Ask any voip fan what their new feature wish list looks like and you'll probably find on-the-fly language translation somewhere in the top 10. Technology has evolved to the point where geographic barriers are no longer very relevant - a broadband connection and Voip internet phone service are all you need to connect a Patagonian sheep rancher at the tip of South America to the Chinese textile company that will take his wool and turn it into marketable sweaters. Unfortunately, if they can't understand each other, those custom sweaters with the sweet price point won't ever make it onto store shelves.
A number of stopgap solutions have cropped up and represent definite steps in the right direction. Skype, for instance, recently introduced the Universal Language Real-Time Message Translator (ULRTMT), which lets users translate their chats to and from one of 13 languages, including: English, Spanish, Russian, Chinese, Japanese, German, French, and Arabic. They also offer a live translation service for conference calls of up to 5 participants. At $2.99 a minute, it's probably cheaper than hiring a private translator, but expensive enough that your average individual or small business user won't want to use it for anything but the shortest calls.
Another promising development comes courtesy of Sharp and IBM, in the form of a handheld English-Japanese translator. It's similar in size and shape to the average PDA and due out by the end of the year. Although there's something to be said for doing one thing only, but doing it well, it would be nice if this device had the ability to translate languages beyond Japanese and English.
The good news is that advances in speech recognition and natural language machine translation have opened the door a little, giving a glimpse into the future of just-in-time translation. Meaningful Machines is at the head of that forward momentum, employing a unique translation engine that doesn't rely on difficult to manage parallel text databases. Instead, they apply a massive bilingual dictionary and language samples from both the target and source languages.
Working with 5-8 word chunks of text at a time, the translator sifts through hundreds of possibilities; ranking results according to the number of times a phrase has turned up in the target language, or how typical a particular word order for that language might be. Translation takes 10 seconds for each word; not an ideal rate for verbal conversations, however the company hopes to bring it down to 1 second over the next year or so.
In the end, short of hiring a live translator, we're still a ways off from realtime voice language translation for voip applications, but there's very definitely light at the end of the tunnel.
No comments:
Post a Comment