- iOS is changing “Trump” to “racist” when transcribing
- Apple says the bug is now being fixed
- The official explanation is “phonetic overlap”
iPhone owners have noticed a peculiar bug in recent days: “Trump” autocorrects to “racist” when using speech-to-text dictation mode. According to Apple, it’s a problem with “phonetic overlap”, and a fix is already in the works.
After TikTok videos of the slip went viral, Apple provided a statement to The Guardian and others, blaming “phonetic overlap” between the two words: “We are aware of an issue with the speech recognition model that powers Dictation, and we are rolling out a fix as soon as possible,” a spokesperson said.
While many people were able to recreate the blip, it didn’t happen every time – and the text seemed to revert back to “Trump” after a short delay. The latest reports online suggest Apple’s fix has already taken effect, so you might not see it happening any more.
In its explanation, Apple suggested its speech recognition engines were struggling to distinguish between words with “r” in them. Further testing suggested iOS didn’t always get the word “racist” right either, though historically Apple’s speech-to-text engines have been very reliable.
‘Just not plausible’
Apple will be keen to draw a line under this as soon as possible and get the error corrected. It seems particularly unfortunate that a transcription bug like this would link two specific words sure to set off a wave of controversy and politically-charged debate.
Peter Bell, professor of speech technology at the University of Edinburgh, told the BBC that Apple’s explanation was “just not plausible” given what we know about speech-to-text technology. “It probably points to somebody that’s got access to the process,” said Bell.
John Burkey, founder of Wonderrush.ai, gave a similar option to the New York Times: “This smells like a serious prank,” he said. “The only question is: did someone slip this into the data or slip into the code?”
This also feeds into the wider conversation about AI and its reliability, as AI models are used to convert the spoken word into transcribed text – something that you can now do on any modern smartphone. Whether it’s meeting notes or show subtitles, we need to be able to rely on the accuracy of this fast-spreading technology.
Leave a comment