It's a polyglot world out there. With corporations thinking global thoughts and Internet-based interactions spanning the world, single-language retrieval systems are severely limiting. The time is ripe for a good multilingual retrieval system.
To answer this need, Manning & Napier Information Services (MNIS) rolled out its newest product, CINDOR, last month at Online Information '99 in London. Like the other MNIS products, this one is based on full natural language processing of both documents and queries. Its strong point is its "Conceptua-Interlingua" underpinning, which matches the ideas in a query in any language to the ideas in documents in any other language in the system. Instead of performing a direct translation, or using a built-in multilingual dictionary as most existing multilingual systems do, CINDOR codes both documents and queries for the concepts they contain. Then it matches the concepts, not just the words. From a retrieval standpoint, the results are greatly improved. An additional advantage to this concept-based approach is that it is possible to retrieve between uncommon pairs of languages, such as Japanese and Spanish, even though a Japanese-Spanish lexicon may not exist.
To use CINDOR, type in a query in your own language, and specify the collections to search. Results are returned grouped by language. They can be viewed in their original language or they can be translated using a document translation system.
At present, CINDOR can be used to search a collection of intellectual property databases through the MNIS MAPIT service, but it is primarily targeted as a stand-alone software module for incorporation into other systems. It should enhance their retrieval capabilities markedly. Paraic Sheridan, director of MNIS's research division, MNIS-TextWise Labs, said, "Conceptual matching ensures more precise results to search requests, while CINDOR's language-independent internal representation makes searching across a wide range of language combinations achievable."
CINDOR is a modular system. It currently supports English, French, Spanish, and Japanese. Additional modules can add more languages. Mary McKenna, CINDOR product manager, anticipates immediate use for CINDOR in searching non-English patents, multilingual collections of documents for corporate intranets, or multilingual e-commerce sites.
Anyone who has ever used a machine translation system can tell some funny stories of mistranslations, such as, "Where can I get my hair cut?" translated as "Oł puis-je obtenir mes cheveux pour couper?" ( ... which means, "Where can I obtain my hair for cutting?") Multilingual systems that rely on machine translation are hindered by their inability to render the same ideas accurately in a second language. By retrieving instead on the concepts involved, CINDOR can avoid these common errors.
The product also takes advantage of its foundation in MNIS's DR-LINK information system. DR-LINK also performs concept mapping and clustering, but it provides, in addition, proprietary lexicons that can expand geographic names such as Europe to include all the constituent countries and cities. It also disambiguates (decides which meaning of a word is the right one) from the context of the sentence or paragraph.
The launch of CINDOR, along with the recent announcements of Oingo and Simpli.com (see the December 20, 1999 NewsBreak), marks the entry of natural language into commercially viable information products and a step into the next generation of natural-language-processing-based information systems.
For more information, contact MNIS at 716/631-2327, extension 230, or read about CINDOR at http://www.cindorsearch.com.