神經(jīng)網(wǎng)絡(luò)機(jī)器翻譯文學(xué)作品的質(zhì)量評(píng)價(jià)：以百度翻譯與谷歌翻譯為例

夏江南

摘要：端到端的神經(jīng)網(wǎng)絡(luò)機(jī)器翻譯系統(tǒng)極大地提高了翻譯的流暢性和可讀性，但同時(shí)也帶來了新的問題：流暢性可能會(huì)犧牲忠實(shí)度，而傳統(tǒng)的機(jī)器翻譯評(píng)價(jià)方法——BLEU已經(jīng)不再適用于翻譯質(zhì)量的評(píng)價(jià)。本文以《水滸傳》及其英譯版本為研究語料，從誤差分析的角度，結(jié)合翻譯質(zhì)量評(píng)估研究的理解，集成一組機(jī)器翻譯質(zhì)量評(píng)估模式，利用文獻(xiàn)研究和定性分析相結(jié)合的方法，針對(duì)同一語料文本，將當(dāng)前主流的兩大機(jī)器翻譯軟件--谷歌翻譯和百度翻譯的譯文與人工翻譯的譯文進(jìn)行比對(duì)，從本體錯(cuò)誤、語段錯(cuò)誤兩個(gè)層面來分析機(jī)器翻譯的成果。研究結(jié)果顯示，即使是作為當(dāng)今最先進(jìn)的神經(jīng)網(wǎng)絡(luò)機(jī)器翻譯軟件，在文學(xué)翻譯方面也依然存在著很大的問題。

Comparison and analysis

Substance errors

Original text： “那清河縣里有一個(gè)大戶人家，有個(gè)使女，小名喚作潘金蓮，年方二十馀歲，頗有些顏色?！?/p>

Google： “There was a large family in Qinghe County，and there was a maid named Pan Jinlian，who was more than 20 years old and she was rather colorful.”

Baidu： “There is a wealthy family in naqinghe county.There is a maid named Pan Jinlian.She is in her twenties and has some color.”

This segment introduces the background of Pan Jinlian's life and the reason why she married Wu Dalang.Baidu incorrectly translated“清河縣”as "naqinghe county " and the Google translation mistakenly identified “武大”as "the University of Wuhan"，which is confusing.

Text errors

Original text： “拿了燭臺(tái)，引著大王，轉(zhuǎn)入屏風(fēng)背后，直到新人房前?！?/p>

Google： “I took the candlestick，led the king，and turned to the back of the screen until the newcomer's room.”

Baidu： “He took the candlestick and led the king to the back of the screen until he was in front of the new man's room.”

In terms of terminology errors，both translation software incorrectly translated “新人”into new people.In China，“新人” refers to the person who is about to get married，especially the bride.

Conclusion Machine translation is a powerful tool for translators.They should accept machine translation with an open mind，learn to use machine translation to help them complete the heavy translation work，rather than blindly pessimistic attitude or boycott.

References

[1]Booth D Andrew，et al.1993.Special issue on Evaluation of MT Systems [J].Machine Translation，8（1-2）： 1-126.

[2]Carl James.1980.Contrastive Analysis [M].London and New York： Longman.

[3]Carl James.2001.Error in language learning and use： exploring error and analysis [M].Beijing： Foreign Language Teaching and Research Press