Preschoolers' Interactions with Social Robots: Investigating the Potential for Eliciting Metatalk and Critical Technological Thinking

In this paper, we present a pilot study investigating whether and how social robots can be used to elicit metatalk in preschool children. We conceive metatalk as communication about communication, which requires children's ability to reflect on what is happening in an interaction. We argue that this ability is essential to develop critical thinking about emerging technologies from an early age, thus empowering children as digitally literate citizens. Despite the variety of educational domains in which social robots have demonstrated their potential as learning partners, there has been little research on how they can be used to support children's language development in terms of their metatalk as a key social skill. The aim of this paper is twofold: First, we report on a method for eliciting metatalk using social robots. Specifically, we explored how small groups of children could first interact with two social robots, whereupon this interaction was used as a conversational prop in guided group discussions with an experimenter to elicit metatalk. Second, we present preliminary results suggesting that, based on the method developed, the children were able to successfully engage in metatalk, thereby potentially providing an entry point for critical reflection on the technology.


INTRODUCTION
Nowadays, there is a burgeoning development and application of advanced technologies, such as social robots, that young children encounter in their everyday environments.This raises the question of how we can fnd ways to support children to become digitally literate so that they can be informed and empowered users as well as responsible developers of future technologies.One way of empowering users is to foster their critical thinking about the use and design of technology [8].Individuals who can develop a critical understanding and refect on how technology afects an interaction will be able to recognize the potential and limitations of technologies, alternative ways of acting, and can more efectively decide on their responsibilities [1,20].However, in early childhood education, we know little about how to stimulate critical thinking about technological resources such as social robots.In fact, most existing work has focused on how to teach basic AI concepts [30], programming skills [28], or is aimed at older children [9].Importantly, critical thinking, e.g., refecting upon information reliability, represents an advanced metacognitive skill because it requires individuals to engage in explicit deliberation and refection that goes beyond the immediate context of an interaction [17, p. 550].More precisely, it has been shown that learning to think critically involves a set of higher-order cognitive skills (e.g., interpreting, predicting, analyzing, evaluating information) underpinned by lower-order skills (e.g., the ability to recall or comprehend information) [12,13,23].At the preschool age, these skills are still under development, however, research suggests that a specifc repertoire of verbal means and use of language seems to mediate this development [2,10].Yet, it remains poorly understood which verbal means are crucial.In this vein, we propose that metatalk, i.e., making communication itself the focus of activity, could lead to refective thinking [3,17,24]: Once children engage in metatalk, they might refect on the efects and characteristics of the digital tools they interact with.However, little is known which interactive contexts might be useful to elicit metatalk in preschool children and whether social robots can provide such contexts while potentially providing an opportunity to stimulate critical refection on the technology.Along these lines, previous research on young children's ability to metatalk has yielded somewhat mixed results.Köymen and Tomasello examined how 5-and 7-year-old peer dyads jointly made collaborative decisions when presented with conficting information that difer in reliability about what a new animal should eat, whereby one child heard a frsthand report from the animal itself while the other child heard a secondhand report from a girl expressing uncertainty [17].
The authors found that whereas all children talked about the appropriate food for the animal, predominantly only school-aged children were able to produce metatalk (e.g., "she was not sure whether it drinks blue soda").However, the authors acknowledged that in this study, the task burden of producing metatalk fell primarily on the child who received the report secondhand, which may have reduced the frequency of metatalk, particularly among the 5-year-olds.In a subsequent study, Köymen et al. reported that 5-year-olds, and even 3-year-olds after discourse training, could successfully engage in metatalk in the form of producing counterarguments, e.g., refuting their partners' arguments when collaboratively engaging in a task on deciding which box contained the item a specifc animal needed [16].Moreover, work by Grazzani et al. involving groups of children talking about mental states has convincingly shown that in such collaborative settings, children as young as two years old can refect on aspects of communication such as others' emotions, suggesting that even toddlers may be able to produce metatalk to some extent [11].Taken together, these fndings underscore the importance of the parameters of the interactive social context for eliciting metatalk.Importantly, so far, approaches have not been explicitly used to elicit metatalk in children for other purposes, such as stimulating critical refection on the characteristics of emerging technologies such as social robots.Therefore, in the present work, we report on a pilot study in which we systematically engaged small groups of preschool children in metatalk by introducing two diferent social robots, one with and one without communicative peculiarities, within an educational setting.The underlying premise of our study was that the systematic elicitation of metatalk through the use of social robots with communicative peculiarities could be harnessed to engage preschool children in critical thinking processes about technology.In contrast to prior studies (e.g., [17]) that elicited metatalk in preschoolers through discussions about the reliability of information, in our study the elicitation of metatalk within the group interaction was approached systematically by addressing diferent communicative dimensions in relation to the behavior of the robotic partner.

METHOD
Building on recent work that aims to use social robots in educational settings in innovative ways that go beyond traditional social roles, such as using the robot as a tutor or peer to teach new words in areas such as language learning [22], we will elaborate below on the rationale of the proposed approach and how we designed the robot's behavior with communicative particularities and designed the implementation of the group discussions to elicit metatalk in children.

Participants
Participating children (N = 17), 7 males and 10 females, ranged in age between 5 and 6 years (mean age = 5;6, SD = 0.5).All children were native German speakers and were recruited from two local kindergartens.Two morning classes from each kindergarten participated in the study, each visiting the experimental rooms at the university as a group.Based on discussions with the participating educators and ethical considerations, the individual groups of children were composed in the same way as they are usually composed within the kindergarten, that is, the children knew each other within the groups.A familiar educator was present during all child-robot interactions as suggested by [27], but was not actively involved in the interaction.None of the children had ever seen the robot before the experiment.In accordance with university ethics procedures for research with children, parents provided written consent prior to their children's participation.Children also provided verbal assent prior to taking part in the interaction with the robot, and the interaction could be discontinued at any time at no disadvantage to the children.In addition, the study design was approved by the university's ethics committee.

The robot(s)
We used two diferent humanoid Nao robots (see Figure 1a) from Softbank Robotics.Specifcally, informed by prior work [6], the elements in which the robots difered included the color of the robot (one red robot and one gray robot), and we further manipulated the voice of the robot using the integrated text-to-speech production, diferentiating the intonation between the robots by lowering the pitch of the synthesized voice for one of the robots, resulting in clearly distinct voices.In addition, the robots introduced themselves with diferent, gender neutral, names ("Nao" and "Nikita").Moreover, a further diference between the two robots was that one robot moderated the event, encompassing tasks such as welcoming the children, introducing the robots, and introducing the storytelling activity.Meanwhile, the other robot served solely as a narrator of the story.

Experimental Procedure
The overall procedure comprised three phases: (i) the warm-up phase, in which both robots were introduced to the children, (ii) the storytelling phase, which served as an interactive event that provided the basis for the subsequent initiation of metatalk (see Figure 1a), and (iii) the group interaction phase, in which the experimenter engaged in an interaction with children about their previous storytelling experience with the robots by asking various questions to systematically elicit metatalk (see Figure 1b).

Design of the warm-up to introduce the robots.
Informed by prior work [26], the interactive setting draws on familiar activities used in kindergarten and was structured in a way that a group of children (5-6 children) were frst introduced to the robots in a switched-of state.The children then drew a picture of the robot while the experimenter told them about some basic capabilities of the robot (e.g., that the robot could talk and move).This aimed both to reduce the novelty efect [21] and to provide the children with some background information, setting the stage for subsequent elicitation of metatalk [7].
2.3.2Design of the robots' interactive behavior to provide a basis for metatalk.The guiding idea when designing this phase was to capitalize on the specifc behaviors of social robots by using them as a stimulus within an initial communicative event that could be used in a subsequent conversational setting to elicit metatalk within the interaction in small groups of children.Specifcally, we sought to leverage the aspect that, compared to other types of technology, social robots, by virtue of their interactive, embodied, and social nature, are particularly useful for prompting or stimulating others to engage in communicative behavior [29], thus ofering potential as a conversational prop for eliciting metatalk.Additionally, since current state-of-the-art robots' interactive capabilities are rather limited in terms of contingent interactions, e.g., due to the persistent lack of robust speech recognition of a child's verbal utterances or visual social signal processing [4], the peculiarities in their interactive behavior provide a unique opportunity for children not only to critically evaluate the technology, but also to become aware of and talk about dialogical structures and features in general.Indeed, as mentioned earlier, preschoolers have been shown to engage in metatalk (e.g., [17]), and we built on the fact that they can refer to aspects of turn-taking, meaning, and acceptability of lexical choices [24].The overall phase followed the structure that frst one of the two robots welcomed the children and introduced that the other robot had prepared a story to tell the children.The other robot then introduced itself and told a short story about a competition between animals, appropriate for preschool children (see Figure 1a).Although initial fndings suggest that children may refect on and adapt to a robot's communicative behavior [25], previous work on human-human interactions indicates that preschool children may struggle to refect on atypical features of communicative behavior, such as those exhibited by their peers [5].Thus, during the narrative event, we even reinforced the peculiar behavior of one social robot by deliberately implementing pronunciation difculties into the robot that acted as the storyteller.Specifcally, this involved the robot speaking some words phonologically atypically and replacing consonants (e.g., "dold medal" instead of "gold medal") while telling the story.By employing two robots instead of just one, we further aimed to provide the children with an additional contrast in the communicative behavior between the two robotic interaction partners, thereby ofering another opportunity for metatalk.The fact that only one of the robots displayed specifc communicative peculiarities was also due to the fact that the participating children had never interacted with robots before and to prevent them from assuming that the implemented features were typical for a robot.During the interaction between the robots and the children, teleoperation was used to allow the robots to act more contingently [15] and to provide a comparable interaction experience for all groups of children.After an interaction time of 8-10 minutes during the storytelling phase, both robots said goodbye to the children and the experimenter transitioned to the next phase.

Design of the group interaction to elicit children's metatalk.
After interacting with the robots, the children were divided into subgroups (2-3 children) to employ the experienced situation within a group interaction with a human experimenter for eliciting metatalk (see Figure 1b).In total, our sample comprised a number of seven subgroups of children.The elicitation of metatalk within the group interaction was approached systematically by addressing diferent communicative dimensions in relation to the robots' behavior and encouraging the children to refect and comment on the verbal, nonverbal, and paraverbal aspects of the robot's dialogical behavior (e.g., the robot's speech style, facial expressions, gestures, emotions).Further, we devised questions (the developed set of questions can be found on the OSF: https://osf.io/exztv/)at both abstract and concrete levels to facilitate that all children could contribute to the discussion.We defned abstract questions as those providing openended response options and ofering no specifc cues regarding the linguistic peculiarities of the robot (e.g., Did you notice anything while the robot was talking?).In contrast, concrete questions included comparisons and contrasts that specifcally indicated the linguistic peculiarities of the robot (e.g., Did you think the robot spoke quickly or slowly?).While we recognize that the questions may have had a guiding character that afected the children's responses, it is important to highlight that the questions were developed on a theoretical-conceptual basis in order to elicit metatalk using specifc linguistic strategies and to provide the children of the target age with the support that was considered to be required [5].After having addressed the dialogical behavior of the robot on a frst meta level, in a second step, the aim was to guide the children to engage in metatalk about the technological characteristics of the social robots and their design and use in general.We particularly considered here to elicit metatalk aiming at critical refection that targets key AI literacy competencies such as programmability, the human role in AI, sensors, and the strengths and weaknesses of AI, which can also be well connected to the domain of social robots [18].

RESULTS
In our preliminary qualitative content analysis [19], we were interested in whether and in what ways children were able to produce metatalk and refect on the interactive behaviors of the robots they encountered, and thus to refect on the communicative peculiarities.As seen in Table 1, we found that after abstract questioning, in most cases the children were able to articulate specifc details about the robots' verbal behavior.More specifcally, although the children reported that they generally understood the robot well, they noted that some words were difcult to understand.Some children were even able to recall the specifc words (e.g., "dold medal") in which pronunciation difculties were implemented.When posed with more concrete questions, children further indicated that the robot's verbal behavior was diferent from that of humans (e.g., their parents or educators).Regarding the metatalk about the nonverbal dimension, the children refected that the robot could move its arms, but lacked mimic expressiveness.For example, one child pointed out: "he couldn't be sad and smile at all." At the same time, the children stated that a distinctive feature of the communicative nonverbal behavior was the lighting up of the eyes.Moreover, they commented that in contrast to the robotic partner, their parents combined verbal and nonverbal modalities diferently during storytelling activities.In addition, with respect to the metatalk about the paraverbal dimension, they pointed out that the pronunciation as well as the temporal structure of the turns were deviant and characterized by delays or pauses.Interestingly, in some cases, we could also observe that the children evaluated the dialogical behavior of the robot and made language-related judgments or refected on the tasks of a communicative partner on a metalevel against the background of the robots' behaviors.In this regard, one child stated: "[the robot is] a good storyteller, I think he's really good at it." Nevertheless, we observed that many of the preschool children encountered difculties in responding to abstract questions.In contrast, by using questions at a concrete level, such as comparisons and contrasts, contributions were elicited from almost all children.Thus, although abstract questions seem to be less guiding and more objective to children's opinion, concrete questions seem to facilitate metatalk in preschool children.Overall, we observed that by involving the children in metatalk, they were able to engage in refective thinking and critically discuss the robot's communicative behavior in a diverse range of aspects.

DISCUSSION AND CONCLUSIONS
In this paper, we introduced an approach that aims to systematically promote metatalk within a collaborative activity in kindergarten, using an interaction with social robots as a conversational prop for subsequent critical refection.Whereas social robots have been used in a range of educational settings [14], examples of using social robots to promote particular forms of communication are still limited, but have previously been suggested by Rohlfng et al. [22].We have presented a promising method and argue that by systematically initiating and allowing children to participate in metatalk, i.e., making communication itself the subject of conversation, children can be provided with an opportunity to refect on the interaction itself in more detail, thereby allowing them a further opportunity to critically look at their interactions with digital technologies such as social robots.In summary, our fndings suggest that preschoolers can actively participate in group discussions and engage in metatalk.Such metatalk seems to give rise to refections on communication on the one hand, and technology on the other.Yet, not all children participated equally in the discussions and needed additional support through concrete questions to express their observations.We assume that although all children made some observations about the communicative behavior of the social robots, not all were able to articulate them efectively.Consequently, in future work, it will be valuable to explore the possibility that metatalk with preschool children requires extended practice and multiple sessions.This might facilitate the establishment of a relevant vocabulary, thereby enabling all groups of children to engage more efectively in discussions about language and technology.In addition, and in light of the fact that children in our study engaged in the group interaction collaboratively, we will examine interactional group dynamics and how children build on each other's contributions, as well as the role that individual diferences play in children's engagement in metatalk.

Figure 1 :
Figure 1: Experimental setting of the study.A) Warm-up and storytelling phase with all children.B) Discussion phase in smaller groups.

Table 1 :
Metatalk dimensions and examples of children's produced metatalk to concrete and abstract questions.