Songs of birds comprise hierarchical sets of vocal gestures. In zebra finches, songs include notes and syllables (groups of notes) delivered in fixed sequences. During singing, premotor neurons in the forebrain nucleus HVc exhibited reliable changes in activity rates whose patterns were uniquely associated with syllable identity. Neurons in the forebrain nucleus robustus archistriatalis, which receives input from the HVc, exhibited precisely timed and structured bursts of activity that were uniquely associated with note identity. Hence, units of vocal behavior are represented hierarchically in the avian forebrain. The representation of temporal sequences at each level of the hierarchy may be established by means of a decoding process involving interactions of higher level input with intrinsic local circuitry. Behavior is apparently represented by precise temporal patterning of spike trains at lower levels of the hierarchy.

Humans regularly produce new utterances that are understood by other members of the same language community 1 . Linguistic theories account for this ability through the use of syntactic rules (or generative grammars) that describe the acceptable structure of utterances 2 . The recursive, hierarchical embedding of language units (for example, words or phrases within shorter sentences) that is part of the ability to construct new utterances minimally requires a 'context-free' grammar 2, 3 that is more complex than the 'finite-state' grammars thought sufficient to specify the structure of all non-human communication signals. Recent hypotheses make the central claim that the capacity for syntactic recursion forms the computational core of a uniquely human language faculty 4,5 . Here we show that European starlings (Sturnus vulgaris) accurately recognize acoustic patterns defined by a recursive, self-embedding, context-free grammar. They are also able to classify new patterns defined by the grammar and reliably exclude agrammatical patterns. Thus, the capacity to classify sequences from recursive, centre-embedded grammars is not uniquely human. This finding opens a new range of complex syntactic processing mechanisms to physiological investigation.The computational complexity of generative grammars is formally defined 3 such that certain classes of temporally patterned strings can only be produced (or recognized) by specific classes of grammars (Fig. 1). Starlings sing long songs composed of iterated motifs (smaller acoustic units) 6 that form the basic perceptual units of individual song recognition 7-9 . Here we used eight 'rattle' and eight 'warble' motifs (see Methods) to create complete 'languages' (4,096 sequences) for two distinct grammars: a context-free grammar (CFG) of the form A 2 B 2 that entails recursive centre-embedding, and a finite-state grammar (FSG) of the form (AB) 2 that does not ( Fig. 2a, b; 'A' refers to rattles and 'B' to warbles).We trained 11 European starlings, using a go/nogo operant conditioning procedure, to classify subsets of sequences from these languages (see Methods and Supplementary Information). Nine out of eleven starlings learned to classify the FSG and CFG sequences accurately (as assessed by d', which provides an unbiased measure of sensitivity to differentiating between two classes of patterns), but this task was difficult (Fig. 2c). The rate of acquisition varied widely among the starlings that learned the task (303.44 ± 57.11 blocks to reach criterion (mean ± s.e.m.), range 94-562 blocks with 100 trials per block), and was slow by comparison to other operant song-recognition tasks 7 .To assess the possibility that starlings learned to classify correctly the motif patterns described by the CFG and FSG grammars through rote memorization of the training exemplars, we further (Fig. 3a). The mean d' over the first 100 trials with new stimuli (roughly six responses to each exemplar) was 1.08 ± 0.50, which is significantly better than chance performance (d' = 0). Over th...

Songbirds such as the white-crowned sparrow memorize the song of conspecific adults during a critical period early in life and later in life develop song by utilizing auditory feedback. Neurons in one of the telencephalic nuclei controlling song have recently been shown to respond to acoustic stimuli. I investigated the auditory response properties of units in this nucleus using a technique that permitted great flexibility in manipulating complex stimuli such as song. A few of the units exhibited considerable selectivity for the individual's own song. In wild-caught birds, song-specific units exhibited intradialect selectivity. In those birds that sang abnormal songs due to laboratory manipulation of song exposure during the critical period for song learning, units were selective for the abnormal songs. By systematic modification of a song, and by construction of complex synthetic sounds mimicking song, the acoustic parameters responsible for the response selectivity were identified. Song-specific units responded to sequences of two song parts but not to the parts in isolation. Modification of the frequencies of either part of the sequence, or increasing the interval between the parts, varied the strength of the response. Thus, temporal as well as spectral parameters were important for the response. When sequences of synthetic sounds mimicking song were effective in evoking an excitatory response, the response was sensitive to the aforementioned manipulations. Wih these techniques it was possible to elucidate the acoustic parameters required to excite song-specific units. All songs of the repertoire eliciting a strong excitatory response contained the appropriate parameters, which were missing from all weakly effective, ineffective, or inhibitory songs.

Memory consolidation resulting from sleep has been seen broadly: in verbal list learning, spatial learning, and skill acquisition in visual and motor tasks. These tasks do not generalize across spatial locations or motor sequences, or to different stimuli in the same location. Although episodic rote learning constitutes a large part of any organism's learning, generalization is a hallmark of adaptive behaviour. In speech, the same phoneme often has different acoustic patterns depending on context. Training on a small set of words improves performance on novel words using the same phonemes but with different acoustic patterns, demonstrating perceptual generalization. Here we show a role of sleep in the consolidation of a naturalistic spoken-language learning task that produces generalization of phonological categories across different acoustic patterns. Recognition performance immediately after training showed a significant improvement that subsequently degraded over the span of a day's retention interval, but completely recovered following sleep. Thus, sleep facilitates the recovery and subsequent retention of material learned opportunistically at any time throughout the day. Performance recovery indicates that representations and mappings associated with generalization are refined and stabilized during sleep.

Songbirds learn a correspondence between vocal-motor output and auditory feedback during development. For neurons in a motor cortex analog of adult zebra finches, we show that the timing and structure of activity elicited by the playback of song during sleep matches activity during daytime singing. The motor activity leads syllables, and the matching sensory response depends on a sequence of typically up to three of the preceding syllables. Thus, sensorimotor correspondence is reflected in temporally precise activity patterns of single neurons that use long sensory memories to predict syllable sequences. Additionally, "spontaneous" activity of these neurons during sleep matches their sensorimotor activity, a form of song "replay." These data suggest a model whereby sensorimotor correspondences are stored during singing but do not modify behavior, and off-line comparison (e.g., during sleep) of rehearsed motor output and predicted sensory feedback is used to adaptively shape motor output.

Neuronal activity in the hyperstriatum ventrale, pars caudale (HVc) is associated with and necessary for the production of song by songbirds. HVc neurons also respond to acoustic stimuli. The present investigation assessed the auditory response properties of neurons in HVc by testing with the individual bird's own (autogenous) song and the songs of conspecific birds. Throughout HVc, multiunit clusters preferentially responded to autogenous song. Selectivity for autogenous song was apparent even when compared to similar intradialect songs, and neuronal clusters preferred autogenous song over the (tutor) song model that birds heard during the impressionable phase early in life. The responses to autogenous song were stable in the adult. HVc neurons were sensitive to the acoustic parameters of autogenous song and consistently exhibited a diminished response to modified song. In contrast, field L neurons, which are presumed to be a source of auditory input to HVc, did not exhibit selectivity for autogenous song and showed no special sensitivity to the acoustic parameters of autogenous song. These observations implicate song (motor) learning in shaping the response properties of HVc, but not field L, auditory neurons. It is proposed that HVc auditory neurons may contribute to a bird's ability to discriminate among conspecific songs by acting as an "autogenous reference" during perception of those songs.

Song learning shapes the response properties of auditory neurons in the song system to become highly selective for the individual bird's own ("autogenous") song. The auditory representation of autogenous song is achieved in part by neurons that exhibit facilitated responses to combinations of components of song. To understand the circuits that underlie these complex properties, the combination sensitivity of single units in the hyperstriatum ventrale, pars caudale (HVc) of urethane-anesthetized zebra finches was studied. Some neurons exhibited nonlinear temporal summation, spectral summation, or both. The majority of these neurons exhibited low spontaneous rates and phasic responses. Most combination-sensitive neurons required highly accurate copies of sounds derived from the autogenous song and responded weakly to tone bursts, combinations of simple stimuli, or conspecific songs. Temporal combination-sensitive (TCS) neurons required either two or more segments of a single syllable, or two or more syllables of the autogenous song, to elicit a facilitated, excitatory response. TCS neurons integrated auditory input over periods ranging from 80 to 350 msec, although this represents a lower limit. Harmonic combination-sensitive (HCS) neurons required combinations of two harmonics with particular frequency and temporal characteristics that were similar to autogenous song syllables. Both TCS and HCS neurons responded much more weakly when the dynamical spectral features of the autogenous song or syllables were modified than when the dynamical amplitude (waveform) features of the songs were modified. These results suggest that understanding the temporal dynamics of auditory responses in HVc may provide insight into neuronal circuits modified by song learning.

Quantitative biomechanical models can identify control parameters used during movements, and movement parameters encoded by premotor neurons. We fit a mathematical dynamical systems model including subsyringeal pressure, syringeal biomechanics, and upper vocal tract filtering to the songs of zebra finches. This reduced the dimensionality of singing dynamics, described as trajectories in pressure-tension space (motor “gestures”). We assessed model performance by characterizing the auditory response "replay" of song premotor HVC neurons to presentation of song variants in sleeping birds, and by examining HVC activity in singing birds. HVC projection neurons were excited and interneurons were suppressed with near-zero time lag, at times of gesture trajectory extrema. Thus, HVC precisely encodes vocal motor output via the timing of extreme points of movement trajectories. We propose that the sequential activity of HVC neurons represents the sequence of gestures in song as a “forward” model making predictions on expected behavior to evaluate feedback.