This story is over 5 years old.


Wolves Have Different ‘Howling Dialects,' Machine Learning Finds

The research could help improve current wolf conservation and management methods.
Researchers have used algorithms to distinguish different wolf dialects. Image: Arik Kershenbaum

Differentiating wolf howls with human ears can prove tricky, so researchers have turned to computer algorithms to suss out if different wolf species howl differently. They think that understanding wolf howls could help improve wolf conservation and management programs.

In a study published in the journal Behavioural Processes, a group of international researchers describe using machine learning for the first time to analyse 2,000 wolf howls gathered from both wild and domesticated wolves and their subspecies from around the world.


"Rather than looking at these howls and saying, 'This one sounds like that and this,' we wanted to remove that kind of subjective analysis." Arik Kershenbaum, a research fellow at the University of Cambridge's Department of Zoology and lead author of the study, told me over the phone. "So we used some mathematical techniques—firstly to characterize the howls, then to give them an accurate and objective mathematical representation."

The machine learning techniques helped the researchers whittle down the 2,000 sound recordings into 21 types of howl based on pitch and fluctuation, which they dubbed 'howling dialects.' They then used mathematical models to find out which patterns of howling different species used.

"What we found was a consistent difference in the way that animals used the different types of howl," said Kershenbaum. "Arctic northern wolves (C. I. arctos) use very long low howls while many of the smaller animals tend to yap a lot more and use up-and-down howls."

Researchers are working on monitoring wolves remotely using their howls. Image: Arik Kershenbaum

Wolves continue to clash with farmers who think they pose a threat to their livestock. But according to Kershenbaum, traditional methods of keeping them at bay—using fencing, shooting them down, or playing back howls of other wolves—do not work.

Currently, the researchers have a project in Yellowstone National Park in the US, where they are using triangulation technology to detect and record howl locations and sounds. In the future, the researchers hope that they will be able to locate where certain species are based on their howls, and tailor conservation methods accordingly. But they admit that studying wolves in the wild is difficult given their elusive nature.

Kershenbaum cited the example of comparing the howls of endangered red wolves to coyotes (Canis latrans) with which they have a tendency to breed and hybridise. Kershenbaum explained that as the two have similar styles of howling, pinpointing the subtle differences might help conservationists keep these two species apart in the future.

"We're looking to monitor wolves remotely by using howls," said Kershenbaum. "If we could understand the difference between a territorial howl and a feeding or hunting howl, then that may enable us to produce more effective barriers to conflict and keep wolves away from human territories in a non-lethal way."