Can indicating translation accuracy encourage people to rectify inaccurate translations?

Abstract

The accuracy of machine translation affects how well people understand each other when communicating. Translation repair can improve the accuracy of translated sentences. Translation repair is typically only used when a user thinks that his/her message is inaccurate. As a result, translation accuracy suffers, because people's judgment in this regard is not always accurate. In order to solve this problem, we propose a method that provides users with an indication of the translation accuracy of their message. In this method, we measure the accuracy of translated sentences using an automatic evaluation method, providing users with three indicators: a percentage, a five-point scale, and a three-point scale. We verified how well these indicators reduce inaccurate judgments, and concluded the following: (1) the indicators did not significantly affect the inaccurate judgments of users; (2) the indication using a five-point scale obtained the highest evaluation, and that using a percentage obtained the second highest evaluation. However, in this experiment, the values we obtained from automatically evaluating translations were not always accurate. We think that incorrect automatic-evaluated values may have led to some inaccurate judgments. If we improve the accuracy of an automatic evaluation method, we believe that the indicators of translation accuracy can reduce inaccurate judgments. In addition, the percentage indicator can compensate for the shortcomings of the five-point scale. In other words, we believe that users may judge translation accuracy more easily by using a combination of these indicators. © 2011 Springer-Verlag.