Source memory’s next top model

Did you watch an interview where Robert Downey Jr. says he’s coming back as Iron Man or did you just read an avid fan theory on Reddit? Was it your mother or anatomy professor who said bubble gum would stay in your stomach for seven years? Whether you’re sharing celebrity gossip or submitting a homework assignment, knowing where and how you got a piece of information can be just as important as the information itself.

Psychologists use the term source memory to refer to memory of the context in which something was learned. Source memory is essential for evaluating how valid a piece of information is, and it also aids in retrieving that information (according to the Tulving and Thompson encoding specificity principle). It’s easier to recall something during exposure to the context in which it was originally learned. For example, students perform better when taking an exam in the same classroom where they learned the material.

Researchers who study source memory are currently locked in a debate over how exactly the retrieval process works. Some believe source memory retrieval is a continuous process, which means that retrieved information varies in its quality and is not an all-or-none process. Others say it is a discrete process, which means only information that passes an all-or-none threshold is retrieved, and if that threshold isn’t passed, then the information is not retrieved, and that’s where guesses come in.

Jason Zhou, Adam Osth, Simon Lilburn, and Philip Smith (pictured below) put these different models of source memory retrieval to the test in a recent paper published in the Psychonomic Society journal, Psychonomic Bulletin & Review. The authors adapted previously-developed circular diffusion models to create three new models that represented continuous, discrete, and hybrid accounts of source memory. 

Zhou2021 fig1
Authors of the featured article

Researchers have used diffusion models to make sense of the cognitive processes underlying binary decision tasks (e.g., was a previously-seen stimulus on the left or right side of the screen?). A circular diffusion model is a special kind of diffusion model that can be applied to continuous response tasks (e.g., having to click on the screen exactly where you remember the stimulus being).

The circular diffusion model assumes that during the retrieval process there is a constant accumulation of evidence from memory before making a decision. The drift rate vector in the model measures the direction the accumulated evidence has moved the decision towards and how long it takes to reach that decision. In source memory models that include a guessing component, this drift rate is set to a value of zero, representing a lack of defined direction in the decision-making. The figure below illustrates an example of evidence accumulation leading to a decision and includes a visual representation of the drift rate vector.

Zhou2021 fig2
Components of a circular diffusion model that account for accuracy and reaction time in continuous-outcome tasks.

Continuous and discrete accounts of source memory make different predictions about decision outcomes and decision times, so the circular diffusion model is especially useful in this case because it includes accuracy and response time and can be applied to tasks with continuous responses.

We make continuous responses to source memory dilemmas all the time. These are scenarios where possible responses lie on a continuous scale, rather than being constrained by binary choices. For example, if someone asks where you last saw your keys on the table, it makes more sense to point to the exact spot you remember rather than saying whether you saw them on the left or right side. As these kinds of scenarios happen so often in everyday life, it is useful to understand the process underlying how people make continuous source memory decisions.

Zhou and colleagues’ primary aim was to use circular diffusion models to test whether source memory decisions, specifically during tasks with continuous outcomes, are a discrete or continuous process. To test this, they compared the predictions made by these models to behavioral data.

Participants completed a simple source memory task where they studied marked locations along a circle and words (i.e., “targets”) together. Next, they took a recognition memory test on the words and a source memory test where they were cued by targets and had to indicate on a blank circle the location that was originally paired with each target. These tasks are illustrated in the figure below.

Zhou2021 fig3
The behavioral task included an encoding phase, a recognition task to assess memory for the studied words, and a source memory task.

Below are the three models and their predictions on source memory errors (measured by distance from original source location) and reaction times.

    1. The variable-precision model assumes source memory retrieval is a continuous process and allows for variability in the drift rate (i.e., directionality of the decision) across trials. This model predicts that the distribution of memory errors will be highly concentrated around the mean, and it also predicts that errors will be accompanied by long response times.
    2. The threshold model consists of a memory-driven state that has a positive drift rate that does not vary across trials, and if this state fails, a guessing state is activated. Like the first model, this model predicts the error distribution will be concentrated at the mean, but it does not have any expectations of response times.
    3. The hybrid diffusion model allows for variability in drift rate across trials like the variable-precision model, and has a guessing process like the threshold model. As with the threshold model, this model predicts a concentrated error distribution and flexibility in response times.

The models were tested on how closely their predictions for response errors and reaction times matched participant data. The distribution of errors was concentrated around the mean, as predicted by all three models. However, the pattern of slow reaction times predicted by the continuous variable-precision model was not observed. This suggests that one of the models that included a guessing state – either the threshold model or the hybrid model – was a better fit.

Some critics of the discrete account say that, rather than being part of the source memory retrieval process, guessing is simply the result of not recognizing the cue items. However, in this experiment, the researchers found that a guessing process for source memory was engaged even for words that were successfully remembered in the recognition task. In these analyses, where source performance was conditioned on recognition confidence, it was determined the best fitting model was one where drift rate did not vary. And a constant drift rate is a key component of the pure threshold model.

Using a novel application of circular diffusion models, the authors present compelling evidence that source memory retrieval is a thresholded process. As both the threshold and hybrid models outperformed the continuous model, these findings indicate that the implementation of guessing is critical for source memory retrieval.

This work demonstrates how circular diffusion models are a useful tool for unpacking the components underlying retrieval processes. These results also set the stage for future work exploring how recognition and source memory could both be accounted for in a single model. Though the discrete vs. continuous debate is still ongoing, this research serves as a  contribution to understanding the source memory decisions we make every day.

Featured Psychonomic Society article

Zhou, J., Osth, A. F., Lilburn, S. D., & Smith, P. L. (2021). A circular diffusion model of continuous-outcome source memory retrieval: Contrasting continuous and threshold accounts. Psychonomic Bulletin & Review, 28(4), 1112-1130. https://doi.org/10.3758/s13423-020-01862-0

The Psychonomic Society (Society) is providing information in the Featured Content section of its website as a benefit and service in furtherance of the Society’s nonprofit and tax-exempt status. The Society does not exert editorial control over such materials, and any opinions expressed in the Featured Content articles are solely those of the individual authors and do not necessarily reflect the opinions or policies of the Society. The Society does not guarantee the accuracy of the content contained in the Featured Content portion of the website and specifically disclaims any and all liability for any claims or damages that result from reliance on such content by third parties.

You may also like