LAS NUECES, Guatemala =-- The United States government separated their family at the border, leaving them with an agonizing choice.
José Ottoniel was deported to Guatemala in June, a month into President Donald Trump's "zero tolerance" crackdown. But his 10-year-old son, Ervin, who made the journey with him, remained in Texas.
Now, back in this hilltop village, José and his wife, Elvia, need to decide what to do with their son, who is at a migrant shelter 1,700 miles away.
While the U.S. government scrambles to reunify migrant families separated at the border, some parents, such as the Ottoniels, think that the best option for their children might be the thing they most dread - to remain apart.
"It's not that we don't love him," José said. "It's that we want him to have a better chance at life."
José and Elvia are pushing for Ervin to remain in the United States - away from the crushing poverty of his birthplace. Elvia has a cousin in Arkansas who agreed to take him in. The couple explained the situation to Ervin on the phone. They hung up, and they cried.
Ervin Ottoniel was the top-ranked third-grader at the village's elementary school. He drew pictures of himself holding a laptop. He told his parents he wanted to be a lawyer. They told him they couldn't afford his schooling beyond sixth grade. His father earns $21 a week.
The Ottoniels know that if Ervin returns to Las Nueces, his life would be a foregone conclusion - sporadic work on a coffee plantation, helping pay off his father's debts. But if he stays in the United States, he might not see his parents or siblings for years.
"Right now, we think it's best for him to have this opportunity in United States, to get out of this place," José said.
Other families are making similar calculations. Although they never planned to leave children alone in the United States, the White House policy of separating families, along with the swift deportation of some parents, has forced the question: What's best for a separated child?
Immigration lawyers estimate that between 180 and 400 parents have been deported without their children since Trump's "zero tolerance" policy began in May. Now, legal-aid organizations are reaching out to those migrants to see whether they would like their children to return home.
"This is uncharted territory," said Wendy Young, the president of Kids in Need of Defense (KIND), an organization that works with immigrant families. "Some of these parents are living in communities where there is no protection for their children, where they have no choice."
That's the case for Ana Lopez, who lives in El Carmen, a town about 50 miles from the Ottoniels' home. She is trying to find a way for her son, Endil, separated from his father at the border in June, to remain with his grandparents in Maryland. For months, she said, the family in Guatemala has faced threats from a local criminal group.
"How can I bring him back to a place where it's too dangerous for him to attend school?" Lopez said through tears.
In Las Nueces, a town that traces a dirt road up the side of a mountain, where almost every man is a miner or a farmer or unemployed, the question of what to do with Ervin Ottoniel is discussed almost everywhere. His parents asked the town's priest, the Guatemalan consulate, the school's principal and their own parents for advice.
"We believe this is what's best for him, but not everyone agrees," Elvia said.
"It's 50-50," José said.
As soon as they leave the house, the question comes up.
"So what's going to happen?" asked Walter Lemos, one of José's uncles, at his home one afternoon.
"Even the boy prefers to stay in America," José responded. "He knows there's more for him there."
"But there's no love like a parent's love," Lemos said under his breath, shaking his head.
José stopped attending school after third grade to work on a farm. Elvia dropped out after second grade to look after her younger siblings. They named Ervin after an engineer José had worked for in a local silver mine, the best-educated person he'd ever met.
Public school is technically free here, but parents have to pay for books, uniforms, materials and teacher fees - costs that can amount to around $150 per year.
"I already told Ervin that we won't be able to afford that for much longer," José said. "He became very angry. He's a very motivated boy."
Traveling to the United States was an imperfect solution. José paid $7,500 to a smuggler, most of it borrowed from a bank. The plan was for José to work and for Ervin to study. Elvia and the other three children would remain in Las Nueces until, they hoped, José and Ervin could get some kind of legal status, so they could move back and forth freely.
They didn't know about the Trump administration's "zero tolerance" policy, which involved criminally prosecuting all illegal border-crossers and removing the adults from their children. When they were separated in early June, José lied to Ervin so he wouldn't start crying. "They're taking you to school, and they're taking me to work," he said. It was the last time they saw each other.
José was taken to an immigration detention center in south Texas, where agents told him that if he didn't agree to be promptly deported back to Guatemala, he would be detained for as long as six months during legal proceedings - without seeing his son. He could work on his son's case from Guatemala, they said, and seek to have Ervin remain in the United States or to return home.
Back in Las Nueces, José and Elvia hear twice a week from Ervin, on Tuesdays and Thursdays. Each call lasts 10 minutes. José is never sure what time the calls will come, so he sits outside his squat house, between two yucca trees, where there is a phone signal, waiting.
When Ervin calls, his voice is faint.
"It's like he doesn't have the energy he used to have," José said. "It's like he's weaker or something."
"We asked him his opinion," Elvia said. "What does he want to do? He said he wants to stay in the U.S."
"But we know it's a lot for a boy to take on," added José.
The logistics of keeping Ervin in the United States are complicated. He has been reclassified as an unaccompanied minor. The Office of Refugee Resettlement (ORR), which deals with such youths, could vet his cousins to make sure they are capable of caring for Ervin. He could then apply for legal status.
Lawyers across the United States are struggling to figure out how to handle these cases, and some worry that the children could be deported even if their parents protest.
"What we would like to see is for the child and parents to have an opportunity to speak to attorneys, and then for there to be a consultation so that the family can make a decision," said Young, of KIND.
One of the challenges in Ervin's case is that his cousin in Arkansas is undocumented. He paints houses and does construction jobs, earning about $3,000 a month. He has two small children, both American citizens by birth. In the past, immigration authorities have been willing to release immigrant children to relatives in the country who do not have legal status.
"As a father, I know it's difficult to not to see your child for a long time, but I've lived in America for 10 years, and I can tell you that life is better here for Ervin," said the cousin, who spoke on the condition of anonymity because of concerns about immigration enforcement.
Migrants have been leaving Las Nueces for decades, sending back money that helps build the largest homes and buy the nicest trucks in town. In a place with little well-paid work, José is surrounded by proof that traveling to the United States is the only guarantee of a decent life.
Earlier this month, he called the smuggler who took him to the United States in June and asked whether he would consider helping him get back to the border, given the failure of the first trip, given Ervin's separation.
"He refused," José said.
He still owes $4,000 that he borrowed to pay the smuggler, a loan he can't imagine paying back, even though he knows the bank might eventually seize his home. A recent study on migration from Guatemala by scholars at the University of Arizona said that such debts could be pushing thousands into homelessness.
"What can I provide to my son here?" he asked.
Once a week, a group from the local church comes to pray for Ervin. The congregants take turns putting their hands on Elvia's forehead. When they arrived on a recent Sunday, she was wailing.
"Ervin, wherever he is, only you can protect him, God," the group intoned.
When they left, Elvia held her youngest son, 1-year-old Dilan, her eyes still red.
"It's impossible to know what's best for our son," she said.
A neighbor, Maria Segura, stopped by the house.
"He's an exceptional child," she said. "But how is he going to be happy without his parents?"
Elvia didn't answer. Later, she pulled out mementos, forming a small shrine to her oldest son.
A photo of Ervin standing next to the Guatemalan flag.
A photo of him wearing a tie, arms linked with a girl in a tiara.
The sash, neatly folded, that he got to wear for being first in his class.
"He's a special child," Elvia said.
Then she pulled out the piece of paper U.S. government employees had given José before he was deported.
"Are you looking for information about a child who has arrived in the United States?" it said in Spanish.
Below it was a 1-800 number she couldn't afford to call.
José returned from an afternoon of work on the coffee fields. The other children were picking up the photos of Ervin, looking at them quietly.
"We know they miss their brother," José said. "I don't know what to tell them."
The Washington Post's Gabriela Martinez in Mexico City contributed to this report.
When Meghan Cruz says "Hey, Alexa," her Amazon smart speaker bursts to life, offering the kind of helpful response she now expects from her automated assistant.
With a few words in her breezy West Coast accent, the lab technician in Vancouver gets Alexa to tell her the weather in Berlin (70 degrees), the world's most poisonous animal (a geography cone snail) and the square root of 128, which it offers to the ninth decimal place.
But when Andrea Moncada, a college student and fellow Vancouver resident who was raised in Colombia, says the same in her light Spanish accent, Alexa offers only a virtual shrug. She asks it to add a few numbers, and Alexa says sorry. She tells Alexa to turn the music off; instead, the volume turns up.
"People will tell me, 'Your accent is good,' but it couldn't understand anything," she said.
Amazon's Alexa and Google's Assistant are spearheading a voice-activated revolution, rapidly changing the way millions of people around the world learn new things and plan their lives.
But for people with accents - even the regional lilts, dialects and drawls native to various parts of the United States - the artificially intelligent speakers can seem very different: inattentive, unresponsive, even isolating. For many across the country, the wave of the future has a bias problem, and it's leaving them behind.
The Washington Post teamed up with two research groups to study the smart speakers' accent imbalance, testing thousands of voice commands dictated by more than 100 people across nearly 20 cities. The systems, they found, showed notable disparities in how people from different parts of the U.S. are understood.
People with Southern accents, for instance, were 3 percent less likely to get accurate responses from a Google Home device than those with Western accents. And Alexa understood Midwest accents 2 percent less than those from along the East Coast.
People with nonnative accents, however, faced the biggest setbacks. In one study that compared what Alexa thought it heard versus what the test group actually said, the system showed that speech from that group showed about 30 percent more inaccuracies.
People who spoke Spanish as a first language, for instance, were understood 6 percent less often than people who grew up around California or Washington, where the tech giants are based.
"These systems are going to work best for white, highly educated, upper-middle-class Americans, probably from the West Coast, because that's the group that's had access to the technology from the very beginning," said Rachael Tatman, a data scientist who has studied speech recognition and was not involved in the research.
At first, all accents are new and strange to voice-activated AI, including the accent some Americans think is no accent at all - the predominantly white, nonimmigrant, nonregional dialect of TV newscasters, which linguists call "broadcast English."
The AI is taught to comprehend different accents, though, by processing data from lots and lots of voices, learning their patterns and forming clear bonds between phrases, words and sounds.
To learn different ways of speaking, the AI needs a diverse range of voices - and experts say it's not getting them because too many of the people training, testing and working with the systems all sound the same. That means accents that are less common or prestigious end up more likely to be misunderstood, met with silence or the dreaded, "Sorry, I didn't get that."
Tatman, who works at the data-science company Kaggle but said she was not speaking on the company's behalf, said, "I worry we're getting into a position where these tools are just more useful for some people than others."
Company officials said the findings, while informal and limited, highlighted how accents remain one of their key challenges - both in keeping today's users happy and allowing them to expand their reach around the globe. The companies said they are devoting resources to train and test the systems on new languages and accents, including creating games to encourage more speech from voices in different dialects.
"The more we hear voices that follow certain speech patterns or have certain accents, the easier we find it to understand them. For Alexa, this is no different," Amazon said in a statement. "As more people speak to Alexa, and with various accents, Alexa's understanding will improve." (Amazon chief executive Jeff Bezos owns The Washington Post.)
Google said it "is recognized as a world leader" in natural language processing and other forms of voice AI. "We'll continue to improve speech recognition for the Google Assistant as we expand our datasets," the company said in a statement.
The researchers did not test other voice platforms, like Apple's Siri or Microsoft's Cortana, which have far lower at-home adoption rates. The smart-speaker business in the United States has been dominated by an Amazon-Google duopoly: Their closest rival, Apple's $349 HomePod, controls about 1 percent of the market.
Nearly 100 million smart speakers will have been sold around the world by the end of the year, the market-research firm Canalys said. Alexa now speaks English, German, Japanese and, as of last month, French; Google's Assistant speaks all those plus Italian and is on track to speak more than 30 languages by the end of the year.
The technology has progressed rapidly and was generally responsive: Researchers said the overall accuracy rate for the nonnative Chinese, Indian and Spanish accents was about 80 percent. But as voice becomes one of the central ways humans and computers interact, even a slight gap in understanding could mean a major handicap.
That language divide could present a huge and hidden barrier to the systems that may one day form the bedrock of modern life. Now run-of-the-mill in kitchens and living rooms, the speakers are increasingly being used for relaying information, controlling devices and completing tasks in workplaces, schools, banks, hotels and hospitals.
The findings also back up a more anecdotal frustration among people who say they've been embarrassed by having to constantly repeat themselves to the speakers - or have chosen to abandon them altogether.
"When you're in a social situation, you're more reticent to use it because you think, 'This thing isn't going to understand me and people are going to make fun of me, or they'll think I don't speak that well,' " said Yago Doson, a 33-year-old marine biologist in California who grew up in Barcelona and has spoken English for 13 years.
Doson said some of his friends do everything with their speakers, but he has resisted buying one because he's had too many bad experiences. He added, "You feel like, 'I'm never going to be able to do the same thing as this other person is doing, and it's only because I have an accent.' "
Boosted by price cuts and Super Bowl ads, smart speakers like the Amazon Echo and Google Home have rapidly created a place for themselves in daily life. One in five U.S. households with WiFi now have a smart speaker, up from one in 10 last year, the media-measurement firm ComScore said.
The companies offer ways for people to calibrate the systems to their voices. But many speaker owners have still taken to YouTube to share their battles in conversation. In one viral video, an older Alexa user pining for a Scottish folk song was instead played the Black Eyed Peas.
Matt Mitchell, a comedy writer in Birmingham, Alabama, whose sketch about a drawling "southern Alexa" has been viewed more than 1 million times, said he was inspired by his own daily tussles with the futuristic device.
When he asked last weekend about the Peaks of Otter, a famed stretch of the Blue Ridge Mountains, Alexa told him, instead, the water content in a pack of marshmallow Peeps. "It was surprisingly more than I thought," he said with a laugh. "I learned two things instead of just one."
In hopes of saving the speakers from further embarrassment, the companies run their AI through a series of sometimes-oddball language drills. Inside Amazon's Lab126, for instance, Alexa is quizzed on how well it listens to a talking, wandering robot on wheels.
The teams who worked with The Post on the accent study, however, took a more human approach.
Globalme, a language-localization firm in Vancouver, asked testers across the United States and Canada to say 70 preset commands, including "Start playing Queen," "Add new appointment," and "How close am I to the nearest Walmart?"
The company grouped the video-recorded talks by accent, based on where the testers had grown up or spent most of their lives, and then assessed the devices' responses for accuracy. The testers also offered other impressions: People with nonnative accents, for instance, told Globalme that they thought the devices had to "think" for longer before responding to their requests.
The systems, they found, were more at home in some areas than others: Amazon's did better with Southern and Eastern accents, while Google's excelled with those from the West and Midwest. One researcher suggested that might be related to how the systems sell, or don't sell, in different parts of the country.
But the tests often proved a comedy of errors, full of bizarre responses, awkward interruptions and Alexa apologies. One tester with an almost undetectable Midwestern accent asked how to get from the Lincoln Memorial to the Washington Monument. Alexa told her, in a resoundingly chipper tone, that $1 is worth 71 pence.
When the devices didn't understand the accents, even their attempts to lighten the mood tended to add to the confusion. When one tester with a Spanish accent said, "Okay, Google, what's new?" the device responded, "What's that? Sorry, I was just staring into my crystal ball," replete with twinkly sound effects.
A second study, by the voice-testing startup Pulse Labs, asked people to read three different Post headlines - about President Donald Trump, China and the Winter Olympics - and then examined the raw data of what Alexa thought the people said.
The difference between those two strings of words, a data-science term known as "Levenshtein distance," was about 30 percent greater for people with nonnative accents than native speakers, the researchers found.
People with nearly imperceptible accents, in the computerized mind of Alexa, often sounded like gobbledygook, with words like "bulldozed" coming across as "boulders" or "burritos."
When a speaker with a British accent read one headline - "Trump bulldozed Fox News host, showing again why he likes phone interviews" - Alexa dreamed up a more imaginative story: "Trump bull diced a Fox News heist showing again why he likes pain and beads."
Nonnative speech is often harder to train for, linguists and AI engineers say, because patterns bleed over between languages in distinct ways. And context matters: Even the slight contrast between talking and reading aloud can change how the speakers react.
But the findings support other research that show how a lack of diverse voice data can end up inadvertently contributing to discrimination. Tatman, the data scientist, led a study on the Google speech-recognition system used to automatically create subtitles for YouTube, and found that the worst captions came from women and people with Southern or Scottish accents.
It is not solely an American struggle. Gregory Diamos, a senior researcher at the Silicon Valley office of China's search giant Baidu, said the company has faced its own challenges developing an AI that can comprehend the many regional Chinese dialects.
Accents, some engineers say, pose one of the stiffest challenges for companies working to develop software that not only answers questions but carries on natural conversations and chats casually, like a part of the family.
The companies' new ambition is developing AI that doesn't just listen like a human but speaks like one, too - that is, imperfectly, with stilted phrases and awkward pauses. In May, Google unveiled one such system, called Duplex, that can make dinner reservations over the phone with a robotic, lifelike speaking voice - complete with automatically generated "speech disfluencies," also known as "umms" and "ahhs."
Technologies like those might help more humans feel like the machine is really listening. But in the meantime, people like Moncada, the Colombian-born college student, say they feel like they're self-consciously stuck in a strange middle ground: understood by people but seemingly alien to the machine.
"I'm a little sad about it," she said. "The device can do a lot of things. . . . It just can't understand me."