Accessible bowdlerization

I’ve been experimenting with screen readers as part of my research on creating accessible documents for people with print disabilities. Popular screen-reading programs include

• VoiceOver, which comes free on a Mac;
• NVDA, which is free to download for Windows; and
• JAWS, a Windows program that costs $179 for a ninety-day licence or $895 for the home edition.

Of course, because I have the mentality of a twelve-year-old, the first thing I did was run VoiceOver on the Strong Language tag cloud, which the software recited with aplomb.

Screen readers have come a long way in the last decade and a half. No longer are they monotonous drones that robotically spit out one syllable after another, although VoiceOver is noticeably better than NVDA’s default voice in this regard. (I tested these two free programs only.) They have much more realistic prosody, with appropriate pauses for punctuation like commas and periods and rising inflection for questions. In short and simple sentences, both VoiceOver and NVDA have no problem distinguishing use as a noun and use as verb, for example, and they adjust their pronunciation accordingly.

But how would they fare in other sweary and not-so-sweary contexts? (Incidentally, VoiceOver and NVDA both pronounce sweary as [swiɹi]—“swee-ry”—rather than [swɛɹi].) Different screen readers react differently to certain symbols and words, so if you have to censor, what flavour of bowdlerization is most accessible?

The grawlix

The grawlix—or the obscenicon, as Ben Zimmer calls it—is perhaps worst of all. VoiceOver reads @#&*! as “at-number-and-asterisk,” whereas NVDA reads it as “at-and-star,” both of which would be quite confusing to someone who wasn’t expecting a censored swear word.

The dash

Neither VoiceOver nor NVDA reads the dash (or hyphens in place of each missing letter) out loud, so bowdlerizations like s— or c— are read as “ess” and “cee.” Bulls— is pronounced “bulls.” With the exception of fuck rendered as f—, which we readily accept because the letter is so commonly used in place of the word, this method of censorship can be very confusing for people who use screen readers, because it gives no audible indication that something has been taken out.

The asterisk

VoiceOver reads * as “asterisk,” whereas NVDA reads it as “star”—so on a Mac, “I don’t give a f***” sounds like “I don’t give a f-asterisk-asterisk-asterisk.” This option is better than the dash, because it tells the reader that letters have been censored, although “f-asterisk-asterisk-asterisk” is far from natural conversation. The reader would also have to keep track of the number of asterisks to distinguish bitch‘s b**** from bastard‘s b******—a distraction from the content of the text.

[Expletive deleted] or [redacted]

With this method of censorship, VoiceOver is better than NVDA at conveying the fact that obscenities have been deleted, pausing at the square brackets. That pause makes all the difference in comprehensibility: I don’t give a [redacted] as “I don’t give a [pause] redacted” is much clearer than “I don’t give a redacted.” Because the screen readers have different aptitudes on this task, it’s best to opt for a more accessible option.

The substitution

We at Strong Language have been known to ridicule the New York Times for its tortuous attempts at swear avoidance:

"'You have to have some stuff to do,' she said (she didn’t say 'stuff'…)" NYT tying itself up in knots of avoidance: http://t.co/R02AKXBb1g

— Strong Language (@stronglang) February 19, 2015

Thanks, @NYT ban on profanity: "And not everyone has the guts to do that (although she didn't use the word guts)." http://t.co/u0aSgN6quB

— Mededitor (@Mededitor) March 22, 2015

Because this method uses full words, though, it is a more accessible form of bowdlerization than dashes, asterisks, or grawlixes.

The initialism

Hiding swears words in abbreviations like WTF or FFS is relatively accessible. Both VoiceOver and NVDA will read them one letter at a time, and it’s not too hard to piece together what the text is trying to convey. Both also read the acronyms snafu and fubar without issue.

The misspelling

This method is used less to soften strong language and more to get past automated censors. VoiceOver and NVDA both pronounce fuk and dik without a problem. Mincing shit as shite is fine, but bitch as biatch produces the hilariously mispronounced [bajətʃ] (“bye-etch”) on both platforms. Test out your spelling before letting it loose.

No bowdlerizing at all

It comes as no surprise that this option is by far the most accessible, albeit the most potentially offensive. Interestingly, although screen reader prosody is better than it used to be, it isn’t perfect, so expressions made to sound almost-but-not-quite offensive, including sofa king (“This is sofa king interesting!”) and the name of Shrek villain Lord Farquaad, sound like straight-up swears (especially if you use the British English voice). Neither VoiceOver nor NVDA will touch retard as a noun with a ten-foot pole, insisting on putting the stress on the second syllable regardless of its function in a sentence.

Playing around with the screen readers has given me a bit more insight into how people with print disabilities interface with the sources of information the rest of us take for granted. I urge you to do your own experiments, and please do get in touch if you ever figure out why VoiceOver has such a fucked-up pronunciation of cumquat.