Chapter 2 : Neuronal ensemble coding of birdsong

In many models of neural population coding, similar sensory or motor states are represented in the brain by similar neural ensembles (Georgopoulos et al., 1999; Lewis and Kristan, 1998; Wilson and McNaughton, 1993). To explore this issue, we measured the activity of large numbers of single neurons in the pre-motor nucleus RA of the singing zebra finch. During singing, individual RA neurons generate precise bursts of action potential spikes (Yu and Margoliash, 1996). We found that highly similar song elements were typically produced by stereotyped but uncorrelated ensembles of RA neurons. Small changes in acoustic structure between two syllables were frequently subserved by entirely different ensembles of RA neurons. Furthermore, the neural activity in RA often changed on a time scale an order of magnitude faster than the acoustic structure in the song, causing constant acoustic outputs to be produced by sequences of rapidly changing neural ensembles. Thus there are dynamics internal to the song control system that occur reliably and precisely yet are not directly correlated with structure in the song itself. These data represent a new type of neural code underlying complex learned behaviors.