A Comparison of Two Corpus-Based Methods for Translingual Information
Retrieval

Michael L. Littman and Fan Jiang

online version: http://www.cs.duke.edu/~mlittman/papers/tr98-cllsi.ps

(local: /u/mlittman/web/papers/tr98-cllsi.ps)


In translingual information retrieval (TIR), ad hoc queries in any of
a set of languages can be used to retrieve documents in any of a set
of languages.  Classical information-retrieval methods such as the
vector-space model cannot be applied to TIR because they base
similarity on the overlap of terms between queries and
documents---this is typically zero in TIR.  The generalized
vector-space model (GVSM) and latent semantic indexing (LSI) are two
variations of the vector-space model that make comparisons outside of
term space.  For this reason, both can be and have been applied to
TIR.  In this paper, we report on a series of experiments comparing
the performance of GVSM and LSI on monolingual and translingual
retrieval tasks.  We find that the performance of both methods depends
crucially on parameter settings, that LSI performs better, and that
GVSM runs more quickly.