Re: [port-peer-review] reviews
> But the main question is: how can they be integrated with
> formal structures? More precisely, could ontologies and
> formal techniques be used to extract formal knowledge
> from associative representations? (01)
You'll have the same problems as from natural language,
may be more on some points because natural language articles
("a", "the", "all", "most", ...), which are hints for the
quantifications, can be lost in concept maps. (02)
When I wrote that concept maps may be an interesting compromise,
I was mainly thinking about only allowing invented concept type names.
If you keep the constraints that
(i) concept nodes must have a known quantifier (with an explicit scope),
(ii) relations types must be declared and have an associated signature,
then the invented labels may be automatically inserted into an ontology,
and some checkings may be done (not enough but that is a start).
This might be acceptable by people given that:
(i) there is a limited list of quantification forms, e.g. "a", "the",
"some", "any", "most", "at least 35%", ... (see the FCG grammar at
http://www.webkb.org/doc/F_languages.html#FCG);
(ii) there is a limited list of common binary basic relation types
(see http://www.webkb.org/doc/RDF/phOntology.html#relation). (03)
Currently, WebKB-2 does not permit invented category identifiers but
permits the use of category names (known in the KB) instead of category
identifiers, as long as each used name can be resolved to a unique
category; this resolution is possible when (i) a name/word refers to
only one category (in the KB), or (ii) the relation signatures permit
to filter out the other candidate categories.
This approach already eases KR a bit. (04)
To illustrate the importance of quantifiers and scoping,
I can't resist to refer to my article for ICCS'02
http://webkb.org/article/iccs02/. The section 7 shows that the
sentence "4 judges have approved 3 laws" can be given at least
16 different interpretations (to have an idea, think about the
differences with "3 laws have been approved by 4 judges") and
even 32 but that begins to be a bit hair-brain.
If quantification is not represented, all we may have is
[judges, approver: laws] instead of [4 judges, approver: 3 laws]
(furthermore, WebKB-2 can remove the 's' at the end of "judges"
and "laws" when numerical/universal quantifiers are used, and
hence find which category is referred to, but there would be
ambiguities without the use of quantifiers). (05)
As this section 7 shows, even with formal structures, keeping the
ambiguity of natuaral languages sentences such as "4 judges have
approved 3 laws" is possible. [4 judges, approver: 3 laws] is ok.
More precisions are possible but not mandatory. (06)
Philippe (07)