Dynamics of the Vocal Imitation Process: How a Zebra Finch Learns Its Song

Thorpe W. H., Ibis 100, 535 (1958).

Konishi M., Z. Tierpsychol. 22, 770 (1965).

F. Nottebohm, in The Design of Animal Communication, M. Hauser, M. Konishi, Eds. (MIT Press, Cambridge, MA, 1999), pp. 63–110.

K. Immelmann, in Bird Vocalization, R. Hinde, Ed. (Cambridge Univ. Press, Cambridge, 1969), pp. 61–74.

R. E. Zann, The Zebra Finch: Synthesis of Field and Laboratory Studies (Oxford Univ. Press, Oxford, 1996).

Tchernichovski O., Nottebohm F., Proc. Natl. Acad. Sci. U.S.A. 95, 8951 (1998).

Eales L. A., Anim. Behav. 33, 1293 (1985).

Jones A. E., ten Cate C., Slater P. J. B., J. Comp. Psychol. 110, 354 (1996).

Troyer T. W., Doupe A. J., J. Neurophysiol. 84, 1204 (2000).

Doupe A. J., Kuhl P. K., Annu. Rev. Neurosci. 22, 567 (1999).

Young males were raised by their mothers (no adult male present) until they were 30 days old. Each juvenile was then placed singly in a soundproof chamber that contained a plastic model of an adult zebra finch male (Fig. 1A). At 35 to 40 days after hatching, birds started to produce soft subsong, recorded for each bird on days 42 and 43. Training started on day 43. Within 36 hours, most birds (42 out of 50) began to peck at either one of two keys (31, 32); the 8 birds that failed to peck by that time were not included in the data. We provided two keys to encourage pecking activity (e.g., to overcome preferences to one side of the cage). Pecking either one of the keys induced the playback of the same short (1.4 s) span of song from a tiny speaker placed inside the plastic bird (32). Each playback consisted of two identical repetitions of a single song motif recorded from an adult bird. Training day 1 was taken as the first day the bird activated the song control keys. Each day consisted of two training sessions (32). In brief, during each session, we reinforced the first 10 key pecks with a song playback. Additional key pecks were allowed but were not reinforced, so that the overall daily quota of model song that a bird could trigger was, at most, 28 s. We trained a total of 42 birds with one of four different model songs, and 12 additional birds were kept in the training apparatus as controls and not trained. A few minutes of song were recorded digitally (16 bits, 44.1 kHz) from each bird at least once a day during days 42 to 52 after hatching and at least once a week thereafter, until day 90 (by which time they produced stable song typical of adults).

Tchernichovski O., Nottebohm F., Ho C. E., Pesaran B., Mitra P. P., Anim. Behav. 59, 1167 (2000).

Clark C. W., Marler P., Beeman K., Ethology 76, 101 (1987).

Fee M. S., Shraiman B., Pesaran B., Mitra P. P., Nature 395, 67 (1998).

Similarity measurements were performed as described in (12) and (18). Briefly, two sounds were considered similar if feature analysis for the two sounds yielded at least 90% similarity. In tracing imitation trajectories, we took a conservative approach and also verified the trajectories a posteriori by visual inspection. For example, a syllable from the mature song was compared to a random sample of 10 s of song recorded on an earlier day (e.g., day 83). If the similarity criterion was met and an earlier version of the sound was identified, then this earlier version was used as a point of entry for the next comparison with a still earlier sample of song, and so on recursively. This approach circumvents the need to partition juvenile sounds, because the similarity section automatically provides boundaries. Although the procedure could trace similarity of one sound to two or more earlier fragments (e.g., to test for sound translocation), robust implementation of this capability would require continuous recording, which was not done in the present study. To ensure that imitation trajectories were reproducible, we repeated the procedure using different 10-s samples of sound from each day that song was recorded. We confirmed that an imitation trajectory was reproducible by calculating the similarity across the two independently identified antecedents. We maintained confidence in the trajectory as long as the significant similarity between antecedent versions on a same day was above 80% [(18), pp. 43–45].

Harmonic stacks are composed of nonmodulated harmonic frequencies, where the fundamental frequency defines the pitch. Harmonic stacks were traced automatically as described (12, 15, 18). Even though period doubling causes a discontinuous change of pitch, tracing is still possible with the other features. Pitch was measured in the middle of the traced sound.

When the initial pitch (at the beginning of the trajectory) was lower than the target, reaching the first harmonic would require that the fundamental be crossed first, matching the pitch to the model sound before period doubling could occur, which then would seem to make period doubling unnecessary.

O. Tchernichovski, P. P. Mitra, Sound Analysis User Manual, version 1.01 (ed. 2, 2000) (available at ).

The feature diversity d(t) is a scaled average (across birds and features) of the standard deviations σ_α(t, b) of song features α = 1, …, 4, computed for a given bird b and for a given time epoch t (t denotes the day of the 10-s sample) with the samples of the features involved during that 10-s epoch. The SDs for different features have different scales; this is taken care of by defining a scale for each feature by pooling all observations for that feature across birds b and measurement windows t

${\bar{\sigma }}_{\alpha }$

where T is the total number of epochs and N_b is the number of birds. For each training day we compute

$d$

The SE of the feature diversity is computed for a given epoch by treating the birds as different statistical samples. The error bars on Fig. 3C are SEs of the mean feature values. To obtain these, we computed the mean feature values for individual birds for a given epoch; the grand mean of these individual means is then plotted. The SEs are computed as before by treating the birds as different statistical samples. This measure should not be confused with the feature diversity, which denotes the spread of feature values within individual birds. The two are not, in principle, related.

Dave A. S., Yu A. C., Margoliash D., Science 282, 2250 (1998).

Hultsch H., Todt D., Anim. Behav. 44, 590 (1992).

R. E. Stark, in Child Phonology, G. H. Yeni-Komshian, J. F. Kavanagh, C. A. Ferguson, Eds. (Academic Press, New York, 1980), pp. 73–92.

D. K. Oller, in Child Phonology, G. H. Yeni-Komshian, J. F. Kavanagh, C. A. Ferguson, Eds. (Academic Press, New York, 1980), pp. 93–112.

M. M. Vihman, in Developmental Neurocognition: Speech and Face Processing in the First Year of Life, B. de Boysson-Bardies et al., Eds. (Kluwer, Dordrecht, Netherlands, 1993), pp. 411–419.

Roug L., Landberg I., Lundberg J., J. Child Lang. 16, 19 (1989).

Period was calculated by implementing the similarity measurement (15) as follows. Starting from a randomly selected frame within a 10-s epoch, the time frame is first moved forward from the point of origin until it encounters a frame that is less than 75% similar to the sound at the selected frame, so as to depart from the original sound. The time window is moved further forward, until it encounters a sound that is at least 90% similar to the original frame. This gives an instantiation of the time elapsed between two similar sounds. This procedure is repeated 50 times, each time starting from a different randomly selected frame within the 10-s epoch, and the median of the durations thus identified is a measure of periodic structure in the song. We call this the period.

To characterize the overall temporal structure of the juvenile song with reproducible statistical results, we partition sounds to syllables (continuous sounds) based on “silent intervals,” where no frequency contours are detected (12). Although the statistical estimate of the number of syllables is reproducible, the units themselves (syllables) are not equally well defined in subsong and song and would not be suitable for automated categorization.

To test for a possible effect of training on the transition from repetitive to serial delivery of syllables, we raised 12 control birds without exposure to the model song under housing conditions otherwise identical to those encountered by the birds that were trained. We found that the period of the mature song in the control birds was significantly shorter than the final period in the trained birds (366 ± 66 ms versus 524 ± 26 ms, P < 0.01) and was within the range of the period soon after the onset of training.

Price P., J. Comp. Physiol. Psychol. 93, 260 (1979).

Our software for automated recording, song recognition, and real-time analysis of the imitation process, as well as the raw sound data, are available (at no charge) at.

Adret P., Anim. Behav. 46, 149 (1993).

Tchernichovski O., Lints T., Mitra P. P., Nottebohm F., Proc. Natl. Acad. Sci. U.S.A. 96, 12901 (1999).

We thank M. Konishi, M. Schmidt, P. Marler, and the reviewers for their useful comments; the staff of The Rockefeller University Field Research Center for their technical support; and the TalkBank project for providing Internet storage. Supported by U.S. Public Health Service (PHS) grant DC04722-01 (O.T.), PHS grant MH18343 (F.N.), the Mary Flagler Cary Charitable Trust, the Herbert and Nell Singer Foundation, and the Phipps Family Foundation.