Context-related group differences in verbalization patterns

Children in the music and academic groups verbalized spontaneously with their social partners for a longer duration compared to children in the robot group (see Figure 5-3). We hypothesized that the amount of verbalization in the movement groups would be comparable to that seen in the standard-of-care academic group. The stationary academic group engaged in activities such as reading and fine motor imitation, which are typically practiced by children in school-based ABA settings. Conventionally, these sedentary contexts have been used to facilitate verbal communication in children (Goldstein, 2002; Paul & Sutherland, 2005). Such table-top contexts are ideal to promote verbal skills, since children are relatively confined at the table, and trainers are provided with abundant opportunities to engage children in reciprocal conversations. In contrast, in the music and robot groups, children engaged in rhythmic synchrony and imitation-based gross motor games with their social partners to the beat of music. In spite of the relatively unconstrained nature of training activities in the music and robot groups, we expected children to engage in high levels of interactions with their social partners, since the activities practiced were tapping into children’s inherent strengths and predilections. In comparison to adult-selected

activities within highly structured and confined table-top settings, children showed greater improvements in language skills following child-led activities that took place within unconfined play-based settings, as suggested by a comprehensive review based on 10 studies (Delprato, 2001). Moreover, interventions that promoted skills such as motor imitation, play, and joint attention led to collateral increases in language skills of children with ASDs (Ingersoll & Schreibman, 2006; Jones, Carr, & Feeley, 2006; Kasari, Paparella, Freeman, & Jahromi, 2008; Kasari et al., 2006). When children engaged in co-constructed and synchronous activities with their social partners, a state of joint engagement was created, which in turn promoted verbal communication skills over sessions (Ingersoll & Schreibman, 2006; Kasari et al., 2008). Along the same lines, our movement groups were designed to promote joint engagement through

imitation-based games, and we expected children to verbalize at levels that were comparable to those seen in the academic group. Although the music group followed our expectations by the late session, the robot

group verbalized at levels that were significantly lower than the music and academic groups even in the late session.

We think that the current limitations of our robotic technology might have restricted the amount of verbalization in the robot group. Behaviors exhibited by the robot influence the social behaviors exhibited by the child within the robot-child interaction context. Specifically, children verbalized more, both towards the robot and towards their parents, when the actions of the robot were contingent to those of the child compared to a non-contingent condition (Feil-Seifer & Mataric, 2008). The Nao robot we used had limited pre-programmed verbiage. Although we utilized the online speech capabilities of the robot, there was some time lag to the responses of the robot, as a result of the trainer having to manually type in responses to the child’s verbiage using control software. Hence, the robot’s speech was not very contingent to the speech of the child. Moreover, many children found the robot’s speech unclear, which required the trainer to repeat the robot’s verbiage to aid children’s comprehension. Overall, we think that the lack of contingent responding by our robot might have limited its ability to act as an effective social mediator to facilitate children’s communication skills, resulting in a lower amount of verbalization in this group compared to the human-delivered music and academic groups. A few previous studies have compared the effects of a human versus robot mediator on the social communication abilities of children with ASDs (Duquette et al., 2008; Huskens, Verschuur, Gillesen, Didden, & Barakova, 2013; Kim et al., 2013). Children directed greater number of utterances towards an adult confederate during a single session of interaction with a robot partner compared to a human interaction partner within a triadic context involving the child, the adult confederate, and the interaction partner (Kim et al., 2013). In the robot condition, children initiated spontaneous conversation with the adult confederate to share their interest and curiosity about the robot (Kim et al., 2013). In another study, following ABA-based training to promote self-initiated questions either using a human or a robot trainer, children in both groups improved equally in their abilities to ask questions during a follow-up session, suggesting that there were no added benefits obtained with a robot trainer (Huskens et al., 2013). Lastly, more in lines with our

findings, Duquette and colleagues demonstrated that children showed better verbal imitation abilities with a human mediator compared to a robot mediator (Duquette et al., 2008). Although the first two studies showed either comparable (Huskens et al., 2013) or in fact superior (Kim et al., 2013) communication skills following interactions with a robot compared to a human, we argue that the effects were limited to a single session of training. In contrast, in our study, we were able to compare the effects of a prolonged 8- week robot-delivered intervention with the effects of a comparable human-delivered intervention on the communication skills of children with ASDs. We observed that once the initial novelty of the robot wore off, children did not find the robotic context as engaging as the music and academic groups, which might have led to lower overall verbalization in the robot group compared to the other groups.

A related finding in the robot group was a greater amount of vocalizations/verbalization towards self across sessions (see Figure 5-3). During the movement-based action game, children were required to synchronize their movements with the adult model. Many children chose to spontaneously count the movements as they were synchronizing with the model. In addition, children in the robot group practiced simple and complex drumming patterns during the drumming condition. Specifically, children practiced quarter and eighth patterns while associating them with verbal labels such as “Pal-Buddy” (quarter note on “Pal” and eighth note on “Buddy”) or “Grape-Lemon” (quarter note on “Grape” and eighth note on “Lemon”) to facilitate the mapping between sounds and their motor actions (Wan, Demaine, Zipse, Norton, & Schlaug, 2010). We found that children effectively learned complex sequences of quarter and eighth notes using this strategy and across sessions, children chose to spontaneously use these verbal cues as they practiced drumming patterns. Since, the verbal cues used during action game and drumming game were not directed towards any of their social partners or to the robot, they were coded as self-directed verbalizations. However, we noticed that children verbalized to themselves not just during the drumming and action game conditions but also during other conditions within training. Our observations suggest that this might be due to the limitations of the current robotic technology. As discussed above, the limited verbal repertoire of the robot prevented it from being a compelling social entity and children quickly lost

interest in the training activities. Instead of engaging in meaningful communicative exchanges with their social partners using the robot as an effective mediator, children engaged in scripting, echolalia, and self- directed vocalizations including squealing, giggling, and whining. Other data from our lab on social attention, affect, and repetitive behaviors also lend support to our hypothesis. Specifically, children showed progressive boredom with an increase in negative affect and negative behaviors across training weeks. Our observations fit with the findings of an existing 18-day trial on typically developing (TD) children, where children were allowed to interact with a humanoid Robovie robot that helped them learn new words. Although some children learnt new words following training, there was a sharp decline in their interest and interactions with the robot during the second week of training (Kanda et al., 2004). We recommend that future efforts be directed towards improving the robotic technology to make the robot’s actions and verbal interactions contingent on the child initiating the exchange, and also towards designing functionally relevant activities that can sustain the engagement of children over multiple sessions.

We also found group differences in the amount of responsive verbalization to social partners (see Figure 5-3). The academic group engaged in greater responsive verbalization across sessions compared to the other two groups. Although the quantity of spontaneous verbalization was greater than the amount of responsive communication within the academic group, there were significant between-group differences in the amount of responsive verbalization (see Figure 5-3). As outlined previously, the environmental setup of the academic group was in some ways similar to that used in traditional ABA settings, in that children were seated at the table and the trainer initiated several bids requiring a response from the child (Delprato, 2001; Goldstein, 2002); however, unlike traditional ABA interventions that are completely instructor-driven, we provided children with multiple opportunities for engaging in spontaneously- initiated communicative exchanges. Nevertheless, we think that the very nature of the context afforded several bouts of responsive communication between children and their social partners. To elaborate, in the academic group, trainers had greater opportunities to engage verbally with the child and to present several prompts to elicit appropriate verbal responses from children. For example, the trainer frequently

questioned children about the book they were reading, asked them to make choices in terms of usage of supplies, and encouraged children to read out listed instructions that were provided for the building and art-craft projects. Similar contexts used during traditional ABA interventions have been shown to increase responsive communication more than spontaneous initiations by children with ASDs (Kasari et al., 2006; Schreibman, 1997; Vismara & Rogers, 2010; Whalen & Schreibman, 2003). In contrast, in the music and robot groups, children engaged in movement-based games to background music during a majority of the conditions, which might have restricted the amount of communication that could be initiated by the trainer. Hence, a majority of the conversation afforded by the movement contexts was spontaneous in nature. Thus, we think that the movement and academic groups afforded different types of

communication between children and their social partners, due to the differences in the nature of the training context.