Figure 4. Simulation 1: The impact of temporal sequence on number learning.

Panel A depicts a simulation of number learning in which object Features predict Labels (FL-learning), while Panel B depicts a simulation in which Labels predict Features (LF-learning). The models learned to associate sets of 2, 4 and 6 objects to the labels “two,” “four” and “six.” In addition to number, each object set had size, shape and color cues that competed as cues with set-size as predictors of number words. These graphs depict the value of mappings between the object features, set-sizes and the label “six” learned in each simulation. In FL-learning, uninformative cues are completely devalued as a result of cue competition, leading to enhanced discrimination.