Participant, Setting, and Materials

Alice was a 6-year-old girl with autism spectrum disorder (ASD). She had previously received behavior analytic services in a clinical setting for approximately two years, although these services were disrupted by the COVID-19 pandemic. The current study was executed as part of a university-based research program and not during clinical services. Alice’s performance on the Expressive Vocabulary Test – 2nd Edition (Williams, 2007) and Peabody Picture Vocabulary Test – 4th Edition (Dunn & Dunn, 2007) was at the 4th and 16th percentile, respectively. Alice’s performance on the Verbal Behavior Milestones Assessment and Placement Program (VB-MAPP; Sundberg, 2014) was consistent with an emerging level 3 learner. She communicated using full sentences, although her intraverbal repertoire was less developed compared to tact and listener response domains. Alice was included in this study as she exhibited deficits in describing complex scenes and past events. For example, when shown a brief video clip, Alice would only tact the movie or character name. This protocol aligned with the caregivers’ reported goals for Alice to share information about others. At the time of the study, she could accurately respond by listing three characteristics of herself and the first author when asked “tell me about you” or “tell me about [author].” Alice showed a pronounced interest in characters from animated movies, so these targets were included in the current study (see Table 1).

Table 1 Target list and alphanumeric notation

All sessions took place in an individual room in a university-based laboratory. The room was approximately 2.6 m x 2 m and included a child-sized table and chairs, toy shelf, and one-way mirror. Window blinds were pulled over the one-way mirror to reduce distraction. Alice sat diagonally from the experimenter and a secondary observer sat behind Alice or in the adjoining observation room. An additional playroom, approximately 5.9 m x 3.4 m, was used for longer breaks throughout her appointment and included several playsets, games, and a large table.

Stimuli included pictures of cartoon characters on a white background printed on 5.1 cm x 7.6 cm laminated cards. These stimuli were used as tact prompts during training. A laminated token board was present on the table during all sessions.

Dependent Variables and Response Measurement

Unprompted correct responses were defined as Alice emitting the target response within 5 s of the antecedent verbal stimulus. Prompted correct responses were defined as Alice emitting the target responses within 5 s of the presentation of a prompt. Each target response was defined before the study and Alice was required to emit the entire response. Target responses were single words except during divergent intraverbal probes (see below), during which Alice was required to emit the entire response as it was presented during training. For example, when instructed “tell me about Shrek,” Alice was required to respond, “He has a friend named Donkey and is an ogre.” The primary dependent measure was the percentage of unprompted correct responses, which was calculated by dividing the number of unprompted correct responses by the total number of trials within a session, multiplied by 100. The mastery criterion was set at two consecutive sessions with 100% unprompted correct responses.

Preference Assessment and Token Economy

Before participating in the current study, a Reinforcer Assessment for Individuals with Severe Disabilities (Fisher et al., 1996) was conducted with Alice’s caregiver. A Multiple Stimulus Without Replacement (DeLeon & Iwata, 1996) preference assessment was then conducted. Items identified through these assessments were available in the experimental room, and Alice was allowed to mand for items during breaks following each session.

A token board was used during all sessions. Alice had used this same token system during all aspects of her instructional programming for more than one year at the time of this study. The number of tokens required to complete the board varied across sessions depending on the number of trials in the session (i.e., if a session had six trials, token-exchange opportunities were made available after six tokens were accrued). After the token board was filled, Alice received a 2-min break and access to a preferred tangible item. Preferred tangible items were restricted to breaks following token-exchange opportunities and could not be removed from the session room to reduce the likelihood of satiation.

Experimental Design

A multiple-probe design (Horner & Baer, 1978) across four instructional sets was used to evaluate the effects of compound intraverbal training on unprompted correct responding. After responding met the mastery criterion in one panel, post-training probes were conducted with the mastered targets and pre-training probes were conducted for the targets in the subsequent panel before the independent variable was applied. Once responding in all four sets met the mastery criterion, remedial training was conducted (see Table 2 for the training sequence).

Table 2 Sequence of training and probe sessionsGeneral Procedure

Each target set included two compound intraverbal targets. During initial training, compound antecedent verbal stimuli included two components: “Who has a friend named [character’s friend] and is a [character type/occupation].” Table 1 shows the target list and alpha-numeric notation. To begin all sessions, the experimenter secured ready behavior (i.e., looking at the experimenter, hands on the table, and sitting in her seat) before presenting the target antecedent verbal stimulus.

Baseline

During baseline procedures, the experimenter presented the antecedent verbal stimulus and Alice was given 5 s to respond. Responses produced no differential consequences. Mastered demands were interspersed after an average of three trials. Correct responses to mastered demands produced a token and praise. All mastered demands were previously targeted or tested relations to which Alice responded correctly during greater than 90% of opportunities.

Compound Intraverbal Training

Training was conducted using a constant prompt delay procedure and tact prompts. The first two sessions were conducted at a 0-s prompt delay. The experimenter presented the compound antecedent verbal stimulus and immediately presented the tact prompt. Prompted correct responses produced praise and a token. Prompted incorrect responses resulted in the re-presentation of the compound antecedent verbal stimulus and the tact prompt at a 0-s prompt delay until a prompted correct response occurred. All subsequent training sessions were conducted at a 5-s prompt delay. Unprompted and prompted correct responses produced praise and a token. This relation is labeled as the baseline relation below.

Stimulus Control Probes

Once the training criterion was met for the baseline relation, stimulus control probes were conducted using procedures identical to baseline for that set. The target relations are shown in Table 2, and an example of the Set 1 baseline relations and subsequent probes are shown in Fig. 1. Each component of the compound antecedent verbal stimulus was presented alone to identify control by individual elements (e.g., “Who has a friend named Gus Gus?”). Probes of the reverse relation were also conducted, which involved the presentation of the same components of the antecedent verbal stimulus presented in reverse order (e.g., “Who is a maid and who has a friend named Gus Gus?”). These were included to assess for irrelevant stimulus control topographies, such as the order of stimulus elements. Divergent intraverbal (DIV) probes were also conducted and served as the reverse intraverbal (i.e., symmetrical) relation to the baseline relation. The experimenter presented the antecedent verbal stimulus “tell me about [character]” and Alice was required to respond with both trained elements in any order (e.g., “She has a friend named Gus Gus and is a maid”). Competing (COMP) relation probes served as an assessment of stimulus competition. In these trials, a component from each trained stimulus was presented in a stimulus compound. For example, “who has a friend named Gus Gus (A1) and who is an ogre (B2)?” Alice’s response (i.e., Cinderella [C1] or Shrek[C2]) was recorded. Although she could have emitted both responses (e.g., “Cinderella and Shrek”), this was never observed.

Fig. 1

Set 1 targets and probes. Note. DIV, divergent intraverbal; COMP, competing stimulus probe

Each type of probe was presented in separate sessions with each relation presented twice. For example, probes of the individual elements were intermixed in an 8-trial session that included two presentations of A1, B1, A2, and B2. Separate sessions were conducted for the individual element, reverse, DIV, and COMP probes. The order of these probes was randomly determined, except for the COMP probe, which was always conducted last. Criterion performance was defined as unprompted correct responding at or above 75% in a given probe.

Remedial and Booster Training

Due to inconsistencies in responding during probes across all four sets, training was introduced for an additional relation beginning with either the DIV or individual elements. We began training with the most recently trained set (i.e., Set 4 followed by Set 3, etc.; see Table 2). We chose not to train additional relations following the initial compound intraverbal training until all sets had undergone the initial training as doing so was hypothesized to affect discriminated performances in later sets. Specifically, training the individual elements in the compound or DIV relation may have impacted performance in subsequent training sets. For Sets 1, 2, and 4, remedial training targeted the DIV relation. Set 3 included remedial training of the single components. Set 1 was supposed to be exposed to individual training along with Set 3; however, due to high levels of emergence of the single component relation following training, the DIV relation was trained instead. All relations except for the COMP were probed again before remedial training was conducted. During remedial training, two sessions were conducted at a 0-s prompt delay using echoic prompts. All subsequent sessions were conducted at a 5-s prompt delay. Training continued until two consecutive sessions with 100% unprompted correct responses were observed.

Remedial training for Sets 1–3 began after booster training of the baseline relation (e.g., A1B1 and A2B2); that is, booster training was used to ensure that Alice’s responding in the baseline relation was at the mastery criterion before beginning remedial training. Booster training was identical to the 5-s prompt delay condition during initial training of the baseline relation. After responding met the mastery criterion for this relation, probes were again conducted, followed by remedial training. Booster training was not conducted for Set 4 as remedial training occurred immediately following post-IV training probes (see Table 2). Once the training criterion was met for the remedial relation (i.e., DIV for Set 4), probes were again conducted. Training of additional relations occurred if low rates of responding were observed during probes as indicated in Table 2.

Interobserver Agreement and Procedural Integrity

An independent data collector was present during 83.7% of training sessions and 54.6% of probe sessions. Trial-by-trial interobserver agreement (IOA) was calculated by dividing the number of trials with an agreement divided by the total number of trials and multiplied by 100. Mean IOA was 100% during training and probes.

Procedural integrity was also recorded by at least one observer during 85.7% of training sessions and 98.2% of probe sessions. Procedural integrity was recorded on a trial-by-trial basis. Percent of trials with integrity was calculated by dividing the number of trials implemented with integrity by the total number of trials in a session, multiplied by 100. Mean procedural integrity was 99.6% (range, 83.3% to 100%) during training and 97.7% (range, 50% to 100%) during probes. Procedural integrity was at 50% during two, four-trial sessions in which the experimenter incorrectly delivered praise following correct responses during probes. A second observer was also present for 81.6% of training and 54.6% of probe sessions, which allowed for trial-by-trial IOA to be calculated for procedural integrity. Mean procedural integrity IOA was 100% during training and 98.3% (range, 50% to 100%) during probes. Procedural integrity IOA was 50% during a single session and occurred when the second observer did not record the experimenter’s praise for two correct responses as an integrity error.