Variable distraction 1 Running head: VARIABLE DISTRACTION Altering the flow of mental time: A test of retrieved-context theory

The flexibility and power of the human memory system are exemplified by its ability to create associations between temporally discontiguous events. Retrieved-context theory proposes that these associations are mediated by indirect associations binding the neural representation of each study event to a temporally sensitive contextual representation, which is used as a retrieval cue during memory search. When past states of this contextual representation are reactivated, this gives rise to temporal organization in recall sequences. Previous empirical and simulation work has established that temporal organization is insensitive to the temporal scale of the study experience: Adding substantial periods of inter-item distraction to a study list does not affect the temporal organization of memories for that list. However, this previous work has only examined experiments in which all periods of inter-item distraction in a study list are increased or decreased by an equivalent amount. Here, we demonstrate and test a previously unexamined prediction of retrieved-context theory: The scale-insensitivity of temporal organization is conditional on how temporal structure is manipulated. Specifically, we show that temporal organization is dependent on the relative duration of the temporal intervals surrounding a study event, in a way predicted by retrieved-context theory. Using Bayesian estimation techniques, we demonstrate that the ability of a model to predict the order of responses in free recall is substantially improved when the duration of inter-item distraction intervals influences the representational structure of temporal context. We contrast the predictions of our retrieved-context model with those of an influential model of temporal distinctiveness. Variable distraction 3 Altering the flow of mental time: A test of retrieved-context theory The importance of understanding how temporal information is represented in the brain has been recognized since the earliest days of psychological research (Ebbinghaus, 1885/1913; James, 1890). In order to explain time-sensitive psychological phenomena in the domain of learning and memory, theorists have developed computational models in which stimulus representations are associated with a temporally sensitive internal representation (Estes, 1955; Bower, 1972; Mensink & Raaijmakers, 1988; Levy, 1996; Wallenstein, Eichenbaum, & Hasselmo, 1998; Howard & Kahana, 2002; Howard, Shankar, Aue, & Criss, in press). Estes laid the foundation for these theories in his work on stimulus sampling theory, showing that a model in which associations are formed between items and a slowly changing internal representation can account for major phenomena in the conditioning literature, including extinction and spontaneous recovery (Estes, 1955). Bower (1972) suggested that a major component of this slowly changing representation could be thought of as reflecting one’s inner mental context, the set of thoughts, goals, plans, and intentions that are active when a stimulus is encountered. Bower’s conception of inner mental context as an ever-changing entity shares much in common with James’ description of the ever-flowing stream of consciousness, the unending cascade of thoughts that make up our experience (James, 1890). Bower used stimulus sampling theory to bring these ideas into contact with experimental data: Under this framework, context is an internal representation that changes gradually as time passes; it is associated with studied material, and can be used as a retrieval cue, to target particular memories that took place in that context. This allowed him to explain a number of time-sensitive behavioral phenomena in both list discrimination paradigms and temporal judgment paradigms. If inner mental context represents things like passing thoughts, goals, and intentions, Variable distraction 4 then this suggests that one can intentionally shift their context, causing events from before the shift to be less memorable. Sahakyan and Kelley (2002) demonstrated this by asking participants to perform an imaginative task during the delay between two study lists (participants imagined what they might do if they had the power of invisibility). This mental activity caused memories for the first list to become less accessible, and memories from the second list to become more accessible, presumably due to reduced proactive interference from the memories of the first list. This suggests that distracting mental activity will cause recently learned things to become less memorable. This has been demonstrated many times in the domain of free recall: A short delay after a list filled with effortful mental activity greatly reduces the recency advantage for end-of-list study items (Postman & Phillips, 1965; Glanzer & Cunitz, 1966; Howard & Kahana, 1999). Petrusic and Jamieson (1978) demonstrated that with increasingly difficult end-of-list distraction tasks, one could progressively dampen memorability of the studied items, to the point where recall of items from all serial positions in a short list were affected. Retrieved-context models provide a computational framework for understanding how contextual change affects performance on memory tests (Howard & Kahana, 2002; Sederberg, Howard, & Kahana, 2008; Polyn, Norman, & Kahana, 2009a; Lohnas, Polyn, & Kahana, in press). Many of the ideas proposed by Bower are preserved in these models: As study items are encountered, they are associated with a gradually changing contextual representation. This contextual representation can be used as a retrieval cue during memory search, supporting the retrieval of memories that were associated with similar contextual states. The critical cognitive operation at the heart of these models involves the retrieval of prior states of the contextual representation, a mechanism that wasn’t described in Bower’s earlier work. In these models, when a particular studied item is recalled, the system also reactivates the contextual state that was associated with that item. This allows the model to perform what Endel Tulving referred to as mental time Variable distraction 5 travel, in which one travels backward on their mental timeline, revisiting, in a sense, their past experience (Tulving, 1993). According to retrieved-context theory, the structure of one’s mental timeline greatly influences the structure of one’s memories. Organizational analyses of memory propose that the associative structure of memory is reflected in the order in which memories are retrieved in a memory search task like free recall (Puff, 1979; Friendly, 1979; Kahana, 1996). When one remembers a particular item, and reactivates the contextual state associated with the study event, this gives rise to temporal organization: The retrieved contextual information becomes part of one’s contextual retrieval cue, causing the memories of the neighboring study events to become more accessible. These dynamics give rise to the contiguity effect of free recall, the general tendency for nearby study events to be recalled successively during memory search. A lag-based conditional response probability analysis reveals the contiguity effect (Kahana, 1996). With this analysis, one calculates the conditional probability of transitions between the studied items, with each transition labeled according to the positional distance (i.e., lag) between the two recalled items. The contiguity effect exhibits itself in terms of the distribution of these lags, with short-lag transitions being much more likely than long-lag transitions, on the whole. In a retrieved-context model, the contextual representation gradually changes throughout the study period. As each item is studied, it is associated with the current state of the contextual representation. During recall, the context representation acts as a retrieval cue, reactivating the representations of studied items. Polyn et al. (2009a) used a spotlight metaphor to describe the dynamics of contextual cuing. The items are spread across a darkened stage, with the placement of each item indicating its position on the mental timeline defined by the contextual representation. The current state of context determines where the spotlight is pointed; if the list just ended, the spotlight illuminates the last few items on the list, giving rise to the recency effect. When a particular item is Variable distraction 6 reactivated, and the context of the study event is retrieved, the spotlight is centered on the just-remembered item. This causes the neighboring items to be illuminated, giving rise to the contiguity effect. The contiguity effect is quite robust; while several seconds of math distraction at the end of a list is sufficient to disrupt the recency effect, adding the same amount of distraction between each and every study item does not affect contiguity at all (Howard & Kahana, 1999; Kahana, 2012; Lohnas & Kahana, 2014). This phenomenon, referred to as long-range contiguity, is challenging for theories of memory in which temporal intervals are bridged by maintaining item information across the delay (Atkinson & Shiffrin, 1968; Raaijmakers & Shiffrin, 1981). Under these theories, math distraction at the end of the list should disrupt maintained item information, leading to a diminished recency effect, as is observed. However, when the same math task is interpolated between the study items, this should disrupt the representation of the item preceding the delay. This should lead to diminished associations between neighboring items, and consequently a diminished contiguity effect, which is not observed (Howard & Kahana, 1999). Sederberg et al. (2008) demonstrated that retrieved-context models can account for the long-range contiguity effect. Under their model, the distracting mental activity disrupts the contextual representation by causing it to change. Going back to the spotlight metaphor, this is like evenly increasing the spacing between all of the items on the stage, as depicted schematically in Figure 1a. When the spotlight is centered on a just-recalled item, the illumination of both of its neighbors decreases by a similar amount. Recall is a competitive process, so the relative likelihood of recalling either of these items next is unchanged. Even when the length of the inter-item distraction intervals are doubled from 8 to 16 seconds, the shape of the contiguity effect is unchanged (Lohnas & Kahana, 2014). Retrieved-context theory proposes that while the contiguity effect is insensitive to this manipulation, the memory system itself is quite sensitive to the amount of inter-item Variable distraction 7 distraction. Each distraction interval perturbs the representation of temporal context, causing the mental timeline to advance. Evenly increasing these distraction intervals does not alter temporal organization, because the relative support for surrounding items is unchanged. However, retrieved-context theory makes a heretofore untested prediction: One should be able to experimentally manipulate the shape of the contiguity effect, by changing the durations of these inter-item distraction intervals in an uneven manner. Figure 1b depicts this in terms of the spotlight analogy. Here, we provide empirical support for this previously untested prediction of retrieved-context theory. Using a retrieved-context modeling framework, we demonstrate that the relative durations of the distraction intervals on a study list influence the organization of one’s memories, suggesting that the content of these intervals are important in determining the structure of one’s mental timeline. These results also bear upon an older debate in the memory literature, regarding the influence of a distraction task on the memorability of studied material. Briefly, Koppenaal and Glanzer (1990) and Thapar and Greene (1993) demonstrated that if a participant must shift from one distraction task to another during a study period, this task shift can decrease the memorability of items studied before the shift. These results raise the possibility that the negative consequences of distraction arise simply because a person must shift from an encoding task to a distraction task; in this scenario the relative duration of distraction intervals may be unimportant in determining item memorability and recall organization. Our modeling framework allows us to address this question directly, and we return to this issue in the discussion. Finally, these results bear upon temporal distinctiveness theories of human memory, under which the memorability of a particular item is affected by its temporal distance from the recall test, as well as the item’s temporal distance from other study events. In order to qualitatively contrast retrieved-context models with temporal distinctiveness Variable distraction 8 models, we examine the memorability of items as a function of the duration of the surrounding distraction intervals. In this experiment, temporally isolated items are less well remembered than temporally crowded items, a finding which may be at odds with the predictions of temporal distinctiveness models. We consider this finding in terms of both our retrieved-context model and an influential model of temporal distinctiveness, the Scale-Invariant Memory, Perception, and Learning model (SIMPLE; Brown, Neath, & Chater, 2007). Insert Figure 1 about here Materials and Methods