cltl/formatconversions

Several conversions between formats that are commonly used by our tools

FormatConversions

Several conversions between formats that are commonly used by our tools

This repository provides several conversion scripts between commonly used formats.

For conversions between NAF and FoLiA, see https://github.com/cltl/NAFFoLiAPy

Included conversions:

• SONAR 2 NAF: tokens, terms and srl
• NAF 2 CAT: tokens, for specific tasks factuality and Dutch SRL (see naf2cat/README.md)
• CAT 2 CAT: cat2cat markables: turns specific attributes of a markable into markables (for comparison and quick inspections)

LICENSE:

All software included in this repository is freely available under the Apache License, Version 2.0 (http://www.apache.org/licenses/)

formatconversions's Issues

Better mmax2conll conversion

@MPvHarmelen The mmax2conll conversion in this repository either (a) throws away most data, or (b) generates invalid output. I've made a new conversion script to address those issues: https://github.com/andreasvc/dutchcoref/blob/master/mmaxconll.py

License missing

mmax2raw.py ignores sentence boundaries

Alpino expects every sentence of tokenized input to be on a separate line (Alpino User Guide, Section 2.5 on page 7).
mmax2raw.py, however, totally ignores this and outputs all words of a MMAX file on a single line of the output file:

FormatConversions/mmax2conll/mmax2raw.py

Line 81 in 6810be2

fd.write(" ".join(word['word'] for word in words))

Fixing this could improve tagging and therefore coreference results.
Alpino also sometimes seems to output cyclic graphs as dependency-"trees". Maybe this is also caused by this issue.