Discussions on Fictional Universes

Your command is my wish...Done.

Look to the Fictional Object and let it be fruitful and multiply exponentially.

This is my first published type, let me know how to improve it in the discussion here or more appropriately for the Fictional Object (aka Fictional Device) so I can be further edified in Meta-datawrangling.

Thanks Gordon. Fictional Object and Fictional Substance should also have the property "Appears In These Fictional Universes" with expected type Fictional Universe to link them together. Fictional Character has the property "Powers or Abilities" (Character Powers) and I think a similar property would work for Fictional Object as well, with a new type Object Powers set as the expected type. I see "Destroyed By" has the unset expected type of "all", is it work in progress?

Is the fictional object type meant to be used only for individual objects (like The One Ring) or for classes of objects (like lightsabers), as well? The description and properties imply the former, but I think it would be useful to have a place for the general classes as well.

For now Fictional Object is for the singular individual objects. One Ring indeed!

Classes are inevitable I guess.

As for Powers, I have the text only Attributes / Properties property for now, should it be a Topic?

OK, but you should talk to Jeff about why it was set up the way it was -- it's an intricate, powerful piece of machinery.

No. Characters should be co-typed as the appropriate character type ("book character", "tv character", etc.), which will provide the appropriate links. Note that "film character" doesn't exist yet, although there's an open bug to resolve that problem.

OK. Like Brendan said some documentation of the expected content for each property would be nice.

I updated the help for this and the fictional universe type, and took a quick peek at the other types. But speak up if there are things that still aren't clear!

whoops, I didn't mean "this page" I meant the documentation on the "fictional universe" type page

http://www.freebase.com/view/filter?id=/fictional_universe/fictional_universe

I see that several 'universes' were added like Dr. Who. I have removed them, thanks for visiting, noticing and especially commenting.

I'm seeing this a lot (Firefly, The Simpsons, Brazil, etc).  Are the statements above still valid, and should we have separate universes for all of these?

It's a bit of a judgement call, well known (ie fictional works that have inspired a myriad of fan-fiction and/or authorized novelizations of cannon-based/non-canon storylines) tend to have a topic instance already in existence...Firefly might have an existing 'verse but I couldn't find that Brazil has one.

 One could mark the Brazil topic for a split or better yet we can remove the fictional universe type association as it seems to have little value at this time.

 The Simpsons topic is based originally upon a Wikipedia article that is about the TV show, history, cultural impact other media, etc...It almost is a universe topic, just heavily weighted towards the TV series aspect. It can easily be marked as a split, but as is, it could remain a multi-typed topic.

I'm not sure what other solution there would be. Some characters can be found in many domains, but most won't be. Many people (real ones) are really cross-typed.

The real issue here is that the 'anime character' and 'video game character' types are actually useful. They:

a) help autocomplete work correctly
b) Provide a grouping that helps users find what they are looking for -- that is, users can filter search results and ask questions about the group like "find me tv characters who have appeared on more than one show."

If the problems is that the UI gets cluttered with so many cross types, then we should fix the UI.

This issue of proliferating types to deal with different fictional universes was solved in Cyc using "microtheories". (Am I allowed to mention Cyc?) I think this thread is about how this works in Freebase.

The problem is this: I want to fully describe a fictional character who graduated from Harvard in 2000. Do I want to create a whole new type system for 'Education' in my fictional universe? No, so I'll use the existing types. But then someone uses Freebase to search for classmates from Harvard and finds my fictional character, then has to manually sort the results for figure out which results are "real" people. Back to square one with Google where you get too many hits. You might say "filter the results to leave out topics that have type TV Character." But as faye mentions, then you discover that you also need to sort out 'Opera Character', etc. etc. The user needs too much omniscience.

In Cyc, the assertions for where my fictional character went to Harvard are in a separate microtheory and aren't included in the "real world' microtheory used for the classmate search, so fictional characters never show up. What is the Freebase solution to prevent the user from needing to manually sort the results for the correct context/universe?

Hmmm. A modifier and/or flag that could ensure that a property of Fictional Character is expressed/displayed for Science Fiction/Fantasy/Games/Anime,etc Genres and not for anything else (like Soap Opera, Drama, Mystery, Romance/Roman à clef, etc).

Or how about two Fictional Character Types?

One 'Fictional Character - Human' (has ethnicity property)
Second 'Fictional Character - Nonhuman' (has species property)

Lets assume that at some point we will be modeling race and/or ethnicity for both real and fictional characters. I can easily see value is being able to search for (for example) "Black characters in science fiction novels" or "Asian-American characters in TV crime dramas". Given that, I'm inclined to keep this as Character Race.

That said, where else do we have a property that's called two or more things in different contexts? How do we handle it there?

Even after combining the type to 'Character Race or Species' (which makes sense), there should be a separate topic for each universe, since each work of fiction gets to define the properties of that race. 'Human' in Star Wars is different from 'Human' in Lord of the Rings. A 'Human' in Harry Potter can fly. This is not a property of the biological species Human, which is Homo sapiens. Also, in the Left Behind books, humans did not evolve from apes. I think 'Character Race or Species' should not be used as a type for the "real" topics.

Part of the problem here is how to distinguish between all the different "humans" -- there are currently around eight or so different kinds of humans (imported from Wikipedia), and even with this few types we have been getting put in all kinds of fictional universes that they don't belong in. But in principle, if this can be worked out, I see no reason not to have different topics for different fictional varieties of humans; this will probably be confusing for some, and there will probably be some contention over which universes have non-mimetic humans, but I think the community will be able to sort that out.

I'm not sure, however, that "character species" (or whatever we call it) should never be used as a type for real topics -- how else would you describe Lassie's species if not as that of an actual dog (Canis familiaris)? Maybe if we renamed the type to something like "Species of Fictional Characters" it would be better?

Correct me if I'm wrong, but the problem is that Freebase needs a way to put the assertion about a topic in a certain context. Right now, all assertions are assumed to be in one base context which is fine if we're only describing the One-And-Only-True-Consensual-Reality (tm) where Paris is in France, etc. Dealing with fictional universes breaks that assumption. And that's why we're having the problem of duplicating every type so that we can separate out the properties for the other contexts.

Wouldn't it be neat: I assert that Lassie is a Dog, but I put this assertion in the context of the Lassie Universe. It's OK that the assertion is in the "real" Dog topic because most of the other assertions (such as "Dog is a species called Canis lupus familiaris") are in the context of base assertions.

Right now, a search returns assertions that are in the context of base assertions, but everything in Freebase is now in this context, which is the problem. Wouldn't it be neat: I change the context for my search to allow assertions in the context of the Lassie Universe wherever they come from. For example, in the Lassie Universe, the city of Yorkshire is saved from disaster by a dog. Want to assert this "fact" to the Yorkshire topic? Go right ahead, just put the assertion in the context of the Lassie Universe, so that it doesn't show up in the base context when people are researching city disasters.

I realize this is a fundamental design topic in Freebase, but the issue of Fictional Universes brings it out. Summary: we don't need to put topics in a special type, we need to put assertions in a special context.

That's exactly the sort of thing we had in mind for this property.

Instances of real people (and deities, for that matter) that appear in fictional works should be of the type "Fictional Character" but not "person". Instead, the person they are based on should be entered in the "based on property". This will create a link back from the actual person to any fictional representations thereof, and allows for accurate fictional information to be recorded without muddying the real-world waters.

Thanks for the quick reply. I fixed the LBJ example. I got lead down that path because many other topics are already typed Person and Fictional Character, including Julius Caesar, Howard Hughes, William Wilberforce, etc. Are there near or long term plans for queries to automate checking when topics conflict with the contract defined by the type (as Freebase grows to millions of topics)?

Yes! Actually we have an "Incompatible Types" type (http://www.freebase.com/view/filter?id=/dataworld/incompatible_types) which lets you store collections of types that cannot logically coexist on a topic. Exactly how we'll enforce these assertions remains to be seen, but certainly you would be right to remove Fictional Character from any topic who is really a Person.

Thanks Alexander. Along the same lines, can the real species Dog (Canis lupus familiaris) also be the character species for the fictional character Dogbert? Or, for that matter, can the real species Human be the species for Harry Potter (who flies and does other non-human things)?

Well... this is a matter of some debate currently. The philosophically ideal solution is to create new topics for every fictional rendition (or interpretation or opinion...) of anything, but this makes for complex schemas and tends to make querying and browsing more difficult, and I'm not sure what it really gives us that's useful in most cases.

I'd be inclined to say that Dogbert's species is plain old Dog, even though real dogs can't talk, because, as it is, we aren't actually asserting that "dogs can talk" when we make that link. Real dogs also don't save children from wells out of their own moral imperative every day of the week, but it seems very pedantic to create a new topic just for "Canines in the Lassie Universe."

The only way I can think of for deciding how to draw these lines is to consider what is the most useful and practical. We also have to consider what pattern of data people will actually input, and I expect that most users wouldn't think to create a new topic called Dog, link it to the Dilbert Universe and assert that it's "based on" the Dog of reality, but rather link directly to Dog because they intuitively know that we never talk about the fictional species to which Dogbert would belong in a philosophically ideal world -- he's one of a kind, and any interesting things about him will be connected to the Dogbert topic itself. Wookies, on the other hand, though also "based on" the canine, really are thought of as a distinct fictional species -- there are many instances of them, and Star Wars geeks would happily discuss all of the species' various fictional social and biological features.

In our framework, it's trivial to look for "animals that are also character species" to see exactly where these cases appear, and you could then (programmatically, if needed), cleanly split all of these cases into two. In other words, I don't see any pollution of either the animal or the character species namespaces with this type of convergence right now, and if we ever want to be a little stricter about it, we can easily migrate there. For now, I don't see how a topic just to represent how Dogbert is dissimilar to real-world dogs would help anybody answer any questions, but I can imagine how it would make things a little more confusing.

Anyway, everything is in flux, and that's sort of the point. We aren't claiming to have an end-all solution to the philosophy of knowledge, we're providing a powerful and flexible framework that can gracefully evolve to accommodate all sorts of different ways of looking at the world. These models will always be imperfect; they wouldn't be models if they didn't carry assumptions. As Freebase evolves, we'll all have a better idea of what modeling patterns work best for solving the problems the community cares about.

Alexander, you say "The philosophically ideal solution is to create new topics for every fictional rendition (or interpretation or opinion...) of anything". But there is another technical, and much cleaner, solution used by other knowledge systems. Instead of creating new topics for everything, allow the user to specify a context for a particular assertion on a given topic. Right now on the Dog page, if you click the drop-down for Dogbert it has 'View', 'Edit' and 'Remove'. Imagine it also had 'Context', and you could specify the 'Dilbert Universe' as the context, right on the assertion itself, without creating a new topic. If Freebase had this feature, arguably we wouldn't need the explosion of duplicate types in the Fictional Universes domain at all. (I've posted this question already several times. I sure hope someone answers this time....)

Actually, I do think that the contextual assertion approach you've been suggesting could lead to a pretty solid, generic solution to some of these sorts of problems. It seems like it would also let you answer the "when was X a valid assertion?" question, as is being discussed on the Domains and Types page, to deal with historical countries and that sort of thing but without duplicating schema, as you said. It does add complexity to the core representation system, though, and would require a lot of changes to pretty much every component of Freebase, so it's really a question of whether this is such a common pattern for the things we want to represent that it's worth integrating. I'm only speaking for myself here, but I'd definitely like to see this idea discussed more... maybe we should think through some really specific, compelling examples of what this would enable.

Thanks for the reply. I can easily come up with lots of examples because:
A. Every assertion has a source and set of assumptions. (Even well-meaning people will come to different conclusions)
B. Every assertion is true at some point in time, and can change later (because the world changes).
So, every assertion is in the context of a source and a time. Freebase can only skirt around this by assuming every assertion has only one universally greed source, and that circumstances will never change. This hampers the knowledge that Freebase can capture. Examples:

1. What is the population of India? Instead of making a special composite type for population, the simple assertion of 1.12 billion should be marked in the context of the source of the information and the time at which it was true.

2. In what viral group is this virus? That depends on what you assume is its genetic function. Reasonable people can come to different conclusions. Allow assertions for both group A and group B, but make the context of each assertion visible.

3. Who are the children of Thomas Jefferson? There are different claims of paternity. Instead of making a different topic for each "fictional" version of Thomas Jefferson, allow the assertion on the main Thomas Jefferson topic and cite the source of the claim on the assertion. (Currently, Freebase tragically obliterates all the footnotes in Wikipedia articles that cite sources, because a citation is a property of an assertion, not a topic. Freebase needs this mechanism.)

4. Who are graduates of Georgetown University? Apparently, one of them is Zoey Bartlet. Why? Because in the explosion of parallel types for Fictional Universes, no one has created a "Fictional Education" type, so the fictional character is made to be a real Person, so that it can be asserted that she went to the real Georgetown University. Is the solution to make a "fictional" Georgetown University and Education type to go with it? No. The assertion that Zoey Bartlet went to Georgetown University is in the context of the West Wing TV show.

5. How much does Oprah Winfrey weigh? Right now, the Weight property is a single value with no source or timestamp. But like all properties (height, religion, even gender nowadays), they change. Even the property Place of Birth can often legitimately be disputed. What does it mean if every month people go in and change the Weight property? Are they correcting a previously incorrect assertion, or are they correctly asserting that at the moment, according to this source this is true? (The Page History on a topic is not the place to track changes over time because it doesn't distinguish between actual changes to a value, and people just trying to correct mistakes.)

I'll stop there, but the list goes on.

I'll copy the example from my amazon.com, it was Kilgore Trout. But this is an attribution to a fictional character created by Kurt Vonnegut. It was really written by Philip José Farmer. How to handle this? Do we need a 'Fictional Book' type so that there can be a topic for the fictional version of the book? No. It's easy if we put the assertion that Kilgore Trout wrote the book in the context of Kurt Vonnegut's fictional universe. This requires Freebase to show properties on assertions, not just on topics.

(Try again...) I'll copy the example from my post below:

6. Who wrote the book Venus on the Half Shell? According to amazon.com, it was Kilgore Trout. But this is an attribution to a fictional character created by Kurt Vonnegut. It was really written by Philip José Farmer. How to handle this? Do we need a 'Fictional Book' type so that there can be a topic for the fictional version of the book? No. It's easy if we put the assertion that Kilgore Trout wrote the book in the context of Kurt Vonnegut's fictional universe. This requires Freebase to show properties on assertions, not just on topics.

We thought about this, and even had it modeled, but decided ultimately that the decrease in usability was greater than the value of the increased granularity of the model. One way that it works better than fictional character is that there's only one outgoing property from the actual location to the fictional works it appears in, whereas with characters, there was the possibility of an extremely large number of extra properties (books appears in, films appears in, operas appears in, comic strips appears in, etc.), some of which we properties specific to a certain domain (the type "comic book character" has several properties beyond just the works it appears in). With location, we were able to get around this by creating a separate type ("work of fiction") to use as a cotype when we want to indicate settings.

This is a complex issue, all right, and I'm sure that Ali G isn't the only such instance. I do think that this is a similar problem to the use of pseudonyms, although with an extra layer of complexity. It's valuable to know under what name a work was produced, both for cataloging-type applications, as well as if you just want to find a book on a shelf somewhere and need to know the name it was published under. But it's also important for people who don't know that a name is a pseudonym to be able to find works by that person published under other names. (Also, sometimes pseudonymous works are reprinted under the author's real name, and we need to be able to assert both that the work has been published under different names, and that it is the same work.) We haven't come remotely close to solving this issue, so any suggestions would be welcome.

See my . I'll add a similar example along these lines that could be solved by making the context of an assertion visible.

6. Who wrote the book Venus on the Half Shell? According to amazon.com, it was Kilgore Trout. But attribution is to a fictional character created by Kurt Vonnegut. It was really written by Philip José Farmer. How to handle this? Do we need a 'Fictional Book' type so that there can be a topic for the fictional version of the book? No. It's easy if we put the assertion that Kilgore Trout wrote the book in the context of Kurt Vonnegut's fictional universe. This requires Freebase to show properties on assertions, not just on topics.

So if I am following you correctly we should have a three John Nash's in the Freebase.

Real John Nash
Fictional John Nash (based on the real John Nash)
Performance John Nash (Performance by Russell Crowe, based on Fictional John Nash)

Um, Faye? There is a "based on" property on "fictional character" for just such a purpose. We tried having it on fictional setting, but it turns out to be much more confusing, and even with documentation lead to more bad data than good.

Hmm, the property "Based on" expects a "Person Or Being In Fiction" in the fictional universe. Is the correct usage, then, to cotype a real Person in the People domain as a "Person Or Being in Fiction", which is then linked to the fictional character(s)?

I don't think there's a standard for this at all. User duck1123 has created a few types that can equate alter egos to each other. See his types Primary Identity and Alternate Persona.

Personally (and I stress personally because it's by no means any kind of standard), I think it makes sense to use the better known name as the name of a topic. I apply this principle to real people as well as fictional characters. Additional names can always be added as topic aliases ("also known as").

There's a practical side to it. People are more likely to talk about (and search for) "Spider-Man", "Mark Twain", and "Sting" by these names, than by the lesser known "Peter Parker", " Samuel Clemens", and "Gordon Sumner". There's less information associated with the lesser known names. As an unfair example, I just searched for music on Amazon using "Gordon Sumner", and not surprisingly none of Sting's well-known albums showed up.

IMO, all egos deserve their own topic pages. How about something like an identity type with an alter ego property that links multiple identities? "Primary identity" would then be determined through other properties and the purpose of the declaration. This would allow Batman to have properties that are distinct from those of Bruce Wayne but they are still linked through an informative property.

Batman is a good case. Many of his story lines focus on whether Batman or Bruce Wayne is the primary identity. Both could be considered "secret identities". One is a caped crusader, the other is a public figure and never the 'tween shall meet (except in the Bat Cave, which is neutral ground). 

I would not attribute Batman as the owner of Wayne Industries. Whether or not Bruce Wayne could be considered a superhero or detective is a matter of debate.

On the flip side, citizens of Gotham are curious about the secret identity of Batman and they don't know to be curious about a secret identity of Bruce Wayne. 

Sting is an interesting case.. I don't really see Sting as an alter ego of Gordon Sumner. Sting is just a pseudonym/alias. Compare that situation to the acting roles that Sting/Sumner has played in film and television.

There are some superheroes that have multiple “primary identities”, and some “primary identities” have donned the costumes of multiple superheroes. Does the model handle those cases? It sort of does but not that well. See Batman vs Batman vs. Batman. Maybe addressing these situations could help refine the model.

Correct me if I'm wrong, but isn't "an identity type with an alter ego property that links multiple identities" essentially what duck1123's types do?

One "primary identity" with multiple superhero identities seems pretty straightforward: see Barbara Gordon. But the more complicated issue is when one superhero identity is assumed by multiple people. Is the Batman of "Batman Beyond" (Bruce Wayne's son in that continuity) really the same as the original Batman? And is either the same as the Batman persona assumed by Jean-Paul Valley while Batman/Wayne was temporarily out of commission?

The question of alternate-Earth Batmen seems more straightforward -- different characters with different alternate identities. So the alternate Batmen would have alternate identies of Earth-One Bruce Wayne and Earth-Two Bruce Wayne as appropriate.

I agree that pseudonyms are different than this. I think Faye's suggestion was simply that common use should determine the topic name, which I think is the right answer if we use the simpler model. But I'm inclined toward the alteregos model if we think it can work.

duck1123's types sort of do this but their names have are somewhat subjective (e.g., "primary") and therefore not as useful as they could be. Also see the third linked Batman topic. That redundant topic has an "Alternate persona" type that contains info that should really be linked to the first Batman topic but can't because that one is attributed with Bruce Wayne characteristics.

I prefer the way that Barbara Gordon is described compared to the way that Batman is described. For the most part, Barbara Gordon is a character that is distinct from Catwoman and Oracle but linked to them through alternate identies.  I would just rename "Primary identity" to "Identity". 

Primary identity is important in terms of who is most commonly associated with the other identity. There should be a way to distinquish that Bruce Wayne is the primary identity of Batman and not Jean-Paul Valley. However, that info  may be better pulled from the data in terms of # of story lines or similar value.  For example, who is the primary identity of Witchblade throughout history?