Boundary-constrained chunking shapes local–global positional representations in working memory

Chunking is a key mechanism by which working memory (WM) compresses sequential information, yet little is known about how it reorganises positional representations. We examined when chunking produces stable and shared local-global positional structure in WM. Across four behavioural experiments, participants encoded nine-item sequences and later reported the original position of a probed item. The sequence types differed in the source and strength of chunk-defining boundaries: structured linguistic sequences, scrambled linguistic sequences, color-segmented shape sequences, and uniformly colored shape sequences. We characterised positional representations by applying parameterised behavioural representational similarity analysis to position-specific response distributions. Structured linguistic sequences produced a near-orthogonal geometry, indicating clearly separable local and global positional structure. Color-segmented shape sequences yielded a reduced-angle, reduced-scale geometry, suggesting weaker local-global organization; individual-level analyses further indicated that this intermediate group-level solution partly reflected participant-level heterogeneity. In contrast, scrambled linguistic and uniformly colored shape sequences were best described by one-dimensional solutions, implying that local and global dimensions were no longer separable when explicit boundary cues were absent. These findings suggest that chunking does not produce a uniform representational format in WM; instead, separable local-global structure depends on the source and strength of chunking cues.