The name of the base system (platform): | Artificial intelligence (AI, Artificial intelligence, AI) |
Developers: | PROMT |
Date of the premiere of the system: | 2019/08/23 |
Last Release Date: | 2020/03/05 |
Technology: | Speech technologies |
PROMT Neural is the machine translation technology capable to study at highly specialized data of the customer — bilingual corpora of parallel texts and at glossaries with highly specialized terminology.
2020: Release of PROMT Professional Neural for Windows
On March 5, 2020 the PROMT company announced release of the program for automatic translation on the basis of neural network technologies – PROMT Professional Neural for Windows. The program integrated the best qualities of cloud and dekstopny solutions. At the heart of the solution – the hybrid technology of PROMT Neural integrating neural network approach and "transfer on the basis of rules" (RBMT). The neural network PROMT Neural model is trained for 100+ of one million parallel segments and tested on considerable volumes of representative selections. Thanks to this PROMT Professional Neural shows qualitatively other level of transfer: texts as a result turn out smooth, have no "machine accent" and are almost indistinguishable from the transfer executed by the person.
Using PROMT Professional Neural it is possible not only to translate words, phrases and small texts, but also documents entirely – the documents MS Office, RTF, PDF, ODF. The program includes PROMT Agent – an important component of all product line of PROMT which allows to transfer fragments of texts directly in all Windows applications or browsers "to one click", translates texts on images, web pages and documents. Besides, in PROMT Professional Neural it is possible to look at the detailed dictionary reference for a separate word or a phrase and also how this word or phrase is translated in different contexts on real examples from books, popular scientific, technical, publicistic literature, encyclopedias and subtitlings to movies. The total amount of contextual examples in the program for the Russian and English languages – more than 40 million offers with transfers. Use of contextual examples online and offline is provided in the program. In the online mode the program addresses for examples on service of free online translation of PROMT.One, and the standalone mode will require installation of all base of contextual examples on the local computer.
Important feature of a product – a guarantee of information security. During the work with confidential information it is possible not to be afraid that sensitive data will flow away in network as PROMT Professional Neural does not require connection to the Internet and gives the chance to work absolutely independently.
the Majority of translation systems on the basis of neural network technologies works in a cloud or can be set in corporate network while the solution of PROMT company is the program installed directly on the personal computer. Unlike the cloud systems, PROMT Professional Neural provides complete confidentiality of the translated information – the solution does not require connection to the Internet and works independently, |
It is important to note that, despite use of neural network technologies, the translator from PROMT has reasonable system demands to the processor and RAM that allows to use it on all modern PCs and notebooks on Windows: the processor of the class Intel Core i5-44xx is required and not less than 8 GB RAM and 10 GB empty seats on a system disk for program installation are more senior (it is recommended with 6 cores more). Use of contextual examples in offline mode will require 20 more GB of an empty seat on a system disk. By default the program works for CPU, but in the presence on the PC or the notebook of the installed graphic processor (GPU) (the developer recommends the class Nvidia GeForce GTX1050 above) will automatically use its opportunities for increase in speed of transfer.
The presented solution of PROMT will allow to simplify work of professional translators, technical and IT specialists, managers of the small and medium companies – all those who in professional activity are faced the need of operational translation of texts and documents and also is obliged to conform to requirements of corporate security.
The first version of the translation program allows to translate from English into Russian and vice versa. In the next versions will appear neural network offline - transfer for the European and Asian languages.
2019: Announcement of the solution
In September, 2019 the PROMT company presented the ready-made solutions for transfers using technologies of neural networks and Big Data - Promt Neural. They in the company are called a new stage of evolution of machine translation.
Initially machine translation systems were based on an analytical method (a RBMT, rule-based machine translation is rule-based machine translation). It means use of a rule set of transfer from source language in target, written by the linguist, and bilingual dictionaries. Then there was statistical machine translation where the method based on the text body is involved. Translation models in it form on the basis of parallel bilingual corpora of texts.
Neural translators became the following step. They use the mechanism of neural networks constructed on matrix calculations which allows to build much more difficult probabilistic models in comparison with statistical machine translators.
As well as statistical translation, neural uses for training the parallel body of the text allowing to compare automatic translation to "human", however when training model it operates not with separate phrases and phrases, and the whole offers. One of premises of appearance of neural translators in Promt call growth of computing powers as training of such system requires much more computing resources.
The Promt Neural technology uses the hybrid approach where the translation variant is offered by both a statistical model, and a neuronet. The algorithms developed by the company analyze the text and decide what technology is suitable for transfer of this or that text fragment better. At the same time according to statements of developers, a system considers a complete context of the offer in a source text and earlier created transfer.
Such approach, claim in Promt, provides the transfer without "machine accent" which is brought most closer to "human".
Promt Neural is intended for the translation of the recognized speech transferred to the text, private correspondence (at the same time the slang and the offensive speech are supported), white papers, orders and resolutions, information messages and publications from different sources, including social networks, document flow of government services and state agencies.
At the time of release of technology Promt proposes solutions for transfers based on universal models for more than 25 languages and also model for industry transfers: IT and telecom, finance, public sector, oil and gas industry, chemical industry, mechanical engineering, mining industry and metallurgy, power, medicine and pharmaceutics.
Proceeding from needs of the customer, translation models specially under specifics of his activity can be in addition created. Transfer can be integrated into business processes and software products and services of the client.
A system is suitable for the translation of multilingual documents, for group correspondence and internal communication in different foreign languages. At text processing it automatically determines source language and a translation profile even if it is several of them.
The solution is available also to deployment of on-premise on the party of the customer and as a cloud service on a subscription, told TAdviser in the company. Installation on own servers is most relevant for customers who have increased requirements to data protection.
Neural transfer can work under OS Linux and Windows in offline mode.
The cost of the solution can vary depending on customer requirements on customizations and numbers of users. The basic set is expected 20 simultaneous connections to a system.