微软智能翻译技术新突破：机器翻译提前7年达到人类专业翻译水平

　　A team of researchers from the Microsoft Research Asia and Research Redmond announced in April that, when reporting and translating general news between Chinese and English, its R & D machine translation system has reached the level of human professional translators. This is the first ever system that could be on a par with human professionals in terms of quality and accuracy.

　　由微软亚洲研究院与雷德蒙研究院的研究人员组成的团队四月宣布称，其研发的机器翻译系统在通用新闻报道的中译英测试中，达到了人类专业译者水平。这是首个在新闻报道的翻译质量和准确率上媲美人类专业译者的翻译系统。

　　"This is a landmark achievement in the field of natural language processing," said Dr. Huang, who is responsible for Microsoft’s voice, natural language and machine translation.

　　在微软负责语音、自然语言和机器翻译工作的黄学东博士表示，这是自然语言处理领域的一项里程碑式的成就。

　　“Breaking through the limitations of current neuromachine translation paradigm, performance is more than one order of magnitude，” he said, adding that they will fulfill that ambition as long as keeping abreast of the most advanced smart technologies.

　　“突破当前神经机器翻译范式局限，让性能再上一个数量级 ，”他补充说，我们会紧跟最先进智能技术，实现这一伟大目标。

　　Microsoft’s breakthrough, which takes the machine translation beyond the time of human amateur translators, has advanced seven years ahead of the expectations of numerous ML researchers.

　　微软的这次突破，将机器翻译超越人类业余译者的时间，提前了整整7年，远远超出了众多ML研究人员的预想。

　　Although the breakthrough is remarkable, Microsoft researchers also remind you that this doesn’t mean that humans have completely solved the problem of machine translation, but we can only explain that we’re closer to the machine.

　　虽然此次突破意义非凡，但微软研究人员也提醒大家，这并不代表人类已经完全解决了机器翻译的问题，只能说明我们离终极目标又更近了一步。

　　In order to be able to obtain a landmark breakthrough in Chinese-English translation, three research groups from the Microsoft Asian Research Institute and the Redmond Research Institute conducted joint innovation across the region and across the area of research.

　　为了能够取得中-英翻译的里程碑式突破，来自微软亚洲研究院和雷德蒙研究院的三个研究组，进行了跨越中美时区、跨越研究领域的联合创新。

　　Finally, they find that our system is completely complementary, so that we can benefit from the system combination and ultimately achieve the goal of machine translation to the human level.

　　最后，他们发现我们的系统是完全互补的，因此可以从系统组合中获益很多，最终实现机器翻译达到人类水平的目标。

　　Based on AI，cloud-computing, smart corpus database and the ongoing development in neuromachine technology, neuromachine translation would be the mainstream media in insofar machine automation translation.

　　有了人工智能、云计算、智能语料数据库和正在研发当中的神经机器技术的加持，神经机器翻译将成今后机器翻译绝对主流。

　　Machine’s Deep Learning Ability (MDLA) is another interesting section that researchers are paying much attention to for a reason that deep learning is the master key to realize strong artificial intelligence

　　机器深度学习能力(MDLA)掌握着实现强人工智能的钥匙，这是研究人员饶有兴趣关注的另一项技术。

　　For other artificial intelligence tasks such as speech recognition, it is judged whether the performance of the system can be comparable to that of humans because the ideal result is exactly the same for humans and machines, and the researchers call this task as a pattern recognition task.

　　对于语音识别等其它人工智能任务来说，判断系统的表现是否可与人类媲美相当简单，因为理想结果对人和机器来说完全相同，研究人员也将这种任务称为模式识别任务。

　　However, machine translation is another type of artificial intelligence task, even if two professional translators have slightly different translations for identical sentences, and neither translation is wrong.

　　然而，机器翻译却是另一种类型的人工智能任务，即使是两位专业的翻译人员对于完全相同的句子也会有略微不同的翻译，而且两个人的翻译都不是错的。

　　Complexity makes machine translation a very challenging problem, but it is also a matter of great significance.

　　复杂性让机器翻译成为一个极有挑战性的问题，但也是一个极有意义的问题。

　　Huang Xudong，a deep researcher in this field, believes that neuromachine translation, or depth learning, is the most exciting place in that it is able to learn from the inside of a natural language, to learn from the textual structure, semantic structure and semantics of the language, and to feed back to the system.

　　深耕该领域的黄学东研究员认为，神经机器翻译，或者说深度学习，最激动人心的地方在于，它能够学会自然语言内部的特征，然后学会文本的结构、语义结构和语义内容，再反馈到系统，从而实现自然语言理解的突破。