POST WRITTEN BY João Graça
If you use translation software from the likes of Google or Microsoft, most of the time you’re going to get a spot-on answer — as long as it’s a word or phrase. Want to know the German for “airport”? Or how to order a beer in French? Perhaps you want to ask for the bill in Japanese to impress your date? Just use Google Translate (surreptitiously or otherwise), and you’ll be all set.
Great, but this doesn’t mean that pure machine translation (MT) is fluent, or that it makes fewer mistakes than humans, or that is indistinguishable from professional translators. Papers claiming that human parity has been achieved for some domains are more often than not slightly misleading, as this evaluation is done on a sentence-by-sentence level and not on the whole document.
Add in any level of complexity, say from the business sphere, and MT is far less capable. Would you be happy sending an important email to a business partner in another language that had simply been put through an online translation engine? Most MT in a generic sense is trained only on web data. It can’t distinguish whether you want a sentence to be informal or formal. It doesn’t understand industry-specific jargon (think finance, law, health), and the wider context of a piece of content is lost on it. General translation engines might translate text, but they are a lousy format for clear communication.
Research from the European Parliament shows that a common language increases trade flows directly by 44%. However, businesses wanting to communicate with foreign markets and improve their user experience have always been held back by language barriers. Professional translators can be costly and time-consuming, and there simply aren’t enough of them to translate all the content being produced today. MT may be far cheaper and quicker, but it’s unable to maintain the tone of voice businesses need across markets and languages to protect their reputation and build meaningful customer relationships, and as such can’t be relied upon to maintain professional communication standards in the world of business.
How then can businesses make the solution fit the problem and unlock the true benefits of machine translation?
We’re seeing our fellow service providers generally take one of two approaches. The first is a managed service line, where companies use an agency model built around a small community of professional translators and large-scale translation projects. Leading players in this group are those like Lionbridge, SDL and Smartling.
Then you have those taking an automated-service approach. Companies like TextMaster, Gengo and my company, Unbabel, provide more streamlined software models that plug into client applications to retrieve content and deliver translations, normally with high levels of data curation in particular. Both approaches have their strengths, but they can run into traditional pain points of heavy project management toil, topic expertise and high delivery costs.
For me, it’s all about using the most advanced intelligence we have available — humans. Placing humans at the right point in the process, to be used only when absolutely necessary, can create an AI-powered translation platform with humanlike quality while still making savings on speed and costs. A post-editing process minimizes the number of human working hours from a business perspective but ensures that the whole translation-as-a-service offering — predominantly focused around machine translation and quality estimation — gives a near-perfect result each time.
Whichever approach you take, if you’re a business leader looking to unlock the benefits of machine translation, here are a few top tips:
- Take the chance to offer your services in a multilingual fashion. Visitors to your website should be able to communicate with you in their preferred language, and a good MT platform will make this process quick and painless while you reap the reward of customer satisfaction.
- Let the machine do what it does best. Humans and AI work really well together, but not if they are treading on each other’s toes. Don’t interfere with the heavy lifting that machine translation can do easily, but make sure that humans can use their instinct and experience to ensure fluency and quality.
- Feed the system. Even the best MT will need human guidance on the jargon and everyday phrases of the environment into which it is about to be placed. A bank would need to provide a list of financial phrases that can be tricky to instantly decipher. A government department would need to teach its MT software about all the acronyms and codenames its employees use every day. If humans are freed up to constantly monitor and update this dictionary, then the machine translation device will become more accurate more quickly. The more data a user sends through their MT system, the more data exists to train the model in a virtuous loop.
This is part of a wider strategy that businesses can consider in terms of their technology stack. Having a fully integrated technology stack for translation — from the MT engine, to the translation tools, to the integrations with customers — has incredible benefits in reusing human-produced data to train AI components. Ultimately, a smooth machine translation pipeline can enable increased customer satisfaction and greater global reach for businesses around the world.Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?