JAMA recently published an article on using mainstream translation software in a health care setting, which was also picked up by Gizmodo. Researchers used Google Translate to translate emergency department discharge instructions prepared in English into Spanish and Chinese.
The results were then retranslated into English by bilingual translators, and lo and behold Google Translate handled the lion’s share correctly—92% of Spanish instructions were rated accurate. Not surprisingingly, Chinese was more difficult and about 81% of instructions were spot on. In each language, a small percentage of the faulty translations were deemed clinically significant with the potential to cause harm.
The punchline: “While GT can supplement (not replace) written English instructions, machine translated instructions should include a warning about potentially inaccurate translations.” The study also noted that the authors “cautiously support its use” in a clinical encounter.
Medical interpretation is the trickiest and most nuanced form of interpretation—we see it every day. To make it more complex, interpretation (the spoken word) and translation (the written word) are different disciplines and the danger is that the public most often doesn’t recognize there is a difference, nor do most clinical personnel who use those words interchangeably. Interpreters are often painting a picture of understanding vs. doing word for word translation, because in many languages a phrase will not directly translate into the same meaning in the other language. In addition, the context or the situation in which the words are spoken matter and needs to be taken into account.
For example, the English-language phrase “my wife is going into labor” becomes “mi esposa va a laborer” in GT (pictured), which, translated back into English is more like “My wife is going into work.” The contextual meaning is lost. Or when a Spanish speaker complains of “mal de orina,” the literal interpretation of the words they are saying is “bad urine,” but in context a skilled interpreter will recognize they may be trying to say they have a UTI. Even short phrases present a huge potential for error.
“Why should I care?” is most often the next question to be asked when considering and how Google Translate or other computerized translation tools might affect outcomes. We all should care for many reasons. First off, we need to assume that a clinician reading JAMA may start using such a tool for all their interpretation and translation needs even though it hasn’t been recommended for such usage. This would lead to significantly increased risk for the patient, as well as potential physician liability. We also need to contend with the fact that no one is QA-ing or supervising tools like Google Translate in the field to measure its effectiveness as has been done in this study and therefore “miles may vary” in terms of its true accuracy and effectiveness.
The study simply tested Google Translate’s ability to handle a sentence or two of fairly clear-cut instructions. An error rate that high on straightforward text is something to be concerned about. It’s a far cry from interpreting through a full medical exam or procedure. Cloudbreak’s average provider/patient video interactions last 10 minutes, and they benefit from non-verbal cues oftentimes communicated. Furthermore, nearly 20% of our encounters are in American Sign Language for the Deaf and Hard of Hearing community, and browser-based transliteration simply can’t address that.
We get it – care teams may like the idea of using a computerized tool such as Google Translate because it’s fast is available at their fingertips, but that’s the experience that Martti users have always had.
But even more important is the quality of the interpretation. Martti is health care-specialized and it’s underpinned by interpreters who are not only specially trained and certified by CCHI, but also participating in the US health care environment rather than being based in another country. In many other nations, the idea of things like co-pays or certain advanced technology may not exist, so U.S.-based interpreters are not only adept at handling the language piece, but also speak the language of health care as its delivered here.
Now, if you’ve read this far, look back on the blog post and consider how many idiomatic English-language phrases were used. Phrases like:
lo and behold
Certified interpreters unfamiliar with these phrases would know to look them up and apply context, or perhaps ask the speaker for clarification. That’s what effective (and ultimately accurate) medical interpretation looks like.
We believe in the power of technology and its power to solve thorny problems such as building a virtual medical interpretation tool. We love Google and use it every day. That being said, the technology isn’t quite there yet to “cautiously support its use” for the root of good healthcare: communication. As a patient, would you want to be one of the statistics of the studies acceptable error-level window? As a doctor, would you want to put your medical license at risk or see one of your patients harmed?
Currently Martti offers our health care partners the ability to instantly access at the push of button Certified Medical Interpreters in U.S.-based language centers who speak over 250 languages, delivered over a reliable and scalable telemedicine infrastructure. We believe this is currently the gold standard offered in the market and remain committed to building patient and provider trust through our more than 85,000 encounters per month at over 1,000 health care venues.