🚀 go-pugleaf

[Note from Jeff (INSOFT-L List Manager): Please see how Glenn perfectly
 changed the original "Subject: " header once the topic shifted from the
 original message.]

  Date: Fri, 11 Jun 93 17:15:54 +0200
  From: hugh_davies.wgc1@rx.xerox.com

  glenn@metis.com writes;

  <...
  Fully automated translation of embedded text in programs is not currently
  possible; some automated translation is possible, however the quality is
  not sufficient to depend on it completely.
.>

  That depends entirely on what you mean by "Fully automated translation".
  If you mean that you point your translation application at the program
  and a fully translated application pops out the other end, then true.

Your response proves my point.  As a human who speaks English, even you
don't know what I mean by the adverb "fully".  Therefore, "fully" automated
translation is impossible.  QED.

I am quite familiar with the capabilites and utility of SYSTRAN, ALPS,
METAL and many other MT systems.  They are not capable of "fully" automated
translation.  They are expensive and difficult to master (at least in
their details -- just try altering the syntactical formalism of any of
these systems).  They are tools for human translators to use, which is how
you use them.  I don't see any disagreement between our statements at
all.

The reason I make this statement is because I commonly encounter the idea
that such "fully" automated translation is possible.  Perhaps the readers
of this list are informed enough not to make such a leap of faith.  In that
case my statement was unnecessary.

When an MT system can resolve the difficulties of lexical and syntactic
ambiguity, of ellipsis, of default reasoning as pertains to pragmatics,
of metaphor and of metonymy, then, and only then, might I agree that
"full" automation is possible.  I'm sure you're not suggesting that these
have been solved in todays systems. Are you?

Glenn Adams