Why Grammarly and other Grammar AI is secretly racist and you shouldn't use them
I wrote this post: https://www.tumblr.com/kimyoonmiauthor/769503985076977664/why-spell-check-and-some-grammar-check-isnt-ai
But right now I'm waiting for a render of the Earth in Blender that's taking forever to finish (16+ hours left)... sooo....
I thought I would write a linguistics/programming essay on why I never used Grammarly and how no matter how you set up a grammar program, it will be racist.
About me, because people don't read that on the blog...
I have a BA in Anthropology, as part of the BA in Anthropology I took a Cultural Linguistics Class. I'm not claiming this makes me an "expert" by any means, but more along the lines of "I have some bare basics"
I'm a terrrriiiblllleee programmer. I have some basics and can explain to you how to do scripting and have done work in php and C++.
English was not my "mother" tongue, though it is my language of fluency. So I have very pointed memories of learning English. I say this because I'm sure someone will try to say, but, you don't understand ESL students at all. No, I do, and when talking to Korean ESL students, I feel relief about the struggles I had learning English. I helped some ESL Koreans in exchange for (re)learning Korean... BTW, my Korean is also frozen in time....
I also learned English while listening to Canadian English, British English and Black American English of the region and American English, which is really 3 regions at once of 3 different class types. My English was very messy for years on end as I mixed different English types up willynilly. I eventually was able to purge the Black American English out of respect for Black people (I can still hear and distinguish it), but couldn't bring myself to speak SAE to the white levels my white mother demanded (which I've alluded to... I can't bring myself to speak white woman and white woman violence speech gives me hives. White woman toxic speech to almost karen levels makes me choke[link] and I always end up asking a white person to write that for me.) I was bullied somewhat into a certain type of English, and then eventually gave up and code switched to whomever was in front of me. Usually badly. (British are upset at an "American" using British English. Toffs. You shouldn't have imported Monty Python and other British shows to the US, then. I read British literature in British English... and Canadian books in Canadian English... and was able to get my hands on Australian books... in Australian English. Messy...) We aren't even getting to the historical Englishes I crunched into my brain. Jane Austen's "quite" is not the current "quite" of SBE. And now you are getting a taste of my problems with grammar programs.
I also know French, Japanese and a Tiiiiinnnyyyy bit of Mandarin not very well and a smattering of other languages.
I also have a weird habit of hearing a person's voice when I read text in their accent and cadence... but I'm not sure if that helps or hurts my case.
I make fun of this regularly, but the only thing I can reliably translate is a confession scene because I've watched so many Romantic Comedies in so many other languages I can tell you when someone is making a love confession, but Hell if I know how to ask for the bathroom in Hindi or Tagalog.
I've also helped translate Korea and Japanese dramas. I joke that I have loved them in the womb.
So I have a decent amount of different language group knowledge, some academic knowledge and some programming knowledge.
I still think someone else could probably outmatch me... but... ya know.
Laymen's understanding of Grammar
Most people start with understanding grammar as a word order thing, though it's more complex than that and then might level up to Chomsky, and then find out Chomsky is wrong and controversial.
But generally, the basics are that grammar is steady and DOES NOT CHANGE within the language.
But in cultural linguistics, once you get down to dialects and socioeconomic differences, and all of those racial differences and then creole, you find out this assumption is wrong.
For US English, it's called SAE but most SAE is comprised of Midwestern American English because that's what they use on the telephone messages. This has specific grammar requirements.
Then, you learn about different white dialects and Southie Boston is entirely different grammatically from North Boston.
Pop v. Soda v. Sodapop v. Coke... and how the surrounding grammar is going to go in context also changes.
So, the problem for programmers is this:
Which grammar structure are you going to choose for the sequence?
Black American English from the South, from California, from Midwest? (I did a paper on the Black American English found in the movie Ray, and made the thesis that there is code switching throughout the movie based on where the main character is. So had to become familiar with Southern to the specific state, NE, and particularly Californian white and Black English. I can still take a stab at where people are from based on those grammar and various linguistic features. BTW, people on the wikipedia page on Southern American English are still bent and say, "I don't speak Black" (though in ruder terms) while not understandin' what the article is sayin' to them is that there were certain features in them's English that come from Black folk because linguists traced the features back to various West African original languages. Yes, yes, even in them white folk English.)
Indigenous American English, but from which region?
White American English, but then from which class?
Oh, Oh, but then there is British English, Canadian English, NZ English, Australian English, Spanglish, Nigerian English, Singaporean English, Konglish, Singlish, Taglish, Filipino English (Filipino English in the Philippines is different from Filipino English in the US which is a huge politics I'm not going to hash. It's a badtrip. This is different from Taglish.), Hinglish, (I haven't listed them all) and all of these have different grammar rules and different vocabulary.
"Something" in Korean makes no sense in English and might have back flowed to US English somehow... "Situationship" v. flirting. Handphone is English, but nothing like a mobile or a cellphone. Knucks... can't decide between mobile and cellphone with the first hits from Bank of Canada listing it as a mobile and other sites listing it as a cellphone.
The thing is that a computer does not have enough if-then structure to recognize all of these different kinds of speech, but a human brain can. Remember, no computer on Earth has as much power as the human brain and the human brain is specifically programmed to recognize patterns, sometimes to the Turin-shroud-Jesus-in-toast detriment of humanity.
Computers, as it stands, can't read context very well. (I've tested LLMs on this) unless they are told what the context is. Humans can usually figure out the context even if they aren't given all of the pieces because human's pattern recognition tools are high.
Since the standard English used in academia is usually (male) SAE, this is what the programmers tend to use. And you aren't supposed to start with And. Gerunds are terrible. All of the academic rules that makes your language sound flat. No fragments.
But, wait, you just cut out, Asians, Indigenous Peoples of the America's English, all of Black American English (Both AAE and African diaspora English), Latines, and so on. You horrible, horrible people.
"You had had gone to the market." is now wrong to SAE, but acceptable in BAE.
Surely AI can crunch this?
No. Not enough RAM without burning the planet down can fix this. We literally have quantum computing and they *still* can't get anywhere near the level of processing of a human.
For example, if you programmed the computer to say Iunno, Levon is Black, and therefore uses BAE, it's going to choke as soon as Levon has to go in for a job interview in front of white people. This is something a human can understand: it's code switching. But how and when and how much code switching one does is something that breaks most humans, if not computers. *I* suck at code switching, how the hell is a computer supposed to do it smoothly when the rules aren't always clear? Then how do you crunch that into a linear grammar instruction on which grammar is correct?
The computer doesn't have enough information to figure out and cross reference what kind of code switching Levon would have to do in each situation.
LLMs are shown to be racist when it comes to resumés. Even with the reduction of this, in my cultural linguistics class said the cultural barriers can be fairly high. Work habits of humans worldwide is not even and a resume with 10 years of service doing the same job for India and maybe? Germany (from what I've read) might not be unusual, but might be flagged in the United States as "not enough variety of experience." LLMs can't crunch the nuances enough and can end up over correcting.
You know, who can crunch nuances? A human.
I have a love of dialects especially since my (natural) Korean language is somewhere between two non mainstream dialects, which is hilarious to my fellow Koreans. When speaking Korean, I might be pulled into speaking Korean with Seoul words, but in a Gyeongsangnamdo dialect cadence when I get too emotional in Korean in the middle of a sentence which makes every Korean I've met laugh from all regions of Korea. So I know for a fact that it's not an easy task to program that in.
So what can a programmer do without enough RAM?
Stick to the "standard" which usually is racist, classist, and sexist.
For American English: Rich, White, Anglo-Saxon Male, Straight, Protestant.
Correct sentences like, "She was married to her wife." (Google) and other queer statements. There's linguistic studies on gay men, gay women, bisexual women, NBs, etc. Here's ONE video with a bunch of international studies on "gay speech" https://www.youtube.com/watch?v=SF7KCsvcw2g and AI can't cover it. LLMs still think all NBs use "they". This NB has open pronouns in English, but I prefer iel in French. Eonni as an honorific in Korean (for long historical linguistic etymological reasons) and so on. This would break any LLM completely. WHAT DO YOU MEAN You use Eonni in Korean, but open pronouns in English. What's the logic? That wouldn't congruent with someone else's reasons. Grammar programs hate us queers. Wait until you see the corrections on polyamory and ambiamory...
Correct "you trippin'" (I took this from a Black commentator on a Black owned channel...)
to "You are tripping." (which, BTW, has a totally different meaning.)
make the "had had" marked wrong and so on.
I believe, though, we shouldn't be writing only white Anglo-saxon, straight males.
White women use more questions to placate issues.
White Jews use questions differently too. More to debate and hash out opinions.
BTW, East Asian Americans use questions in a third way... to ask everyone to self-reflect which, BTW, never goes well outside of those circles. *cough* I'll be over there, not saying how I know this.
When fighting, African Americans take offense to questions (from Thomas Kochman's Black and White Styles of Conflict) and I found out struggle to tell the difference between a white Jew rhetoric question and white Christian woman's placating question—the very hard way.
And the most popular dialect of British English changes by the year, if not the decade. lol (This is a joke I've heard Brits make... so it's fair? Also saw a BBC thing on it. Manchester is in, Manchester is out. The last trendy one was Multicultural British English... which the British Youtube channels keep calling the new English, though there are subtle racial differences the channels don't pick up on within MBE that I can hear.)
But I use it for academia...
I wrote Linguistics papers, as cited, and had to know the difference between the noun affect and the noun effect and affect the verb and effect the verb.
Humans get this wrong ALL of the time. The grammar program constantly wanted me to change affect, the noun to "effect" which BTW, has a different effect.
In linguistics, this is called "jargon" and often jargon uses common words in ways that make only sense in context, thus gets a different set of grammar rules.
AI can't crunch it, but you know who can? A human.
A doctor might say, "I would like an EKG and a CAT stat."
But would the computer think it's really a 🐈?
But the doctor later might say, "I saw the PET and the cat." But the computer wouldn't have enough power to remember that the doctor saw a previous cat in the waiting room because computer grammar looks at one sentence at a time.
Which is it? PET scan and a CAT scan, or they pet the cat, or they saw a PET scan and a 🐈?
Spell checkers and grammar checkers also flag all of my Asian names. Some humans also do it because they are so cross that the name resembles an English word.
And we haven't even gotten to things like baby talk, which is a type of register... and that doesn't have the same grammar rules either.
In Japanese, Japanese argue that anime has its own register separate from regular Japanese. Watch a Japanese person's face screw up ever so slightly if you say you used, "shimatta" in every day speech while they hide their mouth and try to gently correct you with... a question. "Sou?" *Teeth suck.* *Mmmm...* If you're USian, you'll totally miss the hint to self-reflect on your behavior. And even Korean has a different register for variety TV shows than normal Korean. Variety TV shows is a lot more informal than typical. Then news broadcasts have their own register too, usually for all countries.
Phone grammar might change by country. Korean phone grammar is a pain in the butt to remember all of the various rules. US English's becomes less formal. UK's becomes slightly more formal... and so on.
Have we also gone over people's changing grammar based on their relationship to each other which may or may not be based on title? The computer motherboard is overloaded about now.
∴ The best grammar checker is a specialist human.
Mild programming humor. If you know, you know. Even if you don't know, it's a stupid joke and you're probably thinking too much about it.
Humans suck. I get that. But asking another human also can help you refine it and build community. I get the self consciousness of asking another human, but you know what language is for?
look I've seen a lot of people reblogging my anti-AI post (NOT COMPLAINING ABOUT THAT PART!!) who seem to be under the impression that the biggest problem with genai is that...people are being lazy?
hot take but first of all why is that your concern and second of all, I don't think laziness exists - the concept of laziness is largely a capitalist, ableist, and very christian idea; in my experience, people who are "lazy" almost always have a genuinely good excuse to not be doing the thing you want them to do when you look a little deeper/ask them like
not wanting to do it because it takes time is reasonable! time is limited and valuable, especially in the capitalist hellscape we live in
not wanting to expend mental effort on it is also reasonable! as is not wanting to expend the physical effort, spoons, energy, etc, etc, etc
additionally - being "lazy" isn't a fucking crime or sin??? and it's not your job to decide whether someone is putting in enough effort! we are not meant to grind and work and put in 110% in everything we do in fact it's extremely unhealthy to do so
the biggest problem with LLMs isn't that the people don't want to do the work (although, I will admit, it can become a problem when they decide they don't want to learn and would rather use it to breeze through becoming a medical professional or the like - the problem here isn't that they're lazy tho it's that they're risking people's lives callously) the problem with LLMs is that they steal real humans' hard work and destroy the ecosystem, taking all the water while entire cities go through massive dangerous droughts
tl;dr stop calling people lazy, start calling them assholes destroying the planet for stolen content
I keep getting this ad on youtube for what I assume is some kind of AI coding tool or w/e, and it opens with some guy saying "there's a lot of the job that's not fun that I've now automated away" because you know, how dare a job have any aspect that's not fun?
The attitude of "We need to kill people in order to make technological advancements" is terrible. It's terrible. If these AI bros love it so much, they should live in one place and then put the data centers there.
