TedEval: An Architecture for Cross-Experiment Parse Evaluation

User Guide

Using TedEval

Single Experiment Evaluation:
Usage:
  java -jar tedeval.jar -p parse_file -g gold_file -o eval_file -format conll 
Pairwise Experiment Evaluation:
Usage:
  java -jar tedeval.jar -p1 parse_file1 -g1 gold_file1 -o1 eval_file1 -p2 parse_file2 -g2 gold_file2 -o2 eval_file2 -format conll 

Using StatSignTest

Statistical significance testing for a pair of experiments which use the same native gold or the same generalized gold file:
Usage:
  java -jar statsigntest.jar  -i eval_file1.ted -i eval_file2.ted -o results.statsign -n 9999

Command Line Options

TedEval

Usage:
  java -jar tedeval.jar -h for more help and options

help                      ( -h) : Show options
version                   ( -v) : Show version
-----------------------------------------------------------------------------
Single experiment evaluation:
gold_file                  (-g) : path to gold-parses file (single experiment)
parsed_file                (-p) : path to parse-hypotheses (single experiment)
output_file                (-o) : path to result file (single experiment), default file name is eval

Pairwise Experiment evaluation:
first_gold_file           (-g1) : path to gold-parses file (the first of two experiments)
first_parsed_file         (-p1) : path to parse-hypotheses (the first of two experiments)
second_gold_file          (-g2) : path to gold-parses file (the second of two experiments)
second_parsed_file        (-p2) : path to parse-hypotheses (the second of two experiments)
first_result_file         (-o1) : path to result file (the first of two experiments), default file name is eval1
second_result_file        (-o2) : path to result file (the second of two experiments), default file name is eval2
-----------------------------------------------------------------------------
file_format           (-format) : the default is labeled ptb-like bracketed format (ignoring anything after the dash)
  -format conll                   the conll-x format
  -format bracketed               labeled ptb-like bracketed format (empty elements not allowed, ignoring anything after the dash)
-----------------------------------------------------------------------------
labeling_flag     (-unlabeled ) : use the unlabeled measure instead of the default labeled one
avg_format_flag       (-micro ) : use the micro average instead of the default macro average one
-----------------------------------------------------------------------------
normalization_factor    (-norm) : 

StatSignTest

Usage:
  java -jar statsigntest.jar -h for more help and options

help                         (-h) : Show options
version                      (-v) : Show version
-----------------------------------------------------------------------------
ted_file_1                   (-i) : Path to the first output-file from tedeval
ted_file_2                   (-i) : Path to the second output-file from tedeval
output_file                  (-o) : path to result file                        
iterations                   (-n) : number of iterations (default: 10 000)
-----------------------------------------------------------------------------