Combination Approaches for Multilingual Text Retrieval

Abstract

We describe the Eurospider component for Cross-Language Information Retrieval (CLIR) that has been employed for experiments at all three CLEF campaigns to date. The central aspect of our efforts is the use of combination approaches, effectively combining multiple language pairs, translation resources and translation methods into one multilingual retrieval system. We discuss the implications of building a system that allows flexible combination, give details of the various translation resources and methods, and investigate the impact of merging intermediate results generated by the individual steps. An analysis of the resulting combination system is given which also takes into account additional requirements when deploying the system as a component in an operational, commercial setting.

Enriching document representation via translation for improved monolingual information retrieval

“…McCarley [28] trained a statistical MT system from a parallel corpus, applied it to perform QT and DT, and showed that the combination of scores from QT and DT drastically improved either method alone. Similar results have been reported using either a full-fledged MT system [4] or a simple translation algorithm [7]. Kraaij et al [19] used the translation model of IBM Model 1 [5], obtained from an automatically constructed parallel corpus from the web, and also reported that the combination of QT and DT improved either method alone.…”

“…For Chinese translation representation, different values of μ were used for P (w|d) and P (d|q) of Eq. (11), by fixing to 0 and 4 The smoothing parameter μ at the second retrieval was fixed to 1,500 for both representations.…”

“…Even though a common approach in CLIR is to perform query translation (QT) using a bilingual dictionary [32], there were studies showing that combining both QT and document translation (DT) improved retrieval performance in CLIR by using bilingual representations in both the source and target language [28,19,7,4]. McCarley [28] trained a statistical MT system from a parallel corpus, applied it to perform QT and DT, and showed that the combination of scores from QT and DT drastically improved either method alone.…”

Matching meaning for cross-language information retrieval

“…Combining the two ranked lists yielded better MAP than when either approach was used alone. Similar improvements have since been reported by others using variants of that technique (Braschler, 2004;Kang et al, 2004). Boughanem et al (2001) tried one way of pushing this insight inside the retrieval system, simply filtering out potentially problematic translations that were attested in only one direction.…”

Bibliothèques et accès à l’information multilingue : pour en finir avec Babel

“…Ces ressources linguistiques ont chacune démontré leurs forces et leurs faiblesses. Or, il semble que la démarche la plus prometteuse consiste à combiner plusieurs ressources différentes pour la traduction (Braschler 2004).…”