Exploiting a multilingual semantic Machine translation architecture for knowledge representation of patient information for Covid-19

Abstract

AwezaMed Covid-19 is a multilingual speech-to-speech translation application for screening patients for Covid-19. It enables Englishspeaking health care providers (HCPs) to conduct screenings by asking questions to patients in all other official languages of South-Africa. It uses a multimodal computational grammar translation system to enable English speech and screen-based input, which can be translated to produce synthetic speech in the target languages. Grammatical Framework is used for the translation system, utilising a semantic interlingua. Because of this, each utterance translated by the application is represented by a semantic tree, which could be exploited for knowledge representation. In this paper, we describe how the machine translation architecture designed for multilingual speech-to-speech translation can be adapted for knowledge representation consistent with existing knowledge representation formalisms, namely openEHR archetypes and RDF triples, which could be recorded seamlessly by HCPs during the screening.