Forgot your password?
typodupeerror
Cellphones Communications Japan Technology

Japan Getting Real-Time Phone Call Translator App 113

Posted by Soulskill
from the philology-becomes-an-engineering-specialization dept.
another random user writes with news that NTT Docomo, Japan's largest wireless carrier, will be rolling out a real-time translation app for phone calls on November 1. At launch, the app will translate Japanese into English, Mandarin, and Korean, and later that month it will add French, German, Indonesian, Italian, Portuguese, Spanish, and Thai. No word on Klingon. From the article: "The products have the potential to let companies avoid having to use specially trained multilingual staff, helping them cut costs. They could also aid tourism. However, the software involved cannot offer perfect translations, limiting its use in some situations. ... It provides users with voice translations of the other speaker's conversation after a slight pause, as well as providing a text readout. ... NTT Docomo will soon face competition from France's Alcatel-Lucent which is developing a rival product, WeTalk. It can handle Japanese and about a dozen other languages including English, French and Arabic. The service is designed to work over any landline telephone, meaning the company has had to find a way to do speech recognition using audio data sampled at a rate of 8kHz or 16kHz. Other products — which rely on data connections — have used higher 44kHz samples which are easier to process."
This discussion has been archived. No new comments can be posted.

Japan Getting Real-Time Phone Call Translator App

Comments Filter:
  • Re:Star Trek (Score:3, Informative)

    by Anonymous Coward on Monday October 22, 2012 @08:54AM (#41727613)

    The red bars show it's a new story without comments yet. For such a low UID I'm surprised you'd never seen it before

  • My prediction: crap. (Score:5, Informative)

    by OldSport (2677879) on Monday October 22, 2012 @09:49AM (#41728035)

    I currently use the best commercial voice recognition software, and have experience with a variety of machine translation suites that do Japanese-English, so I have somewhat of a frame of reference.

    With respect to the former, the quality is quite good. I can speak at a relatively natural pace and get upwards of 95% accuracy. That said, I still have to adjust my speech in a sometimes unnatural manner to be sure the program "hears" me correctly, and I have heard horror stories from people with different accents/dialects having a terrible time. (Someone made a joke about Welsh, but I know a Welsh fellow and he had some colorful things to say about the commercially-available VR programs.) An additional complicating factor in the J-E scenario is that Japanese has many words that sound nearly identical, and are distinguished only by slight inflections: (hashi; chopsticks) and (hashi; bridge) is a typical example. There are thousands of examples of such words, and from what I understand, Japanese voice recognition software is quite far behind because of this particular trait. Without a UI for speakers to choose which word they are actually trying to say, I can imagine that the VR side of this program has a slew of problems.

    Now, onto machine translation. As it stands, MT works great for some language pairs, but Japanese-English is notoriously problematic. AFAIK part of the reason is the highly contextual nature of the Japanese language. Subjects and objects are often omitted entirely, for example. I don't really have to go into this in detail -- just run any Japanese Wikipedia article through Google translate and see what happens. Other commercially-available and proprietary software I've used has been basically the same (Google actually seems to be a bit better, usually.) English-Japanese is a bit easier because the context (subjects, objects, verbs) are typically "all there" -- so even if the result is Japanese that is horribly unnatural, you may still be able to get the info you need.

    But now, they're going to take the VR-generated input of varying degrees of accuracy, and run THAT through MT software that butchers even the most simple and perfect sentences? I could be wrong, but I'm having trouble seeing how the result will be anything less than a disaster. "The software involved cannot offer perfect translations, limiting its use in some situations" sounds like the understatement of the year. Get ready for synthesized-voice gobbeldygook and an mp3 website called "spoken Engrish."

    (Full disclosure: I am a translator, but in a lot of ways a yearn for the day when MT will be good enough to put me out of a job; I think the idea of people being able to instantly communicate with speakers of other languages is exciting and would lead to a much more open world. However, I've been hearing that tech like this is 3-5 years away for decades now. If this product showcased a revolutionary engine for MT, then I would be singing a different tune, but for now it seems like a mere combination of two imperfect technologies.)

  • Re:Great. (Score:5, Informative)

    by ShanghaiBill (739463) on Monday October 22, 2012 @09:57AM (#41728121)

    Yet another reason for the Japanese not to learn English properly. As if we didn't have enough Engrish already!

    A few years ago I was eating dinner with some gaijin friends in Tokyo, and a Japanese guy came over to our table, introduced himself, and sat down and started talking in a weird, unintelligible language. We couldn't understand a word he was staying. Then the waitress explained that he was trying to tell us that he was an English teacher at the local high school, and the language he was speaking was English.

We warn the reader in advance that the proof presented here depends on a clever but highly unmotivated trick. -- Howard Anton, "Elementary Linear Algebra"

Working...