Upper-level ontologies capture mostly concepts that are basic for the human understanding of the world. They are "grounded" in (supported by, wired to) the common sense that makes it difficult to formalize a strict definition for them. They represent prototypical knowledge using mainly taxonomic relations.

This division seems to be a question of goal and scope of the developers of the ontology rather than a representational or management problem. However, there exists a significant real difference between the two types of ontologies. The domain-specific ontologies that are trying to capture, for example, a market segment or certain scientific area typically consist of well-defined concepts. For example, in the natural sciences (Mathematics, Physics, Chemistry, Biology, Medicine) the knowledge is easy to formalize because it is more or less systematic --- it could be expressed using well-defined scientific terms. In such cases, the objects in the universe of discourse are either purely abstract or they are some idealized/simplified models of the real phenomena in the world.

The so-called lexical knowledge bases (LKB, such as WordNet) are lexicons, thesauri, or dictionaries that attempt to formalize the lexical semantics - the meanings of the words in one or more natural languages. Similar to the upper-level concepts, the meanings of the words are grounded in the common understanding of huge populations - there are no formal definitions, the words can bear a number of different meanings often based on associations, typical uses, collocations, and prototypical knowledge. Going further, the meanings of many words are just primitive concepts. Some upper-level ontologies were developed on the basis of a LKB - such an example is the SENSUS ontology. Other upper-level ontologies were developed in order to give formal semantics to a LKB - such an example is the EuroWordNet Top Ontology.

In fact, LKB are huge knowledge bases with relatively simple relations between the concepts. Usually there is a single concept for each word sense. For example the word "bank" is related to (at least) two concepts corresponding to its different meanings: bank as an institution and also bank as a river bank. Further more, there are various relations between the concepts, but the most important seem to be these between a sub-category and its super-category. For example, the relation between "duck" and "bird", encoding that every duck is also a kind of bird.

During the last years the interest in huge knowledge bases to be used for more efficient text processing is increasing in both scientific and “industrial” society. It is mostly because the “total” natural language understanding requires “true” artificial intelligence – a non-human agent capable of passing the Turing test. That is still a science fiction.

The lexical knowledge bases come as a reasonable simplification – they are already available and provide a serious improvement for many text processing applications. A sample application of a LKB is for improvement of text search in order to enable searching by meaning or concept rather than by word. For example, if a query “gray bird” is put against AltaVista, documents containing “gray duck” will not be judged as relevant. A LKB can be used to correct this as reported in [KiryakovSimov99]. In fact there are already two innovative search engines that are partially emplying this approach (Simpli.com and Oingo.com). Other typical applications are information filtering and message understanding.

Currently the evaluation of feasibility of general-purpose ontologies and upper-level models is expensive mostly because of technical problems such as different representation formalisms and terminologies used. Additionally, there are no formal mappings between the upper-level ontologies that could ease any kind of studies and comparisons. We present the OntoMap, a project with the pragmatic goal to facilitate the access, understanding, and reuse of such resources. A semantic framework on conceptual level is implemented that is small and easy enough to be learned on-the-fly. We tried to design the framework so that it captures most of the semantics usually encoded in upper-level models. Technically, OntoMap is a web-site providing access to several upper-level ontologies and manual mapping between them. Read more in [KiryakovEtAll2000d]

Additionally a mapping between the most comprehensive common-sense knowledge base (Cyc) and the EWN Top Ontology – the system of semantic features used to formally classify the base lexical concepts in EWN. The result is available as MS Access database. Some important theoretical issues related to this work were reported in [KiryakovSimov2000a] and [KiryakovSimov2000b]. The mapping is also available as online service at http://demo.ontotext.com

OntoText Lab. was a sponsor of 

Following the success of the first issue, the second one took place on 27th of May 2002 at Las Palmas, Canary Islands - Spain (preceding the 3rd International Language Resources and Evaluation Conference, LREC2002). It was supported by two EU thematic networks: OntoWeb and ELSNET. The Program committee gained more than ten of the world top researchers in the area.