Selected Newsgroup Message
From: "Sergio Navega" <snavega@attglobal.net>
Subject: Re: More on neuroscience...
Date: 29 Oct 1999 00:00:00 GMT
Message-ID: <3819c43f_1@news3.prserv.net>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<7v9v7f$60u$1@quine.mathcs.duq.edu> <7vbjgc$bpa$1@nntp3.atl.mindspring.net>
X-MimeOLE: Produced By Microsoft MimeOLE V4.71.1712.3
X-Complaints-To: abuse@prserv.net
X-Trace: 29 Oct 1999 15:58:55 GMT, 129.37.182.56
Organization: Intelliwise Research and Training
Newsgroups: comp.ai.nat-lang
Alexander Gross wrote in message <7vbjgc$bpa$1@nntp3.atl.mindspring.net>...
>> Why are you assuming that all neurolinguists are Chomskian?
>
>i'm not. why would i indeed? it just happens to be the Chomskians
>i'm talking about.
>
Hi, Alex!
I would say that, with some (very) noble exceptions, most neuroscientists
are anti-Chomskians. They have access to suggestive evidences that
point toward "no universal grammar" in our heads. Needless to say,
I'm at their side.
>> And an explanation of the hoped-for knowledge can be obtained from
>> any survey article in J. Child Language. In general, response
>> time, sucking rate, preferential looking, and so forth, can all be
>> applied to obtain evidence about what's going on in kids' heads.
>
>Yes, but this is all inference! Merest empricism! At least those
>are the naughty names the Chomskians have hurled at the linguists of
>the past in their haste and fury to exile them. Just recently, even
>Sergio, who admits to being a recovering Chomskian, suggested it was
>even true of Bloomfield's Secondary Responses to Language, which are in
fact
>so "generative" that they even apply to C's theories.
>
>Weren't the Chomskians supposed to do better? Weren't they
>supposed to come up with incontrovertible, measurable, truly
>"scientific" evidence? Or do inference, empiricism, and observed
>behavior suddenly become divine and blameless tools in their hands?
>
>
Chomskians are supposed to do better, and the state of this affair
is blurred today because of them. After all, they are the ones
who claim the existence of such an "organ". They are the ones
who must show strong evidences, not "must be so" stories. The
guys "on the other side" are skepticals: they are not convinced
that such an organ exists. But the burden of "proof" are in the
hands of Chomskians, as it is not easy to prove the inexistence
of something.
Regarding your story, at first I took it seriously (and even
became mildly concerned) but I thank you for leaving the notice
of its fictitiousness so evident.
MRI and fMRI are techniques that are non-invasive (very different
from, say, PET, which requires the ingestion of a liquid with
an oxigen isotope) and, besides the noise and claustrophobia
(reduced in modern machines), don't impair much suffering to
the patient. But it is not easy to find mommies willing to
leave their babies inside the machine (unless, of course, the
mommy herself is a neuroscientist, but I would bet she would
insist to do that with someone else's baby).
But there are other non-invasive techniques that can help in
this task. One of them is ERP (Event-Related Potential) that is
able to show semantic and syntactic effects in adults and children
with only EEG equipment (which means, just plugging some
externally connected wires on the skull, without pain or
discomfort).
Anyway, this is the way things must go, if we want some
progress toward this universal grammar debate. Sooner or later
we'll have enough evidences to paint a clear picture of the
whole process.
Regards,
Sergio Navega.
From: "Alexander Gross" <alexilen@sprynet.com>
Subject: Re: More on neuroscience...
Date: 29 Oct 1999 00:00:00 GMT
Message-ID: <7vg1nn$g31$1@nntp2.atl.mindspring.net>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<7v9v7f$60u$1@quine.mathcs.duq.edu> <7vbjgc$bpa$1@nntp3.atl.mindspring.net>
<3819c43f_1@news3.prserv.net>
X-Server-Date: 31 Oct 1999 00:15:51 GMT
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
Organization: MindSpring Enterprises
Newsgroups: comp.ai.nat-lang
Thanks once again, Sergio, for your excellent clarification of what
i was trying to say, though in a somewhat more provocative manner.
And i am grateful to all of you for your continued help within the
feverished workings of my own neural connections!
Yes, Sergio, i must sadly report that i know what an MRI is through
having undergone one. And my wife has had several. As for their
being non-invasive, that's truly a bit inside-out. The person
invades (literally penetrates) the device! But there is warfare all
the same.
Let me see if I can sum up what Sergio said (and no doubt you will
correct me if I still have it wrong):
The one last hope for Chomsky's theories of innateness, UG,
poverty of stimulus, etc. lies in the work being done by
neuroscientists. But unfortunately even they by and large
are pretty damned skeptical about these theories.
Fair or unfair? You tell me.
Patrick might be interested in looking at the September '99 issue of
the ATA Chronicle, which on page 71 contains an article entitled
"The Translation Portal: The Next Technology Revolution in the
Translation Industry." Though not about MT or AI, it contains an
excellent summary of how the translation industry currently works
and can save him some time in understanding this rather scary
process.
Jeff might also be interested in two other articles in the same
issue: one on translating a Chinese song into English, the other on
the transLITERATION of American firm names and slogans into Chinese.
I wouldn't be at all surprised if we see in coming years the
growing ascendancy of translation, translation theory, and translation
history over several different schools of linguistics, Chomskianism
prominently among them. It's simply that translators and scholars
in this field will have much practical and applied knowledge to
offer in place of dubious and/or sonorous theories.
very
best!
alex
PS--my most profound apologies to Patrik. i sincerely
hope not a one of you lost a single wink of sleep over my Happy
Halloween Card!
PPS--just took a look up on Steven Pinker's home page. Here's what
he thinks he is busy doing. It would be helpful to his cause if he
can pull this off--but can he?
"My research includes both empirical studies of linguistic behavior
and theoretical analyses of the nature of language and its relation
to mind and brain.
"On the empirical side, I study specific modules of grammar from a
variety of disciplines, much as biologists direct focus on a few
"model organisms." Currently, our group is studying inflectional
morphology: the ability to derive walked from walk or mice from
mouse. We are aiming for a unified theory and an extensive database
of how the system works computationally, how it is learned, how it
varies across languages, how it is used in language production and
comprehension, and how it is represented in the brain. In our human
information processing laboratory, we conduct reaction-time and
rating studies of how people produce and perceive inflected forms.
We also borrow from, and contribute to, generative linguistic
theory, examining interactions between syntax, morphology, and the
lexicon in English and other languages. We study people with
neurological and genetic language and memory disorders (aphasia,
Alzheimers, specific language impairment), gathering evidence on how
the different cognitive and linguistic modules underlying morphology
might dissociate. We study the development of inflection in
childrens language, both by analyzing computer-based transcripts of
spontaneous speech and by conducting experiments in child-care
centers; these studies document childrens memorization and
rule-deployment abilities and how they change over time. On the
theoretical side, I have used linguistic and psycholinguistic data
to develop a comprehensive model of the acquisition of grammar and
lexicon, and to analyze issues such as the role of symbolic and
connectionist computational architectures in language, the evolution
of human language, and the nature of conceptual categories."
From: "Sergio Navega" <snavega@attglobal.net>
Subject: Re: More on neuroscience...
Date: 01 Nov 1999 00:00:00 GMT
Message-ID: <381d8cce_3@news3.prserv.net>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<7v9v7f$60u$1@quine.mathcs.duq.edu> <7vbjgc$bpa$1@nntp3.atl.mindspring.net>
<3819c43f_1@news3.prserv.net> <7vg1nn$g31$1@nntp2.atl.mindspring.net>
X-MimeOLE: Produced By Microsoft MimeOLE V4.71.1712.3
X-Complaints-To: abuse@prserv.net
X-Trace: 1 Nov 1999 12:51:26 GMT, slip-32-102-102-87.sp.br.prserv.net
Organization: Intelliwise Research and Training
Newsgroups: comp.ai.nat-lang
Alexander Gross wrote in message <7vg1nn$g31$1@nntp2.atl.mindspring.net>...
>Thanks once again, Sergio, for your excellent clarification of what
>i was trying to say, though in a somewhat more provocative manner.
>And i am grateful to all of you for your continued help within the
>feverished workings of my own neural connections!
>
>Yes, Sergio, i must sadly report that i know what an MRI is through
>having undergone one. And my wife has had several. As for their
>being non-invasive, that's truly a bit inside-out. The person
>invades (literally penetrates) the device! But there is warfare all
>the same.
>
>Let me see if I can sum up what Sergio said (and no doubt you will
>correct me if I still have it wrong):
>
> The one last hope for Chomsky's theories of innateness, UG,
> poverty of stimulus, etc. lies in the work being done by
> neuroscientists. But unfortunately even they by and large
> are pretty damned skeptical about these theories.
>
>Fair or unfair? You tell me.
>
Fair enough. When Chomsky started the "cognitive revolution"
methods to functionally investigate the brain weren't fully
developed. Only in recent years (2, 3 decades) the techniques
improved substantially. But the question transcends fMRI.
Another big question is to come up with plausible computational
methods, able to duplicate the experimental results. This is a
potent point in the development of a theory that can duplicate
comparable mechanisms of our brain. In this regard, symbolic
systems appear to have a greater "research time", but
interestingly their results are somewhat stuck (and limited).
It is in the connectionist arena that important work is being
done today. I'm not a connectionist (I may lay down a lot of
disadvantages of this field) but it is difficult to deny that
they are doing a hell of a good job. Now guess what: most (if not
all) connectionists don't require any kind of innate modules. They
are closer to biological plausibility and their results mimic
experimental data (although in subsets of the whole problem) very
impressively. This is a strong attack to any innatist theory
of language.
Regards,
Sergio Navega.
From: "Alexander Gross" <alexilen@sprynet.com>
Subject: Re: More on neuroscience...
Date: 02 Nov 1999 00:00:00 GMT
Message-ID: <7vnfm6$2fj$1@nntp3.atl.mindspring.net>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<7v9v7f$60u$1@quine.mathcs.duq.edu> <7vbjgc$bpa$1@nntp3.atl.mindspring.net>
<3819c43f_1@news3.prserv.net> <7vg1nn$g31$1@nntp2.atl.mindspring.net>
<381d8cce_3@news3.prserv.net>
X-Server-Date: 2 Nov 1999 19:56:54 GMT
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
Organization: MindSpring Enterprises
Newsgroups: comp.ai.nat-lang
> Another big question is to come up with plausible computational
> methods, able to duplicate the experimental results. This is a
> potent point in the development of a theory that can duplicate
> comparable mechanisms of our brain. In this regard, symbolic
> systems appear to have a greater "research time", but
> interestingly their results are somewhat stuck (and limited).
> It is in the connectionist arena that important work is being
> done today. I'm not a connectionist (I may lay down a lot of
> disadvantages of this field) but it is difficult to deny that
> they are doing a hell of a good job. Now guess what: most (if not
> all) connectionists don't require any kind of innate modules. They
> are closer to biological plausibility and their results mimic
> experimental data (although in subsets of the whole problem) very
> impressively. This is a strong attack to any innatist theory
> of language.
Thanks again, Sergio. Can you tell me where i can find this
material? URLs for on-line papers or journals, or publications i
can sit down with in the library?
i think you're right, that if this approach is working, it might yet
achieve something, though still far from perfect MT or voice-writing
applications.
very
best!
alex
From: "Sergio Navega" <snavega@attglobal.net>
Subject: Re: More on neuroscience...
Date: 02 Nov 1999 00:00:00 GMT
Message-ID: <381f4fa1_4@news3.prserv.net>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<7v9v7f$60u$1@quine.mathcs.duq.edu> <7vbjgc$bpa$1@nntp3.atl.mindspring.net>
<3819c43f_1@news3.prserv.net> <7vg1nn$g31$1@nntp2.atl.mindspring.net>
<381d8cce_3@news3.prserv.net> <7vnfm6$2fj$1@nntp3.atl.mindspring.net>
X-MimeOLE: Produced By Microsoft MimeOLE V4.71.1712.3
X-Complaints-To: abuse@prserv.net
X-Trace: 2 Nov 1999 20:54:57 GMT, 166.72.29.177
Organization: Intelliwise Research and Training
Newsgroups: comp.ai.nat-lang
Alexander Gross wrote in message <7vnfm6$2fj$1@nntp3.atl.mindspring.net>...
>
>> It is in the connectionist arena that important work is being
>> done today. I'm not a connectionist (I may lay down a lot of
>> disadvantages of this field) but it is difficult to deny that
>> they are doing a hell of a good job. Now guess what: most (if not
>> all) connectionists don't require any kind of innate modules. They
>> are closer to biological plausibility and their results mimic
>> experimental data (although in subsets of the whole problem) very
>> impressively. This is a strong attack to any innatist theory
>> of language.
>
>Thanks again, Sergio. Can you tell me where i can find this
>material? URLs for on-line papers or journals, or publications i
>can sit down with in the library?
>
>i think you're right, that if this approach is working, it might yet
>achieve something, though still far from perfect MT or voice-writing
>applications.
>
This will keep you busy for a while:
Risto Miikkulainen, UTCS Neural Nets Research Group
http://www.cs.utexas.edu/users/risto/index.html
Srini Narayanan, International Computer Science Institute
http://www.icsi.berkeley.edu/~snarayan/
Lokendra Shastri, AI Group, International Computer Science Institute
http://www.icsi.berkeley.edu/~shastri/
Jeff Elman, Dept Cognitive Science, Univ. California, San Diego
http://crl.ucsd.edu/~elman/
CogSci 110, Neural Basis of Thought and Language
http://www.icsi.berkeley.edu/~mbrodsky/cogsci110/page.html
CNBC Center for the Neural Basis of Cognition
http://www.cnbc.cmu.edu/
papers:
ftp://cnbc.cmu.edu:/pub/pdp.cns/
Morten H. Christiansen
http://www-rcf.usc.edu/~mortenc/index.html
http://www.siu.edu/departments/cola/psycho/faculty/mhc.html
Mark S. Seidenberg, University of South California
http://www-rcf.usc.edu/~seidenb/
Michiro Negishi
http://cns-web.bu.edu/pub/mnx/negishi.html
Axel Cleeremans
http://164.15.20.1/axcWWW/axc.html
I have more but this is enough for now. I hope you'll
have as much fun as I'm having!
Regards,
Sergio Navega.
From: "Alexander Gross" <alexilen@sprynet.com>
Subject: Re: More on neuroscience...
Date: 03 Nov 1999 00:00:00 GMT
Message-ID: <7vorak$9v0$1@nntp3.atl.mindspring.net>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<7v9v7f$60u$1@quine.mathcs.duq.edu> <7vbjgc$bpa$1@nntp3.atl.mindspring.net>
<3819c43f_1@news3.prserv.net> <7vg1nn$g31$1@nntp2.atl.mindspring.net>
<381d8cce_3@news3.prserv.net> <7vnfm6$2fj$1@nntp3.atl.mindspring.net>
<381f4fa1_4@news3.prserv.net>
X-Server-Date: 3 Nov 1999 08:21:40 GMT
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
Organization: MindSpring Enterprises
Newsgroups: comp.ai.nat-lang
>I have more but this is enough for now. I hope you'll
>have as much fun as I'm having!
>
Thanks again, Sergio. And thanks to Patrick for your thoughts.
i'll let the two of you fight about this one... :-)
best!
alex
From: juola@mathcs.duq.edu (Patrick Juola)
Subject: Re: More on neuroscience...
Date: 02 Nov 1999 00:00:00 GMT
Message-ID: <7vnmgt$dus$1@quine.mathcs.duq.edu>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<3819c43f_1@news3.prserv.net> <7vg1nn$g31$1@nntp2.atl.mindspring.net>
<381d8cce_3@news3.prserv.net>
Organization: Duquesne University, Pittsburgh PA USA
Newsgroups: comp.ai.nat-lang
In article <381d8cce_3@news3.prserv.net>,
Sergio Navega <snavega@attglobal.net> wrote:
>Alexander Gross wrote in message <7vg1nn$g31$1@nntp2.atl.mindspring.net>...
>It is in the connectionist arena that important work is being
>done today. I'm not a connectionist (I may lay down a lot of
>disadvantages of this field) but it is difficult to deny that
>they are doing a hell of a good job. Now guess what: most (if not
>all) connectionists don't require any kind of innate modules.
Actually, this isn't true. The modules are simply more subtle
than the ones that Pinker, et al. would like -- but they're there.
One possibility which has sometimes been seen is that an apparently
module-less network will actually subdivide itself into functional
modules over the course of training. For instance, it's accepted
that the simplest method of implementing an autoencoder is to
perform a principle component analysis (PCA) on the inputs, and
this is the usual solution found by a standard backprop network.
Somewhat less subtle is the use of the nodes-and-boxes modularity
*at the architectural level of the network.* Miikkulainen's
DISCERN is a good example of this; although the individual
nodes don't appear to perform a symbolically meaningful task, they
are explicitly built in pools connected in a topologically
complex and symbolically meaningful fashion. So the innate modules
are there in the divisions between and among the unit pools.
Denis Mareschal has similar architectures in his models of the
acquisition of object permanence. Nakisa even has a well-developed
theory of "genetic connectionism" where the module topology
(and hence the "meaning" of the innate modules) is itself subject
to learning pressures.
There are actually relatively few connectionists working with
pure, unstructured networks.
-kitten
From: "Sergio Navega" <snavega@attglobal.net>
Subject: Re: More on neuroscience...
Date: 03 Nov 1999 00:00:00 GMT
Message-ID: <38209b23_4@news3.prserv.net>
References: <7v9uhg$ii4$1@nntp2.atl.mindspring.net>
<3819c43f_1@news3.prserv.net> <7vg1nn$g31$1@nntp2.atl.mindspring.net>
<381d8cce_3@news3.prserv.net> <7vnmgt$dus$1@quine.mathcs.duq.edu>
X-MimeOLE: Produced By Microsoft MimeOLE V4.71.1712.3
X-Complaints-To: abuse@prserv.net
X-Trace: 3 Nov 1999 20:29:23 GMT, 166.72.21.28
Organization: Intelliwise Research and Training
Newsgroups: comp.ai.nat-lang
Patrick Juola wrote in message <7vnmgt$dus$1@quine.mathcs.duq.edu>...
>In article <381d8cce_3@news3.prserv.net>,
>Sergio Navega <snavega@attglobal.net> wrote:
>>Alexander Gross wrote in message
<7vg1nn$g31$1@nntp2.atl.mindspring.net>...
>>It is in the connectionist arena that important work is being
>>done today. I'm not a connectionist (I may lay down a lot of
>>disadvantages of this field) but it is difficult to deny that
>>they are doing a hell of a good job. Now guess what: most (if not
>>all) connectionists don't require any kind of innate modules.
>
>Actually, this isn't true. The modules are simply more subtle
>than the ones that Pinker, et al. would like -- but they're there.
>One possibility which has sometimes been seen is that an apparently
>module-less network will actually subdivide itself into functional
>modules over the course of training. For instance, it's accepted
>that the simplest method of implementing an autoencoder is to
>perform a principle component analysis (PCA) on the inputs, and
>this is the usual solution found by a standard backprop network.
>
Agreed, but this does not address my previous comment. I was
reflecting over the presence of modules *before* training. Actually,
the wording 'module' is not what I should have said. I was thinking
about 'innate knowledge'. It is clear that NNs start without any
kind of innate knowledge and that's why I put them (connectionists)
closer to the anti-nativist position (which is also my preferred
way to see things).
>Somewhat less subtle is the use of the nodes-and-boxes modularity
>*at the architectural level of the network.* Miikkulainen's
>DISCERN is a good example of this; although the individual
>nodes don't appear to perform a symbolically meaningful task, they
>are explicitly built in pools connected in a topologically
>complex and symbolically meaningful fashion. So the innate modules
>are there in the divisions between and among the unit pools.
Agreed. Certainly this is equivalent to some kind of innate
predisposition, even if each module starts from zero. I guess that
this somewhat smoothes the position of some connectionist (they
can be seen not as entirely anti-nativist), which is apparently
the best position. After all, it is clear that much of the structure
of our brain is genetically determined, if not as much as the
nativists want. The issue that muddles things a bit appears when
we think of language. As a preponderantly cultural construct, I
question the ideas of genetic "knowledge" of language, although
I may accept the genetic predisposition to allow one organism to
develop it (but not specific to this domain!).
>Denis Mareschal has similar architectures in his models of the
>acquisition of object permanence. Nakisa even has a well-developed
>theory of "genetic connectionism" where the module topology
>(and hence the "meaning" of the innate modules) is itself subject
>to learning pressures.
>
I didn't know about Mareschal, I'll try to search a bit. But I'm
a bit unconfortable with innate modules that are domain specific.
Sure, I may see the relevance of the visual cortex in this sense,
but one can't use this as an example of a pure domain specificity,
given the frequent reuse of this area by audition, by the visually
impaired. Perhaps what I question most is the idea of domain
specific intelligences, genetically determined characteristics that
may affect one kind of "intelligence" but not other (like Gardner's
idea of multiple intelligences). I bought the idea that there should
be one kind of "mechanism" whose basic principles are common among
the various "specialties". This leads me often to look for parallels
between contrasting things, like visual perception and language.
Could some basic principles common to both things be found? That's
one of my bets (still to confirm...)
Regards,
Sergio Navega.
Back to Menu of Messages Sergio's Homepage
Any Comments? snavega@attglobal.net