AI series, part 1/3: Artificial intelligence in translation and localization – how it’s changing the face of the industry

AI series, part 1/3: Artificial intelligence in translation and localization – how it’s changing the face of the industry

AI series, part 1/3: Artificial intelligence in translation and localization – how it’s changing the face of the industry 1920 1465 Bartosz Budzynski

In this three-part series, we discuss the role of artificial intelligence in translation and localization. In this first post, we talk about how it’s changing the industry – both within the act of translation and localization – and more broadly. The second post will look at how you can use AI to boost your localization and translation projects. And the third will provide practical advice about integrating it into your workflows.


In translation and localization, when we think of AI, the first thing we think of is machine translation. No wonder. It’s been around a while and has developed considerably. It’s a perfect example of artificial intelligence technology in action.

But it’s far from the only one. In fact, AI crops up in many parts of the industry – both within actual translation and localization work, and in the work that happens around it – the project management, quality assurance, and terminology work. Not to mention in our everyday lives and the way in which we work.

Artificial intelligence in translation and localization, the act itself

Machine translation – adaptive, augmented, blended, hybrid

Machine translation has come on in leaps and bounds. Powered by natural language processing, it now produces text approaching human-quality translation. In response, we’ve seen an abundance of new approaches, from blended translation and augmented translation to adaptive MT and hybrid technology. Let’s look at these in more detail.

In blended translation, either MT or TM is used to pre-populate CAT tool segments for the translator to review. The augmented translation is similar but, in this case, MT and TM suggestions are presented to the translator (rather than pre-populating). With adaptive MT, systems learn from a translator’s edits in real-time, improving MT output on the fly. Note that these AI technologies aren’t mutually exclusive – adaptive MT can be used alongside the augmented or blended translation.

Hybrid technology has recently entered the fray. Used to spot common issues in text segments, such as typos and named entities, and lock segments when needed, it aims to boost translation efficiency.

We’ve also got better at assessing MT quality. Edit distance evaluation allows us to calculate the number of changes translators make to MT-generated content, so we can tell how much work it takes to get it up to scratch.

Data extraction for added context

AI is already being used to extract data from documents, images, and graphics. If tomes of client documents, mammoth style guides, or screeds of website content could be processed in a similar way, extracting key data, and integrating this into the localization workflow, this could help boost efficiencies dramatically. It would give translators and localizers what they’re often lacking – that all-important context. And consequently, it would help produce a better quality translation with fewer client queries.

Speech-to-text and text-to-speech

For some time now, translators have been making the most of speech-to-text technology such as Dragon to dictate their translations. By taking away the need for typing, it allows them to work more quickly and prevent muscle and joint strain. Some claim it also helps produce more natural, conversational texts from the off.

Text-to-speech also has its benefits in translation work. Many use it to have their texts read aloud to them. This is a useful addition to the editing toolbox. Hearing texts read by someone (or something) else makes it easier to spot issues easily missed in a visual review. This can include typos, missing words, or even problems with flow and sentence complexity.

These technologies seem to be growing in importance, with additional platforms, such as Zoom, integrating them into their software. Perhaps this is a sign of things to come in the translation and localization space too. CAT tool integrations with text-to-speech are already available, as with Dragon. But in the future, maybe CAT tools will start to embed speech-to-text or text-to-speech directly within their software? Watch this space.

Artificial intelligence technology in the wider translation and localization workflow

Just as artificial intelligence already affects – or will affect shortly – every business, AI affects not just the act of translation or localization, but also everything that goes on around it.

Automation of translation workflows

Beyond the wordface, AI has a big impact on translation and localization workflows. Countless LSPs are harnessing AI to automate certain stages of their workflows. This frees up precious project manager time, allowing them to work more efficiently or perhaps focus on other areas such as client relationship management.

As AI technology becomes more widespread and accessible, presumably workflow automation will no longer be the sole preserve of large multinationals. Small agencies and individual freelancers will start to make the most of workflow automation too, allowing them to focus more on the work itself rather than the admin, and boost productivity. In fact, this is already happening to some extent, with a plethora of apps for project management, accounting, to-do lists, and time tracking on the market.

Terminology mining

Another area that makes especially good use of AI is terminology management. It can be a huge boon for translation providers. Using AI technology, they can filter, categorize, and extract terms from large swathes of content. Artificial intelligence tech is not only speeding up this work but far surpassing the quantities humans could feasibly achieve.

New roles

As artificial intelligence changes the face of translation and localization, new roles and opportunities are starting to emerge. But for MT quality to reach new heights, we’ll also need more in-depth quality assurance. So we’ll likely see more roles in this area. Similarly, because MT engines are only as good as the data they’re trained on, there’ll be a greater need for people with linguistic expertise who are able to vet and clean that data.

Bartosz Budzynski

Responsible for Professional Services at XTRF. A strong supporter of the open-source movement, sharing economy enthusiast, and a passionate developer. His greatest superpower is transforming business needs into technical requirements.

All stories by : Bartosz Budzynski