Workflow Analysis of Collective Translation

The necessity of translation continues to increase because of the rapid spread of multi-language activity, especially on the World Wide Web. As result, we need new way of translation because the number of professional translators is limited. One way is using the power of crowdsourcing. Crowdsourcing is outsourcing various tasks to workers on the WWW. Crowdsourcing is classified into several categories, especially micro-task is paid attentio. Micro-task platforms such as Amazon Mechanical Turk allow us to gather many workers with scant rewards. Our aim is to obtain high quality translation results from micro-tasks. However , since translation is so difficult, result quality is generally not enough if just one worker is assigned to translate the complete document. Translating a document , which is an assemblage of single sentences, raises the importance of task decomposition because we can outsource only simple tasks to micro-task workers. In this situation, we have to design the workflow carefully and assign micro-tasks to the appropriate workers. For designing effective workflows, we address below three issues. Analysis of the effect of mother language on translation by micro-task It is clear that the mother language of the translators greatly impacts translation performance. Current micro-task markets such as the Amazon Mechanical Turk allow the nationality of workers to be controlled. So we try to make use of strong points of workers of source language native and target language native by knowing characteristics of them. Process design also impacts the quality of the results. The simplest approach to decomposing the translation task is to split the document into single sentences and translating each in isolation. For creating a more effective translation process, we have to analyze the features of this process. Proposal of new collective translation processes Based on new knowledge, we propose a new collective translation processes. For the first issue, we analyzed the effect of workers' mother language ii through an experiment. We obtain new knowledge such as workers whose mother language is the target language can more efficiently improve translation results by other translator. For the second issue, we hold an empirical collective translation on the process and analyze the results. We created a draft translation by the document translation process of combining single translation results and asked a professional translator to revise it. We analyzed the difference between the two translation results, including interviews to the professional translator. At last, we proposed a new collective …