Abstract
<p>This study investigates age-related changes in the mental lexicon using natural language data and network-analytic methods, evaluating whether observed differences reflect cognitive decline or cumulative lifelong learning. Prior research based on free association tasks has suggested that lexical networks become sparser with age; however, such methods may not capture language use in naturalistic contexts. To address this limitation, we analyzed 1,116 English novels written across the lifespan by 110 authors. Contextualized word embeddings were generated using a BERT-based model, and cosine similarity was used to estimate associative strength. Networks were constructed using two complementary thresholding approaches: an above-threshold approach (ATA) retaining only stronger associations and a below-threshold approach (BTA) retaining only weaker associations. In the ATA, age was associated with decreased degree and clustering coefficient and increased average shortest path length, consistent with sparsification. However, when weak-to-moderate associations were included (low-range ATA and BTA), the pattern reversed, indicating denser and more clustered networks with age. Entropy analyses further showed that age-related decreases emerged only when weak associations were excluded. These findings support an enrichment account in which aging expands the underlying associative space, while competitive interference during retrieval yields behaviorally sparser networks.</p>