Jump to content

Wikidata talk:Lexicographical data

Add topic
From Wikidata
Latest comment: 1 month ago by ArthurPSmith in topic One dictionary as source for lexeme

Wikidata:Lexicographical data


What can be a toponym: a Lexeme or a Sense?

[edit]
Latest comment: 2 months ago12 comments5 people in discussion

How to best model words as toponym (Q7884789) was already discussed multiple times, with the most impactful being #Values for lexicographic category by @Fnielsen in 2020, where they suggested cleaning up the lexicographic categories of lexemes and moving toponym (Q7884789) to instance of (P31). This query shows that this statement is set on ~20k Lexemes, and ~100 Senses. I am not really unfamiliar with this hundred of Senses, for I am the one who did them.

In my opinion, a Lexeme cannot be a toponym (Q7884789), rather its Sense(s). As an example, let's take the French proper name Rhône (L26760). Here on Wikidata, it has two meanings: (1) the department (Rhône (Q46130)) and (2) the river (Rhône (Q602)). Does this mean that the Lexeme is both (P31) a toponym (Q7884789) (1) and potamonym (Q21661721) (2)?

Individual Sense(s) are what carries the meanings — therefore, imo, they are toponyms, potamonyms, choronyms… Why would a purely abstract entity (the lexeme) hold something that pertains to its meanings as its nature? It sounds weird…

Any opinions? Poslovitch (talk) 15:05, 19 March 2026 (UTC)Reply

@Poslovitch: You're on the right track here; if an 'instance of' statement applies to only some of the senses of a lexeme, then it should be moved to those senses, but if it applies to all senses of a lexeme, then there is no harm retaining it at the top level. Mahir256 (talk) 18:14, 19 March 2026 (UTC)Reply
@Mahir256 : then you're expected to query the P31s of both senses and lexemes, and infer that "one" (or a few) of the P31 of the lexeme applies to all its senses? I disagree with this, but if that's what we're going for: where is it/should it be documented? Poslovitch (talk) 09:16, 20 March 2026 (UTC)Reply
@Poslovitch: Your disagreement having been acknowledged, I would welcome an effort on your part to move the remaining top-level "instance of" "-nym" statements to the appropriate senses (perhaps adding some on the way if not present on those lexemes). The answer to "where should it be documented"--and perhaps "it" could be expanded to the general nature of where certain statements should go--is on one of the subpages Wikidata:Lexicographical data/Documentation/Lexeme statements or Wikidata:Lexicographical data/Documentation/Senses as appropriate. Mahir256 (talk) 11:02, 20 March 2026 (UTC)Reply
@Mahir256 I will gladly help! My main worry was that I was overlooking something, somewhere, where it could have made sense. I will also edit the documentations. Thanks for your answer :) Poslovitch (talk) 11:53, 20 March 2026 (UTC)Reply
@Mahir256, Poslovitch: I'm not entirely sure (I see a lot of pros and cons and even more uncertainties) but my naive first impression is that indeed it belong to the sense level and maybe it would be best to move all instance of (P31)toponym (Q7884789) (and subclasses) to senses, or maybe at the very least point to sense(s) (perhaps only if there is several?). Luckily - if we agree on a model - it can easily be done with QuickStatements, but does it worth the trouble (especially as 3/4 of the current toponyms only have one sense, but of course there is outlier like the famous Janowo (L1473010) where it applies to all 91 senses). Also, I'm wondering if it really should be explicit as you can already infer that a noun is a toponym from it's value but maybe a bit of redundancy and explicitness is not that bad here? (sorry if I don't bring real answer, I a bit puzzled and I would love more point of view on this) Cheers, VIGNERON (talk) 17:39, 20 March 2026 (UTC)Reply
I may be wrong, but dictionaries regards words as toponyms... I agree with all above arguments, but it may become more difficult to query toponyms if these claims would be at Sense level... Infovarius (talk) 11:25, 26 April 2026 (UTC)Reply
@Infovarius, @Mahir256, @Poslovitch and @VIGNERON: what do you think about introducing a dedicated property instead of using instance of (P31) on Senses? For example, a property such as noun type (or a better name) could describe the semantic class of a noun.
In that case, Rhône (L26760-S1)noun classtoponym (Q7884789), while Rhône (L26760-S2)noun classpotamonym (Q21661721). This would avoid overloading instance of (P31) while still allowing these classifications to be attached to the Sense, where the meaning actually resides.
It could also be useful for other semantic noun classes beyond toponyms, making the modelling more consistent.
JhowieNitnek (talk) 17:23, 31 July 2026 (UTC)Reply
Sounds interesting. Poslovitch (talk) 17:25, 31 July 2026 (UTC)Reply
@JhowieNitnek: +1, it sounds like a good idea (regardless of this specific case in this discussion). Cheers, VIGNERON (talk) 17:27, 31 July 2026 (UTC)Reply
@Poslovitch, @VIGNERON: Great! Could it be possible to help me with this proposal? I haven't really done any lexeme property proposals (except external-id's). JhowieNitnek (talk) 17:32, 31 July 2026 (UTC)Reply
I made a proposal. -- JhowieNitnek (talk) 12:04, 3 August 2026 (UTC)Reply

Items or Lexemes ?

[edit]
Latest comment: 5 months ago3 comments3 people in discussion

Hi,

Why do so many thesauri exist only for Items (Qxxxx) and not for Lexemes? (allowed-entity-types constraint (Q52004125))

Examples: UK Archival Thesaurus ID (P10913), UNESCO Thesaurus ID (P3916), etc.

Cordially. —Eihel (talk) 17:46, 25 April 2026 (UTC)Reply

@Eihel: The two properties you mention seek primarily to organize concepts, rather than particular words or phrases used for those concepts. The latter property you mention, for example, has labels in the six official languages of the United Nations, and the former property appears to borrow concepts from the latter property. It's not impossible for there to be some property for a thesaurus site, but it is likely to have less value than other dictionary properties we have since those dictionary properties often have synonyms listed in the first place. Mahir256 (talk) 19:53, 25 April 2026 (UTC)Reply
+1 with Mahir256. Also, you can have a lexeme linked to an item and said item linked to the thesaurus, so in the end the lexeme is already indirectly linked to the thesaurus. Cheers, VIGNERON (talk) 09:28, 1 May 2026 (UTC)Reply

Adding P5137 to lexemes by using Wiktionary translation clusters

[edit]
Latest comment: 1 month ago10 comments3 people in discussion

I'm running an experiment where I use basic translation data from Wiktionary (via my WikDict project) to add P5137 to lexemes that don't have any such link yet and have exactly one sense.

Basic concept:

  1. Find translation on Wiktionary where one end of the translation already has a P5137. Then the same relation is a candidate for the other end of the translation. Problem: words can be ambiguous.
  2. Get the description of the lexeme's sense and the labels and descriptions of the assigned item (all CC0 from Wikidata). Let an LLM verify this assignment to rule out most bad matches.
  3. Let a human go through the remaining items and assign approve/reject/defer labels. Only approved ones are queued for writing. The results are also stored for evaluation and future improvements.
  4. Write to Wikidata

Licensing: Wiktionary (CC-BY-SA) contributes only facts — which lemmas cluster, and the lemmas themselves. Verification reads only the anchor item's CC0 label/description, never the Wiktionary gloss, and nothing from Wiktionary (gloss, translation, sense numbering) is ever written to Wikidata. To my understanding, this clears all licensing hurdles.

Examples:

I'll use an edit group next time.

Questions:

  1. Do you agree that the licensing is ok?
  2. Do the examples look correct? Is the P887 → Q140308675 reference scheme the right way to disclose the heuristic? I created that one because I could not find a good existing heuristic.
  3. If this is generally ok, what is the best way to proceed? Should I slowly (<100 items/week, even less at first) push more manually approved items as assisted writes?

I'm happy to share review data, code, examples in other languages or any other information! Karlb42 (talk) 16:10, 29 June 2026 (UTC)Reply

It looks generally ok with everything. A little deviation is that "Bosnier" is not absolutely "Bosnians and Herzegovinians" (different stages of a country), but it's very hard to fix. Infovarius (talk) 13:26, 30 June 2026 (UTC)Reply
As a German I never heard citizens of that country being referred to differently outside of news broadcasts. But I agree that it is not an exact match. I intentionally kept it in the examples to get a good feeling for how strict I should be. Thanks for the feedback! Karlb42 (talk) 18:14, 30 June 2026 (UTC)Reply
@Karlb42: it looks great. Not sure why you need the Wiktionary here and the licensing should be ok. Could you do a test for French, Breton and English (to check languages I speak better than German 😉). Also, how do you deal with homographs? (think of the 3 words "tour" in French...) Cdlt, VIGNERON (talk) 07:39, 19 July 2026 (UTC)Reply
Wiktionary provides translations, so that I can carry over the denoted item from one word to a translation of it. Homographs should be identified by the LLM, which gets the sense gloss and the item descriptions to judge. I also have a few heuristics in addition to reduce the likelihood of including homographs, like excluding lexemes that have multiple senses in Wikidata. I want to get a high accuracy on the simple cases before moving to more complicated ones.
I can't do Breton, since that is not my Wiktionary dataset (no Breton Wiktionary content parser yet), but here's the list of English and French items I have written so far:
French (64)
English (63)
See my edit groups for a full list of changes in all languages: https://editgroups.toolforge.org/?user=Karlb42
I'm also doing a test-run of my data at the distributed game, so that I can evaluate the approach against judgements other than mine: https://wikidata-game.toolforge.org/distributed/#/game/96 Karlb42 (talk) 08:29, 19 July 2026 (UTC)Reply
can you look at Russian too please? Note that Russian Wiktionary has very strict structure and I believe it's easilty parseable. Infovarius (talk) 14:02, 20 July 2026 (UTC)Reply
@Infovarius Yes, Russian Wiktionary data is not a problem, I just can't review the results at all myself. Would you do some review on the Russian results? Should I prepare some Russian data for The Distributed Game? Karlb42 (talk) 06:26, 23 July 2026 (UTC)Reply
Yes, I would like to do some review. Infovarius (talk) 22:56, 3 August 2026 (UTC)Reply
@Infovarius I did the first run for Russian, but that added only 8 tiles for the Distributed Game. Most Russian lexemes either have no sense at all or already have a P5137 assigned. The approach is currently limited to lexemes with a single sense, but I'm working on a version that can add a sense to senseless lexemes (similar to Macht Sinn). I'll let you know when that is ready. Karlb42 (talk) 14:24, 13 August 2026 (UTC)Reply
Ah yes, they mostly have no senses without P5137, thanks. Infovarius (talk) 22:26, 14 August 2026 (UTC)Reply

One dictionary as source for lexeme

[edit]
Latest comment: 1 month ago26 comments4 people in discussion

I'm working on a project to create Wikidata Q-items for each transcluded entry of a dictionary which I am proofreading on enWS. (I have created A (1) (Q140790794) as an example.) I would like to connect these to lexemes, but not all exist; for example, アアシヤコシヤ is poorly attested, and a Google search only finds this file, which looks to translate as "アアシヤコシヤ is a mocking laughter". Can I just use the dictionary as a source, or do I need something more? Any other tips on creating the Q-items? I plan to use OpenRefine for the mass creation of Wikidata items. TE(æ)A,ea. (talk) 04:56, 8 August 2026 (UTC)Reply

This seems inappropriate as items in Wikidata; what prompted you to do it this way? This definitely would qualify as mass editing under the proposed Wikidata:Requests for comment/Mass-editing policy. I'm not sure what WikiProject should be consulted - maybe this talk page is as good as any. See also Step 2, the notability and license check. ArthurPSmith (talk) 20:27, 10 August 2026 (UTC)Reply
  • I posed a separate question elsewhere, and was directed here as the best place to ask it. These dictionary items, as far as I can tell, unambiguously pass the notability policy. The main criterion is a valid site-link, and every item here will have a valid site-link on enWS. By way of comparison, a similar policy is done with respect to all articles in all encyclopedias which we transcribe. See 1911 Encyclopædia Britannica/Mutsu Hito (Q84665129), and all Q-items published in (P1433) Encyclopædia Britannica 11th edition (Q867541). I am not native to Wikidata, so I am unaware of any unsolidified general consensus, but I have read both Wikisource-specific Wikidata notability-policy discussions and believe that these items qualify. I am not able to follow the proposed policy as the key Wikidata: page for it does not exist and the proposed policy has not been brought into force, but I intend to fully discuss the proposed mass-edits before they are made in some suitable forum (which I had hoped this would be). TE(æ)A,ea. (talk) 21:32, 10 August 2026 (UTC)Reply
    Each dictionary entry would be its own Wikisource page? ArthurPSmith (talk) 14:08, 11 August 2026 (UTC)Reply
    The elsewhere was in WikiProject, and there was resistance there. Vicarage (talk) 20:57, 19 August 2026 (UTC)Reply
    • I have already provided the Wiki link for this discussion. I also mentioned, as you should know from that discussion, that I was directed to come here in that discussion, which, I may add, offered me no clear answers. My course of action of action is entirely consonant with existing policy and enWS practice, and I have not been convinced otherwise. TE(æ)A,ea. (talk) 02:01, 20 August 2026 (UTC)Reply
@TE(æ)A,ea.: "It would not be possible to associate one lexeme or Q-item with multiple pages on the same project." - sitelinks do not apply to lexemes. You would link them with a property such as Wikimedia import URL (P4656) or perhaps a dedicated URL, and you can link as many wikisource pages as you like to the same lexeme. Please don't add Wikidata items for each dictionary entry! ArthurPSmith (talk) 19:14, 13 August 2026 (UTC)Reply
  • If site links do not connect with lexemes, then wouldn't Q-items for dictionary entries be the logical connection point? P4656 specifically excludes Wikisource source links as valid entry values, and this system would be different from how site links are handled for encyclopedia articles and everything else. Using site links would also mean that the transcluded enWS pages would not be able to be connected to Wiktionary entries, which is one of the intended benefits of my approach. TE(æ)A,ea. (talk) 02:09, 14 August 2026 (UTC)Reply
    Maybe a new property is needed then. I don't understand your comment about connecting to Wiktionary. How would an item prevent that? ArthurPSmith (talk) 16:25, 15 August 2026 (UTC)Reply
I believe items like Q140790794 is quite useful in Wikisource perspective, especially if Wikisource wants to provide external links from s:A Japanese-English and English-Japanese Dictionary/Japanese/A (1). In addition, if Wiktionary want to use such lexicographical data (also applies to external tools using lexicographical data), it provides a way to deep link to this specific Wikisource pages.--GZWDer (talk) 13:52, 20 August 2026 (UTC)Reply
Can you create such a link as an example? This item exists now, so please use it as an example. I really am not following what you all are talking about here. ArthurPSmith (talk) 19:43, 20 August 2026 (UTC)Reply

How to label phrases with varying arguments?

[edit]
Latest comment: 1 month ago3 comments3 people in discussion

How can I make a lexeme for a collocation such as "number (object) from (x) to (y)"?

Should I at all? -wd-Ryan (Talk/Edits) 22:21, 13 August 2026 (UTC)Reply

That doesn't seem to me like it should be a lexeme. ArthurPSmith (talk) 23:57, 16 August 2026 (UTC)Reply
Yeah, this sounds like a sense on number (L37935) (specifically sense 4.a. in the OED). Mahir256 (talk) 00:23, 17 August 2026 (UTC)Reply

Lex-Olympia

[edit]
Latest comment: 1 month ago1 comment1 person in discussion

I'm advertising here a tool I created.

If you're interested, I released a simple tool inspired by the Lexemes Challenge (now unmaintained). I added several new ideas to play with, like senses, audio, and hyphenation challenges. If you're curious, please take a look!

https://lex-olympia.toolforge.org/

Any comments or ideas are welcome. Thanks! -Luca.favorido (talk) 20:09, 17 August 2026 (UTC)Reply

Retrieved from "https://www.wikidata.org/w/index.php?title=Wikidata_talk:Lexicographical_data&oldid=2555601653"