Content-agnostic online segmentation as a core operation

We approach the problem of explaining segmentation --- the human capacity to partition input streams into representations of appropriate form and content for efficient downstream processing --- by exploring a theoretically minimalistic and computationally plausible account of phoneme-to-word chunking. Through computational models, mathematical proofs, algorithm design, and observer model simulations in two languages, we suggest that online segmentation can be guided by content-agnostic properties of internal memory structures (i.e., lexicality and length type frequency). Our theoretical and empirical findings point to a formal link between such properties with practical performance benefits. Together, these contributions make progress on a fully explicit computational- and algorithmic-level account with plausible implementational-level primitives.