The challenge of decoding animal communication using AI

Understanding animal communication requires identifying how signals map onto meaning in the receiver’s perceptual space, yet most studies rely on clustering signals by physical similarity ¹ . It is tempting to assume that the physical and the perceptual spaces are highly correlated, but this may not be the case. We exemplify this by examining a form of communication for which we do have some access to the perceptual space — the communication of pre-linguistic human toddlers directed towards adult humans. We use a small data set of vocalizations recorded in three different semantic contexts which could be described as: (1) distress calls; (2) contact calls with a specific recipient (father or mother); or (3) begging calls (emitted in expectation of feeding). Our results show that acoustic similarity does not necessarily correlate with the receiver’s perceptual space.