Autocorrect function in word processing software
|Submitter Email:||click here to access email|
Last spring, I sent the following inquiry to Linguist:
>I'm looking for information (articles, white papers) on the >autocorrect
>technology built into most word processing software (e.g., Microsoft
>I want to understand (in broad terms) the software engineering side of
>things (how does the technology work?), but just as importantly, the
>theoretical side (e.g., classification of errors [haplography,
>metathesis, etc.] according to current thinking on the physiological
>mechanisms involved in copying/transcribing/typing (perhaps this sort >of
>falls within the purview of visual word recognition?).
I received two off-list replies, one from Mike Maxwell, the other from
Robert Dale provided a bibliographic starting point:
> There's a paper by the Word Grammar Checker team (principally George
> Heidorn) in R Dale, H Moisl and H Somers (eds.) , Handbook of
> Language Processing. Marcel Dekker. I believe that's the most detailed
> exposition around.
> I don't know of any material on aspects of text correction in Word other
> than the grammar checker -- the spelling checker, autocorrects etc are
Mike Maxwell initally wrote to say that the replace-text-as-you-type feature
in Word relies on a hand-built correction list, which one can consult
by selecting "Autocorrect" from the "Tools" pull-down menu:
>My understanding of Autocorrect, is that all there is to it is a list >of
>"from" and "to" spellings: misspellings and corrections, or >abbreviations
>and their translations ("(c)" gets changed to the copyright symbol). >So
>when you type a space or punctuation, there is presumably an action >that
>Word performs in the background: it selects the preceding word (as
>by whitespace and/or punctuation), and looks on the "from" field of >the
>records in its correction list. (You'll notice that list is >alphabetized,
>so lookup is fast.) If Word finds a match, it substitutes the "to" >field
>from that same record. I think it's also smart enough to copy the
>capitalization from your word into the substituted word.
He later wrote back to partially amend that statement:
>It seems to
>have at least the ability to automatically correct metathesized >letters
>certain omitted letters. I say that because I've seen it fix words >that
>sure it doesn't have in its correction list (particularly linguistic
>I just watched it correct 'alolmorph' and 'alomorph' to 'allomorph').
>Presumably it's using its spell checker for a list of correct words,
>together with some notion of common errors. For example, it corrects
>'alomorph' but not 'alloorph', so it seems to know that writing a >single
>for a double 'll' is a common mistake, while omitting an 'm' is not.
>In sum, it's smarter than I thought.
My thanks to Robert Dale and Mike Maxwell for their helpful responses. I'd
be interested in any additional information others could provide.
University of Rochester
|Original Query:||Read original query|
Sums main page