A recent report from the World Health Organization (WHO) interpretation team (INT) caught our eye, as it dives into a crucial question: can AI speech translation effectively support multilingualism, especially when budgets are tight? The report was published in June 2025 and provides valuable insights into the current state of AI speech translation. INT initiated the research in response to requests from WHO Technical Units facing funding shortfalls for meetings.
It is worth noting that the WHO-tested AI interpretation provider does not include the cutting-edge technology that Interprenet uses for AI speech translation. As a full-service language service provider that also offers high-quality AI Speech Translation, we see the need to educate our clients and prospects on both the benefits and limitations of using an AI-powered voice translation tool. Keep in mind that the WHO results reflect the perspective of a high-stakes, diplomatic communication scenario, which may differ in terms of accuracy from your communication setup and goals.
The WHO's thorough study on AI interpretation across all six official languages (Arabic, Chinese, English, French, Russian, and Spanish) revealed some critical limitations within WHO’s unique scenarios. Here's a breakdown of their key findings:
Low Passing Rates: Out of 90 AI interpretations, only one received a passing grade, with overall scores ranging from a dismal 5% to 83%. The overall average score was a mere 46%.
Issues with Proper Nouns and Cultural References: The tested AI interpretation tool struggled significantly with proper nouns, misinterpreting names of countries and people. For instance, "Brunei Dar Essalam" became "the brunette Russel," and "Haiti" was rendered as "Heidy". Cultural references, like Bangladesh's national slogan "Joy Bangla," were also severely misinterpreted, leading to offensive and unacceptable errors.
Problems with Figures and Technical Terms: Figures, especially those with many zeroes and even dates, were often incorrectly transcribed and pronounced, making sentences incomprehensible. Technical terms were also frequently misinterpreted; for example, "transmission of polio" became "transportation" from Arabic.
Reputational Risks: Not a single AI interpretation in this particular test was free of reputational risks, with some speeches having as many as nine such instances. These risks included elements that could threaten the WHO's values or image or even lead to diplomatic fallout.
Monotonous Delivery: Interpretation delivery was described as monotonous and unexpressive, making it difficult for listeners to follow for extended periods.
The report concludes that AI interpretation is not suitable for use in WHO meetings involving mission-critical communication with external stakeholders that requires the maximum level of accuracy.
Despite the current limitations, the WHO interpretation team recognizes the potential for AI interpretation and has put forth several recommendations:
The WHO report provides a valuable, real-world assessment of AI interpretation in a high-stakes, diplomatic scenario where handling nuanced language, cultural context, and specialized terminology is critical to achieve the highest level of interpretation accuracy.
However, while AI technology may not reliably replace human interpreters in such high-stakes environments, our clients are seeing the benefits of AI voice translation and AI captions in less critical scenarios and in hybrid setups where AI technology and human interpreters are used side by side, providing a blend of efficiency and accuracy.
For example: AI language technology is useful for live events or meetings when there aren't enough attendees needing a specific language to justify a human interpreter's cost, or when an interpreter isn't available at short notice. For events with many languages, a hybrid approach could use human interpreters for high-demand languages and AI for others. Additionally, it benefits bilingual participants who prefer listening in their native language.
Selecting an AI-powered voice translation solution requires careful evaluation of the underlying technology. Our CEO, David Medrano, highlights two essential considerations:
"At Interprenet, we use technology that consistently evaluates top AI and neural machine translation engines to ensure the best performance for each language combination. This way, we deliver live speech translations that are tailored to your specific needs, and you don’t have to settle for an out-of-the-box solution."
"Our service also includes the use of a glossary to train the engine on proper nouns, names of countries and people that you expect to come up in your communication to prevent mishaps such as those experienced by WHO."
In our role as a consultative service provider, Interprenet understands the complexity of multilingual events and we empower businesses and event organizers to make informed decisions that meet their specific needs and objectives. If your organization is exploring how to add AI Speech Translation to their language service mix, we’re happy to provide a demo (and a free trial).