similar vs verb_group

[arademaker]

The wn30.xml from https://github.com/bond-lab/omw-data/tree/main/wns/pwn30 has similar used in place of verb_group that is documented in https://globalwordnet.github.io/schemas/. The DTD files do not contain the verb_group:

    <Synset id="pwn-02136271-v" ili="i32435" partOfSpeech="v" dc:subject="verb.perception">
      <Definition >to throw or bend back (from a surface); &quot;Sound is reflected well in this auditorium&quot;</Definition>
      <SynsetRelation relType="domain_topic" target="pwn-06094774-n" />
      <SynsetRelation relType="similar" target="pwn-02136533-v" />
      <SynsetRelation relType="hyponym" target="pwn-02136533-v" />
      <SynsetRelation relType="hyponym" target="pwn-02766925-v" />
    </Synset>

[goodmami]

similar is also in the list of synset relations, but not in the list for senses. @arademaker are you saying that relations that are originally verb_group in the PWN 3.0 are replaced with similar for the wn30.xml file that Francis produces?

I was unaware of verb_group, probably because I'm working from the DTDs, but it has been in the documentation since at least 2016. @jmccrae is it just missing from the DTDs?

[jmccrae]

We decided that similar was the name of the property that Princeton called 'verb group'. This was merged with the 'similar to' property for adjectives in PWN. It was decided that they were essentially the same thing in different part-of-speech.

[goodmami]

Should mention of verb_group then be removed from index.md and example.xml? For example:

https://github.com/globalwordnet/schemas/blob/cab56797473b289186d187515bfdcb7ce0ccfc55/example.xml#L55

[jmccrae]

It probably should! Thanks for catching this.

[arademaker]

similar is also in the list of synset relations, but not in the list for senses. @arademaker are you saying that relations that are originally verb_group in the PWN 3.0 are replaced with similar for the wn30.xml file that Francis produces?

yes

[arademaker]

OK, so Verb Group ($) is merged with Similar to (&) from https://wordnet.princeton.edu/documentation/wninput5wn, right?

[arademaker]

The page https://globalwordnet.github.io/schemas/ also says that verb_group is a relation between senses, which is not true. The Princeton verb group is a relation between synsets. The name similar to was also simplified to similar only, right? But I found now in https://globalwordnet.github.io/gwadoc/#similar the comment about the merging. I believe the page https://globalwordnet.github.io/schemas/ should just point to the https://globalwordnet.github.io/gwadoc/ so we can avoid these inconsistencies.

[1313ou]

The 'similar' relation is central to the adjective model in Princeton wordnet: it is exclusive to adjectives and tags the relation between a core adjective (a cluster head that has an antonym) and its satellites and vice-versa. It is NOT used with another meaning (and specifically it does not hold among synset members)

here is the count:

similar a s 10717  (from head to satellite)
similar s a 10717  (from satellite to head)

a=adjective cluster head s=satellite adjective

The 'verb_group' relation is indeed semantic (though 2 instances are lexical, probably erroneously). It looks like similarity for adjectives in that "satellites" (block up, block off, blockade, barricade, bar, stop) relate to a head (block) thus forming a verb group. But it is not a defining relation among verbs the way similarity is among adjs.

Should they be merged ? If so the new relation should not be described as a Princeton WordNet property without a caveat.

[arademaker]

I agree @1313ou. I would vote not to merge. But you are right that if I they are merged. We need to explicit declare as a different relation not from PWN.