Re: Interoperability and Vagueness
Matthew, Tim, and John,
Your positions are very close to what I've been
trying to say in different ways for a long time.
I'm grateful for some new ways of expressing
many of the same issues.
Matthew's three points are essentially what I did
in my "diamond" diagram of three top distinctions
in the ontology chapter of my KR book:
MW> 1. It seems to me that the "names" we give
> concepts in an ontology should be seen more like
> product codes (identifiers), than words.
>
> 2. An ontology must be extensible, because it will
> always be possible to come up with new concepts.
>
> 3. Given this it is useful to standardise the
> identification of concepts, because concepts don't
> change (only their importance to us).
I very much agree. The only additional point I would
make is that people cannot remember "meaningless"
identifiers, and they very quickly turn them into
word-like strings that they can pronounce.
The chemists have a great system of identifiers built
up from letters assigned to each element, which they
combine to form strings, such as H20. But whenever
they correspond to a familiar word, such as _water_,
even chemists say "water". Whenever the strings are
too long to fit into normal sentences, they use acronyms
such as DDT or coin new words that are outside the
systematic nomenclature, but with an agreed mapping to it.
For the top-level of my lattice, I chose words whose meanings
covered what I wanted to say, such as _Physical_ vs. _Abstract_,
but I emphasized the point that they are technical terms in the
same sense that the common word _force_ becomes the technical
term _force_ in physics. For the combinations, I said that
the preferred identifiers were the acronyms such as MA or RPO.
Since nobody could remember those acronyms, I said that they
could use mnemonics, such as MA = Mediating Abstraction or
if they really need one word, Nexus. And RPO = Relative
Physical Occurrent or Participation. But the acronyms MA and
RPO are preferred for precision *because* they are not words.
The emphasis on the acronyms is essential to prevent changes
in meaning, as Tim noted:
TK> ... as word meaning evolves do some of the names chosen
> at any point become less suitable?
Indeed. Then it may be necessary to use a different word
or phrase as a mnemonic for RPO or MA.
MW> Well I must admit that I tend to have a preference for
> having the formal identifier for a concept as a number...
I'm happy with numbers or very long strings as the official
identifiers. But nobody can remember numbers, and nobody
will use very long strings more than once. They quickly
turn "dichlorodiphenyltrichloroethane" into DDT. It's
important to recognize that tendency right up front and
provide a systematic nomenclature that accommodates it.
I completely agree with John Bateman:
JB> We take the fact that word (and grammatical and semantic
> and textual) distinctions are *inherently* negotiable
> precisely via "complex and dynamic mappings", as Boyan
> says, as an intrinsic, necessary and even ontological
> feature of natural language. Without this feature,
> there would not be natural language.
But I would go one step further and say that this feature
is fundamental to all human thinking about anything and
everything, and it results from the need to accommodate
our cognitive systems (whatever they may be) to the
complexities of the world and to our social and physical
interactions with it and in it.
JB> So it is certainly much more than "indirectly"
> a problem for NLP.
Certainly. When I used the word "indirectly", I was
trying to say that the reason why natural language is
the way it is results from the way our minds work in
enabling us to live and interact with the world and
with one another. I was trying to argue against
artificial systems, such as Wittgenstein's Tractatus,
and in favor of his later philosophy.
I admit that one of the major sources of evidence
for how our minds work is what we say in language.
But language is not the only source of evidence,
and other sources -- such as neurophysiology,
psychology, anthropology, evolution, animal behavior,
and deficits or disturbances, such as autism and
strokes -- corroborate and often clarify or explain
the evidence from language.
JB> I look forward to more interaction concerning
> a nontrivialised version of the relation between
> language and ontology.
Certainly.
John Sowa