Interlingual Information Extraction as a Solution for Multilingual QA Systems
Information extraction systems have been dealt with at length from the viewpoint of users posing definite questions whose expected answer is to be found in a document collection. This has been tackled by means of systems that analyse the user query and try to use the grammar features of each language to find a possible answer. This approach has failed to work for users of other languages or for documents in different languages, save for a few languages for which the query can be machine translated to the target language or all languages. Where there are more languages, however, this approach is impracticable for information extraction in a reasonable time. The massively multilingual approach (> 6 languages) necessarily involves the language-independent representation of the contents, that is, using an interlingua. This paper reports a promising early trial of a method that launches a query in any language against a language-independent representation of the document set using a general-purpose UNL interlingua and receives a precise response.
Keywordsmultilingualism information extraction UNL Interlingua
Unable to display preview. Download preview PDF.