subject: Error modeling Message-ID: > Don Norman has noted that there are basically two types of errors. > Slips and mistakes. Slips are errors where you, "intend to do one > action, but do another" and mistakes, "result from the choice of > inappropriate goals". Norman notes that there are many types of slips: > capture errors, description errors, data-driven errors, associative > activation errors, loss-of-activation errors, and mode errors. > > In ACT-R there seem to be two primary ways to produce errors: errors of > commission and errors of omission. It would seem that both of these map > nicely onto what Norman has called "slips". An error of commission maps > nicely onto a type of slip called associative activation errors and > errors of omission map somewhat nicely onto loss-of-activation errors. > > I am wondering how what Norman calls "mistakes" map to the types of > errors that can be simulated in ACT-R? It would seem like "mistakes" > would result from having inappropriate chunks in declarative memory, but > I would like to hear some other thoughts on this. If you assume that it's always possible to retrieve an incorrect chunk, then another contrast lies in the timing of the incorrect retrieval. If it occurs when cognition is looking for a new goal, this will produce a "mistake" -- an action based on an incorrect goal. If it occurs in service of the current goal, this might be a "slip". Erik. ----------------------- Erik M. Altmann Psychology 2E5 George Mason University Fairfax, VA 22030 703-993-1326 hfac.gmu.edu/~altmann ----------------------- From trafton at itd.nrl.navy.mil Thu Jun 10 10:11:36 1999 From: trafton at itd.nrl.navy.mil (Greg Trafton) Date: Thu, 10 Jun 1999 10:11:36 -0400 (EDT) Subject: Goals and goalstack talks at ACT-R workshop Message-ID: Hi, all. I'm organizing a group of papers on goals and the architecural goalstack to be presented together at the ACT-R workshop. The format will be the same as it always is (people giving talks); we'll just have all the papers together. In addition, we'll have a panel at the end to talk about current ideas, theories, and views of the goalstack. If you have a model or experiment that is relevant to any aspect of goals or goalstacks, we'd love to include you in the group. Currently the group consists of: Erik Altmann John Anderson/Scott Douglass Greg Trafton Please respond to me if you're interested in being part of this group of talks (as well as submitting the normal abstract). thanks, greg -- Greg Trafton (trafton at itd.nrl.navy.mil) http://www.aic.nrl.navy.mil/~trafton/ From Stellan at tigger.cc.uic.edu Mon Jun 21 11:29:59 1999 From: Stellan at tigger.cc.uic.edu (Dr. Stellan Ohlsson) Date: Mon, 21 Jun 1999 09:29:59 -0600 Subject: Error modeling Message-ID: Just returned from a trip and found a brief exchange on my machine about the modeling of errors, particularly mapping Norman's error categories onto ACT-R architectural processes. However, there is another way to look at performance errrors: They are ALL commission errors, by necessity. The reason for this is that behavior never stops; the architecture is always doing *something* in the next unit of time, so if it should have done X but doesn't, it is certain to be doing Y instead; hence every omission error is an commission error. Furthermore, the general reason why we do Y when we should have done X (given that we are at all *capable* of retrieving/doing X), is that the triggering conditions for Y are overly general. If the triggering/production conditions for Y were specified with perfect accuracy, then Y would not, by definition, be considered in a situation in which we should be doing X. It is because the application of Y is underspecified that gives Y the opportunity to compete with X and, sometimes, win. This view of error provides a single, unified account that covers both omission and commission errors. If anyone is interested in pursuing this idea in detail, the relevant references are: Ohlsson, S. (1996). Learning from performance errors. Psychological Review, 103, pp. 241-262.Times Ohlsson, S. (1996). Learning from error and the design of task environments. International Journal of Educational Research, 25(5), 419-448.Times -Stellan Ohlsson -------------------------------------------------------------------------------- Achieving a goal requires work; solving a problem requires intelligence; seeing the truth requires imagination. Stellan Ohlsson Phone: (312) 996-6643 Chair, Cognitive Division Fax: 312-413-4122 Department of Psychology (M/C 285) Email: stellan at uic.edu University of Illinois at Chicago (UIC) 1007 West Harrison Street Chicago, IL 60607-7513 Web page: http://www.uic.edu/depts/psch/ohlson-1.html