An analysis of machine translation errors on the effectiveness of an Arabic-English QA system

The aim of this paper is to investigate how much the effectiveness of a Question Answering (QA) system was affected by the performance of Machine Translation (MT) based question translation. Nearly 200 questions were selected from TREC QA tracks and ran through a question answering system. It was able to answer 42.6% of the questions correctly in a monolingual run. These questions were then translated manually from English into Arabic and back into English using an MT system, and then re-applied to the QA system. The system was able to answer 10.2% of the translated questions. An analysis of what sort of translation error affected which questions was conducted, concluding that factoid type questions are less prone to translation error than others.