[ACT-R-users] to Elaborate last message

Wed Apr 24 10:57:28 EDT 2013

We also get the effect of partial matching in the word recognition component
of our language analysis model by spreading activation from letters and
trigrams in the perceptual input to words in declarative memory which have
slots for the letters and trigrams. 

The problem with partial matching in ACT-R is that when it is turned on, all
DM chunks are candidates for retrieval. There is not even a chunk type
constraint. With 58,000 word chunks in our mental lexicon, the amount of
computation to retrieve a word is prohibitive when partial matching is on.
Our word retrieval mechanism is constrained to at least match on a word
chunk type with a matching first letter (as a computational compromise),
which limits the amount of computation.

The theoretically interesting question is whether or not partial matching is
needed given spreading activation...

Jerry

P.S. I hope someone will correct my comments about ACT-R, if they are
incorrect

-----Original Message-----
From: Lynne Reder [mailto:reder at cmu.edu] 
Sent: Wednesday, April 24, 2013 10:35 AM
To: Laura Hiatt
Cc: Ball, Jerry T Civ USAF AFMC 711 HPW/RHAC; Kelley, Troy D CIV (US);
ACT-R-users at act-r.psy.cmu.edu
Subject: to Elaborate last message

there is no need for similarity values bc I only try to model similarity
when I have manipulated the features explicitly in the experiment (e.g.,
font and fan of font, or the operator).  So the overlap in features is
objectively determined.  I'm sure it is not adequate for most of what one
needs to do however so I inhibited writing for the last week, but my guard
is down...

On Apr 24, 2013, at 10:32 AM, Lynne Reder wrote:

> I have a tiny model based on earlier versions of ACT (no production
system, just nodes and spreading activation).  Many of the assumptions are
the same as the declarative side of ACT-R but some aren't, either by
accident (I used what I thought was in the system) or design.  In any case I
have been modeling partial matching for years.  The way it works in SAC is
that if the component features are activated/primed, they spread activation
to associated nodes.  If enough of the features are primed, it can get
sufficient activation to get the wrong node over threshold.  We have modeled
this in Feeling of Knowing tasks (where we swapped operators), in
recognition tasks where the unique font of the probe word matches a studied
word and the person FAs, etc.  
> 
> If anyone is curious I can point you to the work (some is in the process
of being written up for an invited chapter).
> 
> Cheers,
> Lynne
> 
> On Apr 24, 2013, at 10:24 AM, Laura Hiatt wrote:
> 
>> Thanks for the great feedback, everyone!  We have a new ACT-R-based
approach to similarity that we are developing here, based on base-level /
spreading activation values as the model runs (paper at CogSci this sumer);
I am looking to test it in a variety of situations (such as blending).  
>> 
>> Thanks again,
>> Laura
>> 
>> 
>> On Apr 23, 2013, at 9:11 AM, "Ball, Jerry T Civ USAF AFMC 711 HPW/RHAC"
<Jerry.Ball at wpafb.af.mil> wrote:
>> 
>>> Laura,
>>> 
>>> Troy's comment reminds me that Raluca Bidiu used LSA to set 
>>> similarity values in her research on "the Moses illusion". Here's a
paper:
>>> 
>>> Budiu, R. & Anderson, J. R. (2004). Interpretation-Based Processing: 
>>> A Unified Theory of Semantic Sentence Processing. Cognitive Science 28,
1-44.
>>> 
>>> The paper is available on the ACT-R web site under "Language Processing:
>>> Analogy and Metaphor".
>>> 
>>> Jerry
>>> 
>>> -----Original Message-----
>>> From: act-r-users-bounces at act-r.psy.cmu.edu
>>> [mailto:act-r-users-bounces at act-r.psy.cmu.edu] On Behalf Of Kelley, 
>>> Troy D CIV (US)
>>> Sent: Tuesday, April 23, 2013 8:35 AM
>>> To: Magill, Laura M CIV (US); ACT-R-users at act-r.psy.cmu.edu
>>> Subject: Re: [ACT-R-users] similarity in partial-matching/blending
>>> 
>>> Laura,
>>> 
>>> "Dynamically during runs" sounds interesting.  I have also been 
>>> thinking that using something like Latent Semantic Analysis (LSA) or 
>>> the similarity values in ConceptNet might be a useful way of setting the
initial values.
>>> 
>>> Troy D. Kelley
>>> RDRL-HRS-E
>>> Cognitive Robotics and Modeling Team Leader Human Research and 
>>> Engineering Directorate U.S. Army Research Laboratory Aberdeen, MD 
>>> 21005 Phone
>>> 410-278-5869 or 410-278-6748 Note my new email address:
>>> troy.d.kelley6.civ at mail.mil
>>> 
>>> 
>>> 
>>> On 4/22/13 10:27 AM, "Laura Hiatt" <laura.hiatt at nrl.navy.mil> wrote:
>>> 
>>>> Hi All,
>>>> 
>>>> I am looking for pointers to papers or work on 
>>>> partial-matching/blending that utilizes similarity values between 
>>>> chunks.  Much of what I have read using these mechanisms relies on
>>> numerical values (like blending utility values).
>>>> Lately we have been working on studying similarity between chunks, 
>>>> and how we can calculate it dynamically during model runs; I would 
>>>> love to be able to compare what we've come up with to some other 
>>>> approaches to
>>> similarity.
>>>> 
>>>> Thanks,
>>>> Laura
>>>> 
>>>> 
>>>> _______________________________________________
>>>> ACT-R-users mailing list
>>>> ACT-R-users at act-r.psy.cmu.edu
>>>> https://mailman.srv.cs.cmu.edu/mailman/listinfo/act-r-users
>>> 
>>> Troy D. Kelley
>>> RDRL-HRS-E
>>> Cognitive Robotics and Modeling Team Leader Human Research and 
>>> Engineering Directorate U.S. Army Research Laboratory Aberdeen, MD 
>>> 21005 Phone
>>> 410-278-5869 or 410-278-6748 Note my new email address:
>>> troy.d.kelley6.civ at mail.mil
>>> 
>>> 
>>> _______________________________________________
>>> ACT-R-users mailing list
>>> ACT-R-users at act-r.psy.cmu.edu
>>> https://mailman.srv.cs.cmu.edu/mailman/listinfo/act-r-users
>> 
>> 
>> _______________________________________________
>> ACT-R-users mailing list
>> ACT-R-users at act-r.psy.cmu.edu
>> https://mailman.srv.cs.cmu.edu/mailman/listinfo/act-r-users
> 
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 5621 bytes
Desc: not available
URL: <http://mailman.srv.cs.cmu.edu/pipermail/act-r-users/attachments/20130424/1ff42acc/attachment.bin>