In Wikidata, we describe concepts (Q items) and lexemes (L items), where the lexemes may be nouns, verbs, adjectives, etc.

It is usually straightforward to describe the relation from noun lexemes to concepts, e.g., the Danish noun "bil" (https://www.wikidata.org/wiki/Lexeme:L36385) and the English noun "car" (https://www.wikidata.org/wiki/Lexeme:L3648) can both be linked to the car concept (https://www.wikidata.org/wiki/Q1420).

A problem arise when we want to describe non-noun lexemes. For instance, the English verb "drive" does not have an entirely equivalent concept in Wikidata, - in my opinion. It seems closely related to the driving concept (https://www.wikidata.org/wiki/Q999646), but also to a possible driver concept and the noun "drive" (A trip made in a vehicle). Currently, the verb "drive" links to the driving concept with the P5137 (sense/means/denotes, https://www.wikidata.org/wiki/Property:P5137) property, that is usually used to link nouns with concepts. I wonder if that is not a too broad application of the property, as now it seems that the English verb "drive" and the word "driving" (which in Danish is the noun "kørsel") are synonyms..!?

While drive-driving is not that far off conceptually there are other words that have other relations, e.g., the verb "employ" could relate to employer, employee, employment and hiring. In the case of "employer", there is - at least it seems to me - a nomen agentis relationship between employ and employer: an "employer" is a person that "employs". Also there is a (etymological) derivational relationship between employ and employer, employee and employment, - but not hiring.

For adjectives, there are similar problems, e.g., what does "high" relate to? There is a height concept in Wikidata (https://www.wikidata.org/wiki/Q208826) and the adjective "high" relates to this somehow, but so does its antonym "low". One solution to the problem is to define an "adjective concept" for "high" in Wikidata so we can link the word "high" to that concept, but there is a probably a reluctance to do it this way. And even if we would define a "high" concept, so we can link the "high" lexeme with the high concept, there is still the problem of linking the high concept with the height concept or other possible concepts.

WordNet describes relations between word classes with "derivationally related form" which can link, e.g., "employ" with its related nouns, in Python with NLTK:

>>> from nltk.corpus import wordnet as wn
>>> wn.synsets('employ')[2].lemmas()[2].derivationally_related_forms()
[Lemma('employment.n.01.employ'), Lemma('employment.n.03.employment'), Lemma('employee.n.01.employee'), Lemma('employer.n.01.employer')]

but that seems not to describe the detailed semantics between the employ* nouns and the "employ" verb, and I do not see any other way that WordNet describes word-sense-synsets relationships between word classes.

I haven't yet found any scientific articles that builds a broad formalism that could be used for linking non-nouns lexemes in Wikidata with a high degree of specificity. It is unclear to me whether a FrameNet approach will be of help. A property corresponding to WordNet's "derivationally related form" is already present in Wikidata with the P5191 property (https://www.wikidata.org/wiki/Property:P5191).

  • 1
    What sort of a thing is what you call a "concept"? Why do you think there are such things? Are you making a new linguistic theory, or what?
    – Greg Lee
    Jan 9, 2020 at 23:31
  • 1
    Why don't you just add verbal and adjectival concepts to the database?
    – curiousdannii
    Jan 10, 2020 at 0:15
  • This sounds like it's being built to resemble English, which is probably not the right exemplar. Take a look at Foley's Anthropological Linguistics and Frawley's Linguistic Semantics for the properties that need to be codable (different set for each language, natch); the "concepts" can be linked up with them.
    – jlawler
    Jan 10, 2020 at 2:27
  • @GregLee I am using the word "concept" to mean the Q-items of Wikidata which often correspond to a Wikipedia article. I my opinion these Q-items correspond to synsets of wordnets. It is not necessary to call them "concept". We can call them "synsets" or "Q-items", if one prefers that wording. Jan 10, 2020 at 9:16
  • @curiousdannii It may be that Wikidata ends up with constructing verbal and adjectival concepts, but that still leaves the question on how to link these concepts with "noun" concepts. A few seems "easy", e.g., the triple for a nomen agentis ("baker", "is a person that", "bake"), but, e.g., what about high/height and employ/employee? Jan 10, 2020 at 9:24

1 Answer 1


This is a problem I have been facing for a while too. I have figured out nouns, verbs, and adjectives, and partially words like "I" and "this" and "we" with referential qualities. But your question is about adjectives mainly.

In programming I model this as objects (nouns), actions (verbs), and features (adjectives). Features are the key. All things in the universe are not objects. They are not actions. They are features. The "high" feature is now an entity that you could link to, you could apply it to an object, you could extend/merge it with other features, etc.

I hope I understood your question. I personally find the idea of "concepts" and "lexemes" useless in programming, they just don't capture things in enough detail.

  • Why the downvotes, please explain?
    – Lance
    Jan 15, 2020 at 13:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.