cK¢ takes up the challenge of modeling learners understanding (a response to Guershon Harel - continued)

Several of the questions Guershon Harel [*] asked after my talk at the PME-NA conference in Chicago concern the scope and objectives of the cK¢ modeling framework. In this post, I take each of these questions and give a short answer leaving for specific posts more elaborated answers when needed. So, here it is:

4. Is the cK¢ a model of learning processes or learning states?

The answer is very simple: cK¢ provides a framework for modeling learning states. Indeed learning processes are of a paramount importance, but they are in my opinion more an object of study for psychology than for mathematics education. Indeed, I don't confuse "learning processes" which are of a mental and intellectual nature, and "problem solving processes" which correspond (to make it simple) to the activity the learner engage when he or she has to solve a problem in a given situation. We need to understand and model these processes, but even if they may inform us about learning processes they are only a dimension of them.

5. Is the cK¢ a model of a learner (period), a model of a learner learning mathematics, or a model of a learner in a mathematics classroom setting?

cK¢ provides a framework to model learners' understanding (learning states, as just said) in mathematics from a situated perspective; situations may be set up within a classroom or in an other context. Actually, the objective is slightly larger, I would claim that cK¢ is a framework to model mathematical understanding taking into account the situational characteristics, not being restricted to learners. A key idea when I started the project was to find a way to model mathematical conceptions with the same tools, be it they conceptions of novices or experts, wrong or correct from whatever knowledgeable perspective.

6. What exactly are the challenging aspects of modeling learning relative to modeling content and pedagogy? 

Anyone will expect the content to be in some sense "correct" and explicit enough to be defined precisely as a content to be taught. Still, there are challenging aspects related to its nature; for example, to model algebra or geometry from an epistemic and teaching perspective is not of the same level of difficulty.
Concerning pedagogy, which is in the first place the product of a practice, the related knowledge is largely implicit. Practitioners can share and discuss their expertise within their community or with teacher students in an apprenticeship approach, but the communication register is largely based on a pragmatic co-construction of meaning referring to a shared practice; as a result it is difficult to model. The challenge is to make explicit what is largely knowledge in action. However, there are progresses as witnessed by research on tutoring and adaptive learning systems.
In line with what I wrote before, I will not consider learning but "the best conditions for learning" (best or optimal). So, I would consider the question: what are the challenging aspects of modeling (determining) the best conditions for learning compared to those of modeling content and pedagogy? I would say that these challenges are very close the one to the other. First, part of the challenge comes from the nature of the content at stake, in particular the role of representations and the complexity of validating the related mathematical statements. Arithmetic, algebra, geometry, calculus, probability, discrete mathematics raise their own specific learning challenges within mathematics. Then, determining the best conditions for learning requires knowing some critical things about the initial state of knowledge of the learners (as a matter of fact, we can only teach people who know); in other words it depends on our knowledge of the conceptions which have to evolve or to be rejected. Eventually, one can say that the challenge is of knowing the knowledge at stake from a learning perspective, understanding learners initial understanding, and being able to design situations likely to stimulate, support and validate the construction of new knowledge. Eventually, modeling pedagogy consists in stating principles for designing situations which implements the "best conditions for learning" under multiple constraints: curricula, institutional standard, cultural and economical environment, time and all material means to make the class working properly.

7. What are the interdependent relationships among these three models? 

Indeed, as the above answers suggests it, the three models are tightly related. In particular, from an educational perspective modeling knowledge is under learning constraints because what we need is not a "knowledge model" for itself but a model of the intended learning outcomes. This knowledge (intended learning outcomes) must be learnable (accessible to learners) and teachable (manageable by teachers); actually, the objective of the didactical transposition is exactly to produce this knowledge which is always at a distance from a knowledge of reference which to some extend justify it. 

8. What is the efficacy of such models if they are constructed independently from each other? In particular, can models of content and pedagogy be viable without the presence of a learning model?

Be it explicitly the case or not, any pedagogical model includes a learning model; I mean a model of the (claimed) best conditions for learning. 

Then, a more philosophically oriented, yet critical, question is

9. Are cognitive models of thinking possible?

Once we have agreed on what means "cognitive", "model" and "thinking" my answer would be: yes... but a discussion of this answer may go far beyond my field of expertise and beyond the scope of this blog as well.

A decade after, what is left from Kaleidoscope?

Ten years ago, on March 2004 the 9th, we held the kick-off meeting of Kaleidoscope, a FP6 network of excellence, in the Castle of Sassenage, near Grenoble. A great day for a great ambition. The network initially gathered 76 research teams in Technology Enhanced Learning (TEL), what meant about 850 researchers and PhD students ; by the end of the EC contract we were about an hundred research teams associated in some way, and more than a thousands researchers and PdD students.

The aim of Kaleidoscope was to foster integration of different research disciplines relevant to TEL, bridging educational, cognitive and social sciences, and emerging technologies. To bring this ambition to reality, in a very fragmented European TEL research area, we chosen to involve a large number of contributors of which only a small number were already collaborating, and a large range of different research themes. Hence a very high level challenge. A set of instruments (focussed joint projects, virtual doctoral school, common platform, etc.) was planned to support the integration process at both the content and the infrastructure level (cf. the technical annex of the project [here], and the slides of the general presentation at the kick-off meeting [there]).

In my opinion, situated at equal distance from success and failure, Kaleidoscope was both a human and a scientific venture. Writing a report on the lessons learned with Sten Ludvigsen, scientific director of the network during the last period of the contract, we noted that "the history of these four years is that of the construction of the network in interaction with a process for understanding what to be a Network of Excellence means, and what integration means in the TEL research area. It is also the history of the interactions between the consortium and the reviewers team and the project officers."

Interestingly, this difference in the views about Kaleidoscope may be illustrated with a certain sense of humour by this picture. Above the head of Jens Christensen, our founding project officer, the portrait of Gaspar Baron de Sassenage, above myself the image of a character taking off supported by angels in a blue sky... Ten years after the character has landed. He is back with ideas still ambitious but probably better shaped by experience and a certain sense of pragmatism which he learned in particular in an other TEL network of excellence from the FP7, STELLAR. Some outcomes of this joint academic venture are still there, as the TeLearn Open archive, the TEL dictionary, and the largely disseminated book synthesizing the Kaleidoscope scientific legacy. TELEARC, the association which has taken the challenge of keeping alive and building on Kaleidoscope legacy has organised a new Alpine Rendez-vous conference in collaboration with STELLAR, and may organize an other one. But all this does not really account for what the Kaleidoscope network has changed in the TEL research area, to understand this change the best data we could have is that from your own view and experience, hence the question:

As a participant in the Kaleidoscope network of excellence, either contractor or associated, what in your opinion can be considered as a legacy? What is left or what you miss when looking back to what we did?

You can respond by leaving a commentary on this post. If there are enough comments, I will make a synthesis of your views and publish it on this blog (let's say in a month or two) and possibly find a way to share it with the project officers and the reviewers who have looked after us during these years.

Conceptions et situations

La place de la recherche sur les connaissances des élèves n'est pas tout à fait claire en didactique et est parfois contestée. En témoignent les vifs échanges entre psychologues et didacticiens dans les années 80, années fondatrices de la didactique des mathématiques. Pourtant l'étude de ces connaissances pour leur compréhension et leur modélisation est inséparable de celles engagées dans le cadre de la théorie des situations didactiques, c'est dans ces termes que Guy Brousseau l'évoque dans l'article qu'il publie dans le premier numéro de la revue Recherches en Didactique des Mathématiques alors qu'il déplore que les travaux de Diénès ne conduisent pas le didacticien à "questionner les mathématiques pour y chercher, au-delà des structures, les concepts et au-delà des concepts, éventuellement les conceptions qui pourraient se forger chez un sujet dans des situations historiques ou didactiques particulières."
Il poursuit :

"L'analyse de ces conceptions, qu'il faudra que l'élève possède ou évite, est inséparable de celle de la famille des situations spécifiques où elles prennent leur fonction et utilité. Toutes les deux sont inévitables dans toute entreprise qui prétendrait à la fois fournir une théorie dotée de ses méthodes de confrontation (probablement spécifiques aussi) et de techniques didactiques continument contrôlable par les enseignants" (Brousseau 1980 RDM 1.1 p.46)

Dans le même volume (p.80) Régine Douady insiste :

"Le problème didactique est de reconnaitre et décrire, à travers les actions et démarches des enfants placés dans une situation d'apprentissage, les modèles mathématiques qui expliquent, justifient ces actions et démarches."

En d'autres termes, la proposition de Douady est de produire des modèles mathématiques des conceptions dont Brousseau pose qu'elles sont indissociables des situations. Il faut entendre ici situation au sens de ce qui va, dans l'interaction entre l'élève et le milieu, être la source de problèmes mobilisateur des conceptions. Ces conceptions pouvant être, dans une perspective mathématique, erronées ou inadaptées et ce qui fait problème étant finalement largement déterminé par les conceptions initialement disponibles, la production de modèles tels qu'évoqués par Douady est un défi. C'est celui que relève la proposition de modélisation cK¢ notamment en formalisant la dualité entre problèmes et conceptions.

cK¢ is not a cognitive model (a response to Guershon Harel)

The first question Guershon Harel [*] asked about cK¢ is

3. To what extent is the cK¢ a cognitive model?

Actually, this question comes after a more general one: (1) "What is a cognitive model and what are its purposes?" and a more direct one (2) "Is the cK¢ a cognitive model?
My response is very simple and direct:

cK¢ does not propose a framework to construct cognitive models. It does not pretend to model an "approximation to processes of humans’ mental activities" and do not ambition to be "capable of explaining mental processes or interactions among them", eventually it does not aim at answering a specific question such as "how do we learn to categorize perceptual objects?"

Yet, cK¢ has a very strong relation to the learner by being focused on his or her interaction with a learning environment (more precisely the "milieu"). Indeed, cK¢ could contribute to a cognitive approach, but it is not its objective in the first place.

Based on the evidence we can get from the learner's activity, the objective is to characterize it in terms intelligible from a mathematical perspective and which can serve as inputs to take teaching decisions. Two types of evidence are easy to get: representations manipulated by the learner and operators he or she uses in order to achieve a task or to solve a problem. Actually, these operators are not always explicit but it is not impossible to have an interpretation of the learner's behaviours which makes sense from a mathematical perspective (this corresponds to the Vergnaud coup de force when he coined the concept of "theorem in action"). It is then reasonable to claim that we have a picture of the learner understanding when these representations and operators are stable within a problem-space. This has to be completed by a description of the means the learner uses to take a decision about the validity of his or her activity and the related outcomes. It is the idea of the control structure. Once we have a characterization along these four dimensions, we can conjecture a mathematical meaning, but this does not tell what are the related mental activities or cognitive structures as psychology or neuroscience would understand them. It is very likely that different learning theories would shade different lights on these characterizations. However, my claim is that such characterizations are sufficient to assess the so-called mathematical understanding, and to take teaching decisions or  to design learning environments.

For the rest, cK¢ shares many of the scientific characteristics of "cognitive models": it is based on "rigorous methods", it is "capable of generating testable predictions" and of generating descriptions in "formal, mathematical or computer, languages". It does not describe processes but  nothing prevents it a priori to contribute to such descriptions, this is something to explore.

Eventually, it is important to realize that cK¢ does not ambition to construct models to respond to the question "How does a child transition from additive reasoning to multiplicative reasoning?" but to the question "What are the optimal conditions to initiate and support the child transition from additive reasoning to multiplicative reasoning?"

Teaching, an emergent property of learning environments

Teaching, an emergent property of learning environments - IST 2000

I first presented this view of teaching in the context of the design of learning environment in 1999 on the occasion of a EU-US conference in Stuttgart (see the notes here and there). This new version was prepared for a talk at IST 2000 held in Nice; it includes outlines of the project Baghera which was emerging:

The project Baghera, a leading project of the Leibniz Laboratory, has the objective of shaping and experimenting radically new perspectives on the design of eLearning environments. First, by eLearning environment we mean not only the technology but the whole complex constituted by the machinery, its users and its environment. Second, it is the project basic belief that the complexity of human learning can be faced only if the design of eLearning environments takes the collaboration between artificial and human agents as a foundational principle. This requires a strong pluridisciplinary approach at every stage of the design and of the implementation.

A platform like the one we look for, is structured by several different types of interaction and cooperation: between teachers and artificial agents, between human teachers with the mediation of the technology, but also between learners mediated by the technology. Indeed we must add the interactions between learners and teachers either in an asynchronous mode or in telepresence, and between learners and the learning environment. Learning does not occur because of one specific type of interaction, but because of the availability of all of them. One type of interaction, or one type of agent, being selected depending of the needs of the learner at the time when the interaction is looked for, as well as of the specific characteristics of the knowledge at stake.

Then, the learning environment, constituted by content specific resources and conception specific resources (taking into account the variety of learners possible conceptualisations) gets its teaching power not from the property of one of its components, but the emergent property of the interactions of all the agents involved—either artificial or human, learners or teachers. In this approach the crucial issue is not that of the genericity of the technological environment (which is always obtain to the detriment of its cognitive and epistemological specificity), but of its adaptability and openess to change.

May be this is just rediscovering that education has never been the result of the action of one isolated tutor, or single intitution, but of the Society at large...

By the way, why “Baghera”? Because at the core of the system we intend to develop a society of non-human agents whose interactions will aim at the education of a human learner. But unlike the famous story, this time some human agents will take part in the adventure…

cK¢ , the Chicago talk and Guershon Harel questions

cK¢, a model to reason on learners' conceptions 
 

This presentation at PME-NA 2013 was followed by comments and questions from Guershon Harel on the invitation of the conference organisers.  I publish below with Guershon's permission his reaction on the talk, and will respond to his questions in coming posts on this blog.

Questions Inspired by or Generated from the cK¢ Model Presentation
Guershon Harel, University of California at San Diego

I would like to thank the program committee for inviting me to react to Nicolas Balacheff’s plenary talk. I have known Nicolas for many years, both professionally and personally. I feel honored to have the opportunity to react to his work.
A fundamental human nature is that not only do humans seek to resolve puzzles, but also they seek to be puzzled. Scholarly work, thus, is judged not only by the questions it answers but also by the questions it generates. Nicolas’ paper—of which the talk you have just heard is part—does exactly that: It addresses fundamental questions about learning and thinking and at the same time generates new questions.
A strong feature of Nicolas’ work, in general, and of this paper, in particular, is its attempt to define concepts and ideas rigorously. This puts the reader in a mood to follow suit, by asking questions of rigor as well.
What I will do in the time allocated to me is to share with you some of the questions Nicolas’ paper generated for me as I tried to build a coherent image of the cKc model. It is possible that the image I constructed is entirely idiosyncratic, not coinciding with the image—or better say conception—intended by Nicolas.
Whatever the case may be, I highlight that the sole purpose of the questions I present before you now, is to generate discussions, with the hope that they would further understanding, generate research studies, and advance effective classroom implementations of the cKc model. Balacheff’s paper is about a “[cognitive] model of a learner”. The adjective “cognitive” is important here to differentiate it from other types of models. So, following the rigorous style of the paper, the first question one might ask is:

1. What is a cognitive model and what are its purposes?

Briefly, and aggregately, the essential characteristics of “cognitive model”, as they appear in the literature include the following:

a. Cognitive models are approximation to processes of humans’ mental activities, such as attention, understanding, inferencing, decision making, etc.
b. They are derived from basic principles of cognition, such as a particular theory of learning.
c. They are based on rigorous methods of elicitation of cognition.
d. They are capable of explaining mental processes or interactions among them.
e. They are capable of generating testable predictions, both quantitative and qualitative.
f. They are described in formal, mathematical or computer, languages.
g. They aim at answering a specific question; for example: how do we learn to categorize perceptual objects? Such as:

i. How does a student learn to categorize problems according to their mathematical structure?
ii. How does a child transition from additive reasoning to multiplicative reasoning?
iii. How does one learn to categorize paintings according to the periods to which they belong?

h. They may target cognitive processes or cognitive states.

For example, the question, “What are humans’ categories of perceptual objects?” is a question about product rather than process. Likewise, the question “What are students’ proof schemes?” is a question about state, not process.
To illustrate the difference between these two types of models, I mention two examples of works many of you are familiar with. These are the seminal works of Marty Simon and Jere Confrey. What sets the research programs of Marty and Jere apart from many other works is their focus on the mechanisms that account for conceptual learning: namely, the transition from one conceptual state to another.
So relative to this background and characterizations, the questions one might ask about the cK¢ model are:

2. Is the cK¢ a cognitive model?

Or less rigidly,

3. To what extent is the cK¢ a cognitive model?

4. Is the cKc a model of learning processes or learning states?

Furthermore, given the unique nature of the mathematics discipline among the various disciplines, and given the complexity of the classroom setting, in general, and that of mathematics classroom, in particular,

5. Is the cK¢ a model of a learner (period), a model of a learner learning mathematics, or a model of a learner in a mathematics classroom setting?

As mathematics educators, we are most interested in the interactions among the three models outlined by Balacheff: the model of the learner, the model of the content to be learned, and the model of pedagogy. Nicolas indicates “For the last two [models], research has constantly been very active with some promising progress. On the contrary, modeling the learner proved to be a real challenge.” Two questions of interest, though they perhaps go beyond the scope of the paper, are:

6. What exactly are the challenging aspects of modeling learning relative to modeling content and pedagogy?
7. What are the interdependent relationships among these three models?
8. What is the efficacy of such models if they are constructed independently from each other? In particular, can models of content and pedagogy be viable without the presence of a learning model?

A more philosophically oriented, yet critical, question is

9. Are cognitive models of thinking possible?

This question is derived from the third characteristic of mental models I listed earlier; namely, a mental model is based on a rigorous method of elicitation of cognition. This characteristic is particularly problematic. Here is why. The cK¢ is a model of learning/thinking. As was pointed out by Colin Eden, “if we take seriously Karl Weick’s aphorism that we do not know what we think until we hear what we say, then the process of articulation—that is, the learner’s utterances and behaviors that constitute the data for the construction of the model—is a significant influence on present and future cognition. Since articulation and thinking interact, as is largely accepted, then an elicitation of cognition that depends upon articulation is always out of step with cognition before, during, and after the elicitation process.”

Even if we overcome this philosophical hurdle, an empirical question emerges:

10. To what extent can a general learning model be viable, given human diversity of character, culture, and circumstances?

The fourth component of the cK¢ model is control. Balacheff characterizes control under the general umbrella of metacognitive behaviors. The control component is crucial, and is Balacheff’s significant addition to Vergnaud’s model. It is crucial because it is the place where issues of the learner’s understandings are to be revealed. The set of four examples Balacheff discusses to illustrate the cK¢ models are illuminating, but I still found myself wanting to better understand the cK¢’s definitions and treatment for crucial control constructs such as understanding, meaning, and ways of thinking.

These are crucial constructs with various instantiations. For example, when we talk about “understanding” and “meaning”, we—researchers and teachers—want and need to distinguish, for example, between “understanding in the moment” and “stable understanding”, and between “meaning in the moment” and “stable meaning”. Likewise, we want and need to observe ways of thinking, or habitual anticipations of meanings, both desirable and undesirable. Thus, it is natural to ask:

11. What are “understanding,” “meaning,” and “way of thinking” for the cKc model, and what is a reliable methodology to elicit them?

12. What is “Problem” for the cK¢ model?

Recall that Balacheff’s definition of “conception” is a quadruplet (P, R, L, Σ). Balacheff recognizes that the first component, Problems, is problematic; namely, he faced the question as to how to characterize the set of the problems for a particular conception. After considering two possible characterizations, one by Vergnaud and one by Brousseau, Balacheff describes P as a set of problems prototypical to the field to which the conception belongs. This characterization raises theoretical, methodological, and instructional questions.

Specifically, the cK¢ model postulates that problems are the source and the criteria of learning and knowing. And following Vergnaud and Brousseau, problems are also held as the engine of the teaching process. A consequence of these largely agreed upon positions is that the cK¢  hinges upon the school prototypical problems one chooses to elicit conceptions.

The difficulty that arises here is that many of these prototypical problems are alien, not intrinsic, to the students. The students might be able to solve them, but the kinds of perturbations they engender with the students are didactical, aimed solely at satisfying the will of the teacher. Thus:

13. If the problem is alien to the learner, what meaning can a researcher give to the operations and control components of the model?
14. How is to be determined by researchers, and more importantly by teachers, whether the problem posed to elicit conception is intrinsic or alien to the learner, and how does this determination effect the observer’s conception of the learner’s conception?

The Problem component is also a crucial factor in Balacheff’s definition of generality. Generality is one of the factors in the cK¢ model shaping relations between conceptions. As such, it is crucially important, for the simple reason that it provides a criterion for conceptual development; namely, how one conception is more general than other.

Balacheff defines generality as follows:

C=(P, R, L, Σ) is more general than C’= (P’, R’, L’, Σ’) if there exists a function of representation ƒ: L’→L so that ∀p ∈P’, ƒ(p)∈P.

The examples of relative generality discussed in the paper work nicely according to this definition. Balacheff’s definition also worked well with many of the examples I tested. For some
cases, where P=P’, the definition may need further refinement. Consider the following example:

A 13-year-old girl, Tami, and an 8-year-old boy, Dan, were interviewed in pair.

Interviewer: One pound of candy cost $7. How much would 3 pounds of candy cost?
Tami: Three times seven, 21.
Dan: I agree, three times seven.
Interviewer: What if I changed the 3 into 0.31? What if the problem were: One pound of candy cost $7; how much would 0.31 of a pound cost?
Tami: The same. It is the same problem, you have just changed the number, 0.31 times 7.
Dan: No way! It isn’t the same. Can’t be [angrily]. It isn’t times. Why did you [speaking to the interviewer] agree with her?
Interviewer: I didn’t agree with her, I’m just listening to both of you. How would you solve the problem?
Dan: You take 1 and you divide by 0.31. You take that number, whatever that number is, and you divide 7 by that number.

Indeed:

On the one hand, the set of problems belonging to Tami’s conception is identical to set of problems belonging to Dan’s, and it seems that there is always a translation between the corresponding L and L’ satisfying Balacheff definition of generality. Hence, the two conceptions seem to be equivalent. On the other hand, intuitively, I want to attribute a greater generality to Tami’s conception, with all the great admiration I have for Dan’s conception.

In closing,

Three of Balacheff’s goals for introducing the cK¢ can be summarized as follows:

a. Make more efficient our own research.
b. Clarify concepts and their relationships.
c. Contribute to better understanding of learners’ understanding, so as to support decision making for teachers and learners.

It is against these goals that I chose the questions I have just presented.

Thank you

The Arabic translation of the terms and expressions of the TEL Dictionary has been released

The Arabic translation of the entries of the TEL Dictionary has required a considerable effort which demonstrates once more that questioning the Technology Enhanced Learning vocabulary from a multiple perspective is needed. This time five expressions have not found a satisfactory translation: Constructionism, Pedagogical agent, Virtual pedagogical agent, Animated pedagogical agent and Programmable course. The case of "constructionism" is not a surprise, it is in fact in all languages not translated but transliterated. For the next three cases, it is the term "agent" which is resisting. In my opinion,  the last case should find a solution soon. Suggestions and contributions are welcome, for this purpose LinkedIn members are invited to join the LinkedIn "TEL dictionary initiative" group.

The terms documented by the TEL Dictionary are now available in...
English  • العربية • Български • Dansk • Español • Ελληνικά • Eesti • Français • Magyar • Italiano • 日本語 • 한국어 • Nederlands • Português • Русский • Slovenčina • Türkçe • Tiếng Việt • ‪中文(繁體)‬ • ‪中文(简体)‬