Part of Speech 's'

[fredsonaguiar]

In the PWN:3.0 and PWN:3.1 data, one may find occurrences of PartsOfSpeech s, such as in

[...]
<Lemma writtenForm="well-connected" partOfSpeech="s" />
<Lemma writtenForm="humongous" partOfSpeech="s" />
<Lemma writtenForm="trimotored" partOfSpeech="s" />
<Lemma writtenForm="right-hand" partOfSpeech="s" />
[...]

In https://wordnet.princeton.edu/documentation/lexnames5wn the POS described are: NOUN; VERB; ADJECTIVE; and ADVERB.

That might have been a misleading with the ss_types from https://wordnet.princeton.edu/documentation/wndb5wn: NOUN (n); VERB (v); ADJECTIVE (a) ; ADJECTIVE SATELLITE (s); and ADVERB (r).

[goodmami]

So lemmas with partOfSpeech="s" should probably be partOfSpeech="a"?

It might also be good for https://github.com/globalwordnet/schemas/ to change the partOfSpeech attribute on <Synset> in WN-LMF to sstype or something, but being a backwards-incompatible change it might be harder to get that through.

[fcbond]

Good point, @jmccrae what do you think? In the original OMW, we actually changed them all to 'a', even at the synset level (and lost a little information).

…

On Tue, Jul 20, 2021 at 10:14 PM Michael Wayne Goodman < ***@***.***> wrote: So lemmas with partOfSpeech="s" should probably be partOfSpeech="a"? It might also be good for https://github.com/globalwordnet/schemas/ to change the partOfSpeech attribute on <Synset> in WN-LMF to sstype or something, but being a backwards-incompatible change it might be harder to get that through. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#9 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAIPZRQML3DQ4Z3MDH2CMOLTYV67HANCNFSM5AV2XZRQ> .

-- Francis Bond <http://www3.ntu.edu.sg/home/fcbond/> Division of Linguistics and Multilingual Studies Nanyang Technological University

[jmccrae]

Satellite is a fundamentally different part-of-speech in the structure of Princeton WordNet and certain parts of the structure, as well as related technical implementations (sense key calculation), rely on this. Linguistically, IMHO, it is not a sensible distinction and it leads to all kinds of issues (see OEWN's Issue globalwordnet/english-wordnet#35 for the start of the rabbit hole).

My opinion is that PWN uses 'satellite' as a part-of-speech value on the same level as 'noun' and this should be respected in any export of PWN. OEWN may at some point, I hope, get round to removing this distinction.

My opinion on adding sstype to the schema is that this just leads to the part-of-speech being duplicated for most synsets. It is easier for implementations just to understand s as a value for part-of-speech.

[goodmami]

@jmccrae I think you're addressing a different issue than what @FredsoNerd raised.

Part-of-speech is, linguistically, a syntactic property and not a semantic property, and therefore in WNDB lexical entries (in the index files) have a pos field while synsets (in the data files) have a ss_type field even though their values are mostly the same (compare Index File Format and Data File Format here: https://wordnet.princeton.edu/documentation/wndb5wn). Both pos and ss_type can have n, v, a, and r, but only ss_type may have the fifth: s.

@FredsoNerd was pointing out that partOfSpeech="s" is appearing on <Lemma> elements, not (just) on <Synset> elements. For instance:

$ grep well-connected *.adj
data.adj:00567414 00 s 01 well-connected 0 001 & 00566099 a 0000 | connected by blood or close acquaintance with people of wealth or social position; "a well-connected Edinburgh family"  
index.adj:well-connected a 1 1 & 1 0 00567414

Note that the ss_type is s in data.adj but the pos is a in index.adj, whereas in the WN-LMF conversion of PWN (and OEWN) the lemma seems to copy the ss_type value from the data file instead of using the pos value from the index file:

$ grep  '"well-connected"' wn30.xml 
      <Lemma writtenForm="well-connected" partOfSpeech="s" />

Unless I've misunderstood the WNDB format, it appears this is an error, and the <Lemma> elements should have partOfSpeech="a" while their corresponding synsets retain the partOfSpeech="s" attribute.

[jmccrae]

I guess that is an interpretation... the schema description of the format is kind of clear that that is not the interpretation we have made so far:
https://globalwordnet.github.io/schemas/
I was also checking the first paper we wrote (https://aclanthology.org/2016.gwc-1.9.pdf) and it says that s is for sentence!
On the other hand, I don't think it breaks anything to use a different value of partOfSpeech for the lemma and synset, although I would not be in favour of this in projects like OEWN, as I hope we can remove satellites entirely in the future.

[arademaker]

Satellite is a fundamentally different part-of-speech in the structure of Princeton WordNet and certain parts of the structure, as well as related technical implementations (sense key calculation), rely on this.

But PWN never claimed that s is a part of speech. That is the point, synset types are not part of speech...

[arademaker]

I was also checking the first paper we wrote (https://aclanthology.org/2016.gwc-1.9.pdf) and it says that s is for sentence!

Oh, this was the interpretation at that time or a typo?

[goodmami]

Also, regardless of the interpretation and what we want to do moving forward, we should be careful that we're not changing or losing information in PWN, which is fixed (see #5 (comment)). Currently there's no info loss (in the entropy sense) as we can replace partOfSpeech="s" with ...="a" on Lemmas, but it has been changed (imagine a user searching for "well-connected" with POS "a" and finding nothing).

[jmccrae]

I think s for sentence was a mistake, I guess it was from me.