Message-ID: <36F7C0F6.BE423CA8@andrew.cmu.edu>
Date: Tue, 23 Mar 1999 11:27:34 -0500
From: Rob Mason <rmason@andrew.cmu.edu>
MIME-Version: 1.0
To: Tom Mitchell <Tom.Mitchell@cs.cmu.edu>
CC: Chuck Rosenberg <chuck+@cs.cmu.edu>, danieln+@cs.cmu.edu,
        mitchell+@cs.cmu.edu
Subject: Re: fMRI data/ sentences
References: <Added.AqxsoF200Udb0D0040@andrew.cmu.edu>
Content-Type: multipart/mixed;
 boundary="------------7A2707977EA0DC8B3010B33A"


Tom----
I'll answer each of your questions in turn.
(1)
> Is it easy for you to provide us the list of specific sentences and
> questions given to subject 02882?
Well, yes it's easy, but in an easily digestable format isn't so easy. 
What I'll do is attach two files.  

The frst is a list of all 40 sentences used in the expt, they are
grouped by condition (without questions).  So the first ten sentences
are all from condition 1 in the order that they appear in within that
condition, but in order to get the actual value of s_n you'd have to
look match with the condition number in the overall set.  

The second file is the actual stimulus set used in the experiment.  This
contains the sentences in order along with questions.  The drawback to
this format is that it also contains all the COGLAB codes which are
unnecessary for your purpose.  I'm sorry that there is no inbetween
version (believe me I wish I had one).

(2)
> So something about the stimuli is making the time series very different 
> for each sentence, despite the fact that the voxel correlations remain 
> intact.  
If this is a pattern that varies by condition then the conclusion is
that it is due to the experimental manipulation, however I take it that
is not the case.

If the variation in the time course is not correlated to the conditions,
then I fear that the question will be vary difficult to answer.  Within
each condition the syntax of the sentences is exact but nothing else
is.  There are too many other linguistic variables that could cause
variation (e.g., lexical frequency of actor/patient nouns are not
controlled, theta role recipiants could have been animate/inanimate,
some sentences had location cues some did not, etc.).  As of yet we
don't have a good handle on what linguistic variables should have
localized (in the brain) effects and which might have diffuse effects so
it is unclear where to start if you want to look at specific factors.

(3)
> One question this raises is "if the subject is given the same sentence 
> twice, how repeatable are the voxel time series?"
Behaviorally (classic reading time measures), reading appears to be very
different when a sentence is repeated (at least in isolation).  It would
be very interesting if voxel time series were repeatable for identical
sentences.  But if the activity of a voxel is in a sense probablistic,
you would need many repeatitions of a sentence to get a stable
repeatable time series.  By then I have no doubt that the reading
process would be very different than it was on the first iteration.

(4)
> Can you tell me whether sentences were ever repeated to a subject?
No senetences were repeated.  There was occasionally a verb which was
used in more than one sentence, sometimes even across conditions (e.g.,
"called"). That might be a place to look, but I doubt that a voxel's
time course could be verb specific (although I know some linguists who
would love that idea).

Sorry that was longer than I expected.  Let me know if anything else
comes up.
rob.
-- 
---------------------------------
Robert A. Mason
Center for Cognitive Brain Imaging
Department of Psychology
Carnegie Mellon University
Pittsburgh, PA 15213

e-mail: rmason@andrew.cmu.edu
phone: (412) 268-2402
fax: (412) 268-2804
http://www.andrew.cmu.edu/user/rmason/index.html