Event segmentation ability uniquely predicts event memory

Highlights

• Ability to segment experience into discrete events predicts memory.

• General abilities (e.g., working memory) do not account for this relationship.

• This relationship is observed across the lifespan.

Abstract

Memory for everyday events plays a central role in tasks of daily living, autobiographical memory, and planning. Event memory depends in part on segmenting ongoing activity into meaningful units. This study examined the relationship between event segmentation and memory in a lifespan sample to answer the following question: Is the ability to segment activity into meaningful events a unique predictor of subsequent memory, or is the relationship between event perception and memory accounted for by general cognitive abilities? Two hundred and eight adults ranging from 20 to 79 years old segmented movies of everyday events and attempted to remember the events afterwards. They also completed psychometric ability tests and tests measuring script knowledge for everyday events. Event segmentation and script knowledge both explained unique variance in event memory above and beyond the psychometric measures, and did so as strongly in older as in younger adults. These results suggest that event segmentation is a basic cognitive mechanism, important for memory across the lifespan.

Introduction

Memory for everyday events (event memory) is critical for normal functioning and supports, for example, one’s capacity to understand instructional videos, to give eyewitness testimony and to answer the ubiquitous question: What happened? To perceive the continuous activity of everyday life as discrete events, one must segment ongoing experiences into meaningful temporal units. Consistent with work showing benefits of chunking for human memory (e.g., DeGroot, 1978, Gobet et al., 2001), research into event memory has shown a relationship between how events are segmented and how they are remembered (e.g., Boltz, 1992, Ezzyat and Davachi, 2011, Newtson and Engquist, 1976, Schwan et al., 2000). The current study investigates the possibility that the relationship between event segmentation and event memory is a causal one.

Newtson, 1973, Newtson, 1976 developed a paradigm to assess how an observer segments an everyday activity into meaningful events. Participants watch a video of someone performing the activity—for example, checking out groceries at the store—and are asked to press a button whenever they believe one unit of activity (or event) has ended and another has begun. In this example, a participant might press the button after each successive item is scanned and bagged. Studies using this paradigm have established a connection between event segmentation and event memory. For example, superior recognition and recall memory has been observed for activity occurring near event boundaries (Newtson and Engquist, 1976, Schwan et al., 2000). Editing movies by deleting intervals containing event boundaries impairs memory for the movies more than deleting intervals between event boundaries (Schwan & Garsoffky, 2004). Similarly, inserting commercial breaks or pauses into films at event boundaries can improve memory, and inserting such breaks between event boundaries can impair memory (Boltz, 1992, Schwan et al., 2000).

Recent work has shown that long-term associations are stronger within than between events. In particular, cued recall of target information from a narrative is better if the cue and target come from the same event, compared to when the cue and target come from different events (Ezzyat & Davachi, 2011). Consistent with this finding, memory for details viewed 5 s before testing was reduced if an event boundary occurred during the interval between the appearance of the detail and the memory probe (Swallow et al., 2009, Swallow et al., 2011). In addition, recognition of probes from previous compared to current events was associated with greater activation in brain regions that handle longer-term memory, including the hippocampus and parahippocampal gyrus. These studies provide strong evidence that how experience is segmented into events is important for how that experience is remembered.

Why might event segmentation be predictive of memory, specifically for everyday events? Much of what researchers have learned about episodic memory comes from studies using lists or series of discrete stimuli such as syllables, words or pictures (e.g., Bjork and Whitten, 1979, Buschke, 1973, Deese, 1959). By constructing these memory test materials, experimenters create a series of minor events that are intended to be the “episodes” of later episodic memory. The problem of how the activity is segmented into episodes in these experiments can be safely ignored because the highly structured situation constrains participant’s segmentation almost perfectly: The segments can be assumed to be the words, pictures, lists, etc. On the other hand, for an everyday event such as checking out from the grocery store, the problem of segmentation is immediately evident. Putting milk in a bag, for example, may be perceived as one small event. The beginning and end of this event may be defined by, among other things, the motion characteristics of the clerk’s arm, the interaction with the milk carton, the clerk’s perceived goals, or some weighted combination of these factors. Some people may spontaneously chunk activity into units that are effective for memory encoding and later retrieval; others may fail to identify effective units during perception, and their subsequent memory may suffer as a result. Thus, event segmentation, like other cognitive mechanisms such as spatial attention and memory retrieval, is a process that may vary in its effectiveness across individuals and thus can be studied as an ability. Our concern in this research is this individual difference, that is, segmentation ability.

Better segmentation ability is associated with better subsequent memory (Bailey et al., 2013, Kurby and Zacks, 2011, Zacks et al., 2006). In these studies, participants watched movies of actors engaged in everyday activities (e.g., washing a car) and segmented them by pressing a button whenever they believed one unit of activity ended and another began. Segmentation ability was defined as the degree to which an individual agreed with the sample as a whole about where event boundaries occurred in the movies. In all three studies, individuals showing greater segmentation ability remembered the movies better. This raises an important question: Are segmentation ability and memory correlated because both are supported by a general cognitive capacity, or does segmentation ability uniquely predict memory? This is the primary question addressed in the current study.

To answer additional questions regarding healthy aging, this study examined the relationship between event perception and memory across the adult lifespan. Age related deficits in episodic memory are well documented (for review see Zacks, Hasher, & Li, 2000). Because segmentation appears to be a mechanism that contributes to memory performance, understanding whether and how this contribution changes across the lifespan might be useful for efforts to address age related memory deficits. Previous studies of event segmentation and episodic memory showed poorer event memory in older compared to younger adults (Kurby and Zacks, 2011, Zacks et al., 2006). These studies also showed reduced event segmentation ability in older compared to younger adults. These results lead us to the following questions. If event segmentation ability and event memory both decline with age, does segmentation ability mediate the age – event memory relationship? Furthermore, if there is a unique relationship between segmentation ability and episodic memory, does this relationship persist in healthy aging?

To address the questions posed here, we used an individual-differences approach to test for relationships among event segmentation ability, event memory, and general cognitive abilities in a lifespan sample of cognitively normal adults. The specific measures used to assess segmentation ability and event knowledge are discussed in Section 1.1, and the measures used to assess general cognitive abilities are discussed in Section 1.2.

We refer to variables that measure abilities specific to the perception and understanding of events as event understanding variables, to distinguish them from measures of general cognitive abilities such as working memory and processing speed, which are discussed below in Section 1.2. The selection of event understanding variables used in the present study was motivated by Event Segmentation Theory (EST; Zacks, Speer, Swallow, Braver, & Reynolds, 2007). Briefly, EST proposes that everyday experience is interpreted in the context of event models: mental representations maintained in working memory that describe what is happening right now. Event models contribute to perception by facilitating predictions regarding what is likely to happen in the immediate future. When relevant dimensions of the ongoing event change, the event model becomes outdated, leading to prediction errors. The system uses those prediction errors as a signal that the model needs to be updated. For example, when watching a clerk bag groceries, one forms a mental model that allows predictions, e.g., the clerk has placed an item in the bag and will now reach for the next item. However, when the last item has been bagged and the clerk is ready to take payment, the old model will generate inaccurate predictions and a new model needs to be established. EST posits that when an event model is updated people perceive an event boundary. When an event model is updated, its contents are determined by the current perceptual input, the current state of working memory, and long-term knowledge and memory for previous events.

Segmentation ability has previously been assessed using several measures. The primary measure in this study was segmentation agreement, a measure of the degree to which an individual identifies event boundaries that also are identified by the group (Kurby and Zacks, 2011, Zacks et al., 2006). Event segmentation is inherently subjective, so it is not possible to objectively assess segmentation accuracy. However, previous studies have shown good agreement across observers in where event boundaries occur (Newtson, 1976) and even better agreement within individuals across time (Speer, Swallow, & Zacks, 2003). Given that individuals tend to agree with one another regarding the locations of event boundaries, it seems that normative segmentation is adaptive and reflects segmentation ability. The fact that segmentation agreement predicts subsequent memory supports this proposal.

We also considered two other measures of segmentation ability that have been used previously, Alignment is the degree to which high-level events (e.g., washing your hands) consist of groups of smaller events (e.g., turning on the water, putting soap on your hands, lathering the soap, etc.; Kurby and Zacks, 2011, Zacks et al., 2001). Enclosure reflects the degree to which groups of fine units are “enclosed” by coarse units (Hard, Recchia, & Tversky, 2011). As will be seen, segmentation agreement proved to be a substantially more reliable psychometric measure in this sample, and agreement was therefore the measure of segmentation ability used in the primary analyses. Henceforth, the term segmentation ability refers to segmentation agreement. Details on the computation of each measure are provided in Section 2.2.

Event knowledge measures assess the integrity and depth of an individual’s knowledge regarding what generally happens in certain situations. Drawing on work in narrative comprehension (e.g., Rumelhart, 1975, Schank and Abelson, 1977, van Dijk and Kintsch, 1983), theories of event cognition have proposed that specific event models are informed in part by structured, long term representations of generalized classes of events, known as scripts or schemata (Rosen et al., 2003, Zwaan and Radvansky, 1998). We refer to scripts and event schemata here as event knowledge. For example, during a specific visit to the grocery store one’s event model may include unperceived features that are filled in by an event schema comprising knowledge about what generally occurs at the grocery store. Event knowledge may be considered a specific type of general knowledge. Whereas tests of general knowledge assess vocabulary and memory for specific, isolated facts (e.g., who was Cleopatra?), event knowledge benefits from understanding relationships between the features of generalized events. For example, to describe what generally happens at the grocery store, it helps to know that, as you enter, produce is generally to the right and dairy to the left, and that you must select items before paying, and pay before leaving. Because event schemata encode hierarchical relationships between units of activity within stereotypical events, and because they inform event models, they might also be important for event segmentation. Event knowledge could affect event memory directly, or indirectly, through its effect on segmentation.

According to EST, the perception of event boundaries involves multiple cognitive and neural mechanisms interacting in a specific way. To summarize, perceptual processing leads to predictions about the near future and is biased by event models maintained in working memory. Event models in turn are updated when predictions are erroneous. During updating, event models are influenced by long-term episodic memory, general semantic knowledge, and event-specific semantic knowledge. (For specific proposals regarding the neurophysiological aspects of these mechanisms, see Zacks et al., 2007.)

Age related declines are well established in several of the abilities thought to contribute to event segmentation (e.g., working memory). Therefore, we might expect poorer event segmentation associated with older age. However, the relationship between age and event perception may not be so simple. There may be qualitative differences in how younger adults perceive and understand events they have seen hundreds of times compared to how older adults perceive those same events after thousands of viewings. For example, EST posits that event segmentation is guided, in part, by general knowledge, scripts and schemas, which change as we accumulate life experience. Research suggests that older adults use this type of knowledge in comprehending written narratives to compensate for declines in other areas (Arbuckle et al., 1990, Radvansky and Dijkstra, 2007). We expect that differences in general processing factors such as working memory play a large role in any age differences in event segmentation. However, it is also possible that in segmenting events older adults rely more on scripts and schemas than younger adults. Thus, regardless of age differences in segmentation ability, the relative contributions of component mechanisms, and in essence, the style of event segmentation might change across the lifespan. We ask whether the relationship between segmentation and memory is consistent across the lifespan to test the possibility that age related differences in segmentation style differentially support episodic memory.

There are a number of cognitive abilities that likely contribute to how one understands and remembers events. We administered a battery of cognitive tests chosen to assess theoretically plausible mediators of the relationship between segmentation ability and event memory. The goal was to test the hypothesis that segmentation ability predicts event memory independently of any effects of general cognitive abilities. Below we describe the general cognitive factors included in the current study and why they were chosen.

Working memory (WM) supports the capacity to maintain information in an activated state and manipulate it. The perception and segmentation of even the simplest events involves the ability to integrate information from various sources (e.g., visual and auditory perception, long term memory) and across dimensions (e.g., space, time, characters, goals). Several theories propose that working memory supports multidimensional representations of immediate events, which we refer to as event models (e.g. Baddeley, 2000, Zacks et al., 2007). As described above, event models provide a context that guides the processing of ongoing experience, and are hypothesized to play an important role in the segmentation of experience into events.

Executive function (EF) is the ability to adaptively control behavior in response to goals and task demands. Psychometric measures of WM and EF are highly correlated (e.g., McCabe, Roediger, McDaniel, Balota, & Hambrick, 2010), both explain significant amounts of variance in episodic memory performance (e.g., Bugaiska et al., 2007, Rosen and Engle, 1997), and both generally decline with age (e.g., Moscovitch and Winocur, 1992, Salthouse, 1990). Furthermore, WM and EF have both been shown to mediate the relationship between age and episodic memory (e.g., Bugaiska et al., 2007, Salthouse et al., 2003, Troyer et al., 1994).

Perceptual processing speed is correlated with performance across a range of cognitive tasks (e.g., Faust et al., 1999, McCabe et al., 2010). Age related declines in processing speed are well established (e.g., Park et al., 1996). However, even controlling for age, processing speed has been shown to correlate with high-level cognitive abilities (Coyle et al., 2011, Fry and Hale, 1996).

There is likely to be considerable overlap in the cognitive mechanisms underlying memory for everyday events and those underlying typical laboratory measures of episodic memory (e.g., for lists of words). However, everyday event memory and typical laboratory episodic memory tasks differ in important ways. For example, compared to word lists, everyday events are more likely to be encoded in the context of pre-existing knowledge structures, reflecting a lifetime of experience with similar events. Also, lists or series of discrete stimuli used in laboratory episodic memory tasks present, at least superficially, more explicit cues to segmentation than do everyday events. As a result of these differences, segmentation ability may relate differently to memory for word lists than to event memory. Older adults have been found to perform worse than younger adults on both laboratory and event memory tasks. However, event memory tasks may offer a richer encoding context and therefore ameliorate some age differences (Koutstaal et al., 1998, Zacks et al., 2000, Zacks et al., 2006).

Finally, general knowledge about objects and facts could contribute to constructing effective event models. General knowledge, an expression of crystallized intelligence (Gc), plays a substantial role in many complex cognitive tasks (Carroll, 1993, Friedman et al., 2006). Significantly, general knowledge usually shows gains rather than losses with age (e.g., Park et al., 1996). Therefore, we might expect general knowledge, as well as script and schema knowledge, to mediate age-related differences in segmentation and episodic memory.

The constructs outlined above reflect abilities that are considered general because they predict performance on a range of tasks that humans perform in the laboratory, including tests of fluid intelligence (e.g., Engle et al., 1999, Fry and Hale, 1996, Kane and Engle, 2000, Unsworth and Spillers, 2010). These general abilities likely underlie much of the processing involved in higher-level cognition. Therefore, we might expect individual differences in event segmentation to be explained by these general cognitive factors. On the other hand, if event segmentation is supported by a particular interaction among these general systems then event perception, as a distinct cognitive activity, may not be measured well by individual psychometric tests of basic cognitive abilities. It is also possible that event segmentation ability reflects the operation of neural and cognitive mechanisms that are not captured by established cognitive ability tests. We therefore hypothesized that selective measures of event segmentation would uniquely predict event memory, above and beyond any contribution of basic cognitive abilities.