AI Could Help You Understand Animal Speech

You might finally know what Lassie was trying to tell you

Key Takeaways

  • Researchers say they have been able to use AI to translate pig grunts. 
  • The study was intended to lay the groundwork for systems that can improve the well-being of farm animals.
  • Apps available to 'translate' dog and cat sounds haven't been developed based on scientific facts, one expert said.
A dog and a child sitting face-to-face at a picnic table, appearing to have a conversation.

Jetta Productions / Walter Hodges / Getty Images

A pig's squeal might be worth a thousand words.  

In a recent study, an international team of researchers has used artificial intelligence (AI) to translate pig grunts into emotions. Using more than 7000 audio recordings of pigs, the researchers designed an algorithm that can decode whether an individual pig is experiencing a positive feeling, a negative one, or somewhere in between. 

"With a very large data set like we had of calls produced in known contexts, we could thus train such network and reach a high accuracy, which can then inform us about the emotion of the pigs (so 'translate' pig calls to humans if you want)," associate professor Elodie Floriane Mandel-Briefer of the University of Copenhagen's Department of Biology, who co-led the study, told Lifewire in an email interview. 

A Silicon Dr. Doolittle? 

The researchers recorded pig sounds in both commercial and experimental settings, which, based on the behavior of the pigs, are either associated with positive or negative emotion. Positive situations included, for example, those when piglets suckle from their mothers or when they are united with their family after being separated. The emotionally negative situations included separation, fights between piglets, castration, and slaughter, among others.

In experimental stables, the researchers also created mock scenarios for the pigs, designed to evoke more nuanced emotions in the middle of the spectrum. These included an arena with toys or food and a corresponding arena without any stimuli. The researchers also placed new and unfamiliar objects in an area for the pigs to interact with, and their calls, behavior, and heart rates were monitored and recorded when possible.

The scientists then analyzed the audio recordings to see if there was a pattern in the sounds that communicate emotions and discern the positive situations and emotions from the negative ones. In negative cases, the researchers collected more high-frequency calls (such as screams and squeals). At the same time, low-frequency calls (such as barks and grunts) occurred in situations where the pigs experienced positive or negative emotions.

In the study, researchers compared a supervised automated method (permuted Discriminant Function Analyses, pDFA) based on four vocal parameters and an unsupervised method, a neural network based on the images (spectrograms) of the sounds. 

"The pDFA could classify calls to the correct emotional valence (positive or negative) the pig was experiencing during vocal production 62% of the time, while the neural network reached 92% accuracy," Mandel-Briefer said. 

Translating Animal Emotions

The study was intended to lay the groundwork for systems that can improve the well-being of farm animals. But Mandel-Briefer said the same research could apply to other animals as well. 

"If similar large databases of vocalizations produced in specific contexts and emotions are gathered by scientists, we could develop similar algorithms for other species as well, and that would be more objective than existing apps,” she said. 

A dog sitting on a chair appearing to speak into a boom microphone.

Tim Robberts / Getty Images

There are some apps available that can ‘translate’ dog and cat sounds, such as MeowTalk Cat Translator or Human-to Dog Translator, but they haven’t been developed based on scientific facts and contexts of known emotions, Mandel-Briefer said. 

"Scientists now have established frameworks and methods to study animal emotions in an objective way (e.g., using behavioral, neurophysiological, and cognitive indicators), and this is what we used in our paper,” she added. 

Don’t plan on having conversations with your pets just yet. Even translating between human languages is still a challenge for AI. There are many AI-powered language translation services, including Google Translate and Microsoft Text Translation API. The benefit of AI-driven translation services is that they are more affordable than hiring a human translator.

“While AI-powered translation services are convenient, they’re still limited in their translation capability,” Kavita Ganesan, an AI expert and founder of Opinosis Analytics, told Lifewire in an email interview. “For example, they have difficulty understanding language-specific idioms and sarcasm, often translating them literally.”

Was this page helpful?