Further evidence for the spread of attention during contour grouping: A reply to Crundall, Dewhurst, and Underwood (2008)

Pieter R. Roelfsema, Roos Houtkamp, Ilia KorjoukovAttention 2010 · DOI

Introduction

How does the human brain parse a cluttered visual scene? Consider the common task of identifying which plug belongs to which device in a tangled mess of cables behind a desk. This process, known as "contour grouping," is fundamental to visual perception.

Figure 1: A real-world example of contour grouping: identifying which cable belongs to which device in a cluttered environment.

There are two primary competing theories for how this is achieved:

  1. The Spreading-Attention Model: Proposes that attention begins at one end of a curve and spreads along its length, effectively "highlighting" or labeling the entire object in the visual cortex.
  2. The Moving Zoom Lens Model: Suggests attention acts as a mobile spotlight, moving along the path and focusing on one segment at a time, leaving previous segments unattended.

This paper defends the spreading-attention model against challenges raised by Crundall et al. (2008), who argued that attention moves away from the starting point of a curve as tracing progresses. Roelfsema and colleagues argue that Crundall's conclusions were based on a flawed experimental metric.

Experimental Methods

To test these models, researchers conducted a psychophysics experiment with five subjects. The core task involved "curve-tracing" where subjects identified which shape was at the end of a target curve connected to a central fixation point.

Figure 5: The experimental paradigm showing the sequence of stimulus, mask, and potential change detection (Stimulus 2).

The critical innovation was a direct change-detection task. During the trial, a gray mask would flicker, and in 50% of trials, the curves would undergo a structural change (e.g., an intersection appearing or disappearing). Unlike previous studies that relied on whether subjects updated their final answer (the "response reversal" metric), this study asked subjects directly if they noticed a change. This distinction is vital: a subject might see a change but lack the time to re-calculate their answer before the trial ends.

Key Insights and Results

The study revealed several critical findings that support the spreading-attention model over the zoom lens theory:

The Response Reversal Fallacy

The researchers demonstrated that failing to change a final answer does not imply a failure to perceive the change. Tracing a curve is a serial process, adding approximately 100ms per intersection. If a change occurs late in a trial, subjects often detect it but cannot complete the "re-tracing" required to update their response in time. At long delays, subjects detected changes at the start of the curve with 86% accuracy when asked directly, even when their accuracy in the primary tracing task dropped to 74%.

Persistent Labeling and Object-Based Selection

Subjects maintained high sensitivity to changes at the beginning of the curve, even when their focus was directed toward the end. This suggests that the brain maintains an active "tag" or "label" across the entire selected object. Furthermore, while subjects were highly accurate at detecting changes on the target curve, they were nearly blind to changes on irrelevant curves (~9% accuracy, matching the false-alarm rate). This confirms that attention is object-based; once an object is selected, its entire extent is prioritized.

Conclusion

The results strongly favor the spreading-attention model. Attention does not merely move like a spotlight; it spreads to encompass the entire object, labeling it as a single entity. This persistent highlighting allows the brain to maintain awareness of the whole object even while processing specific segments, providing a robust mechanism for navigating complex visual environments.

Summary updated

BibTeX
@inproceedings{roelfsema2010further,
  author = {Pieter R. Roelfsema and Roos Houtkamp and Ilia Korjoukov},
  title = {{Further evidence for the spread of attention during contour grouping: A reply to Crundall, Dewhurst, and Underwood (2008)}},
  booktitle = {Attention},
  year = {2010},
  doi = {10.3758/app.72.3.849},
}
Figure