WebKB-2 Technical Profile As an Ontology Server

Availability. WebKB is directly usable on-line (www.webkb.org). For firms, WebKB-2 sources/binaries are available for fee.

Modeling Features/Limitations. The model used in WebKB-2 is intended to permit a logical-based and normalized storage of any natural language sentence (hence, expressivity is paramount); however, the inference engine is comparable to the ones of other good ontology servers (e.g. LOOM/Ontosaurus) which means that not all the expressivity is exploited for logical inferences (e.g. contexts are currently exploited as if they were positive contexts and there is no rule-based system). The focus of the inference engine is on constraint checking and knowledge retrieval.

Base Language. FS (i.e. FO, FCG, ...), an extension of CGLF. See http://www.webkb.org/doc/languages/

Web Support. CGI interface, HTML interfaces, internal script language, category identifiers can be URIs, and all categories in the KB can be accessed from the Web via URIs.

Import/Export Formats. Taxonomies can be exported in DAML/RDF, CGIF and KIF; these languages are currently only partially parsed;

Graph View. Only via a connection to Ontorama, i.e. only for taxonomies.

Consistency Checks. Complete syntactic checking; detection of forbidden cycles for transitive links, of exclusion link violation, of relation signature violation.

Multi-user Support. The KB is shared by all users but each element (category, link, nodes, relations, ...) has an associated author; this is exploited for avoiding lexical conflicts, semantic conflicts, filtering on knowledge authors, ... See http://www.webkb.org/doc/papers/iccs01/

Ontology Merging. No automatic support for comparing and merging independently built ontologies, but a framework that permits to store these ontologies into a single KB and hence tightly interconnect their elements (as opposed to approaches where ontologies are kept in different files/modules).

Lexical Support. Each category may have many names (possibly shared by other categories) but has a unique identifier that reuses the most common name and the category author identifier. Graphs (and query graphs) may use names instead of identifiers as long as there is no ambiguity (i.e. when there is only one category for the name or if the relation signatures permit to exclude other candidates). Categories may be searched according to their authors, names, parts of their names.

Information Extraction. There is no data-mining mechanism or natural language parsing mechanism.

Comments. An integration, correction and extension of WordNet 1.7 and various top-level ontologies is proposed as a default ontology in the KB and exploited for easing and guiding the entering of knowledge, generating menus, and checking the entered knowledge.

More Info. Home page: http://www.webkb.org/doc/w2doc.html
Article: http://www.webkb.org/doc/papers/wi02/