Edited by Eric I. Knudsen, Stanford University School of Medicine, Stanford, CA, and approved December 20, 2010 (received for review July 21, 2010)
Vocal ontogeny in songbirds provides a good model for understanding how complex motor behavior, including speech, is learned. For birdsong, as for other motor learning, it has generally been assumed that a subject's motor output at any point during learning represents what the subject has learned to produce by that time. Here, we show, however, that juvenile zebra finches partway through song learning, singing immature song, are capable of producing song with much more mature properties, depending on the behavioral context. In these birds, we were able to elicit courtship (female-directed) song, which young birds normally sing infrequently, and to compare it with the alone or “undirected” song (Undir) predominantly produced during learning as well as with the same bird's subsequent adult song. We found that the juvenile courtship song was much less variable than the immature Undir and as stereotyped as the adult song produced after a further month of practice. More strikingly, the juvenile courtship song was also acoustically much more similar than Undir to the adult song. This finding demonstrates that the Undir that juvenile birds usually produce underestimates the extent of learning and that song structure is learned faster than previously thought. Moreover, the rapid improvement in song quality in response to external social cues supports the idea that courtship singing is a state of motor “performance,” in which the bird selects the best variants of the song learned during singing alone, and suggests that such performance states can reveal unappreciated progression of learning.
Birdsong is a complex learned vocal behavior developed by motor practice in early life, with many striking parallels to human speech (1–3). Young birds first produce highly variable but poorly structured song, and through tens or even hundreds of thousands of immature song productions over many weeks, they gradually develop a stable song that resembles the song of their tutor (4–6). During this process of sensorimotor learning, birds compare their vocalizations with the memorized song model using auditory feedback and guide their song modification using this comparison (7). Given the gradual development of vocal patterns during song learning, it has been widely believed, for songbirds just as for babbling infants or young pianists, that motor output during learning directly reflects the progress of learning. It is assumed that the immature and inaccurate structure of the song that juvenile birds vigorously produce results from the immaturity of their central and/or peripheral nervous system and represents what they have learned to produce by that time (8–10). The young song is also highly variable. Some of this variability is actively generated by the central nervous system (11–13) and has been hypothesized to aid in the bird's exploration of vocal space as it learns (13), but such variability also makes it less clear what the birds actually know.
To examine what young birds have really learned to produce and to test the assumption that their vocal production directly reflects the progress of song learning, we compared the songs that they produce in two different behavioral contexts. Adult male zebra finches [≥90 d posthatch (dph)] are well known to sing in two contexts (14, 15). Courtship song, also known as “directed” song (Dir), is produced while the bird is facing a female bird and performing a rhythmic pivoting dance, and “undirected” song (Undir) is delivered when the bird is not facing any other bird or is alone (Fig. S1A). In adults, Dir is subtly but significantly less variable from rendition to rendition in fine structure than Undir, although the average acoustic properties of song across many renditions are virtually the same between Dir and Undir (15–17) (Fig. 1B and Fig. S1B). This indicates that adult birds can control some of the variability in Undir to produce more stereotyped Dir. In contrast to adults, juvenile male birds learning song primarily sing in a noncourtship context, and such song is thus classified as Undir (14, 15). Juvenile birds can produce Dir (15), however, although only infrequently because of their sexual immaturity, and thus, little is known about the acoustic properties of this adolescent courtship song. Accordingly, it is also unknown whether or how much juvenile birds have learned to control the variability of their variable Undir and what the song structure would be if they could minimize its variability. We therefore elicited both Dir and Undir in juvenile zebra finches in the middle of sensorimotor learning and compared the songs’ acoustic properties. Our data show that juvenile Dir has remarkably higher stereotypy than Undir, similar to that of adult song, demonstrating that juvenile birds at that age can control most of the high degree of variability in Undir. Moreover, when we compared the juvenile song with the fully mature song that the same birds later produced as adults, we found that juvenile Dir has acoustic properties that are not simply the average of the variable Undir properties. Rather, juvenile Dir consists only of variants of song that are “better” (i.e., more similar to the bird's subsequent adult song). These results demonstrate that the song juveniles usually produce in a noninteractive setting underestimates the extent of their vocal competency and highlight the powerful influences of social cues on learned behavior.
Examples of Undir (blue) and Dir (red) of a male zebra finch as a juvenile (A, 56 dph) and an adult (B, 90 dph). At each age, song snippets are from consecutive song bouts. Note that in juveniles, Dir has more stereotyped motifs composed of relatively adult-like and consistent syllables compared with Undir, although the phonology of many juvenile syllables is still more primitive and less structured compared with adult syllables.
To collect a sufficient number of juvenile Dir for quantitative comparison of its acoustic properties with Undir, we presented sexually mature adult female birds to juvenile male birds every day, multiple times a day, starting around the middle of song learning (around 50 dph; Materials and Methods). Because juvenile male birds usually sang more bouts of Dir when female birds exhibited copulation solicitation displays (CSDs), including back arching and tail vibration (Fig. S1A), we selectively used female birds that exhibited frequent CSDs to the juvenile male birds. Most birds younger than 50 dph produced Dir only rarely in response to presentation of a female bird and gradually increased the number of Dir bouts as they became older. We analyzed Dir and Undir from the first day on which the bird sang more than 20 motifs of Dir (mean = 59.5 dph, range: 53–66 dph).
Juvenile birds at this age sing immature Undir that is different from adult song in at least two aspects (Fig. 1 and Fig. S1B). First, in contrast to the very stereotyped adult song (both Dir and Undir), juvenile Undir is highly variable in its structure from one rendition to the next. This variability is present both in the acoustic structure of individual song elements, or “syllables,” and in the temporal structure of song phrases, or “motifs,” which are sequences of syllables. Second, the structure of juvenile Undir is acoustically different from the structure of the adult song that the same bird eventually develops. The phonology of individual syllables is more primitive and less structured than that of adult syllables. The sequence and timing of juvenile syllables are also different from those of adults, making the temporal structure of song motifs different from that of adult motifs. We compared these two aspects of juvenile song (i.e., trial-by-trial variability of song structure and acoustic similarity of juvenile song to adult song) between Dir and Undir to examine differences in the maturity of these two types of song.
When we examined the acoustic structure of juvenile Dir and Undir by listening to them and inspecting their spectrograms, we found that Dir had a strikingly lower degree of trial-by-trial variability than Undir (Fig. 1A, Fig. S1B, and Movies S1 and S2). To quantify this, we first examined distributions of acoustic features of all individual syllables from all songs recorded in a single day. Song bouts were segmented into syllables, and for each syllable, we computed the means and variances of six spectral features (i.e., total 12 features), following previous studies (18, 19): Wiener entropy, amplitude modulation (AM), frequency modulation (FM), pitch, goodness of pitch, and mean frequency (SI Materials and Methods). When these acoustic features of all individual syllables were plotted against syllable durations for juvenile song, the Dir syllable distribution generally showed more distinct and dense clusters than the Undir syllable distribution for many syllable features (e.g., mean FM in Fig. 2A), indicating more stereotyped production of song syllables in directed singing. In contrast, the distributions of adult syllables were similar between Dir and Undir (Fig. S2A), consistent with previous studies showing only subtle differences in adult song variability between the behavioral contexts (17, 20). We quantified how discrete (“peaky”) vs. how scattered (“flat”) the structure of the syllable distribution plots was by calculating the entropy of probability density functions obtained from the distribution plots (Fig. 2A, Middle and Bottom) and compared the entropy between Dir and Undir by taking their ratio (“Undir/Dir syllable variability ratio,” SI Materials and Methods). If this ratio is >1, it indicates that the feature-vs.-duration plot of Dir syllables has more discrete clusters than that of Undir syllables, representing more stereotyped production of Dir syllables than Undir syllables. For most features and most birds, juvenile song showed a syllable variability ratio much higher than 1, whereas the adult songs of the same birds had a smaller variability ratio, close to 1 [mean FM and the average of all acoustic features are presented in Fig. 2B (n = 11 birds); results of other individual features and the statistical analysis are presented in Table S1]. We also confirmed that there was little effect of sample size differences between Dir and Undir on this analysis (Fig. S2B and SI Materials and Methods).
Trial-by-trial variability of syllable structure in juvenile Dir is much lower than that in juvenile Undir and similar to that in adult song. (A) Scatter plot of duration vs. the acoustic feature mean FM in degrees (deg) for individual syllables in juvenile song of the bird shown in Fig. 1 (Top); red and blue dots represent syllables in Dir and Undir, respectively. Smoothed probability density for Undir and Dir syllables (Middle and Bottom, respectively), calculated from the scatter plot at Top. Note that Dir syllables show tighter clusters than Undir syllables, indicating the more stereotyped structure of Dir syllables. Adult song data of the same bird are presented in Fig. S2A. (B) Undir/Dir syllable variability ratios calculated from mean FM vs. syllable duration plots of juvenile and adult songs (Left) and the average of syllable variability ratios for all 12 acoustic features examined (Right). Data for each bird are connected by a line (n = 11 birds). The statistical analysis of all these group data is presented in Table S1. (C) Examples of juvenile syllables of the same type from consecutive renditions of Undir and Dir motifs, showing the highly stereotyped acoustic structure of Dir syllables and more variable Undir syllable structure. (D) Feature variability (feature SD normalized by SD of adult Undir) in individual syllable types. Data of the mean FM (Left) and the all features average data (Right) are shown. Lines connect data from 59 syllable types in 11 birds; arrowheads indicate the mean across all syllable types. The statistical analysis of group data is presented in Table S2. Note that the feature variability of Dir is close to 1 in many syllable types, indicating similar variability of juvenile Dir syllables to adult Undir syllables.
The above analysis of all syllables, without reference to their identity, illustrates a markedly lower degree of trial-by-trial variability in Dir syllable structure than in Undir structure in juveniles; this contrasts with the very small difference between Dir and Undir in adults. This analysis is sensitive to more than differences in syllable variability, however, when we compare juvenile and adult songs. Developmental changes in aspects of song structure, such as the number of syllables composing a motif (e.g., Fig. S1B), will also have some effect on the entropy of syllable probability density. To compare syllable variability across different ages directly, we therefore analyzed the variability of syllable structure in identified syllable types (18, 21) (i.e., in each discrete set of syllables with similar acoustic structure in a bird's songs: for example, there are six “syllable types” in adult song in Fig. 1; see also Fig. 2C). All Dir recorded and a similar number of Undir were manually segmented, and the resulting syllables were classified into syllable types (SI Materials and Methods); variability in syllable features of the same syllable types was then compared between Dir and Undir as well as between juvenile and adult song.
Consistent with the result of the overall variability analysis in unidentified syllables, syllable feature variability in identified syllables of juvenile birds was lower in Dir than in Undir for many syllable types and for many features [Fig. 2 C and D; mean FM and the average data for all acoustic features are shown in Fig. 2D (n = 59 syllable types from 11 birds); results for other individual features and for the statistical analysis are presented in Table S2]. Moreover, the degree of feature variability in juvenile Dir was similar to that in adult song when we compared it with adult Undir (Fig. 2D and Table S2). Together, these analyses thus show that at the syllable level, Dir is much more stereotyped than Undir in juvenile birds and that its stereotypy is close to that of Undir produced by the same birds in adulthood.
In addition to syllable structure, the temporal structure of song motifs was more stereotyped in Dir than in Undir of juvenile birds (Fig. 1A, Fig. S1B, and Movies S1 and S2). When motifs were defined based on the syllable identification used in the syllable variability analysis (SI Materials and Methods), juvenile Dir motifs had more stereotyped amplitude envelopes than juvenile Undir motifs (Fig. 3 A and B). The mean correlation coefficients (CCs) between neighboring motifs were significantly higher in Dir than in Undir in juvenile birds but not in adult birds [Fig. 3C; P < 10−3 for juvenile songs and P = 0.054 for adult songs, Wilcoxon signed-rank test (n = 11 birds)]. Moreover, mean CCs of juvenile Dir were comparable to those of adult Undir (Fig. 3C; P = 0.15), indicating that the degree of stereotypy in juvenile Dir is similar to that of adult song. In addition to the amplitude envelope, we examined the continuous time course of the six acoustic features of song described above and observed similar results to those seen with amplitude envelopes (Fig. S2C shows the mean CCs of juvenile Dir and Undir, normalized by the mean CC of adult Undir, for each feature and then averaged across all features; results of individual features and the statistical analyses are presented in Table S3). All these results demonstrate that juvenile birds in the middle of song learning can control much of the variability normally seen in Undir structure and can produce song with much higher acoustic and temporal stereotypy, similar to that of adult song.
Variability in motif structure is lower in juvenile Dir than in Undir and similar to that in adult song. (A) Overlaid amplitude envelopes of song motifs in Undir (blue) and Dir (red) of juvenile and adult songs, showing high stereotypy of motif structure in juvenile Dir as well as in adult songs. Twenty randomly selected motifs are aligned so as to maximize correlations, allowing for ±10% time warping. (B) Amplitude envelopes of juvenile song motifs are expressed by pseudocolored lines (amplitude level is proportional to the brightness) and aligned in chronological order. The vertical bar to the left of the pseudocolored lines indicates the behavioral context in which the motif was produced: directed singing (red) or undirected singing (blue). Circles connected by lines on the left indicate CCs of amplitude envelopes between neighboring motifs in the same contexts. CCs of Dir and of Undir are shown in red and in blue, respectively. (C) Mean CCs of motif amplitude envelopes in 11 individual birds were compared between Undir and Dir for both juvenile (Left) and adult (Right) song. Mean CCs in juvenile Dir are significantly higher than those in juvenile Undir (P < 10−3, Wilcoxon signed-rank test) and comparable to those in adult Undir (P = 0.15).
We also examined the other aspect of song maturity in juvenile birds, the degree of similarity of song structure, at both syllable and motif levels, to the bird's ultimate copy of the tutor song, i.e., its own final adult song. To examine development of syllable phonology, we first measured the acoustic similarity of juvenile syllables to the adult syllables that the same bird eventually produced. All syllables that were identified for the variability analysis were compared between juveniles and adults for each syllable type. Surprisingly, we found that for many syllable types, the acoustic similarity of juvenile syllables to adult syllables was much higher, on average, in Dir than in Undir (Fig. 4 A and C and Fig. S3). Although the similarity of juvenile Dir to adult song was not all the way to adult levels, it was generally tightly clustered at the highest level of Undir similarity, whereas Undir similarity to adults varied over a wide range. This similarity result was also supported by measures of within-syllable entropy variance (EV), which captures the richness of acoustic structure within individual syllables and is a general measure of syllable maturity and complexity; EV usually increases as song learning progresses (18). For many syllable types, EV of juvenile Dir syllables was much higher than that of juvenile Undir syllables (Fig. 4 B and C), indicating more structured, and thus more mature, phonology of Dir syllables. Therefore, by two measures, Dir syllables in young birds were overall more adult-like than Undir syllables in their phonology, although they are generally not as mature as the syllables of the same bird's final adult song.
Juvenile Dir syllables are more similar to those of adults than juvenile Undir syllables. (A) Acoustic similarity of juvenile (Left) and adult (Right) syllables to adult syllables in Dir and Undir; an example for one syllable type is shown. Mean scores are represented as green dots connected with a line. Representative syllables of juvenile Dir (red) and Undir (blue) as well as of adult song (black) are also shown, and their similarity scores are indicated with arrows (except for the adult syllable). (B) EV of Undir and Dir syllables in juvenile and adult song; the same syllables as those shown in A. Conventions are as in A. (C) Comparisons of acoustic similarity to adult syllables (green crosses) and EV (magenta circles) of juvenile syllables between directed and undirected contexts for all syllables examined (n = 11 birds, 57 syllables). For both measures, mean scores of juvenile syllables (green dots in A and magenta dots in B) were normalized by the mean score of adult syllables (dashed line in A and B). The diagonal line indicates the unity slope. Dir syllables had significantly higher normalized scores than Undir syllables for both measures (P < 10−5 for acoustic similarity and P < 10−4 for EV, Wilcoxon signed-rank test).
To examine the similarity of juvenile song to adult song at a motif level, we compared the amplitude envelope of juvenile motifs with that of adult motifs for each bird. When we computed the maximum values of cross-correlation functions (mCCs) between each juvenile-adult motif pair (SI Materials and Methods), mCCs between juvenile Dir and adult envelopes were higher than those between juvenile Undir and adults in many birds, and Dir vs. adult mCCs were mostly at around the highest level of Undir similarity (Fig. S4 A–C). Similar results were observed for the time course of the other acoustic features that we examined (averaged values across all features are shown in Fig. S4D; individual feature data and the statistical analysis are presented in Table S4). Thus, just as for syllable structure, the motif structure of juvenile song, although still not fully adult, is much more adult-like in Dir than in Undir.
Taken together, our data demonstrate that although juvenile birds in the middle of song learning produce highly variable song in the undirected context, with variable and often low similarity to the final adult song, in the directed context, they are able to sing stereotyped song with both syllable and motif structure much more similar to that of the adult song. Because such juvenile Dir is not identical to the adult song in either syllable morphology or motif structure, however, our results suggest that juvenile Dir represents the best (most adult-like) song variants that the birds can produce at that time in song learning.
Juvenile zebra finches in the middle of song learning, like the birds we study here, have been shown to exhibit daily oscillation in the structure of their (Undir) song. They show marked improvement in Undir syllable structure (increased complexity and increased similarity to the song of the bird's tutor) during the first 2–3 h of daily singing (i.e., morning singing), followed by little further improvement during the rest of the day, and pronounced deterioration of syllable structure during night-time sleep (18). This raised the question of whether our birds showed diurnal improvement in both Dir and Undir, and, if so, how such diurnal improvement relates to context-dependent song changes. To examine this, we compared the maturity of juvenile syllables between songs produced early in the day and those produced at a later time (SI Materials and Methods) using both syllables’ acoustic similarity to adult syllables and their EV.
In birds from which we recorded both Dir and Undir throughout most of the day (n = 5 birds, 24 syllables), we observed diurnal improvement in both Dir and Undir structure. Acoustic similarity significantly increased from early song to later song in both Dir and Undir syllables (Fig. 5 A and B, Left; P < 0.05 for both Dir and Undir), and EV significantly increased in Undir syllables (Fig. S5 A and B, Left; P < 0.05), although the increase in Dir syllables did not reach statistical significance (P = 0.06). The degree of diurnal improvement was not significantly different between Undir and Dir for either acoustic similarity or EV (P = 0.41 and P = 0.67, respectively). The average percentage increase in EV of Undir syllables was 13.4%, which is similar to the previously reported average EV decrease during night-time sleep (16.7%) (18).
Diurnal improvements in syllable structure are observed in both Dir and Undir. (A) Changes in acoustic similarity of Dir (red) and Undir (blue) syllables to adult syllables in a juvenile bird (58 dph) during a single day; an example of one syllable type is shown. Lines indicate the local mean in a ±30-min time window centered around individual syllable renditions. (B) (Left) Diurnal improvement in acoustic similarity of Dir and Undir syllables, expressed as the difference in average similarity scores between early song and later song (n = 24 syllables, 5 birds; mean ± SEM). *P < 0.05 (Wilcoxon signed-rank test) for early-later difference; no significant difference between Undir and Dir (P = 0.41). (Right) Dir-Undir difference in mean similarity scores between early and later song. **P < 0.001 for Dir-Undir difference; no significant difference between early and later song (P = 0.55).
Consistent with the similar diurnal improvement between Dir and Undir syllables, a context-dependent difference in syllable structure was observed in both early and later song. Both acoustic similarity and EV were significantly larger in Dir syllables than in Undir syllables for both early and later song (Fig. 5B, Right and Fig. S5B, Right; P < 0.001 for all cases). The Dir-Undir differences were not significantly different between early and later song (P = 0.55 for acoustic similarity and P = 0.97 for EV). Thus, although juvenile birds exhibit diurnal improvement in Undir syllables, ranging from immature structure in the morning to more mature structure in the afternoon, they can produce more mature syllables in the directed context regardless of the time of day, and such Dir syllables show similar diurnal improvement to that of Undir syllables. Moreover, in the birds we studied, the magnitude of context-dependent difference was similar to or even larger than the degree of diurnal improvement (“Early” vs. “Undir” in both Fig. 5B and Fig. S5B; P = 0.06 for acoustic similarity and P = 0.02 for EV).
Our findings demonstrate that the immature properties of song during vocal practice by juvenile zebra finches partway through song learning are not simply attributable to immaturity of the birds’ vocal motor system. Rather, the presence of a female bird can elicit song with more mature properties (higher stereotypy and more adult-like structure) that is already learned and latently encoded in the neural circuit. This indicates that vocal output in the Undir that juvenile birds usually produce does not directly represent what they have learned to produce by that time, and that juvenile birds, at least by the middle of song learning, have already learned more about song structure than previously thought.
Juvenile birds normally produce numerous renditions of Undir during the process of developing adult song and can develop adult song normally even without any Dir (8, 18). The results here raise questions about the function of the many weeks of “suboptimal” undirected singing during learning, with implications for neural mechanisms of song learning, and motor learning more generally. Vocal learning is thought to proceed by trial-and-error (or reinforcement) learning, in which birds explore vocal space by randomly perturbing the vocal motor circuit and then selectively reinforce the motor activity patterns that generate better song (i.e., song closer to the adult target) (13, 20, 22–24). Previous studies have suggested that highly variable Undir, dependent on neural variability from basal ganglia circuitry, could serve as motor exploration (13, 20, 23, 25), whereas stereotyped Dir may reflect performance without motor exploration. In light of this model, the suboptimal production that we observe in Undir might seem to imply that undirected motor exploration is not random but, instead, highly biased in the direction of worse song (i.e., song with less adult-like structure). Our results point out how complex reinforcement learning is, however, when it is not simply a choice between two alternatives but involves the learner building up complex high-dimensional behavior by elaborately coordinating multiple muscles at the same moment. In such a multiparameter learning situation, even random perturbations of the motor program, like those hypothesized to occur during undirected singing, will predominantly result in less-coordinated muscle movements and degraded song structure, making the likelihood of better song production low during motor exploration (26). The directed context may turn off the random perturbations of the song motor program in juvenile birds, thereby revealing that the motor pathway is already much closer to adult song than the average undirected output suggested. The overall slowness of juvenile song learning in comparison with rapid adult song plasticity models (23, 25), in which only a single feature of otherwise fully mature song is measured and selectively reinforced, may be, at least in part, a consequence of the young bird needing to explore and optimize multiple parameters of vocal space simultaneously, as suggested by previous modeling work (22).
Rather than reflecting song with and without motor exploration, the striking difference in song maturity of Undir and Dir in our adolescent birds could, instead, reflect active shaping of an immature motor pathway by a more rapidly learning basal ganglia circuit, consistent with an instructive learning model of song learning as well as with recent work on adaptive changes in adult vocal patterns (20, 22, 25, 27, 28). By this model, instructive signals from the basal ganglia drive the motor pathway to generate song with more adult-like structure. Thus, during Dir, such instructive signals may be strong and consistent, resulting in stereotyped production of song with relatively adult-like structure, whereas in the undirected context, the instructive information may be weak and variable enough that it only occasionally results in song structure comparable to that of Dir. As has been suggested in homologous mammalian circuits (29), the output of the basal ganglia circuit may gradually “train” slower learning mechanisms in the motor pathway through numerous productions of Undir. Future studies, such as transiently eliminating basal ganglia input and investigating whether Undir structure becomes as close to the future adult song as Dir or whether Dir, instead, becomes less similar to adult song, are needed to distinguish the above two hypotheses.
Our results also have implications for neural mechanisms of much slower changes in juvenile song structure, the night-time deterioration, and the subsequent diurnal improvement described by Derégnaucourt et al. (18) in birds of this age. We found that even in the morning, when young birds produce Undir that is presumably more immature than that produced in the previous evening because of overnight deterioration, the young birds’ Dir is much more mature than their Undir. The magnitude of this context-dependent improvement in morning song was similar to, or even larger than, the subsequent Undir diurnal improvement, which is known to counteract the preceding night-time deterioration (18). This suggests that birds singing deteriorated Undir in the morning still retain much of their ability to produce song as good as that before the night-time deterioration. In addition, however, we found that like Undir syllables, Dir syllables also improve from morning to afternoon. This finding suggests that the mechanisms underlying context-dependent changes and daily oscillation are not the same, although they could be closely related. Further studies at the neural circuit level are clearly required to determine the neural mechanisms of these two kinds of vocal change in learning birds and the functional interactions between them.
Regardless of the mechanisms, our finding that the bird can rapidly change in song quality in response to the appearance of a female bird, ranging from variable production of different song variants to stereotyped production of the most adult-like song variant, provides strong evidence supporting a performance-practice dichotomy. This model suggests that undirected singing is a state of vocal practice in which birds improve or actively maintain their vocal pattern, whereas directed singing is a state of motor performance in which birds exploit what they have already learned through undirected singing to impress the female bird (17, 30). This idea was hypothesized for adult song; however, despite robust differences in neural activity between Dir and Undir (17, 20, 30–33), adult song is fully mature, only subtly different between contexts, and does not normally change in structure over time (15–17), thus not providing strong support for the hypothesis. Our study takes advantage of juvenile Undir, which is clearly motor practice for learning and still highly variable and developing in structure, and therefore was able to reveal marked differences in song variability and structure between social settings.
These results also further strengthen the striking parallels between birdsong and human speech, in a model system in which it is straightforward to examine the underlying neural changes. Social cues are well known to influence both sensory processing and motor production of speech powerfully (34–38). For instance, adult humans alter their speech when they are directing it toward infants rather toward other adults (38). Moreover, human infants increase the complexity and maturity of their babbling in response to brief social reinforcement from their mothers (37). There are, however, intriguing differences between the infant studies and the adolescent songbirds studied here. Those infants maintained the increased vocal quality of their vocalizations after the social reinforcement stopped, at least for the brief extinction period examined, suggesting that they had learned in response to social feedback, similar to what has been seen in cowbirds (39, 40). In contrast, juvenile zebra finches do not appear to change their song quality via learning in our experiment. They increased their song quality to a quite uniformly high level immediately after a female bird was introduced, and when the female was removed, they immediately lowered it again to the level of the preceding undirected singing. Thus, adolescent finches appear instead to have latently learned the motor program of higher quality song while singing Undir and to be able to produce it selectively in the presence of social signals that greatly increase the birds’ motivational state (41). Our results from songbirds highlight the powerful influence of social cues on motor production during learning more generally, and reveal that motor output during learning may underestimate the actual progress of competency.
We used 11 male zebra finches (Taeniopygia guttata) in this study. All birds were raised in individual breeding cages with their parents and siblings until at least 45 dph and were then housed individually in sound-attenuating chambers (Acoustic Systems) for song recording. The care and treatment of experimental animals were reviewed and approved by the Animal Care and Use Committee at the University of California, San Francisco.
We recorded Dir and Undir of juvenile and adult birds in a manner similar to that previously reported (17, 20) and analyzed their acoustic structure using the song-analysis software Sound Analysis Pro (SA+) (21) and custom-written software in Matlab (Mathworks, Inc.). Detailed methods of song recording and analyses are described in SI Materials and Methods. Briefly, we characterized the differences in trial-by-trial variability between Dir and Undir at both syllable and motif levels, based on six acoustic features (Wiener entropy, AM, FM, pitch, goodness of pitch, and mean frequency). We segmented all songs recorded in a single day into syllables and examined distributions of syllable acoustic features plotted against syllable durations. Also, in a subset of songs, syllables were identified and labeled based on their acoustic features and their variability was examined for each syllable type. For motif variability analysis, song motifs were extracted from the song in which syllables had been labeled, and the continuous time course of a set of acoustic features during motifs (including amplitude envelope) was compared between all neighboring motifs to obtain the CCs. Acoustic similarity of juvenile to adult syllables was obtained by comparing labeled juvenile syllables and “canonical” versions of matched adult syllables that were selected from adult syllable populations based on syllable acoustic features. Similarity of juvenile motif structure to that of adults was measured by comparing the continuous time course of the set of acoustic features above between juvenile and adult motifs. For each feature, individual juvenile motifs and the average of 16 randomly selected adult motifs were compared. The best match between each juvenile-adult motif pair was measured by computing the mCCs between the motif pair.
We thank J. R. Chen and members of the A.J.D. and Brainard laboratories for helpful suggestions regarding analysis methods and the manuscript. This study was supported by National Institutes of Health Grants MH055987 and MH550012 (to A.J.D.).
Author contributions: S.K. and A.J.D. designed research; S.K. performed research; S.K. and A.J.D. analyzed data; and S.K. and A.J.D. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1010502108/-/DCSupplemental.
