You’ve met Hai before. Hai is SDL’s personification of SDL Linguistic AI™, our unique technology foundation that fuels our software portfolio and is both the foundation of and a product of our ground-breaking neural machine translation development efforts. SDL Linguistic AI applies deep learning and other advanced techniques to content challenges that prevent organizations from going omnimarket.
One of the biggest drawbacks to implementing a machine-first approach is a perceived lack of quality – a machine can’t “talk” like a human. Not now. And maybe not ever. Of course that doesn’t mean that we abandon hope --- and the revenue that a global customer base can deliver -- of reaching that global customer base with multi-lingual content at the scale and speed a modern business demands.
Recently, SDL Machine Translation was recognized with an “AI Breakthrough” award. This award recognizes innovators in AI across a variety of disciplines. We weren’t surprised – though of course we are absolutely thrilled – to receive this level of recognition. After all, SDL had broken the Russian language barrier just last year and achieved an industry record with over 90% of the system’s output rated “perfect” by professional Russian-English translators.
However, no innovator rests on its laurels. We know that we need to continue to push the boundaries and that we need to help our customers realize their omnimarket aspirations faster because the ability to capture a global market depends on the ability to speak its language – and do it across all channels.
SDL Machine Translation encapsulates our deep expertise in this space into an enterprise-grade solution that breaks language barriers across content types, within various channels, and within any number of content-intensive business processes. When we put forth a vision of a machine-first approach to achieving omnimarket capabilities, it is our machine translation capabilities that lead the way.
As we enter the Intelligent Translation Era, we are trying to understand the role of human effort in a process that has to be machine-first if it is to succeed at all. In the latest release of SDL Machine translation, we give our customers the power over AI. It is the essence of Hai: Human + Artificial Intelligence yielding results that are greater than either can achieve alone.
William Shakespeare wrote that “There is nothing either good or bad, but thinking makes it so”. What a perfect statement about language from someone that mastered the art. When information is in front of us, we take it at face value. We think about what the communication is saying to us and we may or may not consider the language that is being used. Machines can easily translate short phrases and commands and humans won’t be able to tell the difference. Stop is stop. Go is go. Where is ______ is also quite easy and a machine will get it just as “right” as a human.
Unfortunately, the content that we want to put in front of our customers is more complicated than simple instructions and basic phrases. There are nuances. There is organizational jargon. There are likely many different ways to express concepts and there may be instances where machines and humans may not come up with the same answer – at least not right away.
We tend to see quality as an inherent trait of a product or service. That may be true if quality is measured using an explicit metric. For example, when buying linens, thread count is used to measure quality. It’s fairly straightforward. You count the threads and if they are over a certain threshold, you know you have a quality product. It’s concrete and it’s either there – or it’s not.
This isn’t the case for AI-based products. Quality is taught. Machines are trained. When we hear the term “machine learning” – we think of unsupervised processes where machines get better without any human intervention. That’s only partially true. Machines learn from data and humans have to either kick-start the process or monitor the process such that bias isn’t introduced into the system. At the very least, humans are there to assess whether the machine “got it right”.
And this brings us to the latest SDL Machine Translation Innovation – Adaptable Language Pairs. In the latest release of SDL Machine Translation, we introduced a way for our customers to adapt neural language pairs to their own content – without ever compromising the security and confidentiality of their corporate IP. Within our flagship product, SDL Enterprise Translation Server, we have introduced a training facility that anyone can use to adapt a standard language pair to their own domain, project, process, customer bases, etc. And this is something that can be done continuously because we recognize that learning isn’t a one-time event for anyone – including a machine.
This new feature adds to the arsenal of customization options SDL provides and is complemented by our wealth of expertise across a variety of content disciplines. We are hosting a webcast on July 11th where you can hear more about SDL Machine Translation, including the new release, and we hope you can join us to learn more.