Navon letters and composite faces: same or different processing mechanisms?

Abstract

Navon letters and composite faces are two fascinating demonstrations of hierarchical organization in perception. Many researchers believe that the two types of stimuli and their associated tasks gauge comparable holistic mechanisms. This belief is so common that the two paradigms are now being applied in tandem to measure impaired holistic processing in prosopagnosic patients. But are Navon letters and composite faces processed in a similar fashion? In the present study we take a closer look at their apparent affinity. We gain novel insights into their underlying mechanisms by fitting parameters of the linear ballistic accumulator (LBA) model to empirical correct and incorrect response times (RTs). The results reveal major differences in processing between the two tasks. We conclude that despite the presence of a compelling surface similarity, Navon compound letters and composite faces tap into separate psychological processes.

Keywords: LBA; Navon letters; composite face illusion; face recognition; holistic processing; response time models.

Figure 1

(Top) Examples of the Navon compound letters constructed from the English capital letters “H” and “S.” These stimuli were used in Experiment 1. (Bottom) Prototypical results from the Navon paradigm.

Figure 2

(Top) Schematic illustration of the “complete design” version of the composite face task. (Bottom) Prototypical results in this paradigm.

Figure 3

Application of the linear ballistic accumulator model (LBA) to decision in the Navon (A) and Composite face (B) tasks. Top panel: Two LBA accumulators for correct and incorrect responses, respectively, in the Navon task. Participants judge the local letter and ignore the global letter. Bottom panel: Two LBA accumulators for correct and incorrect responses, respectively, in the composite face task. Participants judge the top-half while ignoring the bottom-half.

Figure 4

Experiment 1: Mean correct RTs (left) and Percentage of Error (right) as a function of Congruity (congruent, incongruent) and Task (local, global). Error bars are standard errors of the mean. n.s, non significant; ^*p < 0.05, ^**p < 0.01, ^***p < 0.001.

Figure 5

Experiment 1: Quality of fit of the LBA model parameters to the data in judgments of the local letter (top) and in judgments of the global letter (bottom). Quartile-Quartile plots are depicted for the 1st, 2nd, and 3rd Quartiles in which empirical mean RTs (in seconds) are plotted against predicted mean RTs (in seconds).

Figure 6

Experiment 1: Mean LBA parameters. Top left: drift rates (v) as a function of Task and Congruity, Top right: non-decision times (Ter) as a function of Task and Congruity. Bottom left: starting point (A) parameter as a function of Task (local, global); Bottom right: threshold (b) parameter as a function of Task (local, global). Error bars are standard error of the mean. n.s, non significant; ^*p < 0.05, ^**p < 0.01, ^***p < 0.001.

Figure 7

Experiment 2: mean correct RTs (left) and percentage of error (right) as a function of Task (aligned, misaligned) and Congruity (congruent, incongruent) conditions. Error bars are standard errors of the mean. n.s, non significant; ^*p < 0.05, ^**p < 0.01, ^***p < 0.001.

Figure 8

Experiment 2: Quality of fit of the LBA model parameters to the data in judgments of aligned composite faces (top) and in judgments of misaligned composite faces (bottom). Quartile-Quartile plots are depicted for the 1st, 2nd, and 3rd Quartiles in which empirical mean RTs (in seconds) are plotted against predicted mean RTs (in seconds).

Figure 9

Experiment 2: mean LBA parameters. Top left: drift rates (v) as a function of Task (aligned, misaligned) and Congruity (congruent, incongruent), top right: non-decision times (Ter) as a function of Task (aligned, misaligned) and Congruity (congruent, incongruent). Bottom left: starting point (A) parameter as a function of Task (aligned, misaligned); bottom right: threshold (b) parameter as a function of Task (aligned, misaligned). Error bars are standard error of the mean. n.s, non significant; ^*p < 0.05, ^**p < 0.01, ^***p < 0.001.