[ACT-R-users] Model of writing

[db30 at andrew.cmu.edu]

My message was really only intended as a suggestion to Bonnie as a way
that they may be able to speed up the typing extension which they've
implemented for CogTool given the assumptions that she listed.  I don't
think on its own the peck action would be sufficient for your work, and
I don't know enough about the typing literature to offer any advise as
to how best to model it.

Given that you did run that model however, I want to make sure that
you are comparing the right times from the model to your data.  So,
first let me make sure I understand what you're measuring.  Are the
75ms and 180ms times for typing "rt" that you list the time between
the "r" and the "t" hits or the total time from when the person is
instructed to type such a sequence?  I assume that it's the first of
those, in which case the corresponding times from the two different
mechanisms in the model run would be 300ms and 200ms.  Those are the
differences between the times of the output-key actions which indicate
when the keys are actually hit by the model.  If you wanted to measure
the second of those (total response time until the second press) then
the times would be 750ms and 650ms because that is the time from the
first production in the sequence to the striking of the second key.

Dan

--On Wednesday, August 25, 2010 11:23 AM +0200 Michael Carl <mc.isv at cbs.dk> 
wrote:

> thanks very much for the program and the literature pointer (I will go
> through it)! I have run Dan's peck-test lisp model. The ACT-R press-key
> action in the lisp code
> produces "rt" in 900ms and the peck actions for the same string takes
> 700ms. Comparing this to our (Danish) data: there are two peaks in the
> duration distribution for "rt": one around 75ms and another one around
> 180ms (there are more instances at the second peak), while for instance
> the most frequent bigram "er" (on the keyboard just to the left of "rt",
> and more likely typed with two fingers) has most occurrences around
> 60-90ms. A frequent trigrams like "ing" (even in Danish:-) is most of the
> time produced in 230-250ms and we also have some 6-grams produced in less
> than 600ms (always for n-1 inter-keystroke intervals).
>
> In principle, if two or more fingers are involved there could be any time
> lapse between the keystrokes, zero and even negative when the keys occur
> in the reverse order e.g. frequent typo hte (but maybe you would model
> that differently). If you splash both hands on the keyboard then almost
> all keys could be produced simultaneously - all of which shows that
> typing actions go on in parallel as Susan mentions.
>
> There does not seems to be an easy way to simulate something like this in
> ACT-R, even if we have the distributions and given also that there is no
> indication which fingers were used for a particular typing event in our
> data.
>
> Any suggestions?
>
> Michael
>
>
> 2010/8/24 Bonnie John <bej at cs.cmu.edu>:
>> Susan, what you say is certainly true, but adding my own bit of history,
>> part of my thesis compared my serial approximation to the Rumelhart and
>> Norman PDP model and I matched the inter-keystroke time for digraphs
>> better than they did. My serial model had an average absolute percent
>> error of 0.9% and theirs was 5.7%.
>>
>> That part of my thesis used a model of each finger's position and a very
>> simple approximation of where all the fingers on the hand move to when
>> one finger hits a key. Then it just used Fitts Law for the horizontal
>> movement to the next key. This is at a lower level than what we are
>> currently doing in CogTool/ACT-R, but it could be done in ACT-R if we
>> programmed the fingers to know where they are in space and move
>> according to the algorithm in my thesis. Anybody want to implement that?
>>
>> Bonnie
>>
>> Reference for the serial typing algorithm:
>> John, B. E. (1996) TYPIST: A Theory of Performance In Skilled Typing.
>> Human-Computer Interaction , 11 (4), pp.321-355.
>> Figure 16 and 17.
>> The PDP model I compared to was in
>> Rumelhart, D. E. & Noman, D. A (1982) Simulating a Skilled Typist: A
>> Study of Skilled Cognitive-Motor Performance. Cognitive Science, Volume
>> 6, Issue 1, pages 1?36.
>>
>>
>> Susan Chipman wrote:
>>> I thought I might remind people that typing was one of the first
>>> behaviors modeled by the PDP folks. They had data showing that
>>> multiple typing actions went on in parallel -- that is, the actions of
>>> future fingers were beginning before the current action was completed.
>>> Don't know if these ACT-R models are dealing with that.
>>>
>>> Susan Chipman
>>>
>>> On Tue, Aug 24, 2010 at 3:29 PM, <db30 at andrew.cmu.edu
>>> <mailto:db30 at andrew.cmu.edu>> wrote:
>>>
>>>
>>>
>>>     --On Tuesday, August 24, 2010 1:18 PM -0400 Bonnie John
>>>     <bej at cs.cmu.edu <mailto:bej at cs.cmu.edu>> wrote:
>>>
>>>         We have a lower-level model of typing implemented in ACT-R tht
>>> is         "under-the-hood" of CogTool. It is a mixture of my ages-old
>>>         PhD thesis
>>>         and what we cold do in ACT-R without changing the entire
>>>         structure of
>>>         its hand and fingers. So it still has remnants of ACT-R's typing
>>>         assumptions, like the the hand always goes back to the
>>>         home-row between
>>>         each keystroke, but we have relaxed some of the other
>>>         assumptions in the
>>>         standard ACT-R typing model and so have sped it up to being
>>>         about a 40
>>>         wpm typist instead of the 20 wpm typist it is in the general
>>>         release.
>>>
>>>
>>>     One note to make about that is the only assumptions about fingers
>>>     returning
>>>     to the home-row are with the use of the press-key and peck-recoil
>>>     actions.
>>>     If one programs the specific finger movements with peck and punch
>>>     actions
>>>     then the fingers will stay at the key that was hit. If you're not
>>>     already
>>>     taking advantage of that you may be able to speed up your CogTool
>>>     typist
>>>     even further. Of course the complication is that to do that you
>>> would     also have to have something that computes the necessary
>>> geometry     from the
>>>     current finger position to the target key instead of just the
>>>     home-row to
>>>     target key geometries which are available from press-key.
>>>
>>>     As a simple demonstration of that, attached is a simple model
>>>     which types
>>>     two keys in sequence using the same finger twice. The first time
>>> using     two press-key actions and the second using explicit peck
>>> actions. The     inter-key time for the second pair is less than for
>>> the first and that     should be true for all valid one-finger pairs.
>>>
>>>     Dan
>>>     _______________________________________________
>>>     ACT-R-users mailing list
>>>     ACT-R-users at act-r.psy.cmu.edu <mailto:ACT-R-users at act-r.psy.cmu.edu>
>>>     http://act-r.psy.cmu.edu/mailman/listinfo/act-r-users
>>>
>>>
>>> ------------------------------------------------------------------------
>>>
>>> _______________________________________________
>>> ACT-R-users mailing list
>>> ACT-R-users at act-r.psy.cmu.edu
>>> http://act-r.psy.cmu.edu/mailman/listinfo/act-r-users
>>>
>>
>> _______________________________________________
>> ACT-R-users mailing list
>> ACT-R-users at act-r.psy.cmu.edu
>> http://act-r.psy.cmu.edu/mailman/listinfo/act-r-users
>>
>>
>>
>>
>>
>