Supplementary Methods - Word file (39 KB )

advertisement
Methods
Sample of elephant calls and truck sounds. This paper is based upon a set of
2341 recordings of calls made by JP1,2 of African elephants in the wild and AS-H of
Asian elephants and one African elephant in zoos (see Supplementary Table 1). The full
JP dataset used in this paper included 1522 calls recorded from 10 adult/adolescent
female African elephants, 78 calls recorded from 2 adult male African elephants, 158
calls recorded from 2 female African calves, and 341 calls recorded from 4 male African
calves. Out of this dataset the following calls were measured for the analyses in this
paper: 502 calls recorded from 9 adult/adolescent female African elephants, 64 calls
recorded from 2 adult male African elephants, 66 calls recorded from 2 female African
calves, and 183 calls recorded from 4 male African calves. In addition, 60 truck-like calls
and 23 non truck-like calls were measured from a ten year old female African elephant
named Mlaika. The full AS-H data set included 46 chirp-like sounds recorded during
2001-2002 from Calimero, an adult male African elephant. He rarely made any sounds
other than chirps, and only one non-chirp sound was recorded from him. In addition, the
AS-H data set included 53 chirps recorded from two adult female Asian elephants with
whom Calimero lived at the Rome Zoo from the age of 4 years for 18 years before being
transferred to the Basel zoo in 2000. The AS-H data set also included 142 chirps recorded
from 7 other adult female Asian elephants from the Emmen Zoo. Measurements were
made of all calls from Calimero and the Asian elephants.
The full set of calls recorded from elephants varied from ten to 250 calls recorded
per individual. In order to balance the amount of data per individual, two separate smaller
datasets were constructed for the statistical analyses presented in the paper, using ten high
quality calls randomly selected from each individual elephant. The first data set
compared similarity between Mlaika’s truck-like calls and the recordings of the trucks
and other African elephants. This ‘Mlaika dataset’ included ten calls from each of 9
1
adult/adolescent female African elephants, 2 adult male African elephants, 2 female
African calves, 4 male African calves, Mlaika’s non truck-like calls, Mlaika’s truck-like
calls, and 10 of the 42 truck recordings. The truck recordings were made from the
stockade where Mlaika’s truck-like sounds were recorded. Mlaika was ten years old at
the time the imitations were recorded, an age that is midway between juvenile and adult.
Therefore, we included as many calves and adult/adolescent animals as possible from
both sexes, subject to the requirement of at least 10 high quality calls. The second,
‘Calimero dataset’ compared similarity between Calimero’s chirp-like calls and the calls
of other African elephants or the chirps of Asian elephants. Only 1 non-chirp-like call
was recorded from Calimero, so we were not able to statistically compare his chirp-like
to non-chirp-like calls. This ‘Calimero dataset’ consisted of 10 calls from each of 9
adult/adolescent female African elephants, 2 adult male African elephants, 2 female
African calves, 4 male African calves, the chirps of 9 Asian elephants (including the two
individuals Calimero lived with at the Rome Zoo from the age of 4 years for 18 years
before being transferred to the Basel zoo in 2000), and Calimero’s chirp-like calls
recorded at the Basel zoo in 2001-2002. The Calimero dataset involved all Asian and
African elephants for which recordings of at least 10 high quality calls were available.
Statistical Analysis of Similarity of Imitations to Models in the Mlaika and
Calimero data sets. The parameters used in the initial analyses to describe the acoustic
features of elephant calls were median bandwidth (Hz), duration (ms), and the maximum
and minimum of the fundamental frequency (Hz). These were estimated by JP and AS-H
based upon visual inspection of spectrograms. Bandwidth was eliminated from the final
analyses presented in the paper because it did not provide much discriminatory power
beyond the maximum and minimum of the fundamental frequency, and because it
complicated description of the statistical results, adding unnecessary length to the paper.
Supplementary Table 2 describes the distribution of duration, minimum fundamental
frequency (f0min) and maximum fundamental frequency (f0max) for Mlaika’s truck-like
2
calls and the actual truck sounds and for Calimero’s chirp-like calls and chirps recorded
from Asian elephants.
Figure 2B and 2C use multidimensional scaling to present the results of
discriminant analyses, using only the duration, f0min, and f0max for each call and
excluding bandwidth. The data for each variable in the ‘Calimero’ and ‘Mlaika’ datasets
were transformed (X’=1/ln(X)) to approximate a normal distribution and to decrease the
correlation between the standard deviations and the means5. A 1-way MANOVA was
calculated for each dataset using the manova1 function in Matlab 6.5. This function
calculates the linear discriminant functions for the dataset, chosen to maximize the
separation between groups (individuals), and generates Mahalanobis distances between
each pair of group means. The classifications were bootstrapped (resampled with
replacement) using the classify function in Matlab 6.5 (from R.E. Strauss website
http://www.biol.ttu.edu/Strauss/Matlab/Matlab.htm). Five hundred bootstrap iterations
were run to estimate the frequency distribution of the classification and determine the
percent correct classification per individual. These bootstrap analyses determined that
the correct classification of the discriminant analysis averaged 87% per individual for the
Mlaika dataset classification and 81.5% for the Calimero dataset classification. When the
discriminant analysis was run excluding bandwidth, the percent correct classification
decreased to 73.5% (Fig 2B) and 62.6%(Fig 2C), respectively. This demonstrates that the
multi-dimensional scaling of the discriminant analysis in Figs 2B and 2C provide a good
representation of the full data set, and that deletion of bandwidth only caused a slight
decrease in the correct classification scores.
We designed a statistical test to compare Mlaika’s truck-like calls and Calimero’s
chirp-like calls to the putative models for imitation: truck sounds or Asian elephant
chirping calls vs. average African elephant calls. Another dataset was generated
consisting of the median bandwidth, median duration, median minimum and median
maximum fundamental frequency of each individual African or Asian elephant (from the
3
above ‘Calimero’ and ‘Mlaika’ datasets) plus 10 normal calls from Mlaika, 10 truck-like
calls from Mlaika, 10 truck sounds, and 10 chirp-like calls from Calimero. Due to a small
sample size and non-normal distribution of the variables, we compared the data using
Kruskal-Wallis ANOVA and Mann Whitney U tests (Supplementary Table 3). For each
variable of interest, there were 11 post-hoc comparisons made, therefore the significance
level was set at =(0.05/11)=0.0045.
Acoustic analysis of the complete sample of Mlaika’s truck-like calls. Using the
entire sample of truck-like calls and truck sounds, Mlaika’s truck-like calls (N=60) did
not differ significantly from the sounds of trucks (N=44) in either minimum of the
fundamental frequency (Mann-Whitney U=1177.5; z=-0.67; p=0.07) or maximum of the
fundamental frequency (Mann-Whitney U=1227.5; z=0.61; p=0.54). Though she was not
able to match the truck’s duration (Mann-Whitney U=561; z=-4.99; p<0.001), the longest
truck-like sounds she produced (median=6,984; range 685-14,372ms) were longer by
almost two seconds than any other sounds measured from an African elephant (N=2236
for the full sample of calls for which JP has measured duration). Excluding Mlaika’s
truck-like calls, the duration of measured African elephant calls ranged from 87-12,612
ms (median: 3,192 ms; inter-quartile range: 1,555–4,689). The extreme length of these
truck-like calls emitted in spite of Mlaika’s relatively small size is strong evidence that
she was attempting to match the long sounds of trucks.
In order to test whether Mlaika tended to match the specific truck sound she heard
at the time of her truck-like call, we compared the similarity of acoustic features of 39
truck-like calls and simultaneous truck sounds to the similarity of 60 truck-like calls and
60 randomly selected truck sounds recorded at separate times (see Supplementary Figure
1 for a spectrogram of a paired truck sound and truck-like call, and Supplementary Audio
1 for a wav file of the same example, in which Mlaika imitates a truck within a few
seconds of hearing it). Only 44 truck sounds were recorded, so this analysis required
some resampling of truck recordings. For each of Mlaika’s 60 truck-like calls, we
randomly selected one truck sound recorded at a different time from the truck-like call for
4
the non-paired comparison set. For each parameter, we took recordings where trucks and
calls occurred at the same time and subtracted the bandwidth, f0min, f0max, and duration
measurements between the paired calls. This gave the difference in each of the
parameters for calls recorded at the same time. We then did the same thing for the calls
and truck sounds recorded at different times. We compared the two sets of differences
with a t-test. There were no significant differences in the difference between acoustic
parameters for truck-like calls and paired vs. non-paired truck sounds (t-test, p>0.6 for all
4 variables). Therefore, the truck-like call Mlaika produced upon hearing a truck was no
more similar to the coincident truck sound than the truck-like calls she produced at other
times. Mlaika appears to have matched the overall parameters of the trucks, rather than
matching a particular truck at a given moment.
Acoustic analysis of the complete sample of Calimero’s chirp-like calls.
Calimero’s chirp-like calls (see Supplementary Audio 2 for a wav file of a Calimero
chirp-like call and Supplementary Figure 2 for a spectrogram of this call) are similar in
duration to Asian chirping calls (see Supplementary Audio 3 for a wav file of a chirp
from a female Asian elephant and Supplementary Figure 3 for a spectrogram of this call),
different in all parameters from adult African elephant calls, and different in duration and
fundamental frequency from African calf calls (Supplementary Table 3). Comparing all
46 of Calimero chirps to the 195 chirps recorded from Asian elephants, the durations of
Calimero’s chirps were remarkably similar to the chirps of Asian elephants (Calimero
median=281 ms, Asian median=280 ms; Mann-Whitney-U= 4281; z=-0.483; p=0.629).
The chirping calls of Asian elephants usually include a series of similar chirping sounds.
The difference in the number of chirps in a series produced by Calimero (median=3;
N=17) and the female Asian elephants (median=3.5; N=38) was not statistically
significant (Mann-Whitney-U=227; z=-1.785; p=0.074). The similarity of repetition is
strong evidence for imitation since, other than the pulsated play trumpet3,4, the repertoire
of African elephants does not contain calls emitted in an uninterrupted series. The
minimum fundamental frequency of chirps produced by Calimero (median=130 Hz) was
5
lower than the minimum fundamental and dominant frequency of the Asian females
(median=520 Hz), but Calimero put most energy into the second harmonic of his chirp
(Figure 1d). At almost twice the weight (6800 kg) of the Asian females, Calimero may
have found it easier to attempt to match their calls in this way.
1. Poole, J. H., Payne, K. B., Langbauer, W. Jr. & Moss C. J. The social contexts of some
very low frequency calls of African elephants. Behavioral Ecology and Sociobiology 22,
385-392 (1988).
2. Poole, J. H. Signals and assessment in African elephants: evidence from playback
experiments. Animal Behaviour 58, 185-193 (1999).
3. Berg, J. K. Vocalizations and associated behaviors of the African elephant (Loxodonta
africana) in captivity. Z. Tierpsychol. 63: 63-79 (1983).
4. Poole, J. H. & Granli, P. K. The visual, tactile and acoustic signals of play in African
savannah elephants. In (ed.) Jayewardene, Jayantha. Endangered Elephants, past present
& future. Proceedings of the Symposium on Human Elephant Relationships and
Conflicts, Sri Lanka, September 2003. Biodiversity & Elephant Conservation Trust,
Colombo. Page 44-50 (2004).
5. Sokal
R. R. & Rohlf, F. J. Biometry. 2nd ed. W. H. Freeman, New York, (1981).
6
Download