Chapter 3 - What Has Machine Translation “Mis-Translated” about COVID-19? What “Mistakes” Can Tell Us about Humanity that Machines Cannot

Summary

Abstract

This chapter examines the performance of a Neural Machine Translation (NMT) model in the translations of Chinese language journal articles on COVID-19. The limitations of existing automatic evaluation metrics for translation quality are identified, and both House's (2015) refined translation quality assessment model and the Multidimensional Quality Metric model (Lommel et al., 2014) are applied to identify semantic and pragmatic errors. The authors argue that automatic evaluation metrics would benefit from the introduction of cultural, emotional, or ideological elements as these are necessary to understand the discursive, pragmatic, and social communication rendered by humans. Rather than relying solely on automatic evaluation metrics, a consolidated model of human-intervened evaluation should be applied to pinpoint specific translation problems produced by machine translation.

Keywords: Neural Machine Translation, automatic evaluation metrics, Juliane House's translation quality assessment model, multidimensional quality metric model, machine translation

Introduction

For some time, human beings have used computers to carry out fully or partially automatic processes to translate texts from one language to another. Machine translation (MT) has evolved through different phases since the 1950s from ruleand dictionary-based approaches through to the statistical approach and, finally, to the neural-network-based approach. Whereas the early development of MT endeavoured to mimic human cognition, advancements in machine learning and data science have moved the later development of MT gradually towards mass industrial translation.

Shaken by the outbreak of COVID-19 in late 2019, demands for information have grown exponentially. One of the most significant endeavours of healthcare practitioners, scientists, and researchers in combating this global pandemic is the sharing of information. Faced with the threat of illness and the disruption of our daily lives, we all endeavour to seek remedies, be they medical, social, or economic. Making information accessible across different cultural contexts is essential, as it could constitute an effective approach to fighting COVID-19 and its second- and third-order effects.

During the COVID-19 pandemic, scientific papers and preprints in various research areas, such as epidemiology, biomedical engineering, environmental science, and socioeconomics flooded into academic venues to flesh out our understanding of coronavirus. This could also be seen from the World Health Organization's (WHO) website, which has constantly been publishing COVID-related information (see https://search.bvsalud.org/global-literature-on-novel-coronavirus-2019-ncov/).