The Chinese Diplomat's "the"

Alexander Gross language at sprynet.com
Tue Aug 31 07:59:47 UTC 2004


Good, Rob, glad to hear you think it's impossible, though that's probably
not the
whole story either, and as the source i cited mentioned, all the work that
has
been done (& all the billions of $ spent so far) could end up helping
translators
to work more efficiently, though CAT & TM already do this.  The real kicker
is
that even if they finally perfect MT, the only people who will be able to
handle
the system & make corrections will end up being human translators, or at
least
those human translators willing to work with it.

No, i'm certainly not accusing the authors of fraud.  But i do have to tell
you
that there have been some genuine instances of fraudulent demos in this
field,
documented back in the 80s in the pages of Language Technology (the
precursor of WIRED Magazine), which i wrote for at the time. Also, on one
occasion three of us, the UN's MT & Terminology expert, the president of the
NY Circle of Translators, and myself, had no choice but to show up for a
press
conference promoting a blatantly fraudulent MT system. Fortunately the press
managed to figure it out for themselves, and we didn't have to say very
much.
The person behind that system just might be one of those people you feel is
smarter than you--or perhaps the teacher of some of them.  What's more, all
through the late 80s one MT company ran ads promising that with their system
monolinguals will perform "truly automatic translation .....without
assistance
from bilinguals, polyglots or post-editors.....but meeting the quality
standards of
professional translators-no less." That guy is still quite active in the
field but
now promises no more than further improvements in TM (Translation
Memory).

> I'm no MT advocate -- my personal feeling is that MT is impossible, but
> there are enough people smarter than me who disagree that I hesitate to
> say that in public.  The original motive for the paper that I cited was
> an adaptive communication device.  It had nothing to do with MT.  And,
> in case I didn't make it clear, the "right almost 85% of the time" was
> for a narrowly defined task, namely recovering omitted articles in
> monolingual English texts. For that task, according to the results they
> published, it really is right almost 85% of the time. Unless you are
> accusing the authors of fraud, I don't see there is any evidence of
> "innumeracy" here, spreading or otherwise.
>
> Absolutely no claims are being made about MT, or how well this program
> would perform as a component of an MT system, or really even whether a
> program like this is useful for anything.  However, I am making the
> claim based on this paper (though the authors might not endorse it) that
> most of the time selecting which article to use in a given context isn't
> very hard.

There i certainly agree with you.  But remember, it isn't very hard for you
&
me, but it's bewilderingly difficult for many ESL & translator-training
students.
Anyway, 85% still isn't going to cut it, and i can't help wondering if they
asked
their system to choose only between definite and indefinite articles, in
which
case the law of averages would already credit both alternatives with 50%.
Even if they allowed for zero grade articles, that would still give all
three
alternatives a 33% free boost before the test went further.

very best!

alex

> --
> Rob Malouf <rmalouf at mail.sdsu.edu>
> Department of Linguistics and Oriental Languages
> San Diego State University
>
>
>



More information about the Funknet mailing list