readme.sec23

来自「中心词驱动的短语结构句法分析器。该模型考虑了跟随介词短语的名词短语的中心词的作用」· SEC23 代码 · 共 38 行

# README.sec23 gives the command lines to recreate the scored files for
# section 23 under the parser
#
# It can be run as a script: as input are the three files
# sec23.model1 sec23.model2 and sec23.model3 which were created by
# parsing examples/sec23.tagger.
#
# As output are sec23.model1.scores sec23.model2.scores sec23.model3.scores
# which give the output of the evalb scorer
#
# You will need to download the EVALB scorer from 
#   ftp://ftp.cs.nyu.edu/pub/local/sekine/EVALB.tar.gz
#
# Steps taken in creating the scoring files:
#
# Models 2 and 3 of the parser fail on a few sentences in section 23,
# due to the chart becoming too large. For this reason "merge.pl" is
# used to place Model 1 output in the sec23.model2.merged file where model
# 2 fails to produce output. Similarly, sec23.model2.merged is used as
# a back-up for the sec23.model3 file, producing sec23.model3.merged
#
# proc_pout.prl creates treebank-scorer friendly output.
# For the purporse of cleanSec23.prl, see the comments at the top of this
# file.
#
# Finally, the evalb scorer is used to create the .scores files


./merge.prl sec23.model2 sec23.model1 > sec23.model2.merged
./merge.prl sec23.model3 sec23.model2.merged > sec23.model3.merged
cat sec23.model1 | ./proc_pout.prl | ./cleanSec23.pl > sec23.model1.final
cat sec23.model2.merged | ./proc_pout.prl | ./cleanSec23.pl > sec23.model2.final
cat sec23.model3.merged | ./proc_pout.prl | ./cleanSec23.pl > sec23.model3.final

../../../PARSEVAL/EVALB/evalb -p ../../../PARSEVAL/EVALB/COLLINS.prm sec23.tgrep.clean sec23.model1.final > sec23.model1.scores
../../../PARSEVAL/EVALB/evalb -p ../../../PARSEVAL/EVALB/COLLINS.prm sec23.tgrep.clean sec23.model2.final > sec23.model2.scores
../../../PARSEVAL/EVALB/evalb -p ../../../PARSEVAL/EVALB/COLLINS.prm sec23.tgrep.clean sec23.model3.final > sec23.model3.scores