One chilly day in February 1877, a British cotton baron named Joseph Sidebotham heard what he thought was a canary warbling near his hotel window. He was vacationing with his family in France, and soon realized the tune wasn’t coming from outside. “The singing was in our salon,” he wrote of the incident in Nature. “The songster was a mouse.”

The family fed the creature bits of biscuit, and it quickly became comfortable enough to climb onto the warm hearth at night and regale them with songs. It would sing for hours.

Clearly, Sidebotham concluded, this was no ordinary mouse.

More than a century later, however, scientists discovered he was wrong. It turns out that all mice chitter away to each other. Their language is usually just too high-pitched for human ears to detect.

Today, mouse songs are no mere curiosity. Researchers are able to engineer mice to express genetic mutations associated with human speech disorders, and then measure the changes in the animals’ songs. They’re leveraging these beautifully complex vocalizations to uncover the mysteries of human speech.

* * *

Anecdotal accounts of singing mice date back to 1843. In the journal The Zoologist, the British entomologist and botanist Edward Newman wrote that the song of a rare “murine Orpheus” sounds as “if the mouth of a canary were carefully closed, and the bird, in revenge, were to turn ventriloquist, and sing in the very centre of his stomach.”

To explain this phenomenon, the British natural historians Gerald Barrett-Hamilton and Martin Hinton claimed in a 1921 review that the singing was a pathological behavior resulting from inflammation in a mouse’s windpipe, like a cough or wheeze. They didn’t suppose that illness simply lowered these animals’ voices—the way a bad cold does in people—which might actually be what happened to Sidebotham’s extraordinary mouse.

In 1948, Wolfgang Schleidt, a zoology professor at the University of Vienna, was the first to show that all field mice vocalize in pitches too high for human ears to hear. About two decades later, Gillian Sales, a senior lecturer at King’s College London, who has devoted her career to studying rodent songs, used a tape recorder to listen to the songs by slowing them down to 1/20 normal speed. This had the effect of dropping the mouse’s register into the range of human hearing, but it also made the vocalizations sound a bit like the slow, repetitive glissando of a slide whistle.

Analyzing these songs was incredibly tedious, Sales says. It all had to be done manually. Just measuring the pitch and duration of a few notes could take ten minutes. As a result, even though the following decades saw fruitful research on why and how mice sing, there wasn’t much focus on the structure of the songs themselves, according to Sales. No one thought of mice as a potential way to model human speech.

The tide turned for mouse song in 1999, when a postdoctoral fellow at Harvard made a serendipitous discovery. Timothy Holy was looking for ways of identifying female-mouse sex pheromones, and he reasoned that pheromones in urine might cause males to make sounds. To him, the male’s high-pitched squeaks were simply good indicators of pheromones’ presence; he didn’t even think of them as songs.

To listen in while his mice sniffed at the urine swabs, Holy recorded their vocalizations then lowered the pitch without slowing down the tempo. To his astonishment, he heard music coming from these little furry faces.   

“I noticed basically right away that these vocalizations were a lot more complicated than I had grown to expect based on the reading I had done on the literature,” says Holy, who’s now a neurobiology professor at Washington University, St. Louis. “I remember joking with my postdoc mentor at the time that maybe these were like birdsongs, basically.”

Although Holy’s pheromone experiment never panned out as he hoped, his rigorous characterization of mouse songs led to a seminal paper in 2005 in PLoS Biology. Other labs took notice, and mouse-song research exploded.

* * *

Among the limited number of other mammals that communicate vocally—including humans, whales, elephants, dolphins, and bats—none are ideal for the laboratory. Mice, by contrast, are small, cheap, and easily genetically modified. Labs routinely delete or add genes to the mouse genome; many transgenic mice can even be ordered from a catalog.

When humans speak, their vocal cords vibrate at different frequencies like the strings on a guitar. People shape those sounds with the lips and tongue to form phonemes—the distinct units of speech—which come together to make syllables, words, and phrases. Mice lack vibrating vocal cords and an articulatory apparatus, but they can still volitionally utter streams of distinct syllables. So Holy sees the mouse as useful for exploring the rudimentary aspects of speech production that mice and men share.

For a study published earlier this year, Holy teamed up with Dennis Drayna, the chief of the genetics of communications disorders section at the National Institutes of Health, to introduce a human stuttering gene mutation into mice. Lo and behold, the mice stuttered too. “I think many people are shocked that the mouse could have much relevance to any human-speech condition,” says Drayna. For speech disorders like stuttering or Tourette syndrome, the problem arises from initiating words rather than thinking them up, so mouse songs are a reasonable proxy.

Currently there’s no cure for stuttering. Psychiatric drugs may be able to help, but so far, FDA-approved versions haven’t made any headway. To get new stuttering drugs to market, animal testing is essential, Drayna says. Stuttering mice now make those experiments possible.

Mouse songs aren’t only pertinent to the study of speech initiation. They can also be used to study how complex sequences of speech form from simple building blocks. To examine this notion in mice, the Duke professor of neurobiology Erich Jarvis uses what he calls “sexy songs.” These are relatively elaborate courtship songs males belt out to woo a mate, and the lady mice love them.

Jarvis worked with genetically engineered male mice that expressed a rare genetic mutation on a gene called FOXP2.* Humans who carry a FOXP2 mutation have trouble stringing individual phonemes together to form complex words or sentences. Jarvis discovered that these mutant mice no longer sang their sexy songs, even in the face of a female in heat. Their simple chirps remained normal. It’s essentially the same speech pathology observed in humans with the FOXP2 mutation. His lab published these results in October.

Jarvis takes this as evidence for his more general hypothesis that speech is not all-or-nothing. It evolved gradually, and the rudiments are present even in humans’ distant cousins, the rodents.

Despite promising findings both in his lab and elsewhere, Jarvis, like his peers, is quick to point out that mouse song is not a perfect model for human speech. Most glaringly, unlike humans and some birds, mice have scant abilities to modify their vocalizations through learning. But Jarvis thinks he can change that.

In 2014, Jarvis’s lab published in Science that humans and songbirds both evolved changes in the way certain genes are expressed. These genes are the ones that allow humans and songbirds to alter what they say as a result of experience. Mice lack these evolutionary tweaks, but Jarvis is currently using genetic engineering to add them in. He hopes doing so will make mouse vocalizations more human-like, which would allow mice to serve as better models for human speech.

There are plenty of genetic human speech disorders, and Drayna, Holy’s collaborator, thinks even as-is the mouse could be a reasonable model for many of them. While speech is unique to humans, it is the product of genes that are not unique to humans. Examining the more elaborate social songs may provide new insight into the evolutionary roots of human speech. For instance, some labs house mice together in social groups, and one of Drayna’s colleagues has observed that if you take one of the mice away for a few hours and then return him to the group, the cage bursts into boisterous song to greet him. It’s the most they ever sing.

“You can only imagine what they’re saying,” Drayna says.

* This article originally stated that Jarvis engineered the FOXP2 mice himself. We regret the error.