The effect of meaningful stimuli on eye movements in stimulus equivalence class formation

ABSTRACT

Meaningful stimuli interspersed with abstract stimuli increase the probability of establishing conditional discriminations and responding according to stimulus equivalence in a matching-to-sample procedure, compared to procedures with only abstract stimuli. Test accuracy and reaction time have been previous experiments’ primary dependent variables. However, contemporary research on stimulus equivalence has also included eye movement measures as a means of a fine-grained analysis of behavior. The present experiment investigates meaningful stimuli’s effect on eye movements. The present experiment was arranged as a group design, with 30 adult participants allocated in three groups. All learned 12 conditional discriminations in a one-to-many training structure before testing to establish three 5-member stimulus equivalence classes. One group was trained with meaningful sample stimuli, one with meaningful comparison stimuli, and one with all abstract stimuli. Results show significant differences in fixation durations and gaze transitions between the groups with meaningful stimuli and those with all abstract stimuli in the test for stimulus equivalence. Hence, measuring eye movements with eye-tracking technology can provide information about behavioral differences between conditions not obtained with accuracy scores.

KEYWORDS:

• Meaningful stimuli
• eye movements
• eye-tracking
• stimulus equivalence
• matching-to-samples
• fixation duration
• gaze transitions

In a typical MTS training procedure, participants choose one of several comparison stimuli to match a sample stimulus, learning if-then relations between two stimuli. Once participants learn a minimum of two stimulus-stimulus relations, with one common stimulus, for example, A-B and B-C relations, and at least two classes of stimuli, it is possible to test for stimulus equivalence. Stimulus equivalence is defined based on three features: reflexivity (A-A, B-B, and C-C), symmetry (B-A, and C-B), and transitivity (B-C and C-B), which are untrained relations that emerge when tested. Laboratory research investigating variables affecting stimulus equivalence class formation often uses stimuli with which the participants have no pre-experimental history, called abstract stimuli, to ensure experimental control. In the last two decades, more and more research has explored the effect different stimuli properties have on emergent responding. These investigations have shown that stimuli with a pre-experimental discriminative function in the participants’ repertoire increase the probability of stimulus equivalence class formation compared to stimuli with which the participants have no experience (e.g., Arntzen & Mensah, 2020; Fields & Arntzen, 2018; Marin et al., 2022; Mensah & Arntzen, 2022). Stimuli with a discriminative function are often referred to as meaningful stimuli.

In stimulus equivalence research, results are often reported as yields: the number of participants responding according to a criterion (e.g., Fields et al., 2020). In addition to the stimulus equivalence test result, the number of training trials needed to learn the conditional discriminations and reaction time to sample and/or comparison stimuli are also frequently reported. Furthermore, more advanced technology, such as eye-tracking, provides a more fine-grained analysis of the antecedents of matching behavior and emergent responding. However, to the authors’ knowledge, the effect meaningful stimuli have on eye movements has not yet been investigated. Hence, the present experiment is designed to investigate the effect meaningful stimuli have on eye movements when learning conditional discriminations and testing for stimulus equivalence class formation in an MTS procedure.

Eye movements are visual observing responses (Tomanari et al., 2007). Observing responses are any response that increases the organisms’ contact with the environment and can be seen as a prerequisite for stimulus control (Dinsmoor, 1985). With visual observing responses, eyes move “so that the image of the stimulus falls in the fovea” (Tomanari et al., 2007, p. 29). With eye-tracking technology, one can measure how eyes move on a computer screen as participants conduct a conditional discrimination task or a test for stimulus equivalence. Eyes move continuously, and the movements must be operationally defined in measurable dimensions to be a subject for analysis. The most common measure is fixations. Fixations are relatively stable eye movements to or around a specific spot in the visual field (Holmqvist et al., 2011). No specific time limit defines a fixation within the eye-tracking literature. Each experiment sets a time-based threshold. The gaze measures must be within a spot or area when defining the fixations, varying from 100 to 400 ms (Salvucci & Goldberg, 2000). The fixation threshold also depends on the algorithms used to calculate if fixation has occurred and where (Salvucci & Goldberg). Fixation duration and fixation rate are commonly used eye movement measures (Holmqvist et al., 2011, p. 377 and 416). Research in behavior analysis measuring eye movements has reported correlations between fixation time and rate and accuracy scores in stimulus control experiments.

Dube and colleagues (Dube et al., 2006, 2010) investigated eye movements in delayed MTS procedure with multiple sample stimuli and found in both experiments that participants with high accuracy scores had longer and more frequent fixations to sample stimulus compared to those with low accuracy scores. Huziwara et al. (2016) found that participants had longer fixation time to the sample stimulus compared to the comparison stimuli and longer fixation time to the correct comparison stimulus (S+) compared to the incorrect comparison stimulus (S-). They also found a decline in fixation duration across the serialized training session, which they refer to as an effect of practice, also reported by Pessôa et al. (2009) and Schroeder (1970). Eye movements have also been studied in stimulus equivalence research (e.g., Hansen & Arntzen, 2018; Sadeghi & Arntzen, 2018).

Hansen and Arntzen (2018) examined eye movements in three different arrangements of the training procedure: many-to-one (MTO), one-to-many (OTM), and linear series (LS) training structure testing for the formation of stimulus equivalence class formation. In short, all stimuli are trained in a chain in an LS training structure. In contrast, in the OTM and MTO training structure, all stimuli are trained to one connecting node stimulus, respectively, as the sample or comparison stimuli. In Hansen and Arntzen’s experiment, three participants in each group learned six 3-member stimulus classes of abstract stimuli. Accuracy scores from the test showed no effect of training structure on emergent responding; one of three participants responded in accordance with stimulus equivalence in each group. The eye-tracking measures showed that fixation duration to comparison stimuli decreased from the first to the last block of training and that fixation rates increased from the first to the last training block. In the test for stimulus equivalence, the fixation duration to correct comparison stimuli was higher than the fixation duration to incorrect comparison stimuli. The experiment showed no apparent effect of the three training structures on eye movements.

Sadeghi and Arntzen (2018) also investigated differences between the OTM, MTO, and LS training structures in a group design with 10 participants in each group. All participants were taught conditional discriminations with abstract stimuli and tested for the emergence of three 5-member stimulus equivalence classes. The accuracy scores showed that 100%, 70%, and 0% of the participants in the OTM, MTO, and LS groups, respectively, responded in accordance with stimulus equivalence. The eye movement measures showed decreased fixation durations, fixation rates, and gaze transitions from the start to the end of training for all groups. Gaze transitions are the number of eye movements between stimuli with a registered fixation. The variation in gaze transitions between the training’s start and end was most prominent for the MTO group for all measures.

There were no differences between groups regarding the tested relations (baseline, symmetry, and equivalence) for fixation durations to comparison stimuli in neither Hansen and Arntzen (2018) nor Sadeghi and Arntzen (2018). Although, Sadeghi and Arntzen reported a decrease from baseline to symmetry trials for the MTO and LS groups and a decrease from symmetry to equivalence trials in all groups. At the end of the test, there was a general decrease for all relations and a similar pattern for gaze transitions. At the beginning of the test, the mean fixation rate varied from 1 to 2.5 across all relations. Mean transition rates were much lower overall (approximately 0.03 to 0.7). The discrepancy between the fixation and the transition rates was not discussed. However, the difference could be because participants’ gaze went outside a comparison stimulus area and then back to the same comparison stimulus, not affecting transition rates. Alternatively, it might be that the gaze went from one comparison stimulus to the sample stimulus and back again to the same comparison stimulus, which would not affect transition rates either. In the future, including fixation to sample stimulus may be advantageous in analyzing gaze transitions to understand eye movements in stimulus equivalence research better.

Notwithstanding the number of gaze transitions with sample stimulus fixations in the analysis, it will not provide sufficient information since a transition could be between the sample stimulus and a comparison stimulus or between two comparison stimuli. However, analyzing the number of trials where repeated fixations to the sample stimulus are measured can shed additional nuanced light on the controlling relations in training and testing. For example, trials without a gaze back at the sample stimulus could be analogous to a delayed matching-to-sample paradigm (e.g., Blough, 1959). Hence, eye-movement analysis from the present experiments will include gaze transitions with sample stimulus and the number of fixations to sample stimulus to expand the understanding of the eye movements during the MTS procedure.

Most experiments that have investigated the effect of meaningful stimuli on stimulus equivalence class formation have established conditional discriminations with an LS training structure (e.g., Arntzen & Mensah, 2020; Fields et al., 2012; Mensah & Arntzen, 2016, 2022). In these experiments, the group with one or more meaningful stimuli (nameable picture) is compared with a reference group trained with all abstract stimuli in an LS training structure. The rationale for using the LS training structure in experiments investigating the effects of meaningful stimuli has been that the LS training structure yields are usually low, with few participants forming stimulus equivalence classes (e.g., Mensah & Arntzen, 2022), compared to when conditional discriminations are established with MTO and OTM training structures (e.g., Arntzen & Holth, 1997, 2000; Arntzen et al., 2010; Eilifsen & Arntzen, 2014; Sadeghi & Arntzen, 2018; Saunders et al., 2005).

The literature has discussed two main challenges with using an LS training structure. The first is the effect of the number of nodes created in large stimulus classes. Some experiments have demonstrated an inverse function between the test yields and the number of nodes that separated the stimuli (e.g., Fields et al., 1995, 2012; Kennedy, 1991; Mensah & Arntzen, 2022). Consequently, it has been suggested that using MTO or OTM training structures would avoid the effect of the number of nodes when investigating meaningful stimuli (Mensah & Arntzen, 2022). Second, in the LS training structure, all the nodal stimuli serve two functions: as sample stimulus in one trial and comparison stimuli in another (Fields & Verhave, 1987). In OTM or MTO training structures, each stimulus only serves one function, either as a sample or a comparison stimulus. Research has shown that the dual function has an adverse effect on the formation of stimulus equivalence class compared to when stimuli only have a single function despite an equal number of nodes in the stimulus class (Braaten & Arntzen, 2021; Menédez et al., 2017). As far as the authors know, if and how meaningful stimuli functioning as sample or comparison stimuli affect eye movements has not been investigated.

The present experiment was arranged as a group designed to investigate potential significant differences in participants’ eye movements when establishing conditional discriminations and testing for stimulus equivalence class formation by comparing stimulus classes with and without meaningful stimuli. Utilizing an OTM training structure allowed us to investigate if meaningful stimuli functioning as a sample or comparison stimuli affect eye movements. Also, adding other measures, such as fixations to sample stimuli in gaze transition analysis and the number of trials with refixation to sample stimuli, together with the more common fixation analyses, will potentially gain an increased understanding of how observing responses are affected by meaningful stimuli.

Method

Participants

Sixty-one university students (age range 19–43) participated in the experiment. To be included in the research, participants could not have experience with the MTS procedure, knowledge of stimulus equivalence, or any experience with eye-tracking experiments. Participants could not wear glasses or high-power contact lenses during the experiment. Upon arrival, participants were shown the experimental room and the eye-tracking equipment and asked to read and sign a consent form. In the consent form, the participants were informed about the experiment in general terms, their rights, and the approximate duration (three hours). After signing the form, the experimenter told the participants that they would be offered an optional break every 20 minutes and emphasized the issue of anonymity and that they could withdraw from the experiment at any time without any negative consequences. At the end of the experiment, all participants were offered a chance to view their data and were thoroughly debriefed about the research. Unfortunately, 31 participants were excluded because it was impossible to calibrate the eye-tracker to the participants or because of poor calibration throughout the training and testing, which resulted in eye measurements that could not be analyzed.

Setting

The experimental room was 12 m² and furnished with three tables and chairs. The participants used one table to read and sign the consent form. The experimental computer was on a second table. This table had a chinrest placed 60 centimeters in front of the screen to minimize head movements. The eye-tracking computer was on a third table. Opaque curtains covered the windows on one wall, and the room had dimmed lighting to ensure quality eye-tracking.

Stimuli

Fifteen abstract stimuli and 12 meaningful stimuli were used in this experiment, see Figure 1. Each stimulus set consisted of 15 stimuli arranged into three 5-member classes of stimuli. The stimulus set for the one group, the Node-as-Abstract group (N-ABS), consisted of three abstract stimuli and 12 meaningful stimuli. For the second group, the Node-as-Picture group (N-PIC), the stimulus set consisted of three meaningful stimuli and 12 abstract stimuli. For the third group, the Abstract group (ABS), all stimuli in the set were abstract shapes.

Figure 1. The Stimuli used in each group.

The numbers denote the stimuli classes, and the letters denote class members. N-ABS = node as abstract stimuli group, N-PIC = node as picture stimuli group, ABS= all abstract stimuli group.

On the screen, the width and length of the stimuli were 1.2–3.5 cm, dependent on the shape of the stimuli. The sample stimulus was presented in the middle of the screen, and the comparison stimuli were presented in four corner positions. The distance from the middle of the sample stimulus to the middle of any of the comparison stimuli was 12 cm. There was 11 cm from the middle of each compound stimuli to the outermost corner of the screen. The comparison stimuli were placed in a mid-position between the sample stimuli and the screen’s outermost edge to minimize the angle of eye movement and, in turn, increase precision in eye-tracking measures.

Apparatus

The experiment was conducted on a custom-built PC with Windows 7 Professional (32-bit) (Computer 1) with custom-made MTS software. This software randomized trials in training and testing, randomized comparison stimuli’s position in each trial, presented programmed consequences and registered participants’ responses. The LCD screen on Computer 1 had a 1280 × 1024 resolution. The touch-sensitive area for mouse clicks around each stimulus was 212 × 212 pixels, representing the area of interest (AOI) defining fixation time and count. Another custom-made software (ETAnalyzer) analyzed the data from the MST software with the eye-tracking data generated by Computer 2. Computer 2 was an ISCAN® computer running Windows 7 installed with an ISCAN® DQW 1.2 program to calibrate and track movements of the left eye with an ISCAN® head-mounted pupil/corneal reflection eye-tracking system (ISCAN Corp., Burlington, MA; http://www.iscaninc.com). The ISCAN® DQW 1.2 program processes data at 60 Hz. (16.5 milliseconds) and sent eye-tracking data to Computer 1 via AVerKey 300.

A duplicate screen of Computer 2 was on the outside of the experimental room. On this screen, the experimenter could watch the eye-tracking and evaluate the quality of the calibration. Then, if necessary, the experimenter would do a re-calibration to ensure correct eye movement measures.

Design

A semi-randomized group design was used in the present study. Upon arriving at the experiment, the majority of the participants were randomized to one of three groups. At the end of the experimental period, a few participants were allocated to different groups to get 10 participants in each group, therefore semi-randomizing participants. As described above, the groups differed regarding stimuli used in training and testing conditions.

Procedure

Calibration

Participants were seated in front of the experimental computer (Computer 1). The table and chair were adjusted, so participants sat comfortably with chins on the chinrest. The experimenter placed the eye tracker on the participant’s head and adjusted it for optimal pupil and cornea reflection detection. Participants were asked to look at the five dots on the screen for calibration. The calibration process took approximately 5 minutes. When the eye-tracker point-of-regard accurately represented participants’ gaze, calibration was complete. If, for some reason, the point of regard was off during training or testing, a re-calibration was done.

MTS procedure

Instruction

The MTS training started, and participants were presented with the following instructions on the screen (translated from Norwegian):

A stimulus will appear in the middle of the screen. You must click on this with the mouse. Next, three other stimuli will appear. Select one of these by clicking the mouse. If you choose the one we have defined as correct, words like “good,” “super,” etc., will appear on the screen. If you press incorrectly, “wrong” will display on the screen. Later in the experiment, the computer will not provide feedback on whether your choices are correct or incorrect, but based on what you’ve learned, you can get all the tasks right. Do your best to get everything correct. Good luck!

The participant had to press “START” under the instruction to advance to training.

Matching-to-sample training condition

The purpose of the training procedure was to establish 12 conditional discriminations in an OTM training structure. All participants were taught A-B, A-C, A-D, and A-E relations. See Table 1 for an overview of the trained relations. Each trial started with the sample stimulus in the middle of the screen. Three comparison stimuli appeared once the participants clicked on the sample stimulus with the mouse. The sample stimulus remained on the screen. If the participants chose the experimenter-defined correct comparison stimulus with a mouse click, the response was followed by a programmed consequence such as “correct,” “good,” etc. The word “wrong” appeared if the participants chose one of the experimenter-defined incorrect comparison stimuli. The programmed consequences were on the screen for 0.5-s., followed by a 0.5-s inter-trial interval before the subsequent trial. For each trial, the comparison stimuli appeared randomly in the four comparison stimuli positions on the screen described above. The baseline relations were established concurrently, and each relation was presented five times in each block, with a total of 60 trials in one block. After blocks with 100% programmed consequences, the probability of programmed consequences was reduced to 75%, 25%, and then 0% in the consecutive blocks. A 95% mastery criterion was required to advance in all the training phases. For the N-ABS group, the one abstract stimulus in each class was always the sample stimulus. The meaningful stimuli were always comparison stimuli, exemplified in Figure 2a. For the N-PIC group, the one meaningful stimulus in each class was always the sample stimulus, and the abstract stimuli were comparison stimuli, see Figure 2b. For the ABS group, both sample and comparison stimuli were abstract (see Figure 2c).

Figure 2. A visual display of trials as it appears on the computer screene.

Panel (a) shows an example of baseline trials for N-ABS and symmetry trials for N-PIC. Panel (b) shows an example of baseline trials in N-PIC and symmetry trials in N-ABS, and Panel (c) shows an example of all trials in ABS and equivalence trials in N-PIC.

Table 1. Overview of the training and test phases.

Phases	Trial Types	Number of trial types	Minimum # of Trials	Programmed Consequences (%)
Training Phase 1:	A1B1, A2B2, A3B3, A1C1, A2C2, A3C3, A1D1, A2D2, A3D3, A1E1, A2E2, A3E3	5	60	100
Training Phase 2:	A1B1, A2B2, A3B3, A1C1, A2C2, A3C3, A1D1, A2D2, A3D3, A1E1, A2E2, A3E3	5	60	75
Training Phase 3:	A1B1, A2B2, A3B3, A1C1, A2C2, A3C3, A1D1, A2D2, A3D3, A1E1, A2E2, A3E3	5	60	25
Training Phase 4:	A1B1, A2B2, A3B3, A1C1, A2C2, A3C3, A1D1, A2D2, A3D3, A1E1, A2E2, A3E3	5	60	0
Test Phase:	Baseline
	A1B1, A2B2, A3B3, A1C1, A2C2, A3C3, A1D1, A2D2, A3D3, A1E1, A2E2, A3E3	3	36	0
	Symmetry
	B1A1, B2A2, B3A3, C1A1, C2A2, C3A3, D1A1, D2A2, D3A3, D1A1, D2A2, D3A3	3	36	0
	Equivalence
	B1C1, B2C2, B3C3, B1D1, B2D2, B3D3, B1E1, B2E2, B3E3, C1D1, C2D2, C3D3, C1E1, C2E2, C3E3, D1E1, D2E2, D3E3, C1B1, C2B2, C3B3, D1B1, D2B2, D3B3, E1B1, E2B2, E3B3, D1C1, D2C2, D3C3, E1C1, E2C2, E3C3, E1D1, E2D2, E3D3	3	108	0

Test condition

The purpose of the test was to test for baseline (BSL), symmetry (SYM), and equivalence (EQ) relations. The test consisted of 180 trials: 36 BSL trials, 36 SYM trials, and 108 EQ trials. See Table 1 for detailed information about the tested relations. All relations were presented three times and in random order. As in training, each trial started with the sample stimulus in the middle of the screen. Three comparison stimuli appeared after a mouse click on the sample stimulus while the sample stimulus remained on the screen. The test had no programmed consequences following comparison stimuli selection, only a 1-s inter-trial interval before the subsequent trial. To pass the test, participants had to respond correctly to 90% of BSL, SYM, and EQ relations.

Eye-movement measures

Fixation time

An Area-of-Interest (AOI) algorithm was used to calculate fixations (Salvucci & Goldberg, 2000). A fixation was measured when the point-of-regard was inside one of the AOI for 100 ms or more. Fixation time to sample stimulus was measured as combined fixation time to the sample stimulus after the comparison stimuli appeared on the screen. Fixation time to the correct comparison stimulus was calculated as the combined fixation time on the comparison stimulus defined as correct in each trial. Fixation time to incorrect comparison stimuli was the combined fixation time on the two comparison stimuli defined as incorrect for that trial divided by two (since there were two incorrect comparison stimuli). All fixation times were averaged across trials per participant, either all trials in the first and last training block or the five first and five last tested relations: BSL, SYM, and EQ. The software randomized the five first and last EQ trials, though the program did not ensure these relations differed. In theory, three of the five relations measured could be the same EQ relation.

Number of gaze transitions

Gaze transition was measured by the number of movements between two fixations on different stimuli. One transition could be between a sample and a comparison stimulus or between two comparison stimuli. The number of transitions was averaged across trials, either all trials in the first and last training block or across the five first and five last tested relations: BSL, SYM, and EQ.

Refixation to sample stimulus

The percentage of trials participants fixated on the sample stimulus after the initial fixation is calculated for the first and last training block and the test’s BSL, SYM, and EQ relations. All test trials were included in this analysis. For instance, a fixation at one of the comparison stimuli followed by a fixation at the sample stimulus would be considered a trial with a refixation to the sample stimulus. However, if fixation were not registered at the sample stimulus after the inital sample stimulus fixation before a comparison stimulus choice, this would be a trial without refixation to the sample stimulus.

Statistical analysis

When analyzing the variables, a Kruskal-Wallis test was used to determine the differences between the groups, and a Dunn’s test was used for a post hoc comparison test to indicate which groups differed. A significant level of 0.05 was used in all tests. The non-parametric Kruskal-Wallis test was chosen (1) after a Shapiro-Wilk test showed that the data did not follow a normal distribution, (2) the nature of some of the analysis, and (3) a small sample size.

Results

Number of training trials to criterion and test for stimulus equivalence formation

On average, participants in the N-ABS group, N-PIC group, and ABS group required 666, 510, and 720 baseline trials, respectively, to meet the criterion in training, and there were no statistical differences between the groups. 10 out of 10 participants responded in accordance with stimulus equivalence in the N-ABS and N-PIC groups, and nine out of 10 participants did so in the ABS group. The one participant who did not reach the mastery criterion had more than 90% correct on BSL and SYM trials and 89.8% on EQ trials.

General comment on the eye movements analysis

Figures 3–7 show the results of the eye movement measurements. In all graphs, the dark gray line depicts the N-ABS group’s results, the light gray line the N-PIC group’s results, and the white line the ABS group’s results. Table 2 lists the Kruskal Wallis and the post hoc Dunn’s tests that indicated statistical differences between groups.

short-legendFigure 3.

The graph shows the averaged mean fixation duration to sample stimulus at the beginning and the end of the training, and the first five and the last five test trials of each relation tested. The asterisk represents a statistically significant one-way ANOVA test. BSL=baseline relations, SYM=symmetry, EQ=equivalence, N-ABS = node as abstract stimuli group, N-PIC = node as picture stimuli group, ABS= all abstract stimuli group. Error bars represent the standard error of the mean (SEM). ms=milliseconds.

short-legendFigure 4.

The graph shows the averaged mean fixation duration to correct comparison stimuli at the beginning and the end of the training and the first five and the last five test trials of each relation tested. The asterisk denotes statistically significant post hoc Tukey tests. BSL=baseline relations, SYM=symmetry, EQ=equivalence, N-ABS = node as abstract stimuli group, N-PIC = node as picture stimuli group, ABS= all abstract stimuli group. Error bars represent the standard error of the mean (SEM). ms=milliseconds.

short-legendFigure 5.

The graph shows the averaged mean fixation duration to incorrect comparison stimuli at the beginning and the end of the training, and the first five and the last five test trials of each relation tested. The asterisk denotes statistically significant post hoc Tukey tests. BSL=baseline relations, SYM=symmetry, EQ=equivalence, N-ABS = node as abstract stimuli group, N-PIC = node as picture stimuli group, ABS= all abstract stimuli group. Error bars represent the standard error of the mean (SEM). ms=milliseconds.

short-legendFigure 6.

The graph shows the average mean number of gaze transitions between stimuli at the beginning and end of the training and the first five and the last five test trials of each relation tested. The asterisk denotes statistically significant post hoc Tukey tests. BSL=baseline relations, SYM=symmetry, EQ=equivalence, N-ABS = node as abstract stimuli group, N-PIC = node as picture stimuli group, ABS= all abstract stimuli group. Error bars represent the standard error of the mean (SEM).

short-legendFigure 7.

The graph shows the mean percentage of trials with refixation to sample stimulus at the beginning and the end of the training, and the first five and the last five test trials of each relation tested. The asterisk denotes statistically significant post hoc Tukey tests. BSL=baseline relations, SYM=symmetry, EQ=equivalence, N-ABS = node as abstract stimuli group, N-PIC = node as picture stimuli group, ABS= all abstract stimuli group. Error bars represent the standard error of the mean (SEM).

Table 2. Results from the statistical tests.

	Statistical Test Results
Eye Movement Measures	Kruskal-Wallis	Post Hoc Dunn’s test
Fixation Duration to Sample Stimuli
First five BSL	H = 6.575, p = .0373	N-ABS vs N-PIC	p=.0384
Fixation Duration to Correct Comparison Stimuli
Last Training Block	H = 6.887, p = .032
First five BSL	H = 10.01, p = .0067	N-ABS vs ABS	p=.0063
First Five EQ	H = 6.74, p = .0344	N-ABS vs ABS	p=.0309
Fixation Duration to Incorrect Comparison Stimuli
Last Training Block	H = 11.74, p = .0028	N-ABS vs N-PIC	p=.0022
First five BSL	H = 7.208, p = .0272	N-ABS vs ABS	p=.0287
First Five EQ	H = 7.814, p = .0201	N-ABS vs ABS	p=.0156
Last five SYM	H = 8.266, p = .0160	N-PIC vs ABS	p=.0126
Last Five EQ	H = 8.434, p = .0147	N-ABS vs N-PIC	p=.0156
Gaze Transitions
Last Training Block	H = 11.01 p = .0041	N-ABS vs N-PIC	p=.0241
		N-ABS vs ABS	p=.0072
First five BSL	H = 6.791, p = .0335	N-ABS vs ABS	p=.0282
Last five SYM	H = 7.329, p = .0364	N-PIC vs ABS	p=.0364
Number of Trials With Refixation to Sample Stimulus
First Training Block	H = 6.196, p = .0451	N-ABS vs N-PIC	p=.0424
BSL	H = 14.22, p = .0008	N-ABS vs N-PIC	p=.0103
		N-PIC vs ABS	p=.0013
EQ	H = 6.624, p = .0364	N-ABS vs N-PIC	p=.0457

BSL=baseline relations, SYM=symmetry, EQ=equivalence, N-ABS = node as abstract stimuli group, N-PIC = node as picture stimuli group, ABS= all abstract stimuli group.

Fixation duration to sample stimulus

The averaged mean fixation duration to sample stimulus after the appearance of comparison stimuli is shown in Figure 3. In training, there were no statistical differences between the groups. Regarding fixation duration to sample stimulus in the first five BSL relations, the results of the Kruskal-Wallis test yielded significant differences between groups. The Dunn’s test indicated that N-ABS group scores were significant from those of the N-PIC group, where N-ABS were the highest, and N-PIC were the lowest. No statistical differences existed between the groups in the first five SYM and EQ trials or the last five trials for any of the relations.

Fixation duration to correct comparison stimulus

Training

Figure 4 shows the average mean fixation time to the correct comparison stimulus in the training and test. There were no statistically significant differences between the three groups in the first training block. The Kruskal-Wallis test yielded significant differences between groups in the last training block. The post hoc Dunn’s test did not show statistical differences between any specific group. The N-ABS group had the lowest fixation duration, and the N-PIC group had the highest number. The ABS group scores were somewhere in the middle.

Test

Results of the Kruskal-Wallis test on the first five BSL trials yielded significant differences between groups. A post hoc comparison using Dunn’s test indicates that fixation duration to the correct comparison stimulus between the N-ABS and ABS groups differed significantly. The N-PIC had the lowest fixation duration, and the ABS group had the highest. The N-PIC group scores were somewhere in the middle. The differences between the groups were not statistically different for the SYM trials. The Kruskal-Wallis test yielded significant differences between groups in the first five EQ trials. The post-hoc Dunn’s test showed that the N-ABS and ABS groups differed significantly. Again, the N-ABS had the lowest fixation duration, and the ABS group had the highest. The N-PIC group scores were somewhere in the middle. There were no statistical differences between the groups in the last five trials in the test for any of the relations.

Fixation duration to incorrect comparison stimuli

Training

The fixation duration to incorrect comparison stimuli in training and testing is shown in Figure 5. There were no statistical differences between the groups in the first training block. In the last training block, the Kruskal-Wallis test yielded significant differences between groups, and the post-hoc Dunn’s test showed that the N-ABS and N-PIC groups differed significantly. The N-ABS groups had the lowest fixation duration to incorrect stimuli, and the N-PIC had the highest. The ABS group’s scores were in the middle.

Test

The overall trend in fixation duration at the beginning of the test between the groups is that there were longer fixations to incorrect stimuli for the group with all abstract stimuli, the ABS group. The Kruskal-Wallis test yielded significant differences between the groups for the first five BSL trials. A post hoc Dunn’s test showed that the N-ABS and ABS groups differed significantly. In these trials, the N-ABS group had the lowest fixation duration, and the ABS group had the highest. The N-PIC group’s scores were in the middle. In the first five SYM trials, groups were not statistically different. The Kruskal-Wallis test yielded statistically significant differences between the groups for the first five EQ relations. Dunn’s test showed that the N-ABS and ABS groups differed significantly in EQ test trials, where N-ABS scores were lowest, and ABS groups were highest. The Kruskal-Wallis test showed statistically significant differences between the groups for the last five SYM trials. The post hoc test showed that the N-PIC and ABS groups differed significantly in SYM trials. The N-PIC group had the lowest fixation duration, and the ABS group had the highest. The N-ABS group’s scores were in the middle. In the five last EQ trials, the Kruskal-Wallis test showed statistically significant differences between the groups. Dunn’s test showed that the N-ABS and the N-PIC groups differed significantly. The N-ABS group had the lowest fixation duration, and the N-PIC group had the highest. The ABS group’s scores were in the middle.

Gaze transitions

Training

The averaged mean number of gaze transitions across participants in each group in training and test are shown in Figure 6. The first training block has no statistically significant differences between the three groups. However, the Kruskal-Wallis test yielded statistically significant differences between the groups regarding gaze transitions in the last training block. A post hoc Dunn’s test showed that the N-ABS and N-PIC groups and the N-ABS and ABS groups differed significantly. The N-ABS group had the lowest number of gaze transitions, and the ABS group had the highest number. The N-PIC group scores were somewhere in the middle. Also, there is a decrease in the number of transitions from the first to the last training block.

Test

The Kruskal-Wallis test yielded statistically significant differences between the groups in the first five BSL test trials. A post hoc Dunn’s test showed that the N-ABS and ABS groups differed significantly. The N-ABS group had the lowest number of gaze transitions, and the ABS group had the highest number. The N-PIC group scores were somewhere in the middle. Visually, a similar pattern can be seen in SYM and EQ relations, but the group differences were insignificant. For the last SYM trials, the Kruskal-Wallis showed significant differences between the groups, and Dunn’s test showed that the N-PIC and ABS groups differed significantly. The N-PIC group had the lowest number of gaze transitions, and the ABS group had the highest number. The N-ABS group scores were somewhere in the middle. There were no differences between the groups in the last five EQ test trials.

Number of trials with refixation to sample stimulus

Figure 7 shows the mean percentage of trials when the fixation returned to the sample stimulus after the initial first fixation when the comparison stimuli appeared on the screen in the training and test. The three groups are compared in the first and last training block and the tested relations BSL, SYM, and EQ. Based on visual inspection, the N-PIC group had fewer trials with fixation back to sample stimulus in training and testing. That means the N-PIC group had more trials where eye movements went directly to one or several comparison stimuli before a response and did not fixate on the sample stimulus again. The Kruskal-Wallis test yielded statistically significant differences between the groups in the first training block. The post-hoc Dunn’s test showed that the N-ABS and N-PIC groups differed significantly in the first test block. The N-PIC group had the lowest score, and the N-ABS group had the highest. The ABS group scores were somewhere in the middle. There were no statistical differences between the groups in the last training block. For the tested relations, the Kruskal-Wallis test yielded statistically significant differences between the groups for BSL and EQ trials. The post-hoc Dunn’s test showed that the N-ABS and N-PIC groups and N-PIC and ABS groups differed significantly in BSL trials. Dunn’s test showed significant differences between N-ABS and N-PIC groups for the EQ relations. In both BSL and EQ trials, the N-PIC group scored the lowest and the ABS group the highest. The N-ABS group scores were somewhere in the middle. There were no statistical differences between the groups in SYM trials.

Discussion

The present experiment investigated the effects of meaningful stimuli on eye movements using eye-tracking technology in an MTS procedure. The three groups showed no differences regarding the number of training trials necessary to learn the conditional discriminations, which supports previous research comparing groups with and without meaningful stimuli in three 5-member classes of stimuli (e.g., Arntzen & Mensah, 2020; Arntzen et al., 2014; Mensah & Arntzen, 2016). Also, there were no differences between the groups regarding stimulus equivalence class formation. These results are not surprising as OTM training structure and the inclusion of meaningful stimuli are variables that often provide a high probability of stimulus equivalence class formation (e.g., Arntzen & Holth, 1997, 2000; Arntzen & Mensah, 2020; Arntzen et al., 2010; Eilifsen & Arntzen, 2014; Fields et al., 2012; Saunders et al., 2005). Conceivably, obtaining similar results on one measurable dimension of behavior could be advantageous in examining whether the manipulation implemented can still result in differences in other measurements of behavior (e.g., Dias et al., 2021). In the present study, eye movements are the primary dependent variable investigated, and the results show that in several of the measures, meaningful stimuli affect eye movements significantly, especially when comparing the groups with and without meaningful stimuli. The group with only abstract stimuli, the ABS group, stands out with the highest fixation durations on all measures, the highest number of transitions, and the highest number of trials with refixation to the sample stimulus. Some of these differences were also significant and will be discussed.

Training

In training, meaningful stimuli reduce fixation duration to correct and incorrect comparison stimuli compared to abstract comparison stimuli. The differences are significant in the last training block. Fixation duration to correct and incorrect comparison stimulus is highest in the N-PIC group, with abstract comparison stimuli, and lowest in N-ABS, with meaningful comparison stimuli. Meaningful stimuli might be easier or faster to discriminate due to the color or other physical features or the fact that the stimuli have a previous discriminative function in the participants’ repertoire, reducing the need for long foveal fixations or repeated fixations.

Further, meaningful stimuli reduce the number of transitions and trials with refixations to the sample stimulus. When the sample stimulus is meaningful, participants fixate less on it after the initial fixation in the first training block. Meaningful stimuli might reduce the need to refixate the stimuli due to the individual discriminative history with the stimuli. When the comparison stimuli are abstract (N-PIC and ABS groups), there are significantly more gaze transitions between stimuli. Visual inspection shows a general decrease in fixation duration to incorrect comparison stimulus and the number of transitions from the first to the last training block for all groups. This might suggest that the higher the strength of stimulus control, the fewer and shorter observing responses, as the accuracy increases in the last training block.

Test

Fixation duration to sample stimulus

Compared to abstract stimuli, meaningful stimuli appear to have minimal effect on the length of fixation on sample stimuli., except in the first five BSL trials. In these trials, fixation durations are lower when the sample stimulus is meaningful than abstract (N-PIC and N-ABS). In SYM trials, however, the averages between the N-PIC and the N-ABS groups are similar, and there is no significant effect of meaningful stimuli on fixation duration to sample stimulus between these groups. On equivalence trials, based on visual inspection, the average fixation duration for the ABS group was higher than the other two groups, even though for the N-PIC and ABS groups, all stimuli in the screen are abstract and, therefore, the trials are visually similar (see Panel c, Figure 2). This difference between N-PIC and ABS might indicate that including meaningful stimuli in the stimulus class affects eye movements even when meaningful stimuli are not presented on the screen in those trials. More conclusive data must support these speculations. At the end of the test, there are no differences in fixation time to sample stimulus between the groups.

Fixation duration to correct and incorrect comparison stimulus

Data analysis of fixation duration to correct and incorrect comparison stimuli (Figures 4 and 5) show an effect of meaningful stimuli in the first five BSL and EQ trials. The group with meaningful comparison stimuli (N-ABS) had the lowest fixation duration in the first five BSL and EQ test trials compared to those with abstract comparison stimuli in these test trials. This pattern might be due to the meaningful stimuli discriminability and discrimination history in the individual repertoire discussed above, reducing the need for long fixations. There is no reversed effect in symmetry trials at the beginning of testing; one might have assumed that the fixation duration to meaningful comparison stimuli in the N-PIC group would be lower than the duration of the fixations to abstract comparison stimuli in the N-ABS group. One interpretation might be that stimuli properties have no effect in SYM trials. A second interpretation could be that the reduced difference in fixation duration to meaningful stimuli in SYM trials is due to the novelty of the symmetry trial or, more specifically, to the novelty of the comparison stimuli’s function. In both BSL and EQ trials, the comparison stimuli also function as comparison stimuli in training, whereas in SYM trials, the previously trained sample stimuli function as the comparison stimuli. Testing symmetry trials after conditional discrimination training with an LS training structure could test this hypothesis, as some of the stimuli in a larger stimulus class function as sample and comparison stimuli in training and, therefore, have no new functions in symmetry trials.

Visually, fixation duration is lower for all relations from the first five to the last five trials. Still, meaningful stimuli do not affect fixation duration to correct comparison stimulus in the last five test trials. However, there are differences in fixation duration to incorrect comparison stimuli in the five last SYM and EQ trials (discussed later). The mean averaged fixation duration to incorrect comparison stimuli at the end of the test is very low for all relations, often below 200 ms, which is lower or around the fixation duration threshold for an individual to observe a stimulus. The reason for this low number can be multifaceted. First, there are probably several trials where participants never gaze at the incorrect stimuli. Instead, participants’ gaze moves directly to the correct comparison stimulus before a response, or participants gaze between the correct comparison stimulus and the sample stimulus (discussed below). Therefore, the end score is somewhat artificially low when averaging the duration across trials within participants and across participants. Secondly, the low fixation duration may be an artifact of how we calculate fixation duration to incorrect comparison stimuli. In the present study, we add all the duration (100 ms or above) and divide the number by two because two incorrect stimuli are simultaneously on the screen. This does not mean the participants’ gaze was fixated on both incorrect stimuli in each trial. The authors did not find a way to calculate fixation time to incorrect comparison stimuli to avoid this artifact. We did consider separating it into incorrect stimuli 1 and 2. But we were left questioning which stimuli were 1 and 2 in each trial. Ultimately, this type of analysis would render other artifacts. Visual inspections of the graphs for fixation duration to both correct and incorrect comparison stimuli in the test show a stepwise upward pattern from baseline to symmetry and from symmetry to equivalence trials in all groups, which replicates previous experiments reporting eye movements (Sadeghi & Arntzen, 2018; Steingrimsdottir & Arntzen, 2016).

The number of gaze transitions and refixations to sample stimulus

The test also shows an effect of including meaningful stimuli in the class on the number of transitions in some tested relations. The group with only abstract stimuli has more transitions in BSL trials (significant differences) and equivalence trials (visual inspection) at the beginning of the test. Interestingly, the average number of transitions in EQ trials in the N-ABS and the N-PIC groups are identical, even though all the stimuli presented on the screen in EQ trials are meaningful for the N-ABS groups, and all the stimuli are abstract in EQ trials for the N-PIC group, like the ABS groups. These results might suggest that a meaningful stimulus in an equivalence class reduces the need to gaze back and forth, even though the meaningful stimuli are not present on the screen in EQ trials, which might be related to stimulus control, as discussed above. This interpretation is speculative since there were variations in the number of transitions for each participant in the first five EQ trials for the ABS group (range = 4.2–25.8) and not significantly different from the other groups. Significant differences exist between the groups in the last five SYM trials, between the N-PIC and ABS groups, showing reduced gaze transitions when comparison stimuli are meaningful.

When the sample stimulus is meaningful, participants refixate on the stimulus less than when the sample stimulus is abstract in all the tested relations (significant for BSL and EQ trials). Again, this might indicate that the group trained with a meaningful sample stimulus might have established stimulus control in a way that reduced gaze transitions between comparison stimuli and the sample stimulus in the test, also when the sample stimulus no longer were meaningful stimuli, i.e., in symmetry and equivalence trials.

Synthesis

When extracting data from a continuous behavior stream like eye movements, it is challenging to get a complete picture of what is happening, and often, there are variations across trials. However, each analysis paints a broader stroke, creating a picture piece by piece. Looking at the analysis of incorrect comparison stimuli, the number of transitions, and the number of trials with refixation to the sample stimulus together regarding the tested relations, it is reasonable to assume that most of the transitions were between the correct comparison and the sample stimulus due to the low number of trials with refixations to sample stimulus and low fixation duration to incorrect comparison stimuli. The only relation that stands out regarding gaze transitions is equivalence trials, especially at the beginning of the test. In these trials, many transitions are to the sample stimulus but probably more often between different comparison stimuli due to the high score on fixation duration to incorrect comparison stimuli.

One relation stands out at the end of the test regarding eye movements: symmetry trials for the N-PIC group. The sample stimulus is abstract in these trials, and the comparison stimuli are meaningful (see Panel a, Figure 2). Here, fixation duration to incorrect stimuli is extremely low, with few transitions. This indicates that after the initial fixation on the sample stimulus, participants fixated on the correct comparison stimulus straight after. Only on a few occasions do the participants fixate back on the sample stimulus or any of the incorrect comparison stimuli. This means that when participants fixate on the sample stimulus when the comparison stimuli appear, they probably discriminate the comparison stimuli in the periphery before moving their gaze to the correct stimulus. Their gaze does not scan the screen to locate the correct comparison stimulus.

Additionally, when doing this data analysis, the first author noticed that in some trials, participants did not move their gaze from the sample stimulus before responding to the correct stimuli but matched stimuli using peripheral vision. Both Schroeder (1970) and Pessôa et al. (2009) report on similar observations. In discussions in the literature about the effect of peripheral vision, it is mentioned that peripheral vision probably affects both fixation measures from the beginning of training (Perez et al., 2015) and throughout training as stimulus control increases (Hansen & Arntzen, 2018). Also, the stimuli’s physical features might affect peripheral discrimination (Schroeder, 1970). Based on such observations, we examined the extent of peripheral vision in MTS procedures using a set-up similar to traditional threshold experiments (Braaten & Arntzen, 2022). In that experiment, participants were forced to fixate in the middle of the screen while responding to stimuli varying in size and position. Results showed that participants could, to a large extent, discriminate stimuli in the periphery, and only when abstract stimuli are 0.7 cm in size or smaller and 12–18 cm from the fixation point does discrimination decrease. Also, the results showed that familiar and simplistic stimuli are even more discriminable in the periphery and must be 0.3 cm and 12–18 cm from the fixation point before the discrimination index decreases. Therefore, the present experiment size of stimuli and positions on the screen make it likely that participants discriminated stimuli in the periphery (See Braaten and Arntzen's dicussion section for further elaboration on the implications of interpreting eye-tracking data and procedural suggestions for reducing discrimination in the peripheral vision).

Limitations and future experiments

Determining whether the present effect of meaningful stimuli on eye movements is due to a pre-experimental discriminative function in the participants’ repertoire or whether it is the features of the stimuli, i.e., shape and color, is difficult. One possible way to make this distinction in future research could be to include a group where some abstract stimuli are pre-trained with a discriminative function before entering a stimulus class, as done by Nartey et al. (2015).

Repeated exposure to the trained relations and increasing stimulus control reduce fixation duration to both sample and comparison stimuli. However, the reduced fixation duration might also result from a practice effect or enhanced peripheral discrimination due to repeated exposure. One way to test this is to compare the eye-movement results between participants who learn conditional discriminations and those who do not, similar to what Hansen and Arntzen (2018) and Steingrimsdottir and Arntzen (2016) did. If the fixation measures decrease for both groups, the reduction is due to repeated exposure to the procedure and stimuli. Provided that participants who do not learn the conditional discriminations have not established participant-defined stimulus classes as opposed to experimenter-defined stimulus classes. In the present study, we could not compare the eye movements of those who formed equivalence classes and those who did not. Future experiments can create more variability in the yields while using the OTM training structure by either increasing the number of classes participants learn (e.g., Hansen & Arntzen, 2018) or increasing the number of members in each class to seven or more (e.g., Ayres-Pereira & Arntzen, 2019). Additionally, including a group where all stimuli were meaningful would allow for comparing the effects on eye movements, especially for trials testing symmetrical relations.

Utilizing eye-tracking technology to measure eye movements results in large amounts of data and many different analytic possibilities. The present analysis has been done to compare the results with previous research and expand how eye-tracking measures are analyzed in this line of research. There are possibilities to analyze the data even more fine-grained, for example, by looking into each stimulus or separating each relation to see if one class or relation is different regarding eye movement. Maybe some stimuli are more easily discriminated in the periphery? Or do some specific relations result in many transitions? As our analytical tools get more advanced through software programming, we will have more opportunities to do advanced analysis.

Concluding remarks

The present experiment shows that meaningful stimuli affect eye movements for several measures, both in training and testing. Having at least one meaningful stimulus in the stimulus class seems to reduce fixation time and gaze transitions, also on trials where the meaningful stimuli are not a part of the relation tested compared to when all stimuli in the class are abstract. This may indicate that the degree of stimulus control differs, though accuracy scores are similar. Measuring eye movements with eye-tracking technology can provide information about behavioral differences between conditions not obtained by accuracy scores, giving us a more detailed analysis of behavior.

Ethical approval

All procedures performed in studies involving human participants were in accordance with the ethical standards, the 1964 Helsinki Declaration, and its later amendments or comparable ethical standards. The present research was assessed by the Norwegian Center of Research Data (Ref. No. 922412).

Informed consent

Informed consent was obtained from all participants in the study.

Acknowledgments

We want to thank Jan A. Østby, Julie Larsen, Juan Minguela Cabrera, Anouchka Moflag Candia, and Emma Aasen Holtan for their assistance in the data collection.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research is not funded