Establishing Multiple-Control Responding of Children with Autism to People and Emotions in Context by Utilizing Derived Stimulus Relations

Abstract

This study evaluated the efficacy of a set of procedures in evoking responding under multiple stimulus control (identifying emotions when provided a person and context) in a match-to-sample arrangement. Three participants with autism achieved a mastery criterion following direct training of the target relations, and two of the participants additionally demonstrated derived relations without direct training (identifying people when provided an emotion and context). Corrective procedures were effective in promoting the emergence of derived relations for the third participant. These data suggest that incorporating derived stimulus relation training and testing procedures may have utility for clinical interventions in children with autism.

Responding to a single stimulus rather than multiple compound stimuli in a match-to-sample arrangement is referred to as stimulus overselectivity (Dube, 2009), and is common in children with autism at all levels of intellectual functioning (Brown & Bebko, 2012). If a participant is provided an array of pictures including a red scarf, a blue scarf, a red hat, and a blue hat, and is provided the sample vocal stimulus “which one is a blue scarf”, then a correct response necessitates multiple stimulus control exerted by both the utterance “blue” and the utterance “scarf”, where restricted stimulus control exerted by either of these two terms alone would result in correct responding on only 50% of occasions. Three early studies demonstrated the efficacy of multiple-cue training in teaching participants to select AB compound stimuli given comparison AC and A stimuli (Allen & Fuqua, 1985; Koegel & Schreibman, 1977; Schreibman, Charlop, & Koegel, 1982), and these results have been extended in further evaluations (e.g., Farber, Dube, & Dickson, 2016). A more complex topography of responding under multiple control may occur when aspects of the response are derived rather than directly trained. For example, if a participant is provided reinforcement for identifying how a [Person; A] feels in a given [Context; B] (AB = select feeling C), the person may derive who feels [Emotion; C] in a given [Context; B] (CB = select person A). A correct response necessitates attending to both stimuli present in the compound stimulus, as a person may feel differently in various contexts, and two different people may feel differently within the same context. Overselectivity in the above example would occur when either the subject or the context (compound stimulus) exerted restricted control over the response, or when derived contextually controlled responding failed to occur, both of which may partially account for the inflexible responding exhibited by individuals with autism (e.g., the belief that all people feel happy in all contexts).

The present study evaluated a set of procedures for teaching compound discriminations under multiple control in a match-to-sample arrangement with three individuals with autism. The procedures were adapted from the Promoting the Emergence of Advanced Knowledge – Equivalence Module (PEAK-E; Dixon, 2015), a curriculum designed to develop pivotal language skills. Procedures from PEAK-E were used because prior research has supported the application of PEAK-E as a technology for developing verbal skills in children with autism (e.g., Dixon, Belisle, Stanley, Daar, & Williams, 2016), as well as to aid in replication of the procedures. In addition to evaluating the development of directly trained discrimination responses, we evaluated the derivation of untrained relations, consistent with a stimulus equivalence account of verbal behavior development.

Method

Participants, Setting, and Materials

Two children and one adolescent with autism participated in the study. The study took place at a Midwestern US school specialized for school-age children and adults with autism and related disabilities. Individualized Education Program plans were reviewed to ensure that all participants were receiving specialized services due to an autism diagnosis. All participants were able to (a) vocally read all textual stimuli used in the study (e.g., MALL) and (b) correctly state an emotion (e.g., “happy”) when shown a corresponding picture (e.g., smiling face) prior to the study. Participants’ derived relational repertoire was evaluated using the PEAK-E pre-assessment (PEAK-E-PA; Dixon, 2015), which provides a measure of participants’ ability to derive arbitrary reflexive, symmetrical, transitive, and equivalent relations. Scores on the PEAK-E-PA range from 0 to 48. Joshua was a 17-year-old boy with an IQ of 82 (WISC-IV Short Form) and a PEAK-E-PA score of 41, suggesting he had the ability to demonstrate reflexive, symmetrical, transitive, and basic equivalence relations. Cade was a 12-year-old boy with an IQ of 80 (WISC-IV Short Form) and a PEAK-E-PA score of 18, suggesting he had the ability to demonstrate reflexive as well as basic and intermediate symmetrical relations. Finally, Marita was a 12-year-old girl with an IQ of 62 and a PEAK-E-PA score of 4, suggesting she had the ability to demonstrate basic reflexive relations. Sessions were conducted in a therapy room separate from the other students, which contained stimuli for training, along with a desk and two chairs. Stimuli for training were those specified in PEAK-E program 13C–Equivalence: Feelings in Context, including pictures of unknown people, pictures of facial expressions, and textual stimuli. The pictures of the unknown people were all presented in side view, as shown in Table 1, to ensure that the pictures did not contain a specific facial expression that could have biased responding.

Table 1.

Relational responses trained and tested in the study. Full arrow shows directly trained relations and dashed arrow shows derived relations

Design and Procedure

A multiple-baseline with an embedded multiple-probe design was used to evaluate the efficacy of the procedures described above. Stimuli in the study were arranged in four 3-member classes (A1–3, B1–3, C1–3, D1–3), where A stimuli were pictures of people (A1 = Billy, A2 = Jimmy, A3 = Suzy), B stimuli were pictures of facial expressions (B1 = happy face, B2 = sad face, B3 = bored face), C stimuli were textual stimuli that corresponded with the facial expressions (C1 = HAPPY, C2 = SAD, C3 = BORED), and D stimuli were textual contexts in which the emotions may occur (D1 = MALL, D2 = PARK, D3 = HOME). There were two dependent variables in the study, including the correct selection of an emotion (B) given the conditional multiple stimulus of a person (A) in a given context (D) [DA-B; D1A1-B1, D2A1-B2, D3A1-B3, D1A2-B2, D2A2-B3, D3A2-B1, D1A3-B3, D2A3-B1, D3A3-B2; e.g., How does Susy (A3) feel at the mall (D1)? Correct response: bored (B3)], as well as the correct identification of a person (A) given the conditional multiple stimulus of an emotion (C) and a given context (D) [DC-A; D1C1-A1, D2C2-A1, D3C3-A1, D1C2-A2, D2C3-A2, D2C1-A2, D1C3-A3, D2C1-A3, D3C2-A3; e.g., Who feels bored (C3) at the mall (D1)? Correct response: Suzy (A3)]. For both variables, each permutation was used to ensure that correct responding was not under the restricted stimulus control of either compound stimulus. A summary of the trained and derived relations in the study are in Table 1. Correct responding was reported as a percentage within 10-trial blocks (correct/10, multiplied by 100). Ten-trial blocks were used to remain consistent with procedures described in the PEAK-E curriculum to aid in clinical replication.

In the baseline phase, DA-B and DC-A probes were conducted in an alternating pattern to determine whether the participants were able to demonstrate correct responding prior to training due to a pre-experimental history. All testing was conducted in 10-trial blocks, where the stimulus classes presented during each trial were randomly selected. The number of probes was staggered across participants. For DA-B testing, participants were provided a sample picture of a person (A), a sample textual context (D), and an array of three pictures of facial expressions (B). The experimenter then presented the vocal stimulus, “How does (A) feel at (D)?”, where a response was considered correct if the participant touched or vocally stated the corresponding emotion (B). For DC-A testing, participants were provided a sample textual emotion (C), a sample textual context (D), and an array of three pictures of people (A). The experimenter then presented the vocal stimulus, “Who feels (C) at the (D)?”, where a response was considered correct if the participant touched the correct picture or vocally stated the corresponding name (A). Contingent praise and differential prompting strategies were not used in the baseline phase.

Only DA-B responses were evaluated throughout the DA-B training phase. The stimulus presentation was identical to the baseline phase, and training was conducted in 10-trial blocks. Contingent praise was identified as a potentially reinforcing stimulus by the classroom teachers and paraprofessionals for each of the participants, and was delivered contingent on correct responding on a continuous reinforcement schedule during the training phase. When participants failed to demonstrate a correct response within 5 s of the delivery of the vocal discriminative stimulus, a least-to-most restrictive prompting hierarchy was used. Following the first error, the experimenter stated, “not that one, try again”, and following the second error, the experimenter pointed to the correct stimulus in the array. The discriminative stimulus was not represented. No additional prompts were required to evoke the correct response in the study. For two of the participants, a shaping procedure was used because increases in correct responding were not observed in the preceding five trial blocks. The shaping procedure involved training only two stimulus classes in the initial trial block, and reintroducing a single additional stimulus class in each of the subsequent trial blocks. Therefore, the shaping procedure occurred across nine trial blocks. The same reinforcement and prompting strategies were used in the shaping procedure, and the regular procedure was reintroduced following shaping. A mastery criterion of three consecutive trial blocks with at least 90% correct responding was adopted for the study. Following mastery of the trained relations, a DC-A test probe was conducted to determine whether participants were able to derive the DC-A response in the absence of direct training. One of the participants failed to demonstrate an increase in correct DC-A responses, so three exemplar classes were directly trained (i.e., differential contingent praise and prompting) in three 10-trial blocks (D1C1-A1, D2C3-A2, D3C2-A3), where the presentation of each class was randomized within each block. Afterward, a final DC-A test probe was conducted. Interobserver agreement (IOA) was assessed for 33% of the trials by comparing the scores of two observers for each trial (agreements divided by total number of trials, multiplied by 100), and IOA in the study was 98%.

Results and Discussion

The results of the study are summarized in Fig. 1. In the baseline phase, none of the participants consistently demonstrated correct responding, with mean scores of 20%, 30%, and 25% for DA-B tests probes and 20%, 40%, and 33.3% for DC-A test probes, for Joshua, Cade, and Marita, respectively (a score of 33.3% would be expected given chance alone). Joshua’s scores increased to a mean of 77.8% during DC-A training, and the mastery criterion was achieved on the ninth trial block. In addition, Joshua’s scores on the DC-A test probe increased to 80% following training of the DA-B relation. Cade’s scores also increased in the training phase to a mean of 65.3%; however, his scores failed to show an increasing trend following the 25th trial block. Cade’s mean score increased throughout the shaping procedure, and he achieved the mastery criterion on the 44th trial block. In addition, Cade’s score on the DC-A test probe also increased to a score of 80%. Finally, Marita’s scores increased to a mean of 64.8%, but she also failed to show an increasing trend following the sixth trial block. Her scores increased throughout the shaping procedure, and she was able to achieve the mastery criterion on the 29th trial block. During the DC-A testing phase, Marita failed to demonstrate an increase in correct responding. During exemplar class training of the DC-A relations, Marita’s mean correct responding increased to 76.7%, and in the final DC-A test probe inclusive of all stimulus classes, her correct responding increased to 100%. Slightly different training arrangements were required to meet the terminal performance in this study, complicating the experimental design. Although this is not uncommon in behavioral research with children with autism, experimental control was demonstrated in this study, in that pre-shaping responding during training for Cade served as a control for Josh’s responding in training, and Marita’s responding during training (pre-shaping) served as a control for both Cade and Josh during training (pre-shaping), and her responding during shaping served as a control for Josh’s responding during shaping only.

Fig. 1.

Multiple baseline across subjects with an embedded multiple probe. Data paths show where direct training occurred

Together, the results suggest that the procedures were efficacious in bringing the responding of all three participants under the multiple control of compound stimuli. Following training, the participants were able to correctly identify people’s emotions when provided a sample person and context, and two of the three participants were able to demonstrate the derived response of identifying people when provided with a sample emotion and context, in the absence of direct training. A limitation of the obtained results may be that there were inconsistencies in the amount of training required for participants to obtain mastery, as Joshua required only seven trial blocks, compared to 45 for Cade and 30 for Marita, both of whom also required a shaping procedure embedded in the training phase. Differences in the amount of training required to obtain mastery may be accounted for by differences in the pre-experimental repertoires of the participants prior to the study, as Joshua was older than the other participants and had the highest IQ and PEAK-E-PA scores at the onset of the study. Differences in the participants’ pre-experimental relational abilities may also account for Marita’s inability to demonstrate the derived relation following DA-B training alone. Direct training of three of nine exemplar classes was, however, efficacious in increasing Marita’s percentage of correct responding once the remaining classes were reintroduced, and she achieved a higher percentage of correct responding than either Joshua or Cade. Although this limits the experimental design logic of the study, it does provide case examples of ways to modify instructional procedures to produce mastery that are consistent with basic applied behavior analytic procedures already supported in prior research (i.e., shaping, multiple exemplar instruction). Additional limitations in the current investigation include a small number of test probes in the baseline phase and the use of pictures and textual stimuli in addition to vocal discriminative stimuli. Few test probes were conducted due to the arbitrary nature of the tested relations, where a prior history with the stimuli was unlikely, to avoid testing effects in the subsequent training phase. In addition, picture stimuli were included so that the experimenter could prompt correct responding when it did not occur independently, but we are unsure whether participants responded to the vocal compound stimulus, the picture and textual compound stimulus, or some combination thereof. Similarly, we accepted either a selection response or vocal response as a correct response throughout the study. This was an artifact of the simultaneous vocal-visual sample stimulus; however, we do not necessarily know whether, given a selection response (listener), the participant could provide the vocal response (speaker), or whether, given the vocal response (speaker), the participant could also provide the selection response (listener). In addition to addressing these limitations, future research may evaluate how such training generalizes to non-experimental contexts, as well as to corresponding changes in standardized measures of symptoms common across individuals with autism.

Funding

No external funding was utilized for completion of this research.

Compliance with Ethical Standards

Conflict of Interest

Mark R. Dixon receives small royalties from the sales of the PEAK curriculum. All other authors declare that they have no conflict of interest.

Ethical Approval

All procedures performed in this study involving human participants were in accordance with the ethical standards of the institution and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Informed Consent

Informed consent was obtained for participation in this research.