WebKB-2 helps people build queries or statements (about any subject or domain) in a precise, uniform and machine-understandable way by associating a form-to-fill to each category (word meaning) it knows (WebKB-2 currently knows about 78,000 categories; they represent the meanings of about 112,000 English nouns). Each form proposes relations (or characteristics) that can be used to describe an object of the selected category. To do so, WebKB-2 exploits schemas associated to the category or its generalizations. Any user can add a schema to a category.
For the destination of a relation, you may specify:
#Paris)
or one of its names
(e.g. Paris or
capital_of_France);"+61 7 1234 5678" and
"Paris"); the forms display
"..." where strings have to be entered;
do not use quotes around objects that are not strings);-123,456.789),
a date (e.g. 31/12/2001 and
2001),
a time (e.g. 23:59:59 and
31/12/2001 23:59);a #cat, many cats,
3 #cat,
between 3 and 5 cats,
57% of #cat and
5.5 to 7 kg
(click here for a list and a grammar of
accepted quantifiers) -- if the form is for querying knowledge,
you may omit the quantifier, the existential quantifier some is
then selected (this is not possible when knowledge is stated);
{#New_York, 2 #capital_city};
(2 #person, experiencer of: (a living, place: {#Paris, #New_York}))
(sometimes, a form proposes to access another form to associate
relations to a certain object).Most of the proposed relations have a "signature" that specifies the kinds of acceptable source and destination objects. When you submit the form, WebKB-2 checks that the signatures of the used relations are respected or exploits them to find adequate categories when category names are used instead of category identifiers. Relation signatures are a basic but surprisingly efficient way to detect mis-interpretations of mis-uses of categories. The relations that WebKB-2 proposes and the controls it does are necessary for the entered knowledge to be "understandable" and exploitable by the machine, as opposed to texts in English or French, for example. Knowledge representation is a slow and difficult manual activity (outside very limited domains, the semantic content of a text in a natural language cannot be automatically translated into knowledge representations). However, we believe that WebKB-2 is currently the knowledge server providing the easiest way to enter reasonable amounts of general and expressive knowledge, thanks to the knowledge representation languages we developed and the exploitation of a semantic dictionary of English (plus the relations and schemas we associated to top-level categories) for generating forms and allowing the use of category names intead of category identifiers. (Note: the "concept map" language/model is easier to use than the languages used in WebKB-2 but is also far less precise, formal and exploitable; it may or may not be considered as a knowledge representation language; we may extend WebKB-2 in the near future to accept "concept map" representations too).