%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

CoNLL Shared Task on Semantic Role Labeling
	
Authors: 
	Xavier Carreras and Lluis Marquez
	TALP Research Center
	Technical University of Catalonia (UPC)

Contact: carreras@lsi.upc.edu


PropBank Version : PropBank-1.0

Created :  January 2004
Revised :  March 2005

Revised 2005/12/01 : Added options to process NomBank; see below

This software is distributed to support the CoNLL-2005 Shared Task. It
is free for research and educational purposes. 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


+ 
+  LINKING PropBank TO THE TreeBank : link_tbpb.pl 
+------------------------------------------------------------

The link_tbpb.pl script takes a WSJ tree and PropBank annotations of a
sentence, and outputs the sentece in the CoNLL-2005 shared task column
format, with a column for each predicate of the sentence representing
the arguments of the predicate.

The script expects two parameters: the name of the file containing WSJ
trees, and the name of the file containing the PropBank annotations
corresponding to the trees. The PB annotations must be ordered with
respect to sentence number (2nd column) and predicate position in each
sentence (3rd column).

In the srlconll package, the directory data contains two files, WSJ
trees and PB annotations for the 1500 WSJ file. With such files, we
can play with the link_tbpb.pl script : 

$  link_tbpb.pl data/tb_1500 data/pb_1500 |& less
<...> 

Moreover, a number of options can be specified to control the behavior
of link_tbpb. Executing the script without parameters lists the
options with a brief explanation. The verbosity level can be
controlled by the -v option. The messages always are output through
the STDERR channel. Apart from this, the options control which
propositions are processed (or, actually, which ones are skipped). 

When generating the data for the CoNLL-2005 shared task, only the
options related to traces were used: -st and -ft. The command was: 

$  link_tbpb.pl -noi -st -ft data/tb_1500 data/pb_1500 > data/srltask_1500
<...> 


PREPROCESSING OF THE PropBank "prop.txt" file 
--------------------------------------------------

The PropBank project distributes the propositional annotations for the
WSJ TreeBank in a single file, containing the annotations for all the
WSJ sections and files of the section. However, the link_tbpb.pl
script expects in the second argument the propositions corresponding
to a WSJ TreeBank file. Thus, the PropBank "prop.txt" file must be
split into the TreeBank file structure. 

The "nb_propsbyfiles.sh" script performs this operation for the WSJ
sections which compose the CoNLL datasets. 


+
+  NOMBANK
+-------------------------

- Before using "link_pbtb.pl", it is necessary to split the nombank
propositions file into many files, following WSJ TreeBank structure.
To do so, use the "nb_probsbyfiles.sh" script under "bin".

- With link_pbtb.pl use option "-nb" to read nombank entries. E.g. : 

$ link_pbtb.pl -nb data/tb_1500 data/nb_1500