Slide Slam Q5
Semantic networks across the lifespan: examining local network sensitivity and individual differences
Abigail Cosgrove1, Michele Diaz1; 1The Pennsylvania State University
Introduction Recent evidence suggests that semantic networks undergo age-related changes (Cosgrove et al., 2021; Dubossarsky et al., 2017; Wulff et al., 2019). Previous research has shown that older adults’ semantic networks had longer path lengths, less clustering, and higher modularity values compared to younger adults. These results indicate that younger adults have more efficient networks, less segregation of sub-communities, and greater flexibility for more efficient semantic processing (Cosgrove et al., 2021). Further, differences in retrieval processes can be quantified by measuring how connections between words in a semantic network break apart – the faster connections break apart the weaker the search processing. Such percolation analyses, have shown that older adults’ semantic networks broke down faster compared to younger adults’ networks, suggesting that younger adults’ semantic networks were more robust. However, previous work has focused on group-based global network characteristics, with limited focus on local node characteristics or individual differences. Methods To address these limitations, we conducted two different analyses. First, we examined age-related differences in the local characteristics of each node (i.e., word) from previously examined semantic networks (N = 78 older, 78 younger adults). Specifically, we focused on node centrality measures (degree and clustering coefficient, which represent the number of connections a word has with other words in the network, and the extent to which neighbors of a node are neighbors of each other, respectively. In a separate study (N = 30 older, 30 younger), individual semantic networks were created from semantic relatedness judgements (Benedek et al., 2017). Global network properties including efficiency (path length), connectivity (clustering coefficient), and community structure (modularity) were correlated with individual performance on a verbal fluency task. Results We observed age-related differences in the local node characteristics of degree and clustering coefficient (all p’s <.001), however there was no consistent pattern to these age-related differences. That is, some node degree values increased with age, while others decreased with age. Additionally, there were no significant age effects across individually-calculated network properties of average shortest path length, clustering coefficient, and modularity. These network characteristics also did not significantly correlate with individual differences in language production ability as measured by the total number of verbal fluency responses, number of category clusters, or number of switches between categories. Conclusion Consistent with prior research on global network measures between age groups, our analysis provides evidence of age-related differences in node-based measures, but with some nodes increasing and some decreasing. These findings reflect the sensitivity of processing within a semantic network, which can be influenced by changes in the local structural properties of certain nodes. Our individually calculated networks from semantic relatedness judgments did not show any age-related differences, or, relationships with language production measures. In contrast to our results, others have found individual differences in network structure similar to group findings (Wulff et al., 2018). These researchers found greater variance with increased age related to differences in life experiences among older adults which demonstrates the importance for future work to account for individual differences in the aging mental lexicon.