Be wary of handing over translation to a machine
The following comment appeared in the Financial Times (4 December 2016):
Sir, Microsoft’s Bing search engine has, you report, been “crawling the web” to feed masses of language data into its machine translation algorithms (“Targeting the ‘next big thing’ ”, The Big Read, November 29). But as any reasonably proficient linguist knows, bad translations are just as rife on the internet as fake news, and powerful algorithms are liable to propagate them in exactly the same uncritical way. The risk here is of a feedback loop: as bad translations become embedded in machine translation software, they are fed back on to the web and picked up once more by the algorithm. Before we know where we are, the bad translation will be taken as good, just through the sheer weight of big data supporting it. Like it or not, the complexity of today’s world means that clear, accurate communication across language barriers is arguably more important than ever. Translation matters, and as with anything else that matters, we should be wary of handing it over to machines on the assumption they can do it better than humans.
Stephen Reynolds Financial Translator, Penzance, Cornwall, UK
