Journals of Autocorrection

March 2, 2014 5:08 PM

Can you point me to resources that deal with the thresholds that computers use to overwrite authorial intent? I'm interested in the boundaries between automatic correction and untrammeled thumbs.
posted by Bistle to Technology (2 answers total)

Best answer: This MeFi comment is pretty informative. I doubt there are going to be many scholarly resources on the actual thresholds involved; my guess is they sort of tune them based on feeling and have the software adjust itself over time.
posted by vogon_poet at 8:52 PM on March 2, 2014

Best answer: The comment vogon_poet linked is great. I'd add that there's no one universal algorithm for autocorrect, so different systems (and even different versions of the same system) will behave differently for a given set of inputs. Also, as noted in the linked comment, the inputs may not be as straightforward as "which key was pressed", but may be fuzzier like "what point on the screen was touched".

That said, I could see a relatively straightforward experiment to collect this data, by collecting a corpus of words (correctly spelled and not--the misspellings should probably correspond to errors common on a touch keyboard like fat-fingering the adjacent letter or omitting letters altogether) that you would then input on the system and then see what the output is and how it compares with the intended output.
posted by Aleyn at 4:27 PM on March 3, 2014