Abstract
Language barrier is one key challenge to global collaboration where people speaking multiple languages have to exchange messages and understand each other. To bridge different languages at a large scale, interactive translation that involves non-experts to iterate the inputs for machine translation (MT) and to enhance the overall quality is potentially helpful. One design challenge is how do we support the collaboration between human workers and MT, especially how to provide feedback of translation quality to inform workers’ subsequent editing actions. In a lab study and a field study, we evaluate the effects of different types of interface feedback (back translation, numeric score of estimated translation quality, and anthropomorphic social messages as a way of MT-worker communication). The results confirm the utility of using numeric score and social messages as feedback, and shed light on the design of MT interface and cross-lingual communication support. The results further show that not all social features are useful. Facial expression does not add value while showing emotional valence tends to increase positive experience and perception of the work (with positive emotional cues) or improve the quality of translation (with negative emotional cues).