Market of machine translation technologies: interview with Nikita Shablykov, PROMT

Nikita Shablykov: According to Global Market Insights, Inc the market of machine translation by 2024 should make more than 1.5 billion US dollars. Annual increment is estimated at 14.6%. At the same time, the size of the market of all linguistic services in 2019 will be $53.5 billion. Here interpreting and translation, software localization, websites and multimedia, including games, movies and series and many smaller services enter. Machine translation – one of the main tools on process optimization of transfer, localization and work with text Big Data.

The volume of the world market of machine translation in 2016 - 2024 (in one million US dollars)

Nikita Shablykov: The Segment of private users on sales is small. 20 years ago people were around the world ready to buy programs for the PC for the solution of private translation tasks. Even in Russia, despite piracy, it was the significant market.

With distribution of the Internet, mobile devices, we began to use online services. PROMT started the first in Runet free online service of transfer - Translate.ru^[1] - in 1998, then already in this market other players and quite fierce competition appeared. Therefore the question with transfer for the private purposes is closed. Only on our services are millions of translation requests a day from around the world. However the programs installed on the PC are still important for a part of users. For example, the translation of documents with preservation of formatting or the transfer in offline mode guaranteeing confidentiality of data is important.

Today the main market share of machine translation is occupied by a corporate segment. In 2012 researchers of Boston Language Institute conducted survey during which they found out that 64% of entrepreneurs consider language barriers an obstacle for access to the foreign markets. Most of respondents believes that, having got rid of language restrictions, it is possible to increase considerably number of consumers and business revenues.

Nikita Shablykov: Better to illustrate corporate users, I will tell about four standard business challenges which were created in 30 years of a company performance of PROMT in Russia and the world.

The first such task - the international cooperation.

Most of employees of the large companies well knows foreign languages, and, nevertheless, addresses machine translation. First, it is connected with what to work with texts and documents on a mother tongue much quicker and more conveniently, especially, if it is about large volumes. Secondly, if you need to prepare the report or the presentation, you will cope with a task much quicker, having the word-per-word translation with machine translation. Thirdly, correspondence with the foreign partner becomes simpler when you can check yourself through machine translation. Also you should not forget one more important circumstance – confidentiality of data when translating.

MT in daily work use many large companies in Russia, for example, Norilsk Nickel. Sometimes in the companies even access to well-known online translation services is closed. It is made in order to avoid a "accidental" vykladyvaniye on cloud services of fragments of correspondence or the internal documents representing a trade secret.

The second business challenge – interaction in online space.

There are many online services where business is constructed on providing information on goods and services in the automatic mode. Users from around the world share this information: they leave comments, write comments, discuss goods and services. That content was clear to all, it needs to be translated into different languages. It is impossible to involve in transfer of professional translators – it will demand huge resources and time. Therefore MT in this case – the only exit. It is reasonable at the cost and terms. Quite so the TripAdvisor company resolves an issue with transfer of traveler reviews. The TripAdvisor resources work in 30 countries at 21 languages, and for transfer of the most part of the user content machine translation is used.

The third business challenge which is solved by machine translation – integration of a translation engine into structure of other software product.

End users do not want to have 10 solutions for different tasks, they want to have one working interface in which all tools are near at hand for economy and optimization. For this purpose there are translation engines of PROMT for different platforms: server OS, mobile and desktop.

Our main advantage is that modules of PROMT work offline at high-quality preserving of transfer, and, so ensure information security and confidentiality of data.

Among large projects of integration of language translation technologies into a working environment of the company there are projects with Lukoil and Federal Institute of Industrial Property.

And, at last, the fourth standard task about which I would like to tell is a localization.

The developer companies of the software delivering the products abroad are faced the need of localization of specific documents – manuals, user agreements, marketing descriptions. Similar documents can be translated automatically, and then to edit that will allow to reduce costs for transfer many times. A striking example - Kaspersky Lab where machine translation of PROMT is used for localization of documentation to products.

Nikita Shablykov: Along with PROMT, in a corporate segment it is possible to carry Systran and SDL companies to world technology leaders. Systran is engaged only in machine translation technologies and long ago works at the markets of the USA, France and other countries. For SDL of a machine translation technology – not core business, and, first of all, the company is known for the solutions in the field of "translation memory" and management of translation processes. So products and SDL technologies are expected first of all professional translators.

In the Russian market of PROMT – the only solution provider on machine translation in a corporate segment who proved the high competence and product quality.

Except obvious requirements – the translation quality, the translation of documents in different formats with preservation of formatting – for our clients is important to have the complete solution when in one interface all tools are available: the corporate translator, the corporate dictionary and other tools intended both for department of transfer of the company and for normal users (managers, lawyers, analysts).

Approaches of the companies differ depending on their opportunities and orientation to specific consumers. When developing solutions of PROMT we use also artificial intelligence technologies, neural networks, we combine different approaches and we create new. It allows us to issue the best result for a wide range of users.

Nikita Shablykov: PROMT is completely Russian IT company with the St. Petersburg roots, in the market we 28 years. The state small, up to 100 people, a half of them are those who think out, develop, and tests products. For the last five years the number of developers in the company essentially did not change. In a command there are very experienced specialists, some of them work in the company more than 15 years, there are also those who else study or recently graduated from university.

Nikita Shablykov: Main market, of course, Russian. Also we have the German company PROMT GmbH and the American PROMT Americas. PROMT GmbH – the noticeable player both in private, and in a corporate segment in the market of the German-speaking countries. Last year, for example, the company signed the agreement with the third largest supplier of automotive industry in Germany – ZF Friedrichshafen AG. There is a partner in France who successfully extends products of PROMT under the brand. Among clients – L’Oréal, Air France. There are partners in Japan, China, India, and even in Iran and Saudi Arabia.

Nikita Shablykov: IT market is international, and any company at start of a product gets to very competitive environment. PROMT managed to enter international market in time. It happened at the beginning of the 2000th when we opened offices in Germany and the USA, became partners with the French company. The most interesting and at the same time complex market now for us is China. This market is rather closed, there are serious restrictions on promotion and distribution of IT products. For example, on local social networks only the company registered in the territory of the People's Republic of China can promote. It should be taken into account and national factors – we had a bad experience of partnership just owing to this reason. However we continue to work in this direction.

Nikita Shablykov: In Russia it is the banking sector, the oil and gas industry, power, military industrial complex, a public sector, the transport industry and also mechanical engineering and IT.

Today the product which is most demanded among our customers is PROMT Translation Server, server solution for the translation of documents and texts within corporate network of the client.

Solutions of PROMT are established in corporate network and do not require connection to the Internet, providing complete confidentiality of the translated data. Therefore our products – a safe alternative to online translators – to potential channels of information leak where even in user agreements it is specified that any content translated through cloud services becomes the property of these services. It is very important aspect, especially, for government institutions and large corporations.

One more moment on which I would like to focus attention, - a possibility of saving of time and the budget at the expense of qualitative solutions on automatic translation. As statistics, in the presence of such solutions shows, the user can independently process 60-70% of documents which are usually transferred to department of transfer.

Nikita Shablykov: Last year when in Russia there took place the FIFA World Cup, our company developed the special mobile translation app for conductors of long distance trains. With its help the staff of the Russian Railway could communicate with foreign passengers – guests of the FIFA World Cup – in English, German, French and other popular European languages. The translator does not demand connection to the Internet, it is important at unstable communication in way, there is a Dialog mode at which the user can communicate with the foreigner on a mother tongue. The application was built-in in UKEBY – special devices using which conductors verify authenticity of e-tickets of passengers.

One more interesting project of our company – implementation of transfer of PROMT in the Amadeus booking system. The Amadeus booking system – the leading supplier of technological solutions in the field of tourism – wins first place among the global distributive systems by the number of users. Travel-agenty which use this system should give clients the comprehensive information under the terms of armoring of an avia - and train tickets. Similar data are fully specified in special documentation – "The conditions of application of a rate" which automatically form in English in real time. We developed the solution allowing to translate this documentation into Russian on the fly. This solution is integrated in a booking system and configured on the translation of highly specialized terms that facilitated work to agents.

Of course, I cannot but mention also the joint project with TAdviser - start of the first in Russia, and, perhaps, and in the world, media, the artificial intelligence working at a basis. The new tadviser.com portal is an English-speaking analog of the edition where translation functions are integrated into structures of the website, interact with different databases, and as a result all published materials of the Russian-speaking website quickly appear in English without additional efforts.

PROMT has many interesting projects – for example, implementing solution on automatic translation in the system of the Federal Institute of Industrial Property (FIIP), cooperation with Kaspersky Lab, the large industrial companies – such as Lukoil, Siemens and Norilsk Nickel.

Nikita Shablykov: Already now online services are more popular in a private segment. If the user is not disturbed by confidentiality of the translated information - it is normal option. Unambiguously the share offline - solutions will only grow in a corporate segment. Uses of solutions in the closed circuit will remain important for state structures, military industrial complex, any commercial organization which cares for the information security.

Nikita Shablykov: In PROMT different machine translation technologies, including, since 2018 – neural technologies are used. Neural transfer is the "newest" approach, it gives higher quality of transfer, than other technologies (analytical and a SMT). Though this technology has minuses and features too.

The neural network can make difficult diagnosed mistakes during the work with material which did not get to the training selection. In other words, without knowing the answer, the network "will think up" it that, for example, is especially relevant when processing proper names (a personnel, toponyms and so on). At general "smoothness" of transfer, the user can not notice an error even. Besides, neural transfer is exacting to resources at installation in a local network.

Already now products of PROMT with neural transfer are available to the Linux server platform inside, and in September, 2019 we are going to release PROMT Translation Server with neural transfer under Windows Server. It will be break since in the world there are no similar solutions.

Neural or analytical, machine translation is imperceptible and already irreplaceable technology in the modern world.

One more aspect is a development of case technologies. NMT, as well as statistical MT, is based on the body therefore there are a lot of efforts, including in the companies, should be directed to collecting and storage of multilingual data.

Nikita Shablykov: Our strategy is a development of ready-made specialized solutions and tools for fast adaptation of the basic solution under the customer's task.

Generally neural transfer yields result of higher quality, than other technologies. But after all each business challenge requires accounting of its features, both at the linguistic level, and in the technical plan. High quality of transfer in a corporate segment is possible only at adaptation of the solution under a specific objective and also when using a combination of different technologies (neural, analytical, statistical).

For example, the translation of tariff rules in the Amadeus booking system for a number of reasons will be hardly possible using neural transfer. The basic reason is that there are no parallel corpora for such content, it is a lot of features in the structure of the text: all text is written in an uppercase, the offer is segmented. So it is such specific content which the ordinary person would hardly understand.

Nikita Shablykov: Already now we will be able to show you transfers where you find it difficult to tell, they are executed by machine or the person. A problem of AI not in comparing result of the program and the person and to mislead. The task - whether can replace the machine the person so that replacement was equivalent. At MT several significant advantages.

First, reduction of routine transactions. The translation of technical documentation of product X of version Y is not a creative task which the mankind would like to save for itself, without giving to machines.

Secondly, speed. The machine translates extremely quickly.

Thirdly, availability. You do not need to learn language to understand a text in Finnish or the description of service in Portuguese, and it is not even necessary to look for the translator who will translate it for you. Using technologies all of you will make.

Fourthly, and it is very important point in the modern world, it is privacy. If you can address for transfer technology, then you have no intermediary, and nobody learns what information interests you - a personal letter, the description of medicine, financial information.

At the same time it is necessary to remember that language translation technologies are only technologies, the tool for the person. Kind of well the technology worked, its result should be estimated by the person. Nobody will come to court with the documents translated by machine without check of the specialist. Once again I will repeat, a question not in in general to replace the person, and in leaving it only a creative, analytical part of work, and to transfer a routine to the machine.

Nikita Shablykov: The product PROMT Analyzer SDK works at a basis of the deep semantico-syntactical analysis of natural languages. The solution became one of finalists of the tender "The Best Information and Analytical Tools" in the "Best Information and Analytical Solution for Processing of Unstructured Data" nomination.

Analyzer SDK represents API for integration into EDMS, BI, corporate portals. PROMT Analyzer SDK analyzes the document, selects key objects, for example, in agreements, and gives an idea of its semantic structure. The analysis result remains in the database and can be used for intellectual search in it.

The main features of the solution are a high accuracy and completeness of analysis, multilingual capability (on an input it is possible to file the text or the document in any language) and also work speed – many times quicker than the analogs presented at the market. It is the module for embedding in other BI systems, we invite partners to testing and cooperation.

Nikita Shablykov: This question very much disturbs our many clients. But there are also those who do not realize danger of use of online services yet. In 2017 there was a scandal over the known company Statoil. Many confidential documents of this company were translated on an online service, transfer was indexed and documents became available at issue in search systems.

In the user agreement of the largest company, the providing service in online translation, it is told: "Loading, adding, saving, sending and receiving keeping in our Services, you provide to the company and its partners the license existing around the world which allows us to use this content, to place it, to store, reproduce, change, create on its basis derivative works (for example, transfers, adaptations and other methods of optimization of materials), to exchange it, to publish it, openly to reproduce, display and also to extend".

For the benefit of the company — to provide the employees with corporate solution on transfer which conforms to all requirements of information security. The solution should not adjoin to "external environment" — then the translated information will not leave out of limits of one PC or internal server of the company, a private cloud. The service of automatic translation working in the local area network of the company — the same regulation for commercial and government institution, as corporate mail.

Nikita Shablykov: It is natural process. I think that the European regulations of GDPR about personal data protection of citizens of the European Union will affect also regulation of personal data in our country a bit later.

Nikita Shablykov: There are certain difficulties during the work in international markets which constrain development. For example, in Europe there is a simplified order of purchase of software and other goods and services. If we negotiate terms of the contract with partners, we expose them an invoice – one piece of paper, - and the fact of payment is confirmed by consent with terms of the contract. There are no acts, delivery notes at all. At us reaches ridiculous: now in the legislation regulation which demands from us as selling company "provide the due foreign currency in the terms provided by the agreement" is recorded and if did not provide – a serious penalty. So if suddenly the foreign buyer could not or did not want to pay goods in time, responsibility for it is born by the Russian seller! Except direct losses – also a penalty.

Currency control in the IT sphere needs to be cancelled absolutely, mostly it is atavism. Occasionally it turns out that we increase export of hi-tech products not thanks to the help of the state, and contrary to difficulties which it cannot eliminate yet.