Could AI someday tell us what our pets are saying?

Anyone who lives with a dog or cat has probably wondered what exactly that bark, whine or indignant meow is supposed to mean. One bark may signal someone at the gate, another excitement, while a particular mewl might appear reliably five minutes before dinner.

Artificial intelligence is now being used to examine those sounds on a much larger scale, analysing thousands of vocalisations and identifying patterns that would be difficult for the human ear to detect.

Researchers at Kaunas University of Technology in Lithuania say AI can already group recurring animal sounds and link them with behaviour or environmental context. Turning that into a genuine animal-to-human translator, however, is far more ambitious.

Machine-learning systems are particularly useful when researchers have huge quantities of audio to examine, comparing features such as pitch, duration, rhythm and frequency to identify recurring patterns.

Research published this year in Frontiers in Veterinary Science, for example, trained a deep-learning system on thousands of dog and cat vocalisations and reported high accuracy when classifying certain emotional states.

That may sound tantalisingly close to translation, but there is an important distinction: recognising a pattern is not the same as understanding meaning.

A bark is not necessarily a word

A computer might learn that one type of bark frequently occurs during play, another when a stranger approaches, and a third during distress. What it cannot automatically tell us is whether the animal is conveying a specific “message” comparable to a human sentence.

Separate research has highlighted this difficulty by testing AI on toddler vocalisations, where humans already have some idea what the sounds are intended to communicate.

Even advanced neural networks sometimes grouped sounds with different meanings together, while separating others that conveyed similar things. The researchers concluded that sound alone is not enough; context and the listener’s response also matter.

With animals, that challenge is greater still. A bark accompanied by a stiff body, raised hackles and a stranger at the gate may mean something very different from a similar-sounding bark produced while chasing a favourite toy.

That is why researchers increasingly look beyond sound to facial expressions, posture, gaze, tail movement, activity and surroundings. Studies involving dogs using communication buttons have likewise cautioned against interpreting an isolated signal without considering what happened before and after it.

The eventual “translator” may therefore look less like a device that turns “woof!” into “I would like chicken, please”, and more like a system combining sound, movement and context to estimate an animal’s emotional state or likely intention.

Such technology could still be useful: AI-assisted systems might help owners recognise stress, fear or discomfort earlier, while vets and animal carers could use vocal patterns as another clue when monitoring wellbeing.

Dr Dolittle will have to wait

Scientists nevertheless urge caution over claims that AI will soon allow humans to “talk” to animals. Researchers quoted recently by The Guardian warned that excitement around animal-language AI can easily outrun the evidence, particularly when commercial products promise more than science has demonstrated.

At present, there is no validated system capable of translating a dog’s thoughts into human language, and even researchers enthusiastic about the technology acknowledge that true translation remains a distant goal.

For now, AI may be less of a universal pet translator than an exceptionally attentive listener – one capable of spotting patterns across thousands of barks, whines and meows that humans might otherwise miss.

And until a machine can reliably tell you what your cat is thinking, experience suggests the answer will often remain familiar: “I want food!”

Subscribe to our newsletter and get news delivered to your mailbox.