3 years ago
October 19, 2015
The Wall Street Journal on How Etsy Tackles Machine Translation’s “Eternal Problem”
As the language services market enjoys ongoing growth, its various niches are also seeing a lot of progress. In fact, separate research reports indicate that machine translation (MT), will continue to grow at a compound annual growth rate of 23.53% through to 2019, and that by 2022 it will have carved out a nearly $1 billion niche within the language industry.
With a market of this size, MT has again captured the attention of mainstream media. The Wall Street Journal published an article on MT used by ecommerce companies like buy-and-sell portal Etsy: Businesses Try to Fix Machine Translation. As made apparent by the title, WSJ’s coverage focused on rules-based and statistical MT solutions used to localize ecommerce platforms by automatically translating their content into different languages, and, of course, how they cause a few problems. The article mentions mild problems with translation, such as when Etsy’s MT solution translates “clutch bags” into automotive parts, and also highlights more severe ones, such as misunderstandings that required the company to intervene and even refund purchases.
The rise of ecommerce does indeed drive more demand for translation and localization, and as more ecommerce sites use MT to reach global audiences, it naturally follows that they would want to improve its accuracy and efficiency. What the WSJ article touched upon were typical stumbling blocks that have always held MT back from being fully reliable sans human post-editing. There are, however, related but independent fields that can help bolster MT; fields such as natural language processing (NLP). Already widely used (and growing) in sectors like life sciences and healthcare, NLP is making strides towards better ways to process natural language for machines to understand.
Yet MT, like the larger language services market it belongs to, is still extremely fragmented with various development efforts that do not necessarily intersect. Commercial enterprises have their own MT solutions constantly under development: Alphabet Inc. has Google Translate and IBM has IBM Watson. Even the EU is trying its hand, with the giant Horizon 2020 Research initiative funding development of MT solutions. The EU Patent Office, in fact, is so confident that MT will reach an acceptably high level of accuracy for their needs that it will completely take over patent translations by 2028. And then there are established commercial players like Systran and start-ups like KantanMT that are gaining ground.
Indeed, transcreation is where MT falls short. Too bad, considering some highly profitable and cutting-edge spaces that translation and localization plays a part in like mobile apps and games need a lot of transcreation.
Evidently, even if Etsy hones its MT to the point that it no longer translates “cross-body bags” into something morbidly different in another language, it seems MT will still have a ways to go.