LAUSR

the identity of the phrase two steps previously in the sequence’s history (a ‘second-order’ correlation). Most of the correlations between ROI activity and phrase sequence depended on the identities of preceding (rather than future) phrases, and these ‘past’ correlations were stronger than correlations depending on the identity of future phrases. Thus, the activity of some HVC neurons reflects the sequence of preceding phrases. Next, the authors investigated whether this subset of HVC neurons showed properties that are predicted of a neural basis of long-range syntax. To be able to support long-range syntax, neurons would need to be able to maintain information about actions in a sequence over the course of multiple actions. Indeed, the authors found that some sequence-correlated ROIs carried forward information about the identity of the first of a sequence of four phrases. This was the case even when the final phrase of the sequence was replaced with singing cessation, confirming that the information these neurons carried corresponded to the first phrase itself, rather than future phrase choice. These results suggest that some HVC PNs are able to mediate long-range syntax rules in birdsong. Neurons that mediate long-range syntax are also expected to provide context-specific information parti cularly around the time of the complex transition. In line with this, more sequence-correlated ROIs were observed immediately before or after complex transitions, when information about past context would be most needed. Moreover, the authors found examples in which the activity of multiple ROIs was much more predictive of future and past phrase identity than was the activity of any individual ROI, suggesting that long-range syntax is supported by the synergy of the activity of a network of neurons. Together, this study provides evidence that a subset of HVC PNs maintain information about phrase sequences required to implement long-range syntax rules for complex transitions in birdsong. Natasha Bray