A model of the production effect over the short-term: The cost of relative distinctiveness

Highlights

• For the first time, the production effect was systematically studied in short-term recall.

• Findings show a complex but systematic impact of production on memory over the short-term.

• Serial positions were found to be central in understanding the production effect.

• Reading the items aloud appears to disrupt rehearsal, while increasing their distinctiveness.

• A revised version of the Feature Model accounts well for the data.

Abstract

The production effect relates to the better memory of words read aloud during a study phase compared to silently read items. Here, we examined the production effect for memory over the short-term. In long-term memory tasks, the effect generates a complex pattern of results where production interacts with memory task and list composition. Within an immediate ordered recall paradigm, involving both item and order information, we tested the item-order account, recently called upon to explain the production effect. We also analysed results as a function of serial position. Results of the first five experiments were highly consistent, but hard to reconcile with the item-order account. Instead, we put forward an interpretation based on relative distinctiveness and the costs of the richer encoding associated with production. The predictions we derived from this interpretation were supported in the final experiment. Moreover, we tested the interpretation through a new version of the Feature Model. Overall, the work highlights the value of the production effect as a prototypical distinctiveness phenomenon illuminating the interaction of encoding and retrieval processes, the value of feature-rich representations, and the costs that can be associated with feature-generating distinctive processing.

Introduction

Recent research on episodic memory has seen interest in the production effect increase (see MacLeod & Bodner, 2017, for a review). Simply put, when some of the words within a list are pronounced aloud – i.e. produced – they tend to be better remembered than those read silently. First identified in the sixties and seventies (e.g., Crowder, 1970, Hopkins and Edwards, 1972, Pollack, 1963), the phenomenon was systematically studied more recently by MacLeod and collaborators, who coined the term production effect (MacLeod, Gopie, Hourihan, Neary, & Ozubko, 2010). As is the case for several encoding effects, the overall pattern of findings proved to be complex, with the strength of the effect depending on the memory task called upon, as well as on list composition and design. Other encoding effects include the generation effect (Serra & Nairne, 1993), the bizarreness effect (McDaniel & Einstein, 1986), the enactment effect (Engelkamp & Dehn, 2000), and others (see, e.g., Nairne, 1988a). In comparison to these, the production effect can be seen as relatively simple – it involves little extra processing compared to generating a bizarre image, for example. Instead, production involves a highly compatible and easy to observe response: reading verbal material aloud. In that sense, it is one of the best encoding effects to explore as it makes it easier to get to the fundamentals of what is causing the obtained patterns. In sum, the production effect can be thought of as a prototypical encoding effect, one that can help to elucidate the principles underlying the influence of such effects on memory.

In the present study, we had two main aims: The first was to determine if the production effect also held for short-term memory (STM), as the work alluded to above focussed on long-term memory. Should our results reveal the basic effect, we wished to systematically document the phenomenon in STM. Our second aim was originally to test one of the promising explanations of encoding effects: The item-order account (McDaniel & Bugg, 2008). As applied to the production effect, the account explains the interaction between list composition, memory task, and production by proposing that a) item-specific and order information compete for encoding resources and b) in some tasks, disruption of order encoding leads to predictable recall difficulties due to the role of order information in guiding retrieval (Jonker et al., 2014, McDaniel and Bugg, 2008). We return to this explanation of the production effect later. At this point, we note that our original objective was to test how well these ideas can account for the production effect in STM; to do so, we called upon well-known short-term memory tasks where item and/or order information are considered central. Below, we briefly review the relevant data in the field as well as how the item-order account might explain the known pattern of findings; we then present five experiments as well as a quantitative model of our results.

To anticipate, the impact of production on some aspects of STM performance was spectacular, but unpredicted by the item-order account. The follow-up studies suggest a compelling interpretation of the production effect in STM, where limited encoding resources interact with relative distinctiveness to produce the rich and complex mnemonic behaviour that we observed.

The account we propose emphasises the importance of local distinctiveness effects as well as trade-offs at the point of encoding, i.e., distinctive processing comes at a cost. We use the expression ‘local distinctiveness’ because our findings and modelling underline the fact that differences between contiguous items—a form of contrast effect—have a very significant effect on STM performance. We argue that the principles highlighted in explaining these findings reveal some core truths about encoding effects and STM functioning more generally. Some of these interpretations may also apply to long-term memory, although we did not test this directly; hence, the latter suggestions are more speculative.

In a number of circumstances, relative to silent reading, production enhances future recall and the effect can last up to a week (Grohe and Weber, 2018, Ozubko et al., 2012) and occur even with an incidental learning procedure (see, e.g., Greene & Pearlman, 1996). Moreover, different forms of production have a positive effect, including singing, mouthing items, spelling, writing, typing, and even imagining typing; that said, except for singing, the latter effects are typically smaller than what is observed when items are read aloud (Forrin et al., 2012, Jamieson and Spear, 2014, MacLeod et al., 2010, Quinlan and Taylor, 2013, Quinlan and Taylor, 2019). Importantly however, the size of the effect depends on list composition and task (MacLeod & Bodner, 2017). In recognition tasks, when lists comprise a mixture of items read aloud versus silently, a sizable production effect is obtained (see, e.g., Gathercole and Conway, 1988, Hopkins and Edwards, 1972, MacLeod et al., 2010). However, the effect is much reduced and less reliable when the two item types are studied separately in what are called pure lists – i.e., lists containing only silently read or only read aloud items (see Fawcett, 2013, for a meta-analysis). With recall tasks, the reported pattern is slightly different; with mixed lists, containing both silently read and aloud items, there is again a strong production advantage, but the effect disappears when pure lists are tested (Forrin and MacLeod, 2016a, Jonker et al., 2014, Lambert et al., 2016).

One promising hypothesis to account for the pattern across tasks and list types is the item-order account (Jonker et al., 2014, McDaniel and Bugg, 2008). According to this view, producing items involves more item-specific processing relative to silently reading items; when pronouncing words aloud, the motor features as well as the auditory features involved would enhance/add to the encoded event. In mixed lists, these more elaborate records would create a relative distinctiveness advantage. The item-order account also emphasises the importance of relational information in recall tasks; more specifically, order information is thought to play a central role in retrieval processes by guiding retrieval in tasks like free recall (see, Beaman and Jones, 1998, Grenfell-Essam et al., 2017).

With respect to order information, the suggestion is that producing items has a negative effect (Forrin and MacLeod, 2016a, Jonker et al., 2014, Lambert et al., 2016). The item-specific processing is thought to have a cost: It disrupts order encoding. In item recognition tasks, order information is not relevant but item-specific information is central; hence, according to the item-order account, in pure lists, produced items should be better recognised than silent items because the extra item-specific information will support recognition. In the case of mixed lists, the same is true; however, the relative distinctiveness of produced items compared to silently read words boosts the recognition advantage for the produced items. In recall tasks, the same relative distinctiveness produces an advantage in mixed lists; it is thought that this advantage overwhelms any negative impact of production on order information. In pure lists, however, the influence of relative distinctiveness is much reduced for produced items; moreover, any advantage of item-specific processing is offset by the disruption of order information encoding that production causes. The results can go in any direction depending on the influence of item-specific processing and of how disruptive it is to order encoding. Because the effects of item-specific processing and disruption of order encoding are in opposite directions, any effect will be small and so far, only null effects have been reported (see Forrin & MacLeod, 2016a).

These relatively precise predictions of the item-order account were tested by Jonker et al. (2014). More specifically, they tested the idea that differential order information processing could account for the production effect findings in recall tasks. In their first experiment, they examined the influence of list type (i.e., mixed or pure lists) on performance using an order reconstruction task. In such a task, the studied items are provided anew at the point of recall and the participants attempt to reproduce the order in which the words were studied. Based on the item-order account, order reconstruction performance should be better for lists read silently than for lists read aloud because reading aloud is thought to disrupt order encoding. Moreover, as the items are provided at the point of recall, any item-specific processing advantage is likely to be reduced. The reasoning in the case of mixed lists is that reading items aloud would hinder order encoding of silent items, while encoding of silent items would support the order encoding of produced items. The result of the latter would be that any order advantage associated with silently read words would be reduced or eliminated. Based on this analysis, Jonker et al. (2014) also predicted that order reconstruction for silent items in mixed lists would be poorer than in pure lists. Conversely, they expected more order errors for aloud items in pure relative to mixed lists. Their first experiment involved lists of eight words followed by a thirty-second distractor task after which participants completed the order reconstruction test. The results of the experiment supported the item-order account predictions. For pure lists, participants were able to remember the studied order of silent items better than the order of aloud items; in contrast, in mixed lists, performance for silent and aloud items was equivalent.

In a second experiment, Jonker et al. (2014) wanted to verify that the use of a reconstruction task had not altered encoding and processing strategies in such a way that the typical production effect was not observable. They used the same order reconstruction task for half of the trials but for the other half, the task was free recall. Participants were only told at the point of retrieval which task they needed to accomplish. The results for the reconstruction task were similar to those of the previous experiment. There was better order reconstruction for pure silent lists, as expected based on the item-order account, while for the mixed list, there was no significant difference between the silently read and produced items (although there was a trend toward better order recall for items read aloud). With respect to free recall, aloud items were better remembered than silent items when they were presented in mixed lists but for pure lists, recall of silent and aloud items could not be distinguished. The authors also examined two order recall measures for the free recall task. These analyses showed that the order of the items was better reproduced in the free recall of pure lists read silently than for pure lists read aloud; moreover, silent reading produced better order memory than what was observed in mixed lists. Order memory in pure aloud lists was not distinguishable from order memory in mixed lists.

The Jonker et al. (2014) study provided clear support for an item-order account with the predictions derived from the account buttressed by the main data of both experiments and when measures of order memory in free recall were considered. The type of episodic memory task called upon was somewhat atypical however, in that the list length was reduced (8 items) relative to standard free recall tasks (15+ items). There is some evidence that the importance of order information in recall increases with shorter lists (Grenfell-Essam et al., 2017).

To address this issue, Lambert et al., 2016, Forrin and MacLeod, 2016a followed up on the Jonker et al. (2014) study. They set out to further test the item-order account with a task involving longer lists and free recall. The general pattern of results in all experiments supported the predictions of the item-order account in that there was no difference between silent and aloud lists for the groups studying pure lists whereas a reliable production effect was found for the mixed list group. However, Lambert et al. (2016) found little evidence that order memory could account for the production effect pattern; they examined two measures of order memory and found no support for the expected differences in order memory predicted by the item-order account. However, as mentioned above, Grenfell-Essam et al. (2017) have shown that in free recall, order information is most critical with shorter lists.

From the above, we conclude that the item-order account seems like a promising explanation for the production effect in general and even more in STM where item and order information are heavily called upon. However, there are inconsistencies in the reported findings; replications are needed, and further research should test the specific predictions attached to the item-order account in a variety of settings.

These were an important part of the aims that we initially pursued when conducting the first experiment described below. We set out with two objectives in mind. The first was to test the item-order account of the production effect within paradigms recognised for calling upon item and order information – namely, immediate serial memory tasks. A pattern of results similar to those reported by Jonker et al. (2014) and by Forrin and MacLeod (2016a) would provide support for the item-order account, while a discrepant pattern of results like the one reported by Lambert et al. (2016) would provide evidence against the account. Second, as the tasks are typically associated to memory over the short-term, we wished to better establish the nature of the production effect within such a setting. The presence of a production effect in typical STM tasks analogue to the production effect observed in typical long-term memory tasks would provide support for the view that the same set of principles govern memory irrespective of the tasks (Surprenant & Neath, 2009). In short-term ordered recall tasks, with pure lists, an advantage of produced items has been shown in some studies (e.g., Conrad & Hull, 1968), but not in others (e.g., Kappel, Harford, Burns, & Anderson, 1973). However, mixed list designs have not yet been investigated in immediate serial recall so the complete pattern of findings remains to be established.