Juvenile zebra finches can use multiple strategies
to learn the same song
Wan-chun Liu*, Timothy J. Gardner, and Fernando Nottebohm
Field Research Center, The Rockefeller University, 495 Tyrrel Road, Millbrook, NY 12545
Contributed by Fernando Nottebohm, October 29, 2004
Does the ontogeny of vocal imitation follow a set program that,
given a target sound, unfolds in a predictable manner, or is it more
like problem solving, with many possible solutions? We report that
juvenile male zebra finches, Taeniopygia guttata, can master their
imitation of the same song in various ways; these developmental
trajectories are sensitive to the social setting in which the bird
grows up. A variety of vocal developmental trajectories have also
been described in infants. Are these many ways to learn unique to
the vocal domain or a hallmark of advanced brain function?
learning strategy ? social influence ? vocal learning ? zebra finch
acquired (1–4). The young bird must solve two problems. The
first one is to identify and commit to memory the model it will
imitate. The second one is how to achieve this imitation. The
solution to the second problem, which is the subject of this
report, could occur in various ways. For example, the young,
naive brain could have at its disposal the program for all sounds
used by that species, and then imitation would be a simple
process of selection (5). Or there could be for each class of adult
sound an unlearned starting point or prototype that is similar for
all individuals and that then is modified until a close match with
the model is achieved (4). A further possibility is that early vocal
ontogeny unfolds as a learning exercise that acquaints the young
brain with a range of vocal tract positions and their acoustic
consequences; experience gleaned in this manner can then be
be mutually exclusive, and there may still be others. Thus, there
is no compelling logic that tells us, in advance, how vocal
imitation will proceed in the young vocal learner. Similarly, we
do not know whether every member of a species learns its song
in the same way. If the answer is yes, then the trajectory followed
by any one individual would be akin to the unfolding of a preset
program, with little room for creativity or unexpected solutions.
Conversely, if each individual is capable of following a number
of different trajectories to achieve a same imitation, then the
cognitive level of the task would seem greater. We used juvenile
male zebra finches to explore these issues.
Zebra finches are highly social songbirds that breed colonially.
In nature, male juveniles learn their songs by imitating that of
their father or other adult males with whom they interact, often
copying different parts of the song from different adults (6, 7).
In the laboratory, juvenile males acquire and develop their songs
between 20 and 80 days of age, a time known as the sensitive
period for vocal learning. This period includes two partially
overlapping phases. During the first phase, from 20 to 50 days,
a juvenile male acquires, soon after fledging, the sensory mem-
ory of the song it will imitate (8). Observations we report here
suggest that midway during this same period, the second phase,
vocal imitation, gets under way. This second phase starts in the
context of the soft, highly variable and poorly structured sounds
of subsong. As subsong becomes more frequent and louder, it is
gradually modified into recognizable units of sound (‘‘syllables’’)
separated by silent gaps, and this more structured, yet still
variable sound is called ‘‘plastic song’’; over time the recogniz-
ocal ontogeny in songbirds provides a good model for
studying how a complex set of learned vocal signals is
able units become more and more like the syllables of the song
the bird is striving to imitate, and the song becomes more stable.
Sometime between days 80 and 90 the by now sexually mature
male produces its song with a degree of stability that approaches
that of older adults. This song changes little after this point. The
entire process that culminates with stable imitation of the model
requires auditory feedback (9).
Materials and Methods
Sound Recording. Juveniles reared in a family setting shared the
other cages of this type. One of us observed and recorded the
35 to day 50 after hatching (mostly at 0730–1230 and 1600–1800
hours). The early subsong (days 20–34) and advanced plastic
song (days 50–60) were recorded every other or every third day.
The style of singing did not seem to differ during the day. The
stage during which serial repetition of sound is most commonly
used usually lasts ?5 days. A Marantz (Itasca, IL; model 221)
tape recorder was used to record the songs of birds kept in a
family setting. Sound-triggered recording software was used to
record automatically for 7 h each day (0830–1530 hours) the
songs of juveniles housed individually with an adult male in a
Similarity Measurements. We measured the similarity (‘‘similarity
score’’) between two songs or two syllables by using the default
setting of the latest update of SOUND ANALYSIS 3.28 software (10).
In short, this procedure can detect the similarity of two songs or
entropy, and spectral continuity. The similarity score estimates
the proportion of sound in the song model for which there is a
close correspondence in the pupil’s song (10) or the proportion
of sound that bears such close correspondence between two
Quantifying the Incidence of Serial Repetitions. A song was defined
as a stream of sounds preceded and followed by a silent interval
of ?0.5 s; in the literature, this would correspond to a ‘‘song
bout’’; however, with our definition, it also applies to subsong.
For each song with more than three consecutive syllables, the
first syllable was compared with the second one (see below), and
then the second syllable was compared with the third syllable,
and so on. A song was considered to show serial repetition if
three or more consecutive syllables had a similarity score ?75%,
a phenomenon that rarely occurs in subsong or adult song. Had
repetition would have become too inclusive to be of interest.
Initially we collected 100 s of song (mostly between 1000 and
1200 hours) every other day from each of six males (from three
clutches) between posthatching days 30 and 60, and then we
calculated the percentage of songs that included serial syllable
repetitions. Then, for each of 23 males with similar song devel-
*To whom correspondence should be addressed. E-mail: firstname.lastname@example.org.
© 2004 by The National Academy of Sciences of the USA
December 28, 2004 ?
vol. 101 ?
no. 52 ?
opmental schedule, songs were collected at three different ages:
30–35 days (early subsong), 40–45 days (early plastic song), and
55–60 days (advanced plastic song). We used for sound analysis
the first 100 s of song from each day’s recordings (mostly
between 1000 and 1200 hours) that were not overlaid by the
vocalizations from other birds and used this sample to estimate
for each bird the percentage of songs that showed serial repe-
tition on that day (see Fig. 5).
Multiple Strategies for Vocal Imitation. Our first evidence that zebra
finches have multiple strategies for mastery of a vocal imitation
came from visual inspection of the sound spectrographic record of
song ontogeny in 37 family-reared juvenile zebra finches from 15
different clutches. These birds were kept, until 90 days of age, in
cages that they shared with their parents and siblings. The term
‘‘strategy’’ is used here in a descriptive manner.
The two main strategies highlighted in Fig. 1 were obvious
posthatching days 20–35 but most commonly around day 30, was
similar in all juveniles, but 9–10 days after the beginning of
subsong, a difference in how plastic song evolved could be seen
among individuals. The developmental course of this difference
is shown in greater detail in Fig. 2. In the ‘‘serial repetition’’
strategy (n ? 18), an approximation to one syllable of the model
is repeated many times (4). Often, but not always, different
syllables from the final imitation emerge, through modification,
from repetitions of the single syllable; they do so already in the
order, relative to each other, in which they will appear in the
adult song (4). So, for example, four different syllables from bird
repetition of a same early syllable that was very similar to the
model’s syllable A. The serial repetition stage lasted in this bird
?5 days, with some infrequent recurrence of serial repetition
The second, ‘‘motif’’ strategy (n ? 19 males) is exemplified by
became apparent 8–10 days after onset of subsong. Bird 2
attempted early on a global imitation of its father’s song,
including the periods of sound and silence. The original sounds
were noisy and imprecise, but they became more structured and
stereotyped over a period of days and weeks. In this bird, each
of the four syllables in the adult stable song could be traced back
to four separate precursors delivered, as early as posthatching
days 38–40, in a motif-like serial order, with silent intervals
separating them. There were relatively few serial repetitions of
any syllable in this bird at any point in song ontogeny.
The song development of bird 3 (Fig. 2) revealed another
version of motif strategy. In this second motif strategy, all
that normally separate syllables in adult song. This manner of
singing was already recognizable by posthatching day 40; at that
time, the forerunners of four syllables (B–E) that eventually
became a close match of the model’s syllables were produced as
an uninterrupted sound. Whereas the differences between the
serial repetition strategy and the motif strategy are easily
recognized at times when one or the other predominate, the
differences between the two motif strategies described above for
birds 2 and 3 are bridged by intermediate cases, where silent gaps
between the different syllable precursors can be more or less
obvious. The remarkable facts about the three birds in Fig. 2 are
that they were siblings, they were members of the same clutch,
and they all imitated the same model.
Time is on the horizontal axis, and frequency is on the vertical axis. Shown for each bird is a sequence of sounds and pauses just as they were delivered, except
for the silent interval between consecutive songs, which is not shown. Boundaries between consecutive songs are indicated by yellow dots. The way in which
these two strategies emerge during ontogeny is shown in greater detail in Fig. 2.
This figure uses time-frequency analysis (10) to illustrate the striking difference in strategy followed by two groups of 43-day-old zebra finch juvenile
www.pnas.org?cgi?doi?10.1073?pnas.0408065101Liu et al.
We were concerned that the appearance of strategy differ-
ences based on visual inspection of sound-spectrographs might
be influenced by our own perception of what the birds were
doing, and so we resorted to a more objective, computer-aided
analysis (see Materials and Methods) of song development in the
same 15 clutches discussed above. In this more rigorous quan-
titative analysis, we examined the incidence of syllable repetition
(Figs. 3a and 4) and the time course of song imitation (Fig. 3b).
For this more quantitative approach, we used only 23 birds; these
birds came from clutches that had two to five males, and the song
of each of them had, by posthatching day 60, a 70% or better
similarity with the model song. We restricted the sample in this
manner because we wanted males that had a similar time frame
of song development, that were similarly successful at matching
their model, and that had at least one other male sibling in their
clutch. For a small subset of 6 birds, song samples were taken
the 1-s-long song motif of their father, labeled ‘‘Tutor’’ in the box on top. The vertical axis corresponds to frequency, in kilohertz, and the horizontal axis
corresponds to time, in seconds. The highly stereotyped song motif of the father consisted of five different syllables, identified by letters A–E; each syllable is
of similar duration also occur in juveniles. The vocal ontogeny path that each sibling followed until it achieved a close imitation of the tutor song motif is shown
for each day the best match with the tutor song motif. Further refinements in imitation, not shown here, continue to occur after day 60; by day 90 adult, stable
versions, labeled A?, B?, etc., which were already in place by about 40 days. For more details, see Results.
The ontogeny of the song motif in three juvenile zebra finch siblings (birds 1, 2, and 3). The siblings were members of the same clutch and mastered
Liu et al.
December 28, 2004 ?
vol. 101 ?
no. 52 ?
every other day (Fig. 3a), but it was not possible to do this for
all 23 birds, so song samples for them were taken at three times
(32, 43, and 60 days after hatching; Fig. 4). We obtained for each
of these 23 birds, at each sampling time, a mean similarity score
for contiguous sounds; these were sounds separated from the
previous and subsequent ones by a silent interval of at least 20
ms. During subsong (posthatching day 32), the mean similarity
between contiguous sounds tended to be, in all juveniles, low.
During early plastic song (posthatching day 43), the range of
similarity between contiguous sounds was much broader. In
approximately half of all birds sampled, similarity scores re-
mained at this stage as low as during subsong; in the other half,
mean similarity scores were significantly higher than during
subsong (Fig. 4). Inspection of the sound spectrographs of the
song of birds in these two groups revealed, respectively, birds
using predominantly the motif and serial repetition strategies.
On the last sampling day (posthatching day 60), all birds had
converged on a comparably sparse incidence of serial repetition,
and their songs showed a close approximation to what would be
The results shown in Fig. 4 suggest that there is on day 43 a
clear separation between individuals that follow predominantly
a motif strategy and those that follow predominantly a serial
repetition strategy of song ontogeny. Next, we wanted to deter-
mine the actual incidence of serial repetition during the three
23 birds described in the previous paragraph, and these results
are presented in Fig. 5a. For this test we defined as serial
repetition the occurrence of three contiguous sounds that
in a subset of 6 of our 23 birds that were followed from early subsong (day 32) until the birds produced a clearly recognizable approximation of the tutor song
on day 60; three of the birds sang in the serial repetition manner (Left Upper) from day 40 to day 50; the other three sang during this same time in the motif
manner (Left Lower). In both cases, for each song sample, syllable A is compared with B, B with C, and so on. (b) For the same six birds, we identified the song
that on day 60 (Upper) came closest to the tutor song for that bird. Then we estimated the similarity index obtained when comparing that song and another
song on each of several previous days, going back to day 32. The results for each day are based on a sample of 100 songs per bird. Notice that on day 45, three
(color identification as in a). In this case the two panels in Left emphasize the type of comparison, between earlier and later song that went into the graph.
Two approaches that quantified changes of song during the first 60 days. (a) The graph shows mean similarity scores for contiguous syllables recorded
whole group of 23 birds; scores were obtained by using three recording times
meant to characterize events during early subsong (day 32), early plastic song
(day 43), and later plastic song (day 60). Note that on day 43 similarity scores
clump into two separate clusters that correspond, respectively, to the serial
repetition and motif strategies.
Mean similarity score for contiguous syllables, as in Fig. 3, for the
www.pnas.org?cgi?doi?10.1073?pnas.0408065101Liu et al.
showed at least 75% similarity. The range of serial repetition
varied from 0% to 30% during the early subsong stage (30–35
days old), but it broadened during the next sampling period
(40–45 days old). In one group of birds (n ? 10), 40–80% of all
songs included serial repetitions; these birds were ‘‘high repeat-
ers.’’ In the other group (n ? 11), the incidence was lower,
0–20% (Fig. 5a), and we shall refer to them as ‘‘low repeaters.’’
Two birds straddled the two groups. It was among birds in the
low repeater group that we found our best examples of early
motif organization, which, in turn was much less common among
high repeaters. Subsequently, the performance of these two
groups of birds converged, so that by days 55–60 the incidence
of serial repetition was fairly low in all 23 birds.
In all 15 clutches sampled, members of the same clutch
commonly used different strategies to master their imitation of
the same tutor, as shown by the three siblings in Fig. 2. In all 15
clutches studied, at least one male sibling in each clutch used
predominantly the motif strategy (n ? 18), and one used
predominantly the repetition strategy (n ? 16). In our 15-clutch
sample, the first sibling to start singing adopted a motif strategy
in five cases and a serial repetition strategy in seven cases. (For
three other clutches, we were unsure which bird was the earliest
singer.) It appears that male zebra finches learning their songs
avoid the strategy of the sibling whose song is closest in devel-
Interestingly, juveniles favoring the serial repetition or the
motif strategies were able to acquire a comparably accurate
imitation of the model song. When a clutch included more than
two males, not every juvenile produced a good imitation (11).
The male sibling that produced, in each clutch, the best imitation
(?90% similarity with the model song) used during the 35- to
50-day period either a low-repetition (n ? 7) or high-repetition
(n ? 8) strategy to develop its song, suggesting that both
approaches are similarly effective for mastering an imitation.
Social Influence on Learning Strategy. We speculated that juveniles
within the same clutch use different song development strategies
to mitigate the inhibitory effect of another sibling’s song on the
process of model imitation (11). We tested this idea by removing
12 juveniles aged 20–25 days from their mothers and siblings,
housing each of them in a separate sound-proof chamber with an
adult male zebra finch, and recording how each juvenile devel-
oped its song. Eleven of the 12 males produced, eventually, a
good imitation of the adult model (similarity score of ?75%),
the 40- to 45-day period a pure motif (low repetition) strategy;
one bird used a high repetition strategy and did not sing in the
motif manner during the 40- to 45-day period. The remaining
seven produced during that time a fair approximation of the
entire motif that they would sing at 90 days while showing, at the
same time, a frequent occurrence of serial repetitions; so theirs
was a ‘‘mixed’’ strategy. Moreover, whereas only 1 bird out of the
11 included three or more contiguous repetitions in ?50% of all
out of 23 met this criterion in the family-reared group (Fig. 5a),
and this difference was significant (Fisher’s exact test, P ? 0.01).
We infer from this comparison that, whereas singly reared
juvenile zebra finches can develop their imitations of a song
model by using either the repetition or motif strategies or a
mixture of both, options that therefore are probably available to
all juvenile males, a greater polarization of developmental
trajectories occurs when siblings exposed to a single model grow
up together. It is possible that in nature, where juvenile zebra
finches are normally exposed to a variety of models that they can
copy (7), this polarization is not as great as it is in single-family
The similar effectiveness of high repetition and low repetition
meant to show the poorly structured and diverse sounds of subsong. During
plastic song, some birds follow from early on a motif trajectory (left), others
branching points. Toward the end of the process, all these trajectories con-
verge on the same final output.
A diversity of trajectories can lead to the same end. The top line is
recordings made from 23 males kept with their families during the subsong
(days 30–35), early plastic song (days 40–45), and later plastic song (days
birds were recorded on the same day. (b) Sound recordings made every other
day from day 30 until day 60 from 12 birds kept singly with an adult. The
vertical axis indicates the percentage of all songs produced by a bird during
those times that included three or more serial repetitions that met the
criterion of 75% similarity.
Liu et al.
December 28, 2004 ?
vol. 101 ?
no. 52 ?
strategies in song development can be tested in still another way. Download full-text
Fig. 3b shows that low repeaters (i.e., motif strategy birds) and
32. After this the low repeaters approach the model faster, yet
by day 60, the difference between the two groups has disap-
peared. Because zebra finch males do not breed before age
70–80 days, at the earliest (6), our conclusion stands that both
strategies of song imitation should be equally serviceable to the
birds that use them.
An earlier study (4) suggested that song imitation in zebra
finches commonly started with successive repetitions of a same
precursor syllable. In that study, each juvenile was raised singly
by its mother until it was 30 days old and then housed singly in
a sound-proof chamber, without any access to a social tutor or
tutor song until 43 days of age, when this juvenile was allowed
to peck at a key that resulted in song playback (each time the
same song, identically repeated) emanating from the chest of a
plastic adult male model. In the present study, the variety of
strategies available for song learning became apparent, partic-
ularly in the juvenile males that were raised and acquired song
in a family setting, which is the natural social environment for
juveniles. It is possible that a variety of trajectories for achieving
song imitation are always part of vocal learning and that it is only
under conditions of severe social deprivation, invariant repeti-
tion of the playback signal, and?or delayed exposure to a model
that less diverse trajectories are encountered.
Although we have emphasized two different ways of imitating
a model, one with many repetitions of an early syllable and one
providing a coarse approximation to the whole motif, there are
also cases in which birds seems to use both ways, at times singing
in the motif manner and at other times repeating many times the
same syllable. In addition, even repetition might be used to more
than one end, for example, to rehearse a single syllable (figure
1 in ref. 4) or to give rise to multiple syllables (figure 5 in ref. 4
and this report). Moreover, the route followed by any one bird
can move from one strategy to the other and even explore
variants on which we have not dwelled (Fig. 6). For this reason,
it is probably best to say, quite simply, that song can be imitated
in several different ways.
Zebra finches and other estrildid finches are special in that
song learning occurs while juveniles are still close to other family
members. These birds may encounter, from the time they fledge
until they reach independence at ?40 days, a greater variety of
social circumstances than other less colonial and gregarious
songbirds. For this reason, the variety of vocal learning styles
found in zebra finches need not apply to all oscine songbirds.
in less social oscine song birds (12). Moreover, the variety of
strategies may have evolved not only to cope with variable social
settings, but to enable a young bird to imitate either a whole song
or just parts thereof, or even, in the case of the high repeaters,
provide a means for not committing prematurely to a particular
song model. Because the many syllables of a song occur, from
their earliest expression, in the order in which they will appear
in the adult song, syllable repetition can be thought of as a way
of holding ‘‘slots’’ for syllables that have not yet been acquired,
thereby extending the time during which a juvenile bird can
acquire song material by imitation. In sum, the variety of
strategies for vocal imitation found in zebra finches may be the
norm rather than the exception. Although it had been reported
before that social bonds play an important role in vocal learning,
we are surprised that this interaction starts so early in vocal
ontogeny and that it involves not only the bond with an adult but
also relations among siblings.
As in zebra finch juveniles, infants too show remarkable
variability in the way in which they achieve eventual mastery of
on serial repetitions of a same word, others go through a stage
where they use fairly imprecise short phrases for which, even-
tually, the words become clearer. The manner of speech of the
latter children has been compared with the cadence, in terms of
inflexion and emphasis, of the adult phrase, although it still lacks
recognizable individual words. These two strategies are very
different, and children often use both in a way that cannot be
related to the parents’ efforts to guide speech development
(13–16). Thus, in both infants and zebra finches, vocal learning
does not unfold in a preset manner but rather emerges as an
exercise in problem solving that leaves much room for external
influences and individual learning styles. Overall, our observa-
tions add in unexpected ways to the many known similarities
between vocal learning in birds and humans (17, 18). As in the
past, we are struck by the fact that songbirds, with brains 1,000
times smaller than those of humans and with a very different
evolutionary history, go about vocal learning, nonetheless, in
ways that are not all that different from ours.
We would like to end with a question. Did the multiple ways
to learn sounds evolve to meet the special needs of vocal learning
or do they reflect a general ability of advanced brains to master
skills in a variety of ways?
In addition, Dr. Tchernichovski provided invaluable advice with use of
the SOUND ANALYSIS software that we used for analyzing our zebra finch
song recordings. We also thank Daun Jackson, Sharon Sepe, and Helen
Ecklund for their expert care of our birds. This research was conducted
with the support of Public Health Services Grant MH18343, and W.-c.L.
was supported by a Li Memorial Scholar Fund fellowship. This work was
made possible, too, by the generous support of the Mary Flagler Cary
Phipps Family Foundation.
1. Thorpe, W. H. (1958) Ibis 100, 535–570.
2. Marler, P. J. (1970) Comp. Physiol. Psychol. 71, Suppl., 1–25.
3. Nottebohm, F. (1972) J. Exp. Zool. 179, 35–49.
4. Tchernichovski, O., Mitra, P. P., Lints, T. & Nottebohm, F. (2001) Science 291,
5. Marler, P. & Pickert, R. (1984) Anim. Behav. 32, 673–689.
6. Zann, R. A. (1996) The Zebra Finch: Synthesis of Field and Laboratory Studies
(Oxford Univ. Press, Oxford), pp. 157–195.
7. Williams, H. (1990) Anim. Behav. 39, 745–757.
8. Bohner, J. (1990) Anim. Behav. 39, 369–374.
9. Price, P. H. (1979) J. Comp. Physiol. Psychol. 93, 260–277.
10. Tchernichovski, O., Nottebohm, F., Ho, C. E., Pesaran, B. & Mitra, P. P. (2000)
Anim. Behav. 59, 1167–1176.
11. Tchernichovski, O. & Nottebohm, F. (1998) Proc. Natl. Acad. Sci. USA 95,
12. Baptista, L. F. & Gaunt, S. L. L. (1997) in Social Influences on Vocal
Development, eds. Hausberger, M. & Snowdon, C. (Cambridge Univ. Press,
London), pp. 23–40.
13. Macken, M. A. & Ferguson, C. A. (1983) in Children’s Language, ed. Nelson,
K. E. (Lawence Associates, Hillsdale, NJ) , Vol. 4.
14. Menn, L. & Stoel-Gammon, C. (2001) in The Development of Language, ed.
Gleason, J. B. (Allyn and Bacon, Boston), 5th Ed., pp. 70–124.
15. Nelson, K. (1973) Monogr. Soc. Res. Child Dev. 38, 1–1379.
16. Peters, A. M. (1977) Language 53, 560–576.
17. Marler, P. (1970) Am. Sci. 58, 669–673.
18. Doupe, A. J. & Kuhl, P. K. (1999) Annu. Rev. Neurosci. 22, 567–631.
www.pnas.org?cgi?doi?10.1073?pnas.0408065101 Liu et al.