I believe the argument about the merits of whether to employ machine translation has finally been won. In the 10 short years since Google kicked off machine translation for real, its evolution from SMT through to NMT has been dogged by the argument that machines cannot translate to a sufficiently high standard to become a viable solution. Translators themselves pushed back strongly rightly feeling their superb skills were being taken for granted. This line of argument pitted the machine against the human translator. It presupposed a zero-sum solution. But as I will explain later, we have already moved beyond that binary argument and the good news is – everyone’s a winner with the new translation paradigm being promoted by KantanMT.
Two major dynamics changed the world in which translation companies now compete – the Cloud and Big Data. When these two factors were married with advanced, high speed technologies and the internet it became a no-brainer that the boundaries of our translation world would become a lot wider, and the challenge that widening boundary would present was how to manage the huge volumes of work, across the range of languages that would come with it.
I have spoken in other blogs of how global companies literally look to the world as their market. And they are not wrong to do so. In 2019, online sales are predicted to hit $3.53 trillion and e-retail revenues are projected to grow to $6.54 trillion dollars in 2022 (Source: https://econsultancy.com/ ). It would foolish for any large company to ignore the realties now offered by the global market.
Much of the demand is being driven by newly emerging markets. Asia, for example, is now seen as a slowly awakening powerhouse ready to grow into a major e-market. Artificial Intelligence and machine-learning tools are allowing retailers to leverage a wealth of consumer data from these markets so that the can best position themselves to win large chunks of the sales. In order to position themselves in a market-friendly way, these companies understand that communicating in the language of the customer is essential. This is the landscape in which translation companies are now selling. Companies want multiple languages, large volumes of words and data translated and a lot of time they want it yesterday.
The only solution to this challenge is machine translation. Only MT can handle the huge volumes at the speed required. Which then leads us to the quality dilemma. And there is no doubt that MT alone cannot always provide enough quality (though at times it can – but that’s another debate). Google themselves have admitted that: “Google Says Google Translate Can’t Replace Human Translators”. (Source: https://www.propublica.org/) This quality argument has been accepted by most, if not all MT companies, to ignore it would be myopic. So, the debate is now underway as to how we “humanise and soften” MT output. Of course, post-editing is a service that is already offered and has its rightful place within the L10n workflow. However, a new paradigm has emerged aimed at infusing MT with a more human quality translation output.
Just last month KantanMT.com launched its new product, which is called KantanSkynet (see my last blog https://kantanmtblog.com/2019/08/15/kantanKantanSkynet-to-rain-down-opportunities-for-linguists-worldwide/). The product is described as a “Human-Powered Machine Translation Platform”. KantanSkynet is a global crowd-sourcing translation platform which seeks to integrate human translation skills with Kantan’s neutral machine translation technology. By combining the two, KantanSkynet aims through an ongoing, relentless process to build and augment a human translation quality into the KantanMT NMT engines.
KantanMT launch a very successful recruitment process beginning last August to recruit language experts in all corners of the globe. This highly qualified panel will form the community of language experts who will be paid to undertake the human translation of texts deemed of low quality by the NMT. Once translated by the human translator the texts are fed back in to the NMT system and the quality of the engine is improved accordingly.
So today, instead of sitting in an office in Dublin, London or Amsterdam a translator can be virtually anywhere and still be available for work on KantanSkynet. Ironically, while in-house translation technology has become sophisticated, translators have reverted to working in something of a 19th century cottage industry work model. That is the model that suits so many of them. Most translators choose to work from home. In today’s technological era they have the connectivity they need to work from almost any place and at any time. With today’s seemingly ubiquitous WIFI, no translator is restricted to where and when the translate.
And this is the online community that is now working with KantanMT to help humanise and soften NMT output. It is the essence of KantanSkynet to combine the speed and low cost of machine translation with a layer of human expertise powered by a “global community” of professional translators. So, using this model, KantanMT can offer the scale needed to translate huge volumes of data to a higher quality standard. And working through its KantanSkynet crowd-sourcing solution the native quality translations that many customers require is now an integral feature of KantanMT’s workflow.
Over time, the quality of translations will grow exponentially. This symbiotic solution of human translators working in tandem with high tech machinery is surely the paradigm that will be followed by many others long into the future of our industry.
Aidan Collins is Marketing Manager at KantanMT