Focus
Automation in the Age of Remote Work
Jessica Roland
A strategic account director at SDL since 2014, Jessica Roland works with selected enterprise customers to help them reach global audiences and enhance customer experience, increasingly via AI. Prior roles include leading enterprise software globalization teams and international product management.
Jessica Roland
A strategic account director at SDL since 2014, Jessica Roland works with selected enterprise customers to help them reach global audiences and enhance customer experience, increasingly via AI. Prior roles include leading enterprise software globalization teams and international product management.
emember what it used to be like to get all your work done in an eight-hour workday? Starting with the advent of the internet, where an immediate response first became possible, followed by the exponential growth in content, our formally normal workdays are long gone. Even the pandemic has not slowed this trend. COVID-19 work-from-home has eliminated commuting for many, but the informal consensus is that we’re not working any less. Au contraire, hours have lengthened for many. According to Bloomberg News, stay-at-home workers in the United States are spending three more hours per day on the job than before the pandemic, with some other countries not lagging far behind.
Workers are also attending a plethora of virtual conferences and other webinars designed to compensate for the fact that we’re not meeting in person these days. The communications dial has been turned up, not down, during the current crisis. There’s a continual, ever-faster flow of data (Figure 1) and tasks, to the point where it’s now impossible to keep up without assistance.
Figure 1: Annual size of the global datasphere. Adapted from Data Age 2025, published November 2018.
That assistance used to be human, but with the increasing sophistication of available technology, software programs running on a variety of interfaces have taken over the “assistance” role. Our software assistants have become increasingly intelligent — to the point where they are able to anticipate needs, not just respond to commands.
The Harvard Business Review identifies three ways artificial intelligence can help take on these routine tasks from a real-world perspective, where software is learning to act like humans. The first is process automation at a level of sophistication where the software acts like a human inputting and consuming information from multiple IT systems. The second is cognitive insight — detecting patterns and determining their meaning. The third is cognitive engagement, engaging individuals using natural language processing.
In the global content business, whether creating the original content or offering it in multiple languages, the trends are clear. Traditionally, there have been many routine, monotonous, and rather uninteresting tasks associated with the production of words for commercial purposes and with their translation and adaptation to local languages and cultures. Today, more and more of these routine tasks can be performed by software assistants, leaving writers, translators, and reviewers free to focus on refining texts and tools and on gaining more understanding of future communication needs. Software assistants not only help us humans keep up with the faster pace of work, but also allow us to spend our precious time on the most valuable tasks. That’s an important goal, for as we all realize sooner or later, human time is clearly finite, and every moment spent doing one thing represents a choice not to spend our human creative powers and insights on other activities — a significant opportunity cost.
Let’s look at a few examples of human plus machine efficiencies that are saving valuable time today in the language industry, and which could make our collective 2021 more fruitful and fun (and don’t we all need that these days?).
Content creators
Any business writer will tell you that there’s a lot of pleasure in creating content the first time around. It’s fun to think of the concepts you need to convey, to organize them so that they do the best job of persuading or educating, and to feel your words falling into elegant placement on the virtual page. Less fun may be having to revisit the same material — to do updates or to repurpose it for multiple publications or different channels. For websites, help came along in the form of web content management systems that allow site managers to see every place across a site where a given version of a piece of content is used, and to update it easily, consistently, and simultaneously. Same for technical publications managers, who now have DITA-based component content management systems whose raison d’être is to enable efficient content reuse and updating. This year, with the increasing trend of buyers to consume a blend of commercial and technical information as part of their customer journey, we began to see content management systems that “mashup” (blend) these purposes, achieving a single source of truth for all content. Blending these systems saves content creators from having to reinvent the wheel over and over again… and also to avoid inadvertently creating conflicting information across sources (Figure 2).
Figure 2: Parent publications and different channels. Source: SDL.
In 2020, we also saw the early stages of a technological race to enable “writing assistants.” Now, writing assistance is not a new concept: you can think of mid-century dictation machines as writing assistants. Over the years, that concept morphed into speech-to-text automation, which, combined with translation technology, has enabled both live and virtual conference attendees to better digest event content and more quickly have a record of it. That technology assists with writing in the sense of capture and transmission, but it does not generate content. In recent years, we’ve started to see AI-driven technology that can automatically derive summaries, abstracts, and social media blurbs when given a full piece of content.
AI-generated writing continues to make startling progress. This year’s news of OpenAI GPT 3 technology being able to actually create fluid content from scratch has created a lot of buzz and speculation about future uses. In some cases, this could be quite problematic; for example, in the wrong hands, it could be used to quickly and widely spread disinformation. At the very least, if it becomes common for software to actually generate content, it will likely mean even vaster amounts of content to consume and make sense of. Then, inevitably, we will delegate that task to software as well! In any case, the “state of content” is rapidly changing, and we will continually need to adapt.
Translators
Translators in general are most enthusiastic when they’re able to bring their creativity to bear. For example, in transcreation, target language content is wholly or partly created from scratch in order to resonate locally, rather than being directly and exactly translated. Transcreation remains one of the really fun and beautiful areas of the business.
Direct translation of commercial or technical content can often seem dry and repetitive. Translation memory and machine translation technologies evolved and were eventually accepted, not only because they allowed managers to drive more output faster, but also because translators themselves came to happily depend upon leveraging previous translations and receiving suggestions on translations they know are consistent with past use. This is especially true for commercial and technical content that is ephemeral and purpose-driven. Post-editing a machine-generated translation may not only faster, but less stultifying and repetitive because now, with AI-driven neural machine translation that takes a machine-first, human-optimized approach, the quality of MT is continually improving.
Translators can observe, and marvel at, these continual improvements, and also help make them happen, by participating actively in MT training and testing. In fact, more and more, training MT seems to be the sweet spot for the translator of the future. It will require translators to have more of a computational linguistics background than ever before, but it will be interesting — and fun. In fact, one could look at it as having a partner in translation, only the partner is a machine and requires some on-boarding to be a helpful part of the team. Translation provider companies should be doing all they can to enable translators to adapt to this shift in translation skill set.
Managers
Some of the most routine tasks for localization project and program managers, whether they’re working on vendor-side or client-side localization teams, have to do with:
- Getting files transferred to and from translators
- Answering translator questions, both linguistic and logistical
- Monitoring and enforcing schedules, milestones, and budgets
Managing the time-cost-quality triangle is their core role, and parts of it have, in the past, been excruciatingly time-consuming and routine. Over the past years, automation such as workflow-based translation management systems and integration with content management systems have made it much easier to move files and jobs along through the different stages of translation on time. Their functionality has also been enriched over the years to include managing business tasks such as quoting projects, processing approvals, and invoicing. Online query systems have helped ensure project managers do not have to answer the same questions over and over for different translators. Routine tasks have thus been eased by software assistants, and this has been essential as translation volumes have grown to super-human levels. Ultimately, these developments have saved project teams enormous amounts of time.
Now we are seeing a further transformation in this area via the power of data, as TMS systems offer continuous, real-time monitoring of various project parameters. This allows project managers to catch problems as they occur. But in today’s AI-emergent world, catching issues quickly is simply not enough. Preventing problems in the first place is even better, and it can be enabled by predictive analytics. Predictive analytics and preemptive actions are the sweet spot of localization projects of the future. The future is starting to be visible right now; for example, at SDL we see this internally, as assigning the best translator resources and translation method at a given point in time is transitioning from human to AI-powered work.
Globalization executives
Similarly, in a level up from day-to-day localization project management, globalization leaders have in recent years been pushing for data analysts to join their teams. Their purpose in doing so is to nip problems in the bud or prevent them before they occur, as well as show the business value gained from their teams’ efforts. At this level, it’s key to be able to see summarized data across projects, in order to derive more systemic insights.
Getting budget for, and hiring, such data analysts has been a major challenge for globalization leaders. The alternative — doing the analysis themselves — is often impossible, due to lack of time or skill set. When they do have the knowledge to do some of it, it’s generally at the cost of other priorities… or sleep! And, even though the insights are exciting, the actual task of identifying, assembling, cleaning, and processing data is excruciating work for a globalization leader with relationships up and down the organization to monitor.
Here again, intelligent assistants are coming to the rescue. TMS and MT dashboards are now enabling this important macro view. TMS application program interfaces (APIs) also allow for connection to business insight programs like Sisense to enable calculation of translation profit margins and ROI. API connections to content management systems enable tying language to personalization efforts and to AI-based language optimization tools like Acrolinx. Systematic capture and automated sentiment analysis of online comments or survey data regarding source or translated content are also ways of catching issues early on or gathering positive data for ROI discussions. These connections are all moving from only summarizing past data to also including predicting the future. Human efforts plus machine efforts are providing ever-more valuable insights and control of the business, with far less effort and much more fun.
Translation quality and review
No doubt about it, translation testing and review can be highly repetitive. Review in particular can be perceived as burdensome, because these days it’s increasingly being left to subject matter experts (SMEs). SMEs have their own day jobs to think of first and are often lending a hand to review in their “free” time. Lightening the reviewer’s burden starts with easy to use tools for reviewing material in context (whether source content or translation) and capturing, sharing, and resolving review comments in one interface.
It’s also imperative that they be given the highest quality content to start with — content that has already been thoroughly checked for routine errors — so that the reviewer can focus on the truly value-add aspects. In recent years, automated “assistants” have helped companies handle the increased volumes of testing and review that result from the exponential growth of content. Translators can run a battery of automated quality checks so that basic spelling, grammar, and other syntax errors are caught before review ever starts.
If translators haven’t run automated quality checks on their desktop, server-based translation management systems can do it before the translation is accepted. Upstream from translation, source content can be optimized automatically, whether in batch mode after it’s written, or real-time during the writing process — preventing errors from occurring in the first place and ensuring consistency.
Where AI comes in is with the massive amounts of data that go into training these tools in order to produce cognitive insights that can guide writers and translators. We can anticipate that, as data is collected from writers during the authoring process, it could be used predictively to help determine where training or focused automated checks could be used upstream. This would be key, because the further upstream that errors are caught or prevented, the less costly it is for the organization.
We’ve gone over some of the ways that automation has helped take the burden of repetitive tasks away from humans in the content creation and translation process, making room for higher-value contribution. Most of the automation examples were in the process class, with a few in the cognitive insights class. The emergence of cognitive engagement can most commonly be seen in chatbot applications, where the goal is to make the interaction seem as human as possible. But chatbot use is currently focused on serving consumers and has not yet found a place in helping writers and translators do their jobs.
We’ve covered a lot of ground on changing roles, so let’s recap (Figure 3).
| Role | Areas for intelligent automation | Frees up time for |
|---|---|---|
| Creator | Content management Derivative content creation |
Research Content strategy Creative content Innovative content |
| Translator | Translation memory Machine translation |
Transcreation Training automated tools |
| Project manager | Workflow management Resource management Schedule management Cost management |
Process innovation Research into translation needs |
| Executive | Forecasting Budget management Performance monitoring |
Building stakeholder relationships Developing strategy Communications |
| Reviewer | Quality assessment Collaboration workflow |
Testing strategy Continuous improvement strategy |
Figure 3: Areas for intelligent automation and the roles that go with them.
Of course, there is a growing line of sustainability thinking that questions whether un-
bridled growth in anything, including content, is inevitable or even desirable. But that’s another, longer-term conversation. For the time being, there is no end in sight to exponential content growth. Given that, without intelligent automation, tasks such as those in the middle column above will quickly consume all available time, and workers will be increasingly overwhelmed. The emergence of intelligent tools for process automation and for content insights will enable us to leave repetitive and routine tasks to machines and focus our human time and energy on research, strategy, planning, and optimization, using our full capacity for creative thinking and driving solutions. We’ve seen an acceleration this year in that transition, driven by remarkable advances in linguistic AI. Rather
than feel threatened, we can rejoice that these advances will help carve out the mental space and freedom to focus on the most interesting end of the work spectrum. Human plus machine equals more capacity, but also more fun for all of us in the business of words. 
