Apple's iOS Predictive Text Uses 'Grey List' to Restrict Autocompletion of Over 100,000 Phrases

Discovery of the 'Grey List'

Independent researcher Derik Kauffman uncovered a hidden 'grey list' within Apple's iOS operating system. This list contains over 100,000 words and phrases that Apple's predictive text and swipe-typing features will not autocomplete for users. Kauffman found the encoded table after noticing consistent issues when trying to type specific phrases on his iPhone.

Restricted Content Categories

The identified list includes a range of content. Unsurprisingly, it restricts terms related to illegal drugs and sexual acts. However, it also extends to phrases like 'iMessage isn’t working', and numerous politically charged terms. Examples include 'Dictator Trump', 'NSA spies', 'Trayvon Martin', 'Kashmir dispute', 'Muslim extremist', and 'male chauvinist pig'.

Language-Specific Restrictions

The restrictions also apply to other languages. In Hebrew, the phrase 'the occupied territories' is not autocompleted. Similarly, in Arabic, 'the Hamas movement' is on the restricted list. This indicates a broader application of content filtering across different linguistic contexts within iOS.

Impact on User Experience

When users attempt to swipe-type these restricted phrases, the system often stumbles, leading to incorrect suggestions or requiring manual input. For instance, 'United Ireland' might become 'United Iceland', and 'Tibetan independence' could be changed to 'Tibetan independences'. This forces users to spell out the full phrase if it is on the grey list.

Context of Autocorrect Evolution

This discovery provides insight into the underlying mechanisms of autocorrect technology. Early versions of autocorrect, like those in Unix, relied on simple word lookups. Modern systems, including Apple's, utilize n-grams to statistically predict subsequent words, making the deliberate exclusion of specific phrases a notable intervention in this predictive process.