Friday, January 14, 2011
Autoincorrect from the NYT
By BEN ZIMMER
Pity poor Hannah, who received a startling text message on her cellphone, sent from her father: “Your mom and I are going to divorce next month.”
After Hannah registered her alarm, her father quickly texted back: “I wrote ‘Disney,’ and this phone changed it. We are going to Disney.”
Welcome to the world of smartphone autocorrection, where incautious typing can lead to hilarious and sometimes shocking results. With the rapid success of Apple’s iPhone and Google’s Android phones, more and more people are discovering the pitfalls of tapping on a virtual keyboard. Just as the spell-check feature in a word-processing program tries to save you from your own sloppy typing, either by politely suggesting alternatives or by automatically replacing egregious errors, the latest mobile devices are supposed to take care of your typos — but often fail with comic results.
Back in June, The Times’s technology columnist, David Pogue, blogged about some “autocorrect follies” sent to him by his readers, full of howlers like “Sorry about your feces” when “Sorry about your fever” was intended. Pogue sagely advised, “Especially when your boss, your parents or your love interest is the recipient of your e-mail or text message, it’s worth taking an extra moment to proofread.”
These vast new opportunities for social embarrassment are now being charted by the Web site Damn You Auto Correct! (D.Y.A.C. for short), where victims of autocorrect are invited to submit screen grabs of their most inglorious gaffes. Though D.Y.A.C. wasn’t the first to exploit this concept (a Tumblr feed with an unprintable twist on “iPhone” came first), it has quickly become an online sensation. Within days after Jillian Madison, co-founder of the Pophangover Network, set up the site in late October, D.Y.A.C. started getting a million daily page views, with hundreds of submissions every day. And now Madison has parlayed that success into a D.Y.A.C. book, due out in March.
Madison runs D.Y.A.C. as a one-person operation, slogging through submissions to find at least 25 new ones to post to the site every day, as well as keeping good ones in reserve for the book. “It consumes pretty much every waking moment,” she told me. “For now I’m enjoying it. It’s a crazy ride.”
Hannah’s screen grab of her father’s “divorce” message is one of the most popular (and least risqué) of the recent submissions to D.Y.A.C., but it was also hotly debated by commenters who doubted the possibility that a smartphone could autocorrect “Disney,” however badly typed, into “divorce.” In the comments, Hannah defended the exchange as authentic. “This actually happened to me and almost gave me a heart attack,” she averred, blaming the work of her father’s “fat fingers.”
The problem is that the results of such “fat-finger errors” are often not reproducible: iPhones, Android phones and other smartphones learn from the patterns of individual users so that suggested replacements are tailored to the history of a given phone, with a focus on recent and frequently used words. Since every phone develops a unique textual footprint, automatic corrections can vary from one device to another. (Madison, for her part, says that she and a pool of friends try to recreate the D.Y.A.C. submissions on their own cellphones before they are put up on the site.)
Despite these idiosyncrasies, amusement and frustration at the incorrectness of autocorrect are near-universal, particularly for users of the iPhone. Because the iPhone requires precise tapping to decline a pending suggestion, even seasoned users may miss the opportunity to pull back an autocorrected goof before sending it off. Moreover, the iPhone isn’t always adept at handling words typed with letters repeated for emphasis (a common style in text messaging). Thus yeahhhh will get changed to uranium, simply based on the proximity of letters on the keyboard.
Taboo words are another sticky subject. Designers of autocorrect word lists are clearly careful to avoid certain obscenities, even going overboard by changing hell to he’ll when it’s not warranted. And yet, as a visit to D.Y.A.C. swiftly indicates, there’s a whole universe of questionable vocabulary that can slip through unbidden. To take a relatively benign example, an invitation to a “boardgame night” got changed to a “bisexuals night” — because the iPhone’s dictionary included board game only with a space in the middle, and bisexualshappened to fulfill the “fat finger” proximity criteria.
All of this recalls an earlier era of computer-aided miscorrections. When the 1997 edition of Microsoft Word introduced its background spell-checker, some of the on-the-fly substitutions were a little off the mark. Most notoriously, cooperation was rendered by the autocorrect feature as Cupertino, since the spell-checker dictionary recognized co-operation only with a hyphen. When translators for the European Union started noticing the name of a Northern Californian town (coincidentally, the home of Apple Inc.) creeping into their documents, they coined “the Cupertino effect” to describe such unwanted spell-checker changes.
Microsoft’s Natural Language Processing group subsequently tinkered with its algorithms to make sure that only truly obvious errors are autoreplaced (like teh for the), and it has also continually expanded the spell-checker dictionary to keep up to date. (Instant updates helped when the first release of Office 2007 didn’t contain Obama, unfortunately recommending Osama in its place.) Now much of that R.&D. has been repurposed for Microsoft’s own mobile operating system, Windows Phone 7, a direct competitor to the iPhone and Android platforms. The Windows phone won’t spare us from the cellular counterpart of the Cupertino effect, however. Errant thumbs and fickle spelling will keep autocorrect developers guessing for a long time to come.
Ben Zimmer will answer one reader question every other week. Send your queries to email@example.com. You can follow Mr. Zimmer on Twitter at twitter.com/OnLanguage and Facebook.