Re: Directions for future research
On Monday 24 January 2005 18:11, M. Shamsfard wrote:
> Rob Freeman wrote:
> >Do you mean dynamic in the sense of constant total
> > re-interpretation, or only in the sense that new
> > meanings can be added incrementally?
>
> ...
>
> 3- As the paper describes the world can be conceptualized from
> several different perspectives and we may have various views of
> concepts and relations and so various ontologies. A system may
> contain all or some of these viewpoints in the same time and use
> any one on demand. This property have been considered in the
> design of Hasti. Suppose your ontology as a graph with different
> colored weighted nodes and links. each color may be the
> representative for a perspective or for a domain and weights show
> the importance or relevance factor of the ontel to that perspective
> or domain. when you use the ontology in a specific domain or
> specific perspective or even specific ideology, you can just see the
> corresponding colors. or if you suppose them to be linked by
> various threads it seems like you drag a thread and all linked
> ontels to the thread remains and others will fall. the remained
> ontols (linked by the thread) are the needed ontology for your task.
> This idea is under development in Hasti.
I like this idea. I very much believe we need to maintain alternative
perspectives in the system. My view, however, is that there are just too many
of these alternative perspectives to enumerate them all prior to use. All
that is possible is to store the data from which they can be "learned", and
"learn" the perspective relevant to each new problem at the time the problem
is presented to the system.
I think this is necessary (this keeping perspectives only as potentials in the
data and learning them fresh when a problem is presented) because that is
what seems to be necessary to model the classes (structural perspectives) of
syntax, which can be unique to a particular combination of words (syntactic
"problem".)
Have you considered the possibility that the number of alternate ontological
perspectives might be very large (infinite?), and that you need to consider
not only selecting between pre-learned perspectives (using "colors" or
"threads"), but actually "learning" a new perspective for each new problem?
-Rob