Upload
hans-rudolf-guettinger
View
235
Download
1
Embed Size (px)
Citation preview
ORIGINAL PAPER
Human melody singing by bullfinches (Pyrrhula pyrrula)gives hints about a cognitive note sequence processing
Jurgen Nicolai • Christina Gundacker •
Katharina Teeselink • Hans Rudolf Guttinger
Received: 6 November 2012 / Revised: 23 April 2013 / Accepted: 20 May 2013
� Springer-Verlag Berlin Heidelberg 2013
Abstract We studied human melody perception and pro-
duction in a songbird in the light of current concepts from the
cognitive neuroscience of music. Bullfinches are the species
best known for learning melodies from human teachers. The
study is based on the historical data of 15 bullfinches, raised
by 3 different human tutors and studied later by Jurgen
Nicolai (JN) in the period 1967–1975. These hand-raised
bullfinches learned human folk melodies (sequences of
20–50 notes) accurately. The tutoring was interactive and
variable, starting before fledging and JN continued it later
throughout the birds’ lives. All 15 bullfinches learned to sing
alternately melody modules with JN (alternate singing). We
focus on the aspects of note sequencing and timing studying
song variability when singing the learned melody alone and
the accuracy of listening-singing interactions during alter-
natively singing with JN by analyzing song recordings of 5
different males. The following results were obtained as
follows: (1) Sequencing: The note sequence variability
when singing alone suggests that the bullfinches retrieve the
note sequence from the memory as different sets of note
groups (=modules), as chunks (sensu Miller in Psychol Rev
63:81–87, 1956). (2) Auditory–motor interactions, the
coupling of listening and singing the human melody:
Alternate singing provides insights into the bird’s brain
melody processing from listening to the actually whistled
part of the human melody by JN to the bird’s own accurately
singing the consecutive parts. We document how variable
and correctly bullfinches and JN alternated in their singing
the note sequences. Alternate singing demonstrates that
melody-singing bullfinches did not only follow attentively
the just whistled note contribution of the human by auditory
feedback, but also could synchronously anticipate singing
the consecutive part of the learned melody. These data
suggest that both listening and singing may depend on a
single learned human melody representation (=coupling
between perception and production).
Keywords Songbird �Melody perception and production �Sequencing and timing � Auditory–motor interactions �Feedback between hearing and singing � Internal
representation
Introduction
In this manuscript, we analyze human melody and alternate
singing in bullfinches in terms of current concepts from the
cognitive neurobiology of music perception and production
in humans (e.g. Janata and Grafton 2003; Sakai et al. 2004;
Brown et al. 2004, 2006; Zatorre et al. 2007; Chen et al.
2009; Tierney et al. 2011). Music performance is currently
considered to be one of the most complex and demanding
cognitive challenges that human mind can undertake. It
offers a view of the brain in which acoustic perceptual and
motor systems are coupled across multiple levels of pro-
cessing. Melody singing requires precise timing of several
hierarchically organized actions as well as accurate control
over different pitches and durations of consecutive notes.
Several cortical and subcortical regions including the basal
ganglia, the supplementary motor area and the cerebellum
Jurgen Nicolai: Deceased in 2006.
Electronic supplementary material The online version of thisarticle (doi:10.1007/s10071-013-0647-6) contains supplementarymaterial, which is available to authorized users.
J. Nicolai � C. Gundacker � K. Teeselink � H. R. Guttinger (&)
Abteilung fur Allgemeine Zoologie, FB Biologie, Universitat
Kaiserslautern, P.B. 3049, 67653 Kaiserslautern, Germany
e-mail: [email protected]
123
Anim Cogn
DOI 10.1007/s10071-013-0647-6
have been implicated in the learning and production of note
sequences in humans.
Until recently, comparative neuroanatomy did not rec-
ognize cortex-like structures and functions in the avian
forebrain. However, the Avian Brain Nomenclature Con-
sortium (Jarvis et al. 2005) disproved this classical thinking
about avian brain. In the songbird, neural coding of syn-
tactic structure in learned vocalizations has been detected,
suggesting that individual neurons can acquire syntactic
attributes in the course of vocal learning (Fujimoto et al.
2011). Thus, the bird’s brain is neither a precursor nor a
primitive form of the mammalian brain but a highly and
differently organized structure. Given the new under-
standing of avian brain organization and function, which
demonstrates similarities to mammalian sensory and motor
cortices, our study on human melody singing in bullfinches
promises insights into whether a similar cognitive pro-
cessing may also occur in songbird’s brain.
We focus on the following two aspects:
1. Sequencing and timing (rhythm), studying the vari-
ability (errors) of singing the melody solo, without a
human partner.
We studied whether the bullfinches memorized and
recall the note sequence of the melodies as a coherent,
linear chain or, as humans do (Janata and Grafton 2003), in
smaller sub-sequences (or modules), in chunks. We use
‘‘module’’ here to refer to the objective breakdown of a
melody into sub-melodies and ‘‘chunking’’ for the inferred
cognitive processes of correct memorizing and recalling
the note sequence.
The significance of chunking derives from the ability to
overcome objections to linear models of serially organized
behavior by augmenting linear with hierarchical structures.
Chunking is regarded as a basic cognitive process. It pro-
vides a window onto cognitive processes that guide or
interrupt otherwise automatic processes (Sakai et al. 2004).
Processing beyond the boundary of a chunk requires a
cognitive selection of the next chunk, whereas processing
within chunks is carried out more automatically. Chunks
can be identified by long variable pauses and elevated
transitional error probabilities after these pauses (Terrace
2001). Phrasing by inserting pauses into a sequence can
either facilitate or retard serial learning for humans and
animals (Wallace et al. 2008).
For songbirds, organizing songs in chunks has already
been reported for the Zebrafinch (Taeniopygia guttata)
(Williams and Staples 1972), for Nightingales (Luscinia
megarhynchos) (Hultsch and Todt 1989) and the Bengalese
finch (Lonchura striata) (Suge and Okanoya 2010).
We studied whether the bullfinches retrieve the learned
human melodies in sub-melodies (modules) separated by
pauses according to the note pattern whistled by the human
tutor and how bullfinches assemble modules into a correct
melody.
2. Auditory–motor interaction, that is, coupling between
listening, perceiving and singing melody parts of
bullfinches during alternate singing with its human
partner.
We analyzed the accuracy of the bullfinch’s choices and
how a bird continues after the human partner pauses,
focusing on whether the bird chooses the appropriate note
pattern. We studied the precision of the bird’s responses to
a variable challenge, that is, to the group of notes that had
just been whistled by its human partner. When faced with a
variety of possible entry points into a melody, the bird must
choose the correct note sequence and the exact time to cue
into continue the melody where the human partner just
ended. Alternate singing, when various entry sites are
possible, that is whenever the human stops, provides
information about how the bullfinch acoustically processes
the note sequence whistled by the human and thus how
auditory–motor interactions in the bird’s brain retrieve the
still remaining parts of the notes sequence. It requires a
departure from routinely singing the sequence from the
beginning to the end, to cope with challenges that are not
precisely predictable. Does the bird compare the notes, or
note groups, just heard with its mental representation of the
learned human melody stored in its memory? To be able to
choose the correct consecutive melody module, the bird
probably has to engage in short-range planning.
In order to understand our approach, Melody 1 in the
original recording (Electronic Supplementary Mate-
rial = ESM, S1) and the note sequences of the visual
display both spectrograms and the musical notation (Fig. 1)
should be followed simultaneously. Listening to the mel-
ody and simultaneously following its notation in notes and
spectrograms enables a more stringent recognition of the
phrasing into smaller sub-melodies and better detection of
sequential errors than does examining the visualized pre-
sentations by themselves.
To determine where the bullfinch starts singing after the
human partner stops, the notes uttered by the bullfinch or
whistled by JN must be distinguished with high reliability.
This distinction can be accomplished both by listening and
by analyzing spectrograms (Fig. 2, ESM: S4): The notes of
the bullfinches sound more sonorous and vary far less in their
dynamics. On the visual display, the bird’s notes consist of a
fundamental, which determines the pitch, and a row of
overtones that are responsible for their flute-like timbre.
Teaching birds to imitate human melodies was a popular
hobby in the 18th and 19th centuries and the bullfinch was the
favorite species (e.g. Hamersley 1717; Pernau 1768; Brehm
1832; Holden 1895; Naumann 1900). In his pioneering song
study, ‘‘Family tradition in the song development of the
Anim Cogn
123
bullfinch’’ Nicolai (1959) stated that the bullfinch’s indi-
vidual bonding to its father or to its foster parent (canary or
human tutors) is the crucial factor in focusing the attention of
the juvenile bullfinches and enabling them to learn their song
patterns. He noted later (Nicolai 1969) that his human mel-
ody-singing bullfinches would sometimes omit sections and
if JN began whistling the omitted parts, the birds would begin
singing again, completing the missing modules correctly. All
15 human melody-singing bullfinches could produce con-
secutive modules in the correct-order chronological order
during alternate singing.
Alternate singing has never been reported in wild bull-
finches, but in other species, duets between pair partners
occur regularly, especially in the tropics (e.g. Thorpe 1972;
Seibt and Wickler 2000).
Materials and methods
Animals, melody tutoring program
The study is based on the historical data of 15 bullfinches,
hand-raised by 5 different human tutors kept and studied by
JN within 1967–1975 at the Max Planck-Institut fur Ver-
haltensphysiologie at Seewiesen. All 15 bullfinches learned
to sing melody modules alternately with JN. The five
bullfinches used in this study were bought at the age of
6 months, from 3 different tutors (Christian Grosch, Karl
Moller and Karl Dorr) at Angersbach, near Fulda, a former
center for tutoring bullfinches to sing human melodies
(Lichau 1989).
Melody tuition started before fledging, generally in June
of the birds’ hatching year. The tutors did not tune their
whistling for absolute pitch (frequency). Pitches of their
entire melody performances recorded at different days
varied within the range of 3 semitones. Their performances
were defined by relative pitches. Unlike the humans, the
bullfinches sang the melodies at high-pitch constancy
(Guttinger et al. 2002).
By December, after daily training for 5–6 months, the
bullfinches often sang the tutored note sequence with
excellent accuracy. However, even birds that were excel-
lent melody singers during their first winter often stopped
singing the entire note sequence correctly after the molt
and started to repeat single melody parts (modules) again
and again without completing the entire melody as
Fig. 1 Melody 1, ‘‘Der Jager aus Kurpfalz,’’ sung by Bullfinch 1
shown as spectrogram and in musical notation, using Helmholtz
notation for the names of notes. The notes are numbered according to
their position within the melody; numbers with primes identify notes
sung by the bullfinch, and numbers without primes those whistled by
JN. The beginning of a new module is indicated by an arrow. Note
that each module has a characteristic contour, e.g. rising and falling
pitch: module 1, 2, 3, 5 and 7. Note also that modules 2 and 3 are
followed by extended pauses that act as separators. In addition,
compare the Melody 1 depicted in spectrograms and notes by
listening to the original sound recording of Bullfinch 1 on S1. Notes 4
and 5 are at ‘‘critical positions’’ defined in the text
Anim Cogn
123
documented for Bullfinch 4 by ESM S5e acoustically, by
S5f schematically and by S5g with spectrograms. There-
fore, in an attempt to maintain the birds’ accuracy as
adults, JN continued to whistle the melody to the birds each
day. In Figs. 1, 2 and 4, the notes whistled by human tutors
are numbered without primes and notes sung by bullfinches
by numbers with primes.
Instead of repeating the entire melody as did the tutors
who had hand raised the birds, JN evaluated and tried to
correct each bird’s individual deficiencies. For example,
when a bird did not continue the note sequence, JN spe-
cifically whistled the continuing modules again and again.
We document a training session with Bullfinch 3, recorded
on May 13, 1971, in a film with a sound track (Haanstra
Fig. 2 Alternate singing of the Melody 2 ‘‘Im tiefen Bohmerwald’’ between JN and Bullfinch 4. Notes with primes were sung by Bullfinch 4,
those without primes by JN. Notes of the birds are framed by light gray whereas those of JN are dark
Anim Cogn
123
1972) (ESM S2). ESM S3 is a schematic documentation of
7 repetitions of Melody 3 by Bullfinch 5, based on a
recording of a training session showing the mutual module
contributions for Bullfinch 5 and JN. Note that as Bullfinch
5 often did not continue after note 31 (from module 4–5),
JN repeated the consecutive note group from 32 to 40
(module 5) again and again.
Characterizations of the learned melodies
Studying a cross-cultural data base on folk songs, Tierney
et al. (2011) concluded that segmenting folk songs into
sub-melodies called modules, phrases or chunks (Zatorre
et al. 2007) are a widespread human song feature. One can
generally discern these sub-melodies by the following 3
characteristics: (1) separation from the other sub-units by
lengthened pauses; (2) differences in the melodic shape; (3)
the final note of the modules is usually particularly long.
Segmentation into sub-melodies is a particular promi-
nent acoustic feature of melodies. Therefore, in the Elec-
tronic Supplementary Material, we document the songs
analyzed both acoustically, by the original recordings and
also visually, by spectrograms. Thus, one can simulta-
neously gauge acoustically the temporal organizations of
the note sequence by pauses, the differences in the melodic
shape (contour) and the sequential errors of melody singing
and compare it to its pictorial displays on the spectrograms.
By listening to Melody 1 (ESM S1) and by following the
note sequence simultaneously on the notation and on the
spectrograms (Fig. 1), one can perceive unambiguously the
3 main criteria for modules:
1. Lengthened pauses between modules: On the spectro-
grams, the sequence of 26 notes is segmented into 4
modules by pauses varying between 0.1 and 1 s.
2. Differences in the melodic shape (=contour): Module 1
has rising pitch, modules 2 and 4 have descending
pitch, and module 3 has both rising and falling pitch.
3. Lengthened duration of the module’s terminal notes:
note 6, 20 and 26.
The three learned human melodies differ substantially in
melodic characteristics, in numbers of modules (note
groups separated by longer pauses) and in numbers of
different pitches and durations of notes (see Table 1).
Melody 1 (‘‘Der Jager aus Kurpfalz’’), learned by
Bullfinches 1, 2 and 3, is documented in parallel by sound
spectrograms, in musical notes (Fig. 1, for Bullfinch 1
singing alone) and as a sound file (S1). Alternate singing
between Bullfinch 1 and JN and versatility of entries at
critical position is shown schematically on Fig. 3.
Melody 2 (‘‘Im tiefen Bohmerwald’’ and ‘‘Abend wird
es wieder’’), learned by Bullfinch 4, is documented in
musical notes (S5a), acoustically, as audio files (S5b):
whistled by tutor Moller, S5c: sung correctly by Bullfinch 4
and by the corresponding spectrograms (S5d). Errors and
module repetitions are documented on S5e-h. Alternate
singing with JN is documented by Fig. 2 and S6a.
Melody 3 (‘‘In einem kuhlen Grunde’’), learned by
Bullfinch 5, is documented while alternate singing with JN
in spectrograms on Fig. 3 and as notation on S8.
Recording and analyses
JN recorded the singing using a NAGRA III b tape recor-
der. We analyzed the songs with Avisoft-SASLab Pro
software (Specht 2000).
Data Analysis
1. How accurately do bullfinches sing the sequence of
notes solo?
We start by investigating whether the bullfinches retrieve
the learned note sequences from memory in the same way
as humans, in sub-sequences separated by noticeable pau-
ses (Zatorre et al. 2007), or as a single linear chain.
In order to focus on flexibility, we selected, based on our
acoustic impression, song parts where deviating sequences,
particularly repeats of preceding modules, occurred. For
birds 4 and 5, we analyzed five different sets of melody
performances, each lasting from 20 to 50 s. Note that for
solo singing with correct temporal and sequential organi-
zation, we have previously documented the concordance
between the note sequences of the human teacher and the
songs of Bullfinch 4 for the note sequence from note 1 to
note 15 (Guttinger et al. 2002).
Bird 4 repeated particular modules multiple times and
made sequential errors in singing, as documented by S5d-h.
Table 1 Characterization of the three learned melodies
Melody
number
Number of
notes
Number of
chunks
Duration
in s
1 26 4 17
2 45 7 18
3 50 6 15
Segmenting the note sequence into sub-melodies (modules) by pauses
is a particular prominent acoustic feature of melodies. Listen to
Melody 1 (ESM S1) and simultaneously follow up its notation in
notes and spectrograms (Fig. 1). Additionally, in the electronic sup-
plementing material Melody 2 is documented both acoustically
whistled by the human tutor (S5b), sung by Bullfinch 4 (S5c) and
visually, by spectrograms (S5d) of the song documented acoustically
(S5c) and in the musical notation (5a). Melody 3 is depicted in
musical notation (S8)
Musical notation gives additional information on differences in the
melodic shapes (contour) of modules, on the characteristics of notes
(note duration in relation to other notes, its pitches and intervals
between notes)
Anim Cogn
123
Bird 5 only repeated module 4, but did not make any
detectable errors in the note sequence.
Using the solo singing of Melody 2 by Bird 4, we
investigated the following questions:
1.1 Did the bird phrase the sequence into melodic sub-
units (modules) by lengthened pauses where the
human teacher did (Table 2)?
1.2 Temporal organization of modules by different note
durations (Table 3).
1.3 Characterization of errors (schematic documentation
of errors S5f).
1.4 Are errors disproportionally located at the beginning
and ending of module (Table 2), a criterion for
chunks?
1.5 Did Birds 4 and 5, by repeating modules, enhance the
accuracy of those modules adjusting the pitches and
durations of notes to those of the human model
(Table 3)?
2. Alternate singing
We focus on the correct entries and continuations, dis-
tinguishing between correct and incorrect choices, deter-
mining the rhythmical, sequential order and versatility of
the two partners’ contributions. To produce the specific
note subsequence at the appropriate time in response to the
antecedent melody part, the bullfinch has both to perceive
the antecedent melody part of JN and at the same time to
anticipate its own entry and select the correct continuation.
In the analysis of the recordings of alternate singing, we
asked the following questions (Table 4):
2.1 Did the birds start singing only when JN stopped
(parallel singing), or also while he was still whistling?
2.2 Did the birds enter only at the beginnings of modules,
or also in the middle of modules (Table 4)?
2.3 Did they continue singing at random, beginning at
any note of the melody (H0), or from any point in the
Fig. 3 Versatility of entries at
critical positions for Melody 1
by Bullfinch 1 and JN.
Schematic documentations of
modules whistled by JN are
framed in light gray and the
entry and continuation of
Bullfinch 1 in white. Note that in
repetition 5, the correct entry
and continuation are at the
critical position 5
Table 2 Did bullfinch 4 segment Melody 2 by pauses into modules only where the human teacher did? Points of transitional errors, an
operational criterion for chunking?
Pauses Duration of pauses Errors
Number of
the pause
Between
notes
Tutor
N = 5
Bullfinch 4
N = 5
Singing with repeats
Duration of pause Singing accurately Singing with repeats (errors) Between
modules
Within
modulesM SD M SD M SD
1 6/7 0.35 0.10 0.17 0.04 0.27 0.15 3 0
2 12/13 0.95 0.15 0.53 0.08 1.18 0.82 2 0
3 24/25 1.15 0.07 0.24 0.06 0.53 0.33 1 1
4 30/31 0.35 0.21 0.12 0.02 0.32 0.24 0 0
5 35/36 0.96 0.13 0.67 0.27 1.29 0.44 1 0
6 41/42 0.56 0.11 0.28 0.05 0.46 0.23 9 1
The phrasing into modules by pauses in the songs whistled by the human teacher Karl Moller and by Bullfinch 4 shows that the bird left pauses
only where the human tutor did. With respect whether these pauses may define boundaries of chunks: Sixteen of 18 errors occurred at the
boundaries between modules. Thus, the elevated occurrences of errors closely match the criterion for chunking
Anim Cogn
123
melody, where an identical note (critical position)
that JN had just whistled occurred, or from the correct
point (H1) by continuing the sequence correctly (e.g.
at note 25 after JN just had whistled note 24, as
documented in Fig. 2?
2.4 As regards the bullfinches’ correct choice of consec-
utive note groups we assessed how variably and
flexibly the bullfinches and JN chose the positions of
their entries (Table 4).
2.5 To determine whether the bullfinch is merely
responding to JN’s last note or to longer groups of
notes (syntactical information processing), we focus
on the following two situations, which we define as
‘‘critical positions:’’
• Successive notes, identical in pitch and duration
• Identical notes that recur at different positions within
the melody.
If the bird responded only according to the last note and
did not take into account earlier notes of the melody it had
just heard, the continuation should often be not correct.
Abe and Watanabe (2011) showed that a songbird, the
Bengalese finch (L. striata), can use syntactical informa-
tion processing to discriminate songs.
3. Statistics
To study the relation between pauses and sequential
errors statistically, a criterion for chunking was tested by
binomial test. Do sequential errors occur at random from
any position within the note sequence (H0) or preferentially
at the borders between modules selectively at the first note
of modules (H1)?
To investigate whether, when alternate singing, the
bullfinches entered at random at any note within a melody
having N notes (H0), or selectively at the next note (H1), we
again used a binomial test.
Results
General remarks on the natural songs of bullfinches
The songs of free-living bullfinches are soft and contain
syllables that are similar to the whistled notes human
melodies, but the birds also sing ‘‘squeaking’’ syllables
(Guttinger et al. 2001, 2002). The bullfinches’ natural
Table 3 How did Bullfinch 4 copy the rhythmical organization by
different note values within modules from its human teacher Karl
Moller?
Number
of note
Tutor
N = 5
Bullfinch 4
N = 4
M SD M SD
1 0.32 0.03 0.32 0.01
2 0.39 0.07 0.36 0.01
3 0.18 0.01 0.15 0.01
4 0.27 0.00 0.30 0.01
5 0.33 0.07 0.23 0.00
6 0.71 0.07 0.66 0.02
7 0.25 0.01 0.26 0.00
8 0.21 0.03 0.25 0.00
9 0.26 0.02 0.20 0.01
10 0.42 0.02 0.40 0.03
The first 10 notes of Melody 2 include 5 different note values, from
semiquaver to dotted crochet in musical notation (S5). Bullfinch 4
copied these temporal differences accurately. It sang notes with an
identical pitch, but with different durations (e.g. Note 1 and 6, E flat),
corresponding to its human model firstly at point 1 as eight with
duration of 0.32 s and at point 6 as a dotted crochet with duration of
0.71 s
Table 4 Comparison of the flexibility in the entry points between the bullfinches and JN while they were engaged in alternate singing
Entries of JN Entries of the bullfinch
With the
bullfinch
number
On the
beginning
of a module
Within a
module
Correct on the
beginning of
a module
Incorrect on the
beginning of
a module
Correct within
a module
(parallel singing)
Incorrect within
a module
(parallel singing)
1 13 3 8 0 10 1
2 4 0 7 0 1 0
3 10 1 5 1 1 9
4 15 0 8 1 4 0
5 24 1 15 3 4 3
Sum 66 5 43 5 19 13
The bullfinches entered the melody after JN stopped whistling, but also while he was still whistling (parallel singing). Bullfinches 1, 2, 4 and 5
made a total of 62 correct and 21 incorrect entries. Bullfinch 3, however often confused the similar modules 1 and 3 of Melody 1 which both end
with the note A as a crotchet (see Fig. 1). It entered incorrectly 10 times, compared to only 7 correct entries
In addition, flexibility in the choice of entries of JN and the 5 bullfinches is documented. Note particularly that JN mostly whistled entire modules
as song units, whereas the birds started flexibly and correctly not only after modules, but also when he was still whistling notes within modules
Anim Cogn
123
repertoire size varies between 42 and 52 different syllable
types and is thus similar to the note number in the human
melodies they learned.
How accurately do bullfinches sing the learned melody
alone?
1. Did the bird phrase the note sequence into modules
like the human model?
The segmenting of the 45 notes of Melody 2 into modules
(sub-melodies) by pauses is documented acoustically in
ESM S5b, whistled by the human tutor Karl Moller, and in
ESM S5c, sung by Bullfinch 4. The positions and different
durations of pauses can be examined visually on spectro-
grams (ESM S5d). The temporal segmenting by pauses
whistled by the tutor and sung by Bullfinch 4 was assessed
by measuring the pauses’ durations (Table 2). Both by
listening and by analyzing spectrograms, we can recognize
distinctly three longer pauses (around 1 s) and some shorter
ones. By measuring these pauses on the spectrograms, we
analyze whether Bullfinch 4 leaves long pauses only or
disproportionally often where its tutor did. We examined
two different recordings, one containing correct melody
repetitions and one with songs repeating a single-module
again and including errors (S5e). The positions of the bird’s
pauses and also the differences in the durations of its
pauses correspond closely to those of the human model:
three longer pauses—often longer than 1 s—between notes
12 and 13 (boundary between module 2 and 3), between
note 24 and 25 (the boundary between module 4 and 5) and
between notes 35 and 36 (boundary between module 6 and
7), and 3 shorter ones (in the range of 0.3–0.5 s). Note that,
when singing with errors, the bird pauses at identical
points, but by segmenting into modules by often extended
and much more variable pauses (S5h).
2. Temporal organization of modules by different dura-
tions of notes.
The temporal patterning of the sequence by notes is
documented for Bullfinch 4 by a spectrogram of the entire
melody (S5d) and in more detail for module 6, notes
36–41, in S5g. Compared to the distinct pauses between
modules, the pauses within modules are so short that we
can scarcely perceive them acoustically or measure them in
the spectrograms (a supplementary criterion for a module).
However, we can discern 45 consecutive notes and rec-
ognize their durations and pitches. The musical notation of
this performance (S5a) uses 6 categories on note values for
the ranging from semiquaver (sixteenth note) to minim
(half note) and 9 different pitches. Learning to sing a folk
song is an extraordinary challenging sequencing task. In
order to sing each of the 45 notes which are substantially
determined in duration and pitch, the bird has to time
precisely the execution of a specific motor pattern at a
particular location in the sequence.
We assessed how precisely the Bullfinch 4 copied dif-
ferent note values by comparing the measured values of the
first 10 notes between the teacher and this bird (Table 3):
These values correspond as well between the human and
the bullfinch as they do to the indicated value in musical
notation (e.g.: note 1, notated as a quaver, 0. 32 s and note
6, notated as a dotted crotchet, 0. 71/0.66 s).
Characterizations of errors
We recognized the following irregularities in the repeti-
tions of Melody 2 by Bullfinch 4, documented as Audio
Files in the Electronic Supplementary Material S5e, as
spectrograms in S5g and S5h and schematically depicted in
S5f:
• Repetitions of antecedent note groups, by starting again
after an extended pause at the first note of antecedent
modules, as depicted in sonograms (S5h).
• Skipping ahead, omitting notes, but continuing the
melody correctly at later positions after a lengthened
pause in relation to the duration of the omitted melody
part (S5g and S5h). Correct continuation after omission
is of particular interest, since it appears to indicate
hierarchical representation (see ‘‘Discussion’’).
• Repetition of the antecedent note (S5h).
We counted as follows: 11 repetitions of antecedent note
groups, 7 occurrences of skipping ahead omitting notes and
two repetitions of a note. Furthermore, we could not detect
any other categories of errors.
Skipping ahead, omitting notes but continuing correctly
after lengthened pauses, is documented by spectrograms in
ESM S5g, and S5h. It occurred at 3 different positions: At
the boundary between modules 1 and 2, having skipped
from note 5–7 (S5g), from the middle of module 3 (note
20) to the first note of module 4 (note 25) and at the
boundary between modules 41/42, from note 41 to note 43.
Are errors disproportionally higher located at the
beginnings and endings of modules?
To assess whether the melody modules might be com-
parable to chunks, we studied the positions of errors in note
choices in relation to pauses between modules (Table 2).
Sixteen of 18 observed errors documented acoustically
(S5e), graphically (S5f), and as spectrograms (S5g)
occurred at endings of modules (H1). Melody 2 has 45
notes and includes 7 modules, so the probability that an
error occurs at the end of a module by chance is 7/45, that
is, 0.156; a binomial test against this null hypothesis shows
that the observed tendency for errors to occur at the end of
modules is highly significant (P = 1.3 9 10-11). All 11
Anim Cogn
123
repeats restarted not from any random positions but at the
first note of a module (and there were 6 restarts at note 36,
from the beginning of module 5 to notes 41 or 45). A
similar binomial test shows that this tendency for restarts to
happen at the beginning of a module is again highly sig-
nificant (P = 1.3 9 10-9). The extremely elevated occur-
rence of errors at the border between modules closely
matches the operational criterion for chunking (Terrace
2001).
Finally, we studied whether the Bullfinches 4 and 5 were
trying to enhance the accuracy of the copy, by measuring
the variation of the notes structure (frequency and dura-
tions). The variations of notes values are documented for
the human teacher and Bird 4 on Table 3, (frequencies are
not shown,) for the notes of particular modules repeated
again and again. Because both bullfinches repeated these
notes with high constancy its note values and in its pitches
(not shown), we suggest that these repetitions (used instead
of progressions) were caused by an actual memory
blockage.
In summary, analysis of the bird’s solo singing, specif-
ically the phrasing in sub-melodies marked by pauses,
indicates that the bullfinch did not retrieve the learned
melody as one coherent unit—as a linear chain of notes—
but as modules, containing of much smaller sub-sequences
containing 4–12 notes.
Alternate singing
We focus on the entries and on the consequent continua-
tions, distinguishing between correct and false choices and
continuations, analyzing sequential, rhythmic orders and
assessing the versatility of the two partners’ contributions.
For Bullfinch 4 and JN, correct and incorrect entries for
a single sequence (the notes of Melody 2) are depicted in
spectrograms (Fig. 2) as audio files (S6) and schematically
the variable points of entry (S7). In the spectrograms, the
notes sung by bullfinches can be distinguished from those
whistled by JN by the pronounced overtones, which are
nearly absent in the human’s whistled notes (S4).
In the spectrogram of Fig. 2, JN started and whistled the
first module of the melody (notes 1–6). Bullfinch 4 entered
at position 4 sang notes 4–6 (parallel singing) with its
human partner until the end of module 1 where JN stopped
while the bullfinch continued alone until position 12. JN
took over correctly at note 13 and continued up to position
24. The bird entered again, at position 24 slightly (30 ms)
too early, but with the correct note (note 25)—not con-
sidered as an error—and continued singing until note 35.
From this position, JN continued up to note 41. Instead of
finishing the melody, which contained a total of 45 notes,
Bullfinch 4 then, after a pause of 1 s, started again from the
beginning, which we rated as incorrect. The bird continued
singing up to note 12, when JN took over. S6a documents
additionally the temporal variation of the bird’s entries at
position 24/25 in 3 repetitions of the melody and as audio
files (S6b-S6c).
Data on entries of the 5 bullfinches and JN are given in
Tables 4 and 5. As regards the birds’ choices of the point of
entry, all bullfinches sometimes entered the melody while
JN was still whistling (parallel singing), not only when he
had stopped. They tended to start by correctly singing the
first note of the next module (43 occurrences) or the next
note within a module (19 occurrences). Table 4 shows that
Bullfinches 1, 2, 4 and 5 made a total of 62 correct and 21
incorrect entries. Bullfinch 3, however, often confused the
similar modules 1 and 3 of Melody 1, which end with the
identical note (see Fig. 1) and entered incorrectly 10 times,
compared to 7 correct entries. We assessed the accuracies
of the entries for each bird separately. In a melody of N
notes, the probability of entering at the correct point by
chance is 1/N. The significance of the number of correct
entries made can be assessed by a binomial test. The data
and the corresponding significance levels are shown in
Table 5; it can be seen that even for the least accurate bird
(Bird 3), the probability of the observed number of correct
entries occurring by chance is vanishingly small.
We continue by studying the flexibility of the entry
positions for JN and the birds, documented here in sche-
matic tabular sequences (Bullfinch 1, Melody 1: Fig. 3,
Bullfinch: 4, Melody 2: S7, Bullfinch 5 S3) based on
spectrograms of recorded alternate singing repetitions. The
birds were not only correct with respect to their entries, but
were also flexible in their behavior, as can be seen on S7
and on in Table 4. The beginnings and endings of the note
contributions vary much more in the songs of the bull-
finches than in those whistled by JN. He whistled the entire
modules, whereas the bullfinches started flexibly and cor-
rectly not only at the end of a module (38 times) when JN
stopped, but also when he was still whistling (parallel
singing: 18 times). In this context, the high flexibility in the
choice of entries of Bird 1 (documented in Fig. 3) is par-
ticularly notable. While JN constantly whistled module 1,
from note 1 to 6, the bullfinch entered at 3 different notes
Table 5 Binomial tests for significance for correct entries
Bird N n ce ic Bi
1 26 16 18 01 6.2 9 10-25
2 26 04 08 00 4.8 9 10-12
3 26 08 06 10 1.1 9 10-10
4 45 07 12 01 1.8 9 10-19
5 50 09 19 06 6.0 9 10-21
N = number of notes of the melody, n = number of melody repeats
studied, N n(=the number of positions to enter at random),
ce = number of correct and ic = number of incorrect entries
Anim Cogn
123
within the module by parallel singing and at note 7 by
alternate singing. Thus, given that flexibility is a criterion
for distinguishing decisions and choices based on cognition
from those based on conditioning (Griffin and Speck 2004),
so we exclude inflexibly programmed conditioning as the
main explanation for the birds’ correct choices of entries.
Finally, we determine whether the bullfinch is merely
responding to JN’s last note or to a group of notes by
discriminating syntactic rules of the note sequence (Abe
and Watanabe 2011). We focused on the following two
aspects, which we call critical positions:
1. Successive notes, identical in pitch and duration,
within a melody module: Consider, in the musical
notation of Melody 1 (Fig. 1), positions 3–6 sung as
a’’’. If Bullfinch 1 responded only according to the last
note a’’’ without considering what are preceded in the
note sequence, its singing would deviate substantially
from the melody.
If the bird responds only according to the last note, its
continuation should often be incorrect. However, as
Fig. 3 shows, Bullfinch 1 was able to accurately take
up the melody at four different positions with no
observed errors. This degree of accuracy requires that
the bird attend to pitch, duration and rhythmical timing
of several preceding notes.
2. Identical notes that recur at different positions within the
melody: If the bird considers only the last note when
deciding upon its entries, deviating note sequences
would accumulate at these positions. The most instruc-
tive examples showing that Bullfinch 5 could not only
perceive melody parts, but was also capable of short-
range planning, occurred when JN did not continue the
melody (spectrograms on Fig. 4 and tabular documen-
tation of the contributions of JN and the bird on S3), but
went to an antecedent part, (from position 40 to 32) and
repeated the sequence 11 times again until note 40.
Bullfinches 5 waited until position 40 and entered cor-
rectly with note 40, or later at note 41.
These data show that the decision on entry is at least
based on previously heard note groups, not just on the
last note.
Discussion
Aspects of memorization and recall of the melody (solo
singing)
For bullfinches, learning the long note sequences of human
folk melodies is a most demanding task, achieved by only a
subset of tutored individuals. Holden (1895), who imported
and sold melody-singing bullfinches in America, described
their performances in detail. He listed about 50 different
German, British and American popular tunes. The best
singers sang 2–3 melodies, containing at the maximum
about 50 notes. But only about one-third sang the entire
Fig. 4 Parallel singing of Bullfinch 5 (notes with primes) of Melody
3 with JN (notes without primes). JN repeats the module 5 (notes
32–40) twice. Bullfinch 5 entered correctly twice at position 40 and
finally it started again accurately at the end of module 6 at note 49 and
sang the continuing terminal notes in parallel with JN
Anim Cogn
123
tutored sequence accurately. Another third copied only
about half of the tutored sequence, and the remaining birds
did not sing more than one or two fragments (modules).
JN’ s birds, as a group, performed significantly better,
probably because of the continuing long-lasting individu-
alized interactive tutoring that JN provided. These suc-
cessful birds provide considerable data on studying melody
learning with respect to sequencing.
We distinguish the following 3 hierarchical levels of
organization (Fig. 1) in the songs: Single notes, at the
lowest level; coherent groups of notes as a module at the
intermediate level; and the entire melody, the correct
sequence of modules, at the top level. To sing a melody,
accurately the bird must learn at least the following three
hierarchically organized motor control functions:
1. To sing each note at the pitch and duration of the
corresponding note within the melody.
2. To sing consecutive notes of a module in the specific
temporal pattern (rhythm) for this melody.
3. To assemble single modules into the entire sequence of
the melody: In a melody consisting of 4–7 modules,
each module must be sung in its correct position in the
sequence.
The bullfinches retrieved the melody whistled by the
human model by reorganizing it into much smaller mod-
ules separated from one another by pauses, a criterion for
chunking (Terrace 2001). After chunking individual notes
into modules, the birds merged the modules into the
higher-order level of a hierarchically organized melody.
Hierarchical storage and recall from several levels is much
more demanding than performing a linear chain of
responses. However, chunking, unlike linear storage,
ensures that small sequential errors do not prevent the bird
from continuing correctly: Omitting a note within a module
on the intermediate level should not block retrieval of notes
of the next consecutive module. We counted 7 instances in
which a bird omitted a note or note group (S5h) but then
performed the next module accurately—hinting at hierar-
chical storage. An additional indication that temporal
organization may be represented at a higher level is given
by the durations of the pauses after the bird omitted parts of
the melody. Pauses before continuing correspond to the
length of the omitted part: The pauses were short (in the
range of 2 s) when just one note was omitted (S5h) and
much longer (5 s) when a note sequence (20–25) was
omitted.
We assume that, as the initial step, the bullfinches
started to memorize and to practice just a single-module
containing 4–12 notes and as their performance progressed,
they learned to sing more of the melody, but still as isolated
modules. As a final step, in order to sing the entire, much
longer note sequence correctly, they had to assemble the
modules together in the correct temporal and melodic
order. As untutored bullfinches do not emit songs con-
taining long coherent sequences, and because even the
tutored birds reported by Holden (1895) often failed to
learn an entire melody, we supposed that the final step, the
coupling of the modules to form a sequence of 25–50 notes,
is the upper limit for sequential learning and requires more
than what is required by song learning in nature. Such
extraordinary success seems to be achieved by the
continuing intensive individual, daily tuition by JN
throughout the birds’ lives.
Other species may respond similarly. Note that for
Nightingales (L. megarhynchos), learning of the pattern
structure (at the lower hierarchical organizational level)
and of sequencing different pattern structures (at the higher
level) are two different aspects of song acquisition. ‘‘Low’’
exposure frequencies (10–30 compared to 100 times) lim-
ited learning of sequential orders on the higher level, but
did not influence the number and the accuracy of the copies
of the subordinated patterns (Hultsch and Todt 1989,
1992). Rose et al. (2004) used data from the White-
crowned Sparrow (Zonotrichia leucophrys) to test a con-
ceptual model of a mechanism in which motor programs
link song parts together. Early, in song development, while
still perfecting single-module structures, the birds began to
combine modules; first, primarily into pairs, then into
longer sequences. Their model posits that only module
pairs that match the tutor model are reinforced maximally
over time and refined into final form.
Alternate singing gives insights into how the bird’s brain
processes a melody
We now consider the cognitive processes that allow the
bullfinch to continue to sing the correct melody part when
its human partner stops. To continue singing appropriately,
the birds had to switch from attentive listening to singing,
as an immediate response to the antecedent notes whistled
by JN. As discussed above, the bird’s behavior—appro-
priate entries at critical positions (identical notes which
occur at several positions within the note sequence, e.g.,
Fig. 3—reveals that their choices are not based entirely on
the last note but at least on the preceding note groups, or
even on much longer parts of the melody.
What mechanism is responsible for this behavior? We
consider ‘‘inner singing’’ (imagined) to be a prerequisite
for correct entries. As soon as the human starts, the birds
can match the note sequences it hears to its memorized
representation in the brain. At the point of changeover,
from perception (sensory side) to vocalization (motor side),
cognitive processing is required. Listening attentively to
the note group just whistled by the human may simulta-
neously activate the appropriate motor program to sing the
Anim Cogn
123
consecutive notes together with the human (parallel sing-
ing), but may, as long its human partner whistles, suppress
the transformation into motor commands to sing. The
bird’s brain may compare the currently perceived notes
from its partner with its memorized note sequence repre-
sentation. This comparison can establish ‘‘the inner world’’
that one associates with awareness: As soon as the human
ceases to whistle, the already-activated motor program
(‘‘imagined singing’’) can be translated into motor com-
mands to sing the consecutive notes aloud. In this way,
accurate entry and continuation may be assured. We
interpret the occasional occurrence of parallel singing as
the inability (perhaps because of incomplete learning) to
suppress its own execution during the partner’s singing.
For the bird to cue in accurately, its ability to listen to
the partner’s contributions and its ability to control its own
singing must become tightly coupled by learning. Prather
et al. (2008) identified auditory–motor mirror neurons in
the songbird with properties similar to the visual-motor
mirror neurons first observed in the monkey frontal cortex
(Rizzolatti and Craighero 2004). The songbird auditory–
motor mirror neurons fired both when a song was per-
formed by the bird and when the same song was heard. It
has been suggested that such properties in mirror neurons
suit them for the perception of visual and auditory com-
munication, including both gestures and speech (Koehler
et al. 2002). Activation of auditory mirror neurons of a
bullfinch hearing its human partner whistle may allow the
bird to use the neurons that have already been activated by
listening to continue the note sequence with the corre-
sponding motor commands.
Alternate singing by bullfinches with conspecifics, with
other bird species or with humans, has previously only
been reported as a single anecdote with a canary (Henschel
1903). Thus, this newly observed behavior may have
resulted from the supplementary daily interactive training
by JN provided to individual birds. He specifically tried to
perfect the bird’s performance by repeating missing
continuing modules again and again immediately after a
bird did not continue the melody to the end (see Supple-
mentary Material, S2: film and sound track (Haanstra 1972)
and S3: tabular documentation of the reciprocal song
modules whistled by JN and sung by Bullfinch 5). It was a
kind of ‘‘action-based’’ teaching (Nelson and Marler 1994).
Direct singing interactions between tutor and bird are
considered important for song learning (Burt et al. 2007).
The additional interactive melody rehearsal might have
specifically tuned the auditory–motor neurons in the song
circuit to switch rapidly between auditory following of the
melody and active singing.
Melody learning and alternative singing demonstrate
that as follows: (1) Bullfinches can represent a sequence of
vocal patterns as a set of modules, or ‘‘chunks’’ (Miller
1956) and (2). They have an internal representation of it.
They can cope with the complex and demanding cognitive
challenges of perceiving a human melody in its rhythmic
and melodic complexities and learning to sing it accurately.
Acknowledgments We thank Christina Meier for very helpful and
encouraging comments to improve the presentation of the results. Dr.
Jean–Pierre Stockis from the department of mathematics of TU Ka-
iserslautern gave advices and did the statistical analyses. Thanks are
also due to Tim R. Birkhead, Jessica Boffo, Manfred Gahr, Irene M.
Pepperberg and Jeffry Tesselink for constructive comments on the
earlier drafts and for correcting the English, and to three anonymous
referees.
References
Abe K, Watanabe D (2011) Songbirds possess the spontaneous ability
to discriminate syntactic rules. Nat Neurosci 14(8):1067–1074
Brehm CB (1832) Handbuch fur den Liebhaber der Stuben-, Haus-
und aller werthen Vogel. Voigt, Illmenau
Brown S, Martinez MJ, Hodges DA, Fox PT, Parsons LM (2004) The
song system of the human brain. Cogn Brain Res 20(3):363–375
Brown S, Martinez MJ, Parsons LM (2006) Music and language side
by side in the brain: a PET study of the generation of melodies
and sentences. Eur J Neurosci 23:2791–2803
Burt JM, O‘Loghlen AL, Templeton CN, Campbell SE, Beecher MD
(2007) Assessing the importance of social factors in bird song
learning: at test using computer-simulated tutors. Ethology
113:917–925
Chen JL, Penhune VC, Zatorre RJ (2009) The role of auditory and
premotor cortex in sensorimotor transformation. Ann N Y Acad
Sci 1169:15–34
Fujimoto H, Hasegawa T, Watanabe D (2011) Neural coding of
syntactic structure in learned vocalizations in the songbird.
J Neurosci 31(27):10023–10033
Griffin DR, Speck GB (2004) New evidence of animal consciousness.
Anim Cogn 7:5–18
Guttinger HR, Dobmeyer S, Schwickert S, Nicolai J (2001) Individual
learning and species-uniform song program in the bullfinch
(Pyrrhula pyrrhula). In: Kotrschal K, Muller G, Winkler H (eds)
Filander, Furth, pp 313–329
Guttinger HR, Turner T, Dobmeyer S, Nicolai J (2002) Melodi-
ewahrnehmung und Wiedergabe beim Gimpel: untersuchungen
an liederpfeifenden und Kanariengesang imitierenden Gimpeln
(Pyrrhula pyrrhula). J Ornthol 143:303–318
Haanstra B (1972) Film fragment from: Bij de Beesten af, ‘‘Instinct
for survival’’
Hamersley J (1717) The bird fancyer’s delight. R. Meares, London
(reprinted by Schott, Mainz, 1954)
Henschel G (1903) Bullfinch and canary. Nature 67:609–610
Holden GH (1895) Canaries and cage birds. Holden, Boston
Hultsch H, Todt D (1989) Memorization and reproduction of songs in
nightingales (Luscinia megarhynchos): evidence for package
formation. J Comp Physiol A 165:197–203
Hultsch H, Todt D (1992) The serial order effect in the song
acquisition of birds: relevance of exposure frequency to song
models. Anim Behav 44:590–592
Janata P, Grafton ST (2003) Swinging in the brain: shared neural
substrates for behaviors related to sequencing and music. Nat
Neurosci 6:682–687
Jarvis ED, Gunturkun O, Bruce L, Csillag H et al (2005) Avian brains
and the new understanding of vertebrate brain evolution. Nat
Rev Neurosci 6:151–159
Anim Cogn
123
Koehler E, Keysers Ch, Umilta MA, Forgassi L, Gallese V, Rizzolatti
G (2002) Hearing sounds, understanding actions: action repre-
sentation in mirror neurons. Science 297:846–848
Lichau KL (1989) Zur Geschichte der liederpfeifenden Dompfaffen
im Vogelsberg Ge. Welt 113(17–18):45–47
Miller G (1956) The magical number seven, plus or minus two: some
limits on our capacity for processing information. Psychol Rev
63:81–87
Naumann JF (1900) Naturgeschichte der Vogel Mitteleuropas, Bd III:
Lerchen, Stelzen, Waldsanger und Finkenvogel. Kohler, Gera-
Untermhaus
Nelson DA, Marler P (1994) Selection-based learning in song
development. Proc Natl Acad Sci USA 91:10498–10501
Nicolai J (1959) Familientradition in der Gesangsentwicklung des
Gimpels (Pyrrhula pyrrhula). J Ornithol 100:39–46
Nicolai J (1969) Akustische Gestaltwahrnehmung, Fehlerkorrektur
und Wechselsingen bei Gimpel. In: Bezzel E (ed) Deutsche
Ornithologen-Gesellschaft, 81. Jahresversammlung (1968) zu
Innsbruck (Sitzungsbericht). J Ornithol 110:514
Pernau FA (1768) Grundliche Anweisung aller Arten von Vogeln zu
fangen, einzustellen, abzurichten, zahm zu machen, ihre Eigens-
chaften zu erkennen, ihnen fremden Gesang zu lernen. Monath,
Nurnberg
Prather JF, Peters S, Nowicki S, Mooney R (2008) Precise auditory-
vocal mirroring in neurons for learned vocal communication.
Nature 451:305–310
Rizzolatti G, Craighero L (2004) The mirror-neuron system. Annu
Rev Neurosci 27:169–192
Rose GJ, Goller F, Gritton HJ, Plamondon SL, Baugh AT, Cooper BG
(2004) Species-typical song in white-crowned sparrows tutored
with only phrase pairs. Nature 432:753–758
Sakai K, Hikosaka O, Nakamura K (2004) Emergence of rhythm
during motor learning. Trends Cogn Sci 8(12):547–553
Seibt U, Wickler W (2000) ‘Sympathetic song’: the silent and the
overt vocal repertoire, exemplified with a dueting pair of the
African slate-coloured Boubou. Laniarius funebris Ethology
106:795–809
Specht R (2000) Avisoft-SASLab Pro. Sound analysis and synthesis
laboratory. Version 422
Suge R, Okanoya K (2010) Perceptual chunking in self-produced
songs of Bengalese finches (Lonchura striata var. domestica).
Anim Cogn 13:515–523
Terrace H (2001) Chunking and serially organized behavior in
pigeons, monkey and humans. In Cook RG (ed) Avian visual
cognition. Comparative Cognition Press, Medford, MA.
www.pigeon.psy.tufts.edu/avc/terrace/
Thorpe WH (1972) Duetting and antiphonal song in birds: its extend
and significance. Brill, Leiden
Tierney AT, Russo FA, Patel AD (2011) The motor origins of human
and avian song structure. Proc Natl Acad Sci USA 108:
15510–15515
Wallace GW, Rowan JD, Fountain S (2008) Determinants of phrasing
effects in rat serial pattern learning. Anim Cogn 11:199–214
Williams H, Staples K (1972) Syllable chunking in Zebra finch
(Taeniopygia guttata) song. J Comp Psychol 106:278–286
Zatorre RJ, Chen JL, Penhune VB (2007) When the brain plays
music: auditor-motor interactions in music perception and
production. Nat Rev Neurosci 8:547–558
Anim Cogn
123