Study: Predictive Text Slows Users Down

Predictive text may not make typing messages on phones quicker or easier according to a new study. In many cases, the study says, it will slow things down.

The research comes from Pe Ola Kristennson, a professor of interactive systems engineering at the University of Cambridge, and colleague Thomas Mullners. (Source: acm.org)

They previously ran a study in 2019 asking 37,000 volunteers to copy sentences on a mobile phone. Those who used predictive text averaged 33 words a minutes, slightly slower than the 35 words a minute of those who didn't use any technologies. (Source: theconversation.com)

The comparison was with predictive text, where the phone shows between one and three suggestions and the user can tap one to automatically insert it in the message. Both speeds were well behind the 43 words a minute of people who used autocorrect where they type words in full and the phone corrects anything it detects as a typing error.

Checking Suggestions Slows Users Down

With the results somewhat unintuitive and the figures so close, the pair decided to explore whether the advantages (or lack of) of predictive text were consistent or if they varied in different use cases.

To do this, they gathered data from previous research showing the average user takes 0.26 seconds to press a key (physical button or touchscreen key) and 0.45 seconds to check the suggested predictive text terms and select one.

They then explored the ways people actually use predictive text. Whether or not they do so consciously, many people use one of two main strategies. One is to type shorter words in full and only check for predictive text selections for longer words. The other is to type a certain number of letters each time, then check the suggestions.

A third variable was how many times users were prepared to see incorrect updated suggestions for each word before they stopped looking and simply typed the word in full. However, this proved harder to account for in the study.

Most Cases Bring Little Benefit

Crunching the numbers suggests the biggest benefit from predictive text comes when people use it for words of at least six letters and type the first three before checking the suggestions.

The problem is that this sweet spot is a very narrow range. In most cases either a word is so short it's quicker to type it out in full than check predictive text suggestions even once, or a word is long enough that users need to type several characters before the suggestions become accurate enough to be useful.

What's Your Opinion?

Do you use predictive text? If so, how often do you glance at the suggestions? Is autocorrect reliable enough to ignore predictive text altogether?