Re: Modeling fictional characters in movies and TV from Gregg Kellogg on 2013-01-14 (public-vocabs@w3.org from January 2013)

From: Gregg Kellogg <gregg@greggkellogg.net>
Date: Mon, 14 Jan 2013 05:24:21 -0500
To: Richard Wallis <richard.wallis@oclc.org>
CC: Jason Douglas <jasondouglas@google.com>, "public-vocabs@w3.org" <public-vocabs@w3.org>
Message-ID: <1A5C60FF-B4A9-44CB-9466-8ED9D3CDDE6B@greggkellogg.net>
On Jan 14, 2013, at 10:32 AM, Richard Wallis <richard.wallis@oclc.org<mailto:richard.wallis@oclc.org>> wrote:

Hi Gregg,

Referencing the other current thread on this list about the purpose of Schema.org<http://Schema.org> and the breadth and purpose of the list itself, I am looking at this from the point of view of someone marking up their web pages.  Using the additionalType property, I come up with something like this in Microdata:

<div itemscope itemtype="http://schema.org/Person"<http://schema.org/Person%22>>
   <link itemprop="additionalType" href="http://schema.org/FictionalThing" />
   <span itemprop="name">Harry Potter</span>
   Affiliation:
   <span itemprop="affiliation">Hogwarts</span>
   <div itemscope itemtype="http://schema.org/FictionalThing"<http://schema.org/FictionalThing%22>>
    From the book:  <span itemprop="referencedIn">Harry Potter and the Philosopher’s Stone</span>
  </div>
</div>

Actually, I don't think you want to nest the "referencedIn" property within it's own item.

I am deliberately resisting the urge to create a Character class, as it is an approach that will obviously not scale across all the fictitious things that we could come across.   How is this different to saying that a thing  we are describing is a schema:Book with an additionalType of schema:Offer so that we can so that we can describe both its creative work and product or sale properties.

The reason I proposed a Character type is because I think that a missing property from Movie, TVEpisode and so-forth is, in fact, schema:character. To say that a Movie has a name, director and actors, but not characters, seems to be missing a big part about what people expect from movies, books or TV shows. After all, it's the characters that draw people to these things, it's just the actors that help realize it. Given the need for a character property, what is it's expected range? Thus, my proposal to add a Character class.

I acknowledge that Character may be a special case that requires a type – after all we have a word in English to describe a fictitious person in a creative work, so it may well have enough extra importance to demand one.  But using what is a special case to test a generic theory doesn’t work very well – lets get generic right and then specialise if it is inadequate.

I'm not really trying to solve a generic problem here, and I don't think that the schema.org<http://schema.org> maintainers are that interested in generic things. However, I do think that the vocabulary should be able to express them; adding a FictionalThing class that can be "mixed-in" seems to accomplish this generically.

A bit of turtle always helps me see what I’m modelling, this is what I get for the above which would naturally map to RDFa:
<http://example.com/people/1234>
    a <http://schema.org/Person>;
    a <http://schema.org/FictionalThing> ;
    schema:name "Harry Potter";
    schema:affiliation "Hogwarts";
    schema:referencedIn “Harry Potter and the Philosopher’s Stone”.

Except for the referencedIn I mentioned above, yes. From this perspective, a Character class isn't necessary, but as I mentioned above, if you think (as I do) that a character property is required, then you need a way to describe what the value of that property can be.

I’m no RDFS expert so the nuances of your explanation are not fully clear – however many of our [web page markup] audience are not even aware of RDFS.

The above feels like a simple solution to me (which may be simpler to use, than describe using OWL & RDFS).

The RDFS Entailment stuff is more intended to motivate what the semantics of rangeIncludes and domainIncludes is. This is not something that should be exposed in the actual vocabulary, or anything that web developers need to know. But for those who want to reason over the vocabularies, we need a way of known precisely what is meant by rangeIncludes or domainIncludes. However, this is divergent from the main thrust of the proposal, but at some point we need to understand the meanings of these annotation properties.

Gregg

~Richard.

On 13/01/2013 09:05, "Gregg Kellogg" <gregg@greggkellogg.net<x-msg://335/gregg@greggkellogg.net>> wrote:

Hi Richard,

On Jan 11, 2013, at 5:44 PM, Richard Wallis <richard.wallis@oclc.org<x-msg://335/richard.wallis@oclc.org>> wrote:

Re: Modeling fictional characters in movies and TV
Hi Gregg,

Sorry for replying to this one out of sequence, I have been thinking about it and discussing it ;-)

You say “From my understanding of RDFS, if I have a subject with two different types, then I'm really saying that the properties associated with this subject are the intersection of those two types.”

My understanding is that the result would be the union of the properties of the two types, thus achieving the result I was suggesting.

Yes, of course you're right, I was confusing this with rdfs:subTypeOf. My concern is that if I wanted to describe properties on a fictional person, I would need to define a type which is the union of schema:Person and schema:FictionalThing, such as the following:

:FictionalPerson a [ a owl:Class; owl:unionOf (schema:Person schema:FictionalThing)] .

I am confused by your actor example, as an actor would not normally be a fictional Person, the character they portray would be.

I was using schema:actor as a property of a schema:Character. When I was saying that schema:actor had the domain of wikia:Character, I was saying that a Character may have an associated Actor; I think this is okay. If I had used rdfs:range instead of rdfs:domain, then it would be a problem.

To satisfy your example of an actor portraying a fictional character, I believe we may be looking for a type of Person ‘Actor’ that has a property of ‘portrayed’ or ‘created’  that would have an expected type of Person.  When encoding your data you could then describe that person (your character) as being of type Person and of type FictionalThing.

That "portrayed" is an appropriate inverse property of actor, as if a character has an Actor, an actor portrays a Character.

To be successful we need to find a generic way of enabling any ‘thing’ to be described as a FictionalThing – picking off individual solutions for specific types of things such as a character simply won’t scale.

Yes, I think that having such a generic mechanism is a good idea, it's just that if you want to start to describe properties on the "mixin" of Person and FictionalThing, though, you need to be able to describe the union type. I believe that this is the intent of schema:rangeIncludes/domainIncludes. I think that appropriate rules for these could be the following (although expressing it formally will require some more work.

For all classes ?c* which have a common subject and rdfs:rangeIncludes predicate, construct the following triples:

_:rangeClass a owl:Class; owl:unionOf (?c*)  . # same for domains

Actually constructing these statements is challenging because owl:unionOf takes a list rather than multiple repeated statements for each object.

Note that multiple properties having the same set of range/domainIncludes will result in multiple blank nodes, which entail each other, but are not equivalent nodes. Furthermore, actually using the property in a concrete instantiation uses an entailment rule that claims that the range is of an equivalent class

<JamesTKirk> schema:actor <WilliamShatner> .

would end up entailing

<JamesTKirk> a [ owl:Class; owl:unionOf (schema:Person schema:FictionalThing) ].

due to rule rdfs2

If it is appropriate for schema.org<http://schema.org> <http://schema.org<http://schema.org/>>  to define a something like a character property (in addition to the Performance property suggested by Yves), then I think it does need to define a schema:Character class, so that properties such as schema:actor can be ascribed to it. For schema definition purposes, in the spirit of not getting all pedantic about using OWL to do this representation, it might require adding something like a schema:classIncludes annotation property, similar to rangeIncludes and domainIncludes so we could assert:

schema:Character schema:classIncludes schema:Person, schema:FictionalThing .

Gregg

~Richard.



On 09/01/2013 21:14, "Gregg Kellogg" <gregg@greggkellogg.net<x-msg://335/gregg@greggkellogg.net> <x-msg://178/gregg@greggkellogg.net<x-msg://178/gregg@greggkellogg.net>> > wrote:

Hi Richard,

On Jan 9, 2013, at 1:38 AM, Richard Wallis <richard.wallis@oclc.org<x-msg://335/richard.wallis@oclc.org> <x-msg://178/richard.wallis@oclc.org<x-msg://178/richard.wallis@oclc.org>> > wrote:

Re: Modeling fictional characters in movies and TV
Starting from the point of view that most of the ‘real’ types of things we want to describe could have fictional equivalents – fictional people (characters), fictional organisations, fictional places, fictional products, fictional automobiles, fictional planets, fictional fizzy drinks, fictional races, fictional religions, fictional songs, ...., ....

I am thinking perhaps a more generic solution might be appropriate.

What about a FictionalThing type. Probably a sub-type of Intangible, having  createdIn and referencedIn properties both having an expected type of CreativeWork.

To describe something as fictional you would only have to add FictionalThing as an alternativeType, or add a typeOf attribute in RDFa.

That way we would not have to invent a new type for every new thing that we find a fictional instance for.

I agree that there's a more general need here, and that anything could be fictional, but I think just adding another type doesn't really solve the problem. From my understanding of RDFS, if I have a subject with two different types, then I'm really saying that the properties associated with this subject are the intersection of those two types. It actually is sort of the reverse of this based on rdfs:domain and rdfs:range semantics. If I say the following:

schema:actor a rdf:Property
  rdfs:domain schema:Person, schema:FictionalThing .

What I'm saying is that the property is on _both_ schema:Person and schema:FictionalThing. In general, I don't want 'actor' to be a property of either schema:Person or schema:Thing. Dan's introduced the notion of schema:rangeIncludes to address this problem, but the semantics haven't entirely been worked out yet. This issue is, is there a similar schema:typeIncludes that uses something closer to owl:unionOf semantics. But, using schema:additionalType (or @typeof with RDFa) won't do this, AFAIK.

The way I had actually modeled Character for Wikia was closer to the following:

wikia:Character a owl:Class
  rdfs:subClassOf [ a owl:Class; owl:unionOf schema:CreativeWork, schema:Person ] .

This basically allows me to define properties on wikia:Character that done "infect" schema:Person or schema:CreativeWork.

Gregg

~Richard.


On 09/01/2013 02:04, "Jason Douglas" <jasondouglas@google.com<x-msg://335/jasondouglas@google.com> <x-msg://178/jasondouglas@google.com<x-msg://178/jasondouglas@google.com>>  <x-msg://3694/jasondouglas@google.com<x-msg://3694/jasondouglas@google.com> <x-msg://3694/jasondouglas@google.com<x-msg://3694/jasondouglas@google.com>> > > wrote:




On Tue, Jan 8, 2013 at 6:01 PM, Gregg Kellogg <gregg@greggkellogg.net<x-msg://335/gregg@greggkellogg.net> <x-msg://178/gregg@greggkellogg.net<x-msg://178/gregg@greggkellogg.net>>  <x-msg://3694/gregg@greggkellogg.net<x-msg://3694/gregg@greggkellogg.net> <x-msg://3694/gregg@greggkellogg.net<x-msg://3694/gregg@greggkellogg.net>> > > wrote:
On Jan 8, 2013, at 5:43 PM, Jason Douglas <jasondouglas@google.com<x-msg://335/jasondouglas@google.com> <x-msg://178/jasondouglas@google.com<x-msg://178/jasondouglas@google.com>>  <x-msg://3694/jasondouglas@google.com<x-msg://3694/jasondouglas@google.com> <x-msg://3694/jasondouglas@google.com<x-msg://3694/jasondouglas@google.com>> > > wrote:

> On Tue, Jan 8, 2013 at 3:43 PM, Gregg Kellogg <gregg@greggkellogg.net<x-msg://335/gregg@greggkellogg.net> <x-msg://178/gregg@greggkellogg.net<x-msg://178/gregg@greggkellogg.net>>  <x-msg://3694/gregg@greggkellogg.net<x-msg://3694/gregg@greggkellogg.net> <x-msg://3694/gregg@greggkellogg.net<x-msg://3694/gregg@greggkellogg.net>> > > wrote:
> TL:DR: propose adding schema:Character, schema:Location, and schema:FictionalLocation classes, along with a schema:character property.
>
> In the work I'm doing with Wikia, we're using extensions to schema.org<http://schema.org> <http://schema.org<http://schema.org/>>  <http://schema.org<http://schema.org/> <http://schema.org/> >  <http://schema.org<http://schema.org/> <http://schema.org/>  <http://schema.org/> >  to add structure to Wiki content. Wikia hosts hundreds of thousands of wiki's, mostly related to special-interest subjects. Important classes of these include sites about Movies, TV Shows/Series and Video Games.
>
> Cool!!
>
>
> The schema.org<http://schema.org> <http://schema.org<http://schema.org/>>  <http://schema.org<http://schema.org/> <http://schema.org/> >  <http://schema.org<http://schema.org/> <http://schema.org/>  <http://schema.org/> >  vocabulary is pretty useful in doing this but lacks some important properties and types:
>
> Character class: a Character is a subclass of Person, which is intended to represent some fictional character. This could include fictional human characters, such as Sam Spade, as well as non-human characters, such as "The Cat in the Hat". As such, it could also be considered to be a union of schema:CreativeWork and schema:Person. Alternatively, it may simply be a sub-class of Creative Work which simply has some properties in common with Person (birthDate, colleague, gender, ...). Note that books can also have characters.
>
> Character property: An important characteristic of things such as movies, and TV shows is the characters that are in them. For instance, TVEpisode has actor, director, producer and so forth, but no way to indicate the characters that are in the show. Here is where having Character class comes in handy, so that you might have the following:
>
> <http://en.memory-alpha.org/wiki/Where_No_Man_Has_Gone_Before_(episode)>
>   a schema:TVEpisode;
>   schema:name "Where No Man Has Gone Before"@en;
>   schema:partOfTVSeries <http://en.memory-alpha.org/wiki/Star_Trek:_The_Original_Series>;
>   schema:character <http://en.memory-alpha.org/wiki/James_T._Kirk>;
>   schema:actor <http://en.memory-alpha.org/wiki/William_Shatner> .
>
> One downside of this is that when you have multiple characters you lose the ability to correlate the actor to the character.

In the example below, I set schema:actor on the character

ah, missed that.

, but it's obviously loosing some temporal information. More complex modeling can be more accurate, but within the context of a single graph (e.g., for a specific episode), we can probably simplify it that there is a signal actor for each character, or at least a set of actors.

> An alternative, that Freebase uses, is to create a class for Performance, that has the properties of character (expects Character has you defined) and actor (which works the same as actor on today).

This is sort of like an Event, which acts to join various elements together (characters and actors in this case), perhaps along with elements such as scenes and locations. I like the idea of having a Performance class, but there may be times when simpler modeling works.

> Connecting that to TVEpisode could be done with either a new "performance" property or by making actor accept Person or Performance.
>
>
> <http://en.memory-alpha.org/wiki/James_T._Kirk> a schema:Character
>   schema:name "James T. Kirk";
>   schema:birthDate "2233-03-22"^^xsd:date;
>   schema:deathDate "2371"^^xsd:gYear;
>   schema:actor <http://en.memory-alpha.org/wiki/William_Shatner> .
>
> Another class of properties generally useful for works of media classes is Location. A schema:Location class could be a sub-class of schema:Place, intended to describe locations that might not be real, or not at least not having geographic coordinates you can get to using Goole Maps. A location could also be a Fictional Location, such as Middle Earth.
>
> Lastly, many wiki's concern themselves with Video Games, which have quite deep structure. Logically, a Video Game is probably a sub-class of schema:SoftwareApplication. Of course, there are many other things that could be modeled on video games, such as levels, objectives and weapons, but having a concrete class for describing them would be quite useful.
>
> Since the Wikia communities are the real domain experts here do you think there's any chance they could enumerate these in a public schema.org<http://schema.org> <http://schema.org<http://schema.org/>>  <http://schema.org<http://schema.org/> <http://schema.org/> >  <http://schema.org<http://schema.org/> <http://schema.org/>  <http://schema.org/> >  extension proposal?

This post was an attempt to get that started, and the most obvious missing pieces are for Character and Location, along with the character relationship.

Agreed, those are needed incremental additions.


Video Games is a deeper subject, and we can share the work we're doing, but it will be a more substantive proposal. I'd really like to use it to get a conversation with other people who have worked on modeling games. Our current focus has been on linear first-person shooter style games, but much of the work carries over to open-world/parallel mission environments too. We'll come back with more on this in the future.

Yeah, Video Games is where I was hoping for a more "substantive proposal" as you say.  Sounds good.


Gregg

> -jason
>
>
> Gregg Kellogg
> gregg@greggkellogg.net<x-msg://335/gregg@greggkellogg.net> <x-msg://178/gregg@greggkellogg.net<x-msg://178/gregg@greggkellogg.net>>  <x-msg://3694/gregg@greggkellogg.net<x-msg://3694/gregg@greggkellogg.net> <x-msg://3694/gregg@greggkellogg.net<x-msg://3694/gregg@greggkellogg.net>> >
>
>
>
Received on Monday, 14 January 2013 10:25:07 UTC