When Skype announced its real-time translation program back in May, most of us seized on the sci-fi-ness off it all—Star Trek’s universal translator, Babel fish, etc. But the technology is very real, and has been for years, just it separate pieces. Skype Translator is is the commercial culmination of those efforts, bringing all those things, like speech recognition, automated translation, and machine learning, into one program.
This week Skype began rolling out the “first phase” of Translator, a beta version of the service’s live speech translating feature (between Spanish and English for now) and text translation for 40+ languages.
The promise of breaking down the global language barrier is a lofty one—solving the human speech puzzle with all its nuance and imperfection would give our machines a skill that has forever been uniquely human. Skype Translator doesn’t quite reach it. Not yet, anyway.
To test, I decided to have some lengthy convos with Manuel Méndez, managing editor at Gizmodo Español. Having not spoken a syllable of Spanish since high school, I opted to speak in my native Inglés (that’s one of about five words that I remember) while Manuel, who’s a completely fluent English speaker and smarter than me, checked Skype’s Spanish-to-English accuracy.
In Translator, you’re given a live translation on the right as you’re speaking, both in your native language and whatever language your caller is speaking. Now, picture all of the Skype conversations you’ve ever had. This will not be like that. For Skype Translator to work properly, there is a little mental conditioning involved. For one, you must speak slowly. Skype Translator’s speech recognition is good, and plenty fast, but that accuracy decreases as you speed up in words per minute. “Hey, how is it going?” can change to “Hey is going?” pretty quickly.
Also, you’ll need to make exaggerated pauses when you’re done speaking. Skype Translator will translate pretty quickly. If you’re someone who “ums” and “ahhs” and pauses between phrases, your sentence will appear in little chunks, which can be annoying as hell.
Skype Translator will start the conversation with audio translation turned on, meaning after every translated sentence, your male or female avatar, will hop in and basically ready what was just translated. After about five minutes, I turned this feature off (which turns it off for the other speaker as well) and just read the transcripts.
Once you’re able to rewire your brain to Translator speak, then this program is really quite amazing. The speech recognition is the foundation of all the translation work. It needs to be perfect. Microsoft says headphones with a dedicated microphone will yield the best results, and for the most part, that was true. But even talking unplugged and over loud music, Translator was still able to do its thing pretty accurately.
But where Skype aces speech recognition, Translator might need some extra credit to get a passing grade in translation. For example, during a Skype chat translation of the following sentence:
“I think I have a handle on this guy.”
I was saying to Manuel that I think I understand Skype Translator, calling said program by “guy.” However, Skype Translator didn’t know that (understandably) and translated:
“Pere creo que tengo un mango de este tipo.”
Which literally means “I think I have a dick of this type,” with “mango” meaning “handle” but also being a slang term for “dick.” Your Grandma living in Honduras just became very concerned.
This is probably an outlier in possible translation mishaps, but they do pop up here and there. That’s why Skype Translator Beta really feels like a language assistant that a true translator. According to Manuel, if a Spanish speaker with no English knowledge tried to decipher Skype’s rendition of my beautiful prose, they would have a tough time understanding.
Because I know no Spanish whatsoever, I can relate. Generally, I could get the impression of what Manuel was trying to say, but it would show up somewhat broken. But if you have a basic understanding of the language, not necessarily fluent but know a couple hundred words and general grammar, Skype Translator fills in the blanks.
How It Works
In a follow-up post to Skype’s Monday beta launch, the team create a helpful little infographic showing how exactly the program’s cogs turn:
This an overly simplistic representation of the advanced computer science going on here, but Skype Translator recognizes your voice, corrects for any stuttering or ticks, translates and then delivers to the listener—all in a split seconds.
After some setup—selecting your language, your digital voice avatar—you enter the Translator app, which looks basically like Skype proper on Windows 8.1 but with a few extras. Now, when you chat with a friend, a translation toggle pops up beneath their profile. When you switch on the toggle, Skype will ask you what language the person you’re about to call speaks and writes.
This is important because if you get this confused, Skype will try to translate English phonetically into Spanish, which comes out like jumbled nonsense. Set this up right (and make sure your caller does the same), and place the call like normal.
The verdict? Translator isn’t quite there yet. For now, the language barrier is still here. But Skype has created the battering ram that will one day hopefully breach its walls.