Cross-language information retrieval (CLIR) Track overview

There are three main ways in which cross-language information retrieval approaches attempt to "cross the language barrier" - through query translation, or document translation, or both. (Oard, 1997). CLIR research started out with experiments using controlled vocabularies and associated dictionaries and thesauri, but nowadays free text approaches are most common. These approaches also dominate experiments in past and present CLIR tracks. Free text methods can be further classified according to the resources used to cross the language boundary: machine translation, machine-readable dictionaries, or corpus-based resources. Machine translation (MT) seems an obvious choice for cross-language information retrieval systems. It also played a large role in the TREC-8 experiments of a number of groups. However, CLIR is a difficult problem to solve on the basis of MT alone: queries that users typically enter into a retrieval system are rarely complete sentences and provide little context for sense disambiguation.Corpus-based approaches are also popular. Groups experimenting with such approaches during this or former CLIR tracks include Eurospider, IBM and the University of Montreal.Lastly, a significant number of cross-language retrieval approaches make use of existing linguistic resources, mainly machine-readable bilingual dictionaries. Various ideas have been proposed to address some of the problems associated with dictionary-based translations, such as ambiguities and vocabulary coverage. One of the groups that have investigated the use of such dictionaries is the Twenty-One consortium.