add a check function

I would like to add a module (check.py) that checks for various issues with a wordnet. The goal is to help developers avoid producing wordnets with structural problems. It would duplicate some of the functionality of the validate.py in the https://github.com/globalwordnet/english-wordnet/ and the checks of the https://github.com/globalwordnet/OMW but make them more widely available.

[goodmami]

I would have been hesitant, initially, as this kind of operation is probably only useful for wordnet authors, but there have been many easy-to-detect-if-you-know-what-to-look-for kind of issues with some wordnets of late, so it makes sense to make the tests easy to run. It would be useful if each test could be run individually. It would also be useful if the tests could be run from the command line.

Here’s an initial list of tests:

• Every lexical entry has at least 1 sense
• Every synset has at least 1 member
• Synsets should have the same POS as their members (but s on a synset implies a on the member)
• Synsets should have the same POS as their hypernyms
• Inverse relations are present (e.g., hypernym/hyponym)
• Each ILI is assigned to at most one synset in a lexicon

Some simple things are already checked by wn.lmf or when adding a lexicon to the database:

• POS values are within n, v, a, r, and s (currently a warning is issued when reading the XML file)
• IDs are unique (sqite3 raises an integrity error on the uniqueness violation)
• Relation names are valid (currently wn.lmr throws an error if not)

Some things are not checked at this level but should be:

• POS on Lemma should not include s

Feel free to add more checks.

[goodmami]

After recent discussions I’m starting to have a nagging idea that this should be a separate package. There are some benefits:

• Wn is leaner for normal users
• omw-data can use it for validation without waiting for Wn 0.9.0 (a one-time benefit)
• Dependencies like NetworkX can be added without complications