[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Parser

Neither the parser nor the glosser have been issued as products for any form.
A couple of people have copies of each -call them Beta testers, though in
reality it is because they are doing text reviews of texts posted by others
or writing so much text that it is easier on John, Nora, and me to let
other people use the tools as best as they can than to do it ourselves.

Neither the parser nor the glosser is suitable as a commercial product.
The parser actually has a little bit of a readme file; the glosser has
even less.  If you do perfect Lojban, the parser will confirm it by issuing
a parsed output, but if you aren't familiar with the grammar, you will see
such an output as approval, when it really means that it is time to start
checking for the errors that the parser cannot catch (like the missing
"cu" in the first line of your haiku - first attempt.  Perfect Lojban but it
doesn;t mean what you intended.)

If you actually have a parser error, the parser's error reporting is not
very helpful, and indeed is even misleading, as to what the problem is.

Finally, thge parser John and I use always reflects the latest trial grammar
and is not the baselined version.  We cannot support a lot of people by issuing
a new parser every time we change the grammar, therefore want to limit things
to people who are "inner circle" enough that we generally are consulting with
 them on each such grammar change anyway.  This is virtually synonymous with
the people who write lots of Lojban text or review lots of Lojban text, of

Saying this, the parser is written in fairly portable C, and both Unix and
Dos versions exist.  At the point when we are ready to make it commerically
 avaiavailable, it would not be much of a problem.  However, the natur eof
Unix systems has caused John to mention that we will probably have to distribute
Unix parsers as source, whereas the MS-Dos parser is better distributed as
executable.  This is to some extent depndent on hardware, as well as on the
nature of the type of people on each platform that will be using it.  More
people working on Unix systems are willing and indeed expectant of hacking
around in the code, while people in the PC world tend to prefer pretty
packaged stuff that works right the first time without us having to mess
with it.  Thus, before we isssue the parser, we have to settle policy issues
about what forms we will issue and support it in, what pricing, how to
 distribuite it.  Some people will argue strongly for a Free Software approach;
say that software is our one product that will make money and the people
on-net are getting more than their share of LLG services (in the form of
the list and elctronic distribution) tyhan they are paying for anyway.
(I don't expect to argue these topics now - they've been argued many times
without real resolution.  Our bent is generally to promote more liberal
distribution, but to try to protect our options - right now we have too few
options to bother making a decision that might have to be changed.

The glosser is among other things an attempt to fix the limitations of the
parser.  It presumes a parser otuput as input, and attempts to give an
English fairly-literal translation.  The limitation is that word-for-word
Lojban is unreadbale unless you know the grammar almost well enough to read
it without the glosser; what it does these days is save you time looking
up wprds and rafsi that you know less than perfectly - but the gloss is so
mindless that you need to knwo what the keywords mean grammatically in order
to put the words together in a useful form.  Nora has in mind improving this
to the level of being a crude Lojban-to-English translator that actually uses
the parse tree to intelligently choose expression form, but she has had
little time for any Lojban work in the last year and other things are
higher priority than the glosser.

The glosser is written in Turbo-Pascal, with a highly specific-to-DOS
screen interface module.  You, Matthew, were at one time given an
opportunity to play with the older LogFlash, which we tried to port to
UNIX, and I don't recall hearing that you had any success.  Nora's later
TP code is more sophisticated and also bulkier since she has built this suite
of common routines that make her programming of human interface virtually
table driven.  That earlier code had very detailed documentation; Nora has not
had time to bother documenting the newer versions (and the large number
of people who didn't effectively use the earlier code and documentation
has strongly unmotivated her to bother with it again.

Again, the glosser will probably eventually reach being a usable product,
and we'll distribute it freely (or sell it).  But not when we have more
important things to do, like get some books published.

We have considered putting the parser up installed in the IRC Lojbot, since
we can live with executable dealing with a specific machine accessible to
many netters, and have a commitment from the LojBot author to support
the thing to keep the load of John and me.  This is still in consideration.
It isn't much of an option for the glosser, though.

Comments are welcome on this whole topic, including people's opinions on
what the policy should be when we set such a policy, but as I said above,
I can't take time to argue the policy now, and I doubt that we are prepared to
make any decisions in the near future - it would be a Board decision, and Nora
hasn't had time to write up minutes from the LogFest meeting, much less
prepare for a new decision meeting, even if the rest of us had time.