Clicking the AI Translate button for fast-moving, instantly multilingual research data might seem tempting – but it rarely is worth the risk.
In this article, Raisa McNab, CEO of the UK’s Association of Translation Companies (ATC) explores some of the global challenges and technological solutions around the localization of healthcare market research content, with a solid risk-based approach.
Data, lots of data
In every single sector, the amount of data being generated, consumed, and analysed has grown exponentially, and the insights industry is not an exception. In the same way as new and emerging technologies can help you manage and analyse larger amounts of research data from audio-visual verbatim responses to sentiment analysis on drug reviews, they can help you do the same with multilingual content. The main difference is that, with content in different languages, there is an added layer of complexity around translation accuracy, and cultural context.
AI is just a tool
Let’s just put it out there: AI. We understand that it can do all sorts of amazing stuff, especially when it comes to managing and analysing large amounts of data, in a way that our technology solutions have not done before. But let’s be honest, for most of us normal people, AI is just one big black box. And the problem with black boxes is that unless you do actually understand what they do, you just have to trust them to do what you want them to do.
And this is a challenge if the data we handle is sensitive to bias, cultural contexts, and language variations – or indeed if our language data, multilingual or not, includes specific terminology, colloquial language, and personal information. In all of these cases, we still need human intervention and human insights to ensure that what comes out the box is fit for purpose.
So let’s put this into a multilingual context, and consider the risks using AI-enabled translation for life sciences market research.
AI-enabled translation will enable you to process large amounts of data in different languages, and create raw translations of the data for gisting, that is, understanding what it’s all about. But without human intervention, you will not know that, in fact, all references to ‘water’ were translated to ‘whisky’ (true story, happened in one of my jobs recently). Without human intervention, you will not notice that all the personal pronouns in languages that don’t have genders were translated as ‘he’. And without human insight, you will not know whether the term actually used by the respondent for a medical intervention was the lay term, the scientific term, or even the correct one, if AI has translated it.
To put it simply: AI translation can produce OK results, but that depends very much on the type and risk profile of the data – and without human intervention, you just won’t know whether it is OK or not. Whether these issues are massive ones, well, that depends on what you want out of the box. But like in most cases, and for most of us, AI is really just a tool right now, and the way to leverage its power without compromising the results is through a solid risk-based approach.
A risk-based approach
When it comes to managing multilingual insight content and translation, the same principles to risk apply as they do within the much wider AI context, for example, within the EU AI Act: determine the levels of risk, prohibit practices that pose unacceptable risks and set clear requirements for high-risk scenarios, and put in place the right conformity, enforcement and governance practices.
Within the context of multilingual translation, the primary areas of risk to focus on are in data profiling, data security and protection, and in the interface of human-machine solutions.
You cannot evaluate the risks around specific processes or technologies unless you clearly map out what data you have, and what risk level it carries. Within the insights space, there many, many different types of data categories. Some of these can be treated as information-only and be very low on the risk scale. Others, such as primary survey questions, are absolutely on the high risk level due to the impact on the entire survey of an incorrect translation or inappropriate cultural reference.
There are also key considerations around the processing of personal data, as any data containing personal data, and even more so with sensitive health data, poses of course automatically a higher risk than data that does not contain any personal information. And beyond these types of data, there is data that carries a high risk marker from a business sensitivity point of view.
Once done, a data profiling exercise will help you understand what data you have, and what levels or risk its processing for translation and for data security measures carries. And this puts you in a much better position to make better-informed decisions on what technologies to leverage and how, where to implement and prioritise human insights and interventions, and who to partner with for the delivery of multilingual solutions that match your needs.
Partnering with the right people
Choosing the right partner for your multilingual needs is critical, especially in the intersection of life sciences and market research.
These are my three tips for building a successful partnership for multilingual content management:
Choose a language services provider who speaks both ‘life sciences’ and ‘market research’ and is capable of supporting you in this challenging space.
Talk to your partner about your data profiles and processes, and take their advice on where AI and technology could be leveraged, and where human intervention and professional human translators are absolutely necessary.
Make sure that you and your language service provider (and all of their network) protect your data based on its risk profile.
To help you on your way, the Association of Translation Companies has worked together with EPHMRA friends at MRS to create best practice guidance for buying translation services and data privacy and compliance for translation and transcription services:
MRS/ATC Checklist for Buying Translation Services
Translation and Transcription Procurement: Data Privacy and Compliance
Use these resources to orientate yourself to the challenges and solutions around multilingual research data, and for choosing the right partner who can provide expert, quality-managed services and work with you to find the right solutions for your data.
On the Association of Translation Companies’ online Member Directory, you can search for prospective partner companies in both the life sciences and market research space.
About the Author
Raisa McNab is CEO at the UK’s Association of Translation Companies (ATC), the voice for companies operating in the UK’s language services industry whose 245 member companies provide multilingual translation and interpreting support to businesses and public sector authorities across the UK and the world. Raisa is an expert in ISO standards and data privacy for language services, and her experience as a translator and a language services production, quality and development manager have given her a deep insight into the dynamic language services industry in a changing global landscape.
Hyperlinks to resources
https://atc.org.uk/member-directory/
https://www.mrs.org.uk/resources/translationchecklist
https://www.mrs.org.uk/standards/translation-and-transcription-procurement