Characterizing production: the production effect is eliminated for unusual voices unless they are frequent at study

ABSTRACT

The production effect refers to the finding that items read aloud are better remembered than items read silently. This is often explained with reference to distinctiveness, arguing that aloud items become associated with distinctive sensorimotor features that facilitate retrieval at test. Based on this framework, more distinctive forms of production should result in larger production effects. The present study tested this theory by having participants study items silently or aloud in either their own voice or as a popular character. Participants were then tested for those items using recognition memory. Relative to silent items, aloud items read in the participants’ own voice demonstrated a typical production effect; however, contrary to any predictions, no production effect was observed for the character voices. We next manipulated how frequently the character voice was used relative to the participants’ own voice. This revealed a production effect for character voices only when those voices were more common than the participant’s own voice. This pattern could not be attributed to cognitive demands or performance anxiety but was predicted by a novel computational account based on the Retrieving Effectively from Memory (REM) model. Our results show that the relation between distinctiveness and memory is not necessarily linear.

KEYWORDS:

• Production effect
• memory
• distinctiveness
• recognition memory

Acknowledgments

The authors would like to thank Kathleen Prior, Rachel Hewitt, Emily Buchanan, and Michelle Johnson for contributions to data collection. We would also like to thank our voice actor Garrett Martin for narrating the voices used in our audio recordings. Lastly, we would like to thank The Combine music for lending their recording equipment.

Disclosure statement

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

Open practices statement and data availability statement

None of the data or materials for the experiments reported here are available online, and neither of the experiments were preregistered. However, all data and materials reported in this article are available from the corresponding author on request.

Notes

1 Prior to conducting our study, we generated a large number of prospective character voices, including both male and female characters, which we then had a sample of ∼12 graduate and undergraduate laboratory members attempt to impersonate and rate for similarity to their own voice using a multidimensional scaling task. The final voices were selected on the basis that they were well-known to our target demographic, easily imitated and as differentiated from their typical speaking voice as possible. Regrettably, all female characters were removed because they were either judged to be insufficiently distinctive (e.g., Lois Griffin) or too similar to a more common male character (e.g., Minnie as compared to Mickey Mouse). Importantly, analyses of our data as a function of participant sex demonstrate the same pattern across all measures for male and female participants. Also, without access to acoustic information pertaining to the voices used, we freely admit we are unable to isolate what it is about the individual voices that drives any observed effects. Importantly, the voices used in Experiment 1 (each possessing distinctive acoustic properties) all demonstrate the same pattern described below when analyzed separately, suggesting the specific acoustics are perhaps not important.

2 Between the test phase and questionnaires, participants also completed a phase in which they rated their familiarity with each of the character voices (amongst others) and undertook a multidimensional scaling exercise meant to quantify conceptual similarity between these characters. However, given the unexpected outcome of our analyses (described next) these data were neither processed nor analyzed; as a result, we will not discuss them further.

3 We would like to thank Drs. Aaron Newman and Colin MacLeod for independently proposing this test of the cognitive demands account when this work was presented at the 2018 annual meeting of the Canadian Society for Brain, Behaviour and Cognitive Science.

4 We have chosen to present the Experiment 1 data here, too, rather than earlier, owing to the fact that with so little data we felt it better to consume this analysis as a whole. Also, defining “switching” based on condition (voice, aloud, silent) or persona (self, voice – here assuming silent items are self) produces the same pattern.

5 A corollary of the cognitive demand interpretation would be that the present findings imply that production alone is insufficient to effect the memory benefit; should the difference between groups in Experiment 2 be due to cognitive demands masking the production effect early in the voice trials, it implies that attention plays a greater role in the emergence of the production effect than permitted by classic interpretations of the distinctiveness account, because according to that perspective having someone produce a word in an unusual, effortful manner would be sufficient to undermine the effect, despite the fact that production did occur (for discussion of the role of attention, see Fawcett & Ozubko, 2016; Mama & Icht, 2018; Mama et al., 2018; Ozubko et al., 2012).

6 Concurrent to – and independent of – our own efforts to model the production effect using REM, Kelly, Ensor, Liu, MacLeod and Risko (2022) produced their own, comparable implementation using this model. We learnt of each other only during revision of our respective manuscripts.

7 Here, we are focusing on the combined analysis of Experiments 1a and b, with an effect of voice on the production effect being reflected in the interaction between production condition (listen, repeat) and production “instruction” (own voice, imitate) and higher order interactions, including with accent (American, Dutch). Arguably, one could instead consider the interaction between production condition (listen, repeat) and accent (American, Dutch) in Experiment 1b, wherein participants imitated in all cases, as reflecting an interaction between voice and the production effect because many participants were probably American and therefore imitation would be similar to using one’s own voice; however, imitation in that case would still necessitate some degree of alteration to one’s manner of speaking, and further, it is plausible that not all participants had the same accent as the speaker. Regardless, the conclusions were similar once averaged across intelligibility (easy, hard).

8 We thank an anonymous reviewer for orienting us to this work. Although we were unaware of it, we had spoken to the research team in question preceding their project whilst presenting the current experiments also at the annual meeting of the Canadian Society for Brain, Behaviour and Cognitive Science.

9 Here, one might challenge whether reading a voice as a character rather than one’s own voice even manipulates distinctiveness. While distinctiveness is a difficult concept to define without resorting to circular reasoning (e.g., Hunt, 2006), we would argue that reading a word as a character ought to be distinctive, as it is not something that participants often do (meaning that it should stand out in memory) and further that doing so has been shown to activate neural patterns distinct from one’s own voice (McGettigan et al., 2013). Further, singing was declared a manipulation of distinctiveness with similar reasoning (e.g., Quinlan & Taylor, 2013).

Additional information

Funding

JMF was funded by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant.