real neurologists, Kohonen maps

[Timo Honkela]

Dear Funknetters,

Brian wrote:

>    Models that take the details of neuronal functioning seriously and which
> capture the intricacy of neuroanatomical patterns are going to be tough to
> build.  One area where modelling and real neurological facts seem to be
> coming into good contact is in regard to the details of the wiring of local
> map topology.  For example, models of Kohonen self-organizing feature maps
> closely echo facts of lateral inhibition that are important in setting up
> neuronal fields.

It is my pleasure to tell you about an experiment in which we used
Kohonen's self-organizing maps to process natural language input.  The
result was interesting compared with some neurophysiological findings.
The general organization of the map reflected both syntactical and
semantical categories. All verbs could be found in the top section
whereas the nouns were located in the lower right corner of the map.
Inside the large, syntactically based groups on the map, fine
structures of semantic relationships could also be discerned.  For
instance, the animate and inanimate nouns formed a group of their own
each. The implicit categories emerged in the unsupervised learning
process the input being just the text itself.

A summary as well as the full paper is available at the
address http://nucleus.hut.fi/~tho/lang-som/


Best regards,
Timo


                -------------------   ---------------------------
 Timo Honkela   Timo.Honkela at hut.fi   http://nucleus.hut.fi/~tho/
Neural Networks Research Centre,      Helsinki Univ of Technology
     and        Rakentajanaukio 2 C,  FIN-02150 Espoo, Finland
 Nat Lang Proc  Tel. +358-9-451 3275, Fax +358-9-451 3277