Neural Machine Translation
Systran is a language solutions provider headquartered in Paris, France. Recently, the company released a “Neural Machine Translation” engine, which is primarily targeted at enterprise-level users with its PNMT (Pure Neutral Machine Translation). Systran claims that this translation tool is very close to the fluency level of human translators, and according to the company, “it can even translate better than humans,” and can be used across various industries.
When designing PNMT, Systran primarily considered serving enterprise-level users who deploy business strategies in multiple countries and regions to improve their market efficiency in different countries and reduce the time spent. Systran can adjust the translation tool to meet the needs of users in different industries.
According to Systran, this PNMT tool can translate over 140 combinations of source and target languages. To help potential buyers better understand the product before purchasing, Systran offers free online demonstrations for interested users to test. From the company’s description, this engine first “reads” and “understands” the statements that need to be translated, and only after this step does it translate them into the target language.
The translation tool developed by Systran is based on neural networks as the underlying technology, similar to the way the human brain works. A significant feature of neural networks is that they use thousands of raw materials as their learning resources during training, which for Systran means learning specific rules of various languages from a vast corpus. In fact, human translators also go through the process of “acquiring text – understanding text – translating text”.
Using neural networks as the underlying support technology to develop translation tools has become the choice for major companies to upgrade from traditional statistical and rule-based model frameworks.
This month, Google officially announced that it has applied its neural network machine translation system GNMT to Google Translate;
At the Microsoft Connect 2016 conference, the company also announced that all Microsoft speech translation services have fully adopted the most advanced deep neural network technology, including Skype Translator for real-time voice translation and Microsoft Translator mobile app, etc.
In addition, all developers and users can integrate this technology into various applications and services through the API provided by the company.
Just a few days ago, Sogou also publicly released its own real-time voice translation technology based on deep learning systems.
“We are living in a historic moment of significant change in the field of machine translation,” said Systran’s CTO Jean Senellart. “Communication in languages has become a necessity. We are excited to create such a tool for businesses across various fields and to serve in different niche markets.”
Student Union of the Youth Class, University of Science and Technology of China