Umm
"Current models of predictive text"
Well, SwiftKey doesn't...
An Apple patent application just published by the US Patent Office shows that the masterminds of Cupertino have figured out (or hope to figure out) a cunning way to work out what words you're going to tap into your little Jesus mobe before you type it. The US Patent Office has not yet granted the patent, but in accordance with …
The patent was filed in 2010 if you actually read the patent. Swiftkey came out in 2010. So it's not like they sat around and saw its popularity grow and decided to patent it.
Even if Apple did develop something quickly and patented it quickly there is still a good chance it was in progress before Swiftkey was released.
Fandroids may be all butthurt about it but the only thing it shows is this idea, like most patents, aren't unique ideas which is a case against software patents as a whole But as long as they allow software patents this is goal two groups worked towards and one clearly got to the patent office before the other so it sucks to be the other guy but that's how it works in many cases.
AIUI Swift Key is a Bayesian predictor based on letter inputs. This is a Bayesian predictor based on the exact places you touch the screen when typing. Neither Apple nor Swift Key invented Bayes' Theorem, AFAIAA, and since the inputs to the algorithm differ they are not the same thing. You may as well say they're all ripping off Paul Graham (A Plan for Spam). He definitely did invent Bayes' Theorem, that's why it's called ... oh, wait.
Predictive Text is just about the worst thing on a device. Try using Russian names with it on and see what you get. My Wife's name gets totally messed up. She knows when peole send her messages with it in. Android PT gives different resuls on some words from an iPhone PT.
Do not want. Fail.
and using that to predict the errors that will be generated?
I've got news for them; main stream spelling checkers right back to the very first do this (generally on one or both of the assumptions that 'the user can spell but can't type' or 'the user can type but can't spell') to derive the best order to present corrections.
I did it myself in a dissertation a couple of years ago investigating better correction for OCR.
The internet has many dialects. The /me would be familiar to any user of IRC, while the reference to preening only makes sense to a particular community. Outside of IRC, it's also acceptable to indicate third-person poses by using * delimiters on each end, while * delimiters on a single word are used for emphesis.
Appologies. I guess I should be more clear on where I vent my annoyance. This is just another case where Apple's patent grabbing little hands find another thing they can patent through careful tweaking of something that practially already exists, then they make a big fanfare about how they have a new tech or whatever and all the fanbois squeal and hop about happy. Siri, I'm looking at you
iOS autocorrect doesn't even keep track of cursor relocations. If one moves the cursor back to manually correct something (typically an autocorrect gone wrong), whatever letters you type IN THE MIDDLE OF AN EXISTING WORD (!) are examined without regard to the context (!) and then autocorrected to the point where several words are deleted or overwritten in the process. What amazes me is that the SW Machine that is Apple fails to notice this sort of glaring bug.
Android - many of my coworkers have Android phones. They seem to have many more horrible SW bugs than does iOS. The high points are good (great even), but it seems to be an ecosystem that lets more serious bugs loose. It's great to have alternatives to meet all needs. I'd love to have an Android phone, but I'm not wanting another monthly bill. Perhaps I'll grab a pure Android tablet (as opposed to PlayBook) when they're on sale. Cheers.
Check and see if they're running 2.3. The vast preponderance of Android phones currently out there are on it because the carriers and manufacturers refuse to pull their finger out. Around 2.3.4 or so either Google or whomever baked the carrier's custom ROM were screwing around with how text entry worked... It got better in 4.0, or at least with the Nexus reference device it did. Also, see if they have Swype or not. If not, there's the problem right there.
If you want an Android phone, see if you can score the Galaxy Nexus GSM unlocked version? Google just dropped the price on it a bit, pop your sim in it or the iPhone for the day, swap at lunch, etc...
The main problem with autocorrect, imho, is that the dictionaries are vastly huger than the optimal size. After about 3 autocorrect muff-ups I swyped 'this is ridiculous' only to end up sending 'this is Tuscaloosa'.
Similar problems with spellcheckers in word-processors or browsers. It's much better to question a new word (with the option to add to the user's personal dictionary) than to consider it a correctly spelt obscure or archaic word. Only crossword-solving dictionaries need to be big.
There's loads of research out there which suggests that the optimal for a spelling correction dictionary is one that contains only the words in the words used in the corpus. There's an obvious issue with that, but the way I do it is to use fairly small (90k words or so) dictionary and analyse the text under correction to identify words which both look as if they are following the same general spelling rules of the language used and which are frequently used.
There's a genre of disastrous bad texts apparently (if genuine) generated by the device remembering that you used a very rude word before, and you seem to be using it again.
And I don't suppose that if your cockatiel bird is a male, a cock, that the finger movement for the first four letters is different.
Of course they are trying to avoid you sending a word that you didn't intend to write, so it's near-miss typing that's the issue - making the bad letter a smaller target, maybe.