Today’s neural machine translation engines are capable of translating simple – and ever longer – texts into an increasing number of languages. They can be trained, and they also learn continually. However, without human intervention, the end product remains relatively raw. Which is why MT is only suitable for specific types of text.
Nonetheless, social media platforms have long since begun integrating automatic translation features into their sites. Posts on Facebook, LinkedIn and Instagram can now be translated into the browser’s language with a single click. We wanted to see exactly what they were capable of, so we put some of our own recent posts to the test:
Verdict: The meaning is clear, but the human touch is lacking.
Verdict: Automatic translation? It’s complicated.
Verdict: Could do better.
YouTube
The popular video platform has also gone multilingual: thanks to automatic speech recognition, it can generate subtitles automatically and then translate them directly with the help of NMT. It’s an impressive feature, especially considering that it’s available in 103 different languages. Here’s how it looks:
Verdict: The quality of the automatic speech recognition is good. The quality of the translation, however, varies significantly depending on the context and video content. Not to mention the questionable grammar choices.
What does this mean for you?
The results of our test run speak for themselves, but we’re fans of words, so we’ll say it again: automatic translation is adequate when it comes to understanding posts, and could be useful when you’re just starting out on a social media platform and don’t (yet) want to pay for translation. Why not test it on your own posts? And then talk to the professionals.
Cover image via Twenty20