naf2conll

Script to convert coreference data in NAF format to CoNLL format.

!! NB !! At the moment, this script only supports the following columns:

• 1: Document ID
• 3: Word number
• 4: Word itself
• 12: Coreference

The following CoNLL columns are supported by NAF, but are not (yet) processed (correctly) by this script:

• 5: POS tag
• 6: constituency tree
• ...?
• 11: named entities

See CoNLL-specification.md for an extensive description of the CoNLL format.

Usage

naf2conll.py

To automatically find all (sub)folders that contain NAF files and convert all data in those folders, run:

naf2conll.py path/to/output_dir -d path/to/some/folder [-d path/to/another/folder ...]

To only convert one file, run:

naf2conll.py path/to/output.conll path/to/input.naf

Columns of CoNLL output

By default only Column 1, 3, 4 and 12 are output.

If you choose to output more columns, the following values and place-holders are used.

[1]: The reference spans are not closed in the correct order if they end at the same word. The following is an example of output from naf2conll.py:

          (10
            -
      (52|(55
          52)
            -
10)|55)|(133)

While pedantically correct would be:

          (10
            -
      (55|(52
          52)
            -
(133)|55)|10)

Issues

• 'on_missing' config key is not validated before use
• Raise an error when there is no coref layer in extract_coref_sets