Hi Max.

I've fixed the bug related to abstract attribute of the synset (it was always True, because bool("non-empty-string") is always True)

I've also speeded up synset_relations_where by order of 3-4 magnitudes.

This might be a bit an overkill, but it has two advantages.

First is:

Another is that you can rewrite code like this:

def path_to_top(synset):
    spo = []
    for rel in [11, 107, 171, 172, 199, 212, 213]:

with meaningful names, not numbers

I've used wikipedia (https://en.wikipedia.org/wiki/PlWordNet) to decipher 45 of 54 domains listed on Słowosieć.

There might be more: for example, zwz

Can you try to decipher the rest? My Polish isn't too good (yet ))

So, here is my attempt. I've used standard python stack for now, will let you know if it caused any problems

I've tested it on Africa/Afryka with different combinations, all looked sane to me:

for lu in wn.find("Afryka"):
    for i, pth in enumerate(wn.hypernym_paths(lu.synset, full_searh=True, interlingual=True)):
        print(f"{i + 1}: " + "->".join(str(s) for s in pth))

gave me

1: {kontynent.2}->{ląd.1 ziemia.4}->{obszar.1 rejon.3 obręb.1}->{przestrzeń.1}
2: {kontynent.2}->{ląd.1 ziemia.4}->{obszar.1 rejon.3 obręb.1}->{location.1}->{object.1 physical object.1}->{physical entity.1}->{entity.1}
3: {kontynent.2}->{ląd.1 ziemia.4}->{land.4 dry land.1 earth.3 ground.1 solid ground.1 terra firma.1}->{object.1 physical object.1}->{physical entity.1}->{entity.1}

Sorry, I accidentally blacked your file, so now it has more changes than expected. The important one, though is that:

+        # For cases like Instance_Hypernym/Instance_Hyponym
+        for rel in self.relation_types.values():
+            if rel.inverse is not None and rel.inverse.inverse is None:
+                rel.inverse.inverse = rel

Hello.

The next logical step for me is to implement tree traversal and data export. For tree traversal I'd try to stick to the following algorithm:

• Find the true top-level hypernyms for the english and polish (no interlingual hypernymy)
• Calculate number of leaves under each top level hypernym (and/or number of LUs under it)
• For each node calculate the distance from top-level hypernym

To export I'd like to use the information above and pass some callables for filtering to only export particular nodes/rels. For example, I only need first 3-4 levels of the trees for nouns, that has more than X leaves. This way I'll have a way to export and visualize only parts of the trees I need.

Speaking of export, I'm looking into graphviz (to basically lay top level ontology on paper) and ttl, but in the format, that is similar to PWN original TTL export.

I'd like to have your opinion on two things:

• General approach
• How to incorporate that into code. It might be a part of Wordnet class, a separate file (maybe under contrib section), an usage example or a separate script which I/we do or don't publish at all

Hi Max. I've slightly cleaned up your spreadsheet on domains (replaced TODO and dashes with nones and made POSes compatible to UD POS tagset) and wrapped everything into classes. I've also made two rows out of cwytw / cwyt and moved pl description of adj/adv into english one. I made en fields default ones for str method

It's up to you to replace str domains in LexicalUnit with instances of Domain class as it's still ok to compare Domain to str

I've also added support for bz2 in loader helper.

PlWordNet 4.2 comes with a supplementary file (słownik_anotacji_emocjonalnej.csv) containing sentiment annotations for lexical units. Users should be able to load and access sentiment data.

Currently, nothing is done with the description field in Synset and LexicalUnit. Information about the description format comes in PlWordNets readme.

Parsing should be done lazily to avoid slowing down the initial loading of PlWordNet into memory.

Example description:

##K: og. ##D: owoc (wielopestkowiec) jabłoni. [##P: Jabłka są kształtem zbliżone do kuli, z zagłębieniem na szczycie, z którego wystaje ogonek.] {##L: http://pl.wikipedia.org/wiki/Jab%C5%82ko}

Desired behavior:

A new (memoized) method rich_description returns the following dict:

dict(
  qualifier='og.',
  definition='owoc (wielopestkowiec) jabłoni.',
  examples=['Jabłka są kształtem zbliżone do kuli, z zagłębieniem na szczycie, z którego wystaje ogonek'],
  sources=['http://pl.wikipedia.org/wiki/Jab%C5%82ko'])