- Email: [email protected]
Do infants perceive the social robot Keepon as a communicative partner?
G Model
ARTICLE IN PRESS
INFBEH-1051; No. of Pages 11
Infant Behavior & Development xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Infant Behavior and Development
Full length article
Do infants perceive the social robot Keepon as a communicative partner? Andreea Peca a,∗ , Ramona Simut b , Hoang-Long Cao c , Bram Vanderborght c a b c
Babes¸-Bolyai University, Department of Clinical Psychology and Psychotherapy, Cluj-Napoca, Romania Vrije Universiteit Brussel, Clinical and Life Span Psychology Department, Brussels, Belgium Vrije Universiteit Brussel, Robotics and Multibody Mechanics Research Group, Brussels, Belgium
a r t i c l e
i n f o
Article history: Received 7 April 2015 Received in revised form 27 September 2015 Accepted 3 October 2015 Available online xxx Keywords: Human–robot interaction Contingency Agency Turn-taking Conversation Infants
a b s t r a c t This study investigates if infants perceive an unfamiliar agent, such as the robot Keepon, as a social agent after observing an interaction between the robot and a human adult. 23 infants, aged 9–17 month, were exposed, in a first phase, to either a contingent interaction between the active robot and an active human adult, or to an interaction between an active human adult and the non-active robot, followed by a second phase, in which infants were offered the opportunity to initiate a turn-taking interaction with Keepon. The measured variables were: (1) the number of social initiations the infant directed toward the robot, and (2) the number of anticipatory orientations of attention to the agent that follows in the conversation. The results indicate a significant higher level of initiations in the interactive robot condition compared to the non-active robot condition, while the difference between the frequencies of anticipations of turn-taking behaviors was not significant. © 2015 Published by Elsevier Inc.
1. Introduction In the last decades, there has been an increased research interest related to the infants’ attunement to their social environment and the development of the behavioral reciprocity of the infant and their primary caregiver in the first years of life. These studies have important implications for our understanding of the early social development and the development of the self. Children develop the capability for social communication through physical and social interactions with their caregivers, from the moment of birth. There is a lot of evidence that infants are capable to mimic orofacial actions at birth (Meltzoff & Moore, 1989; Nadel & Butterworth, 1999; Meltzoff, 2005), and recognize the prosodic features of their mothers’ voices (DeCasper & Fifer, 1980; Fernald & Mazzie, 1991); in the first two months of life, the temporal structure of the rhythmic turn-taking originates mainly from the caregiver’s reading the child’s response pattern, while the child establishes eye-contact with the caregiver, along with exchanges of voice and facial expressions (Kozima & Nakagawa, 2006). At 2–3 month, infants are already aware of their mother contingent and emotional appropriate behavior and actively engage with it, according to the results obtained with the still face and the double television replay procedures (Trevarthen, 1993a; Trevarthen, 1993b; Tronick, 1989). Longitudinal studies focused on observing the transformations that appear during the first year of life indicate, at the middle of the first year, “an increased intricate, precise and selective coordination of the infant, with the mother’s richly inflected, rhythmically patterned, and repetitive expressions of communication”
∗ Corresponding author. Tel.: +40 742019045; fax: +40 264434141. E-mail address: [email protected] (A. Peca). http://dx.doi.org/10.1016/j.infbeh.2015.10.005 0163-6383/© 2015 Published by Elsevier Inc.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
2
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
(Trevarthen & Aitken, 2001). During this period, the child gradually learns to predict the caregiver’s behavior, which makes the interaction more and more symmetric (Kozima & Nakagawa, 2006). By the end of the first year, before the emergence of words, the human infant can communicate efficiently by using conventional vocalizations and gestures (Trevarthen & Aitken, 2001). However, what is the role of these fine-tuned abilities to coordinate with another human being? Are these the prerequisites for more sophisticated social abilities? Meltzoff (1990), Meltzoff and Moore (1995), Meltzoff and Brooks (2001) and Meltzoff (2002a) argues that imitation is the test infants use to discriminate between an intentional and an unintentional agent. The innate equipment for imitation and imitation recognition represents a neural mechanism for recognising “congruent with me”, not just “contingent on me”. The temporal contingency is not a sufficient condition, the congruence is also necessary for infants to recognise the object as a “like-me” entity and attribute a “mind” to that entity. This theory has received a growing consensus from different disciplines, from evolutionary psychology to neuroscience (e.g. Gordon, 1995; Tomasello, 1999; Rizzolatti, Fogassi, & Gallese, 2002). Developmental psychology has also addressed the issue of attributing mental states to objects. A few studies suggest that infants attribute mental states only to humans (Field et al., 1982; Legerstee, 1991; Meltzoff, 1995, cited in Arita, Hiraki, Kanda, & Ishiguro, 2005). The results of these studies are questionable in the light of several facts: the object used in these experiments were objects for which infants had already formed expectations, the objects had few human-like features and motion (Legerstee, 1994, 1997, 2001, cited in Arita et al., 2005). On the other hand, several studies found evidence that infants attribute mental states to non-human objects that appear to interact with persons: a box-shaped machine that beeped and flashed lights, or a small fur ball that was making noises and flashed lights (Movellan et al., 1987; Johnson et al., 1999, 2001, cited in Arita et al., 2005). Another interesting study is the one of Arita et al., 2005, who used the looking-time paradigm (Legerstee, Barna, & DiAdamo, 2000) to investigate whether 10-month old, expected people to talk to a humanoid robot, obtained similar results. They found that infants only perceived the interactive robot as a communicative agent and the non-interactive robot (either active or stationary) as an object. Moreover, the results of Johnson and Shimizu (2004) suggest that by the end of the first year of life, infants have a broad and complex definition of what counts as an agent and attribute goals even to non-human and unfamiliar agents. These results imply that interactivity between humans and objects is the key factor in mental attribution. The evidence suggesting that the congruence and contingency of actions may play an important role in the perception and attribution of intentionality, has led to a growing interest in endowing robots with the ability to imitate. One important aspect of contingency is the temporal proximity between the user’s behavior and the robot’s response. In natural conversation, communication partners tend to respond to each other’s linguistic behavior in a time frame of approximately 200–300 ms (Sidnell & Enfield, 2012). Another important aspect refers to the synchrony between movements and sounds. When speaking to infants, parents perform movements that are in a tight temporal synchrony with the speech (Gogate, Bahrick, & Watson, 2010) and are shorter, which results in less roundness and more pauses between the individual segments (Brand et al., 2002; Rohlfing et al., 2006, cited in Fischer et al., 2013). Movellan and Watson (2002) investigated whether infants follow the line of regard of a non-human robot with an abstract pattern that did not resemble the structure of a human face. They found that infants follow the line of regard when the robot responded contingently to different events in the external environment and to the infants’ initiations. In another study, Meltzoff, Brooks, Shon, and Rao (2010) found that infants use the information derived from an entity’s interactions with other agents as evidence about whether that entity is a perceiver. The infants who saw the robot interacting contingently with a human adult were more likely to follow the line of regard of the robot. Other studies have investigated the preference infants have for a contingent or noncontingent interaction. The results of Sidner, Lee, Kidd, Lesh, and Rich (2006) suggest that participants interacted longer with a penguin-shaped robot that was producing contingent, non-verbal feedback, judging it as being more reliable and its movements more appropriate. Similar results were obtained by Kose-Bagci, Dautenhahn, Syrdalm, and Nehaniv (2010), participants showing a preference for a contingency pattern whose temporal dynamics was closer to human–human conversations. These results have important implications for the development of the social robotics domain. While some studies suggest that interactivity is the essential condition for the attribution of intentionality, and other studies suggest that contingency is necessary, another line of studies suggest that both the contingency and the congruence conditions should be met and that the robot should be designed in a way that allows it to be involved in “like-me” actions. On the other hand, robots represent as “an ideal opportunity to study cognitive and social development in infants” (Scasselatti, 1998, 2000; Deák, Fasel, and Movellan, 2001). By creating robots that have different morphological characteristics and by programming them to exhibit precisely controlled contingency structures, researchers can test the strategies infants use to identify their nature and capabilities (Movellan & Watson, 2002). The questions that should be addressed in this approach are: Which are the requirements that robots should meet in order to be treated by infants as conversational agents? Is animacy a sufficient condition? Is contingency absolutely necessary? What about similarity? Do infants need similarity in order to find a correspondence between their actions and the actions of the robot? This study aims to investigate if infants use an unknown agent, the robot Keepon, as a communicative partner after observing a brief conversation between a human adult and the robot. More specifically, our study aims to investigate if the interactivity of the robot Keepon is a sufficient condition for perceiving it as a communicative partner. Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
3
2. Objectives of the present study In this study, we investigate the following research questions:
(1) Do infants manifest more initiations in the interactive robot condition compared with the non-active robot condition? (2) Do infants manifest more anticipation of turn-taking behaviors in the interactive robot condition compared with the non-active robot condition?
3. Experimental platform and system design The experimental test-bed used in this study is the robot MyKeepon (see Fig. 1) a simplified version of the Keepon robot developed by Kozima, Nakagawa, and Yano (2004). While Keepon is a complex research version with complex abilities, the MyKeepon version has limited abilities and cannot be controlled by a PC, while the appearance is identical. Keepon is a small (12 cm) creature-like robot designed for simple, natural, nonverbal and playful interaction with children. It has a simple (like a yellow snowman) and soft (made of silicone rubber) appearance (Kozima et al., 2004). Despite its simplicity, Keepon is capable of displaying two psychological characteristics which are essential for the attribution of intentionality: attention (which suggests that the robot can hear and see the world), and emotion (which suggests that the robot has a “mind” to evaluate what it perceives) (Kozima, Michalowski, & Nakagawa, 2009). Moreover, Keepon’s repertoire of sounds and motions follow some of the requirements for child-directed speech: the pitch variation is high, the utterances are short, the pauses are long, the speech rate is low (Fernald et al., 1989). This pattern of speech seems to facilitate language acquisition, to modulate attention and arousal levels and to communicate affect to the infant (Fernald et al., 1989). Another strength of Keepon is its high speed of execution, an essential ability for achieving a high level of temporal contingency between the acts of the child and those of the robot, a condition for our study. Last, but not least, the low price of the MyKeepon version makes it available for any research group or pedagogical institution. For our study, we have used an enhanced version of the commercial version, which was controlled by an operator through a computer (Cao et al., 2014). The interface used to control the robot (see Fig. 2) is intuitive and therefore, easy to use. In the exposure phase, the movements and sounds of the robot were part of a pre-programmed scenario that was started when the human operator pushed the start scenario button (see Table 1). In the testing phase, the computer operator was trained to correctly identify the initiations of the child and manually push the buttons that encoded the specific movements and sounds in the order they were displayed on the interface. In order to facilitate future research with this robot platform, the manual how to modify the hardware, the electronics and software is made open source available on http://probo.vub.ac.be/HackingKeepon/.
Fig. 1. The MyKeepon robot.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model
ARTICLE IN PRESS
INFBEH-1051; No. of Pages 11
4
A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
Fig. 2. The operator interface program ViKeepon.
Table 1 The scenario for the interactive robot condition. 00:02.000 H: Hi Keepon! 00:03.000 K: Up wakeup (sound) 00:03.500 K: Jump (motion) 00:03.750 K: Jump (motion) 00:04.500 H: I am glad to see you! 00:06.500 K: Initial Boot (sound) 00:06.750 K: Bow (motion) 00:08.000 H: How are you? 00:10.500 K: Sigh (sound) 00:10.750 K: Bend (motion) 00:12.000 H: Oh, but why are you sad? 00:14.500 K: Sneeze up (sound) 00:14.750 K: Jump (motion) 00:16.000 H: Oh, you got a cold! 00:18.000 K: Sneeze up (sound) 00:18.250 K: Jump (motion) 00:19.000 H: You have been trodding through puddles again 00:22.500 K: Yawn down (sound) 00:22.750 K: Bow (motion) 00:23.500 H: What did your mommy say when you came home so wet? 00:27.000 K: Say no (motion) 00:27.500 K: Whine (sound) 00:28.000 H: Keepon, can you do this? 00:32.500 K: Bend (motion) 00:32.750 K: Sigh (sound) 00:33.500 H: Well done! What about this? 00:37.500 K: Bow (motion) 00:37.750 K: Sigh (sound) 00:39.000 H: Keepon, do you want me to bring you jellies next time? 00:45.000 K: Jump (motion) 00:45.250 K: Initial boot (sound) 00:46.000 H: Ok, I will bring you many colorful jellies! 00:50.000 K: Jump (motion) 00:50.250 K: Yawn up (sound) 00:51.500 H: I enjoyed playing with you! 00:53.500 K: Yawn down (sound) 00:53.750 K: Jump (motion) 00:55.000 H: Bye, Keepon! 00:56.500 K: Head hit beep (sound) 00:56.700 K: Head hit beep (sound) 00:56.750 K: Bow (motion)
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model
ARTICLE IN PRESS
INFBEH-1051; No. of Pages 11
A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
5
4. Method 4.1. Participants The participants for this study were 22 children, including 14 boys and 8 girls, aged 9–18 months, (M = 13.68, Sd = 2.67). All children were full-term, had normal birth weight and had no developmental concerns according to parental report. Three children, two boys and one girl, had to be excluded due to intense fussiness. The Bayley Scales of Infant Development (Second Edition, BSID-II) were administered by two experienced clinical psychologists that had demonstrated reliability in scoring (see Table 2). The Bayley score index suggested a typical development for all infants that participated to the study. The infants were exposed to both a contingent and a noncontingent interaction between a human adult and the social robot Keepon. The order of the presentation of the two conditions was randomised. 10 infants started with the interactive robot condition, followed by the non-active robot condition, and 9 children started with the nonactive robot condition, followed by the interactive robot condition. Of the 19 infants (12 boys and 7 girls), 6 infants refused one of the two experimental conditions (3 children refused the non-active robot condition when presented second time and 1 child refused the non-active robot condition when presented for the first time, and 2 children refused the interactive robot condition when presented second time). All participants signed a parental consent form before the study started. 4.2. Setting Each child was tested in a room (4 × 5 m) while seated in his/her parent’s lap, in front of a table (77 × 155 cm). In the exposure phase, at 100 cm distance away from the face of the child, on the other part of the table, the robot and the experimenter were facing each other, so the child saw the robot from the lateral point of view. In the testing phase, the robot faced the child and the same distance was maintained (100 cm) (see Figs. 3 and 4). Two digital cameras recorded the interaction; a frontal camera recorded the child’s upper torso and face, and a lateral one captured the child, the robot and the experimenter in the exposure phase, and the child and the robot in the testing phase. 4.3. Procedure 4.3.1. Habituation with the environment The experimenter, the child and one of the parents entered the experimental room. An assistant of the experimenter, the robot operator, was also in the room. The experimenter introduced some toys to the child (bubbles, a ball, a car, a doll, a small book with animals) and the child explored the environment and interacted with the toys, while the experimenter was chatting with the parent. A grey cardboard screen (29.7 × 42 cm) hid the robot during this time. After a 10 to 15 min interval, when the parent confirmed that the child was relaxed and acclimated to the new environment, the experimenter’s assistant uncovered the robot Keepon, the experimenter took his position on the chair in front of the robot, and the parent was asked to sit on a chair with the child on his/her lap. The experimental procedure involved two phases: the exposure to a scripted experimenter-robot conversation and a subsequent test in which the motor initiations and the vocalizations the child addressed to the robot were recorded. In the interactive robot condition, Keepon responded contingently, by a preprogrammed set of sounds and motions to the adult verbal initiations (see Table 1), while in the non-active robot condition, Keepon remained still. In the testing phase, Keepon responded by the same pre-programmed set of sounds and motions to any intentional vocalization or motor act of the child. The parent received the instruction to keep the verbal interaction to a minimum during the task. In the case the child asked for assurance, the parent had to acknowledge the child by nodding, saying something as “I see it” and redirect the attention to Keepon. We asked the parent to hold the child in his arms and restrain him from climbing the table. 4.3.2. Exposure phase: 4.3.2.1. Contingent interaction condition. The child was exposed to a 65 s conversation between the experimenter and the robot Keepon. While the experimenter was addressing Keepon verbally, Keepon responded contingently by a pre-programmed set of sounds and motions. The conversation consisted of 14 replies for each of the two partners (see Table 1). Table 2 The characteristics of the two randomised groups of children.
Chronological age Bayley score
Interactive robot/ non-active robot group
Non-active robot/ interactive robot group
P
N
M
SD
N
M
SD
P
10 10
13.45 99.8
1.18 11.33
9 9
13.83 96.44
3.38 13.62
.79 .57
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
6
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
Fig. 3. Room diagram for robot-infant interaction.
Fig. 4. An overview of the experimental task during the two conditions.
4.3.2.2. Non-active robot condition. In the non-active robot condition, the behavior of the experimenter did not differ compared to the interactive robot condition. The robot did not produce any sound or motion. The duration of the interaction was the same as in the interactive robot condition. At the end of the conversation, the experimenter repositioned Keepon, so that it faced the child, and told the child that Keepon is playing Peek-aboo, masked the robot with a grey screen and left the room. The robot was covered for 30 s, the time needed for the operator to start the ViKeepon program. Afterwards, the experimenter uncovered the robot and the testing phase began. 4.3.3. Testing phase The testing phase was identical for both the interactive robot condition and the non-active robot condition. In the beginning, Keepon initiated a motion accompanied by a short sound waiting for the child to respond. Keepon repeated the initiation every 20 s, for 120 s, the total duration of the testing phase. The role of these spontaneous initiations was to offer the opportunity to perceive Keepon as animated, to the children who did not produce any initiation in the first part of the testing phase. Keepon responded by producing sounds and movement as a feedback to the vocalizations and to the motor initiations of the child that were directed toward it. The repertoire and the order of motions and sounds produced by the robot was the same for all the children, as specified in the protocol (see Table 1). A snapshot of the two experimental phases is presented in Fig. 4. 4.3.4. Experimental design A 2 × 1 within subjects experimental design (interactive robot condition/non-active robot condition × Keepon partner) was used in this study. In the Exposure phase, the measured variables were the frequency of anticipatory orientations of attention toward the Keepon robot when the robot was next in the conversation. In the Testing phase, the measured variable was the frequency of total initiations produced by the child and directed toward the robot Keepon. The definitions of the dependent variables and the instructions for coding are presented in Table 3.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
7
Table 3 Definitions of the measured variables. Variable
Measure
Specifications
Anticipatory orientations of attention toward Keepon
Frequency
Total initiations of social interaction
Frequency
Score when the child switches attention from the adult to Keepon, after the adult has finished his remark, but exactly before Keepon starts to perform any action If the child maintains attention fixed to Keepon, do not score as an anticipatory behaviour If the child looks in any other direction besides the experimenter before switching to Keepon, do not score an anticipatory behavior Any motor or verbal behavior that is directed toward Keepon and is accompanied by visual contact The behaviors that are very subtle are not scored For some verbal behaviors that were not accompanied by visual contact, but seemed to be directed toward Keepon, a supplementary agreement between two independent coders was obtained Do not score prompted gestures
5. Data analysis 5.1. Data collection The two phases of the experiment, the exploratory phase and the testing phase, were presented in a standardised format. The exploratory phase lasted 64 s and involved 14 conversational replies for each of the agents (experimenter and the robot Keepon) in the interactive robot condition and 14 conversational replies for the experimenter and 14 pauses for the robot Keepon in the non-active robot condition. In the non-active robot condition, the experimenter offered his reply when this appeared on the computer screen after a pause that lasted the equivalent of the according reply in the interactive robot condition. The testing phase lasted 50 s for all the participants. A video coding analysis was performed in order to measure the target variables. The behaviors were manually coded from the videotapes using ELAN transcription software (see Fig. 4) (Sloetjes & Wittenburg, 2008). The video recordings were scored by an independent coder. For the coding of the initiation behaviors, the coder was blind to the infant’s experimental group, since in the testing phase, Keepon had an identical behavior in the two conditions. We assessed the scoring agreement by rescoring 50% of the recordings. We performed an inter-rater agreement analysis on the measured variables. We found a significant Pearson coefficient with a high magnitude for contingent initiations (r = .99, p < .001), noncontingent initiations (r = .98, p < .001), anticipations of turn taking in the interactive robot condition (r = .99, p < .001), and anticipations of turn taking in the non-active robot condition (r = .99, p < .001). 5.2. Statistical analysis 5.2.1. Frequency of initiations We used a Wilcoxon signed-rank test to analyze if there are any differences between the frequency of initiations in the interactive robot condition and in the non-active robot condition. The frequency of total initiations in the interactive robot condition (M = 9.68, Sd = 6.73) was significant higher than the frequency of total initiations in the non-active robot condition (M = 6.2667, Sd = 7.54), Z = −2.363, p = .018, with a medium effect size r = −.682 (see Fig. 5). We have controlled the order effect by randomising the order of the presentation of the interactive robot condition and the non-active robot condition. A Mann–Whitney test was performed in order to verify if the order of the condition influences the distributions of the measured variables. In addition, we performed the same analysis for the non-active robot condition. The Mann–Whitney test indicated that the frequency of the initiations recorded when the interactive robot condition was presented first (Mean rank = 12.06) was significantly higher than when it was presented after the non-active robot condition (Mean rank = 5.56), Z = −2.65, p = .006, r = −.618. On the other hand, the difference between the frequency of the initiations recorded when the non-active robot condition was presented first (Mean rank = 6.50) and when it was presented after the interactive robot condition (Mean rank = 8.25) was not significant, suggesting that the order effect did not influence the frequency of initiations in the non-active robot condition. 5.2.2. Frequency of anticipations of turn-taking The Wilcoxon signed-rank test revealed no significant differences between the frequency of anticipations of turn-takings in the interactive robot condition (M = 3.84, Sd = 2.79) and in the non-active robot condition (M = 4.89, Sd = 2.53), Z = −1.50, p = .133 (see Fig. 6). The Mann–Whitney test indicated that the frequency of anticipatory behaviors in the interactive robot condition presented first (Mean rank = 9.90), did not differ from the frequency of anticipatory behaviors when the interactive robot condition was presented after the non-active robot condition (Mean rank = 10.11), Z = −.083, p = .968. Also, the difference between the anticipatory behaviors in the non-active robot condition presented first (Mean rank = 9.22) and after the Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
8
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
Fig. 5. The graphical representation of the frequency of initiations in the interactive robot condition versus non-active robot condition.
Fig. 6. The graphical representation of the anticipations in the interactive robot condition versus non-active robot condition.
interactive robot condition (Mean rank = 10.70) was not significant too, Z = −.575, p = .604. These data suggest that the anticipatory behavior variable was not influenced by the order effect. 5.2.3. The age impact The Kruskal Wallis test was used to investigate if the age of the infants influenced the frequency of the measured variables. The results suggest that the distribution of both initiations and anticipations was the same across the whole age spectrum in both the interactive robot condition and the non-active robot condition (see Table 4). 5.2.4. Additional observations A series of additional observations that describe the children’s behavior during the task and are not captured by the quantitative analysis are presented below. First, the most of the participants seemed to enjoy the interaction with Keepon. From the 22 initial participants, only one was excluded due to fear of the robot. Taking into account the small age of the participants, this could be a good indicator that Keepon is a non-threatening interaction partner for infants. In the exposure phase, some infants expressed positive affect during the interactive robot condition, a phenomenon that was not observed in the non-active robot condition. In the testing phase, a tendency was observed for the younger infants lose interest for the Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
9
Table 4 The distribution of the dependent variables across the age spectrum. Null hypothesis
Test
Sig.
The distribution of initiations in the interactive robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.161
Kruskal–Wallis The distribution of initiations in the non-active robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.312
Kruskal–Wallis The distribution of anticipation behaviors in the interactive robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.505
Kruskal–Wallis The distribution of anticipation behaviors in the non-active robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.333
Kruskal–Wallis
robot in the non-active robot condition. Some infants recognized whether the robot was responsive to them very quickly, and engaged in the conversation. The most common initiations produced by the infants were: vocalizations, hitting the table, reaching out to Keepon, shaking head, laughing. While in the first seconds, some infants would produce an initiation, and after the response of the robot, would wait for some seconds before producing another one, in the last part of the interaction, the pauses between the initiations became shorter. This phenomenon may indicate that these infants were testing the robot capabilities in the first part, and then accepted the robot as a conversational partner in the second part of the interaction. 6. Conclusions and discussions The purpose of this study was to determine whether infants perceive the interactive robot Keepon as a communicative agent. The results show that infants use the cues from the interaction between the adult and the robot Keepon to make inferences regarding the social agency of Keepon. Therefore, after observing the contingent interaction between the adult and the robot Keepon, infants manifested more initiations in the testing condition compared to the infants exposed to the non-active robot condition. Congruent with the results of Johnson and Shimizu (2004), our results suggest that infants have a broad definition of what counts for an agent and can attribute goals even to simple creatures such as robot Keepon, when animacy is involved. Since our study did not include a fine control for different levels of contingency, we cannot argue that the attunement of the behavior of the robot and human adult was the factor responsible for the inference that the robot Keepon has a mind of its own and is capable of conversation. Future studies should investigate in more detail which are the factors that are responsible for triggering initiations in infants, initiations that become part of a genuine conversation between the two partners. The factors that are worth considering are: the simple animacy, the temporal contingency of the actions of the two agents, the synchrony between movement and sound, and the simplicity of the interaction (quick motions, short and attractive, pitched sounds similar to mothering). In the light of the studies performed by Meltzoff et al. (2010); Johnson and Shimizu, (2004) and Movellan and Watson (2002), it is highly probable that the critical factor involved in the attribution of intentionality to be the contingency between the robot and the human adult’s behavior. In all the mentioned studies, the authors have found that the motion itself could not account for the results across the groups. The real factor that carries a significant weight is not the motion itself, but the nature of the interaction between the two agents. Another question raised by our results is if the measured variable, the infants’ motor and verbal initiations could account for something different than the attribution of intentionality in the robot Keepon. Is it possible that infants were just imitating the behavior observed in the human adult in the exposure phase? Since the human adult behaved identically in the robot interaction condition and in the non-active condition, this hypothesis can be eliminated. Therefore, we consider that the motor and verbal initiation variable is an innovative and reliable measure of intentionality detection in infants. Another result of our study is that there were no significant differences between the frequencies of anticipation of turntaking behaviors in the interactive robot condition compared with the non-active robot condition. This is contrary to our expectation that infants will manifest more anticipatory looking behaviors to the robot in the interactive robot condition. From our point of view this result may suggest that the interactive behavior of the human adult generated in infants the assumption that the robot Keepon is a mentalistic agent, therefore, they have anticipated a response from the robot. Even though our results are congruent with the one obtained by Arita et al. (2005), they interpret the result in a different way, by arguing that the looking time measure indicates the fact that infants did not expect the robot to be talked to by the person. An alternative explanation would be that by looking at the robot, the infants were following the attention of the adult; therefore Keepon might have functioned as a joint attention object for the adult and the infant. In a future study, we should find a more reliable measure of the expectations infants have from mentalistic agents. Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
10
One limitation of our study refers to using a repeated measures design, with infant being exposed to both the interactive robot condition and the non-active robot condition. The analysis of the ordering effect suggests that the frequency of initiations was influenced by the order of the presentation of the two conditions for the interactive robot condition. More specifically, there were more initiations recorded when the interactive robot condition was presented first, compared to the interactive robot condition presented as the second condition. This result suggests that although we have controlled the effect of the repeated exposure by counterbalancing the order of the two conditions, there are still limitations in using such a design. In a future improved version of this study, a design with independent groups will eliminate the bias of the order of the two conditions. As well as the limitations previously mentioned, the technical constraints of the robot could also represent a limitation of the study. Because of its shared autonomy, errors by the operator could influence research outcomes, thus making the robot incapable of adapting to children’s behaviors at this point in time. Future work will consist of developing supervised autonomous interaction, so that an operator is no longer required (Thill, Pop, Belpaeme, Ziemke, & Vanderborght, 2013). Finally yet importantly, the qualitative observations from our study suggest that MyKeepon modification is a useful platform for investigating HRI in infants, since MyKeepon seems to be a non-threatening and attractive interaction partner for this age group. Moreover, the low-cost of the platform represents an additional argument for a long-term use in studies on the effects of specific factors infants use to infer the existence of a mind in a non-human agent. Concluding, our study suggests that the social robot Keepon is an appropriate social interaction partner for infants, a partner that is perceived as a conversational agent when offering communicative cues. More in depth work is needed in order to understand exactly the attributes that make infants assume Keepon is a social agent. Our proposal, inspired by the work of Meltzoff et al. (2010), and Arita et al. (2005), is to include, in a future study, both an interactive and noncontingent robot condition, and an active robot condition, in which the robot behaves like a human, and the person is stationary and silent. Such an experimental design would facilitate the distinction between the effect of animacy and the effect of contingency. This study opens the door and offers incipient answers to important questions regarding the future of HRI domain. When social robots will be part of our lives, will people regard these artificial objects as mentalistic agents? If so, which are the conditions that a robot should respond to, in terms of interaction dynamics, morphology and motion characteristics? This study comes in line with some other studies suggesting that interactivity is an important aspect in the attribution of intentionality. Moreover, our results suggest that morphological similarity is not absolutely necessary for the attribution of intentionality. Acknowledgements The authors thank the children that participated in the study and their parents which made this study possible. This work has been supported by the CNCSIS-Bucharest, Romania project PN-II-IDPCE-2011-3-0484-Exploring Robot-assisted therapy for children with ASD and the European FP7 project DREAM (grant no. 611391). Appendix A. Appendix Table 1 References Arita, A., Hiraki, K., Kanda, T., & Ishiguro, H. (2005). Can we talk to robots? Ten-month-old infants expected interactive humanoid robots to be talked to by persons. Cognition: 95., B49–B57. Cao, H. L., Van De Perre, G., Simut, R., Pop, C., Peca, A., Lefeber, D., et al. (2014). Enhancing MyKeepon robot: A simple and low-cost solution for robot platform in human–robot interaction studies. Proceedings of the IEEE International Symposium on Robot and Human Interactive Communication, 555–560 (RO-MAN) Deák, G. O., Fasel, I. R., & Movellan, J. R. (2001). The emergence of shared attention: Using robots to test developmental theories. In C. Balkenius (Ed.), Proceedings of the first international workshop on epi-genetic robotics: Modelling cognitive development (pp. 95–104). Lund, Sweden: Lund University Cognitive Studies. DeCasper, A. J., & Fifer, W. P. (1980). On human bonding: Newborns prefer their mothers’ voices. Science: 208., (4448), 1174–1176. Fischer, K., Lohan, K., Saunders, J., Nehaniv, C., Wrede, B., & Rohlfing, K. (2013). The impact of the contingency of robot feedback on HRI. In International conference on collaboration technologies and systems San Diego, CA, (pp. 210–217). http://dx.doi.org/10.1109/CTS.2013.6567231 Fernald, A., Taeschner, T., Dunn, J., Papousek, M., Boysson-Bardies, B., & Fukui, I. (1989). A cross-language study of prosodic modifications in mothers and fathers’ speech to preverbal infants. Journal of Child Language: 16., 477–501. Fernald, A., & Mazzie, C. (1991). Prosody and focus in speech to infants and adults. Developmental Psychology: Vol. 27., (2), 209–221. Gogate, L. J., Bahrick, L. E., & Watson, J. (2010). A study of multimodal motherese: The role of temporal synchrony between verbal labels and gestures. Child Development: 71., 878–894. Gordon, R. M. (1995). Simulation without introspection or inference from me to you. In M. Davies, & T. Stone (Eds.), Mental simulation (pp. 53–67). Oxford: Blackwell. Johnson, S. C., & Shimizu, I. A. (2004). Infants’ attribution of a goal to a morphologically unfamiliar agent. Developmental Science: 7., (4), 425–430. Kose-Bagci, H., Dautenhahn, K., Syrdalm, D. S., & Nehaniv, C. L. (2010). Drum-mate: Interaction dynamics and gestures in human-humanoid drumming experiments. Connection Science: 22., (2), 103–134. Kozima, H., Nakagawa, C., & Yano, H. (2004). Can a robot empathise with people? Artificial Life and Robotics: 8., (1), 83–88. Kozima, H., & Nakagawa, C. (2006). Interactive robots as facilitators of children’s social development. Mobile Robots Towards New Applications, 314–315. Source: Mobile Robots Towards New Applications, ISBN 3-86611-314-5, Edited by Aleksandar Lazinica, pp. 784, ARS/plV, Germany, December 2006.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
11
Kozima, H., Michalowski, M. P., & Nakagawa, C. (2009). Keepon a playful robot for research, therapy, and entertainment. International Journal of Social Robotics: 1., 3–18. Legerstee, M., Barna, J., & DiAdamo, C. (2000). Precursors to the development of intention at 6 months: Understanding people and their actions. Developmental Psychology: 36., 627–634. Meltzoff, A. N., & Moore, K. M. (1989). Imitation in newborn infants: Exploring the range of gestures imitated and the underlying mechanisms. Developmental Psychology: Vol. 25., (6), 954–962. Meltzoff, A. N. (1990). Foundations for developing a concept of self: The role of imitation in relating self to other and the value of social mirroring, social modeling, and self practice in infancy. In D. Cicchetti, & M. Beeghly (Eds.), The self in transition: Infancy to childhood (pp. 139–164). Chicago, IL: University of Chicago Press. Meltzoff, A. N., & Moore, M. K. (1995). Infants’ understanding of people and things: From body imitation to folk psychology. In J. L. Bermudez, & N. A. M. Eilan (Eds.), The body and the self (pp. 43–69). Cambridge, MA: MIT Press. Meltzoff, A. N., & Brooks, R. (2001). “Like me” as a building block for understanding other minds: Bodily acts, attention, and intention. In B. F. Malle, L. J. Moses, & D. A. Baldwin (Eds.), Intentions and intentionality: Foundations of social cognition (pp. 171–191). Cambridge, MA: MIT Press. Meltzoff, A. N. (2002). Elements of a developmental theory of imitation. In A. N. Meltzoff, & W. Prinz (Eds.), The imitative mind: Development, evolution, and brain bases (pp. 19–41). Cambridge: Cambridge University Press. Meltzoff, A. N. (2005). Imitation and other minds: The “Like Me” hypothesis. In Perspectives on imitation: From neuroscience to social sciences. pp. 55–77. Cambridge, MA: MIT Press 2. Meltzoff, A. N., Brooks, R., Shon, A. P., & Rao, R. P. N. (2010). “Social” robots are psychological agents for infants: A test of gaze following. Neural Networks: 23., 966–972. Movellan, J. R., & Watson, J. S. J. S. (2002). The development of gaze following as a Bayesian systems identification problem. In Proceedings of the international conference on development and learning (ICDL02) IEEE. Nadel, J., & Butterworth, G. (1999). Imitation in infancy. Cambridge, UK: Cambridge University Press. Rizzolatti, G., Fogassi, L., & Gallese, V. (2002). Motor and cognitive functions of the ventral premotor cortex. Current Opinion in Neurobiology: 12., 149–154. Scasselatti, B. (1998). Building behaviors developmentally: A new formalism. In Proceedings of the AAAI spring symposium on integrating robotics research. Scasselatti, B. (2000). How robotics and developmental psychology complement each other. In NSF/DARP a workshop on development and learning. Sidnell, J., & Enfield, N. J. (2012). Language diversity and social action: A third locus of linguistic relativity. Current Anthropology: 53., 302–333. Sidner, C. L., Lee, C. L., Kidd, C., Lesh, N., & Rich, C. (2006). Explorations in engagement for humans and robots. Artificial Intelligence: 166., (1–2), 140–164. Sloetjes, H., & Wittenburg, P. (2008). Annotation by category—ELAN and ISO DCR. In Proceedings of the 6th international conference on language resources and evaluation LREC, 2008. Thill, S., Pop, C. A., Belpaeme, T., Ziemke, T., & Vanderborght, B. (2013). Robot-assisted therapy for autism spectrum disorders with (partially) autonomous control: Challenges and outlook. Paladyn Journal of Behavioral Robotics: 3., (4), 209–217. Tomasello, M. (1999). The cultural origins of human cognition. Cambridge, Massachusetts, London, England: Harvard University Press. Trevarthen, C. (1993a). The function of emotions in early infant communication and development. In J. Nadel, & L. Camaioni (Eds.), New perspectives in early communicative development (pp. 48–81). London: Routledge. Trevarthen, C. (1993b). The self born in intersubjectivity: The psychology of an infant communicating. In U. Neisser (Ed.), The perceived self: Ecological and interpersonal sources of self-knowledge (pp. 121–173). New York, NY: Cambridge University Press. Trevarthen, C., & Aitken, K. J. (2001). Infant intersubjectivity: Research, theory, and clinical applications. Journal of Child Psychology and Psychiatry: 42., (1), 3–48. Tronick, E. Z. (1989). Emotions and emotional communication in infants. American Psychologist: 44., 112–119.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
ARTICLE IN PRESS
INFBEH-1051; No. of Pages 11
Infant Behavior & Development xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Infant Behavior and Development
Full length article
Do infants perceive the social robot Keepon as a communicative partner? Andreea Peca a,∗ , Ramona Simut b , Hoang-Long Cao c , Bram Vanderborght c a b c
Babes¸-Bolyai University, Department of Clinical Psychology and Psychotherapy, Cluj-Napoca, Romania Vrije Universiteit Brussel, Clinical and Life Span Psychology Department, Brussels, Belgium Vrije Universiteit Brussel, Robotics and Multibody Mechanics Research Group, Brussels, Belgium
a r t i c l e
i n f o
Article history: Received 7 April 2015 Received in revised form 27 September 2015 Accepted 3 October 2015 Available online xxx Keywords: Human–robot interaction Contingency Agency Turn-taking Conversation Infants
a b s t r a c t This study investigates if infants perceive an unfamiliar agent, such as the robot Keepon, as a social agent after observing an interaction between the robot and a human adult. 23 infants, aged 9–17 month, were exposed, in a first phase, to either a contingent interaction between the active robot and an active human adult, or to an interaction between an active human adult and the non-active robot, followed by a second phase, in which infants were offered the opportunity to initiate a turn-taking interaction with Keepon. The measured variables were: (1) the number of social initiations the infant directed toward the robot, and (2) the number of anticipatory orientations of attention to the agent that follows in the conversation. The results indicate a significant higher level of initiations in the interactive robot condition compared to the non-active robot condition, while the difference between the frequencies of anticipations of turn-taking behaviors was not significant. © 2015 Published by Elsevier Inc.
1. Introduction In the last decades, there has been an increased research interest related to the infants’ attunement to their social environment and the development of the behavioral reciprocity of the infant and their primary caregiver in the first years of life. These studies have important implications for our understanding of the early social development and the development of the self. Children develop the capability for social communication through physical and social interactions with their caregivers, from the moment of birth. There is a lot of evidence that infants are capable to mimic orofacial actions at birth (Meltzoff & Moore, 1989; Nadel & Butterworth, 1999; Meltzoff, 2005), and recognize the prosodic features of their mothers’ voices (DeCasper & Fifer, 1980; Fernald & Mazzie, 1991); in the first two months of life, the temporal structure of the rhythmic turn-taking originates mainly from the caregiver’s reading the child’s response pattern, while the child establishes eye-contact with the caregiver, along with exchanges of voice and facial expressions (Kozima & Nakagawa, 2006). At 2–3 month, infants are already aware of their mother contingent and emotional appropriate behavior and actively engage with it, according to the results obtained with the still face and the double television replay procedures (Trevarthen, 1993a; Trevarthen, 1993b; Tronick, 1989). Longitudinal studies focused on observing the transformations that appear during the first year of life indicate, at the middle of the first year, “an increased intricate, precise and selective coordination of the infant, with the mother’s richly inflected, rhythmically patterned, and repetitive expressions of communication”
∗ Corresponding author. Tel.: +40 742019045; fax: +40 264434141. E-mail address: [email protected] (A. Peca). http://dx.doi.org/10.1016/j.infbeh.2015.10.005 0163-6383/© 2015 Published by Elsevier Inc.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
2
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
(Trevarthen & Aitken, 2001). During this period, the child gradually learns to predict the caregiver’s behavior, which makes the interaction more and more symmetric (Kozima & Nakagawa, 2006). By the end of the first year, before the emergence of words, the human infant can communicate efficiently by using conventional vocalizations and gestures (Trevarthen & Aitken, 2001). However, what is the role of these fine-tuned abilities to coordinate with another human being? Are these the prerequisites for more sophisticated social abilities? Meltzoff (1990), Meltzoff and Moore (1995), Meltzoff and Brooks (2001) and Meltzoff (2002a) argues that imitation is the test infants use to discriminate between an intentional and an unintentional agent. The innate equipment for imitation and imitation recognition represents a neural mechanism for recognising “congruent with me”, not just “contingent on me”. The temporal contingency is not a sufficient condition, the congruence is also necessary for infants to recognise the object as a “like-me” entity and attribute a “mind” to that entity. This theory has received a growing consensus from different disciplines, from evolutionary psychology to neuroscience (e.g. Gordon, 1995; Tomasello, 1999; Rizzolatti, Fogassi, & Gallese, 2002). Developmental psychology has also addressed the issue of attributing mental states to objects. A few studies suggest that infants attribute mental states only to humans (Field et al., 1982; Legerstee, 1991; Meltzoff, 1995, cited in Arita, Hiraki, Kanda, & Ishiguro, 2005). The results of these studies are questionable in the light of several facts: the object used in these experiments were objects for which infants had already formed expectations, the objects had few human-like features and motion (Legerstee, 1994, 1997, 2001, cited in Arita et al., 2005). On the other hand, several studies found evidence that infants attribute mental states to non-human objects that appear to interact with persons: a box-shaped machine that beeped and flashed lights, or a small fur ball that was making noises and flashed lights (Movellan et al., 1987; Johnson et al., 1999, 2001, cited in Arita et al., 2005). Another interesting study is the one of Arita et al., 2005, who used the looking-time paradigm (Legerstee, Barna, & DiAdamo, 2000) to investigate whether 10-month old, expected people to talk to a humanoid robot, obtained similar results. They found that infants only perceived the interactive robot as a communicative agent and the non-interactive robot (either active or stationary) as an object. Moreover, the results of Johnson and Shimizu (2004) suggest that by the end of the first year of life, infants have a broad and complex definition of what counts as an agent and attribute goals even to non-human and unfamiliar agents. These results imply that interactivity between humans and objects is the key factor in mental attribution. The evidence suggesting that the congruence and contingency of actions may play an important role in the perception and attribution of intentionality, has led to a growing interest in endowing robots with the ability to imitate. One important aspect of contingency is the temporal proximity between the user’s behavior and the robot’s response. In natural conversation, communication partners tend to respond to each other’s linguistic behavior in a time frame of approximately 200–300 ms (Sidnell & Enfield, 2012). Another important aspect refers to the synchrony between movements and sounds. When speaking to infants, parents perform movements that are in a tight temporal synchrony with the speech (Gogate, Bahrick, & Watson, 2010) and are shorter, which results in less roundness and more pauses between the individual segments (Brand et al., 2002; Rohlfing et al., 2006, cited in Fischer et al., 2013). Movellan and Watson (2002) investigated whether infants follow the line of regard of a non-human robot with an abstract pattern that did not resemble the structure of a human face. They found that infants follow the line of regard when the robot responded contingently to different events in the external environment and to the infants’ initiations. In another study, Meltzoff, Brooks, Shon, and Rao (2010) found that infants use the information derived from an entity’s interactions with other agents as evidence about whether that entity is a perceiver. The infants who saw the robot interacting contingently with a human adult were more likely to follow the line of regard of the robot. Other studies have investigated the preference infants have for a contingent or noncontingent interaction. The results of Sidner, Lee, Kidd, Lesh, and Rich (2006) suggest that participants interacted longer with a penguin-shaped robot that was producing contingent, non-verbal feedback, judging it as being more reliable and its movements more appropriate. Similar results were obtained by Kose-Bagci, Dautenhahn, Syrdalm, and Nehaniv (2010), participants showing a preference for a contingency pattern whose temporal dynamics was closer to human–human conversations. These results have important implications for the development of the social robotics domain. While some studies suggest that interactivity is the essential condition for the attribution of intentionality, and other studies suggest that contingency is necessary, another line of studies suggest that both the contingency and the congruence conditions should be met and that the robot should be designed in a way that allows it to be involved in “like-me” actions. On the other hand, robots represent as “an ideal opportunity to study cognitive and social development in infants” (Scasselatti, 1998, 2000; Deák, Fasel, and Movellan, 2001). By creating robots that have different morphological characteristics and by programming them to exhibit precisely controlled contingency structures, researchers can test the strategies infants use to identify their nature and capabilities (Movellan & Watson, 2002). The questions that should be addressed in this approach are: Which are the requirements that robots should meet in order to be treated by infants as conversational agents? Is animacy a sufficient condition? Is contingency absolutely necessary? What about similarity? Do infants need similarity in order to find a correspondence between their actions and the actions of the robot? This study aims to investigate if infants use an unknown agent, the robot Keepon, as a communicative partner after observing a brief conversation between a human adult and the robot. More specifically, our study aims to investigate if the interactivity of the robot Keepon is a sufficient condition for perceiving it as a communicative partner. Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
3
2. Objectives of the present study In this study, we investigate the following research questions:
(1) Do infants manifest more initiations in the interactive robot condition compared with the non-active robot condition? (2) Do infants manifest more anticipation of turn-taking behaviors in the interactive robot condition compared with the non-active robot condition?
3. Experimental platform and system design The experimental test-bed used in this study is the robot MyKeepon (see Fig. 1) a simplified version of the Keepon robot developed by Kozima, Nakagawa, and Yano (2004). While Keepon is a complex research version with complex abilities, the MyKeepon version has limited abilities and cannot be controlled by a PC, while the appearance is identical. Keepon is a small (12 cm) creature-like robot designed for simple, natural, nonverbal and playful interaction with children. It has a simple (like a yellow snowman) and soft (made of silicone rubber) appearance (Kozima et al., 2004). Despite its simplicity, Keepon is capable of displaying two psychological characteristics which are essential for the attribution of intentionality: attention (which suggests that the robot can hear and see the world), and emotion (which suggests that the robot has a “mind” to evaluate what it perceives) (Kozima, Michalowski, & Nakagawa, 2009). Moreover, Keepon’s repertoire of sounds and motions follow some of the requirements for child-directed speech: the pitch variation is high, the utterances are short, the pauses are long, the speech rate is low (Fernald et al., 1989). This pattern of speech seems to facilitate language acquisition, to modulate attention and arousal levels and to communicate affect to the infant (Fernald et al., 1989). Another strength of Keepon is its high speed of execution, an essential ability for achieving a high level of temporal contingency between the acts of the child and those of the robot, a condition for our study. Last, but not least, the low price of the MyKeepon version makes it available for any research group or pedagogical institution. For our study, we have used an enhanced version of the commercial version, which was controlled by an operator through a computer (Cao et al., 2014). The interface used to control the robot (see Fig. 2) is intuitive and therefore, easy to use. In the exposure phase, the movements and sounds of the robot were part of a pre-programmed scenario that was started when the human operator pushed the start scenario button (see Table 1). In the testing phase, the computer operator was trained to correctly identify the initiations of the child and manually push the buttons that encoded the specific movements and sounds in the order they were displayed on the interface. In order to facilitate future research with this robot platform, the manual how to modify the hardware, the electronics and software is made open source available on http://probo.vub.ac.be/HackingKeepon/.
Fig. 1. The MyKeepon robot.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model
ARTICLE IN PRESS
INFBEH-1051; No. of Pages 11
4
A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
Fig. 2. The operator interface program ViKeepon.
Table 1 The scenario for the interactive robot condition. 00:02.000 H: Hi Keepon! 00:03.000 K: Up wakeup (sound) 00:03.500 K: Jump (motion) 00:03.750 K: Jump (motion) 00:04.500 H: I am glad to see you! 00:06.500 K: Initial Boot (sound) 00:06.750 K: Bow (motion) 00:08.000 H: How are you? 00:10.500 K: Sigh (sound) 00:10.750 K: Bend (motion) 00:12.000 H: Oh, but why are you sad? 00:14.500 K: Sneeze up (sound) 00:14.750 K: Jump (motion) 00:16.000 H: Oh, you got a cold! 00:18.000 K: Sneeze up (sound) 00:18.250 K: Jump (motion) 00:19.000 H: You have been trodding through puddles again 00:22.500 K: Yawn down (sound) 00:22.750 K: Bow (motion) 00:23.500 H: What did your mommy say when you came home so wet? 00:27.000 K: Say no (motion) 00:27.500 K: Whine (sound) 00:28.000 H: Keepon, can you do this? 00:32.500 K: Bend (motion) 00:32.750 K: Sigh (sound) 00:33.500 H: Well done! What about this? 00:37.500 K: Bow (motion) 00:37.750 K: Sigh (sound) 00:39.000 H: Keepon, do you want me to bring you jellies next time? 00:45.000 K: Jump (motion) 00:45.250 K: Initial boot (sound) 00:46.000 H: Ok, I will bring you many colorful jellies! 00:50.000 K: Jump (motion) 00:50.250 K: Yawn up (sound) 00:51.500 H: I enjoyed playing with you! 00:53.500 K: Yawn down (sound) 00:53.750 K: Jump (motion) 00:55.000 H: Bye, Keepon! 00:56.500 K: Head hit beep (sound) 00:56.700 K: Head hit beep (sound) 00:56.750 K: Bow (motion)
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model
ARTICLE IN PRESS
INFBEH-1051; No. of Pages 11
A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
5
4. Method 4.1. Participants The participants for this study were 22 children, including 14 boys and 8 girls, aged 9–18 months, (M = 13.68, Sd = 2.67). All children were full-term, had normal birth weight and had no developmental concerns according to parental report. Three children, two boys and one girl, had to be excluded due to intense fussiness. The Bayley Scales of Infant Development (Second Edition, BSID-II) were administered by two experienced clinical psychologists that had demonstrated reliability in scoring (see Table 2). The Bayley score index suggested a typical development for all infants that participated to the study. The infants were exposed to both a contingent and a noncontingent interaction between a human adult and the social robot Keepon. The order of the presentation of the two conditions was randomised. 10 infants started with the interactive robot condition, followed by the non-active robot condition, and 9 children started with the nonactive robot condition, followed by the interactive robot condition. Of the 19 infants (12 boys and 7 girls), 6 infants refused one of the two experimental conditions (3 children refused the non-active robot condition when presented second time and 1 child refused the non-active robot condition when presented for the first time, and 2 children refused the interactive robot condition when presented second time). All participants signed a parental consent form before the study started. 4.2. Setting Each child was tested in a room (4 × 5 m) while seated in his/her parent’s lap, in front of a table (77 × 155 cm). In the exposure phase, at 100 cm distance away from the face of the child, on the other part of the table, the robot and the experimenter were facing each other, so the child saw the robot from the lateral point of view. In the testing phase, the robot faced the child and the same distance was maintained (100 cm) (see Figs. 3 and 4). Two digital cameras recorded the interaction; a frontal camera recorded the child’s upper torso and face, and a lateral one captured the child, the robot and the experimenter in the exposure phase, and the child and the robot in the testing phase. 4.3. Procedure 4.3.1. Habituation with the environment The experimenter, the child and one of the parents entered the experimental room. An assistant of the experimenter, the robot operator, was also in the room. The experimenter introduced some toys to the child (bubbles, a ball, a car, a doll, a small book with animals) and the child explored the environment and interacted with the toys, while the experimenter was chatting with the parent. A grey cardboard screen (29.7 × 42 cm) hid the robot during this time. After a 10 to 15 min interval, when the parent confirmed that the child was relaxed and acclimated to the new environment, the experimenter’s assistant uncovered the robot Keepon, the experimenter took his position on the chair in front of the robot, and the parent was asked to sit on a chair with the child on his/her lap. The experimental procedure involved two phases: the exposure to a scripted experimenter-robot conversation and a subsequent test in which the motor initiations and the vocalizations the child addressed to the robot were recorded. In the interactive robot condition, Keepon responded contingently, by a preprogrammed set of sounds and motions to the adult verbal initiations (see Table 1), while in the non-active robot condition, Keepon remained still. In the testing phase, Keepon responded by the same pre-programmed set of sounds and motions to any intentional vocalization or motor act of the child. The parent received the instruction to keep the verbal interaction to a minimum during the task. In the case the child asked for assurance, the parent had to acknowledge the child by nodding, saying something as “I see it” and redirect the attention to Keepon. We asked the parent to hold the child in his arms and restrain him from climbing the table. 4.3.2. Exposure phase: 4.3.2.1. Contingent interaction condition. The child was exposed to a 65 s conversation between the experimenter and the robot Keepon. While the experimenter was addressing Keepon verbally, Keepon responded contingently by a pre-programmed set of sounds and motions. The conversation consisted of 14 replies for each of the two partners (see Table 1). Table 2 The characteristics of the two randomised groups of children.
Chronological age Bayley score
Interactive robot/ non-active robot group
Non-active robot/ interactive robot group
P
N
M
SD
N
M
SD
P
10 10
13.45 99.8
1.18 11.33
9 9
13.83 96.44
3.38 13.62
.79 .57
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
6
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
Fig. 3. Room diagram for robot-infant interaction.
Fig. 4. An overview of the experimental task during the two conditions.
4.3.2.2. Non-active robot condition. In the non-active robot condition, the behavior of the experimenter did not differ compared to the interactive robot condition. The robot did not produce any sound or motion. The duration of the interaction was the same as in the interactive robot condition. At the end of the conversation, the experimenter repositioned Keepon, so that it faced the child, and told the child that Keepon is playing Peek-aboo, masked the robot with a grey screen and left the room. The robot was covered for 30 s, the time needed for the operator to start the ViKeepon program. Afterwards, the experimenter uncovered the robot and the testing phase began. 4.3.3. Testing phase The testing phase was identical for both the interactive robot condition and the non-active robot condition. In the beginning, Keepon initiated a motion accompanied by a short sound waiting for the child to respond. Keepon repeated the initiation every 20 s, for 120 s, the total duration of the testing phase. The role of these spontaneous initiations was to offer the opportunity to perceive Keepon as animated, to the children who did not produce any initiation in the first part of the testing phase. Keepon responded by producing sounds and movement as a feedback to the vocalizations and to the motor initiations of the child that were directed toward it. The repertoire and the order of motions and sounds produced by the robot was the same for all the children, as specified in the protocol (see Table 1). A snapshot of the two experimental phases is presented in Fig. 4. 4.3.4. Experimental design A 2 × 1 within subjects experimental design (interactive robot condition/non-active robot condition × Keepon partner) was used in this study. In the Exposure phase, the measured variables were the frequency of anticipatory orientations of attention toward the Keepon robot when the robot was next in the conversation. In the Testing phase, the measured variable was the frequency of total initiations produced by the child and directed toward the robot Keepon. The definitions of the dependent variables and the instructions for coding are presented in Table 3.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
7
Table 3 Definitions of the measured variables. Variable
Measure
Specifications
Anticipatory orientations of attention toward Keepon
Frequency
Total initiations of social interaction
Frequency
Score when the child switches attention from the adult to Keepon, after the adult has finished his remark, but exactly before Keepon starts to perform any action If the child maintains attention fixed to Keepon, do not score as an anticipatory behaviour If the child looks in any other direction besides the experimenter before switching to Keepon, do not score an anticipatory behavior Any motor or verbal behavior that is directed toward Keepon and is accompanied by visual contact The behaviors that are very subtle are not scored For some verbal behaviors that were not accompanied by visual contact, but seemed to be directed toward Keepon, a supplementary agreement between two independent coders was obtained Do not score prompted gestures
5. Data analysis 5.1. Data collection The two phases of the experiment, the exploratory phase and the testing phase, were presented in a standardised format. The exploratory phase lasted 64 s and involved 14 conversational replies for each of the agents (experimenter and the robot Keepon) in the interactive robot condition and 14 conversational replies for the experimenter and 14 pauses for the robot Keepon in the non-active robot condition. In the non-active robot condition, the experimenter offered his reply when this appeared on the computer screen after a pause that lasted the equivalent of the according reply in the interactive robot condition. The testing phase lasted 50 s for all the participants. A video coding analysis was performed in order to measure the target variables. The behaviors were manually coded from the videotapes using ELAN transcription software (see Fig. 4) (Sloetjes & Wittenburg, 2008). The video recordings were scored by an independent coder. For the coding of the initiation behaviors, the coder was blind to the infant’s experimental group, since in the testing phase, Keepon had an identical behavior in the two conditions. We assessed the scoring agreement by rescoring 50% of the recordings. We performed an inter-rater agreement analysis on the measured variables. We found a significant Pearson coefficient with a high magnitude for contingent initiations (r = .99, p < .001), noncontingent initiations (r = .98, p < .001), anticipations of turn taking in the interactive robot condition (r = .99, p < .001), and anticipations of turn taking in the non-active robot condition (r = .99, p < .001). 5.2. Statistical analysis 5.2.1. Frequency of initiations We used a Wilcoxon signed-rank test to analyze if there are any differences between the frequency of initiations in the interactive robot condition and in the non-active robot condition. The frequency of total initiations in the interactive robot condition (M = 9.68, Sd = 6.73) was significant higher than the frequency of total initiations in the non-active robot condition (M = 6.2667, Sd = 7.54), Z = −2.363, p = .018, with a medium effect size r = −.682 (see Fig. 5). We have controlled the order effect by randomising the order of the presentation of the interactive robot condition and the non-active robot condition. A Mann–Whitney test was performed in order to verify if the order of the condition influences the distributions of the measured variables. In addition, we performed the same analysis for the non-active robot condition. The Mann–Whitney test indicated that the frequency of the initiations recorded when the interactive robot condition was presented first (Mean rank = 12.06) was significantly higher than when it was presented after the non-active robot condition (Mean rank = 5.56), Z = −2.65, p = .006, r = −.618. On the other hand, the difference between the frequency of the initiations recorded when the non-active robot condition was presented first (Mean rank = 6.50) and when it was presented after the interactive robot condition (Mean rank = 8.25) was not significant, suggesting that the order effect did not influence the frequency of initiations in the non-active robot condition. 5.2.2. Frequency of anticipations of turn-taking The Wilcoxon signed-rank test revealed no significant differences between the frequency of anticipations of turn-takings in the interactive robot condition (M = 3.84, Sd = 2.79) and in the non-active robot condition (M = 4.89, Sd = 2.53), Z = −1.50, p = .133 (see Fig. 6). The Mann–Whitney test indicated that the frequency of anticipatory behaviors in the interactive robot condition presented first (Mean rank = 9.90), did not differ from the frequency of anticipatory behaviors when the interactive robot condition was presented after the non-active robot condition (Mean rank = 10.11), Z = −.083, p = .968. Also, the difference between the anticipatory behaviors in the non-active robot condition presented first (Mean rank = 9.22) and after the Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
8
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
Fig. 5. The graphical representation of the frequency of initiations in the interactive robot condition versus non-active robot condition.
Fig. 6. The graphical representation of the anticipations in the interactive robot condition versus non-active robot condition.
interactive robot condition (Mean rank = 10.70) was not significant too, Z = −.575, p = .604. These data suggest that the anticipatory behavior variable was not influenced by the order effect. 5.2.3. The age impact The Kruskal Wallis test was used to investigate if the age of the infants influenced the frequency of the measured variables. The results suggest that the distribution of both initiations and anticipations was the same across the whole age spectrum in both the interactive robot condition and the non-active robot condition (see Table 4). 5.2.4. Additional observations A series of additional observations that describe the children’s behavior during the task and are not captured by the quantitative analysis are presented below. First, the most of the participants seemed to enjoy the interaction with Keepon. From the 22 initial participants, only one was excluded due to fear of the robot. Taking into account the small age of the participants, this could be a good indicator that Keepon is a non-threatening interaction partner for infants. In the exposure phase, some infants expressed positive affect during the interactive robot condition, a phenomenon that was not observed in the non-active robot condition. In the testing phase, a tendency was observed for the younger infants lose interest for the Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
9
Table 4 The distribution of the dependent variables across the age spectrum. Null hypothesis
Test
Sig.
The distribution of initiations in the interactive robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.161
Kruskal–Wallis The distribution of initiations in the non-active robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.312
Kruskal–Wallis The distribution of anticipation behaviors in the interactive robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.505
Kruskal–Wallis The distribution of anticipation behaviors in the non-active robot condition is different in the two categories of age (9–14 and 14–18)
Independent samples
.333
Kruskal–Wallis
robot in the non-active robot condition. Some infants recognized whether the robot was responsive to them very quickly, and engaged in the conversation. The most common initiations produced by the infants were: vocalizations, hitting the table, reaching out to Keepon, shaking head, laughing. While in the first seconds, some infants would produce an initiation, and after the response of the robot, would wait for some seconds before producing another one, in the last part of the interaction, the pauses between the initiations became shorter. This phenomenon may indicate that these infants were testing the robot capabilities in the first part, and then accepted the robot as a conversational partner in the second part of the interaction. 6. Conclusions and discussions The purpose of this study was to determine whether infants perceive the interactive robot Keepon as a communicative agent. The results show that infants use the cues from the interaction between the adult and the robot Keepon to make inferences regarding the social agency of Keepon. Therefore, after observing the contingent interaction between the adult and the robot Keepon, infants manifested more initiations in the testing condition compared to the infants exposed to the non-active robot condition. Congruent with the results of Johnson and Shimizu (2004), our results suggest that infants have a broad definition of what counts for an agent and can attribute goals even to simple creatures such as robot Keepon, when animacy is involved. Since our study did not include a fine control for different levels of contingency, we cannot argue that the attunement of the behavior of the robot and human adult was the factor responsible for the inference that the robot Keepon has a mind of its own and is capable of conversation. Future studies should investigate in more detail which are the factors that are responsible for triggering initiations in infants, initiations that become part of a genuine conversation between the two partners. The factors that are worth considering are: the simple animacy, the temporal contingency of the actions of the two agents, the synchrony between movement and sound, and the simplicity of the interaction (quick motions, short and attractive, pitched sounds similar to mothering). In the light of the studies performed by Meltzoff et al. (2010); Johnson and Shimizu, (2004) and Movellan and Watson (2002), it is highly probable that the critical factor involved in the attribution of intentionality to be the contingency between the robot and the human adult’s behavior. In all the mentioned studies, the authors have found that the motion itself could not account for the results across the groups. The real factor that carries a significant weight is not the motion itself, but the nature of the interaction between the two agents. Another question raised by our results is if the measured variable, the infants’ motor and verbal initiations could account for something different than the attribution of intentionality in the robot Keepon. Is it possible that infants were just imitating the behavior observed in the human adult in the exposure phase? Since the human adult behaved identically in the robot interaction condition and in the non-active condition, this hypothesis can be eliminated. Therefore, we consider that the motor and verbal initiation variable is an innovative and reliable measure of intentionality detection in infants. Another result of our study is that there were no significant differences between the frequencies of anticipation of turntaking behaviors in the interactive robot condition compared with the non-active robot condition. This is contrary to our expectation that infants will manifest more anticipatory looking behaviors to the robot in the interactive robot condition. From our point of view this result may suggest that the interactive behavior of the human adult generated in infants the assumption that the robot Keepon is a mentalistic agent, therefore, they have anticipated a response from the robot. Even though our results are congruent with the one obtained by Arita et al. (2005), they interpret the result in a different way, by arguing that the looking time measure indicates the fact that infants did not expect the robot to be talked to by the person. An alternative explanation would be that by looking at the robot, the infants were following the attention of the adult; therefore Keepon might have functioned as a joint attention object for the adult and the infant. In a future study, we should find a more reliable measure of the expectations infants have from mentalistic agents. Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
10
One limitation of our study refers to using a repeated measures design, with infant being exposed to both the interactive robot condition and the non-active robot condition. The analysis of the ordering effect suggests that the frequency of initiations was influenced by the order of the presentation of the two conditions for the interactive robot condition. More specifically, there were more initiations recorded when the interactive robot condition was presented first, compared to the interactive robot condition presented as the second condition. This result suggests that although we have controlled the effect of the repeated exposure by counterbalancing the order of the two conditions, there are still limitations in using such a design. In a future improved version of this study, a design with independent groups will eliminate the bias of the order of the two conditions. As well as the limitations previously mentioned, the technical constraints of the robot could also represent a limitation of the study. Because of its shared autonomy, errors by the operator could influence research outcomes, thus making the robot incapable of adapting to children’s behaviors at this point in time. Future work will consist of developing supervised autonomous interaction, so that an operator is no longer required (Thill, Pop, Belpaeme, Ziemke, & Vanderborght, 2013). Finally yet importantly, the qualitative observations from our study suggest that MyKeepon modification is a useful platform for investigating HRI in infants, since MyKeepon seems to be a non-threatening and attractive interaction partner for this age group. Moreover, the low-cost of the platform represents an additional argument for a long-term use in studies on the effects of specific factors infants use to infer the existence of a mind in a non-human agent. Concluding, our study suggests that the social robot Keepon is an appropriate social interaction partner for infants, a partner that is perceived as a conversational agent when offering communicative cues. More in depth work is needed in order to understand exactly the attributes that make infants assume Keepon is a social agent. Our proposal, inspired by the work of Meltzoff et al. (2010), and Arita et al. (2005), is to include, in a future study, both an interactive and noncontingent robot condition, and an active robot condition, in which the robot behaves like a human, and the person is stationary and silent. Such an experimental design would facilitate the distinction between the effect of animacy and the effect of contingency. This study opens the door and offers incipient answers to important questions regarding the future of HRI domain. When social robots will be part of our lives, will people regard these artificial objects as mentalistic agents? If so, which are the conditions that a robot should respond to, in terms of interaction dynamics, morphology and motion characteristics? This study comes in line with some other studies suggesting that interactivity is an important aspect in the attribution of intentionality. Moreover, our results suggest that morphological similarity is not absolutely necessary for the attribution of intentionality. Acknowledgements The authors thank the children that participated in the study and their parents which made this study possible. This work has been supported by the CNCSIS-Bucharest, Romania project PN-II-IDPCE-2011-3-0484-Exploring Robot-assisted therapy for children with ASD and the European FP7 project DREAM (grant no. 611391). Appendix A. Appendix Table 1 References Arita, A., Hiraki, K., Kanda, T., & Ishiguro, H. (2005). Can we talk to robots? Ten-month-old infants expected interactive humanoid robots to be talked to by persons. Cognition: 95., B49–B57. Cao, H. L., Van De Perre, G., Simut, R., Pop, C., Peca, A., Lefeber, D., et al. (2014). Enhancing MyKeepon robot: A simple and low-cost solution for robot platform in human–robot interaction studies. Proceedings of the IEEE International Symposium on Robot and Human Interactive Communication, 555–560 (RO-MAN) Deák, G. O., Fasel, I. R., & Movellan, J. R. (2001). The emergence of shared attention: Using robots to test developmental theories. In C. Balkenius (Ed.), Proceedings of the first international workshop on epi-genetic robotics: Modelling cognitive development (pp. 95–104). Lund, Sweden: Lund University Cognitive Studies. DeCasper, A. J., & Fifer, W. P. (1980). On human bonding: Newborns prefer their mothers’ voices. Science: 208., (4448), 1174–1176. Fischer, K., Lohan, K., Saunders, J., Nehaniv, C., Wrede, B., & Rohlfing, K. (2013). The impact of the contingency of robot feedback on HRI. In International conference on collaboration technologies and systems San Diego, CA, (pp. 210–217). http://dx.doi.org/10.1109/CTS.2013.6567231 Fernald, A., Taeschner, T., Dunn, J., Papousek, M., Boysson-Bardies, B., & Fukui, I. (1989). A cross-language study of prosodic modifications in mothers and fathers’ speech to preverbal infants. Journal of Child Language: 16., 477–501. Fernald, A., & Mazzie, C. (1991). Prosody and focus in speech to infants and adults. Developmental Psychology: Vol. 27., (2), 209–221. Gogate, L. J., Bahrick, L. E., & Watson, J. (2010). A study of multimodal motherese: The role of temporal synchrony between verbal labels and gestures. Child Development: 71., 878–894. Gordon, R. M. (1995). Simulation without introspection or inference from me to you. In M. Davies, & T. Stone (Eds.), Mental simulation (pp. 53–67). Oxford: Blackwell. Johnson, S. C., & Shimizu, I. A. (2004). Infants’ attribution of a goal to a morphologically unfamiliar agent. Developmental Science: 7., (4), 425–430. Kose-Bagci, H., Dautenhahn, K., Syrdalm, D. S., & Nehaniv, C. L. (2010). Drum-mate: Interaction dynamics and gestures in human-humanoid drumming experiments. Connection Science: 22., (2), 103–134. Kozima, H., Nakagawa, C., & Yano, H. (2004). Can a robot empathise with people? Artificial Life and Robotics: 8., (1), 83–88. Kozima, H., & Nakagawa, C. (2006). Interactive robots as facilitators of children’s social development. Mobile Robots Towards New Applications, 314–315. Source: Mobile Robots Towards New Applications, ISBN 3-86611-314-5, Edited by Aleksandar Lazinica, pp. 784, ARS/plV, Germany, December 2006.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005
G Model INFBEH-1051; No. of Pages 11
ARTICLE IN PRESS A. Peca et al. / Infant Behavior & Development xxx (2015) xxx–xxx
11
Kozima, H., Michalowski, M. P., & Nakagawa, C. (2009). Keepon a playful robot for research, therapy, and entertainment. International Journal of Social Robotics: 1., 3–18. Legerstee, M., Barna, J., & DiAdamo, C. (2000). Precursors to the development of intention at 6 months: Understanding people and their actions. Developmental Psychology: 36., 627–634. Meltzoff, A. N., & Moore, K. M. (1989). Imitation in newborn infants: Exploring the range of gestures imitated and the underlying mechanisms. Developmental Psychology: Vol. 25., (6), 954–962. Meltzoff, A. N. (1990). Foundations for developing a concept of self: The role of imitation in relating self to other and the value of social mirroring, social modeling, and self practice in infancy. In D. Cicchetti, & M. Beeghly (Eds.), The self in transition: Infancy to childhood (pp. 139–164). Chicago, IL: University of Chicago Press. Meltzoff, A. N., & Moore, M. K. (1995). Infants’ understanding of people and things: From body imitation to folk psychology. In J. L. Bermudez, & N. A. M. Eilan (Eds.), The body and the self (pp. 43–69). Cambridge, MA: MIT Press. Meltzoff, A. N., & Brooks, R. (2001). “Like me” as a building block for understanding other minds: Bodily acts, attention, and intention. In B. F. Malle, L. J. Moses, & D. A. Baldwin (Eds.), Intentions and intentionality: Foundations of social cognition (pp. 171–191). Cambridge, MA: MIT Press. Meltzoff, A. N. (2002). Elements of a developmental theory of imitation. In A. N. Meltzoff, & W. Prinz (Eds.), The imitative mind: Development, evolution, and brain bases (pp. 19–41). Cambridge: Cambridge University Press. Meltzoff, A. N. (2005). Imitation and other minds: The “Like Me” hypothesis. In Perspectives on imitation: From neuroscience to social sciences. pp. 55–77. Cambridge, MA: MIT Press 2. Meltzoff, A. N., Brooks, R., Shon, A. P., & Rao, R. P. N. (2010). “Social” robots are psychological agents for infants: A test of gaze following. Neural Networks: 23., 966–972. Movellan, J. R., & Watson, J. S. J. S. (2002). The development of gaze following as a Bayesian systems identification problem. In Proceedings of the international conference on development and learning (ICDL02) IEEE. Nadel, J., & Butterworth, G. (1999). Imitation in infancy. Cambridge, UK: Cambridge University Press. Rizzolatti, G., Fogassi, L., & Gallese, V. (2002). Motor and cognitive functions of the ventral premotor cortex. Current Opinion in Neurobiology: 12., 149–154. Scasselatti, B. (1998). Building behaviors developmentally: A new formalism. In Proceedings of the AAAI spring symposium on integrating robotics research. Scasselatti, B. (2000). How robotics and developmental psychology complement each other. In NSF/DARP a workshop on development and learning. Sidnell, J., & Enfield, N. J. (2012). Language diversity and social action: A third locus of linguistic relativity. Current Anthropology: 53., 302–333. Sidner, C. L., Lee, C. L., Kidd, C., Lesh, N., & Rich, C. (2006). Explorations in engagement for humans and robots. Artificial Intelligence: 166., (1–2), 140–164. Sloetjes, H., & Wittenburg, P. (2008). Annotation by category—ELAN and ISO DCR. In Proceedings of the 6th international conference on language resources and evaluation LREC, 2008. Thill, S., Pop, C. A., Belpaeme, T., Ziemke, T., & Vanderborght, B. (2013). Robot-assisted therapy for autism spectrum disorders with (partially) autonomous control: Challenges and outlook. Paladyn Journal of Behavioral Robotics: 3., (4), 209–217. Tomasello, M. (1999). The cultural origins of human cognition. Cambridge, Massachusetts, London, England: Harvard University Press. Trevarthen, C. (1993a). The function of emotions in early infant communication and development. In J. Nadel, & L. Camaioni (Eds.), New perspectives in early communicative development (pp. 48–81). London: Routledge. Trevarthen, C. (1993b). The self born in intersubjectivity: The psychology of an infant communicating. In U. Neisser (Ed.), The perceived self: Ecological and interpersonal sources of self-knowledge (pp. 121–173). New York, NY: Cambridge University Press. Trevarthen, C., & Aitken, K. J. (2001). Infant intersubjectivity: Research, theory, and clinical applications. Journal of Child Psychology and Psychiatry: 42., (1), 3–48. Tronick, E. Z. (1989). Emotions and emotional communication in infants. American Psychologist: 44., 112–119.
Please cite this article in press as: Peca, A., et al. Do infants perceive the social robot Keepon as a communicative partner? Infant Behavior and Development (2015), http://dx.doi.org/10.1016/j.infbeh.2015.10.005