Machine translation for pros

At the end of May in Moscow, we (ABBYY Language Services) gathered representatives of the translation industry and localization at the TAUS round table, to talk together about the automation of translation: what is it, what is the use, what to do with it and who needs it. The conversation turned out to be productive, which we are very pleased with. Now we will tell you about one of the reports, which was the best at the end of the round table and allowed its author to receive a special award TAUS Excellence Award.



The report, which we dwell on, was devoted to the topic of machine translation (MT). In general, many participants spoke about machine translation. For example, that its popularity is not declining, and many ordinary users and companies have begun to more actively use it in their work - only Yandex.Translate passes about 100 GB of information daily.

Our director of innovation, Anton Voronov, decided to talk about what is needed for productive professional use of machine translation.
')
We already wrote that in the West, we had time to evaluate the benefits of automation, and many organizations and language service providers use different technologies in real work on orders: dictionaries, glossaries, translation memory bases, crowdsourcing and machine translation. Everything is simple: the representatives of the industry understood that, despite the fact that content around the world doubles almost every year, the pace of translation remains the same. Obviously, you need to increase productivity.

It has been proven in practice that machine translation should be used if at least two or three points are fulfilled from the following requirements:

• You have big projects with short deadlines.
• It is necessary to translate the so-called "less noticeable" content: technical documentation, user-generated content, knowledge bases. They are intended for a large number of users, but, as a rule, they are very large in size, and each section of such a text array (knowledge base, for example) is carefully read by only a small number of users.
• As the source you have on hand texts with a clear structure and a high level of repetition;
• Your team has debugged combined production processes with flexible quality requirements - different projects may require different levels of quality: some texts should be translated as accurately as possible, for others it is enough to grasp the general meaning. And the team should be ready to adapt the translation process to such differences. It is assumed that, at the same time, specialists are familiar with how post-editing, crowdsourcing and other operations work, which are different from the traditional TEP-model .

At the same time, it is necessary to take into account the features of the system: in order to achieve high quality of translation from MT, a fair amount of translation memory bases is required, the choice of a suitable “engine” of machine translation, its adjustment to the type of project and deep integration of the MT system into your production process. Otherwise, the miracle does not happen.

How does it look in practice? Imagine that you need to translate a lot of technical instructions to a specific software. First, it is worth stocking up with Translation Memory databases that were compiled during previous translations for this software or left after similar projects - the more the better. Then it makes sense to decide on a suitable machine translation system - perhaps in past projects, some of them showed themselves in the best possible way - and conjure it with its setting: feed the existing bases and parallel texts. In the process of translation, be prepared to monitor the operation of the machine: so that you can quickly make adjustments if something goes wrong.

In our practice, the following scheme of the production process has proven its effectiveness:

• Start with terminology — extract terms from the source text and translate them right away, this will make your life easier;
• Do not forget about linguistic resources - parallel texts, translation memory bases: they are important both for setting up the engine and for the translation itself;
• Choose for each project the appropriate "engines" - for this, of course, you need to constantly monitor the performance of all used MT systems;
• Train your machine translation systems with results that meet the required level of quality;
• By the way, about the quality - constantly monitor the quality of the translation: how much the results meet the expectations;
• Fix what helps to make the translation of a particular segment better: this will help metrics and check for consistency of terminology;
• Measure all the parameters to assess how much you need to refine the text at the stage of editing, and adjust the process accordingly;
• Get a platform that will automatically handle all of this.

Experience has shown that in order to maximize the automation of the translation process in any company, it is necessary to take care of the online CAT tool. It needs to integrate the terminology management module and the MT system. It also makes sense to provide for a flexible production model (in case you have to change something on the go), real-time teamwork for the team, automatic registration of all actions of the editor (this allows you to find "bottlenecks") and built-in quality control.

In our case, this full cycle of automation is performed on the basis of SmartCAT , which we wrote about earlier and which we continue to actively develop.

A little touched and how you can train the "engines" of machine translation. In order for expectations from MT results to be met, it is important to reuse linguistic resources when setting up the system. Extract terminology, monitor its uniformity and give the resulting glossaries to the “engines”. Take the segments that have already been translated and have passed the stage of post-editing, and share them with your MT systems: the last options are important here, since they are more relevant.

Quality control throughout the process of working with machine translation will avoid unpleasant surprises. The history of text changes, elapsed time and the results of the automatic quality check will help you choose segments that require close attention during the final quality assessment. In general, anything can happen, so be prepared for changes in the quality control process when translating MT.

Few talked about the plans. The fact is that we managed to get so deeply immersed in the process, since we ourselves have long been actively testing various automation systems and working methods in search of high performance and flexible quality level management. It became clear to us that in order to work more effectively with MT, there is a lack of an integrated module for extracting terminology, hints when searching in already loaded databases, data on the context of certain terms. And, of course, more quality checks and more metrics. We continue to incorporate this into our products and our own processes.

Of course, linguistic technologies continue to evolve. But content volumes are growing even faster, and existing solutions still require the participation of professional translators in the process. In general, the immediate future of the industry for the joint work of people and machines.