Fig. 2.

Simplified illustration of probabilistic programming applied to human cognition, with pseudocode. (a) Our adaptation of the key ideas in [68]. Characters (e.g., “g” and “q”) are composed of a sequence of parts and subparts and their spatial relations (i.e., where the parts are joined). There is motor variance in the “writing” process, which may lead to visually similar, but conceptually different, character concepts. (b) Language understanding in the Rational Speech Acts framework [70] proceeds via nested reasoning: the pragmatic listener reasons about a nested speaker, which in turn reasons about a nested literal listener.