1 Title: Challenges to Yudkowsky's Pronoun Reform Proposal
4 Tags: Eliezer Yudkowsky
7 > Go, Soul, the body's guest,
8 > Upon a thankless errand:
9 > Fear not to touch the best;
10 > The truth shall be thy warrant:
11 > Go, since I needs must die,
12 > And give the world the lie.
14 > —["The Lie" by Walter Raleigh](https://www.lesswrong.com/posts/trb9HPWFk8Gy9MBdN/less-wrong-poetry-corner-walter-raleigh-s-the-lie)
16 [In a February 2021 Facebook post, Eliezer Yudkowsky inveighs against English's system of singular third-person pronouns](https://www.facebook.com/yudkowsky/posts/10159421750419228). As a matter of clean language design, English's lack of a gender-neutral singular third-person pronoun is a serious flaw. The function of pronouns is to have a brief way to refer back to entities already mentioned: it's more user-friendly to be able to say "Katherine put her book on its shelf" rather than "Katherine put Katherine's book on the book's shelf". But then why couple that grammatical function to sex-category membership? You shouldn't _need_ to take a stance on someone's sex in order to talk about her or him putting a book on the shelf.
18 This affects, for example, science-fiction authors writing about AIs or hermaphroditic aliens (which don't have a sex), or mystery authors writing about a crime suspect whose identity (and therefore, sex) is unknown. In these cases, [_she_ or](/2020/Apr/the-reverse-murray-rule/) _he_ are inappropriate, but the English language offers no alternative lacking its own downsides: _it_ is understood to refer to non-persons, _they_ gets conjugated as a plural, and neopronouns like [_ey/em/eir_](https://en.wikipedia.org/wiki/Spivak_pronoun)—or [_ve/ver/vis_](http://www.urticator.net/essay/0/30.html), as used in some of [Yudkowsky's juvenilia](https://intelligence.org/files/CFAI.pdf)—are hard to rally adoption for because pronouns are a [closed class](https://en.wikipedia.org/wiki/Part_of_speech#Open_and_closed_classes)—not something people are used to new members of being coined, in the way that people are used to seeing new nouns and adjectives.
20 It doesn't have to be this way! If you were fortunate enough to be in the position of intelligently designing a language from scratch, you could just include a singular third-person gender-neutral pronoun (like _it_, but for persons, or like _they_ but conjugating in the singular) in the original closed set of pronouns! If you wanted more pronoun-classes to reduce the probability of collisions (where universal [_ey_](https://en.wikipedia.org/wiki/Spivak_pronoun) or singular _they_ would result in more frequent need to repeat names where a pronoun would be ambiguous), you could devise some other system that doesn't bake sex into the language while driving the collision rate even lower than that of the sex-based system—like using initials to form pronouns (<em>K</em>atherine put <em>k</em>er book on its shelf?), or an oral or written analogue of [spatial referencing in American Sign Language](https://www.handspeak.com/learn/index.php?id=27) (where a signer associates a name or description with a direction in space, and points in that direction for subsequent references).
22 (Although—one might speculate that "more classes to reduce collisions" could _be_ part of the historical explanation for grammatical gender, in conjunction with the fact that sex is binary and easy to observe. None of the other most salient features of a human can quite accomplish the same job: age is continuous rather than categorical; race is also largely continuous [(clinal)](https://en.wikipedia.org/wiki/Cline_(biology)) and historically didn't typically vary within a tribal/community context.)
24 If you grew up speaking English, gendered pronouns feel "normal" while gendered [noun classes](https://en.wikipedia.org/wiki/Noun_class) in many other languages (where, _e.g._, in French, a dog, _le chien_, is "masculine", but potatoes, _la pommes de terre_, are "feminine") seem strange and unnecessary, but someone who grew up with neither would regard both as strange. If you spoke a language that didn't _already_ have gendered pronouns, you probably wouldn't be spontaneously eager to add them.
26 All this seems fine as a critique of the existing English pronoun system! However, I argue that Yudkowsky's prescriptions for English speakers going forward goes badly wrong. First, Yudkowsky argues that it's bad for stances on complicated empirical issues to be baked into the language grammar itself: since people might disagree on who fits into the [empirical clusters](https://www.lesswrong.com/posts/WBw8dDkAWohFjWQSk/the-cluster-structure-of-thingspace) of "female" and "male", you don't want people to be forced to make a call on that just in order to be able to use a pronoun.
28 Fair enough. Sounds like an argument for universal singular _they_ (and eating the cost of increased collisions where it's ambiguous which subject an instance of _they_ would refer to): if you don't think pronouns should convey sex-category information, then don't use pronouns that convey sex-category information! But then, in an unexplained leap, Yudkowsky proclaims:
30 > So it seems to me that the simplest and best protocol is, "'He' refers to the set of people who have asked us to use 'he', with a default for those-who-haven't-asked that goes by gamete size" and to say that this just _is_ the normative definition. Because it is _logically_ rude, not just socially rude, to try to bake any other more complicated and controversial definition _into the very language protocol we are using to communicate_.
32 The problem with this is that [the alleged rationale for the proposal does not support the proposal](https://www.lesswrong.com/posts/i6fKszWY6gLZSX2Ey/fake-optimization-criteria). If your default pronoun for those-who-haven't-asked goes by perceived sex (which one presumes is what Yudkowsky means by "gamete size"—we don't typically observe people's gametes), then you're still baking sex-category information into the language protocol in the form of the default! Moreover, this is clearly an "intended" rather than an accidental effect of the proposal, in the sense that a policy that _actually_ avoided baking sex-category information into the language (like universal singular _they_, or name-initial- or hair-color-based pronouns) would not have the same appeal to many of those who support self-chosen pronouns: _why_ is it that some people would want to opt-out of the sex-based default?
34 Well, it would seem that the motivating example—the historical–causal explanation for why we're having this conversation about pronoun reform in the first place—is that trans men (female-to-male transsexuals) prefer to be called _he_, and trans women (male-to-female transsexuals) prefer to be called _she_. (Transsexuals seem much more common than people who just have principled opinions about pronoun reform without any accompanying desire to change what sex other people perceive them as.)
36 But the _reason_ transsexuals want this is _because_ they're trying to change their socially-perceived sex category and actually-existing English speakers interpret _she_ and _he_ as conveying sex-category information. People who request _he/him_ pronouns aren't doing it because they want their subject pronoun to be a two-letter word rather than a three-letter word, or because they hate the [voiceless postalveolar fricative](https://en.wikipedia.org/wiki/Voiceless_postalveolar_fricative) (_sh_) sound. They're doing it _because_, in English, those are the pronouns for _males_. If it were _actually true_ that _she_ and _he_ were just two alternative third-person pronouns that could be used interchangeably with no difference in meaning, with the only function of the distinction being collision-avoidance, then _there would be no reason to care_ which one someone used, as long as the referent was clear. But this doesn't match people's behavior: using gender pronouns other than those preferred by the subject is typically responded to as a social attack (as would be predicted by the theory that _she_ and _he_ convey sex-category information and transsexuals don't want to be perceived as their natal sex), not with, "Oh, it took me an extra second to parse your sentence because you unexpectedly used a pronoun different from the one the subject prefers, but now I understand what you meant" (as would be predicted by the theory that "_he_ refers to the set of people who have asked us to use _he_ [...] and to say that this just _is_ the normative definition").
38 You can't have it both ways. "That toy is worthless", says one child to another, "_therefore_, you should give it to me." But if the toy were _actually_ worthless, why is the first child demanding it? The problem here is not particularly subtle or hard to understand! If the second child were to appeal to an adult's authority, and the adult replied, "The toy _is_ worthless, so give it to him," you would suspect the grown-up of not being impartial.
40 "Pronouns shouldn't convey sex-category information," is a fine [motte](https://slatestarcodex.com/2014/11/03/all-in-all-another-brick-in-the-motte/), but it's not consistent with the bailey of, "_Therefore_, when people request that you alter your pronoun usage in order to change the sex-category information being conveyed, you should obey the request." Even if the situation is an artifact of bad language design, as Yudkowsky argues—that in a saner world, this conflict would have never come up—that doesn't automatically favor resolving the conflict in favor of the policy of keeping both _she_ and _he_ but asserting that the difference doesn't mean anything.
42 This may be clearer to some readers if we consider a distinction less emotionally and politically fraught than sex/gender in the current year. [Many languages have two different second person singular pronouns that distinguish the speaker's relationship to the listener as being more familiar/intimate, or more formal/hierarchical.](https://en.wikipedia.org/wiki/T%E2%80%93V_distinction) In Spanish, for example, [the familiar pronoun is _tú_ and the formal pronoun is _usted_](https://en.wikipedia.org/wiki/Spanish_personal_pronouns#T%C3%BA/vos_and_usted): one would address friends, family members, children, or personal servants as _tú_, but strangers or social superiors as _usted_. Using the wrong pronoun can be the cause of offense or awkwardness. A speaker switching from _usted_ to _tú_ for an interlocutor who they're getting along might ask if it's okay with _¿Te puedo tutear?_ (Can I call you _tú_?) or _Nos tuteamos, ¿verdad?_ (We call each other _tú_, right?); this is somewhat analogous to an English speaker asking if they may address someone by first name, rather than with a courtesy title or honorific (Ms./Mr. Lastname, or ma'am/sir).
44 One could argue that the _tú_/_usted_ distinction is bad language design for the same reason Yudkowsky opposes the _she_/_he_ distinction: you shouldn't be forced to make a call on how familiar your relationship with someone is just in order to be able to use a pronoun for them. The modern English way is more flexible: you _can_ indicate formality if you want to by saying additional words, but it's not baked into the grammar itself.
46 However, if you were going to reform Spanish (or some other language with the second person formality distinction), you would probably abolish the distinction altogether, and just settle on one second-person singular pronoun. Indeed, that's what happened in English historically—the formal _you_ took over as the universal second-person pronoun, and the informal singular _thou_/_thee_/_thine_ has vanished from common usage. (People still recognize it as a second-person pronoun when encountered in old poetry—"The truth shall be thy warrant", _&c._—but most probably aren't aware of the formality distinction.) You wouldn't keep both forms, but circularly redefine them as referring only to the referent's preferred choice of address (?!).
48 Really, the circular definition shouldn't satisfy _anyone_: people who want someone to call them _usted_ (or _tú_), do so _because_ of the difference in meaning and implied familiarity/respect, in the _existing_ (pre-reform) language. (Where else could such a preference possibly come from?) People who want the ability to dictate whether people address them with familiarity or respect might _think_ the circular definition is what they want, because it implies the behavior they want (other people using the preferred pronoun), but—whether or not the proponent of the changes consciously _notices_ the problem—the redefinition is _functionally_ "hypocritical": it's only desirable insofar as people aren't _actually_ using it internally.
50 This is a pretty basic point, and yet Yudkowsky steadfastly ignores the role of existing meanings in this debate, bizarrely writing as if we were defining a conlang from scratch:
52 > It is Shenanigans to try to bake your stance on how clustered things are and how appropriate it is to discretely cluster them using various criteria, _into the pronoun system of a language and interpretation convention that you insist everybody use!_
54 There are a couple of problems with this. First of all, the "that you insist everybody use" part is a bit of a [DARVO](https://en.wikipedia.org/wiki/DARVO) in the current political environment around Yudkowsky's social sphere. A lot of the opposition to self-chosen pronouns is about opposition to _compelled speech_: people who don't think some trans person's transition should "count", don't want to be coerced into legitimizing it with the pronoun choices in their _own_ speech. That's different from insisting that _others_ use sex-based non-subject-preferred pronouns, which is not something I see much of outside of gender-critical ("TERF") forums. Characterizing the issue as being about "freedom of pronouns", [as Yudkowsky does in the comment section](https://www.facebook.com/yudkowsky/posts/10159421750419228?comment_id=10159421833274228), elides the fact that freedom to specify how other people talk about you is in _direct conflict_ with the freedom of speech of speakers. No matter which side of the conflict one supports, it seems wrong to characterize the self-ID pronoun side as being "pro-freedom", as if there wasn't any "freedom" concerns on the other side. [(Policy debates should not appear one-sided!)](https://www.lesswrong.com/posts/PeSzc9JTBxhaYRp9b/policy-debates-should-not-appear-one-sided)
56 More importantly, however, in dicussing how to reform English, we're not actually in the position of defining a language from scratch. Even if you think the cultural evolution of English involved Shenanigans, it's not fair to attribute the Shenanigans to native speakers accurately describing their native language. Certainly, language can evolve; words can change meaning over time; if you can get the people in some community to start using language differently, then you have _ipso facto_ changed their language. But when we consider language as an information-processing system that we can reason about using our standard tools of probability and game theory, we see that in order to change the meaning associated with a word, you actually _do_ have to somehow get people to change their usage. You can _advocate_ for your new meaning and use it in your own speech, but you can't just _declare_ your preferred new meaning and claim that it applies to the language as actually spoken, without speakers actually changing their behavior. As a result, Yudkowsky's proposal "to say that this just _is_ the normative definition" doesn't work.
58 To be clear, when I say that the proposal doesn't work, I'm not even saying I disagree with it. I mean that it literally, _factually_ doesn't work! Let me explain.
60 The "meaning" of language isn't some [epiphenominal](https://www.lesswrong.com/posts/fdEWWr8St59bXLbQr/zombies-zombies) extraphysical fact that can be declared or ascertained separately from common usage. We can only say that the English word "dog" means [these-and-such four-legged furry creatures _Canis familiaris_](https://en.wikipedia.org/wiki/Dog), because English speakers actually use the word that way. [The meaning "lives" in the systematic correspondence between things in the world and what communication signals are sent.](https://www.lesswrong.com/posts/4hLcbXaqudM9wSeor/philosophy-in-the-darkest-timeline-basics-of-the-evolution)
62 There's nothing magical about the particular word/symbol/phoneme-sequence "dog". In German, they say _Hund_; in Finnish, they say _koira_; in Korean, they say _개_. Germans and Finns and Koreans (and their dogs) seem to be getting along just as well as we Anglophones.
64 Nevertheless, it is a fact _about contemporary English_ that "dog" means dog. If you thought this was bad for whatever reason, and you wanted to change that fact, you'd have to change the behavior of actually-existing English speakers. If you tried to stipulate on your Facebook wall that "blerble" should mean dog now, and all of your Facebook friends nodded in agreement at your clever argument _and then continued to call dogs "dogs" in their everyday life just like they always had_, then your language reform attempt would have, _in fact_, failed—even if the fact that it failed would be less obvious if you only looked at the Facebook thread full of people nodding in agreement.
66 The inseparability of meaning from behavior-and-usage may be clearer if considered in a context other than that of natural language. Take computer programs. Sometimes programmers make bad design decisions. For example, in the C programming language, [it's standard to represent strings (textual data) in memory with a sequence of bytes ending in a zero (null) character](https://en.wikipedia.org/wiki/Null-terminated_string); the machine only knows where the string stops when it reaches the null at the end. This convention has a lot of disadvantages relative to the alternative of [prefixing the string data with the length](https://en.wikipedia.org/wiki/String_(computer_science)#Length-prefixed); a missing or misplaced null character could cause the machine to erroneously read or write data in adjacent memory, causing serious bugs or security vulnerabilities.
68 Given the existence of strong arguments for the length-prefixed string convention, replacing old software that uses null-terminated strings with new software that uses length-prefixed strings, sounds like a good idea! But the thing is, you _do_ have to upgrade or replace the old software. If you _just_ start sending data in a new format to the old software that doesn't understand the new format, your code is not going to yield the expected results. It would be _convenient_ if you could just declare a new semantics for your existing data on your Facebook wall and be done, but that just doesn't work if you're still using the old software, which is programmed to behave according to the old data-interpretation convention. This continues to be true even if the convention you're trying to retire is very bad (like null-terminated strings), and if the old software is widely deployed and would be very expensive to systematically replace. The backwards-compatibility trap [is real and can't be defied away even if it's very unpleasant.](https://twitter.com/ESYudkowsky/status/1216799697907486720)
70 Natural language faces a similar backwards-compatibility trap. The English language, as "software", is _already_ "deployed" [to 370 million brains as native speakers, and another 980 million second-language speakers](https://en.wikipedia.org/wiki/List_of_languages_by_total_number_of_speakers#Top_languages_by_population).
72 And among those hundreds of millions of speakers, there is _already_ a very firmly entrenched convention that _she_ refers to females and _he_ refers to males, such that if you say, "I met a stranger in the park; she was nice", the listener is going to assume the the stranger was female, even if you didn't say "The stranger was female" as a separate sentence. If the listener later gets the chance to meet the stranger and the stranger turns out to be male, the listener is going to be _surprised_.
74 Bad language design? I mean, maybe! You could argue that! You could probably get a lot of Likes on Facebook arguing that! But if 370 million native English speakers _including you and virtually everyone who Liked your post_ are going to _continue_ automatically noticing what sex people are and using the corresponding pronouns without consciously thinking about it, then the criticism seems kind of idle!
76 The "default for those-who-haven't-asked [going] by gamete size" part of Yudkowsky's proposal is _trying_ to deal with the backwards-compatibility problem.
78 [OUTLINE of remainder—
79 * The attempted backwards compatibility measure doesn't work; if people's behavior is still the same, then sex-category information is still being transfered; and, again, that was the motivation for the reform effort all along; in order to _actually_ de-gender language, you have to break the correlation: either go all-in on singular _they_, or somehow get people to establish and declare pronoun preferences that aren't about gender (which isn't what's going on in our world)
80 * Given that sex-category information _is_ being transfered, the "Pronouns are Ryphnol" lady has a point
81 * And preferred pronouns have the same function as the typographic attacks in the multimodal-neurons paper
82 * Yudkowsky's response to all this?—apparently, to play dumb!!
83 * "I don't know what it's like in you head for a pronoun to map onto more than 'doesn't look like an Oliver'"—lies
85 * CFAI footnote 16 example
86 * "It can't be based on feelings"—hypocrisy, the only reason we're talking about this at all is because of genderspecial people's feelings, as explicitly acknowledged in the OP!!!
87 * "Can't imagine a sympathetic protagonist"—lies, imagine a rape victim
88 * "If there were unspeakable arguments against, we couldn't talk about them"—okay, then you and your rationalists are frauds
89 * I know none of this matters, but one would have thought that the _general_ skills of correct argument would matter for saving the world ... right? / brief recap of my Whole Dumb Story, need the correct answer in order to decide
92 * Douglas Hofstader also made fun of gendered pronouns with his "Person Paper"—but notice that he didn't even consider the self-chosen criterion!!
93 * singular they for named individuals undermined indefinite singular 'they'
98 https://www.facebook.com/yudkowsky/posts/10159421750419228
100 https://www.lesswrong.com/posts/onwgTH6n8wxRSo2BJ/unnatural-categories-are-optimized-for-deception
104 * The problem with this is that the proposed convention still transmits sex-category info; you're just not being honest about it
105 * Software is already deployed
108 * Aside: "gamete size"—this is a tic where everyone knows what sex is, but no one is allowed to acknowledge the cluster
109 * Aella https://knowingless.com/2019/06/06/side-effects-of-preferred-pronouns/
110 * Pronouns are ryphenol https://fairplayforwomen.com/pronouns/
111 * Policy debates should not appear one-sided
112 * Rape victim is a sympathetic character
113 * "I don't know what it feels like to 'you don't look like an Oliver'" is a lie; you can use pronouns for someone whose sex but not name you don't know https://web.archive.org/web/20070615130139/http://singinst.org/upload/CFAI.html#foot-15
114 * non-compelled speech is more compelling than clothing freedom
115 * at least Sabbatai Zevi had an excuse: his choices were to convert to Islam or be impaled https://en.wikipedia.org/wiki/Sabbatai_Zevi#Conversion_to_Islam
116 * I need the correct answer
117 * "We can't talk about this"—utterly discrediting of the entire project"
118 * not the woke position—it's an incoherent position
119 * What is the regularity in human psychology such that we end up with "gendered" noun classes? We are sexually dimorphic animals
120 * compare the formal/informal distinction tu/Usted in other languages—that's a case where you obviously want speaker choice, not subject choice
121 * the people aligning language models need to know this!!
122 * he can only speak in terms of abstractions that are very obviously not what's happening—it's true that bathroom usage is not an ontological fact, but the function of bathrooms is _to protect females from males_. If you can't talk about that core issue—the thing that people actually care about—then the smugness is actively derailing the discussion, even if you didn't say anything false
123 * And doesn't EY have this whole thing about how you can't just wish away coordination problems?! (Although, this also makes it harder to escape the self-ID Schelling point)
124 * Schild's ladder—noun classes in other languages are already pretty arbitrary; if the proposal is to make names like that
125 * TODO: buff my "circular definition satisfies no one" argument to not be vulnerable to the anti-Liskov-substitution property of natural language definitions
126 * Amazing World of Gumball, "The Nest", "Who says this pregnant turtle is a her?" and everyone gives him a look. Yudkowsky isn't really claiming not to get the joke?! The show is rated TV-Y7!!! https://rating-system.fandom.com/wiki/TV-Y7 EY is dumber than a 7-year-old
127 * typographic attacks https://openai.com/blog/multimodal-neurons/
130 https://www.ehu.eus/seg/_media/gizt/5/5/brown-gilman-pronouns.pdf
132 > In terms of important things? Those would be all the things I've read - from friends, from strangers on the Internet, above all from human beings who are people - describing reasons someone does not like to be tossed into a Male Bucket or Female Bucket, as it would be assigned by their birth certificate, or perhaps at all.
134 Well, as Yudkowsky says earlier in the post, many "human beings who are people"—which pleonastic construction bizarrely seems to suggest Yudkowsky is opposing someone who somehow believes that some humans are _not_ people, but never mind—have "describ[ed] reasons someone does not like to be tossed into a Male Bucket or Female Bucket, as it would be assigned by their birth certificate, or perhaps at all."
138 For example, I think it's Shenanigans to use the word "roommate" to refer to people who only share a house or apartment and not a literal room; surely you should say "housemate" or "flatmate" if that's what you really mean. However, this claim of mine about the meaning of the word "roommate" is [actually _false_ in American usage](https://en.wiktionary.org/wiki/roommate). (Apparently the British are more sensible about this.) The only way to get the Shenanigans to stop is to get people to _actually_ adopt my usage in their mapping of people's-living-situations to word-used-to-describe-living-situation. If I were to just _pretend_ that my preferred usage was already the actual usage, then I would make worse predictions when my friends in California mention their roommates.
140 https://www.glowfic.com/posts/4508?page=14
141 > Real people have concepts of their own minds, and contemplate their prior ideas of themselves in relation to a continually observed flow of their actual thoughts, and try to improve both their self-models and their selves.
143 What Quakers Can Teach Us About the Politics of Pronouns
144 https://archive.is/bYdde
146 If the software is written so that each computer broadcasts a `{"object_type": "BLEGG"}` JSON message when it detects a [blue egg-shaped object](https://www.lesswrong.com/posts/4FcxgdvdQP45D6Skg/disguised-queries) in front of its webcam, then we can can say that the `{"object_type": "BLEGG"}` message "means" that a blue egg-shaped object was seen;
148 Maybe this was a design mistake!
projects / Ultimately_Untrue_Thought.git / blob