Content uploaded by Csaba Molnar
Author content
All content in this area was uploaded by Csaba Molnar
Content may be subject to copyright.
Human Listeners Are Able to Classify Dog (Canis familiaris) Barks
Recorded in Different Situations
Pe´ter Pongra´cz and Csaba Molna´r
Eo¨tvo¨s Lora´nd University
A
´
da´m Miklo´si
Hungarian Academy of Sciences
Vilmos Csa´nyi
Eo¨tvo¨s Lora´nd University
The authors investigated whether human listeners could categorize played-back dog (Canis familiaris)
barks recorded in various situations and associate them with emotional ratings. Prerecorded barks of a
Hungarian herding dog breed (Mudi) provided the sample. Human listeners were asked to rate emotion-
ality of the vocalization and to categorize the situations on the basis of alternative situations provided on
a questionnaire. The authors found almost no effect of previous experience with the given dog breed or
of owning a dog. Listeners were able to categorize bark situations high above chance level. Emotionality
ratings for particular bark samples correlated with peak and fundamental frequency and interbark
intervals. The authors did not find a significant effect of tonality (harmonic-to-noise ratio) on either the
emotionality rating or situation categorization of the human listeners. Humans’ ability to recognize
meaning suggests that barks could serve as an effective means of communication between dog and
human.
Wild and domesticated canids are well known for their rich
repertoire of vocal signals (e.g., Lehner, 1978; Tembrock, 1976).
Although most of the canine acoustic signals have been found to
be connected to special social situations (Cohen & Fox, 1976),
barking, which is the most characteristic vocalization of the dog,
has been assumed to have no specific communicative role. Dog
barking has been described as notoriously variable and used almost
freely in various situations when dogs interact with each other
(Cohen & Fox, 1976). As other wild canids bark only in their
puppyhood and as adults bark only in specific contexts (e.g.,
during defending territory; Schassburger, 1993), some researchers
have thought that excessive and repetitive barking in adult dogs is
either a neotenic feature with no (or little) communicative function
or that dog barking is a byproduct of relaxed selection during the
domestication process (Cohen & Fox, 1976).
However, recently it has been hypothesized that dog barking
could play a role in interspecific communication between dogs and
humans (Feddersen-Petersen, 2000; Yin, 2002). Dogs form strong
attachments to humans (Ga´csi, Topa´l, Miklo´si, Do´ka, & Csa´nyi,
2001; Topa´l, Miklo´si, & Csa´nyi, 1998), they are attracted by
different human activities, and they are adept at understanding
human communicative signals, both visual (e.g., Miklo´si, Polga´rdi,
Topa´l, & Csa´nyi, 2000; Soproni, Miklo´si, Topa´l, & Csa´nyi, 2002)
and acoustic (Pongra´cz, Miklo´si, & Csa´nyi, 2001). Dogs can learn
to anticipate new, unusual patterns of human behavior (Kubinyi,
Miklo´si, Topa´l, & Csa´nyi, 2003), and they can learn socially from
human demonstrators (Kubinyi, Topa´l, Miklo´si, & Csa´nyi, 2003;
McKinley & Young, 2003; Pongra´cz, Miklo´si, Kubinyi, et al.,
2001; Pongra´cz, Miklo´si, Tima´r-Geng, & Csa´nyi, 2003). Further,
during domestication, humans have selected dogs according to
specific working purposes (Coppinger & Coppinger, 2001) that
could be the basis for abilities responsible for communication from
the dogs toward humans.
There are two nonexclusive possibilities that can provide the
foundation of interspecific communication. Species could rely on
common rules of communication that determine the structure of
the signal. The presence of such rules was assumed by Morton
(1977), who showed that many mammalian and bird signals emit-
ted in specific contexts can be characterized by distinctive acoustic
features. Alternatively, the individuals of the different species
could learn the meaning of the heterospecific signal. One such
natural example for the latter is the realization of the recognition
of another species’ alarm call (e.g., Seyfarth & Cheney, 1990;
Shriner, 1998), and in the laboratory other instances have also
emerged; for example, mammals and birds can learn the referen-
tiality of human pointing and verbal signals (dolphin: Herman et
al., 1999; grey parrot: Pepperberg & McLaughlin, 1996). It is
interesting to note that that there is much less evidence for the
former process.
Pe´ter Pongra´cz, Csaba Molna´r, and Vilmos Csa´nyi, Department of
Ethology, Eo¨tvo¨s Lora´nd University, Budapest, Hungary; A
´
da´m Miklo´si,
Comparative Ethology Research Group, Hungarian Academy of Sciences,
Budapest.
This study was funded by Hungarian Ministry of Education Grants
FKFP No. 127/2001 and OTKA No. T047235, F01/031. We thank the
members of the Magyar Ebtenye´szto˝k Orsza´gos Egyesu¨lete Magyar Mudi
Klub for their participation in the bark recordings with their dogs. We also
thank Sa´ndor Zsebo¨k for the voice-analyzing software and Thomas Riede
for the information on how to calculate the harmonic-to-noise ratio. Karen
McComb gave useful comments on the first version of this article. We
thank Celeste Spadavecchia for checking our English in this article.
Correspondence concerning this article should be addressed to Pe´ter
Pongra´cz, Department of Ethology, Eo¨tvo¨s Lora´nd University, Budapest,
Pa´zma´ny Pe´ter se´ta´ny 1/c, H-1117 Hungary. E-mail: uupeter@ludens
.elte.hu
Journal of Comparative Psychology Copyright 2005 by the American Psychological Association
2005, Vol. 119, No. 2, 136 –144 0735-7036/05/$12.00 DOI: 10.1037/0735-7036.119.2.136
136
In the present study, we investigated whether dog barks can play
a role in dog– human communication. This assumption relies on at
least two conditions: First, dog barks should be acoustically dif-
ferent (Yin, 2002; Yin & McCowan, 2004), and humans should
react to dog barks as communicative signals—that is, they should
be able to judge the emotional state of the sender and/or the
situation in which the bark was emitted.
In the present study, we compared the acoustic structure of dog
barks recorded in different behavioral situations and allowed hu-
mans to categorize these dog barks and describe their emotional
content. Further, we looked at the role of experience by choosing
humans with different experiences of dogs for the playback
experiment.
Method
Subjects
We formed three experimental groups that were based on the listeners’
knowledge about Mudis and on the listeners’ general experiences of dogs.
Mudi owners (n ⫽ 12; mean age ⫽ 40.17, range ⫽ 19 – 60 years; men:
women ⫽ 6:6) either had a Mudi at the time of the experiment or possessed
a Mudi earlier. They were mainly drawn from the members of the Hun-
garian Mudi Club or were the family members of the Mudi owners. Other
dog owners (n ⫽ 12; mean age ⫽ 27.92, range ⫽ 19 –52 years; men:
women ⫽ 6:6) either had a dog at the time of the experiment or possessed
a dog earlier, but they had never owned a Mudi. Nonowners (n ⫽ 12; mean
age ⫽ 41.30, range ⫽ 31– 69 years; men:women ⫽ 6:6) were those that
had never had a dog.
Source of Sound Recordings
Barking vocalizations of a Hungarian sheepdog (Canis familiaris) breed,
the Mudi, were used for this study (see Figure 1). This medium-sized breed
is listed under the 238th Standard of the Fe´de´ration Cynologique Interna-
tional. This breed is used traditionally for herding flocks of sheep and
cattle. It has also been used as a vigilant watchdog in the countryside. The
working style of this breed requires extensive use of barking.
Bark recordings from 19 Mudis (10 males and 9 females; mean age ⫽
3.80 years, range ⫽ 1.1–10.0 years) were collected for this study. All the
dogs were kept as pets (by 15 owners) in family houses or apartments.
Recording Situations
We collected bark recordings in six different behavioral contexts, most
of which could be arranged at the homes of the owners, with the exceptions
of the schutzhund situation, which was staged at dog training schools, and
the alone situation, which was staged on a street or in a park. The six
situations are as follows:
Stranger: The experimenter (male, age 22), who acted as the stranger
for all the dogs, appeared in the garden of the owner or at the front
door of his or her apartment in the absence of the owner. The
experimenter asked the owner by phone to stay in another room or at
a greater distance during the time needed for the recording. The
experimenter recorded the barking of the dog during his appearance
and intrusion into the garden or apartment for about 2–3 min.
Schutzhund: For dogs to perform in this situation, the trainer, who
acted as the “bad guy,” encouraged the dog to bark aggressively and
to bite the bandage on the trainer’s arm. During this situation, the
owner held the dog on a long leash. The experimenter recorded the
barks of the dogs during their training for 1–2 min.
Going for a walk: The owner was asked to behave as if he or she was
preparing to go for a walk with the dog. For example, the owner took
the leash of the dog in her or his hand and told the dog, “We are
leaving now.” The experimenter recorded the barks of the dogs during
such situations.
Alone: The owner tied the leash of the dog to a tree in a park and
walked away, out of sight of the dog. The experimenter recorded the
barks of the dog from a distance of 4–5 m in the absence of the owner
for 3– 4 min.
Ball: The owner held a ball (or some favorite toy of the dog) at a
height of approximately 1.5 m in front of the dog. The barks elicited
in this situation were recorded by the experimenter for 1–2 min.
Play: The owner was asked to play a usual game with the dog, such
as tug-of-war, chasing, or wrestling. The barks emitted during this
were recorded by the experimenter.
We collected as many barks from a given dog and in as many situations
as that dog produced barks. In total, 72 barks were used in this study, 12
from each situation originating from different individuals chosen
randomly.
Recording and Preparing the Sound Material
Tape recordings were made with a tape recorder and a microphone.
During recording of the barks, the experimenter stood in front of the dog
and faced it while holding the microphone within 1– 4 m of the dog.
Figure 1. Portrait of a Mudi. The Mudi is a midsized Hungarian sheep-
dog. It is an alert, intelligent, and obedient breed. Its herding style involves
a lot of barking. Photo credit: Pe´ter Pongra´cz.
137
HUMAN CLASSIFICATION OF DIFFERENT DOG BARKS
The recorded material was transferred to a computer, where we cut out
a 20-s-long sample from each bark recording for further analysis and for
the playback experiment. We used very simple rules for cutting out the 20-s
bark samples from the recordings: (a) The sample could not be the starting
or ending section of the bark; (b) the sample needed to contain at least 10
individual barks; and (c) the first 20-s-long sample from the given bark
recording was used, which fulfilled the first two criteria. These samples
consisted of a various number of individual barks, depending on the
interbark intervals. Barks were digitalized with a 16-bit quantization and a
22.05-kHz sampling rate by using a TerraTec DMX 6-fire 24/96 sound
card. To equate the calls for loudness, barks were normalized by rescaling
each waveform so that its highest amplitude peak was at ⫺6 dB. Barks
were analyzed with ACMS sound-analyzing software (contact Sa´ndor
Zsebo˜k at zsebok@ludens.elte.hu for more information). The program took
100 sequential frequency and amplitude measurements of the dominant
frequency (the frequency band in which the most energy is concentrated)
in a frequency-time spectrum for each individual bark by using a fast
Fourier transform of 1,024 points and a frequency resolution of 22 Hz. The
following parameters were collected or derived from the original samples:
Interbark interval: The average interval between the individual barks
in a 20-s-long sample.
Average peak frequency: The average of the most intense (peak)
frequency components of each individual bark in a 20-s-long sample
sequence.
Average fundamental frequency: The frequency of the fundamental
harmonic of the given bark. We confirmed visually all the individual
fundamentals on narrow-band spectrograms after the software chose
them automatically. We averaged the fundamental frequencies of the
individual barks within the sample sequences.
Harmonic-to-noise ratio (HNR): This parameter served for the de-
scription of the “roughness” of the barking. The calculation of HNR
was performed by the method described by Riede, Herzel, Hammer-
schmidt, Brunnberg, and Tembrock (2001), except that we used a
1,024-point fast Fourier transform. This provided us with twice
greater dissolution for the analysis. The HNR compares the volume of
harmonic tones of the sound with the volume of nonharmonic noise
within the sound. The higher the HNR is, the clearer the sound is. The
calculation of HNR was done as follows: We computed the power
spectrum of a 50-ms segment from the middle of a bark. We estimated
the noise level by calculating the moving average of the spectrum
curve. Then we determined the maximum difference between the
harmonic peaks and the noise level by using a Microsoft Excel macro.
Playback Experiments
Each human listener was provided with a unique set of 18 barks prepared
in advance. No two listeners heard exactly the same 18 barks, and the order
of the barks within the sets was also randomized. There were 12 listeners
in each of the groups and 12 barks coming from different dogs in each of
the six situations. Each listener was given 3 different barks from each
situation, which resulted in the 18 barks per set. At first we randomized the
order of the 12 out of 12 barks in every situation; then we assigned 3 out
of 3 of them into the individual sets by using a method in which each bark
could go into only 3 sets. After we randomized the barks within the
individual situations again and again for the three groups of listeners, we
prepared the sets of 18 barks; we could thus guarantee that every listener
got a truly unique set of barks and that a given individual bark was scored
by only three listeners within a given group.
The sound sets were copied onto a CD and could be played on a
computer. Barks were presented to the participants via a multichannel, soft
flat-panel PC speaker system. Each listener was exposed to one of the
prepared sound sets, which were randomly chosen before the trial. The
barks were played back one by one to the listeners, who were allowed to
listen to every bark twice. The experimenter handled the player software.
The listeners had to fill in the corresponding questionnaire sheet during the
experiment. Two questionnaires were used as the experimenter presented
two playback series to the subjects. These series were done one after the
other, so the listeners handled only one questionnaire at the same time.
After playing back a given bark sample twice, the experimenter stopped the
device and gave listeners approximately 30 s to fill in the corresponding
row on the questionnaire. The experimenter did not give suggestions or any
specific help to the listeners but, if needed, played back the given bark
sample once more. Listeners performed the playback tests alone or in
smaller groups (up to three persons) with the experimenter. In Experiment
1, the listeners had to rate each bark sample for five different kinds of
emotions. Following Experiment 1, the experimenter played the same
sound set once more for the listeners, but now they had to guess the
situation of every bark (Experiment 2). Listeners were not directly in-
formed that they were given the same sound set, but if they asked, the
experimenter told them that the same barks might be in both of the sets.
Questionnaire 1: Emotionality ratings. Listeners had to rate each bark
sample on a 5-item scale for different content of emotionality: (a) aggres-
siveness, (b) fearfulness, (c) despair, (d) playfulness, and (e) happiness.
Low values indicated the absence of that type of emotion, whereas higher
values suggested a predominant presence of the emotion in question. For
example, subjects could scale a given sample for the lack of aggressiveness
(rated as 1 on the Aggressiveness Scale) but indicate high levels of
playfulness (rated as 5 on the Playfulness Scale). Listeners had to rate each
bark sample for the presence of each emotion by using the scale system.
Questionnaire 2: Categorization of situations. On the questionnaire,
subjects categorized each sample into one of the six situations listed on
their sheets. The listeners did not know that each situation could occur
three times.
Data Analysis
At first we averaged the emotionality ranks given by the 3 subjects
within every group who listened to the same bark sequence. Emotionality
ratings were analyzed by a one-way analysis of variance (ANOVA) with
Student Newman–Keuls post hoc tests or by a Friedman repeated measures
test with Dunn’s post hoc tests. We compared the acoustic parameters of
the bark sequences in the different situations by using a one-way ANOVA
with Student Newman–Keuls post hoc tests. Pearson correlation tests were
used to analyze the possible relationship between the emotionality ratings
and the acoustic parameters of the barks. The accuracy with which subjects
categorized the bark situations was examined by one-sample t tests. We
analyzed the effect of different situations on the categorizing of the bark
samples by using an ANOVA with post hoc tests.
Results
Acoustic Comparison of Barks Recorded in
Different Situations
First we wanted to know whether the bark sequences, originat-
ing from the six different situations, differed in the acoustic pa-
rameters (see Figures 2 and 3). One-way ANOVAs showed that
with the exception of the tonality, the situation had a significant
effect on the interbark interval, F(5, 66) ⫽ 8.98, p ⬍ .001; the
average peak frequency, F(5, 66) ⫽ 4.69, p ⬍ .001; and the
fundamental frequency of the bark, F(5, 66) ⫽ 3.40, p ⬍ .01. The
Student Newman–Keuls post hoc tests showed that barks from the
schutzhund situation had the shortest interbark intervals and that
the barks from the alone and ball situations had the longest
138
PONGRA
´
CZ, MOLNA
´
R, MIKLO
´
SI, AND CSA
´
NYI
interbark intervals. The average peak frequency and fundamental
frequency were the lowest in the stranger situation, which had
significantly lower frequency values of both parameters than did
the barks from the walk, play, and alone situations.
Emotionality Ratings
The overall comparison of the three groups’ scores on the five
emotionality scales (Friedman repeated measures test; see Table 1)
showed significant differences in the case of the Despair and
Playfulness scales. The only difference among the groups was that
other dog owners gave significantly higher scores of despair than
did the nonowners.
We analyzed how the listeners rated the emotionality of the
barks from different situations in the different groups (see Table
2). In the Mudi owners group, Friedman ANOVAs showed that
emotionality traits had a significant effect for most situations
except for walk and ball barks. Mudi owners gave the highest
scores of aggressiveness to barks recorded in the stranger and
schutzhund situations (see Figure 4), they gave significantly higher
despair and fearfulness than happiness scores to barks recorded in
the alone situation, and they gave the highest scores of playfulness
to the play barks in comparison to barks recorded in all the other
situations, respectively. Additionally, they gave the lowest scores
of happiness to the alone barks and the lowest scores of fear and
despair to the play barks (Dunn’s post hoc test). The results of the
other dog owners group showed a similar general pattern (see
Table 2). The listeners did not discriminate the walk, ball, and play
situations on any of the emotionality scales. They gave the highest
scores of aggressiveness to the barks recorded in the stranger and
schutzhund situations (see Figure 4). Other dog owners gave high
scores of despair and fearfulness and very low scores of happiness
and playfulness to the alone barks. Listeners in the nonowners
group found two situations difficult to judge by their emotionality:
There were no significant differences in the case of either walk or
ball barks (see Table 2). However, the categorizing habits of
nonowners were very similar to the other two groups in the
remaining situations. Nonowners gave high scores of aggressive-
ness to the stranger and schutzhund situations. Alone barks got
high scores in fearfulness and despair while also being character-
ized by low scores of aggressiveness and happiness. Nonowners
gave the highest scores of playfulness and happiness to the play
barks, and this situation got the lowest scores on the Despair Scale.
Analyzing the groups separately, we found that humans associ-
ated different emotionality as a function of the bark situations (see
Table 3). There was a significant effect of situations in all groups
on all of the emotionality scales (Friedman repeated measures test
with Dunn’s post hoc tests). Listeners of all the groups gave
significantly higher aggressiveness scores to the stranger and
schutzhund situations than to any other emotions. Similarly, all
listeners gave the significantly highest scores of fear and despair in
the alone situation. In the Mudi owner and nonowner groups, the
listeners gave the highest scores of happiness and playfulness to
the play situation (see Figure 4).
Finally, we performed Pearson correlation tests to analyze the
possible relationship between the emotionality ratings and the
acoustic parameters of the barks. For this analysis, the groups were
pooled because we found no fundamental differences between the
ratings of the three groups. We used the mean score on each
emotionality scale for any given bark. Similarly, we calculated the
mean of the acoustic parameters for each bark (see Figure 3 for
typical sonograms of different barks).
Aggressiveness ratings showed negative correlations with the
interbark interval, fundamental frequency, and the average peak
frequency of the barks. Ratings for despair correlated positively
with the average peak frequency. Playfulness showed positive
correlations with the interbark interval, average peak frequency,
and fundamental frequency. Happiness correlated positively with
the fundamental and average peak frequencies of the barks. There
was no evidence of significant correlation in the case of fearfulness
with any of the acoustic parameters (see Table 4).
It is interesting to note that we did not find relevant differences
between the three groups of listeners in the scoring manners of the
Figure 2. Acoustic parameters with significant differences between the bark situations. Different letters refer
to significantly differing values of the given parameter (one-way analysis of variance with Student Neuman–
Keuls post hoc test). Error bars represent standard errors of the mean. Schutzh. ⫽ schutzhund.
139
HUMAN CLASSIFICATION OF DIFFERENT DOG BARKS
emotionality of barks. Group differences would have indicated that
experience with dogs or, more possibly, experience with a partic-
ular breed can strongly influence the listener’s opinion about the
emotional content of a particular vocalization. Listeners found
stranger and schutzhund situations to be mostly aggressive, barks
recorded in the alone situation received high scores on the Despair
Scale, and the majority of the listeners gave high scores of hap-
piness and playfulness to the situations of walk and play. We
should note that the listeners did the emotionality scoring first and
had to categorize the barks later, so it is unlikely that the catego-
rization of a bark influenced the emotionality scoring. We suppose
that the emotionality scoring is more sensitive and free from the
bias of earlier experiences in a broader extent than is guessing the
exact context of a barking. Considering this, we found high agree-
ment between the emotionality scorings of the different groups of
subjects.
The walk and ball situations caused the most difficulty in
scaling their emotionality. We assumed that the vocalizations
emitted in the “begging for a ball” situation went through consid-
erable shaping during previous interactions with the owner—that
is, such barks could have a strongly learned component, which
varies among individuals. A similar explanation could hold for the
walk situation: Besides the high excitement of the dog, individual
prewalking habits of a dog– owner pair could make a difference.
Therefore, the manner of barks in this situation may suggest
emotions varying among more scales because of the emotion of the
individual dog’s desire to go for the walk.
Additionally, we found that some emotional ratings correlated
strongly with the acoustic features of the barks, which could
provide the physical basis for the ability of humans to assign
different emotional ratings to some types of barks. We found it
interesting that the degree of fearfulness did not correlate with any
of the acoustic parameters, but of course this emotion might be
scored on the basis of compound or more complex acoustic fea-
tures. The correlation analysis showed that a bark is likely de-
scribed as being aggressive if it is characterized by low peak and
fundamental frequencies and short interbark intervals. In contrast,
high despair ratings are associated with the opposite acoustic
parameters of high frequencies and longer interbark intervals.
Playfulness and happiness correlated positively mainly with the
frequency components, suggesting that a more high-pitched bark-
ing is considered more likely to relate to happy or playful emo-
tions. We can conclude that the dimension of quickly versus
slowly pulsating barks and high-pitched versus low-pitched fre-
Table 1
Effect of Previous Experience With Dogs on the Scoring of
Different Emotionality Scales (Friedman Repeated
Measures Tests)
Scale
2
(2)
p
Aggressiveness 0.58 .74
Fearfulness 2.73 .26
Despair 10.26 ⬍.01
Playfulness 8.31 ⬍.05
Happiness 5.76 .06
Note. Human listeners were divided into three groups: Mudi owners,
other dog owners, and nonowners.
Table 2
Effect of Emotionality Traits in the Different Barking Situations
(Friedman Repeated Measures Test)
Situation
Mudi owners
Other dog
owners Nonowners
2
(4)
p
2
(4)
p
2
(4)
p
Stranger 58.88 ⬍.001 39.05 ⬍.001 39.48 ⬍.001
Schutzhund 43.79 ⬍.001 69.99 ⬍.001 46.03 ⬍.001
Walk 7.60 .11 5.36 .025 5.14 .24
Alone 24.10 ⬍.001 43.43 ⬍.001 23.30 ⬍.001
Ball 2.77 .60 3.70 .45 7.46 .11
Play 35.29 ⬍.001 6.39 .17 30.92 ⬍.001
Figure 3. Sonograms recorded in different situations.
140
PONGRA
´
CZ, MOLNA
´
R, MIKLO
´
SI, AND CSA
´
NYI
quency could be important for making happy versus sad distinc-
tions, respectively.
Categorization of Situations
First we analyzed the accuracy of recognition of situations
within the groups. We found that all groups performed signifi-
cantly over the chance level (by chance humans could be correct
on 3 out of 18 cases: 16.67%)—one-sample t tests: Mudi owners
(40.74%), t(11) ⫽ 8.22, p ⬍ .001; other dog owners (39.35%),
t(11) ⫽ 4.08, p ⬍ .001; nonowners (39.35%), t(11) ⫽ 7.00, p ⬍
.001 (see Figure 5). The comparison of groups with different
experience with dogs showed that Mudi owners proved to be only
slightly better at guessing the correct situations than listeners in the
other two groups, and these differences were not significant:
one-way ANOVA, F(2, 33) ⫽ 0.06, p ⫽ .94.
As the listeners in the different groups categorized the barks
similarly, groups were pooled and we investigated whether the
situation influenced the accuracy of categorization. We analyzed
the percentage of correct answers with a one-way ANOVA and
Student Neuman–Keuls post hoc tests. The type of situation had a
significant effect, F(5, 210) ⫽ 10.64, p ⬍ .001 (see Figure 6). The
listeners most accurately categorized the barking of dogs in the
stranger, schutzhund, and alone situations. At the same time, they
showed the poorest results in recognition of ball and walk barks.
One could ask whether the acoustic features or merely the amount
of individual barks in a given bark sample caused the differences
between the accuracy of categorization of the different bark sam-
ples. As a sample was 20 s long and the average interbark intervals
were different between some of the situations, it could happen that
human listeners could better categorize those samples that origi-
nated from situations with shorter interbark intervals and therefore
Figure 4. Radar histograms about the effect of situations on the emotionality ratings. The schutzhund
(Schutzh.) and alone situations were characterized by high scores of aggressiveness, the alone situation was
characterized by high scores of fear and despair, and the play situation was characterized by high scores of
playfulness and happiness.
Table 3
Effect of Barking Situations on How Human Listeners Rated
Emotionality Scales (Friedman Repeated Measures Test)
Scale
Mudi owners
Other dog
owners Nonowners
2
(5)
p
2
(5)
p
2
(5)
p
Aggressiveness 64.60 ⬍.001 76.88 ⬍.001 59.50 ⬍.001
Fearfulness 19.39 ⬍.01 20.72 ⬍.001 24.93 ⬍.001
Despair 19.79 ⬍.01 30.87 ⬍.001 25.17 ⬍.001
Playfulness 35.85 ⬍.001 23.83 ⬍.001 44.16 ⬍.001
Happiness 33.63 ⬍.001 26.49 ⬍.001 33.32 ⬍.001
141
HUMAN CLASSIFICATION OF DIFFERENT DOG BARKS
contained more individual barks. Our results show that this is
unlikely, however, as samples from the stranger, schutzhund, and
alone situations were categorized most accurately by the listeners
and, accordingly with Figure 2, the interbark intervals were the
shortest in the schutzhund situation, medium length in the stranger,
and long in the alone situations.
We calculated a so-called confusion matrix, too (see Table 5).
We wanted to know if the listeners did not correctly categorize the
given bark sample, which situations they used most for these
misidentified items. Four situations were quite well identified:
stranger, schutzhund, alone, and play. From these, stranger and
alone had no preferred alternative, as the wrong answers were
distributed quite equally among the other situations. On the other
hand, schutzhund was quite often misinterpreted as stranger, and
play was mistakenly categorized most often as ball.
Discussion
Our results from the two experiments showed that (a) indepen-
dent from their previous experiences with dogs, human listeners
scored the emotional content of barks in similar manner and
accuracy; and (b) the emotional ratings were in accordance with
expectations knowing how the specific situations could affect the
emotions of dogs (i.e., vocalizations of dogs attacking a stranger in
the garden or performing schutzhund training were given high
scores of aggressiveness, or the vocalizations of dogs left alone
and tied to a tree were given high scores of despair). Further, we
found evidence that (c) the emotional content as judged by humans
correlates with particular acoustic parameters of a given bark, but
(d) we did not find a major difference in the accuracy of catego-
rization between the performance of the listeners on the basis of
their previous experiences with dogs; (e) human listeners catego-
rized more accurately those situations for which they found the
emotional content less ambiguous (stranger, schutzhund, alone,
and play), but (f) listeners could categorize over the chance level
the majority of the barking situations on the basis of listening only
to the vocalizations.
Until now there were only a few investigations that compared
dog barks emitted in different situations (Bleicher, 1963; Riede &
Fitch, 1999). Recently, Yin (2002) found that barks from three
situations (to strange noise, dog is left alone in a room, and play)
differed in some acoustic parameters and hypothesized that this
could provide a basis for interspecific communication; similarly,
Feddersen-Petersen (2000) suggested that dogs might communi-
cate with humans mainly by barking. Our experiments were con-
Table 4
Pearson Correlations Among Acoustic Parameters of Barks and Emotionality Scores Given by
Listeners
Scale
Interbark interval
Average peak
frequency
Fundamental
frequency
Tonality
(HNR)
rpr p rprp
Aggressiveness ⫺.31 ⬍.01 ⫺.35 ⬍.01 ⫺.23 .05 ⫺.02 .87
Fearfulness .28 .82 ⫺.13 .27 ⫺.12 .30 .02 .89
Despair .07 .54 .26 ⬍.05 .17 .15 .05 .65
Playfulness .26 ⬍.05 .42 ⬍.001 .32 ⬍.01 .08 .50
Happiness .23 .06 .39 .001 .32 ⬍.01 .09 .44
Note. The three groups of listeners were pooled for this analysis. HNR ⫽ harmonic-to-noise ratio.
Figure 5. Overall percentages of situations categorized correctly. All three groups of listeners performed over
the chance level (16.67%; one-sample t test). There was no significant difference between the groups (one-way
analysis of variance). Errors bars represent standard errors of the mean. *p ⬍ .001.
142
PONGRA
´
CZ, MOLNA
´
R, MIKLO
´
SI, AND CSA
´
NYI
ducted on a relatively large human sample, and we used barks from
more dogs and from more situations than had been collected
before. Our results showed that human listeners are able to cate-
gorize these situations on the basis of auditory cues only. As our
subjects were not given visual cues of the situations, we can
conclude that the acoustic structure of barks provides features that
make them recognizable for human listeners. This suggests that
some dog barks may present means for interspecific communica-
tion. This finding differs from the results by Nicastro and Owren
(2003), who found that human listeners were less skilled in clas-
sifying cat meows and that their abilities were in strong correlation
with their experience and affinity to cats.
It is interesting to note that in our case, there was no significant
difference between humans who had many opportunities for learn-
ing about dogs and those who have never had a dog at home.
Therefore, humans might share some learned knowledge about dog
vocalization and behavior, and/or dog barks may have a strong
emotional content because acoustic features affect homologous
inborn human abilities. According to the latter, it is worth referring
to Morton’s (1977) theory of how the acoustic parameters of the
mammalian and avian vocal signals express the aggressiveness or
subordinance of the signaler. Morton concluded that atonal, low-
pitched signals bear aggressive meaning, whereas tonal and high-
pitched signals express subordinance or the lack of aggressiveness.
Most important, these results were based on several nonrelated
species; therefore, Morton assumed that these rules could be uni-
versal, at least among mammals and birds. Our results partly
support Morton’s theory as we found that high fundamental and
peak frequencies of the dogs’ barks were characteristic to the
nonaggressive situations (walk, play, and alone), whereas low
fundamental and peak frequencies were found in the stranger and
schutzhund situations, which were characterized by the humans as
the most aggressive ones as well. It is interesting that tonality had
no significant effect on how our subjects categorized the barks or
on what kind of emotions they attributed to them (but see Yin &
McCowan, 2004).
It is interesting that humans judged bark sequences of shorter
interbark interval to be more aggressive. This has some parallels
with the findings of McConnell and Baylis (1985), who found that
shepherds’ high-pitched, quickly pulsating whistles have an acti-
vating, encouraging effect on dogs, because in both cases the
pulsing of the sound is associated with some urgency on the part
of the listener.
We found that most of the dog barks bear a very strong emo-
tional content for human listeners. This suggests that basic emo-
tions and the ability to recognize them is an ancient capability
shared by animals and humans. The fact that humans found the
barks to express basic emotions could be a sign that this form of
Figure 6. Percentages of correctly categorized situations in the pooled groups of human listeners. The different
letters refer to significantly differing situations (one-way analysis of variance with Student Neuman–Keuls post
hoc test). Error bars represent standard errors of the mean. Schutzh. ⫽ schutzhund.
Table 5
“Confusion Matrix” of Correct and Incorrect Identifications of Bark Situations
Situation
Answers (%)
Stranger Schutzhund Walk Alone Ball Play
Stranger 58.33
a
12.96 7.41 9.26 5.56 6.48
Schutzhund 30.56 48.15
a
3.70 4.63 3.70 9.26
Walk 11.11 4.63 23.15
a
18.52 20.37 22.22
Alone 12.96 4.63 10.19 47.22
a
14.81 10.19
Ball 16.67 6.48 11.11 25.93 25.00
a
14.81
Play 4.63 12.96 12.96 9.26 23.15 37.04
a
Note. The answers from the three groups were pooled.
a
Indicates percentage of correct answers.
143
HUMAN CLASSIFICATION OF DIFFERENT DOG BARKS
vocalization has communicative relevance for humans. Natural
and artificial selection could have favored the emergence of such
understandable vocalizations in dogs. Macedonia and Evans
(1993) suggested that pressure for predator-specific avoidance
behavior resulted in functionally referential alarm calls in some of
the mammalian species, whereas in those mammal prey species for
which predator avoidance always requires the same behavior,
alarm calls remain nonreferential. Considering this, we may expect
context-specific barks in such situations that it is important for
humans to react accurately after hearing the given bark (e.g., when
the dog expresses aggression or despair). Contrary to this hypoth-
esis, the highly emotional situations, where we collected our vocal
samples, as well as the listeners’ ratings indicate that dog barking
is mainly an emotionally driven communicative process, which
makes referentiality less likely (Hauser, 1996, 2000). We think that
there are at least two key conditions that changed dog barking to
an effective communicative signal between dog and human: First,
domestication processes have resulted in dogs that are more de-
pendent on humans, making them more human oriented (Miklo´si
et al., 2003; Miklo´si, Topa´l, & Csa´nyi, 2004). Second, humans
have selected for dogs that bark reliably and in accordance to
certain behavioral and emotional situations.
References
Bleicher, N. (1963). Physical and behavioral analysis of dog vocalizations.
American Journal of Veterinary Research, 24, 415– 427.
Cohen, J. A., & Fox, M. W. (1976). Vocalizations in wild canids and
possible effects of domestication. Behavioural Processes, 1, 77–92.
Coppinger, R., & Coppinger, L. (2001). Dogs. Chicago: Chicago Univer-
sity Press.
Feddersen-Petersen, D. U. (2000). Vocalization of European wolves (Canis
lupus lupus L.) and various dog breeds (Canis lupus f. fam.). Arch Tierz
Dummerstorf, 43, 387–397.
Ga´csi, M., Topa´l, J., Miklo´si, A
´
., Do´ka, A., & Csa´nyi, V. (2001). Attach-
ment behavior of adult dogs (Canis familiaris) living at rescue centers:
Forming new bonds. Journal of Comparative Psychology, 115, 423–
431.
Hauser, M. D. (1996). The evolution of communication. Cambridge, MA:
MIT Press.
Hauser, M. D. (2000). A primate dictionary? Decoding the function and
meaning of another species vocalizations. Cognitive Science, 24, 445–
475.
Herman, L. M., Abichandani, S. L., Elhajj, A. N., Herman, E. Y. K.,
Sanchez, J. L., & Pack, A. A. (1999). Dolphins (Tursiops truncatus)
comprehend the referential character of the human pointing gesture.
Journal of Comparative Psychology, 113, 347–364.
Kubinyi, E., Miklo´si, A
´
., Topa´l, J., & Csa´nyi, V. (2003). Social anticipa-
tion in dogs: A new form of social influence. Animal Cognition, 6,
57– 64.
Kubinyi, E., Topa´l, J., Miklo´si, A
´
., & Csa´nyi, V. (2003). The effect of a
human demonstrator on the acquisition of a manipulative task. Journal
of Comparative Psychology, 117, 156–165.
Lehner, P. N. (1978). Coyote vocalizations: A lexicon and comparisons
with other canids. Animal Behaviour, 26, 712–722.
Macedonia, J. M., & Evans, C. S. (1993). Variation in mammalian alarm
call systems and the problem of meaning in animal signals. Ethology, 73,
177–197.
McConnell, P. B., & Baylis, J. R. (1985). Interspecific communication in
cooperative herding: Acoustic and visual signals from human shepherds
and herding dogs. Zeitschrift fu¨r Tierpsychologie, 67, 302–382.
McKinley, S., & Young, R. J. (2003). The efficacy of the model–rival
method when compared with operant conditioning for training domestic
dogs to perform a retrieval-selection task. Applied Animal Behaviour
Science, 81, 357–365.
Miklo´si, A., Kubinyi, E., Topa´l, J., Ga´csi, M., Vira´nyi, Z., & Csa´nyi, V.
(2003). A simple reason for a big difference: Wolves do not look back
at humans but dogs do. Current Biology, 13, 763–766.
Miklo´si, A
´
., Polga´rdi, R., Topa´l, J., & Csa´nyi, V. (2000). Intentional
behavior in dog– human communication: An experimental analysis of
‘showing’ behavior in the dog. Animal Cognition, 3, 159 –166.
Miklo´si, A
´
., Topa´l, J., & Csa´nyi, V. (2004). Comparative social cognition:
What can dogs teach us? Animal Behaviour, 67, 995–1004.
Morton, E. S. (1977). On the occurrence and significance of motivation-
structural rules in some bird and mammal sounds. American Naturalist,
111, 855– 869.
Nicastro, N., & Owren, M. J. (2003). Classification of domestic cat (Felis
catus) vocalizations by naive and experienced human listeners. Journal
of Comparative Psychology, 117, 44–52.
Pepperberg, I. M., & McLaughlin, M. A. (1996). Effect of avian– human
joint attention on allospecific vocal learning by grey parrots (Psittacus
erithacus). Journal of Comparative Psychology, 110, 286 –297.
Pongra´cz, P., Miklo´si, A
´
., & Csa´nyi, V. (2001). Owners’ beliefs on the
ability of their pet dogs to understand human verbal communication. A
case of social understanding. Cahiers de Psychologie Cognitive/Current
Psychology of Cognition, 20, 87–107.
Pongra´cz, P., Miklo´si, A
´
., Kubinyi, E., Gurobi, K., Topa´l, J., & Csa´nyi, V.
(2001). Social learning in dogs. The effect of a human demonstrator on
the performance of dogs (Canis familiaris) in a detour task. Animal
Behaviour, 62, 1109–1117.
Pongra´cz, P., Miklo´si, A
´
., Tima´r-Geng, K., & Csa´nyi, V. (2003). Prefer-
ence for copying unambiguous demonstrations in dogs. Journal of
Comparative Psychology, 117, 337–343.
Riede, T., & Fitch, T. (1999). Vocal tract length and acoustics of vocal-
ization in the domestic dog (Canis familiaris). Journal of Experimental
Biology, 202, 2859–2867.
Riede, T., Herzel, H., Hammerschmidt, K., Brunnberg, L., & Tembrock, G.
(2001). The harmonic-to-noise ratio applied to dog barks. Journal of the
Acoustical Society of America, 110, 2191–2197.
Schassburger, R. M. (1993). Vocal communication in the timber wolf,
Canis lupus, Linnaeus. Advances in Ethology (No. 30). Berlin, Germany:
Paul Parey Publishers.
Seyfarth, R., & Cheney, D. (1990). The assessment of vervet monkeys of
their own and another species alarm calls. Animal Behaviour, 40, 754 –
764.
Shriner, W. M. (1998). Yellow-bellied marmot and golden-mantled ground
squirrel responses to heterospecific alarm calls. Animal Behaviour, 55,
529 –536.
Soproni, K., Miklo´si, A
´
., Topa´l, J., & Csa´nyi, V. (2002). Dogs’ (Canis
familiaris) responsiveness to human pointing gestures. Journal of Com-
parative Psychology, 116, 27–34.
Tembrock, G. (1976). Canid vocalizations. Behavioural Processes, 1,
57–75.
Topa´l, J., Miklo´si, A
´
., & Csa´nyi, V. (1998). Attachment behavior in the
dogs: A new application of the Ainsworth’s Strange Situation Test.
Journal of Comparative Psychology, 112, 219 –229.
Yin, S. (2002). A new perspective on barking in dogs (Canis familiaris).
Journal of Comparative Psychology, 119, 189 –193.
Yin, S., & McCowan, B. (2004). Barking in domestic dogs: Context
specificity and individual identification. Animal Behaviour, 68, 343–
355.
Received January 7, 2004
Revision received September 28, 2004
Accepted October 1, 2004 䡲
144
PONGRA
´
CZ, MOLNA
´
R, MIKLO
´
SI, AND CSA
´
NYI
A preview of this full-text is provided by American Psychological Association.
Content available from Journal of Comparative Psychology
This content is subject to copyright. Terms and conditions apply.
