An old but common adage is that “context is everything”, and while this aphorism can be applied to many different topics, it has been especially significant in theories of memory. Even in early psychological research, theorists believed that having knowledge and memory for an item meant connecting it to the “context which the world provides”, as William James famously penned. However, an open question remains: Is temporal context a salient cue during recall? Or is context something else, such as conceptual relationships that connect separate ideas?

Researchers Adam Osth and Julian Fox explored this question in a recent article in the Psychonomic Bulletin & Review (PB&R), continuing a debate that has taken place within the Psychonomic Society’s journals. One view holds that associations in memory can be constructed across temporal contexts – a position supported by a 2008 paper in PB&R, whose results suggest evidence of long-range temporal associations formed between items within a list. Another view, which Douglas Hintzman discussed in a 2016 paper in Memory & Cognition, is that learners only use temporal context as a retrieval cue if it is needed. When presented with a task that affords better mnemonic cues, such as one that allows retrieval of a target item to be guided by the presentation of a cue (e.g., cued recall), learners will not rely on temporal context.

So how should we examine the roles that temporal and associative contexts are thought to play? Osth and Fox do so by first contrasting different theoretical models of episodic memory.

Models of Memory and Context

A prominent model of memory is the temporal context model (TCM). This framework holds that the sequence in which items are learned is an important context, and that items studied over the course of learning develop associations with this temporal information.

Temporal information becomes an important retrieval cue because remembering one piece of information can lead to remembering another piece of information that was studied soon before or after it, which itself can lead to remembering another item – all by virtue of being studied together at relatively the same time.

Under the right conditions, the TCM predicts that retrieval can cascade as a function of the sequence in which the items were studied. This sequencing is what researchers often see in free-recall responses.

This is not without its cons, though: Temporal context can cause the co-activation of other items in memory, and if a learner is not careful or is unable to distinguish between the “correct” memory item and other competitors, she may make a memory error. You could think of this in terms of remembered paired items, such as shown in the figure below:

When attempting to remember the target (“B”) that goes with the cue (“A”), you might also think about the pairs that were studied immediately after A-B, such as C-D or, to a lesser extent, E-F. In that case, you would be more likely to false alarm, or report the wrong answer at test, to items that were studied close in time to the correct pair. Then, you might mistakenly think that “D”, or perhaps even “F”, is associated with “A”.

But what if the temporal relationship between studied items is not the most dominant retrieval context? In opposition to temporal context models of episodic memory are ones that highlight bonds that are formed only between items that are shown together.

The search of associative memory model (SAM), for instance, hypothesizes that relationships are only developed within a pair of items that are studied together and not across an entire set. In that case, it would not matter when other pairs of items were learned compared to A-B; all else being equal, a learner could just as easily false alarm to an item at the end of the study list (“L”) as she could to one studied immediately after A-B if she was unable to remember the correct target.

Putting the Theory to the Test

Osth and Fox constructed a memory task in order to better examine if incorrect responses during retrieval are related to the temporal context of the correct cue-target pair. First, participants were asked to study pairs of associated words. After a brief distractor task, participants were then shown pairs of words and asked to decide whether they were shown during study (i.e., yes/no associative recognition), where half of the pairs were intact cue-target word pairs (e.g., A-B) and the others were rearranged pairs (e.g., A-F). Here, lag was the critical manipulation, defined as the separation between study trials for rearranged test items.

To give an example using the pairings from the above figure: If I first studied “A-B”, then “C-D” second, but had to determine if “A-D” was a pairing that I learned earlier, then that would be considered a lag-1 trial because the cue (“A”) and incorrect target (“D”) were in consecutive study trials. An “A-F” test trial, then, would be lag-2 because “F” was shown 2 trials after “A”.

In this task, there are two potential results: If temporal context is a strong retrieval cue during associative learning, then the proportion of false alarms at test will be related to lag, such that individuals will be more likely to incorrectly say that a cue was paired with a target from a different pair of items that was studied closely before or after it. If the association between two items is the dominant cue, then the rate of false alarms will not change across lags.

Osth and Fox’s results support an association-context account and show that lag between pairs at study has no effect on later memory performance, particularly for false alarm rates (shown in the figure below).

Osth and Fox provide additional support for this view by re-examining the results of three experiments performed outside of their lab and showing that false alarm rates do not change as a function of lag at study. Bayesian analyses of the experimental effects suggest that there is extremely strong evidence against a temporal context effect. Thus, in these associative recognition tasks, temporal context does not appear to be a strong cue.

Continuing the Conversation

If you had high hopes for the temporal context hypothesis in this study, then don’t worry: It is still highly prevalent. Even outside of free recall tasks, where TCM is typically used for interpretation, evidence for cross-pair associations are abundant. For example, Davis and colleagues have demonstrated that errors in cued-recall tasks are related to the original pairs’ proximity to the wrong answer in a study sequence, even though serial position at study should be unrelated to an attempt to recall a target item.

So, despite the lack of evidence for cross-pair associations in this paper, we will continue to look forward to understanding how temporal context is important for memory.

Psychonomics article featured in this post:

Osth, A. F., & Fox, J. (2019). Are associations formed across pairs? A test of learning by temporal contiguity in associative recognition. Psychonomic Bulletin & Review. DOI: 10.3758/s13423-019-01616-7.