Fig 1.
A) Participants initially performed a Time Categorization (also called bisection) task in which a fixation point was followed by an image presented for one of 7 possible log-spaced durations between 300 and 900ms, which subjects then categorized as “short” or “long”. All participants returned the following day to perform a surprise memory test, in which stimuli from the previous day and a set of matched foils were presented with participants judging if they had seen the image before. B) Stimuli from the LaMem dataset with memorability and “speed” (as determined by the A parameter of our simple model); note that lower A values reflect “faster” speeds. Red points indicate those stimuli used in our previous report [21]. Three new sets of stimuli were sampled from these images along three different axes: “Slow Speed” stimuli which increase in memorability yet maintain a slow speed; “Constant Memorability” which increase in speed yet all have a memorability of 0.5; “High Speed + Memorability” which increase in memorability and all have respectively higher (fast) speeds. Images used are available at https://osf.io/5bfvh/.
Fig 2.
Separate and combined effects of speed and memorability on time estimates.
For the either the Slow Speed or Constant Memorability groups, higher memorability stimuli or faster speeds (more negative A parameter values) led to an increase in the probability of responding “long”. In contrast, for the High Speed + Memorability group, higher memorability stimuli led to more “long” responses while faster speeds led the opposite effect. Plotted points represent individual trial responses for each subject, whereas smooth lines represent fits from our generalized linear model.
Fig 3.
Memory recognition performance on the surprise memory test.
For the Slow Speed group, participants were more accurate for stimuli with higher memorability (higher bins). For the Constant Memorability group, where all stimuli had a memorability of 0.5, participants were more accurate at recognizing stimuli with faster speeds (higher bins) and less accurate at recognizing stimuli with slower speeds (lower bins). For the High Speed + Memorability group, participants were more accurate at recognizing images that increased in both memorability and speed (higher bins). Further, accuracy in this group was significantly higher than either the Slow Speed or Constant Memorability groups. Plotted points represent average proportion correct with shaded regions representing within-subject standard error.
Fig 4.
Relationship between speed and memorability.
A) Heatmap of mean proportion of “long” responses for the High Speed + Memorability group as a function of both memorability and speed, each of which were separated into seven equally-spaced bins. B) Same data as in (A) but presented as a line plot with each line representing a successive memorability bin. Within each memorability bin, faster speeds are associated with lower proportions of responding “long”, yet this effect becomes shallower as memorability increases; note also that higher memorability bins are associated with higher average proportions of “long” responses in general. C) Mean proportion “long” but for data combined from the High Speed + Memorability and Constant Memorability groups; these data comprise images at ~0.5 memorability, but across 11 speed bins. Here, the results follow an inverted-U shape with faster speeds leading to progressively higher proportions of responding “long” followed by an inflection – overlaid curve represents a quadratic fit. D) A simple model expanding the quadratic effect between speed and proportion “long” responses to other memorability scores suggests that time dilation/compression with speed depends on the mean level of memorability. This model can accommodate both higher proportions of “long” responses with higher memorability (increase in peak height) and the shift in direction with increasing speed.