NEUROCOGNITIVE BASES OF LANGUAGE
Discussion

December 10-20, 2001
Date: 11 Dec 2001
Subj: A few starting comments on the interesting 
      paper by Katherine Hrisonopulo:

I believe it is appropriate to take a look at linguistic signs
from a neurocognitive perspective, as relational entities
within the neurocognitive systems of people.

From this point of view the sign is a relationship between the
internal representation of a signifier (phonological or graphic
or other) and that of its meaning, a conceptual or perceptual or
motor or other cognitive entity.

Consider KH's statement (p.1) "In semiotic terms the phenomenon
of using a linguistic sign ... may be described as establishing
a triadic relation between the sign as such (its material form),
the object it represents, and the interpretant, that is, a
mental entity associated with the sign...."

Rethinking from a neurocognitive standpoint, we first
distinguish between (1) the sign relationship itself and (2) the
phonomenon of using a sign. (2) can only occur after (1) has
already been established within certain cognitive systems, at
leas that of the producer. Further, it is necessary to
distinguish two kinds of interpretant: the producer, in an
instance of using a sign, and the interpreter(s).

Next, we have to recognize several distinct levels: (1) that of
"the sign as such (its material form)"; (2) the internal
representation of (1), a neurocognitive entity, usually
comprising a subnetwork; (3) the internal (cognitive)
representation of the meaning (conceptual or ...); and (4) the
object it represents, outside the people involved, that is, in
the outside world. The sign as a neurocognitve unit is the
relationship between (2) and (3). We must also assume the
existence of neurocognitive relationships that mediate between
(1) and (2), and between (3) and (4).

Sydney M. Lamb                  http://www.ruf.rice.edu/~lamb/
Linguistics and Cognitive Sciences
Rice University, Houston, TX



Subj:Some comments on the interesting Sydney Lamb's observations on the paper
by Katherine Hrisonopulo:
From:Gian Carlo Buoiano

I have read with great interest Sydney Lamb's observations concerning the paper by Katherine Hrisonopulo, above all because his analysis permits to enlighten the neurocognitive relations among sign, reference, referent. As you all well know, from a neurolinguistic (and linguistic) point of view a 'sign' is composed by two arbitrary entities: expression and content (or signifier and signified). Actually I do not see any way to avoid this double arbitrariness, if we consider that any phonic chain is casually linked to a content (and vice-versa). So we need a device that can define the sign as non-arbitrary within the frame of a neurolinguistic theory in order to explain why neurocognition and language have phylogenetically developed using (also) arbitrary 'signs', since this would appear as an irreducible contradiction in itself. If our brain had to implement continually sounds (phonic chains), physical properties of real-world objects, concepts (content forms) without any compulsory link among these three domains, the human being would be in a state of total arbitrariness and could found a primitive form of communication only on the limbic system, i.e., we would be in a condition similar to that of animals, with a non-structured language based on limbic 'signals'. So, the only item that is not arbitrary, in the elaboration of the sign, is what Hjelmslev calls 'sign function', that is to say that the 'sign function' is a non-arbitrary device that links two (or more) arbitrary items. From a strictly neurolinguistic point of view this device must be implemented, after Norman Geschwind's theories (and recent scientific evidence) in the left IPL (Inferior Parietal Lobule), that matches data coming from secondary auditive, somesthetic, and visual cortices. In this way we have the necessary bridge between 'sounds' and 'physical properties of objects' - hence 'concepts' or 'content forms' - and also a specific cerebral region where, plausibly, this bridge is built. Since this specialization of IPL is characteristic of human species only, its evolutionary origin may exactly have been that of building this bridge that, in this way, would be the only non-arbitrary part of the sign. I am talking only about 'simple names' and not of elaboration of a sign in a given co(n)text. This last neurocognitive mechanism involves prefrontal areas (for instance DLPFC, DorsoLateral PreFrontal Cortex), but this seems anyway a first step toward the aim of establishing what is the non-arbitrary element of the sign.

Gian Carlo Buoiano Neurolinguistics, Linguistics Dept., University of Pisa, Italy g.buoiano@ling.unipi.it


Subj: on the interesting Sydney Lamb's observations  
Date: 13 Dec 2001
From: Jack Ferguson

Gian Carlo Buoiano states:
>"...sign, reference, referent. As you all well know, from a
>neurolinguistic (and linguistic) point of view a 'sign' is composed by
>two arbitrary entities: expression
>and content (or signifier and signified)."..."So we need a device
>that can define the sign as non-arbitrary within the frame of a
>neurolinguistic theory in order to explain why neurocognition and
>language have phylogenetically developed using (also) arbitrary 'signs',
>
>since this would appear as an irreducible contradiction in itself."

I'm not sure I follow this. I cannot relate to the metaphor of "content"
except as
a pattern within another pattern ad infinitum or a fractal movement that
may or may
not have associations to those patterns in the encoder's brain. The only
device I can find is the pattern constructed around associations that
may trigger associations in
the decoder's brain.

Jack Ferguson


Subj:Some words on "Questions of Evidence in Neurocognitive Linguistics"
by prof. Sydney Lamb, concerning parallel distributed models.

The PDP models really lacks of supporting evidence from neuroscience, 
but the authors usually signify that the units in a PDP
network models do not correspond to a single neurons, rather to ensembles
of neurons (presumably cortical columns), though this issue often left out
of discussion. What actually deserves criticism in PDP from the point of
plausibility is 1) the algorithm, widely used back-propagation learning
looks definitely unrealistic on a single neuron-axon level, but this becomes
less clear when we consider a unit as a cortical column, or a part
(a functional layer) within a column. A variant of back-propagation can be
implemented with the elements listed in [subj]. 2) Another point concerns
frequency signal encoding that underlies the attractiveness of back-propagation
- there are few recent models employing synchronization of pattern encoded
neural activity for learning.
One of the key features that must be adopted from the connectivist approach
is that the symbols used to interpret a model, are not mixed up
with the entities that are processed by the model itself. Another
important feature is stress on learning dynamics that relies on
operational plausibility until neurological arguments remain unclear.

Anton Ianchuk,
Institute for Informatics, State university of Kazan.
currently: neuroling@ksu.ru

On the "Evidence in Neurocognitive Linguistics" by Sydney Lamb.

According to linguistic evidences 1-13 from the paper, it may be reasonable to expect that a plausible model of metaphor can be built in network terms. We also can expect that cognitive neurolinguistics will provide a basis for mental spaces theory of Fauconnier and for conceptual blending theory. Is it so? It seems that computer models of interesting linguistic effects may serve as supporting argument for the theory proposed in the paper. The PDP approach is fairly criticized in many respects, still there are PDP-models of acquisition of grammar categories by infants and the learning curve followed U-shape that corresponds to realistic dynamic of a child learning. Was it ever implemented as a computer model the theory proposed in the paper? If so, that are results and if not what are the problems? Valery Solovyev State University of Kazan solovyev@mi.ru


Subj: Responding now to the comments of Gian Carlo Buoiano ("B"),
From:Sydney M. Lamb


> ...
> I have read with great interest Sydney Lamb's observations concerning
> the paper by  Katherine Hrisonopulo, above all because his analysis
> permits to enlighten the neurocognitive relations among sign, reference,
> referent. As you all well know, from a neurolinguistic (and linguistic)
> point of view a 'sign' is composed by two arbitrary entities: expression
> and content (or signifier and signified).


I appreciate these comments, but I am having trouble with B's
use of the term 'arbitrary'. According to Saussure's concept of
the sign, which Hjelmslev considered and revised, it is the
relationship between signifier and signified that is arbitrary,
rather than the signifier and signified themselves.

In keeping with Hjelmslev's analysis, we have to distinguish
between the form and the substance for both expression and
content (i.e. both signifier and signified), so that we have

1       expression substance    external signifier (phonetic)
2       expression form         internal representation of 1
3       content form            internal representation of 4
4       content substance       external signified (referent)

As mentioned in my previous comment, the sign relation is that
between 2 and 3, but the neurocognitive system must also
accommodate the relationships 1-2 (phonological
relationships) and 3-4 (perception).

Now, in my thinking, 2 and 3 are not arbitrary, for they are
based more of less directly upon 1 and 4 respectively.  Rather,
they can be considered abstract, and different language-culture
systems set up different systems for the entities in 2 and 3, so
the abstraction is indeed arbitrary to some extent, but such
arbitrariness is relatively monor in comparison to that
involving the relation 2-3.

I believe that B and I are in general agreement, except with
respect to the use of the term arbitrary, for I find much of
the rest of his comments quite cogent.

The arbitrariness of the relation 2-3 is necessary for a rich
communication system like that of humans, since to be rich it
must be capable of distinguishing among many thousands of
separate objects, relations, processes, etc. A non-arbitrary
relationship would require that all signs be iconic, and there
are not that many iconic expressions available. I believe this
remark is more or less parallel to B's

> ...       i.e., we would be in a
> condition similar to that of  animals, with a non-structured language
> based on limbic 'signals'.

> ...
> links two (or more) arbitrary items. From a strictly neurolinguistic
> point of view this device must be implemented, after Norman Geschwind's
> theories (and recent scientific evidence)  in the left IPL (Inferior
> Parietal Lobule), that matches data coming from secondary auditive,
> somesthetic, and visual cortices. In this way we have the necessary
> bridge between 'sounds' and 'physical properties of objects' - hence
> 'concepts' or 'content forms' - and also a specific cerebral region
> where, plausibly, this bridge is built.


By IPL do you mean specifically the supramarginal gyrus and the
angular gyrus?

>  Since this specialization of IPL
> is characteristic of human species only, its evolutionary origin may
> exactly have been that of building this bridge that, in this way, would
> be the only non-arbitrary part of the sign. I am talking only about ...


I would say, 'the only arbitrary part of the sign'. This
function is of course extremely important, as the region in
question must provide for connections between essentially any of
many thousands of conceptual nodes on the one hand (widely
distributed in the cortex) and any of thousands of phonological
nodes (in the superior posterior temporal lobe). The angular
gyrus and the supramarginal gyrus are ideally located for such a
function, and by having their millions of available latent
connections branching out in both directions, they can make it
possible to provide the connections between units of levels 2
and 3.

Sydney M. Lamb                  
http://www.ruf.rice.edu/~lamb/

Linguistics and Cognitive Sciences
Rice University, Houston, TX

Subj:Responding to the comments of Anton Ianchuk,


> ...  What actually deserves criticism in PDP from the point of
> plausibility is 1) the algorithm, widely used back-propagation learning
> looks definitely unrealistic on a single neuron-axon level, but this
> ...


I agree: Back-propogation as the means of learning seems quite
implausible, both operationally and neurologically.


> ...
> - there are few recent models employing synchronization of pattern
> encoded neural activity for learning.


This is in my opinion a more attractive and more intriguing
approach. It builds (or can be seen to build) upon Hebbian
learning, which I find quite realistic.


> One of the key features that must be adopted from the connectivist
> approach
> is that the symbols used to interpret a model, are not mixed up
> with the entities that are processed by the model itself.


I agree!

In relational network theory, learning is based on Hebb's idea,
and uses also an 'abundance hypothesis'. It operates in
accordance with a principle of 'neural Darwinism' (cf.
Gerald Edelman). There is a summary of the learning principles
on the web at http://www.ruf.rice.edu/~lngbrain/lrnout.htm .


Sydney M. Lamb                  
http://www.ruf.rice.edu/~lamb/
Linguistics and Cognitive Sciences
Rice University, Houston, TX


Re:Responding to the comments of Valery Solovyev:


> According to linguistic evidences 1-13 from the paper,
> it may be reasonable to expect that a plausible model of metaphor
> can be built in network terms. We also can expect that cognitive
> neurolinguistics will provide a basis for mental spaces
> theory of Fauconnier and for conceptual blending theory. Is it so?


Yes, indeed. One of my students did some work on metaphor a few years ago using relational networks. Unfortunately this work has not yet been published. Metaphors are based on shared conceptual properties. In a relational network such properties are implemented as connections. It turns out to be feasible and illuminating to treat some of Lakoff's work, for example, in relational network terms. Similarly, Fauconnier's mental spaces and conceptual blending theory, which I consider quite sound, should be representable in relational network terms, but this work has yet to be done.

> It seems that computer models of interesting linguistic effects > may serve as supporting argument for the theory proposed in the paper. > The PDP approach is fairly criticized in many respects, still > there are PDP-models of acquisition of grammar categories by infants > and the learning curve followed U-shape that corresponds to realistic > dynamic of a child learning. Was it ever implemented as a computer > model the theory proposed in the paper? If so, that are results and > of not what are the problems?

Another of my students has worked extensively on this problem, and in fact it was the subject of his recently completed dissertation. He has been working on getting the material ready for publication but has unfortunately been delayed. As to the problems: The main problems arise from the fact that for the system to be neurologically realistic it has to be quite complicated, and the programming (in Java) is therefore very difficult. It would be nice if we had a team of programmers available to work on it, but unfortunately we are not so blessed. Sydney M. Lamb http://www.ruf.rice.edu/~lamb/ Linguistics and Cognitive Sciences Rice University, Houston, TX


From:Gian Carlo Buoiano
Subject:Response to the comments of Sydney M. Lamb


As for IPL, I mean Supramarginal Gyrus (BA 40) and Angular Gyrus (BA 39) 
of the dominant hemisphere. This is just  an approximation, as 
Geschwind points out.

Sydney Lamb wrote:


>In keeping with Hjelmslev's analysis, we have to distinguish
>between the form and the substance for both expression and
>content (i.e. both signifier and signified), so that we have

>1       expression substance    external signifier (phonetic)
>2       expression form         internal representation of 1
>3       content form            internal representation of 4
>4       content substance       external signified (referent)

>As mentioned in my previous comment, the sign relation is that
>between 2 and 3, but the neurocognitive system must also
>accommodate the relationships 1-2 (phonological
>relationships) and 3-4 (perception).


>Now, in my thinking, 2 and 3 are not arbitrary, for they are
>based more of less directly upon 1 and 4 respectively.  Rather,
>they can be considered abstract, and different language-culture
>systems set up different systems for the entities in 2 and 3, so
>the abstraction is indeed arbitrary to some extent, but such
>arbitrariness is relatively minor in comparison to that
>involving the relation 2-3. 


I agree with many of Sydney Lamb's observations, and he has the merit of 
having raised an important question, but from a strictly hjelmsevian 
point of view, 1 and 2 are arbitrary as 2 and 3. Furthermore, 1 and 2 
have not an autonomous existence - at least if considered from a 
glossematic point of view -  i.e. they cannot exist without 2 and 3:



... Tegnfunktionen er i sig selv en solidaritet; udtryk og indhold er 
solidariske, foruds=E6tter n=F8dvendigvis hinanden. Et udtryk er kun 
udtryk i kraft af det er udtryk for et indhold, og et indhold er kun 
indhold i kraft af det er indhold for et udtryk. ... Vi konstaterer 
altsaa i det sproglige indhold, i dettes forl=F8b, en specifik form, 
indholdsformen, der er uafh=E6ngig af og staar i arbitr=E6rt forhold til 
meningen, og former denne til en indholdssubstans. ...


(.The  sign function is a solidarity in itself; expression and content
form a unity, they mutually and necessarily presuppose each other. An 
expression is an expression only because it is an expression for a 
content and a content is a content only because it is a content for an 
expression. ... Therefore, we observe in the linguistic content, in its 
process, a specific form - the content form - that is independent on - 
and is in arbitrary relationship with - the meaning, and shapes it in a 
content substance. 
Hjelmslev 1943. The translation and underlined 
text are mine in this and in the following quotations).


As we can see, Hjelmslev unmistakably state that 3 is in arbitrary 
relationship with - and non-dependent on the - sense (roughly, the 
meaning)  because the meaning, if we exclude the sign function is merely 
a shapeless continuum that can be perceived only as a nebulous. I am not 
saying that this is my point of view, I am only considering Hjelmslev's. 
The same observations can be made for 2:


... N=F8jagtig det samme kan iagttages i den anden af de to st=F8rrelser 
der er tegnfunktionens funktiver, nemlig udtrykket. ... (we can exactly 
state the same for the other of the two entities that are functives of 
the sign function, namely the expression. .)


Content and expression are arbitrary in the same way: at least, this is 
what Hjelmslev states. But the most striking fact to consider is the 
relationship between 1-2 and 3-4:


Denne unders=F8gelse viser os da, at de to st=F8rrelser der indgaar 
tegnfunktionen: udtrykket og indholdet, forholder sig paa ensartet maade 
over for denne: i kraft af tegfunktionen, og kun i kraft af den, 
existerer dens to funktiver, der nu helt n=F8jagtig kan betegnes som 
indholdsformen og udtryksformen, og kun i kraft af dem, existerer 
henholdsvis indholdssubstansen og udtrykssubstansen, der fremkommer ved 
at formen projiceres paa meningen, ligesom naar et udsp=E6ndt net kaster 
sin skygge ned paa en uinddelt flade. ... (This examine shows that the 
two entities that enter into the sign function: expression and content, 
have a unavoidable relationship. In force of the sign function and only 
in force of it, exist its two functives that now we can designate as 
content form and expression form, and  only in force of these two 
functives, consequently, the content substance and the expression 
substance subsist, and these last items become perceptible when those 
two forms project themselves on the meaning (sense) as a net projects 
its shadow on a shapeless surface. .)


In other words, Hjelmslev says that the substance not only is arbitrary 
- like the form -  but that it cannot exist without a back standing
form, i.e. we can hypothesize a form without a substance but not a
substance without a form. The obligatory linkage between form
(expression/content) and substance (expression/content) is the sign
function: if we exclude the compulsory sign function we cannot 'cut'
real-world entities in structured categories. This last fact is clearly
stated by Hjelmslev in 'Sprog og Tanke' (Language and Thought, 1936; I 
don't know whether it has ever been translated into English). It is true 
that Saussure considers the relationship between signifier and signified 
as arbitrary, but Hjelmslev calls this way of proceeding a 
'tankeexperiment' (experiment of thought), because we cannot state that 
sounds and concepts precede language. In other words, the relationship 
between a phonetic chain - say, [bottle] and the concept 'bottle' is 
arbitrary but compulsory, hence, in a broader optic, the two entities 
that form the sign are chosen arbitrarily, while their linkage is 
motivated by the exigency of analysis of perception and communication.   


Now, I  agree with S. Lamb on the fact that perception is not arbitrary, 
and that to perceive a sound (or a image, or a tactile sensation) is a 
non-arbitrary reality. At the same time, I must say that - as he points 
out - every language forms phonetic expressions and content forms in 
different, often non-overlapping ways. So, entities that are very 
dissimilar in different languages are linked by the same universal tool, 
the sign function that seems, in this optic, the only non-arbitrary part 
of the sign. It is in this logic that I said that in my thinking the 
sign function - being universally the same - seems the only 
non-arbitrary part of the sign. Nevertheless, may be we are dealing with 
a question of pure terminology.  



References:


1)L. Hjelmslev, Omkring Sprogteoriens Grundl=E6ggelse, K=F8benhavn, 
Munskgaard, 1943, now published in "Travaux du Cercle Linguistique de 
Copenhague", vol. XXV, Copenhagen, The Linguistic Circle of Copenhagen, 
1993.

2)"Sprog og Tanke", Sprog og Kultur, V-1936, 24-33, 1936.


Gian Carlo Buoiano

Neurolinguistics, Linguistics Dpt.,  Universita di Pisa Italy

g.buoiano@ling.unipi.it


From: Sydney M. Lamb                  
http://www.ruf.rice.edu/~lamb/

Subject:Responding to the remarks of Gian Carlo Buoiano:


>  I agree with many of Sydney Lamb's observations, and he has the merit of 
>  having raised an important question, but from a strictly hjelmsevian 
>  point of view, 1 and 2 are arbitrary as 2 and 3. Furthermore, 1 and 2
>  have not an autonomous existence - at least if considered from a
>  glossematic point of view -  i.e. they cannot exist without 2 and 3:

>  [quotation]
>  ...
>  As we can see, Hjelmslev unmistakably state that 3 is in arbitrary 
>  relationship with - and non-dependent on the - sense (roughly, the 
>  meaning)  because the meaning, if we exclude the sign function is merely 
>  a shapeless continuum that can be perceived only as a nebulous. I am not 
>  saying that this is my point of view, I am only considering Hjelmslev's. 
>  The same observations can be made for 2:



Yes, I agree. Hjelmslev does have a good point here, and I now understand why GCB used the term arbitrary as he did in the earlier message.

> [quotation] > ... > [quotation] > ... > In other words, Hjelmslev says that the substance not only is arbitrary > - like the form - but that it cannot exist without a back standing > form, i.e. we can hypothesize a form without a substance but not a > substance without a form. The obligatory linkage between form > (expression/content) and substance (expression/content) is the sign > function: if we exclude the compulsory sign function we cannot 'cut' > real-world entities in structured categories.

Yes. When I wrote my earlier comment, I was neglecting this very important point.

> ... > sounds and concepts precede language. In other words, the relationship > between a phonetic chain - say, [bottle] and the concept 'bottle' is > arbitrary but compulsory, hence, in a broader optic, the two entities > that form the sign are chosen arbitrarily, while their linkage is > motivated by the exigency of analysis of perception and communication.

OK, I accept this.

> Now, I agree with S. Lamb on the fact that perception is not arbitrary, > and that to perceive a sound (or a image, or a tactile sensation) is a > non-arbitrary reality. At the same time, I must say that - as he points > out - every language forms phonetic expressions and content forms in > different, often non-overlapping ways. So, entities that ere very > dissimilar in different languages are linked by the same universal tool, > the sign function that seems, in this optic, the only non-arbitrary part > of the sign. It is in this logic that I said that in my thinking the > sign function - being universally the same - seems the only > non-arbitrary part of the sign. Nevertheless, may be we are dealing with > a question of pure terminology.

Yes, on this point I also agree: a matter of terminology. I can accept that there is a nonarbitrary feature of the sign function, as a 'universal tool'. Yet the specific relationships involved in different actual signs are arbitrary in a very important sense: Any given instance of the sign function must provide a link between any of thousands of different (phonological) signifiers and any of hundreds of thousands of units of content. And this powerful property of human language is evidently provided (largely?) by the angular gyrus and the supramarginal gyrus (as Buoiano pointed out).

Sydney M. Lamb http://www.ruf.rice.edu/~lamb/ Linguistics and Cognitive Sciences Rice University, Houston, TX

P.S. My hat goes off to anyone who can understand Hjelmslev's prose in its original Danish!


From: Jack Ferguson Subject:regarding back-propagation learning states:

If we consider the human brain to be the 'ultimate' neural network, then ideally we would like to build a device which imitates the brain's functions.

I see no epistemological parallel between the brain and a computer. I think this premise is a confusion of a well designed tool, the computer, with the designer, the brain. This is like confusing a machine gun with boxing or a sculpture with the artist. The limitations inherent in the biological architecture of the brain/body are overlooked to the detriment of dealing with cognitive processes in the real world, i.e., in the chaos of the classroom and the problems of human communication. To get lost in the metaphor confuses two metalanguages. Machine language may parallel symbolic logic and various algorithms with crystal clarity while semiotic code remains a blur. Neat bifurcations are helpful for understanding machine code while the generating brain becomes a study in computer analogs, but remains hidden in its own shadows.

I think one problem is not to completely review the epistemological assumptions within semiotics. Terms such as referent, sign, symbol, etc. are not examined, but assumed correctly established. In one site, http://pespmc1.vub.ac.be/SEMIOTER.html these term are basically defined within an analytic network. They basically reference their etymology for definition, and look for support because of their own framework. I remain skeptical of this metacode. What is the epistemological foundation of this system?


mailto:neuroling@ksu.ru

Proceedings of the conference

Web Journal of Formal, Computational & Cognitive Linguistics