The following article requires a subscription:



(Format: HTML, PDF)

Significant trial-by-trial variation persists even in the most practiced skills. One prevalent view is that such variation is simply 'noise' that the nervous system is unable to control or that remains below threshold for behavioural relevance 1-3. An alternative hypothesis is that such variation enables trial-and-error learning, in which the motor system generates variation and differentially retains behaviours that give rise to better outcomes. Here we test the latter possibility for adult bengalese finch song. Adult birdsong is a complex, learned motor skill that is produced in a highly stereotyped fashion from one rendition to the next 4,5. Nevertheless, there is subtle trial-by-trial variation even in stable, 'crystallized' adult song 6-8. We used a computerized system to monitor small natural variations in the pitch of targeted song elements and deliver real-time auditory disruption to a subset of those variations. Birds rapidly shifted the pitch of their vocalizations in an adaptive fashion to avoid disruption. These vocal changes were precisely restricted to the targeted features of song. Hence, birds were able to learn effectively by associating small variations in their vocal behaviour with differential outcomes. Such a process could help to maintain stable, learned song despite changes to the vocal control system arising from ageing or injury. More generally, our results suggest that residual variability in well learned skills is not entirely noise but rather reflects meaningful motor exploration that can support continuous learning and optimization of performance.

(C) 2007 Nature Publishing Group