🧠 neuroscience

Characterising semantic prioritisation in visual working memory

By combining hierarchical drift-diffusion modelling with behavioural manipulations, this study demonstrates that semantic representations in visual working memory are more readily accessible than perceptual details, particularly when representations fall outside the focus of attention or face interference, suggesting a shift toward abstract, long-term memory-like formats under limited attentional support.

Original authors: Kerren, C., Gonzalez-Garcia, C., Linde-Domingo, J.

Published 2026-02-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Kerren, C., Gonzalez-Garcia, C., Linde-Domingo, J.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The Big Question: How Does Our Brain Remember Things?

Imagine your Working Memory is like a small, cluttered desk in a busy office. You can only keep a few items on the desk at once (the "focus of attention"). If you need to remember more than a few things, you have to shove the extra items into a drawer or a filing cabinet nearby.

The big mystery scientists have been trying to solve is: When you reach into that drawer to grab a memory, what do you find?

Do you find a high-definition, crystal-clear photo of the object (Perceptual detail)?
Or do you find a rough, abstract sketch with just the main ideas (Semantic meaning)?

Previous research suggested that when our attention is busy, our brain seems to prioritize the "rough sketch" (meaning) over the "high-definition photo" (details). But why does this happen? Is it because the photo gets blurry over time? Or is it because it's just harder to find the photo in the drawer than the sketch?

This paper uses a clever mathematical tool called Drift-Diffusion Modeling (think of it as a "decision stopwatch") to figure out exactly when and how our brain makes these choices.

The Three Experiments: Testing the "Filing Cabinet"

The researchers ran three sets of experiments to test this. Here is what they did, using simple metaphors:

1. The Re-Analysis: The "Busy Desk" Check

First, they looked at old data where people had to remember 1 to 4 items.

The Finding: No matter how many items were on the desk, people were always faster at answering questions about the meaning of an item (e.g., "Is this an animal?") than its appearance (e.g., "Is this a photo or a drawing?").
The "Stopwatch" Insight: The math showed that the brain wasn't just "thinking faster" overall. The advantage happened before the actual thinking started. It was like the "meaning" file was sitting right at the front of the drawer, while the "photo" file was buried at the back.
The Twist: When the brain was under heavy pressure (remembering 3 or 4 items), it also started processing the "meaning" faster during the decision phase. But the main speed boost was always in the "retrieval" phase.

2. Experiment 1: The "Magic Spotlight" (Retro-Cues)

Imagine you are holding three items in your mind. Suddenly, a magic spotlight shines on one of them, telling you, "Hey, you only need to remember this one!"

The Setup: Sometimes the spotlight told them exactly which item would be tested (Valid Cue). Sometimes it just shone on all three, leaving them to guess (Neutral Cue).
The Result:
- When the spotlight shone on just one item, the "meaning" advantage almost disappeared. Why? Because the brain could focus all its energy on that one item, making the "photo" just as easy to grab as the "sketch."
- When the spotlight was vague (Neutral), the "meaning" advantage came roaring back.
The Lesson: The brain doesn't just "prefer" meaning. It prefers meaning when it's overwhelmed. If you can focus your attention perfectly, you can access the details. If you are distracted, your brain defaults to the abstract "gist."

3. Experiment 2: The "Interruption" Test

Here, they wanted to know: Does time make memories blurry, or does distraction make them blurry?

Condition A (Immediate): Test right away.
Condition B (Delayed): Wait a bit, but do nothing else (just stare at a screen).
Condition C (Prospective/Interference): Wait a bit, but do a different hard task in the middle (like matching shapes).
The Result:
- Just waiting (Time) didn't change much.
- Doing a different task (Interference) made the "meaning" advantage huge.
The Lesson: It's not the passing of time that ruins our memory of details; it's interference. When our attention is hijacked by something else, the brain quickly drops the "high-def photos" and switches to "abstract summaries" to survive the chaos.

The "Aha!" Moment: What Does This Mean?

The researchers discovered that Semantic Prioritization (preferring meaning over details) isn't a bug; it's a feature.

Think of your brain like a smartphone battery saver mode:

Normal Mode (Focused Attention): You have plenty of power. You can load high-resolution images (perceptual details) and process them instantly.
Low Power Mode (Distraction/Interference): When your attention is split or you are interrupted, the brain switches to "Battery Saver." It stops trying to load the heavy, high-res images because they take too much energy and time to retrieve. Instead, it loads the lightweight, text-based summaries (semantic meaning).

Why is this cool?

It explains why we remember the story of a conversation but forget the exact font of the text message.
It shows that our memory isn't a static video recorder. It's a dynamic system that changes its format based on how much attention we can give it.
It proves that when we are distracted, our brain doesn't just "forget"; it adapts by becoming more abstract and efficient.

The Bottom Line

When your attention is focused, you can access the fine details of your memories. But the moment you get distracted or have to juggle too many things, your brain automatically switches to "Abstract Mode." It drops the high-definition photos to save energy and gives you the "gist" instead. This isn't a failure of memory; it's a brilliant survival strategy to keep you functioning when things get chaotic.

1. Problem Statement

Working memory (WM) allows for the temporary maintenance of information beyond sensory input, yet the format of these representations and the mechanisms governing their access remain debated. While some theories suggest WM relies on sustained sensory traces (sensory recruitment), others propose that representations are transformed into abstract, semantic formats, particularly when outside the focus of attention.

Previous research indicated a "semantic prioritisation" effect in WM: participants are faster and more accurate at retrieving semantic features (e.g., animacy) than perceptual features (e.g., image style) when maintaining multiple items. However, the underlying cognitive mechanism driving this bias was unclear. It was uncertain whether this advantage stemmed from:

Retrieval dynamics: Faster reactivation of semantic representations from an unattended state.
Decision-level biases: A preference for semantic evidence during the decision process after retrieval.
Attentional factors: Whether the effect is driven by the mere passage of time (temporal decay) or by attentional disruption/interference.

The study aimed to dissociate pre-decisional access processes from evidence accumulation processes to determine the precise locus of semantic prioritisation.

2. Methodology

The authors employed a multi-experiment approach combining behavioral paradigms with Hierarchical Drift-Diffusion Modelling (HDDM). HDDM allows for the decomposition of reaction time and accuracy data into latent decision parameters:

Non-decision time ( $t$ ): Processes occurring prior to evidence accumulation (e.g., stimulus encoding, retrieval, motor preparation).
Drift rate ( $v$ ): The efficiency and speed of evidence accumulation toward a decision.
Threshold ( $a$ ): Response caution.

Experimental Design:

Reanalysis of Kerrén et al. (2022): A dataset involving 1–4 items with varying memory loads and temporal lags was re-analyzed using HDDM to establish baseline patterns of semantic vs. perceptual access.
Experiment 1 (Retro-cue Manipulation):
- Participants: 83 participants.
- Task: Maintain 3 items. A retro-cue indicated either the specific item to be tested (Valid) or all three items (Neutral).
- Goal: To dissociate retrieval demands. Valid cues allow pre-activation of the target, reducing retrieval load; Neutral cues require maintaining all items until the probe.
Experiment 2 (Interference vs. Delay):
- Participants: 77 participants.
- Task: Maintain 3 items. Three conditions were tested:
  1. Immediate: Test follows a short blank delay.
  2. Delayed: Test follows a long blank delay (isolating time effects).
  3. Prospective: Test follows a long delay plus an intervening cognitive task (isolating interference/attentional disruption effects).
- Goal: To determine if semantic prioritisation is driven by temporal decay or the need to shield memory from interference.

Analysis:

Bayesian statistical tests (t-tests, ANOVAs) were used for behavioral metrics (Accuracy, Reaction Time, Balanced Integration Score).
HDDM was used to model the latent decision processes, testing how feature type (semantic vs. perceptual) and experimental conditions modulated $t$ and $v$ .

3. Key Results

Reanalysis of Previous Data

Non-decision time ( $t$ ): Semantic judgements consistently showed significantly shorter non-decision times than perceptual judgements across all memory loads and lags. This indicates a robust advantage in pre-decisional access (retrieval/encoding) that is independent of load.
Drift rate ( $v$ ): Semantic advantages in evidence accumulation were conditional. They emerged primarily under high cognitive demands (Load 3 and Load 4) and longer temporal lags. Under focused attention (Load 1), perceptual features actually accumulated evidence faster.

Experiment 1 (Retro-cue)

Behavioral: Semantic prioritisation (faster RTs) persisted in both Valid and Neutral conditions but was stronger in the Neutral condition.
HDDM Findings:
- Non-decision time: The semantic advantage in $t$ was large in the Neutral condition (high retrieval demand) but eliminated in the Valid condition (low retrieval demand). This suggests the access advantage is driven by the difficulty of retrieving unattended items.
- Drift rate: Conversely, the semantic advantage in drift rate ( $v$ ) was amplified in the Valid condition. When retrieval was easy (Valid cue), attentional selection enhanced the efficiency of accumulating semantic evidence.
Conclusion: Semantic prioritisation is not a fixed bias but an adaptive process. When retrieval is hard (Neutral), semantic info is accessed faster. When retrieval is easy (Valid), attentional selection boosts the accumulation of semantic evidence.

Experiment 2 (Interference vs. Delay)

Behavioral: Semantic prioritisation in accuracy was significantly higher in the Prospective (interference) condition compared to the Immediate condition. The Delayed (time only) condition did not show a significant increase in prioritisation compared to Immediate.
HDDM Findings:
- Non-decision time: The semantic advantage in $t$ was significantly larger in the Prospective condition (interference) compared to both Immediate and Delayed conditions.
- Drift rate: No reliable differences in drift rate were found across conditions.
Conclusion: The enhancement of semantic prioritisation is driven specifically by attentional disruption/interference, not by the mere passage of time. When memory must be shielded from interference, the system shifts to a format where semantic information is more rapidly accessible.

4. Key Contributions

Mechanistic Dissociation: The study successfully dissociates the access phase (non-decision time) from the accumulation phase (drift rate) in working memory. It demonstrates that semantic prioritisation is primarily an access advantage (faster retrieval of semantic features) rather than a decision bias.
Role of Attentional State: It clarifies that semantic prioritisation is an adaptive response to attentional uncertainty. The advantage is most pronounced when items are outside the focus of attention and must be retrieved from an unattended state or shielded from interference.
Time vs. Interference: It provides evidence that temporal delay alone does not drive the shift to semantic formats; rather, the need to protect representations from interference is the critical factor.
Theoretical Integration: The findings bridge working memory and episodic memory literature, suggesting that when attention is withdrawn, WM representations shift toward abstract, long-term-memory-like formats that rely on top-down, reconstructive retrieval mechanisms.

5. Significance

This paper fundamentally shifts the understanding of working memory limitations. It suggests that the "capacity limit" of WM is not just about how many items can be stored, but about which representational formats remain accessible when attentional support is limited.

Cognitive Architecture: It supports models where unattended WM contents rely on mechanisms overlapping with episodic memory retrieval (activity-silent states), where semantic/conceptual features are more resilient and rapidly re-accessible than fine-grained perceptual details.
Adaptive Flexibility: The brain dynamically reconfigures the format of memory representations based on task demands. When attention is scarce or disrupted, the system prioritizes robust, abstract semantic information over fragile perceptual details to ensure behavioral adaptability.
Methodological Impact: The use of HDDM to separate retrieval latency from decision accumulation provides a powerful framework for future research into the temporal dynamics of memory access, moving beyond simple accuracy/RT measures.