NOTE: This repository is currently not managed. Please check the Issues page for bug reports. I will try to fix it later. Sorry for any inconvenience.

This repository contains two scripts:

• conlleval_perl.py: the Python equivalent of the Perl script conlleval, which can be used for measuring the performance of a system that has processed the CoNLL-2000 shared task data.

• conlleval.py: a slight modification on the above script, so that it can be imported and used elsewhere. You will find import evaluate from conlleval useful.

For more information on the original Perl script, see http://www.cnts.ua.ac.be/conll2000/chunking/output.html.

Both scripts can be used to evaluate from a supported file.

Read from output.txt and print the results to the console:

  python conlleval.py < output.txt

or save the results in result.txt:

  python conlleval.py < output.txt > result.txt

And the result is:

   processed 961 tokens with 459 phrases; found: 539 phrases; correct: 371.
   accuracy:  84.08%; precision:  68.83%; recall:  80.83%; FB1:  74.35
                ADJP: precision:   0.00%; recall:   0.00%; FB1:   0.00
                ADVP: precision:  45.45%; recall:  62.50%; FB1:  52.63
                  NP: precision:  64.98%; recall:  78.63%; FB1:  71.16
                  PP: precision:  83.18%; recall:  98.89%; FB1:  90.36
                SBAR: precision:  66.67%; recall:  33.33%; FB1:  44.44
                  VP: precision:  69.00%; recall:  79.31%; FB1:  73.80

Options for conlleval_perl.py (the same as the original Perl script):

• -l: Generate output as part of a LaTeX table. The definition of the table can be found in an example document: latex ps pdf

• -d char: On each line, use this character rather than whitespace (or tab) as delimiter between tokens

• -r: Assume raw output tokens, that is without the prefixes B- and I-. In this case each word will be counted as one chunk.

• -o token: Use token as output tag for items that are outside of chunks or other classes. This option only works when -r is used as well. The default value for the outside output tag is O.

Usage for conlleval.py:

from conlleval import evaluate

# print out the table as above
evaluate(true_tags, pred_tags, verbose=True) 

# calculate overall metrics
prec, rec, f1 = evaluate(true_tags, pred_tags, verbose=False)

NOTE: This script can be used with IOB2 or IOBES tagging scheme. If you are using a different scheme, please convert to IOB2 or IOBES.

EDIT: There has been report that IOB2 support is broken (see Issues page). This has not been fixed yet.

For an example of data format to be used with this script, check out the accompanied output.txt file in this repository, or the original source at http://www.cnts.ua.ac.be/conll2000/chunking/output.txt.gz.

Sentences are separated by empty lines. Each line contains at least three columns, seperated by whitespaces (or a character specified in -d). The second last column is the chunk tag according to the corpus, and the last column is the predicted chunk tag. The other columns are ignored in evaluation.

Example:

   Boeing NNP B-NP I-NP
   's POS B-NP B-NP
   747 CD I-NP I-NP
   jetliners NNS I-NP I-NP
   . . O O
   
   Rockwell NNP B-NP I-NP
   said VBD B-VP B-VP
   the DT B-NP B-NP
   agreement NN I-NP I-NP