Using Targeted Negotiation to Support Students

1

Using Targeted Negotiation to Support Students’ Learning Susan Bull Department of Artificial Intelligence University of Edinburgh 80 South Bridge Edinburgh EH1 1HN Scotland UK [email protected]

Matt Smith Computing and Maths Group School of Science, Technology and Design King Alfred's College of Higher Education Sparkford Road Winchester SO22 4NR UK [email protected] Abstract

Student/system negotiation provides students with the opportunity to assume more responsibility for their own learning. This paper discusses the notion of targeted negotiation, an approach to providing negotiation for specific purposes in intelligent learning environments (ILEs). This targeted negotiation may potentially be aimed at a variety of aspects of ILEs; the object of negotiation will depend on what is appropriate for the particular learning environment. As examples we illustrate two approaches by referring to two systems which employ negotiation for different functions. In one, the object of negotiation is the application of domain knowledge, and the domain knowledge itself. In this system this is necessary because the domain is creative. In the second ILE it is the contents of the student model which are negotiable; here the process is aimed at increasing learner awareness and reflection, thus learning, rather than the domain, is the target of negotiation. Keywords: targeted negotiation, creative domains, student modelling, learning strategies, melody composition, foreign language learning. Introduction Negotiation has been proposed as a means of overcoming some of the limitations of the traditional, directive tutor (e.g. Baker, 1990; Moyse & Elsom-Cook, 1992). There is potential for the negotiation of many (and possibly all) aspects of ILEs; e.g. learning goals, task, interaction style, interface and even the domain (for example in the case of more creative subject areas). Negotiation may also be used to increase metacognition by placing focus on the discussion process as a means of promoting reflection. Nevertheless, in many cases negotiation may not be the most effective approach for all aspects of the interaction. Targets of negotiation must be selected appropriate to the individual case (i.e. taking into account the domain, aims of the system, intended users, etc.). We argue here not that there should be different forms of negotiation, but that the object of negotiation, and when negotiation is used (or offered), form the basis of a notion of negotiation ‘targeted’ towards achieving some educational goal (whether this be more successful understanding, or application of domain knowledge, or encouragement of higher level thinking, etc.) This paper discusses when negotiation may be a useful technique, and what the object of negotiation is. We illustrate the notion of targeted negotiation with two systems which use a mixed-initiative dialogue of student controlled interaction, and negotiation; these systems use negotiation for different purposes. At any time during the student controlled mode, either the student or the system can request a negotiation which the other can accept or reject (though in practice there are few circumstances in which the system would wish to reject a negotiation phase). We argue for the importance of negotiation, but target its use to specific functions and tasks. Preliminary findings suggest that students are willing to enter into intermittent negotiation with a system (Bull & Pain, 1995). Moyse and Elsom-Cook (1992) explain that for some types of domain there is no single correct representation of knowledge. In addition, a tutor may often not be considered an exhaustive source of information about a domain. The first ILE we describe is an example of such a domain: MELODY-ED, a system for novice composers of melody, which is currently being developed (Smith & Holland, 1994; Smith, 1995a). Here, due to the impreciseness of the domain, it is the application of, and contents of the domain knowledge which are targeted for negotiation. The second system we introduce, although allowing discussion of the domain

International Conference on Computers in Education, Singapore, December 1995

2 (pronoun usage in European Portuguese), does not allow the domain itself to be negotiated (the tutor can here be expected to hold an exhaustive knowledge of the domain). Instead, it is the contents of the student model which are negotiable; a process aimed at promoting learner reflection about this ‘correct’ domain knowledge, and also about the learning process. This system is called Mr. Collins1 (Bull, 1994; Bull & Pain, 1995). In both systems the student is the focus. This is because the systems cannot be infallible, for example, as MELODY-ED is concerned with a creative domain, the system may not state "Your composition is wrong". In contrast, Mr. Collins can say "Your pronoun should go here instead", but it cannot insist "You believe X". This is reminiscent of the situation in collaborative learning systems, where the artificial co-learner is to be regarded as a peer who, like the human student, may make mistakes (e.g. Dillenbourg & Self, 1992; Chan, 1991). However in our systems the computerised negotiator is not attempting to learn the domain; the fallibility is not equivalent to the student’s fallibility. In MELODY-ED the limitations are due to the computer negotiator’s necessarily incomplete knowledge of the (creative) domain, and in Mr. Collins limitations are due to the obvious difficulties of accurately inferring a student’s beliefs. These limitations are factors which led to the application of targeted negotiation in these systems. Negotiation in Intelligent Learning Environments Negotiated Tutoring The potential for the use of negotiation in ILEs has been illustrated by Baker (1990, 1992). Baker proposes a form of tutorial dialogue he terms negotiated tutoring, using as a vehicle for the technique an ILE for assisting students in the expressive performance of melodies. Two systems are involved, the KANT dialogue engine, and the GRAF melody analysis program. Students collaboratively parse melodies with the GRAF system, to construct a structural analysis of a melody. This analysis is used as the basis for an expressive performance of the piece to communicate the identified structures. Baker’s GRAF system is based on Lerdahl and Jackendoff’s (1983) theory of music analysis, but like all such music theories it is incomplete, presented in a semi-formal representation, and assumes a substantial amount of musical knowledge in those wishing to apply the theory. Therefore Baker proposes use of the KANT dialogue engine to allow the student and system to collaboratively parse a melody; each agent having to justify their statements, and attempting to come to an agreed analysis through the negotiation process. Targeted Negotiation Although negotiation can be useful, too much negotiation could become overwhelming. There clearly are areas where continuous negotiation would be beneficial, for example, developing metacognitive skills of argumentation (e.g. Baker, 1990). However in many cases there is no need for negotiation the whole time. Examples of candidates for targeted negotiation include situations where there may be a specific, indisputable2 topic to be learnt, and where the necessity of learning this topic is clear. In such cases it may be possible and desirable to negotiate the actual task(s) to be performed to learn this aspect of the domain, as the same tasks may not be equally suitable for all learners. Similarly, as learners may have different preferences regarding the manner in which they interact with the system, aspects of the interface could be negotiated. This would involve student statement of, for example, their preferred method of viewing information (e.g. textually, graphically, etc.), and system explanation of trade-offs, i.e. what could be gained by using one method, and what may be lost. MELODY-ED and Mr. Collins provide specific examples of negotiation targeted to particular functions. In this section we provide an overview of these ILEs, and highlight the objects of negotiation in the two systems, explaining why negotiation is directed to these areas. MELODY-ED A traditional intelligent tutoring system architecture is inappropriate for the domain of melody composition; there are no clear criteria for when a solution has been found. Therefore the MELODY-ED system is based on a domain-independent architecture for creative domains in general3. MELODY-ED is currently under development, and has been designed around two existing computational tools: a parser for melodies (MPARSER) and a musical constraint-based generator (M-CONSTRAINT) (see Smith & Holland, 1994; Smith, 1995a). Novice composers of melody are set the task of composing a melody to meet a set of constraints set by the system (e.g. a piece of twelve bars, in the key of C major, in 4/4 time). Students compose in MELODY-ED by first constructing a generating analysis (a constraint-network describing a class of melodies that would yield the 1 2 3

Strictly speaking it is the student model which is called Mr. Collins (collaboratively maintained, inspectable student model). However for simplification we here also refer to the system itself as Mr. Collins. Indisputable in terms of the ‘correct rules’ of the domain. The CONNIE architecture (Smith, 1995b).


3 analysis when parsed according to a given music theory4). The process of parsing a melody is one that can be negotiated, since some of the aspects of the computational parser in MELODY-ED are based on domain knowledge (i.e. encoding of the musical theory) which are simplistic, or at the least represent one of a number of ways of describing different aspects of musical knowledge. Once the student has a generating analysis, additional constraints can be applied to it, and then a melody (or set of melodies) that meets all the constraints can be generated by a constraint-engine. A second place where negotiation can occur is in the application of constraints to a generating analysis. The system can suggest constraints that would help lead the final melody to meet the task constraints, and can disagree with the choices of the student. It may be that the student wishes to apply constraints that are not represented in MELODY-ED’s constraint database, in which case the student may propose a new constraint, and the system and student will negotiate the details of the constraint and whether it is valid (e.g. as a representation of a particular rhythmic style, or as more specific representation of a style of embellishment a named composer might use). However, many actions of the student in creating the generating analysis, and in choice of constraints to apply to this analysis, will not be contentious, and if the system were to enforce negotiation for every action the task would become very laborious. Therefore interaction between the student and system uses a mixture of direct manipulation (Schneiderman, 1984), and negotiation. The direct manipulation is of the current analysis and any constraints applied to it, and for interactive lookup of the constraint knowledge-base and melody database. The use of negotiation is targeted to circumstances where an agent either is unsure of an action, and will request a negotiation dialogue to gain collaborative assistance, or where a passive agent disagrees with an action by the active agent. The passive agent may request a negotiated dialogue with the goal of either convincing the active agent of their error, or allowing itself to be convinced of the action by the justification given by the active agent. Mr. Collins Mr. Collins does have clearly defined correct domain knowledge (grammar rules), which can therefore not be negotiated. This domain information may be accessed by the student in a number of ways. She may ask direct questions (composed from restricted menu options). She may view example sentences or inspect Portuguese grammar rules and the equivalent rules of other languages (for consideration of language transfer). The learner may also practice by requesting exercises. These take the form of sentences offered by the system, with the instruction to type the given pronoun into the correct position in the sentence. (In European Portuguese the rules for the placement of pronouns are quite complicated.) The student then offers the modified sentences back for parsing. As stated above, because Mr. Collins contains representations of correct domain knowledge, the domain may not be negotiated. However it is possible to enhance learner awareness of this correct domain knowledge5, and indeed, of learning in general. So it is helpful to have the more usual exercises and possibilities for exploration, but also include a meta-level of reflection. This reflection can be achieved through the student model, by means of negotiation of its contents. The student model may be made visible on the screen (presented through templates producing textual and tabular information). Negotiation can concern two aspects of the model: the student’s view of the domain, and the future use of learning strategies. Negotiation over the correct representation for the student’s view of the domain is based on two sets of belief measures (Bull, 1994) relating to 1. the system’s assessment of the student’s perception of the domain (the more conventional component of the student model - the system’s student model), and 2. the student’s own view of the domain (the student’s student model). Belief measures for the student’s student model are obtained by requiring the student to state her confidence in her performance at the same time as entering her input. Negotiation may occur in cases where an agent is unsure of a representation in the student model, or where there exists disagreement between the two agents about the student’s beliefs. In the first case questioning occurs. In the second, a challenge. This challenge requires defence from the other (challenged) agent, or that agent’s acceptance of the challenging agent’s justification6. Negotiation about learning strategies is aimed at encouraging the learner to try appropriate learning strategies depending on the type of strategy she prefers, and has success with. Implementation of the learning strategy component (described in Bull, Pain & Brna, 1993) was based on O’Malley and Chamot’s (1990) description of language learning strategies. Results of Negotiation As negotiation can be targeted for different purposes, it will have different results. The desired results are the factors determining what should be negotiated in a particular system. Figures 1 and 2 illustrate the results of 4 5 6

Narmour (1990); formalised and computationally modelled in Smith (1995a). See Schmidt (1990) for the importance of language awareness. See Bull & Pain (1995) for examples and discussion.


4 negotiation in MELODY-ED and Mr. Collins; these results may be in terms of a product, educational benefits or a change in learner behaviour. Two objects for negotiation are illustrated in figure 1, the first is the application of domain knowledge, and the second is the constraint knowledge base. The domain knowledge is in two parts: knowledge about the analysis of an existing melody, and knowledge about how constraints can be applied to an existing analysis in order to generate new melodies. The negotiation of the first of these types of domain knowledge results in the production of an analysis for a given melody. Then negotiation of how constraints can be applied to an analysis, (and the use of the analysis itself) result in a melody being generated. In addition to negotiation of how to apply the domain knowledge, the contents of the constraint knowledge base itself can be negotiated. If an existing constraint were changed, and were referred to in the generating analysis, the result could be a change in the melody (or melodies) being generated. In all cases negotiation results in a product, (i.e. an analysis, melodies, new or changed constraints), as the aim of MELODY-ED is to assist the student in her creation of a product. key to figures:

product

educational benefits

student behaviour

generating analysis

negotiation of application of domain knowledge

melody

negotiation of constraint knowledge base

changes to constraint knowledge base Figure 1:

MELODY-ED

reflection on domain knowledge

negotiation of student model (beliefs about domain knowledge)

enhanced performance

changes to student model

changes in learning strategy use Figure 2:

negotiation of student model (learning strategy use)

reflection on learning strategies

Mr. Collins

In figure 2 there are also two possible objects of negotiation: two different aspects of the student model. Firstly, the representations of beliefs about the student’s domain knowledge and misconceptions may be negotiated. This leads to a product: changes in the student model. There are also educational benefits resulting from the negotiation process: learner reflection on the domain knowledge. This reflection may, in turn, lead to improved performance (i.e. a change in student behaviour), resulting in further changes to the student model. Secondly, the student’s future use of learning strategies may be discussed. Again there is a product: changes to


5 the learning strategies component of the student model, and again there are educational benefits: student reflection on their approaches to learning, and a change in student behaviour: the learner may try new learning strategies, or drop current ones. In Mr. Collins the product (changes to the student model) and the changes in student behaviour are consequences of the system’s educational aim of increasing reflection. Note that although the negotiation of constraints in MELODY-ED could in some sense be viewed as similar to the negotiation of beliefs about what constitutes a good melody, negotiation is not here aimed directly at learner beliefs as in Mr. Collins, because the student model for MELODY-ED only consists of a set of system beliefs about what are considered to be the current beliefs and dialogue goals of the student. This is because it is a creative task which is being undertaken, and the resulting product belongs to the student. It is therefore not the role of the system to try to change the beliefs of the student (in most cases). Features of Negotiation in MELODY-ED and Mr. Collins We will now describe the manner in which negotiation takes place in MELODY-ED and Mr. Collins. We have already argued that targeted negotiation should have different purposes and results in different ILEs, therefore these descriptions are only in terms of the important features of negotiation in these two systems. These features can be characterised by the following headings: agreement/disagreement; collaboration; compromise; judgement; justification; resolution; symmetry. Some of these aspects are similar in our two systems, and some differ due to the different functions of negotiation; these descriptions are therefore intended as examples of how negotiation may function. agreement/disagreement Agreement and disagreement here refer simply to whether the views of the student and system are identical. MELODY-ED: Agreement is preferable, but it is not necessary in MELODY-ED dialogues. The aim of the negotiated dialogue is to provide the opportunity for the system and student to move towards agreement. Mr. Collins: Agreement is reached in the student model if the student’s own confidence in her use of a particular rule exactly matches the system’s confidence in her ability to use that rule. The aim is to reach agreement, but this may not always be achieved. In contrast there will be no doubt in the student model about which learning strategies have been used, as their use is actually traced (Bull, Pain & Brna, 1993), but there may be disagreement about future use of learning strategies, though the aim is to try to avoid this through negotiation. collaboration Collaboration implies two agents working together on the same tasks towards a common goal. MELODY-ED: Both objects of negotiation in MELODY-ED (the constraint knowledge base, and the set of constraints used to analyse a melody and to be applied to the generating analysis for the current composition) are computationally represented. The system and student have the joint, collaborative aim of working with these ‘objects’ to generate a melody from the starting constraints set out by the system. Mr. Collins: Through discussion the student and system collaboratively maintain the representations in the student model of the student’s beliefs. Future learning strategy use is also collaboratively negotiated. compromise Compromise implies the adoption of a view representing neither agent’s position exactly, but one which is somewhere between what each agent believes, and the compromise position is acceptable to both agents. MELODY-ED: A feature of the negotiation of constraints is that they are not necessarily present or absent; constraints may be relaxed, i.e. applied in a weaker, more general form. If the student and system disagree about the application of a constraint, either agent may suggest a compromise in the form of a relaxation of the constraint in question. The negotiation then focuses around the nature of the relaxation. Mr. Collins: In a situation where there is disagreement between a certain belief held by the system about the student, and the actual belief of the student, and if the distance between these two belief measures is not too great, a compromise (mid-point) can be agreed on if both parties are in favour. In contrast, with learning strategies there is no possibility for compromise, either a student will use a particular strategy, or will not. She may try a new strategy after discussion with the system, but this will either be on her own initiative, or after justified proposal from the system, rather than compromise. judgement Judgement here refers to whether there is room or reason for one agent to judge the views of the other. (Our use of the term judgement does not include any of the moral issues, such as whether a computer system should be allowed to judge a student.) MELODY-ED: Due to the imprecise nature of creative domains, neither agent can make fully confident judgements of the other’s statements. Therefore judgements with MELODY-ED are qualitative, and take the form of supported disagreement and agreement rather than critical statements of correctness.


6 Mr. Collins: Because the negotiation is here about the student’s beliefs, the student will have her own ideas about what the correct representation in the student model should be. The system will have evidence of student performance gathered during the interaction, and from this evidence it is able to conclude its beliefs about the student. Each agent therefore has clear ideas about the correct representation to be used, and in any situation where one agent disagrees with the other, this agent will necessarily be judging the other’s beliefs. However for discussion of learning strategies the situation for judgement is weaker; each agent is aware that the other agent has additional knowledge; i.e. the system’s expert knowledge of learning strategies, and the student’s feelings about what works for her. Because these are different types of knowledge, it is harder for either agent to judge the other’s viewpoint. justification Justification refers to the manner in which participants must defend and argue for their own views during negotiation. MELODY-ED: The type of justification in MELODY-ED dialogues depends on the subject of the negotiation. In the case of negotiated parsing of a melody the student may justify some part of an analysis (e.g. the hierarchical promotion of a note) by referring to some aspect of domain knowledge which the system has flagged as being a simple view (e.g. harmonic consonance) or for which there is no computational implementation (e.g. recognition of a modified phrase from earlier in a piece). The system will justify analysis statements in terms of the implemented analysis theory. The acceptance and form of a new constraint to be inferred from existing pieces of music can be justified in terms of existing, similar constraints, or by reference to a melody in the melody database (not currently being referred to) which also exhibits the constraint. Justification may also take the form of reference to similar previous negotiations in the dialogue record, or drawing the attention of the disagreeing agent to a conflict between the constraint under discussion and an existing, accepted constraint. Mr. Collins: The system can justify its representations of the learner’s beliefs by referring to previous input of the student; the student’s own stated belief measure; her previous statements about the student model; by running its own or the student’s student model to try to prove that the representations contained within can produce evidence supporting its view. The student can justify herself by providing information which is not available to the system, e.g. that she has simply forgotten something; that she had just been guessing before; that she had previously believed she understood something, but now has evidence to make her unsure, etc. Justification applies in a weaker form in the case of learning strategies, e.g. the system may state: "The combination of your use of the strategies of note-taking and summarization show that you are very keen on noting down new or relevant information. It would be useful also to consider the grouping of information; i.e. organising new or important information in some manner which is meaningful to yourself", in its attempt to justify its suggestions to the student, but the system is not able to prove that this would be a useful strategy for that student (in the way it could prove that the student had used a grammar rule correctly by producing evidence from the interaction). resolution Resolution has to do with eliminating conflict. However, it is clear that in both systems there may be no such resolution. Therefore resolution is also concerned with which agent has the power to make the final decision. MELODY-ED: In all cases of unresolvable disagreement the student has the final decision. Partly this is acknowledgement that current musical theories (of both music analysis and composition) are not sufficiently formal for sophisticated computational modelling. Secondly, due to the nature of the domain there are no correct answers, and since the task of composing a melody is ‘owned’ by the student, she should have final control of decisions. Mr. Collins: Negotiation is aimed at resolving conflict in the student model. Where an agreed resolution is not possible, the final decision is made by that agent who owns the view which is being challenged, i.e. the system has ultimate control over the representations in the system’s student model, and the student has final say over the student’s student model. symmetry Our use of the term interaction symmetry is based on Baker’s (1990) notion that each agent should be provided with the same set of dialogue moves. MELODY-ED: With the exception of who can resolve disagreements (see resolution above), all negotiations have symmetry, in that actions can be proposed, agreed with, disagreed with, challenged, justified, and modified, by either system or student. Mr. Collins: The same possibilities for maintenance of the student model are provided to each side. The system may challenge the student about the student’s student model, and the student may challenge the system about the system’s student model. Both participants may provide justifications for their beliefs, and request information from the other. There is also symmetry of power of resolution (see above).


7

What do you tell the student? As discussion in MELODY-ED concerns the domain directly, this is a focus which the student will be expecting, and therefore her attention need not be explicitly directed to the fact that she is performing a negotiative task (though she must be aware that the system is not the more powerful agent). However, in Mr. Collins the focus is on the student’s beliefs. This is an unusual perspective, and because of this ‘surprise element’ it must be ensured that the student understands why her beliefs may be challenged; i.e. the potential educational gains arising from the negotiation process, and the better system adaptation to the student resulting from the improvements in the student model after negotiation. The learner is then more likely to accept, and exploit the approach of the system. An example Imagine a student using Mr. Collins wishes to discover more about learning strategies. She may inform the system about this decision through selection from the permanent 'student menu'. The following is an example discussion, ultimately leading to negotiation about which strategy the student should try next7. ___________________________________________________________________________________________ system: What would you like to discuss? • your use of learning strategies • learning strategies in general • (cancel) student: (selects) learning strategies in general. system: Which type of strategy do you wish to discover more about? • metacognitive • cognitive • social • (cancel) student: (selects) social. system: Which social strategy would you like more information on? • question for clarification • cooperation • (cancel) student: (selects) cooperation. system: 1. Cooperation implies the idea of working with others. Here you will be working together with the system in an attempt to clarify your knowledge and beliefs (both to the system and to yourself). 2. You may state your beliefs about your learning strategy use at any time through selecting 'your use of learning strategies', and also ask questions. You may assert your confidence level in a particular sentence by using the dialog box which appears after you have entered your input. You may discuss grammar (target only, or transfer-related) by selecting 'question grammar' from the 'student menu'. You may discuss the system's representation of your beliefs by selecting 'question model' from the 'student menu'. 3. The system will initiate discussion on any of the above topics if it requires information, and it is helpful if you cooperate with such requests (as this ultimately results in a more accurate representation of your knowledge in the system, which will, in turn, enable the system to anticipate your requirements more effectively). 4. The system will cooperate with all your requests for information. 5. It may be helpful to discuss your use of this learning strategy. Do you wish to do so? • yes • no • (cancel) student: (selects) yes. system: How would you like to proceed? • state your views on your cooperation • see system assessment of your cooperation • (cancel) student: (selects) see system assessment. system: You cooperate very well with system requests for information about your beliefs. You tend not to volunteer information on your own initiative. The concept of cooperation includes offering useful information also at times when it is not explicitly requested. It is good that you ask questions about grammar when you are unsure. It is good that you challenge and discuss the system's representations of your beliefs when you disagree. Would you like to see examples of your previous use of cooperation? • yes • no • (cancel) student: (selects) no. system: Given that you already cooperate well and question the system appropriately, it may be useful to try more resourcing, because this is also a useful method of obtaining information. Do you agree? • yes • no • no (and comment) • (cancel) 7

Examples and analyses of interactions, and student reactions to negotiating the representations of learner beliefs in the student model, can be found in Bull and Pain (1995).


8 student: (selects) no (and comment). system: Please select one of the following: • this does not feel right FOR ME • this does not feel right FOR ME (and comment) • I would prefer to try another strategy • I would prefer to try another strategy (and comment) • I would like to come back to this later • I would like to come back to this later (and comment) • none of the above • (cancel) student: (selects) I would prefer to try another strategy (and comment). system: Please state your reason for preferring to try a different strategy. • resourcing would be easy for me (I would prefer something I may need to think more about) • I would like to try an approach which is very different from my usual approach • none of the above • (cancel) student: (selects) resourcing would be easy for me. system: Is there a specific strategy you would like to try? • yes • no • (cancel) student: (selects) no. system: 1. As you already use the information seeking social strategies of questioning and cooperation, and information seeking cognitive strategies of deduction and inferencing, it would be useful for you to consider some of the other cognitive strategies: note taking or summarization. This is because it is useful to reorganise the new knowledge you gain through these strategies, in a manner that suits your own needs. 2. As you already effectively obtain new information, you may like to try the metacognitive strategies of self monitoring or self evaluation, with reference to this information. Would you like to try any of these strategies? • note taking • self monitoring • other strategies • summarization • self evaluation • (cancel) student: (selects) self monitoring and self evaluation. ___________________________________________________________________________________________ In this example the student has chosen to seek more information to clarify her understanding of a particular learning strategy (cooperation), before considering her own actual use of the strategy. The system responds with a short definition of cooperation, including explanations of how she may initiate discussion with the system of various aspects of this strategy. Following this, the system views it as an appropriate moment to prompt discussion of the student's own use of this learning strategy (with a view to subsequently suggesting another strategy). The student agrees to discuss her use of cooperation, choosing at this point to take the more passive role of hearing the system's views. Following its assessment, the system suggests another strategy (resourcing), which could be useful for that student, and also easy for that student to adapt to, given that she already uses other information seeking strategies. However the student disagrees, but cooperates with the system nevertheless by choosing to inform it of why she does not wish to try out this strategy (at least, at the present time). Note that this latter evidence of cooperation will increment the system's record of the student's cooperation despite the fact that the student does not agree to follow the system's suggestion. Her choice of trying a different strategy is valid, and indeed, she chooses two which are subsequently proposed by the system as appropriate for her (thereby again indicating cooperation). The system's record of the student's questioning (with regard to learning strategies) will also be incremented, as the initial request for information came from the student. Targeted Negotiation in Other Contexts We have already suggested the potential for different objects of negotiation (task, teaching strategies, etc.), in different types of ILE. Specifically, we have discussed the music domain as an example of a creative task for which there is no single correct view of the domain knowledge or of the criteria for what constitutes a ‘good’ melody, and foreign language learning as an example of a domain for which one correct viewpoint can be defined: the grammar rules. It is interesting to note that the reverse situation can be found for different aspects of these domains. We have described melody composition as a task which, when performed in collaboration with a system, must be achieved through negotiation due to the lack of precise definition of the task. However, in some cases music composition can be presented as a domain where there is a single, correct view. One example of this is 16th century strict counterpoint, a set of rules for the construction of pieces of music with three voices (notes playing simultaneously). In general practice, when novice composers are first taught this counterpoint, the rules (which can be viewed as constraints) are presented as if they are always correct, and should never be broken. The task for the student becomes one of finding notes for the harmonisation which do not break any of the rules. An ILE designed to assist the student in learning to perform such harmonisations would have the dual purpose of highlighting any rule violations, and possibly, negotiating with the student over alternative notes which do not violate the rules. If a student were to use MELODY-ED to compose a melody, then use the second type of ILE


9 to harmonise the melody according to 16th century counterpoint, the student would experience ILE negotiation from both these points of view. Foreign language learning has been treated in this paper as a domain made up of definite rules which may not be negotiated. Nevertheless, certain tasks in language learning may be more open to negotiation, for example, translation. There is a clear necessity for a translation to match the content of the original. This is not negotiable. However, once the correctness of the content of the translation has been assumed, the question ‘is this a good translation?’ remains. There is scope to encourage student/system negotiation about the style of the translation, for example. There is a need for the style of the two texts to be similar (e.g. formal or informal), but this similarity may be achieved through different surface forms. Another important consideration is that the translation reads naturally in the second language, i.e. it does not sound like a translation. Negotiation could be used to attain the final 'fluent' version. So intelligent computer assisted language learning system design for tasks involving stylistic aspects of translation must be guided by similar considerations to MELODY-ED, since there is no single correct view of such translation domain knowledge. Finally, it should be noted that although we have in this paper been describing two systems and two domains, the two approaches described are themselves general. The possibility of implementing the music approach has also been considered for another creative domain: juggling pattern design (Smith, 1995b), and a negotiated student model could also be used in other areas. This has been suggested for the domain of electrical circuits (Bull, Brna & Pain, in press). Here, again, it is the requirements of the proposed systems which dictate the objects of negotiation (i.e. creative domain in the first, and educational philosophy of promoting reflection in the second). Summary and Conclusion In this paper we have identified the potential for targeted negotiation; situations where negotiation may be usefully turned on or off. Objects of negotiation in an ILE will vary depending on the aims and important features for that particular system. Two systems were described as vehicles for illustrating different methods (and reasons) for targeting negotiation. Both use negotiation, but the objects of negotiation, and the aims of the use of negotiation, differ. References Baker, M.J. Negotiated Tutoring an Approach to Interaction in Intelligent Tutoring Systems, unpublished PhD thesis, Open University, Milton Keynes, UK, 1990. Baker, M. Design of an Intelligent Tutoring System for Musical Structure and Interpretation, in Balaban, M., Ebcioglu, K. & Laske, O. (eds), Understanding Music with AI, AAAI Press/MIT Press, Cambridge, MA, 1992. Bull, S. Student Modelling for Second Language Acquisition, Computers and Education vol. 23, no. 1/2, 1994, pp. 13 20. Bull, S., Brna, P. & Pain, P. Extending the Scope of the Student Model, User Modeling and User Adapted Interaction, in press. Bull, S. & Pain, H. "Did I say what I think I said, and do you agree with me?": Inspecting and Questioning the Student Model, Proceedings of the World Conference on Artificial Intelligence in Education, Washington DC, 1995. Bull, S., Pain, H. & Brna, P. Student Modelling in an Intelligent Computer Assisted Language Learning System: The Issues of Language Transfer and Learning Strategies, Proceedings of the International Conference on Computers in Education, Taipei, Taiwan, 1993, pp. 121 - 126. Chan, T-W. Integration-Kid: A Learning Companion System, Proceedings of International Joint Conference of Artificial Intelligence, 1991, pp. 1094 - 1099. Dillenbourg, P. & Self, J.A. PEOPLE POWER: A Human-Computer Collaborative Learning System, Proceedings of ITS 1992, Montreal, 1992, pp. 651 - 660. Lehrdahl, F. & Jackendoff, R. A Generative Theory of Tonal Music, MIT Press, Cambridge, MA, 1983. Moyse, R. & Elsom-Cook, M.T. Knowledge Negotiation: An Introduction, in Moyse & Elsom-Cook (eds), Knowledge Negotiation, Academic Press, Harcourt Brace Jovanovich, Publishers, London, 1992, pp. 1 - 19. Narmour, E. The Analysis and Cognition of Basic Melody Structures, University of Chicago Press, Chicago, 1990. O’Malley, J.M. & Chamot, A.U. Learning Strategies in Second Language Acquisition, Cambridge University Press, Cambridge, 1990. Schmidt, R. W. The Role of Consciousness in Second Language Learning, Applied Linguistics, vol. 11, no. 2, 1990, pp. 129 - 158. Schneiderman, B. The Future of Interactive Systems and the Emergence of Direct Manipulation, Behaviour and Information Technology, No. 1, 1982, pp. 237 - 256. Smith, M. MOTIVE: A Constraint-Based Tool for Novice Composers of Melody. PhD thesis (submitted April 1995), Open University, Milton Keynes, UK, 1995a. Smith, M. CONNIE: An Intelligent Learning Environment for Creative Tasks Based on the Negotiation of Constraints, Proceedings of the World Conference on Artificial Intelligence in Education, Washington DC, 1995b. Smith, M. & Holland, S. MOTIVE: Development of an AI Tool for Beginning Melody Composers, in Smith, Smaill & Wiggins (eds), Music Education: Artificial Intelligence Approach, Springer Verlag, Berlin, 1994, pp. 41 - 55.


10 Acknowledgements

We would like to thank Paul Brna and Geraint Wiggins for their comments on this paper. Mr. Collins was funded by the Economic and Social Research Council, award no. R00429124033. The systems associated with MELODY-ED were funded by the Science and Engineering Research Council, award no. 90311096.
