User-assisted query translation for interactive CLIR

Abstract

We view interactive Cross-Language Information Retrieval (CLIR) as an iterative process in which the searcher and the retrieval system collaborate to find documents that satisfy the searcher’s needs, regardless of the language in which those documents are written. Our motivation is that humans and machines can bring complementary strengths to this process. Machines are excellent at repetitive tasks that are well specified; humans bring creativity and exceptional pattern recognition capabilities. Properly coupling these capabilities can result in a synergy that greatly exceeds the ability of either human or machine alone. Designers of CLIR systems can select from a variety of fully automatic techniques to overcome problems with unknown terms and translation ambiguity, but automatic processing of this sort risks reducing the searcher’s understanding of system operation. This, in turn, tends to work against the synergy that we seek to accomplish. We are therefore exploring more transparent approaches to support interactive CLIR.