Statistical learning in visual search: ‘contextual cueing’ reflects the acquisition of an optimal, ‘one-for-all’ oculomotor scanning strategy

Seitz, Werner; Zinchenko, Artyom; Müller, Hermann; Geyer, Thomas · 2023 · Crossref

DOI: 10.21203/rs.3.rs-2561645/v1

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 mechanism behind "contextual cueing," a phenomenon where visual search for a target becomes more efficient when the target appears repeatedly within a stable arrangement of distractors. The standard account posits that observers acquire display-specific long-term memories (templates) that guide attention to the target location upon re-encountering a specific context. In contrast, the authors propose an alternative "procedural-optimization" account, suggesting that contextual facilitation arises from the acquisition of generic, task-general oculomotor scanning strategies. Under this view, repeated displays benefit more from learning simply because they accrue greater weight in optimizing a single, "one-for-all" scanning strategy, rather than through the formation of unique memory templates for each display. To distinguish between these accounts, the authors conducted an eye-tracking experiment with 46 participants performing a visual search task. Participants searched for a target letter "T" among distractor letters "L" in arrays that were either repeated (constant distractor layout) or non-repeated (random distractor layout). The experimental design ensured that all participants viewed the same set of repeated and non-repeated displays to allow for fine-grained comparisons. The study replicated standard measures, such as reaction times (RTs) and fixation counts, but primarily focused on novel scanpath similarity metrics: Dynamic Time Warping, Discrete Fréchet Distance, and Area Between Curves. These metrics quantified the spatio-temporal similarity of fixation sequences both across different participants viewing the same display and within individual participants viewing different displays. The results confirmed standard contextual cueing effects: participants exhibited faster reaction times, fewer fixations, shorter scanpath lengths, and reduced saccade variability for repeated compared to non-repeated displays. Crucially, the scanpath similarity analyses revealed that fixation sequences became increasingly similar over the course of the experiment for both repeated and non-repeated displays. However, scanpaths for repeated displays showed significantly higher similarity than those for non-repeated displays. This pattern held true whether comparing different participants viewing the same display or the same participant viewing different displays. These findings contradict the display-specific template hypothesis, which would predict that scanpaths for different repeated displays should diverge as unique templates are formed. Instead, the data support the procedural-optimization hypothesis, indicating that learning leads to a convergence toward a generic, optimal scanning strategy that is more heavily tuned to frequently encountered displays. The significance of these findings lies in challenging the dominant view that contextual cueing relies on stimulus-specific associative memories. By demonstrating that performance improvements are driven by the proceduralization of task-general visual search operations, the study suggests that training-related improvements in search tasks depend on changes in generic procedural memories rather than unspecific, individual display representations. This implies that the brain optimizes oculomotor scanning strategies globally across a set of displays, with statistical learning weighting frequent contexts more heavily in this optimization process.

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.

StageOutcomeToolModelPromptAttemptsCompleted
discover success Crossref 1 2026-06-11
archive success canonical_url 1 2026-06-25
extract success cached 2 2026-06-25
clean success clean 1 2026-06-11
chunk success chunk 1 2026-06-11
embed success embed Qwen/Qwen3-Embedding-8B 1 2026-06-11
promote success 1 2026-06-11
summarize success llm qwen3.6-27b-prismaquant summ-v5 1 2026-06-25
tag success vector_similarity 6 2026-06-11
verify success 1 2026-06-26

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.