Y'all have a Texas accent? Siri (and the world) might be slowly killing it
Y'all have a Texas accent? Siri (and the world) might be slowly killing it
Voice recognition tools such as Apple’s Siri still struggle to understand regional quirks and accents, and users are adapting the way they speak to compensate
By Tom Dart in Houston Wednesday 10 February 2016 10.30 EST
It was a simple enough question, at least in this part of the world.
“How can we mosey on down to the rodeo?” my friend Ben Crook drawled, sat in a rocking chair on his front porch, a can of Lone Star beer in his left hand on a humid night in Houston.
Only one thing jarred with this otherwise stereotypical Texas scene: Crook was asking Siri, the voice-activated digital personal assistant on his iPhone, rather than, say, a passing sheriff on horseback with a cowboy hat wider than the Buffalo Bayou.
Siri understood the individual words but didn’t know how to respond. But Crook had other questions. He was hungry; heck, so hungry he coulda eaten the north end of a southbound billy goat.
“We’re fixin’ to eat brisket, where should we go?” he asked Siri. She offered a list of 15 restaurants – though not all appeared to serve Texas barbecue. Siri was also helpful when asked where to find crawfish, but baffled about kolaches, the pastries of central European origin that are hugely popular in Texas, calling them “Colotchies”.
Meanwhile, though the free Dragon dictation app performed admirably when fed lines from the 2004 movie The Alamo, it did turn “Davy Crockett feller” into “David Rockefeller”, and evoking a family of Yankee industrialists is no way to describe a hero of the battle for Texas independence.
The upshot of this brief and decidedly unscientific experiment is that Siri is at her best when addressed in standard English, with accents toned down and slang avoided where possible.
The writer Julia Reed came to a similar conclusion in an essay for the latest issue of the southern lifestyle magazine Garden & Gun, when she turned to dictation apps after breaking her left elbow in New Orleans. She wrote:
Like the iPhone’s highly temperamental Siri, Dragon and the rest of the dictation apps I tried steadfastly refused to understand pretty much everything I had to say. Apparently none of [Dragon’s] coders have spent a natural minute below the Mason-Dixon Line. A smart person could make a lot of money by inventing a Siri for Southerners.
(An Apple spokesperson directed the Guardian to a Siri user guide and a list of the countries and languages where it is available).
“I’ve had a bunch of people from Australia and India say they only really get along with Siri if they fake an American accent,” said Lars Hinrichs, a sociolinguist at the University of Texas at Austin.
Most people have what we would call a telephone voice … they also have a machine voice
Apple has refined and developed Siri since the tech giant bought it from a small startup in 2010. It is improving fast. But while voice-recognition programs will continue to evolve when hearing non-standard speech such as regional accents, a question that neither human nor machine can answer for certain is whether they will need to. People are adapting to virtual assistants, as well as the other way round.
“Most people have what we would call a telephone voice, so they actually change away from their local family accent when they’re speaking on the telephone to somebody they don’t know,” said Alan Black, a Scottish computer scientist who is a professor at the Language Technologies Institute at Carnegie Mellon University in Pittsburgh.
They also have a “machine voice”, he said. “People speak to machines differently than how they speak to people. They move into a different register. If you’re standing next to somebody in an airport or at a bus stop or something, you can typically tell when they’re talking to a machine rather than talking to a person.”
Black speculated that “one of the reasons they designed Siri to be fundamentally a polite, helpful agent who isn’t your friend but works for you, is to encourage people to be somewhat polite and explicit to her, rather than being very colloquial. Because speech recognition is always hard when you drop into colloquialisms.”
With speech recognition ever more widespread and efficient, our younger generation will grow believing chats with Siri and the new Amazon Echo are routine and genuinely useful; a far cry from when calls to utility companies became stilted shouting matches with machines that had trouble understanding “yes” and “no”, never mind “put me through to a real person, for God’s sake”.
But aside from Siri, proud Texans should worry about something else.
Long before machines who could understand you, cultural and demographic shifts were already moving people towards standardized English. In fact, mass media and migration are slowly killing the Texas twang.
“The way young people in Dallas or Houston speak nowadays is a lot closer to a regional common denominator accent than to what it was 50 years ago,” said Hinrichs, who is originally from Germany and directs the Texas English Project. “I never hear any of my students sound ‘Texan’ in class anymore; but they can when they go home. The accent modularizes because people are more mobile and connected with the world.”
In effect, Texans are using the “telephone voice” in everyday life, partly thanks to the effects of TV and social media and partly because the influx of arrivals from around the country and overseas, so that everyone can understand each other. It is called “accent levelling”.
Quirky expressions such as 'doggone it', 'shucks' and 'drat it' are dying out, replaced with more mainstream dialect
“Vocabulary is the first thing to go. Then syntax and pronunciation,” he said. Double modals such as “might could” and “oughta should”, and quirky regional expressions such as “doggone it”, “shucks” and “drat it” are dying out, replaced with more mainstream dialect and an accent often described as “midwestern”.
Hinrichs and some colleagues conducted an experiment to see if Texans could tell west Texans and east Texans apart.
“Basically the answer’s no,” he said – even though in such a vast state, landscapes and lifestyles differ widely depending on location. “People from Texas cannot hear the difference, but most will tell you they can.”
Still, he suggested, while globalization might make life easier for voice-recognition programmers, flexibility also poses challenges. “A frontier for speech processing is to deal with multilingualism – switching between English and Spanish,” he said. About 40% of Texas’ population is Hispanic.
Black thinks that in coming years, programs such as Siri will go from being aloof in style to more familiar, understanding your language patterns as if they were a close friend rather than a casual acquaintance.
“Dialogue systems at the moment work pretty well, speech recognition has got substantially better,” he said. “I think what’s probably going to happen is a much more long-term rapport. It will know more about you. It will be able to answer the question sort of before you ask it – this is one of the things that Google Now’s aim is, answer the question before you actually ask. You’ll find that you can be less specific when you’re talking because it will know the sort of things that will be relevant. If you ask the time, the machine might say something like, ‘it’s OK you’ve still got three minutes before your meeting’, because it knows that you ask the time when you’re worried about the meeting, that’s what you always do.”
In other words, the future holds less southern charm, but fewer problems getting to the rodeo.