Explicitness of Declarative Chunks

[Troy Kelley]

Good reference Frank, so Ericsson and Simon review some of the expert
elicitation methods and note that some of the research supports the view
that complementary methods can uncover some of the deeper more
proceduralized knowledge of experts, even though Ericsson and Simon still
would rather just use verbal protocol analysis elicitation method.  They
reference Burton et al. (1988) who compared four elicitation methods
(formal interview, protocol analysis, laddered grid and card sort) and
found that protocol analysis produced fewer of the rules given by domain
experts than the other three elicitation methods.  They also note that this
task was a perceptual task (recognition of rocks by geologists) - so, much
like riding a bike it is difficult to discuss perceptual tasks.  They also
reference Brown and Day (1983) who found that experts in English, "were
unable and unwilling to give precise rules for summarization during a
standard interview.  However, summarization rules could be readily
extracted from these subjects' thinking-aloud protocols, produced while
they were actually summarizing sample text".   Ericsson and Simon's general
viewpoint here is that concurrent verbal protocol analysis is a productive
methodology compared to other techniques, but they also suggest that given
the right knowledge elicitation process, experts will elicit more of their
underlying procedural knowledge.  So the distinction between what experts
are aware of, and what they are not aware of becomes more complex.

Ericsson and Simon go on to say that, "one kind of information that
subjects clearly cannot report, because it is not available in STM, are the
cues that allow them to recognize stimuli.  The result of the process of
recognition (i.e. the thing recognized) is heeded and can be reported, but
not the intermediate steps in the recognition process.  Inability to report
recognition cues must not be confused with failure to report contents of
STM."  But here again, the recognition of the stimulus is assumed or
implied.  But I am almost certain, that if asked, an expert could surely
offer details about the recognition of stimuli important stimuli.  So
again, is the recognition procedural?  I would say yes.  Is the expert
unaware of the rules governing the recognition of the procedural knowledge?
In at least some cases, depending on the task, I would say no.

Troy Kelley
Army Research Laboratory.






"Frank E. Ritter" <ritter at ist.psu.edu> on 12/14/2001 11:40:16 AM





To:    Troy Kelley <tkelley at ARL.ARMY.MIL>, Jerry.Ball at williams.af.mil
cc:    act-r-users at andrew.cmu.edu
Subject:    Re: Explicitness of Declarative Chunks


A very good book in this area is Ericsson and Simon's Verbal Protocol
Analysis.  It is essentially a theory of how and what people can talk
about, and how they do it.  It is worth getting in hardcover, if you
can, because it is very useful.

Cheers,

Frank

At 10:25 -0500 14/12/01, Troy Kelley wrote:
>Jerry,
>
>First, it is good to see someone posting to the ACT-R list.  I think you
>might want to look at the literature and see if any there are any
>experiments which look at just what people are aware of as explicit
>knowledge and what they are not aware of.  This seems to be a pretty good
>question for a few experiments because, off the top of my head, I don't
>think many people have looked at what "explicit knowledge" constitutes and
>what it excludes.  There is some evidence, from knowledge elicitation
while
>developing expert systems, that experts are not completely aware of their
>knowledge, but I don't think anyone has really looked at this in any great
>detail, i.e. *exactly what* are they not aware of, and why they not aware
>of some things, but aware of other bits of their knowledge.
Theoretically,
>to carry this a step further, I have some trouble with the notion that
>people are completely unaware of procedural knowledge.  Sure it is
>difficult to explain how to ride a bike, but that is because it is a motor
>skill, and motor skills might not translate into words as easily as
>something more cognitive, like playing chess for example.  I think if you
>force people to think about what they know, more and more of the knowlege
>comes out.  I am not sure if the simple fact that some things are
difficult
>to talk about, and difficult to quantify, perhaps because someone has
never
>had to talk about some aspect of their expertise before, that this
>necessarily means that this knowledge is somehow unavailable or not
>explicit.
>
>These are interesting questions though, and a good area for research.
>
>Troy Kelley
>Army Research Laboratory
>
>
>
>
>
>
>Jerry.Ball at williams.af.mil on 12/13/2001 02:59:20 PM
>
>
>
>
>
>To:    act-r-users/@andrew.cmu.edu
>cc:
>Subject:    Explicitness of Declarative Chunks
>
>
>It has been suggested that declarative chunks are explicit in that humans
>are consciously aware of the contents of declarative chunks and can
reflect
>on their content.  However, it is not clear to me that the "type" of a
>declarative chunk is something that can necessarily be reflected on. Thus,
>if a human has a declarative chunk type  "ISA noun", it is not necessarily
>the case that the human can reflect on that type. Humans may know that
>words
>belong to various categories without explicitly being able to reflect on
>what those categories are. If, on the other hand, the Part of Speech of a
>word is encoded in a slot with the type of the declarative chunk being
>something like "ISA word", then the same argument holds for the slot
>containing the POS.
>
>For example, given
>
>(man isa noun
>    word man)
>
>or
>
>  (man isa word
>      word-form "man"
>      word-root man
>      word-type noun)
>
>Although the type "noun" is encoded in the declarative chunks, knowledge
of
>the type "noun" remains implicit.
>
>Jerry


--

Frank  Ritter at ist.psu.edu
School of Information Sciences and Technology
The Pennsylvania State University
004 Thomas (Basement)
University Park, PA  16801-3857
ph.  (814) 865-4453   fax (814) 865-6426
http://ritter.ist.psu.edu