Society has long been concerned with how the adoption of machine translation (MT) might impact our jobs and livelihoods. Could the rapid developments in MT diminish the role of translators or, as the technology evolves further, negate the need for them completely?
Once TBSJ began looking seriously at MT as a tool for helping to deliver high-quality translations more efficiently, we reconfirmed that nothing could ever replace really good translators. It was therefore heartening to hear similar affirmations on the importance of people in the translation industry at MT Summit 2021. TBSJ’s own experts in the field, Paul O’Hare, CTO and co-founder, and Yury Sharshov, chief scientist, attended the global event for users, providers, and researchers of MT. Here, they share their findings and reflect on how TBSJ approaches the sometimes controversial topic of people and machine translation.
People are more involved in the development and use of MT than they have ever been. Unlike in the early years of this technology, when it offered only a one-size-fits-all generic approach, professional use of MT today involves post-editing by translators as well as the application of natural language processing and other tools to make the overall translation process as efficient as possible.
MT would not be possible without the input of highly skilled linguists. At MT Summit, a team from translation firm Welocalize emphasized the importance of the human-in-the-loop approach to the complex workflows that have evolved around MT in their half-day tutorial, “A Deep Learning Curve for Post-Editing.”
“At TBSJ, when we do use MT, we consider it merely the first step in any translation project,” says Yury. “In line with the industry’s human-in-the-loop model, we are focused on making post-editing activities as efficient and effective as possible.” This involves a wide range of processes and automated steps supported by expert translators, editors, and technologists: from MT customization, post-editing setup, and training, right through to quality and performance evaluation.
Ahmad Taie from language services provider Lengoo gave a hands-on tutorial for newcomers, explaining that acquiring and preparing datasets for training MT models requires a special set of skills, even if the training process itself has become considerably easier in recent years. Furthermore, as echoed at the roundtable “Building MT Capacity and Competence In-House,” it is indispensable to have a skilled team capable of setting up a production environment, leveraging the results of state-of-the art industry research, and applying linguistic and customer knowledge to every step. “TBSJ adopted this approach from the get-go and acquired highly valuable experience in the process,” explains Yury.
One of the key challenges for modern, neural network-based MT is terminology integration, and people help with that, too. The core tasks, such as terminology management, terminology identification, and translation with terminology, were discussed in presentations given by various well-known companies. Mārcis Pinnis of translation and localization firm Tilde spoke about their terminology-related challenges and solutions. Randy Scansani and Loïc Dugast of translation provider Acolad discussed how bilingual glossaries can help to improve translation quality. And Konstantin Savenkov of AI meta-provider Intento talked about the technical aspects of interfering with neural MT—such as when handling tags in technical content—and experiments aimed at preserving high MT quality.
At TBSJ, it was clear even before the advent of neural MT that glossary creation and refinement is—despite the considerable time and effort required—highly worthwhile. During our 11 years of operation, internal linguists with advanced subject-matter expertise have hand-built a wide range of high-quality data resources, such as termbases, translation memories, and custom quality-assurance checks, on a client-specific basis.
Looking at various areas of the industry, therefore, it is clear that people have a critical role to play in the development and use of MT. Indeed, many summit presenters noted that skilled translators are now in high demand, not just to support MT, but also for regular translation, as the majority of professional translation produced globally is still not yet receiving a MT boost.
Even as technology evolves, MT will never be a means to fully replace translators. Rather, it is simply another way to help make them more efficient. MT can easily meet today’s low-end demand, but there will always be complicated and critical translation tasks where the expertise of the human professional is indispensable. “At TBSJ, our aim with MT is to enable specialists in language and translation to fully utilize their abilities by releasing them from the more menial tasks where MT excels,” says Paul.
TBSJ works closely with translators, editors, project managers, and developers to create, evaluate, and refine the tools and processes involved in the application of MT. Looking ahead, Sarah Bull, chief relationship officer and director of legal services, says those relationships “will likely become even more collaborative as we navigate the unchartered territory of MT’s evolution.”