This thread is taken from topic #43680

Because Danbooru-style tagging does not comprehensively capture relationships between characters, new tags have come up like male_penetrating_human and female_rimming_male that can end up leading to a lot of tagging bloat. Extremely specific tags have come up like clothed_male_nude_andromorph, which just adds to the list of tags to maintain compared to just using clothed_male nude_andromorph.

Tags from 2017 like anthro_penetrating_feral have implications anthro_penetrating and feral_penetrated. Following in this direction and replacing x_verbing_y tags with x_verbing and y_verbed would be a significant tagging project/bulk update request. However, the resulting 26-ish tags will be easier to manage and keep track of compared to handling implications for 25 x_verbing_y tags for anthros/ferals/etc and 64 x_verbing_y tags for male/gynomorph/ambiguous/etc for every verb.

A second point is that tags that cross gender/form distinctions (like male_penetrating_human) should be invalidated because they further exacerbate this problem.

Some thoughts from the other thread:

Obviously [these tags] may not return the exact results people want, but that's because I don't think there's a perfect solution with the danbooru tagging system. There's a tradeoff between ease of tagging, manageability of tags, and specificity of tags. I think specificity is less important than making searching easy for the average user and tagging easy for the average uploader. Making taggers' lives easier makes searchers' lives easier. Encouraging taggers/searchers to use extremely specific tags through the wiki may be nice for a few dedicated fetish taggers, but makes it harder for everyone else to use the site.

Relevant: topic #42656

That's a good thread. My intent is for this thread to be about proactively curbing the number of tags that go in that direction now that <gender>_<form> is more commonly used.

The original thread is pretty long and involved, so much encouraged for anyone looking for context to give it a read. I can't really hope to summarize individual points like I would've liked to.

thegreatwolfgang said: (forum #400827)
Say we invalidate tags like male_rimming_female, people will still continue to tag male_rimming and female_rimmed because they follow the precedent of *_penetrating & *_penetrated.

I don't understand this point. If we invalidate male_rimming_female, then wouldn't we also invalidate male_rimming and female_rimmed simultaneously? Under what circumstances would we wish to invalidate only the combo tag, if we don't want people to also be tagging the decomposed tags?

I do think this suggestion is generally harder to implement without support for aliasing one entered tag to multiple target tags—alias male_penetrating_female to both male_penetrating and female_penetrated simultaneously, so familiar taggers (or just taggers who want a shorthand!) can tag the combo and get the decomposed tags applied (with the combo skipped out). AFAIK that's not a feature yet. Would love to be corrected on this though!

abadbird said: (forum #400843)
Decomposition can only produce less powerful searching, assuming that the proper implications are in place. Exact result = powerful search. Mix-and-match tags = larger pile of less relevant results.

I saw a few people basically agreeing with this point / family of points, but I feel like the definition (which abadbird really conveniently identified) is a bit of a limited view on what searching is for.

"Exact result = powerful search." No, exact result = exact result. It's great if you are looking for exact results—but that's not the only thing people use searching for. IMO there is a lot of power in deliberately reducing the specificity of your search. I don't believe it's possible to find e.g. female_penetrating feral_penetrating right now—here we're composing two tags for the penetrator, and don't care about who is penetrated.

(AFAIK in principle, wildcard searching should accomplish what I'm looking for, but it seems to be a bit quirky, or not support combination? Like, female_penetrating_* *_penetrating_female should be equal to female_penetrating_female, but it clearly isn't. female_penetrating_* feral_penetrating_* is also a no-go.)

It is definitely true that decomposed tags are inherently less good at finding specific compositions, because danbooru tags aren't suited this way. We have combination tags because we want to find specific combinations. That's OK, but it has obvious problems with scaling (see main post here, plus forum #400856) so it needs to have a well-defined curtail point—IMO it should at least have commonly agreed guidelines that apply across different kinds of combination-tag circumstances that come up.

With decomposed tags you can't search for feral_penetrating_feral anthro_penetrating_anthro (the closest is feral_penetrating feral_penetrated anthro_penetrating anthro_penetrated, which can't identify the pairings)... but across 17k and 113k posts respectively, there are only 32 posts with overlap.

And if you're irritated that sometimes your anthro_penetrating feral_penetrated includes, say, stuff that shows in anthro_penetrating_feral anthro_penetrating_anthro... well, there is a very obvious tag to use for exclusion: anthro_penetrating feral_penetrated -anthro_penetrated.

You can make similar fine-grain adjustments of other sorts, too: you thought you were looking for anthro_penetrating feral_penetrated but turns out you don't want male_penetrated in the end, and the appropriate exclusion is right there. No need to groan and go anthro_penetrating_feral -male_penetrating_male -female_penetrating_male -intersex_penetrating_male -gynomorph_penetrating_male and on and on...

IMO danbooru-style tagging lends well to adapting your searches by using decomposed tags, and lends poorly to finding particular combinations of tags. Maybe there are good justifications for certain combined tags, but IMO, in the context of searching, they're more stifling and annoying than useful. (I do think we avoid 99.9% of issues by having the combination tags implicate their decompositions, but if that's the case, we still need to identify rules or guidelines for curtailing new-and-existing combination tags.)

everybodyknowsuradog said:
And if you're irritated that sometimes your anthro_penetrating feral_penetrated includes, say, stuff that shows in anthro_penetrating_feral anthro_penetrating_anthro... well, there is a very obvious tag to use for exclusion:

I generally agree and wanted to highlight this point--because searching permits tag exclusion (and, as https://github.com/e621ng/e621ng/pull/625 shows, potentially arbitrary boolean combinations of tags) users have plenty of ways to find exact posts if they want them. Searching will be fine with either decomposed or non-decomposed tags. The focus should move towards having a relatively small, well-organized set of tags that taggers know how to use, maintain, and extend.

everybodyknowsuradog said:
"Exact result = powerful search." No, exact result = exact result.
It's great if you are looking for exact results—but that's not the only thing people use searching for. IMO there is a lot of power in deliberately reducing the specificity of your search.

Part of the reason that being less specific can be better is because as a rule tag coverage is incomplete (so the more terms you require, the more false negatives mount up)

I don't believe it's possible to find e.g. female_penetrating feral_penetrating right now—here we're composing two tags for the penetrator, and don't care about who is penetrated.

(AFAIK in principle, wildcard searching should accomplish what I'm looking for, but it seems to be a bit quirky, or not support combination? Like, female_penetrating_* *_penetrating_female should be equal to female_penetrating_female, but it clearly isn't.

Why?
Shouldn't it be equivalent to one of the following (terms populated from tag searches; note that there are more *_penetrating_female than female_penetrating_* tags, this isn't a mistake)

• female_penetrating_ambiguous female_penetrating_andromorph female_penetrating_anthro female_penetrating_female female_penetrating_feral female_penetrating_gynomorph female_penetrating_herm female_penetrating_human female_penetrating_humanoid female_penetrating_intersex female_penetrating_male female_penetrating_maleherm ambiguous_penetrating_female andromorph_penetrating_female animal_penetrating_female anthro_penetrating_female female_penetrating_female feral_penetrating_female gynomorph_penetrating_female herm_penetrating_female human_penetrating_female humanoid_penetrating_female intersex_penetrating_female male_penetrating_female maleherm_penetrating_female taur_penetrating_female tentacles_penetrating_female (the literal expansion of both of the wildcards) -- Clearly it isn't this (0 posts)... or,
• The union of ~female_penetrating_ambiguous ~female_penetrating_andromorph ~female_penetrating_anthro ~female_penetrating_female ~female_penetrating_feral ~female_penetrating_gynomorph ~female_penetrating_herm ~female_penetrating_human ~female_penetrating_humanoid ~female_penetrating_intersex ~female_penetrating_male ~female_penetrating_maleherm (4.6k posts) and ~ambiguous_penetrating_female ~andromorph_penetrating_female ~animal_penetrating_female ~anthro_penetrating_female ~female_penetrating_female ~feral_penetrating_female ~gynomorph_penetrating_female ~herm_penetrating_female ~human_penetrating_female ~humanoid_penetrating_female ~intersex_penetrating_female ~male_penetrating_female ~maleherm_penetrating_female ~taur_penetrating_female ~tentacles_penetrating_female (maximum results ie. >187500 posts)-- yeah, based on returned results and post counts, it's this.

I guess you were expecting it to do some kind of algebraic inference because those two sets of tags are related, but, well, aside from it factually not doing that kind of thing, if we changed it, I'm not sure what principle we could appeal to to explain it working the way you seem to expect it to.

EDIT: actually, one of the terms is probably just discarded; see this quote from the cheatsheet:

african_*
Search for posts with any tag that starts with african_, such as african_wild_dog or african_golden_cat. May not work well when combined with other syntaxes.
Limit one wildcard per search.

ie. yes, it does NOT support combination . So female_penetrating_* *_penetrating_female == *_penetrating_female

The focus should move towards having a relatively small, well-organized set of tags that taggers know how to use, maintain, and extend.

This. A strict, minimalist ontology promotes fluency and efficiency. Loose ontology promotes having a 'box of tricks' but little understanding (because making a large ontology consistent and predictable is an infeasible amount of work).

savageorange said:
I guess you were expecting it to do some kind of algebraic inference because those two sets of tags are related, but, well, aside from it factually not doing that kind of thing, if we changed it, I'm not sure what principle we could appeal to to explain it working the way you seem to expect it to.

EDIT: actually, one of the terms is probably just discarded; see this quote from the cheatsheet:
ie. yes, it does NOT support combination . So female_penetrating_* *_penetrating_female == *_penetrating_female

Thanks, good point. I shouldn't have indicated (and thought!) it's equal to female_penetrating_female — really it would ideally be equal to female_penetrating female_penetrated, and this would be equal to the "union" you described. (Wouldn't it be an intersection though? union = A + B, here we require both tags so it's intersection.)

xerxes_i said:
I generally agree and wanted to highlight this point--because searching permits tag exclusion (and, as https://github.com/e621ng/e621ng/pull/625 shows, potentially arbitrary boolean combinations of tags) users have plenty of ways to find exact posts if they want them. Searching will be fine with either decomposed or non-decomposed tags. The focus should move towards having a relatively small, well-organized set of tags that taggers know how to use, maintain, and extend.

:+1: Also, I'm really happy to hear PR #625 exists. It's so cool to see that's in (as far as I can tell) very active development too—no clue how often big PRs are merged from outside developers, but this is still exciting.

savageorange said:
This. A strict, minimalist ontology promotes fluency and efficiency. Loose ontology promotes having a 'box of tricks' but little understanding (because making a large ontology consistent and predictable is an infeasible amount of work).

Yeah, this too. Minimalism is just lower maintenance, in every sense of the word:

• Expansion maintenance—no explosions of new, rarely-used tags
• Basically the greatest strength of decomposition, maybe second to nicer searching
• Decomposition doesn't get you comprehensive minimalism on its own, it's just a more minimal form that inherently avoids two awkward kinds of tag expansion (gender_verb_gender, gender_verb_form)
• More comprehensive minimalism also needs to identify what kinds of pairs of actor-verb are acceptable: gender_verb is good, form_verb is good, but what about the various role_verb sorts of tags, etc? These remain compatible with x_verb / y_verbed tags, so need addressing beyond just avoiding x_verb_y.

• Cognitive tagging maintenance—tags are "lower weight" to remember exist, e.g. you're thinking about the penetrator, not about the relationship—IMO "male penetrating" is a cognitive cue as much for "anthro penetrating" (different attribute, same subject) as it is for "female penetrated" (same attribute, different subject)—lower cognitive load = better tagging across the board, hopefully!
• For comparison, going from "male penetrating female" to "anthro penetrating feral" only leaves you with consistency in the form, i.e. subject verbing object. IMO it's a more abstract leap, mostly grammatical as compared to social concepts of who's doing the thing, who's having the thing done to them. You can form a cognitive connection but it's less natural.
• IMO wiki pages need to serve as an effective reference for taggers: we need a tag group for actions in general, and it needs to clearly identify which kinds of pairs (in a minimal selection) are welcomed. E.g. yes we have "gender sex-act" tags -> "male_rimming", "female_penetrated"; no we don't have "role non-sex-act" tags, don't tag "teacher_holding", "prey_held", etc.

• Cognitive searching maintenance—as we've gone over there are a lot of ways decomposition makes searching easier, but to be specific, IMO encouraging/empowering the use of exclusions necessarily makes users more likely to perform more meaningful, thoughtful searches, which broadly improves interest in tags which are effective for searching.
• Searching is the crux of what tags are good for, after all: it makes sense that tags which are good for users searching will automatically heighten the broadly perceived and expected value of tags in general.
• Easier searching = better searching; the same case for cognitive leap from "male penetrating" to either "anthro penetrating" or "female penetrated" applies here, and easier finer-grained searching encourages anyone who both tags and searches (hopefully a lot of people!) to add tags more consistently!
• The wiki should not be written with only taggers in mind, i.e. it should serve as a very handy reference for users who are exploring, too. This obviously means we need to make sure the overall reference of "what kinds of action tags can you tag and search for?" is very easy to discover (surface it in wiki pages for tags like male_penetrating, anthro_rimmed, etc). It also means the reference should be accessible and welcoming, not written in a harsh tone like "all you fools better stop adding new clothed_male_nude_andromorph tags, crivens!!"
• Ways a wiki page helps a searcher: offering example searches, both featuring exclusion and not featuring exclusion
• Ways a wiki page doesn't help a searcher: espousing at length on how great x_verb y_verbed tags are—the searcher is already stuck with whichever form we end up using!

—

Also, if we want tag decomposition, which form makes more sense? "x_verbs y_verbed" or "x_verbing verbing_y"? IMO the former is nicer but both forms include a kind of symmetry:

• female_penetrated female_penetrating -> likenesses between actor/actee
• female_penetrated female_held -> multiple actions for same character
• penetrating_female penetrating_male -> multiple actees for same action
• penetrating_female penetrating_feral -> multiple attributes for same action
• All four above show related tags which would be grouped together in e6's alphabetical tag list!

Iit's probably arbitrary which symmetry you prefer. Maybe "x_verbing y_verbed" feels more "natural", and I do wonder if there's precedent for that form (or the other) in existing tags, but there's a real benefit to "x_verbing verbing_y" too: no confusion over reversing male_holding and female_held, nor in decomposing male_holding_female to those.

everybodyknowsuradog said:
Thanks, good point. I shouldn't have indicated (and thought!) it's equal to female_penetrating_female — really it would ideally be equal to female_penetrating female_penetrated, and this would be equal to the "union" you described. (Wouldn't it be an intersection though? union = A + B, here we require both tags so it's intersection.)

Intersection would probably be the most intuitive way for it to work if multiple-wildcard-terms support was actually implemented; but clearly that wasn't the outcome you were getting (that would be kind of like female_penetrating_female but with also a bunch of other posts tagged group that had both a female_penetrating_[something] and a [something]_penetrating_female). So union was just the conclusion I fell back to (incorrectly)

The bulk update request #7725 is pending approval.

Reason: There are a lot of these, and hopefully I did this BUR right. These are only the tags where the anthro is the active partner. The rest of the form tags and all of the gender tags still need to be covered, but I wanted to put this one up for a vote now. I'm not super happy with tags like anthro_top, and I wonder if tags like anthro_birthing are too specific, but I wanted to first get all of these x_verb_y tags out of the way. I also did not mess with the anthro_penetrating_* tags, even when the * was a gender and not a form.

The bulk update request #7726 is pending approval.

Reason: part 2

The bulk update request #7727 is pending approval.

remove implication anthro_dominating_humanoid (31) -> humanoid_on_anthro (13182)
remove implication anthro_dominating_anthro (253) -> anthro_on_anthro (228637)
remove implication anthro_dominating_human (1730) -> human_on_anthro (86347)
remove implication anthro_dominating_feral (0) -> anthro_on_feral (31909)

Reason: run this before part 2

These could alternatively be aliased like anthro_toying_feral -> anthro_on_feral. Either way ends up losing something if it's not otherwise tagged, what the second character being interacted with is (e.g. anthro_on_feral to indicate bestiality), or the type of interaction (e.g. anthro_raping to indicate rape), so it could go either way I suppose.

Yeah, I don't have strong feelings one way or the other regarding how to alias these. If someone presents a compelling argument I'm happy to change.

The bulk update request #7737 is pending approval.

Reason: some male tags. updated male_carrying_other to male_carrying_another. i don't really care what happens to male_carrying_another, but i do think the double tags have to go

The bulk update request #7738 is pending approval.

Reason: part 2

The bulk update request #7739 is pending approval.

create alias male_dominating_ambiguous (12) -> dominant_male (52492)
create alias male_dominating_herm (7) -> dominant_male (52492)
create alias male_dominating_gynomorph (40) -> dominant_male (52492)
create alias male_dominating_anthro (31) -> dominant_male (52492)
create alias male_dominating_female (1115) -> dominant_male (52492)
create alias male_dominating_andromorph (34) -> dominant_male (52492)
create alias male_dominating_male (560) -> dominant_male (52492)
create alias male_fellating_male (30) -> male_fellating (17)
create alias male_fellating_anthro (9) -> male_fellating (17)
create alias male_licking_female (2) -> male_licking_other (0)

Reason: part 3. before i do more, i think its better to gauge interest first

xerxes_i said:
create alias male_carrying_female (1) -> male_carrying_other (0)
create alias male_carrying_male (1) -> male_carrying_other (0)

I'd prefer something more like male_carrying_another (if not aliased to carrying_another).

You guys are still missing a few things... or choosing to ignore them.

For people who only want the decomposed halves, the utility of keeping the composed whole is greatly increasing the likelihood of both halves getting tagged on the same post. One tag, with all expected implications, is much more likely to get tagged than both of a pair of related tags. Taggers are forgetful, lazy, biased, etc and will not tag both halves nearly as often than with a composed whole, especially once that whole gets fully implicated. Your theorized decomposed ideal is blinding you to this reality. If we were all better taggers who lived up to that ideal, then we would not rely on implications at all.

Stop looking at this as a "problem to solve" with a universal solution. These tags are unwanted in any form. Nothing about them is good. What they identify does not matter. The decomposed halves are just as unwanted as the composed whole. This isn't about designing a system. This is using experience to make value judgements. Suggesting a one-size-fits-all solution betrays a lack of experience and general disregard for the content.

Go tag by tag and ask yourself if they or their decomposed halves accomplish a useful, needed, or interesting purpose. Is the content diverse enough such that specialization creates unique piles or is everything kinda the same? From that question, I can easily make a sweeping generalization that leads to reining in a chunk of the tag explosion: anything involving ferals is most interesting and diverse but those same things involving anthros, humanoids, and humans are mostly the same. Thus, as a compromise to full invalidation, anthro, humanoid, and human combo tags could be combined into *biped* tags with minimal loss (not perfect, but close). Arguably, taurs are too rare to need specialized tags.

I kinda feel like it might be better to alias most of these to <verbing>_partner or <verbing>_another since those tend to be the most undertagged of the lot and, depending on perspective, those are kind of the closest adjacent tags.

abadbird said:
You guys are still missing a few things... or choosing to ignore them.

I like (and agree with) your angle and happily admit a lack of experience working with most of these tags. I don't know where you're getting that anyone is thinking this is a "problem to solve" with a one-size-fits-all solution, though. Like, quoting part of my latest post:

everybodyknowsuradog said:
More comprehensive minimalism also needs to identify what kinds of pairs of actor-verb are acceptable: gender_verb is good, form_verb is good, but what about the various role_verb sorts of tags, etc? These remain compatible with x_verb / y_verbed tags, so need addressing beyond just avoiding x_verb_y.

I explicitly acknowledged that addressing the syntax form doesn't tell you anything about whether or not the tags are desirable, and that needs to be decided separately. Doesn't matter if we go x_verbing, y_verbed or x_verbing, verbing_y or x_verbing_y, we're fundamentally still pairing the same kind of subject with the same kind of action. The only thing getting rid of x_verbing_y does is make geneder_verbing_form impossible, and that's something we can rule out without deleting x_verbing_y, anyway.

I don't know if you're missing that because I'm talking about a lot of things besides "we need to address this beyond just the form". Yes, the form is what's the most interesting to me, and I do think we should decompose x_verbing_y, even if we keep x_verbing_y as an implication-shortcut where appropriate (yes! we should do that!). I also think we need to figure out what tags are good and why. But I don't have much to say on that because I'm inexperienced.

abadbird said:
Go tag by tag and ask yourself if they or their decomposed halves accomplish a useful, needed, or interesting purpose. Is the content diverse enough such that specialization creates unique piles or is everything kinda the same?

And yeah, this seems like a good guideline for addressing things so that we have a minimum of tags, which, you know...

xerxes_i said:
Searching will be fine with either decomposed or non-decomposed tags. The focus should move towards having a relatively small, well-organized set of tags that taggers know how to use, maintain, and extend.

savageorange said:
This. A strict, minimalist ontology promotes fluency and efficiency. Loose ontology promotes having a 'box of tricks' but little understanding (because making a large ontology consistent and predictable is an infeasible amount of work).

...is what people, in this thread, are saying.

abadbird said:
For people who only want the decomposed halves,

I mean, maybe I'm just misreading the tone of your entire post, because it makes more sense if everything you're saying is targeted only to people who 1) want to completely destroy the well-established existing tags like, you know, male_penetrating_Female, 2) replace them 100% with x_verbing, y_verbed (or similar) tags, 3) not establish implications, for god's sakes, kill the x_verbing_y tags, rah!! and 4) pretend there are no other problems. But like, I don't think anyone here actually wants that.

everybodyknowsuradog said:
I do think this suggestion is generally harder to implement without support for aliasing one entered tag to multiple target tags—alias male_penetrating_female to both male_penetrating and female_penetrated simultaneously, so familiar taggers (or just taggers who want a shorthand!) can tag the combo and get the decomposed tags applied (with the combo skipped out). AFAIK that's not a feature yet. Would love to be corrected on this though!

If this isn't acknowledging the reality of the situation, I don't know what is.

I could absolutely be misreading these BURs, but I strongly disagree with aliasing e.g. male_fingering_anthro -> only male_fingering. It needs to either be aliased to both male_fingering and anthro_fingered/fingering_anthro, or be made an invalid tag (or just a non-tag). I can't imagine a situation where we would have good justification for tagging only one side of the equation, and if we do want to tag both sides, then the tag relationships need to reflect this, or else not exist at all.

Again, could be misreading the BURs, I just can't tell if these are addressing the character receiving the action at all. If they aren't, I don't think we should go ahead with them.

everybodyknowsuradog said:
I could absolutely be misreading these BURs, but I strongly disagree with aliasing e.g. male_fingering_anthro -> only male_fingering. It needs to either be aliased to both male_fingering and anthro_fingered/fingering_anthro, or be made an invalid tag (or just a non-tag). I can't imagine a situation where we would have good justification for tagging only one side of the equation, and if we do want to tag both sides, then the tag relationships need to reflect this, or else not exist at all.

Again, could be misreading the BURs, I just can't tell if these are addressing the character receiving the action at all. If they aren't, I don't think we should go ahead with them.

Well you can't alias to both, and making it an invalid tag because you can't alias to both is just splitting the baby

snpthecat said:
Well you can't alias to both, and making it an invalid tag because you can't alias to both is just splitting the baby

Then if it's possible for that feature to be added, this should be done. If it's reasonable for the BURs here to wait on that then like, *raises hand* I'm OK with waiting for that.

But I think realistically a better approach (at least until multi-alias is supported) would be to not alias these tags at all, just fill in the appropriate implications and let the combination x_verbing_y tags keep existing. It's very bad vibes to make tag relationships (whether aliases or implications) represent something different than what we actually want taggers to tag. If a non-alias relationship is appropriate it should go both ways; if an alias relationship is appropriate it should not be added until it can go both ways.

everybodyknowsuradog said:
Then if it's possible for that feature to be added, this should be done. If it's reasonable for the BURs here to wait on that then like, *raises hand* I'm OK with waiting for that.

You're going to have to wait for a very long while.

snpthecat said:
You're going to have to wait for a very long while.

Forever even, since that isn't compatible with how aliases are designed. They are 1:1, not 1:many. When you try to get into 1:many you get into the problem of the later targets of these multi aliases no longer existing on the posts. If we keep track of the original tags before aliases on posts, they're just a worse implication system. This issue would happen even in the initial aliasing due to how BUR's are designed, there is no guaranteed order in execution, one instruction could be executed now, and another in 20 minutes depending on the current load. A special instruction could be made to counteract this or something, but in the end it just isn't worth it