[ACT-R-users] Model of writing

[db30 at andrew.cmu.edu]

--On Wednesday, August 25, 2010 6:02 PM +0200 Michael Carl <mc.isv at cbs.dk> 
wrote:

> Thank's Dan!
> yes, you are right one would have to take into account the hands-to-home
> overhead so that the difference of the data and the ACT-R model becomes
> only 75/180ms to 200ms (for the "rt" example), which sounds less serious.

That initial time isn't hands-to-home overhead because the fingers start
on the home row.  It's the preparation and execution of that first keypress
itself -- 450ms to hit the "r" from a cold start.  In the press-key example
it also takes 450ms from the start of the production which requests the
"t" be hit until that key is struck.  However those times are overlapping
because the motor module is allowed to prepare one action while executing
the previous, and in this case that resulted in an inter-key time of only
300ms.

> Nevertheless, I have also tried to use different fingers in the lisp code
> which would type the same sequence, but the 200ms seems to be hard wired.
> Does that mean that currently in ACT-R a one-finger typist has the same
> performance than a
> two finger typist than a 10 finger typist?
>

200ms would be the lower bound for successive peck actions with the default
motor module parameters, but it's not a fixed value because the time for
the motor action includes a Fitts's law calculation for the distance the
finger has to move.  Thus if the model were to only use one finger it would
have higher execution costs because of the increased travel distance vs
a model which used multiple fingers traveling shorter distances.

Interestingly however, if the model were to continually peck with that one
finger (instead of pointing-at and punching) the one finger model would 
have
lower preparation costs since it'd always be the same action.  I think the
added execution cost would dominate that preparation savings on average, 
but
I'll actually have to run a test now because that's an interesting question.

> I just read on page 293 of the ACT-R manual (section Isa press-key) that
> you describe a theory and a learning model for typing, maybe this might
> resolve those problems?
>

If it were possible to write the model which did what's described there
(essentially learn how to do press-key instead of have it as a built in
construct) then that might indeed be useful to you.  However, a critical
piece of that hypothetical model is something which ACT-R's motor module
currently isn't able to do -- learn.

Dan