Teach-Back

I am Ethiopian, I speak Afaan Oromo and Amharic, and for most of my life I have been the one who goes along to the clinic or the immigration office to interpret. Errors there are invisible to everyone but me: a dropped "not", a dose off by ten, a date from the Ethiopian calendar written as if it were Gregorian. Teach-Back checks the interpreter in both directions. When the patient speaks, two Meta models, Omnilingual ASR and Muse Spark, hear the Amharic and code compares what they heard. When the provider speaks, Teach-Back records the English, then hears the interpreter's Amharic through the same listeners and compares, fact by fact, what the provider said with what the patient was told. SAM 3.1 reads the medicine the patient brought. A human confirms each flag. It advises no one.