Representations of geometric shapes have syntactic structure

Recent research suggests that humans use language-like mental representations for manystimuli, from auditory sequences to visual shapes. However, evidence has been largelyindirect, relying on stimulus compression as a proxy for internal representation. Usingconstituency tests, we probed representational structure in the domain of geometry moredirectly. Across three preregistered experiments (n= 136), we find robust evidence for treestructure in human adults’ shape representations. First, the same shape can receivedifferent structural representations depending on how a preceding animation organizes it.Second, subparts of shapes are easier to detect when they belong to the same subtree thanwhen spanning different subtrees. Third, shape fragments are easier to reconfigure thehigher in the tree they are split. Unlike humans, state-of-the-art deep networks show nosyntactic effects whatsoever. Thus, humans—and so far only humans—encode geometricshapes in hierarchical structures, mirroring the representations used in natural languageprocessing.