Education shapes the structure of semantic memory and impacts creative thinking

The study asks whether and how educational approaches shape the internal organization of children’s semantic memory and, in turn, impact higher cognition, particularly creative thinking. Montessori and traditional schooling differ in how concepts are learned (self-directed, interdisciplinary, discovery-based vs teacher-directed, sequential, and often rote). Building on evidence that the structure of semantic memory supports higher cognition and that flexible semantic networks are associated with creativity, the authors hypothesize that children in Montessori classrooms will show more enriched, interconnected, and flexible semantic networks (higher clustering, shorter paths, lower modularity) and higher creative thinking relative to peers in traditional classrooms. Groups were matched on socioeconomic status and nonverbal intelligence to isolate educational effects.

Prior work highlights that early experience strongly influences cognitive development and semantic knowledge acquisition, with socioeconomic status relating to vocabulary growth and semantic memory expanding substantially during elementary years. Cognitive network science provides tools to model semantic memory as networks of concepts (nodes) and their relations (edges), with metrics such as clustering coefficient, average shortest path length, and modularity indexing network organization. Adult studies show that more flexible, highly connected networks with shorter paths are linked to openness to experience and higher creative ability, while rigid, modular networks are associated with lower creativity. Montessori education has been associated with improved academic, socio-emotional, and creativity outcomes compared to traditional schooling. These lines of research suggest that educational environments could shape semantic memory structures with consequences for creativity, yet direct evidence in children has been lacking.

Design and participants: Cross-sectional study of 67 Swiss children (mean age 9.31 years, SD 2.23; 47.8% girls) enrolled either in Montessori (n=36) or traditional (n=31) classrooms from early years or for at least 3 years. Inclusion: age 5–14, enrollment in Montessori or traditional system; exclusion: parent-reported learning disabilities or sensory impairments. Groups were matched on socioeconomic status (SES) and nonverbal intelligence (short version of Raven’s Progressive Matrices). SES was computed from parents’ education (1–5) and job (1–4), summed and averaged across parents (max 9). No significant between-group differences were found for gender, age, nonverbal intelligence, or SES.

Tasks and measures:

• Verbal fluency (semantic): Children named as many animals as possible in 60 seconds; responses were recorded and transcribed. Preprocessing via the SemNA pipeline in R included collapsing repetitions/word variants and excluding non-category responses. Metrics: total number of responses and number of unique responses.
• Creativity assessment: Divergent and convergent creativity were measured using the EPOC (Evaluation of Potential Creativity). Divergent task: produce as many different drawings as possible from an incomplete abstract form within 10 minutes; score = count of valid drawings. Convergent task: select three abstract forms (out of eight) to create one original drawing within 15 minutes; three blind judges rated originality per EPOC manual (Krippendorff’s alpha = 0.905).

Semantic network construction:

• For each group, a binary response matrix (participants × unique exemplars) was created (1 = exemplar produced, 0 = not produced). To equate networks, only responses produced by at least two participants in each group were retained, and node sets were matched across groups.
• Word association matrices were computed using cosine similarity between exemplar vectors (range 0–1).
• Networks were filtered using the Triangulated Maximally Filtered Graph (TMFG) to retain the most informative edges while holding constant the number of edges across groups (3n−6). Final networks contained 68 nodes and 198 edges for both groups.

Network metrics:

• Clustering Coefficient (CC): interconnectivity among neighbors (higher = more interconnected network).
• Average Shortest Path Length (ASPL): average number of steps between node pairs (lower = more compact network).
• Modularity (Q): extent of partitioning into subcommunities (higher = more modular/segregated network).

Statistical analyses:

• Group comparisons on demographics/nonverbal intelligence: chi-square and t tests (Welch’s as needed).
• Verbal fluency and creativity: independent-samples t tests; Pearson correlations between creativity scores and total fluency responses.
• Network validation: (1) Comparisons against 1000 Erdös–Rényi random networks with same nodes/edges to test whether empirical CC, ASPL, Q differed from random; (2) Case-wise bootstrap (1000 resamples) to estimate sampling distributions of CC, ASPL, Q for each group, followed by independent-groups t tests on bootstrap distributions. Effect sizes reported as Cohen’s d (0.20 small, 0.50 moderate, 0.80 large).

Procedure: Ethical approval by CER-Vaud; written parental consent and child assent obtained. SES questionnaires completed online by parents. Tasks were administered individually at school in a quiet room; task order was randomized. Data and materials available upon request; study not preregistered.

• Groups were comparable on gender, age, nonverbal intelligence, and SES (no significant differences).
• Verbal fluency: Total number of animal names did not differ significantly (Montessori M=16.4, SD=5.0; Traditional M=14.8, SD=5.8; t=1.23, p=0.225, d=0.30). Montessori children produced more unique responses overall (105) than traditional (87), with more responses unique to Montessori (37) than unique to traditional (19); statistical test for unique responses p=0.023 (effect size reported as 0.28).
• Creativity: Montessori group scored higher on both measures: Divergent creativity M=9.6 (SD=3.7) vs Traditional M=7.1 (SD=5.1), t(64)=2.34, p=0.02, d=0.58; Convergent creativity M=5.3 (SD=1.7) vs 2.6 (SD=1.3), t(64)=7.10, p<0.001, d=1.74. Creativity scores correlated with total fluency responses: divergent r=0.27, p=0.03; convergent r=0.36, p=0.003.
• Semantic network structure: Empirical networks for both groups differed from random networks on CC, ASPL, and Q (all p<0.001). Bootstrap comparisons showed Montessori networks were more flexible: higher CC (M=0.707, SD=0.01) than traditional (M=0.701, SD=0.01), t(1998)=11.33, p<0.001, d=0.51; lower ASPL (M=3.07, SD=0.26) vs traditional (M=3.12), t(1998)=-5.32, p<0.001, d=0.24; lower modularity Q (M=0.582, SD=0.03) vs traditional (M=0.589, SD=0.02), t(1998)=-7.07, p<0.001, d=0.32. Visualizations showed Montessori networks to be more condensed and less modular.

Findings indicate that educational experience shapes children’s semantic memory organization and creative thinking. Children in Montessori environments exhibited more interconnected and less modular semantic networks, a structure theorized to facilitate access to and combination of remotely associated concepts, aligning with their higher divergent and convergent creativity scores. As groups were matched on SES and nonverbal intelligence, differences are attributable to educational context rather than extraneous factors. The results likely reflect pedagogical differences: Montessori’s interdisciplinary, self-directed, discovery-oriented learning, peer-to-peer teaching in multi-age settings, uninterrupted work periods, and active learning may cultivate dynamic, richly connected conceptual representations and foster creativity. These outcomes mirror adult findings linking flexible semantic networks to creativity. The study underscores the cognitive significance of educational practices and motivates investigation into mechanisms linking classroom practices, semantic network development, and higher cognition.

This study provides initial evidence that schooling practices shape the structure of semantic memory and influence creative thinking in children. Compared to traditional classrooms, Montessori education is associated with more flexible semantic networks (higher clustering, shorter paths, lower modularity) and superior performance on divergent and convergent creativity tasks. These results suggest that child-centered, interdisciplinary, and self-directed learning may enrich conceptual integration with downstream benefits for creativity. Future research should use longitudinal designs to track how semantic networks evolve with different educational experiences, examine additional semantic domains beyond animals, identify specific pedagogical features driving these effects, link individual-level network metrics to creativity and other cognitive abilities, and test potential transfer to out-of-class knowledge acquisition.

• Cross-sectional design prevents causal inference; longitudinal studies are needed to establish developmental trajectories and causality.
• Semantic networks were built from a single category (animals); generalizability to other domains is unknown.
• Group-based network modeling precluded linking individual semantic network metrics to individual creativity scores.
• The specific Montessori features responsible for observed differences were not isolated; multiple classroom factors may contribute.
• Although groups were matched on SES and nonverbal intelligence, unmeasured confounds may remain.