As life sciences companies’ international ambitions grow, and global regulatory scrutiny in parallel, managing translated content is becoming ever more complex. So the promise of neural machine learning as a potential solution is appealing. But just how far can the technology go in reducing workloads and improving accuracy and turnaround times?

There are few industries where the need for transformation of translation management is more acute than life sciences. New drug development is happening at an accelerating pace, companies’ global ambitions are growing and, in parallel, the industry is subject to increasingly rigorous and demanding regulatory standards internationally. Indeed, the scale of documentation now needed to bring drugs to market is immense. Medical device manufacturers now face similarly strict controls too, as governments act to improve patient safety in the wake of some high-profile public scares.

In the meantime, the profile of translation automation technology is rising sharply. As with so many digital developments today, consumer experience is shining a light on what’s possible. Instant phrase translation and real-time conversations between people from different countries, enabled by tools such as Google Translate, iTranslate and Waygo, have raised expectations of what could – and should – be possible in a business context. This is especially the case given how much budget and time is allocated to maintaining international consistency and messaging, and containing the risk of meaning being lost or skewed as content is adapted for different markets.

The risks of ad-hoc approaches

All aspects of the life sciences product life cycle, from development and clinical trials through to post-marketing compliance and safety vigilance, must be tracked and documented in very specific ways in every market, in line with local as well as regional and global marketing authorisation and reporting standards. Non-compliance in content can result in delay to market, recalls, fines and most notably risk to patient safety, and associated reputational damage.

Traditionally, translation-related activities in this area have been managed somewhere between local market affiliates and professional translation agencies or language service providers (LSPs), but almost always in a decentralised way, largely out of view and beyond the reach of corporate quality control teams. Processes are labour-intensive, costly and inefficient in terms of speed to market, and there is considerable risk of inconsistency and compliance failure – especially as the international regulatory environment is so volatile. Requirements across every stage in the product life cycle are being enhanced and updated with regularity, introducing new work, new cost and new risk at every juncture if life sciences companies aren’t on top of things.

Yet, moving to a centralised content management model is onerous too. Trying to link previously unconnected systems so that they can talk to each other is a significant and expensive undertaking, and something companies can’t expect to achieve overnight, however valuable the exercise.

Another approach has been to create regional capabilities – teams structured to look after the content needs of groups of countries, which share at least some of the same characteristics or requirements. But these plans place too much emphasis on people to handle all of the work and quality checks, incurring considerable expense and processing time.

Absorbing specialist vocabularies at speed

In the meantime, translation technology has moved on at a phenomenal rate. It is now much better able to cope with specialised vocabularies, for instance, thanks to the coming-of-age of neural machine translation. Now, automated systems can recognise, learn and adapt to new terminology at high speed – to quickly achieve a translation accuracy and quality that can make a meaningful difference to life sciences workloads, and to associated time and cost metrics.

Already, today, neural machine translation technology is at such a point that the overwhelming majority of international life sciences organisations are planning for its usage – and putting pressure on their translation agencies to adopt it. In clinical trials, where the timelines for producing localised content are extremely tight, neural machine translation is already having an impact today.

Notifications of adverse events, for example, can now be translated and understood almost immediately. Even for documents which must be flawless in their translation accuracy, like Patient Informed Consent Forms, potential savings from automation range from 30%-50%, with just the final checks passed to human editors or compliance and quality control teams.

The benefits become more pronounced as volumes of content rise and where the language pairing is favourable (English to Spanish being more common than English to Malay, for example).

Enabling central visibility & quality control

The scope for combining neural machine translation with regional and eventually more centralised content processes is considerable. For instance, as companies look to take a more holistic and efficient approach to monitoring market authorisation requirements and compliance, they may opt to translate everything into English, or standardise in another corporate language, for the purpose of corporate visibility and quality control checks. Such an approach offers those with overall responsibility the chance to verify what the equivalent document says for each market.

Alternatively, in the case of pharmacovigilance, central responsible teams can collectively view all of the real-world feedback/adverse event reporting about a product from across geographical boundaries, enabling speedier and more precise decision-making, with a positive impact on risk control. Although companies can’t (yet) rely entirely on machines alone to pick up everything, neural networks are already having an impact on data mining – by quickly learning the signs to look out for. Human translators still have a role to play though – for instance, in validating whether red flags in translated feedback warrant further exploration and action.

Life sciences companies are already making good headway with intelligent automation of data mining in pharmacovigilance, for instance for combing the published literature and public patient forums for red flags. So their desire to extend the same capabilities to cross-market content translation is logical and rapidly becoming established.

As they have already seen, the cost, time and risk-management benefits are potentially very impressive, especially for high-volume translation services for which raw machine translation output is adequate. The processing cost per word through a neural machine engine is nominal compared to translation costs by a human. For the more exact needs of standard publishable content, the scope for savings are still substantial because of the time and labour saved in initial rounds of translation.

Redefining translation services

Where translation needs are highly specialist, companies may prefer to invest in custom-trained engines that use the company’s preferred terminology, more precise vocabulary and so on. They can seek advice from their LSP partner which is likely to have a good understanding of all the evolving technology options.

While, for now, other industries may be further along with neural machine adoption, necessity is the mother of innovation and, with all of the challenges currently facing international life sciences organisations, this is rapidly driving change in the sector’s approach to translation management.

Monika Vytiskova is a global solutions architect within the Global Content Solutions business at AMPLEXOR Life Sciences