# Wikidata talk:WikiProject Ontology

## WP kickstart

@TomT0m, Emw, Danneks, Infovarius: Wikiproject started after TomT0m's complaint about the lack of a central space for discussing edits, changes, and rearrangements that have an impact on the structure. Personally I'm dissatisfied with most upper ontologies out there. BFO for me is one of the best but unfortunately it has some leanings that make it incompatible with other ones. In any case I think our priority should be usefulness, clarity, and ease of use. Speaking as engineer, better to have something useful for building cool things like Watson, rather than something strictly compatible but not useful.--Micru (talk) 22:02, 24 August 2014 (UTC)

Yep, the initial question good to have a central place to talk.
I'd like thoughts about Help:Classification (an essay about classification in Wikidata) for example :). The intended applications are to be basis for implementation of `{{Classification}}`, and primitives for building query with `{{WDQ}}` with subtemplate to handle classes and metaclasses. Not everyone have to agree it's a good idea though :)
I guess one open question is how this project will interract with other one.
@TomT0m: My main "problem" is that I work with machines and physical systems, and that makes me totally biased :) So whenever I see the word "class", specially when referring to material ever-changing entities, I wonder, "how has this class originated?", "what are the boundaries?", and "how much does it depends on other classes to convey meaning?". I know that the text is an approximation to some key concepts, but those key concepts are very volatile in the real world...
Perhaps instead of giving just the OWL interpretation, we should offer several compatible interpretations and that would make easier to relate to everyday concepts and to find equivalent readings in other ontologies. For me the starting point is "system representation", how things are perceived, how to conceptualize signal input, output, feedback loop, etc. We talked many times about it but never did any effort, and it might bring interesting insights. I am writing an initial document about it, I hope it doesn't get too long :P
I like the idea of bringing more attention to concept trees instead of to individual concepts, so whatever goes in that direction I will welcome it :) Another tool that I would like to have is one to "collapse" all the statements of a tree in a lower item, that way I would be able to see if the item is inheriting useful statements from its upper classes. Maybe it is already possible, but I don't know how.--Micru (talk) 13:12, 25 August 2014 (UTC)

## Pattern for representing relations brought in from external ontologies?

I'm not sure if this is the right place to ask, but has any consensus arisen about how to go about relating wikidata concepts to concepts in external ontologies? I'm working with some people that would like to bring some elements of the human disease ontology http://disease-ontology.org/ into the wikidata framework. --Genewiki123 (talk) 16:36, 26 August 2014 (UTC)

@Genewiki123: Maybe this question is more related to Wikidata:WikiProject Medicine. The property Disease Ontology ID (P699) is available, but currently only used once. An important first step would be to get the permission of the database owner to copy their identifiers to Wikidata. Then we would need a bot to match our items with those from their database. Tobias1984 (talk) 19:01, 26 August 2014 (UTC)
Thanks for the reference. It seems like this is more of a general ontology question though as the same problems will arise whenever some one wants to bring an ontology from outside into the wikidata concept space. As an example, I know there is a lot of interest in doing this for the gene ontology http://geneontology.org/ . They key issues would be first identifier mapping as you describe, relationship modeling, and updates. In this particular case, we are working directly with the owners of the ontology and hope to deploy bots such as this nascent gene bot https://www.wikidata.org/wiki/User:ProteinBoxBot . My main concern is to get the relations mapped properly. Most importantly I think, how should hierarchy be modeled? subclass? Will wikidata support any level of inference ? --Genewiki123 (talk) 20:02, 26 August 2014 (UTC)
@Micru, TomT0m, Tobias1984: Maybe we can create "type of a class" property? So, for example, <bacterial infectiuous disease> <type of a class> <disease by infectious agent> and <Homo sapiens> <type of a class> <taxon>? Danneks (talk) 06:38, 27 August 2014 (UTC)
@Danneks: I am in favor of the idea of using types (or metaclasses as TomT0m calls them) because that would give us more flexibility when linking with external databases. We could also define the metaclass <disease by infectious agent according to Source 1> which might be different than other sources, and as such might have more or less classes linked. It is worth noting that the concept of type/metaclass has other names in other knowledge domains and as such I don't know what would be the best name for this. Would "type of" be appropriate to link the metaclass with the class or would it bring unwanted uses?--Micru (talk) 08:40, 28 August 2014 (UTC)
@Micru: Actually, the more I think on that matter, the less I'm sure about anything. Maybe it would be more precise to obtain such lists of classes via queries, but it would require more properties of course. Danneks (talk) 13:51, 28 August 2014 (UTC)
Perhaps this pattern is worth a look. Here is how one person brought the gene ontology into freebase. http://www.freebase.com/biology/gene_ontology_group?schema= --Genewiki123 (talk) 17:43, 28 August 2014 (UTC)

I think we would be better off without a new "type of class" or "type of" property distinct from instance of (P31) and subclass of (P279). No other Semantic Web ontologies use such a property. They all get by with the standard W3C properties rdf:type (P31) and rdfs:subClassOf (P279). Innovating at the foundations of ontology vocabulary by adding a new basic membership property would not be a good idea.

The convention among Semantic Web ontologies is to use rdfs:subClassOf -- i.e. P279 -- to construct hierarchies. Genewiki123 refers to the Disease Ontology. Take a look at how Disease Ontology models cancer as "cancer subclass of disease of cellular proliferation" (see bottom of page). This is reflected in the OWL export of the Disease Ontology, doid.owl (warning: big), which uses rdfs:subClassOf to link to the two concepts.

Gene Ontology (GO) also uses subclass of extensively. See e.g. the Gene Ontology record for pathogenesis, which classifies pathogenesis as a subclass of multi-organism process. We could import this to Wikidata by simply adding the claim "subclass of (P279) multi-organism process" to pathogenesis (Q372016). (The claim "instance of disease" that's currently there is ontologically and biologically incorrect.)

Both the Disease Ontology and the Gene Ontology are part of the Open Biomedical Ontologies (OBO). I recommend anyone interested in modeling biological concepts to read The OBO Foundry: coordinated evolution of ontologies to support biomedical data integration by Barry Smith et al. The OBO uses the Basic Formal Ontology (BFO) as its upper ontology, which enables interoperability among many domain-specific ontologies in biomedicine and beyond. Emw (talk) 16:44, 1 September 2014 (UTC)

@Emw: Well, Semantic Web ontologies, after all, classify resources... and Wikidata also does, with a restriction to notable resources? Maybe it would be the best solution, because it answers all existential questions — we will just say that our entity (Q35120) corresponds to the class rdfs:Resource. Or maybe we shouldn't classify anything at all (only to associate data with Wikipedia pages), it is also a solution. Danneks (talk) 13:34, 2 September 2014 (UTC)
Danneks, owl:Thing, not rdfs:Resource, is the standard root class in Semantic Web ontologies. It corresponds to the "top concept" -- ${\displaystyle \top }$ -- in literature on knowledge representation and description logic. The mapping of entity (Q35120) to owl:Thing is a core relation in Wikidata's ontology. It's unclear how mapping entity (Q35120) to rdfs:Resource "answers all existential questions". Indeed, it's unclear how that resolves any questions at all.
For reference, the OWL specification describes owl:Thing as follows: "Two OWL class identifiers are predefined, namely the classes owl:Thing and owl:Nothing. The class extension of owl:Thing is the set of all individuals. The class extension of owl:Nothing is the empty set. Consequently, every OWL class is a subclass of owl:Thing and owl:Nothing is a subclass of every class (for the meaning of the subclass relation, see the section on rdfs:subClassOf)." See http://www.w3.org/TR/owl-ref/#ClassDescription.
You write "Or maybe we shouldn't classify anything at all (only to associate data with Wikipedia pages), it is also a solution." This strikes me as either very unclear or an especially odd non-starter. Classifying things into concept hierarchies is a fundamental mode of knowledge representation in Semantic Web ontologies. Much work on Wikidata has gone into classification and building infrastructure to support it. This work has been quite fruitful.
There are certainly ontological issues to be resolved, but using an idiosyncratic top concept or abandoning classification itself would be throwing the baby out with the bathwater. Emw (talk) 04:09, 3 September 2014 (UTC)
@Emw: When I say "to classify something", I mean "to specify type of that thing." OWL calls those classified things "individuals" — all right, but what are those individuals in case of Wikidata? For me, the simplest answer would be to say that individuals are the unions of Wikipedia/Wikisource/etc. articles which "have the same topic". It implies that every item with a sitelink may be, eventually, classified as an instance. I hope it makes my first point more clear. The second point expresses my concern whether specifying type of everything could be done unambiguously, since there may exist very different views on the nature of things (although most of these views are almost inapplicable in the narrow domain of describing resources). Other properties seem to be less subjective and they are more often provided with a reference; all of them may describe type of the thing even more precisely than a single property could. I agree that it is not the most constructive approach at the moment, but eventually — maybe. Danneks (talk) 15:42, 3 September 2014 (UTC)

This discussion is overlapping one that I think it about to start at Wikidata_talk:WikiProject_Medicine#Adding_disease_properties, but I really think its a generic problem so I want to bring it back here. I think I want to support the idea of avoiding creating new relations like "type of" that are very close to those that exist in the OWL, RDFS, and SKoS worlds. My concern is how we are going to handle the presence of multiple hierarchies within the context of wikidata. For property modeling, when we are bringing in hierarchies that are in use externally and have already been given substantial modeling attention to support real use cases, we should try to leave their structures in place as much as possible. To me, this would entail the use of the same relations that they use. If one ontology uses subclass, we import it using that relationship. If another uses narrower-than, than we use that. For the poly-hierarchy problem, I propose that we bring in any ontology that has users, maintainers and applications. When we have multiple subclass relations coming from one entity in wikidata, we differentiate them by specifying the source for each in relation statement (or as a qualifier?). For those that want to fight the hard ontological battles about which is the correct or best ontology, lets fight that battle with the 'preferred' construct. Thoughts? @Emw, Danneks, Emitraka, Micru, TomT0m, Tobias1984: @Lschriml: --Genewiki123 (talk) 17:39, 19 September 2014 (UTC)

Not sure of what we are talking about, class hierarchies ? for the narrower than versus subclass of, I guess this might be essentially the same property just with a different label, if we look carefully. 100% OK to source hierarchies from where they come from, that's the spirit. I think metaclass modeling is a powerful tool to deal with different hierarchies : imagine a class of class <Class of this very good Ontology>, with
< Class of this very good Ontology > subclass of (P279) < class >
, then any class of the very good ontology is an instance of it. Then it's easy in a query to filter the classes of this ontology by quering only those who are instance of this metaclass. 17:58, 19 September 2014 (UTC)
Narrower than is really not the same as subclass of. See skos-broader-vs-subclassof for some discussion. As an example, the MeSH hierarchy makes sense to be modeled with narrower-than (rooted in "MeSH Category") but would not really make sense to convert to subclasses. My point is that we should be bringing the external semantics in as unchanged as possible for any reasonably well-established ontology (or controlled vocabulary, directed graph, etc.). I'm not really getting your example @TomT0m:. Is your point to add a construct that simply indicates that a particular wikidata entity is an instance of an ontology Class? Perhaps you could fill in with a realistic example so I can understand? --Genewiki123 (talk) 19:52, 19 September 2014 (UTC)
Genewiki123, as you know, there are many ontology languages out there. (For a moment, let's set aside purported differences among ontologies, controlled vocabularies, terminologies, knowledge representations, concept models, simple knowledge organization systems, etc.) OWL, SKOS, CycL, KIF, OBO, etc.: these languages are all skinning essentially the same cat. Wikidata has an ambitious goal to structure all human knowledge from over 250 natural languages. I think it would be best to do so with one ontology language, namely OWL 2 DL, the most current tractable version of the W3C standard web ontology language.
I agree that we should avoid creating additional basic membership properties like type of or narrower than. instance of (P31) and subclass of (P279) -- which have the semantics of rdf:type and rdfs:subClassOf -- are enough. Those properties are how the rest of the Semantic Web ontologies out there build concept hierarchies and instance layers. They give us great tool support like that available in reasoners, Protege, NCBO Portal, etc. At least as importantly, sticking with only the two basic membership properties used in OWL avoids fragmenting Wikidata's terminology among the many competing knowledge representation standards out there -- including our own. https://www.google.com/search?q=mesh+hierarchy+in+owl reveals many reasonable castings of MeSH into OWL, so I think we can handle that adequately given just instance of and subclass of, as we can with other ontologies.
Your suggestion about a polyhierachy is worth considering. I am not opposed to importing certain external subsumption hierarchies, but what I do want to avoid is Wikidata becoming a hosting service for the concept hierarchy of every notable research organization under the sun. For the particular domain under discussion here -- biomedical ontology -- the NCBO project at Stanford is probably better suited for that. I'd prefer to not have to sift through dozens (or, really, even a handful) of subclass of claims to navigate up a concept tree of, say, a given kind of disease.
Here's how various ontologies classify the same major diseases:
Alzheimer's disease
• subclass of dementia, taupathy (DOID:10652 (unlinkable?), Disease Ontology)
• subclass of abnormality of the central nervous system (HP:0002511, Human Phenotype Ontology)
• subclass of dementia (C10.228.140.380.100, MeSH)
• subclass of central nervous system (!? MTHU004582, OMIM)
• subclass of cerebral degeneration presenting primarily with dementia, dementia (26929004, SNOMEDCT)
• subclass of other degenerative diseases of the nervous system (G30, ICD-10)
• ...
Cancer
Do we really want all of these hierarchies in Wikidata? I lean towards no. I think it's too much of a cost for enabling what's likely to be a rare use case: comparative analysis of how various ontologies classify concepts in their respective taxonomies. I think our approach with authority control properties is a great model for pluralism in ontologies and would be a better, less cluttered solution that could facilitate that use case of Wikidata as a data hub. Academics who want to do ontology comparisons could use the various authority or ontology IDs on a given subject to pull in third-party ontologies and do analysis outside Wikidata.
I think we should emulate the OBO Foundry here. As explained in this paper, OBO covers many domains, but each concept is generally handled by one domain ontology. There are many ontologies, but they are orthogonal in subject matter. "The orthogonality principle helps to reduce the need for arbitrary decisions between equivalent-seeming terms drawn from different ontologies". The OBO Wiki's notes on inter-ontology links with is_a seems relevant here. (is_a in OBO is equivalent to rdfs:subClassOf in OWL.)
In summary, I think we should avoid creating more basic membership properties beyond the OWL ones we have now, even if they are used in outside ontologies language. Also, I support the idea of importing certain ontologies, but I think we should generally choose one ontology for each domain. I'm enthusiastic to see where this goes with medical ontologies in particular. Emw (talk) 02:56, 20 September 2014 (UTC)
@Emw: A question, there is some stuffs I think do not fit well into the OWL ontology class/instance pattern, I'm thinking of sciences or fields of studies, for example maths subdomains. To express this, I think maybe some more fuzzily defined properties like those in SKOS relationships could be useful. For example lets take the example of abstract algebra (Q159943) :
In algebra, which is a broad division of mathematics, abstract algebra is a common name for the sub-area that studies algebraic structures in their own right.
This is a definition of abstract algebra, I do not think it's easily translatable in OWL, at least without adding some properties like study of to link a field to the object it studies. Yet it's useful to link a domain to close domain (seems obvious that there is a relationship beetween algebra and abstract algebra), and maybe doing things more like in a thesaurus like fashion would release some of the modeling headheach. The meaning of these properties are fuzzier by design, yet they have proved to be useful at least for human navigation in information.
@Genewiki123: We should clarify some things : it seems the SKOS relationships are mainly designed to build thesaurus. Thesaurus links terms to other closely related terms, for example to terms with closely related meanings. This is a different thing than the Wikidata model where items relates to concepts, which may be terms, for a wiktionary item for example, but usually not. Wikidata concepts can have different labels in any language, and the important thing is mainly the definition, not the terms used to relate to the concept. Which means it's ontologically complicated imho to use skos-like relationships in Wikidata items : we can't refer to the label, as there is many labels, and labels in one language may have different relationships to related labels in other languages. So if we use skos relationship we have to be careful : for example it seems OK to say a field of science is broader or narrower than another one, or that a term in a language is broader or narrower in his meaning than another one, but we can probably do better than that with more specific properties like hyperhonym or antonym for terms, or subclass of and complement of for objects. Of course we can already use instance of (P31) to identify an item to a term or instance of (P31) <concrete object> or instance of (P31) <science>, and we will be able to know that terms have meanings and make the link from a term to the item of its meaning. 09:58, 20 September 2014 (UTC)
I pretty much agree that we don't need any new properties, if anything, we need to take a look at those that we have right now and see if we really need those and if they are used. Another thing to consider is that they also need to be properly defined. I am trying to compile some data on that right now. emitraka (talk)
@Emw: I don't think we should attempt to limit wikidata to use only OWL constructs. IMHO the main point of OWL is to enable inference based on logical definitions. That sort of thing is not part of what I see in the wikidata technology. I mean.. we can't even run inference on a directly asserted subclass hierarchy in here. What I do see is a very nice way to lay out directed graphs with evidence and that structure fits a lot of different things including SKoS. @TomT0m: Sure: I think its important to avoid thinking of wikidata items as "terms" : I would not want to see "synonym" as a property linking items in here. I do think there are situations that are better modeled with narrower-than than with subclass so I propose to keep it on the table. --Genewiki123 (talk) 00:03, 23 September 2014 (UTC)
@Emw: Regarding wikidata and the OBO Foundry concept. I think this is a pretty fundamental question about what we think wikidata is or should be. Is this a place where one 'truth' is sought or is this a place where we record claims and evidence? I tend toward the latter. In deciding whether a particular ontological relationship (or any relationship really) should be recorded here, my only question is "is it useful"? I don't see a problem with the long list of different subclass relations you put up there because this is a database we are talking about - not an end-user application. The applications that will soon sit on top of this database can, assuming that appropriate evidence is also recorded, decide which of those subclasses is their preference. In terms of supporting the construction of Watson-like systems (which can handle disagreements/noise), the richer and larger the network we assemble here the better. The OBO Foundry and OWL work well together and have their place and purpose: to enable coordinated work on knowledge representation and to support logical inference based on axioms. We should build on their work but not be constrained by them. --Genewiki123 (talk) 00:03, 23 September 2014 (UTC)
@Genewiki123: Then do we have a good definition of narrower than ? It can be great to be a little fuzzy, but it can also be a weakness. When used for a thesorus we know that it imply some term refer to something more specific, but for something else ? What would be the domain of this property, stuffs like sciences ? 08:57, 23 September 2014 (UTC)
@TomT0m: The W3C definitions for the SKoS semantic relations are good place to start. They also provide a list of use cases in many domains they produced when coming up with the standard. In relation to OWL, you might think of it as a bit 'looser'. While it does have semantics that can be used for logical inference (e.g. transitive relations) they are nothing like what you can do with OWL. The model is also different in that its central units are skos:concepts rather than Classes and Instances. Like I said above, I'm an inclusionist at heart.. This is just one more important scheme for knowledge organization that we should be aware of and think about supporting - when there are users and applications that could benefit. Not trying to say that everything should be done like this. --Genewiki123 (talk) 16:28, 23 September 2014 (UTC)

## A better way to model causation

Danneks, Genewiki123, Infovarius, Micru, TomT0m, Tobias1984, Wikidata currently has one very narrow property to model causation. A broader and more robust approach would help build interesting applications. Please see Property_talk:P828#A_better_way_to_model_causation and give your feedback! Thanks, Emw (talk) 03:20, 9 September 2014 (UTC)

## Modelling systems

It is something that I have been working on for some time, and now it is ready for discussion: Modelling systems on Wikidata. The questions I wanted to address are:

• what does the top "entity" means for wikidata?
• how to represent emergent systems?
• what are the intrinsic qualities of any entity?

I hope you find this approach satisfactory.--Micru (talk) 20:01, 3 October 2014 (UTC)

@Micru: If I understand, you are saying that "entity" always represents a matter-energy structure, including an abstract "class" which is a matter-energy process of an observing system during perception (in someone's brain). An important part of this process is "instantiation" where the observing system selects a model to represent the signals it is getting. So, the Wikidata "X instance of Y" property is really a claim about the behavior of an observing system: When it gets the signals for X it selects the model Y. Finally, if we understand that Wikidata classes and "instance of" properties really describe the process of an observing system, then we can more easily model when observing systems observe other observing systems ("recursion"). For example, in Predatory lending (Q949033) like in the United States housing bubble (Q2928006), "the bank understands that the borrower doesn't understand the terms of the loan." To represent what caused the United States housing bubble (Q2928006), we must represent the misunderstanding of the borrower as an observing system. Is that right? Jefft0 (talk) 11:50, 26 October 2014 (UTC)
@Jefft0: That's correct. I still have some doubts about what is the best way to represent that a certain observer uses its own tree of classes. I am also a bit stuck characterizing cohesion in dissipative systems. In a way, what in natural language we call "not understanding" means that a certain concept still doesn't have functional cohesion with the required supporting concepts, but under which circumstances can be said that such cohesion has been reached? These are tough questions that will require time, thanks anyhow for taking a look.--Micru (talk) 17:34, 27 October 2014 (UTC)
@Micru: Yes, the lack of cohesion is why (in this example) the borrower doesn't understand. But the bank person infers this lack indirectly by the positive observation that the borrower is sitting calmly and nodding. The bank person knows that the borrower understanding the terms of the loan implies that the borrower runs out the door. So, the bank person observer system selects the model sitting (Q1144593) which implies not running (Q105674) which implies the borrower not understanding. Can Wikidata express that models for sitting (Q1144593) and running (Q105674) cannot be selected at the same time for a signal? (They are disjoint.) Jefft0 (talk) 19:57, 27 October 2014 (UTC)
@Jefft0: sitting could be described as "exerted muscles:none", plus body position. Running has "exerted muscles:X, Y, Z", so both models are already mutually exclusive. Apart from that, no, there is no way to express "disjoint", "some of these", or "none of these".--Micru (talk) 16:22, 28 October 2014 (UTC)
@Micru: For "disjoint", I notice that opposite of (P461) is used a lot, and it is not just a strict opposite (it is more like "disjoint"). For example, monotheism (Q9159) opposite of (P461) atheism (Q7066). Jefft0 (talk) 10:23, 2 November 2014 (UTC)

## Process vs. event

What is the difference between upper ontology items process (Q3249551) and occurrence (Q1190554)? An event has a specific time and place, but a process does not. We have many Wikipedia articles and Wikidata items like eating (Q213449) as a general process. But what about an individual event of eating a poison soup at a particular time which was a cause of someone's death? This particular event is not an instance of a general process, it is an instance of an event. Should there be a different item for the event of eating vs. the general process of eating? Or if all events are processes, should occurrence (Q1190554) be a subclass of process (Q3249551)? Jefft0 (talk) 15:23, 28 October 2014 (UTC)

According to wiktionary a process is "a series of events to produce a result", so I guess it would be right to say . "Eating a poison soup" is also a process, as it can be subdivided in smaller events. I do not reach to see your reasons to state that "a process doesn't have a specific time and place".--Micru (talk) 16:54, 28 October 2014 (UTC)
@Micru: I'm only trying to understand the difference between process and event. For example, football (Q1081491) is a subclass of process, but association football match (Q16466010) is a subclass of event. 2014 FIFA World Cup Final (Q15926885) is an instance of association football match (Q16466010), therefore an instance of event. This makes sense. Is there anything that is an instance of the process football (Q1081491) so that 2014 FIFA World Cup Final (Q15926885) is a part of it? On the other hand, we have 1985 Brixton riot (Q4582222) which is instance of riot (Q124757) which is a subclass of process, not event. Why is 1985 Brixton riot (Q4582222) a process, but 2014 FIFA World Cup Final (Q15926885) is an event? Jefft0 (talk) 18:01, 28 October 2014 (UTC)
One could say that whether something is an event or a process depends on the observer. Eating some soup could be a process if you are interested in just how it happen at that specific instance, but if you are interested in a specific persons life then a specific meal is just an instance event. Another way to imagine the difference is like this; a process contains subprocesses, while a state machine contains events. Eating a soup can trigger an event, the person is poisoned and undergoes a state transfer into state "dead". It is all in the observer. Jeblad (talk) 02:05, 30 October 2014 (UTC)
@Jeblad: "It depends on the observer." But for representing items in Wikidata, we have to choose a point of view which can be publicly verified. It sounds like we can always divide an event and "zoom in" so that it looks like a process containing smaller events. So "events in a series" in a process are themselves subprocesses. The things that are true of process are also true of an event if examined closely. Therefore, I propose that occurrence (Q1190554) is a subclass of process (Q3249551). (I believe SUMO does this.) Jefft0 (talk) 10:47, 30 October 2014 (UTC)
If some small process changes the state of the object, then I guess you can say it is an event. But hey, whether something changes the object also depends on the observer! =) Jeblad (talk) 12:48, 30 October 2014 (UTC)

## Spatial region, spatial position, coordinates

@Jeblad: suggested some spatial properties on WD:PP/PLACE: spatial length, spatial width, and spatial heigth. This is quite interesting, but how can the system of coordinates or a main axis be defined? And how to represent the position of one part respect another? Is a geo-shape enough or 3d models are required for more general uses like anatomy?--Micru (talk) 17:26, 28 October 2014 (UTC)

The object itself has a location, this is the extent of the object at that position. It is not the shape, it is like a rotated bounding box. There might be an extension along the axis to a reference plane, but it is not clear how this is best expressed. There might also be a rotation around the axis, but usually this will be "unknown value".
I'm not sure, but I wonder if it is better to reformulate this as a generic extent that is an instance of width, height, length, etc. A bit more typing but would be more flexible.
My suggestions can encode the extent of 3D object, but encoding the actual shape is a lot more complex. When you start to relate parts to each other you need local coordinate systems that relates to other such systems, and when you go from static to dynamic systems you both physical factors and limits to movements. Imagine the difference between modeling a w:en:chair and a w:en:pogo stick. Jeblad (talk) 23:35, 29 October 2014 (UTC)

## Higher order classes

I'm trying to work out how Wikidata represents higher-order collections, and I'm a little confused. For example, we have:

``` egg (Q17147) --instance of (P31)--> development stage of animal (Q20056177) --subclass of (P279)--> animal (Q729)
```

This suggests that the class "eggs" is an instance of the class "animals" (like Douglas Adams (Q42)), but I would have considered them both to be first-order classes.

Similarly, we have:

``` animal (Q729) --parent taxon (P171)--> Filozoa (Q1131559)
parent taxon (P171) --subproperty of (P1647)--> subclass of (P279)
```

Now it makes sense for animals to be a sub-class of Filoza, but what is a taxon?

``` animal (Q729) --instance of (P31)--> taxon (Q16521)
```

This implies that "taxon" is a second-order class that the class "animals" is an instance of.

``` taxon (Q16521) --has part (P527)--> organism (Q7239)
```

This implies that the class of organisms (presumably a super-class of animals) is a part of taxon.

I expected to find a property like "type of", relating a higher-order class to some other class, whereby the instances of the former are sub-classes of the latter. Are we trying to make do without higher-order classes here?

Cheers, Bovlb (talk) 12:45, 15 October 2015 (UTC)

Don't rely on taxonomy items, taxonomists on Wikidata follow their own logic and the result is awful in my opinion. Hard to talk to them. I wrote an essay on the topic : Help:Classification. There is a long history of debate in Wikidata, and beetween POVs like the one of taxonomists with a lot of specific classification properties, which we deleted for a lot of them except a few like "parent taxon", and the fact that some people wanted to follow hardcorely type–token distinction (Q175928) with no higher order class at all, with the consequence that they actually used higher order class in the same hierarchies than non higher order one, things have been kind of tough on that matter.
You're right about everything, I'd express with something like
instance of (P31) < metaclass >
of (P642) < organisms >
for example. author  talk page 13:01, 15 October 2015 (UTC)
Thanks for the response. I note that inconsistent hierarchy for taxons is suggested in passing by the data model description:
The exact meaning of an Item cannot be captured in Wikidata (or any technical system), but is discussed and decided on by the community of editors, just as it is done with the subject of Wikipedia articles now. It is possible that an Item has multiple "aspects" to its meaning. For example, the page Orca describes a species of whales. It can be viewed as a class of all Orca whales, and an individual whale such as Keiko would be an element of this class. On the other hand, the species Orca is also a concept about which we can make individual statements. For example, one could say that the binomial name (a Property) of the Orca species has the Value "Orcinus orca (Linnaeus, 1758)."
Bovlb (talk) 22:37, 16 October 2015 (UTC)
@Bovlb: I too am somewhat confused about how Wikidata is actually trying to handle higher-order classes. The way it should happen is clear enough, I think. First-order classes in some domain, e.g., diseases like malaria (Q12156), are organized into a hierarchy using subclass of (P279) links. There may or may not be a "universal" class, e.g., disease (Q12136). There will also generally be intermediate classes in this hierarchy, e.g., no label (Q18555201). Then there are second-order classes, e.g., a slightly different notion of disease like "named disease" which has as instances some of the first-order classes, here malaria (Q12156) but not no label (Q18555201). However, this is not how the domains that I have looked at actually work. Instead there there are instance of (P31) links between the first-order classes, e.g., from malaria (Q12156) to parasitic infectious diseases (Q1601794) which is itself a subclass of disease (Q12136). This mixing of levels makes it very hard to determine what is going on and even harder to figure out how to add new classes. The second-order classes might not be present, eliminating a very useful abstraction that can be used by both users and tools. There are sometimes also redundant subclass of (P279), e.g., from malaria (Q12156) and other diseases to disease (Q12136). The purpose of these redundant links are not given. In all, quite a mess. Peter F. Patel-Schneider (talk) 00:33, 20 October 2015 (UTC)
Community lacks good rules about this, and quality ways to enforce the consistency of the class and higher order class trees. One step would be to create a WikiProject Desease do not exist (yet?) Will you be the its creator ? click here., to amend and expand Help:Classification, and to vote in this RfC to make this a community official guideline for everyone to accept the principle of higher order class and the distinction between instance of (P31) and subclass of (P279) hich has been hard to explain to community correctly (especially because some people at the beginning was denying the principle of higher order class at all. There is two opened RfC to make good classification principles guidelines of Wikidata : Adopt Help:Classification as an official help page and Help:Basic membership properties. Then we can develop constraint and queries to check the consistency of the class tree, like queries who checks that no order-1 class in in the orderé2 class tree for example. author  talk page 11:02, 20 October 2015 (UTC)

I would like to invite comment for this RFC: Wikidata:Requests for comment/Are colors instance-of or subclass-of color. --Tobias1984 (talk) 14:21, 15 October 2015 (UTC)

## WikiProject for constraints checking

I was trying to work out which WikiProject is responsible for the constraints checking. Is that WikiProject Ontology or WikiProject Reasoning or another one? Cheers, Bovlb (talk) 19:13, 21 October 2015 (UTC)

@Bovlb: There is no WikiProject actually responsible for all constraint checking right now. How this works right now is that some Wikidatans are especially interested in a property, or thematics WikiProjects try to enforce (and define) "their" constraints. We might lack consistency beetween project however with that system. I don't really know if a dedicated constraint checking WikiProject would help. There might be statistic pages on constraint if I remember well. author  talk page 09:25, 22 October 2015 (UTC)
I think the Wikiprojects should manage their properties and work on the Constraint violation reports. The problem is that there is very little activity on those projects. We really need to get more people to activly participate on those projects. It is maintaince work and answering questions, but it is a vital part of the ecosystem. - The other problem is that a lot of things are very tedious things like correcting "color -> champagne (region)" to "collor -> champagne (color)". And that means that each WikiProject would at least need 1 person that can write a bot and can run it within a week. Otherwise people just give up on correcting hundreds of wrong claims. --Tobias1984 (talk) 09:56, 22 October 2015 (UTC)
On your example, no need to be able to write a bot. Autolist2 (alone) or Autolist2 + PagePile if needed can do miracles. We need to make those tools more popular. P
PS: on frwiki people get more and more used to Wikidata, so some of them might come here and maybe eventually we can expect more people here :) I heared that dewiki finally had a RfC on Wikidata usage, so the same might arise from there. author  talk page 10:22, 22 October 2015 (UTC)
Thanks for the response. (And thanks for the pointers to AutoList 2 and PagePile.) I understand that specific WikiProjects are responsible for handling the lists of known violations. My question was about where we discuss how constraints checking takes place. For example, if I wanted to find out what type of constraint checking we do and why, or if I wanted to propose a new type of constraint checking (say, just for example, disjointness), where would I do that? Cheers, Bovlb (talk) 01:11, 23 October 2015 (UTC)
Oh, I see. For the existant : this had been Ivan A. Krestinin (talkcontribslogs) talk page as its bot KrBot was the one who wrote constraint reports (but we sadly never could make him publish his code), and from what I understood the metadatas Wikidata:Property_proposal/Property_metadata but I have not really been involved in this discussions and I don't know what's happening here, and I don't know how the special constraint checking page works. This seems a little bit stalled atm. For say disjointness, I asked Ivan on Wikidata:Property proposal/Generic § subclass and if I recall well he's been pretty clear that his paradigm was property based, not class based, so he would not implement this. There might be a trick to translate a class disjointness constraint to a set of property constraint, something like "Any property that apply only to class A might never be used to any property only appliable to B", or if we have one characteristic property for class A and class B we can limit that to one constraint. But this implies some kind of property number inflation distinguished sometime only with the type they apply to, and it's not especially user friendly afai can see. It may also expression of some "disjoint union of" constraint not especially esay to do. In particular if the types have no really specific properties.author  talk page 10:36, 23 October 2015 (UTC)
But you're right, we definitely needs help page for constraints and to organize the effort on building that modelling tools. author  talk page 10:37, 23 October 2015 (UTC)

## Metaclass -- merge?

Is there any reason to have Wikidata metaclass (Q19361238) ("class of wikidata classes") distinct from metaclass (Q19478619) ("in knowledge representation, a class of classes") ?

Items are metaclasses if the concepts they represent are metaclasses, which surely is a property of the concept itself (so Q19478619).

Does it make any sense to have the confusingly separate Q19361238, when we're trying more generally to move away from wiki-specific conceptualisations, towards defining class items in ways that are more broadly meaningful (eg away from "Wikimedia history article", towards aspect of history (Q17524420)) ?

I did this, and it's a conservative choice to avoid theorical problems. The metaclass concept in general have a lot of implementations, which might or might not be equal to the Wikidata concept, which is an implementation of this, that might or might not be equal to others. I choose what seemed to be a safe option : don't mix Wikidata implementations with the generic concept they implement. But indeed this might be confusing without a clear guideline in description. author  talk page 10:19, 23 October 2015 (UTC)
Q19361238 currently has part of (P361) Wikidata (Q2013), which is obviously incorrect given its current usage. automobile model (Q3231690) is not part of Wikidata. --Yair rand (talk) 01:18, 27 October 2015 (UTC)
This is disputable /o\. Wikidata is about concepts, and that's what "car model" is, a concept, abstract. But I must admit this is conceptually not clear. author  talk page 08:58, 27 October 2015 (UTC)
More to the point, automobile model (Q3231690) is not just part of Wikidata -- and that is surely the key test. Otherwise one could say that anything with a WD article was "part of" Wikidata. Jheald (talk) 11:37, 27 October 2015 (UTC)
Yes, but the purpose of this item is actually to denote the metaclass which actually are used in Wikidata, not to other kind of metaclasses. This does not imply that they do not denote something else. author  talk page 11:51, 27 October 2015 (UTC)
I guess a metaclass in python could have an article in Wikidata, but it would not be an instance of this item. author  talk page 11:54, 27 October 2015 (UTC)
For that sort of metaclass there is metaclass (Q1924819)   Jheald (talk) 12:23, 27 October 2015 (UTC)
Yes, but if we choose to follow some principle "class are classes of tokens, metaclasses of order 1 are classes of classes of tokens and should only be subclasses of order 1 metaclasses, order 2 ..." then our metaclasses may be different from classes used in other modelling languages who would not follow this principle. Plus the move to have separate items is also conservative. author  talk page 18:54, 27 October 2015 (UTC)
I have recently been thinking (for other reasons) that it would be good to introduce a qualifier "applies to aspect" (synonym: "in relation to", "with respect to") for more general cases of qualification than should be handled with applies to part (P518).
Perhaps the right expression is automobile model (Q3231690) is an instance of (P31) metaclass (Q19478619) "in relation to / with respect to" Wikidata ? Jheald (talk) 12:19, 27 October 2015 (UTC)
I don't get if that would help. What for ? author  talk page 12:23, 27 October 2015 (UTC)
My understanding was that you were concerned that in some ontology other than Wikidata, automobile model (Q3231690) might not be expressed as a metaclass, so you had reservations about making automobile model (Q3231690) is an instance of (P31) metaclass (Q19478619) as a general statement, using the general item for "metaclass in a knowledge representation system".
It seemed to me that adding the qualifier on the statement might mitigate that concern, that I thought had motivated your 'conservative' creation of the separate item Wikidata metaclass (Q19361238). Jheald (talk) 12:29, 27 October 2015 (UTC)
I don't think so. For one thing it's higly redundant to add this to any Wikidata metaclass. Secondly the "metaclass" notion in any representation system might not map the definition we use here, who might be a special case, or even inconsistent with the way metaclasses are used in other system. author  talk page 12:41, 27 October 2015 (UTC)
Comment "The metaclass concept in general have a lot of implementations, which might or might not be equal to the Wikidata concept". Sorry TomT0m, but I think all the problem is in that sentence "which might or might not". We don't know about what we are discussing so all the work done is useless. Snipre (talk) 13:21, 27 October 2015 (UTC)
Well, I think I did my part. Help:Classification might not be what you expect, but it seems to me that it's usable. author  talk page 13:46, 27 October 2015 (UTC)
It is not me who started this section. Seems that your document is not so clear for others persons too. Snipre (talk) 17:50, 27 October 2015 (UTC)
The clarity of the concept was not really the point of the discussion. It's wether or not we should have a specific item for Wikidata metaclasses as used in our data model wrt. the concept of metaclass in modelling languages in general. Wikidata is a particular case, so it's safe to have subclass anyway and it could avoid problems. author  talk page 18:49, 27 October 2015 (UTC)
The clarity of the concept is the point. If the concept is clearly defined no one will come here and ask for a fusion. Snipre (talk) 19:05, 27 October 2015 (UTC)
With all due respect, the meaning of "Wikidata metaclass" is bloody obvious:
An item represents a "Wikidata metaclass" if instances of it also have subclass of (P279) statements.
The question is whether it is worth distinguishing this specific operational criterion on Wikidata from the encompassing notion of a metaclass in any knowledge-management system. Jheald (talk) 19:18, 27 October 2015 (UTC)
The concept of metaclass is a nightmare if we consider inheritance because we have to be able to take them out when doing inferences. I take the examples form User:TomT0m/Classification to illustrate
HMS Aboukir (Q5631188) instance of Albion-class ship of the line (Q4121227)
Albion-class ship of the line (Q4121227) subclass of ship (Q11446)
Albion-class ship of the line (Q4121227) instance of ship class (Q559026)
According to inheritance concept which is the fundamental of subclass structure we can say:
HMS Aboukir (Q5631188) instance of ship (Q11446) (this is ok), HMS Aboukir (Q5631188) instance of ship class (Q559026) (this is not ok).
Metaclass is a good thing for maintenance purpose or queries but when mixed with the ontology tree this implies some filtering when doing inferences.
And by the way the core of the discussion is not metaclass concept but creation of several items with not enough described characteristics to allow a clear identification of each one. Snipre (talk) 21:55, 27 October 2015 (UTC)
instance of (P31) is not transitive, so your assertion is wrong.
In general, you cannot assume that statements about a class also apply to instances of that class. Albion-class ship of the line (Q4121227) is a subclass of ship (Q11446); but the fact that HMS Aboukir (Q5631188) is an instance of Albion-class ship of the line (Q4121227) does not imply that HMS Aboukir (Q5631188) is a subclass of ship (Q11446).
Similarly, Albion-class ship of the line (Q4121227) is an instance of ship class (Q559026); but the fact that HMS Aboukir (Q5631188) is an instance of Albion-class ship of the line (Q4121227) does not imply that HMS Aboukir (Q5631188) is an instance of ship class (Q559026). Jheald (talk) 22:18, 27 October 2015 (UTC)
@Jheald: "instance of (P31) is not transitive". Why ?
instance of (P31) is not transitive in the traditional model because you can apply it only to instance which doesn't have any child so the question of the transitive mode is not relevant. When you start to apply it to subclasses, you enter in a new world where instance of (P31) can be transitive, this is just a question of definition. The problem is the extension of the use of instance of (P31) from a well structured system to a new system with new rules.
The main problem is the definition of the property instance of (P31) which changes according to the item which uses it. Each time you find instance of (P31) in a structure you have to check on what kind of item it is applied in order to apply the correct process to the data.
Is it wrong ? No, this is just a new structure with new rules and perhaps less intuitive than the traditional one. But what is important is to know the consequences of that choice and this is described in a quite well done document I found (see this pdf. This is even different ontologies according to the use of metaclasses or not (see that document). In both document use of metaclass implies loss of consistency especially in data check.
What is the best solution ? For me we should use a new property for relation between class and metaclass. Things will be clear and more easy to understand. Snipre (talk) 15:36, 2 November 2015 (UTC)
@Snipre: No. It's not because P31 is set membership, and subclass of (P279) is substetting in set theory. This is it because by definition we do not suppose it's transitive. No need to take weird and esoteric and not functional definition. If i'm an (instance of) human, and human is a(n instance of class) class, then nothing should imply that I'm a class, this would not make any sense. But if I'm a human, and all humans are animals, then I'm an animal. That's how it works. You'll find nowhere set membership is transitive, see https://fr.wikipedia.org/wiki/RDF_Schema#Classes_et_sous-classes. Subclass of is because we defined it as transitive (every "a" is also an instance of "B" if "a" is an instance of A and B is a subclass of A).author  talk page 15:44, 2 November 2015 (UTC)

## Use this project for discussing modelling methodology for classes?

It was suggested by Snipre that a project would be a good vehicle for the development of a modelling methodology for classes. Is this project a reasonable place for the work to happen? How can a project be used to develop a guideline? Peter F. Patel-Schneider (talk) 20:59, 23 October 2015 (UTC)

If I propose to work within a wikiproject this is to find people who have knowledge and interest in a common topic. Starting a policy in a corner of Wikidata is the bad way to get in touch with people who can bring their ideas, because Wikidata has many corners and often developments/discusions stop during a certain time and the advantage of a project is its ability to keep connected different pages and to allow everyone to continue the work later.
If I propose to use this project for the development of a policy about classes this is to be sure that we set up the list of the good questions first before starting to answer them.
I am nor an expert but for me classes definition is a part of ontology definition and we should check that any choice we are doing is done with the knowledge of the consequences. The worst case will be to define some policy to see later that some possibilities we never suspect previously are closed due to the choice we did.
I propose to start with 2 subpages Wikidata:WikiProject Ontology/Ontology and Wikidata:WikiProject Ontology/Class. For each page we should start with a list of the main characteritics (with their definition) of ontologies and classes. We should first create a "dictionary" to be sure than people are using the same language. I don't propose to creat something new, if you have good reference, mention them and when possible provide a link. Snipre (talk) 14:44, 24 October 2015 (UTC)

## Wikidata:Wikimania 2016

There is already proposal for an Ontology talk. If you know who could give the talk please comment on Wikidata:Wikimania 2016. - Hopefully all members of this project can comment on the different points on that page. --Tobias1984 (talk) 13:35, 4 November 2015 (UTC)

## Problem with has part (P527) and part of (P361)

I found an interesting problem in the use of part of (P361) which leads to a problem with the inverse characteristic with has part (P527):

Here is a problem for the inverse relation between has part (P527) and part of (P361): this is not always true and we shouldn't have an automatic addition of part of (P361) to an item connected to another item with has part (P527). In practice we need to differentiate the cases by using a new relation. Comment ? Snipre (talk) 15:13, 4 November 2015 (UTC)

@Snipre: I think I addressed this problem in my modified Help:Basic membership properties page. The inverse does not necessarily apply when the item is a class, rather than a non-class object. The meaning of X part of (P361) Y in the case of classes is "an instance of class X is (almost always) part of an instance of class Y". Whereas the meaning of X has part (P527) Y is "In all or most cases, an instance of X has an instance of Y among its parts or components". These are not inverse statements such as in the chloride example you just gave. ArthurPSmith (talk) 16:32, 4 November 2015 (UTC)
Thanks a lot for your link because it shows something I was disputing a lot with user:TomT0m about the use of the same properties for both instances and classes but with different meanings. The sense of the relation is not more depending on the property but depend on the kind of items linked by the property too. This leads to a bigger effort to express inferences because we have to analyze in details the type of items to recover the correct meaning. The big question is then if the use of new properties depending on the type of items instead of the same properties with different meanings is better. This will lead to bigger errors in the use of the properties but at least we can detect them more easily by putting constraints for the use of properties. Snipre (talk) 17:44, 4 November 2015 (UTC)
Yeah. I have always been dubious about using part of (P361) to link a class to another class. For me it should always link an instance to a bigger instance. The consensus seems to be against me on this however. :( Joe Filceolaire (talk) 01:14, 5 November 2015 (UTC)
Actually so do I, the biggest supporter of this is Emw and I disputed with him a few times before surrendering. author  talk page 18:32, 6 November 2015 (UTC)
Instance part of instance and class part of class aren't the only two options. There's also class part of instance. For example, observable universe (Q221392) has part (P527) matter (Q35758) and administrative territorial entity of the United States (Q852446) part of (P361) United States of America (Q30). I think that has part (P527) and part of (P361) should be split up into many more properties. A bunch of different situations:
• X (instance) is made up of distinct components (instances) A, B, and C.
• X (instance) is made up of an unspecified amount of instances of (classes) A, B, and C.
• Every instance of X (class) is made up of precisely one instance of A, B, and C each. (Maybe allow for quantity in cases where there's an exact multiple of one or more of the items listed.)
• Every instance of X (class) is made up of an unspecified amount of instances of (classes) A, B, and C.
The first is the only one that has a proper inverse. The other three sometimes, but not always, have reverse statements. Things get a little more complicated when there are components that don't divide so cleanly. --Yair rand (talk) 02:32, 5 November 2015 (UTC)
For cardinalities, I used in the past :
< X > has part (P527) < A [ B, C] >
quantity (P1114) <  2 >
which works. 2 is an example, but it could be whatever. We could also use min and max as qualifiers. author  talk page 10:29, 7 November 2015 (UTC)
@Yair rand, Filceolaire, ArthurPSmith: I found that *paper speaking about the problem and the proposed solution is to delete the property "part of" and to use only 4 different properties for "has part": has_determinate_part, has_substituent, has_granular_part and has_ingredient. These properties are quite related to chemistry but we can generalize them. If someone have another example from an external ontology this will be more easy to define a new set of properties. Snipre (talk) 12:34, 5 November 2015 (UTC)
@Snipre, Filceolaire, Yair rand: for a group that is trying to be precise I'm a little disappointed you are all using "instance" to mean what I've called a "non-class" - an item for which there is no possible interpretation of it as representing a set having more than one element. Classes of course can be instances of metaclasses, and can be thought of as singular things in some contexts (particularly for things like particles, molecules etc. in physics and chemistry) - there is only one kind of electron, but of course there are many of them. In some contexts 'electron' is an "instance", but it is also definitely a class representing the set of electrons in the universe. There is also some issue here with how you represent time in the ontology - a single entity that exists for a period of time can be considered a class representing the set of instances of that entity at specific points or intervals in time (for example, at least some properties of every concrete object change over time). If the "part of" property for example applies at one point in time but not at another, should it be applied to the entity over all time, or only to the "instance" associated with a particular time interval for which it is actually true, or only for that entity if it is true at the present time, and we remove the property when it is no longer true? And even for an entity that represents something at an instant or short interval of time, for example the 1906 San Francisco earthquake (Q211386) you can have differing perspectives for other reasons - we also have Category:1906 San Francisco earthquake (Q8125160) which lists many items under essentially the same topic.
Ok I should probably get to the point here - basically what I feel is that something like "part of" is extremely useful to represent these sorts of relationships even when there are ambiguities or it isn't precise for all instances of a class or for all times for a physical object. People add it because it makes sense to them, and it makes sense for good reasons. In most cases the ambiguities can be resolved just by looking at the claims on the subject and object items (whether an item is a class or not for wikidata purposes is easy to find). But there is also a place for more precise properties where ambiguities could cause trouble - let's come up with some real examples where it's an issue and see what properties we really need. The only example mentioned so far I see as being wrong or ambiguous is chloride ion (Q108200): part of (P361) sodium chloride (Q2314) - I think that relationship should be removed as it is not always or typically true. Maybe we do need a "sometimes part of" or "ingredient of" property to represent a case like that. I do agree we cannot rely on the inverse relationship between "part of" and "has part", if that's been asserted anywhere it should probably be removed. ArthurPSmith (talk) 16:21, 5 November 2015 (UTC)
We are not going to eat this elephant at one sitting. As a first step I suggest we propose new properties:
• has component part
• component part of
• series element
• part of series (We have series (P179))
• has ingredient
• ingredient of
proposing all six at the same time should help us explain to people why 'has part' and 'part of' are not quite enough and lack nuance. Also it makes us look like a proper organised wikiproject.
@ArthurPSmith, Snipre, Yair rand: anyone support this? Joe Filceolaire (talk) 05:23, 6 November 2015 (UTC)
@Filceolaire: : that lacks a little explanation of the properties and examples. I don't for example see that "series element" is interesting. For orderable elements, we should explain how to note the order. The "cardinality" problem should also be addressed in our proposal. Does "has component part" / ingredient part dichotomy maps to "determinate / "granular part" of Snipre's paper ? @Snipre: By the way, you should have noted that they use the token/type relationship the same way I and others try to push and that they note that CheBi is not really clean on this. And that the "molecule/substance" dichotomy should be explained, something I point for quite some time now :) author  talk page 11:14, 7 November 2015 (UTC)
TomT0m All these are designed to replace "part of" and "has part" as currently used in some places on wikidata. I had hoped they were pretty self-explanitory, at least to the people who hang out on this page.
At the moment "has part" is used as the inverse of series (P179). You could I suppose use "has component part" for book, film and sports series but that seemed (to me) to be against the material/abstract split discussed here so I added "series element" as another easy win. We would note the order exactly the way we note the order now on the items for the elements for these series - not via this property; though series ordinal (P1545) could be used as a qualifier if you really wanted to move this info from the series element items to the series item.
I know what "has component part" and "has ingredient" mean but I am not completely clear what "determinate / "granular part" mean. I hope they are the same.
These five properties were proposed as they can be slotted in as like-for-like replacements for "part of" and "has part" in a wide number of instances on wikidata leaving the more difficult cases to be dealt with later. Joe Filceolaire (talk) 23:42, 7 November 2015 (UTC)
I know what "has component part" and "has ingredient" mean but I am not completely clear what "determinate / "granular part" mean. I hope they are the same. : thanks for the context but I'm aware of it, as you supposed :) My point was that for me, no, I don't know what "has component part" and "has ingredient" means precisely, and that's for what I expected clarification. There is an explanation about determinate part is : a part like a finger, who may have its own name (index, ring, ...) which when removed from the whole disminished the whole. Granular part : a part that when removed from the whole, do not change anything. If you remove a molecule of water in a mixture, or a little bit of the mixture, usually you still have the same mixture as a result, and the whole is not really disminished.
I also don't understand what are exactly supposed to mean counted/weighted in the sentence I think properties 1,2,5,6 I proposed above correspond to "physical part of" but with the distinction as to whether the part is weighed (property 5,6) or counted (Property 1,2). : In chemistry for example the molar mass (Q145623) notion is used to convert a weight to a count, so this dichotomy is pretty naïve and not really suitable for use in chemistry. This needs more explanation to be usable to me.
TomT0m In chemistry a mixture or solution is weighed - the parts are specified by % or by gr/gr or mg/litre. Even if the unit of measurement is the mol it can still have decimal places so we would use 'has ingredient/has granular part'.
A compound is counted - the parts are specified by whole numbers of molecules - so we could use 'has component part/has determinate part'.
But this distinction is, as you said, naïve and probably not suitable for chemistry. I would expect chemistry to have it's own defined properties. The properties I proposed above are for simpler cases. Joe Filceolaire (talk) 16:29, 9 November 2015 (UTC)
@Filceolaire: Yes, actually if you read Snipre's paper, they define "ingredients" as a mixture of "grains retaining their identity", which means I guess not involed in chemical reaction with each other, that can be separated later by simple mechanical mechanisms such as chromatography. author  talk page 16:52, 9 November 2015 (UTC)

Please see Refining "part of" for previous discussions about this, especially Markus' comments who are especially interesting. For mine : I'll comment all this later this week end. author  talk page 17:20, 6 November 2015 (UTC)

I'll quote Markus' proposal :
Physical part of :. A relationship between two concepts that describe things you can touch: A is a part of B if the atoms of B are a superset of the atoms of A. This subsumes the suggested "organic part of" and "manufactured part of". It also subsumes the example "Africa-Earth" from "subsystem".

Process/event part of : A relationship between two concepts that describe events or processes. "Part of" holds if the events that make up one of the concepts are also part of the events that make up the other concept. The example above is "Third Punic War part of Punic wars" and maybe also "1397 part of 1380s".

Metaphorical/figurative part of : Abstract concepts that do not have a direct manifestation as physical objects or events may still be defined to consist of parts. This is natural when a physical/temporal relation is "almost" there: "Chapter 1 is part of the book" (if you print the book, the pages of Chapter 1 are a physical part of the book), "The first movement of the symphony is part of the symphony" (if you perform the piece, the events associated with the first movement are part of the events of the overall performance). This case covers the examples given under "creative work" above (I don't think that the idea is in any way specific to creative works though, e.g., "adolescence is part of childhood").

See his subsequent comment for a complete argumentation. author  talk page 18:29, 6 November 2015 (UTC)

I think properties 1,2,5,6 I proposed above correspond to "physical part of" but with the distinction as to whether the part is weighed (property 5,6) or counted (Property 1,2).
Property 3 is for all of those series of TV shows, books, sports leagues which use "has part" at the moment. I have no idea how this relates to "Metaphorical/figurative part of" which is my problem with TomT0m's examples above. It is not clear how they should be applied. Until we are clear lets leave them aside and do the bits we are clear on. Eh? Joe Filceolaire (talk) 09:12, 7 November 2015 (UTC)
@Filceolaire: Markus' explain later the concept of "figurative part of" : a book's introduction is a part of a book (as an artwork) in the sense that in every printed exemplar of this book countains pages that contains the book, or that reading the introduction is part of the experience of reading the book. But "the book introduction" is an abstract object (by the way, we fail to make the distinction beetween the book as an artwork and the book as a physical object in Wikidata atm.) author  talk page 11:34, 7 November 2015 (UTC)
TomT0m: There are few, if any, wikidata items for book introductions and where these exist they are, I believe, by other authors and are treated (by wikidata) as separate literary works. I don't think using that as your example helps me understand how "figurative part of" would be used on Wikidata.
If by "book as an artwork" you mean "book as a literary work" then I think there is fairly widespread understanding of this concept with a lot of books getting statements that they are <instance of:literary work> (or some specialised subclass of literary work such as novel, biography, textbook etc.) and <subclass of:book>. Joe Filceolaire (talk) 23:42, 7 November 2015 (UTC)
@Filceolaire: Well it seems we definitely fail to understand each other. Say a work is a collective creation, with several authors, and that someone wanted to comment the specific part of the work written by someone who got so famous that it was commented as a work as such, and that that part is famous and has its own item or wp article. It's clearly, at least at start, a part of the collective work. We can also take the, probably more convincing, example of a fiction figure that appears in a work. In which sense is this a part of a work ? author  talk page 11:16, 8 November 2015 (UTC)
Oh, I also note that when I find an element about a book, I usually see it as an instance of book. It is really unclear to me what are the guidelines of WikiProject Books are and that there is a real concensus on this. author  talk page 11:30, 8 November 2015 (UTC)
I am not sure what sense a chapter is "part of" a book. That is why I proposed we start with stuff where we can probably achieve agreement and look at the more metaphysical cases later. On the question of <instance of:book> versus <instance of:literary work> I may be imagining things. Maybe the consensus doesn't exist. If you want to start a discussion at [Wikidata talk:WikiProject Books]] I would support a more specific guideline. Joe Filceolaire (talk) 11:45, 9 November 2015 (UTC)
@Filceolaire: Well, this was defined but Markus :) "a manifestation of the chapter is a part of a manifestation of the script", manifestation is
Manifestation is "the physical embodiment of an expression of a work. As an entity, manifestation represents all the physical objects that bear the same characteristics, in respect to both intellectual content and physical form."[1] The performance the London Philharmonic made of the Ninth in 1996 is a manifestation.
(en:FRBR). This seems pretty precise an usable to me. Also please answer about my questions on your propositions, this is clear to you, not to me what your proposed property means (sorry, I did try to ping but made a typo on your username). author  talk page 12:30, 9 November 2015 (UTC)
TomT0m Well, for me, the distinction between Expression and Manifestation seems a bit arbitrary and I favour combining these as edition (Q3331189), using wikidata properties 'edition (P747)' and 'edition or translation of (P629)' to link these to the item for the literary Work but otherwise I think we should follow en:FRBR. See above for my answers to your other questions. Joe Filceolaire (talk) 16:29, 9 November 2015 (UTC)
@Filceolaire: In what is this arbitrary ? There clearly can be several manifestations of the same expression. For example, "The performance the London Philharmonic made of the Ninth in 1996", to take the article's example, and "The performance the London Philharmonic made of the Ninth in 1997" if they reiterate the same performance, are several manifestations but the same expression if I understand well, as if we recorded the result, it's supposed to be almost the same. author  talk page 16:45, 9 November 2015 (UTC)
Making the ebook, paperback, hardback, audiobook different manifestations of the same expression, rather than treating each as an edition, seems a bit unneccessary for wikidata but I agree with you that a performance is something else and probably does need a different item. Discussing this here is all a bit pointless however as such a detailed discussion belongs in Wikidata:WikiProject Books. Joe Filceolaire (talk) 05:15, 10 November 2015 (UTC)
@Filceolaire: Unfortunately WikiProject Books is not totally the only place to discuss this as FRBR is concerned by artworks in general. :) Maybe we should start a "WikiProject Artworks" but what's become difficult as usual in Wikidata is how to make more generic project talk and make agreement with more specific ones. This also seems a bit pointless to make such project if we have no volunteer to make them live. author  talk page 20:45, 11 November 2015 (UTC)
TomT0m we have Wikidata:WikiProject sum of all paintings which is dealing with original paintings so it's not much bothered with editions of the paintings. Wikidata:WikiProject Music and Wikidata:WikiProject Movies deal with mass-produced artworks. These need separate editions for movies dubbed into other languages (with different voice actors) but as far as I can see there doesn't seem to be much need for separate items for the various manifestations (LP, CD, DVD, BluRay etc.) of these.
For performance art (plays and symphonies) however there might one day be a need for 'Performance' items to go with the item for the 'work' and items for the various language 'editions' though, in practice, very few performances will be notable enough to get their own item. This is a pretty specialised niche compared to the much larger quantities of original artworks, books, movies, songs and music albums on wikidata which seem to be OK without separating 'Expression' and 'Manifestation'. At least that is what it looks like to me. Joe Filceolaire (talk) 14:26, 12 November 2015 (UTC)

I made some related comments on Property_talk:P361 about the implied quantification inherent in using individual-level properties between classes. My suggestion was that we use qualifiers to specify the quantification (cf. CycL). Bovlb (talk) 18:58, 29 December 2015 (UTC)

## New properties to keep "instance of" more specific

Again a question about the use of instance of (P31). In some case we have several instance of (P31) in some items especially when we use parallel classifications on a item. The classical example is the structure classification and the application classification like for DDT (Q163648) which is

In that case the third relation can be replaced using a special property or a general relation "has_role" and the the fourth using "has_application" relation. New properties to create more specific relations can be a solution. Do you have other examples ? Snipre (talk) 16:02, 5 November 2015 (UTC)

We can replace the second statement with specific statements about safety measures and melting point and just leave the first statement which is about what it is or, in wikipedia terms, what infobox should we use?. Joe Filceolaire (talk) 05:05, 6 November 2015 (UTC)
This cannot be a universal solution since every WP has its own set of infoboxes. author  talk page 18:21, 6 November 2015 (UTC)
Plus : It's not a problem to query with subclasses, so there should be no harm in creating a subclass. See for example `{{WDQ/instances}}` and `{{Query instances}}` for WDQ and SPARQL respectfully.
That depends if we're talking of :
the substance
a subclass of "chemical subtance" or "pure substance",
the molecule
then this is both a "type of molecule" and a subclass of "molecule", or a subclass of "organical molecule".

But this is not a instance of (P31) pure substance, as any substance of this kind is a pure substance, so this is covered by transitivity of "subclass of". author  talk page 17:18, 6 November 2015 (UTC)

Suppose we have two items in wikidata, one representing a molecule and the other representing the substance, how would we express the relationship between the two? A slightly related example perhaps is between different states of the substance - ice (Q23392) and water vapor (Q190120) are considered subclasses of water (Q283) - but does water (Q283) represent the liquid state, or the molecule? Is this the right hierarchy relationship? Should there perhaps be a separate "liquid water" item and all be considered subclasses of "H2O" as a molecule? ArthurPSmith (talk) 20:23, 6 November 2015 (UTC)
@ArthurPSmith: Chemistry has a concept for "substance where all the molecule are in some state" : phase (Q104837), or several, such as "liquid water", which would be a subclass of "water substance", I guess. Water liquid water could be an instance of "phase of water", for example. It has also a concept for some possibilities for "simple corps" to exists : allotropy (Q81915)  . For example, we could have an article about "Liquid dioxygen", which could be a combination of the two.
But your question is interesting. In current Wikidata state, I'd say that "Dioxygen" as a molecule has part : Oxygen atom. Dioxygen substance would be a subclass of oxygen simple corp (or oxygen taking enwiki definition of element) and would be <has part : Dioxygen molecule>. I think "Oxygen simple corps" would be the class of all substances composed only of atom of oxygens, and would be subclassed by classes for the different phases and allothropes of dioxygen. We would also need properties for the phase and the cristallographic structure to me more specific about their properties. author  talk page 09:46, 7 November 2015 (UTC)

Comment 2, on the principle : I personaly don't like "opposing" specific properties with instance of (P31) and subclass of (P279). We can and should define classes wrt. and not in opposition to properties about specific aspect of the instances of the classes. If an occupational carcinogen (Q21074597) is by definition a substance that can cause cancer by contact usually in the workplace, then we must be able to express with definition, or that a pesticide is a substance that can be used to kill animals or plants. For this we should have specific properties. But in turn we should know that any instance meets the definition (for which we need inferences however). But if we know that any subclass of pesticide is also a chemical substance that can be used to kill stuffs, then we don't have to repeat this in all the subclasses It's then useless to state the chemical compound (Q11173)   statement about all of them. If for some reason pesticide and occupational carcinogen occurs often together, then it's relevant to create a class carcinogen pesticide with

< carcinogen pesticide > subclass of (P279) < occupational carcinogen >

and

< carcinogen pesticide > subclass of (P279) < pesticide >

, you can further compress the statement with Your substance type subclass of carcinogen pesticide. And you'll know any substance of this kind meets all the definitions of both. author  talk page 12:12, 7 November 2015 (UTC)

TomT0m can I suggest that the carcinogenic and anti-pest properties of a chemical compound are exactly the kind of info that should go in specialist statements (separate specialist statements) and not in 'instance of' or 'subclass of' statements. This means we have something like:
• <instance of:chemical compound>
• <OSHA warning:carcinogen>
• <OSHA long term exposure dose limit:3mg/day>
• <usage:pesticide> (applies to:boll weevil)(applies to:corn weevil)
I can think of no good reason why we would want to compress or combine these and doing so would seem, to me, to make the ontology more confusing and more difficult to update and navigate.
'instance of' should, in my opinion, be about what it is with specialist properties used to expand this. My comment that this translates to what infobox should we use was a bit facetious. As you said the wikipedias use different infoboxes, at the moment, but I, personally, do think that reuse of infoboxes across wikipedias is a thing that will happen more as info from wikidata is used more. Joe Filceolaire (talk) 05:41, 10 November 2015 (UTC)
I totally don't get the what it is point as this seems highly not engouh. Typically to express what it is I would see the encoding of the whole first sentence of the Wikipedia article, if not the whole introduction, at least. So what it is to me refers more to ... the whole infobox. This also assumes that the set of infoboxes is complete wrt. the kind of items we have in Wikidata, which is imho not really true. But you're right my point is valid for an infobox if there is a strong inference engine in the core of the system, which is not true at least at this point. Nonetheless a typical classification ontology as far more precise classes than Wikidatas one, and more specific classes than the "type of infobox" kind are imho useful and allows to be efficient in the process of defining an item in only a few clicks. And there is still A LOT of work needed at this point to roughly at least class items. author  talk page 09:41, 10 November 2015 (UTC)

## class / wikidata class | métaclass / wikidata métaclass | property / wikidata property

(follows up of the "merging metaclass" topic) Hi, I think that using instance of (P31) / subclass of (P279) for wikidatas implementation of concepts needs to be separated of the different items in the ontology about the concepts may lead to inconsistencies, for example by browsing Wikidata taxonomy browser one of the loop in the ontlogy was "metaclass : subclass of -> class : subclass of -> metaclass" which should be solved.

For that purpose I created class or metaclass of Wikidata ontology (Q21522864)   intended to be a root item for class/metaclass hierarchy, and I created also a Wikidata instance class (Q21522908)   item. I think we should have a property to link those items to their outside world counterpart, in the same manner we have "fictional analog of". What do you think ? author  talk page 13:28, 21 November 2015 (UTC) @Markus Krötzsch:

I don't fully understand the proposal yet. I think it is a good idea to have a root element for the ontological class hierarchy, which all ontological classes can be instances of. I am surprised that none of our many "class" items already has this meaning. After some thought, I also tend to agree that this root class is "project-specific" and should have Wikidata in its label. Some questions and remarks:
• It seems clear to me that all classes as we use now would then be instances of class or metaclass of Wikidata ontology (Q21522864) (or one of its subclasses). They would not be subclasses of class or metaclass of Wikidata ontology (Q21522864). Therefore, I don't think I understand what you mean by "root item for class/metaclass hierarchy": the classes would just have a direct instance-of link and there would not be much of a hierarchy on this level. It would mostly be like declaring some items to be (Wikidata) classes.
• I would suggest to avoid the term "metaclass" in the label of Wikidata instance class (Q21522908). A metaclass is just a special (sub)type of class, not a thing "next to" class that deserves being mentioned in the root. Since Wikidata does not have a strict separation of items and classes, the distinction is not as strict as in languages like OWL, which have this separation (there, the term "metaclass" makes sense to refer to a new type of entity that is to "class" like "class" is to "instance"). In Wikidata "instance", "class", "metaclass", "metametaclass", ... are all on the same level, so if one wanted to include this into the label of the root, one would have to list all of those "metametameta...classes" there as well. I think we will certainly need instance-of chains that are longer than two steps.
• Do you suggest that Wikidata instance class (Q21522908) subclass of class or metaclass of Wikidata ontology (Q21522864) and that some classes are specifically marked as Wikidata instance class (Q21522908)? This could be useful for quality control, yet we will always have many classes that are not of type Wikidata instance class (Q21522908) and still correctly modelled. The problem with this modelling is that Wikidata instance class (Q21522908) refers to the absence of certain statements ("instances that are not classes"), which might be tricky in an inherently incomplete system like Wikidata -- anything could become a class in the future if instances are entered into the system.
• There are further "types of ontological classes" one could define, e.g., classes that are not expected to have any instances. Another approach is the OntoClean framework proposed many years ago, which groups classes into four types as a way of supporting quality control. I would suggest to discuss such things independently from the general decision of having a main class that all our "Wikidata classes" should be instance of.
• What is the "outside world counterpart" of a class? Can you give an example?
--Markus Krötzsch (talk) 17:33, 24 November 2015 (UTC)
• What is the "outside world counterpart" of a class? Can you give an example?
@Markus Krötzsch: you are right, I need to clarify my thoughts and be more precise
We don't agree about the definition of the "class" concept - mine is non orthodox :/ and that explains a lot of the misanderstanding here : I took the en:Type–token distinction as a basis for the definition of class as a synonym of "type" the "token/type" distinction principle: a class is a collection of token. Then metaclasses are not classes and so on. This is (should be) reflected in the labels and descriptions
About the global scheme, I'll continue with my definitions if you don't mind[1] : let one of the item in our database be labelled "token". This item `Wikidata Token` would take the role of the "token" term in the "token/type relationship" expression. `Wikidata Token` is as a subclass of `Entity`. Also let another item be labelled `Wikidata Class`. That one would take the role of the "type" term in the "token/type relationship" expression. Instances of the class `Token` would be subclasses of no item, and required to have a corresponding real world event or object counterpart - except for fictional, mathematical or similar one. Any `Wikidata Class` instance would be a `Wikidata Token` subclass. `Wikidata Class` would be the root of `Wikidata Metaclass`, and similarly any `Wikidata Class`-subclass would be a `Wikidata Metaclass` instance - and add more meta-levels if needed. Then we can define the "Non-Token" class as the difference beetween `Entity` and `Token`. author  talk page 19:12, 24 November 2015 (UTC)
For the question about the property to link "WIkidata Class" to "class" : I consider Wikidata classes to be an implementation of the concept of class. On the other hand there is Wikipedia articles and therefore items about the general concept of class in knowledge representation. The idea was to link the hypothetical item about wikidata classes to that item.
I think we will certainly need instance-of chains that are longer than two steps.
=> That needs careful examination :) First it's not easy to find example that exceeds that depth, and secondly I'm aware of some theorems about metamodelling that probably do not apply in our case but that exists in other fields like UML related modelling. If we take a (field) model like an UML model or a programming language type structure - or a wikibase instance - then the metamodel of that models can be created (that would be Wikibase data model) you can also build a meta-metamodel, and finally a whole model hierarchy. They are called M0/M1/M2/M3 in this article : en:Meta-Object Facility. There is no level M4, and there is a mathematical theorem for this : the M3 model is a model of itself, in short (I always fail to find the reference for this, it's annoying :). author  talk page 19:12, 24 November 2015 (UTC)
1. and if you do mind that's the same as you can't answer
• I strongly disagree with referencing Wikidata itself in the descriptions or label. These are not Wikidata classes of Wikidata entities, they are real-life classes of real-life things. --Yair rand (talk) 21:30, 24 November 2015 (UTC)
Real life classes ? What does that mean ? author  talk page 21:35, 24 November 2015 (UTC)
When we look for a source for the statement that X is an instance of Y, we don't search for someone saying "X is a Y, the Wikidata class". Real numbers, dogs, and US states are classes that exist completely independent of Wikidata. --Yair rand (talk) 00:23, 26 November 2015 (UTC)

## Subpage on class modelling

In October I asked a bunch of questions about classes in Wikidata, and volunteered to do some work in the area. This project was suggested as the place to do some of the work.

I put together a page on modelling with classes as a subpage of this page, at Wikidata:WikiProject_Ontology/Modelling. I started this page a while ago, stopped further work on it because I didn't like how it was going, and then was caught up in other work. On the weekend I put some effort into cleaning up the page.

The first part of the page is a description of how I think classes work (or should work) in Wikidata. The second part is some background to support this view, and some issues with classes in Wikidata. Comments are welcome on whether this is a reasonable way to go.

user:Peter F. Patel-Schneider Just a comment: don't start with "...is a description of how classes work (or should work) in Wikidata". This will just be your view of the question among a lot of other ones. And at the end people will oppose their views, bunch of discussions on small details and at the end nothing more. We already had some previous experiences. I can't say how you can do it but be aware of that before starting to spend long hours of discussions for perhaps nothing.
PS: be careful when using specific terms because people in WP and WD have a large background and what is clear for you can be different for others. Snipre (talk) 13:03, 24 November 2015 (UTC)
@Snipre: I do understand your concerns about the whether this is my view or not. However, this is hopefully eventually going to be a recommendation on how classes work in Wikidata, isn't it? I suppose that for now I could add "proposed" to the sentence, but that is already covered in the section title. Peter F. Patel-Schneider (talk) 16:49, 24 November 2015 (UTC)
It can become a recommendation but just keep in mind that you will put your ideas in a community page and then modifications will occur. If you really want to share your opinion and to keep it without deep modifications better use a subpage like User:Peter F. Patel-Schneider/Class. Snipre (talk) 20:15, 24 November 2015 (UTC)
@Peter F. Patel-Schneider: I'd be interested in what you can say on Help:Classification. author  talk page 13:09, 24 November 2015 (UTC)
@TomT0m: I will leave some comments on the discussion page for that. Peter F. Patel-Schneider (talk) 17:10, 24 November 2015 (UTC)

@Peter F. Patel-Schneider: Pardon me, I don't follow you. You claim to have worked on something and to have put it in a subpage, I thought you were announcing this, but in fact it seems there is nothing yet ??? There is nothing to comment ! author  talk page 20:38, 24 November 2015 (UTC)

@TomT0m: I forgot to put an explicit link to the subpage I created. My apologies, it is Wikidata:WikiProject_Ontology/Modelling. Peter F. Patel-Schneider (talk) 20:51, 24 November 2015 (UTC)
@Peter F. Patel-Schneider: Weird, I checked with the special prefix index page, and all I could find was /Class that Snipre created. (oh, there is a two hour gap beetween you post here and the creation of the page) author  talk page 21:15, 24 November 2015 (UTC)
Actually you created the page after 20:38 when I asked for a link, at 20:46 looking at the history. This is odd. author  talk page 21:19, 24 November 2015 (UTC)

## Wikimania 2016

Only this week left for comments: Wikidata:Wikimania 2016 (Thank you for translating this message). --Tobias1984 (talk) 12:07, 25 November 2015 (UTC)

## Some principles we can adopt

I was reading some literature about ontologies and I found in one book (Building ontologies with basic formal ontology) some principles to use when defining an ontology. I propose to discuss some of them and if there is an agreement to propose them in a RfC for adoption by the community:

1. Use singular nouns
2. Use lower format for common nouns
3. Avoid acronyms
4. Use essential features in defining terms
5. Avoid circularity in defining terms (example: hydrogen = anything having the same atomic composition as hydrogen)
6. Structure every ontology around a backbone instance of/subclass of hierarchy
7. Ensure instance of/subclass of completeness (every term is linked to the root entity)
8. Ensure asserted single inheritance (every term has one and only one instance of or subclass of statment)

Any comment ? Snipre (talk) 21:48, 19 June 2016 (UTC)

2 is already mandated by WD:L, 5 sounds obvious enough that I don't think it needs to be stated explicitly, 7 is enforced by property constraints. I strongly disagree with 8, classes can't always work like that. 1 sounds reasonable. Re 3, acronyms should probably be aliases. I don't understand what 6 means. --Yair rand (talk) 22:03, 19 June 2016 (UTC)
1. I think that is already common practice.
2. Also already common practice, but does not apply for, e.g. German, where nouns are capitalized.
3. I agree, except for cases where the acronym is much better known than the full name. For example, in Germany everyone knows the ADAC (motorist club), but I had to look up that that stands for "Allgemeiner Deutscher Automobil-Club".
4. I am not really sure what that means. It sounds reasonable, though...
5. Obvious.
6. I think we are kind of doing that in a way. But this needs a broader discussion I think, especially since this needs to be discussed in detail for every subject area. Which is why I would like to use this chance to plug my idea for Wikidata:Best practices again.
7. We don't have a root object at this point, although we could consider creating one. This would force us to think about the instance of/subclass of each object and not having one would be a constraint violation. I think this would be good to ensure at least basic categorization of items.
8. This will not and can not work on Wikidata (or the real world for that matter).
--Srittau (talk) 06:37, 20 June 2016 (UTC)
• Just one comment about the "Best Practices" page : it seems to be once more a page which try to reinvent the wheel in term of topic classification. It is bound to be the same mess of WD:PP, or Help:Modelling. We should find a way to use wikidata class tree in such a task and keep connected to it. author  talk page 06:44, 20 June 2016 (UTC)
• The root class is entity (Q35120). --Yair rand (talk) 06:46, 20 June 2016 (UTC)
• @Srittau: I agree with TomT0m that the creation of "Best Practices" page is untimely: your proposition is based on cases studies and lacks some principles we should first defined. To have a coherent structure we should have a top-down approach which will limits later the possibilities but will ensure similar models in all fields. Snipre (talk) 09:18, 20 June 2016 (UTC)
• How is it untimely? Wikidata is several years old, and we should have had a collection of those guidelines long ago. Your proposal for general guidelines comes only after my proposal. To be hones, we should probably head over to Help:Modelling and have an in-detail discussion there as that page is a very good starting point, which already formulates many principles. --Srittau (talk) 11:50, 20 June 2016 (UTC)
@Srittau, Yair rand, TomT0m: I agree there are some obvious principles but I think we miss a simple list with all principles located on the same page. There are a lot of information but this information is distributed in several pages and contributors don't spend hours to read and understand everything. Wikidata is a wiki and people have the habit to work by trials and errors instead of spending some time by looking at the documentation which is not always up-to-date.
Again I just want to point that these principles are used by quite a bunch of ontologies (see a list there). So just saying "classes can't always work like that" or "This will not and can not work on Wikidata" is quite light as argument. I agree that BFO is not the bible and we can define other rules but in all cases we will have to define strong rules with a lot of constraints in order to be able to have a coherent structure for all fields. Snipre (talk) 09:10, 20 June 2016 (UTC)
Yes, we should have a list of those guidelines. Let's work on Help:Modelling. If you doubt my claim that this will and can not work on Wikidata, you should probably have a look around some of the items that are already on Wikidata and familiarize yourself with a few of the key principles of Wikidata, one of which is that there can be multiple conflicting statements. Keep in mind that the vast majority of existing ontologies are limited to a fairly specific domain. Building a general ontology of the scope of Wikidata, edited by many people, using many different sources has never been (successfully) tried before. You yourself admit that "BFO is not the bible", so try not to dismiss criticism of the points you brought up by pointing to it. --Srittau (talk) 11:57, 20 June 2016 (UTC)

## What is nothing?

Q154242 (nothing) is defined as the pronoun, but it's relationship to Q2165236 (everything) and to Q15893266 (former entity) suggests that it's being used as the concept of not-anything. I suppose the confusion stems from the fact that en:Nothing says "Nothing is a pronoun denoting the absence of anything", and then goes on to discuss the concept, not the word. Maybe that should be changed to "Nothing is the concept of the absence of anything". -- Duesentrieb (talk) 22:41, 9 September 2016 (UTC)

Agree. I've change to concept (Q151885), though I don't feel the difference from notion (Q595523). --Infovarius (talk) 18:04, 11 September 2016 (UTC)
@Infovarius: I feel unconfortable to link to "concept" or "notion" as it seems that basically anything absstract is a concept. Or "Abstract idea" or whatever. This is very little informative.
Also opposite of (P461) is kind of undefined here. I'd very well see some kind of complement (Q242767)   notion as a kind of opposite however. This is doable using disjoint union of (P2738) for classes and the empty class. For example we could define "nothing" as something that has no instance at all. If anything is an instance of https://www.wikidata.org/wiki/Q35120 search then maybe we could say that "nothing" is disjoint with "entity". Something like
disjoint union of (P2738) object or value >
of (P642)
of (P642) < nothing >
. But if we refer to vacuum (Q11475)   this might be different ;) author  talk page 18:27, 11 September 2016 (UTC)

## located in

Do we have a generic property for located in? I want to express

located in search

. --Succu (talk) 11:00, 23 October 2016 (UTC)

It's not a whole-part-relationship. The objects are physically independent and share no boundary like eggs in a bag. --Succu (talk) 20:59, 24 October 2016 (UTC)
Yeah, I see that. So maybe we should have a new proposal for "inside of" or something like that as a property? All the properties that use the word "location" are expecting a concrete geographic location as the target item which isn't what you want here either. ArthurPSmith (talk) 15:12, 25 October 2016 (UTC)
Or "container" which could be very useful for transport as well ... I'd support this. I found (searching rapidly) a 2013 paper that seems to be aware of the distinction and analyses meronomy (Q2465551)   - the search is not so easy as the word "container" is used for data structures like array a lot in literature of computer modelling - author  talk page 11:46, 26 October 2016 (UTC)

## Patterns

I've been researching ways to sort pattern (Q2083958) and its subclasses. AAT has two separate concepts "pattern" (design element) and "pattern" (guide). I think I could sort these items using the AAT hierarchy as a reference.

Today I found this newly created superclass no label (Q30249464), which makes no sense to me and seems very un-Wikidata-like. Does this make sense to anyone else? - PKM (talk) 00:33, 17 June 2017 (UTC)

Sorry, maybe I did not understand the concept of patterns and its name in English. I want to clarify: in the sequence "1-23-1-52-1-62-1-27-1", "1" is a pattern or the sequence "1-23-1-52-1-62-1-27-1" is a pattern? Or the structure "1-...-1-...-1-...-1-...-1" is a pattern? What is "1"? --Fractaler (talk) 14:20, 17 June 2017 (UTC)
You're right, "pattern" means several vaguely related things in Engliah. I think your examples should be sequence (Q133250) in English - see en:Sequence. I think of mathematical sequences as a <subclass> of pattern (Q2083958).
My concern is that we have agreed (in general) that Wikidata items should be singular. So pattern (Q2083958) is itself the class that contains all patterns, and making a class above it called no label (Q30249464): no description is unnecessary and confusing. If you want to separate "patterns in nature" from "human-made patterns", then I think those should be subclasses of pattern (Q2083958). - PKM (talk) 19:14, 17 June 2017 (UTC)
Wikidata items should be singular? I do not understand. Should not be items "All Q5", "All components of clothing", "All days of the month", "All atoms of a molecule", etc? --Fractaler (talk) 08:27, 19 June 2017 (UTC)
the *labels* should be singular, not plural (in general), for a wikidata item representing a class of things (even though there are many things in the class thought of as a set, each thing is singular and an instance of the (singular) class). ArthurPSmith (talk) 19:43, 19 June 2017 (UTC)
I mean: molecule (Q11369) = "all atoms of a molecule", month (Q5151) = "all days of the month", etc. --Fractaler (talk) 10:01, 20 June 2017 (UTC)

## Scholia profile for ontology (Q324254)

This is starting to become useful but needs more work: https://tools.wmflabs.org/scholia/topic/Q324254 . --Daniel Mietchen (talk) 23:31, 26 July 2017 (UTC)

I was happily linking WordNet (Q533822) synset identifiers up to Wikidata items via exact match (P2888) when I ran into mosque (Q32815). I have linked over 100 Wordnet synsets (most associated with ImageNet (Q24901201)). I have done that with exact match (P2888) as I believe the URIs in Wordnet are to be regarded as instances/individuals, e.g., here it is regarded as an instance of skos:Concept. Now with mosque (Q32815) the Wikidata item is linked to DBpedia via equivalent class (P1709), i.e., as a class. From https://www.wikidata.org/wiki/Q32815#P1709 we now have both equivalent class (P1709) and exact match (P2888) links. Does that make sense? I suppose that Wikidata is class-instance-agnostic and it is not a problem to represent it in Wikidata, but are we making havoque in the linked open data (Q18692990) cloud? — Finn Årup Nielsen (fnielsen) (talk) 13:37, 24 August 2017 (UTC)

a concept can be a class - most concepts are abstract things that would naturally represent a set of physical instances (mosque certainly fits in that category). But equivalent class (P1709) doesn't make sense for everything so I'd stick with exact match (P2888) for these relationships, I think it's best to be consistent in these mappings if we can be (for mosque (Q32815) I'd suggest to just keep both). ArthurPSmith (talk) 14:29, 24 August 2017 (UTC)

## object of group (Q36809769)/group (Q16887380) and element (Q379825)/set (Q36161)

In mathematics we have element (Q379825)/set (Q36161). Outside of mathematics we have object of group (Q36809769)/group (Q16887380)? --Fractaler (talk) 08:29, 15 September 2017 (UTC)

## Use of qualifiers in subclass relation

I have not found any discussion of using qualifiers together with subclass of (P279). A SPARQL query shows that properties are use to state details about subclass statements and even allow problematic relations. -- JakobVoss (talk) 06:19, 28 October 2017 (UTC)

I think it can make sense in some cases, but I am not sure it's uniformly helpful. The qualifier could for example better define how this class is a subset (via its instances) of the superclass. I'm even more unsure of the usefulness of qualifiers on instance of (P31) relations, of which there are also quite a few. I think most of these have been added by just one or two people so maybe we should engage them in discussion on the purpose of this? ArthurPSmith (talk) 08:26, 28 October 2017 (UTC)
Right, the use of qualifiers should not be encouraged in general, so the exceptions where it makes sense should be documented. -- JakobVoss (talk) 19:50, 28 October 2017 (UTC)

## Merged help pages about properties

Notified participants of WikiProject Ontology I merged Help:Properties and Wikidata:Properties to avoid project page inflation. We still have too many project pages about similar topics so it is hard to find the current state of development and the pages get outdated quickly. I also try to redirect all related talk pages to Help talk:Properties to avoid fragmentation of discussion about properties. The current help page still needs cleanup and rewording - working in this is sure more useful to WikiProject ontology than philosophical discussion. Once we have good documentation of the current state, it will be easier to discuss ontological changes and implications. -- JakobVoss (talk) 07:54, 31 October 2017 (UTC)

+1. Thanks Snipre (talk) 09:18, 31 October 2017 (UTC)

## Slight page re-org

Hi all

Notified participants of WikiProject Ontology,

Would anybody object if I placed all the problems into there own space/place here on the page so that the WikiProject Ontology/Property wouldn't be miles away from the other sub-pages? I'm new here and I can tell you it confused me so I'm sure it must confuse other n00bs. --Golden herring (talk) 21:26, 26 November 2017 (UTC)

@Golden herring: That's generated from a template that just searches for all subpages matching the prefix, but there's probably a better template to use so it only catches the top level. Go ahead and explore that. I don't think it would be helpful to move subpages to another prefix, where would you put them? Other wikiprojects here have quite similar subpages. ArthurPSmith (talk) 16:15, 27 November 2017 (UTC)

## X is instance of "type of X"...!?

How can an object X be an instance of "type of X"? Please answer also here. --Horcrux92 (talk) 10:08, 23 December 2017 (UTC)

We have object of group (Q36809769) and group (Q16887380) (consist of object of group (Q36809769) and only). Type = group (Q16887380) (synonym (Q42106)) --Fractaler (talk) 17:27, 23 December 2017 (UTC)
1. I don't think this is right. A "type" is an abstract concept, therefore is more like a class than a group:
2. Anyway, this doesn't answer the question.
--Horcrux92 (talk) 18:17, 23 December 2017 (UTC)
herd (Q209542) is a group of animals. group of humans (Q16334295) also is a group of animals. Merge?
What is the name of the set/group of all chordophones? What is the name of the element/object of this set/group?
What is class? --Fractaler (talk) 10:35, 26 December 2017 (UTC)
A group is like a band, a set of individuals linked something, friendship for example. If you lose a friend, the group of your friends changes. Group membership is a whole part relationship. A class is an abstract description that may have instances. « male american in their eighties » in a class. If an american man turns eighty, the class (the abstract description) does not change. The class has just gained an instance. The set of all instances of a class is said to be its « extension ».
It’s possible to go from a group, say the music band « pink floyd » who has human members, to a class, by giving an abstract description. Let’s call « Pink Floyd Member » the class of people who were Pink Floyd members at some point ». But « pink floyd » and « Pink Floyd Member » are very different stuffs.

Maybe an interesting concept to raise is the one of « occurrence », that accounts for the multiple times a type can appear in the description of another type. The abstract type of « car » can be described as « vehicle with one rear left wheel, one front left wheel, one rear right wheel, one front right wheel ». Pierce accounts for that in it’s type token distinction (see the enwiki article) as « rear left wheel » being an occurrence of the « wheel » type in the description of the « car » type (as opposed to « my car’s rear left wheel » who is a token of wheel into a token of car). I don’t think we have such a problem in Wikidata as we have restriction to use « part of » only from type to type or from token to token, but never from token to type (this could be generalized to « link metaclasses only to metaclasses of the same order with part of »). « has part of type » is to be used to link token level to class level.

On « X is instance of "type of X" » please see articles like Metaclass (Semantic Web) — Wikipedia. author  talk page 11:53, 16 January 2018 (UTC)

class is an abstract description: why "abstract"?
I’m beginning to feel tired. Please read stuffs about the type/token distinctions. Types are abstract descriptions of their concrete tokens. « Pink Floyd » is a concrete organization, hence it’s not a type. It has no tokens. An item represents either abstract or concrete stuffs, never both at the same time.
Pink Floyd (Q2306) is rock band (Q5741069). rock band (Q5741069) is vocal-musical ensemble (Q29881657). vocal-musical ensemble (Q29881657) has part (P527) two parts: 1) vocal part, 2) instrumental part. Vocal part has part (P527) singer (Q177220), instrumental part has part (P527) instrumentalist (Q1278335). instrumentalist (Q1278335) has part (P527) human (Q5) and musical instrument (Q34379) (etc). musical instrument (Q34379) is "Pink Floyd member"? --Fractaler (talk) 14:11, 16 January 2018 (UTC)
«  has part (P527) human (Q5) » Definitely wrong. Any instrumentalist (Q1278335) is also a human. This matches the « subclass of » definition, definitely not the « part of » one. An instrumentalist is a human who plays an instrument. This is an abstract description, and given any concrete object you can either answer « this object fits with this description », and its a token of the type, or « it does not fit ». It’s definitely not a matter of knowing if the musician in yourself is a different person and is informally « a part of you », it’s about knowing if you as a concrete individual fits the definition (the abstract description describes yourself or not) of the type. Of course you could as well be a fictional puppet some real person is pulling the string, but then you would not meet the definition of a person, hence not of an instrumentalist. author  talk page 14:40, 16 January 2018 (UTC)
instrumentalist (Q1278335)-musical instrument (Q34379)=human (Q5) (without instrument) --Fractaler (talk) 19:58, 16 January 2018 (UTC)
No. Looks like you do not understand the properties has part (P527) / part of (P361). --Succu (talk) 20:13, 16 January 2018 (UTC)
Do you mean that you understand better and can prove it? Then just prove it with the example of instrumentalist (Q1278335). --Fractaler (talk) 08:59, 17 January 2018 (UTC)
TomT0m explained it allready to you. The instrument someone plays is not part of the player. --Succu (talk) 22:40, 17 January 2018 (UTC)
A group is like a band...The set of all instances of a class is said to be its « extension » - this is an erroneous attempt to statically (using the result) describe what is happening in the dynamics (simulate the process). Ie, write (object-oriented programming (Q79872)) object-oriented source code (create object-oriented programming (Q79872)-classes), compile it and "look", as an instance of the class (an (virtual) object in the computer's memory, object (Q216601)) appears. This is an erroneous approach at this stage. While the Wikidata do not allow modeling the process, only describe it (ie, create a result, a static picture, a taxonomy, a classification). You can describe the process in steps: an empty set of 79-year-olds, an empty set of 80-year-olds, etc .; the appearance of the first 79-year-old (the first element of the set of 79-year-olds), the appearance of the second 79-year-old (the second element of the set of 79-year-olds), etc .; the life of the first 79-year-old, the life of the second 79-year-old, etc .; the disappearance of the first 79-year-old, the disappearance of the second 79-year-old); the appearance of the first 80-year-old, etc. --Fractaler (talk) 09:19, 17 January 2018 (UTC)
The object-oriented programming paradigm (=object-oriented programming (Q79872)) has ony a small overlap - e.g. both are using the terms class and instance - with ontology modeling "languages" (e.g. OWL). This maybe confusing. Your example describes an aging human (a process). If he can play an instrument (an ability) this is still not part of him/her. In OOP you need to have a age property to simulate aging. Does this property describes the proces aging of a human? --Succu (talk) 23:23, 17 January 2018 (UTC)
If someone claims: "I'm instrumentalist (Q1278335)", how will you check it? By definition instrumentalist (Q1278335): "person who plays a musical instrument". Ie, if a person does not play the instrument, this person is not instrumentalist (Q1278335), but a person who can play (has this ability) a musical instrument. instrumentalist (Q1278335) = "part 1" (human (specialist) part/component) + "part2" (tool (musical) part/component). If someone claims: "I'm instrumentalist (Q1278335)", just ask him: "When? Who are you without musical instrument (Q34379)? So, no musical instrument (Q34379), no instrumentalist (Q1278335).
Now here we can not simulate the process, and just describe it with static points (items) in time (take "snapshots of the process"). And only when the soft-tool (that can show the dynamics, "movie", the transition from the first point to the second, then to the third, etc) appears, then we can only say that we simulate the process. Now we have here only "snapshots". OOP, unlike WD, has a tool that can not only create the same points (snapshots: age 1, age 2,... etc.) as in WD (source code (Q128751)), but also carry out the transition from 1 point to another (process, execution (Q1077724) with run time (program lifecycle phase) (Q288510): demonstrate the very process of aging, etc.).
Everything happens as predicted by the known law (no label (Q4184872)) of the transition of quantity (Q309314) (group of snapshots/items) into quality (Q1207505) (process/dynamics). Ie, while we are at the stage of quantity creation (the first point of the process, the second process, ... the last point of the process). We are waiting for the appearance of the modeling (OOP-?)tool for "show": "the first point/item of the process-> the second point/item of the process, -> ... the last point/item of the process" and the subsequent results of using such a soft-tool.
Types are abstract descriptions: is there any other descriptions? Non-abstract descriptions, for example?
Pink Floyd » is a concrete organization When? At what point in time: 1) at the time of creation, 2) at the time of its first performance, 3) at the time of the change of one of the members, 4) at the moment, or 5) in all these moments? So, what is concrete organization? --Fractaler (talk) 07:51, 18 January 2018 (UTC)
Your individual "someone" is simply acting in the role "playing an instrument". No whole/part relationship is involved. --Succu (talk) 22:20, 18 January 2018 (UTC)
For a person who can play a musical instrument, there can be two situations: 1) when he is with a musical instrument, 2) when he is without a musical instrument. Correctly? --Fractaler (talk) 08:07, 19 January 2018 (UTC)
Last night when trying to get some sleep it occurred to me that this could somewhat be correct if you model not a person, but a person life. « My life » can be divided into periods, adulthood, childhood, and so on. There is professional periods of time, parenting one and so on. There is also periods in which someone plays music. Then the periods in which an instrumentist plays music are part of its life. It’s OK to use part of if we model stuffs that way as we divide a lifetime of a person into smaller parts of the same time, following the idea that stuffs of some kind can be composed of stuffs of the same kind., and we could actually use this model for « history of ̃…» articles, for example « history of french first republic » is a part of « history of france ». For a public person with distinct life, say Ronald Reagan, he had politics periods an acting periods in it life. But they are distinct items than the item about the person itself. So this actually seem a practical model for us as it corresponds to an actual distinction used in Wikipedia article to consider that the item about the person is distinct from a description of his life. A person is considered an instrumentalist (by that I mean there is a statement « occupation ː instrumentalist, or equivalently being an instance of instrumentalist) if there is some parts of its life in which he plays a musical instrument. The history of this life may have an item, as well as its instrumentalists periods. ̴̴̴̴
Agree. There is 1) "real/physical person", "source of person's events" and 2) WD/WP-person (representation of events of a real person, historical, eventual aspects of a person). So, when a person who knows how to play an instrument is playing, he generates an event that (if this event is "WD:Notability" for WP/WD), which will say: this person at the moment was instrumentalist (Q1278335) (he was able to play the instrument and play). If nothing is recorded, he is just a person who knows how to play the instrument, he is not instrumentalist (Q1278335). Fractaler (talk) 12:48, 19 January 2018 (UTC)
We have options: 1) a person can play and play constantly (ie, he has a tool), because he earns it by this (it is his job), 2) a person can play and sometimes play (ie, he has a tool), because he is an amateur 3) a person knows how to play and does not play (because he does not have a tool) 4) a person can play, but does not play (so the circumstances turn out), 5) etc. Fractaler (talk) 13:02, 19 January 2018 (UTC)

## Help pages and new stuffs

Notified participants of WikiProject Ontology There may be pages to update, or decide their worth considering advances in classification tools on Wikidata, please see Wikidata talk:Item classification#Metaclassification, union of and so on. author  talk page 11:21, 16 January 2018 (UTC)

Thanks for the notification! However, it would be nice if you could provide a bit more context when pinging a whole wikiproject - if people have no idea what this is about, they are not likely to engage much. If the relevant information is at Wikidata talk:Item classification#Metaclassification, union of and so on, why don't you use `{{Ping project}}` there directly? − Pintoch (talk) 11:31, 16 January 2018 (UTC)
Well, Wikidata tends to be not very well equiped from transversal discussion. And there is not a single page that could be relevant here, as the issue is more global. The starting point of this discussion is a reorg of the help page redirect, so the question is both how we organize our help pages. an illustration is that I think I created this page actually and it was not a redirect, the content was different, it was replaced by a totally different page. So the question « what are we doing here and are we sanely cooperating is at sake. I don’t really care if there at least is consistency in the result and between pages, which is why I post it here. We might have several visions not totally consistent, healthier to talk together about this than to editwar over a redirect. See this as an example of Jakob’s work to make the doc pages better, but with an attempt to reach a consensus and go beyond personal initiative. Subsidiarity principle, I think the issue is worth for the whole project, so I’m also posting here.
Sorry, I don't understand how your reply relates to my comment. − Pintoch (talk) 11:49, 22 January 2018 (UTC)

## A nasty anti-pattern ?

"It's a floor wax and a dessert topping"

Here's a query `tinyurl.com/ydz6f6gd` [better version below] that I think reveals quite a widespread issue - classes that are subclass of (P279) both physical object (Q223557) and abstract object (Q7184903).

(Note that the query is meant to find the highest class that is in both subtrees -- it doesn't necessarily manage this, as the right hand columns show. It does, I think, remove quite a lot of lower classes; but it doesn't seem to get quite to the top of the shared subtree, perhaps because I had to truncate my searches of both to fit within the time).

Modified version of query, that gets further up the tree: `tinyurl.com/yb3jeyy6`. Also now sorted by the abstract class, which is probably where the key issues are. Jheald (talk) 23:59, 17 February 2018 (UTC)
And variant that produces slightly more of the abstract classes: `tinyurl.com/ya4spc62`. Interesting that metaclass (Q19478619) itself is one of the classes that can contribute to the issue. Jheald (talk) 00:10, 18 February 2018 (UTC)

Here's what I think may be happening in at least some of these cases; I think it may be quite a nasty anti-pattern, that may be quite widespread.

We are I think now fairly used to the idea of a metaclass -- so one can have a class of concrete objects, eg consumer goods, that is also an instance of an abstract concept, eg a brand.

What happens now if this class of concrete objects is divided into a set of subclasses? On the concrete side it's no problem, it's subclasses all the way up. But on the abstract side, now there is a problem, because now it looks as if there is also a class of abstract items -- eg once one has Lego Ninjango, Lego Technic, Lego Knights etc, then the original Lego, rather than being an instance of a brand and a class of toys, instead now looks like being something that has multiple instances that are each a brand, making it a subclass of brand.

The above example is all about products and brands, but from the query I suspect this may be an anti-pattern than turns up again and again, in many different domains. And I'm not sure where the fix is. When you think about it, it does seem quite natural to regard Lego as a class of brands, as well as a class of toys. Jheald (talk) 22:49, 17 February 2018 (UTC)

Thanks for digging into this. Your Lego example is similar to Levi's 501 (Q3237209), which is both <instance of> model (Q17444171) and <subclass of> jeans (Q83363). I wonder if we need to distinguish between "Lego Ninjago = brand of toys" (abstract) and "Lego Ninjago-branded toys" (concrete), or "Levi's 501 = model of jeans" (abstract) and "Levi's 501 jeans" (concrete)?
Related to all this are items which are defined as <subclass of> '"object (Q488383): technical term in modern philosophy often used in contrast to the term subject" when what is wanted (I think) is some sort of generic "object" that is parent of both concrete and abstract objects. object (Q17553950), defined as "tangible object", doesn't quite fit the bill.
Also, I note that some of the query results are a failure to distinguish between a concrete object and the process for which it's used or by which it's made. I see these a lot. Example breaking wheel (Q692003) is the device, but the associated methods of execution (Q3966286) should be something like "breaking on the wheel", that is, a process rather than an object, linked by <use>. - PKM (talk) 20:21, 19 February 2018 (UTC)
@PKM: Yes. The latter is a classic, of course, because most of the time, each language's Wikipedia probably only has a single article, covering the entirety of process, practitioner, product, and associated equipment -- all of which get categorised together, and then that single category gets classified under any/all of the above. And one has to add, that there are also a lot of the classes in the 'abstract' column that just look very very odd there (eg 'pendulum' ?), and surely just must be the result of a mistake somewhere,
It's User:TomT0m who is apt to talk the most about the theory of all of this, but certainly there are knowledge systems out there that try to keep a very clear line of different classes between the concrete and the abstract -- I think the CIDOC CRM ontology, used by the British Museum and some other big museums, is one of these, with its clear division between E24 "Physical man-made thing" and E28 "Conceptual Object", with all thesaurus items being classed under E55 "Type", in turn under E28; while all collection items are under E24.
The taxon people here take a similar stance, I think. But they seem to have made quite a separate world for themselves to live in. In most of the project we seem to use instance of (P31) as a breaker cut-out between the physical and the abstract. But in many classes, there's been such a muddle between P31 and P279, it seems rather a weak and uncertain division to rely on. I was having a look at a few classes at the end of last week, and was finding it really hard, without opening up every item (often available only in Russian), to work out which items in "sword" represented a single named sword, and which a class of swords; or which in "computer" represented a one-off unique computer, and which a multiple- or mass-produced computer type. Even after I'd done some clearing up, I still wasn't sure whether anyone would trust the P31s and the P279s enough, by themselves, that they would be confident, faced with a label in Cyrillic, as to what they would get. And now, with the thought of the Lego example above, and the pattern it seems to represent, I wonder whether I should have even less faith in the barrier being secure.
But on the other hand, the present system of allowing the same item to be both a subclass of (P279) and a instance of (P31), with the two being understood to refer to different senses of the item, seems now so embedded, that perhaps that has to be considered what we have chosen to commit to. (?) Jheald (talk) 21:52, 19 February 2018 (UTC)
I guess there are a lot of different issues here, but one of the major problems I see running Jheald's query is there are a lot of classes in the subclass tree of physical object (Q223557) that simply should not be classes of physical objects - all the "works" for example. How is an "anime" a physical object in any sense? Other cases there are issues with different senses of an item as discussed above, but I think there's a lot of straightforward cleanup that would help reduce the problem first... ArthurPSmith (talk) 15:28, 20 February 2018 (UTC)

## Subclass_of statements for terms about migration

Dear Ontology wizards,

I'm looking for some help and second-opinions about the "subclass_of" property used on the 61 terms that are being translated as part of the Wikidata:Europeana migration vocabulary campaign right now. The project is to have all these concepts - from "deportation" and "cultural identity" through to "visa" and "map" - have labels and descriptions in at least all the 24 European Union official languages (and any others that request to join to (see the project page for details).

As an added "stretch" goal, I'm also targeting the labels of all the Superclass items (those which appear in the subclass_of statement for each of these terms). This is not only a question of translating, but also of adding/improving the statement in the first place. Many of the terms had no "subclass_of" statement at all when we began the campaign. At the time of writing this, only 1 is still missing: homeland (Q16513600). This, as you can see, is a quite abstract concept, which is why no one has been able to think of a good item to link it to yet. I would appreciate your advice on improving that specific item, but also to check the consistency and accuracy of the other 60 terms in the list for their subclass_of statement. Here is the table of all targeted terms. Please take a look at the final column and see if there's anything that should be changed/improved. Sincerely, Wittylama (talk) 11:58, 13 March 2018 (UTC)

On that note... I'm having difficulty differentiating the concepts homeland (Q16513600), Heimat (Q629593), fatherland (Q642555), and Motherland (Q1371446). I suspect that homeland (Q16513600) and Motherland (Q1371446) can be merged - because the only language that has a WP article for both is the Dutch article "Thuisland (begrip)", which is about the origin-country of multinational companies, which is something else entirely I think... Wittylama (talk) 16:13, 13 March 2018 (UTC)
as a quick addendum to this, the Table WittyLama posted only shows one subclass_of per vocabulary item, to keep it simple. If there are more subclass_of instances for an item, it will only show one (the query uses the sample() call). Here's a table that shows every subclass_of for every item (rows are duplicated when there is more than one subclass_of). CalvinBall (talk) 16:35, 13 March 2018 (UTC)
The subclass for asylum seeker (Q564811) looks wrong - it should perhaps be an instance of legal status, but subclass of person or human (Q5) - every instance of "asylum seeker" is a person, not a legal construct. ArthurPSmith (talk) 18:01, 13 March 2018 (UTC)
This is exactly the kind of difficulty I've been having in structuring these items... Given that this item does not have an equivalent concept in the UNESCO controlled vocabulary - which is what I've been using to match the various terms against (and also AAT when possible), I think we can adjust it as necessary. Perhaps it should follow the modelling framework set by refugee (Q131572) which is a highly related concept - a Refugee is an Asylum seeker whose claim has been granted. Wittylama (talk) 19:48, 13 March 2018 (UTC)
refugee (Q131572) looks good, yes. ArthurPSmith (talk) 20:26, 13 March 2018 (UTC)
@ArthurPSmith:: I've made that change - and also added a 'different from' tag to each of them referring to the other, to avoid any confusion. Can you suggest any other ontology improvements to the 'subclass of' property on these 61 items? Wittylama (talk)
Well, I'm not too familiar with this area. decolonization (Q230533) subclass of colonialism (Q7167) seems wrong, maybe group action (Q3533467) should be the superclass? ecological crisis (Q2382319) probably should subclass crisis (Q381072). There wasn't anything else that struck me as obviously wrong. ArthurPSmith (talk) 14:56, 14 March 2018 (UTC)
Done for 'crisis'. Not sure about decolonisation though... "group action" implies that it's a consciously coordinated series of events whereas it would be better defined as a process that has, historically, inspired others in similar situations to push for the same process to take place. Perhaps 'facet of' colonialism (if not subclass) - since it's clearly related! Wittylama (talk) 15:12, 14 March 2018 (UTC)