07 Aug Can machine translation do its job in the case of a difficult language such as Polish?
Machine translation based on neural networks is a hot topic, especially when it comes to languages with complex syntax and grammar, such as Polish. A case study presented at the LocWorld37 conference has demonstrated that it is possible to use English–Polish neural machine translation effectively for business purposes.
In 2014, neural machine translation (NMT) was merely a subject of scientific research. By 2016, it became the dominant approach to automated translation. Today, it is an area of interest for giants such as Google, Microsoft, Baidu, Amazon, Facebook and many other companies that work on language technologies.
During the LocWorld37 2018 conference—the most important translation and localization event worldwide, hosted for the first time in Warsaw—Jarosław Ganczarenko from GET IT and Andrejs Vasiļjevs from Tilde (our technology partner from Latvia) presented the results of implementing an NMT engine for English–Polish legal translations.
Polish is particularly difficult for machine translation algorithms because of its complex syntax and grammar, and an extensive inflection system. For this reason, conventional (non-neural) translation does not perform well in this language.
GET IT and Tilde had carried out successful tests which have shown that NMT improves the quality of translation into complex languages, such as Polish, ensuring greater fluency and better context matching than the previously used statistical machine translation (SMT) solutions.
In practice, this means that neural machine translation represents a genuine breakthrough for automated translation. NMT forms part of a wider trend, which has seen developments in neural networks, artificial intelligence and machine learning, taking place since 2016. These technologies are applied in areas such as speech and image recognition, autonomous car design, medical test automation, etc.
The application of NMT yields tangible benefits to clients:
1. It greatly improves the speed of translation—to as many as 100 pages per hour (translation intended for gisting or for further processing).
2. Machine translation with post-editing (revision by a translator) can be completed much faster (by up to 50%) than conventional translation.
NMT engines are now being successfully used by GET IT for legal and financial texts as part of the SMART translation offering. One more benefit provided by our solution (as opposed to free-of-charge translation tools) that is crucial to business clients is data security. SMART translation projects are performed using secure internal servers, ensuring complete security of client content and data (such as financial or personal information), and application of rigorous processes to guarantee full data confidentiality.