The digital transformation is touching every aspect of global business, just like it promised to. In the translation and localization sector, we talk a lot about the changes that tools like neural machine translation (NMT), machine learning (ML) and translation memory (TM) are sparking.
For localization teams embedded within software companies or large global businesses, embracing the digital transformation means thinking about the impacts of the digital transformation across the whole business.
So how can you harness the power of AI and data-driven applications to bring additional value to localization work? Below, we’ve gathered tips from data scientists working in localization to help you optimize digital transformation in your own organization.
Data, data, data
One of the ongoing challenges of digital transformation is one of data: having the right amount of data and recording it consistently across each part of a business.
For example, when a translator changes a code string or a phrase in a large-volume translation we are pretty good at tracking it. But the metadata for the code might need to change as well. The per-word cost might change, affecting the finance side of things; or a change in the code might trigger further changes for the engineers. And all of these changes need to be recorded across the business in a consistent way if one wants to leverage AI in these different areas as well.
But different parts of a business ecosystem might track data in different ways, or one unit might collect more granular data than a partner unit. And that’s before you address differences in how you store data across a business.
To realize the gains of machine learning and AI, you need large volumes of high-quality data. For example, NMTs work best when they can work with a large, accurate and diverse corpus of text. That corpus has to be properly annotated, recorded and coded in order to develop a machine translation model.
In some cases, you may need to go back and modify records or processes in order to produce consistent data all along the pipeline. Since there aren’t industry standards for data consistency (yet), it’s important to set standards within your own business ecosystem.
Context is everything
We already know that AI and machine translation are helping localization teams validate their translations and code.
But machine learning and machine translation are also increasingly being used to provide translators with valuable context. By connecting the work of business units within the internationalization pipeline, MTs that offer context and track metadata can lead to productivity gains.
Decision-making becomes easier when translators have context from other teams, from the software engineers creating code for internationalization, to the marketing team creating an in-country campaign.
Scale it down
So many things are possible with machine learning that it can sometimes be difficult to determine where to start. In a business context, there’s one key factor to help you make the decision: will the application give you business results?
With this question in mind, you can scale your applications down to a manageable size. Setting clear business priorities will also help you determine where to strategically use data and AI applications.
Prioritize your post-editors
In order for any business to realize all the productivity and financial gains that machine translation and neural networks have to offer, it’s essential to apply skilled language professionals to the MT output.
AI doesn’t create meaning — it processes language data and produces language-approximate results. But because AI can’t think and interpret, it can’t create meaning. That’s where highly skilled post-editors come in. Too many businesses, though, assume that post-editing should be treated as low-skill, low-cost labor.
Devaluing the work of language professionals usually only ends up deteriorating the quality of the result. In these cases, easily fixable blunders and errors make their way into the translation memory, and mistakes that a human would never make stay in the text.
If you’re feeding an MT output into translation memory, skilled post-editors and reviewers can keep your translation memory from deteriorating as well. If you feed low-quality outputs into translation memory, the memory quickly gets polluted and the TM stops making sense.
Want to hear more about the digital transformation in localization? Check out the full Global Ambitions episodes. Get in touch if you would like to discuss how to leverage your terminology assets.
This post is based on the Global Ambitions episodes “Applying AI in Localization Systems” with Augstín Da Fieno Delucchi, Principal Data & Applied Scientist at Microsoft, and “NMT and the Last Mile” with Serge Gladkoff, CEO of Logrus Global. To hear more about the latest advances in AI and big data, check out the full episodes!