Asking the Right Question: How to Transform Multilingual Unstructured Data to Query Semantic Databases

Ontology engineers have long tried to develop mechanisms to automatically transform natural language statements into queries knowledge-systems can deal with. This has been an enormous challenge as natural languages are highly ambiguous and contexts for disambiguating are seldom identifiable through simple linguistic patterns. To circumvent these difficulties, developers of knowledge bases have often opted for the use of a restricted vocabulary and syntax. Normal users, nevertheless, prefer to express themselves in their language. Special languages or schemas tend to reflect one language – the developer’s – and make extensibility more difficult. Also multilingual access can be more difficult to handle in that way. In this article we present strategies for transforming queries of natural languages into language-neutral representations that can be more easily transformed into semantic queries. We describe a tool that combines a multilingual database and natural processing modules with a semantic database in order to transform queries in Dutch, French and English into queries from which ambiguity at syntactic and semantic levels have been reduced. We focus on certain aspects of natural language such as negation and collocation preferences to deal with semantics.