Wikidata:Requests for comment/Are colors instance-of or subclass-of color

An editor has requested the community to provide input on "Are colors instance-of or subclass-of color" via the Requests for comment (RFC) process. This is the discussion page regarding the issue. If you have an opinion regarding this issue, feel free to comment below. Thank you!

The following discussion is closed. Please do not modify it. Subsequent comments should be made in a new section. A summary of the conclusions reached follows.

stale --Pasleim (talk) 08:35, 20 August 2016 (UTC)

---

I have recently tried to fix the constraint violations of color (P462) (Wikidata:Database reports/Constraint violations/P465). I found some colors missing the subclass of (P279) claims, but saw that various editors remove such claims in favour of instance of (P31) = color (Q1075) claims.

I therefore would like to invite comments for the following questions. I also invite edits to my questions if you object with their neutrality.

Comments

• I don't think that one can answer this question without first determining what color is supposed to be. There are quite a few different potential meanings for color, including particular RGB values, particular print colors, particular wavelengths of electromagnetic radiation, and descriptive colors. In all but the last, it appears to me that colors are true individuals and instances of color (Q1075) - each color is a particular value and there are no generalization relationships between them. In the last, however, it appears to me that colors are best described as classes, as there are many generalization relationships between the colors - crimson is a specialization of red, for example. Some colors, perhaps the named color of some trademarked object, in this particular account might somehow be different from other colors in that they have no class-like behavior. Whether this means that colors are instances or subclasses of color (Q1075) then depends on what color (Q1075) is supposed to be. If color (Q1075) is the universal color then colors are probably subclasses of it. If color (Q1075) is the class of human-notable colors then human-notable colors are instances of it. If color (Q1075) is a vague class somehow related to colors then maybe some colors are instances of it, others are subclasses of it, and yet others are both. The description of color (Q1075), "visual perception of light wavelengths", appears to rule out the first sort of interpretation. Peter F. Patel-Schneider (talk) 21:45, 19 October 2015 (UTC)
• Interesting. BTW, the third question goes beyond the title of the RFC. --- Jura 16:40, 21 October 2015 (UTC)

• @Jura1: Good point. I moved the 3rd question into a comment-only section. --Tobias1984 (talk) 17:16, 21 October 2015 (UTC)

• Red is a class of light. When we see light of that king, we say (actually we don't, but that's what happens) "oh, I see red light coming from that object" (we say "that object is red). So red is a class of light :
  ⟨ Red light ⟩ subclass of (P279) ⟨ light ⟩
  . Then colors are specific classes of lights, characterized for example by a range of wavelenth. So we can pretty well say
  ⟨ red ⟩ instance of (P31) ⟨ color ⟩
  (taking "red" as a synonym of "red light") just as we say Porshe 911 is a car model, I saw a Porshe 911 in the street yesterday. Color is a metaclass in the sense of Help:Classification. author  TomT0m / talk page 19:09, 21 October 2015 (UTC)

  Pushing this a little more, we could define the "color" property by "type/class of (visible) light this object reflects". The range of the property would then be "color" as a class of class of light, any of its instances like "red" are then types/classes of light. All of this is consistent. author  TomT0m / talk page 20:07, 21 October 2015 (UTC)

• From a programming standpoint, a specific color is an instance of some Color class, for example the color blue is Color(0, 0, 255). However, when you speak of blue as a group of colors as opposed to a specific color (such as royal blue is still "a shade of blue" versus "royal blue" is darker than "blue"), that takes on a different meaning. I'd probably lean to how Q953243 has it now: royal blue is an instance of color (Color(0, 35, 102)), but a subclass of the group of colors generally referred to as blue.  Hazard SJ  22:28, 23 December 2015 (UTC)

But there is a contradiction in Q953243, Hazard-SJ:

• royal blue (Q953243) instance of (P31) color (Q1075)
• royal blue (Q953243) subclass of (P279) blue (Q1088)
• blue (Q1088) instance of (P31) color (Q1075)

Currently an item that has a sRGB color hex triplet of 002366 is the instance of a color that has the sRGB color hex triplet of 0000FF. To me that doesn't ake sense. There should be two concepts for blue. One for 0000FF and one for the hypernym of 002366 and 0000FF.ChristianKl (talk) 13:23, 26 May 2016 (UTC)

So royal blue is apparently both an instance of and subclass of color. That is like saying "Honda Accord instance of car" and "Honda Accord subclass of car", which is widely recognized as incorrect. Emw (talk) 02:10, 4 January 2016 (UTC)

@Emw: What I should have said was that blue is a subclass of color, and royal blue is an instance of blue (and hence an instance of color). Royal blue really wouldn't be a subclass, unless we have different types of royal blue. Blue itself, though, is an instance of color (considering blue to be Color(0, 0, 255)), but also a subclass if you consider blue as a group of colors (tints and shades included). In a sense, this shifts the above from royal blue to blue, but what I think is that "blue, the color" (Color(0, 0, 255)) is different from "blue, the group of colors" (sky blue, royal blue, navy blue, etc.). Is the sky the color blue, or is the sky a tint/shade of blue? Our entity for blue seems to merge the two concepts in one, so strictly speaking, our blue entity is both an instance and a subclass (strange, but that's at least what I'm thinking at the moment).  Hazard SJ  04:48, 4 January 2016 (UTC)

Question: p31, p279, both

Should colors be classified treating them as instances (instance of (P31)), as subclasses (subclass of (P279)), or are there reasons to use both properties? As an example I would like to give heptane (Q150668) which, like many other chemicals, uses both properties. A chemical can be seen as an instance of a certain class of chemicals, but also as a subgroup of other groupings.

p31

1. . Colours which are uniquely defined by sRGB color hex triplet (P465) or similar property are instance of (P31) color (Q1075), simply just because they are uniquely defined. Something unique can not be a subgroup. Colours without unique and unambiguous definitions (like shade of red (Q7460345)) on the other hand can absolutely be subclass of (P279) color (Q1075). /ℇsquilo 13:41, 26 October 2015 (UTC)
2. I agree. <instance of:color> where a colour has a unique definition (which should generally be recorded in a statement). <Subclass of:color>, or <subclass of:purple> for example, for colours where this doesn't apply. Joe Filceolaire (talk) 13:39, 30 November 2015 (UTC)

   @Esquilo, Filceolaire: I would more use a scheme were we still use subclass. There is indeed a lot of instances of "red", even in the unique definition HTML like (255,0,0) example. Each pixel can be or not in these color. I would not really agree that this correspond to a unique color because ... this could be rendered in way different means in different screen. BUT I'll agree there is "some kind" of uniqueness at sake here : I would tag this as

   ⟨ red/(255,0,0) ⟩ instance of (P31) ⟨ primary color type ⟩

   . I think with human eye we could assimilate some colors to this one, those who are undiscernable to this one with a human eye, but that a sensible physical equipment would notice different. it's still a range, even if very small. author  TomT0m / talk page 14:44, 30 November 2015 (UTC)

p279

1. . I prefer colors as subclasses of one another, i.e., crimson as a subclass of red. Even particular colors, e.g., the red color of the soles of shoes made by Christian Louboutin, can be considered to be degenerate classes. Peter F. Patel-Schneider (talk) 21:45, 19 October 2015 (UTC)
2. . Per my reasoning in the discussion section. Crimson lights are all red lights, so
   ⟨ Crimson ⟩ subclass of (P279) ⟨ red ⟩
   . author  TomT0m / talk page 19:13, 21 October 2015 (UTC)
3. Philosophical meanderings aside, the widely-used upper ontology BFO would model red (Q3142) with subclass of (P279), not instance of (P31). I vote for trivially easy compatibility with them. I would also prefer not to use both instance of and subclass of on colors and many other domains. We can apply "instance of foo type" and "subclass of foo" everywhere, but that does not mean we should. Such pervasive explicit metamodeling is unnecessary bloat and would overcomplicate our ontology for the vast majority of users. Emw (talk) 01:57, 4 January 2016 (UTC)

both

1. . The relationship between colors should be subclass, but there is no reason to forbid instance relationships from colors to metaclasses, for example primary colors. Peter F. Patel-Schneider (talk) 21:45, 19 October 2015 (UTC)
2. . Agree with Peter, per my reasoning above I'd add
   ⟨ Primary color ⟩ subclass of (P279) ⟨ color ⟩
   . author  TomT0m / talk page 19:15, 21 October 2015 (UTC)

Question: Usage of both properties

In the larger picture Wikidata should allow building multiple ontologies. One source might speak of something as an instance, while another source will treat something as a subclass. Should we allow usage of both properties for every item on Wikidata (Which would mean that removing either p31 or p279 would be vandalism (but assuming good faith))?

Yes

1. . It is always possible to create a metaclass for any group of classes, e.g., color type for colors, so it is not possible to rule out instance links from classes. Peter F. Patel-Schneider (talk) 21:45, 19 October 2015 (UTC)
2. . Yes, with some precautions : classes (of individuals) should never be subclasses of oher things than classes (of individuals), classes of classes (of individuals) should never be subclasses of oher things than classes of classes (of individuals), ... This implies that an instance of something can never be a subclass of the same thing, either directly or following a subclass chain. author  TomT0m / talk page 20:23, 21 October 2015 (UTC)

   In general I agree, but there can be exceptions. For example, the class of all classes has as instances both first-order classes (classes of individuals) and higher-order classes (classes of classes) and is both a subclass and instance of itself. Peter F. Patel-Schneider (talk) 14:07, 22 October 2015 (UTC)

   I don't think it's really useful to really have a class of all (meta)ⁿclass on Wikidata, so this could be left as a theorical exception or as a query pretty easily. author  TomT0m / talk page 16:05, 22 October 2015 (UTC)

No

1. . Although it is always possible to create a metaclass for any group of classes, e.g., color type for colors, it may not always be a good idea. Peter F. Patel-Schneider (talk) 21:45, 19 October 2015 (UTC)
2. . As the question is asked : I tend to think the definitions needs to be consistent. I don't really get what "one source might talk of something as an instance, another as a subclass" means in practice. We have to check that they use the same definitions of instance and subclass that we use. Overwise we have to translate their definitions in our language. Or use different properties than our "instance" and "subclass". The fact that we will import other ontologies won't mean that we have to take other definitions for our properties. The imports might not be 1 to 1 mappings. author  TomT0m / talk page 20:23, 21 October 2015 (UTC)
3. Per Peter: "Although it is always possible to create a metaclass for any group of classes, e.g., color type for colors, it may not always be a good idea." I go a step further: it is usually not a good idea. Also, the idea that we should accommodate myriad fundamentally conflicting ontologies is unworkable in practice and a bad idea. Accommodating contradictory information is essential in some domains, but we should be able to have a simple, consistent preferred interpretation of what color is. Emw (talk) 01:41, 4 January 2016 (UTC)
4. No. It's useful if different conceptions of color have their own concept. It's bad to mix multiple slightly different concepts in the same concept. ChristianKl (talk) 13:14, 26 May 2016 (UTC)

Not part of RfC but for those interested

On second thought, this should not be part of this particular RfC. You can leave comments, but they will not be part of the RfC outcome. The question is intended to connect the principles of the questions above to the rest of Wikidata. But no decisions will be made based on the below comments. --Tobias1984 (talk) 17:15, 21 October 2015 (UTC)

Question: p31 only for individuals

Should instance of (P31) only be used for individuals? Meaning: A fixed object in time and space? If we use for example "honda accord" instance of (P31) "car", does that not dillute the meaning of "Freddy Mercury" instance of (P31) "human". The first example talks about a car that has millions of instances, while the second example is a true individual.

Yes

1. .
2. .

No

1. . Classes can always also be considered as individuals and can be usefully made instances of metaclasses, e.g., the red color class as an instance of primary colors. Such abstract individuals are everywhere, encompass even more than classes, and should not be ruled out. Peter F. Patel-Schneider (talk) 21:45, 19 October 2015 (UTC)
2. . I agree - "instance of" is useful in both abstract and concrete contexts, and the distinction between individuality and multiplicity is never entirely clear-cut, as any given object having a continuing existence can also be thought of as multiple instances of the "object at a given time", among other ambiguities... Also in the above case, "Honda accord" can't be an "instance of" "car", it's an instance of "car model" perhaps. One has to think about the meaning of the relationships - if the target term is ambiguous then some way of disambiguating (via descriptions in wikidata) is needed. ArthurPSmith (talk) 18:05, 21 October 2015 (UTC)
3. . No. Per Help:Classification, once more, it's the only way to use correctly some items like "car model". btw. it's incorrect to say "honda accord" instance of (P31) "car" because any "honda accord" instance is an instance of "car" is defiitely a class, and that would make two individual classes instance of each other. "honda accord" instance of (P31) "car model" however is cool, this would make "car model" a class of class of individual. classes of individual should always be used with subclass of (P279) with other classes of individuals, metaclass of individuals always with metaclass of individuals, and so on. Otherwise, as transitivity of "subclass of", we would risk stuffs like "The Porshe I say yesterday is a car model", which we do not want. Porshe 911 is a car model, never individual cars. author  TomT0m / talk page 20:02, 21 October 2015 (UTC)
4. For example civil parish (Q4976993) is a subclass of (P279) administrative territorial entity (Q56061), but an instance of (P31) designation for an administrative territorial entity (Q15617994). The latter relationship can be used for example with the is a list of (P360) property, eg to specify that the content of an article is a list of the types of administrative subdivisions in a country, ie instances of subclasses of Q15617994, rather than a list of the boroughs, counties, municipalities themselves, that would be instances of subclasses of Q56061. Jheald (talk) 14:43, 22 October 2015 (UTC)

Other

Most of the "No" votes above seem to be "Yes" votes for the question "Does the statement 'Honda Accord instance of car', dilute the meaning of 'Freddy Mercury instance of human'?" In other words, if we say that Honda Accord is an instance of car, then the meaning of "instance of" changes when we say that Freddy Mercury is an instance of human. We should avoid that.

No, instance of should not always be limited to concrete individuals; statements like "Honda Accord instance of car model" are fine. However, instance of overwhelming gets used incorrectly where subclass of should be used (even in the proposal being voted on!), so I think the questions under consideration are misguided.

All classes are implicitly individuals, but not all individuals are classes. This an extension of what Peter F. Patel-Schneider says above. In the case of things we typically think of as "fixed objects in space and time" -- a.k.a. physical object (Q223557) -- one should indeed not use instance of (P31) unless the subject has a unique or same spatial location for every point in time.

Regarding the ambiguity that ArthurPSmith asserts can exist -- "any given object having a continuing existence can also be thought of as multiple instances of the 'object at a given time'" -- I do not recall ever seeing that assertion made in any actual ontologies or philosophical literature. Widely-used top-level ontologies like BFO (https://github.com/BFO-ontology/BFO) have considered properties like point in time (P585) (which they label "at time"), but do not consider each usage of that property to comprise a distinct individual. There is no ambiguity; noone actually believes that an instance of a physical object existing at multiple points in time entails it is multiple instances.

Finally, please do not base proposals on syntactically negative questions ("does that not dillute"). The logically correct answer is often the opposite of what speakers respond. Emw (talk) 01:36, 4 January 2016 (UTC)

I think I mostly agree with what Emw says above - if I understand it. On the ambiguity issue though, I was maybe thinking of the case of 1997 Cannes Film Festival (Q252688) vs Cannes Film Festival (Q42369) - which is an event repeating in time rather than a physical object; however I'm not certain there are hard-and-fast boundaries one can draw that prevent leaking of that sort of time-based instance-of/class relationship into actual physical objects - for example why treat an event that way but not the venue in which it is held - buildings change over time too. As does anything that's not an elementary particle. The issue here is what does "instance" mean. We seem to be relying on an intuition that is not at all clear to me at least. ArthurPSmith (talk) 15:50, 4 January 2016 (UTC)

@ArthurPSmith: Not really, the token/type distinction is a philosophical principle that's far from just beeing an intuition ( https://en.wikipedia.org/wiki/Type%E2%80%93token_distinction - see cited some well known philosophers ) with this plus metaclasses (see w:en:Metaclass_(Semantic_Web)) and metaclasses levels, we are far from just intuitive ideas : we have a philosophical model who tight reality to our model, plus maths to ensure some properties of the model. author  TomT0m / talk page 13:29, 10 April 2016 (UTC)

@TomT0m: I think we're carrying along similar conversations in several different places, maybe we ought to consolidate somewhere? :) I do agree this is important to understand, but I disagree that there is real philosophical clarity on these matters. The Type-Token distinction is one philosophical stance, but the wikipedia page you link to hardly does it justice (and I'm surprised there's no French version?) See this Stanford Encyclopedia of Philosophy entry for a much more nuanced view of the matter. Most of discourse concerns what seem by those arguments to be types, not tokens (I think the only "tokens" in our discussion here are the references to ourselves as individual people) so the distinction may not even be helpful in most situations - if almost everything is a type, why does the type/token distinction matter? It becomes even more confusing now with electronic devices: words written on a web page are "types", locally stored on the web server(s?) as tokens, which get copied to a different spatial location with each download (if the page is ever downloaded). There is presumably one token consisting of the rendering as presented by the browser on the screen at any given occasion of viewing the page, but there's another token corresponding to the local storage by the browser of a cached copy of the page, etc. etc. And all that seems pretty irrelevant to wikidata's needs. What are the real practical questions we are trying to address here? Here's a list of the questions I think that matter:

• what does "instance of" mean within wikidata?
• what does "subclass of" mean?
• do we need to distinguish classes and metaclasses, and allow for several levels of metaclasses?
• does a given item in wikidata (i.e. some particular Qxxxx) always correspond to a single logical/philosophical semantic construct, or can it represent several different things at once (as ordinary language often does)?

On the first question (and also addressing some of the others), I think this paper by Doug Foxvog on how Cyc handled it is instructive: "Instances of Instances Modeled via Higher-Order Classes". It describes their use of some clear metaclass levels with "fixed order", i.e. every metaclass of that type had instances that were of the metaclass below. But they also needed variable-order metaclasses. Not everything fits into the neat fixed-order metaclass structure that a philosopher might like.

On the last question, I have really enjoyed Douglas Hofstadter's writings on level-crossings and the complexities of representation; I just recently finished reading his book with Emmanuel Sander, "Surfaces and Essences: Analogy as the Fuel and Fire of Thinking". The book was apparently written in French and English simultaneously -I'd encourage you to locate a copy. To me they seem to strongly advocate for quite a different philosophical stance than the token-type distinction: more important to both our everyday thinking and to creativity is generation of categories through analogy, a spontaneous and constantly evolving process that starts from an individual instance of some concept presented to us, and grows over time as we are each exposed to more examples that appear similar in some fashion. To that extent, the category is, initially at least, represented specifically by the archetypal instance. Albert Einstein, for example, is both clearly an individual person who did specific things in his life, but also a representative of a type of genius (and likely other types, such as hair style) and one can talk of an "Einstein" of (as Google autocomplete suggests): "money", "India", "the 21st century", etc. Should wikidata contain an entry both for Einstein as a person and also for Einstein as a type? Maybe, but this is such a ubiquitous phenomenon that I believe this is both impractical and contrary to the way human understanding actually works. We need to live with at least some ambiguity to make our tools (like wikidata) as useful as things like language, which have these ambiguities built in. At least that's my feeling on this right now - I'm willing to be convinced otherwise of course but I think it needs some very clear thinking and not just relying on particular philosophical principles that are not universally held. ArthurPSmith (talk) 18:26, 11 April 2016 (UTC)

@ArthurPSmith: To your question "can Albert Einstein be a type", I think we already have everything we need. The answer is something similar to type specimens for taxons and minerals, that used to be the proof that a taxon could be instantiated. Let the class of all "Albert Einstein" genius be the class of all person that exhibits Einstein-like behavior (that qualifies the reasons for which we consider a genius). Then Einstein will be our "type specimen" that prove that this class make sense. This is not really a difficulty for the class/token distinction, nor to the definition by extension of such a class : we define the class by similarity of an individual. Indeed the Cyc paper is interesting, integrated into en:metaclass (Semantic Web) almost as I read it, but it's not really new material for my personal reflection. The levels allows us to consider classes as if they were tokens, so it's not really a problem that a lot of stuffs are classes : they are instances of something, and on a materialistic perspective they probably have tokens. So we can do exactly the same by considering them as instance of some metaclass that by considering them tokens instance of a class. Except we force us to answer the question "what are the tokens at sake here". No big deal, we just gain in ontological clarity and into foundation of our ontology. Indeed this should be consolidated, I'm tired of repeating (almost) the same things over and over. I wrote Help:Classification for this, there is all those ontology wikiproject and classification noticeboard, not of much use we must admit at that point. Plus it's competing with the own project discussions. author  TomT0m / talk page 19:49, 11 April 2016 (UTC)

To be more precise about the "levels" idea, the reason why I did not really insisted on this earlier is to not bother wikidatans with this because ... there is very very little use for levels higher than the first metaclass level, the only one that can come to my mind at this point is the "taxon rank" metaclass. Cyc came to the same conclusion, I wonder why they needed the fifth level. I'll look at it. author  TomT0m / talk page 20:01, 11 April 2016 (UTC)

@ArthurPSmith: Found. Actually the only instance of "fourth order collection" seems to be ... "third order collection", see http://sw.opencyc.org/concept/Mx4rHUFI8h_TEdaAAABQ2rksLw Actually amongst the third order we find something related to color classification, so it's relevant here : http://sw.opencyc.org/concept/Mx4rHUFI8h_TEdaAAABQ2rksLw author  TomT0m / talk page 20:15, 11 April 2016 (UTC)

Ok. I'm not sure how much we're agreeing and disagreeing here then. In my view trying to apply perfect mathematical reasoning to this problem is not going to work; nevertheless I think we can get pretty far with common-sense reasoning. You seem to have a different definition of "token" from what I've seen elsewhere - but I don't believe the terminology there is the important thing. It seems to me we do agree on (at a common-sense level) what "instance of" means - the Cyc paper shows pretty clearly what can happen. I think I'd like to work out some concrete examples (like "taxon rank" or the third-order color classification or maybe some of the things related to chemistry or manufactured objects) that clearly demonstrate what we're looking for. The Stanford encyclopedia article suggested that the meaning of "instance of" is actually rather domain-specific. Is there a good place in Wikidata:WikiProject_Ontology to work on these things? ArthurPSmith (talk) 20:35, 11 April 2016 (UTC)

@ArthurPSmith: Actually the second metaclass order, the meta-meta-class level, is quite instrucive I think and reflects totally my view : http://sw.opencyc.org/concept/Mx4rHQdVmB_TEdaAAABQ2rksLw One particularly interesting example : movies. See http://sw.opencyc.org/concept/Mx4rv973YpwpEbGdrcN5Y29ycA Movies are conceptual work, and can be instantiated into movie showing : http://sw.opencyc.org/concept/Mx4rvVjVH5wpEbGdrcN5Y29ycA movie showing would be what I call a "token", a localized spatio temporal thing/event. Every movie is then abstract, that can be seen as the class of all these showing - for a specific film. Its subtypes are movie "genre" in wikidata : http://sw.opencyc.org/concept/Mx4rv973YpwpEbGdrcN5Y29ycA . There is several kind of way to classify movies: by genre, see http://sw.opencyc.org/concept/Mx4rv2VcW5wpEbGdrcN5Y29ycA which is a metaclass of higher level, or by MPAA rating, which is a metaclass of the same level that the "genre one" : http://sw.opencyc.org/concept/Mx4rv2VcW5wpEbGdrcN5Y29ycA . Which is totally consistent with what I argue for quite some time now ... author  TomT0m / talk page 20:53, 11 April 2016 (UTC)

@TomT0m: I have started some edits to reflect some of the things we've talked about here at Wikidata:WikiProject_Ontology/Modelling - perhaps we can work there or on related subpages to clarify what we're trying to do a bit more? ArthurPSmith (talk) 21:19, 12 April 2016 (UTC)

@TomT0m: See also Wikidata:WikiProject_Ontology/Problems; I've started listing some issues with the existing wikidata ontology that we should work on fixing. ArthurPSmith (talk) 21:22, 13 April 2016 (UTC)

Same problem with religion (Q9174)

When trying to make an ontology in the "Religions" project, we face the very same problem. And it tends to be really difficult to decide what is an instance of a given religion: should we have two elements for manichaeism (Q131165), one for a sub-class, one for an instance? Should we use both properties on the same element? etc. — nojhan (✐) 08:21, 18 February 2016 (UTC)

• @nojhan: Previously discussed here (without a conclusive result though). 92.100.111.33 21:17, 9 April 2016 (UTC)
• @nojhan: Should we use both properties on the same element Yes. This is justtified both by existing technologies who allows punning like OWL2 and by everiday language practice in which we does that pretty naturally. author  TomT0m / talk page 13:25, 10 April 2016 (UTC)

---

The above discussion is preserved as an archive. Please do not modify it. Subsequent comments should be made in a new section.