Technology in favor of efficiency

Machine translation is on the rise. As more companies adopt machine translation, there is growing evidence that translating first with a machine and adjusting with human revision is faster and more cost-effective than human-only translation in many domains while still capable of achieving human translation quality where this is the objective.

This article looks at machine translation within life sciences, why it is accelerating and what to expect in the near future.

What is machine translation?

Machine translation (MT) is a method of translation performed by a computer. The MT engine “reads” the input text and uses algorithms to generate a translation. Apart from defining the source and target languages, MT is essentially automated translation. Getting to an acceptable level of automation takes some effort and depends on the technology.

Various MT technologies have been sent into the race, but neural machine translation (NMT) is pulling ahead. NMT outperforms other paradigms like statistical MT and rule-based MT in terms of fluency and accuracy. MT engines can also be “trained” with specific domain and customer data to achieve better results regardless of the underlying technology. The quality and amount of data impact the MT engine performance.

Google Translate is perhaps the most well-known example of MT and one that you have likely used as a free public web-based service many times. But others have improved significantly in the last few years, such as Microsoft Translator, Amazon Translate, DeepL, Google Cloud, Alibaba, Promt, Yandex, and so on.

Advances in artificial intelligence (AI) and big data capabilities have helped drive development in the MT market. The number of available MT engine options has increased substantially in a year. New language pairs become available regularly, and quality scores are continuously improving.

Why use machine translation?

Increasingly, MT has the potential to offer a more affordable and viable way to translate. The overall machine translation output quality has improved to the point that large brands and language solution providers now use MT. It works well for an increasing number of language pairs, domains, and applications. The benefits of using MT are cost and time savings today without disappointing on quality.

There is more to gain from MT than just the cost-time-quality performance equation. MT would hardly be innovative if it did not also create new opportunities.

Companies that start using MT usually observe multiple changes in their translation ordering behavior:

• MT replacing traditional translation processes (speed and cost),
• MT is creating new demand, such as new languages for internal communication (ease and speed),
• MT as a new service, such as instant translation for screening market research and tender documentation to get quick insights or identify specific information that facilitates the following steps (new value).

What is also clear is that MT has its limitations. Creative content or content with high degrees of contextual variation remains squarely within human hands. This includes brand and marketing content. On the other hand, easy-to-read and structured content is ideal for MT.

Life sciences in transformation

The life sciences industry is among the most dynamic industries today. Translation technology is just one piece in a massive jigsaw puzzle of transformation. Let’s look at the big picture first and then zoom into translation.

Nearly all the major trends in healthcare are digital by default, with AI playing a pivotal role. We see new fields such as AI-driven diagnosis and expanded use of telemedicine and virtual treatment. Digitalization is changing regulations, leading to international harmonization of product information for regulatory assessment and clinical research.

The Medical Devices Regulation (MDR) and In Vitro Diagnostic Medical Devices Regulation (IVDR) of the European Union are prime examples. Both directly impact translation volume, language scope, and the type of information to be translated or updated.

In the digital world, data is becoming an even more significant asset. Life sciences organizations have declared digitalization, cloud-based processing, interoperability, and international communication as strategic goals. The COVID-19 pandemic has only increased the urgency to digitalize and be able to communicate more through new, virtual channels.

Pharmaceutical companies had to enhance international collaboration like no other industry to react to the pandemic and achieve in months what would usually take years. Life sciences leaders today are confident in the capabilities of AI to solve significant industry problems, including pandemic-related challenges.

In short, digitalization and adoption of AI in translations, like in many other disciplines, are inevitable.

Adopting new language technology

Life sciences companies are traditionally cautious about adopting new technology, and rightfully so. Highly sensitive data combined with demanding standards and regulations makes change more difficult than usual. Besides quality, a typical concern is that new translation technology might cause issues documenting the process for regulatory frameworks and ISO certifications.

Furthermore, data privacy and intellectual property, along with highly specific terminology, have been sources of doubt and difficulty when it comes to refactoring translation processes. As a result, MT adoption in life sciences has been slow so far.

While in 2021, less than 16% of life sciences organizations have used MT, we expect this to increase significantly in short order.

Rather than starting with the core product information, companies find it easier to adopt MT by starting from a different corner. An excellent sandbox to try out MT is internal communications that need a quick, low-cost translation that serves the purpose as a minimum viable product, as it were.

When it comes to technical, patient, or physician-facing information, MT is taking on some of the human translation tasks in order to speed up the whole process. The Post Editing review is carried out by human subject matter experts and ensures translations consistently meet the same high-quality standards.

In pharmacovigilance, MT can be used to speed up and/or automate specific steps. It can be used effectively when there is a vast amount of content in various languages to be screened or a high number of adverse event reports (reported serious and non-serious side-effects) across multiple markets. The respective pharma companies need to address them all and quickly. Still, the amount of multilingual content can be a bottleneck as the majority of such reports are often not in English. Well-positioned MT in the report handling process can dramatically improve throughput.

Other suitable candidates for MT-powered processes include:

• Translating time-boxed communications: urgent messages, adverse reaction case reports, post-market surveillance, social media, and similar content with legally binding timelines
• Translating with restricted budget: low risk, low cost but otherwise required translated content
• Translating exceptionally large volumes of content: projects that require multiple translators to work simultaneously to meet deadlines
• Screening large content volumes: for getting quick insights and gisting, automated translation allows understanding or summarizing the main aspects.

What is the key to MT-driven success?

There isn’t just one. Various factors are essential. Without them, success is a matter of luck rather than a certainty.

• Skilled human experts: training and improving the engines, and post-editors reviewing the MT output
• Customization: wherever standard MT is not enough, training and improving your MT engine with high-quality data is necessary to reach a quality level where human translators are able to just post-edit as opposed to re-translate
• Localization strategy: when does it make sense to use professional human translation, and when does an MT-powered solution reach the required translation quality in the most cost- and time-efficient way?

The required translation quality is the same whether MT is involved or not. With the advances in MT technology, the question is no longer whether we can achieve the required level of quality with MT but rather how.

A German multinational pharmaceutical company shifted six language pairs (English into Portuguese, Dutch, French, German, Mandarin, and Spanish) from human-only translation to an MT process. They made this move after a six-month pilot with Portuguese. The pilot showed them they could reduce their translation effort by 30% without compromising quality. A skilled human team and customization were considered key to success in this case.

Like that of other highly regulated industries, healthcare content requires a careful choice of MT. It is essential to understand that MT engines are not a one-size-fits-all technology. Some MT engines perform better for certain language pairs.

What’s next for MT in life sciences?

AI is one of the tech trends that will touch all aspects of healthcare — from diagnosis to drug development and preventive care. The crux remains the data needed to properly train the algorithms.

The MT market is forecast to register a CAGR of 7.1% over the next five years. When we consider that life sciences are starting to accelerate the adoption of new language technology, the demand for advanced, high-caliber language services providers will explode.

The standards of healthcare, pharmaceutical, clinical research, biotechnology, and medical device companies will all be putting pressure on MT technology. Providers need to train and develop MT engines that are ready to tackle the life sciences content without extensive additional customization.

With the adoption of MT-driven translation processes comes the need for experienced post-editors. Post Editing of MT engine output is increasing demand in the translation services industry. Post-editing requires a different skillset and mindset than traditional translation. Translators are being trained and certified as post-editors, learning to recognize typical errors, assess the behavior of an MT engine, distinguish between light and post-editing, etc. Post-editing can be maximized with well-written source text, terminology management, and accurate translation quality estimation.

By running automatic metrics such as BLEU, METEOR, and TER, the post-editing effort can be predicted and an indication of overall MT quality assessed. It is important to note that automatic metrics can only offer an indication of MT quality. The only reliable method for evaluating MT is still human evaluation.

MT is limited when used as a standalone solution. When incorporated into the overall content process and combined with other technologies (translation memory, terminology, other AI, speech recognition, sentiment analysis), MT reveals its full power.

The dedicated Vistatec Life Sciences Division works closely with some of the world’s leading organizations to deliver safe, accurate, and high-quality solutions. When combined with other Vistatec services, we help you scale up.

Looking to involve MT In your localization strategy? Please fill out our discovery questionnaire, and one of our MT Experts will be in touch.