Between-item similarity frees up working memory resources through compression: A domain-general property.

Abstract

Compression, the ability to recode information in a denser format, is a core property of working memory (WM). Previous studies have shown that the ability to compress information largely benefits WM performance. Importantly, recent evidence also suggests compression as freeing up WM resources, thus enhancing recall performance for other, less compressible information. Contrary to the traditional view positing that between-item similarity decreases WM performance, this study shows that between-item similarity can be used to free up WM resources through compression. Across a series of four experiments, we show that between-item similarity not only enhances recall performance for similar items themselves, but also for other, less compressible items within the same list, and this in the semantic (Experiment 1), phonological (Experiment 2), visuospatial (Experiment 3), and visual (Experiment 4) domains. Across these different domains, a consistent pattern of results emerged: between-item similarity proactively–but not retroactively–enhanced WM performance for other items, and this as compared with a condition in which between-item similarity at the whole-list level was minimized. We propose that between-item similarity in any domain may impact WM using the same underlying machinery: via a compression mechanism, which allows an efficient reallocation of WM resources.

Similarity-based compression in working memory: Implications for decay and refreshing models

“…Indeed, the results in Figure 5 show that the effect is relatively modest compared to what happens in humans (see Figure 2). These results nonetheless support the idea that the increased transposition errors induced by similarity may partially depend on the joint activation of similar items (Kowialiewski et al, 2022;Kowialiewski, Gorin, et al, 2021;Parmentier & Maybery, 2008). In the next section, we performed a more thoughtful exploration of the role of each compression parameter.…”

“…Finally, in no way this work advocates in favor of the theoretical superiority of the TBRS* architecture over other models. Other explanations are also plausible and have been discussed elsewhere (Kowialiewski et al, 2022).…”

“…Another phenomenon related to compression has recently been observed when analyzing the position of recalled items (Kowialiewski, Gorin, et al, 2021;Kowialiewski et al, 2022). It has been shown that similarity constrains the pattern of transposition errors.…”

“…J. Peterson & Berryhill, 2013;Quinlan & Cohen, 2012;Sanocki & Sulman, 2011), phonological (Fallon et al, 1999;Gupta et al, 2005;Nimmo & Roodenrys, 2004), visuospatial (De Lillo, 2004;Parmentier et al, 2005Parmentier et al, , 2006 and semantic WM AND SIMILARITY 4 (Monnier & Bonthoux, 2011;Neale & Tehan, 2007;Poirier & Saint-Aubin, 1995;Saint-Aubin & Poirier, 1999) domains, suggesting that this beneficial effect of similarity is a domain-general property of the human cognitive system. A recent work suggests an indirect benefit of compression (Kowialiewski et al, 2022): when the first items of a list to be remembered are similar, they are obviously better recalled than dissimilar items. Most importantly, subsequent items, although dissimilar from each other, also benefit from being preceded by similar items.…”

“…This phenomenon is illustrated in Figure 1. An example of the material for each domain used in Kowialiewski, Lemaire & Portrat (2022) is reported in Appendix A.…”