Re: response to J. Prothero book review and comments of 12 Oct 90

[uunet!math.ucla.edu!jimc]

> Date:  Tue, 30 Oct 90 21:49:50 +0800
> To:  lojban-list@snark.thyrsus.com
> From:  major@pta.oz.au (Major)
> Subject:  response to J. Prothero book review and comments of 12 Oct 90
   
major@pta.oz.au (Major) writes: 
> lojbab@snark.thyrsus.com (Bob LeChevalier) writes:
> 
> > Sheldon Linker has claimed that he could do a Lojban conversing
> > program with heuristic 'understanding' a la HAL 9000 in 5 man-years.
> 
> If this is true, then he should be able to do English in 6; understanding
> the chaos of natural language would not be particularly difficult after
> understanding the chaos of natural thought.

Don't underestimate the effect of language chaos!  My experience with
-gua!spi may be relevant here.  I am writing the preprocessor for such
a program, which maps language into phrases, with cases identified,
pronouns linked to antecedents, replicated modal cases inserted a la
pc's proposal about tenses, etc.  Simple, eh?  Particularly since
-gua!spi is very similar to Loglan in grammar but super-simplified to
make this step easier.  

In two weeks I had the lexer and parser down cold, and I have hardly
touched them since.  After about six months the phrases would almost
always go into the right cases, taking into account placeholders and
the like.  The rest of the time has been spent mostly in nailing down
nitpicky little details of what should be the antecedents of pronouns. 
For example, why is Kira's girlfriend the speaker (antecedent of "ji")
in the subphrase of a direct quote while Kira is the speaker in the
main phrase?  I translate a little more (deliberately using difficult
relations), run it through the program, and discover another point I
hadn't thought of.  Perhaps some theory will become evident but so far
it's been slog, slog, slog.  

-gua!spi is pre-planned to be easier than Loglan to pronounize, while
Lojban retains a lot of complicated structure which, every now and
then, I moan about to Lojbab's annoyance.  Imagine if you had to handle
English! I think it would be a major achievement to reliably
distinguish direct from indirect objects -- note: reliably, not just
95% of the time.  I'm not sure that an algorithm exists for reliably
dealing with antecedents in something like this:  "Joe likes Jerry.  He
says that he tells cool jokes."

And after that's done, you can start on the main project, the
understander.

		-- jimc