Transifex & Amazon Translate: Bringing Machine Translation to Enterprise Content Management

Guest Blog Post by Dimitris Glezos, Founder & CEO, Transifex

Software developers need localization tools that integrate with their systems and a translation process that is fast without compromising quality. Without these tools, developers are forced to spend time sending files around, sometimes through inefficient spreadsheets or FTP servers. In doing so, content release cycles can slow down, compromising agile product development altogether. At Transifex, we’re constantly asking ourselves what building products can look like when localization becomes an integrated part of the standard developer stack and agile workflows. What technology do we need to enable companies to go global from day one?  How can we make localization seamless?

When translating customer-facing enterprise content, quality matters. Local cultural and linguistic nuances make all the difference for companies who want to offer a great user experience for their global users. Specifically speaking, incorporating proper local slang and jargon shows a strong understanding of the local market and reinforces a deep connection with your local audiences. On the flip side, improper translations can result in brands being perceived as unprofessional and, in worst-case scenarios, can even offend local users. One choice for developers is translating through language providers with human translators, but this option can be slow, prone to inconsistent content, and the cost might be hard to justify. For example, translating your product documentation pages could cost tens of thousands of dollars for a single language. On the other hand, Machine Translation (MT) is an alternative that is fast and scalable but can lack quality for content where potential mistakes can’t be afforded.

Today, we’re excited to announce a step in that direction: a deeper integration between Transifex and Amazon Translate to help bring the power of collaboration and contextual metadata into the fast-paced translation workflows powered by MT. Here’s a closer look at the inspiration behind this next-generation localization technology and how it works.

Machine Translation is very compatible with the fast-paced workflows followed by developers. It’s fast, it’s predictable, and it’s scalable. However, it’s tough for developers to adopt it for content where mistakes can’t be afforded: they need to build internal databases with their global content, which are integrated with all their source systems, create triggers for API calls to the MT with the right content, and build interfaces to post-edit the content by external translators together with permission controls and glossary UIs, among other things.

So what would it look like if we brought together our translation management tools with an established MT solution like Amazon Translate? The result is a switch from a solely human-centric workflow to an MT-centric one (with human-in-the-loop, when needed). Together, the two drive a vital shift towards increased accuracy and seamlessness.

The New Approach

In order to leverage powerful Machine Translation into the localization of enterprise content, two key elements are needed: integrations into source content systems and a quality control layer.

With this integration, users can leverage Transifex’s content management layer to have context-aware metadata, post-editing, and glossary functionality on top of MT. In addition, users now have developer-friendly deep integrations into source content systems for wide adoption inside an organization. Especially for developers, Transifex can easily integrate with an organization’s source content systems via native connectors (for example, GitHub), client libraries, command-line tools, APIs, and client-side JavaScript tools.

As a result, content is automatically and continuously synced centrally in a global content repository and stored together with valuable contextual metadata (such as developer comments, URLs the phrase shows up in, “do not translate” tags, max string length, and whether the word “File” is a verb or a noun). We trigger the right workflows based on the content localized, and we have all the UIs for post-editing the content, adding metadata, creating glossaries, and validating content.

How It Works

This integration merges Amazon Translate’s functionality with Transifex Glossary and creates a space where a company and translators can work together to define the most important terminology used in their content. There, they can answer questions like “how do we want to consistently translate the phrase ‘Like’ in a button?” and “do we want a casual tone or a formal tone in our translations?” Amazon Translate is trained to respect the company’s more accurate and human user-generated content and context, provided by the Transifex Glossary.

Here’s an example. Let’s say we are translating to Greek and training the MT service to understand keeping a brand name untranslated. This is especially relevant if the brand name, like the one in this example, also happens to be a translatable phrase.

• MT alone: “Transifex In-Context” will translate to “Transifex στο πλαίσιο” (literally meaning “Transifex in the frame”)
• MT with Glossary: “Transifex In-Context” will remain “Transifex In-Context” in the translated content, while all other content will be properly translated to Greek

To activate this integration, users who already have an Amazon Translate account can send their Glossary to their existing Amazon Translate integration with the simple click of a button. Then, whenever new Glossary terms or translations are added, the project owner will resend the Glossary to Amazon Translate to feed the machine with more accurate terms.

Driving the Next Generation of Localization Technology

As a translation management system that has been taking innovative companies global for over a decade, Transifex keeps a close eye on the best solutions out there for fast-moving, global companies to achieve the most optimal localization results. Naturally, time and experience have brought with it larger and larger strides of innovation in the field of localization technology. What started out as a fully manual process driven by spreadsheets and emails has now evolved into a solution that provides continuous, quality, and cost-effective localization. With this new integration, we are moving one step closer towards building a fast, high-quality, and transparent tool for developers and their localization teams.

To learn more and connect with other developers who have used tools like this one, join the Transifex community. If you are an AWS user looking to take this next-generation localization technology out for a spin, sign up for our free trial.