The push to stamp out galling ethnic name bias on phones and computers
June 14, 2024 7:20 AM   Subscribe

Is autocorrect racist? The push to stamp out galling ethnic name bias on phones and computers. A new campaign — called I Am Not A Typo — is urging tech companies to fix ethnic bias in their algorithms to stop autocorrect mangling so many people's names.
posted by chariot pulled by cassowaries (35 comments total) 14 users marked this as a favorite
 
I don't blame autocorrect for trying to "fix" less common names, but I think it's a major feeling of the tool that it doesn't learn names that you use often and add them to its dictionary quickly. At a bare minimum, if you have saved your name to your phone, that name should not be autocorrected!

Similarly, I often type with speech to text for a variety of reasons, and I wish that that was able to learn words you use that aren't common, especially proper nouns.
posted by LSK at 7:25 AM on June 14 [14 favorites]


I use the Gboard swipe keyboard, which takes the simple step of not autocorrecting anything that starts with a capital letter. I'm surprised other providers don't do the same thing -- it would be useful for more than just names, too.
posted by Rhaomi at 7:29 AM on June 14 [3 favorites]


It is never duck, Autocorrect. It is never duck.
posted by Melismata at 8:18 AM on June 14 [21 favorites]


I have deck ducks… 🤔

But on topic, I had to turn off autocorrect on the device I used for taking notes in project meetings because it wouldn’t leave internal acronyms alone.
posted by funkaspuck at 8:39 AM on June 14


For whatever reason it seems like autocorrect on phones is less flexible than the spellcheck that's been in word processor for decades. I can tell my computer "no that isn't a typo" with a click but I've got no idea how to tell my phone the same
posted by thecjm at 8:40 AM on June 14 [6 favorites]


It's also way more aggressive.

I remember when autocorrect was added to Microsoft Word, it was a stupid little thing that existed apart from the spellchecker to automatically correct a handful of i-before-e mistakes or places where people frequently transpose letters while typing certain words. You could even view the ruleset it used and make custom additions. Words the spellchecker thought were mispelled still got that squiggly blue line and you had to intercede with a right-mouse click in order to apply any corrections.
posted by RonButNotStupid at 9:02 AM on June 14 [5 favorites]


on the other hand, my phone insists on always correcting "this" to "Thai"
posted by bendybendy at 9:14 AM on June 14 [3 favorites]


I can tell my computer "no that isn't a typo" with a click but I've got no idea how to tell my phone the same

On my Android phone, you do it by choosing it in the autocorrect menu. Let's say I have a friend named Flord. On my phone, I can type out Flord, and the autocorrect list above the keyboard offers me Ford, Flord, or Florida. Florida is bolded to tell me that it's going to autocorrect to that. If I tap on 'Flord' instead, it stays as 'Flord' and adds 'Flord' to the dictionary, and it'll begin being available for swipe typing and even autocorrect.
posted by Hatashran at 9:14 AM on June 14 [5 favorites]


there is an option in iphones to not autocorrect words as you are typing, but if you are typing quickly, or just don't realize it, you will miss it. Its the first one on the left with the quote marks around it. I think if you click it, it will add the word to a dictionary and not automatically try to autocorrect it, but I'm not sure. I agree it's a terrible interface.
posted by The_Vegetables at 9:28 AM on June 14 [2 favorites]


It also doesn't fix what the article is about, which is that they should be in the dictionary, same as common anglophone names are.
posted by The_Vegetables at 9:29 AM on June 14 [7 favorites]


Gboard's swipe keyboard is sometimes really good with unusual person and place names, to the point where I think it must be learning from what I type or from my contacts, but then sometimes it seems to completely forget a proper noun it did impressively well with before, even if I use it frequently.

At a bare minimum, if you have saved your name to your phone, that name should not be autocorrected!

Given how much location data phones tend to collect on you, they should also be able to do local addresses, local sites, and local major business names...
posted by trig at 9:29 AM on June 14


My problem has been similar to bendybendy's. Often I type something, autocorrect will fuck it up, then I go back and retype it correctly, and autocorrect will fuck it up again, the word I am typing is the far more common one and autocorrect is fucking it up to something *really* strange and super specific. Like. WHY.

A couple of days ago I went to the setting on my Android phone and supposedly turned off autocorrect but this issue has not stopped. Autocorrect lives on. Duck me, I guess. It seems I must resign myself to sounding like an ungrammatical nonsense writer who cannot spell basic shit and who confuses your/you're and so on. Of all the things I expected to happen to me in The Golden Future Of Computing, I expected this - being perpetually embarrassed and made to look like a fool by AI - least of all.
posted by MiraK at 9:35 AM on June 14 [11 favorites]


An algorithm miscorrecting African and Asian names is not racist.

A software developer saying, "This is not a high priority bug, ship it." is racist.

Let's not frame the discussion by putting the focus on the machines.
posted by AlSweigart at 10:36 AM on June 14 [3 favorites]


Por que no los dos? Yes, devs refusing to prioritize a fix are being racist. but algorithms miscorrecting non-anglo names are also racist; the algorithm was clearly trained on biased data. Remember that racism does not require conscious or deliberate prejudice. Racism is defined by impact, not by intention.
posted by MiraK at 10:51 AM on June 14 [2 favorites]


My family is full of autocorrect-tormented names. I've switched to FUTO, which is one of the best apps I have ever used. Whatever magic they have summoned should be replicated across the board. The project to fix the implicit bias in tech is very welcome.
posted by criticalyeast at 11:23 AM on June 14 [2 favorites]


It also doesn't fix what the article is about, which is that they should be in the dictionary

I have an uneducated notion that adding all languages' proper names to spell-check dictionaries would mean spell-checking stops catching any spelling errors, but perhaps that's not true? Whatever, I think filters at some level are needed.
posted by Rash at 11:38 AM on June 14


Well I mean it took Data that long to be able to use contractions, so…
posted by St. Peepsburg at 11:47 AM on June 14


Seems like a non issue. My name isn't a common spelling either so auto correct often might not recognize it. No big deal.
posted by Liquidwolf at 12:23 PM on June 14


> A software developer product manager saying, "This is not a high priority bug, ship it." is racist.

FTFY.

Developers don't get to decide what ships.
posted by Aardvark Cheeselog at 12:29 PM on June 14 [5 favorites]


My name gets autocorrected to a common over the counter drug but that also happens in people's heads when they just read it quickly. I can't count the number of times a server or bouncer or whatever reads my license and is like "Is your name really...? Ohh..." and then acts disappointed.
posted by downtohisturtles at 12:33 PM on June 14


I don't think I would call it "racist". There's a limited amount of space available and close to 10 billion potential names and in some cases, translating a name to another language or character set inflates that number. I can't read a Chinese name in its native character set, and likewise some Chinese people probably think my name looks like deranged scribbling.

It's one of those things I think we have to accept in all its well-meaning imperfection. There are things that are impossible for automation. I think people may be expecting too much perfection.
posted by JustSayNoDawg at 1:57 PM on June 14 [3 favorites]


I Am Not A Typo is such a cool project, and a great example of exposing the consequences of Hanlon's Razor en masse. Creating a good word frequency analysis for a language is hard. Which data sources do you include? How far back in time do you go? How many regions do you include? How do you handle dialects? Is there even an available dataset you can use or do you have to create one yourself? The result will always be incomplete and subjective, and wrong in some significant way.

So instead of going through the hassle, everyone uses someone else's word lists, because of course they do. For example, Android's autocomplete dictionaries are opensource, so you can see how well they're maintained. The answer: not at all. Only two people have even touched the word lists themselves in the last 15 years. And the default lists are embarrassingly small. There are only 160K words for en_US, and the entries look like they've cribbed from some other even more terrible and arcane dataset. But your phone probably doesn't use these because even if you're an Android user, 98% of Android phones don't run the stock OS. Instead, you probably have your own even older word lists that are all also awful in their own unique ways.

Having a baseline "you need to meet this bar to be considered good autocorrect" is such a great idea it's embarrassing that such a campaign even needs to exist. Improving autocorrect won't create a perfect solution, but it can be vastly better than what we have now. Fingers crossed the people in charge actually pay attention for a change.
posted by lock robster at 3:48 PM on June 14 [7 favorites]


I think I mentioned here before, but Jermaine would always autocorrect to Hermione on android. Made texting about basketball pretty annoying when your phone is substituting someone's real name for a wizard princess from a kids book.
posted by lkc at 3:54 PM on June 14 [3 favorites]


A software developer product manager saying, "This is not a high priority bug, ship it." is racist.

But who sets the product manager’s priorities? It is not product managers all the way down, but nor is it racists all the way down. You can start with the racist preferences of users (perhaps expressed as a sort of latent racism in not demanding a better autocorrect in terms expressed in dollars) or the racist indifference of advertisers in the same way, trace that up to developers, and then their managers, and their directors, all the way up to the CEO and the board. But this is not an ouroboros, or at least not exactly a clean one.

The top of the corporate hierarchy answers to the owners of the corporation. The rules for the way we’re doing capitalism right now continue to almost entirely operate under the Friedman doctrine: “an entity's greatest responsibility lies in the satisfaction of the shareholders.” But shares are not evenly distributed, they are allocated by capital.

When I was reading the FPP, my head immediately exploded with possible technical solutions to an interesting problem. All of which are irrelevant, as several people here have identified, if the corporation lacks the will to implement them. That will can be coerced through market signals. Customers, whether they be advertisers or users, could coordinate to withhold spending and make their demands for autocorrect a business priority for a corporation.

But even this is refined corporatist bullshit of the highest degree, a bit of MBA jujitsu like the oil industry engineering, focus-grouping, and disseminating the concept of an individual’s “carbon footprint” so citizens would point their guns at individuals spending money on carbon instead of the corporations monetizing carbon on a global industrial atmosphere-shaking scale. This isn’t an issue of users being racist, or consumers being racist, or any other individual citizen being racist. This is about shareholders being racist.

There are many individual racists, but the voices of shareholders are amplified proportionately to their capital when it comes to dictating the priorities of a company. Insofar as a company is racist, it is reflected in the ownership structure of that company. There may be some racist end users, but capital is more racist louder. They are being the mostest racistist by several orders of magnitude. A small number of racist individuals who control vast amounts of capital in the form of shares have a much greater voice than the average individual.

It’s not us. It’s them. Even if some of the racism comes from the many, more of it is coming louder from the few. We can coordinate our spending in clever games of attrition, but as 𝕏 (formerly Twitter) shows, at the end of the day, the holder of capital, the owners of companies, they do what they will. The rest of us suffer what we must.

Point your guns accordingly.
posted by 1024 at 7:40 PM on June 14 [2 favorites]


I think I mentioned here before, but Jermaine would always autocorrect to Hermione on android. Made texting about basketball pretty annoying when your phone is substituting someone's real name for a wizard princess from a kids book.
Hermione is a perfectly real and ancient name - comes up in Homer, even.
posted by kickingtheground at 9:57 PM on June 14 [2 favorites]


The thing that always tips me over from "they're just a bit incompetent, there's too many bells and whistles they have to implement, it's my fault for picking an obscure english noun for my name, my interests are too niche to have technical terms already in the dictionary" to "their priorities are fucked and I'm justified in feeling offended at their suggestions" is when the dictionary lets slip that it has been taught the icorrect iway of icapitalising iphone, along with a slew of corporate trademarks that includes perfectly normal everyday common nouns that people are going to be texting about eating, and next-word suggestions that reveal an extremely unadventurous americocentric media diet.
It knows those words and phrases because of ongoing deliberate choices, if not of what to explicitly place in the dictionary then of what types of material to train their small language model on. That's when I start to hold it against tech companies that I still have to type every letter of my name, others' names, and technical terms.
posted by polytope subirb enby-of-piano-dice at 2:19 AM on June 15 [3 favorites]


I hate that it's seemingly impossible to remove a word from the autocorrect magic. Fat finger an incorrectly spelled word into the dictionary and you can't get it out.
posted by Mitheral at 6:30 PM on June 15 [3 favorites]


I have an uneducated notion that adding all languages' proper names to spell-check dictionaries would mean spell-checking stops catching any spelling errors, but perhaps that's not true?

Even if e.g. "Teh" was in the dictionary as a name, "teh" would still be subject to autocorrect/spellcheck - it would just offer "Teh" as an option as well as "the". And that's without even considering basic context rules.
posted by Dysk at 4:18 AM on June 16


Personally, I wish phone spellcheck worked more like PC spellcheck. Stop autocorrecting my shit, just put a squiggly red line under things that are deemed to possibly be wrong, and let me go back and pick one of the suggested corrections if I agree. But just silently changing something I typed -which is what phones do - is like the worst possible interface for a spellchecker.
posted by Dysk at 4:32 AM on June 16 [1 favorite]


I can hear the pearl clutching from here. "But what about duck?" "Someone using our product might be exposed to a bad word, maybe even intentionally, and that would degrade the user experience and devalue our brand."
posted by Mitheral at 4:52 AM on June 16 [1 favorite]


But just silently changing something I typed -which is what phones do - is like the worst possible interface for a spellchecker.

I think that's specific to keyboards and keyboard settings rather than phones. I use swipe keyboards which have to guess what I'm writing (and sometimes they guess wrong and I don't catch it, as my comment history demonstrates in abundance), but if I just tap out individual letterszxg nothing gets autocorrected at all.
posted by trig at 4:59 AM on June 16


I think that's specific to keyboards and keyboard settings rather than phones. I use swipe keyboards which have to guess what I'm writing (and sometimes they guess wrong and I don't catch it, as my comment history demonstrates in abundance), but if I just tap out individual letterszxg nothing gets autocorrected at all.

Ah maybe. I also use swipe keyboards, but between having two languages installed (both of which I use) and the general inability of the keyboard apps I use to learn words I use regularly that aren't in the dictionary, I find myself typing words pretty often. It does sometimes leave explicitly typed stuff alone, but if it's close enough to something it thinks is correct, it will correct it. (Swiftkey for example will not let me type OR swipe "pic" without automatically correcting it to "Pic" regardless of its position in a sentence, and regardless of how often I try to remove "Pic" from the dictionary altogether. GBoard is not really any better, but I can't recall any specific nonsensical annoyances.)
posted by Dysk at 5:03 AM on June 16


Hmm, I use gboard with multiple languages and dont hsve thar problen. Probably a settings thing?

There's one way I know of to get rid of specific suggested corrections with gboard, though I'm not sure if the word gets deleted from the dictionary altogether, or just isn't suggested as a correction for that specific word anymore, or what. Anyway: if you long-press a suggested correction you don't want, a trash can icon labeled "remove suggestion" should show up and you can drag the word there.


I just realized that it's not just phones vs. keyboards: apps can change the UX too. For example, I just tried 3 different apps on Android (Gmail, Whatsapp, and a random basic notes app): misspelled words are underlined in red and if I tap them I get suggestions and the option to add them to the dictionary. But in this text field I'm writing in right now using Firefox, that isn't happening.

(All of this is kind of a derail from the main topic but maybe it's helpful to someone?)

Happily today gboard has remembered how to let me swipe my non-English street name, which it's mostly done fine for years but was completely failing at a few days ago.
posted by trig at 5:39 AM on June 16


Anyway: if you long-press a suggested correction you don't want, a trash can icon labeled "remove suggestion" should show up and you can drag the word there.

Yep, and it's a temporary solution at best - it reverts after a while, in my experience.
posted by Dysk at 6:02 AM on June 16


I don't think I would call it "racist"
no, of course not. why would you? why would anyone? we all know racism can only ever be something someone does, like saying a slur, right? because we all know there's no such thing as structural racism or anything else.

it's not like coding in assumptions that names like charging*, use*, proto*, or sodding* are typos somehow could make people frustrated as they and their contacts are being told, subtly, that their names are too strange or uncommon.

and obviously that's not really a problem, since we all know that if people are really that bothered by it there's always a "normal" american english name that these weirdos could use to make them less foreign and strange.

after all, we're seeking efficiencies. over 10 billion names with so many looking like deranged scribbling, why not just tell those fucking weirdos with foreign, chinky names to pick "kevin" and "grace" to make programming jobs easier?

it's definitely not racist. not at all. indeed, only I am the racist one, for having used a slur to make a point, for being asian, for mildly irritated at being once again corrected that something wasn't actually racist.

because remember, we all just have to accept imperfection. some of us chinks more than others.

*these were all first names, autocorrected, that celebrities who have careers in the west have, and in order, their origins: Korean; German; Indian; Arabic
posted by i used to be someone else at 8:14 AM on June 16 [3 favorites]


« Older Robber barons in the food system   |   The war on truth Newer »


You are not currently logged in. Log in or create a new account to post comments.