[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [lojban] Re: parsing with error detection and recovery
On 8/17/06, Jorge Llambías <jjllambias@gmail.com> wrote:
On 8/15/06, Chris Capel <pdf23ds@gmail.com> wrote:
>
> For instance, the morphology rules in the BPFK Peg Morphology[1] will
> only parse consonants that don't appear in invalid consonant clusters.
> If a consonant cluster is invalid, it will stop parsing. But by adding
> error rules for consonants that don't check the validity (that only
> get matched if the ones that do check don't match) or that check for
> specific kinds of invalid pairs, the output of the parser could be
> more likely to finish,
That part seems relatively easy to do:
Define a new top rule:
tolerant-text <- text / text-without-phonotactic-constraints
Make a copy of the full grammar with each rule name tagged with
"-without-phonotactic-constraints".
Eliminate the phonotactic constraints from the second set of rules.
These appear only in a few rules. for example, instead of:
c <- comma* [cC] !h !c !s !x !voiced
you will have:
c-without-phonotactic-constraints <- comma* [cC]
That seems like much more than is necessary. I think this can be done
only with local modifications, and then a post-parse scan. I had in
mind something along the lines of this:
c <- c-valid / c-err
c-valid <- c-core !h !c !s !x !voiced
ERR c-err <- c-core
c-core <- comma* [cC]
(Most of that is folded together by Rats!.) Once parsing is finished,
all you have to do is walk the parse tree looking for any rule marked
with ERR, and display an error message written for the rules that are
found.
> and could tell the user why the cluster is
> invalid.
That may be harder to achieve.
Replace c-err in the above with:
c-err <- c-err-unvoiced-voiced / c-err-excluded-pairs / c-err-no-apos
/ c-err-no-doubles
ERR c-err-unvoiced-voiced <- c-core voiced
ERR c-err-excluded-pairs <- c-core s x
ERR c-err-no-apos <- c-core h
ERR c-err-no-doubles <- c-core c
Then when you find that one of those rules matched, you can display a
quite specific message. I actually think that c-err-* except for
c-err-excluded-pairs should be able to be done once for all
consonants, so we're not talking about septupling the size of the
consonant section, either, maybe just doubling or tripling it.
This shouldn't hurt efficiency too much because all the error rules
can be safely marked transient, and shouldn't normally be invoked
unless there's actually an error.
This are actually pretty substantial changes, and it raises the
possibility that some of the changes might change the behavior of the
grammar on valid texts, which is undesirable. Which is why I was
asking in my original post about what class of changes can be made
without risking this.
And I've been reading the papers on Pappy and Rats! recently, and so I
discovered a few techniques for error reporting without postive error
rules like I do above. I'm sure there's a way to integrate those into
the current grammar. (I think one big improvement would be making
"text -> ... EOF?" into "EOF". That way if you can't finish the input,
the parser reports some sort of error.)
And, to be clear, I'm not proposing that these should be changes in
the official grammar. Ideally, there'd be a second file containing
error rules that's optionally merged into the first one to get all
this error stuff. But if that's not practical, there'd just be an
unofficial version of the grammar with this error stuff in it.
Chris Capel
--
"What is it like to be a bat? What is it like to bat a bee? What is it
like to be a bee being batted? What is it like to be a batted bee?"
-- The Mind's I (Hofstadter, Dennet)