The more information an interruption shares with the primary task, the more reconstruction work resumption requires

What it is

Interruptions of the same length are not equally costly — if the interrupting task uses the same kind of information as the primary task (both require remembering numbers, both operate on the same table, both involve the same category judgment), resumption is harder than after an interruption whose content is entirely unrelated. This shows that resumption cost depends not only on interruption duration but also on how much information the interruption shares with the primary task: the more overlap, the more reconstruction work resumption requires.

This conclusion is easy to get backwards — intuition suggests "related things should be easier to pick back up," but what actually happens is the opposite: interference. The more similar the content, the more easily the primary task's information gets mixed up with the interrupting task's, requiring an extra step at resumption to pull the two apart again.

Why it happens

Working memory maintains task state using a set of representations that are distinguishable from one another — the primary task's numbers, categories, and locations each occupy their own separate slot. If the interrupting task uses the same type of information (also numbers, also category labels, also spatial positions), it competes with the primary task's representations along the same dimension, and the two sets of information start interfering with each other during the maintenance phase itself, rather than coexisting peacefully.

At resumption, this interference slows reconstruction in two ways. First, part of the primary task's representation may already be overwritten or blurred by the interrupting task's same-type information, requiring an extra verification step to determine which number, which status, belongs to which task. Second, even when representations aren't directly overwritten, two highly similar sets of information are more prone to mix up during retrieval (retrieval interference), producing more "mistook it for the other task" errors during resumption.

An interruption with entirely unrelated content is much less prone to this kind of interference — different types of information occupy different dimensions and don't compete for the same resource, so resumption only needs to reactivate the primary task's state without any extra work to disentangle the two.

Studying it

A common approach manipulates the content overlap between the interrupting task and the primary task: one condition has high overlap (e.g. both require remembering a sequence of numbers), another has entirely different content (e.g. the primary task involves remembering numbers while the interruption involves shape matching), while interruption duration and placement are held constant. Post-resumption reaction-time cost and error type are then compared.

Common independent variables: the type and amount of information shared between the interrupting task and the primary task, and the shared dimension (numbers, categories, spatial location, etc.). Common dependent variables: post-resumption reaction-time cost, error rate, and the nature of errors (pure forgetting versus mixing up information from the two tasks).

In HCI research, this manipulation is commonly used to evaluate multitasking designs within a single interface: interleaving different sub-tasks within the same table or the same type of operation, compared with switching between entirely different interface modules — which kind of switch produces higher resumption cost has direct implications for information architecture and task partitioning.

A methodological caution: "content overlap" itself can be defined along multiple dimensions (surface-form similarity vs. underlying structural similarity), and different studies don't always operationalize overlap the same way — when comparing effect sizes across studies, check which dimension overlap was defined on first.

Where it stops holding

• The effect of overlap is not obvious when the interruption is very brief. If the interruption itself lasts only a few seconds, the two sets of information don't have time to compete substantially during the maintenance phase, and the extra cost from overlap is masked by the interruption's baseline cost.
• Interference from overlap can be partly offset by external recording. If the primary task's state is explicitly displayed on the interface (rather than relying entirely on working memory), representational confusion caused by overlapping content is noticeably reduced, since verification can rely on the external record instead of internal memory to tell things apart.
• This effect targets overlap at the level of content, not overlap at the level of task nature. Two tasks that are similar in kind (both "review" work) but handle entirely different types of information do not necessarily produce the interference described here.
• This describes a moderating factor for resumption cost, not what resumption cost itself is made of — that is a separate question.

Applying it

• When scheduling interrupting tasks (confirmation dialogs, briefly inserted secondary actions), avoid using the same type of information as the primary task where possible. If the primary task is processing a batch of order numbers, inserting a confirmation dialog that also foregrounds numbers as its main content noticeably worsens resumption cost; switching to a graphical or textual confirmation reduces the interference.
• For sub-tasks within the same interface that require frequent switching and share an information dimension (the same batch of numbers, the same set of categories), make a clear visual distinction between them (different color schemes, different layout structures) to reduce mutual interference during the maintenance phase.
• For interruptions with high content overlap, prefer external recording over reliance on user memory: keep the primary task's key numbers and state continuously visible on the interface, rather than requiring the user to resume from memory after handling the interruption.
• How to check: design two interruption scenarios with different content overlap but the same duration, and compare the distribution of error types after resumption. If the high-overlap condition produces noticeably more "mixed-up information" errors (rather than simply slower responses), overlap is indeed at work, and the related design should be adjusted to reduce information overlap.

Related

• Same group: A5.08.1 Interruption timing should be chosen at subtask boundaries · A5.08.2 The system should preserve the state and position from before the interruption · A5.08.3 A measurable resumption lag exists before performance returns to pre-interruption levels · A5.08.4 Resumption cost has two components — re-executing the primary task and remembering the interrupting task — that must be accounted for separately · A5.08.5 Self-initiated interruptions have lower resumption cost than externally imposed ones
• Nearby: A6.02 Working memory capacity
• Search terms: interruption similarity · retrieval interference · task resumption · working memory interference