Syntactic tagging (was: Re: software Bible study)
KirkLowery at xc.org
Fri Jul 10 10:56:53 EDT 1998
On Wed, 8 Jul 1998 13:44:48 -0400 B Rocine <596547 at ican.net> wrote:
> Now here's a wish I would like to pass onto our computer text-taggers Dale
> Wheeler, Kirk Lowery, et al.: Tag the boundaries of clauses.
Bryan, the good news is that your wish shall be granted....the bad news
is that we're at least five to seven years away from it. I'm not simply
talking about MORPH (which Dale Wheeler, Alan Groves, I and others are
working on), but having any generally available Hebrew Bible tagged at
the syntactic level.
I assume you don't just want a little code inserted at the clause
boundaries, do you? You'd like to know what the constituents of the
clause are, subject, object, etc.? From a practical viewpoint, manual
encoding is very expensive. Also, as good as anyone might be, human
encoding is inconsistent, and error prone--but we don't necessarily know
where the errors are! If one is doing any kind of statistical analysis
of syntax, we don't know exactly what we're measuring. The other choice
is to do the encoding *algorithmically*, i.e., write a computer program
to do it. This has the advantage of absolutely consistent encoding, for
the program will follow the rules given it. But if one doesn't describe
all the conditions, handle all the ambiguities, the program will
stupidly follow the rules it was given and won't tell the user that
there were situations it couldn't handle.
As we move from the phonological level of a language on up through its
morphology and syntax, the amount of ambiguity and variation increases
exponentially. For example, what do you do with words like ':A$ER and
K.IY which sometime function as constituents of two clauses (a main
clause and a dependent clause) at one and the same time! Where do you
put the clause boundary? In the middle of the word? Then there are
verbless clauses. Frequently embedded in other clauses, distinguishing
their boundaries is a function of the lexicon: you have to know the case
frame arguments of the main clause verb (a semantic feature), in order
to discover which noun phrases can't be linked to it. As it turns out,
the clause initial boundary is much easier than clause final to
distinguish. The algorithms I've worked with generally tend to correctly
determine the beginning of main clauses in a narrative, but tend to miss
embedded and verbless clauses. And poetry! All bets are off there.
And you're right, the definition of things becomes a problem. The great
hassle (and great benefit) of using computers for analysis of texts is
that a computer won't let one ignore problems. For example, L + inf cs.
Is it a dependent clause or a noun phrase, especially when our western
languages require us to translate it as a dependent clause if it
expresses purpose? Or distinguishing between the nominal use of the
participle and its verbal use? One has to do something or the computer
Well, that gives you a taste of the problem. It's not impossible to
solve, but we have to bootstrap our way upwards. Vince DeCaen and I have
been trying to conceptualize a software tool which would allow us to
experiment in real time with algorithms for syntactic analysis, so that
we could quickly cycle through various possibilities quickly and
immediately see their effect.
The problem is going to be solved, that's for sure. I hope you can see
that a lot of thought and effort has already been given to it, but it is
far from trivial to solve. (And we all have our day jobs! ;-)
> While I'm at it, another wish: Tag the boundary between narration and
> quoted speech. How beneficial! The resulting research would make it seem
> like we're learning Hebrew all over again.
What am I, your fairy godmother? (I wish! ;-)
This will take *real* magic: imagine all the problems of citation
(direct quote) boundaries in the prophets!
Kirk E. Lowery, Ph.D. "A scholar is just a library's way of making
516 Argyle Ave. another library."
Ambler, PA 19002 --Daniel Dennett, "Darwin's Dangerous Idea"
More information about the b-hebrew