Cell Phone Autocorrect: How It Works and Why You'll Never Avoid Embarrassing Texts
The fundamental algorithm driving auto-correction software is, in fact, not much different from your run-of-the-mill spell checker; it recognizes text input, and then crosschecks completed words against a built-in dictionary. Some software will attempt to guess the word you're typing before you have completely entered it. However, Scott Taylor, vice president of mobile solutions at Nuance, testifies that building a correct dictionary and an accurate prediction system are no easy tasks. Privacy policies prohibit companies from retrieving and analyzing words that actual people have typed and sent. Instead, companies like Nuance are forced to analyze a massive pool, or "corpus" of articles from the popular media, and then create their own language model based on the articles' "structure of language [and] frequency of use," Taylor says.
While basing a corpus on the language of news is sufficient, it is limited. Rarely do people text as if their words are going to be published. To compensate, avid texters frequently have to find work-arounds to use slang in their texts. Luckily, most modern smartphone operating systems treat their users to the luxury of being able to add words to their dictionaries.
The future of auto-correcting text seems to lie in crowdsourcing. When phones can build their corpus off of actual Web language or blog posts, real texting progress can emerge. Until then, you'll have to keep up that love/hate relationship with your phone's T9, and keep guessing what the sender really meant when you get a text that reads, "You lucky duck." [From: Slate]