Welcome back to The Lab, where we take a look at what’s cooking in life sciences localization. This month, we are talking about some key principles, best practices, and real-world examples of how artificial intelligence (AI) and machine translation (MT) are being used to reduce costs and turn-around times, while improving both quality and security of life sciences-related translations.

First off, you may have noticed that I added an extra “leg” to the traditional “three-legged stool” that is typically discussed and considered to be the dominant paradigm for the sales or procurement of translation services. That paradigm says that price, quality, and speed (turnaround time) are the three criteria for purchasing translation, and that you can only get what you want for two of the items. For example, you can get low-cost and fast translation, but then quality will suffer. Likewise, you can achieve high-quality, fast translations, but you will have to pay a higher price.

This model, in my opinion, is outdated for two reasons. First, AI-powered MT, in a growing number of domains and applications, is able to achieve all three objectives for quality, speed, and price. Additionally, security and data protection is increasingly becoming a must have for the buyers of translation. This is especially true in the life sciences industry.

Thus, we have an additional leg called “security and data protection” on the stool. And again, using AI-powered MT, with the appropriate hosting options and workflows, we can oftentimes achieve the requirements for security and data protection along with the requirements for the other three legs of the traditional translation paradigm.

Of course, due to compliance, regulatory, and quality concerns, there is some reluctance by life sciences companies to roll out MT for all workflows. That is why a common approach for deploying MT is to start with non-critical translations where speed and price take a priority over quality or where high quality can be achieved using MT without post editing (PE).

For example, many firms are using MT to translate internal communications such as emails and instant messages and non-regulated financial documents such as invoices, receivables and payables, or other non-critical texts such as posts on social media and community forums.

According to Bill Young, director of life sciences sales at Systran, “There’s been a general apprehension by life sciences companies to adopt MT because of regulatory concerns. However, we are consistently seeing MT being used for internal communications, and this serves as a way for companies to get started with MT, and from there, the application spreads across other text and document types.”

This approach to deploying MT is becoming almost an industry norm. And once companies grow accustomed to the use of MT for less sensitive documents and content, they look for ways to deploy MT for regulated texts such as labels and manuals. “What often happens is that once a company gets comfortable using MT and sees that the engine learns the corporate terminology and can leverage regulatory and medical dictionaries to support translation, they start to use it for other applications including patient-facing texts,” Young said.

A great example of this is the work that a leading LSP is doing for Medtronic to reduce turn-around time and costs for the translation of post-market vigilance (PMV) text using MT. PMV is required to track performance and complaints after a product is released to the market. It requires that some types of complaints be reported to regulators within 48 hours of receipt.

Since complaints can be received in any language from any country around the world, a fast and reliable translation process is needed. Medtronic, for instance, utilized an automated workflow in combination with MT to reduce turnaround times by 14 days and cut costs by over 80% for the PMV of one product cited in their example.

Of course, PE and human-in-the-loop (HITL) applications for MT is still the standard workflow for texts that have higher readability and quality requirements. And while the reductions in price and turnaround times will not be as significant as fully automated MT, most LSPs lead their MT offering with this model.

According to Lara Tosoni, chief revenue officer at Meinrad.CC: “Using AI in the translation process is becoming a standard practice, and if we look at the types of text used in the life science industry, it’s typically technical documentation and product descriptions with full post editing.” Lara also pointed out the importance of having the relevant ISO certifications, including ISO 18587, which provides the requirements for the processing of human post-editing of machine translation output and the post-editors’ competencies.

Barbara Peralta, director of life sciences solutions at Acolad, offers similar feedback.

“MT is an invaluable tool for our med device clients to handle the volume of labeling updates necessary for MDR and IVDR compliance,” she said. “It also is used for adverse event translations where speed is important. In both use cases a post edit by qualified life sciences linguists is essential to ensure quality.”

A best practice to achieve quality, speed, and pricing targets is to invest on the front end to train and customize MT engines. This reduces the amount of post-editing (which means time and cost) required on the back end.

To achieve targets for security and data protection, LSPs and enterprise buyers of MT services should select a secure MT provider that provides a suitable hosting option. Options include fully on-prem self-hosted and secure cloud hosted. Another consideration are the geolocation regulatory requirements that may disallow the hosting of patient data outside of particular region or jurisdiction. To that end, the use of a hosting provider that has data centers and hosting options in a variety of areas could be important.

Lastly, while the terms AI and MT are becoming increasingly common and interchanged in the localization industry, I’d like to argue that, while the most popular types of MT being used in the life sciences industry do use AI, good MT tools are comprised of more than simply an AI algorithm. For example, MT tools should provide an intuitive dashboard, a variety of plugin and integration options with a translation management system (TMS) and other tools to facilitate workflow automation, business reporting, and security.

At the same time, AI itself is much more broadly used than just MT applications. Even, if we limit the scope to life sciences translations, companies are using AI for applications outside or parallel with translation. These additional applications include being able to detect adverse reporting events, sentiment analysis and business intelligence reporting, and resource planning.

For instance, Elizabeth Milkovits, who formerly served as the director of AI solutions for iQvia, told me, “AI can help us more quickly identify adverse events, but also can help us understand workflow requirements for clinical trials and thus assist with resource planning.”

In summary, AI and MT are both becoming increasingly common and increasingly important in life sciences translation, as they help to achieve targets for all four legs of speed, quality, cost, and security. And, while the benefits are most significant for text that can be translated in a fully automated manner, these benefits are also significant for HITL MT usage.