Re: a request (with a bit more)

Judy Diamondstone (diamonju who-is-at rci.rutgers.edu)
15 Mar 1998 14:55:30 -0000

John, a million thanks for the thoughts, leads, and ending provocation

["My Question: Why isn't the material basis for learning (in-the-head
learning) interesting?" ]

Judy

At 09:06 PM 3/14/98 -0400, you wrote:
>Judy (and xmcaites),
>
>Though I am not sure, I suspect that your student has run across references
>to "Parallel Distributed Processing" possibly through references to a
>semi-famous debate in cog science journals centered on modeling the
>acquisition of the english past tense.
>
>There is a fair amount to be untangled here: Parallel Distributed
>Processing has become a term applied only to one school: centered around
>Rumelhart and McClelland and the former PDP group originally at UC San
>Diego. (I've always wondered about this.) Connectionism has become the
>favored academic term and that is the way that the most recent references
>will be indexed. It should be distinguished from "Neural Networks" which is
>a highly technical, and technologized offshoot which is unlikely to be
>helpful.
>
>Literacy acquisition is not my field, and my interest in these ideas are
>through my focus on trying to understand just what it is that changes when
>we say something has been learned-and how that necessarily material process
>constrains theories of learning and instruction. So if I say the obvious
>just let me know.
>
>The seminal reference in the past tense debate is: McClelland, James L. and
>David E. Rumelhart. "On Learning the Past Tenses of English Verbs." In
>Parallel Distributed Processing, Volume 2: Psychological and Biological
>Models, eds. David E. Rumelhart, James L. McClelland, and the PDP Research
>Group, 216-271. Cambridge: MIT Press, 1986. It is in part a reaction to an
>earlier Pinker book which relies on hidden explicit rules. A variant is
>supposed to have appeared in MacWhinney's "Mechanisms of Language
>Acquisition."
>
>A reference that discusses the debate and should help open out to other
>sources and lines of research is: Bechtel, William and Adele Abrahamsen.
>Connectionism and the Mind. Cambridge, MA: Blackwell, 1991.
>
>Hope this helps, and touches on what your student is interested in.
>
>____Some laignappe--something a bit extra.
>There is an interesting bit of history here: Rumelhart is generally
>acknowledged as the formulator of the most recent version of the
>influential concept of "schemata." Rumelhart's work was then focused on how
>children learn to read for meaning and more specifically to understand how
>it is possible to "make up" and extend stories in various genres--an
>ability that he felt had to be tied to children's ability to read a text
>creatively and to insert their own meanings. (I suspect he might have been
>more cautious in his characterization but that is the way I read the
>direction.) From this, pulling on Chomsky's generative grammar, and the
>emerging formalism of Artificial Intelligence he developed a structuralist
>concept of schemata as an acquired data structure 'filled' with differing
>specifics from experience. You'll still see many pro forma references to
>his 77 or 80 work in the educational literature using this concept. (Though
>he always felt that his intuition about schemata was never matched by the
>models because the representations he was using were not active enough.)
>
>His radical revision of this idea and the reason for it has received very
>little play in educational or other practice-oriented fields. In
>Rumelhart's view the problems with his early attempts to instantiate the
>concept lay in the inflexibility of the data structures; it was hard to see
>how anything approaching the flexibility of human learning could result.
>For Rumelhart and his colleagues the solution lay in abandoning the idea of
>data structures, their problematic storage and retrieval, and dubious
>filling of default assignments altogether. Inspired by the massively
>parallel and distributed nature of the architecture of the brain they
>postulated a massively interconnected system linked to the world in such a
>deep way that representations were better thought of evoked by experience,
>with each evocation being an unique response to the situation based on
>prior experience. With the caveat that we can provide ourselves with
>evocative models and replays of experience, the notion is that the entire
>system is based on constraint satisfaction, it reorganizes in significant
>ways in response to each experience and that we don't retrieve memories but
>recreate them in terms of the immediate problematic situation. This gave
>Rumelhart the dynamic, active schemata he wanted but at the cost of what
>most folks would understand as representation. The fundamental operation is
>pattern completion, not formal logic over basically static representations.
>
>_____and some ruminations:
>When I ran across connectionist ideas it seemed to me that they were almost
>tailor-made to fit into the framework put forward by the pragmatists and I
>still think so. But the pragmatist vision was much broader and included a
>strong sense of the role of activity and "ends in view" (goals) than
>connectionists have put forward. (I have come to understand those operating
>out of Vygotskian or several other practice-oriented traditions might make
>the same objection.) In the first chapter of the book cited above Rumelhart
>et al. introduce issues of external representation, formal logic and goal
>seeking and do a credible job of laying out a program to research these
>from within a connectionist perspective. To my knowledge the issue of
>formal logic has been rather handily dealt with but the other two crucial
>elements, goals and external representation, have languished. This,
>especially external representation seems ideal for situated perspectives to
>comment on. Pragmatism certainly had a finely graded series of "ends in
>view" to add to the tool box. -----But few working in practice oriented
>traditions seem interested in the material basis of learning within folks
>(as opposed to communities) and the ways that it might powerfully constrain
>the way we theorize. The political, policy and cultural implications are
>very large--facts vanish in the connectionist story and the structure of
>testing that organizes schooling will make little "scientific sense" if
>this view prevails (and it is my judgement that it will prevail.) Nate a
>bit back introduced some suggestive material and got no response.
>
>My Question: Why isn't the material basis for learning (in-the-head
>learning) interesting?
>
>Thanks for your patience, John
>
>
>---------------------------------------------------------------------------
>John St. Julien (stjulien who-is-at udel.edu)
>Department of Educational Development
>University of Delaware
>
>
>
>

Judith Diamondstone (732) 932-7496 Ext. 352
Graduate School of Education
Rutgers, the State University of New Jersey
10 Seminary Place
New Brunswick, NJ 08901-1183