Trial-Level Sequence Modeling Reveals Hidden Dynamics of Dual-Task Interference

den Otter, Rick; Dame, Anna; Stuit, Sjoerd; van Maanen, Leendert · 2025 · Crossref

DOI: 10.1101/2025.11.14.688220

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the neural dynamics of dual-task interference, specifically testing the long-standing assumption that the same cognitive operations underlie multitasking regardless of stimulus timing. Traditional theories, such as the central bottleneck and capacity sharing models, rely on behavioral averaging which obscures trial-by-trial variability. The authors address this limitation by combining Hidden Multivariate Pattern (HMP) analysis with deep spatiotemporal sequence modeling of single-trial EEG data within the psychological refractory period (PRP) paradigm. The goal was to determine if cognitive operations remain consistent across conditions and to identify how the specific sequencing of these operations influences performance. The researchers analyzed an open dataset comprising 24 participants performing a visual flanker task (Task 1) followed by an auditory pitch discrimination task (Task 2). Stimulus-onset asynchrony (SOA) was manipulated into Long (1200 ms) and Short (300 ms) conditions. Using HMP on the non-overlapping Long SOA trials, the authors identified three distinct cognitive operations for each task: Encoding, Central, and Response. They then trained a deep spatiotemporal sequence model based on the Mamba architecture on these Long SOA trials to decode single-trial EEG patterns. This model was subsequently applied to the Short SOA condition to identify the presence and timing of these operations without retraining, allowing for a direct comparison of neural representations across conditions. The results demonstrated that the same Encoding, Central, and Response operations identified in the Long SOA condition were present in the Short SOA condition, evidenced by high representational similarity in the model’s embedding space. However, the sequence of these operations varied significantly. While the Long condition predominantly featured a serial sequence (Task 1 completes before Task 2 begins), the Short condition exhibited six distinct sequences, including five where Task 2 operations initiated before Task 1 was complete. Crucially, these specific sequences predicted behavioral outcomes. Linear mixed-effects models revealed that the sequence type significantly influenced Task 1 reaction times and accuracy, as well as Task 2 reaction times. This indicates that the precise timing of interference within the cognitive operation sequence directly affects performance efficiency. These findings challenge static bottleneck accounts of multitasking by revealing that dual-task interference is dynamic and heterogeneous. The study establishes that while the fundamental cognitive building blocks remain stable, their temporal arrangement varies across trials and individuals, directly impacting behavioral performance. By demonstrating that trial-level sequence modeling can uncover these hidden dynamics, the authors provide a powerful methodological tool for investigating the nuanced, adaptive nature of multitasking, moving beyond the limitations of averaged behavioral data.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• dual task multitasking
• dual task performance
• task switching