Machine Learning and Machine Translation

[coredump2]

A heuristic approach is bound to fail. Evidence of this is current MTs are very poor translators at best... actually "useless" may be the closer assessment. It baffles me that you guys are still trying an algorithmic approach to MT. It is well known since at least the 70s that it requires more than just mapping words from one language to another. I'm sure you know this. So why try going down a path that leads to no where knowing there's a dead end?

[coolguitar007]

is there anyway that we can get these slides?

[mysterons1]

Yes, contact me at my gmail account (milesosb+talk@ ...) . Phil --who is speaking-- is my postdoc.

Miles

[misiuqrde]

@GoogleTechTalks, please put the high quality video of it, slides are hardly readable.

[mysterons1]

Don't worry about not being able to read the first slide --it just shows how complicated and ad-hoc most MT systems are

Miles

[kobi2187]

Nice, anyone knows of more well defined architectures for machine translation, or other knowledge about this matter?

[theannoyingdogpoo]

[yawning]