[ACT-R-users] Model of writing

[Dan Bothell]

To test the question about 1 fingered, 2 fingered, and 10 fingered
typists in ACT-R I created some test models (if you could even call
them that because they're mostly just Lisp code) which just push motor
requests through to type out sentences repeatedly for 60 seconds to
get a words/minute score (where a word is every 5 keypresses).  Those
models were then tested across the three possibilities for pipelining
of motor actions: "state free", "processor free", and "preparation free".

There were 5 total models:

One-finger is a good "hunt and peck" typist using only one finger.

Two-fingers is a good "hunt and peck" typist using both index fingers
  keeping each hand on its own side of the keyboard.

Ten-fingers is a model which uses the default press-key action to
  touch-type using all fingers.

One-finger-savant is a perfect touch-typist using only one index finger
  i.e. it can move that finger from any key to hit any other key
  perfectly as a single action, without looking.

Two-finger-savant is a perfect touch-typist using both index fingers
  where each finger stays on its own side of the keyboard.

Here's the average WPM I got based on 3 simple sentences which
each have all the letters of the alphabet at least once:

                   state free   processor free    preparation free
one-finger            13.0           19.1                X
two-fingers           13.8           20.7                X
ten-fingers           25.3           40.9              47.5
one-finger-savant     30.5           44.6                X
two-finger-savant     28.3           44.1                X

The code is attached if anyone wants to look at the individual
sentence results (the function run-all-tests will run the models
through all the conditions), but I wouldn't recommend it as a
guide for how to write an ACT-R model.  :)

Here are the things which I found interesting.

- The fastest overall was the ten fingered model in the "preparation
free" case at 47.5 wpm, which is faster than I expected.

- Testing "preparation free" actually lead to typing errors for the one-
and two-fingered models since it was modifying the features before the
last action had begun (the finger was trying to do two things at once).
So, those models are skipped for that condition.

- In the other cases the ten fingered model beats the "hunt and peck"
models as expected, but the "savant" models were faster than the
ten fingered one.  So the savings in preparation time is better than
the cost of the extra movement relative to the press-key actions
with the default motor module parameters.  However, from a plausibility
standpoint what those savant models do seems pretty super human to me.


Dan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: wpm-tests.lisp
Type: text/x-lisp-source
Size: 11043 bytes
Desc: not available
URL: <http://mailman.srv.cs.cmu.edu/pipermail/act-r-users/attachments/20100825/376ad5d4/attachment.bin>