外刊精读 | 机器翻译的崛起，未来还需要人工翻译吗？

今天是小芳老师陪你精读外刊的第29天

难度指数7级（共8级）

读前想想，下面内容用英语怎么说？

1. 混合体 2. 翻译
3. 差一点 4. 有趣的

第一步：泛读原文（盲听1遍，泛读1遍，对照文本听1遍）

The translator of the future is a human-machine hybrid. There will still be a place for humans with specialist knowledge

This first paragraph can read a bit oddly. As an experiment, Johnson wrote it in English, and then ran it through Google Translate to turn it into French, before running it through the same process in reverse, to render it back to English.

But the prose is only slightly odd. The software has turned the original “may” into “can”, which made the first sentence sound a little off. In the second, the original “has written” became “a écrit” in French, which was then translated as “wrote” in English—accurately, but with a small change in emphasis.

Once such round-trip translations were an endless source of humour. Some people took serious texts and put them through this two-way process yielding hilarious results. And bad computerised translation is hardly consigned to the past. “Chinglish” signs are still a source of amusement for English-speakers visiting the Chinese-speaking world. On one menu, sour black-eyed peas with snails becomes “acid fry cowpea screw meat”. The joke cuts both ways: Westerners sometimes botch the Chinese characters in their tattoos.

Yet as Johnson’s experiment shows, for most workaday purposes machine translation (MT) has got a lot better in the past five years or so. The biggest source of improvement has been the deployment of “deep learning” in training such systems, which are meant to mimic the action of a brain’s neurons.

But improvement in automated translation for business purposes has also come about through a proliferation of clever companies solving individual tasks. Some have created specialist dictionaries so that important terms can be translated accurately and consistently. An entire MT engine can now be trained only on data from a professional domain (such as law or medicine), improving accuracy further. Other gizmos integrate MT with a human translator’s editing tools. This means many of those humans spend most of their time checking and refining MT output, rather than doing the bulk of the translating themselves.

第二步：精读原文（精读1遍，对照文本听1遍，精读1遍）

读原文（精读1遍，对照文本听1遍）

01

The translator of the future is a human-machinehybrid. There will still be a place for humans with specialist knowledge.

1.hybrid: 混合体, e.g. So it's a strange hybrid of a phone. 真是一款奇怪的混合体。

翻译：未来的翻译是一个人机混合体。具有专业知识的人类仍将有一席之地。

02

This first paragraph can read a bit oddly. As an experiment, Johnson wrote it in English, and then ran it through Google Translate to turn it into French, before running it through the same processin reverse, torenderit back to English.

1.in reverse: 相反地

2.render: 翻译

翻译：这第一段读起来有点奇怪。作为一个实验，约翰逊用英语写了这段话，然后通过谷歌翻译把它翻译成法语，然后再通过同样的过程反向翻译，把它翻译成英语。

03

But theproseis only slightly odd. The software has turned the original “may” into “can”, which made the first sentence sounda little off. In the second, the original “has written” became “a écrit” in French, which was then translated as “wrote” in English—accurately, but with a small change in emphasis.

1.prose:散文

2.a little off: 差一点，e.g. Your, your form was a little off. 你的姿势差了那么点。

翻译：但是，这篇散文只是稍微有点奇怪。软件将原来的 "可能"变成了"可以"，这使得第一句话听起来有点不对劲。在第二句中，原来的"has written "变成了法语中的 "a écrit"，然后被翻译成英语中的 "wrote"--准确无误，但在重点上有一点变化。

04

Once suchround-trip translations were an endless source of humour. Some people took serious texts and put them through this two-way process yieldinghilariousresults. And bad computerised translation is hardly consigned to the past. “Chinglish” signs are still a source of amusement for English-speakers visiting the Chinese-speaking world. On one menu, sour black-eyed peas with snails becomes “acid fry cowpea screw meat”. The joke cuts both ways: Westerners sometimesbotchthe Chinese characters in their tattoos.

1.round-trip translation: Round-trip translation is the process of translating a word, phrase or text into another language, then translating the result at least once more without reference to the original text, until it ends up back in the language it started in.

2.hilarious: 有趣的. E.g. a hilarious party: 狂欢会

3. 句型学习：“Chinglish” signs are still a source of amusement for English-speakers (visiting the Chinese-speaking world). 括号部分 ing形式做定语。

4.botch: 弄坏

翻译：这样的往返翻译曾经是幽默的无尽源泉。有些人把严肃的文本，通过这种双向的过程，产生了搞笑的结果。而糟糕的电脑翻译也很难被归于过去。"中式英语 "标志仍然是英语使用者访问华语世界的一个乐趣来源。在一份菜单上，酸豇豆加蜗牛变成了 "酸炒豇豆螺丝肉"。这个笑话是双向的。西方人有时会把他们纹身中的汉字弄错。

05

Yet as Johnson’s experiment shows, for most workaday purposes machine translation (MT) has got a lot better in the past five years or so. The biggest source of improvement has been thedeploymentof “deep learning” in training such systems, which are meant tomimicthe action of a brain’s neurons.

1.deployment: 部署，e.g. He has inquired out the deployment of the enemy troops. 他已查出敌军的兵力部署情况。

2.mimic: 模仿, e.g. He used to mimic speech peculiarities of another. 他过去总是模仿别人讲话的特点。

翻译：然而，正如约翰逊的实验所显示的那样，在过去五年左右的时间里，对于大多数日常工作而言，机器翻译（MT）已经有了很大的进步。改进的最大来源是在训练此类系统时部署了 "深度学习"，其目的是模仿大脑神经元的行动。

06

But improvement in automated translation for business purposes has also come about through aproliferationof clever companies solving individual tasks. Some have created specialist dictionaries so that important terms can be translated accurately and consistently. An entire MT engine can now be trained only on data from a professional domain (such as law or medicine), improving accuracy further. Othergizmos integrate MT with a human translator’s editing tools. This means many of those humans spend most of their time checking and refining MT output, rather than doing the bulk of the translating themselves.

1.proliferation: 增加, e.g. The proliferation of cars also encourages us to drive rather than walk汽车的激增也促使我们开车而不走路。

2.gizmo: 小玩意

e.g. 当幸福来敲门中有个桥段，He's been waiting outside the front of the building with some 40-pound gizmo for over a month. 没错，他拎着个 40 磅重的玩意儿，在公司门口等了一个多月了。

翻译：但是，用于商业目的的自动翻译的改进也是通过解决个别任务的智慧公司的激增来实现的。一些公司已经创建了专业词典，以便重要的术语能够被准确和一致地翻译。现在，整个MT引擎可以只根据专业领域（如法律或医学）的数据进行训练，进一步提高准确性。其他的小工具将MT与人类译者的编辑工具结合起来。这意味着许多人把大部分时间用于检查和完善MT输出，而不是自己做大部分的翻译工作。

第三步：阅读双语文章

上下滑动阅读双语全文

The translator of the future is a human-machine hybrid. There will still be a place for humans with specialist knowledge

未来的翻译是一个人机混合体。具有专业知识的人类仍将有一席之地。

This first paragraph can read a bit oddly. As an experiment, Johnson wrote it in English, and then ran it through Google Translate to turn it into French, before running it through the same process in reverse, to render it back to English.

这第一段读起来有点奇怪。作为一个实验，约翰逊用英语写了这段话，然后通过谷歌翻译把它翻译成法语，然后再通过同样的过程反向翻译，把它翻译成英语。

But the prose is only slightly odd. The software has turned the original “may” into “can”, which made the first sentence sound a little off. In the second, the original “has written” became “a écrit” in French, which was then translated as “wrote” in English—accurately, but with a small change in emphasis.

但是，这篇散文只是稍微有点奇怪。软件将原来的 "可能 "变成了 "可以"，这使得第一句话听起来有点不对劲。在第二句中，原来的 "has written "变成了法语中的 "a écrit"，然后被翻译成英语中的 "wrote"--准确无误，但在重点上有一点变化。

Once such round-trip translations were an endless source of humour. Some people took serious texts and put them through this two-way process yielding hilarious results. And bad computerised translation is hardly consigned to the past. “Chinglish” signs are still a source of amusement for English-speakers visiting the Chinese-speaking world. On one menu, sour black-eyed peas with snails becomes “acid fry cowpea screw meat”. The joke cuts both ways: Westerners sometimes botch the Chinese characters in their tattoos.

这样的往返翻译曾经是幽默的无尽源泉。有些人把严肃的文本，通过这种双向的过程，产生了搞笑的结果。而糟糕的电脑翻译也很难被归于过去。"中式英语 "标志仍然是英语使用者访问华语世界的一个乐趣来源。在一份菜单上，酸豇豆加蜗牛变成了 "酸炒豇豆螺丝肉"。这个笑话是双向的。西方人有时会把他们纹身中的汉字弄错。

Yet as Johnson’s experiment shows, for most workaday purposes machine translation (MT) has got a lot better in the past five years or so. The biggest source of improvement has been the deployment of “deep learning” in training such systems, which are meant to mimic the action of a brain’s neurons.

然而，正如约翰逊的实验所显示的那样，在过去五年左右的时间里，对于大多数日常工作而言，机器翻译（MT）已经有了很大的进步。改进的最大来源是在训练此类系统时部署了 "深度学习"，其目的是模仿大脑神经元的行动。

But improvement in automated translation for business purposes has also come about through a proliferation of clever companies solving individual tasks. Some have created specialist dictionaries so that important terms can be translated accurately and consistently. An entire MT engine can now be trained only on data from a professional domain (such as law or medicine), improving accuracy further. Other gizmos integrate MT with a human translator’s editing tools. This means many of those humans spend most of their time checking and refining MT output, rather than doing the bulk of the translating themselves.

但是，用于商业目的的自动翻译的改进也是通过解决个别任务的聪明公司的激增来实现的。一些公司已经创建了专业词典，以便重要的术语能够被准确和一致地翻译。现在，整个MT引擎可以只根据专业领域（如法律或医学）的数据进行训练，进一步提高准确性。其他的小工具将MT与人类译者的编辑工具结合起来。这意味着许多人把大部分时间用于检查和完善MT输出，而不是自己做大部分的翻译工作。

第三步：回答问题（评论区回答问题）

Reading Questions：

1. Who will find Chinglish signs amusing?

2. What is the biggest source of improvement in machine translation?

滑动查看答案

Q1

English-speakers visiting the Chinese-speaking world.

Q2

It has been the deployment of “deep learning” in training such systems.

今天是小芳老师陪你精读外刊的第29天

检查一下，你学到了吗？

1.hybrid: 混合体 2.render: 翻译

3.a little off: 差一点 4.hilarious: 有趣的

【特别声明】以上内容由小芳老师原创整理，转载请注明来源，本平台采用的音视频资源均出于传递信息和英语学习之目的，并没有任何商业目的。本公号尊重知识产权，如无意中侵犯了您的权益，请及时联系后台，本公号将及时删除。