Behavioural science allows reading out the computations performed by the brain, whereas neuroscience reveals how they are implemented. In my talk, I will reveal the algorithm songbirds use to match their songs to an auditory template. We tested how zebra finches cope with the computational complexity of song learning, by prompting juveniles to modify their song to correct conflicting phonological and sequential mismatches in song syllables. Birds matched each syllable to the most acoustically similar sound in the target, regardless of its temporal position, resulting in unnecessary sequence errors that were later corrected. Thus, birds prioritised efficient learning of syllable vocabulary, at the cost of inefficient syntax learning. Overall, we find that birds learn their songs by solving a linear assignment problem. This is the type of problem that taxi companies solve to optimally dispatch their taxis to waiting customers.
In the second part of my talk I will show how insights obtained from observation compare with insights gained from trial-and-error. We find that birds can learn to discriminate auditory stimuli by observing expert performers. These findings agree with social learning theories showing that copying others’ behaviour is a successful strategy. However, our results indicate that the benefit of rapid learning from observation comes at the cost of poor generalisation, revealing a sensory analogue to the common view that the best means to learn a (motor) skill is rigorous practice.