Predictive text systems work by looking at the letters already keyed and finding possible matches for a complete word, then offering a way to select from the possible word completions.
On devices with a standard phone keypad, these systems also need to disambiguate - each key press may represent any of three or four letters, and systems need to decide which combinations of these letters might represent a word.
Using predictive text with a phone keypad, some words are keyed identically and so cannot be distinguished
Around one in eight common words are confused in this way. So the user needs to check each word to see if it is guessed correctly and if not, needs to select from alternatives.
Predictive text systems can only work with words on their list. Most proper names (people, places etc), technical terms and abbreviations will be missing and need to be added manually (typically using Multitap).
Textra allows words from multiple languages to be easily mixed, including accented letters, and allows any text abbreviations to be keyed. Textra is ideally suited to web and email addresses, which cannot be entered using predictive systems.
Currently, most predictive text systems have to concentrate on disambiguating and rarely on word completion.
Using Textra, every letter and number is just one press, so disambiguating is not required. Consequently, predictive word completion can be better and the number of choices reduced. Moreover, all the choices are relevant as they start with the letters you really have keyed and not letters that you just might have meant.
Indeed, prediction systems could be further improved to predict common next words.
