How to efficiently edit the translation results of the machine?

In recent years, thanks to the development of software technology, the emergence of new algorithms, and the improvement of computer performance, machine translation technology has made great progress. It has been recognized by the translation industry and customers, and has become an important part of high -efficiency translation process. However, the quality of translation generated by machine translation still cannot reach the degree of manual translation of professional translators. In order to achieve the balance between the quality of the translation and the efficiency of translation and the advantages of human -computer interaction, post -translation editors become the translation embodiments actively adopted by the translation service industry.

1. What is post -translation editing

To accurately understand what is post -translation editor, you need to accurately understand the definition of machine translation. According to the definition of ISO/DIS 17100: 2013 standards, machine translation is "using a computer system to automatically translate text or voice from one natural language into another." The post -translation editor is "check and correct the output of machine translation." Post -translation editors can be divided into narrow post -translation editors and general post -translation editing. The former refers to post -translation editing directly through the translation obtained by the translation of the machine, and the latter refers to post -translation editing of the translation of translation and translation management system composed of translation memory, machine translation and translation management system.

2. Post -translation editing development trend and motivation

The 2012 Global Free Translator Report in the global well -known translation community ProZ.com (2013) showed that the number of translated editors as a service provided as a service increased from 30% in 2011 to 40%. According to data from the TOP100 list recently released by NIMDZI, the number of companies providing machine translation and post -editing (MTPE) services in 2021 increased by 7.5%.

The development prospects of post -translation editors are broad, and some important factors have promoted the rapid development of post -translation editors. The first is the vigorous development of product globalization and localization, which has generated a large number of translation needs, and this demand is far from meeting artificial translation. The decrease in the quality expectations of machine translation, the emergence of the contemplate translation, etc. In addition, the increasing and shorter product release cycle and the reduction of the company's translation budget are also an important driving force for MTPE development.

3. Introduction to the editing function of CAT software after translation

MEMOQ, Matecat, Trados Studio, MEMSource and other desktop translation software or cloud translation software provides integrated translation environments, that is, at the same time, it has machine translation, translation memory and post -translation editing functions. These translation software is similar, and its post -translation editing functions have their own characteristics.

1. Memoq post -editing function

MEMOQ software integrates 19 machine translation plugins such as Google, Microsoft, Deepl, and Kantanmt. Through Intento or TMXMALL machine translation plug -ins, users can use foreign machine translation engines such as Google, Microsoft, Deepl, and domestic machine translation engines such as Baidu, Sogou, Youdao, Ali, Tencent, and no longer need to register account one by one.

MEMOQ software supports pre -translation with machine translation, and then uses mild post -editing or complete post -editing method to complete the editing work of the translation according to the quality requirements of the translation. The post -translation editor's work interface is the editor interface that translators and reviewers are very familiar with. In this interface, post -translation editors can add, delete, replace the initial translation of machine translation, or refer to the term to modify the term referring to the terms to modify the term Essence The quality assurance function can be played in real time, and the preview function is convenient for editors to view context context and translation format. All editors can be displayed intuitively in tracking and revising.

2. Matecat post -editing function

Matecat is a free open source cloud translation software. It currently integrates 13 machine translation engines such as Google, Microsoft, Modernmt, Mymemory, and a public translation memory containing 12 billion pairs for users to use for free.

Each sentence segment will automatically add a machine translation or translation memory library to match the translation in each sentence segment when the editing interface of Matecat is opened. No pre -translation operation is required. In addition, the source of translation matching is listed below the translation sentence segment, that is, machine translation matching, private translation memory mixing, or public translation memory library matching. This method of translation provides great convenience for post -translation editors. Post -translation editors can compare and choose between machine translation translation and translation memory library matching translation. The content of the term matching is also listed below each sentence. The editing modification, text format and tag (TAG) processing of text content are very easy.

3. Trados Studio post -translation editing function

Trados STUDIO software only integrates SDL Language Cloud and SDL machine translation engines, but users can add and use almost all machine translation engines through RWS AppStore (app store).

In the RWS Language Cloud page, you can select "Translation Engine" or "SDL Machine Translation". The former integrates machine translation engines and users' personal cloud translation memory and cloud language library resources, while the latter is a pure machine translation engine. The specific editing operation process is similar to the Matecat introduced in the previous article. The difference is that in the Trados STUDIO machine translation, in addition to choosing a universal nerve machine translation model, you can also choose an adaptive machine translation model. At present, the biggest pain point of machine translation is that it has not been learned during the translation process, so the same translation error will repeatedly appear. Adaptive machine translation is a solution for this problem. It can record the modification of each editor made by the post -editor personnel in real time and apply it to the subsequent translation. The figure below is the translation result of the adaptive machine translation engine, including the initial translation and modification of editing by sentence by sentence. Translation engine.

4. MEMSOURCE post -translation editing function

MEMSOURCE integrates 35 machine translation engines, with "AI -driven non -translation elements" and "MTQE) function.

AI -driven non -translation element function can not only identify traditional non -translation content, but also identify new content such as people and product names. Its biggest advantage is that it can provide a matching value like a translation memory library, indicating that a sentence segment " It is determined that the "non-translation element segment (100%matching value) or" may be "non-translation elemental segment (95%-99%). Using this function can automatically identify the sentence segment that only contains non -translation elements in the original text before translation, and pre -translation of these sentence segments, thereby improving post -translation editing efficiency.

The MTQE function can provide a sentence -level quality assessment for the output of machine translation, which is similar to the matching of translation memory. Although the output quality of machine translation is greatly improved in general, the translation quality is not stable because of the different content type and language. Therefore, post -translation editors often face many low -quality machine translation translations, and editing such translations are very wasteful. Similar to the matching value of the translation memory, the automatic score of the quality of the machine translation can help estimate the workload of post -translation edit, and reduce the work intensity of post -translation editing. In addition, the MTQE function can also be preferred for machine translation engines for specific content and language pairs.

Fourth, summary

The combination of machine translation and post -translation editing has developed rapidly. It not only gives full play to the speed of machine translation, but also gives full play to the accuracy of manual translation. The post -translation editing model not only meets the needs of the rapid development of the translation market, but also promotes the development of translation technology. It also promotes the exchanges and cooperation between academic and industry, and enriches the composition of the language service industry chain. For translators, mastering the post -editing function of CAT software will provide great help for translation work.

Transfer: Translation Technology Education and Research

Author: Wang Lin

- END -