Our Analysts' Insights

The topic of automation has taken the interpreting industry by storm. On the one hand, enthusiasts believe in artificial intelligence as a way to broaden language access at an affordable cost. On the other hand, skeptics worry about all the things that could go wrong in the implementation. But where does it all settle when it comes to organization-level implementations?

End-Users’ Perceptions on AI Suitability to Provide Language Access

As we’ve seen in “Perceptions on Automated Interpreting,” use cases remain limited for now. An analysis of answers from 243 end-users finds that a little over half of service recipients and frontline professionals (54%) think AI would be suitable to provide language access in low-risk conversations.

Procurement Teams’ Plans for the Next 12 Months

At the same time, out of the 40 buy-side procurement teams who participated in the same study, we do not see mass implementation plans for automated interpreting solutions any time soon. By contrast, the written modalities for AI – captioning, transcription, and subtitling – tend to be the first areas that organizations experiment with when dealing with the spoken word.

Strongest Use Cases to Date

Automated interpreting is not perfect, but it can add value. Where? It finds a home for lower-stakes use cases: “Good enough” interpretation for an internal meeting is much less of an issue than for a CEO’s speech. It’s also much more adequate at the information desk of a hospital than in the oncology ward when delivering bad news.

The buzz in the industry is about risk. Every conversation involves a degree of risk, but significant variations exist regarding the impact of issues and their odds of occurring. Beyond possible discomfort from working with automated services or occasional confusion with irrelevant translations, the concern is that one party may end up making a wrong decision, struggling to make a decision, or may be left with a wrong impression. Possible negative outcomes include unsafe situations, physical harm, financial or legal impact, loss of freedom, impact on equity or dignity, and brand damage.

So where does CSA Research observe the most implementations to date? Definitely not in high-risk sectors such as health care, legal, or social services. Instead, the bulk of technology vendors target online, in-person, or hybrid events. These make attractive use cases because there is often already an infrastructure – such as high-end microphones – to capture good-quality audio and speakers tend to use full-sentences and standard language with proper grammar, and strive to enunciate well. Such conditions make it easier for speech recognition to do a good job, which then sets the machine translation part of the process on the right foot. Yet even so, not all events are suitable for AI use, even with a an engine trained for the topic discussed or with the people, product, or company names that will be mentioned.

Other organization-level implementations focus on contact centers, where end-users are often already conditioned to talk to automated agents, and at information kiosks – such as the concierge desk of a hotel.

We have yet to find organization-approved deployments of AI as the main source of interpreting for any critical interactions in the legal or healthcare sectors.

Want to Be Interviewed?

Our analyst team is currently doing research with organizations that either already use automated interpreting or are considering its implementation in a business-to-business, business-to-consumer, or government-to-consumer scenario. Whether you love it or find it too buggy, we’d love to talk to you. Your insights will help us establish best practices on criteria to consider when selecting use cases for automated interpreting implementations.

If you’re not a user yourself but know of an organization that is, we’d welcome the referral.