September 13, 2022
Machine translation has well and truly become a worthwhile solution for businesses looking to expand their reach to consumers around the world. And AirBnB is perhaps the latest example of a well-known business that has realized this.
Its proprietary translation engine (called, well, Translation Engine) was first revealed in the app’s 2021 Winter update, but the company has periodically announced updates that dramatically increased the quality of translations and expanded its use cases within the app, with the latest update adding machine translation functionality to customer reviews.
Of course, it comes as no surprise that AirBnB of all companies would appreciate the importance of language when it comes to providing the best consumer experience. After all, the app is geared toward people from all over the world who are looking for a place to stay in their travels—people who need to communicate with each other, and who all speak different languages.
AirBnB has previously claimed that its Translation Engine has improved over 99% of its listings—a bold claim if ever there was one—and continues to improve and do better over time.
If you’ve been keeping up with our blog, you probably already know how that works. Machine translation today uses state of the art machine learning models that learn from large quantities of language data. AirBnB definitely has access to that kind of data, claiming to process over 3.5 million messages daily and having over 500 million reviews on its listings to date.
Another notable thing about this data they have on hand is that it’s very domain-specific. What this means is that the same kinds of language and terminology of a given sector—in this case, the hospitality industry—tend to pop up and be translated more often. This makes the machine translation system much better at translating texts within that sector compared to generic models like, say, Google Translate.
AirBnB’s data has the advantage of being hyperspecific, which means that the quality of machine translation for its specific purpose would also tend to be higher. You wouldn’t use AirBnB’s Translation Engine for, say, technical documents, but it’s perfectly tailored to translate rental listings.
And that’s just the back end of things. On the front end, Translation Engine has provided quality of life upgrades not just for the app’s guests, but also for its hosts.
Before AirBnB’s Translation Engine, hosts would have to create listings in different languages manually. They would translate text in their language to another, often using a generic machine translation system like Google Translate, then do it again for a different language.
With recent updates, all that trouble is no longer necessary. Hosts can just create a listing in their own language, and let AirBnB’s Translation Engine do the rest.
In fact, users don’t even have to click a button anymore to translate—the app automatically detects the user’s preferred language and translates listings, reviews, and messages accordingly.
And because it’s specially trained, the quality of the translated text is definitely better than what could be achieved with generic machine translation systems.
Parting thoughts
AirBnB’s Translation Engine is a very good example of a machine translation workflow that is well-thought out from end to end. It plays to the company’s strength, which is the availability of a vast corpus of multilingual data, in order to improve the quality of translations in its specific domain. These gains are then immediately brought to the app’s users through seamless UX design. This is something that other businesses can learn from when deciding on how to leverage machine translation in their own operations.