[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[lojban] Re: peg experiment with changing clauses to better support sa
On 11/23/08, Jorge Llambías <jjllambias@gmail.com> wrote:
> I didn't check the details, but if what you are doing is a SA that
> erases everything to the previous appearance of the same selma'o,
> Robin already had that working in a previous version of his PEG.
Yes mostly that because I noticed in his notes that he didn't have
BRIVLA+SA or CMENE+SA working. In addition, by inspection and testing,
he doesn't have many others working either . Like in #jbosnu : " i sa
i" worked but " ui sa ui" failed.
I didn't know about any earlier working version. I also don't know any
url where I can find earlier working versions.
> The main problem with SA is not how to write the grammar though, the
> problem is deciding what exactly we want it to do. SA-selma'o is not
> very pretty. SA-construct (where "construct" can be "sumti", "selbri",
> and a few others) is only slightly better.
If I understand you; Yes my approach is basicly SA-selma'o, except my
naming is backwards of that -- ${sm}-SA . I decomposed the problem
into about 124 easier subproblems(hopefully my selma'o count is right)
; about 100 of the 124 subproblems share the same template. So I have
A-SA, BAhE-SA, BAI-SA, BE-SA, BEhO-SA, BEI-SA, ... , ... , ...,
ZEI-SA, ZI-SA, ZIhE-SA, ZO-SA, ZOhU-SA, ZOI-SA, BRIVLA-SA, and
CMEVLA-SA . Not sure why you would have sumti-SA and selbri-SA, so I
think we are maybe talking about completely different things. Also for
me "SA-selma'o" isn't one rule it's over a hundred. ${sm} is the
varible that you substitute in for each selma'o that can use the
template.
Each selma'o-SA is only used in two spots:
1) In selma'o-clause stuff
2)recursively to allow " x ... x ... sa sa x" and friends.
If it wasn't for the recursive definition then each selma'o or
pseudo-selma'o could just put it's relevant selma'o-SA directly into
it's self; Each selma'o clause handles it's own sa and su issues.
Other reasons to have it separate is maybe readability, and if you
wanted to have a SA-clause by recomposing the rule from the decomposed
subproblems. I, in the comments, gave an example of how you could
reform a universal "SA-clause" but noted that you don't really ever
need the answer to the general question if a piece of text is part of
*any* SA clause just one of the particular subproblems.
I also noted that most(about 100/124 or about 80%) of the $[sm}-clause
and $[sm}-SA can be formed by using the same template. ${sm}-clause is
already very boiler plate stuff in rlpowell's peg grammar. The ones
which were not able to use the same template were SI, SA, SU, ZO, ZOI,
LOhU, LEhU, ZEI, BU, LAhO, FAhO, NIhO, LU, TUhE, TO, ui, cai, nai,
da'o, fu'o , BY, and ba'e . A lot of those are the so called "magic"
words. That also leaves around 100 selma'o that can use the same
template. Another change slightly unrelated to the sa fixes is that
some of the magic words clauses that quote stuff, now also match and
consume the text so quoted.
Also I think correctness over prettyness, might be a priority; at
least have a complete but ugly version and a pretty but incomplete
version. Also the peg grammar is a more exacting specification for
"deciding what exactly we want it to do". Right now there is some hand
waving imho.
PS I noticed that the first version of the below should probably be
changed into the second version of the below. So that "x ... z .... sa
z .... sa x" works. That should allow some SA nesting.
;old nonnested
${sm}-clause1 <- spaces? (${sm}-SA)* (spaces? SA)* spaces? !anticmavo ${sm}
${sm}-SA <- spaces? !anticmavo ${sm} (quote-clauses / !${sm} !SA !FAhO
any-word)* ${sm}-SA? SA &( (spaces? SA)* spaces? ${sm})
${sm}-clause <- BAhE-clause? ${sm}-clause1 spaces? post-clause
;should be nestable
;can maybe be slightly optimized in case of
; hopefully rare multiple sa's in row within a inner sa thing
that should be skipped over
${sm}-clause1 <- spaces? (${sm}-SA)* (spaces? SA)* spaces? !anticmavo ${sm}
${sm}-SA-end <- (spaces? SA)* spaces? ${sm}
${sm}-SA <- spaces? !anticmavo ${sm} (quote-clauses / spaces?
!${sm}-SA-end !FAhO any-word)* ${sm}-SA? SA &(${sm}-SA-end)
${sm}-clause <- BAhE-clause? ${sm}-clause1 spaces? post-clause
I'll also have to fix up the other one which don't follow the
template, and finish the BY-clause, BY-SA stuff. plus do more review
to see what else I've missed.
PPS
wc cmavo_selmaho.txt
122 122 1213 cmavo_selmaho.txt
122 + the brivla pseudo selma'o and the cmevla pseudo selma'o is how I
derived the number 124. I might be off a tiny bit.
To unsubscribe from this list, send mail to lojban-list-request@lojban.org
with the subject unsubscribe, or go to http://www.lojban.org/lsg2/, or if
you're really stuck, send mail to secretary@lojban.org for help.