The emergence of richly organized semantic knowledge from simple statistics: A synthetic review

Abstract

As adults, we draw upon our ample knowledge about the world to support such vital cognitive feats as using language, reasoning, retrieving knowledge relevant to our current goals, planning for the future, adapting to unexpected events, and navigating through the environment. Our knowledge readily supports these feats because it is not merely a collection of stored facts, but rather functions as an organized, semantic network of concepts connected by meaningful relations. How do the relations that fundamentally organize semantic concepts emerge with development? Here, we cast a spotlight on a potentially powerful but often overlooked driver of semantic organization: Rich statistical regularities that are ubiquitous in both language and visual input. In this synthetic review, we show that a driving role for statistical regularities is convergently supported by evidence from diverse fields, including computational modeling, statistical learning, and semantic development. Finally, we identify a number of key avenues of future research into how statistical regularities may drive the development of semantic organization.

Transformer networks of human conceptual knowledge.

“…Most relevant to this article, Bhatia (2017) has found that cosine similarity between word vectors provides a good measure of association in simple judgment tasks, and can thus predict association-based biases in probability judgment, factual judgment, and forecasting (Evans, 2008;Kahneman, 2011;Sloman, 1996). Finally, on the basis of these and related findings, some scholars have suggested that word distribution statistics also play a critical role in human semantic development (see Unger & Fisher, 2021, for a recent review).…”

Finding Structure in One Child's Linguistic Experience

“…On the other hand, language seems to follow a different path: words for superordinate categories are acquired comparatively late relative to words for basic-level categories (Murphy, 2016). Additionally, the developmental timecourse of taxonomic relatedness, compared to more associative and thematic forms of relatedness, is still debated and seems to vary according to the task (Gelman & Markman, 1986;Markman & Hutchinson, 1984;Sloutsky, Yim, Yao, & Dennis, 2017;Unger, Savic, & Sloutsky, 2020;Unger & Fisher, 2021). Our results suggest that information regarding taxonomic (including superordinate) categories can be readily extracted from a small subset of the linguistic input to one child (up to age 3), as found in other modeling work using broader aggregate data (Sloutsky et al, 2017).…”

Early-life language deprivation affects specific neural mechanisms of semantic representations

“…On the one hand, these findings about how language experience is critical for specific semantic neural structures (dATL) are in line with the rich evidence and discussion in the behavioral literature about the role of language in semantic development. For instance, developmental behavioral evidence shows that young human infants are sensitive to language labels and statistical patterns in forming semantic categories and other types of semantic relations (Perszyk and Waxman, 2018;Unger and Fisher, 2021;Spelke, 2017; see similar discussions in Gelman and Roberts, 2017; Lupyan et al, 2020). On the other hand, consistent with previous behavioral studies of semantic processing in delayed adult signers (Baus et al, 2008;Davidson and Mayberry, 2015;Skotara et al, 2012), in our study native and delayed signers exhibited similar semantic space for words that they are familiar with (Figure 1a) and did not significantly differ in two additional reaction time tasks (Supplementary file 1).…”