Talk:Q23958852

subclass or instance, not both

@TomT0m: So please clean the item: an instance can't be at the same time a subclass. A source is not a argument when defining a logical classification. Snipre (talk) 11:53, 12 July 2016 (UTC)

@Snipre: Please add an exception to the constraint, then. It makes perfect logical sense. Do you have an idea of what we are talking of before beeing authoritative ? author  TomT0m / talk page 16:41, 12 July 2016 (UTC)

@TomT0m: It is logical if we choose to have metaclass. The metaclass concept has to be discussed at the community level in order to weight advantages and disadvantages. And don't speak of authoritative when you revert with as sole explanation "there is a source, don't touch". Snipre (talk) 13:30, 13 July 2016 (UTC)

@Snipre: What is a chemical elements if we choose not to have metaclasses ? Here is how it is modelled in Cyc : http://sw.opencyc.org/2012/05/10/concept/en/ElementStuffTypeByNumberOfProtons This is not a matter of choice. This is needed to represent correctly the frwikienwiki definition. Cyc does this, and I came independantly with the same conclusion. author  TomT0m / talk page 13:40, 13 July 2016 (UTC)

Not to mention : Cyc is a serious and public reference. We don't need more to justify those statements. author  TomT0m / talk page 13:52, 13 July 2016 (UTC)

@TomT0m: I don't care of particular classification problems, I want a general discussion about the general advantages and disadvantages first because allowing metaclass leads to some limitations (classification without metaclass have others) and I want to be sure that the choice is made by the community with full knowledge of the facts. I want to have the discussion presented in that paper in the framework of Wikidata. Once we do that choice we will go to particular problems and solve them using the good tricks.

Et stop avec les références, parce que cela montre que tu ne sais pas de quoi tu parles: si je te parles de OWL-DL ou de OWL-full, tu sais de quoi je parle ? Si non, merci de lire ceci pour comprendre que le problème n'a rien à voir avec quel système à le plus de références ou est utilisé par telle ou telle base de données. Le problème est de savoir quelle logique on veut utiliser pour pouvoir plus tard faire des inférences complexes de manière automatisée et si c'est le but de Wikidata.

En résumé, l'utilisation de métaclass est un foutoir pour les algorithmes lorsque l'on leur soumet des recherches complexes. C'est un des paramètres qui rend une ontologie undécidable. Est-ce que c'est ce qu'on veut sur WD ? Voilà la question que tu ne sembles pas vouloir porter sur la place publique. Je n'ai aucune prétention et encore moins une connaissance approfondie de la question, mais j'estime que le choix de telle ou telle orientation ne relève pas d'un individu mais d'une décision collective, et c'est pourquoi je m'oppose à tes contributions qui imposent ce choix de facto. Snipre (talk) 14:28, 13 July 2016 (UTC)

@Snipre: Bien sûr que je sais ce que sont OWL-DL et OWL-Full. Tu sais ce qu'est le "punning", une fonctionnalité d'OWL2 ? C'est justement une fonctionnalité d'OWL2 qui permet d'avoir des instances d'instances sans perdre la décidabilité et sans utiliser OWL-Full. La question sur les éléments chimiques reste pertinente. Les classifications avec métaclasses sont juste des superensembles plus expressifs des classifications sans métaclasses : tu peux simplement faire plus de déclarations, toute déclaration valable dans une classifiation sans métaclasse est valable dans une classification avec. Y a t'il besoind de concensus pour simplement insérer des déclarations sourcables, sourcées et solides ? author  TomT0m / talk page 14:34, 13 July 2016 (UTC)

Corrolaire : si tu veux une ontologie sans métaclasse, limite toi en SPARQL à toutes les sous-classes et les instances d'individu et tu seras pas embêté avec les métaclasses. C'est pas compliqué. author  TomT0m / talk page 14:37, 13 July 2016 (UTC)

Ah, c'est bien, on sort enfin de l'argumentation gamine de la référence ici ou de la base de donnée là que tu m'a sorti jusqu'ici.

Mais au lieu de me dire de fermer les yeux et d'aller voir ailleurs, merci d'expliquer ce qu'implique le punning ? Parce que le punning ne résout pas tout, c'est juste un patch qui élimine une partie des limitations du méta-modelling sans toutes les enlever. Et négliger l'impossibilité de vérifier la consistence des déclarations fausses ([1]) sur WD alors que WD est ouverte à tous et que cette possibilité d'introduire des erreurs sur un système à forte diffusion sur toutes les WP est un argument fort envoyé par les opposants à WD, c'est un peu léger.

Corrollaire: tu n'en a rien à cirer d'une approche communautaire et ta pédagogie ne vaut pas tripette. Snipre (talk) 15:39, 13 July 2016 (UTC)

Moi au moins j'évite de me lancer dans des guerres d'édition et j'essaye de discuter quand tu vires des trucs arbitrairement. Tu as déjà tenté une approche constructive sur ces sujets ? Moi j'ai essayé, plein de fois, et je n'en aurai rien à cirer des approches communautaires ? Tu te moques de qui ? author  TomT0m / talk page 15:48, 13 July 2016 (UTC)

Dernière chose, j'ai abondamment discuté là dessus, et lancé des RfC. Conclusion : nada. À un moment il faut avancer et là j'ai clairement suffisamment de billes pour le faire en étant assez confiant pour dire que je fais pas de conneries. Après on peut en discuter, mais si tu veux contester tu dois être à la hauteur et rentrer dans les détails, et tu dois aussi contester les sources qui sont des POVs valables. Le NPOV est une règle de base. author  TomT0m / talk page 14:41, 13 July 2016 (UTC)

C'est bien ce que j'ai dis: ta pédagogie ne vaut pas grand chose et tu viens de le démontrer dans cette discussion. Ce n'est pas en assomant ton interlocuteur de references et ensuite en lui disant de se débrouiller pour ne pas tomber dans les cas qui posent problème que tu vas convaincre du monde. Au fait, je serai curieux de savoir où tu as présenter les différents languages ontologiques possibles. Snipre (talk) 15:39, 13 July 2016 (UTC)

Eh bah c'est constructif ça, on enchaîne sur le terrain personnel. tu me reproche de balancer des références alors que tu viens de faire de même avec "tu sais ce qu'est OWL et OWL2" de manière ultra provocante, tu me reproche de faire de même ? T'as pas fini avec les attaques personnelles, tu peux pas poser des questions calmement ? J'ai juste carrément écrit des articles Wikipédia pour expliquer ces trucs, si tu les comprends pas soit constructif et fait des remarques dans leur PdD. Tu crois vraiment faire avancer les choses en me balançant des trucs pareils et en étant aggressif ? author  TomT0m / talk page 15:48, 13 July 2016 (UTC)

Sinon c'est hyper cohérent de me demander de balancer des liens après m'avoir reproché de balancer des liens. author  TomT0m / talk page 17:13, 13 July 2016 (UTC)

@Snipre: There is no logical reason that an entity can't be both a subclass and an instance, even of some one particular entity. RDFS, in particular, allows this, where both of rdfs:Class rdf:type rdfs:Resource . and rdfs:Class rdfs:subClassOf rdfs:Resource . are true in all RDFS interpretations.

OWL 2 DL, on the other hand, does not have entities that are both an instance and a subclass at the same time. Instances in OWL 2 DL denote elements of the object domain; classes denote subsets of the object domain. There is nothing in OWL 2 DL to prevent elements of the object domain from also being subsets of the object domain, but there is nothing in OWL 2 DL that can require that an element of the object domain is a subset of the object domain, so it is not possible to infer that an entity is both an instance and a subclass. One might think that punning can do this but it can't - punning is simply a way of having one name denote an element of the object domain when it is used as an individual and a subset of the object domain when it it used as a class.

RDFS is decidable, so it is possible to have a decidable logic that allows entities to be both a subclass and an instance of some entity. RDFS is non-trivial, so it is possible to have a non-trivial (i.e., free of paradoxes) logic that allows entities to be both a subclass and an instance of some entity.

OWL 2 Full is like RDFS in that it allows entities to be both a subclass and an instance of some entity. OWL 2 Full is undecidable, but not because of this.

What does any of this have to do with Wikidata? Well Wikidata is not a logic, so discussions of decidability are not germane at all. Whether Wikidata should allow entities to be both a subclass and an instance (of some entity) is a question that should have been discussed, but as Wikidata already has quite a few metaclasses whose intended meaning, at least as far as can be determined by their limited descriptions, would indicate that they are both instances and subclasses of some entity, starting with entity (Q35120). Peter F. Patel-Schneider (talk) 21:50, 13 July 2016 (UTC)

Many thanks for this :) I was really not in the mood to write something like that, and I would have done worse anyway. author  TomT0m / talk page 07:56, 14 July 2016 (UTC)

Instance of itself

I'm not sure where the conversation above ended up really, but surely it's not correct for this to be an instance of itself? I've removed it, but revert if I'm wrong. Sam Wilson 01:50, 21 September 2018 (UTC)

@Samwilson: The source makes it clear that it is in fact an instance of itself. Re-adding. --Yair rand (talk) 22:07, 5 November 2018 (UTC)

@Yair rand: Fair enough. Thanks for looking into it. Sam Wilson 00:20, 6 November 2018 (UTC)

Subclass of Wikidata metaclass

Our English description of metaclass (Q19361238) says "class of classes, class whose instances are classes". Variable-order class (Q23958852), on the other hand, represents "classes whose instances may or may not be classes themselves" (my emphasis in each case). Given that, why is variable-order class a subclass of Wikidata metaclass?

Instances of Instances Modeled via Higher-Order Classes (Q23959013), which is used as a reference here, says, which is intuitively correct: "Variable-Order Class is not a subclass of Meta-Class since some of its instances, e.g. Thing, have instances that are Individuals as well as instances that are Classes." Mortee (talk) 21:34, 23 July 2023 (UTC)

I think it is the description of metaclass (Q19361238) that is incorrect. In my view, if variable-order classes are allowed then a metaclass is a class that can have (does have?) instances that are classes, i.e., not all instances of instances of metaclass (Q19361238) have to be classes. Peter F. Patel-Schneider (talk) 23:11, 23 July 2023 (UTC)

@Peter F. Patel-Schneider: interesting. That's different from the Cyc model as I understand it and my own intuitive view, but it would work logically. One possible downside might be for consistency checking - one wouldn't be able to check that all instances of a metaclass should be classes to spot some kinds of bad data. Given that, would we want another class of 'pure metaclasses' for those classes whose instances are always and only classes (i.e. my reading of what metaclass means)? Then we'd have:

• variable-order class → metaclass → class
• second-order class, third-order class, ... → pure metaclass → metaclass -> class
• first-order class -> class
• fixed-order class -> class

(I don't know how exactly class (Q23960977) fits in because from the discussion on that entity it seems identical to class. Before I read that discussion, my best justification for it was that it meant what you mean by metaclass, i.e. classes that must contain classes but may also contain non-classes, but that doesn't seem to be what's meant by it)

While I have you, two other things that strike me as puzzling at the moment:

1. metaclass (Q19361238) is an instance of third-order class (Q24017465), which looks wrong on either view because it includes third- and higher-order classes. It should be a variable-order class (Q23958852) instead, I'd think.
2. fixed-order class (Q23959932) is a subclass of metaclass (Q19361238), which looks wrong on either view because it includes first-order classes, i.e. purely non-meta classes.

It's always possible that I'm just confused, but does that look right to you? Mortee (talk) 10:13, 24 July 2023 (UTC)

What should happen here is taking all the metaclasses in Wikidata and rationalizing them. This is not a trivial undertaking but should probably be done. I've always found it weird that there are Wikidata versions of several metaclasses - I don't understand why. Maybe someone in the unpingable Ontology project group can comment. I've put a comment in https://www.wikidata.org/wiki/Wikidata_talk:WikiProject_Ontology

I agree that metaclass (Q19361238) appears to be the cleanest class for non-obligatory metaclasses.

Peter F. Patel-Schneider (talk) 13:22, 24 July 2023 (UTC)

@Mortee: 1. Yes, this looks wrong, I agree with replacement. 2. Why (pure)classes are in fixed-order class (Q23959932)? If they are regarded as metaclasses of some order, this looks wrong. Infovarius (talk) 18:59, 30 July 2023 (UTC)

description

The English description of this item was "abstract entity representing classes whose instances may or may not be classes themselves". But this is not correct. A class that contains only first-order classes and second-order classes is a variable order class but its instances are all classes. I have changed the description to be correct. Peter F. Patel-Schneider (talk) 21:42, 24 July 2023 (UTC)