New Method For Building Multilingual Ontologies That Can Be Applied To The Semantic Web

An ontology is a structured set of terms and concepts underpinning the meaning of a subject area. Artificial intelligence and knowledge representation systems are the principal users of ontologies. Researchers from all over the world are now working on applying ontologies to the Internet with the aim of building a Semantic Web and providing users with a intelligent tool for using the information on the web.

The importance of the proposal by Jesús Cardeñosa, Carolina Gallardo, Luis Iraola and Miguel Ángel de la Villa is that it revolutionizes current ontology building systems. Until now, these systems have had a major stumbling block: the multilingual component. The application of ontologies to the Internet comes up against serious problems triggered by linguistic breadth and diversity. This diversity stands in the way of users making intelligent use of the web.

So far all approaches to solve the multilingual component have run into serious difficulties. Some were based on experts agreeing on the terms to be used in each language. These endeavours failed to take off due to the difficulty of finding experts in many languages. Other approaches set out to use one language (almost always English) as a pivot. Again the results were held back, this time because the use of a natural language as an interlanguage occasions ambiguity. As the researchers note, early attempts at using a natural reference language to build machine translation systems go back 20 years ago, and the results were no good.

Multilingual thesauruses

Neither has the 1985 international standard on the creation and development of multilingual thesauruses (ISO 5964:1985) solved the problem. A thesaurus is a list of hierarchically interrelated (general and subordinate terms) terms, possibly containing more than one word, used to index (for archiving purposes) and retrieve documents.

Although an ontology is not exactly the same thing as a thesaurus, the researchers claim that it is very similar because a great many ontologies are actually a simplified version of its huge potential for representation, confined to three basic relations (“is-part-of”, “is-a-type-of”, “is-a”). The above ISO standard on thesauruses covers these three basic relationships.

According to these researchers, the current approaches for solving the multilingual component in ontology building are mostly applied to ontologies with the same representation power as a thesaurus, without even applying the above ISO 5964:1985 standard to deal with multilingualism. But, even if this standard is applied, the approach always has to use a reference language, which, as mentioned, has never worked.

Language-independent ontologies

The method proposed by the School of Computing researchers solves this problem because it is based on building ontologies that can represent information irrespective of the language. They are therefore applicable to multilingual systems.

This method is an advance in that, after analysing the natural language, the method looks for linguistic patterns (grammatical structures) that match the exact ontological structures. This it does multilingually, as the linguistic patterns are capable of building language-independent structures.

The innovative thing about what these researchers are proposing is the construction of multilingual ontologies using what are known as universal words as the concept name. The concept of universal word stems from the United Nations University’s UNL Project (Universal Networking Language). This project was set up to break down the linguistic barriers on the Internet. The researchers claim that the characteristics of UNL also very close match the features of an ontology.

These researchers assume that ordinary texts contain more information than what can be extracted using just the domain terms, as any text has implicit ontological relations that can be extracted by analysing certain grammatical structures of the sentences making up the text.

A case study

In an article presented last July at The 2008 International Conference on Semantic Web and Web Services (SWWS'08), these researchers describe their approach and explain a case study demonstrating the validity of their method. The case study used the contents of the current catalogue of Spanish monuments as part of the Patrilex project, funded by the Spanish National Research Plan in conjunction with the Underdirectorate General of Cultural Heritage.

In this case study, the sentences from the catalogue of Spanish monuments were coded in UNL language. This codification is a semantic representation of the catalogue contents. The researchers then searched for predefined linguistic patterns in the semantic representation. After identifying the contents matching the patterns, they instantiated the contents as ontological structures.

The big advantage of using the UNL system is that the universal words are independent of the language and are not ambiguous. The non-ambiguity makes the translation of the ontology built this way to any language extremely precise.

The universal words used can be viewed and used publicly at the public access repository developed by the researchers. Universal words are usually used to develop multilingual dictionaries, as explained in another press release.