In the following message, M. Zeleny takes me (validly) to task for
claiming a syntactic model of the olfactory system (based on Walter
Freeman's work) is a semantic model.
>To: erwin at trwacs.fp.trw.com>Subject: Re: Theories of meaning not relying solely on sym
>"Objects reported" are syntactic by any definition of syntax; likewise
>for attribution of synonymy as determined by similarity or identity of
>activations of the frontal cortex. Since I am at a loss as to what
>might constitute a coherent notion of semantic grounding, I see no more
>way to demonstrate it in primitive mammals, than to demonstrate the holy
>nature of a self-professed Himalayan yogi via lab measurements.
>Physicalism depends on reducing semantics to physical primitives, or on
>eliminating it altogether. The former is simply impossible, as there is
>no way to partition the causal structure of the physical universe, so as
>to obtain the requisite "intentional stance". In other words, there is
>no coherent way to differentiate the physicalist universe into subject
>and object, as required for semantics. Reductionism is the sole viable
>physicalist alternative. End of this discourse.
Freeman's model for the function of the olfactory system is a syntactic
model, but it also has semantic validity, in much the same way that
syntactic models of growth and development have semantic validity.
Consider the process by which a zygote becomes a multi-cellular adult. It
can be described formally as a continuous process fully specified by the
internal and external environments and the genome of the organism. Hence,
it has a syntactic model. However, that model is subject to natural
selection against alternative models, with the best adapted having a
statistical advantage of surviving. Hence, that model can also be
interpreted semantically in terms of the interactions with the environment
and other organisms that have led that model to have superior fitness.
Likewise, Freeman's model can be interpreted in terms of what aspects have
led it to have superior fitness. It appears that early correlation of
environmental features with components of the internal models being
maintained by the cerebral cortex has selective advantage. Thus we can say
that the olfactory system is semantically grounded in that the internal
models are projected out from the cerebral cortex so that the connection
to sense data occurs as early as possible in the processing chain. The
olfactory nucleus does not report measurements of chemical concentrations,
or detections of "dog," but rather track reports on "the wet dog that has
been sniffing around the mouth of the burrow for the last few minutes."
Now I intend to really put my foot into it.
"How the brain works"
I would like to suggest that the following types of modules have a
major role to play in the cerebral cortex:
This module operates similarly to the olfactory nucleus to couple pattern
data to a "semantic" interpretation of those data. It relates three
subsystems: A, B, and M1, interfacing as follows:
A simply presents pattern data to M. This presentation is invariant. B, on
the other hand, interacts with M to interpret the pattern data presented.
B "downloads" objects in real-time to M for matching to the patterns
presented by A. M is quite capable of handling hundreds of individual
patterns, using a chaotic process similar to that seen by Freeman in the
olfactory nucleus. (Individual pyramidial cells can retain hundreds of
features for matching purposes and can be "downloaded" to identify
specific features.) The input data to B are correlated to specific objects
by the pattern of the carrier wave that is modulated to provide the data
reports. Whether B updates the models in M over time or those models have
the capability of evolving over time is unclear.
This module evaluates the data from a set of pattern-matching modules. It
maintains a (dynamic?) model of some aspect of the environment, and
discrepancies and novelties are superimposed on that model. If the system
is to operate as a novelty detector (the frontend to a orientation
response), the output of M1 modules serve as an efferent input, "playing a
spotlight" on significant discrepancies. This system may also operate as a
modulator of a motor process.
This is the simplest module. It is a non-linear readout of a
non-stationary Hopfield net. The output is a multi-dimensional frequency
modulation of a carrier wave (which serves to key the output and match it
to other outputs associated with the same feature at some level).
Consider the speech/hearing system. The speech end is probably a M1 module
that monitors a "train of thought." The B system downloads candidate words
and phrases to the M1 system for matching to the train of thought. The
quasiperiodic dynamics within the M1 system may generate the word
sequence, or some translation process further downstream may be involved
in stringing things together. This process, driven by pattern recognition
eventually results in motor actions and audible speech.
On the other side, (at some level) we have a process similar to that in
the olfactory system. A proposed speech generator (similar to the B
process in the speaker) downloads words to the auditory system for
matching to the heard speech. This then drives a model, which is probably
similar to that in the B system above, which is presented to the "train of
thought" processing using an M1 module. Hence the result of speech is the
entrainment of one mind by another, making use of processes similar to
those used by Pecora and Stafford in their demonstration of the
synchronization of two chaotic processes over a low bandwidth channel.
Internet: erwin at trwacs.fp.trw.com