Research Of Crowdsourcing Quality Management Mechanism Of Machine Translation

The technology of machine translation have been developed in many years,it made a change from rule-based translation technology to statistical machine translation technology,and then it becomes the models that based on neural network and cloud translation,more and more languages are supported,and the quality of translation has been improved gradually.But the quality of semantic and contextual still has many problems to be researched and solved.In particular in the translation of small languages,due to the lack of resources,resulting in the use of statistical machine translation technology is difficult to obtain satisfactory translation results.Crowdsourcing have began to be widespread concerned by people,as a flexible and effective way to solve the problem.Our main purpose of research is that puts crowdsourcing into semantic correction and quality in small language translations,Improving the quality of the translation.But crowdsourcing takes a work way that facing the unknown public groups,and the worker who undertakes the task has the characteristics of independent anonymity.So while crowdsourcing technology has been applied,it's key point for results screening and quality control in the process of crowdsourcing.This paper is based on the research of filter result and quality control method on crowdsourcing,by introducing the median voter theory,design a mechanism of quality management and results screening in machine translation.Firstly,The mechanism is based on the principle of crowdsourcing to complete the task design,decomposit and distribute,and incorporating Median Vote and iterative methods,then to filter a better answer from crowdsourcing results and improve the quality of translation.In order to obtain a better translation effect than the machine translation.Finally,using the Chinese and Thai translation corpus as research object,the method is verified by experiments.The experimental results show that applying the technology of crowdsourcing in relevant process of machine translation,using the paper results to filter and control the feedback results could improve the semantics and the quality.