Stimulus Clarity and the Emission of Descriptive Autoclitics

Abstract

The autoclitic is among the least studied and most complex verbal operant named and described by Skinner. The descriptive autoclitic is one subtype, which among other functions can describe the strength of the response. If the clarity of the stimulus is one source of response strength for tacts, manipulating stimulus clarity should evoke different frequencies of descriptive autoclitics. In an experiment with adults, digitally distorting pictures of common objects predicted the relative frequency of descriptive autoclitics that accompanied tacts. The most distorted images evoked twice as many autoclitics as moderately distorted images, and low-distortion images evoked no autoclitics. We encourage other researchers to interpret Skinner’s conceptualization of the autoclitic and its various forms and test them empirically to evaluate how their functional definitions can be clarified, refined, or altered.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40616-023-00184-1.

Self-described as his most important work (Skinner, 1977), Skinner’s Verbal Behavior (1957) has made an enormous philosophical, conceptual, and clinical impact on the field of behavior analysis (e.g., Axe et al., 2019; Moore, 2022; Petursdottir, 2018). The effectiveness of Skinner’s system of examining the function of verbal operant classes is perhaps best demonstrated in work with individuals with deficient repertoires (e.g., Sundberg & Michael, 2001; Belisle et al., 2022). However, the analysis of verbal behavior is not just important for teaching individuals to communicate; it is an explanation and interpretation of all responses reinforced by listeners—from the first functionally differentiated cries of infants to the most complex behavior of the human organism.

Skinner’s functional categories, the verbal operants, have been quite useful in understanding socially mediated behavior and developing effective technologies. The autoclitic is one of the least studied operants (Petursdottir & Devine, 2017) and perhaps the most complex. Broadly, the autoclitic is a secondary response which modifies the effect of a primary response. For example, the response “rain” could be modified by specifying the source of control (“I heard it is raining”) or the speaker’s confidence in the response (“I think it is raining”). Skinner describes subtypes of autoclitics such as the qualifying and the descriptive autoclitic. These two subtypes are not easily defined because Skinner refers to descriptive autoclitics often when writing about qualifying autoclitics. For instance, the clearest definition of the descriptive autoclitic appears as the first sentence of the qualifying autoclitic section (Skinner, 1957, p. 322). To summarize, the qualifying autoclitic is primarily controlled by motivative variables that make certain behavior of the listener more reinforcing than alternatives, whereas the descriptive autoclitic (the type targeted in this study) is controlled by something about the speaker’s own behavior or how it is evoked.

In Cengher et al. (2020), comparisons were made between one-word mands and what were called “mands with autoclitic frames” defined as a first-person pronoun, a verb, and a noun (the mand). However, when Skinner defined autoclitic frames he repeatedly referred to the novel emission of a “frame” that had previously been conditioned in other contexts. When an individual is taught what may topographically look like an autoclitic frame, this may just be a mand (i.e., the response may not function as an autoclitic). Skinner makes this point about various operant classes several times in Verbal Behavior (e.g., Skinner, 1957, pp. 100–101, p. 325).

Other studies have taught autoclitic functions, such as suffixes/grammatical conventions (e.g., Greer & Yuan, 2008; Speckman & Greer, 2012). In these studies, suffixes (-ed and -er, respectively) were applied to novel verbal responses after instruction on a variety of responses with these forms. Martins et al. (2015) investigated the speaker-as-listener effects of having participants use what is often a qualifying autoclitic (i.e., is), However, participants were instructed to say “is” when describing sample stimuli. Similarly, Howard and Rice (1988) taught preschool children to engage in autoclitic responses in the presence of visual stimuli whose shape or color was distorted compared to training stimuli (e.g., “like a square” for a trapezoid). Kuroda et al. (2014) manipulated the size of reinforcers for a delayed matching-to-sample task where pigeons could peck keys analogous to the qualifying autoclitics “definitely” or “maybe” before pecking the comparison stimulus. Kuroda et al. showed pigeons more frequently pecked the “maybe” key when the selection was delayed and infrequently pecked when the selection was available immediately.

Nevertheless, we could not find empirical evidence that descriptive autoclitic responses are under the control of stimuli Skinner presumed without specific training to evoke it. Modeling, prompting, and reinforcing a response in the presence of specific stimulus conditions will increase the frequency of that response under control of those stimuli. However, it may be valuable to examine the practices of organisms with typical or uncontrived learning histories to see if the relevant autoclitics are a function of the variables described by Skinner. Thus, the present study examined the “strength of the response” aspect of the descriptive autoclitic. One variable that may affect the strength of a tact of a visual stimulus is how clear the stimulus is. Visual stimuli that are unclear should evoke more descriptive autoclitics that modify the tact in comparison to visual stimuli that share similarities to exemplars in the stimulus class. Specifically, we assessed the effects of distortion of pictures of common objects on the emission of descriptive autoclitics modifying tacts in adult participants.

Method

Participants

Participants were 32 undergraduate students at a private liberal arts college in the southeast United States, all of whom were enrolled in a Psychology 101 course. Twenty-six participants were women and six were men, all of them were aged 18–20. Seven were Black, three were Asian, two were Latina, and 20 were White. Participants were offered extra credit in their course contingent on participation in the study. Participation required 6–10 min per participant, including informed consent, data collection, and debriefing.

Materials

Stimuli were created on a desktop computer with the free-to-use, open-source image manipulation software Glimpse. Stimuli were presented on a 10 in. tablet. Sessions were recorded on a digital camcorder for data collection, interobserver agreement, and treatment integrity purposes. Sessions took place in a 9 m by 15 m meeting room within an academic building on campus, which contained a conference table and chairs.

Experimental stimuli were created by digitally manipulating 35 colored images of common objects (a complete list can be found in the Supplemental Materials). Three sets of stimuli were created. Each set contained one version of each image (e.g., a highly distorted apple). Across the three sets, each image was represented at each distortion level (i.e., set A contained a low-distortion apple, set B contained a moderate-distortion apple, and set C contained a high-distortion apple). Three levels of distortion were used by combining different filters in the Glimpse application (see Table 1). Examples of two images at all three distortion levels can be seen in Fig. 1.

Table 1.

Filter settings for creating stimuli in Glimpse software

Distortion level	Ripple filter	Mosaic filter
Low	N/A	Tile geometry: triangles Tile size: 10 Tile height: 4 Neatness: 0.65 Color variation: 0 Uncheck “color averaging” Uncheck “rough tile surface” Uncheck “splitting tiles” Tile spacing: 0.25
Moderate	Amplitude: 55.5 Period: 175 Angle: 1	Same as low distortion
High	Amplitude: 85.5 Period: 175 Angle: 1	Same as low distortion

Fig. 1.

Example stimuli (mailbox, balloon) shown at each distortion level

Dependent Variable and Experimental Design

The dependent variable was the occurrence/nonoccurrence of descriptive autoclitics emitted by participants in each trial. We scored a response as an autoclitic if one or more words indicating tentativeness followed and/or preceded the tact. Examples included saying, “I guess,” “sort-of,” “I think,” “like a __,” “I don’t know”, and answering in the form of a question (e.g., “what is it, a ___?”). Non-examples included tacts accompanied by vocal space fillers such as, “um” or “uh,” saying, “I don’t know” without a tact, and non-vocal responses such as shrugging, head tilts, or facial expressions. These topographies were selected based on Skinner’s examples (1957, pp. 313; 315).

Each participant was exposed to 35 trials in which they were asked to tact a picture. The independent variable was the level of distortion, which had three conditions. In each trial, a participant was exposed to either a high-, moderate-, or low-distortion image. By measuring the occurrence of autoclitics as we rapidly alternated between conditions, the experimental design was akin to an adapted alternating treatments design (Cariveau et al., 2021). Because our dependent variable was binary (occurrence/nonoccurrence) and opportunity bound, we converted the dependent variable to the percentage of trials with autoclitics evoked by each distortion level for each participant and calculated averages for all participants by dividing the number of trials with autoclitics divided by the total number of trials for each distortion level and multiplying by 100. A test of significance was used to determine if the number of autoclitics evoked was associated with level of distortion. A robust chi-square test was used to account for dependence within participants.

Procedure

Each participant was randomly assigned one of three sets of stimuli. The experimenter sat across a table from the participant with the camcorder over their right shoulder. After agreeing to participate in the study, the experimenter started video recording and read the following script.

Thank you for participating in the study. Today I’m going to be showing you a set of distorted images of common objects. What we’re trying to see is how well that we’ve distorted these images. So, we’ll be going through slides of these images, and for each picture, tell me what you see. I need to video record so I can look back at it later. It would really help me if you gave the correct response to each image, but after you tell me what each image is, I’m just going to say “ok” or “all right” regardless of whether you got it right or wrong. Are you ready to start?

During each trial, the experimenter held the tablet up to the camcorder, then turned it around to face the participant approximately 0.5–1 m in front of them. The experimenter waited 10 s for the participant to tact the image. When the participant tacted the image, the experimenter responded with neutral feedback, turned the tablet around, swiped to the next image, and began the next trial.

If the participant asked a question such as, “what if I don’t know?” the experimenter replied, “I need a response.” If the participant said, “I don’t know” or requested to skip the image (e.g., saying “pass”), the experimenter said, “ok” and continued to the next trial. After the last trial, the experimenter thanked the participant for their participation, debriefed them by explaining the purpose of the study, and provided proof of participation for course extra credit.

Interobserver Agreement and Procedural Fidelity

A second observer scored video recordings of sessions and compared those with the original measures to assess interobserver agreement (IOA), which was calculated by dividing the number of agreements by the total number of trials, multiplied by 100. We calculated IOA for 34.4% of sessions, which averaged 98.5% (range, 94.2% to 100%). Procedural fidelity was assessed by scoring the following critical features of the procedure either once per session or per trial: reading the instructions script to the participant without error, asking the participant if they were ready to begin, accessing the correct stimulus set for participant, turning the image to the camera, showing the participant each picture, waiting for the participant’s response before advancing to next picture, giving a neutral response following answers, thanking the participant for their participation, sharing the debriefing form with the participant, and providing an extra credit slip. Procedural fidelity was calculated for 34.4% of sessions by dividing the number of steps completed correctly by the total number of steps and multiplying by 100. Procedural fidelity was 100%.

Results and Discussion

Figure 2 shows each participant’s percentage of trials with autoclitics for each distortion level. Figure 3 shows the averages and ranges of responses evoked by each condition. Across a total of 1120 trials, 108 autoclitics were emitted. No autoclitics were emitted in any of the 372 trials with low-distortion stimuli. In 373 trials with moderate-distortion stimuli, autoclitics were emitted in 36 trials, or 9.8% of trials. In 375 trials with high-distortion stimuli, autoclitics were emitted in 72 trials, or in 19.2% of trials. Eight participants did not emit autoclitics during any trial. The highest mean percentage of trials with autoclitics for the low-, moderate-, and high-distortion conditions was 0%, 36.4%, and 58.3%, respectively. A chi-square test found the correlation between stimulus distortion and the emission of descriptive autoclitics was statistically significant (DF 2, Chi-sq 29.6, P < 0.001).

Fig. 2.

Individual data

Fig. 3.

Averages and ranges across all participants who emitted autoclitics. Note. The eight participants who never emitted an autoclitic are not included in these summary data. The dotted line within the box is the mean, and the solid line is the median

As a proportion of total autoclitics, the high-distortion stimuli evoked double the number of autoclitics evoked by moderate-distortion stimuli. In total, 72 of the 108 autoclitics across all participants were emitted in the presence of high-distortion stimuli (66.7%), and 36 were evoked by moderate-distortion stimuli (33.3%). Of the participants who emitted at least one autoclitic, one participant emitted an autoclitic in the same percentage of trials for moderate- and high-distortion stimuli. Three participants emitted autoclitics in a higher percentage of trials with moderate-distortion stimuli than high-distortion stimuli. Individual data can be found in Supplemental Materials.

Our results showed different observed frequencies of autoclitics emitted in the presence of stimuli of different salience. These results correspond with Skinner’s predictions that descriptive autoclitics would be emitted when the tact “[was] based upon insufficient stimulation” (1957, p. 315). The most salient stimuli in this experiment never evoked any autoclitics, and for those participants who emitted an autoclitic (24 of 32), 83% were more likely to modify their tact with a descriptive autoclitic in the presence of highly distorted stimuli than moderately distorted stimuli.

Regardless of the differences in response patterns for most participants in this study and averages across all participants, these results are not conclusive. Eight participants did not emit any autoclitics under any conditions. Behavior, perhaps especially verbal behavior, is multiply controlled (Michael et al., 2011) and the clarity of the stimulus is not the only variable that could affect participant’s responses. The experimental setting did not resemble a conversation or typical verbal interaction and could have suppressed autoclitic behavior. Audience effects (such as the experimenter, the room, being in an academic building rather than a dorm) could also influence autoclitic behavior; although all sessions were conducted by the same experimenter, the experimenter read the initial instructions from a script which could suppress extraneous words in the participants answers to match the exactness of their speech. Participants were also being video recorded, and if that made them mildly uncomfortable it could have reduced the probability of adding any qualifiers to their answers; the effect of all these variables is likely highly idiosyncratic based on individual learning histories.

Additionally, our operational definition was a limitation. Shrugging, facial expression, and other body language indicators likely do have a descriptive autoclitic function, as does tone and inflection. Excluding those topographies from our operational definition likely artificially reduced the frequency of behavior, but it was a choice that allowed for reliable data collection and less subjectivity. Furthermore, we preferred to under-estimate the occurrence of autoclitics to ensure our conclusions were conservative.

More research is needed to examine variables that affect autoclitics. By solely manipulating antecedents to evoke behavior, any regularity in the data is primarily due to commonalities in the learning histories of the participants. By showing participants are more likely to emit descriptive autoclitics modifying tacts when stimuli are less salient, we are showing there may be practices in the verbal community which shape these patterns of behavior. As such, groups selected from different types of communities (e.g., geographic, age,) may show different relative rates of descriptive autoclitics, and varied topographies of those responses.

We called our dependent variable a descriptive autoclitic rather than a qualifying autoclitic because the only manipulation was stimulus clarity with no changes to the programmed consequences of responding with or without an autoclitic. Although this experiment demonstrated that stimulus clarity may control this class of responses to a certain degree, future basic research on autoclitics should demonstrate that these two functions (descriptive vs qualifying autoclitics) are indeed separable if they are to be useful distinctions in our behavior analytic repertories. Past research and practice teaching responses under control of less salient stimuli should be called descriptive autoclitics only once the field has demonstrated that the verbal community engages in autoclitic behavior in the way Skinner described. We would encourage the field to conduct research (basic or applied) on autoclitics with a close look at function, rather than topography. Topographies are ephemeral and extremely idiosyncratic; therefore, creating operational definitions for autoclitics should be carefully considered.

Data Availability

The datasets generated during and/or analyzed during the current study are available in Supplemental Material and from the corresponding author on reasonable request.

Declarations

The Institutional Review Board for Human Subjects Research at Berry College examined the procedures of this study and found them to be in compliance with all standards and regulations.

Conflict of Interests

The authors have no known conflicts of interest.

Informed Consent

All participants consented to the study.