A dog's bark may contain far more information than many people realize, according to researchers using artificial intelligence to study animal communication. While pet owners often interpret dogs barking as a simple sign of excitement, frustration or alertness, researchers at Kaunas University of Technology (KTU) in Lithuania say artificial intelligence and machine-learning tools are helping uncover patterns in dog vocalizations that could reveal clues about a pet's emotional state and the situation it is responding to. "The most surprising thing I've learned is how much information can actually be encoded in a dog's vocalizations," Rytis Maskeliūnas, a professor in the university's Faculty of Informatics, told Newsweek.
"At first, it is easy to think of barking simply as a general expression of excitement or arousal, but if we look at the whole context, including the general behavior of the dog, we can see patterns that are associated with different emotional states and different situations." To explore those patterns, Maskeliūnas and his colleagues analyzed recordings of German shepherds to see whether machine-learning systems could distinguish between emotional states based on sound alone. In research highlighted by KTU, the team examined whether AI could identify different emotional states from dog vocalizations. Researchers categorized the recordings into four emotional states associated with the dogs: happiness, anger, crying and loneliness.
The team analyzed the frequency and timing of the sounds, and the results suggest that dog vocalizations can contain acoustic patterns linked to different emotional states. The team found they could especially identify angry barking with relatively high reliability. But Maskeliūnas said one of the most important lessons from the research is that recognizing emotion and understanding communication are not the same thing.
"That distinction is probably the most interesting part of this field for me and probably most pet owners, as the more closely we look at our pet behaviors, the more we realize how much information may be there, and also how much we still have to learn about what that information actually means," he said. The researchers believe dog vocalizations likely contain both emotional expressions and communicative signals, though proving the latter remains a significant challenge. One hypothesis, Maskeliūnas said, is that barks contain information not only about a dog's emotional state but also about the situation in which the vocalization occurs.
Whether dogs deliberately use particular sounds to communicate specific information, however, remains an open question. "The key is to look not only at the sound itself, but also at who produces it, in what context and how other dogs and animals around them respond," he said. If a particular vocalization consistently produces a particular response from other dogs, that would provide stronger evidence that the sound is being used as a communication signal rather than simply expressing an emotional state.
For now, researchers aren't suggesting that AI can translate a bark into a sentence such as "I'm hungry" or "I want to go outside." Instead, the technology could help identify patterns that reveal when something about a dog's emotional state or behavior has changed. Maskeliūnas said some differences between vocalizations may be difficult for humans to consciously hear, but could potentially be detected by machine-learning systems. Over time, such a system could even learn an individual dog's normal vocal patterns and identify changes that could be associated with stress, discomfort, fear or other unusual behavior.
Extract — continue reading at the source.
