[sldev] Please help testing the SL Viewer with integrated universal translation

Mike Monkowski monkowsk at watson.ibm.com
Thu Feb 26 13:10:22 PST 2009


Henri, I hope you didn't take my reply too seriously.  I'm sure my 
French is much worse than your English.  I just felt that a sentiment 
such as "simple languages such as English into more complex and subtle 
languages such as French" deserved a friendly rejoinder.

I have worked on speech recognition and natural language understanding 
and have been suprised by the power of statistical methods.  CMU has 
been a leader in language processing, so I expect that their CMU 
Statistical Machine Translation will continue to improve.  Getting it 
into the hands of more users is a good way to facilitate that.

Mike

Henri Beauchamp wrote:
> I never said it was a bad example of user-ran project neither that the
> request for feedback on this list was a bad idea.
> 
> I simply said it was pretty doomed to failure because of the poor
> (and actually pitiful) results returned by the automatic translators
> it is relying on.
...
> My point was not about simple mistakes, such as using the wrong pronoum
> and/or gender, but about the very meaning of the translated text (which
> can become right out the opposite of what the original text was all about),
> something automatic translators are very prone to produce.
...
> It is good enough to clearly express my thoughts and avoid misunderstandings,
> which is really all what I care about. I don't pretend writing a perfect and
> 100% grammatically correct English, but if I were to write this very email
> in French and then run it through an automatic translator before sending it
> on this list, my guess is that you won't understand a single thing...
> 
> However, judging someone's skills, based on an email they wrote in five
> minutes and did not re-read, is not very reliable a method...




More information about the SLDev mailing list