Comparing Google's Neural Machine Translation And Human Translation Of Conceptual Metaphors

Recent decades have seen enormous advances in artificial neural network,machine learning and deep learning.Artificial intelligence has entered a new development stage.Hence,machine translation also underwent transformative changes and witnessed great improvement in the translation quality.In 2016,Google launched its new translation system Google Neural Machine Translation.The GNMT system uses end-to-end approach and most advanced training techniques,and embraces an encoder-attention-decoder architecture with 8 encoder LSTM layers and 8 decoder layers,which reduces 55%-85%translation errors among several major language pairs.However,when translating Chinese into English,the GNMT system still encounters a set of problems,especially in metaphor translation.This study aims to explore how GNMT deals with conceptual metaphors on the basis of Conceptual Metaphor Theory(CMT),in comparison with the human translation.In total,5,111 metaphors were collected from Xi Jinping:The Governance of China II.The categories of the 5,111 metaphors were identified based on Pragglejaz Group's metaphor identification procedure(MIP)and the frequencies of different categories of metaphors were calculated.The results show that of all conceptual metaphors collected from the book,JOURNEY metaphor has the highest frequency(58.07%),and BUILDING metaphor has the second highest frequency(19.40%),while ANIMAL metaphor has the least occurrence(0.13%).When translating Xi Jinping:The Governance of China II,GNMT can be applied to BUILDING metaphor,JOURNEY metaphor,MACHINE metaphor,PLANT metaphor,WAR metaphor,and WEATHER metaphor,while GNMT cannot be applied to ANIMAL metaphor,DISEASE metaphor.FAMILY metaphor,and HUMAN metaphor,due largely to cultural factors.It is hoped that these results may enable language service providers to have a deeper understanding on how to produce metaphor translation with the use of the GNMT system.The results may also be helpful for machine translation R&D scientists to improve their system on the front of metaphor translation...