I believe that concepts play different kinds of roles when they
interact with other concepts.  These roles can have functional
implications.  So from there I would say that concepts have types.
There are types in programming of course but because concepts are more
dynamic, the typing tends to be more elusive.  The recognition that
concepts do have types might make it easier to distinguish what is
happening when concepts are used together.

Looking at the categories that words belong to can provide some
insight into what I am talking about.  The idea that the most general
semantic classes of grammatical analysis might be used to better
understand how sentences are formed has created a number of exciting
theories.  But this idea can be taken much further into the narrower
semantic categories that words can take, and I believe that the
evidence is there that it should be.

For instance, a word like, "captain" might be classified as a proper
noun but it is obvious that the possible associations with the term
are much more extensive than that.  I believe that if the semantic
categories that are relevant to a particular sentence become more
apparent, then that information could be used to type the particular
usage of the word and the nature of the concept that goes along with
it.  Once the program forms a better idea of how the term, 'captain'
can be used in a sentence, the better it can be used to determine how
the other words in the sentence are used.  This reasonable insight
implies that words can have functional relationships with other words
and this is important because it provides evidence that word-meanings
can be typed.  Word types are going to be involved in both semantic
and semantic-related syntactic relations that go way beyond the most
general types talked about in most modern grammatical analysis.  For
example, some words may refer to a central subject of a comment while
others are directed toward other objects in the commentary.  So the
idea that all the words in a comment should be treated as equal tokens
in some way is going to be insufficient.  Similarly, words that are
only evaluated based on their co occurrences are not going to be
appropriately evaluated for their functional roles within a comment.
While attention to this has been paid in natural language analysis, it
has typically been done in such a way as to minimize the typing of the
words so that the analysis can apply to the broadest grammatical
types.  And it always based on an elaborate, carefully designed model
and almost never as something the program has to discover.

When I tried to think of some of the particular categories that the
word 'captain' could belong to I ran into more trouble than I expected
because my attempt to articulate those categorical types with precise
terms did not come easily.  I am not claiming that we precisely define
the categories that words can belong to but that we can make little
fragmented stories about some of those meanings.  Like 'the captain of
an airplane,' or 'the captain of a ship,' or 'a nickname we might give
the owner operator of a boat,' or 'a nominal leader amongst the
players of a team' and so on.

The initial problem in linguistic analysis is discovering the intended
meaning of the terms used in a comment.  So then, would this idea,
which is dependent on the discovery of a more narrow range of possible
meanings for a term, really help?  I think it would because the
mechanisms of learning would be based on acquiring insight into these
associations and it could be done in a way that seems somewhat more
natural than establishing precisely defined grammatical categories.

The typing of words and the functional power of words is different
than the distinct typing and functional-predicate relations of
specialized 'languages' like the operators of programming languages
and logical-computational languages. For example, each word is both a
type related to its usage in a comment and also a variable.  Ignoring
the meaning of term 'predicate' in natural language grammar for a
second, I can say that a word can be both a predicate and an object of
the predicate because words are interactively functional with other
words.  This is a very difficult computational 'language' and that is
why this theory has been effectively suppressed in NLP.
(Incidentally, this kind of theory can be reduced to a simpler
computational language for a precise interpretation of how a word is
used in a comment, but my whole argument is that until an AGI/AI
program can figure the meaning of a particular sentence fragment it
needs to deal with the more complicated system.  My argument is: That
is just the way it is and until we deal with the reality there is very
little hope that we going to work it out.)

But this process is not just limited to words, phrases and stories but
to all concepts.  The idea behind conceptual typing is that the
interaction and functional roles of concepts are partially determined
by the typing of the concepts in a particular usage.  Without dealing
intelligently with this problem in some manner it becomes less likely
that the dynamic interactions of concepts will be brought under
control to be used effectively.

This system is unwieldy and at this time I cannot predict whether it
can be used effectively.   However, one thing is clear; it must be
dependent on an extensive system of indexing.  The problem of making
the most efficient indexing system seems to be NP-Complete.  However,
by using an indexing system where the more frequently used paths are
given some preeminence in the indexing, along with the use of multiple
paths, it should be possible to create an effective indexing strategy.
- Jim Bromer


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to