Context plays a crucial role in episodic memory. Reinstating the encoding context in the test phase may enhance memory retrieval relative to changing the context (Hockley and Bancroft, 2015, Isarida and Isarida, 2014, Roediger et al., 2017, Smith, 2014). For example, going back to an old place such as our university campus often brings back memories of the past that might never be remembered in other places. Numerous studies have previously established the positive effects of context reinstatement on episodic memory, particularly in recall tasks (Godden and Baddeley, 1975, Smith, 1985, Smith and Vela, 2001, Smith et al., 1978). According to the encoding specificity principle (Tulving & Thomson, 1973), context information presented concurrently with the to-be-remembered information at study can serve as a retrieval cue for accessing the target information from stored memory traces during the test phase.

It is noteworthy that context reinstatement effect is less consistent in recognition memory tasks (Isarida and Isarida, 2014, Roediger et al., 2017). The contextual dependence of recognition discrimination, which pertains to the ability to distinguish between old and new test items in memory, was not supported in early studies (Godden and Baddeley, 1980, Murnane and Phelps, 1995, Murnane and Phelps, 1994, 1993; Smith et al., 1978). Specifically, these studies generally revealed higher rates of both hit and false alarm responses to test items when the test context was old relative to the new context, which is commonly referred to as the concordant effect. There were no notable discrepancies in memory discrimination accuracy between the conditions of the old and new contexts. The global matching model (e.g., Clark & Gronlund, 1996) explains the concordant effect by assuming that the context automatically acts as a retrieval cue to match with the memory traces, thus augmenting the familiarity of both old and new test items. Fortunately, subsequent research has demonstrated reliable context reinstatement effects on recognition discrimination accuracy, especially when the context is intentionally encoded together with the to-be-remembered items (Hanczakowski et al., 2014, Hockley, 2008, Hockley et al., 2012, Isarida et al., 2020, Murnane et al., 1999, Shahabuddin and Smith, 2016). Therefore, it is possible that reinstated context in a test is more likely to improve recognition memory when individuals successfully associate the study items with their contexts during encoding (Hockley & Bancroft, 2015).

However, the neurocognitive mechanisms underlying context reinstatement effects in recognition memory are not yet clear. It is well-accepted that recognition memory can be underpinned by two dissociable cognitive processes, namely familiarity and recollection (Yonelinas, 2002). While familiarity is a rapid process that occurs without the need for retrieval of specific details of a previously experienced event, recollection is a conscious process that necessitates the retrieval of relevant event details. To date, two competing views have attempted to explain which cognitive processes support the beneficial impacts of context reinstatement on memory performance.

The item, associated context, and ensemble (ICE) model, an extended version of the global matching model, proposes that not only the context itself but also the ensemble information influences the recognition memory performance (Murnane et al., 1999). Ensemble information, as a unique memory representation formed by integrating the study item with context during encoding, would additionally increase the hit rates to study items when the context is reinstated at test, thus resulting in a recognition discrimination advantage in the old context owing to the enhanced familiarity-based recognition judgments during the test.

The dual-process account offers an alternative perspective (Gruppuso et al., 2007, Macken, 2002). It proposes that the beneficial effects of context reinstatement on item recognition memory are driven by facilitated recollection during retrieval. Specifically, individuals are more likely to recognize the study items when the exact context paired with the item during encoding is reinstated during the test because they have bound the study items to their contexts during encoding and subsequently consciously recollect the specific item–context associations. Several studies support this account by showing that context effects in recognition tests are accompanied by recollection-based rather than familiarity-based recognition judgments using the remember/know procedure (Diana et al., 2013, Hockley, 2008, Gruppuso et al., 2007, Macken, 2002).

Notably, most of the previous studies investigating context effects in recognition memory have used recognition tests containing studied and completely new items (Ensor et al., 2023, Hanczakowski et al., 2015, Isarida et al., 2020, Shahabuddin and Smith, 2016). It remains unclear whether context reinstatement effects also occur for recognition discrimination between old items and physically similar new items. It is relatively difficult to discriminate among similar items in recognition memory, which is thought to depend on pattern separation, a neurocomputational process by which similar items sharing overlapping features with old items are transformed into less similar ones (Leal and Yassa, 2018, Yassa and Stark, 2011). To date, three studies have addressed this question by examining the impacts of background scene pictures on object recognition memory (Doss et al., 2018, Racsmány et al., 2021, Szőllősi et al., 2023). All these studies found that picture context reinstatement increased both correct responses to old objects and false alarm rates for similar objects, while not affecting recognition discrimination between old and similar objects.

The lack of beneficial outcomes of context reinstatement on the ability to distinguish between old and similar objects in these studies may be attributed to several methodological issues. On the one hand, participants were engaged in an incidental encoding task during the study phase and were unaware of the memory test that would be conducted in the future (Doss et al., 2018, Racsmány et al., 2021, Szőllősi et al., 2023). On the other hand, participants were directed to concentrate solely on the objects during encoding (Racsmány et al., 2021, Szőllősi et al., 2023). These manipulations may have impeded participants from establishing unique object–context associations during encoding, thus potentially diminishing the memory advantage in the old context condition. Interestingly, despite Doss et al. (2018) instructing participants to create associations between the objects and their contexts during the study phase, no context reinstatement effects on the ability to discriminate between old and similar objects were observed. This may be due to the contextual manipulation by which a particular context had to be linked with multiple study items instead of exclusively linking it with one study item, thus compromising the benefits of context reinstatement on recognition memory because of the interference caused by competing associations. In support of this view, it has been shown that it becomes more difficult for participants to retrieve specific associations as the number of associations between different study items and the same context increases (Hockley and Bancroft, 2015, Reder et al., 2013).

The first objective of the present study was to examine the contextual influence of background scenes on recognition memory in situations that require differentiation between old and similar objects, using the context reinstatement paradigm. We used common and familiar objects as study items and semantically unrelated background pictures of real-world scenes as contexts (see Figure 1 for example). Each object was paired with a unique background scene during encoding. Critically, the participants were required to associate objects with background scenes intentionally during encoding. Afterward, they undertook a recognition memory task that comprised of studied objects and similar objects that were physically and conceptually similar to the studied objects (e.g., two different suitcases) regardless of their background contexts. In the old context condition, the test objects were presented with pictures of the old background scene from the study phase. Notably, we manipulated the new context condition by presenting the test objects together with new background scenes that were physically and categorically similar to the scenes from the study phase (e.g., two different bookshelves).

It has been suggested that representations of associative information in episodic memory can be remembered at different levels of specificity (Greene et al., 2022, Greene and Naveh-Benjamin, 2020, Greene and Naveh-Benjamin, 2022, Greene and Naveh-Benjamin, 2023). The highly specific representation of an association refers to a representation that retains precise information about which specific components had been associated together during encoding (e.g., remembering that the helmet had been paired with the particular desert scene). The gist representation of an association refers to a representation of association at a less specific or detailed level (e.g., remembering that the helmet had been paired with a desert scene, but not remembering specifically which desert scene). In the present study, it is supposed that highly specific representations of associative information would be retrieved for the old objects in the old context condition, while less specific or gist representations of associative information would be retrieved for the old objects in the similar context condition. As a result, by taking the similar context as a control condition, we can examine the effects of the specificity of object-context associations on recognition memory.

Furthermore, the present study aimed to elucidate the cognitive mechanisms that underlie the beneficial effects of context reinstatement on memory performance. To this end, electroencephalogram (EEG) was captured, and event-related potentials (ERP) were extracted during the test phase. The ERP technique is widely recognized for its ability to offer a reliable measure of the subprocesses involved in the retrieval of episodic memory (Rugg et al., 1998). Two ERP old/new effects have been identified in previous ERP studies, which can be viewed as potential indicators of familiarity and recollection, respectively. The early mid-frontal old/new effect that occurs approximately 300 ms after stimulus onset—maximal at frontal recording sites—is associated with familiarity-based recognition (Mecklinger and Bader, 2020, Rugg and Curran, 2007). The late parietal old/new effect that emerges around 500 ms post-stimulus is indicative of recollection-based remembering (Friedman and Johnson, 2000, Rugg and Curran, 2007). The topography of this effect exhibits a parietal maximum for words but is more widespread for pictures (Gutchess et al., 2007, Höltje and Mecklinger, 2020). By comparing the ERP old/new effects between reinstated and similar contexts, we can explore the specific cognitive processes underlying the advantages of context reinstatement in discrimination among similar items in recognition memory. Of importance, we can investigate the electrophysiological correlates of the specificity of object-context associations in episodic memory.

Behaviorally, we predicted that reinstatement of the encoding context relative to the similar context during retrieval would lead to an enhanced ability to distinguish between old and similar objects. According to the ICE model, context reinstatement effects on object recognition would be accompanied by an increased mid-frontal old/new effect related to familiarity during object recognition in the old context in comparison to the similar context condition. Conversely, according to the dual-process account and specificity principle of memory, if individuals successfully recognized the study objects because they recollected highly specific details about object–context associations under the old context condition, there would be a more pronounced parietal old/new effect indexing recollection during object recognition in the old context as opposed to the similar context condition.