Introduction

Relationships between people and dogs, which represent the earliest domesticated animals, attract the attention of researchers in many ways1,2,3. The genetic history of dogs extends into the Palaeolithic, when at least five major ancestral lineages had diversified4. The exact timing of the emergence of the dog lineage remains unknown5. Current genetic studies estimate a time of dog–wolf divergence between 25,000 and 40,000 years ago4,6. Domestication of dogs caused differences from wolves in several ways. Adaptations included alterations in sequences of ritualized behaviour, and changes in motivational context for certain behaviours including changes in response thresholds7,8. Dogs are better at cooperating with humans than wolves are. They are more able to recognize our facial expressions and our communication signals; therefore, they work better with humans than wolves do9. Dog puppies display more communicative signals to facilitate social interactions, in contrast to wolf pups9,10. Thus, dogs show a unique complex of skills acquired for communication with humans11. Dogs bark frequently and use them in a wider range of contexts than their close relatives the wolf12,13 and coyote12, and the barks seem to have evolved from the low-frequency barks of wolves, which are mainly produced during agonistic interactions14. Barks changed quantitatively and qualitatively during the domestication process15. The complexity of the dog’s vocal repertoire has been extended by using mixed sounds in barking context. Various barking forms are generated via a mix of transitions and gradations of harmonics, intermediates and noisy subunits3. Some authors suggest that the original function of barks is mobbing (alerting other pack members and calling them in to defend the territory together)16. From this, hunting barks may have also be derived, as their function is to alert humans and lead them to the prey that the dog has found. Hunting-dog breeds were originally bred to fulfill some kind of hunting work. Humans artificially selected some breeds to bark frequently14,17,18. These facts indicate a strong selection force on barking performance in hunting dogs, which are also mentioned for several dog breed standards14. Additionally, sport-hunting breeds have been adapted to specific hunting work via an improved physiology, e.g. cardiac function, blood flow, and cognitive performance19. Some breeds have been bred for specific kinds of hunt, e.g. pointing breeds (pointers) were bred from dogs that were able to stand quietly and maintain its position in the face of the animal's scent until the human counterpart reaches the place where the animal is hiding. In contrast, other breeds were developed for multiple purposes or to be versatile and able to perform a number of tasks (e.g. hounds, retrievers, spaniels). Spaniels and retrievers will find and bring a shot animal to a hunter20,21. The dachshund is considered according to Fédération Cynologique Internationale one of the most versatile hunting breeds and not just for hunting below ground. Cooperative hunting dogs keep close contact with the hunter during the hunt (e.g. retrievers) whilst non-cooperative hunting dogs perform independent work, either chasing (e.g. beagles) or attacking (e.g. terriers) the animal22. Small terriers locate and hunt smaller mammals, while larger terriers are able hunt larger animals. Selected hunting breeds were bred to follow prey while barking, and some are even capable of specialised barking; on the other hand, other breeds have to stand silently and motionless near the found animal until the arrival of the hunter20,23. Specific forms of barking produced by some hunting dogs are even requested in the dog breed standards of international cynological organizations14. Recent studies have also shown that barks contain meaningful information based on context17,24,25,26,27, individual identity24,26,27, inner states25, and emotionality28,29.

We aimed to test whether hunting dogs produce barks differentially during encounters with different animal species. In order to test the barking of dogs at animals of different sizes, we needed to choose a universal dog breed. The choice of breeds for such a purpose was determined by the legislation of the Czech Republic. The Hunting Act distinguishes and defines four types of work performance. Dachshunds and terriers are the only groups of hunting dogs that can pass all four tests and be used for all types of hunting work in the Czech Republic. Although hounds are better suited to hunting wild boar, our law prohibits the use of dogs of a height of 55 cm or more for hunting ungulates. Dachshunds and terriers are no longer bred for earth-hunt work only, but are used for their independence and ability to adapt to surface work. For these reasons, dachshunds and terriers belong among the most common breeds for hunting all kinds of game in the Czech Republic. These breeds are considered to be independently working breeds that are able to work without visual contact with the hunter.

We used two different dog-breed groups: (A) dachshunds and (B) terriers. The hunting style of both breeds is as follows: looking for an animal, starting to bark, following in the footsteps of the animal, continuing to bark and chasing the animal to the hunter. In order to test potential bark differentiation, we recorded barks elicited by encounters with four different animal species. We selected animal models that would represent both (1) potentially dangerous animals (red fox, wild boar) and (2) non-dangerous animals (fowl, rabbit). Encounters with wild boars represent, for the small-bodied dog breeds used in our study, a real life-threatening situation accompanied by increased levels of emotion. Emotions with high-arousal are associated with a high sympathetic tone and a low parasympathetic tone30. Emotional arousal changes the muscular actions required for vocal production (e.g. diaphragm, vocal and intercostal muscles) which affect the way air flows through the vocal system and thus alter the quality of the sounds produced31. Expression of emotions informs other group members about the probable behavior intentions32. Vocal responses of dogs to these animal models could show us if acoustic structure of barks allow us to make predictions about how such signals change according to emotional arousal. Review on vocal correlates of emotions revealed that vocal signals of mammals become longer with increased arousal, louder and harsher, with higher and more variable frequencies and produced at faster rates30. Expression of emotions and perception of emotional states during hunting could play an important role in dogs as social species. Expression of emotions thus should benefit dogs by regulating social interactions within groups during hunting, whether it is a group of more dogs or dogs and hunters.

We have postulated the following hypotheses:

  1. (1)

    Barks produced during encounters with different animal species will have different acoustic structures. We can predict this based on previous robust literature showing that dog barks can be categorized based on various types of context17,24. Barks during hunting should vary with the demands of the situation, e.g. based on urgency or a common species-specific animal response, such as running away in hares, flight in pheasants, active defence in wild boar, etc.

  2. (2)

    Barks will show a different acoustic structure depending on the arousal of the caller. We predict that barks produced during the presence of non-dangerous species might differ from those produced in the presence of potentially dangerous species. We tested whether the degree of threat from the species of animal the dog encounters (e.g. wild boar versus rabbit) is reflected in the structure of the acoustic parameters based on the valence-arousal model33.

Results

Animal species context

We analyzed 1888 barks of 19 individual dogs belonging to two breeds—(1) dachsund and (2) terrier—which were produced in response to four different animal species (Fig. 1): wild boar, red fox, rabbit and fowl (Table 1). Figure 1 shows the bark spectrograms of the four types of barks (audio files: Additional Files 1–4).

Figure 1
figure 1

Spectrograms showing barking responses to wild boar, red fox, rabbit and fowl. Barks in each panel were produced by the same individual: Dachshunds Pecka (a) and Vendula (b). Fox Terriers Hard (c) and Gam (d).

Full size image
Table 1 Tested dogs and number of analyzed barks per context.
Full size table

To investigate whether acoustic parameters of barks differ in response to encountered animal species, we performed three separate stepwise discriminant function analyses (DFA) for three datasets (1) dachshunds only, (2) terriers only and (3) pooled data of both breeds.

The resulting models show a similar pattern for all three datasets independently (Table 2), including classification results (Table 3–4): the dachshund model (N = 9, n = 810, Wilks’ lambda = 0. 495, p < 0.001) included 8 acoustic parameters, the terrier model (N = 10, n = 1078, Wilks’ lambda = 0. 528, p < 0.001) included 10 parameters, and the pooled model (N = 19, n = 1888, Wilks’ lambda = 0. 549, p < 0.001) included 12 variables (Table 5). DFA for all three datasets showed that barks of dachshunds and terriers are recognizable based on the animal encountered with a higher probability than would be classified by chance (Table 2). The randomization procedure confirmed that these results were significant (pDFA, p < 0.001) for all three DFA models. Barks evoked by wild boar were classified better than those evoked by other animals (dachshund model: 60.6%; terrier model: 80.5%; pooled model: 73.3%), which is much higher than classification by chance (22.2%; 19.5% and 20.7% respectively) (Table 3). The percentages of correctly classified barks evoked by the other three animals were similar in all three models: dachshund model (43.3–52.1%), terrier model (27.8–56.2%), pooled model (35.7–49.8%). Classification outputs were significantly higher in comparison to classification by chance (dachshund model: Chi-Square = 128.7, df = 3, p < 0.001; terrier model: Chi-Square = 232.9, df = 3, p < 0.001; pooled model: Chi-Square = 170.9, df = 3, p <  0.001).

Table 2 The resulting discrimination function models. (Classif Orig/Valid /a priori) percentage of correct classification based on stepwise DFA, cross-validated DFA and a priori probability (classification by chance); (DF1, DF2) variable mostly correlated with the first and second discrimination function.
Full size table
Table 3 Confusion matrix for the animal species categorization task.
Full size table
Table 4 Classification results of bark subcategories in other studies. (Categ) classified categories, (No categ) number of classified categories, (Classif) correct classification percentage, (Chance) classification by chance, (Differ) difference between correct classification and classification by chance.
Full size table
Table 5 Measured acousticparameters.
Full size table

According to the arousal hypothesis it is expected that frequency parameters and call duration will differ between species which differ in their life-threatening level to hunting dog. We selected four frequency-related parameters (F5, Q1F, F95, Q3F) and call duration to test whether differences in the arousal state are encoded in barks.

Barks in response to a wild boar showed significantly lower F5 than in response to other animals: F5 (GLM: F 3,18 = 8.3, multiple comparisons: wild boar vs. all animals: p < 0.001). Differences between other animals (rabbit, fowl and red fox) were not significant (p ≥ 0.52), (Fig. 2). The lowest frequencies in the wild boar in comparison to other animals were also shown by the other three frequency parameters (Fig. 2).

Figure 2
figure 2

Univariate GLM comparison of acoustic parameters for arousal testing. Frequency parameters F5 and Q1F show the proportion of the acoustic energy in low frequencies (a–b), and F 95 and Q3F (c–d) show signal components in higher frequencies and temporal parameter Duration shows bark duration (e).

Full size image

Parameter F 95 showed significant differences between all animal pairs (GLM: F3,18 = 12.8, wild boar vs. all animals: p < 0.001, rabbit vs. red fox: p < 0.001, rabbit vs. fowl: p  = 0.002, red fox vs. fowl: p = 0.010).

Parameter Q3F also showed the biggest differences between wild boar and all other animals (GLM: F3,18 = 6.99, p < 0.001). The significant difference also showed comparison of the rabbit vs. fowl (p = 0.001) and rabbit vs. red fox (p < 0.001). Pair comparison of the red fox with rabbit and wild boar showed a significant difference (p < 0.001), while comparison with fowl was not significant (p = 0.073). Model of parameter Q1F did not show a significant effect of animal species (GLM: F3,18 = 1.52, p  = 0.222).

Temporal parameter Duration has been the longest in barks in response to a wild boar in comparison with all other animals (GLM: F 3,18 = 32.4, p < 0.001). The comparison showed a significant difference for all other pairs (p < 0.001), only red fox and fowl did not differ (p = 1.0).

Discussion

We aimed to test for potential differences in the barking of dogs when they encounter four different animal species—wild boar, red fox, rabbit and fowl—which represent models of various size and danger level for dogs. We used two groups of dogs—dachshunds and terriers. Classification results of a discrimination analysis showed that barks of dachshunds and terriers can be categorized based on the animal species they encountered with a higher probability than would be expected if classification was random. It was revealed that the most distinctive barks were made during encounters with the most dangerous animal, the wild boar. The same pattern was shown when we pooled both datasets together. On the contrary, barks evoked by red fox encounters were classified with a similar result to the other smaller and non-dangerous animals—here, the rabbit and fowl. Although the red fox represents a potentially dangerous species for small dog breeds, the provoked barks were not classified with a much higher success rate than barks at animals that pose no threat. This would indicate that the key parameter might be the body size of the animal the dog meets. When we compare the average success in the classification of barks at different animals with the classification results of different bark-classifying methods from different contexts we can see comparable results (see Table 4); however, different classification methods with different numbers of individuals and chance levels were used in these studies. When we take into account the resulting classification by chance level using weighting classification outputs by chance level, such a comparison may give us a general overview of the classification success generated by different classification methods.

The question is why hunting dogs bark differently at different animal species; is it because of a different inner state or it is a signal directed at their human companions? The barks investigated in previous studies were collected in different distinct social scenarios such as disturbance, isolation, play, presence of a stranger, fight training, begging, walk preparation, etc.14,24, and the acoustic structure of barks presumably reflects the inner states of dogs34 associated with these social contexts, which can be also recognized by humans25,28,35. Across these different contexts the emotional state of a dog may likely differ. We tested therefore whether the degree of threat from the species of animal the dog encounters (wild boar versus rabbit, red fox and fowl) can be reflected by the acoustic parameter structure according to the valence-arousal model33.

Barks produced during contact with a wild boar showed significantly lower frequency parameters. Frequency parameters seems to be not such a strong reliable indicator of emotions as they may both increase and decrease with an increase in arousal30, e.g. a shift in energy distribution towards higher frequencies was found in Weddel seal36, silver fox37, house cats38, red-fronted lemur39, squirrel monkey40,41, while towards lower frequencies in sheep42 and baboon43. Temporal parameters seem to be more consistent predictors of arousal. Vocalizations of mammals become longer with increase in arousal more frequently30. It is consistent with our result of the longest duration of barks produced during encounters with the most dangerous of tested animals, the wild boar.

In our case, it seems that the variability of barking, which depends on the species of animal the dog encounters, is an expression of a dog's inner state rather than functionally reference information. In addition, the expression of the inner state in barking appears to depend on the size of the potential threat. Barking in the case of a great threat (wild boar) is more specific than barking in the case of a smaller threat (red fox) or no threat (rabbit, fowl). This phenomenon could then indicate an innate ability, as it has been reported in the case of naive dogs, without previous experience with wild boar.

Both dog owners and non-owners, including adults and young children, are able to categorize a dog’s emotional state and barking context above the level of chance17,24,25,28. Domestic dogs bark frequently in comparison to feral dogs, which produce barks relatively rarely44. This fact could indicate that barking is at least in some way used for communication with humans22. Some authors have considered barks to be an exaggerated by-product of the domestication process that has no specific function12. Previous studies have shown that dog barks are able to express a wider range of emotions compared to those of wolves. Such a change in the acoustic communication of dogs has resulted from their association with humans45. Recognizing dog barks may be advantageous in inter-specific interactions since dog domestication occurred at least 30,000 years ago46,47,48. This process was initiated by European hunter-gatherers47. Mutualism between dog and hunter probably took place early after domestication49 when dogs assisted in the hunting of prey47. The ability to draw attention to different animal species complements hitherto known communicative skills like human–dog communication via eye contact50, changes in facial expressions of dogs affected by human attentional state51, the ability of dogs to understand the communicative cues of humans52 or communication using eye gaze53,54, and the widely known ability of dogs to understand human pointing gestures52,55.

Hunting dogs were bred to follow the trail of an animal. Some of these breeds were probably selected for a specific type of barking14. Such specialization could lead to the forming of an additional ability in comparison to other breeds not selected for hunting abilities. The hearing system of canids has primarily evolved to optimize predation, especially to localize sounds produced by potential prey56. Recognition of animal species could be favoured in reciprocal cooperation during hunting, e.g. recognition of potentially dangerous vs. non-dangerous animals could be especially favoured. Understanding the regulation mechanisms of mutual communication between humans and animals is especially important for animals such as dogs living in close contact with their human partners, depending on them for food, care and health57. The hunting activity of dogs with humans is considered to be derived from the cooperative behavior of wolves22. In hunting dogs, we might suppose that animal-encounter-specific barking may significantly increase the effectiveness of hunting events.

Methods

Ethics statement

This is a statement to confirm that all experimental protocols were approved by a named institutional or licensing committee. The authors declare that the present study complies with the current laws of the Czech Republic. The research was carried out in accordance with recommendations in the Guide for Care and Use of Animals of the Czech University of Life Sciences, Prague. This study focused on the recording of sounds, which was not considered an invasive experimental technique by The Professional Ethics Commission of the Czech University of Life Sciences Prague (project no 14/19) and did not require a special permit.

Subjects

We recorded barks from 19 dogs (nine dachshunds—two males and seven females, eight fox terriers—four males and four females, one male Welsh terrier and one male jagdterrier) (Table 1) during December 2016 and March 2017. The age of both dachshunds and fox terriers ranged 1 to 11 years, the Welsh terrier was three years old and the jagdterrier two years old. Some dogs had previous experience with tested animal species and others were naive, with no previous experience (Table 1). The dog owners were coauthors of this study (KB and JA) and their colleagues.

Experimental procedure

Recordings were conducted under semi-controlled conditions during outdoor experiments, not during huntig events. The experimental site was chosen in isolation from other objects and potential noise. No vegetation other than low grass was present during the winter and early spring. The experiments were performed in sunny weather without rainfall and almost no wind. Each dog was tested only once per day. The interval between experimental days was longer than fourteen days. Each of 19 dogs was randomly assigned to one of the four treatments (wild boar, red fox, rabbit and fowl). Recorded barks were elicited by encounters with live four different animal species through the fence mesh. There was no direct contact between the tested animals. Only one individual dog was tested during the experiment. Each tested dog was brought to the fence, released and left alone for 5–15 min depending on the frequency of barking required. The recording microphone was placed at a distance of two metres from the fence. We used an Olympus Linear PCM LS-5 audio-recorder with a Sennheiser ME 67 microphone (frequency response 20 Hz–20 kHz) with a K6 powering module.

Acoustic analyses

We randomly selected a maximum of 30 barks per individual. These were chosen from a sample of barks of the best quality: non-overlapping barks with low background noise and a good signal-to-noise ratio. We did not obtain a full matrix as some dogs gave fewer than 30 barks. A total of 1888 barks were analyzed: 390 barks from the wild boar experiment, 508 barks from the red fox experiment, 510 barks from the rabbit experiment and 480 barks collected during the fowl experiment.We analyzed recordings using Raven Pro Sound Analysis Software (Cornell Lab of Ornithology, New York, USA) from which spectrograms were generated using the following parameters: Hann window type with a 1050 point window size, an overlap of 50%, a hop size of 11.9 ms, and grid spacing of 21 Hz. We measured 20 acoustical parameters (Table 5).

Statistical analyses

From the measured parameters we excluded those that were highly correlated (r ≥ 0.90: F95 Rel and Avg Entropy), and the remaining variables were entered into the discriminant function analysis (DFA) (Table 3). We performed three types of analyses: (1) for dachshunds, (2) for terriers and (3) for both breeds together (pooled model). We performed a stepwise DFA in order to test whether dog barks can be classified based on the animal species that they were produced in response to. The procedure selected predictors using the Wilks’ lambda criterion. We used F values as a criterion for entering or removing an acoustic parameter from a classification model (F to enter = 3.84; F to remove = 2.71). For external validation of this model we used leave-one-out cross-validation using IBM SPSS 20 (IBM Corp., Armonk, USA). Animal species were used as a group identifier and the acoustic variables were used as discriminant variables. We normalized measured variables using Z score transformation (by subtracting the mean and dividing by the variable’s standard deviation), which avoids the false attribution of weights in relation to variables measured in different units (IBM Corp., Armonk, USA). We then performed a permuted DFA (pDFA) for nested designs, which serves as a randomization procedure for non-independent data58. We calculated pDFAs using a script written in software “R” (provided by Roger Mundry) using 100 random selections and 10,000 permutations. This procedure gave a p-value which was used to determine the significance of the correct classification rate of barks to the test factor (animal species), while controlling for a single nested factor (individual). N refers to number of individuals (dogs), n refers to number of calls (barks).We used univariate general linear models (GLM) for the motivational–structural test to see if barks differ between the animal species they were produced in response to. Acoustic variables were used as dependent variables, animal species as a fixed factor, and individual dogs as a random factor. We used Bonferroni corrected post hoc multiple comparison. It was also used Chi-Square test based on the observed versus expected values to test classification outputs of DFA models.