The associative structure of memory for multi-element events.

Aidan J. Horner, Neil BurgessJournal of Experimental Psychology General 2013 · DOI

TL;DR

This paper investigates how our brains bind together different pieces of information (like a person, a place, and an object) to form a single memory of an event. They found that retrieving one part of an event makes it easier to retrieve other parts from the same event, suggesting a shared underlying memory structure. This "associative structure" is influenced by how we're asked to recall the memory and how confident we are.

What Problem Does This Paper Address?

We experience the world as a rich tapestry of sights, sounds, and sensations, yet we can recall discrete events as unified wholes. A leading theory suggests the hippocampus acts as a "convergence zone," binding these multimodal elements into a single "engram." However, direct behavioral evidence for this binding – specifically, how the retrieval of one event element influences the retrieval of others from the same event – has been limited. This paper aims to fill this gap by testing the prediction that retrieving one element of a multi-element event should increase the probability of retrieving other elements from that same event, demonstrating a dependency in recall.

Methods

The researchers designed a series of experiments using a novel approach to measure "episodic dependency." Participants were asked to vividly imagine multi-element events (person-location-object triplets). Memory for these events was then tested using cued retrieval over multiple trials. They developed a statistical framework, comparing an "independent model" (where elements are recalled independently) against a "dependent model" (which accounts for shared factors influencing recall of elements within the same event). This allowed them to quantify the degree of dependency between element retrievals. They also explored how this dependency varied based on the retrieval task (cued recall vs. cued recognition) and subjective confidence.

Datasets

The study used custom-generated datasets of 24 (Experiment 1) and 36 (Experiments 2 & 3) unique location-person-object triplets. These stimuli were presented as words. No pre-existing benchmark datasets were used; the stimuli were created specifically for this research.

Key Insights and Ideas

  • Events are more than the sum of their parts: The core finding is that retrieving one element of an event significantly increases the likelihood of retrieving other elements from the same event. This directly supports the idea that events are stored as interconnected structures, not just isolated pieces of information.
  • Retrieval strategy matters: The way we're asked to recall information has a big impact. Cued recall (where you have to actively produce the answer) showed stronger dependency than cued recognition (where you choose from options). This suggests that the effort involved in recall might tap into the shared event representation more effectively.
  • Confidence is key: Higher subjective confidence in recalling an event was associated with stronger dependency between its elements. This implies that when we feel more certain about an event, its constituent parts are more tightly bound in our memory.
  • A common factor links event elements: The researchers propose a "common factor" that influences the retrieval of all elements within a single event. This factor could be related to how strongly the event was initially encoded (e.g., due to attention) or a process of "pattern completion" during retrieval, where recalling one part triggers the recall of others.

Key Results

  • Significant dependency found: Across three experiments, the researchers consistently rejected the "independent model," demonstrating significant episodic dependency. For example, in Experiment 1, the dependency measure (D) for the observed data was significantly higher than predicted by the independent model.
  • Cued recall > Cued recognition: Dependency was significantly higher in cued recall (Experiment 1) compared to cued recognition (Experiment 2). For instance, the difference between observed data and the independent model was larger in Experiment 1 (0.20) than in Experiment 2 (0.05).
  • Confidence modulates dependency: In Experiment 3, high-confidence events showed significantly greater dependency than low-confidence events, even when controlling for overall accuracy.
  • Guessing plays a role: The difference in dependency between recall and recognition tasks was partly explained by increased guessing in recognition tasks. A "dependent model with guessing" provided a better fit to the data in these conditions.

Limitations and Future Work

  • Distinguishing between models: The study acknowledges that their data cannot definitively distinguish between two proposed mechanisms for dependency: a common encoding factor (like attention) or a pattern completion process at retrieval. Future experiments are suggested to tease these apart, perhaps by manipulating encoding strategies or measuring reaction times.
  • Generalizability of the "common factor": While the dependent model with guessing provided a good fit, it still overpredicted dependency in some low-confidence recognition scenarios. This suggests that the proposed "common factor" might not fully capture all nuances of memory retrieval, especially in less confident recall situations.
  • Investigating other event structures: The current study focused on three-element events. Future work could explore dependency in events with more elements or different types of elements.
  • Neuroscientific validation: While the findings align with theories of hippocampal function, direct neuroimaging or lesion studies would be valuable to confirm the neural underpinnings of this observed episodic dependency.

Summary updated

Abstract

The hippocampus is thought to be an associative memory "convergence zone," binding together the multimodal elements of an experienced event into a single engram. This predicts a degree of dependency between the retrieval of the different elements comprising an event. We present data from a series of studies designed to address this prediction. Participants vividly imagined a series of person-location-object events, and memory for these events was assessed across multiple trials of cued retrieval. Consistent with the prediction, a significant level of dependency was found between the retrieval of different elements from the same event. Furthermore, the level of dependency was sensitive both to retrieval task, with higher dependency during cued recall than cued recognition, and to subjective confidence. We propose a simple model, in which events are stored as multiple pairwise associations between individual event elements, and dependency is captured by a common factor that varies across events. This factor may relate to between-events modulation of the strength of encoding, or to a process of within-event "pattern completion" at retrieval. The model predicts the quantitative pattern of dependency in the data when changes in the level of guessing with retrieval task and confidence are taken into account. Thus, we find direct behavioral support for the idea that memory for complex multimodal events depends on the pairwise associations of their constituent elements and that retrieval of the various elements corresponding to the same event reflects a common factor that varies from event to event.
BibTeX
@article{horner2013associative,
  author = {Aidan J. Horner and Neil Burgess},
  title = {{The associative structure of memory for multi-element events.}},
  journal = {Journal of Experimental Psychology General},
  year = {2013},
  doi = {10.1037/a0033626},
}
Figure