Discussions > May be a question for computer-scientists...
May be a question for computer-scientists...
|
Nicolas Balacheff
105 days ago |
Let's imagine that we have a model, say a learner model (LM). We give it to n teams of develpers. It is very likely that we will get n implementations differing by some aspects. Right?
If LM is complex enough, I would claim some of these implementations (at least 2) will be significantly different. Imagine that we "model back" the output of these implementations, the question is: will they be identical to LM?
|
|
Fridolin Wild
104 days ago |
Why are you asking? Sure: the abstraction of the real world (=model) is not equal to the real world and any model is a reduction. If - when mapping back - the original model is not resulting again, then maybe the original model did not have much fitness?
|
|
Nicolas Balacheff
104 days ago |
The issue is not to question the relation between the "real" world and a model of it, but between a model and the outcome of its implementation. You can look at that as a question of the relation between two representation. e.g. a learner model (M1) is a formal specification, or can be seen as such, which is implemented in a system using certain methods or technics, a language contextualised by a given OS. This implementation leads to (a component of) a software (S). You then obtain a system which has a certain behaviour which you can model (M2). The issue is the relation between M1 and M2...
|
|
Christian
104 days ago |
Hi Nicolas - sorry to disagree - from a CS perspective you are probably right but in TEL the fitness between "real" world and formal specification is crucial. Formal specifications have already a difficult time in TEL, they are often obsolete or incomplete (see IMS-LD).
A question and a comment:
- why would you want a model (M2) of a software S for which you already have a model (M1)?
- despite the question above - I think you have already given the answer: models are (always) underdetermined, hence translation back and forth is problematic – a classical issue in philosophy of language. http://en.wikipedia.org/wiki/Indeterminacy_of_translation
|
|
Nicolas Balacheff
104 days ago |
@Christian: I don't think we disagree.The challenge of the adequacy of a model to the reality it refers to is certainly essential, if not critical. But it happens that for once I am looking at another problem (again, that does not mean that I don't recognise the importance of he one you point). So, for a while I will focus on the relation between M1 and M2, where M2 is the model of the artifact which results from the implementation of M1.
Then your questions:
- There is no a priori reason to trust that the behaviour of a software is the behaviour asserted by its specifications (or seminal model M1). This is rarely questionned, and I wonder why since it is an important part of the evaluation of the learning environment. What could be the methodology to decide on the distance between M1 and M2, then to understand the impact of this difference of the expect learning.
- A model is what it is, and I don't ignore the translation indetermination, but this is not the point. The point is the impact of the implementation on the nature of the model, to express it in an other way.
@Fridolin: Where is my questionning from? I was discussing the concepts related to ITS, and mentionned the existence of two expressions "learner model" and "learner module". Do we need both? Some may say "no", but other would say "yes" -- and I am one of them for the reasons this discussion will hopefully clarify. At a certain stage, this issue will be important to understand the complexity and the stake of multidisciplinarity in our research area.
|
|
Fridolin Wild
103 days ago |
For me this is always a question what the model is supposed to explain. I think it's crystal clear, that e.g. a sequence and use case model are fundamentally different to - say - an instructional design model (although, given the expressiveness of its standard representations in UML and LD, in this case it would not be too challenging to map one to the other; it may just not be natural). Similarly, a use case model is fundamentally different from a user interface model (be it Balsamiq or Himalia). All three models - software specification, instructional design, user interface - may be implemented in the same software, but their models would cover different aspects and perspectives.
But even if you only consider the software modeling aspects, there are huge differences between model specifications and their actual implementation models. For example, Zuzana Bizonova has written her dissertation about model driven architectures and LMS integration, here is a short article of hers introducing into that topic of computation independant models vs. platform independant models vs. platform specific models in LMS integration.
Btw: in general, I think this topic is very well covered in Pierre Tchounikine's recent book "Computer Science and Educational Software Design".
|
|
Nicolas Balacheff
103 days ago |
I appreciate your remark pointing that different models are cooperating in a learning environment, so just to make it simple I considered just one (whichever). If we agree that there are such differences between a model and its implementation, then which methodology to evaluate it? and which consequence on the way we organise our research. Eventually, it means that we need different concepts for the design and the implementation and be able to relate them and document the impact of this relation. Hence, "learner model" and "learner module" are not synonymous and their relation not that easy to qualify. We need some more examples of that kind.
|
|
Nicolas Balacheff
102 days ago |
@Fridolin -- You refer to Tchounikine book, I think that the most relevant excerpt of this book in relation to this conversation is: "We define traceability as the maintenance of the correspondence between (1) the theoretical element used as foundation (if any), (2) the used model and (3) the pieces of code (components, architecture, etc." ([there] p.154). Pierre clearly points that to express a model we need means of representation, and he comments a bit on the fact that these means must ensure a certain richness of expressivity while unlike computational models being accessible to human reading (as it were). As Christian emphasised, the relation between the model expressed at a knowledge level and the model at a computational level is under the general constraints of "translation". Steiner would even speak of a "transmutation". But the point is, beyond this remark, that we may need to include in our approach a systematic questioning of this relation and be prepared to see emerging different conceptualisations of both sides.
|
