A study found that AI models could not sort pre-verbal toddlers’ vocalizations by meaning or detect rising urgency, despite outperforming a conventional acoustic method. The researchers say animal calls also require understanding context and the receiver’s perception.
The recordings covered distress, calls to a specific parent and requests for food. Researchers compared a conventional acoustic method with two deep neural networks, one trained on animal vocalizations and the other on adult speech. The networks sometimes grouped different messages together and split different vocalizations with the same intended meaning.
Study supervisor Yosef Yovel said the toddler sample was small and intended only to illustrate the challenge; the team also tested a dataset of real human words and found a different result. The researchers say decoding animal communication will require AI alongside behavioral observations, playback experiments and, sometimes, brain-activity measurements.
