Enumeration takes time: Accuracy improves even after stimuli disappear

Numerical cognition is widespread among animal species and present from birth in humans. Laboratory studies show that longer stimulus duration improves numerical discrimination accuracy (Inglis & Gilmore, 2013; Wood & Spelke, 2005). Inglis and Gilmore (2013) suggested that longer durations allow subjects to resample the stimulus image multiple times, resulting in a more accurate final estimate. The current study tested this "multiple sampling" model alongside two competing models - "serial foveal accumulator" (Cheyette & Piantadosi, 2019) and "longer processing time" - to determine how stimulus duration relates to accuracy. Adult subjects completed a fully within-subject, 2AFC task in which they judged which of two arrays had more dots; accuracy was the dependent measure. Experiment 1 revealed higher accuracy in the 500 ms stimulus duration condition compared to the 100 ms condition, replicating previous results. Experiments 2 and 4 extended this finding. When stimulus duration was held constant at 100 ms, adding a 400 ms delay between stimulus offset and mask onset improved accuracy. Accuracy was similar in the 500 ms stimulus duration condition (E1) and the 100 ms duration + 400 ms mask delay conditions (E2 and E4), indicating that post-stimulus processing time improves accuracy, rather than stimulus duration per se. This contrasts with both the multiple sampling model and the serial foveal model, although we cannot exclude the possibility that sampling continues in iconic memory. Numerosity was a significant factor in all four experiments, suggesting a serial component in the enumeration process. These findings shed light on the cognitive processes underlying nonverbal number representation in humans.