David,
On 28/05/14 16:35, David Cuenca wrote:
Markus,
Ok, now I understand that "same as" wouldn't be a good name for the confusion it would cause. However the property "subject of" as it is now wouldn't be a good candidate either. Its meaning is that a certain statement is represented by another item (that is why it is only allowed to be used as qualifier).
Ok.
Perhaps a better name would be "corresponds with item" and the inverse "corresponds with property". Just by having these connections, a lot of information can be inferred from the connected item.
Consider the following example with "occupation (P106)", and "occupation (Q13516667)":
- I cannot find any clear "subproperty of" for p106, but there is a
clear "subclass of:human behaviour" for the item
- "human behaviour" is "part of" human
I don't understand this use of "part of". Maybe I would say "having an occupation is part of being human" but not that "occupation is part of human". I would not use either of these and restrict "part of" to clear, undisputed statements like "the steering wheel is part of the car". Otherwise, anything could be part of human ("head"?, "sadness"?, "singing"?, "birth"? -- entering this in Wikidata would not lead anywhere).
"Part of" is quite problematic in general. You can see it from the discussion on its property page, and also from the uses it sees in the wiki, that this property is severely misunderstood and/or misused. At the very least, one should distinguish "physical part of" from "meronym" (both are aliases of the property now!). And then one should realise that meronyms are in the domain of Wiktionary, which we cannot capture in Wikidata properly since we do not have items for words but for concepts. One alias for an item might be a meronym of something else, while another alias for the same item is not. Using statements for linguistic properties in Wikidata will not be successful. I am not saying that Wikibase is not able to capture some ideas of a thesaurus (we have actually discussed this), but this is not how it is used in Wikidata.
- "human" can have a statement "intrinsic property" (property proposal
still under discussion) with values "birthday (Q47223)" and an "(eventual) date of death". It can be expanded in the future to include newly created properties like "height", "weight", "eye color", etc
Yes, this again makes sense to me. It is basically a variant of the constraint "Item" which allows you to say that items that are instance of human should also have a birthday. But again, this is schematic information (like constraints) and it should not be mixed up with actual data. It is the same conceptual difference that I have explained for properties vs. items earlier. Moreover, I think this information (even if correct in some sense) has very little utility as a piece of information about an item; it is much more useful for constraints about properties (which are not items).
- birthday (Q47223) <corresponds with property> date of birth (P569)
It should be the other way around: the correspondence says something about P569, not about Q47223. There cannot be any reference for this. It should therefore be a claim on the page of P569 rather than a statement on the page of Q47223.
Out of this I reach the following conclusions:
- the taxonomy of properties is going to be weak, since there is not
always a clear subpropertyOf unless created artificially (more work)
I agree.
- the standard taxonomy of items (subclass of/part of) is sufficient
to automatically reach meaningful constraints and inference (less work)
I agree that the taxonomy will be helpful in constraints. This is what constraints already do when using instance of/subclass of. However, I do not agree that the constraints can or should be stated as part of this taxonomy. Constraints are too complex, and they are conceptually different (they say how a property should be used, not how something in the Real World relates to something else). Constraints interact nicely with the taxonomy and help to get useful conclusions, but they are not "part of" taxonomy ;-). We must keep content organisation separate from content.
- by adding manually the constraints to the property itself we are
duplicating information which will require volunteer effort to maintain (more work)
I disagree. Constraints refer to the property, not to the Wikidata item, and it would be conceptually wrong to mix these things up. We already have agreed that properties and items need to remain distinct for technical reasons. Once this is clear, there is no reason to move information that refers to properties (constraints) to item pages. This will not be a duplication of information: it is enough to have the constraints on the property pages only. If you look at the constraints we have, you can see many examples that are specific to Wikidata and certainly not a general thing about the concept (take the "allowed values" for "sex or gender"). We really want to keep editorial helpers (constraints) distinct from sourced information (statements about items).
My recommendation is to rely mainly on the main taxonomy instead of creating a parallel property taxonomy, and then think of ways to extract information from the main taxonomy to convert it automatically into constraints. All the maintenance takes effort, so the more it can be automated, the more efficient volunteers will be. And if we can simplify the maintenance of properties, we will be able to simplify the creation of properties too, specially when we face the next surge which will come with the datatype "number with units".
I agree with the general goals, but I don't think that things become any easier if we confuse information about properties with information about items. We can still re-use information we have about items (like the class hierarchy that we already use in constraints) to avoid duplication, but some things are clearly not part of the item taxonomy.
Cheers,
Markus
On Wed, May 28, 2014 at 2:48 PM, Markus Krötzsch <markus@semantic-mediawiki.org mailto:markus@semantic-mediawiki.org> wrote:
David, Regarding the question of how to classify properties and how to relate them to items: * "same as" (in the sense of owl:sameAs) is not the right concept here. In fact, it has often been discouraged to use this on the Web, since it has very strong implications: it means that in all uses of the one identifier, one could just as well use the other identifier, and that it is indistinguishable if something has been said about the one or the other. That seems too strong here, at least for most cases. * In the world of OWL DL, sameAs specifically refers to individuals, not to classes or properties. Saying "P sameAs Q" does not imply that P and Q have the same extension as properties. For the latter, OWL has the relationship owl:equivalentProperties. This distinction of instance level and schema level is similar to the distinction we have between "instance of" and "subclass of". * Therefore, I would suggest to use a property called "subproperty of" as one way of relating properties (analogously to "subclass of"). It has to be checked if this actually occurs in Wikidata (do we have any properties that would be in this relation, or do we make it a modelling principle to have only the most specific properties in Wikidata?). * The relationship from properties to items could be modelled with the existing property "subject of" (P805). * It might be useful to also have a taxonomic classification of properties. For example, we already group properties into properties for "people", "organisations", etc. Such information could also be added with a specific property (this would be a bit more like a "category" system on property pages). On the other hand, some of this might coincide with constraint information that could be expressed as claims. For instance, person properties might be those with "Type" (i.e., "rdfs:domain") constraint human. By the way, our constraint system could use some systematisation -- there are many overlaps in what you can do with one constraint or another. Cheers, Markus On 28/05/14 12:14, David Cuenca wrote: Markus, The explanation about the implications of renaming/deleting makes most sense and just that justifies already the separation in two. It is equally true that when we create a property, we might have "cleaned" the original concept so much that it might differ (even slightly) with the understood concept that the item represents. However, even after that process, the "new" concept is still an item... The process of imbuing a concept with permanent characteristics (adding a datatype) and the practical approach, also seems to recommend keeping items and properties separate. Thanks for showing me that reasoning :) I am still wondering about how are we going to classify properties. Maybe it will require a broader discussion, but if they are the same (or mostly the same) as items, then we can just link them as "same as", and build the classing structure just for the items. OTOH, if they are different, then we will need to mirror that classification for properties, which seems quite redundant. Plus adding a new datatype, "property". All in all, my conclusion about this is that properties are just concepts with special qualities that justify the separation in the software (even if in real life there is no separation). many thanks for your detailed answer, and sorry if I'm bringing up already discussed topics. It is just that when you stare long into wikidata, wikidata stares back into you ;) Cheers, Micru On Wed, May 28, 2014 at 11:39 AM, Markus Krötzsch <markus@semantic-mediawiki.org <mailto:markus@semantic-mediawiki.org> <mailto:markus@semantic-__mediawiki.org <mailto:markus@semantic-mediawiki.org>>> wrote: Hi David, Interesting remark. Let's explore this idea a bit. I will give you two main reasons why we have properties separate, one practical and one conceptual. First the practical point. Certainly, everything that is used as a property needs to have a datatype, since otherwise the wiki would not know what kind of input UI to show. So you cannot use just any item as a property straight away -- it needs to have a datatype first. So, yes, you could abolish the namespace Property but you still would have a clear, crisp distinction between property items (those with datatype) and normal items (those without a datatype). Because of this, most of the other functions would work the same as before (for example, property autocompletion would still only show properties, not arbitrary items). A complication with this approach is that property datatypes cannot change in Wikibase. This design was picked since there is no way to convert existing data from one datatype to another in general. So changing the datatype would create problems by making a lot of data "invalid", and require special handling and special UI to handle this situation. With properties living in a separate namespace, this is not a real restriction: you can just create a new property and give it the same label (after naming the old one differently, e.g., putting "DEPRECATED" in its name). Then you can migrate the data in some custom fashion. But if properties would be items, we would have a problem here: the item is already linked to many Wikipedias and other projects, and it might be used in LUA scripts, queries, or even external applications like Denny's Javascript translation library. You cannot change item ids easily. Also, many items would not have a datatype, so the first one who (accidentally?) is entered will be fixed. So we would definitely need to rethink the whole idea of unchangeable datatypes. My other important reason is conceptual. Properties are not considered part of the (encyclopaedic) data but rather part of the schema that the community has picked to organise that data. As in your example, "emissivity" (Q899670) is a notion in physics as described in a Wikipedia article. There are many things to say about this notion (for example, it has a history: somebody must have defined this first -- although Wikipedia does not say it in this case). As in all cases, some statements might be disputed while others are widely acknowledged to be "true". For the property "emissivity" (P1295), the situation is quite different. It was introduced as an element used to enter data, similar to a row in a database table or an infobox template in some Wikipedia. It does probably closely relate to the actual physical notion Q899670, but it still is a different thing. For example, it was first introduced by User:Jakec, who is probably not the person who introduced the physical concept ;-) Anything that we will say about P1295 in the future refers to the property -- a concept of our own making, that is not described in any external source (there are no publications discussing P1295). This is also the reason why properties are supposed to support *claims* not *statements*. That is, they will have property-value pairs and qualifiers, but no references or ranks. Indeed, anything we say about properties has the status of a definition. If we say it, it's true. There is no other authority on Wikidata properties. You could of course still have items and properties "share" a page and somehow define which statements/claims refer to which concept, but this does not seem to make things easier for users. These are, for me, the two main reasons why it makes sense to keep properties apart from items on a technical level. Besides this, it is also convenient to separate the 1000-something properties from the 15-million something items for reasons of maintenance. Best regards, Markus On 28/05/14 09:25, David Cuenca wrote: Since the very beginning I have kept myself busy with properties, thinking about which ones fit, which ones are missing to better describe reality, how integrate into the ones that we have. The thing is that the more I work with them, the less difference I see with normal items.... and if soon there will be statements allowed in property pages, the difference will blur even more. I can understand that from the software development point of view it might make sense to have a clear difference. Or for the community to get a deeper understanding of the underlying concepts represented by words. But semantically I see no difference between: cement (Q45190) <emissivity (P1295)> 0.54 and cement (Q45190) <emissivity (Q899670)> 0.54 Am I missing something here? Are properties really needed or are we adding unnecessary artificial constraints? Cheers, Micru ___________________________________________________ Wikidata-l mailing list Wikidata-l@lists.wikimedia.org <mailto:Wikidata-l@lists.wikimedia.org> <mailto:Wikidata-l@lists.__wikimedia.org <mailto:Wikidata-l@lists.wikimedia.org>> https://lists.wikimedia.org/____mailman/listinfo/wikidata-l <https://lists.wikimedia.org/__mailman/listinfo/wikidata-l> <https://lists.wikimedia.org/__mailman/listinfo/wikidata-l <https://lists.wikimedia.org/mailman/listinfo/wikidata-l>> ___________________________________________________ Wikidata-l mailing list Wikidata-l@lists.wikimedia.org <mailto:Wikidata-l@lists.wikimedia.org> <mailto:Wikidata-l@lists.__wikimedia.org <mailto:Wikidata-l@lists.wikimedia.org>> https://lists.wikimedia.org/____mailman/listinfo/wikidata-l <https://lists.wikimedia.org/__mailman/listinfo/wikidata-l> <https://lists.wikimedia.org/__mailman/listinfo/wikidata-l <https://lists.wikimedia.org/mailman/listinfo/wikidata-l>> -- Etiamsi omnes, ego non _________________________________________________ Wikidata-l mailing list Wikidata-l@lists.wikimedia.org <mailto:Wikidata-l@lists.wikimedia.org> https://lists.wikimedia.org/__mailman/listinfo/wikidata-l <https://lists.wikimedia.org/mailman/listinfo/wikidata-l> _________________________________________________ Wikidata-l mailing list Wikidata-l@lists.wikimedia.org <mailto:Wikidata-l@lists.wikimedia.org> https://lists.wikimedia.org/__mailman/listinfo/wikidata-l <https://lists.wikimedia.org/mailman/listinfo/wikidata-l>-- Etiamsi omnes, ego non
Wikidata-l mailing list Wikidata-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata-l