Topic: Removal of loli tag?

Posted under Tag Alias and Implication Suggestions

Hello, this may have been discussed at some point or another, but has the loli tag been removed? There used to be several dozon pages featuring the tag, but now there doesn't seem to be any. Have I missed something?

Updated by ikdind

Kenjitiger said:
Hello, this may have been discussed at some point or another, but has the loli tag been removed? There used to be several dozon pages featuring the tag, but now there doesn't seem to be any. Have I missed something?

Tag Aliased To Reason User Date
loli (-267) young (9555) don't need gender specificity null0010 7 hours ago

It's been aliased away.

...upon thinking about this, I really disagree with this decision. It can only make searching more difficult. But it can't be put back now without a ton of effort.

Updated by anonymous

Snowy said:
Tag Aliased To Reason User Date
loli (-267) young (9555) don't need gender specificity null0010 7 hours ago

It's been aliased away.

...upon thinking about this, I really disagree with this decision. It can only make searching more difficult. But it can't be put back now without a ton of effort.

It is a little confusing since this is apparently not a furry only site.

*shrug*

Updated by anonymous

Kenjitiger said:
The Loli tag was kind of specific to certain images.

Search for "cub female" tah-freakin'-dah.

And before you say that'll turn up some shota pictures too: You don't have to click on every thumbnail that pops up.

Updated by anonymous

Test-Subject_217601 said:
Search for "cub female" tah-freakin'-dah.

more liek:
young human female

Updated by anonymous

It was decided (after much dramatic debating) that the tags don't serve a purpose here as you can still string together a search to find those types of images.

Updated by anonymous

Rainbow_Slash said:
It was decided (after much dramatic debating) that the tags don't serve a purpose here as you can still string together a search to find those types of images.

While true, that is also true about many accepted tags here lol

Updated by anonymous

Test-Subject_217601 said:
Search for "cub female" tah-freakin'-dah.

And before you say that'll turn up some shota pictures too: You don't have to click on every thumbnail that pops up.

Alias suggestions:
african_golden_cat asian_golden_cat caracal cat cheetah clouded_leopard cougar jaguar leopard lion lynx maine_coon ocelot panther serval snow_leopard tiger -> feline

You don't have to click on every thumbnail that pops up, after all. So why bother with species specificity?

Updated by anonymous

species specificity is not gender specificity. we don't have lioness, vixen, mare, or tigress tags for the very same reason.

Updated by anonymous

The point I'm making is that getting rid of those tags would make searching more difficult for no good reason. What's the good reason for not having gender specificity?

Updated by anonymous

There's no reason for it to exist, so we removed it. Simple as that. Especially when you factor in how often it was misused, which was pretty much always.

Updated by anonymous

ippiki_ookami said:
Especially when you factor in how often it was misused, which was pretty much always.

That's a decent reason.

null0010 said:
Because those tags can easily be replaced by "female lion," "female fox," "female horse," "female tiger," or "female young"

Except for all the false positives that you have to dig through that way. Especially if you're looking for, say, a straight image with a female of a particular species. There's no way to do that with the current tagging system, which I think is one of its main flaws.

Updated by anonymous

How do you find pictures with a young male, and a normal female, or vice-versa?

Young female, regular male (post #155815)

Regular female, young male (post #98024)

And the ones where there's a young male/female and a character of unknown gender? (post #168903, post #50076)

If you search for something like female male young, how do you know which of the two would be loli, or shota?

Is there a way to narrow it down?

Edit: What Snowy said

Updated by anonymous

Like null said, it's like not having vixen, mare, etc. We don't need such excessive tags when you can put female fox together and get the same result. Loli was pretty abused and sometimes not even used, so it was aliased.

Updated by anonymous

Rainbow_Slash said:
Like null said, it's like not having vixen, mare, etc. We don't need such excessive tags when you can put female fox together and get the same result.

Except that you don't get the same result.

Updated by anonymous

Rainbow_Slash said:
It was decided (after much dramatic debating) that the tags don't serve a purpose here as you can still string together a search to find those types of images.

male_renamon versus "male" and "renamon"

Updated by anonymous

Snowy said:
Except that you don't get the same result.

When you search "female fox" you might get a picture of a male fox with a female of a different species, fucking tragedy. :V

Updated by anonymous

Test-Subject_217601 said:
When you search "female fox" you might get a picture of a male fox with a female of a different species, fucking tragedy. :V

Think of every combination

Adrian_Blazevic said:
use pools

Pool? I'll get the 8 ball

Updated by anonymous

Test-Subject_217601 said:
When you search "female fox" you might get a picture of a male fox with a female of a different species, fucking tragedy. :V

TL;DR: It sure sounds cool to make the tagging system smarter than that, but it's actually really, really hard.

Them's the limits of single-word tag systems instead of more hierarchical relation-y systems. You want female to describe fox as a piece of metadata about the fox tag for an image, but the system doesn't store tag information that way.

We don't tag tags, and probably won't until Google or someone can provide an extremely reliable automated image tagger that can not only apply tags to images but tags to the image's tags in a deep hierarchy, and then come up with a way to store and search that information in a time-efficient manner.

It only took 40,000 computers to semi-reliably identify cats in videos, so we probably aren't more than 30 or 40 years from automatic tagging in a laboratory environment. Or 40-50 years to have that in the hands of someone like e621.

Then someone will have to invent the query language for searching these suckers. You can already see the problems some people have with the blacklist, what with separating tags by hitting the enter key, as opposed to only separating them with spaces (this is because the blacklist is actually really, really powerful, which I've discussed at length on several occasions).

So imagine the pain and agony that will be felt when people try to wrap their heads around the syntax for searching for images tagged with "female" which is, itself, also tagged with "fox". Or vice versa, because there's no reason the relationship has to go in that direction (and that's certainly true of many tags and may even be true of all tag-on-tag relationships).

Updated by anonymous

Test-Subject_217601 said:
When you search "female fox" you might get a picture of a male fox with a female of a different species, fucking tragedy. :V

Try to contribute to the conversation :)

Updated by anonymous

ippiki_ookami said:
That's not what pools are for.

Just uncheck "public"? (except less useful)

Updated by anonymous

Test-Subject_217601 said:
When you search "female fox" you might get a picture of a male fox with a female of a different species, fucking tragedy. :V

That's always how it's been because putting together every single combination of gender, species, etc. would make searching really difficult not to mention tagging it. Young female gets the same result as loli. Besides, loli and shota were used correctly one in fifty times.

While we're at it, can we alias colt, filly and foal to young?

Updated by anonymous

AbsebaroKoon said:
Well, *frisks through the pool help page* it says that it can be used for comics AND "groups of posts with a common theme." "This makes pools ideal for SUBJECTIVE TAGS, or for posts that are part of a series."

Oh neat, an outdated help entry in the wiki. You could either go by that, or go by the staff that currently run the site. Up to you.

Updated by anonymous

If we pooled loli then that would be one huge pool, so huge that you would need to narrow it with a search, which is why it would be redundant to group such images.
Aside, themes are things like the Krystal recolor or the "Ew gay" pose. If a single tag can suffice then it doesn't need a pool.

Updated by anonymous

ikdind said:
TL;DR: It sure sounds cool to make the tagging system smarter than that, but it's actually really, really hard.

Them's the limits of single-word tag systems instead of more hierarchical relation-y systems. You want female to describe fox as a piece of metadata about the fox tag for an image, but the system doesn't store tag information that way.

We don't tag tags, and probably won't until Google or someone can provide an extremely reliable automated image tagger that can not only apply tags to images but tags to the image's tags in a deep hierarchy, and then come up with a way to store and search that information in a time-efficient manner.

It only took 40,000 computers to semi-reliably identify cats in videos, so we probably aren't more than 30 or 40 years from automatic tagging in a laboratory environment. Or 40-50 years to have that in the hands of someone like e621.

Then someone will have to invent the query language for searching these suckers. You can already see the problems some people have with the blacklist, what with separating tags by hitting the enter key, as opposed to only separating them with spaces (this is because the blacklist is actually really, really powerful, which I've discussed at length on several occasions).

So imagine the pain and agony that will be felt when people try to wrap their heads around the syntax for searching for images tagged with "female" which is, itself, also tagged with "fox". Or vice versa, because there's no reason the relationship has to go in that direction (and that's certainly true of many tags and may even be true of all tag-on-tag relationships).

It can't be too terribly hard. A query language would be different, but not too terribly hard. Think how CSS selects tags with attributes. It might look like:

mikhaila[hair]
[fox female] [rabbit male]

etc. Where the contents of the brackets means the tags must be decedents of a certain parent, and siblings of each other. If you used a tag without brackets, it would look at all the tags in the entire post, same behavior as there is currently. A nested set model could search a database for this without much additional performance hit.

(Now, if you wanted a really advanced query, like "[straight couple [fox female] [rabbit male]]" where we're describing actions and properties between multiple entities, this might require more than just a hierarchy, but a directed graph. This is a bit harder to properly tag and index.)

Updated by anonymous

idkind said:
We don't tag tags

We don't tag a lot of things that would make searching easier, and it's a real shame

ThenIThought said:
It might look like:

That would be great to see in action

Updated by anonymous

ThenIThought said:
(Now, if you wanted a really advanced query, like "[straight couple [fox female] [rabbit male]]" where we're describing actions and properties between multiple entities, this might require more than just a hierarchy, but a directed graph. This is a bit harder to properly tag and index.)

You've almost realized the complexity of the problem.

The problem is that tag relations cannot be described by a single taxonomy. So any sort of nested set model would have to be instanced for each and every image, and may contain duplicate entries for a given tag which exists unrelated to another instance of that tag. But even then, you have problems.

Sibling relationships aren't necessarily obvious because someone may search solo \[female] but images may be tagged tsampikos \[mikhaila \[female \[solo]]], tsampikos \[mikhaila \[solo female], tsampikos \[solo mikhaila \[female]], solo tsampikos \[mikhaila \[female]], etc., etc. All of these permutations would need to be returned by the search solo [female].

So even if you're given a specific relationship to search for, your search has to be exhaustive among all parents, all parents' siblings, and all children for any given tag relationship, or it's too specific and falls prey to tagging inconsistencies. The alternative is to duplicate entries across all possible relations.

Searching for parent-child relationships takes a lot of time, because you can't know whether a tag is related to another tag until you've looked at all other relations, and you'll still have errors because someone tagged an image "couple female \[fox] male \[rabbit]", and we can no longer distinguish female \[fox] from female \[rabbit] in our search algorithm, because parents' siblings will result in false positive matches. At which point we might as well make all tags global and unrelated again.

Duplicating entries is impractical, even for crowd-sourced tagging. There's just too many potential direct parent-child relationships to exhaust. You'd need an automated tagging system that can intelligently determine these relationships, and turns your nested set model into something closer to a massive directed graph which probably contains close to n-factorial edges, where n is the number of global tags in the image. But then you have a storage requirements problem, even for as few as 10 tags. And 20 tags is potentially 10^18 edges, or roughly 2^60. Think "exabytes".

So the best answer is probably to leave global tags as they are, but throw an artificial neural network at the problem of taking some input syntax defining a set of tags with certain relationships, and letting the ANN figure it out. That technology is decades away, however.

But really, I can make a much shorter argument than all of that: If it were practical, Google Images would already be doing it.

The search syntax itself isn't too hard, but I set a low bar for our "average users" simply because there are people who struggle to grasp how the blacklist works. Then they complain about how it doesn't work when they blacklist "mlp scat gore vore nightmare_fuel" all on one line.

Edit: And seriously, this is a helluva lot of work for gender-specificity (searching for male rabbits or female foxes), which is probably going to amount to 99% of its use.

Updated by anonymous

ippiki_ookami said:
Oh neat, an outdated help entry in the wiki. You could either go by that, or go by the staff that currently run the site. Up to you.

Seems reasonable to go by the help, really. I mean, it's there, there's no sign on it that says "P.S., the mods have changed their minds these days." If it's so outdated, maybe fix it to reflect the current views of the staff that run the site in order to avoid confusion in the future. We can't read your minds, only your wiki entries.

Ik: Dang. I knew anything more complex than our current system would be like...SUPER complex, but I had no idea just how ridiculously in depth it would have to be. Fascinating.

Updated by anonymous

RedOctober said:
Ik: Dang. I knew anything more complex than our current system would be like...SUPER complex, but I had no idea just how ridiculously in depth it would have to be. Fascinating.

Well, my napkin math may err on the side making it look worse than it will be, but I believe the logic is sound and factorials would be the approximate scaling of tag relationships.

As for "decades until ANNs", that was trying to be reasonable as well. Google achieved 15.8% accuracy identifying cats from a self-trained ANN. Though I had the starting numbers wrong - it was 1,000 computers, containing 16,000 cores total. arXiv has the paper they wrote, guess I should have looked for that sooner than going from the news source I had been reading from.

I still think it's reasonable to estimate, using Moore's Law as a guide, that it would take 30-40 years before ANNs could solve this kind of problem in a laboratory environment, and another decade past that before it becomes practical in smaller-scale commercial applications like this one.

Updated by anonymous

  • 1