Responses to spoken words by domestic dogs: A new instrument for use with dog owners

https://doi.org/10.1016/j.applanim.2021.105513Get rights and content

Highlights

  • We developed a detailed inventory of words dogs reportedly respond to consistently.

  • Our measure was modeled after parent word checklists used to assess infants’ language.

  • According to owners, dogs responded to a mean of 89 words (range 15–215).

  • Dogs’ work status and breed type predicted the number of words reported.

  • Commands make up the majority of words to which dogs reportedly respond.

Abstract

Dogs serve many roles in people’s lives, whether as valued family members or trained workers. Their ability to respond to human nonverbal and verbal cues is central to these roles. We aimed to develop a comprehensive owner-reported inventory of words to which owners believe their dogs respond differentially and consistently. Our tool was modeled after the MacArthur-Bates Communicative Development Inventory (Fenson et al., 2007), a parent-report checklist that assesses infants’ understanding and development of early language. The validity, reliability and usability of our measure was assessed with 165 owners of dogs of various ages and breeds. Owners identified words to which they believed their dogs responded. They reported that, on average, their dogs responded to 89 words (78 from the lists provided plus 11 added by them), half of which were commands. To examine potential owner- and dog-related predictors of words reported, owners also answered questions about their dogs and themselves. None of the owner characteristics measured uniquely predicted the number of words reported, but dogs’ professional work status, how quickly they reportedly learn new tricks, and their breed type did. A Principal Component Analysis revealed that words to which dogs reportedly respond encompassed three distinct types: General Words, Verbs (Commands), and Nouns. Thus, based on owner reports, dogs seem to vary greatly not only in the number but also in the kinds of words to which they purportedly respond. Our inventory is intended for use in research on human-dog communication designed to address a variety of research questions.

Introduction

Interactions between domestic dogs and human beings stretch back thousands of years (Miklósi, 2014; Vilà et al., 1997). Although the manner and timing of domestication is debated (Perri, 2016, Range and Virányi, 2015), the nature of the dog-human relationship has unquestionably evolved (Udell and Wynne, 2008). For example, nowadays dogs hold specialized jobs in various fields, including search and rescue, police and military forces, conservation, agriculture, medical assistance, and scent detection (for a review see Helton, 2009). Their ability to perform these jobs is, in large part, the result of their sensitivity and responsiveness to human social cues and communicative intents (Hare et al., 2002, Hare and Tomasello, 2005, Miklósi et al., 1998, Miklósi and Topál, 2013).

Indeed, due to their evolutionary history, the domestication process, and/or their close association with humans, dogs use nonverbal cues from humans to gain information about new situations (Merola et al., 2012) or to successfully complete tasks (Kaminski and Nitzschner, 2013, Udell and Wynne, 2008). For example, when a reward is hidden in one of two containers, dogs use nonverbal cues produced by humans, such as pointing (with hand, arm or leg), head turning, bowing, nodding, and eye gazing towards the target container to successfully find the reward (e.g., Miklósi et al., 1998; for reviews see Kaminski and Nitzschner, 2013; Miklósi and Soproni, 2006). However, Gácsi et al. (2009) and Wobber et al. (2009) found that although dogs from many breeds follow pointing and other nonverbal cues at above-chance levels, breeds bred to interact with humans in a working capacity (e.g., herding dogs) perform better than other breeds bred to work independently (e.g., guard and hound dogs).

Humans also use words to engage in interspecies communication and to solicit behavior from dogs. For instance, Mitchell and Edmonson (1999) found that in 4-minute play sessions, dog owners said 208 words on average to familiar or unfamiliar dogs. In general, human talk to dogs across various contexts (e.g., playtime, waiting, walking, home observations) tends to consist of short and highly repetitive utterances, including a high proportion of imperatives (i.e., commands such as, Come here! or Go get the ball!) and attention-getting words and/or phrases (e.g., Dog’s Name, Ready?; Hirsh-Pasek and Treiman, 1982; Mitchell, 2001). Indeed, dogs attend more to and seek more proximity to an experimenter whose speech both contains dog-relevant content (e.g., Good boy/girl!) and is spoken with elevated pitch and exaggerated prosody (Benjamin and Slocombe, 2018) than whose speech is missing either or both of these characteristics.

As early as 1928, scientists have sought to assess empirically dogs’ ability to comprehend what humans say to them (Kaminski et al., 2004, Pilley and Reid, 2011, Warden and Warner, 1928). For example, Warden and Warner documented the ability of Fellow, a young male German Shepherd, to respond to spoken commands by his owner. Fellow was observed to respond appropriately to roughly 68 words or phrases, including phrases such as, Go outside and wait for me, after which Fellow left the room and sat outside the door. Furthermore, Fellow responded to his owner’s commands even when the owner was not in the room (therefore eliminating the possibility that Fellow relied on behavioral cues). More recently, Kaminski et al. (2004) reported on the ability of Rico, a Border Collie, to retrieve over 200 items (e.g., stuffed toys, balls) by their unique names. Likewise, Griebel and Oller (2012) demonstrated that Bailey, a Yorkshire Terrier, could retrieve over 100 items in similar circumstances. Another Border Collie, Chaser, responded selectively to the names of over 1000 toy items (Pilley and Reid, 2011) that she was taught to distinguish over 3 years of extensive training.

Most striking, Chaser also demonstrated the ability to distinguish between verbs and commons nouns in experimental contexts (see also Ramos and Ades, 2012, for another dog with similar abilities). That is, Pilley and Reid (2011) presented Chaser with three toys and requested that she perform one of three actions (paw—place her paw on top of the item; nose—touch her nose to the item; and take—pick up the item in her mouth) with the toys. The researchers randomly combined the action commands with toy items therefore requesting combinations that Chaser had likely never encountered. More important, the researchers were blind to Chaser’s actions as they requested action/item combinations from another room. Despite relying only on verbal instructions, Chaser’s performance was significantly above chance. On the basis of these results, Pilley and Reid argued that Chaser understood spoken words referentially; specifically, that toy items exist independently from the commands directed towards them (see also, Pilley, 2013, for additional evidence; but see Bloom, 2004; Markman and Abelev, 2004, for arguments against this rich interpretation).

The studies discussed above examined exceptional dogs with years of intensive training. Consequently, Fellow, Rico, Bailey, and Chaser may not be representative of typical dogs. Indeed, Fugazza et al. (2021) recently reported similar skills in 7 additional exceptional dogs identified only after a two-year world-wide search for such dogs and attempting to train close to 50 more. Although it is far from clear on what basis these dogs respond to words—be it on the basis of the words’ meanings or associative learning—what is clear is that at least some dogs can respond consistently and differentially to spoken words and phrases. Understanding the extent of more typical dogs’ word-related responses may prove useful for determining their potential as working dogs. Dogs that respond appropriately to many spoken words or phrases should be more likely to excel in their given field (Helton, 2009). Currently, however, no reliable assessment tool exists for identifying words (and how many of them) to which dogs might respond.

In an effort to identify words purportedly responded to by more typical dogs, Pongrácz et al. (2001) asked 37 dog owners to list all spoken words and phrases that they used in interactions with their dogs as well as the frequency with which their dogs obeyed each utterance (i.e., every time, in contextually appropriate situations, or occasionally) and the situations in which they used them. Owners listed 29 utterances on average, with most words and phrases requesting postures and disallowance (e.g., sit, come, down). Furthermore, owners reported that their dogs responded to the majority of these words appropriately every time or in contextually appropriate situations.

However, because Pongrácz et al. (2001) relied on owners’ ability to recall words outside of the context in which they used these words with their dogs, they may have underestimated the true number of words to which dogs might respond consistently. Indeed, in similar studies with parents of preverbal infants asked to free recall all words their infants understand or respond to, parents have been shown to forget words (see Fenson et al., 1994, for a review).

Using owners as informants, the current study aimed to develop a reliable and valid tool designed to measure the number of words to which dogs might respond consistently and differentially that also minimized demands on owners to conjure up these words unassisted. Our measure was inspired by the MacArthur-Bates Communicative Development Inventory (MB-CDI; Fenson et al., 2007) and other parent-report instruments designed to assess young children’s understanding (i.e., their receptive vocabulary) and production (i.e., their expressive vocabulary) of early words and phrases. For example, the Words and Gesture version of the MB-CDI, designed for infants up to age 16 months, contains lists of words separated by categories (e.g., animal names, vehicles), and parents are asked to identify each word or phrase that their child seems to understand (i.e., by responding to it consistently) or understands and says. The MB-CDI has been adapted into almost 100 languages and dialects and is widely used in research on young children’s language development. Indeed, infants’ early receptive scores on the MB-CDI predict their later language abilities (e.g., Patrucco-Nanchen et al., 2019). The Language Development Survey is another example of a 310-word vocabulary checklist for use with parents of children between 1.5 and 5 years, but it focusses on expressive vocabulary (Rescorla, 1989).

Although one could argue that parents are likely to be biased and unreliable, particularly when evaluating their infants’ receptive vocabulary (cf. Tomasello and Mervis, 1994), research with infants as young as 18 months suggests otherwise. For example, Styles and Plunkett (2009) presented 18-month-olds with 12 picture pairs of concrete objects whose names are reported as understood by 50% of infants at this age on the MB-CDI. Each pair was from the same semantic category (e.g., animals; dog vs. horse) and one picture was designed as the target while the other as the distractor. Infants were shown each pair for 5 s, and the target picture was named at the half-way point (e.g., Look at the dog!). The researchers compared looking time at the target and distractor pictures before and after the target was named as a function of whether parents reported on the MB-CDI that their infants understood the target and/or the distractor words. Styles and Plunkett found a significant increase in looking time at the target but only after it was named and only for targets that parents reported their infants understood. Thus, despite measuring receptive language in different ways, parent reports and infants’ looking behavior both provide converging evidence of infants’ early comprehension of words.

In fact, to obtain a comprehensive estimate of infants’ vocabulary size, parents may be better suited for that task than laboratory measures or trained observers. That is, before infants speak, they demonstrate clear receptive language abilities based on how they respond when spoken to by others. For example, they direct their eye gaze towards appropriate pictures of objects in response to hearing word labels (Fernald et al., 1998, Styles and Plunkett, 2009), or they perform recognizable actions or gestures like waving bye-bye in response to spoken words (Fenson et al., 1994). Through everyday interactions with their infants, parents observe them in various contexts and as such, can pick up on the statistical regularity between words uttered and consistent responses that follow. As such, parents can report on these indicators of their infants’ comprehension (or use) even if these occur only sporadically or in limited contexts. In contrast, trained observers would be unlikely to observe infants in sufficient contexts to obtain comprehensive estimates of their overall receptive vocabulary size. In addition, the observational approach would be prohibitively expensive, labor intensive, and time consuming, not to mention that the mere presence of an observer in the home or laboratory could prevent infants from behaving naturally due to fear, anxiety, or shyness (Fenson et al., 1994).

Interestingly, Patrucco-Nanchen et al. (2019) found that parent reports of 16-month-old infants better predicted their language abilities at 29 months than did the infants’ own performance on a laboratory task in which they were shown pairs of pictures, heard a word referring to one of the pictures and had to touch the corresponding picture labeled on each trial (e.g., Where is the dog? Touch the dog!). Even though this laboratory task used the same words as the MB-CDI, and even though this more direct measure eventually became a better longitudinal predictor of language in older infants, the authors suggested that in very young infants “this direct measure may be too complex for most children [at 16 months] to perform given their cognitive and attentional limitations (p. 7).”

As described previously, like human infants, dogs can respond differentially, consistently, and appropriately when various words and phrases are spoken by humans (e.g., Kaminski et al., 2004; see also, Dalibard, 2009; Ramos and Mills, 2019). Moreover, Pongrácz et al.’s (2001) results indicate that, like parents, dog owners can report on words they believe their dogs respond to, although Pongrácz et al.’s (2001) reliance on owners’ word recall may have led owners to forget some words. In addition, like parents, dog owners can report on words they believe their dogs respond to sporadically or in limited contexts, or might be difficult to assess in a laboratory setting.

Hence, to obtain more comprehensive estimates of the possible number and kind of words that dogs may respond to, we relied on dog owners. In doing so, our aim was not to determine whether dogs understand words referentially or learn to respond to them associatively (see Markman and Abelev, 2004; Pilley and Reid, 2011 for discussion), nor did we attempt to determine whether dogs differentiate words and phrases on the basis of their semantic or auditory properties, on speakers’ accompanying nonverbal gestures or intonation, or on other contextual cues (e.g., location, time of day or object cues; cf. Mills, 2005). Indeed, owners are unlikely to know on what basis their dogs respond to specific words. However, we do believe that, like parents, owners can pick up on statistical regularities between when they utter certain words or phrases and specific responses exhibited by their dogs following these utterances.

Relying on dog owners to obtain reliable information about dogs’ temperament and behaviors is already well-established in the field, as evidenced by the recent trend to recruit owners as citizen scientists (e.g., Stewart et al., 2015; see Hecht and Spicer Rice, 2015, for a review) and by the panoply of owner questionnaires that currently exist (e.g., Monash Canine Personality Questionnaire, Ley et al., 2009; Canine Behavioral Assessment and Research Questionnaire, Hsu and Serpell, 2003; see Wiener and Haskell, 2016, for a review). Online owner questionnaires also have the advantage of potentially reaching a larger and more representative sample than relying solely on local dog owners willing to participate in laboratory or home-based research.

Our first goal was to assess the reliability, usability, and content validity of our online instrument in a large sample of 165 dog owners. To obtain a range of responses about typical dogs, we advertised primarily in venues that catered to dog owners. However, we also targeted dog trainers and experts, and owners of professional dogs to assess content validity (i.e., whether we had identified most words commonly used with dogs to which they are also believed to respond), as we hypothesized that they might engage in more activities with their own dogs than the typical dog owner and therefore, teach them to respond to words related to these activities.

In addition to obtaining estimates of the number and kind of words or phrases to which dogs might respond, we examined whether owner and dog characteristics predict words and phrases reported. To do so, owners were asked questions about themselves and their households (e.g., educational level, dog training experience, household member composition), as well as their dogs (e.g., breed, age, sex, training background). Based on previous research, we expected that some owner and dog characteristics would predict the number of words to which dogs purportedly respond, including age of dogs and owners, and owner education. For example, Kubinyi et al. (2009) found that dogs’ training history predicted their current trainability. More relevant, Pongrácz et al. (2001) found that older dogs and dogs of older owners reportedly responded to more words, whereas more educated owners reported that their dogs responded to more 3-word utterances and fewer 1-word utterances than less educated owners. However, Pongrácz et al. (2001) examined each of their predictor variables in separate analyses, making it difficult to identify which of these variables uniquely predicted owner estimates of dogs’ word-based responses once other predictors had been taken into account. Consequently, in addition to examining zero-order correlations, we used regression analyses to control for shared variance to determine whether any predictors accounted for unique variance.

Furthermore, based on breed differences in performance on nonverbal pointing tasks (e.g., Dorey et al., 2009; Gácsi et al., 2009; Wobber et al., 2009), we expected breed differences could also exist in dogs’ responses to words (or to their accompanying nonverbal cues, e.g., gestures, intonation). For this reason, we examined whether owners of different breeds reported that their dogs respond to varying numbers of words. Although we had insufficient numbers of dogs of individual breeds to assess these differences, we categorized breeds by breed type as others have done (e.g., Gácsi et al., 2009; Wobber et al., 2009). That is, major kennel clubs, such as the American Kennel Club (www.akc.org), the Kennel Club (www.thekennelclub.org.uk), and the Fédération Cynologique Internationale (FCI; http://www.fci.be) differentiate roughly seven major breed types based on the function for which dog breeds were initially bred (Mehrkam and Wynne, 2014), although kennel clubs differ slightly in their nomenclature, and on the exact number of breed groups they distinguish and to which they assign certain breeds. In general, however, breed types typically include Herding (dogs bred to herd sheep and cattle), Toy-Companion (dogs bred for human companionship), Hound (dogs bred for chasing prey by sight or scent), Working-Guardian (large dogs bred to labor or guard), Terrier (dogs bred to hunt vermin in their burrows), Sporting-Gun (dogs bred to assist hunters to find and retrieve game), and Spitz-Primitive (also referred to as Northern; dogs bred for northern climates).

Finally, we examined whether response patterns for individual words and phrases could be reduced to a few meaningful components. To do so, we conducted an exploratory Principal Component Analysis (PCA) on words and phrases reported to see whether a pattern emerged in how owners selected items.

Section snippets

Participants

Recruitment was done with online advertisements on social media platforms (e.g., Facebook™), printed posters placed around our university campus, or emails sent to individuals whose dogs were currently enrolled in ongoing studies. We targeted some of our advertisements to online groups that catered to dog trainers and groups of individuals interested in scientific research, assuming that we would recruit some dog trainers and experts, and owners of professional dogs. We targeted these groups

Words and phrases provided in the word lists

Owners were given 172 words and phrases presented in word lists with response options indicating how much they believed their dog responded to each word. However, they selected primarily the extremes response options with middle options chosen infrequently. Consequently, words reportedly responded to Never (19%) or No Answer (32%), Did at One Time (1%), and Rarely (3%) were assigned 0, whereas words believed to be responded to Selectively (11%) or Often/Most Times (34%) were given a score of 1.

Discussion

The goal of the current study was to develop a reliable, valid, and comprehensive instrument designed to estimate the number of words and phrases to which domestic dogs reportedly respond consistently and differentially. Multiple ways exist in which dogs might learn to respond in particular ways when specific words or word phrases are uttered by humans without assuming that dogs are responding to these utterances on the basis of their meaning. For example, they could acquire conditioned

Conclusion

In sum, we developed an inventory designed to capture the breadth of words and phrases to which dogs might respond as reported by their owners. None of the owner characteristics measured uniquely predicted the number of words they reported, but dogs’ reported learning speed, breed group and professional work status did, suggesting that formal training might be required for dogs to learn to respond to many words. The labor-intensive work of comparing owner reports with observational and

Ethics

Data were collected in accordance with the Canadian Tri-Council Policy Statement (TCPS2) Ethical Conduct for Research Involving Humans and approved by the Human Research Participants & Ethics Committee, Department of Psychology and Neuroscience at Dalhousie University.

Funding

This work was supported by a Natural Sciences and Engineering Research Council of Canada Postgraduate Scholarship-Doctoral award to the first author (which had no direct involvement in the research presented here).

CRediT authorship contribution statement

Both authors conceived and designed the study, CR conducted the study, SJ analyzed the results, and both co-wrote the paper.

Acknowledgements

We thank Dr. Simon Gadbois for useful discussion and advice. Portions of this work were presented at the Society for Research in Child Development Biennial Convention (2019) and at local conferences.

Conflicts of Interests/Competing Interests

None.

References (57)

  • L.R. Mehrkam et al.

    Behavioral differences among breeds of domestic dogs (Canis lupus familiaris): current status of the science

    Appl. Anim. Behav. Sci.

    (2014)
  • Á. Miklósi et al.

    What does it take to become “best friends”? Evolutionary changes in canine social competence

    Trends Cogn. Sci.

    (2013)
  • T. Patrucco-Nanchen et al.

    Do early lexical skills predict language outcome at 3 years? A longitudinal study of French-speaking children

    Infant Behav. Dev.

    (2019)
  • A. Perri

    A wolf in dog’s clothing: initial dog domestication and Pleistocene wolf variation

    J. Archaeol. Sci.

    (2016)
  • J.W. Pilley et al.

    Border collie comprehends object names as verbal referents

    Behav. Process.

    (2011)
  • J.W. Pilley

    Border collie comprehends sentences containing a prepositional object, verb, and direct object

    Learn. Motiv.

    (2013)
  • D. Ramos et al.

    Limitations in the learning of verbal content by dogs during the training of OBJECT and ACTION commands

    J. Vet. Behav.

    (2019)
  • C. Santolin et al.

    Constraints on statistical learning across species

    Trends Cogn. Sci.

    (2018)
  • P. Wiener et al.

    Use of questionnaire-based data to assess dog personality

    J. Vet. Behav.

    (2016)
  • A. Benjamin et al.

    ‘Who’s a good boy?!’ Dogs prefer naturalistic dog-directed speech

    Anim. Cogn.

    (2018)
  • P. Bloom

    Can a dog learn a word?

    Science

    (2004)
  • R.B. Cattell

    The scree test for the number of factors

    Multivar. Behav. Res.

    (1966)
  • J. Delanoeije et al.

    Do dogs mind the dots? Investigating domestic dogs' (Canis familiaris) preferential looking at human‐shaped point‐light figures

    Ethology

    (2020)
  • L. Fenson et al.

    Variability in early communicative development

    Monogr. Soc. Res. Child Dev.

    (1994)
  • L. Fenson et al.

    MacArthur-Bates Communicative Development Inventories

    Baltimore, MD

    (2007)
  • A. Fernald et al.

    Rapid gains in speed of verbal processing by infants in the 2nd year

    Psychol. Sci.

    (1998)
  • C. Fugazza et al.

    Word learning dogs (Canis familiaris) provide an animal model for studying exceptional performance

    Sci. Rep.

    (2021)
  • M. Fukuzawa et al.

    More than just a word: Non-semantic command variables affect obedience in the domestic dog (Canis familiaris)

    Appl. Anim. Behav. Sci.

    (2004)
  • 1

    https://orcid.org/0000-0002-8920-0417

    View full text