Friday, April 29, 2022

Autocorrect Explained: Why Your iPhone Adds Annoying Typos While Fixing Others

Ben Lovejoy:

But yeah, people do complain, and even the guy who created the feature – Ken Kocienda – admits that sometimes it can be more of a hinderance than a help. Indeed, if you think that it’s gotten worse rather than better over the years, that can actually be true …

The reason, paradoxically, is that autocorrect has grown more intelligent over the years, and the more we ask it to do, the more potential there is for new types of errors.

The WSJ’s Joanna Stern went on a mission to learn more about why autocorrect can sometimes be ducking annoying.

[…]

The personalized dictionary looks for words not in the static dictionary, which you have typed three times. That’s the point at which it decides you know what you’re doing, and it’s a real word. However, if you make the same typo three times, it learns that instead!

John Gruber:

Don’t miss the video, which involves some actual ducking ducks.

Louie Mantia, Jr.:

You ever get that absolutely horrible bug in macOS messages where you’re typing a long message and at a certain point it starts chopping up everything you’ve written, replacing characters, moving entire passages, and duplicating text?

Steve Troughton-Smith:

This is the serious autocorrect issue Apple introduced in macOS 12 to Catalyst apps. Twitter also suffers from it. macOS 11, no problem

Previously:

Update (2023-03-28): Todd Thomas:

“There’s no possible way I could guess what you were trying to type…”

7 Comments RSS · Twitter


I still find the new ML-powered autocorrect to be more frustrating than what I remember the initial autocorrect implementation was. I’m not sure if that’s due to rose-tinted glasses or not.

But I don’t think it’s obvious which overall approach is better. It seems that the smarter ML comes at the expense of predictability. And maybe a more predictable system is better in this case. Or maybe if it gets even smarter it’ll be OK.


Autocorrect should be a case study of how machine learning can make features worse. As far as I'm concerned it was a solved problem in 2010 and worked great then, and it's just been steadily getting worse since.


Nigel I 1000% agree with you, and I don’t think it’s rose-tinted glasses at all. Autocorrect has always been a little annoying sometimes, but pre-ML it was also predictable and its suggestions made some sense. Now it’s a minefield and most of the suggestions are absolutely baffling, occasionally making up words and phrases that do not even exist in the language.


It's been a mixed bag for me. Sometimes, it retroactively messes up something I wrote. But sometimes, it fixes something without me having to take care of it.

Either way, though, I don't understand why it can't highlight the words when it does that?

Also, I can't edit what Ken calls the dynamic dictionary, can I?


Auto-correct never worked well for me,
so I have it off all the time. Once a year I turn it on for a day or two and hope that it would be a quantifiable improvement, but it's still at the level that I better off without it.
And it's that bad for English, while in other languages it's just shamelessly much worse.


When you disable Slide to Type in the settings things seem to improve somewhat.


Adam Maxwell

I'm with Bri 100% on this. Autocorrect was super helpful when I bought my first iPhone (2011), but it became unusable with the predictive insertions and grew to be a liability by replacing text unnoticed before I hit send. I disabled it entirely on my phone over a year ago, and stopped feeling the urge to smash it with a framing hammer.

I think Michael's post on "Autocorrect and the delete key" highlighted the most annoying misfeature of the "new" autocorrect. https://mjtsai.com/blog/2020/12/17/ios-autocorrect-and-the-delete-key/

Leave a Comment