Lost in translation: how automated machine translation can address the shortage of medical interpreters
Research in Progress
Background: Healthcare providers typically
use live interpreters to communicate with patients who either do not speak
English or who have limited English proficiency. With the Affordable Care Act
(ACA) encouraging uninsured patients to obtain health insurance, the anticipated
demand for language interpretation will increase, as the proportion of LEP
patients for Medi-Cal and Healthy Families will increase to 41% by
2019. However, a shortage of interpreters exists because
medical interpretation incurs a typical $30 to $50 per hour overhead cost, there is a 35% decrease in earning potential for
medical interpreters versus government or scientific and technical
and using live interpreters increases doctor-patient encounter times by
57% versus phone interpreters. The quality of live
interpretation can also vary as there are legally and professionally no
mandated requirements for certification or quality control.
A possible solution is the use of automated digital interpretation of clinical encounters, but there are no descriptions of such systems in the medical literature, though some public health agencies and clinicians may already be employing such tools.
Methods: We constructed a prototype of translation software that runs on Google Glass, using Nuance NDEV speech recognition technology to convert speech to text, and Google Translate for converting text in other languages into English. We additionally tested Google Translate’s ability to convert Spanish-language speech to another language on Android tablets using sample patient-doctor interview scripts.
Results: Our initial acceptability trials demonstrate that these tools are feasible for use in clinical settings. The next stage is to test their accuracy.
Conclusion: Using speech recognition and machine translation is feasible. Accuracy trials to determine errors of addition, omission, substitution, and false fluency are essential, as is usability comparison with live interpreters.