Re: Corpora: Bralich and Ergo

Philip A. Bralich, Ph.D. (bralich@hawaii.edu)
Thu, 19 Feb 1998 08:34:52 -1000

At 11:43 PM 2/18/98 -1000, Jem Clear wrote:
>Surely Philip A. Bralich, Ph.D. is being deliberately crass in order to
>draw attention to himself and Ergo, isn't he?

I think it is not so much crassness that I am looking at as presenting
a very real and challenging problem to a community of people who
are largely overworked. If I didn't put a bit of an edge on my
posts the point I have to make would not be noticed. Granted some
may find it difficult to entertain because of the challenging
manner in which it is phrased, but there are some very real
issues that need to be addressed:

1) The Penn Treebank II guidelines is an accepted standard yet
no one besides Ergo can generate the trees and labeled brackets.
I guess Satoshi Sakine has something does less than we do with
that same genre, but his program requires a c compiler to open
and is a bit confusing. I can report more on this if you'd like.
2) Mainstream theories are the worst at generating these trees
even though they are the central focuse of research
3) The standards which I proposed were deliberately chosen to
be standards which anyone would accept as bare requirements
for any NLP system. Asking a parser to identify questions and
statements and to do so in the Penn Treebank format is not
asking much. I am sure MANY readers believed that most theories
could do this already and that it is news that they cannot.
4) any genuine parse of the constituent structure of a string should
trivially be able to generate all the items in the standards or
it is reasonable to argue the basic work has not been done.

Finally, the fact that I am associated with the private sector rather
than academia does raise some interesting questions. We could just
as well say that any post is just trying to draw attention to the
individual and his institution, and we might want to ask, didn't
we sign up to find out about developments in the field whatever their
source.

Please, look closely at those standards and then ask yourself if you
did not assume this could already be handled by mainstream theories, and
then ask yourself if it is acceptable that they cannot.

Phil Bralich

Philip A. Bralich, Ph.D.
President and CEO
Ergo Linguistic Technologies
2800 Woodlawn Drive, Suite 175
Honolulu, HI 96822

Tel: (808)539-3920
Fax: (808)539-3924