PROMT machine translation has shown excellent results at WMT21

On November 11, 2021, a conference on machine translation, Workshop on Machine Translation (WMT21) took place. It was organized by the International Association for Computational Linguistics (ACL). The conference aims to bring together academic scientists, researchers and industry representatives to exchange and share their experiences and research results. WMT plays a key role for the entire industry of computational linguistics and machine translation.

This year's conference featured the following shared tasks: a news translation task, a biomedical translation task, an automatic post-editing task, an MT efficiency task, an MT using terminologies task, a chat translation task and many others. In addition to the shared tasks, the conference also featured scientific papers on topics related to MT.

The conference was held in a mixed format this year. Most participants attended online, a small part were present offline in Punta Cana (Dominican Republic).

PROMT has been taking part in the conference since 2013, usually in the news translation task. This year PROMT chose the Terminology Translation Task. Language domains that require very careful use of terminology are abundant. As the wealth of research on domain adaptation shows, such language domains are not adequately covered by existing data and models, all the while new domains appear and models need to be changed to suit them. For this task the organizers proposed to adapt the the terminology of the new COVID-19 domain to neural machine translation to improve the quality of translation of critical information regarding pandemic handling and infection prevention strategies. 

This shared task invited participants to explore methods to incorporate terminologies into either the training or the inference process, in order to improve both the accuracy and consistency of MT systems on a the new COVID-19 domain. The PROMT team worked on two out of the five language pairs, that the shared task was focused on. These language pairs were English to French and English to Russian. In each direction, the company's experts presented the results obtained using two technologies for translation of terminology: PROMT SmartND and Soft-constrained terminology translation. PROMT also took part in the poster session, where Alexander Molchanov, Neural Machine Translation Team Lead at PROMT, detailed the features of each technology and answered their questions.

At the conference, PROMT machine translation technologies were highly appreciated. Both technologies showed the best results for both translation from English into Russian and into French. The results were assessed only using automatic metrics.

“Clients do not always have large high-quality parallel data. But a significant number of companies have either terminological glossaries or competent employees who can determine the correct translation of a particular term. Therefore, the task of adjusting neural translation through the integration of a corporate glossary or direct dictionary editing is very relevant. Our product allows any specialist in this field to create a translation without having any special knowledge in the field of machine translation or programming”, says Alexander Molchanov.

PROMT poster
presented at WMT 21

Back to the list