The machine learning technology that Apple is using for autocorrect has been improved in iOS 17. Apple says it has adopted a “transformer language model,” that will better personalize autocorrect to each user. It is able to learn your personal preferences and word choices to be more useful to you.
After using iOS 17 for a few weeks, most users will notice that the autocorrect suggestions are much better at predicting what you want to say and presenting words for you to tap to autofill. When you use acronyms, shortened words, slang words, and colloquialisms, autocorrect is not as aggressive with the automatic correcting, but it is still able to correct accidental typos.
Imagine my relief, then, when I realized that iOS 17’s brand new autocorrect feature based on a transformer model was not just marketing speak but actually works and makes typing on the iPhone’s (and iPad’s) software keyboard a…pleasant experience. I can’t believe I’m saying this but I love Apple’s new autocorrect system in a way I never even remotely appreciated typing on a software keyboard.
The new autocorrect is so good, it allowed me to write a good chunk of this review without using the Magic Keyboard for iPad at all.
[…]
All of these traits are complemented by the refreshed user experience Apple designed for the system keyboard in iOS 17. For starters, when a word is automatically corrected, it gets underlined; tap the underlined word, and a new popup appears, allowing you to revert to what you wanted to type in the first place or choose between different suggestions.
After using it for a while now, I think autocorrect is definitely improved in iOS 17, but I’m not seeing as dramatic an improvement as others are reporting. I’m not sure that it’s better than the old iOS 10 system that didn’t use machine learning.
The main issues are that it still changes correct entries to be incorrect and still suggests garbage words. At the same time, it is sometimes surprisingly unhelpful at what would seem to be the easy cases. For example, I was recently trying to type the word “didn’t” and had entered “didn”. iOS’s suggested completions were “don’t” and “doesn’t”. Other times, this same example has worked, though. I cannot predict what the predictive text model is going to do, which makes it require more attention, and it sometimes seems dumber than a simple prefix-based approach.
I do really like the new user interface, as it makes it much easier to deal with the system’s foibles.
The keyboard on iOS 17 is a nightmare. It was already going down since few releases but now it’s unbearable.
- can’t keep up if you type too fast
- key precision is worst than ever by triggering the wrong letter.
- sometimes the system takes your typing as touches instead and you are sent back on the springboard with tap triggered everywhere like it was catching up
- keyboard freezes more and more
I have not been seeing these problems, thankfully.
Previously:
- Dictation and Predictive Text in macOS Sonoma and iOS 17
- Apple’s New Transformer-Powered Predictive Text Model
- Autocorrect Explained: Why Your iPhone Adds Annoying Typos While Fixing Others
- iOS 16 Wishes
- iOS 15 Autocorrect
Update (2024-02-05): Mario Guzmán:
I am having to fucking go back and edit my text messages more than ever before due to how fucking aggressive and awful auto-correct on iOS is these days. I’ve tried resetting my dictionary but even that doesn’t help.
Update (2024-03-08): Pierre Igot:
In my experience, lots of Mac users, even if they are regular iPhone users, DO NOT LIKE autocomplete in macOS, for fairly obvious reasons. Yet, unless they happen to know what the official name for the feature is (“inline predictive text”), how are they supposed to find how to turn it off in the jungle of System Settings? The most commonly known term, “autocomplete”, yields NO RESULTS. For some reason, “autoc” and “autoco” work, but NOT “autocom” or anything longer.
Update (2024-03-11): Todd Thomas:
So what really happened to autocorrect between the iOS 17 betas where it looked amazing and the current state of things right now?