The previous Damn You Autocorrect FPP had a really, really good explanation of how autocorrect works.Oh god that explanation, I think, got sidebared but it was actually terrible. It was a complete guess and assumed, for whatever reason, that including an entire dictionary of words was somehow 'too big' to fit on an iPhone, even though it could fit on a 386 PC with a 100Mb hard drive.
So Autocorrect here is taking a very different approach. Rather than ask, "Is that a real word, yes or no?" it asks, "How probable is it that that's a word? How word-like is it?"First of all, that's just a complete guess, but he states it as a fact. A fact that happens to be wrong and then he goes off on some total nonsense about hidden markov models which is just ridiculous when it comes to simple spellcheckers. You would never write a spell checker without using a real dictionary file, because there are all sorts of strangely spelled words in the English language. Typically you use HMMs on entire words not letters. You can use HMMs to generate 'gibberish' words that look like they could be real, though.
dcameron: Go to hell eat my fsckn dickposted by XMLicious at 9:58 PM on December 10, 2011 [15 favorites]
amerkel: that's it. no deal.
dcameron: OMG I meant got to help the EU with fiscal discipline
« Older Are sex offenders and lads’ mags using the same la... | Bloody hockey cards celebrate ... Newer »
This thread has been archived and is closed to new comments
posted by ericb at 3:01 PM on December 10, 2011 [1 favorite]