Enhancing Cross-Language Question Answering by Combining Multiple Question Translations

Abstract

Abstract.One major problem of state-of-the-art Cross Language Question Answering systems is the translation of user questions. This paper proposes combining the potential of multiple translation machines in order to improve the final answering precision. In particular, it presents three different methods for this purpose. The first one focuses on selecting the most fluent translation from a given set; the second one combines the passages recovered by several question translations; finally, the third one constructs a new question reformulation by merging word sequences from different translations. Experimental results demonstrated that the proposed approaches allow reducing the error rates in relation to a monolingual question answering exercise.

Exploiting Wikipedia and EuroWordNet to solve Cross-Lingual Question Answering

“…[42] proposes a candidate question translation obtained by means of different on-line machine translators and also implements some heuristics in order to correct and combine the on-line translations to achieve an acceptable one. Finally, the three methods exposed in [1,2] try to improve the quality of the translation of the query. The first one focuses on selecting the most fluent translation from a given set; the second combines the passages recovered by several question translations; and the third constructs a reformulation is a new question by merging word sequences from different translations.…”

“…Combining Multiple Query Translation. These systems process different translations in order to obtain an optimal translation [42,30,27,1,2].…”

Graph-Based Answer Fusion in Multilingual Question Answering

“…We believe this behavior is consequence of having several incorrect answer translations. In order to reduce the errors on these translations we plan to apply, as future work, some techniques for combining the capacities of several translation machines [1]. It is important to point out that the proposed scheme allows easily integrating several translations for each answer, and therefore, incrementing the possibility of retrieving the correct one.…”

“…They mainly consider the selection of the best in- Figure 1. General architecture of a multilingual QA system stance from a given set of translations [1,11] as well as the construction of a new query reformulation by gathering terms from all of them [10,14].…”

“…The first problem, namely, the translation of the questions from one language to another, has been widely studied in the context of cross-language QA 1 [1,10,11,14]. In contrast, the second task, i.e., the fusion of answers obtained from different languages, has only recently been addressed [2].…”