Titles of related interest KIBBY COMPUTER ASSISTED LEARNING: Selected Proceedings from the CAL '89 Symposium KIBBY & MAYES COMPUTER ASSISTED LEARNING: Selected Proceedings from the CAL '87 Symposium SMITH COMPUTER ASSISTED LEARNING: Selected Proceedings from the CAL '83 Symposium SMITH COMPUTER ASSISTED LEARNING: Selected Proceedings from the CAL '81 Symposium Cover photograph: A Computers in Teaching Initiative Stand at CAL91 (photo: Michael R. Kibby) COMPUTER ASSISTED LEARNING Selected Contributions from the CAL91 Symposium 8-11 April 1991, Lancaster University Edited by MICHAEL R. KIBBY and J. ROGER HARTLEY PERGAMON PRESS OXFORD · NEW YORK · SEOUL TOKYO U.K. Pergamon Press pic, Headington Hill Hall, Oxford OX3 OBW, England U.S.A. Pergamon Press Inc., 395 Saw Mill River Road, Elmsford, NY 10523, U.S.A. KOREA Pergamon Press Korea, KPO Box 315, Seoul 110-603, Korea JAPAN Pergamon Press Japan, Tsunashima Building Annex, 3-20-12 Yushima, Bunkyo-ku, Tokyo 113, Japan Copyright © 1992 Pergamon Press pic All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means: electronic, electrostatic, magnetic tape, mechanical, photo- copying, recording or otherwise, without permission in writing from the publisher. First edition 1992 Library of Congress Cataloguing-in-Publication Data Symposium on Computer Assisted Learning (1991: Lancaster University) Computer assisted learning: selected contributions from the CAL91 Symposium, 8-11 April 1991, Lancaster University/edited by Michael R. Kibby and J. Roger Hartley. p. cm. Includes index. ISBN 0-08-041395-1: $110.00 1. Computer-assisted instruction—Congresses. I. Kibby, Michael. II. Hartley, J. Roger. III. Title. LB1028.5.S95 1991 371.3'34—dc20 91-42861 CIP ISBN 0 08 041395 1 Published as Volume 18, Number 1-3 of the journal Computers ά Education and supplied to subscribers as part of their 1992 subscription. Also available to non-subscribers. Typeset in Great Britain by BPCC Techset Ltd, Exeter. Printed by BPCC Wheatons Ltd, Exeter Computers Educ. Vol. 18, No. 1-3, p. ix, 1992 Pergamon Press pic. Printed in Great Britain PREFACE These Proceedings of the CAL series of biennial Symposia have a somewhat different origin to previous ones. The reason is that CAL91 had a somewhat different structure to its predecessors. The papers included here were not formally presented during the Symposium; however, they are all related to sessions held during the Symposium, reflect the contemporary concerns of participants and were refereed in the usual way. Some words of explanation regarding the content and stucture of CAL91 are necessary. Regarding content, we believed it was important to identify a strong restricted theme rather than allow contributions to range too widely. A suitable theme emerged from the view that after 20 years of mainly pragmatic experience of using information technology to support learning, the time had come for an emphasis on hard research evidence, to which could be added what was judged to be substantially innovative explorations. Certainly not reports on the 501st good way to teach genetics! We also believed it worth experimenting with a new structure. I remember some years ago being taught that one technique for checking the validity of a mathematical model was to examine its behaviour in extreme cases. Whilst many features of a model behave quite well in ways which seem to be reasonable in modest cases, it is rather complex to check general validity by such methods. Some quite useful and quick tests can be undertaken by examining the model's behaviour when it is pushed to the limit of some, even all, of the variables. Our approach to CAL91 was a bit like that. We questioned the value of conventional sessions during which authors speak about, sometimes even read, their papers for 15-20 min and then respond to questions for 5 min or less. There was also a concern about the lottery involved in selecting plenary speakers (and whether the Symposium Dinner was going to provide better or worse, and more expensive, food than on other evenings!). We decided to experiment with another model for symposia and took an alternative model to the extreme. In doing so we hoped that the rather detailed evaluation, which was undertaken independently of the Programme Committee (and is reported in this publication), would show the strengths and weaknesses of the model and would benefit our successors. Nothing ventured, nothing gained! After much debate, the comment of one member of the Committee gave us courage to stick to the innovative plan. We became convinced that participants would value a few substantive and interactive sessions rather than a larger number of passive, superficial ones. So the model for CAL91 emerged, helped by the structure of the space to be used at Lancaster. Over 300 participants (nearly 400 attended for at least part of the time) in groups of no more than 20 people implied more than 15 parallel sessions throughout. It also implied 100 organisers of sessions, mostly of half-a-day. Such a structure depended very strongly on session leaders and their ability to engage participants in interaction on the theme of the session. We expected that the mixture of seminars and workshops of usually three hours duration would allow substantive debate to take place. In order to assist participants in making difficult choices, almost all sessions were provided with a "shop-window" in the form of a poster display. The displays were live at the opening of the Symposium and at advertised times later and in the evenings. How successful where we? What lessons did we learn? Let me leave the reader to decide on the answers by reading the evaluators' report. Also, to wish the CAL93 Chairman with wisdom enough to keep most of the participants happy for most of the time! PROFESSOR R. LEWIS Chairman, CAL91 Programme Committee IX Computers Educ. Vol. 18, No. 1-3, p. x, 1992 Pergamon Press pic. Printed in Great Britain INTERNATIONAL PROGRAMME COMMITTEE Mike Aston, AUME, Hatfield Bob Lewis, ESRC InTER Programme Jonathan Briggs, Kingston Polytechnic David Martin, British Council Jonathan Darby, CTISS Oxford Peter McKenna, Dublin City University John Gillingham, Department of Employment Michele Norton, Pergamon Press Roger Hartley, Leeds University Gordon Reece, Bristol University Corinne Hermant, DELTA, Brussels Ken Tait, Leeds University Terry Hinton, Surrey University Vitor Duarte Teodoro, Universidade Nova de Mike Kibby, Strathclyde University Lisboa LANCASTER UNIVERSITY Maureen Boots, ESRC InTER Programme Barry Forde, Computing Service David Bradley, Department of Engineering Peter Goodyear, Department of Education Mark Bryson, ESRC InTER Programme Research Dorothy Callis, ESRC InTER Programme Dennis McCaldin, School of Creative Arts Jeanette Davies, Commercial & Industrial Michael Twidale, Department of Computer Development Bureau Science Computers Educ. Vol. 18, No. 1-3, p. x, 1992 Pergamon Press pic. Printed in Great Britain INTERNATIONAL PROGRAMME COMMITTEE Mike Aston, AUME, Hatfield Bob Lewis, ESRC InTER Programme Jonathan Briggs, Kingston Polytechnic David Martin, British Council Jonathan Darby, CTISS Oxford Peter McKenna, Dublin City University John Gillingham, Department of Employment Michele Norton, Pergamon Press Roger Hartley, Leeds University Gordon Reece, Bristol University Corinne Hermant, DELTA, Brussels Ken Tait, Leeds University Terry Hinton, Surrey University Vitor Duarte Teodoro, Universidade Nova de Mike Kibby, Strathclyde University Lisboa LANCASTER UNIVERSITY Maureen Boots, ESRC InTER Programme Barry Forde, Computing Service David Bradley, Department of Engineering Peter Goodyear, Department of Education Mark Bryson, ESRC InTER Programme Research Dorothy Callis, ESRC InTER Programme Dennis McCaldin, School of Creative Arts Jeanette Davies, Commercial & Industrial Michael Twidale, Department of Computer Development Bureau Science Computers Educ. Vol. 18, No. 1-3, pp. 1-9, 1992 0360-1315/92 $5.00 + 0.00 Printed in Great Britain Pergamon Press pic REASONING SUPPORTED BY COMPUTATIONAL TOOLS JOAN BLISS,1* JON OGBORN,2 RICHARD BOOHAN,2 JONATHAN BRIGGS,3 TIM BROSNAN,2 DEREK BROUGH,4 HARVEY MELLAR,2 ROB MILLER,4 CAROLINE NASH,2 CATHY RODGERS1 and BABIS SAKONIDIS1 1 King's College London, University of London, London WC2R 2LS,2 Institute of Education, University of London, 20 Bedford Way, London WC1H 0AL,3 Kingston Polytechnic, Penrhyn Road, Kingston upon Thames KT1 2EE and 4 Imperial College, University of London, South Kensington SW7 2AZ, England Abstract—This paper sets out the work of the Tools for Exploratory Learning Programme within the ESRC Initiative Information Technology in Education. The research examines young secondary children's reasoning with computational tools. We distinguish between exploratory and expressive modes of learning, that is, interaction with another's model and creation of one's own model, respectively. The research focuses on reasoning, rather than learning, along three dimensions: quantitative, qualitative, and semi-quantitative. It provides a 3 x 2 classification of tasks according to modes of learning and types of reasoning. Modelling tools were developed for the study and descriptions of these are given. The research examined children's reasoning with tools in all three dimensions looking more exhaustively at the semi-quantitative. Pupils worked either in an exploratory mode or an expressive mode on one of the following topics: Traffic, Health and Diet, and Shops and Profits. They spent 3-4 h individually with a researcher over 2 weeks, carrying out four different activities: reasoning without the computer; learning to manipulate first the computer then later the tool and finally carrying out a task with the modelling tool. Pupils were between 12 and 14 yr. Research questions both about children's reasoning when working with or creating models and about the nature of the tools used are discussed. Finally an analytic scheme is set out which describes the nature of the causal and non-causal reasoning observed together with some tentative results. INTRODUCTION The Tools for Exploratory Learning Programme is part of the ESRC Initiative, Information Technology in Education, and is a development of work in the London Mental Models Group (see Appendix). The aim of the programme is to study children's reasoning when they are interacting with different types of computer tool. Their reasoning is examined through a series of tasks in which learners are asked to work in one of two modes: either to model a situation for themselves or to explore another person's model of a situation. So we were concerned with looking at: (i) whether interaction with tools containing representations of a domain could facilitate reasoning in that domain; (ii) whether learners can be helped to reason about a domain by representing and exploring the consequences of their own ideas about the domain. The specific focus of the research is to examine how pupils reason in tasks using tools, which together require or permit quantitative, semi-quantitative or qualitative reasoning. We have chosen to study pupils in the age range 12-14 yr. An approach to such questions requires analysis of: learning and reasoning the characterisation of different dimensions of reasoning choice of, and distinction between, different types of tools the research strategy LEARNING AND REASONING In our original proposal for the research [1] we had categorised tools as being either exploratory or expressive. Such a definition, however, became too restrictive. The focus needed to be on types of learning with tools, thus we now distinguish between exploratory and expressive modes of learning. Tools can be used in either mode. The Exploratory mode permits pupils to investigate *Author for correspondence. 1 2 JOAN BLISS et al the views of a teacher or an adult about a given domain, views which will often be quite different from their own spontaneous ideas. The Expressive mode permits pupils to represent aspects of their own ideas about a domain, and to explore and reflect on these. In our earlier paper [1] we developed the characteristics of these two modes. It seems to us to be a valid criticism of many studies of learning that they do not allow enough time for significant learning to occur. Learning of the kind discussed above seems likely to need weeks or months rather than hours. Our constraints prevented us studying children over such a period of time. It therefore seemed clearer for us to limit the scope of the study to an investigation of children's reasoning with tools. We have chosen to look at reasoning which would seem to be necessary to learning, and which can be seen as of value in itself, rather than to claim to be investigating learning. Our research is thus based on specially constructed tasks designed to elicit and encourage kinds of reasoning which are relevant to learning. We have distinguished three types of reasoning with tools: quantitative, semi-quantitative and qualitative. A given task is designed to elicit one of these types of reasoning, although of course other forms of reasoning may be used spontaneously by pupils in the task. Thus when we describe a type of reasoning we are describing a task-tool interaction, in which the tool contains or allows the expression of, for example, a semi-quantitative model of a situation, and in which the cognitive demand of the task also requires the learner to use this same type of reasoning. We have created, therefore, three types of task-tool combinations which permit these three kinds of reasoning in both exploratory and the expressive modes. We now give a brief description of the three kinds of reasoning. DIMENSIONS OF REASONING Quantitative reasoning Quantitative reasoning can involve a variety of aspects, from recognising simple numerical relationships, through working with sets of numbers and comparing sizes and magnitudes, to manipulating algebraic relationships. The problem may be to know how changing a quantity by a given amount will affect another quantity or quantities. Thus if the population close to the only large supermarket in a given area doubles, how may this affect queuing times in the store? Other quantitative problems can include questions about possible values variables can take, given constraints upon some of them (how can the price of a meal made from a range of ingredients vary if its nutritional value is to be more than a given minimum?), or they can be dynamic problems about the evolution of a system (how does the number of bacteria in yoghurt vary with time?). Our task-tool situations are limited to quantitative reasoning about variables linked by simple algebraic relationships ( + , x , —, /), providing means for constructing and manipulating algebraic relationships between variables. Qualitative reasoning Qualitative reasoning involves making categorical distinctions and decisions. Thus it may require considering a set of choices or decisions and taking into account their consequences or, given a certain goal, formulating what is necessary to reach that goal. It may require noting and taking account of alternatives, weighing up evidence, or considering if a certain condition is realised then what follows, etc. In our qualitative task-tool combinations the reasoning concerns problematic situations, the actions which are possible in each situation, and the further situations and associated actions to which these might lead. Semi-quantitative reasoning It seemed to us that these two distinctions did not capture some essential aspects of reasoning, in particular reasoning in which the direction but not the size of effects of one part of a system on another is known. For example, Piaget's work [2] on causality showed that young pupils were capable of apprehending problems in this manner when quantitative methods eluded them. Recent work in cognitive science, for example, on mental models [3] indicates the importance of what is often called "qualitative" but is actually semi-quantitative reasoning: thus this type of reasoning Computer-supported reasoning 3 involves seeing how in a complex system the rough and ready size of something has an eifect on the rough and ready size of something else, which may in turn affect other things and might in the end feed back to affect the first quantity. CHOICE OF TOOLS The dimensions exploratory/expressive and quantitative, semi-quantitative, and qualitative produce a 3 x 2 matrix which allows a task-tool combination to be characterised (see our 1988 paper [1]). While some tools can be used in either mode of learning they usually require a specific type of reasoning. Quantitative tools are probably the most widely used. Those used in exploratory mode include simulations, while those able to be used in expressive mode include spreadsheets and modelling systems. Qualitative tools usable in exploratory mode include many expert systems, decision games, logic programs, data bases, and some simulations. Qualitative tools usable in an expressive mode include story-making programs, adventure game shells, data base shells and expert system shells. Semi-quantitative tools are less known. Graphic structures for semi-quantitative models can be expressed in the modelling system STELLA but the tool itself is essentially quantitative. The Alternative Reality Kit (ARK) is another tool of this kind, but not one which is easily accessible. For exploratory mode we note that a number of simulations, in which the user askes for "more" or for "less" of (for example) water or a pesticide, have a semi-quantitative flavour. This flavour is usually introduced to simplify the simulation, but can be seen as valuable in focusing attention on the essentials of the quantitative relations without involving the complexities of the exact relationships. Developments in hardware and software currently change faster than research can deliver results. For this reason, beginning in 1988, we chose to develop or experiment with software which might only be available to schools later, nearer the time at which our results and software would be ready. We chose to work with Apple Macintosh machines, in part because of the possibilities they offered for investigating tools with direct manipulation facilities, given the evidence that directly manipulating the icons representing primitives to construct models was likely to help children to learn the tool quickly and to grasp its nature. The research programme is not intended to be either tool or task driven. We chose one tool for use in each of the areas of reasoning: quantitative, semi-quantitative and qualitative; these tools being designed to be able to be used in both exploratory and expressive tasks. Quantitative tool For quantitative tasks we developed a prototype tool which allows models to be built which use simple algebraic relations between variables. Quantities can be combined by addition, subtraction, multiplication or division, with consistency of units being partially provided for. Quantities can be given maximum and minimum values, so that variables can be altered in steps between these values by moving a graphic slider attached to the box representing the variable. If an independent variable is altered, the tool shows which variables will be affected, while calculating them. Graphs of values of any one dependent variable against any one independent variable can be obtained, thus enabling exploration of sections across the "space" of values of variables. In essence the tool is much like the Algebraic Proposer [4]. However, by using much of the code and facilities of our other tool IQON (see below), it works much more by direct manipulation. A HyperCard version was also used in some of the empirical work. Semi-quantitative tool No suitable practical semi-quantitative tool existed when we began the research. One source of ideas was the notion of causal loop diagrams, which form part of the metaphor used in STELLA and of the system dynamics thinking behind it [5]. Another more general source was thinking in Artificial Intelligence about qualitative reasoning about processes of causal change, as described by, for example, Kuipers[6], Forbus[7] and de Kleer and Brown [8]. 4 JOAN BLISS et al. A tool (IQON) was developed in Smalltalk for use with semi-quantitative reasoning tasks. IQON allows the user to represent a system in terms of interacting variables, specifying the relations between them. Variables are depicted as boxes and relations between them as arrows linking one box to another. Figure 1 gives an example. All variables such as "nomad population" take semi-quantitative "values" of "above or below normal". A variable which is above or below normal drives another variable gradually up or down, depending if the first is linked to the second "positively" or "negatively". Thus in Fig. 1 a high level of "modern medicine" has driven down the variable "amount of disease". In turn, a low level of "amount of disease" has driven up "nomad population". The strength of the links affecting any one box may be varied relative to one another. As can be seen in Fig. 1, relationships can be quite complex, including interdependent variables such as cattle population and amount of grassland, or nomad population and food available. The whole diagram of boxes and links is animated, so that levels of variables can be watched rising and falling in response to each other's values. Qualitative tool We explored the possibilities of an expert system shell, Knowledge Pad [9], and a node and link story builder, LINX88[10], both for IBM machines, and built a HyperCard version of LINX. It proved more difficult than we expected to provide them with an appropriate content for use in exploratory tasks. The final qualitative tool "Explore your options" was a simplified and specialised HyperCard version of LINX. It provides for building a decision structure or tree, out of situations and a set of actions associated with each situation, each of which can lead to another situation. The tree is drawn graphically on screen. Situations and actions are entered as text, and no computation is done on them. The system maintains the tree and displays situations and actions as they are selected. Actions once taken can be undone, returning to an earlier situation to explore other courses of action. Thus in a task pupils are asked to think about a situation and to consider the possible actions allowed in that situation. They can look at all possible actions before deciding which to explore, or can follow up just one action. It is possible for some of the alternative, different paths (arising from the initial situation) to converge on the same end point, whilst other paths give different results. Thus a task presents a starting situation, and a final goal, and the pupil has to consider alternatives among a set of coherent choices for reaching that goal. RESEARCH STRATEGY Research Questions As mentioned previously, our two initial questions were: In what way can interactions with tools containing representations of a domain facilitate reasoning in that domain? Are pupils helped with their reasoning by representing and exploring the consequences of their own mental models of a domain? These questions are elaborated in greater detail below. Elements from which models are built. To what extent do pupils understand, think about and use the primitive modelling elements provided in the system (e.g. variables and links)? What kinds of interacting entities and connections do they spontaneously use? Seeing a model as a whole. To what extent do pupils see models they are given, or the models they make themselves, as complex sets of interacting entities, with a structure seen as a whole? Relationship between model and reality. To what extent do pupils treat a model as a formal structure with its own necessary behaviour? In what ways do they see the relationship between the model and the real world: for example is the model seen as predicting what will happen, or is the known behaviour of the world seen as testing the model?