Google Technology RoundTable: Human Language Technology

Human language technology experts at Google, Franz Josef Och and Mike Cohen discuss their exciting research in machine translation and speech technology with Alfred Spector, Google VP of Research and Special Initiatives.
http://www.google.com/jobs/youtubevideovig

Category:

Science & Technology

Tags:

• Google
• human
• language
• technology
• research
• R&D
• Franze
• Josef
• Och
• Mike
• Cohen
• Alfred
• Spector

[3145mimosa]

Franz Och is more than an expert. He together with PF Brown are two fathers of modern statistical machine translation. i love both of them so much!

[majoyful]

This thing has limitations.

[mikedelgado8888]

I can't wait for this to come out,but i might tell someone to F off ,when I wanted to say hello, i need directions to the nearest coffee shop.

[siglmgga]

For millions of us who cannot hear clearly, please DO fix the automatic captioning on this of all videos online...yes? We'd like to talk to you soon.

CCACaptioning

[Trancecend]

Did he just say there were little data on danish (13:53)?

That sounds pretty odd.

[shitydrumer]

if its my personal information than i do not agree but public information than i think it is a good idea

[salmiak911]

@nennettnet: Probably the user says the same over and over again until the duck quacks ( yeilds the desired result ), then that is feed back thought the system; thus we have unsupervised voice recognition.

Most problems that seems impossible/unsolvable can be resolved with a little bit of fantasy.

Could talk about this all day long but does only have 500 characters of space here.

Cheers!

[feebdaed]

Och's homepage has many links to papers coauthored by him (about handling large language models and so on), I think it's enough to give a look there to go deeper

[nennettnet]

Mike Cohen talks about the importance of "unsupervised learning". What does this mean to iPhone voice search?

I guess, the iPhone app sends each query to Google's voice recognition server, where it is changed from audio to textual data. But for permanent innovation of voice recognition it must be essential too, analysing each query afterwards, whether the system could identify the correct search term.

Is this these days really possible with unsupervised learning or does Google do this by hand?

[bamboowarrior]

When I heard about these videos, I was excited. After watching, less so. Maybe I misunderstand the intended audience--or maybe there isn't one? I'm a bioinformatics researcher, and search is obviously hugely important for us--as are many of the sorts of models Google uses in most of its technologies. But where is the meat? These videos lack any real details about major innovation. (And yes, it is possible to include such details without giving away everything.)