Automatic machine translation selection scheme to output the best result
Keiji YASUDA (ATR Spoken Language Translation Research Laboratories,Doshisha University)
Fumiaki SUGAYA (ATR Spoken Language Translation Research Laboratories)
Toshiyuki TAKEZAWA (ATR Spoken Language Translation Research Laboratories)
Seiichi YAMAMOTO (ATR Spoken Language Translation Research Laboratories)
Masuzo YANAGIDA (Doshisha University)
An automatic selection method for an integrated multiple MT system is proposed. This method employs a machine learning approach to build an automatic MT selector. The selector learns based on the parameters of MT systems and the evaluation result provided by a human evaluator. An experiment is conducted on two MT systems developed in our laboratories. Experimental results show the effectiveness of the proposed method. The ratio of correct selection is 76%. According to the system performance evaluation result, the integrated MT system using the proposed method gives a better performance than each individual MT system.
Speech translation, Automatic selection, TOEIC, Translation paired comparison