PROMT Neural Translation Server DE

The reliable and secure innovative solution for integration of neural machine translation into software products or web sites on premises or as a cloud.
OVERVIEW
Languages
CUSTOMIZATION
Requirements
Integration into Existing Workflow
Secure to the Core
Easy to Deploy and Easy to Use on-premises and cloud-based
Document Translation with Maintaining the Original Layout and Formatting
State-of-the art Neural Machine Translation
Brand Language and Terminology Management
FEATURES
API Features for Developers
  • Text Translation (plain text string, string with tags, array of strings)
  • Document Translation (XML, HTML, DOCX, XLSX, PPTX, PDF (text), JPEG, PNG, TIFF)
  • Source Language Detection
  • Dictionary Lookup (multiple translations for a given word)
  • Website translation
  • Ready-to-use module connectors for GMS systems (SDL Worldserver, SDL TMS, GlobalSight) and Translation Memory systems (Trados, Déjà Vu)
Web-Interface for Users:
  • Text and Document Translation
  • Look-up Dictionary
  • Human-translated Examples
  • User Dictionary and Translation Memory
Administration Features for Administrators:
  • User Management
  • Detailed Statistics on Usage
  • Scalability and Load balancing
  • Linguistic Customization



CUSTOMER STORIES


Arabic
Armenian
Azerbaijani
Bulgarian
Chinese
Simplified & Traditional
Croatian
Czech
Danish
Dutch
English
Estonian
Farsi
Finnish
French
Georgian
German
Hebrew
Hindi
Hungarian
Italian
Japanese
Kazakh
Kirghiz
Korean
Latvian
Lithuanian
Malaysian
Polish
Portuguese
Romanian
Russian
Serbian
Slovak
Spanish
Swedish
Tajik
Tatar
Turkish
Turkmen
Ukrainian
Uzbek

The translator provides  more than 40 languages ​ ​ and 100 language combinations. The list of language pairs developed by PROMT is constantly expanding. Language package is selected at purchase and affects the price of the solution.

For a complete list of language pairs, see  Supported Languages.

Customization is critical for high quality machine translations. Customization provides additional context to generic translation models so that translations can reflect a client’s tone and unique terminology.

To get better translation quality with PROMT Neural Translation Server (PNTS) you can use the following:

  • User dictionaries/glossaries

Many companies have either terminology glossaries or the competence of employees to determine the correct translation of a particular term. In this regard the challenge of tuning the neural translation via integration of corporate glossary or direct dictionary editing is essential.

To meet this challenge PROMT developers have created a special technology – Neural Dictionary (ND). The technology is implemented in PROMT Neural Translation Server for both Windows and Linux versions (in Linux version you can use neural dictionaries, but the dictionaries can be created only in Windows-based desktop and server products).


SmartND and SimpleND

The Smart ND technology is available for several language pairs: English to Russian and Russian to English, English to French and French to English, English to German and German to English.This technology is able to take the context into account and put terms in the right gender, number and case. For other language pairs the SimpleND technology is used. With this technology the terms will be given in the same form as they were added to the dictionary.

  • Parallel data

Parallel data are original texts and their translations in a structured form, where each sentence in the original language corresponds to a sentence in the target language. When tuning the system with parallel data, it is trained so that it adapts the style and vocabulary of the data it was trained on. Parallel data effectively train the system to translate terminology or stylistic clichés that are domain-specific, such as in technical texts, legal or financial documentation.

To train PROMT Neural Translation Server on parallel data you need a special solution – PROMT Neural Translation Server – Training Addon (can be purchased separately).

PROMT Neural Translation Server – Training Addon is:

  • ability to quickly train PNTS on your data
  • translation of texts from your domain with high terminological and stylistic accuracy

Training data

  • Data must be presented as one or more files in TMX format, or as multiple UTF-8 files. 
  • Recommended volume – from 50 000 parallel segments (i.e. sentences in the original language and their translations).

Install PROMT Neural Translation Server – Training Addon, upload your data and train the neural network model to translate highly specialized or industry-specific content.

System requirements

PNTS can be installed on a dedicated server (physical or virtual) or a desktop computer with the following characteristics:

Windows

COMPUTER/VM CHARACTERISTICS

Neural translation module operation mode

CPU

CPU

GPU

Type pf processor (CPU)

 

Intel Core i9 (or higher), Xeon E5 (or higher),
AMD Ryzen 7 (or higher) or Epyc 2 (or higher)

Number pf processor cores

8

16

16

Discrete video card (graphics accelerator, GPU)

N/A

 

OS-compatible CUDA 12 supported graphic processor(GPU) with a minimum of 8 GB** dedicated memory

RAM

32 GB**

Disk space

100 GB***

TECHNICAL LIMITATIONS

Number of  neural models used simultaneously

up to 2

up to 4

Depending on the amount of video memory*

Estimated speed of document translation with one model

3 pages per minute

5 pages per minute

90 pages per minute

 
* in GPU mode with default settings each model (translation direction) requires 2 GB of video memory. The model is loaded into memory when translation in the corresponding translation direction starts, and it is unloaded automatically after 10 minutes of  inactivity in the translation directions provided by this model. The latest version of the driver for the corresponding OS must be installed on the server. Working with vGPU is supported only in Virtual Dedicated Graphics Acceleration mode (vDGA, also known as GPU passthrough).
** for up to 200 users and up to 8 neural models working simultaneously. For more users or neural models, please contact PROMT Company for system requirements.
*** additional disk space may be required, depending on the amount of files stored by users on the server.
**** for configurations that allow users to work through the web interface or PROMT Agent.
The following OS are supported:
– Microsoft Windows Server 2022;
– Microsoft Windows Server 2019;
– Microsoft Windows Server 2016;
– Microsoft Windows Server 2012 R2.
The OS shall include the following system libraries and components:
– Microsoft Internet Information Services 6.0 or higher
** If you plan to use Windows authorization, you must put the server in the same domain as the users (or in a domain that has trust relationships with user domains).

Linux

COMPUTER/VM CHARACTERISTICS

Neural translation module operation mode

CPU

CPU

GPU

Type of processor (CPU)

 

Intel Core i9 (or higher), Xeon E5 (or higher),
AMD Ryzen 7 (or higher) or Epyc 2 (or higher)

Number of processor cores

8

16

16

Discrete video card (graphics accelerator, GPU)

N/A

 

OS-compatible CUDA 12 supported graphic processor(GPU) with a minimum of 8 GB** dedicated memory

RAM

32 GB**

Disk space

100 GB***

TECHNICAL LIMITATIONS

Number of  neural models used simultaneously

up to 2

up to 4

Depending on the amount of video memory*

Estimated speed of document translation with one model

3 pages per minute

5 pages per minute

90 pages per minute

 
* in GPU mode with default settings each model (translation direction) requires 2 GB of video memory. The model is loaded into memory when translation in the corresponding translation direction starts, and it is unloaded automatically after 10 minutes of inactivity in the translation directions provided by this model. The latest version of the driver for the corresponding OS must be installed on the server. Working with vGPU is supported only in Virtual Dedicated Graphics Acceleration mode (vDGA, also known as GPU passthrough).
** for up to 200 users and up to 8 neural models working simultaneously. For more users or neural models, please contact PROMT Company for system requirements.
*** additional disk space may be required, depending on the amount of files stored by users on the server.
**** for configurations that allow users to work through the web interface or PROMT Agent.
The following OS are supported:
– Ubuntu 20.04/22.04/24.04;
– Debian 10/11;
– AstraLinux SE 1.7/1.8;
– Alt Linux 10;
– RED OS 7.3/8.
The OS shall include the following system libraries and components:

  • GLIBC version 2.28 or higher
  • libicu versions 63 to 70
  •  libgcc_s.so.1
  •  libstdc++.so.6
  • systemctl service manager
  • .NET6 (it is included in the distribution set and is installed with the product).