Awesome / sugoi すごい!
« previous post | next post »
From Diane Moderski:
Is this the beginning of the end of the need for interpreters?
October 12, 2017 @ 2:48 pm · Filed by Victor Mair under Information technology, Language and computers, Speech technology, Translation
« previous post | next post »
From Diane Moderski:
Is this the beginning of the end of the need for interpreters?
October 12, 2017 @ 2:48 pm · Filed by Victor Mair under Information technology, Language and computers, Speech technology, Translation
Powered By WordPress
Mark Liberman said,
October 12, 2017 @ 4:51 pm
Rule #1 for canned demos: Don't believe them. Especially if they are obviously rehearsed and/or edited.
This may be a great communications tool. But more likely, it has imperfect coverage of a fairly narrow range of topics — as the FAQ puts it, "Currently, ili is designed for travel phrases (such as shopping, eating, reservations, and getting around). We're working hard to add more phrases and make ili the most convenient translation device for your travels." Also "As of now, ili is best suited for travel phrases (such as shopping, restaurants, making reservations, getting a haircut, and so on). It's not suited for business communication and negotiations."
And "Why ili is one-way, not two-ways?
For the time being, ili is focused on you, the traveler. ili's one-way translation is optimized for quick and efficient communication to help get you to where you want to go, tell what you want to do. " Note that the video falsely implies two-way communication — which is a bad sign for other aspects of its credibility.
There are promising speech-to-speech translation systems out there. It's not clear that this is one of them — by the company's own testimony, it's of no value outside of a narrow range of contexts, and my bet would be that many will find it disappointing even in dealing with transport systems and restaurants.
Ethan said,
October 12, 2017 @ 5:26 pm
The Japanese output sounds pre-recorded rather than synthesized. That's probably a good thing as far as being easily understood, but it means all bets are off if you want to convey a question that can't be assembled from pre-recorded bits. I have no guess as to how it is processing the English input recognition. Also, I'm trying to imagine a circumstance in which one needs to say "take me to Los Angeles" specifically in Japanese rather than English.
E.T said,
October 12, 2017 @ 7:25 pm
Google Translate (in its app form) has had two-ways speech-to-speech translation for a while now and can actually be useful if one is aware of its limitations—it can of course get a bit clunky and awkward in practice. It does require an internet connection, but getting that in order abroad is arguably no more hassle than purchasing this product.
bks said,
October 13, 2017 @ 9:51 am
It's been over 30 years since I worked in speech synthesis, but synthesizing a high-pitched feminine voice would seem to be a difficult row to hoe. I agree with Ethan's theory of pre-recording.
Philip Taylor said,
October 14, 2017 @ 6:54 am
The iframe src="https://www.facebook.com/plugins/video.php?href=https%3A%2F%2Fwww.facebook.com%2Fawesomestuff365%2Fvideos%2F1974282469480969%2F&show_text=0&width=476" does not render here, although its content is visible in the same browser when not sourced via an iframe element.
Philip Taylor said,
October 14, 2017 @ 6:55 am
A possible explanation — an attempt to embed https content in a page served as http.
Tim Martin said,
October 17, 2017 @ 9:20 pm
In addition to the need for skepticism about how useful the pre-recorded phrases are, some of the translations here aren't even that good.
They translated "can I take a picture with you" as if it were a question about physical possibility, rather than permission. The natural ways of asking this question in Japanese include "issho ni shashin wo totte mo ii desu ka," or "issho ni shashin wo totte moratte ii desu ka," etc.
A perhaps bigger problem is the translation of "I don't know how to buy this ticket." The translation provided is great, except they left off the extended predictate ("no desu") at the end. One use of the extended predicate is to imply information that technically your sentence left out. In this case, the speaker says "I don't know how to buy this ticket," but what they're implying is "…so will you help me?" If you don't include "no desu" in your sentence here, you are failing to imply that.
The difference between a proper translation using the extended predicate ("kippu no kaikata ga wakaranai ndesu ga") and the given translation ("kippu no kaikata ga wakarimasen") is actually the difference between asking for help, and being the creepy guy who tells you things about his day for no reason.