Why ‘Neural Fluency’ is a Must for Globalization

Do more with less—this has been the mantra within localization departments of multinational companies in the past few years.

Advances in language technology have indeed been enabling various parts of the translation workflow to do just that. Translation memory and computer-assisted translation, for instance, offer linguists a productivity boost. Meanwhile, translation management systems give project managers and various stakeholders in the localization pipeline the ability to seamlessly handle complex project requirements.

Then there’s neural machine translation (NMT), which has moved automated translation from the era of statistical models to deep learning and artificial intelligence. NMT’s progress has been so rapid, its effects on the localization pipeline so pronounced, that it has de facto replaced statistical MT as the industry standard.

Advertisement

Today, the globalization efforts of localization departments within multinational companies is increasingly relying on “neural fluency”—or how adept they are at NMT and how well they can use it—to scale their business.

Achieving Global Objectives through Local Language Expectations

The more a company expands globally, the more it needs to pay attention to local language expectations in the new markets it is trying to enter.

Their localization departments need to pay close attention to the linguistic, cultural, and even socio-economic nuances of the markets they want to expand into—this makes effective translation a major part of a company’s global growth strategy, even more so for isolated markets.

This is a daunting task that is near impossible without technological leverage and momentum. This is exactly what NMT provides.

With NMT, localization departments can tune localization requirements, adapt content, and focus on local language expectations that make region transfers appear seamless. Granted, predecessor technology—statistical MT—can also help in this regard. But NMT offers several advantages over SMT. Indeed, with the emergence of NMT, localization departments can finally truly do more with less.

Banking on Neural Fluency

The very first thing that sets NMT apart from SMT is the quality of its output. NMT is often vastly more fluent than SMT, which means, right off the bat, most of the advantages afforded by SMT, NMT simply amplifies.

The easiest example is that because NMT provides more accurate translation compared to SMT, it enhances and reduces human effort. Expanding this analogy, this means where SMT allows for local competitiveness and accelerates time to market, NMT does the same thing much better.

NMT is not just an upgrade, however, but more of a paradigm shift. Its fast pace of development has allowed it to overtake SMT in a fraction of the time it took SMT to become as mature as it is today, not to mention NMT is still evolving. Translation quality is still expected to go up, and costs related to translation, hardware, and overhead are still expected to go further down.

The “Localization Goldmine”

Yet another dimension to consider is the training data. NMT thrives on quality training data, even more so than SMT where the quality of the training data is not as big a concern. Localization departments have an abundance of this data in the form of aligned Translation Memories.

Training engines with client data not only helps ensure a remarkable improvement in fluency and quality, it also guarantees the translation output is in the “Client’s Voice.” Many localization departments are finding significant cost savings in the MT process by leveraging their own Localization Goldmine.

Where does all this leave the localization departments within global companies that need to constantly do more with less as they continuously break into new markets?

They need to know their neural. They need to understand how to take advantage of the high quality training data they accumulated over the years. They need to understand how best to make the most out of NMT: from the latest technologies to the best practices to what developments to expect next.

They need neural fluency: the know-how to use the still-developing technology to its fullest.

From Nice-to-Have to Must-Have

NMT is gradually disrupting a few of the ways the language industry have always operated.

Take for instance the fact that since NMT output is already quite fluent, it reduces manual post-editing of grammatical syntax, but shifts the brunt of the work to making sure the translations maximize cultural appeal. It’s less post-editing and more post-translation messaging quality control.

This means fewer man-hours spent on repetitive errors. This, in turn, means more hours spent on messaging refinement. Additionally, the extra linguistic manpower can be turned towards data processing—translators work less on bulk translation of customer-facing material and more on creating data sets for better NMT engines.

The disruption is already under way, even in terms of academic research, where researchers are finding that the usual way MT output is evaluated no longer works as well for NMT because of its fluency. Many researchers have been looking for better ways to evaluate MT output. It was previously nice-to-have a better alternative. Now, it’s a must-have.

The same argument applies to localization departments and their knowhow of MT in general. Two years ago, it was nice to have knowledge about SMT, how it works, and how to leverage it for a limited number of applications. Today, it’s a must have.

Localization teams need neural fluency to gain a competitive edge today and to make sure their companies can continue to scale globally tomorrow. With the rate of research poured into deep learning AI and NMT, they cannot afford to stagnate.

Throughout its 50-year history, SYSTRAN has been thriving in the research and development sector, leveraging machine translation progress and innovation to deploy systems like the Systran Pure Neural NMT for localization teams.

Systran’s VP of Sales and Marketing, Ken Behan, will be talking about effective MT processes during XTM Live USA, XTM International’s knowledge-sharing conference where end-clients can get a 360-degree view of solutions, best practices, and latest localization technology.