The Role of Dialogue in Computer-Based Learning and Observing Learning: an evolutionary approach to theory

: This paper examines two sides of a coin that relate to learning from dialogue. The first side of our coin relates to the role of dialogue in learning; the second side is related to the part that observations of learning can play in the design of computer-based learning environments. In order to define the scope of the paper two complementary research question are examined. The first question is how and why does one learn from dialogue? The second question is how, or to what extent, can theories and studies of dialogue and interaction be exploited in a concrete way by designers of interactive media for education? Following a review of related literature, we investigate the above questions by drawing on a useful, if somewhat simplified, conception of the role of theory and models in learning technology development. There are three aspects to what we are terming an evolutionary approach to learning technology theory: (i) theories/models of learning, (ii) empirical observations of learning, and (iii) interactive learning environment design and implementation. The purpose of this evolutionary approach is the mapping out of not a specific theory, but how people are working towards the creation of theories. The evolutionary approach involves a constant process which slowly takes the educational technology field forward in iterative steps. In order to concretize our evolutionary approach we examine the work of selected researchers, in the field of dialogue in learning, in the context of the identified three points of evolution. We conclude by suggesting that our evolutionary model can help designers of, and researchers into, learning technology in various important ways.


Introduction
This paper examines two issues that relate to learning from dialogue.The first issue concerns the role of dialogue in learning.The second issue pertains to the role that observations of learning can play in the design of computer-based learning environments.Two research questions are examined in order to facilitate our investigation.First, how and why does one learn from dialogue?Second, how, or to what extent, can studies of dialogue and interaction be exploited in a concrete way by designers of interactive media in education?What this paper does not try to put forward is one theoretical basis for learning technologies.The purpose of the evolutionary approach presented below is, rather, to provide a primarily descriptive tool that enables us to map out how different researchers are working towards the creation of theories.Our evolutionary approach is a useful, if somewhat simplified, conception of the role of theory in learning technology development.Following a literature review in the next two sections, we focus in on the relationships between the three points in what we are calling an evolutionary approach to new media theorising: (i) theories/models of learning, (ii) empirical observations of learning and (iii) learning environment implementations.The evolutionary approach involves a constant process which slowly takes the field forward in iterative steps.Our perspective is that no big bang or revolution is imminent with respect to theorising and modelling.Instead, we advocate cycles around the three points of evolution; this will, we suggest, bring about the gradual development of the field of learning technology, allowing us to adapt theories and models to suit our own perspectives.Before examining our evolutionary approach, however, in the next two sections we examine the literature that is relevant to the research questions outlined above.

How and why does one learn from dialogue?
Dialogue between teachers and students may be important in promoting learning (e.g.Vygotsky, 1978;Leontiev, 1975;Lipman, 1991;Jones and Mercer, 1993;Freire, 1993;Pilkington and Mallen, 1996).But exactly how and why does one learn from dialogue?In order to structure this section we will look at dialogue by viewing it as potentially belonging to one of three levels: (i) rote-learning, question and answer, (ii) adaptive dialogue, which mediates between learner and resource and (iii) responsive dialogue, i.e. dialogue about the aim and structure of the educational experience.The major focus of this section is the adaptive dialogue layer.

Rote-learning dialogue layer
Computers can support dialogue at the lowest of our layers, although this is an area that is often ignored.Freire (1993) introduces, and dislikes, the 'Banking' concept, which is conceived as a teacher full of knowledge depositing knowledge in the students through a process of transmission.Banking is similar to the concept of knowledge communication (Wenger, 1987), where knowledge is seen as being transmitted to the learner in a pre-determined format, to be absorbed by the learner.For Freire, dialogue is a horizontal relationship in which one individual is with the other, i.e. the adaptive dialogue layer discussed below.It is positive, hopeful, trusting and critical.It involves two-way communication.Transmission, on the other hand, is a vertical relationship in which one person is higher than the other.To borrow Freire's words, transmission is loveless, arrogant, hopeless, mistrustful and acritical.Transmission does not communicate but rather, issues communicate.
An example of the rote-learning dialogue layer would be drill-and-practice software.These systems perform a useful role in that they can automate assessment in memory recall quizzes.This layer is not the focus of our paper.

Adaptive dialogue layer
In this sub-section we describe the way that a teacher mediates between learner and learning resource, i.e. the adaptive dialogue layer.Below we will also discuss the possibility of automating the middle layer.
Students who are placed in a learning environment will usually need to interact with a teacher or learning facilitator at some point, in order to receive guidance (Elsom- Cook, 1990), feedback and explanations.The adaptive role of a teacher is of central importance to learning (Laurillard, 1993) because learning resources and media, such as books, journals, CD-ROMS, online databases or World Wide Web resources, etc., are rarely able to adapt to a particular group or individual's learning requirement.Students bring different histories of learning with them to a particular situation and therefore have different learning needs (Laurillard, 1993;Ramsden, 1992).Furthermore, these resources and media, typically, do not provide guidance on how they should be integrated and embedded in a coherent fashion so that learning can occur.For example, the tutor may be required to mediate between the learner and their understanding of the way in which they should use learning resources in order to meet the assessed learning outcomes of a particular programme of study.Consequently, in a learning environment, we get a complex set of relationships between how a learner thinks, i.e. cognition, how the learner interacts with teachers and peers, and the various media and resources that are available to support learning.The institution and society in which the learning takes place will also exert an influence on learning in more subtle ways.
The teacher is often more than a source of information.As was pointed out above, the teacher plays a key role in mediating a student's learning, acting as a kind of go-between or guide for the learner as they engage with the various elements of the learning environment, i.e. as they engage with other learners and tutors, learning resources and media, programme learning outcomes and assessment methods (Knight, 1995).The teacher can also help the learner to become more autonomous, to learn how to learn, and to reflect on his or her own problem-solving.The way that such explanation and guidance is provided by a teacher is usually through dialogue, either face-to-face, written or virtual, since this enables the teacher's help to be adapted and individualised to a particular student's needs.Dialogue also enables the student to verbalise and articulate his or her needs and understanding.This latter process of making knowledge explicit, and reflecting on it may itself be an interactive learning mechanism (e.g.Chi, Bassok, Lewis, Reimann and Glaser, 1989).Providing computer-based learning support that is able to acquire aspects of the role of 'teacher as mediator' is a growing area of research and development.However, in this paper what we are claiming is that, although there are many references in the literature on interactive learning mechanisms as they relate to computer-based learning (e.g.van Joolingan and de Jong, 1991;Baker, 1994;Baker and Bielaczyc, 1995), we still do not have sufficient detailed knowledge concerning the relationships between theory, empirical work and implementation of learning environments.
In contrast, Constructivism (described by Wasson, 1996) sees the major goal of education as the creation of rich sets of cognitive tools to help learners explore and interact with their environment and is closely associated with Piaget's (1971) genetic epistemology theory of cognitive development.Papert's (1980) Turtle Logo is a classic example of a learning environment that attempts to embody cognitively relevant tools in the environment.In the case of Logo, the cognitive tools, or cognitive hooks as Papert called them, are claimed to be the graphical immediacy of geometry drawn in realtime.These cognitive hooks are intended for the young learner to use as a tool to enhance the motor skills which they have acquired from birth.Cognitive tools are generalisable tools used to engage learners in meaningful cognitive processing, knowledge construction and facilitation.For example, computer-based cognitive tools are in effect cognitive amplification tools that are part of the environment.Environments that employ cognitive tools are described as distributing cognition; they are constructivist because they actively engage learners in the creation of knowledge that reflects their comprehension and conception of the information rather than focusing on the presentation or transmission of objective knowledge.It is this last item that contrasts with the behavioural approach (see, Hartley, 1998, for a description) which would focus on content selection, sequencing, structuring and presentation.
Taking a different perspective, Jones and Mercer (1993) have argued that a theory of learning, e.g.Behaviourism or Constructivism, is not the best framework for analysing what goes on in understanding the use of media like computers in education, rather a theory and analysis of teachingand-learning is needed.The evolutionary framework, described in this paper, takes a similar approach, drawing as it does on theory and analysis of teaching and learning.Jones and Mercer are in favour of approaches to understanding teaching and learning that have been based on Vygotsky's culturalhistorical theory of human activity.For Vygotsky (1978), human mental functions appear first as interindividual and then intra-individual, that is, by the use of socially developed tools, both technological and psychological ones.For Vygotsky, however, the unit of analysis was still the mediated action of an individual and how that individual developed.Vygotsky also put forward the concept of the zone of proximal development (ZPD), which is the difference between a learner's real level of development and their potential level of development.It is for the previously stated reasons that we place Vygotsky in our adaptive layer, although there may be an argument for also placing him in the responsive dialogue layer (such a discussion is, however, beyond the scope of this paper).
Recent work with computer-based simulations (Twigger, Byard, Draper, Driver, Hartley, Hennessy, Mallen, Mohammed, O'Malley, O'Shea and Scanlon, 1991;van Joolingan and de Jong, 1991), which are used to help students acquire explanatory accounts of the real world, shows that students may fail to generate deep causal models of the behaviour under simulation because they concentrate on manipulating the simulation objects.With respect to the previously stated finding, Pilkington and Mallen (1996) make a strong case for a more Vygotskian (1978) perspective in interaction, i.e.where the teacher mediates knowledge about the society and culture so that it can be internalised by the learner.Interestingly this raises the following question: is this knowledge already formed and finding its way into the learner?If so, this would suggest Freire's Banking model of education.In fact we do not mean to suggest this interpretation.In our interpretation, interaction is seen as an important component of the learning environment, helping students to recognise and resolve inconsistency, i.e. it has an adaptive mediating role.Furthermore, Pilkington and Mallen have also point out that: …if we are to improve the quality of the interaction, then we need to understand the mechanisms by which dialogues work ... We need to know how and why, some kinds of dialogue … seem able to trigger reflective engagement and conceptual change.(Pilkington and Mallen, 1996, p. 213-4) Recently, some researchers have suggested that dialogue with a teacher may be required if the goal is to promote reflection and conceptual change: … self-reflection, or even reflective discussion between students may not be effective in changing beliefs and their 'organisation' into conception.This requires dialogue with a teacher.But … can a computer system be improved/designed to assist the reflective process, and if so, what are the requirements of its improvements?(Hartley and Ravenscroft, 1993, p. 3).
The above researchers (Hartley and Ravenscroft, 1993) go on to describe a system called SCILAB, which was designed to explore one approach, for the domain of science, to providing dialogue that encourages reflection.Lipman (1991) has proposed an approach to learning through dialogue and has suggested that we must stipulate that education should include reasoning and judgement about knowledge.Education in the Lipman sense of the word is not 'simply' learning, it is a Vygotskian-like teacher-guided community of inquiry that places an emphasis on social interaction and cooperative learning.Lipman calls this the reflective model of education practice.As we will see in subsequent sections, Lipman's work has been influential on the author's own work in the area of promoting learning through dialogue.

Responsive dialogue layer
To be truly equal and transformative, dialogue should not just be about content or about making appropriate use of a learning environment; it has to extend to the choice of what is to be learned, decisions about how it is to be learned and even institutional questions (Moore and Kearsley, 1996).This relates to the way that some distance learning writers talk about transactional distance: the perceived degree of separation during interaction between and among students and teachers.Moore and Kearsley (1996) describe transactional distance as having two components: dialogue and structure.Dialogue refers to communication between students and their teacher, i.e. our adaptive dialogue layer, and structure refers to the "responsiveness" of the educational plan to the individual student.By educational plan we refer to an orchestrated learning environment, e.g. a module that draws upon problem-based learning or an institution that is based on a particular educational school of thought.The educational plan can thus include theory and models of learning and interaction.Leontiev (1975) expanded Vygotsky's cultural-historical theory to an activity theory approach to human interaction where reality consists of mediated, social, hierarchically organised, developing, internal and external, object-oriented activities.For Leontiev the unit of analysis was extended to include the collective activity, something done by the community with a motive (which need not be consciously recognised), which is composed of individual actions directed towards a goal.The individual's mediated actions could still be analysed, but there was now a social dimension which could be used to understand the individual's actions.It is for this reason, and the recognition of a motive, that we place Leontiev in the responsive dialogue layer.In section 4 we will present an approach examining this highest layer of dialogue.
To conclude this section, we suggest that in order to improve our understanding of learning in the context of the use of new media like computers, we need to link theory to the analysis of teaching-andlearning interactions.Furthermore, it would appear that certain types of learning may not occur unless dialogue takes place between a tutor and learner(s).Interaction has an adaptive mediating role, helping students to recognise and resolve inconsistency.To be truly equal and transformative, dialogue has to be responsive to institutional questions and educational plans.

System design based on studies of human communicative interaction
In computer-based learning there is very little work that is based on dialogue analysis.The work that has been done tends to examine students' interactions with existing computer-based systems (e.g.Pask, 1976;Recker, 1994;Pilkington and Parker-Jones, 1996).Laurillard (1993, p. 102) has, however, proposed a template for conversations that aims to map out, at a very high level of abstraction, the steps that are required for the design of interactive and adaptive media.However, Hartley (1998) has pointed out that although the applications of technology in education are becoming more numerous, they tend to be "disparate, pragmatically oriented, and largely descriptive in the accounts they present" (Hartley, 1998, p. 20), and that we still need systematic development frameworks that are able to "link theories to methodologies and practice" (Hartley, 1998, p. 36).
The idea that we can somehow base system design on a study of dialogues is a separate concern to building systems that promote dialogue, although the former may lead to the latter.The analysis of communicative interactions may lead to important insights that guide interactive media development on a philosophical and theoretical level.Alternatively, the results may be used to suggest useful tutoring strategies that can be used in a particular learning situation.However, educational research on interactions has tended to focus on a level of analysis and description that is of limited value for the types of models and theories that we wish to construct and use as the basis of learning environment design.This level of description claim does not suppose that educational research is, or has been, carried out at the wrong level of detail.Rather, the claim is that the gap between the level at which educational research is conducted and the fine-grained detail required for learning technology approaches has, up to the present, been too great to be bridged.Support for this claim can be found in the literature: ...  Cook, 1991, pp. 76-77) Over a quarter of a century ago the designers of the WHY system attempted to formalise the Socratic method for tutoring about the rainfall processes on the basis of a study of human tutoring (Stevens, Collins and Goldin, 1982).More recently, the AutoTutor system has been designed to assist college students on a computer literacy course (Graesser, Wiemer-Hastings, Wiemer-Hastings, Kreuz & the Tutoring Research Group, 1999).AutoTutor uses an analysis of human tutors as the basis for its dialogue moves and discourse patterns in a curriculum script.The AutoTutor researchers attempt to use speech act theory (Austin, 1962;Searle, 1969) as the basis for their system's dialogue planning.MetaMuse (Cook, 2001) is a system that attempts to promote a Lipman-like community of inquiry (Lipman, 1991) in the context of undergraduate musical composition.MetaMuse is based on a theoretical and dialogue analysis approach (Cook, 1998) that makes used of higher-level, goal-based interaction analysis and communicative act theory.
In this section we have suggested that further work is needed that explores the systematic relationships between theories and models, empirical work of a fine-grained nature and the implementation of learning environments.Although some work has been done in area, we further claim that there is a need for a clearer mapping out of the problem space, both in descriptive terms and from an analytical perspective.Such a mapping exercise should enable us to draw conclusions and take the field of learning technology forward.In the next section we investigate the research questions presented in the introduction by drawing on a simplified conception of the role of theory and models in learning technology development.

Evolutionary approach to theory
In this section we present an evolutionary approach to theorising about what we called, in Section 2, the adaptive and responsive dialogue layers.Specifically, the purpose of this evolutionary approach is the mapping out of not a specific theory, but a mapping out of how different researchers are working towards the creation of theories.
A model of an educational process, with its attendant theory, can be used to form the basis for the design of a computer tool for education (Baker, 2000).For example, Baker and Lund (1997) describe a model of task-oriented dialogue that forms the basis of design and implementation of tools for computer-mediated communication between learners and teachers in a computer-supported collaborative learning environment.If we accept Baker's (2000) argument that models are not, by their nature, necessarily computational, this opens up a wide range of possible ways in which theories and models can form the bases of design of educational artefacts.As Baker (2000) also points out, what is required of such an endeavor is that the specific nature of the relations between theory, model, corpus (i.e.transcriptions of interaction data), and design of learning environments be made as explicit as possible as legitimate objects of scientific discussion and as means of generalising findings towards redesign.The author's previous work (Cook, 1998;Cook, 2001) describes precisely such a principled relation for the case of a pedagogical agent for learning musical composition.This previous work by the author, which is summarised in Figure 1, explored the systematic relationships involved when moving from theory, i.e. the Knowledge Mentoring framework at the top of Figure 1, to an analysis of corpus data (Cook, 1998).Briefly, the Knowledge Mentoring framework is a theoretical framework of mentoring which includes the following sub-components: (i) categories of goals drawn from theory (Vygotsky and Lipman, who were described above), (ii) a three level analysis framework of goals, subgoals and communicative acts, (iii) a theoretical model of teaching agents (values, wants, commitment, intention and an action cycle).The categories and the three level framework were then used to guide the analysis of empirical data and to thus generate various results.Figure 1 also shows that the empirical findings were in turn used in the design of interactive system that is able promote learning through dialogue, i.e.MetaMuse in the bottom right of Figure 1.The dotted line in Figure 1 means that no explicit link exists between the two evolutionary points for the work under examination.We have argued (Cook, 2001) that the use of human expertise, when modulated, rather than transferred, to the computational medium, is an appropriate starting point for interactive media design; this argument is represented on the bottom line of Figure 1.By modulated we refer to the ability to pass from one state into another using a logical progression.In our own work the initial state was the corpus of data that resulted from the observation of learning-teaching interactions.The target state was the incorporation of this data into a pedagogical agent design.A logical progression, i.e. the modulation, in our context involved the use of dialogue analysis and modelling techniques to enable aspects of interaction data to be converted into a computational model that was used in a learning support system.We were not attempting to transfer human expertise, which would involve attempts to computationally simulate, in a cognitive science sense, the human teacher.Instead, our goal was to modulate interaction data into the design of a computer-based pedagogical agent.
We now provide further explanation of the above argument.The issue of taking descriptive basis for system design, i.e. basing design on a study of dialogue and interaction, can be restated as the question: What is the nature of the argumentative link between the analysis-description of what a human teachers and learners did and the design of a system?The relation can not be one of direct transfer of expertise, for a number of reasons.On the purely dialogue side, you have open-ended spoken dialogue versus constrained human-computer dialogue.And then, artificial agents are not meant to be copies of human ones.The interaction analysis framework and the study described in Cook (1998Cook ( , 2001) ) are part of a pedagogical agent design approach that aims to make practical use of empirical research in pedagogical agent development.We have argued, therefore, that because very few studies have examined how to develop an artificial agent in this way, i.e. to systematically link empirical data to agent design, the best starting point was to look at what human teachers and learners did, and to then implement descriptive models of that (in our case state transition networks).Refinements to the agent and to guiding theories or frameworks, e.g. the Knowledge Mentoring framework, can then take place on the basis of what happens in the real target dialogue environment when students use the system.Any refinement would thus take place as a result of formative evaluations.
Baker has consider different issues with respect to the evolutionary approach.His relevant work is summarised in Figure 3. Unlike Cook, Baker (2000) argues that there should not be any modulation in the case of Computer Supported Collaborative Learning; rather, a corpus is a means of validating a model or theory of interaction and learning (i.e.theories of argumentation).Baker does not attempt to move from empirical work and modulate the findings into the design of a system (this is shown by the dotted line at the bottom of Figure 3).Instead, for Baker the problem is then shifted to that of understanding how exactly (quasi-)formal models of interaction and theories of cooperative learning can 'give rise to' computer-based learning support.That is to say, Baker starts with theories and models, has then run sessions to validate or disprove the model or theory, and has then uses the revised theory or model to influence the implementation of computer-based learning support (the 'gives rise to' link in Figure 3).In fact the dotted lines in our descriptive diagrams, Figures 1 and 2, provide a limited analytical facility, in that we can see what aspect of the evolutionary approach is being omitted by a researcher.This can in turn highlight areas for future work.The interaction was designed as a prescriptive 'dialogue game' (e.g.Levin and Moore, 1997) which modelled features of a tutorial process.Within the developed scheme the learner adopts the role of an explainer whilst the system plays a facilitating role, and these participants collaborate to develop a shared explanatory model of a qualitative, causal domain.A prototype system CoLLeGE (Computer based Lab for Language Games in Education) implements this theoretical framework and currently operates as a dialogue modelling 'workbench' for demonstrating, investigating and developing the approach.Furthermore, an empirical study was conducted which showed that performing this dialogue game supported the dialogue process in ways which stimulated students to revise and refine their beliefs, leading to conceptual change and development in science (Hartley and Ravenscroft, 1999).The empirical work was reused to improve the redesign of the workbench (the bottom line in Figure 4).
The above examples have shown how the evolutionary approach is essentially one of iterative design in slow motion.The researchers described in this section have started from theory and models and proceeded to test them out by going either way around the evolutionary cycle.Not all researchers have completed a full cycle, but they may do so in the future.The following questions now arise: • Are the approaches presented above complementary, or do they reflect inherently different value systems or approaches?
• Can one approach be used to extend the other?For example, do the gaps in the options that were highlighted above illustrate or warrant further investigation?
With respect to the first question, we would claim that very little work has been done on how to modulate data form studies to learning support system design.This is a contentious claim and we accept that there have been some notable exceptions.Furthermore, although it is primarily a descriptive model, the evolutionary approach can help to explain the contributions of different studies and illustrate what further work needs to be done.Consequently, we suggest that it is possible to use our model -particularly the idea of gaps -to shape future work.For example, the author has, more recently, investigated educational dialogues form the opposite direction to Baker.In this recent work (Cook, 2000) we briefly describe how MetaMuse was designed to 'structure interactions' in such a way that would, it was predicted, facilitate creative problem-solving dialogues, i.e. we were starting from the bottom left of our evolutionary model.We then used MetaMuse to generate interaction data (the gap in Baker's work).However, although the users in this initial evaluation of MetaMuse reacted favorably, the initial evaluation of the pedagogical agent did not give much insight into the following question: what are the interactive means by which learning agents engage in cooperative, creative problem-solving?Consequently, we addressed this question by a detailed analysis of a transcribed corpus of the face-to-face interactions that took place between cooperating students when engaged with the pedagogical agent MetaMuse, i.e. we followed the link from empirical work up to theory.Specifically, the analysis results were used to clarify, at a fine level of granularity a model of cooperative, creative learning.We proposed that, in cooperative dialogue, the interactions will not focus on 'winning the argument' or 'persuading your partner', it will involve an acceptance by both participants that they will attempt to 'find and refine' a problem specification; where a problem specification is a description of a problem that is interesting or novel.Thus, in this recent work we have traveled clock-wise around the evolutionary framework.The computer-based agent, MetaMuse, gave rise to dialogues in an empirical study, which when analysed gave raise to a fine-grained model of interaction and cooperative learning (Cook, 2000).This model may contribute to future theorising in the area of communities of cooperative, creative learning.
In this section we have illustrated that our evolutionary framework can be helpful in identifying the theoretical work, and attendant dialogue analysis approach, that is being undertaken in the area of interactive media for education development.Furthermore, the evolutionary approach appears to have a high level of descriptive generality due to its simplicity and some potential as an analytical tool.

Conclusions
In this paper we have examined the role of dialogue in computer-based learning and observing learning.We have proposed that in order to improve our understanding of learning in the context of the use of new media in education, we need to link theory to the analysis of teaching-and-learning.Furthermore, we have posited that certain types of learning may not occur unless dialogue takes place between a tutor and learner(s).Interaction has an adaptive mediating role, helping students to recognise and resolve inconsistency.Dialogue may also be responsive to institutional questions and educational plans.Finally, we have presented a three-part evolutionary approach to investigating theories and models of interaction and learning.
We conclude by suggesting that our evolutionary approach can help designers of, and researchers into, learning technology in three important ways.First, it allows them to examine and compare the theories and models that are available for the design or research task in hand.For example, in the previous section we have demonstrated how our evolutionary approach can be used to compare different research projects.This should in turn enable a critical debate to take place, within the learning technology community, about what the different theoretical models and design approaches have to offer.Second, the evolutionary approach enables researchers and designers to develop models and theories that are appropriate to their own context, particularly in terms of educational objectives and the level of granularity.The use of interaction analysis is particularly useful in obtaining fine-grained data that can be used as the basis for the design of new forms of interactive media (i.e.modulated) or to verify a theory.Granularity is an issue for reuse of learning materials.For an approach to be valuable it is necessary for the learning objects themselves to be "identifiable, discoverable and useful at the smallest appropriate level of granularity" (MEG, 2001).A big problem with a lot of the current reuseable objects research and development is that the theoretic basis from a pedagogical perspective is very rudimentary, with much of the development being on the technical level.The evolutionary approach could be used in helping to clarify the theoretical base for learning object reuse.Furthermore, if a fine-grained level of detail is not appropriate for a particular project then the evolutionary model can be used to zoom out, as it were, and take a higher level view of what is happening in terms of the three evolutionary points; for example, to examine the political and social aspects of learning.The third advantage of the evolutionary model is that it helps us to assess the ways in which different researchers are taking the area of learning technology theorising and modelling forward.In our view this last advantage is a key point that should lead to evolutionary, and not revolutionary, theorising in the evolving 'discipline' of learning technology.

Figure 1 :
Figure 1: Cook going anti-clockwise around the evolutionary approach

Figure 2
Figure 2 gives an example of a MetaMuse interaction with a learner.The most recent dialogue output from MetaMuse is shown at the top of Figure 2 in the "MetaMuse output" box.

Figure 2 :
Figure 2: Example of MetaMuse interaction with a learner

Figure 3 :Figure 4 :
Figure 3: Baker's focus in the context of the evolutionary approach most of this work [educational research on interactions] is descriptive and statistical in nature.It tells us that a teacher spends 40% of his or her time responding to student-initiated activity (or whatever) but offers no help in understanding the processes and mechanisms involved.Similarly, the nonquantitative work, based on sociological and anthropological approaches, is of limited value for the types of models and theories which we wish to construct in AI and education ... we must obviously look at education if we are to find out about educationally specific goals.It is not clear, however, whether we can derive the information we need from existing work.There is a large gap to be bridged in terms of levels of description.If the gap cannot be bridged, then it is necessary for AI and Education to include repetitions of previous research at finer levels of detail.(Elsom-