Translingual mining from text data

Abstract

Like full-text translation, cross-language information retrieval (CLIR) is a task that requires some form of knowledge transfer across languages. Although robust translation resources are critical for constructing high quality translation tools, manually constructed resources are limited both in their coverage and in their adaptability to a wide range of applications. Automatic mining of translingual knowledge makes it possible to complement hand-curated resources. This chapter describes a growing body of work that seeks to mine translingual knowledge from text data, in particular, data found on the Web. We review a number of mining and filtering strategies, and consider them in the context of statistical machine translation, showing that these techniques can be effective in collecting large quantities of translingual knowledge necessary for CLIR.