~/research/emptynodes/README 

Mark Johnson, 17th July 2002


Usage:
-----

restorer restorer.dat < parser-trees > parser-trees-with-empty-nodes

Note: You may want to give an absolute path name to restorer.dat so
that the restorer program can be called from any directory.


Description:
-----------

This directory contains programs for inserting and coindexing empty
nodes in parse trees, as described in my ACL 2002 paper.  My original
program (in ~/research/emptynodes) extracted patterns, restored empty
nodes and evaluated the resulting trees all in one process.  

I'm creating the programs in this directory from those programs, but
I'm doing it about 9 months after those programs were written, so I
hope I'm not introducing any bugs!

I have only tested this program under GCC 3.1, but I think it should
compile under GCC 3.0 as well.

While making these programs, I noticed that my original code used
the Wordnet morphological analysis routines in identifying transitive
verbs, and that contrary to what I said in the paper, it was the
morphological roots of verbs that were classified as transitive, not
the inflected forms.

The Makefile can be used to build all of the programs mentioned below.
The programs and all code contained in this directory are in the
public domain, but I request acknowledgement in all other programs or
publications that utilize any of the code contained in this directory.


Programs and data files:
-----------------------

extractor:  Extracts patterns from treebank trees.  Reads from 
standard input; writes transitive verbs and patterns onto standard
output.

restorer: Restores and coindexes empty nodes in Charniak-style parses.
Reads parses from standard input and writes parses with coindexed
empty nodes onto standard output.  It expects a single argument, which
is the name of a patterns file created by the extractor program.

restorer.dat: Data file required by restorer.  It contains a list of
transitive verbs and the patterns used to restore and coindex empty
nodes.