Verbal Interactional Synchronization between Therapist and Children with Autism Spectrum Disorder during Dolphin Assisted Therapy: Five Case Studies

Simple Summary This study investigates the synchrony in conversations (i.e., turn-taking) between a therapist and five children with Autism Spectrum Disorder during dolphin-assisted therapy. Videos of the first and last dolphin-assisted therapy sessions were analyzed with regard to turn-taking between the therapist and child in the presence of a dolphin. The results show that adequate turn-taking seemed to increase over time, but mainly for children who had reasonable verbal communication skills at the start of the therapy sessions. Abstract Synchronizing behaviors in interactions, such as during turn-taking, are often impaired in children with Autism Spectrum Disorder. Therapies that focus on turn-taking generally lead to increased social skills, less interruptions, and silent pauses, however a positive non-demanding environment is therefore thought to be beneficial. Such an environment can be achieved by incorporating animals into therapy. Our study was guided by the following research questions: (1) How can we characterize the interaction between child and therapist during dolphin-assisted therapy, with regard to synchrony in verbalizations (turn-taking) and (2) does synchrony change over the course of six sessions of therapy? To answer these questions, we performed a cross-recurrence quantification analysis on behavioral data of five children, to give a detailed view of the interaction between therapist and child in the context of dolphin-assisted therapy. We were able to detect synchrony (i.e., adequate turn-taking) in all dyads, although not all children improved equally. The differences might be explained by a delayed reaction time of some children, and their level of language development.


Introduction
Children that are diagnosed with Autism Spectrum Disorder (ASD) have limitations with regard to their functional and effective communication and are impaired in initiating and sustaining reciprocal social interactions [1]. While there is large variation in the language and communication skills of children with ASD, there is often a limitation in the ability to adapt their language to the interaction Our study was guided by the following research question: How can we characterize the interaction between child and therapist during dolphin-assisted therapy, with regard to synchrony in verbalizations (turn-taking), and does synchrony change over the course of six sessions of therapy? To answer these questions, we performed a cross-recurrence quantification analysis (see methods section), to give a detailed view of the shared dynamics between the therapist and child in the context of dolphin-assisted therapy.

Ethical Considerations
In this study, five children with ASD were followed for six weeks while they received dolphin-assisted therapy. At the start of the study, written consent was obtained from the parents of the children. The parents were also informed about the aim of the study, as well as about the protocol of the therapy. It was made clear to them that they were free to stop their participation at any moment. During the research, the dolphins' welfare was continually monitored by a veterinarian. All of the necessary precautions were taken to prevent possible harm to the dolphins, such as limiting the time of interacting with people to a minimum and not allowing people to be in the water (swim) with the dolphins. The study was evaluated by the Health Care Inspectorate (IGZ) of the Dutch Ministry of Health, Welfare, and Sport, and assessed as a low-risk study.

Participants
Five children (four males, one female) with ASD participated in this study. The children were recruited through the website of the SAM Foundation, (www.stichtingsam.nl), which facilitated dolphin-assisted therapy (DAT) in the Netherlands. Parents voluntarily registrated their children for therapy after reading newspaper articles about this program. Upon registration, the parents were asked for permission to participate in the study. After registration, the children were placed on a waitlist and were called in for an extensive intake as soon as there was room to start therapy. The exclusion criteria for participation included fear of water, epilepsy, poor vision, and a history of aggressive behavior toward animals. The therapy program was partly sponsored, and therefore parents paid a reduced price.
We obtained information regarding the speech language development of the children from intake forms that were completed by the parents at the start of the therapy program, such as the number of words that they used and whether they were able to express their feelings. The parents of three children indicated that they only used single words to communicate, and the parents of the remaining two children indicated they could produce short sentences (3-4 words). Table 2 provides a detailed description of the participants.

Procedure
The therapy took place in the Netherlands at the Dolphinarium Harderwijk. The dolphins live in an artificial lagune of 7000 square meters and are not used for shows. They live together as a small pod (17 dolphins, including males, females, juveniles, and babies). The SAM foundation was responsible for the organization of the program. They provided the trained speech and language therapists. All therapists were trained at universities of applied science in the Netherlands and had followed an intensive training on how to work with the dolphins, organized by the foundation.
This professional team, consisting of three female therapists and three female dolphin trainers, guided the children through this specific DAT program. Each child worked with one therapist and one dolphin trainer during six sessions. All of the therapists had at least ten years of experience in working with children with disabilities. The therapy setting with the child, therapist, dolphin trainer, and dolphin was set up to create an atmosphere in which the child felt safe [40,47,48]. According to the literature, children feel attracted by the dolphin because of its anatomically shaped smile and its gentle movements in the water [42]. The interaction between child, dolphin, and therapist enables triangular communication (transmission triangle or carry-over) to provoke children's verbal or non-verbal responses [34]. This means that the interaction between dolphin, therapist, and child creates a triangle in which a redirection of communication via the animal is established. While exercising turn-taking behavior, the therapist teaches the child how to act toward the dolphin and in return the child will respond with attention to the therapist question.
That is, observing the animal's behavior, interacting with the animal, and watching other people interact with the animal can stimulate children's social and communication skills.
We worked with three dolphins for five children, and each child worked with the same dolphin during the sessions. At the start of a session, the dolphin trainer would call the dolphins to the raft, and one dolphin was chosen to work with during the session. After the first session, the preferred choice of dolphin was always the dolphin the child had worked with before. Due to pregnancy of one dolphin, we had to change dolphins for two children in the last session. In addition, because the dolphins were not forced to interact with the children, we only once switched dolphins when the regular dolphin did not come to the raft when the dolphin trainer called. We did not have to cancel a session due to stress signals of the dolphins. In an earlier study, the stress signals of the dolphins were observed before, during, and after therapy sessions [34]. In this study, the respiratory frequency of the dolphins was measured as a parameter of arousal. Observations were made by means of a digital camera and by an observer, who noted the breathing frequency and sounds of the dolphins. The videotape was watched afterward and all of the behaviors were recorded in an ethogram. No significant differences between therapy sessions and control sessions were found.
The dolphin trainer selected the dolphin at the start of the therapy sessions (depending on the dolphin's condition and behavior at that moment), and supervised the behavior and possible stress signals of the dolphin. Before the therapist and child wanted to engage with the dolphin, they always had to ask for permission from the dolphin trainer and no confusing situations for the dolphin would occur. Apart from that, the dolphin trainers were instructed not to interfere in the interaction between child and dolphin or therapist, unless the situation called for immediate action (e.g., when the dolphin showed stress signals or inappropriate behavior).
During the six sessions, the child, therapist, and dolphin trainer were standing or sitting on a raft at the waterfront. At the start of the session, the therapist introduced the child to the dolphin and the dolphin-trainer. Subsequently, the therapist started a conversation about the dolphin and its behavior, inviting the child to verbally express his/her observations. The children were then taught to use gestures to make the dolphin perform certain actions. For example, children learned to signal, so that the dolphin would jump or clap with the fins. The child was only allowed for making the gestures if he/she first made eye contact and verbally asked permission. In this way, the children's interactions with the therapist were reinforced by their wish to play with the dolphin. Furthermore, to help the children express their emotions, 40 cm square boards with emotion pictures were shown, and the therapist started a conversation about these emotions. As recognizing, naming, and describing emotions is something that children with ASD often have trouble with, the therapy program incorporated a task in which these skills are practiced. To include the dolphin in this task, the boards were thrown into the water for the dolphin to return.

Coding of Verbal Behavior
Videos of the first and last session of the therapy were systematically observed by means of timeserial event coding, while using the program MediaCoder, University of Groningen, Groningen, The Netherlands (a proprietary transcoding program for Microsoft Windows) [49]. This means that both the category of the behavior as well as the timing of that behavior were registered. We coded the following four behavioral categories: (1) Therapist speaks, (2) Therapist is silent, (3) Child speaks, and (4) Child is silent. Three student-raters first completed a training in which they coded one therapy session, and compared their codes with those of an expert-rater, who constructed the codebook with coding rules. Inter-rater reliability was considered to be sufficient when at least 80% of the codes of the rater and expert-rater were similar with regard to both the timing and the chosen category. If this percentage was not reached, the raters received an additional explanation of the coding rules and coded a second therapy session, after which the percentage of agreement was determined again. All of the raters reached sufficient inter-rater reliability (>80%) after coding two sessions and proceeded with the coding of the videos.

Data Analysis
We separately present the data for each participant. We first transformed the codes and accompanying times to a time series with a sampling rate of 2 Hz and then applied Cross Recurrence Quantification Analysis (CRQA; see [50][51][52]. CRQA originates from the natural sciences, and it has recently been introduced to the study of human interactions (see e.g., [53][54][55][56][57], and recently also human-animal interactions [58]. This study [58] investigated the course of behavioral synchrony on movement patterns (and the outcome on problem behaviors) of children with Down syndrome (DS) and children with ASD while interacting with a dog. We used CRQA to operationalize the movement synchrony between a dog and child and found an increase in the coupling between child and dog during the final session, which means that the child and dog became mutually attuned in their movements.
CRQA focuses on the shared dynamics of two coupled systems by detecting repeatedly occurring matches between the behavior of two interaction partners [59]. In this case, matching behavior was defined as correct turn-taking behavior in the conversation between child and therapist, that is, speaking when the other person is silent, and vice versa [16,60]. CRQA detects these matches across all possible timescales, which range from seconds to the duration of the entire interaction [59]. This means that, even when the appropriate turn is (slightly) delayed, it will still be detected.
It is possible to calculate several measures from CRQA. We looked at the Diagonal Cross-Recurrence Profile (DCRP; see e.g., [59,[61][62][63]), allowing for leader-follower analysis of the interaction. Figure 1 shows the average DCRP of the first session for all five children, as an example and to highlight the main measures. Concretely, we focused on matching behaviors (adequate turn-taking) within an interval of 30 s around the Line of Synchrony (LOS). The LOS represents instances of matching behavior of child and therapist at the exact same time. This either means that the child speaks while the therapist remains silent, or vice versa. The percentage of matching behavior at the exact same moment (i.e., on the LOS) is called the percentage of synchrony (% Sync). Yet, the patterns of matching behavior do not always occur at the exactly same moment, there is often some delay [28]. Therefore the recurrence rate (RR) represents the proportion of matching behavior within an interval of 30 s around the LOS (15 s before/after the LOS). The highest proportion of matching behaviors that can be measured within that interval is called RR peak . A final measure that can be derived from the DCRP is Q los , which is the proportion of matching behavior on the left side of the LOS divided by the proportion of matching behavior on the right side. If Q los is higher than 1, the therapist temporally leads the interaction. This means that the behavior of the therapist (speaking or listening) influences the child's behavior to a greater extent than vice versa. If Q los is smaller than 1, then the child temporally leads the interaction. proportion of matching behaviors that can be measured within that interval is called RRpeak. A final measure that can be derived from the DCRP is Qlos, which is the proportion of matching behavior on the left side of the LOS divided by the proportion of matching behavior on the right side. If Qlos is higher than 1, the therapist temporally leads the interaction. This means that the behavior of the therapist (speaking or listening) influences the child's behavior to a greater extent than vice versa. If Qlos is smaller than 1, then the child temporally leads the interaction. The Qlos measure can be calculated by dividing the proportion of recurrence on the left side of the LOS by the proportion of recurrence on the right side. Table 2 shows descriptive statistics rlatad to the CRQA analysis. Below, the results will be separately discussed for each participant.  The Q los measure can be calculated by dividing the proportion of recurrence on the left side of the LOS by the proportion of recurrence on the right side. Table 3 shows descriptive statistics rlatad to the CRQA analysis. Below, the results will be separately discussed for each participant.

Participant 1
According to her parents, participant 1 has good verbal skills. She speaks a lot, but does make grammatical errors. Formulating a coherent sentence takes more time, and she has difficulty expressing and interpreting emotions and feelings. She feels comfortable around animals.

Analysis of Turn-Taking
During the first session, we recorded 232 s total time of spoken language, of the participant and 542 s of spoken language of the therapist. During the last session, the seconds of spoken language of the participant increased to 278 (19.8%) and decreased for the therapist to 490 (9.8%). Figure 2 shows the Diagonal Cross-Recurrence Profile (DCRP) of participant 1 during the first and final session. The recurrence rate increased from 0.18 to 0.26. This means that the proportion of behavioral matches (i.e., adequate turn-taking) increased with 44.4%. The RR peak measure, which is the highest proportion of recurrence observed, also showed an increase, from 0.24 to 0.29. The Q los measures show that, during both sessions, the child more often temporally led the conversation than the therapist (both values < 1). When compared to the first session (Q los = −0.03), the dominance of the child's behavior in the interaction appeared to be slightly less (Q los = −0.02).

Analysis of Turn-Taking
During the first session, we recorded 232 s total time of spoken language, of the participant and 542 s of spoken language of the therapist. During the last session, the seconds of spoken language of the participant increased to 278 (19.8%) and decreased for the therapist to 490 (9.8%). Figure 2 shows the Diagonal Cross-Recurrence Profile (DCRP) of participant 1 during the first and final session. The recurrence rate increased from 0.18 to 0.26. This means that the proportion of behavioral matches (i.e., adequate turn-taking) increased with 44.4%. The RRpeak measure, which is the highest proportion of recurrence observed, also showed an increase, from 0.24 to 0.29. The Qlos measures show that, during both sessions, the child more often temporally led the conversation than the therapist (both values < 1). When compared to the first session (Qlos = −0.03), the dominance of the child's behavior in the interaction appeared to be slightly less (Qlos = −0.02).

Participant 2
As reported by his parents, participant 2 started speaking when he was three years old. His language development is delayed and he has difficulty producing meaningful sentences and expressing his emotions.

Analysis of Turn-Taking
During the first session, we recorded 199 s of spoken language of the participant and 599 s of

Participant 2
As reported by his parents, participant 2 started speaking when he was three years old. His language development is delayed and he has difficulty producing meaningful sentences and expressing his emotions.

Analysis of Turn-Taking
During the first session, we recorded 199 s of spoken language of the participant and 599 s of the therapist. During the last session, the number of participants seconds of spoken language decreased to 138.5 (30.4%) and decreased for the therapist to 450 s (24.8%). The Diagonal Cross-Recurrence Profile (Figure 3) shows the recurrence rate, the proportion of behavioral matches (i.e., adequate turn-taking), slightly increased from 0.17 to 0.19. The highest proportion of recurrence (RR peak ) increased from 0.23 to 0.24. The Q los changed from −0.066 to −0.032. This means that the child temporally led the interaction, but that his dominance became less apparent over time. Overall, one could conclude that the conversational dynamics in terms of these measures did not significantly change between the two sessions.

Participant 3
According to his parents, participant 3 is able to speak, but it is hard for him to communicate with the right words. He has weak mouth motor skills, which results in difficulty to formulate sounds with his mouth and slow speech. He uses single words and sometimes 3-4-word sentences, with regular echolalia. He is quickly over-stimulated and is only able to communicate with supportive gestures. His parents often let him finish sentences, such as: "We go to" "the dolphins".

Analysis of Turn-Taking
This participant's total time of spoken language increased from 113.5 to 126 s (11.0%). Additionally, for the therapist, this increased from 562.5 to 605.5 s (7.6%). As Figure 4 shows, the recurrence rate remained relatively stable (a slight increase from 0.08 to 0.09). RRpeak also remained stable (a slight increase from 0.11 to 0.12). Qlos increased from 0.005 to 0.02, which means that the therapist temporally led the interaction, and even more so during the final session. As for participant 3, also for this child, the conversational dynamics in terms of these measures remained relatively stationary.

Participant 3
According to his parents, participant 3 is able to speak, but it is hard for him to communicate with the right words. He has weak mouth motor skills, which results in difficulty to formulate sounds with his mouth and slow speech. He uses single words and sometimes 3-4-word sentences, with regular echolalia. He is quickly over-stimulated and is only able to communicate with supportive gestures. His parents often let him finish sentences, such as: "We go to" "the dolphins".

Analysis of Turn-Taking
This participant's total time of spoken language increased from 113.5 to 126 s (11.0%). Additionally, for the therapist, this increased from 562.5 to 605.5 s (7.6%). As Figure 4 shows, the recurrence rate remained relatively stable (a slight increase from 0.08 to 0.09). RR peak also remained stable (a slight increase from 0.11 to 0.12). Q los increased from 0.005 to 0.02, which means that the therapist temporally led the interaction, and even more so during the final session. As for participant 3, also for this child, the conversational dynamics in terms of these measures remained relatively stationary.
Additionally, for the therapist, this increased from 562.5 to 605.5 s (7.6%). As Figure 4 shows, the recurrence rate remained relatively stable (a slight increase from 0.08 to 0.09). RRpeak also remained stable (a slight increase from 0.11 to 0.12). Qlos increased from 0.005 to 0.02, which means that the therapist temporally led the interaction, and even more so during the final session. As for participant 3, also for this child, the conversational dynamics in terms of these measures remained relatively stationary.

Participant 4
Participant 4 speaks unclearly according to his parents and rather points to what he wants. He speaks with single and simple words, or very short sentences in combination with gestures.

Analysis of Turn-Taking
There is a minor decrease in the participant's total time of spoken language from 176 to 174.5 s (0.8%). The therapist's total time of spoken language increased from 854 to 943 s (10.4%). The DCRP of participant 4 ( Figure 5) showed lower RR peak and RR los during the final session as compared to the first session, with RR peak decreasing from 0.19 to 0.07 and RR los decreasing from 0.06 to 0.04. Q los changed from −0.02 during the first session to −0.01 during the last session. This means that the child temporally led the interaction, but that his dominance became less apparent over time.

Participant 4
Participant 4 speaks unclearly according to his parents and rather points to what he wants. He speaks with single and simple words, or very short sentences in combination with gestures.

Analysis of Turn-Taking
There is a minor decrease in the participant's total time of spoken language from 176 to 174.5 s (0.8%). The therapist's total time of spoken language increased from 854 to 943 s (10.4%). The DCRP of participant 4 ( Figure 5) showed lower RRpeak and RRlos during the final session as compared to the first session, with RRpeak decreasing from 0.19 to 0.07 and RRlos decreasing from 0.06 to 0.04. Qlos changed from −0.02 during the first session to −0.01 during the last session. This means that the child temporally led the interaction, but that his dominance became less apparent over time.

Participant 5
The fifth participant has trouble making certain sounds and is therefore not always well understood, but has a good vocabulary according to his parents. He uses short sentences of three to four words and supports this with gestures.

Analysis of Turn-Taking
We recorded 153 s of spoken language of the participant and 509 s of the therapist during the first session. During the last session, the number of participant's seconds decreased to 34.5 (77.5%) and also for the therapist to 276.5 (45.6%). Figure 6 shows the DCRP of this participant during the first and final session. The RRpeak and RRlos decrease during the final session, respectively, from 0.2 to 0.08, and from 0.16 to 0.04, respectively. The Qlos is higher during the final session and it changes

Participant 5
The fifth participant has trouble making certain sounds and is therefore not always well understood, but has a good vocabulary according to his parents. He uses short sentences of three to four words and supports this with gestures.

Analysis of Turn-Taking
We recorded 153 s of spoken language of the participant and 509 s of the therapist during the first session. During the last session, the number of participant's seconds decreased to 34.5 (77.5%) and also for the therapist to 276.5 (45.6%). Figure 6 shows the DCRP of this participant during the first and final session. The RR peak and RR los decrease during the final session, respectively, from 0.2 to 0.08, and from 0.16 to 0.04, respectively. The Q los is higher during the final session and it changes from negative to positive (from −0.04 to 0.04), which means that, during the first session, the child temporally led the interaction, while the therapist followed, whereas the therapist temporally led during the final session.

Discussion
Therefore, a positive non-demanding environment is thought to be beneficial to increase synchronizing verbal behavior, such as turn-taking, for children with ASD [29][30][31]. Such an environment can be created by incorporating animals into therapy situations [34,43]. However, until now, studies have not investigated whether synchronizing behaviors, and especially turn-taking, increase in the context of animal-assisted therapy (AAT). Therefore, this study examined videos of the interaction between child and therapist during dolphin-assisted therapy (DAT) in terms of turn-taking, while using data of five child-therapist interactions during both the first and final (sixth) session of DAT.
Because this exploratory study intended to investigate whether DAT influences changes in turn-taking behavior, it was beyond our scope to compare the effect of the therapy to a control group. However, in an earlier study [34], we reported the positive effect of this program on verbalization and social interaction measures, by comparing DAT to two control groups (using a radio-controlled boat in a swimming pool, and a waitlist). As advised by Marino [64] we minimized construct confounding, as we exposed the control group and the DAT intervention group to highly similar procedures.
However, the results of this study seemed only partially in line with our expectations that synchronization of verbalizations (i.e., turn-taking behavior) would improve during DAT. The data revealed increases in turn-taking for participants 1 and 2. They showed an increase in RRpeak and RRlos, which implies that the proportion of "matching behaviors" (i.e., speaking while the other person remains silent, and vice versa) between the participant and the therapist increased from the first to the last session. The Qlos measures for participants 1 and 2 indicate that these children temporally led the interaction during both the first and last session, but that they became less dominant leaders during the final session. However, participant 3 showed no changes in RRpeak between the first and the last DAT session. The Qlos measures indicate that the therapist temporally led the interaction, and even more so during the final session. The final two participants (4 and 5) showed a decrease in synchronization of verbalization (verbal turn-taking behavior). Their RRpeak and RRlos measures decreased between the first and the sixth DAT session. The Qlos measures revealed that participant 4 temporally led the interaction during the first session, but this became

Discussion
Therefore, a positive non-demanding environment is thought to be beneficial to increase synchronizing verbal behavior, such as turn-taking, for children with ASD [29][30][31]. Such an environment can be created by incorporating animals into therapy situations [34,43]. However, until now, studies have not investigated whether synchronizing behaviors, and especially turn-taking, increase in the context of animal-assisted therapy (AAT). Therefore, this study examined videos of the interaction between child and therapist during dolphin-assisted therapy (DAT) in terms of turn-taking, while using data of five child-therapist interactions during both the first and final (sixth) session of DAT.
Because this exploratory study intended to investigate whether DAT influences changes in turn-taking behavior, it was beyond our scope to compare the effect of the therapy to a control group. However, in an earlier study [34], we reported the positive effect of this program on verbalization and social interaction measures, by comparing DAT to two control groups (using a radio-controlled boat in a swimming pool, and a waitlist). As advised by Marino [64] we minimized construct confounding, as we exposed the control group and the DAT intervention group to highly similar procedures.
However, the results of this study seemed only partially in line with our expectations that synchronization of verbalizations (i.e., turn-taking behavior) would improve during DAT. The data revealed increases in turn-taking for participants 1 and 2. They showed an increase in RR peak and RR los , which implies that the proportion of "matching behaviors" (i.e., speaking while the other person remains silent, and vice versa) between the participant and the therapist increased from the first to the last session. The Q los measures for participants 1 and 2 indicate that these children temporally led the interaction during both the first and last session, but that they became less dominant leaders during the final session. However, participant 3 showed no changes in RR peak between the first and the last DAT session. The Q los measures indicate that the therapist temporally led the interaction, and even more so during the final session. The final two participants (4 and 5) showed a decrease in synchronization of verbalization (verbal turn-taking behavior). Their RR peak and RR los measures decreased between the first and the sixth DAT session. The Q los measures revealed that participant 4 temporally led the interaction during the first session, but this became less apparent during the final session. Participant 5 also led the interaction during the first session, but this was reversed in the final session, when the therapist mostly led the interaction.
For three out of five children, the total time of spoken language increased during the final session of DAT (see Table 3), while the total time of spoken language of the therapist measured in seconds decreased.
Some authors have suggested that turn-taking behavior requires a reasonable good vocabulary [31]. Taking an active role both as speaker-initiator or listener requires sufficient communication skills while children with disabilities mostly lack such skills [65]. Therefore, some differences in our results can be explained by initial difference in communication skills between the children. While participants 1 and 2 had less trouble speaking, the parents of participants 3, 4, and 5 reported that their children were only able to speak while using supportive gestures (i.e., substituting words for gestures, frequent use of pointing instead of verbal indications). Participants 1 and 2, with better communication skills, indeed seemed to improve with regard to their turn-taking skills, which confirms the suggestion of Yoder and Stone [31] that a reasonable vocabulary is a prerequisite for adequate turn-taking. Future studies on animal-assisted interventions and turn-taking are therefore advised to take a baseline level of verbal skills into account, e.g., using videotapes of the participants prior to the study.
To our knowledge, this study is the first to investigate the effect of DAT on verbal synchronizing behaviors (turn-taking). While a bigger sample would aid in the generalization of our findings, our small sample size enabled us to perform an in-depth analysis. The most important result revealed in this study, that is, turn-taking during DAT seems to increase for children with reasonable verbal communication skills, could be further examined in larger studies, which could also include other variables, such as the age of the children, the severity of the disorder, or gender.
Note that the five children in this study worked with three different therapists, and that the dolphin changed depending on its condition and availability. Although the same therapist always guided each child during the sessions, differences in the therapists' behaviors might affect children's synchronizing behaviors and turn-taking skills. Given that this was an exploratory study (do we see any changes in synchrony?), and because of the labour-intensive nature of the coding, this study only compares the child-therapist synchrony in two sessions. Future research could investigate more sessions to track the development of therapist-child synchrony over time. Moreover, differences in the dolphin's behavior could have influenced the communication between the therapist and child, and the child's experience of the therapy setting. Although this might be hard to incorporate in practice, future studies are advised to let the same therapist guide all children, and to use the same dolphin throughout the sessions. Alternatively, therapists could receive more extensive instructions with regard to what they say to the participants and instructions with regard to taking delayed reaction times into account.
There is an indication that children with ASD show additional attention deficits and a relatively delayed reaction time when compared to typically developing children. This may result in the need for more time to verbally respond to others [66,67]. These attentional deficits and delayed reaction time were not taken into account in this study, but may have influenced the results on turn-taking behavior. For example, when a question takes more time to answer for the child, and the therapist is already moving on, a turn is "missed". According to Levinson [26], "turn-taking precedes language in ontogeny, but when language is acquired children struggle for years to squeeze complex language into short turn sizes with adult response time." This skill might require even more time to master for children with ASD.
In future studies, the effect of Animal-Assisted interventions on synchrony could be further investigated, while using conditions of therapy with and without animals, or by comparing synchrony between child and therapist in the presence of different animals, such as dogs or horses. This will allow us to check the synchrony of turn-taking behavior with and without an animal and compare the different species. The results of this research can be useful in future research, because it makes clear that there is a great variation in language development in children with ASD. Finally, future studies could also investigate the optimal duration of DAT programs. It could be that a longer DAT program yields better results for children with weaker verbalization skills.

Conflicts of Interest:
In accordance with the journals policy and our ethical obligation as researchers, we are reporting that during the research the first author was president of the SAM Foundation, a non-profit foundation offering dolphin assisted therapies in the Netherlands for children with Down Syndrome and children with Autism Spectrum Disorder. The SAM Foundation may be affected by the research reported in the enclosed paper in a non-financial or non-commercial manner. By not interfering in data collection (carried out by professional therapists) and by having an independent third party to do the statistical analyses, we believe we have done our utmost to secure scientific objectivity of design, procedures and results under all circumstances.