Domestic dogs respond correctly to verbal cues issued by an artificial agent

Human-canine communication technology for the home-alone
domestic dog is in its infancy. Many criteria need to be fulfilled in
order for successful communication to be achieved remotely via artificial
agents. Notably, the dogs' capacity for correct behavioural responses to
unimodal verbal cues is of primary consideration. Previous studies of
verbal cues given to dogs alone in the test room have revealed a
deterioration in correct behavioural responses in the absence of a source
of attentional focus and reward. The present study demonstrates the
ability of domestic pet dogs to respond correctly to an artificial agent.
Positioned at average human eye level to replicate typical human-dog
interaction, the agent issues a recall sound followed by two prerecorded,
owner spoken verbal cues known to each dog, and dispenses food
rewards for correct behavioural responses. The agent was used to elicit
behavioural responses in three test conditions; owner and experimenter
present; experimenter present; and dog alone in the test room. During the
fourth (baseline) condition, the same cues were given in person by the
owner of each dog. The experiments comprised a familiarisation phase
followed by a test phase of the four conditions, using a counterbalanced
design. Data recorded included latency to correct response, number of
errors before correct response given and behavioural welfare indicators
during agent interaction. In all four conditions, at least 16/20 dogs
performed the correct recall, cue 1 response, and cue 2 response
sequence; there were no significant differences in the number of dogs who
responded correctly to the sequence between the four conditions (p =
0.972). The order of test conditions had no effect on the dogs'
performances (p = 0.675). Significantly shorter response times were
observed when cues were given in person than from the agent (p = 0.001).
Behavioural indicators of poor welfare recorded were in response to
owners leaving the test room, rather than as a direct result of agent
interaction. Dogs left alone in the test room approached and responded
correctly to verbal cues issued from an artificial agent, where rapid
generalisation of learned behaviours and adjustment to the condition was
achieved.