Six- but not three-month-old infants follow the gaze of a humanoid robotic talking head

• In Event: Seeing is Believing: Infants' and Toddlers' Social Learning from Visual Cues

Abstract

Gaze direction reveals what other people look at, and hence learning to follow others’ gaze plays a major role in referential communication and infant language development (Senju & Johnson, 2009; Mundy et al., 2007). Infants are first found to follow adults’ gaze around 6 months of age, when preceded by ostensive cues such as direct eye contact or infant-directed speech (Senju & Csibra, 2008). However, infants do not follow only the gaze of humans, but also of other intentional agents such as robots that display face-like features and contingent reactivity (Johnson, Slaughter, & Carey, 1998). Previous screen-based studies comparing human vs. robot gaze following in infants have shown that 12-months-olds follow the gaze of both human and robotic agents, but that they only show subsequent preferential looking to the objects gazed by humans (Okumura et al., 2013; 2015). In the present study, we explore gaze following skills and subsequent preferential-looking and manual choice to the target object in a real contingent robot-infant interaction at the younger age of 3 and 6 months. In an effort to enhance the naturalistic nature of the interaction, we opted for a “Furhat robot”, designed to seem human-like and act as a social agent (i.e., a 3D lamp avatar with a back-projected face texture, that orients towards the infant and speaks naturally, escaping the Mona Lisa effect: Al Moubayed, Edlund, & Beskow, 2012). Infants (N=42) sat on their parents’ laps, facing the robot placed on a table with similar-sized toys at each side (see Figure 1). In the first phase, the robot made eye contact with the infant, oriented towards her, greeted in infant-directed speech, and silently gazed at the target object for 5 seconds before gazing back at the infant (6 trials, ABABBA side counterbalancing). The second phase was a preferential looking test to the two objects in the absence of the robot (30 sec), and the third phase was a manual choice test (experimenter prompting the child to grab one of the two objects). Results from the 3-month-old groups showed no evidence of gaze following in First Look (β = -0.16, z = -0.86, p > .1) or Looking time measures (β = 262, t = 0.81, p > .1, see Figure 2). At 6 months of age, however, infants displayed evidence of significant gaze following in both measures (β = 0.93, z = -4.77, p < .01; β = -952, t = -3.28, p < .01, respectively). The 30-sec test phase yielded no significant differences in either group (ps > .1), however 6 months tended to pick the target object more often than distractor (p = .05). These results suggest that human infants can follow the gaze of a social robot at 6 months of age, similar to prior screen-based studies in human gaze following (Senju & Csibra, 2008). These results lay a solid foundation for follow-up studies that further explore more complex infant-robot language learning interactions.

Authors

• Joan Birules, University of Grenoble

• Cristina Ioana Galusca, Centre National de la Recherche Scientifique (CNRS)

• Jérémie Raymond Josse, Université Grenoble Alpes
  Presenting Author

• Olivier Pascalis, Centre National de la Recherche Scientifique (CNRS)

• Mathilde Fort, Université Claude Bernard Lyon 1