6  General Discussion

Interfering with stimulus perception during encoding can sometimes improve later explicit memory. The mixed data on perceptual disfluency has called into question the utility of such manipulations in the learning domain. One of the main aims of the current set of experiments was to examine the underlying mechanisms of the perceptual disfluency effect to better understand when perceptual disfluendey aids memory and when it does not. To this end, our study delved into the impact of one type of perceptual disfluency, blurring (i.e., low blurring and high blurring), on the process of encoding, as assessed through a LDT (Experiments 1a and 1b), and a semantic categorization task (Experiment 2). RT distributions were analyzed with an ex-Gaussian model and DDM (Experiments 1a and 1b) to better understand how perceptual disfluency affects encoding. These models offered a comprehensive descriptive and theoretical framework through which to examine the perceptual disfluency effect.

To recapitulate our findings, during encoding, high blurred words showed greater distributional shifting and skewing compared to clear and low blurred words. In addition, DDM fits indicated high blurred words had a higher \(T_{er}\) and lower \(v\). Conversely, low blurred words compared to clear words showed greater distributional shifting, but there was no difference in skewing. DDM fits showed higher \(T_{er}\), but had no effect on \(v\).

Turning to recognition memory, high blurred words were more likely to be recognized at test compared to clear words and low blurred words. This pattern arose regardless if context was reinstated at test (Experiment 1b). This pattern replicates the results from Rosner et al. (2015). In addition, we showed word frequency (Experiment 2) also modulates the disfluency effect. Namely, low frequency words did not show a disfluency effect. In fact, the effect seemed to be reversed for clear words and low blurred words. However, high frequency words did show a disfluency effect.

These findings have several implications. At a theoretical level, the current data suggests that in order for perceptual disfluency to benefit memory it has to be disfluent enough to affect both early and late stages of processing. A manipulation that only produces a general slowing of responses is not sufficient to enact an mnemonic effect. However, an important caveat to this is that processes during encoding of the word itself are not enough to produce an menominc benefit. In Experiment 2, we did not observe better memory for low frequency-high blurred words which are the hardest and presumably receive the most top-down processing. We only observed a disfluency effect for high frequency-high blurred words. This points to the importance of control processes and processing limitations in producing the disfluency effect.

We argue the current findings align more closely with the stage-specific account proposed by Ptok et al. (2019). While the account was proposed to explain memory effects that require conflict during encoding, like the semantic interference effect, we feel that it is a useful theoretical framework to explain the current findings. In fact, Ptok et al. (2019) and Ptok et al. (2020) suggested a connection between their framework and perceptual disfluency effects and desirable difficulties, more broadly.

Within the stage-specific account, memory performance depends on the nature of processing during encoding and the utilization of cognitive control mechanisms. In our experiments, participants were tasked with determining whether letter strings represented words or non-words (Experiments 1a and 1b), or whether a word belonged to the category of animals (Experiment 2). For skilled readers, these tasks are executed automatically and smoothly. Coupled with perceptual disfluency, this combination is believed to lead to memory advantages seen with perceptual disfluent stimuli.

When we manipulated word frequency, however, the process of recognizing low-frequency words demanded more effort and attentional resources in addition to the perceptual disfluency of blurring. Consequently, this lead to the task becoming more challenging and difficult. Thus, the increased processing demands from recognizing low freqeuncy words may have countered the benefits from high-blurred words, as more attentional resources were allocated to recognizing low-frequency words. This explanation supports our findings of a disfluency effect for high-frequency words, but not low-frequency words.

There are other examples that support this capacity-limited view of perceptual disfluency. For instance, Geller et al. (2018) showed that easy-to-read cursive words and hard-to-read cursive words are better remembered than computer print words, but the memory effect is much larger for easier to read cursive words. As another notable example, participants with low working memory capacity do not seem to benefit from perecptaul disfluency as much as those with higher working memory capacity (Lehmann et al., 2015). At a broader level, Wenzel & Reinhard (2019) suggested that intelligence is an important factor for when desirable difficulties are desirable for learning.

At a methodological level, our experiments demonstrate that a straightforward blurring manipulation can benefit memory, which we observed whether or not we reinstated the context during testing. However, blurring has to be sufficiently difficult do so. If the secondary task requires too much attentional control

More significantly, our current experiments underscore the benefits of using the ex-Gaussian model and DDM to look at RT distributions. For instance, while previous research often validates perceptual manipulations by examining accuracy and overall response time (Geller et al., 2018, 2020). Our experiments revealed that words with high and low levels of blurring affect response time distributions differently. Specifically, low blurred words influenced early or non-decision stages of processing, whereas high blurred words affected both early and late stages. These findings suggest that T_er and v are sensitive to manipulations of perceptual disfluency (also see (gomez?) )

Furthermore, our modeling of RTs appears to be a more sensitive method. Although we found weak evidence for differences between low blurred and high blurred words in accuracy and average response times between low-blur and clear words in Experiments 1a and 1b, we did notice variations in non-decision time and a shift in the response time distribution for low blurred words compared to clear words. We recommend that future studies employ distribution modeling and DDM to decompose response times and directly quantify the impact of perceptual disfluency on encoding.

Important to note here, there is one case of the DDM being applied to the study of perceptual disfluency. In one recent study, Hu et al. (2022) examined RTs during retrieval. Hu et al. (2022) were primarily interested in how perceptual disfluency (i.e., Sans Forgetica typeface and blurring) influenced DDM parameters during recognition and how they relate to confidence judgments. At test,they found a non-significant difference in mean RTs between Sans Forgetica typeface and Arial typeface. However, looking at the DDM parameters, Sans Forgetica and Airal typefaces differed on \(T_{er}\), but not drift rate. They also looked at how parameters of the DDM were related to confidence judgments. Higher \(T_{er}\) was related to lower confidence and higher drift rate was associated with higher confidence. While their focus was on retrieval and not encoding, this corresponds with what we observed for our low blurred words in Experiments 1a and 1b–a weak manipulation affected non-decision time, but not drift rate. Overall, this further highlights the utility of the the DDM in studying perceptual disfluency (and other encoding conflict effects) during encoding and test.

Finally, at a practical level, we do show that blurring can benefit later memory. However, caution needs to be taken here. First, the current experiments were conducted online using simple materials (i.e., list learning). It is unclear how these effects would generalize to a classroom setting. Second, participants were not told about the upcoming test. Geller & Peterson (2021a) has showed that low test expectancy is an important moderator for this effect. Third, while we did not establish a region of practical equivalence, the size of the disfluency effect appears to be be small in nature. Looking at the default region of proximal equuivalence (ROPE) from the bayestestr package (here -0.10 -.10 in standardized units) many of the critical contrasts either were completely inside this region or overlapped this region, suggesting negligible differences. For more applied work, this might be well below the smallest effect size of interest. Now this is not to say all research on perceptual disfluency is superfluous. As noted in Geller & Peterson (2021b), a fruitful avenue for future work might be to investigate how perceptual influenced effects processing in every day life where memory is largely incidental (Castel et al., n.d.).

These results provide some context for the large number of replication failures. Many of the studies looking at disfluency do not take care in ensuring the disfluency manipulation is actually disfluent. Most times, studies use only two levels (disfluent/fluent) and perform more traditonal analyses. As we have hopefully shown here, it is important to take into account the entire RT distribution. By examining the RT distributions of different levels of disfluency we obtained a richer better understanding of the stages or loci manipulations have during encoding. It is our hope that learning and memory researches will begin to use these tools to help understand encoding processes involved in perceptual disfluency effects but also encoding contexts where there is considerable conflict.

6.1 Conclusion

Our paper contributes nuanced insights to the intricate relationship between perceptual disfluency and memory encoding. We have shown that perceptual disfluency can aid in memory retention, but its efficacy is contingent upon the degree of disfluency and other contextual factors such as word frequency. Our findings endorse the stage-specific account, emphasizing the role of cognitive control mechanisms in the observed memory advantages with perceptual disfluency. Furthermore, our methodological contributions, employing an ex-Gaussian model and DDM, not only validate the benefits of examining RT distributions, but also open new avenues for future research in learning and memory studies. We caution, however, that the applicability of these findings in real-world educational settings remains an open question, and the effect sizes observed were relatively small, thus warranting further investigation.

Ultimately, this work stands as a call to action for a more comprehensive, nuanced approach to studying perceptual disfluency, incorporating both advanced statistical methods and a more exhaustive range of experimental conditions to better elucidate when and how disfluency can facilitate memory.

Castel, A. D., Nazarian, M., & Blake, A. B. (n.d.). Attention and incidental memory in everyday settings PROPERTY OF THE MIT PRESS FOR PROOFREADING, INDEXING, AND PROMOTIONAL PURPOSES ONLY.

Geller, J., Davis, S. D., & Peterson, D. J. (2020). Sans forgetica is not desirable for learning. Memory, 28(8), 957–967. https://doi.org/10.1080/09658211.2020.1797096

Geller, J., & Peterson, D. (2021a). Is this going to be on the test? Test expectancy moderates the disfluency effect with sans forgetica. Journal of Experimental Psychology: Learning, Memory, and Cognition, 47(12), 1924–1938. https://doi.org/10.1037/xlm0001042

Geller, J., & Peterson, D. (2021b). Is this going to be on the test? Test expectancy moderates the disfluency effect with sans forgetica. Journal of Experimental Psychology: Learning, Memory, and Cognition, 47(12), 1924–1938. https://doi.org/10.1037/xlm0001042

Geller, J., Still, M. L., Dark, V. J., & Carpenter, S. K. (2018). Would disfluency by any other name still be disfluent? Examining the disfluency effect with cursive handwriting. Memory and Cognition, 46(7), 11091126. https://doi.org/10.3758/s13421-018-0824-6

Hu, X., Yang, C., & Luo, L. (2022). Retrospective confidence rating about memory performance is affected by both retrieval fluency and non-decision time. Metacognition and Learning, 17(2), 651–681. https://doi.org/10.1007/s11409-022-09303-0

Lehmann, J., Goussios, C., & Seufert, T. (2015). Working memory capacity and disfluency effect: an aptitude-treatment-interaction study. Metacognition and Learning, 11(1), 89–105. https://doi.org/10.1007/s11409-015-9149-z

Ptok, M. J., Hannah, K. E., & Watter, S. (2020). Memory effects of conflict and cognitive control are processing stage-specific: evidence from pupillometry. Psychological Research, 85(3), 1029–1046. https://doi.org/10.1007/s00426-020-01295-3

Ptok, M. J., Thomson, S. J., Humphreys, K. R., & Watter, S. (2019). Congruency encoding effects on recognition memory: A stage-specific account of desirable difficulty. Frontiers in Psychology, 10. https://doi.org/10.3389/fpsyg.2019.00858

Rosner, T. M., Davis, H., & Milliken, B. (2015). Perceptual blurring and recognition memory: A desirable difficulty effect revealed. Acta Psychologica, 160, 11–22. https://doi.org/10.1016/j.actpsy.2015.06.006

Wenzel, K., & Reinhard, M.-A. (2019). Relatively unintelligent individuals do not benefit from intentionally hindered learning: The role of desirable difficulties. Intelligence, 77, 101405. https://doi.org/10.1016/j.intell.2019.101405