There are many ways we are seeing AI used today in our customer and partner ecosphere. Localization professionals will be familiar with some of the basic AI building blocks and current applications, but there may be new ideas for AI paths you may want to pursue in the future, whether for your current job or a future one. After all, “AI is going to have as big an impact on our society as electricity,” according to Risto Siilasmaa, chairman of the board at Nokia.
In Figure 1, we show some of the most commonly used AI-related terms and how they fit together. Computer science is depicted as the general study of computers and computing concepts. Although the exact definition of AI is a matter of debate, for this discussion let’s define AI as systems that can perform complex tasks autonomously and adaptively; that is, without constant guidance by a user and with the ability to improve performance by learning from experience. Machine learning is a process by which systems improve their performance in a given task with more and more experience or data. Deep learning is a class of machine learning algorithms that uses multiple layers to calculate high-level outputs from raw inputs. With the increasing power of modern computers, more and more layers are added for greater complexity.
Neural machine translation (NMT) is an AI application that uses machine learning of the deep learning type to perform automatic translation tasks. NMT is the example of AI that most localization professionals are aware of today. Its ROI is clear, its adoption is growing rapidly, and it is a safe and helpful enterprise tool as long as it is deployed securely.
Now that we are clear on the vocabulary, let’s see how AI enables a machine-first approach that can have real human-first impact. Today, AI is already breaking down language barriers that stand in the way of communication and collaboration, eCommerce, customer support and ediscovery. How? Let’s look at some real-life AI stories.
One of the most well-known US designers and producers of semiconductors and integrated circuits had a problem. With an increasingly global workforce, key research and development (R&D) teams had been located overseas, speaking different languages but still needing to communicate and collaborate closely with colleagues back in the US. Some of the overseas workers, while comfortable collaborating with their on-site colleagues, had trouble in written communication exchanges with the US teams, and vice versa. It became clear that translation would be helpful in ensuring both sides fully understood each other. The company already had a translation management system in place and a solid translation supply chain. But of course, the speed of putting text through a traditional translation workflow was not conducive to real-time collaboration, and the volume of communications made a traditional approach cost prohibitive. Real-time machine translation was a natural solution.
This was successful from the start, but even more so today, now that AI-based NMT is available. Using deep layers of nodes between input and output layers, NMT achieves a greater amount of accuracy and fluidity than previous generations of translation systems. Deployed within a system that keeps all the translation queries and results within the company’s firewall, NMT has enabled this manufacturer’s research teams to communicate accurately and instantaneously, increasing their R&D productivity while protecting their intellectual property.
NMT is making its way into many other domains. The eCommerce space is fiercely competitive, and time is of supreme value here. One of the largest online travel companies in the world determined that a key part of their competitive advantage was in instantaneously adapting incoming and updated hotel property descriptions into other languages, at low cost and very high volume, using NMT. Same for a household name in online retail, which, in addition to ensuring product descriptions are updated in real time, also uses NMT to enable their very large global workforce to react quickly to changing corporate guidelines. They find the best way to do that is to send instructions in the employees’ own language.
The fluidity of NMT is especially helpful when humans are trying to have a live conversation. Many companies are exploring AI chatbots as a way to simulate a human conversation in support of conducting day-to-day business. One of the most well-known US banks is deploying a sophisticated multilingual chatbot system to ensure that non-English speaking customers fully understand information they are receiving. This is very important to them not only from a customer satisfaction point of view, but also for legal protection, to avoid accidentally landing in a fraudulent situation.
Also in the legal domain, NMT is making a huge difference in ease of ediscovery. Many government and corporate organizations are faced with an insurmountable volume of incoming multilingual data that would be too slow and cost-prohibitive to translate with traditional methods. For government agencies, secure NMT (not indexed by internet search engines) is critical to international diplomacy and security.
But as we know, NMT is just one example of AI applications that are currently in use today, both commercially and experimentally. Figure 2 shows other applications of the technology.
Credit card fraud detection, self-driving vehicles, personalized web content and facial recognition are all in the news today. They catch our attention as human beings anticipating, often apprehensively, a highly automated, unknown future. But for localization professionals, in our day-to-day work, are there other applications of AI beyond NMT that are actually real and emerging in practical use today?
Before we get to that, let’s leap forward into the localization future that we all can imagine, but that is not quite here yet on a practical level. In the same way that AI has drastically changed the world for translators, the collective localization-sphere expects to see AI transform the world of the localization project manager. There are many problems in the localization project world that would be aided by faster and more accurate, less human-error prone decision making or status monitoring. Autonomous and adaptive methods of assigning language service providers and assigning individual translators, reviewers or other project resources are all easily imaginable. So is predicting delivery dates and quality, and predicting issues in those areas for specific jobs, at a high volume with low human touch. Adaptive and autonomous job routing to maximize quality and on-time delivery, while minimizing cost, is a clear and desirable application from the enterprise perspective. Localization project managers’ jobs will be enhanced, as they gain the ability to handle job volumes that today would be overwhelming. Even one level up from the actual project management, at the data aggregation and reporting level, one can imagine this becoming highly automated and adaptive, and low-human touch (Figure 3). With human efforts augmented by machine automation, project managers will have more time for localization strategy definition, project exception management and tools strategy.
Whether it’s translation output, translator performance data or project management data, one thing is certain: your data holds the secret to your success. Companies need to think carefully about deliberate sharing of this data and be extremely wary of inadvertent sharing via nonsecure tools.
Coming back to actual emerging applications, an insight has already started to alter the work of many localization professionals: that translation is just a step in the overall content process of creating, transforming and distributing content. In order to deal with the exponentially increasing demands for producing more content — continuously, faster and more personalized — there is no place for friction in the content supply chain, including during global transformation (Figure 4).
As part of surviving in an accelerated content supply chain, localization professionals need to be aware that AI is also presenting itself in applications that today are able to streamline content creation, management and distribution. For example, the very same linguistic building blocks that are used to create NMT are also helping transform the professional lives of content creators upstream.
As an example, a well-known worldwide entertainment company and long-time customer for both web content management and translation management was traditionally known as a master communicator always on top of their global marketing game. Yet their digital marketing team was starting to feel crushed by the weight of continual demands to create ever-increasing variations of content for social media postings, blogs, website blurbs and other publicity channels. The required content was no longer creative — it was just a reformulation of existing content. Human workers who had previously enjoyed the work of content transformation were simply becoming numbed by the sheer volume of transformation required.
It turns out this is a perfect application for AI, given the existence of the above building blocks. An original longer piece of content can now be analyzed, summarized and transformed into “snackable content” that is consumed in shorter formats such as tweets or LinkedIn postings, but is all part of the same messaging.
AI presents the content creator with the available AI transformation result and provides tools for the human to adjust the AI output, automatically calling out aspects the human might want to highlight. AI permits a much faster reaction to personalization requirements. Effectively supporting a web content personalization process has resulted in marketers achieving, on average, a 19% uplift in sales. This is just one example of how linguistic AI can influence website functionality. Search and predictive text are other well-known areas.
For localizers who support marketing or documentation teams, the accelerated creation of short-form content variations will make implementation of continuous localization and use of NMT even more important.
We’ve seen that linguistic AI can serve both localization and content management purposes, with human-plus-machine being able to accomplish more and faster, together. There are several notable examples of localization leaders who have expanded into content management, as the two fields are have a common touchpoint in words. Linguistic AI is extremely effective in both domains — already used in current applications such as NMT and emerging in content creation assistance. Being familiar with current and emerging AI applications; thinking ahead to what types of problems lend themselves to autonomous and adaptive solutions; and protecting company data that is the lifeblood of AI can help localization professionals expand beyond their current horizons and increase their contribution to their company’s success in the exciting and challenging times to come.