Wikidata talk:Item classification

Synecdoche, metonymy, and other good friends

@TomT0m: I think in our past conversation about teachers we missed figures of speech. As you said "teacher" can indeed be used as a replacement of "instance of:human", but it is not the same item as in "occupation:teacher". The two distinct items more or less are:

• teacher: subclass of human, occupation: teacher
• teacher: subclass of occupation, field of this profession:teaching

We would need to split teacher (Q37226) and problem solved :)--Micru (talk) 15:09, 18 August 2014 (UTC)

Actually there are two items already! See: school teacher (Q3406814)--Micru (talk) 15:15, 18 August 2014 (UTC)

@Micru: Not quite :/ Actually those two items labels shows that one of these items is about a specific kind of teachers. But both refers to persons. Actually the ontological question about a teacher and the « class / token » relation is that a teacher is a person who teaches. The tokens, in this view are : the teacher, who is a real world token obviouly, and the teaching hours he gives he gives, as a class is localized in time and space. The ontological nature of an «occupation» itself with respect to the class/token relationship is less clear. We say someone is a farmer if he owns a place in which he cultivates something and sells the result, for example, but we could eqully say he has « farmer » as occupation. Do we rally need two items here or can we use « punning » to use the same item as both a subclass of person and as an instance of sométhimg more abstract, an « occupation», as an occupation seems to be neither a token, nor a class of token, but rather is used to state that someone spends a large part of its life realizing events belonging to a class of events associated to this class ? For example if the «teacher» occupation is associated to the «teaching» event class, then if someone has «teacher» has an occupation this would imply that (assuming we kow all the events in its life) there is a lot of teaching events ? It does not séém very useful though. Another approach would be to take, as real world objects' the contract we signs or the assignment we recieve. Then «teaching assignment» would be the class we need for the range of the "occupation" (renamed assigment or contract property.) I would love to read Emw on that. TomT0m ^(talk) 20:06, 18 August 2014 (UTC)

@TomT0m: In practice the boundaries of classes seem to be extremely observer-dependent, and by splitting the item we provide several observation pathways for arriving to similar conclusions. Has the observer seen a person teaching? Or has s/he this person stated or provided a document stating their field of work? The field of work is an abstract information encoding and needs someone stating it (even falsely or without direct knowledge). OTOH, the mere observation of events is direct, but the class itself doesn't provide a clear-cut definition of how many teaching events make a teacher, so again there is the need of an external party to define it. Even if you say "only when I see anyone performing a teaching act I shall identify them as teacher", there too you are bringing in a subjectivity that we might not share. Both concepts are closely related to one another but they are not the same. It can be compared to the FRBR model, with a information-only abstraction of the class "teacher" (cf. work), a multiple observation of the subject of "teaching events" (cf. items), and an embodiment, that is a person that defines therself or is defined as a teacher and engages or has a record of "teaching events" (cf. manifestation).

I am glad enough collecting abstractions, but if you ask the radical reductionists of BFO they will probably tell you that observer-dependent encodings do not exist, only what is directly observable, which is also fine, but just a part of the picture. Perhaps they will change their mind while preparing their information ontology, who knows. In any case, I am not sure that our structure would be suitable for condensing all perspectives into one item (subclass of:human OR occupation?), but I am open to talk about possible ways. Also looking forward to reading Emw's perspective.--Micru (talk) 20:54, 18 August 2014 (UTC)

@Micru: I must not have made myself fully understood :/ The question was mainly about the ontological concept of an «occupation», I don't think we really have good ontological perspectives about what an occupation ontlogically is. As the concepts of «someonone has teacher as an occupation» an «someone is a teacher» seems to be almost redundant, this may imply that the «occupation» concept is not really useful. Or that it may map more easier map to concrete stuff like «if the school hired a teacher and the contract is entitled teaching chari then the person will be said to be a teacher. Subjectivity is not really a question here, as there will always be room for it as we won't always find ready to cite datasets suitable for our needs of organisation of information in a somewhat consistent way. TomT0m ^(talk) 10:00, 19 August 2014 (UTC)

@TomT0m: For "occupation" the reasoning is exactly the same. There is the void concept representing "occupation" which is related to other void concepts (information), an event or a group of events of something that can be considered as someone having a occupation (instance, extension of a class), and the set of conditions for considering that someone has an occupation (class). In the class you can define as many conditions as you wish for characterizing what means "teacher" (teacher contract, teaching chari, datasets, sources, etc), but you will never reach an absolute definition of "teacher", and yes, that is subjectivity. Even if you considered the aggregate of all subjectivities of every person of the planet, that "objective" definition of teacher would only last for a few microseconds.

However, I don't share your pessimism, the term "teacher" is useful even if it cannot be defined as an absolute, it is just a matter of finding a balance between the efforts that we invest modeling a concept and the benefits we get out of that effort.--Micru (talk) 11:02, 19 August 2014 (UTC)

@Micru: We still do not speak the same language, I don't feel you replied to my comment :) TomT0m ^(talk) 11:18, 19 August 2014 (UTC)

@TomT0m: You are asking "what is the relationship between the elements class and token wrt occupation?" and I'm telling you "your question is malformed because there are not two but three elements: information, class and token". Of course we are not speaking the same language! :) --Micru (talk) 11:26, 19 August 2014 (UTC)

@Micru: You did not define that «information» stuff. Actually class/tokens as in the database are informations about stuffs. The question is how best to represent information about stuffs in Wikibase. You answer «information», which is not an answer, but a part of a question. Seems like we're circling around. :) TomT0m ^(talk) 11:37, 19 August 2014 (UTC)

@TomT0m: Actually the basic unit is identifier (Q853614), you need a code to represent something within a system. Information is derived from it and represents how a constellation of identifiers are interpreted or related to each other within the system. An identifier has no meaning except for internal uses of the system that defines them or for others that agree to use them, same for information. Having in mind the three distinct domains, for crossing from one domain to the other there is the need of special properties, from token to class there is "instance of", and from class to information there should be a "manifestation of".

Any "identifier" requires subjectivity to be interpreted or materialized, but the intended meaning, materialization, or interpretation are system-dependent. Also note that "identifiers" are void, and as such there are two "identifiers": the ideal identifier "all the same of this", and its class which is defined by what a particular system defines wrt a shared reality. A "word" is a "manifestation of" identifier, same for Q-numbers, etc., just in case that you were about to say that we already have "identifiers" :)

Makes sense to have items that represent ideals? No idea, but it is useful, specially when dealing with everything that is "artificial" (consciousness-derived output, like mathematics, language, etc). Neither an ideal of "lord of the rings" nor the concept "novel" exist, but you can say "lord of the rings is a novel" and that will have some meaning for you and some other meaning for others, depending on how much have you both have agreed on the definition of "class:novel" and on your personal experiences. These concepts are very old, they are fully characterized, and they do not affect class/token relationships, so we would keep compatibility with other ontologies.--Micru (talk) 12:39, 19 August 2014 (UTC)

@Micru: Sorry, can't extract anything helpful in this comment. Not either a definition of information. So I won't reply any further (If I started an answer I could write ten lines for every line you write, so since this seems unrelate to my current concerns I won't go there, that would make the world explode) TomT0m ^(talk) 13:16, 19 August 2014 (UTC)

@TomT0m: I will have to learn to write better :) The abridged version is: "information is how a set of identifiers are interpreted or related to each other within a system". Thanks for not making the world collapse for now :) If you want we can talk about it, it might be more productive and shorter.--Micru (talk) 13:23, 19 August 2014 (UTC)

Still not. Take a random set of identifiers and a random set of relations amongst them. What we get is random noise, not information about anything, this still meet your information. Still don't know if you are going anywhere but random meaningless jabber. TomT0m ^(talk) 13:38, 19 August 2014 (UTC)

@TomT0m: You are missing the word "interpreted", if the system cannot interpret those random identifiers and relations, then they are random identifiers and relations, not information.--Micru (talk) 13:45, 19 August 2014 (UTC)

@Micru: What about it ? TomT0m ^(talk) 13:47, 19 August 2014 (UTC)

@TomT0m: The verb "to interpret" has a subject and an object. For information to exist there is not enough condition with having an object (the set of identifiers and relations within a system), but the system itself has to be able to act as a subject for it to happen. Wikidata on its own is not information, because Wikidata itself cannot interpret its own set of identifiers and relations. However, when you access Wikidata, then your contact with the set of identifiers and their relations causes it to become information, because you become the necessary subject which can interpret the set of identifiers and relations.

Summing up, for information to come into being an agent (possibly able to integrate information into itself and interpret it) has to come into contact with something (be it a traffic light, or a set of identifiers and relations on the internet). I can elaborate more if you wish, but sometimes it is better to keep it short.--Micru (talk) 14:08, 19 August 2014 (UTC)

@Micru: sometimes it is better to keep it short When you're going nowhere ? /o\ TomT0m ^(talk) 14:17, 19 August 2014 (UTC)

@TomT0m: The problem is that you are used to static definitions, and "information" can only be defined in dynamic terms, you cannot use the same paradigms. Rack your brains for a while, maybe it helps :)--Micru (talk) 14:28, 19 August 2014 (UTC)

@Micru: Hehe, no I don't think that's the problem. TomT0m ^(talk) 14:37, 19 August 2014 (UTC)

@TomT0m: Then you will have to explain me where *you* want to go :) --Micru (talk) 14:40, 19 August 2014 (UTC)

Mmm the original problem was «is it enough to identify an occupation with the class of all people who have them ?» The «painter» class is the set of all persons who paints. What is the nature and usefulness of the «occupation» concept then ? Seems cumbersome. TomT0m ^(talk) 14:59, 19 August 2014 (UTC)

It is the basis for linguistic data, it has no usefulness now, but it might be useful in the future for Wiktionary. At least if we have a clear division we can detect when a property is using it and decide if it is worth it or not. --Micru (talk) 15:36, 19 August 2014 (UTC)

Seems speculative. TomT0m ^(talk) 15:42, 19 August 2014 (UTC)

There is a long way before that, so it is irrelevant.--Micru (talk) 18:41, 19 August 2014 (UTC)

Absolutely not, it is now and it is totally relevant ! Or not. TomT0m ^(talk) 19:01, 19 August 2014 (UTC)

@TomT0m: I was writing an answer, and then it became an essay: Wikidata:Lounge/Growing items.--Micru (talk) 13:28, 20 August 2014 (UTC)

Categories, templates and other boring things.

A point I have been too lazy to raise for a long time. What should we do with items that are about Wikipedia categories, help pages etc. Currently they are (indirect) instances of "Wikimedia page" but I do not think it makes sense. Discussion at Talk:Q17442446. --Zolo (talk) 08:07, 7 October 2014 (UTC)

Metaclassification, union of and so on

A number of stuffs have been done to try to better model classifications : we found literature stuffs on metaclassification ( metaclass (Q19478619)   , which User:TomT0m/Classification tries to explain ). We created union of (P2737) , disjoint union of (P2738) to express stuffs about a set of subclasses in which every instance of a class belongs. We also created properties like next lower rank (P3729) and the converse to model ranking in a classification. What can we do with those ? author  TomT0m / talk page 11:12, 16 January 2018 (UTC)