Piaget made strong claims about children's cognitive competence at different ages (e.g. 4-yr-olds typically fail at 3-mountains problem and are thus egocentric thinkers) -- but more recent work shows that children possess some of these competencies if the demands of the task are reduced (e.g. 3-yr-olds can take other perspectives on this problem if they are given ways to express the terms 'left' and 'right') -- this reminds of me of vygotsky's zones of proximal dvpmt where children can accomplish more with help than they can do by themselves. I wonder if there is variability in how much a child can do on his own vs how much he/she can do when receiving help at different ages -- perhaps children at age 5 will benefit from help much more than children at age 9 -- is there evidence that children have variability in the amount they benefit from receiving help or is it assumed that children will benefit from assistance the same amount regardless of their age. This also relates to implications for instruction -- when children solve problems, how much help should we give them and how much should we let them struggle through the problem on their own at different ages? -bryan
How much can we really conclude if we have a computational model of how some developmental phenomenon occurs whose performance over time matches the behavior of children? Can we say that young children's minds must work the way the model does or could they be using some fundamentally different architecture that arrives at the solution through a different path? How necessary is it that the learning algorithms of the model be biologically plausible if we wish to use it as a model of real children's thinking?
Piaget asserted that sensorimotor actions prepare the infant for intellectual operations. If so, do infants with sensorimotor disorders acquire intellectual operations differently from healthy infants?
How does one discriminate between a single rule which underlies multiple behaviours (e.g., conservation of mass / size) versus the simultaneous development of multiple related rules?To what degree is it possible to causally manipulate performance on a very specific task and conclude that there is a unique rule underlying that task ( as opposed to simply having primed the rule in that particular context)?
Fischer & Hencke (1996) contrast the "nature" vs. "nurture" perspectives present in Piaget's work.To what degree do these arguments still play out today?Why is the argument that there is an interaction which produces something special and orthogonal to the individual contributions of nature and nurture so difficult to appreciate.Is this question even worth asking anymore?
Gopnik (1996) suggests that modular (i.e., innate cognitive structures) systems can account for development because there are multiple possible combinations of modules which can be activated via different input parameters.Cant this theory easily be falsified then by showing that monozygotic twins could be differentially trained t o perform differently on two tasks thought to rely on the appearance of the module (e.g., teach one to perform correctly on a mass conservation task)?
From the Munakata chapter.A universal form of computational modeling has yet to emerge, and the different flavors of modeling which currently exist each have their particular strengths and weaknesses.Why is it so difficult to reconcile these models, such that a single (or all) frameworks can be used to model any task.If they cannot be reconciled at some level, does this mean that some of these architectures are fundamentally flawed (e.g., abstraction might not be possible in a pdp network because the brain does not abstract in the way some cognitive psychologists would have us believe we do? )?
How much can we really conclude if we have a computational model of how some developmental phenomenon occurs whose performance over time matches the behavior of children? Can we say that young children's minds must work the way the model does or could they be using some fundamentally different architecture that arrives at the solution through a different path? How necessary is it that the learning algorithms of the model be biologically plausible if we wish to use it as a model of real children's thinking?
My take on this would be that models aren't sufficiently sophisticated to be a definitive explanation of behaviour. Rather, they are existence proofs that the brain could work in this manner. Thus, to the extent that they are at least not biologically unplausible and correctly capture behavioural data they can be useful. If we don't aren't forgiving in this fashion, modeling efforts would by default have to halt because we simply don't know enough about the brain to do anything but a very coarse approximation of how it functions.
I was wondering how different strategies might affect the processing of information in working memory. We have evidence from Ericsson, K. A. & J. (1989) as well as others, that working memory is more or less flexible and is influenced by expertise. I was wondering if people vary in their working memory capacity/processing based not only expertise, but also on other strategies and if that could explain part of the variance in the population as dissociated from cognitive ability, or innate individual working memory capacity.
Most information processing theories try to give the description or explanation of development for "all" children. How to explain the huge individual difference in the development?
In the Munakata review, the comparison of different developmental modeling approaches takes pains to describe examples of so-called "unified models", even while discussing the relative modeling strengths and weaknesses of the various approaches. If an approach (such as a production rule modeling) does not deal appropriately with emergent cognitive phenomena or other characteristics of development, or provide a general method for generating predictions about those data, can the model be considered "unified" in any sense*?
*I'm not suggesting that the only "unified" model is a perfectly specified one, just that eliminating whole grain sizes of analysis seems unwise especially given the evidence in the dynamic systems model section describing such emergent phenomena in development! --Nora
This is a general question about Klahr's Production system model. 1) I was wondering how to test or verify the steps of thinking (in the time line component). Although it often occurs automatically, it still seems that it requires a large memory load to track the events in the time line and to eliminate redundant processes. 2) Is there any room for the explicit feedback that children can get to facilitate their learning with limited experience (such as 3 year olds learn as 4 year olds do)?
One aspect of development that I do not see addressed explicitly in some of the information processing models is the interaction between multiple developing systems, or between multiple aspects of a single system. This is perhaps most clear in the example of MacWhinney's connectionist model of how children learn the German system of articles (1989). It is not the experience of a German child to have novel instruction all at once, as when you turn on the model. The child has had many more exposures to German articles before they finally "get it", and perhaps more importantly, these exposures are spread out over a long time. Children may be encoding some features during this time period that will allow facilitated processing change at the time at which they begin to use the articles correctly. How well the eventual processing change will be facilitated will depend on a number of factors, including what kinds of features they may have encoded from their past exposures to the articles, and also the constraints of other partially developed systems, such as linguistic understanding. For example, if a child has an underdeveloped semantic understanding of the noun, then the development of correct article use for that noun may naturally be delayed. Understanding the meaning of various nouns is a gradual process, and it is not the case that all words are understood equally. This is unlike the connectionist model, which seems to assume that all input to the model is equal, or "understood" to the same level. A more general example is the question of the impact on the development of otherwise unrelated systems, like attention and focus and how rate of development in this system might affect development of the process in question, correct German article use. It seems like multiple systems must be interacting to form the resulting developmental trajectory for each individual child, even when the outcome may appear similar across the group. Ultimately, this connectionist model revealed many interesting behaviors that seem to map onto the behavioral data, but it seems like it could do an even better job with some kind of long-term experience variable and the inclusion of factors relating to the developmental state of other potentially related processes. (or maybe people do this and I am unfamiliar with the work?)
A more practical question: In a Production System, how does one set parameters for "age", as in the Simon & Klahr (1995, p.91) system to solve number conservation problems? How are the 3-year and 4-year old models different? How are they defined?
