Birdsong modification with food reward

Abstract

Songbirds, such as Bengalese finches (Lonchura striata domestica) produce syntactically organized vocal sequences, composed of individual syllables strung together in variable order. Adult birds can learn to modify transition probabilities between syllables through sequence modification training: when punishing one transition with manipulated auditory feedback, birds will gradually reduce the targeted transition. This protocol is thought to rely on a circuit for vocal-auditory monitoring of the birds own song. Despite the overwhelming usefulness of food rewards for other kinds of trained animal behavior, reinforcing birdsong features with food rewards has remained elusive, possibly because the slow timescale of food reward does not match the fast and precise vocal-auditory feedback loops underlying learned birdsong. Here, we use second-order conditioning to selectively increase the frequency of target transitions in song with food rewards. Bengalese finches learned to associate a delayed primary reinforcer (food delivery by an automated feeder after song ends) with a secondary reinforcer (a short click sound). We then reinforced specific target syllables with the click during ongoing song, leading birds to selectively increase the frequency of the targeted syllable transitions. Learned changes were specific to the target and evident in catch trials without reinforcement, indicative of a learning process. Our results demonstrate that song control circuits can learn from different feedback modalities, including learned associations with food reward.