Confidence Value Calculation for Cultural-Based Misunderstanding Warning in Multilingual Chat

Abstract

Previously, the Cultural Difference Detection (CDD) concept had been introduced to predict possible misunderstandings when machine translation (MT) users communicate across cultures. CDD compared images that linked to a concept in two languages with the assumption that the low similarity of images can indicate cultural misunderstanding. Its result has been used for warning MT users of words that could possibly cause misunderstanding. However, how it should be applied in the warning system has not yet been studied. One of the complications with the warning system is the accuracy of CDD. This paper proposes a method to calculate confidence value and to use this value to predict the accuracy of CDD. This confidence value could be used to decide whether to warn or not to warn the user when a suspect word is typed in the text box. This confidence value is calculated from the similarity among the images used during the CDD calculation. To examine the proposed confidence value, we conducted an experiment by comparing the accuracy of CDD and the confidence value. The experiment confirms that our proposed confidence concept reflects the CDD accuracy as there is a strong correlation between the confidence value and the CDD accuracy.