Human Translators Pitted Against Machines in New Language Study from Quill
Although conventional wisdom says AI is coming for our jobs, new research by Quill, the leader in Primary Content production for ecommerce, has revealed that machines still have a long way to go before they can rival their human counterparts in translating content.
Pitting human translators against machine translations, respondents were asked to review the product descriptions of items being sold online side-by-side, without knowledge of their human or machine translation status. They were then asked which description made them more likely to buy a product, and which description left them with the best impression of the retailer.
For both these metrics, the human translations dominated across all markets and verticals. Four in five people (79%) stated that the human-translated copy made them more likely to buy the product in question, while 80% of respondents agreed that the human-translated descriptions gave them a significantly more positive impression of the retailer.
Ed Bussey, CEO at Quill commented on the research, saying, “Recent innovations in neural machine translation have made a material difference in improving the accuracy of translated content – enabling the technology to provide an output that is, according to 50% of consumers in our research, functionally as easy as a human translation to understand. This is a big step forward, but – and it’s a significant “but” – the technology still has a long way to go before it replaces hand-crafted content. Whilst consumers can understand pure machine translations, they don’t find them as persuasive, convincing or brand-elevating as human-translations. People with an understanding of culture, locale and colloquialisms are still required to write trustworthy and inspiring product content that achieves ecommerce goals and accurately replicates a brand’s tone of voice. And that’s before we even consider the limitations machines face in optimising content for search engines or avoiding SEO-damaging content duplication when different retailers are selling the same or similar products.”
Bussey continued, “Google has recently claimed that it can reduce errors in translation by 60%, but that means a staggering 40% of errors remain unresolved. Whether due to legal requirements around miss-selling or customer demand, retailers would never accept their in-store product labels having any errors. Equally online product information needs to be 100% accurate. I don’t think translators need to worry about machines taking their jobs for decades to come as relying too heavily on machine translations at this early stage of the technology risks directly damaging a retailer’s bottom line. Instead, brands should look to adopt a ‘humans in the loop’ approach, utilising machine translation technology to speed up and streamline processes, while ensuring that a skilled human editor reviews and amends the content to ensure it is unique, properly reflects the brand, resonates with the consumer and delivers results.”