Rapidly developing markets of IT, video games, e-commerce, mobile applications, and others call for more flexibility in their processes. As more software teams transitioned to agile cycles, localization needed to “keep up,” and a couple of years ago a trend of agile localization emerged.
The agile localization approach is unlikely to disappear in the foreseeable future – since small localization (l10n) projects in the fast production cycle become more and more widespread. Clients are no longer willing to pay minimal batch fees for the execution of those tiny daily requests. And language service providers (LSPs) are trying to provide faster and smoother services using online l10n tools, such as Crowdin, Smartling, XTM, Onyx plugin for Google Chrome, and many others.
But as l10n buyers confirm, not all suppliers, however mature they may be, are really able to quickly process small requests. And while the pain of a localizer who is tasked with an out-of-context request of only one word or several UI elements is understandable, one can also agree with the perspective of the buyer side – that a turnaround time of a day or more for these jobs is somewhat inadequate.
Even IT giants are still trying to implement the best strategies for managing translation projects with a myriad of minor tasks. A potential solution lies in the concept of continuous localization. With this approach, the work of the localization manager decreases, as everything happens automatically. A continuous l10n tool:
However, in practice, this kind of “magic” does not always happen.
According to a number of LSPs, “continuous l10n” is already in use. But with that, quality suffers due to a lack of awareness: developers and customer project managers are unaware of the legwork translators have to do in order to make sense of isolated UI strings with no reference or clarification.
A potential reason for this is that some of the automated l10n platforms which call themselves continuous, are initially built on top of concepts offered by translation management systems (TMS) and CAT platforms. While there are evangelists of a continuous l10n approach who don’t agree that TMS, CAT, and QA functionalities are all necessary for processing small jobs, there also exists a counterargument, which comes directly from those who have to execute these tasks. Their point of view: however small the job, we need not only context, but a glossary, a translation memory, a workflow, a query system, and a solid QA system in order to guarantee quality output.
Not all current tech representatives of continuous l10n have all of those features. As a result, their solutions are paired with 3rd party tech companies when implemented on the buyer side. But does it really speak in favor of flexibility? What if a buyer wants to change the CAT tool? What if they would like to use MT engines which are not yet available in this specific CAT (for instance, those which have been developed or trained in-house)? And will they be able to apply pseudo-localization? These and many other questions leave potential users doubting whether current solutions are as continuous as advertised.
To help you navigate your way around the tools available, we will be looking at Serge (open source) and Transifex, two tools that have already proven to be viable solutions for l10n buyers. This time we will not talk about Smartling or Crowdin.
SERGE stands for “String Extraction and Resource Generation Engine.” This command-line utility gathers new source content, sends it out for translation, acquires ready translations, and integrates them back into the product. It pulls and pushes changes and will also synchronize with an external CAT tool of your choice. Equipped with Smartcat as a frontend translation tool, Serge is used by Xsolla, a global video game distributor and publisher.
Evernote, from where Serge originates, uses a CAT tool named Zing, which is a forked version of Pootle. With Serge+Zing, localized resource files have the same structure as the source ones. Serge also handles cases where source files change while their older versions are still in translation. This spares engineers from merging in changes and unblocks the development cycle.
Evernote wanted to simplify the l10n process as much as possible. That’s why the Serge+Zing pair doesn’t feature a regular l10n workflow functionality.
Speaking of workflows, from a translator’s point of view, proper content adaptation for different languages should become an essential part already in the developers’ workflow: it’s the developers who need to be aware not to use extra-short strings in the first place. And to resolve this internationalization pain, l10n managers have to:
Agreeing that developers need to be educated about internationalization, and that context is key, continuous platforms offer:
Unlike Serge, Transifex is not free (in fact, in case of such a need, Serge may be actually paired with Transifex). It comes with a number of tools like glossaries, translation memory, and reports. It also has the functionality to assign roles like “Reviewer” or “Project Admin” to contributors, thus enabling a logical system for translating, reviewing, and taking content live.
Transifex is a platform onto which Microsoft has released open source products. Transifex also has many integrations with systems like Slack and WordPress. It uses a REST API and a command line client tool to auto-detect new content.
In practice, some APIs may result in only a partially localized website. You’ve probably seen examples of this approach in action on the Internet. That means, linguistic sign-off is still advisable for multilingual quality content. And a continuous l10n approach requires either a solid team effort with the end goal communicated and understood across the organization where it is implemented, or APIs which are not based on the principles of uploading files and creating jobs. Because an “automated” system – one where developers have to actually create the whole workflow in-house, as well as understand how to place orders and feed results back to the system while avoiding file conflicts – sort of discredits the concept of automation itself.
As recently shown in a survey conducted specifically for Nimdzi’s upcoming SW l10n report, buyers familiar with the concept of continuous integration (from which continuous localization has originated), turn to continuous l10n solutions regardless of their limitations.
Xsolla with Serge, Eventbrite, Prezi and others with Transifex, or Jive Software with Smartling – these and many other SW developers look into this area, because indeed it does make manual localization management (exporting, converting files, sending them for translation, doing reverse conversion, committing changes to version control, etc.) unnecessary.
At the same time, for the majority of the implementations, serious integration and support efforts from the developers on the buyer side are still required. Under the hood, one may still have to manually create processes and workflows. That’s why it makes sense to keep your eyes on the actual use cases of the continuous localization approach.
We all know that human input is still invaluable when reviewing localized content. But with ever-improving localization technologies, where does a manual approach to auditing matter most?
The ongoing coronavirus outbreak has been affecting the way businesses and individuals work. What does it mean for the localization industry?