Home > News content

Tencent releases artificial intelligence assisted translation for tribute to human translation

via:博客园     time:2018/11/14 15:15:06     readed:195

November 13, Shenzhen-Tencent AI Lab today launched a AI assistant translation product

In recent years, with the rapid development of machine translation technology based on neural network, the quality of machine translation translation has been continuously improved, and the argument that machine translation replaces manual translation has become very popular. Tencent's assistant translation team believes that in practice, machine translation at this stage is often full of errors and omissions. Without human intervention, the accuracy of machine translation can not meet the translation requirements of specific application areas, such as commercial contracts, legal provisions, professional books, academic literature, tourism. Text and so on. A qualified professional translator needs not only long-term professional and hard training, but also a lot of practical experience, in order to be able to compete in specific professional fields of translation work. In the actual application scenario, the accuracy of machine translation is still not comparable to that of professional translation.

However, due to the enhancement of computing power, the innovation of machine translation model and the dramatic improvement of corpus size, the quality of machine translation has improved significantly, and the translation speed is far faster than that of manual translation. At the present stage, machine translation can provide translation reference, help manual translation to complete some relatively mechanical tasks, free manual translation from complicated and inefficient typing, word search and other manual work, and devote energy to higher-level translation creation. In the process of artificial intelligence technology such as machine translation assisting human translation, human translation will obtain personalized machine translation results. The efficiency and quality of translation can be further improved through interaction between machine and machine. Machine translation can not replace manual translation, but it will reconstruct the process of manual translation. Tencent-assisted translation is a technological innovation product based on this concept, aiming at improving the efficiency and quality of manual translation and meeting the growing needs of translation.

Tencent Auxiliary Translation products use team self-developed human-computer interactive machine translation technology, is an artificial intelligence assisted translation of the Internet landing product. The human-computer interactive machine translation technology is the core technology of artificial intelligence assistant translation system. Compared with general machine translation, the important feature of human-machine interactive machine translation is that it allows users to intervene in translation generation in real time, provide interactive machine translation, and translate input method. In order to improve the efficiency of manual translation, real-time translation suggestions and other efficient interactive means. Tencent implements two assistant translations

1, the kernel is the first open interactive Machine Translation Internet landing product in China.

2, equipped with the industry's first open fusion of Machine Translation's auxiliary translation input method. This product integrates artificial intelligence and traditional assistant translation technology, and provides users with more convenient and smooth experience of artificial translation.

The highlights of Tencent's auxiliary translation are as follows:

Interactive Machine Translation

The first publicly available interactive machine translation Internet landing product, which updates the automatic translation in real time according to the input content, significantly improves the translation efficiency.

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413577-365014309.gif

Interactive machine translation (IMT) is a constrained translation and decoding technology based on Autonomous implementation. While improving the accuracy of translation, the decoding speed is also optimized to meet the requirements of real-time interaction in the Internet environment. Real time decoding speed is the key factor affecting the landing of interactive Machine Translation products. Considering the existing accumulation, Tencent AI Lab has designed and implemented a neuro-machine translation system dedicated to human-computer interaction. Unlike ordinary machine translation, the challenge of interactive machine translation lies in the inability to predict user actions, which makes it difficult to speed up the response by establishing caches in the original text. Tencent AI Lab continues to overcome technical difficulties, and has been committed to creating interactive machine translation products with better user experience.

Auxiliary translation input method

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413979-556406654.gif

The first MTI product incorporating machine translation has fewer keystrokes than the conventional input method and significantly reduces typing time. More speedy pinyin to Chinese characters, more accurate word completion, allowing users to enjoy the translation process.

Because it is difficult to perceive the context information in the translation scene, the general input method still has much room for improvement in accuracy and individualization. Compared with the general input method, the assisted translation input method can make use of the knowledge of statistical machine translation and neural network machine translation, greatly improve the accuracy of automatic word formation and recommendation phrases, and make input prediction more personalized.

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144414053-225288585.gif

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413564-795587032.gif

Independently implemented input method knowledge mining algorithm continuously extracts fragmental translation knowledge from massive Internet data and billions of double sentence pairs as the basis of assisted translation input method. In the process of translation, the machine translation context information of the current sentence is fused in real time to assist the user to complete the translation task quickly.

For example, translating English sentences

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413323-430632230.jpg

When you use the QQ Pinyin input method, you need to ente

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413305-70059782.jpg

However, with the support of data mining and machine translation knowledge, only input is required after the adoption of the auxiliary translation input method.

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413327-1724886361.jpg

It can be seen that the auxiliary translation input method can improve the efficiency of manual translation by reducing the number of typing keys.

Real time translation proposal

Based on the self-developed Neural Network Machine Translation Engine, it can provide flexible and accurate real-time translation fragments intelligent recommendation, and continuously learn according to users'translation history and habits, providing more personalized translation suggestions that meet users' expectations.

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413355-394296274.jpg

Due to the limitation of machine translation technology at present, the accuracy of machine translation is not high, so the real-time translation suggestions can not be too long, otherwise the efficiency of manual translation will be reduced. The difficulty of real-time translation suggestions is how to provide phrases or clauses with moderate length and high accuracy. To this end, Tencent AI Lab proposes and implements an intelligent translation recommendation algorithm, which integrates statistical machine translation, neural machine translation and other technologies to provide users with the most appropriate translation suggestions.

For example, while translating sentences,

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413355-394296274.jpg

Tens of millions of professional terms, 100 million examples.

Compared with the traditional assistant translation software, the source of intelligent translation suggestions is no longer limited to the terminology library imported by users, but comprehensive integration of massive Internet data. Tencent AI Lab has excavated tens of millions of professional terminology translations and billions of bilingual examples from hundreds of TB Internet texts to provide users with translation reference information. Then, the reference information such as terminology library and example sentence library is displayed with sentence as dimension correlation to meet users'multi-domain and multi-style translation needs.

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413826-2121200042.gif

Auxiliary translation API and customized service

To provide customized services for partners by opening multiple application interfaces such as interactive machine translation, assistant translation input method, personalized machine translation API, customized machine translation, etc.

orgsrc=//img2018.cnblogs.com/news/66372/201811/66372-20181114144413534-1278870275.gif

Tencent assisted translation was completed by the internal team of Tencent AI Lab for one year. Tencent AI Lab is Tencent's enterprise-level AI laboratory, which was established in April 2016. At present, there are 70 top research scientists and 300 application engineers in China and the United States. Natural Language Processing (Natural Language Processing, NLP) is one of the four major research areas. The NLP team is dedicated to tracking and studying cutting-edge natural language text understanding and generation technologies, and incubating the next generation of natural language processing technologies and business application scenarios. Team scientists have made many innovative achievements in frontier theoretical research. In application exploration, the technologies of text understanding, text generation, dialogue and machine translation have been applied to many products of the company to enhance product intelligence to better serve users.

Experience website:Http://transmart.qq.comEmail enquiry: transmart@tencent.com

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments